Split (1)

train · 521k rows

text stringlengths 254 1.16M
--- title: 'Maternal androgen excess significantly impairs sexual behavior in male and female mouse offspring: Perspective for a biological origin of sexual dysfunction in PCOS' authors: - Nina M. Donaldson - Melanie Prescott - Amy Ruddenklau - Rebecca E. Campbell - Elodie Desroziers journal: Frontiers in Endocrinology year: 2023 pmcid: PMC9975579 doi: 10.3389/fendo.2023.1116482 license: CC BY 4.0 --- # Maternal androgen excess significantly impairs sexual behavior in male and female mouse offspring: Perspective for a biological origin of sexual dysfunction in PCOS ## Abstract ### Introduction Polycystic ovary syndrome (PCOS) is the most common infertility disorder worldwide, typically characterised by high circulating androgen levels, oligo- or anovulation, and polycystic ovarian morphology. Sexual dysfunction, including decreased sexual desire and increased sexual dissatisfaction, is also reported by women with PCOS. The origins of these sexual difficulties remain largely unidentified. To investigate potential biological origins of sexual dysfunction in PCOS patients, we asked whether the well-characterized, prenatally androgenized (PNA) mouse model of PCOS exhibits modified sex behaviours and whether central brain circuits associated with female sex behaviour are differentially regulated. As a male equivalent of PCOS is reported in the brothers of women with PCOS, we also investigated the impact of maternal androgen excess on the sex behaviour of male siblings. ### Methods Adult male and female offspring of dams exposed to dihydrotestosterone (PNAM/PNAF) or an oil vehicle (VEH) from gestational days 16 to 18 were tested for a suite of sex-specific behaviours. ### Results PNAM showed a reduction in their mounting capabilities, however, most of PNAM where able to reach ejaculation by the end of the test similar to the VEH control males. In contrast, PNAF exhibited a significant impairment in the female-typical sexual behaviour, lordosis. Interestingly, while neuronal activation was largely similar between PNAF and VEH females, impaired lordosis behaviour in PNAF was unexpectedly associated with decreased neuronal activation in the dorsomedial hypothalamic nucleus (DMH). ### Conclusion Taken together, these data link prenatal androgen exposure that drives a PCOS-like phenotype with altered sexual behaviours in both sexes. ## Introduction Fertility and sexuality are controlled by the brain and dependent upon complex neuronal circuits that are organized early in life and then activated by sex steroid hormones in adulthood. Prenatal exposure to testosterone, aromatized into oestradiol in the brain, is required for the development of male-typical brain circuitry and behaviors in rodents [1]. Over the last 10 years, the dogma that the female brain develops by default in the absence of sex steroids has been challenged with the discovery that peri-pubertal oestradiol and progesterone are also required for feminization of the brain and sexual behaviors (2–5). Noteworthy, the role of androgen signaling through androgen receptor (AR) on the development of the brain and behavior has been recently highlighted in the male through the study of the neural deletion of AR in male mice showing a central role for AR in the development of proper male copulatory behaviors [6, 7]. However, the role of AR-mediated signaling remains poorly investigated in the female despite AR being present in the developing female brain and behavior [8]. In addition, very little is currently understood about male- and female-typical sexual behaviors that are programmed by in utero androgen excess states such as in PCOS [9, 10], congenital adrenal hyperplasia (CAH) [11, 12] and in the case of environmental exposure to androgenic compounds (13–15). Polycystic ovary syndrome (PCOS) is a common endocrine disorder characterized by androgen excess. PCOS affects roughly 1 in 8 women of reproductive age and is the most common form of anovulatory infertility [16]. In addition to hyperandrogenism, PCOS is characterized by menstrual irregularities and polycystic ovarian morphology (16–18) and is associated with a wide range of comorbidities, including obesity, diabetes and cardiovascular disease [16]. The hypothalamo-pituitary-gonadal (HPG) axis, that controls fertility and reproductive behavior, is disrupted in many women with PCOS. In particular, luteinizing hormone (LH) pulse frequency, which mirrors gonadotropin-releasing hormone (GnRH) neuron activity and secretion, is significantly elevated. An elevated LH to follicle stimulating hormone (FSH) ratio contributes downstream to polycystic ovarian morphology, elevated androgen production and infertility [19, 20]. Steroid hormone feedback to the HPG axis, that would ordinarily slow GnRH/LH secretion is diminished in PCOS patients, suggesting that PCOS originates from a miscommunication between the brain and the ovaries (20–22). The etiopathogenesis of PCOS is most likely multifactorial with genetic susceptibility and environmental exposure playing predominant roles [9]. In line with this, recent studies in men suggest the existence of a male PCOS equivalent in the brothers of women with PCOS (23–26). These men share common endocrine, metabolic and cardiovascular comorbidities with their sisters such as an elevated free androgen index, a low level of FSH leading to an elevated LH/FSH ratio, insulin resistance, type II diabetes and hypertension [23, 25, 26]. Among the current hypotheses of PCOS origins, in utero androgen excess has been highlighted by human and animal-based study as a substantial contributor [9, 27]. Indeed, pregnant women with PCOS show high levels of circulating androgens during gestation, and this is correlated with an increased likelihood of having a daughter diagnosed with PCOS [28]. Maternal androgen excess has also been linked to the development of PCOS-like features in a wide range of female mammalian species [27]. For example, exposure of female mice to elevated levels of the non-aromatisable androgen dihydrotestosterone (DHT) during late gestation programs the development of hyperandrogenism, irregular oestrous cycles and theca cell hyperplasia [29]. These prenatally-androgenized (PNA) female mice exhibit impaired steroid hormone feedback associated with reduced progesterone receptor (PR) expression, elevated LH pulse frequency [29, 30] and elevated GnRH neuronal activity [31], associated with programmed changes in the GnRH neuronal network (29–33). There is some evidence indicating that prenatal androgen excess also alters reproductive function in males. Rams born to mothers exposed to testosterone propionate or dihydrotestosterone exhibit altered testicular function and disrupted neuroendocrine axis function in adulthood (34–38). However, similar disruptions do not appear to be evident in the male siblings of PNA mice modeling PCOS [39]. Epidemiological studies indicate that women with PCOS are more likely to experience sexual dysfunction, including low sex drive and sexual dissatisfaction, that can negatively impact their quality of life (40–48). Interestingly, men diagnosed with early onset androgenetic alopecia (AGA), now considered as a clinical sign of male PCOS equivalent [23, 24, 26], also indicate experiencing sexual dysfunction [49, 50]. In women, the cause of PCOS-related sexual dysfunction is frequently attributed to psychological factors, including reduced self-esteem related to hirsutism and/or obesity, a higher prevalence of anxiety, depression and mood disorders and decreased interest in sexual activities due to infertility issues (40–42, 44–46, 48, 51–56). Similarly, in men with AGA, the early onset of baldness is also often discussed in epidemiological studies as a potential factor for sexual dysfunction [49, 50]. However, it is not unreasonable to imagine that prenatal androgen exposure that is associated with the programming of PCOS-like reproductive features also might impact the development of male and female sexual behaviors. Obviously, human sexual behavior is incredibly complex and difficult to model, however, the PNA mouse model of PCOS, exposed to the non-aromatisable androgen dihydrotestosterone, provides a powerful reductionist approach to tease apart whether prenatal androgen exposure is associated with changes in sex behavior in male and female mice. To investigate how prenatal androgen excess impacts adult sexual behaviors in female and male mice, we performed a series of sexual behavior tests in PNA male (PNAM) and female (PNAF) mice. Finding a significant impairment in PNAF mice, we then further investigated the potential neural substrates involved in the PNA-related female sexual dysfunction. ## Animals Male and female C57BL/6J mice were generated and housed in the Otago Biomedical Research Facility at the University of Otago until adulthood. Mice were kept under a 12 h light/dark cycle with food and water ad libitum. All mice were kept in same-sex housing from weaning and hence were not exposed to the opposite sex before sexual behavior testing. Adult mice were moved to the Otago Behavioural Phenotyping Unit (BPU) for subsequent behavioral testing. In the BPU, mice were kept under a 12 h reverse light/dark cycle with food and water ad libitum. Sodium lamps permitted observation of the mice during the dark phase. All protocols were approved by the University of Otago Animal Ethics Committee. ## Generation of prenatally-androgenized mice modeling PCOS Control (VEH) and prenatally-androgenized (PNA) male and female mice were generated using the well-characterized prenatally androgenized (PNA) mouse model protocol (29–33, 57). Adult male and female C57BL/6J mice were paired overnight on the day of proestrus. Gestational day 1 was recorded as the following day after overnight mating and the male was removed from the cage. Females were then monitored for signs of pregnancy such as increased body weight and increased belly circumference. From gestational day 16-19, pregnant dams received a daily subcutaneous (s.c.) injection in the nape of the neck of either 100 µL dihydrotestosterone (DHT, 250 µg/100µL) in sesame oil as the PNA treatment or 100 µL of sesame oil only as the vehicle control. This window of prenatal androgen exposure has been shown to lead to PCOS-like features in mice and largely avoid the critical period for the differentiation of external genitalia (29–33, 57).The male (M) and female (F) offspring of dams injected with DHT (PNAM or PNAF) and vehicle control (VEH) mice were studied from adulthood (postnatal day (PND) 60 onward) in the following experimental protocols. Oestrous cyclicity of VEH and PNA female mice was assessed to establish the expected loss of oestrous cyclicity in PNA mice by collecting daily vaginal smears over a 20-day period (PND 60–80) (Figure S1) as previously described (29–31, 33, 58). ## Experiment 1: Phenotyping male and female-typical sexual behaviors in prenatally-androgenized mice A cohort of C57Bl6 male ($$n = 9$$-12/group) and female ($$n = 6$$-11/group) control and prenatally-androgenized (PNA) mice underwent the following behavioral tests as previously described (3–5) (Figure 1). Figure 1: Experimental design to study the effect of prenatal DHT exposure on sexual behaviors in male and female mice. In this study, we used two batches of mice. (A) The first batch of mice was used to determine if male and female-typical sexual behaviors could be affected by prenatal androgen exposure (experiment 1 and 2). Dams were treated with either dihydrotestosterone (DHT) or oil vehicle (VEH) at gestational day (GD) 16, 17 and 18. After birth, the male and female offspring entered the experimental protocol at adulthood around postnatal day 60 (PND60). PNAF and VEH females (magenta line) were ovariectomized and implanted with a silastic capsule containing estradiol in order to normalize hormonal status and artificially induced receptivity (OVX+E2). PNAM and VEH male mice (blue line) were left intact. Then, the animals entered a series of behavioral procedures to test their anxiety-like behavior and locomotion within an open-field (OPF) and an elevated-plus maze (EPM) tests, their abilities to discriminate a sexual partner within a three-compartments box partner preference test (3CB) and their typical sexual behavior (SB, i.e. lordosis behavior for female over 6 consecutive weeks and male copulatory behavior within 2 consecutive weeks). Finally, the PNAF and VEH female mice had their estradiol implant replaced by an implant containing testosterone (OVX+T) in order to activate and therefore test male-like sexual behavior (MSB) and verify if prenatal DHT exposure could have led to masculinization of the brain. (B) The second batch of mice was used to identify the neural target of prenatal androgen in female mice associated with impaired lordosis behavior. Only the female offsprings were used in this experiment 3. The PNAF and VEH female mice underwent ovariectomies and implantation of a silastic capsule of estradiol followed by 6 consecutive weeks of lordosis behavior testing. At the last test, PNAF and VEH female mice were euthanized 90 min after the beginning of the test in order to study neural activation induced by lordosis behavior. VEH, control male; PNAF, prenatally androgenized female; PNAM, prenatally androgenized male; PND, postnatal day; OPF, open-field test; EPM, elevated plus-maze; 3CB, three-chamber test; SB, sexual behavior test; MSB, male-like sexual behavior. ## Anxiety and locomotion tests Male and female-typical sexual behavior in mice are highly impacted by anxiety and dependant upon locomotor activity. Since previous studies in the PNA mouse model highlighted anxiety-like behaviors during the diurnal/inactive phase (59–61), we decided to test the basal level of anxiety and locomotion in PNA mice under the same conditions used for testing sexual behavior: during the nocturnal/active phase and under a sodium light imperceptible by mice. Mice were tested in Open-field tests and Elevated-plus maze tests to determine basal locomotor activity and anxiety as detailed below. Open-field test: Mice were placed in the center of a plexiglass aquarium (40 x 40 x 30 cm) under sodium lamps and their movements were recorded for 10 min. Between each animal, the aquarium was cleaned using $10\%$ ethanol. Video recordings were analyzed using TopScan® software. Two virtual zones of the aquarium floor were demarcated: the center zone ($60\%$ of the total floor area) and the periphery zone (surrounding area). TopScan® recorded the distance travelled by the mice and the amount of time mice spent in each zone over 10 min (600 s). Locomotion was measured as the distance travelled (mm). Anxiety behavior was determined by comparing the time (s) spent in the center versus peripheral zone, with less time in the center reflecting heightened anxiety. Elevated-plus maze: The maze apparatus was comprised of four arms (30 cm long, 5 cm wide) elevated 40 cm above the ground by metal legs and arranged in a plus shape. Two arms were open without walls and two arms were enclosed by high walls (15 cm). Mice were placed at the junction of the four arms (center) and their movements were recorded for 10 min. Between each animal, the maze was cleaned using $10\%$ ethanol. Video recordings were analyzed using TopScan®. Each of the four arms and the center zone were virtually demarcated. The distance travelled by the mice and the amount of time spent in each arm and center over the 10 min (600 s) period was traced and recorded. Locomotion was measured as the distance travelled (mm) and anxiety behavior was measured by the time (s) spent in the center, the open arms and the closed arms, with less time in the open arms reflecting heightened anxiety. ## Ovariectomies and hormones replacement for induction of female receptivity All female mice in this study were ovariectomized and implanted with a silastic capsule of oestradiol as previously described (3–5). This ensures uniform hormone concentrations across females and mimics an oestrus hormonal level of estradiol in order to artificially trigger receptivity (3–5). Briefly, all females were bilaterally ovariectomized under general anaesthesia with isofluorane. At the same time, a 5-mm-long Silastic capsule (inner diameter: 1.57 mm; outer diameter: 2.41 mm) containing crystalline 17β-oestradiol (E2758, Sigma-Aldrich, USA) (diluted 1:1 with cholesterol (C8667, Sigma-Aldrich, USA)) was inserted under the skin at the nape of the neck to induce oestrous levels of oestradiol. Mice received Carprofen (5 mg/kg) and were allowed to recover for two weeks before the onset of behavioral tests. On the day of testing, the females (either tests or stimuli) were administered progesterone (P) (P0130, Sigma-Aldrich, USA) (500 µg/mL, s.c.) 3 hours prior to test commencement. ## Male and female partner discrimination test Three-Chambers Box Partner preference: A partner preference test was carried out in virgin animals prior to sexual behavior experiments. Partner preference testing was conducted in a plexiglass box divided into three compartments (60 x 13 x 30 cm) by opaque partitions with fenestrations that allow for odours to diffuse throughout the arena. The day before testing, test females were allowed to habituate in the central compartment for 10 min. On the day of testing, females were administered P (500 µg/mL s.c.) 3 h prior to test commencement. For the test, two other mice, an intact sexually experienced stimulus male and an oestrous stimulus female (OVX + E2 female injected with P 3 h prior to the experiment, OVX+E+P) were placed in the two lateral compartments with their own bedding in order to make their respective compartments as odorous as possible. The test animal (PNAM/PNAF or VEH counterpart), was then placed in the middle compartment containing no bedding and observed for 10 min (the experimenter was blinded to group at the time of experiment). The time (s) that the test female spent actively sniffing each partition was recorded with a stopwatch. Between each test, the middle compartment was cleaned using $10\%$ ethanol to eliminate the previous test subject’s odours. A preference score was calculated by dividing the time spent investigating the male compartment minus the time spent investigating the female compartment by the total time spent investigating both compartments. A positive value of the preference score indicates a partner preference toward the stimulus male, whereas a negative value of the preference score indicates a partner preference toward the stimulus female. ## Male sexual behaviors All male mice were gonadally intact and sexual behavior tests were conducted in a transparent plexiglass aquarium (35 x 25 x 19 cm) filled with a layer of fresh sawdust as previously described (3–5). At the beginning of each test, the male was placed alone in the cage and allowed to adapt for 15 min. A receptive female (OVX+E+P) was then introduced into the cage and the latencies to the first mount and intromission, the latency to ejaculate, as well as the number of mounts, intromissions, and pelvic thrusts, were recorded. The test lasted until ejaculation occurred or 30 min if no ejaculation was achieved. If a male never displayed a certain behavior within the 30 min test, the latency was scored as 1800 s. ## Female sexual behaviors Female typical lordosis behavior: The lordosis behavior test was carried out over six consecutive weeks allowing seven days of rest between each test. Lordosis behavior testing was conducted in the same Plexiglas aquarium as described above. For the test, a sexually experienced male was placed in the aquarium and allowed to habituate for 15 min. Subsequently, a test female was introduced to the aquarium and the pair was observed. The number of mounts exhibited by the male and the number of lordosis behavior displays exhibited by the female was recorded for 15 min. The lordosis quotient corresponds to the number of lordosis postures recorded following a trial mount by the male, divided by the number of mounts attempted by the male throughout the duration of the test, and multiplied by 100 (i.e. (number of lordosis/numbers of male mount trials) * 100). Anogenital sniffing/aggressive behaviors: During all lordosis behavior tests, bouts of anogenital investigation and aggressive behavior toward the male were counted. Bouts of anogenital sniffing was recorded every time the female nose was in contact with the male genitalia. Bouts of aggressive behaviors were recorded every time the female attempted to kick or to bite the male. ## Experiment 2: Determining if prenatal androgen excess drives masculinized sex behavior Following the female typical sexual behavior tests, a random subset of VEH female ($$n = 6$$) and PNAF ($$n = 5$$) mice were anesthetized again with isofluorane in order to remove the oestradiol implant, and replaced it with a testosterone implant as previously described [4]. The testosterone implant was made of a 5-mm-long Silastic capsule (inner diameter: 1.57 mm; outer diameter: 2.41 mm) filled with crystalline testosterone (T1875, Sigma-Aldrich, USA) diluted 1:1 with cholesterol (C8667, Sigma-Aldrich, USA). This procedure mimics typical testosterone levels of adult male mice by 2 weeks and can induce male-like sexual behavior in female mouse [4]. Mice received Carprofen (5mg/kg) and were allowed to recover for 10 days before additional behavior testing. Following recovery, male-like sexual behavior (i.e. mounting, intromission and pelvic thrust) was tested in a plexiglass aquarium (35 x 25 x 19 cm) filled with a layer of fresh sawdust as previously described [62]. Briefly, a test female (PNAF or VEH) was placed alone in the aquarium to habituate for 15 min. Subsequently, an oestrous stimulus female (OVX + E + P) was introduced to the aquarium and the pair was observed. The initial latency to mount, number of mounts and number of pelvic thrusting movements shown by the test female (PNAF or VEH) were scored over 30 min. ## Experiment 3: Mapping sex behavior-related neuronal activation in prenatally-androgenized mice A second cohort of VEH female ($$n = 5$$) and PNAF ($$n = 5$$) mice were tested for lordosis behavior as described above over 6 consecutive weeks. After the last lordosis test, i.e. week 6, female mice were euthanized 90 minutes after the beginning of the lordosis test in order to study neuronal activation following lordosis behavior. In addition, we also euthanized a group of adult females C57Bl6 mice (Basal $$n = 3$$-4) who went through the same procedure as the VEH and PNAF mice except that no male stimulus was introduced into the aquarium for the last lordosis behavior test. This basal group allowed us to determine the basal neural activation without lordosis behavior. ## Tissue processing for immunostaining Upon completion of all behavioral tests, female mice were anesthetized with a lethal i.p. injection of pentobarbital (150 mg/kg/mice) and perfused transcardially with $4\%$ cold paraformaldehyde. Brains were removed and post-fixed in $4\%$ paraformaldehyde overnight. Brains were then cryoprotected in $30\%$ sucrose/tris-buffered saline (TBS) solution over 72 h. Free-floating brain sections were cut at 30µm-thickness on a freezing microtome and collected in cryoprotectant. Forebrains were cut coronally from the rostral telencephalon to the posterior hypothalamus. Sections were collected in four different series and stored at -20°C until immunostaining. ## Immunohistochemistry procedures All the following immunohistochemistry procedures were performed on brain sections from the second cohort of female VEH ($$n = 5$$) and PNAF ($$n = 5$$) mice after lordosis behavioral testing. One set of sections (i.e every fourth section) was rinsed for 10 minutes in TBS six times to remove cryoprotectant. Sections were then treated with $3\%$H2O2 and $40\%$ methanol in TBS for 10 minutes to quench endogenous peroxidases, and then washed a further three times in TBS. Sections were incubated in blocking solution with $2\%$ normal donkey/goat serum and $1\%$ bovine serum albumin in TBST (TBS + $0.3\%$ Triton-X) for 1 hour. Sections were then incubated with primary antibodies against Kisspeptin, Progesterone Receptor and cFOS (Table 1) for 96 hours at 4 degrees. Then, the sections were washed again three times 10 minutes before incubation with IgG biotinylated secondary antibodies (donkey anti-sheep IgG biotinylated/goat anti-rabbit IgG biotinylated) diluted in blocking solution for 90 minutes at room temperature. Sections were then washed before being incubated with the avidin-biotin complex diluted in TBST ($\frac{1}{200}$, ABC, Vector Laboratory, Burlingham, CA). To finish, after rinsing, sections were reacted with 3,3’diaminobenzidine tetrahydrochloride in TBS with nickel ammonium sulfate and glucose oxidase. After a last wash, the sections were mounted onto gelatin-coated slides, dried for 48h, dehydrated in ethanol followed by xylene and then coverslipped with DPX. Table 1 \| Proteins of interest \| Primary antibodies \| Dilutions \| \| --- \| --- \| --- \| \| Kisspeptin \| #Kp052, INRAe \| 1:10000 \| \| Progesterone \| #63605, Abcam \| 1:800 \| \| cFOS \| #sc-52, Santa Cruz \| 1:1000 \| ## Image analysis Photomicrographs of cFOS and PR immunohistochemistry in different brain regions were captured using a 20x objective on a bright field microscope (Olympus BX51). The representative images of each brain region analyzed were identified using the mouse brain atlas from Paxinos and Franklin, 3rd Edition. An experimenter blinded to treatment counted the number of cFOS- and PR-immunoreactive nuclei using Image J software®. For the kisspeptin immunohistochemistry, photomicrographs of the anteroventral periventricular nucleus (AVPV) were captured using a Nikon Eclipse TiE2 inverted microscope at x20 magnification. Z-stacks of 1um focal thickness were captured across three representative sections of the AVPV. The images were then analyzed by a blinded experimenter using the NIS-element software (RRID: SCR_014329). The experimenter counted the number of kisspeptin immunoreactive cells on three representative sections of the AVPV. ## Statistical analysis Statistical analysis was performed with PRISM® software 9.0 (Graph Pad Prism, RRID:scr_002798). Normal distribution and homogeneity of variance were determined using a Shapiro-Wilk test and Fisher’s test, respectively. The percentage of animals performing either male sexual behaviors or female typical lordosis behavior were compared by a Fisher exact tests. All data are represented as the mean +/- SEM. Analysis of lordosis, investigative and aggressive behaviors was performed by a 2-way ANOVA mixed models for repeated measures. All other data were analyzed by an unpaired Students t-tests when the normal distribution and variance homogeneity parameters were met. Otherwise, a Mann-Whitney test was used. A p-value < 0.05 was considered statistically significant and a p-value <0.07 was considered a tendency. ## Male sexual behavior is slightly altered by prenatal androgen exposure Locomotion was not affected by prenatal androgen excess in male, indicated by both the open-field test (10155 +/- 652.8 mm for VEH and 9318 +/- 554.4 mm for PNAM; $t = 0.98$, df=19, $$p \leq 0.34$$) and the elevated-plus maze test (6663 +/- 367.9 mm for VEH and 7004+/-279.6 for PNAM; $U = 52$, $$p \leq 0.92$$). Basal anxiety was also not affected by prenatal DHT exposure, as indicated by the percentage of time spend in the center of the open-field (Figure 2A; $t = 0.697$, df=19, $$p \leq 0.49$$) and in the open-arm of the elevated plus maze (Figure 2B; $U = 52$, $$p \leq 0.92$$). Figure 2: Prenatal DHT exposure modifies male-typical sexual behavior. (A, B) Histogram representing the percentage of time spent in the center zone of the Open-field test (A) and the percentage of time spent in the open arms of the Elevated-plus maze (B) for the VEH male (black bar) and PNAM (dark blue bar). (C, D) Histograms representing the percentage of animals performing each component of male copulatory behavior (mount, intromission, pelvic thrusts and ejaculation) while naive (C) (i.e. during the first test) or experienced (D) (i.e. during the second test) for the VEH male (black bars) and PNAM (dark blue bars). (E, F) Histograms representing the latencies to first mount, first intromission and ejaculation while naive (E) (i.e. during the first test) or experienced (F) (i.e. during the second test) for the VEH male (black bars) and PNAM (dark blue bars). (G, H) Histograms representing the intromission rate, corresponding to the number of mounts with intromission divided by the total number of mount trial, while naive (C) (i.e. during the first test) or experienced (D) (i.e. during the second test) for the VEH male (black bar) and PNAM (dark blue bar). (I) Histograms representing the percentage of time spent sniffing either the male of the female compartment in the three-compartment box partner preference test for the VEH male (black bars) and PNAM (dark blue bars). p<0.01, p<0.05, #p<0.07. VEH, control male; PNAM, prenatally androgenized male, s, seconds. Mean +/- SEM.* Two sexual behavior tests were performed because male sexual behavior is subject to learning with an increase in sexual behavior efficiency after a first experience [63, 64]. In the first test, $55\%$ of VEH males exhibited complete sexual behavior with mounting, intromission, pelvic thrust and ejaculation (Figure 2C). In contrast, only $25\%$ of PNAM exhibited complete sexual behavior (Figure 2C). Significantly fewer PNAM performed mounting ($$p \leq 0.02$$) or thrusts ($$p \leq 0.03$$) compared to VEH males (Figure 2C), while the percentage of animals performing intromission ($$p \leq 0.08$$) and ejaculating ($$p \leq 0.20$$) by the end of the first test was not statistically different (Figure 2C). After sexual experience, in test 2, $55\%$ of VEH males and $33\%$ of PNAM exhibited complete sexual behavior (Figure 2D). While the percentage of PNAM performing mounting behavior remained significantly reduced compared to VEH males ($$p \leq 0.0046$$, Figure 2D), the percentage of animals exhibiting intromission ($$p \leq 0.08$$), thrusts ($$p \leq 0.08$$) and ejaculation ($$p \leq 0.39$$) were not different between PNAM and VEH during the second test. To further investigate the effect of prenatal DHT exposure on male sexual behavior, we analyzed different components of male typical sexual behavior such as the latencies to perform (Figures 2E, F) and numbers (Table 2) of each behavior were also measured. In the first test, we observed that the latency to first intromission increased significantly for the PNAM compare to the VEH (Figure 2E; $t = 2.68$, df=19, $$p \leq 0.03$$). In addition, we observed a tendency to an increase latency to mount ($t = 1.704$, df=19, $$p \leq 0.07$$) and to ejaculate ($t = 1.841$, df=19, $$p \leq 0.07$$) for the PNAM compared to their VEH counterparts (Figure 2E). These increased latencies are associated with significantly fewer mounts with intromission ($U = 27$, $$p \leq 0.0467$$) and fewer pelvic thrusts ($U = 25$, $$p \leq 0.0251$$) in PNAM compared to VEH males (Table 2). In addition, the intromission rate of the PNAM tended toward being decreased compare with VEH males (Figure 2G; $U = 28.5$, $$p \leq 0.057$$). During the second test, only the latency to first mount was significantly increased for the PNAM compared to the VEH males (Figure 2F; $t = 3.972$, df=19, $$p \leq 0.001$$). PNAM and VEH had the same latencies to first intromission ($t = 1.621$, df=19, $$p \leq 0.13$$) and ejaculation ($t = 0.08$, df=19, $$p \leq 0.65$$) during the second test. Interestingly, PNAM displayed significantly fewer mounts ($U = 13$, $$p \leq 0.0018$$), mounts with intromission ($U = 23$, $$p \leq 0.0185$$) and pelvic thrusts ($U = 24$, $$p \leq 0.0233$$) compared to the VEH males (Table 2) which was also associated with a trend toward a lower intromission rate (Figure 2H; $U = 28$, $$P \leq 0.052$$) during the second test. Table 2 \| Number of \| Naive Male(Test 1) \| Naive Male(Test 1).1 \| Experienced Male(Test 2) \| Experienced Male(Test 2).1 \| \| --- \| --- \| --- \| --- \| --- \| \| Number of \| Control (n=9) \| PNAM (n=12) \| Control (n=9) \| PNAM (n=12) \| \| Mounts \| 13.22 +/- 3.227 \| 8.417 +/- 3.607 \| 22.67 +/- 5.649 \| 3.667 +/- 1.982** \| \| Mounts with Intromission \| 6.889 +/- 2.664 \| 2.417 +/- 1.598* \| 10.67 +/- 3.078 \| 2.25 +/- 1.415* \| \| Pelvic Thrusts \| 153.2 +/- 55.46 \| 58.42 +/- 36.14* \| 211 +/- 65.71 \| 54.58 +/- 26.08* \| ## Male partner discrimination is not affected by prenatal androgen exposure Male sexual behavior is dependent upon the odor recognition of a partner. Therefore, the ability of male mice to recognize a female was assessed using the three compartments box partner preference tests. PNAM and VEH male mice spent a similar percentage of time sniffing the male ($$p \leq 0.4117$$) and the female ($$p \leq 0.4116$$) compartments (Figure 2I). Noteworthy, the total time spent sniffing both compartments was also not different between PNAM and VEH male mice (413 +/- 10.45 s for VEH and 389.25+/- 10.91 s for PNAM; $t = 1$,529, df=19, $$p \leq 0.22$$). ## Female sexual behavior is significantly impaired by prenatal androgen exposure Before to test for the female typical sexual behavior, lordosis, we needed to verify that prenatal androgen exposure was not altering locomotion and anxiety in the PNAF compare to control female mice (VEH). Surprisingly, we observed a slight increase of the distance travelled by PNAF compare to VEH female mice in the open-field test (5336 +/- 438.7 mm for VEH and 6923 +/- 944.7 mm for PNAF; $U = 13$, $$p \leq 0.048$$) while no effect of prenatal DHT exposure on locomotion was observed in the elevated-plus maze test (2199 +/- 151.8 mm for VEH and 2436 +/- 330.4 for PNAF; $t = 0.75$, df=15, $$p \leq 0.46$$). Despite this discrepancy in the locomotion, basal anxiety, represented by the percentage of time spend in the center of the open-field (Figure 3A; $t = 0.44$, df=15, $$p \leq 0.67$$) and in the open-arm of the elevated plus maze (Figure 3B; $U = 32$, $$p \leq 0.96$$), was not affected by prenatal androgen exposure. Figure 3: Prenatal DHT exposure significantly impairs female-typical sexual behavior in the prenatally-androgenized female mouse modeling PCOS. (A, B) Histograms representing the percentage of time spent in the center zone of the Open-field test (A) and the percentage of time spent in the open arm of the Elevated-plus maze (B) for the VEH female (black bars) and PNAF (magenta bars). (C) Histograms representing the percentage of animals showing lordosis behavior for the VEH female (black bars) and PNAF (magenta bars) over 6 consecutive weeks of testing. (D) Graphical curve representing the percentage of lordosis quotient, i.e. the number of lordosis behavior divided by the number of male mount trial, over the 6 consecutive weeks of testing for the VEH female (black line) and PNAF (magenta line). (E) Histograms representing the percentage of time spent sniffing either the male or the female compartment in the three-compartment box partner preference test for the VEH female (black bars) and PNAF (magenta bars). #p<0.06, p<0.05, p<0.01, p<0.001. VEH, control female; PNAF, prenatally androgenized female. Mean +/- SEM. The female typical sexual behavior, lordosis, is a learned process. Therefore, we tested lordosis behavior over 6 consecutive weeks. As expected, during the first three weeks, the percentage of VEH female mice exhibiting lordosis behavior increased to reach $100\%$ and then remained over $80\%$ until the last week of testing. In contrast, the percentage of PNAF mice showing lordosis behavior remained lower throughout the 6 consecutive weeks of testing compared to VEH females: $16.67\%$ at week 1 and 2, $66.67\%$ at week 3, $50\%$ at week 4 and only $40\%$ at week 5 and 6. The percentage of PNAF displaying lordosis behavior was significantly reduced compare to VEH females at week 2 ($$p \leq 0.03$$) and 4 ($$p \leq 0.03$$) (Figure 3C). These results were associated with a significant reduction in the lordosis quotient in PNAF mice compared to VEH females (Figure 3D). Statistical analysis with repeated measures mixed model ANOVA revealed a significant effect of the prenatal androgen treatment (F [1,15] = 25.60, $$p \leq 0.0001$$). Following post hoc comparisons, the lordosis quotient was significantly reduced in PNAF mice compared to VEH females in week 2 ($$p \leq 0.0091$$, $t = 3.988$, df=13.14), week 3 ($$p \leq 0.002$$, $t = 5.731$, df=14.99) and week 4 ($$p \leq 0.0271$$, $t = 3.383$, df=13.75). For the last test in week 6, the post-hoc comparison detected a trend toward a decreased lordosis quotient in PNAF compared to VEH females ($$p \leq 0.0583$$, $t = 3.016$, df=12.95). ## Investigative and rejection behaviors during lordosis Sexual motivation was assessed by quantifying the number of times the female sniffed the anogenital region of the male: anogenital sniffing bouts (Table 3). Prenatal androgen treatment had no significant effect on the number of anogenital sniffing bouts (F[1,15] = 0.01529, $$p \leq 0.9032$$). Table 3 \| Unnamed: 0 \| Anogenital Sniffing bouts \| Anogenital Sniffing bouts.1 \| Aggressive behaviors bouts \| Aggressive behaviors bouts.1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| Control (n=11) \| PNAF (n=5/6) \| Control (n=11) \| PNAF (n=5/6) \| \| W1 \| 0.30 +/- 0.20 \| 1.17 +/- 0.65 \| 3.0 +/- 0.75 \| 5.17 +/- 1.51 \| \| W2 \| 1.45 +/- 0.62 \| 1.50 +/- 0.56 \| 3.45 +/- 1.36 \| 4.0 +/- 1.81 \| \| W3 \| 1.91 +/-0.72 \| 1.50 +/- 0.50 \| 0.91 +/- 4.17 \| 4.17 +/- 1.80 \| \| W4 \| 0.73 +/- 0.36 \| 1.00 +/- 0.63 \| 2.36 +/- 0.89 \| 1.67 +/- 0.84 \| \| W5 \| 1.18 +/- 0.40 \| 1.00 +/- 0.55 \| 1.09 +/- 0.74 \| 3.40 +/- 1.40 \| \| W6 \| 2.55 +/- 0.79 \| 2.20 +/- 0.86 \| 2.64 +/- 0.86 \| 7.20 +/- 2.69 \| Rejection behaviors toward the male were assessed by the number of times the female rejected male approaches by kicking or biting (Table 3). Prenatal androgen treatment was found to have no significant effect on the number of aggressive behavior bouts (F [1,15] = 3.296, $$p \leq 0.0895$$). ## Female partner discrimination is not affected by prenatal androgen exposure The female-typical sexual behavior, lordosis behavior is also dependent upon the odor recognition of a male partner. Therefore, the ability of mice to recognize the male was assessed using the three compartments box partner preference tests. PNAF and VEH female mice spent a similar percentage of time sniffing the male ($$p \leq 0.39$$) and the female ($$p \leq 0.39$$) compartments (Figure 3E). The total time spent sniffing both compartments was not different between PNA and VEH female mice (332.81 +/- 18.16 s for VEH and 328.60 +/- 29.09 s for PNA; $U = 18$, $$p \leq 0.30$$). ## Experiment 2: Prenatal androgen exposure with DHT, a non-aromatisable androgen, does not masculinize sexual behavior in female mice To determine whether prenatally androgenized females exhibited masculinized sexual behaviors, adult PNAF and VEH female mice were subjected to elevated testosterone which is triggers male-like sexual behaviors in presence of a receptive stimulus female (Figure 4). The latency to first mount was not different between PNAF (991.4 +/- 346 s) and VEH (726.0 +/-342.4 s) female mice ($U = 10$, $$p \leq 0.3853$$) (Figure 4A). The number of mounts performed by the female over an oestrous stimulus female was also not statistically different between PNAF (11.40 +/- 4.95 mounts) and VEH (12.50 +/- 4.24 mounts) female mice ($t = 0.1699$, df=9, $$p \leq 0.8688$$) (Figure 4B). Finally, the number of pelvic thrust-like movements performed by the female during mount of the stimulus female was also not different between PNA (58.20 +/- 31.44 pelvic thrusts) and VEH (61.83 +/- 31.42 pelvic thrusts) female mice ($t = 0.810$, df=9, $$p \leq 0.9372$$) (Figure 4C). Figure 4: Prenatal DHT exposure does not masculinize behavior in the prenatally-androgenized female mouse modeling PCOS. (A) Histogram representing the latency to first mount for VEH female (black bar) and PNAF (magenta bar). (B) Histograms representing the number of mounts for VEH female (black bar) and PNAF (magenta bar). (C) Histograms representing the number of pelvic thrusts for VEH female (black bar) and PNAF (magenta bar). VEH, control female; PNAF, prenatally androgenized female. Mean +/- SEM. ## AVPV Kisspeptin neuron population size is not affected by PNA exposure AVPV Kisspeptin neurons have been demonstrated to play a major role in the expression of lordosis behavior [65]. In addition, AVPV kisspeptin neurons are sexually dimorphic [66] and sensitive to prenatal testosterone rise [67]. Here, we found that the number of kisspeptin immunoreactive cells per section of the AVPV was not different between PNA (7.27 +/- 1.35 kisspeptin positive cells) and VEH (8.74 +/- 1.28 kisspeptin positive cells) female mice. ## Progesterone receptor immunoreactivity is not different between VEH and PNA female mice after priming with ovarian hormones (OVX+E+P) In adulthood, lordosis behavior is dependent upon oestradiol and progesterone signaling in the brain (68–70). As reduced PR immunostaining has been reported in intact PNA mice [30], we aimed to determine if reduced lordosis behavior observed in PNA females might correspond with reduced PR expression in brain regions regulating female sexual behavior. PR-positive cells were counted in the anteroventral periventricular nucleus (AVPV), the median arcuate nucleus (mARN) and the ventrolateral part of the vendromedial hypothalamus (VMHvl) (Figures 5A–D). Robust PR-immunoreactivity was observed in both groups throughout the regions analyzed and no differences were found in the number of PR-positive cells in any of the regions analyzed between PNA and VEH mice that had previously been OVX and steroid primed for sexual behavior analysis (Figures 5A–D). Figure 5: Prenatal DHT exposure does not alter adult progesterone receptor expression in after hormonal normalization. (A–C) Representative photomicrographs of PR immunoreactivity in the anteroventral part of the periventricular nucleus of the third ventricle (A), the medial arcuate nucleus (B) and the ventrolateral part of the ventromedial hypothalamus (C) in VEH and PNA mice. (D) Histograms representing the number of Progesterone receptor (PR) immunoreactive cells per sections analyzed between ovariectomized and hormone replaced PNAF mice (magenta bars) and VEH female controls (black bars) within the three brain regions analyzed. AVPV, anteroventral part of the periventricular nucleus of the third ventricle; ARN, arcuate nucleus; VMHvl, ventrolateral part of the ventromedial hypothalamus. VEH, control female; PNAF, prenatally androgenized female modeling PCOS. Mean +/- SEM. Scale bar: 100μm. ## Neural activation is reduced only in the dorsomedial hypothalamus after lordosis behavior in prenatally-androgenized female mice In a second cohort of PNA and VEH female mice, that were tested for lordosis behavior again during 6 consecutive weeks, we detected again a significant impairment of lordosis behavior in PNA female mice over the 6 consecutive tests (F [1,8] =13, $$p \leq 0.0069$$). Ninety minutes following the last lordosis behavior test, animals were euthanized and neural activation was assessed by cFOS immunostaining in several brain regions known to be implicated in sexual behaviors and/or fertility regulation in female mice (Figures 6A–C). As expected, the number of cFOS-positive cells increased between control females that did not participate in lordosis behavior (Basal, $$n = 3$$-4) and control females who underwent lordosis behavior tests (VEH, $$n = 5$$) (Figure 6A). Indeed, a significant increase in the number of cFOS-ir positive cells was observed between the Basal and VEH female mice in the majority of the brain regions analyzed and known to be implicated in the regulation of lordosis behavior: the olfactory tuberal nucleus (TU; $$p \leq 0.03$$), the median preoptic area (MnPOA; $$p \leq 0.03$$), the posterior median part of the bed nucleus of the stria terminalis (BNSTpm; $$p \leq 0.0009$$), the paraventricular nucleus of the hypothalamus (PVN; $$p \leq 0.01$$) and the dorsomedian part of the ventromedial hypothalamus (VMHdm; $$p \leq 0.007$$) (Figure 6A). In addition, we found a trend toward an increase in cFOS-ir positive cells between Basal and VEH female mice in three other regions: the Piriform Cortex (PirCx; $$p \leq 0.06$$), the anteroventral part of the periventricular nucleus of the third ventricle (AVPV; $$p \leq 0.06$$) and the periaqueductal gray nucleus (PAG; $$p \leq 0.067$$) (Figure 6A). In contrast, the numbers of cFOS-ir positive cells were not different between PNAF and VEH mice in the majority of brain regions analyzed except for the dorsomedial hypothalamus where we counted significantly lower cFOS-ir positive cells in PNAF mice (14.2 +/- 2.25) compared to VEH female mice (41 +/- 14.15) ($U = 1$, $$p \leq 0.0159$$) (Figures 6A, C). Figure 6: Prenatal DHT exposure decreases neural activation only in the dorsomedial hypothalamus of the PNAF modeling PCOS (A, B) Representative photomicrographs of cFOS immunoreactivity in the ventrolateral part of the ventromedial hypothalamus (VMHvl) (A) and the dorsomedial hypothalamus (DMH) in VEH and PNA mice. (C) Histograms representing the number of cFOS immunoreactive cells per section analyzed for each brain regions studied for Basal female who did not undergo lordosis behavior (grey bars), VEH control females (black bars) and PNAF (magenta bars) who were tested for lordosis behavior during the 6th test. Mean +/- SEM. p<0.05, Mann-Whitney test. VEH, controls; PNA, prenatally androgenized; Pir Cx, piriform cortex nuclei; Tu, olfactory tuberal nucleus; AcC, Accumbens nucleus core; MnPOA, median preoptic area; AVPV, anteroventral part of the periventricular nucleus of the third ventricle; BNSTpm, posterior median part of the bed nucleus of the stria terminalis; PVN, parventricular nucleus of the hypothalamus; ARN, arcuate nucleus; DMH, dorsomedial hypothalamus; VMHvl, ventrolateral part of the ventromedial hypothalamus; VMHdm, dorsomedian part of the ventromedial hypothalamus, MeA, medial amygdala; VTA, ventral tegmental area; PAG, periaqueductal nucleus. # p<0.07, * p<0.05, p<0.01, p<0.001, **p<0.0001. The grey bars and grey statistical signs represent the differences between the Basal and VEH groups while the black bars and black statistical signs represent the differences between the PNAF and VEH groups. VEH = control female; PNAF= prenatally androgenized female modeling PCOS. Mean +/- SEM. Scale bar: 100μm. ## Discussion This study aimed to investigate how maternal androgen excess impacts adult sexual behavior in male and female offspring. We demonstrate here, that prenatally androgenized female mice that model several PCOS features (29–33, 57–61) and their male siblings exhibit altered adult sexual behaviors. We found that PNAM displayed reduced performance in some components of sexual behavior, mostly the mounting and intromission behavior parameters. However, PNAM were able to perform complete sexual behavior similarly to control males independently of their sexual experience. In contrast, PNAF exhibited impaired lordosis behavior throughout the 6 weeks of testing. In addition, female PNA mice displayed similar male-like sexual behaviors to VEH female mice after hormonal replacement with testosterone to induce male-like sexual behavior. This result demonstrates that the observed sexual dysfunction in PNAF is not likely due to masculinization of the brain. This finding is supported by the AVPV kisspeptin population size remaining unaffected in PNAF mice. These results suggest, instead, a specific impairment in the feminization of the brain in the PNA mouse model of PCOS. Noteworthy, in our study condition with PNA and VEH female mice being ovariectomized and supplemented with artificial high level of oestradiol, the PR expression is similar between the PNAF and VEH in three brain regions known to regulate lordosis behavior. This result suggests that the sexual dysfunction of the PNA female mice could have an organizational origin. Lordosis behavior-induced cFOS expression was largely unchanged, but apparently reduced activation of the DMH suggests potential avenues for future investigation. Interestingly, the sexual dysfunction experienced by women with PCOS and men with AGA has long been dismissed as an unfortunate symptom of impaired body image and self-esteem [42]. Our findings suggest an alternative theory regarding the origins of PCOS-related sexual dysfunction. These data in an animal model indicate programming effects of prenatal androgen excess on the regulation of adult sexual behavior. The effect of maternal androgen excess appears to vary from a pernicious effect for the male offspring to a detrimental effect for the female offspring, leading to an inability to copulate with a male. These findings may be applicable beyond PCOS, to other diseases where androgen signaling is enhanced from early life such as in congenital adrenal hyperplasia (CAH) [11, 12]. Moreover, exposure to endocrine disrupting chemicals may lead to increased androgen production or androgen receptor activity at early stages of life (13–15). ## Prenatal androgen exposure in male mice altered male copulatory behavior but does not alter their fertilization capabilities. To date, most of the studies on the role of androgens on male copulatory behavior have focused on the role of testosterone, aromatized into estradiol and therefore acting on estradiol receptors for the development but also the activation of male sexual behavior [1]. Studies on the role of androgens acting through the androgen receptor remains sparse [6]. A set of recent studies using androgen receptor and estradiol receptors deletion within the central nervous system in mice highlighted that neural androgen receptor signaling throughout life is required for the full expression of male copulatory behavior [7, 71, 72]. Studies in sheep injected with either testosterone propionate or DHT during gestation indicate that maternal androgen excess compromises the reproductive function of ram offspring (34–38, 73) without altering sexual behavior [38]. Here, we investigated for the first time, the effect of prenatal exposure to the non-aromatisable androgen, DHT. We observed an alteration only in certain sexual behaviors parameters. Indeed, naïve and experienced PNAM needed more time to perform their first attempt to mount and intromit the stimulus female. They also performed fewer mount with and without intromission as well as fewer pelvic thrusts. These results could suggest a lack of motivation for socio-sexual interaction which remains to be determined. Interestingly, at the end of both tests, a similar number of PNAM and control males were able to reach ejaculation, the endpoint of male copulatory behavior. In addition, the PNAM were more similar to VEH males after experience with only the mounting behavior remaining significantly reduced during the second test. These findings are aligned with previous work demonstrating no significant impairments in the neuroendocrine hypothalamo-pituitary-gonadal axis in PNAM mice [39]. Altogether, these findings suggest that while prenatal DHT exposure disrupts some parameters of adult male copulatory behavior, mostly motivational components, the endpoint capability of the PNAM to fertilize a female mouse is unaffected. Similarly, sexual dysfunction in men with androgenic alopecia, a clinical sign of PCOS equivalent in men, has been also associated with motivation such as decreased desire and decreased sexual arousal without affecting erectile function [50]. ## Prenatal androgen exposure in mice specifically impairs lordosis behavior In females, PNA resulted in a sustained impairment in lordosis behavior. In an attempt to decipher the cause of this sexual dysfunction, several behaviors related to sexual behavior that might explain the observed reduction of sexual receptivity in the PNA mouse were also examined. Anxiety and the physical ability to perform sexual behaviors are considered to be potential confounding factors that might influence an animal’s ability to perform normal sexual behaviors. PCOS patients are more likely to develop anxiety and depression which have been shown to negatively impact their quality of life (41, 45, 51–54). It was therefore pertinent to determine changes in anxiety and locomotion evident in the PNAF mouse model. Previous studies have reported anxiety-like behaviors in the PNA mouse model of PCOS (59–61). However, here, no significant differences were detected in anxiety-like behaviors in PNAF mice. This discrepancy is likely to be experimental as the previous studies were performed during the light phase of the light-dark cycle i.e. the inactive phase of the animals. In the present experiment, tests were performed under a sodium lamp, which cannot be seen by mice, during the dark phase of the light-dark cycle i.e. the active phase of the animals. Those conditions are the same conditions that were used to test lordosis behavior, partner preference and male-like sexual behavior. Therefore, the decrease in lordosis behavior observed in the PNAF mice is unlikely to be due to basal anxiety or deficits in locomotion during their naturally active phase. However, we cannot rule out that the introduction of the male stimulus could have triggered an anxiety-like response similar to the one observed when the animals where tested during their inactive phase in the previous studies (59–61). Like lordosis, male-oriented sexual partner preference is a female-typical. Behavioral tests found no overt difference in the partner preference of the PNAF mice compared to VEH females, suggesting that prenatal androgen does not impact preference for male or female scent. Ano-genital sniffing and defensive behaviors were also unchanged in PNAF, suggesting that reduced sexual motivation and/or an increase in aggression toward the male are unlikely to explain the PNA related sexual dysfunction. In agreement with other studies, repeated testing of lordosis yielded a steady increase in the lordosis quotient over time in VEH females (3–5). PNAF mice, however, did not exhibit this same increase in lordosis quotient over time. Further investigation is needed to determine if prenatal androgen exposure modifies the neuroplasticity occurring in the brain to learn lordosis behaviors [74, 75] or other cognitive behavioral outcomes such as learning, memory, or social interactions. Impaired sexual behaviors have also been reported in the female offspring of dams exposed to anti-mullerian hormone (prenatal AMH or PAMH mice) [76], a paradigm that also models PCOS features [77]. In addition to impaired lordosis, PAMH females also demonstrated altered partner preference and increased rejection/aggression behaviors [76]. Differences in partner preference and aggression behaviors between the two PCOS-like models may reflect the nature of the androgen exposure. PAMH likely drives elevated maternal testosterone, which when aromatized in the fetal brain will result in masculinization [1]. ## Prenatal DHT does not masculinize the female brain and behavior How excess non-aromatisable androgens like DHT impact the development of the female brain and behavior remains poorly understood [78] despite androgen receptor being present in the female brain [6]. Here, we determined that prenatal androgenization with DHT that models PCOS does not masculinize the female brain or sex behaviors. The levels of three domains of male-like sexual behaviors (latency to mount, mounting behavior and pelvic thrusting) were indistinguishable between PNAF mice and control females. These behavioral findings were supported by anatomical data demonstrating a feminized AVPV kisspeptin population as already described [79]. AVPV kisspeptin neurons are a clearly sexually differentiated population in the brain, with the number of neurons decreased by the male-typical prenatal testosterone rise [67]. ## Organizational versus activational effects of sex steroid hormones? Oestradiol and progesterone are crucial for the expression of female-typical rodent sexual behavior [69]. In PCOS patients, progesterone and estradiol levels can be abnormal in association with impaired folliculogenesis and ovulation. To overcome differences in circulating oestradiol and progesterone levels between PNA and VEH, animals were ovariectomized and the hormones required for lordosis behavior were replaced. This also effectively removed the adult hyperandrogenism observed in this model [29], and allowed us to investigate the potential organizational effects of prenatal exposure to DHT on sexual behaviors. Impaired lordosis in PNA mice following a normalization of circulating steroid hormones, therefore, suggests an earlier, organizational impact of sex steroids or other downstream signals on the neural substrates controlling lordosis behavior. The organizational effect of oestradiol and progesterone for feminization of the brain has been demonstrated to occur during the peripubertal period (3–5). As hormone replacement in the present study occurred in adults, after the expected rise in endogenous testosterone levels in the PNA model (at 40-50 days of age) [33], we cannot rule out an impact of this pubertal rise in androgens on the circuits mediating female-typical sexual behavior. PCOS is associated with impaired estradiol and progesterone feedback to the reproductive axis to slow GnRH/LH secretion [21, 80, 81], suggesting a central insensitivity to steroid hormone signaling. Knowing that PNA mice also model this impaired steroid hormone feedback to the reproductive axis by exhibiting mainly a reduced number of PR expressing cells in several hypothalamic nuclei [29, 30], we investigated whether impaired lordosis may be the result of an impaired ability to respond to the artificially delivered hormones (implant of estradiol and injection of progesterone). PR was robustly expressed and not different between VEH and PNAF mice in any of the areas investigated, suggesting that exogenous hormone treatment together with the absence of hyperandrogenism can overcome the impaired progesterone feedback observed in the intact PNAF mice [29], and that this is not a likely explanation for impaired lordosis behavior. ## Which neural target could explain the lordosis behavior impairment? In an effort to identify the neural target of prenatal DHT exposure correlated with lordosis behavior impairment, cFos immunoreactivity, a proxy for neural activation, was measured after lordosis behavior in a wide range of brain regions known to be implicated in the neuronal circuitry controlling female sexual behaviors [68]. Although cFos expression was significantly elevated in all of the expected regions in mice experiencing lordosis compared to a basal control group, cFos expression patterns were largely unaffected by PNA. In view of the recent evidences highlighting the role of RP3V Kisspeptin and VMHvl nNOS neurons in lordosis behavior in healthy female mice [65] as well as in the PAMH mouse model of PCOS [76], we would have expected some changes in cFOS in the RP3V and VMHvl region. As noted earlier RP3V Kisspeptin neurons remain unchanged in the PNA mice (data showed here and recently published [79]. Similarly, a previous study from our group showed no changes in arcuate nucleus nNOS neurons in the PNA mice contrary to the PAMH mouse model [76]. Finally, our data showed an absence of cFOS and PR changes in the VMHvl in the PNA mouse model compare to control therefore it is unlikely that VMHvl nNOS neurons could be involved in the sexual dysfunction observed in the PNA mouse model of PCOS. The changes in Kisspeptin and nNOS observed in the PAMH likely results from a masculinization of the brain circuit controlling sexual behavior since prenatal AMH treatment leads to increase testosterone level [77]. Noteworthy, our cFOS data highlighted a change in the DMH, with PNAF showing reduced cFOS immunoreactivity compare to controls suggestive of reduced activation in this area. The role of the DMH in female sexual behaviors remains unclear, however, RF-amide related-peptide 3 (RFRP-3) neurons in the DMH have recently been identified as potential novel factors of female sexual motivation and behaviors in addition to their known roles in energy balance and GnRH neuron function [82]. Interestingly, recent studies also showed that injection of RPRP-3 leads to suppression of either sexual motivation or receptivity in female hamsters, rats and eusocial mammals (83–85). Thus, further investigation on RFRP-3 in the PNAF mice would be of interest to decipher the potential role of RFRP-3 in the suppression of lordosis behavior. ## Conclusion These new data provide a significant step forward in our knowledge of how prenatal androgenization can influence adult sexual behaviors in male and female mice. These findings are aligned with the “Developmental Origins of Health and Diseases” (DOHaD) hypothesis in which early-life environment can increase sensitivity or risk toward developing adverse outcomes later in life. Combined with evidence that prenatal androgen exposure leads to reproductive disorders in both sexes, these findings suggest a critical sensitive period of development where both the neuroendocrine regulation of reproductive function and behavior are sensitive to androgen specifically through the androgen receptor. The underlying central mechanisms underpinning impaired sexual behavior in PCOS-like female mice and their male siblings remains to be elucidated. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Ethics statement The animal study was reviewed and approved by University of Otago Ethical committee. ## Author contributions ND, as the first author, was a honours student under the supervision of ED (co-last author). She performed most of the experiments under ED supervision. MP is an ARF within the Campell Laboratory. She provided her technical assistance throughout the study as well as provided feedbacks on the manuscript. AR is a PhD student working on progesterone receptor changes in PCOS under supervision of ED and RC has funded ED salary throughout the duration of this study and provided feedbacks on the manuscript. ED and RC received joint funding to start the experiments in this study then ED received her own funding to complete the experimental work in this study. RC gave feedbacks on the manuscript before submission. ED has designed the experimental procedure, performed some experiments herself, supervised the first author undertaking most of the experiments, co-analyzed the results with the first author and wrote the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2023.1116482/full#supplementary-material ## References 1. McCarthy MM, Arnold AP. Reframing sexual differentiation of the brain. Nat Neurosci (2011) 14. DOI: 10.1038/nn.2834 2. Bakker J, Brock O. Early oestrogens in shaping reproductive networks: evidence for a potential organisational role of oestradiol in female brain development. J Neuroendocrinol (2010) 22. DOI: 10.1111/j.1365-2826.2010.02016.x 3. Bakker J, Honda S, Harada N, Balthazart J. The aromatase knock-out mouse provides new evidence that estradiol is required during development in the female for the expression of sociosexual behaviors in adulthood. J Neurosci (2002) 22. DOI: 10.1523/JNEUROSCI.22-20-09104.2002 4. Brock O, Baum MJ, Bakker J. The development of female sexual behavior requires prepubertal estradiol. J Neurosci (2011) 31. DOI: 10.1523/JNEUROSCI.0209-11.2011 5. Desroziers E, Brock O, Bakker J. Potential contribution of progesterone receptors to the development of sexual behavior in male and female mice. Horm Behav (2017) 90. DOI: 10.1016/j.yhbeh.2016.05.008 6. Mhaouty-Kodja S. Role of the androgen receptor in the central nervous system. Mol Cell Endocrinol (2018) 465. DOI: 10.1016/j.mce.2017.08.001 7. Raskin K, de Gendt K, Duittoz A, Liere P, Verhoeven G, Tronche F. Conditional inactivation of androgen receptor gene in the nervous system: effects on male behavioral and neuroendocrine responses. J Neurosci (2009) 29. DOI: 10.1523/JNEUROSCI.0296-09.2009 8. Brock O, De Mees C, Bakker J. Hypothalamic expression of oestrogen receptor alpha and androgen receptor is sex-, age- and region-dependent in mice. J Neuroendocrinol (2015) 27. DOI: 10.1111/jne.12258 9. Abbott DH, Dumesic DA, Levine JE. Hyperandrogenic origins of polycystic ovary syndrome - implications for pathophysiology and therapy. Expert Rev Endocrinol Metab (2019) 14. DOI: 10.1080/17446651.2019.1576522 10. Lizneva D, Gavrilova-Jordan L, Walker W, Azziz R. Androgen excess: Investigations and management. Best Pract Res Clin Obstet Gynaecol (2016) 37 98-118. DOI: 10.1016/j.bpobgyn.2016.05.003 11. Kepczynska-Nyk A, Kurylowicz A, Nowak A, Bednarczuk T, Ambroziak U. Sexual function in women with androgen excess disorders: Classic forms of congenital adrenal hyperplasia and polycystic ovary syndrome. J Endocrinol Invest (2021) 44. DOI: 10.1007/s40618-020-01332-3 12. Schernthaner-Reiter MH, Baumgartner-Parzer S, Egarter HC, Krebs M, Kautzky-Willer A, Kirchheiner K. Influence of genotype and hyperandrogenism on sexual function in women with congenital adrenal hyperplasia. J Sex Med (2019) 16. DOI: 10.1016/j.jsxm.2019.07.009 13. Gray LE, Wilson VS, Stoker T, Lambright C, Furr J, Noriega N. Adverse effects of environmental antiandrogens and androgens on reproductive development in mammals. Int J Androl (2006) 29 96-104. DOI: 10.1111/j.1365-2605.2005.00636.x 14. Ho SM, Cheong A, Adgent MA, Veevers J, Suen AA, Tam NNC. Environmental factors, epigenetics, and developmental origin of reproductive disorders. Reprod Toxicol (2017) 68 85-104. DOI: 10.1016/j.reprotox.2016.07.011 15. Mhaouty-Kodja S, Naule L, Capela D. Sexual behavior: From hormonal regulation to endocrine disruption. Neuroendocrinol (2018) 107. DOI: 10.1159/000494558 16. Lizneva D, Suturina L, Walker W, Brakta S, Gavrilova-Jordan L, Azziz R. Criteria, prevalence, and phenotypes of polycystic ovary syndrome. Fertil Steril (2016) 106 6-15. DOI: 10.1016/j.fertnstert.2016.05.003 17. Azziz R. PCOS: a diagnostic challenge. Reprod BioMed Online (2004) 8. DOI: 10.1016/S1472-6483(10)61644-6 18. Farquhar CM, Birdsall M, Manning P, Mitchell JM, France JT. The prevalence of polycystic ovaries on ultrasound scanning in a population of randomly selected women. Aust N Z J Obstet Gynaecol (1994) 34 67-72. DOI: 10.1111/j.1479-828X.1994.tb01041.x 19. Coyle C, Campbell RE. Pathological pulses in PCOS. Mol Cell Endocrinol (2019) 498 110561. DOI: 10.1016/j.mce.2019.110561 20. Ruddenklau A, Campbell RE. Neuroendocrine impairments of polycystic ovary syndrome. Endocrinol (2019) 160. DOI: 10.1210/en.2019-00428 21. Eagleson CA, Gingrich MB, Pastor CL, Arora TK, Burt CM, Evans WS. Polycystic ovarian syndrome: evidence that flutamide restores sensitivity of the gonadotropin-releasing hormone pulse generator to inhibition by estradiol and progesterone. J Clin Endocrinol Metab (2000) 85. DOI: 10.1210/jcem.85.11.6992 22. Moore AM, Campbell RE. Polycystic ovary syndrome: Understanding the role of the brain. Front Neuroendocrinol (2017) 46 1-14. DOI: 10.1016/j.yfrne.2017.05.002 23. Cannarella R, Condorelli RA, Mongioi LM, La Vignera S, Calogero AE. Does a male polycystic ovarian syndrome equivalent exist. J Endocrinol Invest (2018) 41 49-57. DOI: 10.1007/s40618-017-0728-5 24. Di Guardo F, Ciotta L, Monteleone M, Palumbo M. Male Equivalent polycystic ovarian syndrome: Hormonal, metabolic, and clinical aspects. Int J Fertil Steril (2020) 14 79-83. DOI: 10.22074/ijfs.2020.6092 25. Liu DM, Torchen LC, Sung Y, Paparodis R, Legro RS, Grebe SK. Evidence for gonadotrophin secretory and steroidogenic abnormalities in brothers of women with polycystic ovary syndrome. Hum Reprod (2014) 29. DOI: 10.1093/humrep/deu282 26. Sanke S, Chander R, Jain A, Garg T, Yadav P. A comparison of the hormonal profile of early androgenetic alopecia in men with the phenotypic equivalent of polycystic ovarian syndrome in women. JAMA Dermatol (2016) 152. DOI: 10.1001/jamadermatol.2016.1776 27. Stener-Victorin E, Padmanabhan V, Walters KA, Campbell RE, Benrick A, Giacobini P. Animal models to understand the etiology and pathophysiology of polycystic ovary syndrome. Endocr Rev (2020) 41. DOI: 10.1210/endrev/bnaa010 28. Cesta CE, Mansson M, Palm C, Lichtenstein P, Iliadou AN, Landen M. Polycystic ovary syndrome and psychiatric disorders: Co-morbidity and heritability in a nationwide Swedish cohort. Psychoneuroendocrinol (2016) 73 196-203. DOI: 10.1016/j.psyneuen.2016.08.005 29. Moore AM, Prescott M, Campbell RE. Estradiol negative and positive feedback in a prenatal androgen-induced mouse model of polycystic ovarian syndrome. Endocrinol (2013) 154 796-806. DOI: 10.1210/en.2012-1954 30. Moore AM, Prescott M, Marshall CJ, Yip SH, Campbell RE. Enhancement of a robust arcuate GABAergic input to gonadotropin-releasing hormone neurons in a model of polycystic ovarian syndrome. Proc Natl Acad Sci U S A (2015) 112 596-601. DOI: 10.1073/pnas.1415038112 31. Sullivan SD, Moenter SM. Prenatal androgens alter GABAergic drive to gonadotropin-releasing hormone neurons: Implications for a common fertility disorder. Proc Natl Acad Sci U S A (2004) 101. DOI: 10.1073/pnas.0308058101 32. Marshall CJ, Desroziers E, McLennan T, Campbell RE. Defining subpopulations of arcuate nucleus GABA neurons in Male, female, and prenatally androgenized female mice. Neuroendocrinol (2017) 105. DOI: 10.1159/000452105 33. Silva MS, Prescott M, Campbell RE. Ontogeny and reversal of brain circuit abnormalities in a preclinical model of PCOS. JCI Insight (2018) 3. DOI: 10.1172/jci.insight.99405 34. Recabarren SE, Lobos A, Figueroa Y, Padmanabhan V, Foster DL, Sir-Petermann T. Prenatal testosterone treatment alters LH and testosterone responsiveness to GnRH agonist in male sheep. Biol Res (2007) 40. DOI: 10.4067/S0716-97602007000400007 35. Recabarren SE, Recabarren M, Sandoval D, Carrasco A, Padmanabhan V, Rey R. Puberty arises with testicular alterations and defective AMH expression in rams prenatally exposed to testosterone. Domest Anim Endocrinol (2017) 61. DOI: 10.1016/j.domaniend.2017.06.004 36. Recabarren SE, Rojas-Garcia PP, Recabarren MP, Alfaro VH, Smith R, Padmanabhan V. Prenatal testosterone excess reduces sperm count and motility. Endocrinol (2008) 149. DOI: 10.1210/en.2008-0785 37. Rojas-Garcia PP, Recabarren MP, Sarabia L, Schon J, Gabler C, Einspanier R. Prenatal testosterone excess alters sertoli and germ cell number and testicular FSH receptor expression in rams. Am J Physiol Endocrinol Metab (2010) 299 E998-E1005. DOI: 10.1152/ajpendo.00032.2010 38. Scully CM, Estill CT, Amodei R, McKune A, Gribbin KP, Meaker M. Early prenatal androgen exposure reduces testes size and sperm concentration in sheep without altering neuroendocrine differentiation and masculine sexual behavior. Domest Anim Endocrinol (2018) 62 1-9. DOI: 10.1016/j.domaniend.2017.07.001 39. Holland S, Prescott M, Pankhurst M, Campbell RE. The influence of maternal androgen excess on the male reproductive axis. Sci Rep (2019) 9 18908. DOI: 10.1038/s41598-019-55436-9 40. Davis SR, Worsley R, Miller KK, Parish SJ, Santoro N. Androgens and female sexual function and dysfunction–findings from the fourth international consultation of sexual medicine. J Sex Med (2016) 13. DOI: 10.1016/j.jsxm.2015.12.033 41. Elsenbruch S, Hahn S, Kowalsky D, Offner AH, Schedlowski M, Mann K. Quality of life, psychosocial well-being, and sexual satisfaction in women with polycystic ovary syndrome. J Clin Endocrinol Metab (2003) 88. DOI: 10.1210/jc.2003-030562 42. Fliegner M, Richter-Appelt H, Krupp K, Brunner F. Sexual function and socio-sexual difficulties in women with polycystic ovary syndrome (PCOS). Geburtshilfe Frauenheilkd (2019) 79 498-509. DOI: 10.1055/a-0828-7901 43. Hashemi S, Ramezani Tehrani F, Farahmand M, Bahri Khomami M. Association of PCOS and its clinical signs with sexual function among Iranian women affected by PCOS. J Sex Med (2014) 11. DOI: 10.1111/jsm.12627 44. Kowalczyk R, Skrzypulec V, Lew-Starowicz Z, Nowosielski K, Grabski B, Merk W. Psychological gender of patients with polycystic ovary syndrome. Acta Obstet Gynecol Scand (2012) 91. DOI: 10.1111/j.1600-0412.2012.01408.x 45. Mansson M, Norstrom K, Holte J, Landin-Wilhelmsen K, Dahlgren E, Landen M. Sexuality and psychological wellbeing in women with polycystic ovary syndrome compared with healthy controls. Eur J Obstet Gynecol Reprod Biol (2011) 155. DOI: 10.1016/j.ejogrb.2010.12.012 46. Nohr EA, Hansen AB, Andersen MS, Hjorth S. Sexual health in parous women with a history of polycystic ovary syndrome: A national cross-sectional study in Denmark. Int J Gynaecol Obstet (2021) 157. DOI: 10.1002/ijgo.13911 47. Noroozzadeh M, Ramezani Tehrani F, Bahri Khomami M, Azizi F. A comparison of sexual function in women with polycystic ovary syndrome (PCOS) whose mothers had PCOS during their pregnancy period with those without PCOS. Arch Sex Behav (2017) 46. DOI: 10.1007/s10508-016-0919-8 48. Thannickal A, Brutocao C, Alsawas M, Morrow A, Zaiem F, Murad MH. Eating, sleeping and sexual function disorders in women with polycystic ovary syndrome (PCOS): A systematic review and meta-analysis. Clin Endocrinol (Oxf) (2020) 92. DOI: 10.1111/cen.14153 49. Haber RS, Gupta AK, Epstein E, Carviel JL, Foley KA. Finasteride for androgenetic alopecia is not associated with sexual dysfunction: A survey-based, single-centre, controlled study. J Eur Acad Dermatol Venereol (2019) 33. DOI: 10.1111/jdv.15548 50. Molina-Leyva A, Caparros-Del Moral I, Gomez-Avivar P, Alcalde-Alonso M, Jimenez-Moleon JJ. Psychosocial impairment as a possible cause of sexual dysfunction among young men with mild androgenetic alopecia: A cross-sectional crowdsourcing web-based study. Acta Dermatovenerol Croat (2016) 24 51. Chaudhari AP, Mazumdar K, Mehta PD. Anxiety, depression, and quality of life in women with polycystic ovarian syndrome. Indian J Psychol Med (2018) 40. DOI: 10.4103/IJPSYM.IJPSYM_561_17 52. Cooney LG, Lee I, Sammel MD, Dokras A. High prevalence of moderate and severe depressive and anxiety symptoms in polycystic ovary syndrome: A systematic review and meta-analysis. Hum Reprod (2017) 32. DOI: 10.1093/humrep/dex044 53. Damone AL, Joham AE, Loxton D, Earnest A, Teede HJ, Moran LJ. Depression, anxiety and perceived stress in women with and without PCOS: A community-based study. Psychol Med (2019) 49. DOI: 10.1017/S0033291718002076 54. Dokras A, Stener-Victorin E, Yildiz BO, Li R, Ottey S, Shah D. Androgen excess- polycystic ovary syndrome society: Position statement on depression, anxiety, quality of life, and eating disorders in polycystic ovary syndrome. Fertil Steril (2018) 109. DOI: 10.1016/j.fertnstert.2018.01.038 55. Liu M, Murthi S, Poretsky L. Polycystic ovary syndrome and gender identity. Yale J Biol Med (2020) 93 56. Wahlin-Jacobsen S, Pedersen AT, Kristensen E, Laessoe NC, Lundqvist M, Cohen AS. Is there a correlation between androgens and sexual desire in women. J Sex Med (2015) 12. DOI: 10.1111/jsm.12774 57. Roland AV, Nunemaker CS, Keller SR, Moenter SM. Prenatal androgen exposure programs metabolic dysfunction in female mice. J Endocrinol (2010) 207. DOI: 10.1677/JOE-10-0217 58. Sati A, Prescott M, Holland S, Jasoni CL, Desroziers E, Campbell RE. Morphological evidence indicates a role for microglia in shaping the PCOS-like brain. J Neuroendocrinol (2021) 33 e12999. DOI: 10.1111/jne.12999 59. Manti M, Fornes R, Qi X, Folmerz E, Linden Hirschberg A, de Castro Barbosa T. Maternal androgen excess and obesity induce sexually dimorphic anxiety-like behavior in the offspring. FASEB J (2018) 32. DOI: 10.1096/fj.201701263RR 60. Risal S, Manti M, Lu H, Fornes R, Larsson H, Benrick A. Prenatal androgen exposure causes a sexually dimorphic transgenerational increase in offspring susceptibility to anxiety disorders. Transl Psychiatry (2021) 11 45. DOI: 10.1038/s41398-020-01183-9 61. Risal S, Pei Y, Lu H, Manti M, Fornes R, Pui HP. Prenatal androgen exposure and transgenerational susceptibility to polycystic ovary syndrome. Nat Med (2019) 25. DOI: 10.1038/s41591-019-0666-1 62. Brock O, Bakker J. Potential contribution of prenatal estrogens to the sexual differentiation of mate preferences in mice. Horm Behav (2011) 59. DOI: 10.1016/j.yhbeh.2010.10.012 63. Jean A, Bonnet P, Liere P, Mhaouty-Kodja S, Hardin-Pouzet H. Revisiting medial preoptic area plasticity induced in male mice by sexual experience. Sci Rep (2017) 7 17846. DOI: 10.1038/s41598-017-18248-3 64. Jean A, Mhaouty-Kodja S, Hardin-Pouzet H. Hypothalamic cellular and molecular plasticity linked to sexual experience in male rats and mice. Front Neuroendocrinol (2021) 63 100949. DOI: 10.1016/j.yfrne.2021.100949 65. Hellier V, Brock O, Candlish M, Desroziers E, Aoki M, Mayer C. Female sexual behavior in mice is controlled by kisspeptin neurons. Nat Commun (2018) 9 400. DOI: 10.1038/s41467-017-02797-2 66. Clarkson J, Herbison AE. Postnatal development of kisspeptin neurons in mouse hypothalamus; sexual dimorphism and projections to gonadotropin-releasing hormone neurons. Endocrinol (2006) 147. DOI: 10.1210/en.2006-0787 67. Clarkson J, Busby ER, Kirilov M, Schutz G, Sherwood NM, Herbison AE. Sexual differentiation of the brain requires perinatal kisspeptin-GnRH neuron signaling. J Neurosci (2014) 34. DOI: 10.1523/JNEUROSCI.3061-14.2014 68. Blaustein JD. Neuroendocrine regulation of feminine sexual behavior: lessons from rodent models and thoughts about humans. Annu Rev Psychol (2008) 59 93-118. DOI: 10.1146/annurev.psych.59.103006.093556 69. Blaustein JD, Tetel MJ, Ricciardi KH, Delville Y, Turcotte JC. Hypothalamic ovarian steroid hormone-sensitive neurons involved in female sexual behavior. Psychoneuroendocrinology (1994) 19. DOI: 10.1016/0306-4530(94)90036-1 70. Pfaff D, Frohlich J, Morgan M. Hormonal and genetic influences on arousal–sexual and otherwise. Trends Neurosci (2002) 25 45-50. DOI: 10.1016/S0166-2236(00)02084-1 71. Naule L, Marie-Luce C, Parmentier C, Martini M, Albac C, Trouillet AC. Revisiting the neural role of estrogen receptor beta in male sexual behavior by conditional mutagenesis. Horm Behav (2016) 80 1-9. DOI: 10.1016/j.yhbeh.2016.01.014 72. Trouillet AC, Ducroq S, Naule L, Capela D, Parmentier C, Radovick S. Deletion of neural estrogen receptor alpha induces sex differential effects on reproductive behavior in mice. Commun Biol (2022) 5 383. DOI: 10.1038/s42003-022-03324-w 73. Bormann CL, Smith GD, Padmanabhan V, Lee TM. Prenatal testosterone and dihydrotestosterone exposure disrupts ovine testicular development. Reproduction (2011) 142. DOI: 10.1530/REP-10-0210 74. Herrera-Morales WV, Herrera-Solis A, Nunez-Jaramillo L. Sexual behavior and synaptic plasticity. Arch Sex Behav (2019) 48. DOI: 10.1007/s10508-019-01483-2 75. Meisel RL, Mullins AJ. Sexual experience in female rodents: cellular mechanisms and functional consequences. Brain Res (2006) 1126 56-65. DOI: 10.1016/j.brainres.2006.08.050 76. Silva MSB, Decoster L, Trova S, Mimouni NEH, Delli V, Chachlaki K. Female sexual behavior is disrupted in a preclinical mouse model of PCOS via an attenuated hypothalamic nitric oxide pathway. Proc Natl Acad Sci U S A (2022) 119. DOI: 10.1073/pnas.2203503119 77. Tata B, Mimouni NEH, Barbotin AL, Malone SA, Loyens A, Pigny P. Elevated prenatal anti-mullerian hormone reprograms the fetus and induces polycystic ovary syndrome in adulthood. Nat Med (2018) 24. DOI: 10.1038/s41591-018-0035-5 78. Handa RJ, Pak TR, Kudwa AE, Lund TD, Hinds L. An alternate pathway for androgen regulation of brain function: Activation of estrogen receptor beta by the metabolite of dihydrotestosterone, 5alpha-androstane-3beta,17beta-diol. Horm Behav (2008) 53. DOI: 10.1016/j.yhbeh.2007.09.012 79. Jamieson BB, Moore AM, Lohr DB, Thomas SX, Coolen LM, Lehman MN. Prenatal androgen treatment impairs the suprachiasmatic nucleus arginine-vasopressin to kisspeptin neuron circuit in female mice. Front Endocrinol (Lausanne) (2022) 13. DOI: 10.3389/fendo.2022.951344 80. Sir-Petermann T, Codner E, Perez V, Echiburu B, Maliqueo M, Ladron de Guevara A. Metabolic and reproductive features before and during puberty in daughters of women with polycystic ovary syndrome. J Clin Endocrinol Metab (2009) 94. DOI: 10.1210/jc.2008-2836 81. Stener-Victorin E, Holm G, Labrie F, Nilsson L, Janson PO, Ohlsson C. Are there any sensitive and specific sex steroid markers for polycystic ovary syndrome. J Clin Endocrinol Metab (2010) 95. DOI: 10.1210/jc.2009-1908 82. Anjum S, Khattak MNK, Tsutsui K, Krishna A. RF-amide related peptide-3 (RFRP-3): A novel neuroendocrine regulator of energy homeostasis, metabolism, and reproduction. Mol Biol Rep (2021) 48. DOI: 10.1007/s11033-021-06198-z 83. Peragine DE, Pokarowski M, Mendoza-Viveros L, Swift-Gallant A, Cheng HM, Bentley GE. RFamide-related peptide-3 (RFRP-3) suppresses sexual maturation in a eusocial mammal. Proc Natl Acad Sci U S A (2017) 114. DOI: 10.1073/pnas.1616913114 84. Rahdar P, Khazali H. Rfamide-related peptide-3 suppresses the substance p-induced promotion of the reproductive performance in female rats modulating hypothalamic kisspeptin expression. Exp Brain Res (2020) 238. DOI: 10.1007/s00221-020-05860-5 85. Schneider JE, Benton NA, Russo KA, Klingerman CM, Williams WP, Simberlund J. RFamide-related peptide-3 and the trade-off between reproductive and ingestive behavior. Integr Comp Biol (2017) 57. DOI: 10.1093/icb/icx097
--- title: Predisposing factors for admission to intensive care units of patients with COVID-19 infection—Results of the German nationwide inpatient sample authors: - Karsten Keller - Ioannis T. Farmakis - Luca Valerio - Sebastian Koelmel - Johannes Wild - Stefano Barco - Frank P. Schmidt - Christine Espinola-Klein - Stavros Konstantinides - Thomas Münzel - Ingo Sagoschen - Lukas Hobohm journal: Frontiers in Public Health year: 2023 pmcid: PMC9975593 doi: 10.3389/fpubh.2023.1113793 license: CC BY 4.0 --- # Predisposing factors for admission to intensive care units of patients with COVID-19 infection—Results of the German nationwide inpatient sample ## Abstract ### Background Intensive care units (ICU) capacities are one of the most critical determinants in health-care management of the COVID-19 pandemic. Therefore, we aimed to analyze the ICU-admission and case-fatality rate as well as characteristics and outcomes of patient admitted to ICU in order to identify predictors and associated conditions for worsening and case-fatality in this critical ill patient-group. ### Methods We used the German nationwide inpatient sample to analyze all hospitalized patients with confirmed COVID-19 diagnosis in Germany between January and December 2020. All hospitalized patients with confirmed COVID-19 infection during the year 2020 were included in the present study and were stratified according ICU-admission. ### Results Overall, 176,137 hospitalizations of patients with COVID-19-infection ($52.3\%$ males; $53.6\%$ aged ≥70 years) were reported in Germany during 2020. Among them, 27,053 ($15.4\%$) were treated in ICU. COVID-19-patients treated on ICU were younger [70.0 (interquartile range (IQR) 59.0–79.0) vs. 72.0 (IQR 55.0–82.0) years, $P \leq 0.001$], more often males (66.3 vs. $48.8\%$, $P \leq 0.001$), had more frequently cardiovascular diseases (CVD) and cardiovascular risk-factors with increased in-hospital case-fatality (38.4 vs. $14.2\%$, $P \leq 0.001$). ICU-admission was independently associated with in-hospital death [OR 5.49 ($95\%$ CI 5.30–5.68), $P \leq 0.001$]. Male sex [OR 1.96 ($95\%$ CI 1.90–2.01), $P \leq 0.001$], obesity [OR 2.20 ($95\%$ CI 2.10–2.31), $P \leq 0.001$], diabetes mellitus [OR 1.48 ($95\%$ CI 1.44–1.53), $P \leq 0.001$], atrial fibrillation/flutter [OR 1.57 ($95\%$ CI 1.51–1.62), $P \leq 0.001$], and heart failure [OR 1.72 ($95\%$ CI 1.66–1.78), $P \leq 0.001$] were independently associated with ICU-admission. ### Conclusion During 2020, $15.4\%$ of the hospitalized COVID-19-patients were treated on ICUs with high case-fatality. Male sex, CVD and cardiovascular risk-factors were independent risk-factors for ICU admission. ## Introduction During December 2019 first pneumonia cases of unknown origin were detected in China. The causative pathogen was identified as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1, 2]. Patients with SARS-CoV-2 infections, also shortly named as coronavirus disease 2019 (COVID-19), presented in in- and outpatient settings [1, 3]. First COVID-19 cases in Germany were detected at the end of January 2020 in Bavaria [3, 4] and a strong and fast spread from this initial cluster in the German population was observed [3, 5]. This spread of SARS-CoV-2 infections was accompanied by a previously unprecedented strain on healthcare systems worldwide [6]. In the first wave of the disease in 2020, the German healthcare system had the advantage that several European countries faced this strain some weeks before. Thus, German hospitals were in part able to benefit from experiences made in other healthcare systems in terms of risk stratification for outpatient care, hospital and ICU admissions. Nevertheless, in the early phase of the COVID-19-pandemic, decisions for risk stratification and ICU admission of COVID-19-patients were primarily based on physicians' experience regarding health care management of critical care and the unsorted reports of colleagues all over the world in the light of pending study results. In previously published studies analyzing also the German nationwide inpatient sample, we have shown that the in-hospital case-fatality rate of hospitalized patients with confirmed COVID-19-infection was ~$18\%$ in Germany during the year 2020 [3, 7] and the case-fatality rate increased dramatically if treatment on intensive care units (ICU) and/or mechanical ventilation were needed (3, 7–9). Since a large number of patients with severe respiratory and cardiovascular complications of COVID-19-infection had to be treated on ICU, in some areas, ICUs were completely overloaded [3, 6, 7, 10, 11]. Thus, ICU has to be considered as a bottleneck regarding health care planning and threatening critical overload of the national healthcare systems (3, 6–11). ICU availability, admission policy and health care structure vary across Europe as additionally the demographics and government-policies do [3, 6]. Beside the previously published results, it is of outstanding interest to understand determinants of ICU admission and outcome, which are both crucial factors for adequate health care planning, decision making and pandemic management [3, 6]. Therefore, we aimed to analyze the ICU admission and case-fatality rate in Germany and to identify characteristics and outcomes of patient admitted to ICU in order to identify predictors and associated conditions for worsening and case-fatality for this critical ill patients' group. ## Data source The Research Data Center (RDC) of the Federal Bureau of Statistics (Wiesbaden, Germany) calculated the statistical analyses and provided aggregated statistic-results on the basis of our SPSS codes (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, Version 20.0. IBM Corp: Armonk, NY, USA), which were previously supplied by us to the RDC (source: RDC of the Federal Statistical Office and the Statistical Offices of the federal states, DRG Statistics 2020, own calculations) [12, 13]. With this computed analysis of the German nationwide inpatient sample, we aimed to analyze temporal trends of all hospitalized patients with a confirmed COVID-19 diagnosis. All patients, who were treated in German hospitals with a COVID-19 infection confirmed by a laboratory test (ICD-code U07.1) during the observational period between January 1st and December 31st of the year 2020 were included in the present study. The COVID-19 patients were stratified for ICU treatment and we identified independent predictors of ICU admission during hospitalization (Figure 1). Figure 1: Flow-chart. COVID-19, coronavirus disease 2019; ICU, intensive care unit. ## Study oversight and support Since in the present study the investigators did not accessed directly individual patient data but only summarized results provided by the RDC, approval by an ethics committee as well as patients' informed consent were not required, in accordance with German law [12, 13]. ## Coding of diagnoses, procedures, and definitions After introduction of a diagnosis- and procedure-related remuneration system in Germany in the year 2004, coding according the German Diagnosis Related Groups (G-DRG) system with coding of patient data on diagnoses, coexisting conditions, and on surgeries/procedures/interventions and transferring these data/codes to the Institute for the Hospital Remuneration System is mandatory for German hospitals to get their remuneration [10, 11]. Therefore, patients' diagnoses are coded according to the International Statistical Classification of Diseases and Related Health Problems, 10th revision, with German modification (ICD-10-GM) [10, 11]. In addition, surgical/diagnostic/interventional procedures are coded according to OPS-codes (Operationen- und Prozedurenschlüssel). In our present of the German nationwide inpatient sample, we were able to identify all hospitalized patients with a confirmed COVID-19 diagnosis (ICD-code U07.1) in Germany during the year 2020 (COVID-19 as main or secondary diagnosis). Post-COVID-19 was defined as a status of previous survived COVID-19-infection before the patient's hospitalization with the actual COVID-19 infection. ## Study outcomes Primary study endpoint was admission on ICU. In addition, we analyzed occurrence of all-cause in-hospital death and the prevalence of major adverse cardiovascular and cerebrovascular events [MACCE, composite of all-cause in-hospital death, acute myocardial infarction (ICD-code I21), and/or ischemic stroke (ICD-code I63)]. Furthermore, we analyzed the occurrence of the aggravated respiratory manifestations pneumonia (ICD codes J12-J18) and acute respiratory distress syndrome (ARDS, ICD code J80) as well as other adverse events during hospitalization such as cardio-pulmonary resuscitation (OPS-code 8-77), venous thromboembolism (ICD codes I26, I80, I81, and I82), acute kidney failure (ICD-code N17), myocarditis (ICD code I40), myocardial infarction (acute and recurrent, ICD codes I21 and I22), stroke (ischemic or hemorrhagic, ICD codes I61-I64), intracerebral bleeding (ICD code I61), gastro-intestinal bleeding (ICD code K92.0, K92.1, and K92.2) and transfusion of blood constituents (OPS code 8-800). The outcomes were defined according current guidelines (14–23). The acute respiratory distress syndrome (ARDS) was defined in 1994 by the American-European Consensus Conference (AECC) and revised in 2011 with the Berlin Definition [15]. ## Statistical analysis Differences in patient characteristics between the groups of hospitalized COVID-19-patients with ICU treatment vs. without ICU treatment were calculated with Wilcoxon-Whitney U-test for continuous variables and Fisher's exact or chi2-test for categorical variables, as appropriate. Temporal trends regarding hospitalizations of COVID-19-patients with ICU treatment and in-hospital mortality over time and as well as trend-changes with increasing age were estimated by means of linear regression analyses. Results were presented as β-estimates and $95\%$ confidence intervals (CI). Logistic regression models were calculated to investigate associations between (I) patients' characteristics and ICU-admission as well as (II) associations between adverse events and ICU-admission. Furthermore, we calculated logistic regression models to analyse (III) the associations of patients' characteristics and in-hospital death in ICU-patients as well as (IV) the associations of adverse events during in-hospital course and in-hospital death in ICU-patients. In order to warrant that the results of the mentioned logistic regressions are not substantially biased by other influencing factors and therefore, guarantying a widely independence of important different cofactors during hospitalization, the multivariable logistic regressions were adjusted for age, sex, diabetes mellitus, cancer, heart failure, coronary artery disease, chronic obstructive pulmonary disease, essential arterial hypertension, chronic renal insufficiency (glomerular filtration rate <60 ml/min/1,73 m2), atrial fibrillation/flutter, hyperlipidemia, and obesity. Results were presented as Odds Ratios (OR) and $95\%$ CI. All statistical analyses were carried out with the use of SPSS software (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, Version 20.0. IBM Corp: Armonk, NY, USA). Only two-sided P-values <0.05 were considered to be statistically significant. No adjustment for multiple testing was applied. ## Baseline characteristics During the year 2020, 176,137 hospitalizations ($52.3\%$ males; $53.6\%$ aged 70 years or older) of patients with confirmed COVID-19-infection were reported in German hospitals. Of these inpatients, 27,053 ($15.4\%$) were admitted to ICU, while overall, 31,607 ($17.9\%$) died during hospitalization (Figure 1). ICU admission in COVID-19 patients was associated with increased case-fatality [univariate regression: OR 3.76 ($95\%$ CI 3.65–3.87), $P \leq 0.001$; multivariate regression: OR 5.49 ($95\%$ CI 5.30–5.68), $P \leq 0.001$]. The monthly percentage of COVID-19 patients admitted to ICUs of German hospitals decreased over time from $32.8\%$ in January 2020 to a minimum of $14.8\%$ in November 2020 [β −0.89 ($95\%$ CI −0.93 to −0.85), $P \leq 0.001$], while highest absolute numbers of total ICU admissions were observed in spring and winter of the year 2020 (Figure 2A). ICU admissions related to all COVID-19 hospitalizations increased with inclining age [β 0.11 ($95\%$ CI 0.09–0.14), $P \leq 0.001$] with a maximum in the 8th decade of life ($27.8\%$; Figure 2B). Figure 2: Temporal trends regarding total numbers of patients with COVID-19-infection admitted to ICU. (A) Temporal trends regarding total numbers of hospitalized patients with COVID-19-infection admitted to ICU (absolute numbers: blue bars; relative numbers: blue line) stratified for months. (B) Temporal trends regarding total numbers of hospitalized patients with COVID-19-infection admitted to ICU (absolute numbers: blue bars; relative numbers: blue line) stratified for age decades. COVID-19, coronavirus disease 2019; ICU, intensive care unit. ## Comparison of COVID-19-patients admitted to ICU vs. those without ICU treatment As aforementioned, ~$15\%$ of the hospitalized patients with COVID-19 in Germany were admitted to ICUs who were in median 2 years younger [70.0 (Interquartile range (IQR) 59.0–79.0) vs. 72.0 (IQR 55.0–82.0) years, $P \leq 0.001$] and more often of male sex (66.3 vs. $48.8\%$, $P \leq 0.001$) compared to those hospitalized, but treated outside the ICU (Table 1). COVID-19 patients admitted to ICU had more frequently cardiovascular risk factors (CVRF) and cardiovascular diseases (CVD) as well as lung and kidney diseases than those without ICU-treatment resulting in higher Charlson comorbidity index in ICU treated patients [5.0 (IQR 3.0–7.0) vs. 4.0 (IQR 1.0–6.0), $P \leq 0.001$; Table 1]. As expected, the aggravated respiratory manifestations of COVID-19-infections such as pneumonia (89.2 vs. $55.5\%$, $P \leq 0.001$) and acute respiratory distress syndrome (ARDS, 35.4 vs. $1.4\%$, $P \leq 0.001$) were more frequently found in patients, who were in need of intensive care treatment (Table 1). Table 1 \| Parameters \| COVID-19 without ICU (n = 149,084; 84.6%) \| COVID-19 with ICU (n = 27,053; 15.4%) \| P-value \| \| --- \| --- \| --- \| --- \| \| Age \| 72.0 (55.0 / 82.0) \| 70.0 (59.0 / 79.0) \| <0.001 \| \| Age ≥70 years \| 80,277 (53.8%) \| 14,052 (51.9%) \| <0.001 \| \| Female sex \| 74,834 (50.2%) \| 9,115 (33.7%) \| <0.001 \| \| In-hospital stay (days) \| 7.0 (3.0 / 12.0) \| 16.0 (9.0 / 26.0) \| <0.001 \| \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| \| Obesity \| 6,557 (4.4%) \| 2,826 (10.4%) \| <0.001 \| \| Diabetes mellitus \| 35,581 (23.9%) \| 9,651 (35.7%) \| <0.001 \| \| Essential arterial hypertension \| 68,080 (45.7%) \| 14,400 (53.2%) \| <0.001 \| \| Hyperlipidemia \| 22,651 (15.2%) \| 4,922 (18.2%) \| <0.001 \| \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| \| Coronary artery disease \| 20,174 (13.5%) \| 5,400 (20.0%) \| <0.001 \| \| Heart failure \| 20,521 (13.8%) \| 6,598 (24.4%) \| <0.001 \| \| Peripheral artery disease \| 4,398 (3.0%) \| 1,242 (4.6%) \| <0.001 \| \| Atrial fibrillation/flutter \| 26,478 (17.8%) \| 7,682 (28.4%) \| <0.001 \| \| Chronic obstructive pulmonary disease \| 9,486 (6.4%) \| 2,668 (9.9%) \| <0.001 \| \| Chronic renal insufficiency (glomerular filtration rate <60 ml/min/1,73 m2) \| 22,494 (15.1%) \| 4,878 (18.0%) \| <0.001 \| \| Cancer \| 7,416 (5.0%) \| 1,585 (5.9%) \| <0.001 \| \| Vasculopathy \| 218 (0.1%) \| 112 (0.4%) \| <0.001 \| \| Charlson comorbidity index \| 4.0 (1.0 / 6.0) \| 5.0 (3.0 / 7.0) \| <0.001 \| \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| \| Pneumonia \| 82,784 (55.5%) \| 24,129 (89.2%) \| <0.001 \| \| Acute respiratory distress syndrome \| 2,025 (1.4%) \| 9,569 (35.4%) \| <0.001 \| \| Markers of acute organ failure \| Markers of acute organ failure \| Markers of acute organ failure \| Markers of acute organ failure \| \| Sepsis \| 9,423 (6.3%) \| 4,042 (14.9%) \| <0.001 \| \| Encephalitis \| 14 (0.01%) \| 20 (0.07%) \| <0.001 \| \| Mild liver disease \| 1,267 (0.8%) \| 378 (1.4%) \| <0.001 \| \| Severe liver disease \| 2,216 (1.5%) \| 1,923 (7.1%) \| <0.001 \| \| Mechanical ventilation \| 2,720 (1.8%) \| 9,422 (34.8%) \| <0.001 \| \| Extracorporeal membrane oxygenation (ECMO) \| 65 (0.04%) \| 1,389 (5.1%) \| <0.001 \| \| Proteinuria \| 93 (0.1%) \| 42 (0.2%) \| <0.001 \| \| Dialysis \| 1,522 (1.0%) \| 4,053 (15.0%) \| <0.001 \| \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| \| In-hospital case-fatality \| 21,216 (14.2%) \| 10,391 (38.4%) \| <0.001 \| \| MACCE \| 23,696 (15.9%) \| 11,328 (41.9%) \| <0.001 \| \| Cardio-pulmonary resuscitation \| 1,099 (0.7%) \| 1,760 (6.5%) \| <0.001 \| \| Venous thromboembolism (VTE) \| 2,992 (2.0%) \| 1,995 (7.4%) \| <0.001 \| \| Acute kidney failure \| 12,144 (8.1%) \| 9,931 (36.7%) \| <0.001 \| \| Myocarditis \| 126 (0.1%) \| 100 (0.4%) \| <0.001 \| \| Myocardial infarction \| 1,624 (1.1%) \| 1,129 (4.2%) \| <0.001 \| \| Stroke (ischemic or hemorrhagic) \| 2,206 (1.5%) \| 990 (3.7%) \| <0.001 \| \| Intracerebral bleeding \| 279 (0.2%) \| 297 (1.1%) \| <0.001 \| \| Gastro-intestinal bleeding \| 2,133 (1.4%) \| 815 (3.0%) \| <0.001 \| \| Transfusion of blood constituents \| 5,906 (4.0%) \| 7,968 (29.5%) \| <0.001 \| ## Outcomes of COVID-19-patients admitted to ICU vs. those without ICU treatment MACCE (41.9 vs. $15.9\%$, $P \leq 0.001$) and in-hospital case-fatality (38.4 vs. $14.2\%$, $P \leq 0.001$) rates were substantially higher in patients with COVID-19-infection treated in ICU than in those without ICU treatment (Table 1). ICU treatment was independently associated with increased in-hospital case-fatality rate [OR 5.49 ($95\%$ CI 5.30–5.68), $P \leq 0.001$]. It has to be pointed out that the in-hospital case-fatality rate of COVID-19 patients on ICU was highest in months with high numbers of ICU admissions of COVID-19 patients (Figure 3A). In addition, the case-fatality rate of COVID-19 patients treated on German ICUs increased substantially with patients age (Figure 3C). Highest proportion of ARDS cases were observed in the initial phase of the pandemic during spring 2020 (March-April) and in the 6st to 8th decade of patients' life (Figures 3B, D). Figure 3: Temporal trends regarding total numbers of patients with COVID-19-infection admitted to ICU, in-hospital case-fatality, MACCE, and VTE rate. (A) Temporal trends regarding total numbers of patients with COVID-19-infection admitted to ICU (absolute numbers: blue bars) and rates of case-fatality, MACCE, and VTE stratified for months (lines). (B) Temporal trends regarding proportion of ARDS and pneumonia in patients with COVID-19-infection admitted to ICU stratified for months. (C) Temporal trends regarding total numbers of patients with COVID-19-infection admitted to ICU (absolute numbers: blue bars) and rates of case-fatality, MACCE, and VTE stratified for age decades (lines). (D) Temporal trends regarding proportion of ARDS and pneumonia in patients with COVID-19-infection admitted to ICU stratified for age decades. COVID-19, coronavirus disease 2019; ICU, intensive care unit; VTE, venous thromboembolism; MACCE, major adverse cardiovascular and cerebrovascular events; ARDS, acute respiratory distress syndrome. The rates of the following acute organ failures were increased in ICU patients: the rate of myocardial infarction was nearby 4-fold, whereas the rate of myocarditis was 3-fold increased and the stroke rate more than doubled in patients treated on ICU. While the rate of sepsis was more than doubled, occurrence of encephalitis was 7-fold increased and that of severe liver disease was nearby 5-fold inclined. Additionally, all investigated bleeding events and need for transfusion of blood constituents occurred significantly more often in ICU admitted patients (Table 1). Beside the bleeding events, the rate of venous thromboembolism was also more than 3-fold higher in ICU patients. Furthermore, the risk for acute kidney injury was more than 4-fold higher and, consequentially, dialysis was 15-fold more often performed in patients with COVID-19-infection treated on ICU (Table 1). ## Predictors of ICU admission of COVID-19-patients Male sex [OR 1.96 ($95\%$ CI 1.90–2.01), $P \leq 0.001$] and age younger than 70 years [OR 1.47 ($95\%$ CI 1.43–1.52), $P \leq 0.001$] were independent risk factors of ICU admission (Table 2). Table 2 \| Unnamed: 0 \| Univariate regression model \| Univariate regression model.1 \| Multivariate regression model * \| Multivariate regression model .1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| OR (95% CI) \| P -value \| OR (95% CI) \| P -value \| \| Age (per year) \| 1.003 (1.002–1.004) \| <0.001 \| 0.994 (0.993–0.995) \| <0.001 \| \| Age ≥70 years \| 0.926 (0.903–0.951) \| <0.001 \| 0.681 (0.660–0.702) \| <0.001 \| \| Female sex \| 0.504 (0.491–0.518) \| <0.001 \| 0.511 (0.497–0.526) \| <0.001 \| \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| \| Obesity \| 2.536 (2.421–2.655) \| <0.001 \| 2.201 (2.097–2.310) \| <0.001 \| \| Diabetes mellitus \| 1.769 (1.721–1.819) \| <0.001 \| 1.479 (1.436–1.524) \| <0.001 \| \| Essential arterial hypertension \| 1.354 (1.319–1.390) \| <0.001 \| 1.276 (1.240–1.314) \| <0.001 \| \| Hyperlipidemia \| 1.241 (1.200–1.284) \| <0.001 \| 0.908 (0.874–0.942) \| <0.001 \| \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| \| Coronary artery disease \| 1.594 (1.541–1.648) \| <0.001 \| 1.088 (1.048–1.130) \| <0.001 \| \| Heart failure \| 2.021 (1.958–2.085) \| <0.001 \| 1.719 (1.658–1.783) \| <0.001 \| \| Peripheral artery disease \| 1.583 (1.484–1.688) \| <0.001 \| 1.041 (0.972–1.114) \| 0.254 \| \| Atrial fibrillation/flutter \| 1.836 (1.783–1.891) \| <0.001 \| 1.566 (1.514–1.620) \| <0.001 \| \| Chronic obstructive pulmonary disease \| 1.610 (1.539–1.684) \| <0.001 \| 1.263 (1.204–1.324) \| <0.001 \| \| Chronic renal insufficiency (glomerular filtration rate <60 ml/min/1,73 m2) \| 1.238 (1.196–1.281) \| <0.001 \| 0.872 (0.839–0.906) \| <0.001 \| \| Cancer \| 1.189 (1.124–1.257) \| <0.001 \| 1.175 (1.110–1.245) \| <0.001 \| \| Charlson comorbidity index \| 1.146 (1.141–1.152) \| <0.001 \| – \| \| \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| \| Pneumonia \| 6.609 (6.352–6.877) \| <0.001 \| 6.421 (6.164–6.689) \| <0.001 \| \| Acute respiratory distress syndrome \| 39.746 (37.791–41.802) \| <0.001 \| 35.906 (34.104–37.803) \| <0.001 \| \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| \| Cardio-pulmonary resuscitation \| 9.370 (8.680–10.115) \| <0.001 \| 7.431 (6.862–8.047) \| <0.001 \| \| Venous thromboembolism \| 3.887 (3.668–4.120) \| <0.001 \| 3.782 (3.560–4.018) \| <0.001 \| \| Acute kidney failure \| 6.540 (6.341–6.746) \| <0.001 \| 5.987 (5.790–6.191) \| <0.001 \| \| Myocarditis \| 4.386 (3.372–5.705) \| <0.001 \| 3.744 (2.848–4.922) \| <0.001 \| \| Myocardial infarction \| 3.954 (3.661–4.271) \| <0.001 \| 3.158 (2.908–3.429) \| <0.001 \| \| Stroke (ischemic or hemorrhagic) \| 2.529 (2.344–2.729) \| <0.001 \| 2.277 (2.104–2.465) \| <0.001 \| \| Sepsis \| 2.603 (2.503–2.708) \| <0.001 \| 2.529 (2.427–2.634) \| <0.001 \| \| Encephalitis \| 7.878 (3.979–15.598) \| <0.001 \| 7.384 (3.652–14.929) \| <0.001 \| \| Mild liver disease \| 1.653 (1.473–1.856) \| <0.001 \| 1.326 (1.176–1.495) \| <0.001 \| \| Severe liver disease \| 5.072 (4.764–5.399) \| <0.001 \| 4.129 (3.868–4.408) \| <0.001 \| \| Intracerebral bleeding \| 5.920 (5.025–6.975) \| <0.001 \| 5.485 (4.626–6.504) \| <0.001 \| \| Gastro-intestinal bleeding \| 2.140 (1.972–2.322) \| <0.001 \| 1.907 (1.751–2.076) \| <0.001 \| \| Transfusion of blood constituents \| 10.121 (9.755–10.502) \| <0.001 \| 10.131 (9.735–10.542) \| <0.001 \| Regarding CVRF, obesity [OR 2.20 ($95\%$ CI 2.10–2.31), $P \leq 0.001$] as well as diabetes mellitus [OR 1.48 ($95\%$ CI 1.44–1.53), $P \leq 0.001$] were independent predictors of an increased need of ICU treatment (Table 2). Interestingly, the association of atrial fibrillation/flutter as well as heart failure with ICU treatment were stronger than that of coronary artery disease and chronic obstructive pulmonary disease, which were also associated with ICU admission (Table 2). The severe respiratory manifestations of COVID-19 pneumonia [OR 6.42 ($95\%$ CI 6.16–6.69), $P \leq 0.001$] and ARDS [OR 35.91 ($95\%$ CI 34.10–37.80), $P \leq 0.001$] were strongly and independently associated with ICU-admission. As expected, all adverse in-hospital events and acute organ failures were also associated with ICU-admission (Table 2). ## Risk factors for in-hospital death in COVID-19-patients treated on ICU Increasing age, male sex, obesity, diabetes mellitus and the CVD heart failure, atrial fibrillation/flutter as well as peripheral artery disease, but also chronic obstructive pulmonary disease, chronic renal insufficiency (glomerular filtration rate <60 ml/min/1.73 m2) and cancer were independent risk factors for in-hospital death in COVID-19-patients treated on ICUs in Germany (Table 3). Aggravated respiratory manifestations of COVID-19-infection including pneumonia as well as ARDS were associated with more than 3-fold risk for in-hospital death in ICU-patients. In addition, as expected, adverse events during hospitalization were also accompanied by increased risk for in-hospital death (Table 3). Table 3* \| Unnamed: 0 \| Univariate regression model \| Univariate regression model.1 \| Multivariate regression model * \| Multivariate regression model .1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| OR (95% CI) \| P -value \| OR (95% CI) \| P -value \| \| Age (per year) \| 1.069 (1.067–1.072) \| <0.001 \| 1.068 (1.065–1.070) \| <0.001 \| \| Age ≥70 years \| 4.071 (3.861–4.293) \| <0.001 \| 3.553 (3.352–3.765) \| <0.001 \| \| Female sex \| 0.964 (0.915–1.015) \| 0.161 \| 0.763 (0.720–0.809) \| <0.001 \| \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| Cardiovascular risk factors \| \| Obesity \| 0.814 (0.750–0.884) \| <0.001 \| 1.127 (1.028–1.236) \| <0.001 \| \| Diabetes mellitus \| 1.291 (1.227–1.358) \| <0.001 \| 1.107 (1.045–1.172) \| 0.001 \| \| Essential arterial hypertension \| 0.965 (0.919–1.014) \| 0.161 \| 0.711 (0.672–0.752) \| <0.001 \| \| Hyperlipidemia \| 1.040 (0.976–1.108) \| 0.225 \| 0.725 (0.674–0.779) \| <0.001 \| \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| \| Coronary artery disease \| 1.649 (1.553–1.751) \| <0.001 \| 0.997 (0.929–1.070) \| 0.937 \| \| Heart failure \| 2.037 (1.926–2.155) \| <0.001 \| 1.267 (1.187–1.352) \| <0.001 \| \| Peripheral artery disease \| 1.952 (1.741–2.189) \| <0.001 \| 1.293 (1.141–1.465) \| <0.001 \| \| Atrial fibrillation/flutter \| 2.283 (2.163–2.409) \| <0.001 \| 1.295 (1.219–1.376) \| <0.001 \| \| Chronic obstructive pulmonary disease \| 1.577 (1.456–1.709) \| <0.001 \| 1.195 (1.095–1.303) \| <0.001 \| \| Chronic renal insufficiency (glomerular filtration rate <60 ml/min/1,73 m2) \| 2.257 (2.119–2.403) \| <0.001 \| 1.337 (1.245–1.435) \| <0.001 \| \| Cancer \| 1.738 (1.570–1.924) \| <0.001 \| 1.697 (1.520–1.895) \| <0.001 \| \| Charlson comorbidity index \| 1.410 (1.394–1.425) \| <0.001 \| – \| \| \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| Respiratory manifestations of COVID-19 \| \| Pneumonia \| 2.725 (2.480–2.994) \| <0.001 \| 3.441 (3.104–3.815) \| <0.001 \| \| Acute respiratory distress syndrome \| 2.106 (2.001–2.217) \| <0.001 \| 3.049 (2.872–3.237) \| <0.001 \| \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| Adverse events during hospitalization \| \| Cardio-pulmonary resuscitation \| 6.164 (5.497–6.913) \| <0.001 \| 7.130 (6.291–8.082) \| <0.001 \| \| Venous thromboembolism \| 1.075 (0.979–1.180) \| 0.129 \| 1.367 (1.235–1.513) \| <0.001 \| \| Acute kidney failure \| 4.305 (4.084–4.538) \| <0.001 \| 3.972 (3.747–4.211) \| <0.001 \| \| Myocarditis \| 0.686 (0.447–1.053) \| 0.085 \| 0.921 (0.562–1.507) \| 0.743 \| \| Myocardial infarction \| 1.383 (1.227–1.559) \| <0.001 \| 1.002 (0.876–1.145) \| 0.979 \| \| Stroke (ischemic or hemorrhagic) \| 1.755 (1.545–1.993) \| <0.001 \| 1.851 (1.611–2.127) \| <0.001 \| \| Intracerebral bleeding \| 2.679 (2.116–3.391) \| <0.001 \| 4.676 (3.620–6.041) \| <0.001 \| \| Gastro-intestinal bleeding \| 1.825 (1.587–2.099) \| <0.001 \| 1.543 (1.325–1.797) \| <0.001 \| \| Transfusion of blood constituents \| 2.321 (2.201–2.449) \| <0.001 \| 2.339 (2.203–2.482) \| <0.001 \| ## Discussion One of the most critical determinants in the worldwide health care management of the COVID-19 pandemic are the local ICU capacities [3, 8, 9]. This was impressively obvious in several epicenters of the COVID-19 pandemic with dramatic high case-fatality rates due to overloaded local health care and especially ICU capacities (3, 10, 11, 24–26). Our study analyzing more than 175,000 hospitalizations of inpatients with COVID-19 infection revealed a substantially higher in-hospital case-fatality rate ($24.2\%$ higher) and MACCE rate ($26.0\%$ higher), if an ICU treatment was required in the not-vaccinated German population. As the vaccination program started in Germany not before late December 2020, the wide majority of hospitalized patients with COVID-19 during the year 2020 were not vaccinated and vaccination has no influence on the outcomes of our present study. The need for ICU-treatment was independently associated with 5.4-fold elevated risk for in-hospital case-fatality rate. In accordance with our results, previously published studies have also revealed high case-fatality rates of COVID-19-patients admitted to ICUs and emphasized the importance of accessible ICU beds, ventilator capacities and trained staff to manage the COVID-19 pandemic adequately [3, 8, 27]. This was underlined by data of the United States of America showing that $79\%$ of the hospital beds at the ICUs were occupied by COVID-19-patients at the peak of the pandemic during January 2021 [28]. The proportion of patients with COVID-19-infection, who were transferred to an ICU in Germany, was $15.4\%$ and therefore, comparable to proportions in France ($16.4\%$) [29], United Kingdom ($17.0\%$) [30], and in the Unites States of America (10.2–$19.6\%$) (31–33), but lower than in other countries such as Spain ($26.3\%$) [34] or Iran ($19.0\%$) [35]. Pooled ICU admission rate among 17,639 hospitalized COVID-19 patients meta-analyzed from eight studies worldwide was reported as $21\%$ [36]. While the highest absolute numbers of total ICU admissions due to COVID-19-patients in the year 2020 were observed in spring and winter, the monthly percentage of COVID-19 patients treated at the ICU of German hospitals decreased over time from $32.8\%$ in January 2020 to a minimum of $14.8\%$ at November 2020 and revealed only small variations between May and December 2020. In accordance with our finding of a high case-fatality rate of COVID-19-patients treated in German ICUs during spring and winter of the year 2020 when ICU demands were highest (Figure 3A), other studies have also shown that the ICU capacities and the ICU demand are important factors for COVID-19 patients' outcome [33, 37]. COVID-19 patients who needed ICU-treatment during periods of increased COVID-19 ICU demand had an increased risk of mortality compared with patients treated during periods of low COVID-19 ICU demand, whereas no association between COVID-19 ICU demand and mortality was observed for patients with COVID-19 treated outside the ICUs [37]. This finding is of outstanding interest for adequate pandemic management. In addition, significant variations regarding in-hospital case-fatality rate across European countries were observed [38, 39]. The in-hospital case-fatality rate of COVID-19 patients treated on ICU during the year 2020 in Germany identified by our study was $38.4\%$ and therefore higher than the rates reported in studies from Spain, Andorra and Ireland ($30.7\%$) [27], France ($31.0\%$) [40], United Kingdom ($32.0\%$) [30], Spain (16.7–$34.0\%$) [41, 42], Netherlands (23.4–$32.0\%$) [43], United States of America (21.0–$29.7\%$) [32, 44], Sweden [$17.4\%$ (in-hospital mortality)−$32.1\%$ (60-day mortality after ICU discharge)] [45, 46], Iceland ($14.8\%$) [47], was similar to rates in China ($37.0\%$) [48] as well as in Denmark ($37.0\%$) [49] and lower than in Iran ($42.0\%$) [35], Russia ($65.4\%$) [50], and Brazil ($59.0\%$) [51]. Two large review article including data of ICUs around the world reported summarized worldwide ICU mortality rates of $28.3\%$ [36] and $35.5\%$ [52], the second close to the value we calculated for Germany. As aforementioned, inter-country differences regarding the COVID-19 patients' outcome are strongly impacted by ratio of ICU capacities and ICU demand [33, 37]. In addition, patients' age, sex-distribution and comorbidities are important for these observed differences [27, 31, 43, 46, 53]. In particular, the age-dependency of COVID-19 case-fatality is well-known and very important in this context [3]. Therefore, variations in median age of the different COVID-19-cohorts in the different countries influence the case-fatality rates and might contribute to these variations. For example, in the Swedish COVID-19 intensive care cohort [46] as well as in cohort studies of the United States of America [31], the median age of the ICU patients was more than 10 years lower [31, 46] and in the cohort study in Spain, Andorra and Ireland the median age was 8 years lower than in Germany [27]. In line with this age-comparison, age ≥70 years was a strong predictor of in-hospital death of COVID-19-patients admitted to ICUs in Germany. In addition, aggravated respiratory status such as pneumonia and ARDS as well as acute kidney injury were strongly and independently associated with increased in-hospital case-fatality. However, since not all COVID-19 patients without dyspnoea, who were treated on normal ward, will be and were examined with X-ray, the proportion of pneumonia in this patient group might be underestimated. The prevalence rates of all investigated acute organ failures were substantially higher in ICU patients than in the COVID-19 patients treated on normal ward. Especially, rates of cardiac involvement, but also stroke, encephalitis and sepsis were substantially elevated. Sepsis is a common complication in COVID-19 patients and was detected in $7.6\%$ of all hospitalized COVID-19 patients and in more than $14\%$ of the COVID-19 patients treated on ICU in Germany in the year 2020. However, studies indicate that this number might be underestimated and the real rate of viral sepsis in hospitalized COVID-19 patients might be significantly higher [54]. Cardiac involvement is a known phenomenon and complication in patients suffering from COVID-19-infection (3, 55–58) and comprises predominantly myocardial infarction as well as myocarditis. COVID-19 was identified as a risk factor for acute myocardial infarction and myocarditis (3, 55–60). In studies, SARS-CoV-2 was associated with an increased risk of both arterial and venous thrombotic complications and in particular the risk of myocardial infarction was approximately doubled in the first 7 days after COVID-19 diagnosis [60]. Myocarditis incidence of hospitalized patients was reported ranging between 2.4 and 4.1 cases per 1,000 COVID-19 patients in a multi-center study of centers of different European countries and the United States of America [56, 58]. Cerebral complications such as stroke and encephalitis were reported in studies [59, 61], but our data underlines the importance and impact of these widely overlooked complications. These different acute organ failures are key drivers of in-hospital mortality and therefore, the early detection of impending complications as well as prevention and treatment of these acute organ failures is of major interest for adequate management of COVID-19 patients. It is additionally of outstanding interest, that all investigated bleeding events occurred more often in COVID-19 patients admitted to ICUs. Bleeding events during hospitalization were in different studies strongly associated with increased in-hospital death [62, 63]. Although the COVID-19 pandemic was primarily managed by vaccination programme after the year 2020 (vaccination program started in Germany at late December 2020 and accelerated in the following years) [64], nevertheless, ~$\frac{1}{4}$ of the German population has still no basic immunization by a COVID-19 vaccination at the beginning of the year 2023 [65]. Therefore, risk factors for ICU admission in not-vaccinated patients are still important for health care management. While the understanding of impacting factors on ICU admission and outcome is crucial for adequate health care planning, decision making, and pandemic management [3, 6, 66], the understanding of these factors remains still unsatisfying [66]. We identified male sex as well as obesity and diabetes mellitus as independent predictors for an increased probability of ICU treatment. These findings is consistent with previously published study results in which obesity and diabetes mellitus were associated with aggravated outcome in hospitalized patients with COVID-19 (3, 7, 67–70). In accordance with contemporary literature [71], the prevalence of CVD is distinctly higher in COVID-19-patients requiring ICU care. Interestingly, the association of atrial fibrillation/flutter as well as heart failure with ICU-admission were stronger than the associations of coronary artery disease and chronic obstructive pulmonary disease with ICU-admission, respectively. Atrial fibrillation is the most common arrhythmia in patients with COVID-19-pneumonia affecting ~$20\%$ of the patients with severe COVID-19 pneumonia during ICU stay [72]. In patients with COVID-19, arrhythmias and chronic heart failure are key factors of the development of acute heart failure [73] and heart failure diagnosis is associated with aggravated mortality in COVID-19 patients [3, 73]. COVID-19-infection is associated with increased risk of arterial und venous thrombosis with resulting ischemic events and patients with myocardial infarction infected by COVID-19 had an unfavorable outcome in comparison to those patients without COVID-19 infection (3, 60, 71, 74–77). Among COVID-19 patients, higher proportions of patients with COPD have to be admitted to ICU and treated with mechanical ventilation [78]. Consequently, COPD is an independent risk factor for ICU admission and all-cause mortality in COVID-19 patients [78, 79]. In line with our results, other studies identified patients age, arterial hypertension, diabetes mellitus, chronic renal failure, bronchial asthma, obesity and immunosuppression as independently associated with ICU admission during COVID-19-infection [46, 80]. Nevertheless, it has to be mentioned, that we were not able to distinguish whether the adverse in-hospital events occurred during, before or after ICU treatment during the hospitalization of the COVID-19 patients. However, the aim of the study was to emphasize and illustrate the stress and strain of the ICU in Germany. Total number of patients necessitating ICU admission in relation to provided ICU capacities have to be taken into account for adequate health care and pandemic planning [3, 8, 9]. Based on the knowledge regarding regional differences, epicenters of the COVID-19 pandemic with very high mortality rates, it is of outmost importance to identify trends and factors affecting ICU admission to avoid a critical overload of the healthcare system and particularly of the ICUs with increasing mortality rates [5, 8, 9]. ## Limitations Certain limitations of the present study merit consideration: First, as our results are based on administrative data, we cannot exclude misclassification or inconsistencies. Additionally, our analysis of the German nationwide inpatient sample was not pre-specified and thus, findings of the study can only be considered to be hypothesis-generating. Second, patients with confirmed COVID-19 infection, who died out of hospital, were not included in the German nationwide inpatient sample. Third, the German nationwide inpatient sample does not report follow-up-outcomes after the discharge from hospital. Fourth, coding on medical treatments is only incompletely captured (especially regarding immunotherapy such as dexamethasone, tocilizumab, anakinra, and baricitinib). ## Conclusion During the year 2020, $15.4\%$ of the hospitalized COVID-19-patients were admitted to ICUs in German hospitals. Important and independent risk factors for ICU admission are male sex, CVRF such as obesity and diabetes mellitus as well as several cardio-pulmonary diseases including atrial fibrillation/flutter, heart failure, coronary artery disease and chronic obstructive pulmonary disease. ICU-admission was accompanied by a case-fatality rate of $38.4\%$ and for this substantially higher than the rate $14.2\%$ on normal-ward treatment. COVID-19 patients who were treated in ICUs during periods of increased ICU demand had an increased risk of mortality compared to patients treated during periods of low COVID-19 ICU demand. These findings highlight to draw more attention to predictors for ICU admission in patients hospitalized with COVID-19 in order to optimize monitoring and prevention strategies, avoid critical overload of the healthcare system and particularly of the ICUs in order to prevent the subsequent increase in mortality rates. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Ethics statement Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent from the participants' legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements. ## Author contributions KK and LH conceived this study, led the writing of the paper, accessed and verified the data, and contributed to the study design. LH, IF, LV, SKoe, JW, SB, FS, CE-K, SKon, TM, and IS commented on the paper, oversaw the analysis, and edited the final manuscript. KK led the data analysis with support from LH. All authors had full access to all the data, contributed to drafting the paper, revised the manuscript for important intellectual content, and had final responsibility for the decision to submit for publication. ## Conflict of interest SB received lecture/consultant fees from Bayer HealthCare, Concept Medical, BTG Pharmaceuticals, INARI, Boston Scientific, and LeoPharma; institutional grants from Boston Scientific, Bentley, Bayer HealthCare, INARI, Medtronic, Concept Medical, Bard, and Sanofi; and economical support for travel/congress costs from Daiichi Sankyo, BTG Pharmaceuticals, and Bayer HealthCare, outside the submitted work. CE-K reports having from Amarin Germany, Amgen GmbH, Bayer Vital, Boehringer Ingelheim, Bristol-Myers Squibb, Daiichi Sankyo, Leo Pharma, MSD Sharp & Dohme, Novartis Pharma, Pfizer Pharma GmbH, and Sanofi-Aventis GmbH. SKon reports institutional grants and personal lecture/advisory fees from Bayer AG, Daiichi Sankyo, and Boston Scientific; institutional grants from Inari Medical; and personal lecture/advisory fees from MSD and Bristol Myers Squibb/Pfizer. TM is PI of the DZHK (German Center for Cardiovascular Research), Partner Site Rhine-Main, Mainz, Germany. LH received lecture/consultant fees from MSD and Actelion, outside the submitted work. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Guan WJ, Ni ZY, Hu Y, Liang WH, Ou CQ, He JX. Clinical characteristics of coronavirus disease 2019 in China. N Engl J Med.* (2020) 382 1708-20. DOI: 10.1056/NEJMoa2002032 2. Huang C, Wang Y, Li X, Ren L, Zhao J, Hu Y. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet. (2020) 395 497-506. DOI: 10.1016/S0140-6736(20)30183-5 3. Hobohm L, Sagoschen I, Barco S, Schmidtmann I, Espinola-Klein C, Konstantinides S. Trends and risk factors of in-hospital mortality of patients with COVID-19 in Germany: Results of a large nationwide inpatient sample. Viruses. (2022) 14 20275. DOI: 10.3390/v14020275 4. Böhmer MM, Buchholz U, Corman VM, Hoch M, Katz K, Marosevic DV. Investigation of a COVID-19 outbreak in Germany resulting from a single travel-associated primary case: A case series. Lancet Infect Dis. (2020) 20 920-8. DOI: 10.1016/S1473-3099(20)30314-5 5. Ullrich A, Schranz M, Rexroth U, Hamouda O, Schaade L, Diercke M. Impact of the COVID-19 pandemic and associated non-pharmaceutical interventions on other notifiable infectious diseases in Germany: An analysis of national surveillance data during week 1-2016—week 32-2020. Lancet Reg Health Eur. (2021) 6 100103. DOI: 10.1016/j.lanepe.2021.100103 6. Qian Z, Alaa AM, van der Schaar M, Ercole A. Between-centre differences for COVID-19 ICU mortality from early data in England. Intensive Care Med. (2020) 46 1779-80. DOI: 10.1007/s00134-020-06150-y 7. Keller K, Sagoschen I, Schmitt VH, Sivanathan V, Espinola-Klein C, Lavie CJ. Obesity and its impact on adverse in-hospital outcomes in hospitalized patients with COVID-19. Front Endocrinol. (2022) 13 876028. DOI: 10.3389/fendo.2022.876028 8. Karagiannidis C, Windisch W, McAuley DF, Welte T, Busse R. Major differences in ICU admissions during the first and second COVID-19 wave in Germany. Lancet Respirat Med. (2021) 9 e47-e8. DOI: 10.1016/S2213-2600(21)00101-6 9. Karagiannidis C, Mostert C, Hentschker C, Voshaar T, Malzahn J, Schillinger G. Case characteristics, resource use, and outcomes of 10 021 patients with COVID-19 admitted to 920 German hospitals: An observational study. Lancet Respir Med. (2020) 8 853-62. DOI: 10.1016/S2213-2600(20)30316-7 10. Voci D, Fedeli U, Farmakis IT, Hobohm L, Keller K, Valerio L. Deaths related to pulmonary embolism and cardiovascular events before and during the 2020 COVID-19 pandemic: An epidemiological analysis of data from an Italian high-risk area. Thromb Res. (2022) 212 44-50. DOI: 10.1016/j.thromres.2022.02.008 11. Goyal P, Choi JJ, Pinheiro LC, Schenck EJ, Chen R, Jabri A. Clinical characteristics of COVID-19 in New York City. N Engl J Med. (2020) 382 2372-4. DOI: 10.1056/NEJMc2010419 12. Reinöhl J, Kaier K, Reinecke H, Schmoor C, Frankenstein L, Vach W. Effect of availability of transcatheter aortic-valve replacement on clinical practice. N Engl J Med. (2015) 373 2438-47. DOI: 10.1056/NEJMoa1500893 13. Keller K, Hobohm L, Ebner M, Kresoja K-P, Münzel T, Konstantinides SV. Trends in thrombolytic treatment and outcomes of acute pulmonary embolism in Germany. Eur Heart J. (2020) 41 522-9. DOI: 10.1093/eurheartj/ehz236 14. Olson G, Davis AM. Diagnosis and treatment of adults with community-acquired pneumonia. An official clinical practice guideline of the American Thoracic Society and Infectious Diseases Society of America. Am J Respir Crit Care Med. (2019) 200 e45-67. DOI: 10.1164/rccm.201908-1581ST 15. Definition Task Force ARDS, Ranieri VM, Rubenfeld GD, Thompson BT, Ferguson ND, Caldwell E. Acute respiratory distress syndrome: The Berlin definition. J Am Med Assoc. (2012) 307 2526-33. DOI: 10.1001/jama.2012.5669 16. Papazian L, Aubron C, Brochard L, Chiche J-D, Combes A, Dreyfuss D. Formal guidelines: Management of acute respiratory distress syndrome. Ann Intensive Care. (2019) 9 69. DOI: 10.1186/s13613-019-0540-9 17. Soar J, Böttiger BW, Carli P, Couper K, Deakin CD, Djärv T. European resuscitation council guidelines 2021: Adult advanced life support. Resuscitation. (2021) 161 115-51. DOI: 10.1016/j.resuscitation.2021.02.010 18. Konstantinides SV, Meyer G. The 2019 ESC guidelines on the diagnosis and management of acute pulmonary embolism. Eur Heart J. (2019) 40 3453-5. DOI: 10.1093/eurheartj/ehz726 19. Mehran R, Rao SV, Bhatt DL, Gibson CM, Caixeta A, Eikelboom J. Standardized bleeding definitions for cardiovascular clinical trials: A consensus report from the Bleeding Academic Research Consortium. Circulation. (2011) 123 2736-47. DOI: 10.1161/CIRCULATIONAHA.110.009449 20. McDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, Böhm M. 2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J. (2021) 42 3599-726. DOI: 10.1093/eurheartj/ehab368 21. Ibanez B, James S, Agewall S, Antunes MJ, Bucciarelli-Ducci C, Bueno H. 2017 ESC Guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation: The Task Force for the management of acute myocardial infarction in patients presenting with ST-segment elevation of the European Society of Cardiology (ESC). Eur Heart J. (2018) 39 119-77. DOI: 10.1093/eurheartj/ehx393 22. Collet JP, Thiele H, Barbato E, Barthélémy O, Bauersachs J, Bhatt DL. 2020 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J. (2021) 42 1289-367. DOI: 10.1093/eurheartj/ehaa575 23. Bosco E, Hsueh L, McConeghy KW, Gravenstein S, Saade E. Major adverse cardiovascular event definitions used in observational analysis of administrative databases: A systematic review. BMC Med Res Methodol. (2021) 21 241. DOI: 10.1186/s12874-021-01440-5 24. Grasselli G, Pesenti A, Cecconi M. Critical care utilization for the COVID-19 outbreak in Lombardy, Italy: Early experience and forecast during an emergency response. J Am Med Assoc. (2020) 323 1545-6. DOI: 10.1001/jama.2020.4031 25. Maruotti A, Jona-Lasinio G, Divino F, Lovison G, Ciccozzi M, Farcomeni A. Estimating COVID-19-induced excess mortality in Lombardy, Italy. Aging Clin Exp Res. (2022) 34 475-9. DOI: 10.1007/s40520-021-02060-1 26. Fagiuoli S, Lorini FL, Remuzzi G, COVID-19 Bergamo Hospital Crisis U. Adaptations and lessons in the province of Bergamo. N Engl J Med. (2020) 382 e71. DOI: 10.1056/NEJMc2011599 27. Carbonell R, Urgelés S, Rodríguez A, Bodí M, Martín-Loeches I, Solé-Violán J. Mortality comparison between the first and second/third waves among 3,795 critical COVID-19 patients with pneumonia admitted to the ICU: A multicentre retrospective cohort study. Lancet Reg Health Eur. (2021) 11 100243. DOI: 10.1016/j.lanepe.2021.100243 28. Gupta A, Gonzalez-Rojas Y, Juarez E, Casal MC, Moya J, Falci DR. Early treatment for COVID-19 with SARS-CoV-2 neutralizing antibody sotrovimab. N Engl J Med. (2021) 385 1941-50. DOI: 10.1056/NEJMoa2107934 29. Gaudart J, Landier J, Huiart L, Legendre E, Lehot L, Bendiane MK. Factors associated with the spatial heterogeneity of the first wave of COVID-19 in France: A nationwide geo-epidemiological study. Lancet Public Health. (2021) 6 e222-e31. DOI: 10.1016/S2468-2667(21)00006-2 30. Docherty AB, Harrison EM, Green CA, Hardwick HE, Pius R, Norman L. Features of 20 133 UK patients in hospital with COVID-19 using the ISARIC WHO Clinical Characterisation Protocol: Prospective observational cohort study. Br Med J. (2020) 369 m1985. DOI: 10.1136/bmj.m1985 31. Oliveira E, Parikh A, Lopez-Ruiz A, Carrilo M, Goldberg J, Cearras M. ICU outcomes and survival in patients with severe COVID-19 in the largest health care system in central Florida. PLoS ONE. (2021) 16 e0249038. DOI: 10.1371/journal.pone.0249038 32. Richardson S, Hirsch JS, Narasimhan M, Crawford JM, McGinn T, Davidson KW. Presenting characteristics, comorbidities, and outcomes among 5700 patients hospitalized with COVID-19 in the New York City Area. J Am Med Assoc. (2020) 323 2052-9. DOI: 10.1001/jama.2020.6775 33. Karaca-Mandic P, Sen S, Georgiou A, Zhu Y, Basu A. Association of COVID-19-related hospital use and overall COVID-19 mortality in the USA. J Gen Intern Med. (2020) 2020 7. DOI: 10.1007/s11606-020-06084-7 34. Roso-Llorach A, Serra-Picamal X, Cos FX, Pallejà-Millán M, Mateu L, Rosell A. Evolving mortality and clinical outcomes of hospitalized subjects during successive COVID-19 waves in Catalonia, Spain. Glob Epidemiol. (2022) 4 100071. DOI: 10.1016/j.gloepi.2022.100071 35. Hesni E, Sayad B, Khosravi Shadmani F, Najafi F, Khodarahmi R, Rahimi Z. Demographics, clinical characteristics, and outcomes of 27,256 hospitalized COVID-19 patients in Kermanshah Province, Iran: A retrospective one-year cohort study. BMC Infect Dis. (2022) 22 319. DOI: 10.1186/s12879-022-07312-7 36. Chang R, Elhusseiny KM, Yeh YC, Sun WZ. COVID-19 ICU and mechanical ventilation patient characteristics and outcomes—A systematic review and meta-analysis. PLoS ONE. (2021) 16 e0246318. DOI: 10.1371/journal.pone.0246318 37. Bravata DM, Perkins AJ, Myers LJ, Arling G, Zhang Y, Zillich AJ. Association of intensive care unit patient load and demand with mortality rates in US department of veterans affairs hospitals during the COVID-19 pandemic. J Am Med Assoc Netw Open. (2021) 4 e2034266. DOI: 10.1001/jamanetworkopen.2020.34266 38. Tragaki A, Richard JL. First wave of SARS-COV2 in Europe: Study and typology of the 15 worst affected European countries. Popul Space Place. (2021) 2021 e2534. DOI: 10.1002/psp.2534 39. Souris M, Gonzalez JP. COVID-19: Spatial analysis of hospital case-fatality rate in France. PLoS ONE. (2020) 15 e0243606. DOI: 10.1371/journal.pone.0243606 40. Guillon A, Laurent E, Godillon L, Kimmoun A, Grammatico-Guillon L. In-hospital mortality rates of critically ill COVID-19 patients in France: A nationwide cross-sectional study of 45 409 ICU patients. Br J Anaesth. (2021) 127 e180-e2. DOI: 10.1016/j.bja.2021.08.006 41. Garcia-Vidal C, Cózar-Llistó A, Meira F, Dueñas G, Puerta-Alcalde P, Cilloniz C. Trends in mortality of hospitalised COVID-19 patients: A single centre observational cohort study from Spain. Lancet Reg Health Eur. (2021) 3 100041. DOI: 10.1016/j.lanepe.2021.100041 42. Ceccato A, Perez-Arnal R, Motos A, Barbe F, Torres A, Ciberes UC. One-year mortality after ICU admission due to COVID-19 infection. Intensive Care Med. (2022) 48 366-8. DOI: 10.1007/s00134-021-06611-y 43. Dongelmans DA, Termorshuizen F, Brinkman S, Bakhshi-Raiez F, Arbous MS, de Lange DW. Characteristics and outcome of COVID-19 patients admitted to the ICU: A nationwide cohort study on the comparison between the first and the consecutive upsurges of the second wave of the COVID-19 pandemic in the Netherlands. Ann Intensive Care. (2022) 12 5. DOI: 10.1186/s13613-021-00978-3 44. Auld SC, Harrington KRV, Adelman MW, Robichaux CJ, Overton EC, Caridi-Scheible M. Trends in ICU mortality from coronavirus disease 2019: A tale of three surges. Crit Care Med. (2022) 50 245-55. DOI: 10.1097/CCM.0000000000005185 45. Strålin K, Wahlström E, Walther S, Bennet-Bark AM, Heurgren M, Lindén T. Mortality trends among hospitalised COVID-19 patients in Sweden: A nationwide observational cohort study. Lancet Reg Health Eur. (2021) 4 100054. DOI: 10.1016/j.lanepe.2021.100054 46. Ahlstrom B, Frithiof R, Hultstrom M, Larsson IM, Strandberg G, Lipcsey M. The swedish COVID-19 intensive care cohort: Risk factors of ICU admission and ICU mortality. Acta Anaesthesiol Scand. (2021) 65 525-533. DOI: 10.1111/aas.13781 47. Kristinsson B, Kristinsdottir LB, Blondal AT, Thormar KM, Kristjansson M, Karason S. Nationwide incidence and outcomes of patients with coronavirus disease 2019 requiring intensive care in Iceland. Crit Care Med. (2020) 48 e1102-e5. DOI: 10.1097/CCM.0000000000004582 48. Alimohamadi Y, Tola HH, Abbasi-Ghahramanloo A, Janani M, Sepandi M. Case fatality rate of COVID-19: A systematic review and meta-analysis. J Prev Med Hyg. (2021) 62 E311-E20. DOI: 10.15167/2421-4248/jpmh2021.62.2.1627 49. Haase N, Plovsing R, Christensen S, Poulsen LM, Brøchner AC, Rasmussen BS. Characteristics, interventions, and longer term outcomes of COVID-19 ICU patients in Denmark—A nationwide, observational study. Acta Anaesthesiol Scand. (2021) 65 68-75. DOI: 10.1111/aas.13701 50. Moiseev S, Avdeev S, Brovko M, Bulanov N, Tao E, Fomin V. Outcomes of intensive care unit patients with COVID-19: A nationwide analysis in Russia. Anaesthesia. (2021) 3 11-2. DOI: 10.1111/anae.15265 51. Ranzani OT, Bastos LSL, Gelli JGM, Marchesi JF, Baião F, Hamacher S. Characterisation of the first 250,000 hospital admissions for COVID-19 in Brazil: A retrospective analysis of nationwide data. Lancet Respir Med. (2021) 9 407-18. DOI: 10.1016/S2213-2600(20)30560-9 52. Armstrong RA, Kane AD, Kursumovic E, Oglesby FC, Cook TM. Mortality in patients admitted to intensive care with COVID-19: An updated systematic review and meta-analysis of observational studies. Anaesthesia. (2021) 76 537-48. DOI: 10.1111/anae.15425 53. Garg S, Patel K, Pham H, Whitaker M, O'Halloren A, Milucky J. Clinical trends among U.S. adults hospitalized with COVID-19, March to December 2020: A cross-sectional study. Ann Intern Med. (2021) 174 1409-19. DOI: 10.7326/M21-1991 54. Karakike E, Giamarellos-Bourboulis EJ, Kyprianou M, Fleischmann-Struzek C, Pletz MW, Netea MG. Coronavirus disease 2019 as cause of viral sepsis: A systematic review and meta-analysis. Crit Care Med. (2021) 49 2042-57. DOI: 10.1097/CCM.0000000000005195 55. Berlin DA, Gulick RM, Martinez FJ. Severe COVID-19. N Engl J Med. (2020) 383 2451-60. DOI: 10.1056/NEJMcp2009575 56. Basso C. Myocarditis. N Engl J Med. (2022) 387 1488-500. DOI: 10.1056/NEJMra2114478 57. Block JP, Boehmer TK, Forrest CB, Carton TW, Lee GM, Ajani UA. Cardiac complications after SARS-CoV-2 infection and mRNA COVID-19 vaccination—PCORnet, United States, January 2021–January 2022. Morb Mortal Wkly Rep. (2022) 71 517-23. DOI: 10.15585/mmwr.mm7114e1 58. Ammirati E, Lupi L, Palazzini M, Hendren NS, Grodin JL, Cannistraci CV. Prevalence, characteristics, and outcomes of COVID-19-associated acute myocarditis. Circulation. (2022) 145 1123-39. DOI: 10.1161/CIRCULATIONAHA.121.056817 59. Katsoularis I, Fonseca-Rodriguez O, Farrington P, Lindmark K, Fors Connolly AM. Risk of acute myocardial infarction and ischaemic stroke following COVID-19 in Sweden: A self-controlled case series and matched cohort study. Lancet. (2021) 398 599-607. DOI: 10.1016/S0140-6736(21)00896-5 60. Fanaroff AC, Garcia S, Giri J. Myocardial infarction during the COVID-19 pandemic. J Am Med Assoc. (2021) 326 1916-8. DOI: 10.1001/jama.2021.19608 61. Islam MA, Cavestro C, Alam SS, Kundu S, Kamal MA, Reza F. Encephalitis in patients with COVID-19: A systematic evidence-based analysis. Cells. (2022) 11 162575. DOI: 10.3390/cells11162575 62. Chhatriwalla AK, Amin AP, Kennedy KF, House JA, Cohen DJ, Rao SV. Association between bleeding events and in-hospital mortality after percutaneous coronary intervention. J Am Med Assoc. (2013) 309 1022-9. DOI: 10.1001/jama.2013.1556 63. Spyropoulos AC, Raskob GE, Cohen AT, Ageno W, Weitz JI, Spiro TE. Association of bleeding severity with mortality in extended thromboprophylaxis of medically ill patients in the MAGELLAN and MARINER trials. Circulation. (2022) 145 1471-9. DOI: 10.1161/CIRCULATIONAHA.121.057847 64. Schaffer DeRoo S, Pudalov NJ, Fu LY. Planning for a COVID-19 vaccination program. J Am Med Assoc. (2020) 323 2458-9. DOI: 10.1001/jama.2020.8711 65. Zusammen gegen Corona. Internetseite des Bundesministerium für Gesundheit. (2023) 66. Quah P, Li A, Phua J. Mortality rates of patients with COVID-19 in the intensive care unit: A systematic review of the emerging literature. Crit Care. (2020) 24 285. DOI: 10.1186/s13054-020-03006-1 67. Sorake CJ. Double trouble: A pandemic of obesity and COVID-19. Lancet Gastroenterol Hepatol (2021) 6 608. DOI: 10.1016/S2468-1253(21)00190-4 68. Mohammad S, Aziz R, Mahri SA, Malik SS, Haji E, Khan AH. Obesity and COVID-19: What makes obese host so vulnerable?. Immun Ageing. (2021) 18 1. DOI: 10.1186/s12979-020-00212-x 69. Gao M, Piernas C, Astbury NM, Hippisley-Cox J, O'Rahilly S, Aveyard P. Associations between body-mass index and COVID-19 severity in 69 million people in England: A prospective, community-based, cohort study. Lancet Diabetes Endocrinol. (2021) 9 350-9. DOI: 10.1016/S2213-8587(21)00089-9 70. Diedisheim M, Dancoisne E, Gautier JF, Larger E, Cosson E, Fève B. Diabetes increases severe COVID-19 outcomes primarily in younger adults. J Clin Endocrinol Metab. (2021) 106 e3364-e8. DOI: 10.1210/clinem/dgab393 71. Clerkin KJ, Fried JA, Raikhelkar J, Sayer G, Griffin JM, Masoumi A. COVID-19 and cardiovascular disease. Circulation. (2020) 141 1648-55. DOI: 10.1161/CIRCULATIONAHA.120.046941 72. Beyls C, Hermida A, Bohbot Y, Martin N, Viart C, Boisgard S. Automated left atrial strain analysis for predicting atrial fibrillation in severe COVID-19 pneumonia: A prospective study. Ann Intensive Care. (2021) 11 168. DOI: 10.1186/s13613-021-00955-w 73. Rey JR, Caro-Codón J, Rosillo SO, Iniesta ÁM, Castrejón-Castrejón S, Marco-Clement I. Heart failure in COVID-19 patients: Prevalence, incidence and prognostic implications. Eur J Heart Fail. (2020) 22 2205-15. DOI: 10.1002/ejhf.1990 74. Giustino G, Pinney SP, Lala A, Reddy VY, Johnston-Cox HA, Mechanick JI. Coronavirus and cardiovascular disease, myocardial injury, and arrhythmia: JACC focus seminar. J Am Coll Cardiol. (2020) 76 2011-23. DOI: 10.1016/j.jacc.2020.08.059 75. Xie Y, Xu E, Bowe B, Al-Aly Z. Long-term cardiovascular outcomes of COVID-19. Nat Med. (2022) 28 583-90. DOI: 10.1038/s41591-022-01689-3 76. Hobohm L, Sagoschen I, Barco S, Farmakis I, Fedeli U, Koelmel S. COVID-19 infection and its impact on case-fatality in patients with pulmonary embolism. Eur Respir J. (2022) 2022 ehac544. DOI: 10.1093/eurheartj/ehac544.1888 77. Saad M, Kennedy KF, Imran H, Louis DW, Shippey E, Poppas A. Association between COVID-19 diagnosis and in-hospital mortality in patients hospitalized with ST-segment elevation myocardial infarction. J Am Med Assoc. (2021) 326 1940-52. DOI: 10.1001/jama.2021.18890 78. Lee SC, Son KJ, Han CH, Park SC, Jung JY. Impact of COPD on COVID-19 prognosis: A nationwide population-based study in South Korea. Sci Rep. (2021) 11 3735. DOI: 10.1038/s41598-021-83226-9 79. Graziani D, Soriano JB, Rio-Bermudez CD, Morena D, Díaz T, Castillo M. Characteristics and prognosis of COVID-19 in patients with COPD. J Clin Med. (2020) 9 242. DOI: 10.20944/preprints202009.0242.v1 80. Kim L, Garg S, O'Halloran A, Whitaker M, Pham H, Anderson EJ. Risk factors for intensive care unit admission and in-hospital mortality among hospitalized adults identified through the US coronavirus disease 2019 (COVID-19)—Associated hospitalization surveillance network (COVID-NET). Clin Infect Dis. (2021) 72 e206-14. DOI: 10.1093/cid/ciaa1012
--- title: Celastrol-regulated gut microbiota and bile acid metabolism alleviate hepatocellular carcinoma proliferation by regulating the interaction between FXR and RXRα in vivo and in vitro authors: - Dequan Zeng - Lipen Zhang - Qiang Luo journal: Frontiers in Pharmacology year: 2023 pmcid: PMC9975715 doi: 10.3389/fphar.2023.1124240 license: CC BY 4.0 --- # Celastrol-regulated gut microbiota and bile acid metabolism alleviate hepatocellular carcinoma proliferation by regulating the interaction between FXR and RXRα in vivo and in vitro ## Abstract Celastrol, a triterpene derived from Thunder God Vine (Tripterygium wilfordii Hook f; Celastraceae), a traditional Chinese herb, has promising anticancer activity. The present study aimed to elucidate an indirect mechanism of celastrol-mediated alleviation of hepatocellular carcinoma (HCC) via gut microbiota-regulated bile acid metabolism and downstream signaling. Here, we constructed a rat model of orthotopic HCC and performed 16S rDNA sequencing and UPLC-MS analysis. The results showed that celastrol could regulate gut bacteria; suppress the abundance of Bacteroides fragilis; raise the levels of glycoursodeoxycholic acid (GUDCA), a bile acid; and alleviate HCC. We found that GUDCA suppressed cellular proliferation and induced the arrest of mTOR/S6K1 pathway-associated cell cycle G0/G1 phase in HepG2 cells. Further analyses using molecular simulations, Co-IP, and immunofluorescence assays revealed that GUDCA binds to farnesoid X receptor (FXR) and regulates the interaction of FXR with retinoid X receptor a (RXRα). Transfection experiments using the FXR mutant confirmed that FXR is essential for GUCDA-mediated suppression of HCC cellular proliferation. Finally, animal experiments showed that the treatment with the combination of celastrol/GUDCA alleviated the adverse effects of celastrol alone treatment on body weight loss and improved survival in rats with HCC. In conclusion, the findings of this study suggest that celastrol exerts an alleviating effect on HCC, in part via regulation of the B. fragilis-GUDCA-FXR/RXRα-mTOR axis. ## 1 Introduction Celastrol is a natural pentacyclic triterpene derived from the root of Thunder God Vine, a traditional Chinese medicinal plant (Ng et al., 2019; Chen et al., 2018). Celastrol is considered as one of the most potential five natural medicinal compounds (others include artemisinin, triptolide, capsaicin, and curcumin) (Corson and Crews, 2007). The extracts of Thunder God Vine have been shown to exert hepato-protective effects and have been used to alleviate liver injury (Qi et al., 2020; Yan et al., 2021), hepatic inflammation (Luo et al., 2017), and hepatocellular carcinoma (HCC) (Chang et al., 2016). The direct effect on the signaling process in hepatocytes is considered the main molecular mechanism of celastrol-mediated alleviation of hepatic diseases. Our previous study revealed that celastrol could repair acute liver injury by directly targeting the nuclear receptor Nur77 in the liver tissue and clearing inflamed mitochondria (Hu et al., 2017). Furthermore, our recent finding also showed that celastrol regulated fecal morphology and changed the gut microbiota community structure during alleviating HCC proliferation in an orthotopic HCC rat model. Emerging evidence shows that gut microbiota regulates the progress of carcinoma, wherein celastrol has been shown to regulate gut microbiota to maintain the immune balance of ulcerative colitis and inhibit lipid absorption in obesity. These studies imply that the gut microbiota may, in part, contribute to the HCC-alleviating effect of celastrol. As an important micro-ecosystem of the human body, gut microbiota affects liver bile acid metabolism via enterohepatic circulation (Chávez-Talavera et al., 2017; Staley et al., 2017) and regulates the development of hepatic diseases such as hepatitis (Li et al., 2020), cirrhosis (Ridlon et al., 2015; Kakiyama et al., 2013), and HCC (Yamada et al., 2018). Gut microbiota usually adheres to the intestinal mucosa or excretes with feces, making it difficult to enter the liver. However, the occurrence of unbalanced gut homeostasis with changes in microbiota community structure and an increase in the abundance of pathogenic bacteria damage the intestinal mucosa, leads to the migration of the microorganisms to the liver, affects the metabolism of bile acids, and eventually leads to the development of hepatic diseases, such as HCC. Recently, several studies have revealed that gut microbiota and liver bile acid metabolism regulation are involved in HCC progression (Yamada et al., 2018; Wu et al., 2021; Chiang and JessicaFerrell, 2018). Additionally, bile acid is an important endogenous active molecule regulating liver function and exerts its function via binding to the farnesoid X receptor (FXR) and regulating the heterodimer of FXR (Chiang and JessicaFerrell, 2020; Radun and Trauner, 2021; Molinaro and Marschall, 2022). Therefore, we speculated that exploring the mechanism underlying the regulation of gut microbiota associated with liver bile metabolism, regulation of binding to FXR, the heterodimer of FXR, and HCC proliferation could contribute to the understanding of the alleviating effect of celastrol against HCC. To test this hypothesis, we constructed a rat model of orthotopic HCC induced by diethyl nitrosamine (DEN) and evaluated the effect of celastrol on HCC in this study. We performed 16S rDNA sequencing of rat feces to identify potential active bacterial species and UPLC-MS analysis of liver tissue to identify active bile acids. The effect of active bile acid on the proliferation of hepatoma carcinoma HepG2 cells was confirmed using MTT assay and clone formation. We also evaluated the mTOR/S6K1 proliferation pathway and cell cycle distribution using western blotting and flow cytometry, respectively. Molecular simulations and SPR assays were performed to evaluate the binding of active bile acid to FXR, and a reporter assay was employed to study the effect of active bile acid on the transcriptional activity of FXR. Co-IP and immunofluorescence assays were performed to elucidate the effect of active bile acid on the heterodimer of FXR with retinoid X receptor a (RXRα), and transfection experiments with the FXR mutant were performed to elucidate the role of FXR in active bile acid-regulating cellular proliferation and affecting mTOR/S6K1 pathway-related cell cycle distribution. Finally, a rat model with orthotopic HCC was used to evaluate the synergistic effect of the active bile acid/celastrol combination and to analyze the effect difference between the active bile acid/celastrol combination and celastrol alone. ## 2.1 Materials The materials used in this study are listed in Table 1. TABLE 1 \| Materials \| Source \| Identifier \| \| --- \| --- \| --- \| \| Antibodies \| Antibodies \| Antibodies \| \| Anti-mTOR \| Abcam, Cambridge, UK \| Ab32028 \| \| Anti-p-mTOR \| Abcam, Cambridge, UK \| Ab109268 \| \| Anti-S6K1 \| Abcam, Cambridge, UK \| Ab32529 \| \| Anti-p-S6K1 \| Abcam, Cambridge, UK \| Ab59208 \| \| Anti-p-Rb \| Abcam, Cambridge, UK \| Ab184796 \| \| Anti-myc \| Abcam, Cambridge, UK \| ab9106 \| \| Anti-flag \| Abcam, Cambridge, UK \| Ab205606 \| \| Anti-FXR \| Abcam, Cambridge, UK \| Ab129089 \| \| Anti-cyclin D1 \| CST, Boston, United States \| 55506 \| \| Anti-CDK4 \| CST, Boston, United States \| 3136 \| \| Anti-CDK6 \| Abcam, Cambridge, UK \| AB151247 \| \| Anti-cdc25A \| Abcam, Cambridge, UK \| Ab79252 \| \| Anti-p21 \| Abcam, Cambridge, UK \| Ab109520 \| \| Anti-ki67 \| Abcam, Cambridge, UK \| Ab16667 \| \| Anti-β-actin \| Abcam, Cambridge, UK \| Ab8226 \| \| Goat anti-mouse lgG Secondary antibody, HRP conjugate \| Abcam, Cambridge, UK \| Ab205719 \| \| Goat anti-rabbit lgG Secondary antibody, HRP conjugate \| Abcam, Cambridge, UK \| Ab205718 \| \| Chemical and Reagents \| Chemical and Reagents \| Chemical and Reagents \| \| DEN (Diethyl nitrosamine) \| Sigma, Saint Louis, UK \| N0756 \| \| ALT assay kit \| Nanjingjiancheng, Nanjing, China \| C009-2-1 \| \| AST assay kit \| Nanjingjiancheng, Nanjing, China \| C010-2-1 \| \| MTT (methyl thiazolyl tetrazolium) \| Abcam, Cambridge, UK \| Ab211091 \| \| Fecal DNA Isolation Kit \| Vazyme, Nanjing, China \| DC103 \| \| SYBR green dye \| Vazyme, Nanjing, China \| Q131 \| \| Giemsa dye \| Solarbio, Beijing, China \| G8220 \| \| PI (Propidium Iodide) \| Solarbio, Beijing, China \| P8080 \| \| ECL detecting kit \| Thermo, Waltham, United States \| 32109 \| \| BCA protein assay kit \| Solarbio, Beijing, China \| PC0020 \| \| DAPI dye \| Solarbio, Beijing, China \| C0060 \| \| Protein Ladder \| Thermo, Waltham, United States \| 26616 \| \| Protein G Agarose \| Thermo, Waltham, United States \| 15920010 \| \| DMEM culture medium \| Gibco, New York, United States \| 11965092 \| \| Fetal bovine serum \| Gibco, New York, United States \| 12484028 \| \| Eosin dye \| Solarbio, Beijing, China \| G1100 \| \| Hematoxylin dye \| Solarbio, Beijing, China \| H8070 \| \| Transfection reagents \| Thermo, Waltham, United States \| 11668019 \| \| Dual-Luciferase reporter assay system \| Promega, Madison, United States \| E1910 \| \| CM5 chip \| GE Healthcare Pittsburgh, United States \| BR100530 \| \| Experimental Models \| Experimental Models \| Experimental Models \| \| HEK 293T cell lines \| Cell bank, Shanghai Cell Biology Institute, Shanghai, China \| \| \| HepG2 cell lines \| Cell bank, Shanghai Cell Biology Institute, Shanghai, China \| \| \| Rat \| Xiamen University Laboratory Animal Center, Xiamen, China \| \| \| Software \| Software \| Software \| \| Autodock \| MGL \| AutoDock 4 \| \| PyMol \| DeLano Scientific LLC \| PyMol V2.2.0 \| \| ImageJ \| NIH, United States \| ImageJ V2.3.0 \| \| Prism GraphPad \| Insightful Science, United States \| GraphPad5 \| \| Instruments \| Instruments \| Instruments \| \| Paraffin Embedding Center \| Leica, Germany \| EG1150C \| \| Slicer \| Leica, Germany \| RM2235 \| \| Microplate reader \| Cmax Plus, United States \| Molecular Devices Cmax Plus \| \| Protein Detecting System \| Bio-Rad, United States \| 1645050 \| \| Multimode reader \| BioTek, United States \| Cytation 5 \| \| Cell incubator \| ESCO, Singapore \| CLM-170B-8-NF \| \| Microscope \| Motic Electric, China \| AE2000 \| \| Laser scanning confocal microscope \| Carl Zeiss, Germany \| LSM880 \| \| Flow cytometry \| Beckman, United States \| CytoFLEX \| \| Centrifuge \| Scilogex, China \| D3024R \| ## 2.2 Animal experiments and sampling Sprague-Dawley rats (6–8 weeks old) were purchased from Xiamen University Laboratory Animal Center (Xiamen, China). After acclimation for 1 week in a controlled atmosphere of 12 h light/dark cycle at 22°C, rats were randomly divided into two groups: normal ($$n = 10$$) and model ($$n = 30$$). Rats in the model group were treated by gavage with DEN at a dose of 10 mg/kg body weight once a week for 3 months, and those in the normal group were treated with normal saline as a control. The DEN solution was prepared by dissolving 1 g DEN in 100 mL normal saline to a final concentration of 10 mg/mL. The rats in the model group were randomly divided into three subgroups: model, celastrol, and celastrol/GUDCA ($$n = 10$$/group). Rats in normal and model groups were gavaged with normal saline with $1\%$ DMSO and $5\%$ tween-80 six times per week for 10 weeks, wherein the rats in celastrol and celastrol/GUDCA groups were treated (via gavage) with celastrol at 0.5 mg/kg body weight and celastrol/GUDCA at 0.5 mg/kg +20 mg/kg body weight, respectively, 6 times a week for 10 weeks. The celastrol solution was prepared by dissolving 25 mg celastrol in 1 mL DMSO and following diluting with 99 mL normal saline to a final concentration of 0.25 mg/mL. The body weights of the rats were measured once a week, and the survival number was recorded in real-time. At the end of the 10th week of treatment, blood samples were collected from the eyes to estimate serum ALT and AST levels. Fecal samples were collected for extraction of DNA for gut microbiota community analysis. Subsequently, the rats were sacrificed and liver samples were collected for pathological examination and protein expression analysis. All animal experiments were approved by the Animal Care and Use Committee of Xiamen University. ## 2.3 Tissue processing and histological analysis The liver tissues of the rats were collected and cut into approximate 8.0 mm × 8.0 mm squares, put into $4\%$ paraformaldehyde phosphate buffer solution, and fixed overnight at 4°C. The fixed tissues were then embedded in paraffin and sliced into 4 μm-thick sections. For hematoxylin and eosin (H&E) staining, the sections were deparaffinized using a gradient xylene-alcohol-distilled water solution, stained with H&E stain, and dehydrated using the gradient alcohol-xylene solution. Afterward, the tissue sections were mounted on slides, and images were acquired using a microscope, the images were then analyzed using ImageJ software. To assess the expression of ki67, immunohistochemical analyses were performed. The sections were deparaffinized using the gradient xylene-alcohol-distilled water solution, which were then retrieved using citrate solution (10 mM sodium citrate, pH 6.0) and blocked using $10\%$ goat serum. Afterward, the sections were incubated with a primary antibody of ki67, stained with DAB and hematoxylin, and dehydrated with the gradient alcohol-xylene solution. The sections were then mounted on slides for imaging using a fluorescence microscope and analyzed using ImageJ software. ## 2.4 Estimation of the levels of serum biochemical indicators The blood samples were collected from the eyes of rats and allowed to stand at room temperature (20°C–25°C) for 30 min, centrifuged at 3,000 rpm (centrifugal radius: 7 cm) for 20 min, and the supernatants were transferred to fresh 1.5 mL tubes. Then the levels of glutamic-oxaloacetic transaminase (AST) and glutamic-pyruvic transaminase (ALT) in the supernatants were detected using micro-plate methods according to the serum biochemical indicator assay kits (Nanjingjiancheng, China) following the manufacturer’s protocol. ## 2.5 Western blotting For tissue samples, 100 mg liver was cut into pieces on ice, ground into powder in liquid nitrogen, and lysed with 500 μL RIPA lysis buffer containing protease and phosphatase inhibitors. For cell samples, approximately 1 × 106 HepG2 cells per well of 6-wells plate were collected and added 100 μL RIPA lysis buffer containing protease and phosphatase inhibitors. All samples were lysed for 30 min on ice; afterward, the lysis solution was centrifugated at 12,000 rpm and 4°C for 10 min. The supernatants were collected to estimate total protein concentration using a BCA protein assay kit. The samples were diluted to a final concentration of 2 μg/μL and boiled at 105°C for protein denaturation. The denatured protein samples (20 μL) were loaded onto an $8\%$–$12\%$ SDS-PAGE gel for electrophoresis and transferred to a PVDF membrane. The membrane was blocked with $5\%$ defatted milk powder in TBST solution containing tween-20. Afterward, the membrane was incubated with the primary antibodies against mTOR, p-mTOR, S6K1, p-S6K1, p-Rb, myc, flag, FXR, RXRα, cyclin D1, CDK4, CDK6, cdc25A, p21, ki67, and β-actin, followed by incubation with the corresponding secondary antibodies. All primary antibodies were diluted to 1:1,000 or 1:2,000, and secondary antibodies were diluted to 1:10,000. After incubation, the PVDF membranes were visualized using an ECL solution and imaged on the films via exposure. The images in the film were quantified for the gray value using ImageJ. ## 2.6 16S rDNA sequencing The fecal samples were collected and transferred to Xiamen Treatgut Bio-Technology Co., Ltd. for 16S rDNA sequencing and the analysis of the gut microbiota community. In brief, the total DNA of fecal samples was extracted using a fecal DNA isolation kit. The concentration and purity of DNA were quantified using a Multiskan GO microplate reader (Thermo, United States), and the integrity of DNA was detected by agarose gel electrophoresis. After quality testing, the library was constructed using primers 341F (sequence: CCTACGGGNGGCWGCAG) and 806R (sequence: GGACTACHVGGGTATCTA AT) to amplify the 16S v3–v4 regions. The concentration of the library was quantified using Qubit 3.0, the integrity of the library was detected using an Agilent 2,100 bioanalyzer (Agilent, United States), and qPCR was performed to confirm and obtain accurate quantification. After the library was quantified, different libraries were pooled to flow cells and sequenced using a high-throughput sequencer (Illumina). Subsequently, various bioinformatic analyses were performed, including OTU cluster, α-diversity, PCA, cluster heatmap, species difference, and related metabolism. ## 2.7 Estimation of bile acid contents Liver tissue samples (200 μg) were collected, cut into pieces on ice, ground into powder in liquid nitrogen, and extracted with 2 mL HPLC-grade chloroform/methyl alcohol solution. The extracted solution was centrifuged at 12,000 rpm at 4°C, and the supernatant was harvested and dried under vacuum. The dry bile acid was dissolved in HPLC-grade methyl alcohol, and chlorpropamide was added as an internal standard for quantifying bile acid content. The bile acid samples containing the internal standard were used to assess data from 50 to 800 m/z using an Easy-nLC1000 UPLC-Q Exactive MS system (Thermo, United States). The standards of the bile acids tested in the present study were used to identify the UPLC-MS data. ## 2.8 Cells culture Human embryonic kidney HEK-293T and human hepatoma HepG2 cells were maintained in DMEM supplemented with $10\%$ fetal bovine serum and $1\%$ penicillin-streptomycin solution. HEK-293T cells were used to detect the effects of GUDCA on FXR transcriptional activity using a reporter assay. Before testing, cells were seeded into 48-well plates, transfected with two plasmids, pBind FXR LBD and pGL5 Luc, and treated with CDCA and GUDCA. HepG2 cells were used to detect the effect of GUDCA on cellular proliferation and the interaction between FXR and RXRα. Before the tests, the cells were seeded into the responding multi-well plate, such as 96-well plates for MTT assay, 6-well plates for western blotting, 100 mm plates for Co-IP assay, and 24-well plates for immunofluorescence assay, and then treated with 50 or 100 μM GUDCA for the corresponding times. ## 2.9 MTT assay HepG2 cells treated with 50 or 100 μM GUDCA for 48 h were replaced with fresh DMEM medium containing 0.5 mg/mL MTT and incubated in a CO2 incubator for 4 h. GUDCA stock was prepared by dissolving 45 mg GUDCA in 1 mL DMSO and MTT stock was prepared by dissolving 100 mg MTT in 20 mL PBS. The stock solutions were filtrated through a 0.22 μm microporous filter. After incubation, the cellular supernatant was removed and replaced with 100 μL DMSO to dissolve the precipitate. The optical density (OD) of the dissolving solution was read at a wavelength of 492 nm using a microplate reader (Thermo, United States). ## 2.10 Cell clone formation HepG2 cells treated with 50 or 100 μM GUDCA for 10 days were washed twice with PBS and fixed with 3 mL methyl alcohol for 15 min. Then, the methyl alcohol was removed, and the cells were washed twice with PBS and stained with 2 mL Giemsa dye for 30 min. After staining, clones were washed with distilled water and imaged using a camera (Canon, Japan). ## 2.11 Flow cytometry to detect cell cycle distribution HepG2 cells treated with 50 or 100 μM GUDCA for 24 h were washed twice with PBS and digested with $0.25\%$ trypsin to the single cells. The cells were harvested via centrifugation at 1,500 rpm (centrifugal radius: 7 cm) for 10 min, resuspended in pre-cooled $70\%$ alcohol, fixed overnight at −20°C, and stained with a solution containing 50 μg/mL PI and 1 mg/mL RNase A at room temperature in the dark for 30 min. The stained cells were subjected to flow cytometry (Beckman, United States), and the cell cycle distribution was determined. ## 2.12 Molecular simulation Molecular simulations were performed as described previously (Hu et al., 2017; Luo et al., 2016). Briefly, docking of GUDCA to FXR (PDB ID:3DCT) was performed using AutoDock software, and molecular visualization was displayed using PyMOL software. ## 2.13 Transfection experiment For the reporter assay in HEK-293T cells, plasmids containing 100 ng pBind FXR LBD and 200 ng pGL5 luc were added to 50 μL Opti-MEM medium as solution A, and 1.5 μL lipofectamine 3,000 reagent was added to 50 μL Opti-MEM medium as solution B. After incubation of solutions A and B for 5 min, the two solutions were mixed and incubated for another 20 min, and then the mixture was gently added to the cells with 150 μL DMEM medium. The cells were maintained for 12 h, and the medium was replaced with normal DMEM. For HepG2 cells, the plasmid containing 1 μg flag-FXR and RXRα was added to 500 μL Opti-MEM medium as solution A, and 15 μL lipofectamine 3,000 reagent was added to 500 μL Opti-MEM medium as solution B. After incubation of solutions A and B for 5 min, the two solutions were mixed and incubated for another 20 min, and then the mixture was gently added to the cells with 1.50 mL of DMEM. The cells were maintained for 12 h, and the medium was replaced with normal DMEM. ## 2.14 Dual-luciferase reporter assay A dual-luciferase reporter assay was performed as described in our previous study (Hu et al., 2017; Luo et al., 2016). Briefly, after being transfected with pBind FXR LBD and pGL5 luc and treated with CDCA and/or GUDCA, HEK-293T cells were lysed with 50 μL 1× passive lysis buffer. Then LAR II and Stop/Glo reagent were added respectively according to the manufacturer’s instructions. Luciferase fluorescence values were read using a multimode reader (Thermo, United States), and relative activity was normalized to Renilla luciferase fluorescence. ## 2.15 SPR assay His-FXR LBD proteins (30–50 μg) or its mutants were coupled to CM5 chip, and different doses (2, 5, 10, 20, and 50 μM) of GUDCA were injected into the flow cells of the samples. The chip was submitted to a Biacore T200 system (GE Healthcare, United States), and the curve of binding-dissociation between GUDCA and FXR LBD was drawn. ## 2.16 Co-immunoprecipitation Co-immunoprecipitation was performed as previously described (Hu et al., 2017). HepG2 cells transfected with flag-FXR/myc-RXRα plasmids and treated with GUDCA in a 100 mm culture plate were lysed with a gentle lysis buffer (500 μL) on ice for 30 min. The lysate was divided into two parts:50 μL for the input and 450 μL for IP tests. The input solution was boiled for denaturation, while IP solution was added with 1 μg anti-flag antibody, incubated at 4°C for 2 h, and then precipitated with protein A/G beads. The precipitated beads were washed and boiled for denaturation. Details of the procedure are described in Section 2.5. ## 2.17 Immunofluorescence Immunofluorescence analysis was performed as previously described (Hu et al., 2017). In brief, seeded HepG2 cells transfected with flag-FXR/myc-RXRα plasmids and treated with GUDCA were permeabilized with $0.1\%$ Triton X-100, blocked with $1\%$ BSA and incubated with primary antibodies against FLAG and myc followed by secondary antibodies of anti-goat IgG conjugated with Cy3 and Cy5. The cells were then stained with DAPI and imaged using a laser scanning confocal microscope (Carl Zeiss, Germany). All primary antibodies were diluted at a ratio of 1:50, and the secondary antibody was diluted at 1:200. ## 2.18 Statistical analysis Experimental data are shown as the mean ± SEM using Prism GraphPad version 5.0. A two-tailed unpaired Student’s t-test and one-way ANOVA with Dunnett’s Multiple Comparison Test was used to analyze the differences between groups in the present study. The gray value of the bands obtained by western blotting was analyzed using ImageJ version 2.3. p-values <0.05 were considered statistically significant at $p \leq 0.01$ as highly significant and $p \leq 0.001$ as extremely significant. ## 3.1 Celastrol alleviates HCC in an orthotopic HCC rat model We constructed a rat model with orthotopic HCC by treating the rats with DEN and administrated the model rats with celastrol or normal saline in gavage for 10 weeks. The survival curve analysis revealed a decreased survival in the model group than in the celastrol (Figure 1A). The death of rats in the model group occurred on the third week of administration ($$n = 1$$), which consistently increased until the end of the 10th week and reached 4. In contrast, in the celastrol group, the decrease in survival started in the eighth week ($$n = 1$$) which increased at the end of the 10th week ($$n = 2$$). During treatment, the body weight of rats in the model group was reduced compared to that in the normal group, and celastrol did not relieve the reduction (Supplementary Figure S1A, B). Biochemical analyses of the blood samples at the end of the 10th week of administration showed that celastrol suppressed the DEN-induced increase in serum ALT and AST levels (Figures 1B,C). The number of nodules in the liver tissue was also suppressed in the celastrol group (7.2) compared to that in the model group (20.2) (Figure 1D). Pathology images showed that the liver tissue in the model group displayed inflammatory infiltration, which was reduced in the celastrol group (Figure 1E). Furthermore, the DEN-induced increased expression of ki67, a classical proliferation marker indicating carcinoma, was suppressed by celastrol (Figure 1F). Western blot analysis revealed higher mTOR phosphorylation in the liver tissue of rats in the model group than that in the normal group, wherein this increase in mTOR phosphorylation was suppressed in the celastrol group (Supplementary Figure S1C). FIGURE 1: Celastrol alleviates HCC in an orthotopic rat model. (A) The survival number of rats. Serum levels of (B) ALT and (C) AST. (D) Nodule numbers of liver tissue, (E) H&E staining for pathology analysis. (F) IHC for probing ki67 expression. *p < 0.001 vs. normal group; ##p < 0.01, ###p < 0.001 vs. model group. ## 3.2 Celastrol regulates gut microbiota and bile acid metabolism in rats with HCC The rats administered celastrol had loose bowels. To evaluate the effect of celastrol on gut microbiota, we performed 16S rDNA sequencing and qPCR experiments. A Venn diagram of OTU distribution showed 273 overlapping OTUs, 69 individual OTUs in the gut microbiota of the normal group, and 26 individual OTUs in the model group (Supplementary Figure S2A). The PCA scatter plot revealed significant differences between the gut microbiota composition of the normal and model groups (Supplementary Figure S2B). Analysis of the bacterial community structure of fecal samples from the normal and model groups also revealed differences (Supplementary Figure S2C and Supplementary Figures S3A, S3B). Cluster analysis of bacterial community structure showed that the abundance of different Bacteroides species was highly different in the gut of the normal and model group rats (Figure 2A). To further identify the different strains, we detected the richness of four sub-strains of Bacteroides, Bacteroides fragilis, Bacteroides finegoldii, Bacteroides massiliensis and Bacteroides uniformis. As shown in Figure 2B, DEN increased the richness of B. Fragilis, B. Massiliensis and B. Uniformis. However, celastrol significantly suppressed the DEN-induced increase in B. fragilis richness, demonstrating that B. fragilis could be essential for celastrol-mediated alleviation of HCC (Figure 2B). Subsequent analysis of the B. fragilis-associated metabolism pathway showed that primary bile acid biosynthesis was one of the most promising pathways for HCC progression (Figure 2C). In addition, data from several liver bile acids showed that celastrol increased the levels of GCDCA, UDCA, TUDCA, and GUDCA, while the increase in GUDCA was the most evident (Figure 2D). The structure of GUDCA is shown in Supplementary Figure S3C. FIGURE 2: Celastrol regulates gut microbiota and bile acid metabolism in rats with HCC. (A) Cluster diagram of bacterial structure. (B) Different species abundance of Bacteroides, p < 0.05, *p < 0.001 vs. normal group; ### p < 0.001 vs. model group. (C) Metabolism pathway analysis. (D) Bile acids level in liver, p < 0.001 vs. model group. ## 3.3 GUDCA suppresses the cellular proliferation in HepG2 cells To evaluate the effect of GUDCA on the proliferation of hepatoma carcinoma cells, MTT assay, clone formation test, western blotting assay for probing the mTOR/S6K1 pathway, and flow cytometry assay for detecting cell cycle distribution were performed in HepG2 cells. The MTT assay showed that 50 and 100 μM GUDCA inhibited the proliferation of HepG2 cells in a dose-dependent manner, and the proliferation inhibition ratio at 100 μM was statistically different (Figure 3A). The clone formation test showed a similar effect of GUDCA with the MTT assay, in which GUDCA inhibited the cellular clone number in a dose-dependent manner in HepG2 cells (Figure 3B). Western blot analysis showed that GUDCA suppressed the phosphorylation of mTOR, S6K1, and Rb (Figure 3C), demonstrating that GUDCA regulates the mTOR/S6K1/Rb pathway, which regulates the cell cycle distribution and affects cellular proliferation. Furthermore, flow cytometry data revealed that GUDCA arrested the G1 phase of the cell cycle in HepG2 cells (Figure 3D), wherein expression of G0/G1 phase-related proteins, including cyclinD1, CDK4, CDK6, and cdc25A was detected using western blotting, indicating that these proteins are also regulated by GUDCA (Supplementary Figure S5). FIGURE 3: GUDCA suppresses the cellular proliferation in HepG2 cells. (A) MTT assay for cellular proliferation. (B) Clone formation experiment to assess the clone number. (C) Western blot for probing the expression of mTOR, p-mTOR, S6K1, p-S6K1, and p-Rb. (D) Flow cytometry for detecting cell cycle distribution. p < 0.05 and *p < 0.01 vs. DMSO group. GU means Glycoursodeoxycholic acid (GUDCA). ## 3.4 GUDCA is identified as an antagonist of FXR Bile acid is an important bioactive metabolite that plays a crucial role in the physiological and pathological processes by interacting with nuclear receptors containing FXR. Molecular simulations and reporter assays were performed to evaluate the interaction between GUDCA and FXR. Molecular docking images showed that GUDCA bonded to FXR in the previously reported cave (3DCT), which was the location of GW4064 binding (Figure 4A; Supplementary Figure S6A), a classic FXR ligand, and displayed an extremely similar binding conformation to GW4064 (Supplementary Figure S6B). The lowest binding energy of GUDCA with FXR reached −11.59 kJ/mol (Supplementary Figures S6C–E), revealing the well-binding potential of GUDCA with FXR. The key binding residue analysis showed that FXR M265, R331, H447, and W469 residues had H-bond interactions with GUDCA (Figure 4B), and FXR L348 and F461 residues had hydrophobic interactions (Figure 4C). Figure 4D shows the amino acid sequence of FXR LBD. Furthermore, the interaction between GUDCA and FXR was confirmed using SPR and reporter assays. The binding-dissociation curve obtained by SPR revealed that GUDCA binds to the FXR LBD in a dose-dependent manner (Supplementary Figure S7). A reporter assay revealed that GUDCA suppressed the CDCA-induced transcriptional activity of FXR (Figure 4E). Furthermore, FXR mutant data showed that FXR R331 affected the effect of GUDCA on FXR transcriptional activity (Figure 4F). FIGURE 4: GUDCA is an antagonist of FXR. (A) The docked conformation of GUDCA with FXR (3DCT); (B) H-bond interaction of GUDCA with FXR residues; (C) Hydrophobic interaction of GUDCA with FXR residues; (D) Amino acid sequence of FXR LBD; (E) Reporter assay for detecting the effect of GUDCA on the transcriptional activity of FXR, p < 0.001 vs. normal group; ## p < 0.01, ### p < 0.001 vs. CDCA group; (F) Reporter assay for detecting the effect of GUDCA on the transcriptional activity of FXR mutant, p < 0.01 vs. CDCA group; ns, not significant. GU means Glycoursodeoxycholic acid (GUDCA).* ## 3.5 GUDCA affects FXR interaction with RXRα in HepG2 cells RXRα is an important nuclear receptor that participates in several bio-behaviors, including proliferation via its interaction with itself (forming homodimers) or with many other nuclear receptors such as FXR (forming heterodimers). To confirm whether GUDCA affects the interaction of FXR with RXRα, we performed co-immunoprecipitation and immunofluorescence analyses in HepG2 cells. Co-immunoprecipitation data showed that GUDCA suppressed the interaction of FXR with RXRα in the transfection conditions of FLAG-FXR, FLAG-FXR L348A, and FLAG-FXR M265; however, the suppressive effect of GUDCA on the interaction of FXR with RXRα weakened in the transfection conditions of flag-FXR R331A (Figure 5A). Immunofluorescence analysis revealed that GUDCA significantly inhibited the colocation of transiently transfected FXR and RXRα, wherein the inhibitory effect of GUDCA on the colocation of transiently transfected FXR R331A and RXRα was slight (Figure 5B). FIGURE 5: GUDCA affects FXR interaction with RXRα in HepG2 cells. (A) Co-IP for detecting the interaction of FXR with RXRα; (B) Immunofluorescence assay for detecting the colocation of FXR with RXRα. ## 3.6 FXR is essential for GUDCA-mediated mTOR pathway-associated proliferation in HepG2 cells To further elucidate the mechanism of GUDCA-alleviating hepatoma carcinoma cell proliferation and clarify the role of FXR in the effect of GUDCA, the plasmids of wild-type and mutant FXR were transfected into HepG2 cells, and the effect of GUDCA on proliferation was evaluated using MTT assay, western blotting, and flow cytometry. Proliferation inhibitory data demonstrated a proliferation inhibitory effect in HepG2 cells subjected to transient transfection of FXR siRNA/flag-FXR in response to GUDCA, which was similar to that in wild-type HepG2 cells; however, HepG2 cells subjected to transient transfection with FXR siRNA/flag-FXR R331A lost their proliferation inhibitory effect (Figure 6A). Consistently, mTOR phosphorylation was inhibited, and cell cycle G0/G1 phase was arrested in HepG2 cells subjected to FXR siRNA/flag-FXR transfection in response to GUDCA; however, contrasting results were observed in HepG2 cells subjected to transient transfection of FXR siRNA/flag-FXR R331A (Figures 6B,C and Supplementary Figure S8). FIGURE 6: FXR is essential for GUDCA-alleviating mTOR pathway-associated proliferation in HepG2 cells. (A) MTT assay for detecting cellular proliferation, *p < 0.001 compared to DMSO group; (B) Western blot for probing the expressions of p-mTOR, mTOR, and Flag; (C) Flow cytometry for detecting cell cycle distribution. ## 3.7 GUDCA improves the alleviating effect of celastrol on HCC in a rat model To explore the role of GUDCA in the celastrol-mediated alleviation of HCC, we constructed a rat model of HCC using DEN and evaluated the combined effect of celastrol and GUDCA. The rats treated with the combination of celastrol and GUDCA displayed significant inhibitory effects on serum levels of ALT and AST, liver nodule number, mTOR phosphorylation, and inflammation infiltration, which was similar to the effect of celastrol alone (Figures 7A–E). Moreover, the combination of celastrol and GUDCA alleviated body weight loss to a certain extent compared to celastrol alone during treatment (Figure 7F). The combination of celastrol and GUDCA improved the survival rate of rats; only one death was recorded at the end of the 10th week in the combined treatment group, wherein three deaths (one at the eighth week and two at the end of the 10th week) by the end of the 10th week were recorded in the only celastrol treatment group (Figure 7G). FIGURE 7: GUDCA improves the alleviating effect of celastrol on HCC in an orthotopic rat model. Serum levels of (A) ALT and (B) AST; (C) Nodule numbers of liver tissue; (D) Western blot for probing the expressions of p-mTOR, mTOR, and FXR; (E) H&E staining for pathology analysis; (F) The curve of body weight; (G) Survival number of rats. ## p < 0.01 and ### p < 0.001 compared to model group. GU means Glycoursodeoxycholic acid (GUDCA). ## 4 Discussion For decades, celastrol has been used in folk medicine and clinical practice to treat cancer, including HCC. Because of its broad beneficial biological effects, such as inflammation suppression, cellular proliferation inhibition (Fang and Chang, 2021; Chen et al., 2022), the lipid absorption alleviation (Zhang et al., 2019; Hu et al., 2021), celastrol has garnered the research focuses of natural medicinal agents. Studies unveiling the mechanism of the hepato-protective effect of celastrol have demonstrated that its functions are mainly regulated via caspase-dependent apoptosis (Shen et al., 2021) and mTOR/AKT-associated proliferation (Li et al., 2018). Our previous study revealed that celastrol directly binds to the Nur77 nuclear receptor, clears inflamed mitochondria, and alleviates liver injury (Hu et al., 2017). All these mechanisms of action of celastrol on hepatic diseases focus on its direct effect on the signaling process in hepatocytes and liver tissues. Furthermore, the gut microbiota has been reported to play an important role in liver function and hepatic diseases. The present study demonstrates that celastrol regulates the bacterial community structure of the gut and suppresses the abundance of B. fragilis, which may have an important role in celastrol-mediated HCC alleviation. This result extends the direct mechanism of HCC alleviating effects of celastrol and provides a new direction to explore the mechanism of its other functions. Bacteroides fragilis is an important member of Bacteroides and plays an essential role in the pathological progression of cancers. It has also been shown that B. fragilis could cause intestinal inflammation and tissue injury, eventually leading to colorectal cancer (Cheng et al., 2020; Haghi et al., 2019). In addition, B. fragilis has also been shown to be associated with gastric carcinoma, lung abscess, renal diseases, etc. In this study, we found that celastrol that alleviated HCC suppressed the abundance of B. fragilis in rats. Several studies have revealed that B. fragilis participates in liver bile acid metabolism and regulates hepatic function. Concordantly, the present study showed that celastrol is involved in maintaining the levels of liver bile acids; it significantly increased the level of GUDCA while suppressing the abundance of B. fragilis. These results imply that B. fragilis–GUDCA may be one of the mechanisms underlying celastrol-mediated alleviation of HCC. However, the present study could not explain the inter-relationships of B. fragilis, bile acids metabolism, and HCC alleviating effects of celastrol; therefore, further studies are required to gain further insights. In the future, we propose to evaluate the detailed relationship between therapeutic effects of celastrol against HCC and the alterations in bile acid levels and its upstream pathways via employing B. fragilis to treat rat with HCC. Bile acids is an important component of bile, mainly in the enterohepatic circulation system. In response to the gut microbiota, the contents and types of bile acids are changed in the intestine, absorbed into the liver via enterohepatic circulation, and exert a regulatory effect on hepato-metabolism. Bile acid metabolism plays an important role in the protection of hepatotoxicity. FXR, a member of the nuclear receptor family, is highly expressed in the liver and intestinal tissues to maintain metabolic homeostasis. Several studies have shown that FXR interacts with bile acids such as CDCA, regulates the heterodimerization of FXR with other nuclear receptors, and exerts its multifunction, including lipid metabolism (Zhang et al., 2019; Ramírez-Pérez et al., 2017), hepato-inflammation (Han et al., 2018), and cellular proliferation. This study revealed that the bile acid GUDCA could bind to FXR and suppress its transcriptional activity. Furthermore, FXR was confirmed to be essential for the effects of GUDCA on the mTOR-associated cell cycle G0/G1 arrest and cellular proliferation. These findings enrich the understanding of the function of bile acids and the mechanisms regulating FXR. RXRα is a member of the nuclear receptor family; RXRα interacts with approximately $\frac{1}{3}$ of the nuclear receptor family members to form a heterodimer or homodimer and acts in many physiological and pathological processes. Numerous studies have confirmed that several antitumor agents can target RXRα to regulate cell cycle distribution and inhibit the proliferation of tumor cells (Zhang et al., 2016). However, it is rarely reported that the heterodimer of RXRα/FXR in response to medicinal agents participates in the progression of HCC. In the present study, we verified that GUDCA inhibits the interaction of RXRα with FXR during the suppression of HCC proliferation in HepG2 cells and that the FXR R331 mutant weakened the effect of GUDCA on the interaction of RXRα with FXR and cellular proliferation. In conclusion, the findings of this study demonstrate that the RXRα/FXR interaction is essential for GUDCA-mediated suppression of hepatocarcinoma cell proliferation. Combining our previous research that celastrol could directly bind to the Nur77 nuclear receptor to clear up inflamed mitochondria and repair liver injury, the present study elucidated that the effects of celastrol are regulated via a mechanism of synergistic functions via directly and/or indirectly targeting three nuclear receptors, FXR, RXRα, and Nur77, which provides a novel strategy for the design of antitumor agents. ## 5 Conclusion The present study revealed a mechanism of celastrol-alleviating HCC that celastrol could regulate the gut microbiota and liver bile acid metabolism, inhibit the interaction of FXR with RXRα in the liver, induce mTOR/S6K1-related cell cycle G0/G1 phase arrest, and eventually alleviate the proliferation of HCC (Figure 8). FIGURE 8: Mechanism summary of celastrol alleviating proliferation of hepatocellular carcinoma via regulation of the interaction of FXR with RXRα modulated by gut microbiota-associated bile acid metabolism. ## Data availability statement The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author. ## Ethics statement The animal study was reviewed and approved by the Animal Care and Use Committee of Xiamen University. ## Author contributions DZ, LZ, and QL performed the experiments and analyzed the data. DZ and QL designed and supervised the study. QL wrote the manuscript. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2023.1124240/full#supplementary-material ## References 1. Chang W., He W., Li P-P., Song S-S., Yuan P-F., Lu J-T.. Protective effects of Celastrol on diethylnitrosamine-induced hepatocellular carcinoma in rats and its mechanisms. Eur. J. Pharmacol. (2016) 784 173-180. DOI: 10.1016/j.ejphar.2016.04.045 2. Chang Y., Han H., Rho S., Kim A., Kim T H., Jang K.. FXR inhibits endoplasmic reticulum stress-induced NLRP3 inflammasome in hepatocytes and ameliorates liver injury. Cell Rep. (2018) 24 2985-2999. DOI: 10.1016/j.celrep.2018.07.068 3. Chávez-Talavera O., Tailleux A., Lefebvre P., Staels B.. Bile acid control of metabolism and inflammation in obesity, type 2 diabetes, dyslipidemia, and nonalcoholic fatty liver disease. Gastroenterology (2017) 152 1679-1694. DOI: 10.1053/j.gastro.2017.01.055 4. Chen P., Wang B., Meng L., Chunxue C., Liu F., Gao Y.. Celastrol inhibits the proliferation and migration of MCF-7 cells through the leptin-triggered PI3K/AKT pathway. Comput. Struct. Biotechnol. J. (2022) 20 3173-3181. DOI: 10.1016/j.csbj.2022.06.042 5. Chen S-R., Dai Y., Zhao J., Lin L., Wang Y., Wang Y.. A mechanistic overview of triptolide and celastrol, natural products from tripterygium wilfordii Hook F. Front. Pharmacol. (2018) 9 104. DOI: 10.3389/fphar.2018.00104 6. Cheng W T., Kumar Kantilal H., Davamani F.. The mechanism of Bacteroides fragilis toxin contributes to colon cancer formation. Malays J. Med. Sci. (2020) 27 9-21. DOI: 10.21315/mjms2020.27.4.2 7. Chiang J Y. L., JessicaFerrell M.. Bile acid metabolism in liver pathobiology. Gene Expr. (2018) 18 71-87. DOI: 10.3727/105221618X15156018385515 8. Chiang J Y. L., JessicaFerrell M.. Bile acid receptors FXR and TGR5 signaling in fatty liver diseases and therapy. Am. J. Physiol. Gastrointest. Liver Physiol. (2020) 318 G554-G573. DOI: 10.1152/ajpgi.00223.2019 9. Corson T W., Crews C M.. Molecular understanding and modern application of traditional medicines: Triumphs and trials. Cell (2007) 130 769-774. DOI: 10.1016/j.cell.2007.08.021 10. Fang J., Chang X.. Celastrol inhibits the proliferation and angiogenesis of high glucose-induced human retinal endothelial cells. Biomed. Eng. Online (2021) 20 65. DOI: 10.1186/s12938-021-00904-5 11. Haghi F., Goli E., Mirzaei B., Zeighami H.. The association between fecal enterotoxigenic B. fragilis with colorectal cancer. BMC Cancer (2019) 19 879. DOI: 10.1186/s12885-019-6115-1 12. Hu H., Zhang Y., Zhao F., Chen K., Wu T., Liu Q.. Celastrol inhibits intestinal lipid absorption by reprofiling the gut microbiota to attenuate high-fat diet-induced obesity. iScience (2021) 24 102077. DOI: 10.1016/j.isci.2021.102077 13. Hu M., Luo Q., Alitongbieke G., Chong S., Xu C., Xie L.. Celastrol-induced Nur77 interaction with TRAF2 alleviates inflammation by promoting mitochondrial ubiquitination and autophagy. Mol. Cell (2017) 66 141-153. DOI: 10.1016/j.molcel.2017.03.008 14. Kakiyama G., Pandak W M., PatrickGillevet M., Hylemon P B., DouglasHeuman M., Daita K.. Modulation of the fecal bile acid profile by gut microbiota in cirrhosis. J. Hepatology (2013) 58 949-955. DOI: 10.1016/j.jhep.2013.01.003 15. Li X., Zhu G., Yao X., Wang N., Hu R., Kong Q.. Celastrol induces ubiquitin-dependent degradation of mTOR in breast cancer cells. Onco Targets Ther. (2018) 11 8977-8985. DOI: 10.2147/OTT.S187315 16. Luo D., Guo Y., Cheng Y., Jia Z., Wang Y., Rong J.. Natural product celastrol suppressed macrophage M1 polarization against inflammation in diet-induced obese mice via regulating Nrf2/HO-1, MAP kinase and NF-κB pathways. Aging (Albany NY) (2017) 9 2069-2082. DOI: 10.18632/aging.101302 17. Luo Q., Wang Z., Chen H., Fang H., Xie S., Qian X.. Identification of anticancer drug candidate targeting nuclear receptor retinoid X receptor-α from natural products using receptor-ligand recognition. Chemistryselect (2016) 1 3909-3913. DOI: 10.1002/slct.201600907 18. Molinaro A., Marschall H-U.. Bile acid metabolism and FXR-mediated effects in human cholestatic liver disorders. Biochem. Soc. Trans. (2022) 50 361-373. DOI: 10.1042/BST20210658 19. Ng S W., Chan Y., Kumar Chellappan D., Madheswaran T., Zeeshan F., Chan Y L.. Molecular modulators of celastrol as the keystones for its diverse pharmacological activities. Biomed. Pharmacother. (2019) 109 1785-1792. DOI: 10.1016/j.biopha.2018.11.051 20. Qi Z., Tang P., Zhang T., Huang J-F., Xiao X-R., Zhu W-F.. Celastrol ameliorates acute liver injury through modulation of PPARα. Biochem. Pharmacol. (2020) 178 114058. DOI: 10.1016/j.bcp.2020.114058 21. Radun R., Trauner M.. Role of FXR in bile acid and metabolic homeostasis in NASH: Pathogenetic concepts and therapeutic opportunities. Semin. Liver Dis. (2021) 41 461-475. DOI: 10.1055/s-0041-1731707 22. Ramírez-Pérez O., Cruz-Ramón V., Chinchilla-López P., Méndez-Sánchez N.. The role of the gut microbiota in bile acid metabolism. Ann. Hepatol. (2017) 16 S21-S26-s20. DOI: 10.5604/01.3001.0010.5672 23. Ridlon J M., Kang D-J., Hylemon P B., Jasmohan S. B.. Gut microbiota, cirrhosis, and alcohol regulate bile acid metabolism in the gut. Dig. Dis. (2015) 33 338-345. DOI: 10.1159/000371678 24. Shen B., Chen H-B., Zhou H-G., Wu M-H.. Celastrol induces caspase-dependent apoptosis of hepatocellular carcinoma cells by suppression of mammalian target of rapamycin. J. Tradit. Chin. Med. (2021) 41 381-389. DOI: 10.19852/j.cnki.jtcm.2021.03.006 25. Staley C., AlexaWeingarden R., Alexander K., Sadowsky M J.. Interaction of gut microbiota with bile acid metabolism and its influence on disease states. Appl. Microbiol. Biotechnol. (2017) 101 47-64. DOI: 10.1007/s00253-016-8006-6 26. Wang X., Chen L., Wang H., Cai W., Xie Q.. Modulation of bile acid profile by gut microbiota in chronic Hepatitis B. J. Cell. Mol. Med. (2020) 24 2573-2581. DOI: 10.1111/jcmm.14951 27. Wu L., Jiao F., Li J., Yu Q., Ji J., Wu J.. The gut microbiome-bile acid axis in hepatocarcinogenesis. Biomed. Pharmacother. (2021) 133 111036. DOI: 10.1016/j.biopha.2020.111036 28. Yamada S., Takashina Y., Watanabe M., Nagamine R., Saito Y., Kamada N.. Bile acid metabolism regulated by the gut microbiota promotes non-alcoholic steatohepatitis-associated hepatocellular carcinoma in mice. Oncotarget (2018) 9 9925-9939. DOI: 10.18632/oncotarget.24066 29. Yan C-Y., Ouyang S-H., Wang X., Wu Y-P., Sun W-Y., Duan W-J.. Celastrol ameliorates Propionibacterium acnes/LPS-induced liver damage and MSU-induced gouty arthritis via inhibiting K63 deubiquitination of NLRP3. Phytomedicine (2021) 80 153398. DOI: 10.1016/j.phymed.2020.153398 30. Zhang T,, Qi Z., Xiao X., Yang R., Hu D., Zhu X.. Modulation of lipid metabolism by celastrol. J. Proteome Res. (2019) 18 1133-1144. DOI: 10.1021/acs.jproteome.8b00797 31. Zhang X., Hu Z., Su Y.. Targeting truncated RXRα for cancer therapy. Acta Biochim. Biophys. Sin. (Shanghai). (2016) 48 49-59. DOI: 10.1093/abbs/gmv104
--- title: A systematic analysis of the global and regional burden of colon and rectum cancer and the difference between early- and late-onset CRC from 1990 to 2019 authors: - Liu-Bo Li - Li-Yu Wang - Da-Ming Chen - Ying-Xia Liu - Yuan-Hui Zhang - Wei-Xiang Song - Xu-Bo Shen - Sheng-Quan Fang - Zheng-Yuan Ma journal: Frontiers in Oncology year: 2023 pmcid: PMC9975717 doi: 10.3389/fonc.2023.1102673 license: CC BY 4.0 --- # A systematic analysis of the global and regional burden of colon and rectum cancer and the difference between early- and late-onset CRC from 1990 to 2019 ## Abstract The burden of colorectal cancer (CRC) varies substantially across different geographical locations. However, there was no further quantitative analysis of regional social development and the disease burden of CRC. In addition, the incidence of early- and late-onset CRC has increased rapidly in developed and developing regions. The main purpose of this study was to investigate the trends in CRC burden across different regions, in addition to the epidemiological differences between early and late-onset CRC and their risk factors. In this study, estimated annual percentage change (EAPC) was employed to quantify trends in age-standardized incidence rate (ASIR), mortality rate, and disability-adjusted life-years. Restricted cubic spline models were fitted to quantitatively analyze the relationship between trends in ASIR and Human Development Index (HDI). In addition, the epidemiological characteristics of early- and late-onset CRC were investigated using analyses stratified by age groups and regions. Specifically, meat consumption and antibiotic use were included to explore the differences in the risk factors for early- and late-onset CRC. The quantitative analysis showed that the ASIR of CRC was exponentially and positively correlated with the 2019 HDI in different regions. In addition, the growing trend of ASIR in recent years varied substantially across HDI regions. Specifically, the ASIR of CRC showed a significant increase in developing countries, while it remained stable or decreased in developed countries. Moreover, a linear correlation was found between the ASIR of CRC and meat consumption in different regions, especially in developing countries. Furthermore, a similar correlation was found between the ASIR and antibiotic use in all age groups, with different correlation coefficients for early-onset and late-onset CRC. It is worth mentioning that the early onset of CRC could be attributable to the unrestrained use of antibiotics among young people in developed countries. In summary, for better prevention and control of CRC, governments should pay attention to advocate self-testing and hospital visits among all age groups, especially among young people at high risk of CRC, and strictly control meat consumption and the usage of antibiotics. ## Introduction Estimates from the International Agency for Research on Cancer in 2020 suggested that, globally, colorectal cancer (CRC) constituted over 1.9 million new cases and 900,000 deaths annually [1], ranking the third most commonly diagnosed malignancy and the second leading cause of cancer death [2]. Meanwhile, CRC resulted in 19.0 million disability-adjusted life-years (DALYs) globally, with an age-standardized rate (ASR) of 235.7 DALYs per 100,000 person-years [3]. In addition, CRC showed obvious regional and economic differences [4], which resulted in significant geographic heterogeneity in the age-standardized incidence rate (ASIR), mortality rate (ASMR), and DALYs rate (ASDR). As such, CRC accounted for a higher proportion of cancer burden and gradually surpassed infection-related cancers in the regions with high Human Development Index (HDI) [5], while ASIR showed a stable and even downward trend [6]. However, the incidence of early-onset CRC (usually defined as CRC patients younger than 50 years old) has gradually been increasing since the mid-1990s for unknown reasons [7] in developed countries [8, 9]. Meanwhile, the incidence of late-onset CRC (usually defined as CRC patients older than 50 years old) has increased sharply in developing countries due to the industrialization, the widespread acceptance of unhealthy lifestyles and diets, and environmental deterioration [10]. Therefore, it is necessary to conduct a comprehensive study on economic development and the CRC burden to explain the geographic heterogeneity. In addition, there is a need for in-depth epidemiological research on CRC burden in different regions. Meanwhile, evidence on the risk factors for early- and late-onset CRC is also needed. The contributions of the risk factors of CRC varied with time and by geographical region. For example, smoking and alcohol use have gradually decreased with the implementation of anti-smoking and alcohol policies and campaigns (11–13). In contrast, the influence of diet patterns, especially meat consumption, is affected by economic development and vary widely globally. In addition, antibiotic use was thought to be related to the incidence of early-onset CRC [14]. This study investigated these two risk factors in early- and late-onset CRC. As a non-communicable disease, reducing the global burden of CRC will help to achieve the Sustainable Development Goals (SDGs) of reducing by one-third premature mortality from non-communicable diseases by 2030 [15, 16]. The Global Burden of Diseases (GBD), Injuries, and Risk Factors Study 2019 assessed the CRC burden in 195 countries and territories globally, providing a unique perspective to help understand the general situation and time-varying landscape of CRC. Based on the 2019 report of the GBD Colorectal Cancer Collaborators [17], this study focused on the trends in ASR using estimated annual percentage changes (EAPCs) [18] from 1990 to 2019. In particular, to elucidate the underlying reasons for the geographic heterogeneity of CRC burden, regression analysis and restricted cubic spline analysis (RCS) [19] were implemented to analyze the relationship between HDI and EAPC in ASR. In addition, this study analyzed EAPCs in ASR across different age groups and regions and visually showed differences in the burden of early- and late-onset CRC. Furthermore, the correlation analysis of ASIR between dietary structure changes (represented by global meat consumption) and potential risk factors (represented by antimicrobial usage) was used to explore potential risk factors for early-onset and late-onset CRC. As an extension and complementary study, this analysis could potentially promote improvements in CRC healthcare and clinical guidelines. ## Data source Data on incidence, mortality, DALYs, age-standardized incidence rate (ASIR), age-standardized mortality rate (ASMR), age-standardized DALYs rate (ASDR), and risk exposures of CRC were obtained from the Global Burden of Disease Study 2019 (GBD 2019) on the GHDx (Global Health Data Exchange) data source (http://ghdx.healthdata.org/gbd-results-tool). Risk exposures, including behavioral factors (alcohol use and smoking), dietary factors (including high fasting plasma glucose, diet low in calcium, diet low in milk, diet low in fiber, diet high in red meat, and diet high in processed meat), and metabolic factors (including high fasting plasma glucose and high body-mass index) were included in this study. According to geographical features, the world was divided into 21 geographic regions, including Australasia, East Asia, and Eastern Europe. Based on GBD 2019 data, countries and territories were categorized into five regions based on the socio-demographic index (SDI), including high, high-middle, middle, low-middle, and low SDI. SDI reflects the level of medical health. Meanwhile, comprised by macro indicators of human development, the Human Development Index (HDI) was obtained from the United Nations Development Program (UNDP; http://hdr.undp.org/en/data). The relationship between human development and disease burden was investigated using correlation analysis of GBD data and HDI. Trends in disease burden were assessed across different age groups (10–14, 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84, 85–89, 90–94, 95 plus). In addition, this study used data on antibiotic use (https://www.tropicalmedicine.ox.ac.uk/research/oxford/microbe/gram-project/antibiotic-usage-and-consumption) and data on the economic statistics on the nutrition and health industry-global meat consumption per capita from 1990 to 2029 (https://www.ckcest.cn/default/es3/detail/4004/dw_dataset/C9A75565E5D00001E17C1C012A571CE2). ## Statistical analysis EAPC was used to quantify trends in ASRs, calculated using a generalized linear model with a Gaussian distribution. A regression line was used to estimate the natural logarithm of the rates; for example, y = α + βx + ϵ, where y = ln(ASR) and x = calendar year. The EAPC was calculated as 100 × [exp(β)-1] alongside the $95\%$ confidence interval (CI) using linear regression [18]. An increasing trend was observed when the EAPC value and its $95\%$ CI were larger than 0. In contrast, a decreasing trend was observed when the EAPC value and its $95\%$ CI were less than 0. In addition, a correlation analysis between HDI and disease burden was analyzed using Pearson’s correlation coefficient. Regression analysis and restricted cubic spline models (RCS) models were fitted to explore a nonlinear relationship between the 2019 HDI and EAPC in ASR. Finally, Pearson’s correlation analysis was used to analyze the relationship of ASIR with global meat consumption and antimicrobial usage. All statistics were performed using R (Version 3.6.0). A p-value of less than 0.05 was considered statistically significant. ## Patient and public involvement Patients or the public were not involved in the design, data collection, analyses, or interpretation of this research. ## Trends in the incidence of CRC Globally, the absolute number of colorectal new cancer cases reached 2166.17 × 103 ($95\%$ UI = 1996.29 × 103 to 2342.84 × 103) in 2019, representing a 1.57 times increase from 842.10 × 103 ($95\%$ UI = 810.41 × 103 to 868.57 × 103) in 1990. The ASIR increased gradually (EAPC = 0.58, $95\%$ UI = 0.52–0.65) from 22.25 ($95\%$ UI = 21.30–22.97) per 100,000 persons in 1990 to 26.71 ($95\%$ UI = 24.58–28.89) per 100,000 persons in 2019 (Table 1). The largest increasing trend occurred in the middle SDI region (EAPC = 2.61, $95\%$ UI = 2.44 to 2.77, Figure 1). The absolute number of new cases increased by 4.67 times in 30 years, and the ASIR increased fastest in East Asia (EAPC = 3.61, $95\%$ UI = 3.33–3.89). The ASIR overran 40 per 10,000 persons in the Asia Pacific-high income, Australasia, Western Europe, and North America-high income regions. Among them, ASIR declined the most (EAPC = -0.62, $95\%$ UI = -0.72 to -0.51) in the North America-high income region. In Africa, the overall incidence rate showed a slowly growing trend, with the highest growth trend in Western sub-Saharan Africa (EAPC = 1.21, $95\%$ UI = 1.10 to 1.32). In particular, the EAPC exceeded 3.6 in Equatorial Guinea, Vietnam, and China. ## Trends in mortality due to CRC In 2019, 1085.80 ($95\%$ UI 1002.80 to 1149.68) × 103 deaths were attributed to CRC (Table 2), corresponding to a $109.56\%$ ($95\%$ UI = $96.20\%$ to $121.74\%$) increase compared to 1990. In addition, the ASMR showed a decreasing trend during the last three decades (EAPC = -0.21, $95\%$ UI = -0.28 to -0.14). Meanwhile, the high SDI region showed a downward trend (EAPC=-1.09, $95\%$ UI= -1.17 to -1.00), while the middle and low-middle regions showed obvious increasing trends (EAPC = 1.24, $95\%$ UI = 1.10–1.37; EAPC = 1.15, $95\%$ UI = 1.11–1.19, respectively). Table 2 \| Characteristics \| No. of deaths \| Unnamed: 2 \| Unnamed: 3 \| DALYs \| Unnamed: 5 \| Unnamed: 6 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Characteristics \| Number in 2019, cases X 103 (95% UI) \| Percentage change in absolute number (%) \| EAPC (95% CI) \| Number in 2019, cases X 103 (95% UI) \| Percentage change in absolute number (%) \| EAPC \| \| Characteristics \| Number in 2019, cases X 103 (95% UI) \| Percentage change in absolute number (%) \| EAPC (95% CI) \| Number in 2019, cases X 103 (95% UI) \| Percentage change in absolute number (%) \| (95%CI) \| \| Total \| 1085.80 (1002.80–1149.68) \| 109.56 (96.20–121.74) \| -0.21 (-0.28 to -0.14) \| 24,284.09 (22,614.92–25,723.22) \| 95.71 (82.15–108.59) \| -0.21 (-0.26 to -0.15) \| \| Sex \| Sex \| Sex \| Sex \| Sex \| Sex \| Sex \| \| Male \| 594.18 (550.18–638.03) \| 130.92 (110.50–150.51) \| 0.1 (0.03–0.17) \| 13,959.59 (12,859.72–15,045.72) \| 115.64 (93.77–136.18) \| 0.13 (0.06–0.2) \| \| Female \| 491.62 (437.55–532.38) \| 88.50 (74.84–102.42) \| -0.59 (-0.67 to -0.52) \| 10,324.5 (9494.91–11,149.97) \| 73.79 (60.22–87.63) \| -0.65 (-0.71 to -0.6) \| \| SDI \| SDI \| SDI \| SDI \| SDI \| SDI \| SDI \| \| Low \| 34.66 (30.96–38.61) \| 157.80 (112.08–210.23) \| 0.58 (0.52–0.64) \| 942.42 (835.79–1059.27) \| 147.32 (101.56–200.78) \| 0.44 (0.38–0.5) \| \| Low-middle \| 116.55 (105.51–128.33) \| 223.28 (171.75–267.87) \| 1.15 (1.11–1.19) \| 2998.93 (2703.93–3314.97) \| 189.96 (171.75–267.87) \| 1.01 (0.98–1.05) \| \| Middle \| 279.78 (251.15–306.14) \| 235.57 (196.69–275.09) \| 1.24 (1.1–1.37) \| 6990.43 (6308.90–7671.29) \| 195.47 (159.84–275.09) \| 1.11 (0.99–1.24) \| \| High-middle \| 326.64 (299.66–349.53) \| 100.78 (85.89–115.23) \| -0.04 (-0.15 to 0.06) \| 7174.86 (6649.07–7693.26) \| 81.16 (67.40–95.16) \| -0.18 (-0.28 to -0.08) \| \| High \| 327.57 (294.90–345.58) \| 47.35 (39.21–52.76) \| -1.09 (-1.17 to -1.00) \| 6164.66 (5754.65–6435.90) \| 32.28 (27.55–36.63) \| -1.11 (-1.17 to -1.04) \| \| Regions \| Regions \| Regions \| Regions \| Regions \| Regions \| Regions \| \| East Asia \| 275.60 (238.24–317.89) \| 230.83 (174.13–295.52) \| 1.4 (1.17–1.64) \| 6712.86 (5774.28–7735.91) \| 181.40 (132.43–238.11) \| 1.24 (1.02–1.46) \| \| South Asia \| 94.85 (81.52–109.07) \| 247.31 (177.91–314.50) \| 0.91 (0.79–1.04) \| 2419.10 (2078.02–2782.57) \| 207.50 (146.30–267.01) \| 0.8 (0.68–0.92) \| \| Southeast Asia \| 82.02 (67.62–94.61) \| 246.99 (189.60–300.06) \| 1.25 (1.19–1.31) \| 2142.43 (1780.49–2482.29) \| 218.40 (166.60–265.82) \| 1.12 (1.05–1.18) \| \| Central Asia \| 7.47 (6.82–8.17) \| 45.80 (33.65–60.40) \| 0.33 (0.17–0.5) \| 199.84 (182.01–219.94) \| 35.20 (23.23–49.85) \| -0.37 (-0.53 to -0.21) \| \| High-income Asia Pacific \| 76.93 (64.82–83.6) \| 124.03 (98.74–138.23) \| -0.68 (-0.75 to -0.62) \| 1327.82 (1186.12–1414.81) \| 64.65 (52.3–72.59) \| -0.88 (-0.96 to -0.8) \| \| Oceania \| 0.55 (0.44–0.68) \| 167.41 (230.06–114.82) \| 0.41 (0.34–0.49) \| 16.31 (12.93–20.56) \| 162.99 (108.35–230.10) \| 0.36 (0.3–0.42) \| \| Australasia \| 8.38 (7.57–8.98) \| 48.65 (38.89–57.93) \| -1.73 (-1.88 to -1.59) \| 163.25 (150.87–173.96) \| 29.41 (22.03–37.23) \| -1.89 (-2.03 to -1.17) \| \| Eastern Europe \| 63.48 (57.18–70.01) \| 27.39 (15.58–39.68) \| -0.31 (-0.51 to -0.12) \| 1419.10 (1287.54–1571.37) \| 15.25 (4.68–27.06) \| -0.57 (-0.79 to -0.34) \| \| Western Europe \| 172.45 (155.34–181.81) \| 36.74 (25.02–36.67) \| -1.09 (-1.25 to -0.92) \| 3008.23 (2815.06–3152.89) \| 16.01 (11.24–20.10) \| -1.17 (-1.3 to -1.03) \| \| Central Europe \| 51.57 (45.64–57.75) \| 67.28 (48.95–85.76) \| 0.32 (0.21–0.42) \| 1052.15 (922.92–1184.25) \| 46.84 (29.53–64.72) \| 0.17 (0.07–0.27) \| \| High-income North America \| 95.66 (88.32–99.69) \| 33.04 (28.93–36.55) \| -1.22 (-1.32 to -1.13) \| 1987.11 (1895.87–2059.77) \| 31.74 (28.13–35.37) \| -1.07 (-1.16 to -0.97) \| \| Andean Latin America \| 5.63 (4.59–6.79) \| 269.75 (201.86–349.77) \| 1.16 (1.02–1.3) \| 125.58 (101.75–151.80) \| 230.17 (163.34–321.74) \| 1.04 (0.89–1.18) \| \| Central Latin America \| 22.47 (19.54–26.00) \| 292.01 (241.04–349.36) \| 0.97 (0.93–1.01) \| 539.64 (465.2–627.07) \| 264.23 (214.09–322.09) \| 1.11 (1.07–1.15) \| \| Caribbean \| 7.99 (6.94–9.18) \| 143.26 (112.23–175.87) \| 0.63 (0.57–0.68) \| 172.02 (147.19–200.17) \| 126.54 (94.75–160.28) \| 0.6 (0.54–0.66) \| \| Tropical Latin America \| 27.7 (25.67–29.09) \| 226.92 (208.22–241.92) \| 0.60 (0.44–0.75) \| 660.13 (625.56–687.74) \| 197.23 (182.41–211.04) \| 0.68 (0.51–0.85) \| \| Southern Latin America \| 17.93 (16.77–18.97) \| 103.09 (92.66–114.21) \| 0.15 (0.05–0.25) \| 366.44 (347.73–385.44) \| 87.63 (77.90–97.75) \| 0.17 (0.09–0.24) \| \| Eastern Sub-Saharan Africa \| 12.72 (10.94–15.00) \| 159.54 (109.72–221.98) \| 0.69 (0.62–0.76) \| 356.43 (301.93–425.61) \| 154.30 (101.54–225.87) \| 0.54 (0.47–0.62) \| \| Southern Sub-Saharan Africa \| 5.92 (5.33–6.58) \| 130.50 (105.16–168.87) \| 0.45 (0.24–0.66) \| 147.78 (132.44–165.54) \| 125.01 (98.84–158.74) \| 0.49 (0.28–0.70) \| \| Western Sub-Saharan Africa \| 13.77 (11.70–16.07) \| 165.41 (117.27–232.00) \| 1.05 (0.95–1.15) \| 353.24 (295.57–420.70) \| 165.74 (113.86–238.56) \| 0.88 (0.79–0.97) \| \| North Africa and Middle East \| 39.15 (34.76–44.11) \| 199.32 (140.51–280.08) \| 0.81 (0.62–1.00) \| 1013.63 (896.16–1146.53) \| 177.60 (120.11–249.96) \| 0.56 (0.38–0.74) \| \| Central Sub-Saharan Africa \| 3.54 (2.7–4.61) \| 133.84 (60.65–233.68) \| -0.11 (-0.36 to 0.14) \| 100.99 (75.75–131.45) \| 131.68 (58.49–224.76) \| -0.14 (-0.38 to 0.10) \| The absolute new number of deaths in East Asia increased by 2.31 times, with 275.60 (UI = 238.24–317.89) × 103 deaths in 2019 and showed the highest increase in the trend for ASMR (EAPC = 1.4, $95\%$ UI = 1.17–1.64) within 30 years. On the contrary, Australasia showed the lowest decrease in ASMR (EAPC = -1.73, $95\%$ UI = -1.88 to -1.59). Meanwhile, in Africa, Western sub-Saharan Africa showed the highest increase trend (EAPC = 1.05, $95\%$ UI = 0.95–1.15). Among the 195 countries and territories, China had the highest absolute number of new deaths in 2019 (261,776 new cases), followed by the United States, India, and Japan. Meanwhile, Austria and Singapore had the fastest decline in ASMR within 30 years, with EAPCs of -3.09 and -2.32, respectively (Figure 2). Figure 2: The absolute number of deaths and age-standardized rate of mortality and its EAPC value in 195 countries/territories and 21 regions. (A) the ASMR and its corresponding EAPC value in 21 regions of the world in 2019, in which red bar chart represents ASMR, blue bar chart represents EAPC of ASMR; (B) ASMR of CRC in 195 countries/territories in 2019; (C) The absolute number of deaths in 21 regions in 1990 and 2019; (D) The EAPC of ASMR in 195 countries/territories in 2019. ## Trends in DALYs due to CRC Globally, the absolute number of DALYs increased by $95.71\%$ ($95\%$ UI =$82.15\%$–$108.59\%$) from 1990 to 2019, reaching 24284.09 × 103 ($95\%$ UI = 22614.92 × 103 to 25723.22 × 103) in 2019. Meanwhile, the ASDR decreased slightly from 1990 to 2019 (EAPC = -0.21, $95\%$ UI = -0.26 to -0.15). In the 21 geographic regions, the absolute number of DALYs in 2019 was highest in East Asia (6712.86×103, $95\%$ UI = 5774.28×103 to 7735.91×103), and the largest increase in the number of DALYs, $264.23\%$ ($95\%$ UI = $214.09\%$–$322.09\%$), occurred in central Latin America (Table 2). In addition, the DALYs in Africa increased significantly, with an increase of $117.60\%$ in North Africa and the Middle East. However, the DALYs in Western Europe increased by only $16.01\%$ ($95\%$ UI = $11.24\%$–$21.10\%$). Singapore and Austria had the greatest decline in trends of ASDR (EAPC less than -2.3) (Figure 3). Figure 3: The absolute number of DALYs and age-standardized rate of DALYs and its EAPC value in 195 countries/territories and 21 regions. (A) the ASDR and its corresponding EAPC value in 21 regions of the world in 2019, in which red bar chart represents ASDR, blue bar chart represents EAPC of ASDR; (B) ASDR of CRC in 195 countries/territories in 2019; (C) The absolute number of DALYs in 21 regions in 1990 and 2019; (D) The EAPC of DALYs in 195 countries/territories in 2019. Obvious polarization in the variation track of DALYs across the 21 GBD regions between 1990 and 2019 was observed (Figure 4). In the lower left corner of the figure, a relatively positive correlation was observed in regions such as South Asia, Southeast Asia, Western Europe, Western sub-Saharan Africa, and Eastern sub-Saharan Africa, where their ASDR increased with the fast-growing SDI. On the other hand, an inverse association was observed in Eastern Europe, Western Europe, and Australasia, where the SDI was growing slowly from 1990 to 2019, accompanied by rapidly declining ASDRs. Among the 21 SDI regions, the ASDR declined by $36.25\%$ in Australasia and increased by $41.58\%$ in Southeast Asia in 3 decades. Figure 4: Age-standardized DALYs of CRC for 21 GBD regions by SDI between 1990 and 2019, in which different colors and shapes represent different regions. ## The relationship between HDI and ASR Pearson’s correlation analysis showed an exponential and positive correlation between ASIR and HDI (Figure S1 in the supplementary material), with a correlation coefficient of 0.78. On the other hand, this study tested the overall and nonlinear association between the 2019 HDI and EAPC of ASR using RCS. The test for the overall association between the 2019 HDI and EAPC of ASR was significant, meaning that regardless of the shape of the association, 2019 HDI and EAPC were significantly correlated. In addition, the test of a nonlinear association was significant, indicating that the association was significantly nonlinear. The RCS curve on EAPC of ASIR and 2019 HDI showed an inverted V-shape, with the turning point at an HDI of 0.74 (Figure 5). Figure 5: the relationship between HDI 2019 in 195 countries/territories and EAPC of ASR of incidence, mortality, and DALYs. Scatter plot and fitting curve described the corresponding relationship between HDI and EAPC of ASIR (A), ASMR (B), ASDR (C), in which the size of the scatter represents the number of incidence (A), mortality (B), DALYs (C). Display of nonlinear relationship between ASIR (D), ASMR (E), ASDR (F) and HDI 2019 in 195 countries/territories in restricted cubic spline models, shaded areas represent 95% CIs around incidence trend. ## The correlation analysis of meat consumption and ASIR To further investigate the relationship between dietary changes and ASIR, we included global data on the economic statistics of the nutrition and health industry-global meat consumption per capita (1990–2029) in this study. As shown in Figure 6, the relationship between meat consumption and ASIR showed a linear correlation in different countries from 1990 to 2019, especially for late-onset CRC in developing countries. In the United States, Australia, and New Zealand, meat consumption and the ASIR of CRC were both at high levels during the past 30 years, with little change and weak correlation. On the other hand, in China and Vietnam, meat consumption and ASIR showed a linear correlation and rapid increase during the past decades. In addition, India had a low-meat diet and low lower ASIR. Figure 6: Scatter chart of the relationship between age-standardized rate of incidence and average meat consumption in selected countries around the world from 1990 to 2019. ## Trends in the burden of CRC in different age groups Trends in ASR are shown in Figure 7. There were substantial differences in disease burden between different SDI regions. The ASIR showed an increasing trend in the global population over 20 years old, and the largest increasing trend occurred in the 25–49 age group (EAPC larger than 1.0), which illustrated that the incidence of early-onset CRC was in a stage of rapid growth than that of middle-aged and elderly population. Those in the 20–54 age group experienced an increasing trend in ASIR, while a downward trend was observed among those in the 55–90 age group. Among all age groups, the highest increasing trend occurred in the 30–34 age group (EAPC = 1.11), and the most obvious downward trend was observed in those aged 70–74 years old (EAPC = -0.47). Figure 7: Trends of incidences, mortality, and DALYs in different age groups and SDI regions around the world. (A) EAPC of ASDR in different age and SDI groups; (B) EAPC of ASMR in different age and SDI group (C) EAPC of ASIR in different age and SDI groups. There was a steady increase of ASIR in high-middle SDI regions (EAPC >1), with two peaks among those in the 25–29 and 95-plus age groups (EAPC equal to 1.89 and 1.93, respectively). The fastest increase in the ASIR of early-onset CRC occurred in the 25–29 age group. In the middle SDI regions, the incidence increased steadily in all age groups, maintaining an abnormal increase trend in the 40–89 age group (EAPC >2.0). This histogram of EAPC was similar to a normal distribution, where EAPC peaked at the 65–69 age group (EAPC equal to 2.91). In the low-middle SDI regions, the fastest increase in trend was observed in the 80–84 age group (EAPC = 2.15). In the low-SDI areas, EAPC showed a steady increase by age, with a peak at the 95-plus age group (EAPC equal to 1.79). The ASMR and ASDR showed the same downward trend globally, except for those over 90 years old. However, in the high SDI regions, the decreasing trend of ASMR and ASDR between the ages of 20 and 44 was slower compared to that of the 45–75 age group, especially in the age group of 30–34 (EAPC of ASMR = -0.18). In the high-middle SDI region, the ASMR and ASIR gradually decreased in age groups under 75 and increased among those over 75 years old. In the middle SDI region, ASMR and ASDR increased in those over 35 years old. ## The correlation analysis of antibiotic use and ASIR Pearson’s correlation analysis of data on antibiotic use was conducted further explore the risk factors. Figure 8 shows the correlation between the increasing trend of ASIR in those 20–49, 50–75, and over 75 years old and the EAPC of antibiotic use. The early-onset CRC group had the highest correlation (correlation coefficient = 0.21). In addition, we conducted a correlation analysis between ASIR and antibiotic use in different age groups. Similar correlation relationships were found in early-onset and late-onset CRC. In addition, the correlation coefficient of ASIR and antibiotic use increased with patients’ age in early-onset CRC but remained stable in late-onset CRC (Figure 9). Figure 8: The correlation between the EAPC of ASIR in 20–49 (A), 50–75, (B) and over 75 years old (C) and the annual percentage changes of antibiotic usage globally. Figure 9: The correlation between antibiotic usage and ASIR in 2000, 2010, and 2018. (A). The correlation analysis between ASIR and the global antibiotic usage in two age groups of 25–29 in 2010; (B). The correlation analysis between ASIR and the global antibiotic usage in two age groups of 65–69 in 2010; (C). The correlation analysis between ASIR and the global antibiotic usage in two age groups of 25–29 in 2018; (D). The correlation analysis between ASIR and the global antibiotic usage in two age groups of 65–69 in 2018; (E). Three lines graphs shown the value of correlation coefficient between ASIR of different age groups and global antimicrobial usage in 2000, 2010, and 2018. ## Discussion Although the age-standardized incidence, mortality, and DALYs of CRC have declined in some countries or regions around the world in the past 30 years, the absolute number has increased due to the development of global population growth, population aging, and the gradual increase of early-onset disease. Previous reports have shown that, to a certain extent, the incidence was related to the regional economy and seemed to be positively associated with the level of socioeconomic development [4, 20]. This study reached similar conclusions that ASIR has increased exponentially with the growth of HDI. The reason why the RCS curve of 2019 HDI and ASIR showed an inverted V-shape might be that an increase in a country’s HDI leads to improvements in medical resources and medical insurance policy, resulting in amelioration of screening and medical treatment for diseases. This has brought about positive screening results of CRC over time, along with exponential increases in ASIR. However, after HDI increased to 0.74, the trend in ASIR remained stable or started declining. Globally, the incidence of CRC varied greatly across different regions, with decreasing trends in developed countries, represented by North America and Australia, and gradually increasing trends in developing countries, represented by South America and Southeast Asia. It is worth noting that the incidence in Asia was generally increasing, while the mortality rate within this region was quite different. In the past 3 decades, Vietnam experienced socioeconomic development, rapid urbanization, and lifestyle changes, and CRC was projected to be the most predominant cancer in Vietnamese males and the second most common cancer in females [21]. However, the trend of ASDR and ASIR in China was significantly lower than in Vietnam. Such intra-regional differences could further be ascribed to socioeconomic development and improved healthcare resources, such as price negotiation on anti-cancer drugs, extensive medical insurance, and exempt tariffs on imported cancer drugs in China’s medical insurance policy [22, 23]. In addition, the fastest decline in the trend of ASIR occurred in Austria, which was bound up with long-standing screening programs using colonoscopy and fecal tests (fecal test and colonoscopy screening programs were started in 1980 and 2005, respectively). About $68.6\%$ of the population received chemical detection of fecal occult blood test (gFOBT) within 2 years or a colonoscopy within 10 years [24]. The DALYs caused by CRC ranked third among the cancer-related causes of DALYs globally in 2019 (GBD 2019 Disease and Injury Incidence and Prevalence Collaborators, 2019), and the DALYs varied greatly on a global scale [3]. In this study, the global imbalance in DALYs was manifested as an obvious two-level differentiation. The burden of CRC in low-SDI regions increased annually, largely due to the huge population and insufficient medical resources [25]. Nevertheless, minor decreasing trends were observed in low and low-middle SDI regions, such as Southern sub-Saharan Africa, which might be an optimistic achievement of local health infrastructure, international cooperation, and health aid [25, 26]. On the other hand, ASDR has dropped by $36.26\%$ in the past 30 years in Australasia, mainly due to The Australian National Bowel Cancer Screening Program fully rolled-out 2-yearly screening using the immunochemical Fecal Occult Blood Testing (iFOBT) in people aged 50–74 years [27]. About 92,200 new cases and 59,000 deaths are estimated to be prevented from 2015 to 2040. Furthermore, a colorectal screening for people aged 55–59 years was included in the national bowel screening program in New Zealand, which significantly reduced the incidence and mortality of CRC [28]. In Singapore, CRC screening included a fecal immunochemical test (FIT) performed annually, while a colonoscopy could be performed every 10 years. Moreover, since 2003, the community health assistance program under the census program has been implemented annually for citizens and permanent residents over the age of 50, and censuses on CRC were carried out with free screening kits provided by general practice clinics [29, 30]. In addition, the recommended age for screening varies in different regions worldwide. Japan provides an annual FIT screening service for people aged 40 and over [31], which is one of the reasons for low mortality. Therefore, to reduce CRC mortality, the key lies in supporting a medical insurance system, public participation in colorectal cancer screening, and early screening. Although the overall incidence of CRC in developed countries such as Europe and the United States was gradually decreasing, early-onset CRC now constitutes a substantial cancer burden among younger adults [32]. The highest incidence of early-onset CRC occurred in the rectum, accounting for $42\%$ [7]. The incidence of colon cancer in adults aged 20–49 years increased by 6.4–$9.3\%$, while that of rectal cancer increased by 1.6–$3.5\%$ annually in Europe [9]. Our study found that the ASIR among the 25–34 age group in high SDI regions was fast-growing. This study suggests CRC in young adults should be monitored in future studies. The health administration departments in developing countries should not only popularize screening for CRC in middle-aged and elderly persons but also be alert to the prevalence of early-onset CRC. In addition, from a global perspective, the mortality rate of early-onset CRC patients in the 35–50 age group has decreased less than that of the middle-aged and elderly group, which was related to the more invasive molecular characteristics of early-onset CRC and high frequency of advanced stage at first diagnosis [33]. Although the causes of early-onset CRC remain unclear, earlier studies have explored disease pathogenesis, in which antibiotics use was considered a potential trigger [14], particularly in early-onset CRC [34]. This study reached a similar conclusion to our work by showing that exposure to antibiotics might lead to CRC, especially in those under 50 years. In addition, the use of penicillin and anti-anaerobes in patients with CRC was higher than that in healthy people in a descriptive study. Another study showed that oral antibiotics, especially ampicillin/amoxicillin, increased the risk of colon cancer (OR = 1.09, $95\%$ CI 1.05–1.13) [35]. In this study, gradient differences in the association between the incidence of early-onset CRC and antimicrobial use in different age groups were found, possibly because the intestinal flora of patients under the age of 50 may show different changes under the stimulation of antibiotics. In addition, the correlation has gradually weakened in recent years, possibly due to the emergence of more factors that can interfere with bacterial diversity. Among the potential risk factors, obesity, enteritis, and diabetes need to be paid more attention. The global prevalence of age-standardized incidence of obesity increased from $0.7\%$ in 1975 to $5.6\%$ in 2016 [36], which may also be other latent risk factors for early-onset CRC [37]. An environment of chronic tissue inflammation has been associated with malignant transformations, while diagnosing Crohn’s disease before 40 years old could increase the risk of CRC and even death [38]. Diabetes increases the risk of CRC, especially early-onset CRC [39, 40]. In addition, among diabetic patients, those not taking diabetes medications are more likely to develop adenomas than those taking medications [41]. In our study, the risk proportion of fasting hyperglycemia has sharply increased by $41.8\%$ globally in the past 3 decades (Figure S2 in the supplementary material). Especially in high SDI regions, high blood glucose accounted for $9.6\%$ of all risk factors, and there was a clear rise in trend in recent years. In addition, excess intake of high-fat foods containing a large amount of saturated fatty acids could easily lead to an imbalance in the gut microbiome, promoting the production of carcinogens and the occurrence of CRC [42]. On the other hand, reduced exercise time and increased sedentary sitting affect telomere length and have profound effects on gut microbiota, which could increase the risk of CRC [43]. The incidence of late-onset CRC increased rapidly in developing countries, which was associated with local economic growth. With regional economic growth, diets have also undergone substantial changes. This study focused on the changing relationship between meat consumption and ASIR of CRC. In Asian developing countries, represented by China and Vietnam, with rapid economic growth in the last 30 years, meat consumption and the ASIR of CRC have shown rapid and synchronized growth. Contrastingly, in developed countries like the United States and Australia, meat consumption has not changed significantly in the past 3 decades with a steady HDI, and a weak correlation was found between CRC incidence and meat consumption in these regions. This finding shows that regional economic growth and improvements in HDI may significantly affect CRC incidence through changes in the dietary structure. Moreover, studies have reported that a whole grain diet could reduce the risk of gastrointestinal cancer [44], and daily intake of dietary fiber up to 10 grams can reduce the risk of CRC by $10\%$ [45]. This is because phenolic compounds from microbial degradation from whole-grain diet catabolism can prevent CRC [46]. In addition, evidence suggests that a daily intake of dietary calcium up to 200 mg can reduce the risk of CRC by $6\%$ [42]. Smoking and alcohol were still key risk factors, and their proportions were related to the implementation of tobacco and alcohol policies in different regions (Supplementary Figure S3). For example, Indonesia has the world’s largest tobacco market. The proportion of Indonesian men who smoked accounted for about $76.2\%$ in 2015 and was still rising, leading to a sharp increase in the burden of tobacco-related diseases [47]. As a result, anti-smoking policies were likely to effectively reduce the burden of CRC. ## Conclusion In conclusion, this study found a geographical heterogeneity in the disease burden of CRC. The HDI in 2019 was exponentially and positively correlated with ASIR in all countries, and HDI was nonlinearly associated with the EAPC of ASR, demonstrating an inverted V-shaped relationship. Affected by HDI, changes in dietary patterns, such as meat consumption, were influenced by the economic level and strongly correlated with the incidence of CRC. In addition, antibiotic use has a potential effect on the occurrence of CRC, especially for early-onset CRC. Trends in the ASIR on early-onset CRC across different age groups, especially for young adults in high- and high-middle SDI regions, showed a fast-growing trend. Meanwhile, the mortality and DALYs of late-onset CRC were growing fast in the middle and middle-low SDI regions. Therefore, CRC prevention programs in developing countries should not only popularize screening and control the consumption of tobacco and alcohol but also strictly control meat consumption and antibiotic use. Specifically, countries in low SDI regions should rationally regulate the use of antibiotics, reduce the abuse of antibiotics, strengthen the control of the tobacco and alcohol consumption market, and popularize screening for middle-aged and elderly persons. Meanwhile, developed countries should advocate for self-testing with fecal occult blood kits and hospital visits among young people, especially those at high risk for early-onset CRC. ## Strengths and limitations of this study The present study used the 2019 GBD database to describe the incidence, mortality, DALYs, and risk factors for CRC in various countries and regions worldwide. The limitations of this study are mainly related to the limitations of the 2019 GBD database itself, which is generated using an algorithm based on existing data in every country and depends largely on the quality and quantity of data. Therefore, the data of countries at low levels of development were scarce and of low quality, which is a significant limitation. In addition, due to the lag in data reporting, our estimates may not include recent changes in health data across regions. Disease risk factors are numerous and unclear, especially early-onset colon cancer. This study did not explore all the risk factors for CRC, which require further research. ## Data availability statement The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material. ## Author contributions Project development: S-QF and Z-YM. Data collection: Y-HZ and Y-XL. Data analysis: W-XS and X-BS. Manuscript writing/editing: L-BL and L-YW. Language polishing and manuscript revision: D-MC. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc.2023.1102673/full#supplementary-material ## References 1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin (2021) 71. DOI: 10.3322/caac.21660 2. Keum NN, Giovannucci E. Global burden of colorectal cancer: Emerging trends, risk factors and prevention strategies. Nat Rev Gastroenterol Hepatol (2019) 16. DOI: 10.1038/s41575-019-0189-8 3. Safiri S, Sepanlou SG, Ikuta KS, Bisignano C, Salimzadeh H, Delavari A. The global, regional, and national burden of colorectal cancer and its attributable risk factors in 195 countries and territories, 1990–2017: A systematic analysis for the global burden of disease study 2017. Lancet Gastroenterol Hepatol (2019) 4. DOI: 10.1016/S2468-1253(19)30345-0 4. Wong MCS, Huang J, Lok V, Wang J, Fung F, Ding H. Differences in incidence and mortality trends of colorectal cancer worldwide based on sex, age, and anatomic location. Clin Gastroenterol Hepatol (2021) 19 955-966.e61. DOI: 10.1016/j.cgh.2020.02.026 5. Seitz HK, Stickel F. Molecular mechanisms of alcohol-mediated carcinogenesis. Nat Rev Cancer (2007) 7 599-612. DOI: 10.1038/nrc2191 6. Bailey CE, Hu CY, You YN, Bednarski BK, Rodriguez-Bigas MA, Skibber JM. Increasing disparities in the age-related incidences of colon and rectal cancers in the united states, 1975-2010. JAMA Surg (2015) 150 17-22. DOI: 10.1001/jamasurg.2014.1756 7. Akimoto N, Ugai T, Zhong R, Hamada T, Fujiyoshi K, Giannakis M. Rising incidence of early-onset colorectal cancer — a call to action. Nat Rev Clin Oncol (2021) 18. DOI: 10.1038/s41571-020-00445-1 8. Saad El Din K, Loree JM, Sayre EC, Gill S, Brown CJ, Dau H. Trends in the epidemiology of young-onset colorectal cancer: A worldwide systematic review. BMC Cancer (2020) 20 1-14. DOI: 10.1186/s12885-020-06766-9 9. Vuik FER, Nieuwenburg SAV, Bardou M, Lansdorp-Vogelaar I, Dinis-Ribeiro M, Bento MJ. Increasing incidence of colorectal cancer in young adults in Europe over the last 25 years. Gut (2019) 68. DOI: 10.1136/gutjnl-2018-317592 10. Arnold M, Sierra MS, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global patterns and trends in colorectal cancer incidence and mortality. Gut (2017) 66. DOI: 10.1136/gutjnl-2015-310912 11. Flor LS, Reitsma MB, Gupta V, Ng M, Gakidou E. The effects of tobacco control policies on global smoking prevalence. Nat Med (2021) 27. DOI: 10.1038/s41591-020-01210-8 12. Rumgay H, Shield K, Charvat H, Ferrari P, Sornpaisarn B, Obot I. Global burden of cancer in 2020 attributable to alcohol consumption: A population-based study. Lancet Oncol (2021) 22. DOI: 10.1016/S1470-2045(21)00279-5 13. Alattas M, Ross CS, Henehan ER, Naimi TS. Alcohol policies and alcohol-attributable cancer mortality in U.S. states. Chem Biol Interact (2020) 315. DOI: 10.1016/j.cbi.2019.108885 14. Cao Y, Wu K, Mehta R, Drew DA, Song M, Lochhead P. Long-term use of antibiotics and risk of colorectal adenoma. Gut (2017) 67. DOI: 10.1136/gutjnl-2016-313413 15. Transforming our world: The 2030 agenda for sustainable development. (2015). DOI: 10.1201/b20466-7 16. Cao B, Bray F, Ilbawi A, Soerjomataram I. Effect on longevity of one-third reduction in premature mortality from non-communicable diseases by 2030: A global analysis of the sustainable development goal health target. Lancet Glob Heal (2018) 6. DOI: 10.1016/S2214-109X(18)30411-X 17. Sharma R, Abbasi-Kangevari M, Abd-Rabu R, Abidi H, Abu-Gharbieh E, Acuna JM. Global, regional, and national burden of colorectal cancer and its risk factors, 1990–2019: a systematic analysis for the global burden of disease study 2019. Lancet Gastroenterol Hepatol (2022) 1253 1-21. DOI: 10.1016/s2468-1253(22)00044-9 18. Ou Z, Yu D, Liang Y, He W, Li Y, Zhang M. Analysis of the global burden of disease study highlights the trends in death and disability-adjusted life years of leukemia from 1990 to 2017. Cancer Commun (2020) 40 598-610. DOI: 10.1002/cac2.12094 19. Magliano DJ, Chen L, Islam RM, Carstensen B, Gregg EW, Pavkov ME. Trends in the incidence of diagnosed diabetes: A multicountry analysis of aggregate data from 22 million diagnoses in high-income and middle-income settings. Lancet Diabetes Endocrinol (2021) 9. DOI: 10.1016/S2213-8587(20)30402-2 20. Fidler MM, Bray F, Vaccarella S, Soerjomataram I. Assessing global transitions in human development and colorectal cancer incidence. Int J Cancer (2017) 140. DOI: 10.1002/ijc.30686 21. Tran BT, Choi KS, Nguyen TX, Sohn DK, Kim SY, Suh JK. The direct and indirect costs of colorectal cancer in vietnam: An economic analysis from a social perspective. Int J Environ Res Public Health (2021) 18 1-11. DOI: 10.3390/ijerph18010012 22. Liu H, Hu T. Evaluating the long-term care insurance policy from medical expenses and health security equity perspective: Evidence from China. Arch Public Heal (2022) 80 1-15. DOI: 10.1186/s13690-021-00761-7 23. Zhang Y, Wushouer H, Han S, Fu M, Guan X, Shi L. The impacts of government reimbursement negotiation on targeted anticancer medication price, volume and spending in China. BMJ Glob Heal (2021) 6 1-11. DOI: 10.1136/bmjgh-2021-006196 24. Cardoso R, Guo F, Heisser T, Hackl M, Ihle P, De Schutter H. Colorectal cancer incidence , mortality , and stage distribution in European countries in the colorectal cancer screening era : an international population-based study. Lancet Oncol (2021) 22. DOI: 10.1016/S1470-2045(21)00199-6 25. Burger H, Wyrley-Birch B, Joubert N, Trauernicht CJ, Valentim JM, Groll J. Bridging the radiotherapy education gap in Africa: Lessons learnt from the cape town access to care training programme over the past 5 years (2015–2019). J Cancer Educ (2021) 37. DOI: 10.1007/s13187-021-02010-5 26. Morel T, Maher D, Nyirenda T, Olesen OF. Strengthening health research capacity in sub-Saharan Africa: Mapping the 2012-2017 landscape of externally funded international postgraduate training at institutions in the region. Global Health (2018) 14 1-10. DOI: 10.1186/s12992-018-0395-0 27. Lew JB, St John DJB, Xu XM, Greuter MJE, Caruana M, Cenin DR. Long-term evaluation of benefits, harms, and cost-effectiveness of the national bowel cancer screening program in Australia: A modelling study. Lancet Public Heal (2017) 2. DOI: 10.1016/S2468-2667(17)30105-6 28. Cox B, Sneyd MJ, Hingston G, McBride D, Bagshaw PF. Enhancing bowel screening: Preventing colorectal cancer by flexible sigmoidoscopy in new Zealand. Public Health (2020) 179 27-37. DOI: 10.1016/j.puhe.2019.09.021 29. Fu W-P, Kam M-H, Ling W-M, Ong S-F, Suzannah N, Eu K-W. Screening for colorectal cancer using a quantitative immunochemical faecal occult blood test: A feasibility study in an Asian population. Tech Coloproctol (2009) 13. DOI: 10.1007/s10151-009-0515-1 30. Tan WS, Tang CL, Koo WH. Opportunistic screening for colorectal neoplasia in Singapore using faecal immunochemical occult blood test. Singapore Med J (2013) 54. DOI: 10.11622/smedj.2013077 31. Shaukat A, Levin TR. Current and future colorectal cancer screening strategies. Nat Rev Gastroenterol Hepatol (2022) 7. DOI: 10.1038/s41575-022-00612-y 32. Wu CWK, Lui RN. Early-onset colorectal cancer: Current insights and future directions. World J Gastrointest Oncol (2022) 14. DOI: 10.4251/wjgo.v14.i1.230 33. Siegel RL, Jakubowski CD, Fedewa SA, Davis A, SA N. Colorectal cancer in the young: Epidemiology, prevention, management. Am Soc Clin Oncol Educ B (2020) 40 1-14. DOI: 10.1002/ijc.31314 34. McDowell R, Perrott S, Murchie P, Cardwell C, Hughes C, Samuel L. Oral antibiotic use and early-onset colorectal cancer: Findings from a case-control study using a national clinical database. Br J Cancer (2022) 126. DOI: 10.1038/s41416-021-01665-7 35. Zhang J, Haines C, Watson AJM, Hart AR, Platt MJ, Pardoll DM. Oral antibiotic use and risk of colorectal cancer in the united kingdom, 1989-2012: A matched case-control study. Gut (2019) 68. DOI: 10.1136/gutjnl-2019-318593 36. Bentham J, Di Cesare M, Bilano V, Bixby H, Zhou B, Stevens GA. Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: A pooled analysis of 2416 population-based measurement studies in 128·9 million children, adolescents, and adults. Lancet (2017) 390. DOI: 10.1016/S0140-6736(17)32129-3 37. Li H, Boakye D, Chen X, Jansen L, Chang-Claude J, Hoffmeister M. Associations of body mass index at different ages with early-onset colorectal cancer. Gastroenterology (2021) 1–10. DOI: 10.1053/j.gastro.2021.12.239 38. Olén O, Erichsen R, Sachs MC, Pedersen L, Halfvarson J, Askling J. Colorectal cancer in crohn’s disease: A Scandinavian population-based cohort study. Lancet Gastroenterol Hepatol (2020) 5. DOI: 10.1016/S2468-1253(20)30005-4 39. Ottaviano LF, Li X, Murray M, Frye JT, Lung BE, Zhang YY. Type 2 diabetes impacts colorectal adenoma detection in screening colonoscopy. Sci Rep (2020) 10 1-7. DOI: 10.1038/s41598-020-64344-2 40. Khan UA, Fallah M. Personal history of diabetes as important as family history of colorectal cancer for risk of colorectal cancer: A nationwide cohort study. Am J Gastroenterol (2021) 116. DOI: 10.14309/ajg.0000000000000929 41. Yang WT, Yang HJ, Zhou JG, Liu J. Relationship between metformin therapy and risk of colorectal cancer in patients with diabetes mellitus: A meta-analysis. Int J Colorect Dis (2020) 35. DOI: 10.1007/s00384-020-03704-w 42. Zhou E, Rifkin S. Colorectal cancer and diet: Risk versus prevention, is diet an intervention. Gastroenterol Clin North Am (2021) 50. DOI: 10.1016/j.gtc.2020.10.012 43. Kerr J, Anderson C, Lippman SM. Physical activity, sedentary behaviour, diet, and cancer: An update and emerging new evidence. Lancet Oncol (2017) 18. DOI: 10.1016/S1470-2045(17)30411-4 44. Zhang XF, Wang XK, Tang YJ, Guan XX, Guo Y, Fan JM. Association of whole grains intake and the risk of digestive tract cancer: A systematic review and meta-analysis. Nutr J (2020) 19 1-14. DOI: 10.1186/s12937-020-00556-6 45. Aune D, Chan DSM, Lau R, Vieira R, Greenwood DC, Kampman E. Dietary fibre, whole grains, and risk of colorectal cancer: Systematic review and dose-response meta-analysis of prospective studies. BMJ (2011) 343 d6617. DOI: 10.1136/bmj.d6617 46. Guo H, Wu H, Sajid A, Li Z. Whole grain cereals: The potential roles of functional components in human health. Crit Rev Food Sci Nutr (2021) 62. DOI: 10.1080/10408398.2021.1928596 47. Holipah H, Sulistomo HW, Maharani A. Tobacco smoking and risk of all-cause mortality in Indonesia. PloS One (2020) 15. DOI: 10.1371/journal.pone.0242558
--- title: Effects of a mobile health nutrition intervention on dietary intake in children who have autism spectrum disorder authors: - Tanja V. E. Kral - Lauren O’Malley - Kelsey Johnson - Teresa Benvenuti - Jesse Chittams - Ryan J. Quinn - J. Graham Thomas - Jennifer A. Pinto-Martin - Susan E. Levy - Emily S. Kuschner journal: Frontiers in Pediatrics year: 2023 pmcid: PMC9975727 doi: 10.3389/fped.2023.1100436 license: CC BY 4.0 --- # Effects of a mobile health nutrition intervention on dietary intake in children who have autism spectrum disorder ## Abstract ### Background Children who have Autism Spectrum Disorder (ASD) show preferences for processed foods, such as salty and sugary snacks (SSS) and sugar-sweetened beverages (SSB), while healthier foods, such as fruits and vegetables (FV), are consumed less. Innovative tools are needed that can efficiently disseminate evidence-based interventions and engage autistic children to improve their diet. ### Aim The aim of this 3-month randomized trial was to test the initial efficacy of a mobile health (mHealth) nutrition intervention on changing consumption of targeted healthy (FV) and less healthy foods/beverages (SSS, SSB) in children who have ASD, ages 6–10, who were picky eaters. ### Methods Thirty-eight parent-child dyads were randomly assigned to either an intervention (technology) group or a wait list control (education) group. The intervention included behavioral skills training, a high level of personalization for dietary goals, and involved parents as “agents of change.” Parents in the education group received general nutrition education and the dietary goals but did not receive skills training. Children's intake was assessed at baseline and at 3 months using 24-hour dietary recalls. ### Results While there were no significant group-by-time interactions ($P \leq 0.25$) for any of the primary outcomes, we found a significant main effect of time for FV intake ($$P \leq 0.04$$) indicating that both groups consumed more FV at 3 months (2.58 ± 0.30 servings/day) than at baseline (2.17 ± 0.28 servings/day; $$P \leq 0.03$$). Children in the intervention group who consumed few FV at baseline and showed high engagement with the technology increased their FV intake by 1.5 servings/day ($P \leq 0.01$). Children's taste/smell sensitivity significantly predicted their FV intake ($$P \leq 0.0446$$); for each unit of lower taste/smell sensitivity (indicating greater sensory processing abnormalities), FV intake increased by 0.13 ± 0.1 servings/day. ### Discussion This mHealth intervention did not yield significant between-group differences for changing consumption of targeted foods/beverages. Only children who consumed few FV at baseline and highly engaged with the technology increased their FV intake at 3 months. Future research should test additional strategies to expand the intervention's impact on a wider range of foods while also reaching a broader group of children who have ASD. This trial was registered at clinicaltrials.gov as NCT03424811. Clinical Trial Registration: This study was registered at clinicaltrials.gov as NCT03424811. ## Introduction Children who have Autism Spectrum Disorder (ASD)1 are five times more likely to have mealtime challenges (4–6) and be picky eaters (7–10), which in part has been attributed to restrictive and ritualistic behaviors and heightened sensory sensitivity [7]. Picky eating among autistic youth has been shown to not be associated with a lack of appetite as many parents report that their children show a healthy appetite for the foods that they like [7], which often include highly processed foods (9–13). Consumption patterns that favor energy-dense, nutrient-poor foods and beverages in lieu of healthier options put children on the autism spectrum at an increased risk for excess weight gain. In fact, children who have ASD show a four-fold increased risk for overweight and obesity [14] and are over three times more likely to develop metabolic syndrome [15] than children with neurotypical development. Children on the autism spectrum have been reported to show preferences for highly processed foods, such as salty and sugary snacks (SSS) and sugar-sweetened beverages (SSB), while healthier foods, such as fruits and vegetables (FV), are consumed less often (9–13). There is strong evidence for an independent role of SSB intake (16–18) and increased snacking frequency (19–22) in the promotion of obesity in youth. Increased FV consumption has been shown to reduce dietary energy density [23, 24], moderate energy intake [25, 26], and play an important role in the prevention and treatment of childhood obesity [27, 28]. Reducing consumption of SSB and SSS and increasing consumption of FV in children are therefore considered important targets for intervention. The pronounced core impairments of social and communication skills and presence of restricted and repetitive behaviors in children who have ASD complicate traditional treatment options for healthy eating available to children with neurotypical development. While behavioral interventions are considered effective treatments overall and, in particular, for selective eating in children who have ASD (29–31), it is often behavioral therapists, and not parents or caregivers, who work with children one-on-one in highly structured clinical settings, which are highly resource dependent and less generalizable to everyday life without structured mealtimes. Parents are often removed from the treatment initially and then trained to re-enter mealtimes as discharge approaches [see Sharp and colleagues [31] for a review of the most common intensive multidisciplinary interventions]. Evidence suggests that treatments that involve family members as treatment providers are effective and may be more applicable to the home environment [32, 33]. A small number of studies exist aimed at teaching caregivers, during in-person sessions, behavior modification strategies which they can use to address feeding problems in their children. While some of these interventions indicated a high degree of social validity and caregiver satisfaction, many showed limited success in changing child mealtime behaviors or dietary variety, included small sample sizes, and suffered from high attrition rates (34–38). Barriers to parent-directed interventions include high time commitment and transportation costs, shortage of trained professionals, and lack of childcare [39]. New and innovative tools are needed that are scalable, can efficiently disseminate evidence-based parent interventions, and effectively engage children who have ASD. The use of mobile technology is rapidly increasing in children, including for those on the autism spectrum, across all ages. Children who have ASD, in particular, are engaging with mobile devices on a daily basis and technology has been shown to help with their learning and communication by providing structure and dependability and opportunities for visual learning [40, 41]. Research has shown that multisensory interactions and the ability to individualize instructions are some of the features that can assist children who have ASD when working with technology [42] and smart technologies are effective tools for improving social, functional, and communications skills [40, 43]. We aimed to harness the lure of technology by developing an innovative, evidence-based mobile health (mHealth) nutrition intervention to teach children who have ASD about healthy eating and to motivate them to make healthy food choices in their daily lives. mHealth technologies provide opportunities for remote coaching and skills training, which has proven to be an effective and efficient training tool for parents of autistic youth [44, 45]. Core behavior change strategies that have been used in family-based nutrition and childhood obesity prevention research in children with neurotypical development include the specification of target behaviors, self-monitoring, goal setting, stimulus control, positive parenting strategies, and promotion of self-efficacy and self-management skills [46, 47]. Parents and caregivers act as important “agents of change” for promoting healthy eating among their children. There are parallels in the application of similar core behavior strategies in providing supports for autistic children [48, 49] and many parents are already being trained in these behavior change strategies to address behavioral and communication challenges their children experience [50, 51]. Our study is among the first to empirically test the efficacy of incorporating these same core behavior change strategies into a mHealth nutrition intervention to affect changes in the intake of targeted healthy and less healthy foods and beverages in children on the autism spectrum who are picky eaters. The aim of this 3-month randomized controlled trial was to test the initial efficacy of a mHealth nutrition intervention on changing consumption of targeted healthy (FV) and less healthy foods (SSS) and beverages (SSB) in children on the autism spectrum who were picky eaters. We hypothesized that, by the end of the intervention, children in the technology (intervention) group, but not children in the education (wait list control) group, would show a significant increase (expressed as % change from baseline) in intake of FV and a decrease in calories consumed from SSS and SSB. ## Study design In this exploratory 3-month trial, parent-child dyads were randomly assigned to either a mHealth intervention group (technology group) or a wait list control group (education group) using a randomized block design to produce groups that were comparable in weight status. Parent trainings in both groups were matched for in-person contact time. All study visits were conducted in families' homes to reduce burden and increase child comfort. The study protocol, including screening and recruitment procedures, was approved by the Institutional Review Boards of the University of Pennsylvania and The Children's Hospital of Philadelphia (CHOP). Parents and children were asked to provide voluntary informed consent (parents) and assent (children) to participate in the study by signing the consent and assent forms. During the informed consent/assent process, families were informed that they will be randomized into one of two groups (technology or education group) for the duration of the study. Families were also informed that if they were assigned to the education group, they would be given access to the mobile app for free after their participation in the study ended. ## Study population Participants for this study included boys and girls with ASD, ages 6–10 years, and their parent or legal guardian (referred to as “parent” for ease of use). This age range is consistent with ages that children would engage in similar technology. Parents had to be the children's primary caregiver (i.e., person responsible for grocery shopping and/or feeding). For this exploratory trial, we recruited a well-defined, homogenous group of children to determine the efficacy of this intervention. Only children without significant intellectual disability (see definition in Recruitment and Screening Process) were enrolled to increase the likelihood of comprehension and engagement with the technology. Children with a range in weight were included to explore if the intervention is equally effective in children with normal weight, overweight, and obesity. A power analysis was conducted (PASS software, Version 11, NCSS LLC, Kaysville, UT) for the primary aim using 3-month changes in intake of targeted healthy foods (FV) as primary outcome variable. The estimated mean (± SD) for baseline FV intake (2.57 ± 1.20 servings/day) was derived from our pilot work with children on the autism spectrum [52]. Based on this estimate, a sample size of 46 children with an attrition rate of $10\%$ yielded $80\%$ power to detect a statistically significant increase in FV intake of 1.1 servings/day (or $43\%$) in the intervention group relative to the control group at an alpha level of 0.05 based on a 2-sample t-test. The magnitude of this increase is comparable to that achieved in behavioral interventions in children with neurotypical development [28, 53] and represents ∼one-fifth of the Recommended Dietary Allowance (RDA) of FV for children of that age [54]. ## Recruitment and screening process Recruitment of study participants was carried out in collaboration with the CHOP Center for Autism Research (CAR). Our multi-pronged recruitment plan utilized the following strategies: (a) CAR's research study page of its website, social media accounts, and email listserv; (b) the CHOP Recruitment Enhancement Core, which leverages recruitment via the electronic medical record system; and (c) community-based organizations. Interested families completed a telephone screening. Parents provided information about their child's age, sex, height, weight, autism diagnosis, medical and developmental history, and medication use and provided verbal authorization per the Health Insurance Portability and Accountability Act (HIPAA) for the phone screening and to have their child's medical information and previous diagnostic evaluations reviewed. Parents were asked to complete the Picky Eating subscale of the Child Feeding Questionnaire (CFQ) [55], which consisted of the following items: [1] “My child's diet consists of only a few foods”; [2] “My child is unwilling to eat many of these foods that our family eats at mealtimes”; [3] “My child is fussy / picky about what she eats.” The items on this subscale have shown adequate internal consistency (α = 0.85). Parents were asked follow-up questions to confirm that a child's pickiness was related to the intervention's targeted healthy and less healthy foods to ensure that the intervention goals would be relevant. Picky eating was confirmed if parents endorsed at least two out of the three items on this subscale. To be included in the study, children had to be between ages 6 to 10 years; fluent in English; have an ASD diagnosis using the DSM-IV-TR or DSM-5 criteria [56, 57] and cognitive skills within the low average (or higher) range with IQ scores of ≥80 and comparable verbal ability; meet the definition of picky eater with pickiness related to the intervention's targeted healthy and less healthy foods; and have access to a mobile device. Medical records were reviewed to confirm documentation of ASD diagnosis by an expert clinician (i.e., developmental pediatrician, clinical psychologist) as well as cognitive and language ability at a level sufficient for comprehension of and engagement with the mHealth technology. When specific IQ or language assessment scores were not available in medical records, school records such as Individualized Education Program documents were reviewed by the expert clinician, and descriptions of skills across academic domains and educational goals were used as a proxy. Children were excluded from participation in the study if they had moderate-severe hearing/visual or motor impairment (e.g., were non-ambulatory); were taking antipsychotic medications which may be associated with uncontrolled eating; were on a special diet (e.g., gluten/casein-free diet); or had underweight [i.e., body mass index (BMI)-for-age <5th percentile]. ## Description of the behavioral intervention Parents and children in the technology group received the mHealth nutrition intervention plus training in behavior change strategies. The intervention incorporated core behavior change strategies that have been tested extensively in family-based behavior modification research, including obesity prevention trials [46], and were familiar to families with children who have ASD [58, 59]. The unique features of this mHealth intervention were that it [1] reinforced healthy food choices in autistic youth by using behavioral strategies tailored to the specific needs and learning styles of children on the autism spectrum (e.g., visual depictions, concrete descriptions with “scripts”, and routines for abstract concepts), [2] included a high level of personalization to align dietary goals with individual food preferences and sensory sensitivities, and [3] involved parents as “agents of change.” Specifically, children were reinforced for making healthy food choices while limiting less healthy food choices in their daily lives by earning points towards a prize. Targeted healthy foods included fresh, canned, and frozen FV. Parents were instructed to omit energy-dense toppings and sauces on FV. Targeted less healthy foods included SSS (e.g., all types of chips, popcorn, pretzels, party mixes, ice cream, candy, cookies, cakes/pies, sweet rolls, pastries) and SSB (e.g., sugar-sweetened sodas, fruit drinks/punches and fruit juices, sport drinks, and energy drinks). The intervention also included behavioral training for children via an interactive nutrition education game to facilitate healthier food choices. This involved a “Nutrition Ninja” virtual character, which was directed by the parent and interacted by the child to set and reinforce dietary goals. The inclusion of this virtual character who acted positively towards the child aimed to make the child feel comfortable with technology-mediated communication and offer support for performing the desired behavior, consistent with the Proteus effect (60–62). The goals were very prescriptive and were tailored to children's food preferences and sensory sensitivities by letting them customize dietary targets (e.g., add dip to a vegetable, eat a vegetable raw or cooked, have it served hot or cold). Even small goals, such as touching or smelling a novel food before tasting it, were encouraged. Children received frequent visual and personalized feedback and positive reinforcement to support their unique learning styles. The individual training components for parents and children are summarized in Table 1. Table 1 \| Training type \| Training components \| \| --- \| --- \| \| Parent training \| Parent training \| \| Behavioral skills training \| • Specifying target behaviors and goal setting• Self-monitoring• Stimulus control• Positive reinforcement• Self-efficacy \| \| Nutrition training \| • Explanation of nutritional goals and targeted foods and beveragesa• Training on how to present feeding opportunities during meal and snack times• Substitute unhealthy target foods and beverages for healthier options• Strategies to overcome resistance in children trying new foods using differential attention and positive reinforcement• Strategies for promoting children's intake of healthy target foods and limiting intake of unhealthy target foods and beverages daily for 3 months \| \| Technology training \| • Explanation of the layout and functions of the mHealth technology• Setting dietary goals, selecting rewards, and awarding points• Adding custom foods• Reviewing child's progress, goals, and rewards \| \| Child training \| Child training \| \| Behavioral skills training \| • Specifying target behaviors and goal setting• Positive reinforcement \| \| Nutrition training \| • Explanation of the health benefits of fruits and vegetables and undesirable properties of sugary drinks and salty and sugary snacks• Explanation of nutritional goals and targeted foods and beverages• Explanation of “Go” and “Whoa” foods and beverages \| \| Technology training \| • Explanation of the layout and functions of the mHealth technology• Setting dietary goals and selecting a reward• Learning how to play the educational Nutrition Ninja game• Communicating with the Nutrition Ninja• Reviewing their progress \| ## Description of the mHealth technology The technology-based intervention, a mobile application, was developed in collaboration with Skyless Game Studios, a Philadelphia-based software company. We used an iterative design process for prototype development, testing and refining of the intervention during which stakeholders (i.e., parents and caregivers of children with autism and children with autism) and technology developers were engaged and asked to provide continuous feedback on the functionality and acceptability of the intervention during the development process. During the baseline study visit, trained research staff assisted families with downloading and installing the mobile application onto their smartphones or tablets. They also provided families with a brief technology tutorial which explained to the parent how [1] to view, create, and edit rewards and goals; [2] view their child's progress and award points for eating and drinking goal foods/beverages; [3] add or edit custom foods; and [4] view and send messages to the child. The tutorial explained to the child how to [1] choose their goals; [2] select their reward; [3] send messages to the “Ninja” (i.e., parent) about a food they ate or ask for a food or drink for their next meal; [4] view their progress and awarded points; and [5] play the Nutrition Ninja educational game. Parents and children were also instructed to view a training module, which was built into the app and explained the nutritional targets of the intervention and provided behavioral skills training. Throughout the training, parents and children completed quizzes which reinforced the training content. Parents and children were encouraged to engage with the mHealth technology on a daily basis for 3 months to promote children's intake of the healthy target foods and limit their intake of the less healthy target foods and beverages. Examples of targeted healthy and less healthy target foods were incorporated in the mHealth technology. Figure 1 provides examples of screenshots from the mHealth intervention. The mHealth technology provided research staff with an activity log, which they accessed regularly. For families who showed no activity with the mHealth technology for some time, research staff followed up with them via email (after 2 weeks) or a phone call (after 3 weeks) to assess if they had any technical difficulties with the mobile app and reminded them of the goals of the study. Figure 1: Sample screenshots of the mobile health (mHealth) intervention including goal setting (panel A), parent behavioral skills training (panel B), and child nutrition training (panel C). ## Education control group Parents in the Education group received a printed handout which provided general education about healthy eating and explained the nutritional goals and targeted foods and beverages but did not provide any skills training. Parents were encouraged to promote children's intake of healthy target foods and limit intake of unhealthy target foods daily for 3 months. Examples of targeted healthy and less healthy target foods were included on the handout. Families were offered access to the mHealth intervention after they completed the study. Families in both groups were instructed to try increasing their children's intake of FV and decreasing their intake of SSS and SSB for the 3-month study duration. They were not instructed on what to purchase and make available to their children specifically because we wanted to give families the flexibility to tailor the food and beverage choices to their children's preferences and sensory sensitivities. ## Assessment of child dietary intake Children's intake was assessed at baseline (before the first study visit) and at the end of the 3-month intervention using the telephone-based 24-hour dietary recall method; the gold standard for self-reported intake [63]. Each assessment consisted of three unannounced recalls (two weekdays, one weekend day), conducted by research dietitians from the CHOP and Penn Dietary Assessment Unit at the Center for Human Phenomic Science. Dietitians were blinded to families' group assignment. During each recall, they asked parents, with help from their child, to describe all foods and beverages consumed by their children during the prior 24 h and to provide detailed information about portion sizes and preparation method. Individual food items were manually coded to derive SSS, SSB and water intake. Data were analyzed for the primary outcomes measures which included daily FV intake (servings/day) and calories consumed per day from SSB and SSS using the University of Minnesota Nutrition Coordinating Center's Food and Nutrient Database. All data were averaged across the three days. ## Assessment of child weight status During both study visits (i.e., at baseline and at the end of the intervention), children had their heights and weights measured using a digital scale (SECA 876, Chino, CA) and portable stadiometer (SECA 217, Chino, CA). Measurements were taken in triplicate by a trained staff member in children's homes with children wearing light clothing (no shoes). Child age- and sex-specific BMI percentiles and z-scores were calculated using the CDC Growth Charts 2000 [64]. Children were classified as normal-weight (BMI-for-age 5–84th percentile), overweight (BMI-for-age 85–94th percentile), or obese (BMI-for-age ≥95th percentile) [65]. ## Demographic questionnaire During the baseline home visit, parents completed a demographic questionnaire which included questions about their age, race/ethnicity, marital status, education, household income, and food security status. Families' food security status was assessed using the 6-item short form of the U.S. Household Food Security survey module [66]. The survey's raw score, which is the sum of affirmative responses to the six survey questions, describes four levels of food security: high food security (i.e., no indications of food access problems or limitations); marginal food security (i.e., 1–2 indications of food access problems such as anxiety over food sufficiency or shortage of food in the house); low food security (i.e., reports of reduced quality, variety, or desirability of diet); and very low food security (i.e., reports of multiple indications of disrupted eating patterns and reduced food intake) [67]. This short form of the U.S. Household Food Security survey module has shown to have acceptable conceptual validity, specificity ($99.5\%$) and sensitivity ($85.9\%$) for determination of overall food insecurity for households with children [66] as well as good internal consistency (Cronbach's alpha = 0.86) [68]. ## Sensory profile Parents were also asked to complete the 38-item Short Form Sensory Profile [69], which measures children's sensory processing. For this study, we limited the analysis to the taste/smell sensitivity domain only, which is comprised of the following items: “Avoids certain tastes or food smells that are typically part of children's diets,” “Will only eat certain tastes,” “Limits self to particular food textures/temperatures,” and “Picky eater, especially regarding food textures.” Parents were asked to indicate, on a 5-point scale ranging from “always” [1] to “never” [5], the frequency with which their child responds to these sensory experiences. Using a reverse scoring system, lower scores corresponded to more frequent child behavioral responses. Children were categorized into “Typical Performance,” “Probable Difference,” or “Definite Difference” in oral sensory sensitivity based on the classifications specified by Dunn [70]. Construct validity has been established for the Short Form of the Sensory Profile [71] and the tool has been shown to have acceptable reliability and excellent validity [72]. ## Statistical analysis Data were analyzed using the statistical software SAS (Version 9.4; SAS Institute Inc, Cary, NC). The Shapiro-Wilk test was used in combination with histograms and summary statistics to confirm normal distribution of all outcome variables. Outcome variables included intake of fruits and vegetables (servings/day; with and without French Fries); vegetables (servings/day; with and without French Fries); fruit (servings/day); sweet and savory snacks (kcal/day); sweet snacks (kcal/day); savory snacks (kcal/day); sugar-sweetened beverages (kcal/day and fl oz/day); and water (fl oz/day). Participants' baseline demographic, clinical and anthropometric characteristics and dietary intake were compared between the technology and education group using Chi-Square and t-tests for categorical and continuous variables, respectively. To test the primary aim, an intent-to-treat approach was used. We constructed separate linear mixed effects models for each outcome measure with an unstructured covariance matrix to account for within-subject variance. The effects of time (baseline, 3-month follow-up) and treatment group were included in all models. A time-by-group interaction was included to assess changes in outcomes over time by group. Furthermore, we created adjusted models which controlled for children's age, BMI z-score, and level of taste/smell sensitivity. Secondary analyses assessed the intervention utilizing a dose-response approach. We explored the extent to which level of engagement with the technology impacted participants' dietary outcomes. In this exploratory analysis, we assessed both planned behavior (i.e., completion of nutrition training and setting goals) and performed behavior (i.e., number of points earned for meeting goals). We then categorized children into those with a high planned engagement (i.e., children who completed the nutrition training and set at least one goal) and those with low or no engagement (i.e., children who completed the nutrition training but did not set any goals or children who neither completed the nutrition training nor set any goals). In terms of performed behavior, we categorized children's engagement with the technology using a tertile split (low/no, medium, high). Analyses tested the effect of children's level of engagement with the technology and the time-by-engagement interaction on dietary outcomes. Analyses were conducted with participants in the education (control) group included in the statistical model; their level of engagement was set to “low/no engagement” because they did not have access to the technology. Additional exploratory analyses examined the effects of the intervention on changes in dietary outcomes with the sample stratified into low or high consumers at baseline based on a median split in intake. Stratified analyses were conducted for analyses of both the primary predictor (intervention group) and secondary predictor (engagement level) with and without controlling for participants' age, BMI z-score, and level of taste/smell sensitivity. All analyses were evaluated at the alpha level of 0.05 and are considered exploratory. Results are presented as model-based means (± standard error). ## Child and parent characteristics Child and parent demographic and anthropometric characteristics are shown in Table 2. The majority of children were male ($95\%$), and many were White ($68\%$); $13\%$ were Hispanic. Half of the children had either overweight ($16\%$) or obesity ($34\%$), respectively. Among parents, $84\%$ were married; $42\%$ had a college degree; and $61\%$ had household incomes greater than $100,000. Groups did not differ significantly in any demographic or anthropometric characteristics ($P \leq 0.14$), nor in their baseline intake of the target foods ($P \leq 0.099$). Table 2 \| Characteristic \| Technology group (N = 19) \| Nutrition education group (N = 19) \| P-value \| \| --- \| --- \| --- \| --- \| \| Children \| Children \| Children \| Children \| \| Age (years), mean ± SD \| 8.9 ± 1.2 \| 8.4 ± 1.4 \| 0.26 \| \| Sex, male/female, n (%) \| 18 (94.7%)/1 (5.3%) \| 18 (94.7%)/1 (5.3%) \| 1.00 \| \| Race, n (%) \| Race, n (%) \| Race, n (%) \| Race, n (%) \| \| Asian \| 0 \| 2 (10.5%) \| 0.16 \| \| Black or African American \| 4 (21.1%) \| 1 (5.3%) \| \| \| White \| 11 (57.9%) \| 15 (78.9%) \| \| \| More than one race \| 3 (15.8%) \| 1 (5.3%) \| \| \| Unknown \| 1 (5.3%) \| 0 \| \| \| Ethnicity, n (%) \| Ethnicity, n (%) \| Ethnicity, n (%) \| Ethnicity, n (%) \| \| Hispanic or latino \| 4 (21.1%) \| 1 (5.3%) \| 0.20 \| \| Not hispanic or latino \| 12 (63.1%) \| 17 (89.5%) \| \| \| Unknown \| 3 (15.8%) \| 1 (5.3%) \| \| \| Weight status, n (%) \| Weight status, n (%) \| Weight status, n (%) \| Weight status, n (%) \| \| BMI z-score, mean ± SD \| 0.8 ± 1.36 \| 0.6 ± 1.35 \| 0.64 \| \| BMI-for-age percentile, mean ± SD \| 68.1 ± 34.2 \| 63.1 ± 37.2 \| 0.67 \| \| Has normal weight \| 10 (52.6%) \| 9 (47.4%) \| 0.82 \| \| Has overweight \| 2 (10.5%) \| 4 (21.1%) \| \| \| Has obesity \| 7 (36.8%) \| 6 (31.6%) \| \| \| Sensory profile (score), mean ± SD \| Sensory profile (score), mean ± SD \| Sensory profile (score), mean ± SD \| Sensory profile (score), mean ± SD \| \| Taste/Smell Sensitivity \| 9.7 ± 4.9 \| 11.9 ± 4.0 \| 0.14 \| \| Taste/smell sensitivity classification, n (%) \| Taste/smell sensitivity classification, n (%) \| Taste/smell sensitivity classification, n (%) \| Taste/smell sensitivity classification, n (%) \| \| Typical performance \| 4 (21.1%) \| 5 (26.3%) \| 0.48 \| \| Probably difference \| 3 (15.8%) \| 6 (31.6%) \| \| \| Definite difference \| 12 (63.2%) \| 8 (42.1%) \| \| \| Parents \| Parents \| Parents \| Parents \| \| Academic degree, n (%) \| Academic degree, n (%) \| Academic degree, n (%) \| Academic degree, n (%) \| \| High school \| 3 (15.8%) \| 4 (21.1%) \| 1.00 \| \| College \| 8 (42.1%) \| 8 (42.1%) \| \| \| Master’s \| 6 (31.6%) \| 6 (31.6%) \| \| \| Doctorate \| 2 (10.5%) \| 1 (5.3%) \| \| \| Household income, n (%) \| Household income, n (%) \| Household income, n (%) \| Household income, n (%) \| \| <$50,000 \| 2 (10.5%) \| 4 (21.1%) \| 0.73 \| \| $50,000–$100,000 \| 4 (21.1%) \| 5 (26.3%) \| \| \| $100,000–$150,000 \| 5 (26.3%) \| 5 (26.3%) \| \| \| >$150,000 \| 8 (42.1%) \| 5 (26.3%) \| \| \| Marital status, n (%) \| Marital status, n (%) \| Marital status, n (%) \| Marital status, n (%) \| \| Single \| 2 (10.5%) \| 0 \| 0.54 \| \| Married, remarried \| 15 (78.9%) \| 17 (89.5%) \| \| \| Divorced, separated, widowed \| 2 (10.5%) \| 2 (10.5%) \| \| \| Food security status, n (%) \| Food security status, n (%) \| Food security status, n (%) \| Food security status, n (%) \| \| High/marginal food security \| 18 (94.7%) \| 16 (84.2%) \| 0.74 \| \| Low food security \| 0 \| 2 (10.5%) \| \| \| Very low food security \| 1 (5.3%) \| 1 (5.3%) \| \| ## Efficacy of mHealth intervention on changing target dietary behaviors Table 3 shows participants' mean intake of FV, SSS, and SSB (primary dietary outcomes) at baseline and at 3 months (end of intervention) by group. There were no significant group-by-time interactions ($P \leq 0.25$) for any of the primary dietary outcomes. There were also no significant main effects of group for intake of FV, SSS, SSB ($P \leq 0.15$) or time for intake SSS and SSB ($P \leq 0.83$). We did find a statistically significant main effect of time for FV intake ($$P \leq 0.04$$) indicating that all participants, irrespective of random assignment, on average, consumed significantly more FV at the end of the intervention (2.58 ± 0.30 servings/day) than at baseline (2.17 ± 0.28 servings/day; $$P \leq 0.03$$). This change over time in FV intake remained significant after adjusting the model for children's age, BMI z-score and taste/smell sensitivity. When analyzing fruit intake and vegetable intake as separate outcomes and including or excluding French fries in the vegetables category, the main effect of time was not statistically significant ($P \leq 0.08$). We also explored potential changes in water intake over the course of the intervention but did not find a significant group-by-time effect ($$P \leq 0.99$$) or main effects of group or time ($P \leq 0.21$) on participants' water intake. Table 3 \| Intake \| Technology (N = 19) \| Technology (N = 19).1 \| Nutrition education (N = 19) \| Nutrition education (N = 19).1 \| Results from mixed-model analysis \| Between-group comparison at baseline \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Intake \| Baseline (mean ± SEM) \| Month 3 (mean ± SEM) \| Baseline (mean ± SEM) \| Month 3 (mean ± SEM) \| Results from mixed-model analysis \| Between-group comparison at baseline \| \| Fruits and vegetables \| Fruits and vegetables \| Fruits and vegetables \| Fruits and vegetables \| Fruits and vegetables \| Fruits and vegetables \| Fruits and vegetables \| \| Fruits and vegetables, servings/day \| 1.79 ± 0.40 \| 2.16 ± 0.43 \| 2.56 ± 0.40 \| 3.01 ± 0.41 \| GroupTime: P = 0.82; Group: P = 0.15; Time: P = 0.04 \| P = 0.15 \| \| Fruits, servings/day \| 0.82 ± 0.36 \| 0.84 ± 0.40 \| 1.48 ± 0.36 \| 1.54 ± 0.37 \| GroupTime: P = 0.93; Group: P = 0.18; Time: P = 0.84 \| P = 0.18 \| \| Vegetables (french fries included), servings/day \| 0.97 ± 0.20 \| 1.26 ± 0.25 \| 1.08 ± 0.20 \| 1.45 ± 0.21 \| GroupTime: P = 0.83; Group: P = 0.55; Time: P = 0.08 \| P = 0.66 \| \| Vegetables (french fries excluded), servings/day \| 0.71 ± 0.21 \| 0.84 ± 0.25 \| 0.95 ± 0.21 \| 1.29 ± 0.21 \| GroupTime: P = 0.53; Group: P = 0.19; Time: P = 0.19 \| P = 0.32 \| \| Salty and sugary snacks \| Salty and sugary snacks \| Salty and sugary snacks \| Salty and sugary snacks \| Salty and sugary snacks \| Salty and sugary snacks \| Salty and sugary snacks \| \| Salty and Sugary snacks, kcal/day \| 401.6 ± 49.0 \| 420.8 ± 58.4 \| 456.9 ± 49.0 \| 385.7 ± 51.0 \| GroupTime: P = 0.25; Group: P = 0.87; Time: P = 0.50 \| P = 0.42 \| \| Sugary snacks, kcal/day \| 252.8 ± 38.2 \| 225.3 ± 45.7 \| 251.9 ± 38.2 \| 244.3 ± 39.9 \| GroupTime: P = 0.74; Group: P = 0.85; Time: P = 0.56 \| P = 0.99 \| \| Savory snacks, kcal/day \| 148.8 ± 34.8 \| 195.3 ± 41.8 \| 205.0 ± 34.8 \| 141.9 ± 36.3 \| GroupTime: P = 0.06; Group: P = 0.98; Time: P = 0.77 \| P = 0.22 \| \| Beverages \| Beverages \| Beverages \| Beverages \| Beverages \| Beverages \| Beverages \| \| Sugar-sweetened beverages, fl oz/day \| 7.8 ± 1.6 \| 7.6 ± 1.7 \| 7.1 ± 1.6 \| 7.3 ± 1.6 \| GroupTime: P = 0.75; Group: P = 0.83; Time: P = 0.95 \| P = 0.78 \| \| Water, fl oz/day \| 16.1 ± 3.1 \| 17.5 ± 3.5 \| 21.5 ± 3.1 \| 22.9 ± 3.2 \| GroupTime: P = 0.99; Group: P = 0.21; Time: P = 0.45 \| P = 0.24 \| ## Individual differences in response to nutrition intervention Overall, the results also remained consistent when adding participants' BMI z-score, age, or taste/smell sensitivity as covariates to the statistical models. We, however, did find that taste/smell sensitivity significantly predicted children's intake of FV when French Fries were included ($$P \leq 0.0446$$); for each unit of lower taste/smell sensitivity (indicating greater sensory processing abnormalities), estimated mean FV intake increased by 0.13 ± 0.1 servings/day. We also found a significant effect ($$P \leq 0.03$$) of BMI z-score on intake of SSB indicating that for each unit increase in children's baseline BMI z-score, estimated mean SSB intake increased by 1.86 ± 0.8 fl oz/day during the study. Exploratory stratified analyses revealed that among children who were high consumers of savory snacks at baseline, those in the education group showed a significant decrease in calories consumed from savory snacks over time (296 ± 37 kcal/day vs. 165 ± 37 kcal/day; $$P \leq 0.02$$) while children in the technology group showed a trend for a significant increase in calories consumed from savory snacks over the course of the intervention (286 ± 43 kcal/day vs. 439 ± 61 kcal/day; $$P \leq 0.06$$; GroupTime: $$P \leq 0.007$$). ## Efficacy of technology-based intervention by engagement In this exploratory analysis, we assessed the extent to which level of participants' engagement with the technology impacted their dietary outcomes by examining both planned behavior (i.e., completion of nutrition training and setting goals) and performed behavior (i.e., number of points earned for meeting goals). Exploratory stratified analyses revealed that among children who were low consumers of FV (with French Fries excluded) at baseline, those who showed high performed engagement with the technology-based intervention significantly increased their FV intake by 1.5 servings/day over the course of the intervention (engagement-by-time: $P \leq 0.01$) even when adjusting for children's age, BMI z-score and taste/smell sensitivity (Figure 2). Figure 2: Model-based means (±SEM) of fruit and vegetable intake (servings/day), sweet and savory snack intake (kcal/day), and water intake (fl oz/day) for children in the technology group (n = 19) by level of performed engagement (low, medium, high) with the technology and consumption status (low, high) at baseline and at the end of the intervention. With respect to savory snacks, findings related to performed and planned engagement were consistent with those seen when using treatment group as predictor. That is, among children who were high consumers of savory snacks at baseline, those who were engaged with the technology exhibited an increase in calories consumed from savory snacks, while those who were not engaged with the technology exhibited a decrease in savory snacks (performed engagement-by-time interaction: $$P \leq 0.01$$; planned engagement-by-time interaction: $$P \leq 0.02$$). Further, we found a statistically significant planned ($$P \leq 0.04$$) and performed ($$P \leq 0.03$$) engagement-by-time interaction for water intake in our stratified analysis. That is, among children who were low consumers of water at baseline, those who showed little planned or performed engagement with the technology exhibited a significant increase in water intake over the course of the intervention (adjusted pairwise comparison for planned ($$P \leq 0.001$$) and performed ($$P \leq 0.002$$) engagement). ## Discussion This exploratory trial was among the first to test the preliminary efficacy of a technology-based mHealth intervention for improving dietary intake among autistic youth who are picky eaters. While this technology intervention produced few between-group differences in the consumption of targeted foods and beverages, subgroup analyses revealed that some children, namely those who consumed few FV at baseline but showed high engagement with the technology, significantly increased their FV intake by the end of the intervention. These findings remained significant when adjusting for taste and smell sensitivity, which was identified as a significant predictor of FV intake among the study sample. Exploratory stratified analyses further revealed that among children who were high consumers of savory snacks at baseline, those in the education group showed a significant decrease in calories consumed from savory snacks over time while children in the technology group showed a trend for a significant increase in calories consumed from savory snacks. Also, among children in the intervention group who were low consumers of water at baseline, those who showed little engagement with the technology exhibited a significant increase in water intake over the course of the intervention. Contrary to our hypotheses, we did not find a significant group-by-time interaction for any of the primary dietary outcomes. We did, however, find a statistically significant increase in FV intake by 0.4 servings/day (or $19\%$) in both groups over the course of the intervention. The magnitude of this increase is approximately one third smaller than that seen in behavioral interventions in children with neurotypical development [53, 73, 74] and represents approximately one-sixth of the FV intake recommendations for children of that age [75]. The finding that children in the control group also increased their FV intake was unexpected. It is possible that providing families in the control group with a brief nutrition education and printed handout which summarized the dietary targets, having groups matched for research staff contact time, and simply being part of a nutrition research study led to improvements in FV intake among children who did not receive the technology intervention. The finding may also suggest that even low-level efforts and outreach on topics related to public health and nutrition may be useful for making a small but meaningful impact on health behaviors in this population. With respect to individual differences in children's response to the intervention, we found that children's taste and smell sensitivity significantly predicted their intake of FV for all participants. That is, for each unit of lower taste/smell sensitivity, FV intake increased by 0.13 ± 0.1 servings/day. These data are in agreement with findings from prior research. For example, Coulthard and Blissett [76] reported that children with neurotypical development who were more sensitive to taste and smell stimuli, as measured by the Sensory Profile, consumed fewer FV. A similar relationship between sensory sensitivity and FV intake has been found previously in children who have ASD [77, 78]. Thus, promoting healthy eating is likely particularly challenging for children with increased sensory sensitivities. Future interventions should consider intervening at both the level of support for sensory sensitivities (e.g., occupational therapy, exposure and desensitization, sensory-specific diets) and the level of accommodation by parents (e.g., skills training for parents to modify the taste and/or texture of foods and meals through creative preparation and cooking methods). We further found a significant effect of BMI z-score on intake of SSB indicating that for each unit increase in children's baseline BMI z-score, SSB intake increased by approximately 2 fl oz/day over the course of the intervention. Several systematic reviews and meta-analyses point to a significant positive association between intake of SSB and BMI during childhood (17, 79–81). Even though SSB intake was a dietary target in the current research, the mHealth intervention did not lead to a reduction in SSB. In our stratified analysis, however, we found statistically significant planned and performed engagement-by-time interactions for water intake. Children who drank little water at baseline significantly increased their water intake over the course of the intervention, even in the absence of engaging much with the technology. Given that SSB intake did not decrease over the course of the intervention, it can be assumed that water intake added on but did not replace SSB among children. Additional efforts will be needed for future intervention design and refinement that target the substitution of SSB with water, especially among children with a higher weight status. In an exploratory analysis, we tested if children in the technology group who were more engaged with the mHealth intervention showed greater improvements in their intake of the target foods and beverages when compared to children who were less engaged with the technology. With respect to healthy target foods and beverages, we found that children who were low consumers of FV at baseline and showed high performed engagement with the technology-based intervention significantly increased their FV intake by 1.5 servings/day over the course of the intervention. Interestingly, we also found that among children who were low consumers of water at baseline but showed little planned or performed engagement with the technology also significantly increased their water intake over the course of the intervention. This suggests that (a) participating in a mHealth nutrition intervention may be particularly useful for children with low baseline intakes of the target foods and (b) having children engage even a little with the technology-based intervention can lead to significant improvements in intake (as was the case with water intake). Indeed, prior research in adults has confirmed that engagement with a mHealth technology was a significant predictor of dietary intake behavior change [82]. It will be critically important to identify components of the mHealth intervention, such as increased interactivity, personalization, or individual feedback, which keep children and families engaged with the technology long-term [83]. With respect to unhealthy target foods and beverages, children in the technology group who were [1] high consumers of savory snacks at baseline and [2] showed high planned and performed engagement with the technology exhibited a significant increase of 352 calories consumed from savory snacks over the course of the intervention, while children who were not engaged with the technology showed a decrease in calories consumed from savory snacks. Interestingly, we also showed that children in education group, who did not have access to the technology and were high consumers of savory snacks at baseline, showed a significant decrease of 131 calories consumed from savory snacks over time. Reducing children's intake of less healthy foods or substituting unhealthy foods for healthy foods may represent a more challenging intervention target. While this intervention showed some improvements in increasing intake of healthy foods and beverages (e.g., FV, water), the benefits of this increased intake will be limited if there is no concomitant decrease in intake of less healthy foods (e.g., SSB, SSS) and may actually increase overall energy intake. The strengths of the study include the racial/ethnic diversity of our study sample and the inclusion of children with a range in weight status (∼$50\%$ with overweight/obesity). To our knowledge, this study is also among the first to test a technology-based nutrition intervention in autistic youth who are picky eaters. The study also had limitations. One, enrollment of study participants fell short of our recruitment goal ($83\%$) and had a higher attrition rate ($18\%$) than anticipated due to the restrictions of the COVID-19 pandemic (e.g., social distancing, stay-at-home orders) which took place towards the later part of the study. Some families were unable to complete their final visits due to stay-at-home orders, and others indicated that priorities had shifted given new childcare, etc. demands. Not reaching our recruitment goal likely impacted the statistical power for the study. While the current analyses should therefore be regarded as exploratory, it still provides important effect size estimates which will help inform future technology-based interventions. Second, the mHealth technology included a larger number of examples for both healthy and less healthy foods and beverages than did the handout for the education group. This may have differentially affected families' food choices. Also, given the fairly short duration and low intensity of the current intervention, it will be important for future studies to determine if any changes seen in children's dietary intake are maintained long-term. In addition, future studies that aim to promote healthier eating and increased water intake should also evaluate gastrointestinal function in children. In summary, this mHealth nutrition intervention did not yield a significant increase in intake of targeted foods and beverages among children in the technology group relative to a waitlist education group. Subgroup analyses, however, revealed that some children, namely those who consumed few FV at baseline but showed high engagement with the technology, significantly increased their FV intake by the end of the intervention. Future research should test additional strategies to expand the intervention's impact on a wider range of foods while at the same time reaching a broader group of children who have ASD. ## Data availability statement Deidentified data will be made available upon publication to scientists who provide a methodologically sound proposal for use in accomplishing the aims of the approved proposal. Proposals should be sent to the corresponding author at [email protected]. ## Ethics statement The studies involving human participants were reviewed and approved by Institutional Review Boards of the University of Pennsylvania and the Children's Hospital of Philadelphia. Parents and children were asked to provide voluntary informed consent (parents) and assent (children) to participate in the study by signing the consent and assent forms. ## Author contributions The authors' responsibilities were as follows – TVEK: study design, interpretation of the results, and writing of the manuscript; LOM, KJ, TB: data collection, interpretation of the results, and critical revision of the manuscript; JC, RJQ: statistical analysis, interpretation of the results, critical revision of the manuscript; and JPM, SEL, GT, ESK: study design, interpretation of the results, and critical revision of the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest TVEK discloses a financial conflict of interest related to the intellectual property of the mHealth nutrition intervention that was tested as part of this clinical trial. The conflict was managed by the University of Pennsylvania and all research data collection, data management, and statistical analyses were carried out by individuals who had no related conflicts of interest. JGT is a member of the Scientific Advisory Board and receives consulting fees from Lummé Health, Inc. and Medifast, Inc. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Skyless Game Studios, the software company which developed the technology, did not provide funding for the study and did not receive data from this study to support its marketing. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Bottema-Beutel K, Kapp SK, Lester JN, Sasson NJ, Hand BN. Avoiding ableist language: suggestions for autism researchers. Autism in Adulthood (2021.0) 3 18-29. DOI: 10.1089/aut.2020.0014 2. Kenny L, Hattersley C, Molins B, Buckley C, Povey C, Pellicano E. Which terms should be used to describe autism? Perspectives from the UK autism community. Autism (2016.0) 20 442-62. DOI: 10.1177/1362361315588200 3. Buijsman R, Begeer S, Scheeren AM. ‘Autistic person’ or ‘person with autism'? Person-first language preference in Dutch adults with autism and parents. Autism (2022.0) 13623613221117914. DOI: 10.1177/13623613221117914 4. Sharp WG, Berry RC, McCracken C, Nuhu NN, Marvel E, Saulnier CA. Feeding problems and nutrient intake in children with autism Spectrum disorders: a meta-analysis and comprehensive review of the literature. J Autism Dev Disord (2013.0) 43 2159-73. DOI: 10.1007/s10803-013-1771-5 5. Bourne L, Mandy W, Bryant-Waugh R. Avoidant/restrictive food intake disorder and severe food selectivity in children and young people with autism: a scoping review. Dev Med Child Neurol (2022.0) 64 691-700. DOI: 10.1111/dmcn.15139 6. Baraskewich J, von Ranson KM, McCrimmon A, McMorris CA. Feeding and eating problems in children and adolescents with autism: a scoping review. Autism (2021.0) 25 1505-19. DOI: 10.1177/1362361321995631 7. Cermak SA, Curtin C, Bandini LG. Food selectivity and sensory sensitivity in children with autism spectrum disorders. J Am Diet Assoc (2010.0) 110 238-46. DOI: 10.1016/j.jada.2009.10.032 8. Kral TV, Eriksen WT, Souders MC, Pinto-Martin JA. Eating behaviors, diet quality, and gastrointestinal symptoms in children with autism Spectrum disorders: a brief review. J Pediatr Nurs (2013.0) 28 548-56. DOI: 10.1016/j.pedn.2013.01.008 9. Schmitt L, Heiss CJ, Campdell E. A comparison of nutrient intake and eating behaviors of boys with and without autism. Top Clin Nutr (2008.0) 23 23-31. DOI: 10.1097/01.TIN.0000312077.45953.6c 10. Bandini LG, Anderson SE, Curtin C, Cermak S, Evans EW, Scampini R. Food selectivity in children with autism spectrum disorders and typically developing children. J Pediatr (2010.0) 157 259-64. DOI: 10.1016/j.jpeds.2010.02.013 11. Evans EW, Must A, Anderson SE, Curtin C, Scampini R, Maslin M. Dietary patterns and body mass Index in children with autism and typically developing children. Res Autism Spectr Disord (2012.0) 6 399-405. DOI: 10.1016/j.rasd.2011.06.014 12. Schreck KA, Williams K. Food preferences and factors influencing food selectivity for children with autism spectrum disorders. Res Dev Disabil (2006.0) 27 353-63. DOI: 10.1016/j.ridd.2005.03.005 13. Hubbard KL, Anderson SE, Curtin C, Must A, Bandini LG. A comparison of food refusal related to characteristics of food in children with autism spectrum disorder and typically developing children. J Acad Nutr Diet (2014.0) 114 1981-7. DOI: 10.1016/j.jand.2014.04.017 14. Broder-Fingert S, Brazauskas K, Lindgren K, Iannuzzi D, Van Cleave J. Prevalence of overweight and obesity in a large clinical sample of children with autism. Acad Pediatr (2014.0) 14 408-14. DOI: 10.1016/j.acap.2014.04.004 15. Messiah SE, Vidot DC, Somarriba G, Haney K, Aytur S, Natale RA. Obesity and cardiometabolic disease risk factors among US adolescents with disabilities. World J Diabetes (2015.0) 6 200-7. DOI: 10.4239/wjd.v6.i1.200 16. Malik VS, Schulze MB, Hu FB. Intake of sugar-sweetened beverages and weight gain: a systematic review. Am J Clin Nutr (2006.0) 84 274-88. DOI: 10.1093/ajcn/84.2.274 17. Malik VS, Pan A, Willett WC, Hu FB. Sugar-sweetened beverages and weight gain in children and adults: a systematic review and meta-analysis. Am J Clin Nutr (2013.0) 98 1084-102. DOI: 10.3945/ajcn.113.058362 18. DeBoer MD, Scharf RJ, Demmer RT. Sugar-sweetened beverages and weight gain in 2- to 5-year-old children. Pediatrics (2013.0) 132 413-20. DOI: 10.1542/peds.2013-0570 19. Kuhl ES, Clifford LM, Stark LJ. Obesity in preschoolers: behavioral correlates and directions for treatment. Obesity (Silver Spring) (2012.0) 20 3-29. DOI: 10.1038/oby.2011.201 20. Barlow SE. Expert committee recommendations regarding the prevention, assessment, and treatment of child and adolescent overweight and obesity: summary report. Pediatrics (2007.0) 120 S164-92. DOI: 10.1542/peds.2007-2329C 21. Davis MM, Gance-Cleveland B, Hassink S, Johnson R, Paradis G, Resnicow K. Recommendations for prevention of childhood obesity. Pediatrics (2007.0) 120 S229-53. DOI: 10.1542/peds.2007-2329E 22. Larson N, Story M. A review of snacking patterns among children and adolescents: what are the implications of snacking for weight status?. Child Obes (2013.0) 9 104-15. DOI: 10.1089/chi.2012.0108 23. Vernarelli JA, Mitchell DC, Hartman TJ, Rolls BJ. Dietary energy density is associated with body weight status and vegetable intake in U. S. Children. J Nutr (2011.0) 141 2204-10. DOI: 10.3945/jn.111.146092 24. Ledikwe JH, Blanck HM, Khan LK, Serdula MK, Seymour JD, Tohill BC. Low-energy-density diets are associated with high diet quality in adults in the United States. J Am Diet Assoc (2006.0) 106 1172-80. DOI: 10.1016/j.jada.2006.05.013 25. Leahy KE, Birch LL, Fisher JO, Rolls BJ. Reductions in entree energy density increase children's vegetable intake and reduce energy intake. Obesity (Silver Spring) (2008.0) 16 1559-65. DOI: 10.1038/oby.2008.257 26. Leahy KE, Birch LL, Rolls BJ. Reducing the energy density of an entree decreases children's energy intake at lunch. J Am Diet Assoc (2008.0) 108 41-8. DOI: 10.1016/j.jada.2007.10.015 27. McGuire S. Institute of medicine (IOM) early childhood obesity prevention policies. Washington, DC: the national academies press; 2011. Adv Nutr (2012.0) 3 56-7. DOI: 10.3945/an.111.001347 28. Epstein LH, Gordy CC, Raynor HA, Beddome M, Kilanowski CK, Paluch R. Increasing fruit and vegetable intake and decreasing fat and sugar intake in families at risk for childhood obesity. Obes Res (2001.0) 9 171-8. DOI: 10.1038/oby.2001.18 29. Lukens CT, Silverman AH. Systematic review of psychological interventions for pediatric feeding problems. J Pediatr Psychol (2014.0) 39 903-17. DOI: 10.1093/jpepsy/jsu040 30. Sharp WG, Jaquess DL, Morton JF, Herzinger CV. Pediatric feeding disorders: a quantitative synthesis of treatment outcomes. Clin Child Fam Psychol Rev (2010.0) 13 348-65. DOI: 10.1007/s10567-010-0079-7 31. Sharp WG, Volkert VM, Scahill L, McCracken CE, McElhanon B. A systematic review and meta-analysis of intensive multidisciplinary intervention for pediatric feeding disorders: how standard is the standard of care?. J Pediatr (2017.0) 181 116-124.e4. DOI: 10.1016/j.jpeds.2016.10.002 32. Najdowski AC, Wallace MD, Doney JK, Ghezzi PM. Parental assessment and treatment of food selectivity in natural settings. J Appl Behav Anal (2003.0) 36 383-6. DOI: 10.1901/jaba.2003.36-383 33. Najdowski AC, Wallace MD, Reagon K, Penrod B, Higbee TS, Tarbox J. Utilizing a home-based parent training approach in the treatment of food selectivity. Behav Interv (2010.0) 25 89-107. DOI: 10.1002/bin.298 34. Sharp WG, Burrell TL, Jaquess DL. The autism MEAL plan: a parent-training curriculum to manage eating aversions and low intake among children with autism. Autism (2014.0) 18 712-22. DOI: 10.1177/1362361313489190 35. Cosbey J, Muldoon D. EAT-UP Family-Centered feeding intervention to promote food acceptance and decrease challenging behaviors: a single-case experimental design replicated across three families of children with autism Spectrum disorder. J Autism Dev Disord (2017.0) 47 564-78. DOI: 10.1007/s10803-016-2977-0 36. Johnson CR, Foldes E, DeMand A, Brooks MM. Behavioral parent training to address feeding problems in children with autism Spectrum disorder: a pilot trial. J Dev Phys Disabil (2015.0) 27 591-607. DOI: 10.1007/s10882-015-9437-1 37. Johnson CR, Handen BL, Butter E, Wagner A, Mulick J, Sukhodolsky DG. Development of a parent training program for children with pervasive developmental disorders. Behav Interv (2007.0) 22 201-21. DOI: 10.1002/bin.237 38. Johnson CR, Brown K, Hyman SL, Brooks MM, Aponte C, Levato L. Parent training for feeding problems in children with autism Spectrum disorder: initial randomized trial. J Pediatr Psychol (2019.0) 44 164-75. DOI: 10.1093/jpepsy/jsy063 39. Spoth R, Redmond C. Research on family engagement in preventive interventions: toward improved use of scientific findings in primary prevention practice. J Prim Prev (2000.0) 21 267-84. DOI: 10.1023/A:1007039421026 40. Odom SL, Thompson JL, Hedges S, Boyd BA, Dykstra JR, Duda MA. Technology-Aided interventions and instruction for adolescents with autism Spectrum disorder. J Autism Dev Disord (2015.0) 45 3805-19. DOI: 10.1007/s10803-014-2320-6 41. Bosseler A, Massaro DW. Development and evaluation of a computer-animated tutor for vocabulary and language learning in children with autism. J Autism Dev Disord (2003.0) 33 653-72. DOI: 10.1023/B:JADD.0000006002.82367.4f 42. Hopkins IM, Gower MW, Perez TA, Smith DS, Amthor FR, Wimsatt FC. Avatar assistant: improving social skills in students with an ASD through a computer-based intervention. J Autism Dev Disord (2011.0) 41 1543-55. DOI: 10.1007/s10803-011-1179-z 43. Joshi P. Finding good apps for children with autism, in The New York Times (2011.0) 44. Pickard KE, Wainer AL, Bailey KM, Ingersoll BR. A mixed-method evaluation of the feasibility and acceptability of a telehealth-based parent-mediated intervention for children with autism spectrum disorder. Autism (2016.0) 20 845-55. DOI: 10.1177/1362361315614496 45. Jang JN, Dixon DR, Tarbox J, Granpeesheh D, Kornack J, de Nocker Y. Randomized trial of an eLearning program for training family members of children with autism in the principles and procedures of applied behavior analysis. Res Autism Spectr Disord (2012.0) 6 852-6. DOI: 10.1016/j.rasd.2011.11.004 46. Faith MS, Van Horn L, Appel LJ, Burke LE, Carson JA, Franch HA. Evaluating parents and adult caregivers as “agents of change” for treating obese children: evidence for parent behavior change strategies and research gaps: a scientific statement from the American heart association. Circulation (2012.0) 125 1186-207. DOI: 10.1161/CIR.0b013e31824607ee 47. Wadden TA, Crerand CE, Brock J. Behavioral treatment of obesity. Psychiatr Clin North Am (2005.0) 28 151-70. DOI: 10.1016/j.psc.2004.09.008 48. Tonge BJ, Bull K, Brereton A, Wilson R. A review of evidence-based early intervention for behavioural problems in children with autism spectrum disorder: the core components of effective programs, child-focused interventions and comprehensive treatment models. Curr Opin Psychiatry (2014.0) 27 158-65. DOI: 10.1097/YCO.0000000000000043 49. Myers SM, Johnson CP. Management of children with autism spectrum disorders. Pediatrics (2007.0) 120 1162-82. DOI: 10.1542/peds.2007-2362 50. Ingersoll B, Wainer A. Initial efficacy of project ImPACT: a parent-mediated social communication intervention for young children with ASD. J Autism Dev Disord (2013.0) 43 2943-52. DOI: 10.1007/s10803-013-1840-9 51. Scahill L, Bearss K, Lecavalier L, Smith T, Swiezy N, Aman MG. Effect of parent training on adaptive behavior in children with autism Spectrum disorder and disruptive behavior: results of a randomized trial. J Am Acad Child Adolesc Psychiatry (2016.0) 55 602-609.e3. DOI: 10.1016/j.jaac.2016.05.001 52. Kral TV, Souders MC, Tompkins VH, Remiker AM, Eriksen WT, Pinto-Martin JA. Child eating behaviors and caregiver feeding practices in children with autism spectrum disorders. Public Health Nurs (2015.0) 32 488-97. DOI: 10.1111/phn.12146 53. Raynor HA, Osterholt KM, Hart CN, Jelalian E, Vivier P, Wing RR. Efficacy of U.S. Paediatric obesity primary care guidelines: two randomized trials. Pediatr Obes (2012.0) 7 28-38. DOI: 10.1111/j.2047-6310.2011.00005.x 54. 54.United States Department of Agriculture. ChooseMyPlate.gov. Available at: http://www.choosemyplate.gov/ 55. Birch LL, Fisher JO, Grimm-Thomas K, Markey CN, Sawyer R, Johnson SL. Confirmatory factor analysis of the child feeding questionnaire: a measure of parental attitudes, beliefs and practices about child feeding and obesity proneness. Appetite (2001.0) 36 201-10. DOI: 10.1006/appe.2001.0398 56. 56.American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 4th edn. Washington, DC: American Psychiatric Press Inc. (1994).. Diagnostic and statistical manual of mental disorders (1994.0) 57. 57.American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 5th edn. Arlington, VA: American Psychiatric Press Inc. (2013).. Diagnostic and statistical manual of mental disorders (2013.0) 58. 58.Autism Speaks Treatment Network. ATN/AIR-P: Parent's Guide to Applied Behavior Analysis. 59. 59.Autism Speaks Treatment Network. ATN/AIR-P: An Introduction to Behavioral Health Treatments. 60. Yee N, Bailenson J. The Proteus effect: the effect of transformed self-representation on behavior. Hum Commun Res (2007.0) 33 271-90. DOI: 10.1111/j.1468-2958.2007.00299.x 61. Li BJ, Lwin MO, Jung Y, Wii M. And size: the influence of Proteus effect and stereotype threat on overweight children's exercise motivation and behavior in exergames. Games Health J (2014.0) 3 40-8. DOI: 10.1089/g4h.2013.0081 62. Kandalaft MR, Didehbani N, Krawczyk DC, Allen TT, Chapman SB. Virtual reality social cognition training for young adults with high-functioning autism. J Autism Dev Disord (2013.0) 43 34-44. DOI: 10.1007/s10803-012-1544-6 63. 63.Institute of Medicine. Dietary reference intakes: Applications in dietary assessment. Washington, DC: The National Academies Press (2000).. Dietary reference intakes: Applications in dietary assessment (2000.0) 64. Kuczmarski RJ, Ogden CL, Guo SS, Grummer-Strawn LM, Flegal KM, Mei Z. 2000 CDC growth charts for the United States: methods and development. Vital Health Stat (2002.0) 11 1-190. PMID: 12043359 65. Ogden CL, Kuczmarski RJ, Flegal KM, Mei Z, Guo S, Wei R. Centers for disease control and prevention 2000 growth charts for the United States: improvements to the 1977 national center for health statistics version. Pediatrics (2002.0) 109 45-60. DOI: 10.1542/peds.109.1.45 66. Blumberg SJ, Bialostosky K, Hamilton WL, Briefel RR. The effectiveness of a short form of the household food security scale. Am J Public Health (1999.0) 89 1231-4. DOI: 10.2105/AJPH.89.8.1231 67. 67.United States Department of Agriculture. Definitions of food security. (2015).. (2015.0) 68. Hamilton WL, Cook JT, Thompson WL, Buron LF, Frongillo EA, Olson CM. Household food security in the United States in 1995 technical report of the food security measurement project, U.S.D.o. Agriculture, editor (1997.0) 69. McIntosh DN, Miller LJ, Shyu V, Dunn W. Development and validation of the short sensory profile. Sensory profile user's manual (1999.0) 70. Dunn W. The sensory profile: examiner's manual (1999.0) 71. Miller LJ, McIntosh DN, McGrath J, Shyu V, Lampe M, Taylor AK. Electrodermal responses to sensory stimuli in individuals with fragile X syndrome: a preliminary report. Am J Med Genet (1999.0) 83 268-79. DOI: 10.1002/(SICI)1096-8628(19990402)83:4<268::AID-AJMG7>3.0.CO;2-K 72. Ahn RR, Miller LJ, Milberger S, McIntosh DN. Prevalence of parents' perceptions of sensory processing disorders among kindergarten children. Am J Occup Ther (2004.0) 58 287-93. DOI: 10.5014/ajot.58.3.287 73. Looney SM, Raynor HA. Are changes in consumption of “healthy” foods related to changes in consumption of “unhealthy” foods during pediatric obesity treatment?. Int J Environ Res Public Health (2012.0) 9 1368-78. DOI: 10.3390/ijerph9041368 74. Epstein LH, Paluch RA, Beecher MD, Roemmich JN. Increasing healthy eating vs. Reducing high energy-dense foods to treat pediatric obesity. Obesity (Silver Spring) (2008.0) 16 318-26. DOI: 10.1038/oby.2007.61 75. 75.United States Department of Agriculture. Dietary Guidelines for Americans 2020–2025: Make every bite count with dietary guidelines (2020). (2020) Available at: https://www.dietaryguidelines.gov/sites/default/files/2020-12/Dietary_Guidelines_for_Americans_2020-2025.pdf. (2020.0) 76. Coulthard H, Blissett J. Fruit and vegetable consumption in children and their mothers. Moderating effects of child sensory sensitivity. Appetite (2009.0) 52 410-5. DOI: 10.1016/j.appet.2008.11.015 77. Chistol LT, Bandini LG, Must A, Phillips S, Cermak SA, Curtin C. Sensory sensitivity and food selectivity in children with autism Spectrum disorder. J Autism Dev Disord (2018.0) 48 583-91. DOI: 10.1007/s10803-017-3340-9 78. Dunn W, Little L, Dean E, Robertson S, Evans B. The state of the science on sensory factors and their impact on daily life for children: a scoping review. OTJR (Thorofare N J) (2016.0) 36 3S-26S. DOI: 10.1177/1539449215617923 79. Abbasalizad Farhangi M, Mohammadi Tofigh A, Jahangiri L, Nikniaz Z, Nikniaz L. Sugar-sweetened beverages intake and the risk of obesity in children: an updated systematic review and dose-response meta-analysis. Pediatr Obes (2022.0) 17 e12914. DOI: 10.1111/ijpo.12914 80. Keller A, Bucher Della Torre S. Sugar-sweetened beverages and obesity among children and adolescents: a review of systematic literature reviews. Child Obes (2015.0) 11 338-46. DOI: 10.1089/chi.2014.0117 81. Bucher Della Torre S, Keller A, Laure Depeyre J, Kruseman M. Sugar-sweetened beverages and obesity risk in children and adolescents: a systematic analysis on how methodological quality may influence conclusions. J Acad Nutr Diet (2016.0) 116 638-59. DOI: 10.1016/j.jand.2015.05.020 82. West JH, Belvedere LM, Andreasen R, Frandsen C, Hall PC, Crookston BT. Controlling your “app"etite: how diet and nutrition-related mobile apps lead to behavior change. JMIR Mhealth Uhealth (2017.0) 5 e95. DOI: 10.2196/mhealth.7410 83. Lim SY, Lee KW, Seow WL, Mohamed NA, Devaraj NK, Amin-Nordin S. Effectiveness of integrated technology apps for supporting healthy food purchasing and consumption: a systematic review. Foods (2021.0) 10 1861. DOI: 10.3390/foods10081861
--- title: Adult first-generation immigrants and cardiovascular risk factors in the Veneto Region, Northeast Italy authors: - Teresa Dalla Zuanna - Erich Batzella - Gisella Pitter - Francesca Russo - Teresa Spadea - Cristina Canova journal: Frontiers in Public Health year: 2023 pmcid: PMC9975734 doi: 10.3389/fpubh.2023.956146 license: CC BY 4.0 --- # Adult first-generation immigrants and cardiovascular risk factors in the Veneto Region, Northeast Italy ## Abstract ### Introduction The health condition of immigrants traditionally follows a transition from a low disease occurrence to the epidemiological profile of the deprived groups in the host country. In the Europe, studies examining differences in biochemical and clinical outcomes among immigrants and natives are lacking. We examined differences in cardiovascular risk factors between first-generation immigrants and Italians, and how migration pattern variables could affect health outcomes. ### Material and methods We included participants between 20 and 69 years recruited from a Health Surveillance Program of the Veneto Region. Blood pressure (BP), total cholesterol (TC) and LDL cholesterol levels were measured. Immigrant status was defined by being born in a high migratory pressure country (HMPC) and subdivided by geographical macro-areas. We used generalized linear regression models to investigate differences between these outcomes among immigrants compared to native-born, adjusting for age, sex, education, BMI, alcohol consumption, smoking status, food consumption, salt consumption in the BP analysis and the laboratory in charge for cholesterol analysis. Within immigrant subjects, the results were stratified by variables of the migration pattern: age at immigration and length of residence in Italy. ### Results Thirty seven thousand three hundred and eighty subjects were included in the analysis, $8.6\%$ were born in an HMPC. Heterogeneous results were seen by the macro-areas of origin and sex, with male immigrants from CE Europe (β = 8.77 mg/dl) and Asia (β = 6.56 mg/dl) showing higher levels of TC than native-born, while female immigrants from Northern Africa showed lower levels of TC (β = −8.64 mg/dl). BP levels were generally lower among immigrants. Immigrants residing in Italy for more than 20 years had lower levels of TC (β = −2.9 mg/dl) than native-born. In contrast, immigrants who arrived <20 years ago or arrived older than 18 years had higher levels of TC. This trend was confirmed for CE Europeans and was inverted for Northern Africans. ### Conclusions The large heterogeneity in the results depending on sex and macro-area of origin indicates the need for targeted intervention in each specific immigrant group. The results confirm that acculturation leads to a convergence toward the epidemiological profile of the host population that depends on the starting condition of the immigrant group. ## 1. Introduction The unprecedented flow of immigrants toward Europe in the last few decades has turned most European countries into multiethnic societies. Italy, like other South European countries, has experienced a relevant increase in this phenomenon in the last 20 years, with the percentage of the immigrant population reaching $8.5\%$ in 2018 [1]. The health condition of immigrants in comparison to natives is traditionally expected to follow a transition from a low disease occurrence in the first period after arrival (so-called “healthy migrant effect”) [2] to a progressive convergence toward the health behaviors and epidemiological profile of the lowest socioeconomic groups of the host population [3]. This acculturation process, which entails an increase in risky behaviors [4] and the adoption of a Westernized diet and a more sedentary lifestyle [5], represents a threat to their physical and mental health. Furthermore, the accessibility of health services for immigrants is often undermined by cultural and language barriers, creating new challenges for health systems. In Europe, limited information is available on ethnic differences concerning biochemical and clinical parameters known as CVD risk factors, such as elevated plasma lipid levels and elevated blood pressure levels [6, 7]. Most of the data are derived from comparisons between studies conducted in the United States [8, 9]. A study conducted in the Netherlands found large ethnic differences in lipid components, both in unadjusted models and in models adjusted for multiple covariates known to affect lipid metabolism. These results suggested that, next to lifestyle factors, intrinsic differences in lipid metabolism may contribute to the observed differences in plasma lipid levels [10]. A review on ethnic differences in blood pressure levels in Europe found higher blood pressure levels in immigrants from Sub-Saharan Africa over decades and lower levels in the Muslim population, suggesting the limited efficacy of prevention in some groups and that untapped lifestyle and behavioral habits may reveal advantages toward the development of hypertension [11]. In Italy, studies based on health administrative data or nationwide surveys are improving the knowledge concerning the health status of immigrant populations compared to native-born Italians. The differential disease prevalence has been widely studied, it has been shown that diabetes mellitus prevalence is higher among immigrants than among native-born Italians [12]. The prevalence of other cardiovascular diseases is comparable, showing heterogeneous patterns for immigrants from different countries of origin but with worse indicators of the clinical management of the disease [12]. On the other hand, studies examining differences in biochemical and clinical outcomes, known as cardiovascular risk factors, among immigrants and natives are lacking. Increased levels of total and LDL cholesterol and high blood pressure are the most prevalent conditions increasing the risk of cardiovascular disease (CVD) [13]. Since biochemical alterations are detectable before the onset of the disease itself, we focused the analysis on these risk factors that could help in detecting health differences earlier than the analyses on the confirmed diseases. Early detection of these differences, especially in young populations, could provide more chances for early treatments and reduce future health inequalities. The objective of this study is to examine the differences in the lipid profiles and blood pressure levels between first-generation immigrants and native-born Italians in a large population of the Veneto Region, Northeast Italy. This piece of information could provide insight into understanding the mechanisms behind emerging differences in the occurrence of cardiovascular diseases. The immigrant population is very heterogeneous and should possibly be analyzed considering differences within this category. Therefore, we also analyzed these data by geographical macro-area of origin to investigate how variables of the migration pathway could affect health outcomes. ## 2.1. Participants and study design We analyzed data from a publicly funded health surveillance program implemented by the Veneto Region in 30 municipalities in this Region, located in Northeast Italy [14]. This program is a population-based screening program with the aim of the prevention, early diagnosis, and treatment of chronic disorders possibly associated with the high perfluoroalkyl substances exposure–PFAS, manufactured chemicals with grease-, stain-, and water-repelling properties–that was discovered in this area in 2013. The program started in 2017, and it is still ongoing, with no cost for participants. The target population included 105,000 residents of the contaminated area born between 1951 and 2014. Eligible subjects were identified through the regional health registry, which contains personal and residency data for the entire population of the Veneto Region. Residents who decided to participate in the program completed a structured interview administered by a trained public health nurse, followed by blood pressure measurement and blood and urine sampling. Program visits were performed at public health facilities located throughout the contaminated area to ensure easy accessibility. Data were collected using centralized web-based software connected with the regional health registry. The software allows the extraction of lists of eligible residents, online compiling of interview and blood pressure data, and retrieval of laboratory test results. To maximize data quality by minimizing errors and missing values, standard data checks and cleaning procedures (e.g., range and consistency checks) were performed. All recruited subjects until May 2021, aged between 20 and 69 years old, were included in this analysis ($$n = 38$$,292, participation rate $61\%$). Pregnant women at the time of participation in the study and participants with missing data on relevant variables were excluded, leaving 37,710 subjects included in the analysis (Supplementary Figure S1). No missing data on exposure and outcome variables were present. ## 2.2. Outcome assessment Nonfasting blood and urine samples collected from participants were sent to three local health unit laboratories within the study area (Arzignano, San Bonifacio, Legnago). Blood pressure was measured according to the European Society of Hypertension recommendations [15]. The outcome variables include the following: ## 2.3. Exposure: Country of birth and residential history Immigrant status was defined as the country of birth reported by the participants. Immigrants born in high migratory pressure countries (HMPCs) were further grouped into 5 geographical macro-areas of origin: Central-Eastern (CE) Europe, Central and Southern (CS) America, North Africa, Sub-Saharan (SS) Africa and Asia (except for Israel and Japan). Immigrants born in highly developed countries (HDC) were a very small percentage (330, $0.87\%$ of the study population) and were excluded from this study (Supplementary Figure S1). Each participant of the surveillance program was asked about his or her residential history, including all episodes of transfer of residence to a different country/municipality. The year of immigration to Italy was calculated for immigrants from HMPC countries as the first date of residence in any Italian municipality. The age at immigration (categorized as ≥18 years old and <18 years old, as it is in Italy the age of majority) and length of residence were subsequently calculated. The latter was categorized as <10 years, between 10 and 19 years and ≥20 years, because of the distribution of our population and the need of a sufficient statistical power. Subjects with missing information on residential history, as well as subjects with nonlinear migration pathways—such as long returns to their country of origin—were excluded from these analyses (Supplementary Figure S1). ## 2.4. Covariates The following range of potential confounders were considered based on prior literature: age (years), sex, education [primary/middle school, high school, university or higher), BMI (from self-reported height and weight and classified as normal weight (<25), overweight (25–29.9), or obese (≥30)], alcohol consumption (0, 1–2, or 3+ alcohol units per week), smoking status (current smokers, previous smokers, or nonsmokers), and food consumption. Data on food consumption (meat, fish/seafood, milk/yogurt, cheese, eggs, bread/pasta/cereals, sweets/snacks/sweet beverages, fruits/vegetables) were transformed from the number of servings per day/week/month to the number of servings per week and categorized into tertiles or quartiles to allow a harmonized diet pattern classification. Furthermore, salt consumption (categorized as low, medium, or high) was considered a possible covariate in the blood pressure analyses. Finally, information on the laboratory in charge of the analyses of serum lipids (Arzignano, Legnago, and San Bonifacio) was considered a possible confounder in the statistical analyses of cholesterol levels. ## 2.5. Statistical analysis First, the demographic and lifestyle characteristics of HMPC-born residents and natives were compared. Participants who had reported using cholesterol-lowering medications such as statins, fibrates and red rice were excluded for serum lipid outcomes ($$n = 1$$,676), and participants with a self-reported diagnosis of hypertension or under treatment with antihypertensive medications were excluded for continuous blood pressure outcomes ($$n = 4$$,859), leaving 35,704 and 32,521 subjects included for the lipid and blood pressure analyses, respectively (Supplementary Figure S1). We used generalized linear regression models (LMs) to investigate the differences between cardiometabolic outcomes (SBP, DBP, TC and LDL-C) among HMPCs and the specific macro-areas of birth compared to native-born Italians. Basic models were adjusted only for age (continuous) and sex, while fully adjusted models were additionally adjusted for the whole set of covariates. For TC and LDL-C, a random intercept was added to the models, running linear regression mixed models (LMMs) to account for the laboratory in charge of the serum analyses. For the analyses of the association of migratory status with hypertension prevalence, a log link function was used in the models, and prevalence ratios (PRs) were calculated. Estimates and $95\%$ confidence intervals ($95\%$ CI) were reported. Each HMPC category based on age at arrival and length of stay in Italy was then compared to native-born Italians, using the same previously defined models for each outcome. This analysis was also conducted for each of the three macro-areas with sufficient sample sizes (CE Europe, Northern Africa, Asia). All analyses were stratified by sex. Analyses were performed using the statistical software Stata/SE version 13.0 (Stata Corp LP, College Station, TX, USA) and R (R Development Core Team 2010, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL: http://www.R-project.org/). We employed the “lme4” and “prLogistic” packages to run LMMs and calculate prevalence ratios, respectively. ## 2.6. Ethical aspects The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Regional (Veneto Region) Ethics Committee (24 maggio 2017 prot. No. 203638). Informed consent was obtained from all subjects involved in the study. ## 3.1. Characteristics of studied populations according to immigrant status Overall, 37,380 subjects were included in the analysis, and 3,249 ($8.7\%$) of them were born in an HMPC. Half of the immigrants came from CE Europe, $20.1\%$ from Northern Africa, $17.7\%$ from Asia, $6.1\%$ from CS America and $6\%$ from SS Africa. The characteristics of native-born Italians and immigrants are presented in Table 1. Immigrants were younger than native-born, and their educational attainment was much lower than that of their native-born counterparts, especially for males. Immigrants reported considerably higher percentages of nonsmoking and not drinking alcohol. The percentages of obese subjects were higher among subjects born in an HMPC than among native-born Italians. Table 1 \| Characteristics \| Characteristics.1 \| HMPC (n = 3,249) \| HMPC (n = 3,249).1 \| HMPC (n = 3,249).2 \| HMPC (n = 3,249).3 \| HMPC (n = 3,249).4 \| HMPC (n = 3,249).5 \| Italy (n = 34,131) \| Italy (n = 34,131).1 \| Italy (n = 34,131).2 \| Italy (n = 34,131).3 \| Italy (n = 34,131).4 \| Italy (n = 34,131).5 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| \| Total \| Total \| Males (n = 1,317) \| Males (n = 1,317) \| Females (n = 1,932) \| Females (n = 1,932) \| Total \| Total \| Males (n = 16,493) \| Males (n = 16,493) \| Females (n = 17,638) \| Females (n = 17,638) \| \| \| \| Median (IQR) \| Min-Max \| Median (IQR) \| Min-Max \| Median (IQR) \| Min-Max \| Median (IQR) \| Min-Max \| Median (IQR) \| Min-Max \| Median (IQR) \| Min-Max \| \| Age (years) \| Age (years) \| 40 (33–47) \| 20–66 \| 41 (34–48) \| 20–66 \| 39 (33–46) \| 20–66 \| 42 (32–50) \| 20–68 \| 42 (31–50) \| 20–67 \| 43 (32–51) \| 20–68 \| \| \| \| N \| % \| N \| % \| N \| % \| N \| % \| N \| % \| N \| % \| \| BMI \| Normal weight \| 1464 \| 45.1% \| 510 \| 38.7% \| 954 \| 49.4% \| 19620 \| 57.5% \| 7762 \| 47,1% \| 11858 \| 67,2% \| \| \| Overweight \| 1154 \| 35.5% \| 574 \| 43.6% \| 580 \| 30.0% \| 10103 \| 29.6% \| 6388 \| 38,7% \| 3715 \| 21,1% \| \| \| Obese \| 631 \| 19.4% \| 233 \| 17.7% \| 398 \| 20.6% \| 4408 \| 12.9% \| 2343 \| 14,2% \| 2065 \| 11,7% \| \| Smoking habit \| Non-smoker \| 2211 \| 68.1% \| 747 \| 56.7% \| 1464 \| 75.8% \| 19917 \| 58.4% \| 8101 \| 49,1% \| 11816 \| 67,0% \| \| \| Current-smoker \| 593 \| 18.3% \| 301 \| 22.9% \| 292 \| 15.1% \| 7634 \| 22.4% \| 4444 \| 26,9% \| 3190 \| 18,1% \| \| \| Previous smoker \| 445 \| 13.7% \| 269 \| 20.4% \| 176 \| 9.1% \| 6580 \| 19.3% \| 3948 \| 23,9% \| 2632 \| 14,9% \| \| Alcohol intake \| \| 1700 \| 52.3% \| 564 \| 42.8% \| 1136 \| 58.8% \| 9074 \| 26.6% \| 2260 \| 13,7% \| 6814 \| 38,6% \| \| \| 1–2 \| 826 \| 25.4% \| 291 \| 22.1% \| 535 \| 27.7% \| 11892 \| 34.8% \| 5017 \| 30,4% \| 6875 \| 39,0% \| \| \| 3+ \| 723 \| 22.3% \| 462 \| 35.1% \| 261 \| 13.5% \| 13165 \| 38.6% \| 9216 \| 55,9% \| 3949 \| 22,4% \| \| Education \| Elementary/Middle \| 1586 \| 48.8% \| 725 \| 55.0% \| 861 \| 44.6% \| 10531 \| 30.9% \| 5204 \| 31,6% \| 5327 \| 30,2% \| \| \| Highschool \| 1353 \| 41.6% \| 505 \| 38.3% \| 848 \| 43.9% \| 17313 \| 50.7% \| 8780 \| 53,2% \| 8533 \| 48,4% \| \| \| University \| 310 \| 9.5% \| 87 \| 6.6% \| 223 \| 11.5% \| 6287 \| 18.4% \| 2509 \| 15,2% \| 3778 \| 21,4% \| \| Laboratory \| Arzignano \| 1880 \| 57.9% \| 798 \| 60.6% \| 1082 \| 56.0% \| 19792 \| 58.0% \| 9622 \| 58,3% \| 10170 \| 57,7% \| \| \| Legnago \| 771 \| 23.7% \| 320 \| 24.3% \| 451 \| 23.3% \| 6527 \| 19.1% \| 3130 \| 19,0% \| 3397 \| 19,3% \| \| \| San bonifacio \| 598 \| 18.4% \| 199 \| 15.1% \| 399 \| 20.7% \| 7812 \| 22.9% \| 3741 \| 22,7% \| 4071 \| 23,1% \| \| Central-Eastern Europe \| 1628 \| 50.1% \| 569 \| 43.2% \| 1059 \| 54.8% \| \| \| \| \| \| \| \| \| Sub-Saharan Africa \| 196 \| 6.0% \| 105 \| 8.0% \| 91 \| 4.7% \| \| \| \| \| \| \| \| \| Northern Africa \| 652 \| 20.1% \| 303 \| 23.0% \| 349 \| 18.1% \| \| \| \| \| \| \| \| \| Asia \| 576 \| 17.7% \| 279 \| 21.2% \| 297 \| 15.4% \| \| \| \| \| \| \| \| \| Central-Southern America \| 197 \| 6.1% \| 61 \| 4.6% \| 136 \| 7.0% \| \| \| \| \| \| \| \| \| Age at arrival (years)* \| <18 \| 495 \| 16.1% \| 222 \| 18.0% \| 273 \| 14.9% \| \| \| \| \| \| \| \| \| ≥18 \| 2571 \| 83.9% \| 1015 \| 82.1% \| 1556 \| 85.1% \| \| \| \| \| \| \| \| Length of stay in Italy (years)* \| 0–9 \| 458 \| 14.9% \| 131 \| 10.6% \| 327 \| 17.9% \| \| \| \| \| \| \| \| \| 10–19 \| 1795 \| 58.5% \| 682 \| 55.1% \| 1113 \| 60.9% \| \| \| \| \| \| \| \| \| 20+ \| 813 \| 26.5% \| 424 \| 34.3% \| 389 \| 21.3% \| \| \| \| \| \| \| We retrieved the year of immigration from $94.7\%$ of all immigrants (3,066 subjects). Over half of them resided in Italy for between 10 and 19 years ($58.6\%$), and the majority of all immigrants were younger than 18 years of age upon their arrival in Italy ($83.9\%$) (Supplementary Table S1). ## 3.2. Associations between immigrant status and cardiovascular risk factors Table 2 provides estimates (β coefficients) and $95\%$ confidence intervals ($95\%$ CI) for models assessing the associations between immigrant status and the selected cardiovascular risk factors. Table 2 \| Unnamed: 0 \| Total cholesterol \| Total cholesterol.1 \| LDL cholesterol \| LDL cholesterol.1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| β1 (95% CI) \| β2 (95% CI) \| β1 (95% CI) \| β2 (95% CI) \| \| Italy \| 138.93 \| 137.02 \| 59.83 \| 56.33 \| \| HMPC overall \| 0.2 (−1.03; 1.43) \| 0.23 (−1.03; 1.5) \| 1.96 (0.86; 3.06) \| 1.09 (−0.04; 2.22) \| \| Central-Eastern Europe \| 4.39 (2.69; 6.09) \| 3.31 (1.61; 5.01) \| 5.90 (4.37; 7.42) \| 4.50 (2.98; 6.02) \| \| Sub-Saharan Africa \| −4.7 (−9.43; 0.04) \| −5.00 (−9.73; −0.28) \| −2.17 (−6.41; 2.07) \| −3.57 (−7.78; 0.65) \| \| Northern Africa \| −8.64 (−11.27; −6.01) \| −7.69 (−10.4; −4.99) \| −5.20 (−7.55; −2.84) \| −6.19 (−8.61; −3.77) \| \| Asia \| −0.43 (−3.24; 2.38) \| 0.85 (−2.01; 3.71) \| 1.25 (−1.27; 3.77) \| 1.02 (−1.54; 3.57) \| \| South America \| 2.17 (−2.59; 6.93) \| 1.39 (−3.32; 6.11) \| −0.21 (−4.48; 4.05) \| −1.02 (−5.23; 3.18) \| \| \| Systolic blood pressure \| Systolic blood pressure \| Diastolic blood pressure \| Diastolic blood pressure \| \| \| β1 (95% CI) \| β2 (95% CI) \| β1 (95% CI) \| β2 (95% CI) \| \| Italy \| 109.17 \| 105.61 \| 64.76 \| 65.28 \| \| HMPC overall \| −0.46 (−0.99; 0.07) \| −1.27 (−1.81; −0.72) \| −0.2 (−0.56; 0.16) \| −0.52 (−0.88; −0.15) \| \| Central-Eastern Europe \| −0.48 (−1.21; 0.26) \| −1.35 (−2.08; −0.63) \| 0.19 (−0.3; 0.69) \| −0.24 (−0.73; 0.25) \| \| Sub-Saharan Africa \| 0.53 (−1.64; 2.71) \| −0.27 (−2.4; 1.86) \| 0.02 (−1.45; 1.49) \| −0.43 (−1.86; 1.01) \| \| Northern Africa \| −0.09 (−1.19; 1.02) \| −1.23 (−2.36; −0.1) \| −1.06 (−1.81; −0.31) \| −1.49 (−2.25; −0.73) \| \| Asia \| −1.02 (−2.23; 0.19) \| −1.21 (−2.43; 0.01) \| −0.22 (−1.04; 0.6) \| −0.1 (−0.92; 0.72) \| \| South America \| −0.95 (−2.98; 1.08) \| −1.64 (−3.62; 0.34) \| −0.57 (−1.94; 0.8) \| −1.02 (−2.35; 0.31) \| In basic-adjusted models (adjusted for age and sex), no significant differences were observed for the TC levels between foreign-born adults and their native-born counterparts, while the LDL-C levels of immigrants were significantly higher than those of native-born Italians. Considering the results for each geographical macro-area of origin, immigrants from CE Europe had significantly higher TC and LDL-C levels than native-born Italians, while immigrants from Northern Africa showed significantly lower levels of both TC and LDL-C than native-born Italians. The higher levels of LDL-C among immigrants disappeared in the fully adjusted model (1.09 mg/dl, $95\%$ CI: −0.04; 2.22). Effect estimates on cholesterol levels for each geographical macro-area with the full adjustment were similar to those unadjusted, although slightly attenuated. Immigrants from CE Europe had significantly higher levels of both TC and LDL-C (TC: 3.31 mg/dl, $95\%$ CI: 1.61; −5.01, LDL-C: 4.50 mg/dl, $95\%$ CI: 2.98; −6.02), and immigrants from Northern Africa had significantly lower levels of both TC and LDL-C (TC: −7.69 mg/dl, $95\%$ CI: −10.4; −4.99, LDL-C: −6.19 mg/dl, $95\%$ CI: −8.61; −3.77). No significant difference was observed for BP levels between foreign-born adults and their native-born counterparts in basic adjusted models. In fully adjusted models, immigrants overall had significantly lower levels of SBP and DBP (SBP: −1.27 mmHg, $95\%$ CI: −1.81; −0.72, DBP: −0.52 mmHg, $95\%$ CI: −0.88; −0.15) than native-born Italians. Significantly lower levels of BP were shown for the subgroups of immigrants from Northern Africa (SBP and DBP) and CE Europe (SBP only). Figure 1 presents β estimates and $95\%$ CI results for the associations between immigrant status (overall and by macro-areas) and TC (Panel a) and SBP (Panel b), stratified by sex and adjusted for the full set of covariates. Immigrant males had significantly higher levels of TC than native-born Italian males (4.29 mg/dl, $95\%$ CI: 2.22; −6.36). Significantly higher levels also were seen for males from CE Europe (8.77 mg/dl, $95\%$ CI: 5.8; −11.75) and from Asia (6.56 mg/dl, $95\%$ CI: 2.26; −10.85). In contrast, females from Northern Africa had significantly lower levels of TC than native-born Italian females (-8.64 mg/dl, $95\%$ CI: −12.2; −5.07). Figure 1: β estimates and 95% confidence intervals of the associations between country of birth (overall and by macro-areas) and total cholesterol (A) and systolic blood pressure (B), stratified by sex and adjusted for the full set of covariates. Male and female immigrants, overall and from CE Europe, had lower levels of SBP than their native-born Italian counterparts. No significant difference was seen for immigrants from other macro-areas when stratified by sex. Supplementary Table S2 presents the results of the prevalence ratio for hypertension in the basic and fully adjusted models, overall and stratified by sex. The results confirmed the findings observed for blood pressure levels. ## 3.3. Associations between immigrant status and cardiovascular risk factors in relation to the migratory pathway Figures 2A, B shows the association between country of birth and TC and SBP in relation to duration of residence and age at migration by sex. Figure 2: β estimates and 95% confidence intervals of the associations between country of birth and total cholesterol (A) and systolic blood pressure (B) in relation to duration of residence and age at migration by sex. Immigrants who resided for <10 years in Italy had higher levels of TC compared to their native-born Italian counterparts. This difference gradually decreased with a longer stay, reaching significantly lower levels of TC for those who lived in Italy for more than 20 years. This trend was particularly clear for males, with those living in Italy <20 years showing higher levels of TC than Italian-born adults. Among females, the trend was less clear, with significantly lower levels only for those who resided in Italy between 10 and 20 years. Additionally, considering age at arrival, differences were seen for males only, with higher TC levels for males who arrived in Italy and were older than 18 years compared to native-born Italians (Panel a). When the macro-areas of origin were considered (Supplementary Table S3), immigrants from CE Europe and Asia had patterns similar to the overall pattern. For immigrants from CE Europe, higher levels of TC also were seen for those who arrived older than 18 years. In contrast, immigrants from Northern Africa showed an opposite trend: their advantage eroded with an increasing stay in the country and with a younger age at arrival. Immigrants who arrived under 18 years old had lower levels of SBP compared to native-born Italians, overall and for both sexes. This advantage persisted but was reduced for subjects who arrived after 18 years of age. The same advantage was seen in all groups when duration of residence was considered, but with a less clear gradient (Figure 2B). The greater advantage of those who arrived younger is maintained in subjects coming from all macro-areas of origin (Supplementary Table S3). ## 4. Discussion In our study, immigrant adult males showed higher cholesterol levels than their native-born Italian counterparts, while both immigrant males and females had lower BP levels. The higher levels of TC are mainly driven by the subjects from CE Europe, representing more than $50\%$ of all immigrants. Immigrant males from Asia had higher levels of TC, while immigrants from Northern Africa showed lower levels of cholesterol than native-born Italians, although the results were significant for immigrant females only. Immigrants from CE Europe and Northern Africa had lower levels of SBP than native-born Italians. After adjustments for known determinants (lifestyle factors and social determinants) of CV risk factors, the results maintained the same direction, although differences were attenuated in some cases and increased in other cases. These differences probably depend on the deep heterogeneity of the lifestyle behaviors of immigrants from different macro-areas. Regardless, the preserved direction in the adjusted results suggests that these results are only partially driven by environmental factors of the host country. Genetic factors and early living conditions can play a role in explaining these differences. The prevalence of CV risk factors, especially a high-calorie diet and cardiovascular disease, is high in Eastern Europe [16]. Therefore, when immigrants from these countries arrive in Italy, they already have a background of worse health behaviors than the native population. This could explain the higher levels of TC in subjects from CE Europe. In contrast, the reduced BP levels in immigrants from CE Europe are less expected. Our results are in contrast with the European estimates of BP levels, which indicate a $10\%$ higher prevalence of raised BP in Albania and Romania (the two most represented countries of birth in our sample; see Supplementary Table S4) than in Italy in 2014 [16]. A study conducted in the same Italian region found results similar to our study, with Italian citizens having the highest rates of hypertension compared to all groups of immigrants except those coming from SS Africa [5]. Regardless, it should be noted that the differences in BP levels between immigrants and natives are small in magnitude and that no differences were seen between immigrants from CE Europe and native-born Italians when considering the adjusted results for the prevalence of hypertension. A distinct pattern of cardiovascular risk could be identified among Asians, with higher rates of TC among males than among their native-born Italian counterparts. When interpreting the results of Asian immigrants, the composition within this category is worth noting. In fact, $85\%$ of this population in our study consisted of subjects born in the Indian subcontinent (Supplementary Table S4). The findings of our study are consistent with previous reports on ethnic differences in plasma lipid levels in the United States, Canada, the United Kingdom and the Netherlands, which have shown that South Asians are characterized by high LDL-C and triglyceride levels and low HDL-C compared to the reference population (10, 17–19). The reasons for this unfavorable lipid pattern seem to be both environmental and genetic. It has been demonstrated that South Asians in the US were less physically active and had lower adiponectin and higher resistin levels than Caucasians, resulting in higher levels of LDL-C [20]. Regarding immigrants from Northern Africa, in our cohort, they are mostly represented by subjects born in Morocco ($96\%$, see Supplementary Table S4). The low prevalence of dyslipidemia in Moroccans is consistent with the low LDL-C and TG levels found by Gazzola et al. in Moroccan descendants compared to the Dutch reference population [10]. Additionally, Moroccans are traditionally known to have a lower prevalence of hypertension in Europe, although this health advantage seems to be changing unfavorably through the acculturation process [21]. The favorable lifestyle and behavioral habits that are linked with the Muslim population (highly prevalent in Morocco) may represent an advantage reflecting a lower predisposition toward the development of dyslipidemia and hypertension. Differences in BP levels related to the prevalent religion have been previously found [11], and such an explanation also could be valid for the lipid pattern. Our results on cholesterol also showed relevant gender differences: males from CE Europe and Asia had higher TC levels than native-born Italians, while no differences were seen for their female counterparts. In contrast, females from Northern Africa had lower levels of TC than native-born Italians, and this difference was less clear for males. Additionally, the trends related to acculturation, with results approaching those of the native population in immigrants with a longer stay, are stronger in males than in females. These differences may depend on the prevalence of risky behaviors: smoking and alcohol intake are more prevalent in immigrant males than in females in our population. Additionally, in the group of immigrants from CE Europe, the rate of overweight or obese people is doubled among males compared with females. This is not true for immigrants from Northern Africa, though. Additionally, these differences could be related to different approaches toward health care. In a Dutch study, women were found to have higher levels of awareness, treatment and control of hypertension than men in all ethnic groups, and this has been attributed to frequent use of health care by women [21, 22]. Although environmental and socioeconomic factors play only partial roles in determining these disparities, an effort should be made to reduce the risky health behaviors and potential mediators of these differences. Many interventions in primary health care settings have been demonstrated to be effective for the early prevention of cardiovascular diseases. In particular, the higher BMI of the immigrant population could be reduced with interventions on nutritional habits and physical activity. Additionally, reduced access to primary health care (PHC) could lead to a later diagnosis and a later start of adequate therapy. Immigrants from HMPCs have a probability of an annual LDL-C test reduced by half compared to native-born Italians [12]. PHC services in Italy are, in theory, widely accessible and mostly free at the point of use; however, different economic resources might provide access to more timely private services, and different levels of education might affect people's approaches when managing their health problems [23]. Furthermore, the cultural backgrounds and health practices of immigrants may be dissimilar to those of European people and health professionals, and it is essential to take particular care when dealing with these patients. A clear gradient was seen for what concerns the levels of cholesterol by length of stay: immigrants who stayed longer than 20 years in Italy also had lower levels of TC than native-born Italians, while those who arrived <10 years ago had similar or higher levels of TC compared to native-born Italians. Additionally, immigrants who arrived younger than 18 years had lower levels of SBP than native-born Italians, and these results were attenuated in those who arrived older than 18 years. These results, too, were driven by the large group of immigrants from CE Europe in our population. Several studies suggest that acculturation is associated with a decline in healthy behaviors, resulting in an increase in CV risk factors [24, 25], although the evidence concerning the relationships among acculturation, lifestyle behaviors, and cardiovascular risk factors is not uniform, with some examples of convergence from initially higher levels of risk factors down to local levels (26–28). These different findings mainly depend on the ethnic groups considered [29] and the starting conditions in their country of origin. In our study, the acculturation of immigrants from CE Europe seems to be a protective factor against CV risk factors. As previously said, this could reflect the higher prevalence of the aforementioned CV risk factors in CE Europe. A longer stay in Italy, as well as an arrival at a younger age, could lead to an earlier and longer adoption of healthy behavior and result in lower CV risk factors. For other immigrant groups, the process of acculturation could lead to an increased risk. The health advantage that was seen for immigrants from Northern Africa, with lower levels of TC compared to native-born Italians, declines with an increasing length of stay and disappears for immigrants who arrived younger than 18 years. In this case, the obesogenic environment of the host country probably plays a major role in changing the virtuous lifestyle habits of immigrants. For the Asian group, the results are more heterogeneous, and it is difficult to draw any conclusion concerning this specific category. To our knowledge, this is one of the first studies conducted in Italy analyzing differences in biochemical and clinical parameters among native-born Italians and immigrants and the first that evaluates the effect of the migratory pathway in modifying these outcomes. It has been conducted on a large number of individuals and accounts for a wide number of potential characteristics associated with CV risk factors. The results also pave the way to future analysis considering other CV risk factors, such as glycated hemoglobin as indicator of the risk of developing diabetes mellitus. This could be an interesting point, to early identify subjects and groups at risk, and to tailor interventions before the development of the disease itself, which has been shown to be more prevalent in immigrants than natives [12]. The study also has some limitations. First, we do not know the response rates of immigrants and natives. The questionnaire was conducted in Italian, and it was very complex, so there could have been a selection bias in the responders. The questionnaires were administered by trained nurses and not self-administered, so nurses could have mediated some language barriers by explaining the questions. Additionally, when communication was hindered by language problems, the interviewed subject was invited to return a second time with an interpreter. Second, we do not have information on the citizenship of the responders, and the “immigrant” category was built on the country of birth alone. Therefore, we could have included in this category Italian citizens but incidentally born abroad, although it is probably a quite exceptional occurrence when considering developing countries. Third, the Surveillance Program does not include people born before 1951, and this study therefore does not include elderlies. This limitation can be relevant in the analysis with hypertension as outcome, since the prevalence of this disease raises with age, although it must be pointed out that the prevalence of immigrant subjects in Italy of this age group is minimal. In conclusion, it is clear that immigrants should not be considered a homogenous group when exploring their health outcomes. Tailored prevention and follow-up programs are required to address differences and to target the worse-off groups. Immigrants from CE Europe, especially those who arrived at older ages, arrive with a health disadvantage compared to native-born Italians, and programs raising awareness of worse habits can be useful to accelerate the process of acculturation. In contrast, immigrants from Northern Africa have a health advantage that should be preserved to avoid future increases in CV diseases. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Ethics statement The study was approved by Regional (Veneto Region) Ethics Committee (24 maggio 2017 prot. No. 203638). Written informed consent for participation was obtained from all subjects involved in the study. ## Author contributions TDZ wrote the manuscript, discussed, and interpreted the results of the data. EB performed the statistical analysis. TS, FR, and GP reviewed and edited the manuscript. CC is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data, and the accuracy of the data analysis. All authors have read and agreed to the published version of the manuscript. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh.2023.956146/full#supplementary-material ## References 1. 1.Demo-Geodemo. - Mappe, Popolazione, Statistiche Demografiche dell'ISTAT. Available online at: http://demo.istat.it/ (accessed September 23, 2019). 2. Razum O, Zeeb H, Rohrmann S. The “healthy migrant effect'–not merely a fallacy of inaccurate denominator figures. Int J Epidemiol. (2000.0) 29 191-2. DOI: 10.1093/ije/29.1.191 3. Rosano A. Access to Primary Care and Preventive Health Services of Migrants: Springer Briefs in Public Health (2018.0) 4. Lara M, Gamboa C, Kahramanian MI, Morales LS, Bautista DEH. Acculturation and Latino health in the United States: a review of the literature and its sociopolitical context. Annu Rev Public Health. (2005.0) 26 367-97. DOI: 10.1146/annurev.publhealth.26.021304.144615 5. Fedeli U, Avossa F, Ferroni E, Schievano E, Bilato C, Modesti PA. Diverging patterns of cardiovascular diseases across immigrant groups in Northern Italy. Int J Cardiol. (2018.0) 254 362-7. DOI: 10.1016/j.ijcard.2017.12.014 6. Uitewaal PJM, Manna DR, Bruijnzeels MA, Hoes AW, Thomas S. Prevalence of type 2 diabetes mellitus, other cardiovascular risk factors, and cardiovascular disease in Turkish and Moroccan immigrants in North West Europe: a systematic review. Prev Med. (2004.0) 39 1068-76. DOI: 10.1016/j.ypmed.2004.04.009 7. Ujcic-Voortman JK, Baan CA, Verhoeff AP, Krol A, Seidell JC. Ethnic differences in systemic inflammation: an investigation of C-reactive protein levels among Moroccan, Turkish and Dutch groups in the Netherlands. Atherosclerosis. (2011.0) 218 511-6. DOI: 10.1016/j.atherosclerosis.2011.06.051 8. Fryar CD, Fakhouri TH, Carroll MD, Frenk SM, Ogden CL. The association of nativity/length of residence and cardiovascular disease risk factors in the United States. Prev Med. (2020.0) 130 105893. DOI: 10.1016/j.ypmed.2019.105893 9. Galderisi A, Giannini C, Weiss R, Kim G, Shabanova V, Santoro N. Trajectories of changes in glucose tolerance in a multiethnic cohort of obese youths: an observational prospective analysis. Lancet Child Adolescent Health. (2018.0) 2 726-35. DOI: 10.1016/S2352-4642(18)30235-9 10. Gazzola K, Snijder MB, Hovingh GK, Stroes ESG, Peters RJG, van den Born B-JH. Ethnic differences in plasma lipid levels in a large multiethnic cohort: The HELIUS study. J Clin Lipidol. (2018.0) 12 1217-24. DOI: 10.1016/j.jacl.2018.06.015 11. Modesti PA, Reboldi G, Cappuccio FP, Agyemang C, Remuzzi G, Rapi S. Panethnic differences in blood pressure in Europe: a systematic review and meta-analysis. PLoS ONE. (2016.0) 11 e0147601. DOI: 10.1371/journal.pone.0147601 12. Valore Project, Buja A, Gini R, Visca M, Damiani G, Federico B. Prevalence of chronic diseases by immigrant status and disparities in chronic disease management in immigrants: a population-based cohort study, Valore Project. BMC Public Health (2013.0) 13 504. DOI: 10.1186/1471-2458-13-504 13. O'Donnell CJ, Elosua R. Cardiovascular risk factors. Insights from Framingham Heart Study. Rev Esp Cardiol. (2008.0) 61 299-310. DOI: 10.1016/S1885-5857(08)60118-8 14. Pitter G, Da Re F, Canova C, Barbieri G, Zare Jeddi M, Daprà F. Serum Levels of Perfluoroalkyl Substances (PFAS) in adolescents and young adults exposed to contaminated drinking water in the Veneto Region, Italy: a cross-sectional study based on a health surveillance program. Environ Health Perspect. (2020.0) 128 027007. DOI: 10.1289/EHP5337 15. Williams B, Mancia G, Spiering W, Agabiti Rosei E, Azizi M, Burnier M. 2018 ESC/ESH Guidelines for the management of arterial hypertension: The Task Force for the management of arterial hypertension of the European Society of Cardiology and the European Society of Hypertension: the task force for the management of arterial hypertension of the European Society of Cardiology and the European Society of Hypertension. J Hypertens. (2018.0) 36 1953-2041. DOI: 10.1097/HJH.0000000000001940 16. Wilkins E, Wilson L, Wickramasinghe K, Bhatnagar P, Leal J, Luengo-Fernandez R. European Cardiovascular Disease Statistics 2017 (2017.0) 17. Frank ATH, Zhao B, Jose PO, Azar KMJ, Fortmann SP, Palaniappan LP. Racial/ethnic differences in dyslipidemia patterns. Circulation. (2014.0) 129 570-9. DOI: 10.1161/CIRCULATIONAHA.113.005757 18. Anand SS, Yusuf S, Vuksan V, Devanesen S, Teo KK, Montague PA. Differences in risk factors, atherosclerosis, and cardiovascular disease between ethnic groups in Canada: the Study of Health Assessment and Risk in Ethnic groups (SHARE). Lancet. (2000.0) 356 6. DOI: 10.1016/S0140-6736(00)02502-2 19. Whitty CJM, Brunner EJ, Shipley MJ, Hemingway H, Marmot MG. Differences in biological risk factors for cardiovascular disease between three ethnic groups in the Whitehall II study. Atherosclerosis. (1999.0) 142 279-86. DOI: 10.1016/S0021-9150(98)00239-1 20. Shah AD, Kandula NR, Lin F, Allison MA, Carr J, Herrington D. Less favorable body composition and adipokines in South Asians compared with other US ethnic groups: results from the MASALA and MESA studies. Int J Obes (Lond). (2016.0) 40 639-45. DOI: 10.1038/ijo.2015.219 21. Agyemang C, Kieft S, Snijder MB, Beune EJ, van den Born BJ, Brewster LM. Hypertension control in a large multi-ethnic cohort in Amsterdam, The Netherlands: the HELIUS study. Int J Cardiol. (2015.0) 183 180-9. DOI: 10.1016/j.ijcard.2015.01.061 22. Gerritsen AA, Devillé WL. Gender differences in health and health care utilisation in various ethnic groups in the Netherlands: a cross-sectional study. BMC Public Health. (2009.0) 9 109. DOI: 10.1186/1471-2458-9-109 23. Di Cesare M, Khang YH, Asaria P, Blakely T, Cowan MJ, Farzadfar F. Inequalities in non-communicable diseases and effective responses. Lancet. (2013.0) 381 585-97. DOI: 10.1016/S0140-6736(12)61851-0 24. Rechel B. Migration and Health in European Union (2011.0) 25. Koya DL, Egede LE. Association between length of residence and cardiovascular disease risk factors among an ethnically diverse group of United States immigrants. J Gen Intern Med. (2007.0) 22 841-6. DOI: 10.1007/s11606-007-0163-y 26. Alkerwi A, Sauvageot N, Pagny S, Beissel J, Delagardelle C, Lair M-L. Acculturation, immigration status and cardiovascular risk factors among Portuguese immigrants to Luxembourg: findings from ORISCAV-LUX study. BMC Public Health. (2012.0) 12 864. DOI: 10.1186/1471-2458-12-864 27. Sundquist J, Winkleby MA. Cardiovascular risk factors in Mexican American adults: a transcultural analysis of NHANES III, 1988-1994. Am J Public Health. (1999.0) 89 723-30. DOI: 10.2105/AJPH.89.5.723 28. Juárez SP, Drefahl S, Dunlavy A, Rostila M. All-cause mortality, age at arrival, and duration of residence among adult migrants in Sweden: a population-based longitudinal study. SSM Popul Health. (2018.0) 6 16-25. DOI: 10.1016/j.ssmph.2018.07.007 29. Oza-Frank R, Venkat Narayan K. Effect of length of residence on overweight by region of birth and age at arrival among US immigrants. Public Health Nutr. (2010.0) 13 868-75. DOI: 10.1017/S1368980009992084
--- title: Dietary methionine restriction alleviates oxidative stress and inflammatory responses in lipopolysaccharide-challenged broilers at early age authors: - Xiyuan Pang - Zhiqiang Miao - Yuanyang Dong - Huiyu Cheng - Xiangqi Xin - Yuan Wu - Miaomiao Han - Yuan Su - Jianmin Yuan - Yuxin Shao - Lei Yan - Jianhui Li journal: Frontiers in Pharmacology year: 2023 pmcid: PMC9975741 doi: 10.3389/fphar.2023.1120718 license: CC BY 4.0 --- # Dietary methionine restriction alleviates oxidative stress and inflammatory responses in lipopolysaccharide-challenged broilers at early age ## Abstract In this study, we investigated the effect of dietary methionine restriction (MR) on the antioxidant function and inflammatory responses in lipopolysaccharide (LPS)-challenged broilers reared at high stocking density. A total of 504 one-day-old male Arbor Acre broiler chickens were randomly divided into four treatments: 1) CON group, broilers fed a basal diet; 2) LPS group, LPS-challenged broilers fed a basal diet; 3) MR1 group, LPS-challenged broilers fed a methionine-restricted diet ($0.3\%$ methionine); and 4) MR2 group, LPS-challenged broilers fed a methionine-restricted diet ($0.4\%$ methionine). LPS-challenged broilers were intraperitoneally injected with 1 mg/kg body weight (BW) of LPS at 17, 19, and 21 days of age, whereas the CON group was injected with sterile saline. The results showed that: LPS significantly increased the liver histopathological score ($p \leq 0.05$); LPS significantly decreased the serum total antioxidant capacity (T-AOC), superoxide dismutase (SOD), and glutathione peroxidase (GSH-Px) activity at 3 h after injection ($p \leq 0.05$); the LPS group had a higher content of Interleukin (IL)-1β, IL-6, and tumor necrosis factor-α (TNF)-α, but a lower content of IL-10 than the CON group in serum ($p \leq 0.05$). Compared with the LPS group, the MR1 diet increased catalase (CAT), SOD, and T-AOC, and the MR2 diet increased SOD and T-AOC at 3 h after injection in serum ($p \leq 0.05$). Only MR2 group displayed a significantly decreased liver histopathological score ($p \leq 0.05$) at 3 h, while MR1 and MR2 groups did so at 8 h. Both MR diets significantly decreased serum LPS, CORT, IL-1β, IL-6, and TNF-α contents, but increased IL-10 content ($p \leq 0.05$). Moreover, the MR1 group displayed significantly increased expression of nuclear factor erythroid 2-related factor 2 (Nrf2), CAT, and GSH-Px at 3 h; the MR2 group had a higher expression of Kelch-like ECH-associated protein 1 (Keap1), SOD, and GSH-Px at 8 h ($p \leq 0.05$). In summary, MR can improve antioxidant capacity, immunological stress, and liver health in LPS-challenged broilers. The MR1 and MR2 groups experienced similar effects on relieving stress; however, MR1 alleviated oxidative stress more rapidly. It is suggested that precise regulation of methionine levels in poultry with stress may improve the immunity of broilers, reduce feed production costs, and increase production efficiency in the poultry industry. ## Introduction Bacterial diseases are an increasingly severe problem that restricts the development of the poultry industry. With increased stocking density in intensive production, the immune competence of birds decreases, making the body much more susceptible to bacterial diseases (Averós and Estevez, 2018; Goo et al., 2019). Lipopolysaccharide (LPS) is one of the main pathogenic factors of bacterial diseases in poultry. It is a component of the cell wall of Gram-negative bacteria, which can reduce the antioxidant capacity of poultry, cause inflammation, and produce inflammatory cytokines. ( Leshchinsky and Klasing, 2001; Takahashi et al., 2011). After infection, LPS molecules in pathogens are recognized and regulated by lipopolysaccharide binding protein (LBP)/CD14 in the cell membrane to form the LPS-LBP complex, which binds to CD14/TLRs receptors on the cell surface to induce tumor necrosis factor (TNF)-α, interleukin (IL)-1β, and other cytokines, resulting in inflammation (Bryant et al., 2010). In addition, LPS activates phospholipase C (PLC) and releases diglycerol and inositol 1, 4, 5-Trisphosphate (IP3) in Kupffer’s cells. Inositol 1, 4, 5-Trisphosphate induces an increase in intracellular Ca2+ concentration. The increase in intracellular Ca2+ increases the activity of reactive oxygen species (ROS)-producing enzymes and accelerates the synthesis of free radicals through the mitochondrial respiratory chain, causing an oxidative stress response (Jin et al., 2006). It has been reported that LPS challenge reduces the antioxidant level and antioxidant enzyme activity and improves the level of inflammatory factors such as IL-1β and IL-6 in the serum of broilers (Zheng et al., 2016; Zheng et al., 2020). In addition, in vitro studies reported that excessive accumulation of inflammatory cytokines induced by LPS can damage pig intestinal epithelial cells and lead to intestinal dysfunction (Li et al., 2022). Therefore, alleviating the inflammatory response and oxidative stress caused by avian bacterial diseases is an urgent problem in the feeding industry. Methionine (MET), the first restricted amino acid in broilers, plays a crucial role in regulating the poultry’s growth and development, antioxidant capacity, and immune function (Xie et al., 2007). However, recent studies have discovered that reducing the dietary methionine level, known as methionine restriction (MR), can also improve the anti-inflammatory and antioxidant capacities of animals (Fang et al., 2022). It has been reported that proper MR (the methionine level in the diet is $60\%$–$80\%$ of the NRC level) can significantly improve intestinal antioxidant capacity, gene expression of the intestinal innate immune system, and intestinal flora structure of broilers (Yu, 2013). MR has been shown to reduce the plasma concentrations of LPS and LBP and the mRNA expression of ileal TNF-α and IL-6 genes in mice fed a high-fat diet, indicating that MR could reduce the inflammatory response by limiting the expression of inflammatory factors (Wu et al., 2020a). Further, MR activates nuclear factor erythroid 2-related factor 2 (Nrf2), and the increased activity of Nrf2 and the enhanced expression of the accompanying genes related to the antioxidant response will improve the antioxidant and detoxification abilities of the body (Brown-Borg and Buffenstein, 2017). In addition, MR alleviates the infiltration of inflammatory factors and tissue damage in LPS-challenged mice by promoting hydrogen sulfide production (Duan et al., 2022). However, whether MR can protect broiler chicks against LPS-induced stress has not been reported. Based on these findings, a double-stress model was used to study the effect of MR on the growth performance, liver antioxidant function, and inflammatory response in LPS-induced stress in young broilers reared at high stocking density. Our findings will help further optimize the methionine nutrition scheme of broilers infected with bacterial diseases under high-density feeding conditions and provide useful insights for alleviating production stress in broilers. ## Reagents, animals, and experimental design LPS (Escherichia coli 055: B5) was obtained from Sigma Aldrich (St. Louis, MO, United States) and reconstituted in saline at a dose of 1.0 mg/mL. A total of 504 one-day-old male Arbor Acre broiler chickens were randomly assigned to four treatments (six replicates of 21 birds per cage) in a completely randomized design. The experimental diets consisted of 1) Control group with a basal diet (CON), 2) LPS challenged group with basal diet (LPS), 3) LPS challenged group with a $0.3\%$ methionine restriction diet (MR1), and 4) LPS challenged group with a $0.4\%$ methionine restriction diet (MR2). All broilers were reared under high stocking density (30 birds/m2) in 0.7 m2 pens. At 17, 19, and 21 days, the LPS-challenged groups were intraperitoneally injected with 1 mg/kg body weight (BW) LPS solution, whereas the CON group received an equivalent volume of sterile saline injection ($0.9\%$). The birds were provided access to mash feed and water ad libitum during the 21 days of the experiment. For the first 3 days, the chicken house was kept under 24 h of continuous light, followed by 20 h of continuous light and 4 h of darkness from day 4 to day 21. The temperature was maintained at 33°C from days 1–3, and was gradually decreased by 2°C–3°C every week until it reached at 26°C and then was kept constant. In addition, all birds were vaccinated according to a routine immunization program. The present study was approved by the Animal Health and Care Committee of the Shanxi Agricultural University (Shanxi, China) and conducted according to the Guidelines for the Experimental Animal Welfare of Ministry of Science Technology of China (approval code. SXAU-EAW-2022Po. SD.01129001). ## Sample collection On days 1, 17, and 21 of the experiment, the BW and total feed consumption of broilers were recorded to calculate average daily feed intake (ADFI), average daily gain (ADG), and feed conversion ratio (FCR) before (from 1 to 16 days of age) and after (from 17 to 21 days of age) the LPS challenge. The FCR = ADFI: ADG. At 21 days of age, one bird from each replicate was selected for slaughter 3 and 8 h after LPS injection. Blood samples from the vein under the wing were collected into 5 mL anticoagulant-free vacutainer tubes. After centrifugation at 3,000 × g for 10 min at 4°C, the serum in the tube was collected and stored at −80°Cfor analysis. After blood sampling, broilers were euthanized by cervical dislocation and exsanguination for dissection of the liver, spleen, bursa, and thymus. The liver tissues were flash-frozen in liquid nitrogen and then preserved at −80°C for further analysis. ## Formulation of basal diets The basal diet met the National Research Council [1994] nutrient requirements for broilers. The ingredient composition and nutrient levels are listed in Table 1. TABLE 1 \| Items \| Contents, % \| \| --- \| --- \| \| Ingredients \| Ingredients \| \| Corn \| 54.70 \| \| Soybean meal \| 33.50 \| \| Corn protein powder \| 3.00 \| \| Flour \| 2.00 \| \| Soybean Oil \| 2.40 \| \| Dicalcium phosphate \| 1.40 \| \| Limestone \| 1.35 \| \| Lys \| 0.38 \| \| NaCl \| 0.30 \| \| 50% Choline chloride \| 0.20 \| \| Thr \| 0.11 \| \| 98.5% arginine \| 0.04 \| \| Trace mineral 1 \| 0.20 \| \| Vitamin permmix 2 \| 0.03 \| \| DL-Met(99%) \| 0.20 \| \| Phytase \| 0.01 \| \| Zeolite powder \| 0.22 \| \| Total \| 100.00 \| \| Cacµgated nutrient level 3 \| Cacµgated nutrient level 3 \| \| ME (Kcal/kg) \| 2979 \| \| CP \| 21.80 \| \| Ca \| 0.90 \| \| Available phosPhorus \| 0.35 \| \| Total phosPhorus \| 0.67 \| \| Lys \| 1.24 \| \| Met \| 0.50 \| \| Lys+Cys \| 0.71 \| \| Thr \| 0.91 \| \| Trp \| 0.23 \| ## Organ index The organ indices of the liver, spleen, bursa, and thymus were calculated using the following equation (Deng et al., 2022) organ index=organ weight g÷live body weight kg As shown in Table 4, at 3 h after injection, the spleen organ indices of broilers injected with LPS were increased ($p \leq 0.05$) and the thymus organ indices were decreased ($p \leq 0.05$) compared with the non-challenged broilers. The methionine-restricted diet significantly increased the thymus organ index ($p \leq 0.05$). The MR2 group showed a significantly increased in the spleen organ index ($p \leq 0.05$). At 8 h after injection, LPS challenge increased the spleen organ index ($p \leq 0.05$). The liver and bursa indices were not significantly different among the groups ($p \leq 0.05$). TABLE 4 \| Items 1 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| liver \| bursa \| spleen \| thymus \| liver \| bursa \| spleen \| thymus \| \| CON \| 30.82 ± 1.43 \| 1.28 ± 0.10 \| 0.86 ± 0.13b \| 3.11 ± 0.10a \| 28.03 ± 1.66 \| 1.38 ± 0.11 \| 0.90 ± 0.18b \| 2.61 ± 0.28 \| \| LPS \| 30.35 ± 1.02 \| 1.43 ± 0.15 \| 1.34 ± 0.05a \| 2.23 ± 0.28b \| 31.17 ± 0.88 \| 1.64 ± 0.13 \| 1.39 ± 0.17a \| 2.10 ± 0.29 \| \| MR1 \| 35.27 ± 1.28 \| 1.26 ± 0.16 \| 1.09 ± 0.10ab \| 2.90 ± 0.33a \| 33.03 ± 1.36 \| 1.79 ± 0.15 \| 0.86 ± 0.04ab \| 2.59 ± 0.33 \| \| MR2 \| 31.78 ± 1.21 \| 1.47 ± 0.07 \| 1.27 ± 0.15a \| 2.83 ± 0.22a \| 30.35 ± 0.42 \| 1.45 ± 0.08 \| 1.11 ± 0.14ab \| 2.83 ± 0.23 \| \| p-value \| 0.085 \| 0.563 \| 0.020 \| 0.031 \| 0.072 \| 0.101 \| 0.032 \| 0.339 \| ## Serum biochemistry The serum levels of aspartate aminotransferase (AST), alanine aminotransferase (ALT), γ-glutamyl transpeptidase (γ-GT), total protein (TP), and the albumin to globulin ratio (A:G) were quantified using a chemistry analyzer BS-800 (Shenzhen Mindray Bio-medical Electronics Co., Ltd., Shenzhen, China). ## Liver histology analysis The liver of each bird was harvested, stored in $10\%$ buffered formalin, and embedded in paraffin. For histological investigations, 3 μm sections were cut, deparaffinized, dehydrated, and stained with hematoxylin and eosin (H&E) (Ma et al., 2022). The tissue sections were observed for inflammatory changes under a light microscope, Leica DM 2500 M (Leica Microsystems, Wetzlar, Germany). Liver tissues were observed for focal inflammation, mononuclear infiltration, loss of normal architecture, and necrosis. Acute liver injury was assessed using a composite inflammation and necrosis score as described by Siegmund et al. [ 2002]. Lobular inflammation (0, no inflammation; 1, mild; 2, moderate; 3, severe), portal inflammation (0, no inflammation; 1, mild; 2, moderate; 3, severe), and necrosis (0, no necrosis; 1, ˂$10\%$ of liver parenchyma; 2, $10\%$–$25\%$ of liver parenchyma; 3, ˃$25\%$ of liver parenchyma) were scored and summed up to determine the overall histopathology score. Six 3 μm sections from each treatment group were scored by a trained, blinded researcher. ## Antioxidant capacity Serum samples were used to assay total antioxidant capacity (T-AOC, #A015-2-1) and malondialdehyde (MDA, #A003-1) concentrations. The activities of the antioxidant enzymes superoxide dismutase (SOD, #A001-3), glutathione peroxidase (GSH-Px, #A005-1-2), and catalase (CAT, #A007-1-1) were measured using commercial assay kits (Nanjing Jiancheng, Institute of Bioengineering, Nanjing, China) according to the manufacturer’s instructions. One Gram of liver tissue preserved at −80°C was homogenized with 9 mL of $0.9\%$ ice-cold sodium chloride buffer and then centrifuged at 4,000 × g for 10 min at 4°C, and the supernatant was collected and analyzed immediately. The total antioxidant capacity (T-AOC, #A015-1-2) and activities of superoxide dismutase (T-SOD, #A001-1-2), glutathione peroxidase (GSH-Px, #A005-1-2), and catalase (CAT, #A007-1-1), as well as malondialdehyde (MDA, #A003-1) content, were measured. All readings were obtained using a microplate reader (Synergy H1; BioTek, Winooski, VT, United States). ## Serum stress indices and inflammatory factors The concentrations of corticosterone, endotoxin, IL-1β, TNF-α, IL-6, and IL-10 in serum samples were measured using chicken-specific quantification ELISA kits purchased from Shanghai Enzyme-linked Biotechnology Co., Ltd. (Shanghai, China). ## Quantification of messenger ribonucleic acid (mRNA) by real-time PCR Total RNA from liver samples (about 50–100 mg) was extracted by adding 1 mL of TRIzol-regent (#R601-03; TaKaRa Biotechnology, Dalian, Liaoning, China) according to the manufacturer’s instructions. The concentration and purity of total RNA were assessed using a spectrophotometer (NanoDrop Technologies, Wilmington, DE, United States). Subsequently, the RNA was reverse transcribed to cDNA using the PrimeScripte™ RT reagent Kit (Perfect Real Time, SYBR® PrimeScriP™ TaKaRa, China, #3894A) according to the manufacturer’s instructions. The cDNA samples were amplified by qRT-PCR using the SYBR Premix Ex Taq reagent (Takara Biotechnology). The real-time PCR cycling conditions were as follows: 95°C for 30 s, 40 cycles of 95°C for 5 s, and 60°C for 30 s. The mRNA expression of target genes relative to beta-actin (β-actin) was calculated using the 2−ΔΔCT method (Livak and Schmittgen, 2001). All the primer sequences are listed in Table 2 and were synthesized by Sangon Biotech Co., Ltd. (Shanghai, China). TABLE 2 \| Genes \| Primers sequence (5′-3′) \| Accession number \| \| --- \| --- \| --- \| \| Nrf2 \| F: TTCGCAGAGCACAGATACTTC \| NM_205117.1 \| \| Nrf2 \| R: TGGGTGGCTGAGTTTGATTAG \| NM_205117.1 \| \| Keap1 \| F: CTGCTGGAGTTCGCCTACAC \| XM_025145847.1 \| \| Keap1 \| R: CACGCTGTCGATCTGGTACA \| XM_025145847.1 \| \| SOD \| F: GGTCATCCACTTCCAGCAGCAG \| U28407.1 \| \| SOD \| R: TAAACGAGGTCCAGCATTTCCAGTTAG \| U28407.1 \| \| CAT \| F: TTGCTATACGGTTCTCCACTGTTGC \| NM_001031215.2 \| \| CAT \| R: GTAAAGACTCAGGGCGAAGACTCAAG \| NM_001031215.2 \| \| GSH-Px \| F: GACCAACCCGCAGTACATCA \| NM_001277853.3 \| \| GSH-Px \| R: GAGGTGCGGGCTTTCCTTTA \| NM_001277853.3 \| \| β-Actin \| F: GCTACAGCTTCACCACCACA \| NM_205518.2 \| \| β-Actin \| R: TCTCCTGCTCGAAATCCAGT \| NM_205518.2 \| ## Statistical analysis All data were preliminarily processed using the Excel 2010 software. Statistical analyses were performed using the statistical software SPSS 26.0, with replicates ($$n = 6$$) as an experimental unit, and the results are presented as mean ± SEM. Data were normally distributed (Shapiro-Wilk test) and evaluated for homogeneity of variance (Levene’s test). Statistical analyses were performed using one-way analysis of variance (ANOVA) followed by Tukey’s test. Differences were considered statistically significant at $p \leq 0.05.$ ## Growth performance During the LPS challenge (17–21 days of age), the ADG of broilers were decreased significantly compared with that of the CON group ($p \leq 0.05$). Moreover, the FCR of the MR1 group was significantly lower than that of the LPS group ($p \leq 0.05$). No significant difference was observed in the growth performance of broilers within the trial period (0–21 days of age), ($p \leq 0.05$) (Table 3). TABLE 3 \| Items 1 \| 17-21d \| 17-21d.1 \| 17-21d.2 \| 0-21d \| 0-21d.1 \| 0-21d.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| ADFI/g \| ADG/g \| FCR \| ADFI/g \| ADG/g \| FCR \| \| CON \| 87.98 ± 0.09 \| 57.13 ± 1.17a \| 1.54 ± 0.03ab \| 49.04 ± 1.43 \| 32.55 ± 1.22 \| 1.51 ± 0.01 \| \| LPS \| 78.16 ± 4.71 \| 46.06 ± 2.04b \| 1.69 ± 0.03a \| 48.07 ± 0.79 \| 31.43 ± 0.21 \| 1.53 ± 0.03 \| \| MR1 \| 77.87 ± 4.54 \| 51.36 ± 3.05ab \| 1.52 ± 0.03b \| 44.07 ± 1.01 \| 28.41 ± 1.22 \| 1.55 ± 0.03 \| \| MR2 \| 77.90 ± 3.40 \| 50.79 ± 1.44ab \| 1.60 ± 0.07ab \| 46.06 ± 2.23 \| 30.15 ± 1.27 \| 1.53 ± 0.03 \| \| p-value \| 0.218 \| 0.033 \| 0.089 \| 0.159 \| 0.113 \| 0.704 \| ## Serum biochemical parameters As shown in Table 5, serum biochemical indicators did not differ significantly among the four groups at 3 h ($p \leq 0.05$). However, 8 h after the LPS injection, the serum biochemical indicators of AST and ALT were significantly higher in the LPS group than in the CON group ($p \leq 0.05$). In contrast, the MR1 group displayed significantly decreased AST activity and TP content compared with that of the LPS group ($p \leq 0.05$). TABLE 5 \| Items 1 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 3 h after LPS stimulation.4 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| 8 h after LPS stimulation.4 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| ALT (ng/L) \| AST (ng/L) \| A/G \| TP (g/L) \| γ-GT (U/L) \| ALT (ng/L) \| AST (ng/L) \| A/G \| TP (g/L) \| γ-GT (U/L) \| \| CON \| 4.73 ± 0.46 \| 301.13 ± 7.05 \| 0.62 ± 0.03 \| 27.52 ± 0.89 \| 13.40 ± 0.75 \| 3.43 ± 0.40b \| 268.78 ± 10.00b \| 0.61 ± 0.10 \| 28.42 ± 1.11a \| 14.40 ± 2.24 \| \| LPS \| 4.30 ± 0.30 \| 294.78 ± 14.17 \| 0.61 ± 0.02 \| 26.40 ± 1.56 \| 13.52 ± 1.49 \| 4.52 ± 0.54a \| 309.22 ± 9.96a \| 0.62 ± 0.02 \| 28.97 ± 0.88a \| 12.5 ± 0.20 \| \| MR1 \| 4.12 ± 0.30 \| 294.43 ± 2.80 \| 0.63 ± 0.02 \| 24.10 ± 1.08 \| 12.90 ± 0.84 \| 43.5 ± 0.37ab \| 272.27 ± 8.05b \| 0.58 ± 0.02 \| 24.73 ± 1.07b \| 11.97 ± 0.97 \| \| MR2 \| 3.93 ± 0.34 \| 301.37 ± 8.93 \| 0.63 ± 0.03 \| 24.05 ± 1.27 \| 12.43 ± 1.05 \| 4.45 ± 0.31ab \| 290.68 ± 7.91ab \| 0.61 ± 0.02 \| 28.73 ± 1.38b \| 15.93 ± 1.40 \| \| p-value \| 0.448 \| 0.912 \| 0.955 \| 0.149 \| 0.885 \| 0.089 \| 0.018 \| 0.956 \| 0.047 \| 0.208 \| ## Liver damage Three hours after the LPS injection, the CON group presented with a normal liver architecture with little or no cell infiltration and minimal vacuolar degeneration, as well as neatly arranged hepatic cords. In LPS-challenged chickens, liver injury is characterized by infiltration of inflammatory cells such as heterophilic cells and lymphocytes, moderate and diffuse vacuolar degeneration, disorderly arrangement of hepatic cords, and more cells with nuclear fragmentation and necrosis. Compared with the LPS group, the hepatic cord structures of the MR1 and MR2 groups were more evident, no significant inflammatory cells were noted, and vacuolation degeneration was reduced (Figure 1). The assessment of liver sections from LPS-challenged chickens exhibited a significantly increased mean histopathology score of 4.67 ± 0.21, compared to a mean score of 2.50 ± 0.22 ($p \leq 0.05$) for the CON group. The MR2 diet showed significant suppression of acute liver injury with mean histopathology scores of 2.50 ± 0.22. The MR1 group also had a tendency to alleviate liver damage with scores of 3.50 ± 0.43 (Figure 2). FIGURE 1: Effects of methionine restriction on histopathological analysis of liver in broilers stimulated by LPS (H and E staining, 400 ×). The yellow arrows point to inflammatory cells and the red scissors point to the vacuolar degeneration. CON, a basal diet plus intraperitoneal administration of sterile saline; LPS, a basal diet plus intraperitoneal administration of LPS; MR1, 0.3% mehtionine restriction diet plus intraperitoneal administration of LPS; MR2, 0.4% mehtionine restriction diet in combination with intraperitoneal administration of LPS. 3 h, 3 h after LPS stimulation; 8 h, 8 h after LPS stimulation. FIGURE 2: Effects of methionine restriction on histopathological score of liver in broilers stimulated by LPS. Data was shown as the mean ± SEM from six independent experiments. a–cThese bars without the same letter indicate differences significant at p < 0.05. CON, a basal diet plus intraperitoneal administration of sterile saline; LPS, a basal diet plus intraperitoneal administration of LPS; MR1, 0.3% mehtionine restriction diet plus intraperitoneal administration of LPS, MR2, 0.4% mehtionine restriction diet in combination with intraperitoneal administration of LPS. 3 h, 3 h after LPS stimulation; 8 h, 8 h after LPS stimulation. At 8 h, necrosis and moderate vacuolar degeneration were observed in the LPS group, around which different levels of inflammatory cell infiltration were observed. In comparison, the livers of the MR-treated chickens showed significant inhibition of the parameters of inflammatory damage and maintained normal liver architecture (Figure 1) ($p \leq 0.05$). Compared with a mean score of 2.50 ± 0.22 for the CON group, the LPS group showed a significantly increased mean histopathology score of 4.17 ± 0.41 ($p \leq 0.05$). However, MR1 and MR2 diets significantly reduced the degree of liver damage, shown in mean histopathology scores of 2.50 ± 0.22 and 2.33 ± 0.21 respectively (Figure 2) ($p \leq 0.05$). ## Liver and serum antioxidant capacity At 3 h after LPS injection, the LPS challenge had no significant effect on liver antioxidant capacity ($p \leq 0.05$). However, the MR1 group significantly showed increased GSH-Px activity compared to that of the LPS group ($p \leq 0.05$). Lipopolysaccharide decreased T-AOC and CAT activity, and increased MDA content 8 h after LPS injection ($p \leq 0.05$). Compared to the LPS group, the MR1 group displayed significantly increased T-AOC levels and CAT, T-SOD, and GSH-Px activity, and decreased MDA content in the liver of broilers ($p \leq 0.05$). In addition, broilers in the MR2 group had higher CAT activity and lower MDA content than those in the LPS group ($p \leq 0.05$) (Table 6). TABLE 6 \| Items 1 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 3 h after LPS stimulation.4 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| 8 h after LPS stimulation.4 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| TAOC (U/mgprot) \| MDA (nmol/mgprot) \| CAT (U/mgprot) \| T-SOD (U/mgprot) \| GSH-Px (U/mgprot) \| TAOC (U/mgprot) \| MDA (nmol/mgprot) \| CAT (U/mgprot) \| T-SOD (U/mgprot) \| GSH-Px (U/mgprot) \| \| CON \| 1.60 ± 0.09 \| 0.49 ± 0.05 \| 10.12 ± 2.87 \| 6.89 ± 0.35 \| 42.11 ± 2.49ab \| 1.22 ± 0.13a \| 0.32 ± 0.05b \| 8.43 ± 0.36a \| 5.70 ± 0.42b \| 39.64 ± 5.28ab \| \| LPS \| 1.87 ± 0.11 \| 0.50 ± 0.04 \| 5.02 ± 1.34 \| 5.95 ± 0.48 \| 37.29 ± 2.45b \| 0.67 ± 0.06b \| 0.72 ± 0.08a \| 6.28 ± 0.58b \| 5.85 ± 0.14b \| 30.67 ± 2.89b \| \| MR1 \| 1.60 ± 0.09 \| 0.54 ± 0.06 \| 10.03 ± 2.07 \| 6.69 ± 0.59 \| 47.86 ± 4.22a \| 1.16 ± 0.07a \| 0.40 ± 0.04b \| 8.51 ± 0.68a \| 8.16 ± 0.83a \| 43.84 ± 4.46a \| \| MR2 \| 1.86 ± 0.14 \| 0.60 ± 0.10 \| 7.14 ± 1.85 \| 6.97 ± 0.61 \| 37.19 ± 2.72b \| 0.94 ± 0.08ab \| 0.40 ± 0.09b \| 7.96 ± 0.48a \| 7.46 ± 0.42ab \| 38.57 ± 3.71ab \| \| p-value \| 0.149 \| 0.660 \| 0.384 \| 0.502 \| 0.024 \| 0.001 \| 0.003 \| 0.047 \| 0.006 \| 0.040 \| In the case of serum, LPS treatment significantly decreased T-AOC, SOD, and GSH-Px activity at 3 h after LPS injection ($p \leq 0.05$). However, broilers in the MR1 group had higher CAT and SOD activities and T-AOC levels than those in the LPS group ($p \leq 0.05$). In addition, broilers in the MR2 group had higher SOD activity and T-AOC levels than those in the LPS group ($p \leq 0.05$). At 8 h after stimulation, LPS significantly increased ($p \leq 0.05$) MDA content and decreased ($p \leq 0.05$) T-AOC and GSH-Px activity compared with that of the CON group. The MR1 group showed significantly increased T-AOC and SOD activity but decreased MDA content compared to that of the LPS group ($p \leq 0.05$). Meanwhile, the MR2 group showed significantly increased T-AOC and decreased MDA content ($p \leq 0.05$) compared to that of the LPS group (Table 7). TABLE 7 \| Items 1 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 3 h after LPS stimulation.4 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| 8 h after LPS stimulation.4 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| T-AOC (U/mL) \| MDA (nM) \| CAT (ng/L) \| SOD (pg/mL ) \| GSH-Px (IU/L) \| T-AOC (U/mL) \| MDA (nM) \| CAT (ng/L) \| SOD (pg/mL) \| GSH-Px (IU/L) \| \| CON \| 1.44 ± 0.03a \| 2.34 ± 0.19 \| 2.29 ± 0.28ab \| 110.29 ± 4.59a \| 311.39 ± 8.96a \| 1.63 ± 0.05a \| 2.34 ± 0.19b \| 2.20 ± 0.21 \| 104.22 ± 19.33ab \| 340.18 ± 7.01a \| \| LPS \| 1.21 ± 0.05b \| 2.66 ± 0.18 \| 1.44 ± 0.34b \| 80.07 ± 3.70b \| 258.22 ± 7.37b \| 1.36 ± 0.05b \| 2.84 ± 0.17a \| 2.20 ± 0.24 \| 92.86 ± 5.26b \| 270.73 ± 8.93b \| \| MR1 \| 1.55 ± 0.07a \| 2.75 ± 0.13 \| 2.87 ± 0.33a \| 107.49 ± 6.26a \| 269.69± 4.44b \| 1.59 ± 0.04a \| 2.39 ± 0.11b \| 2.34 ± 0.18 \| 119.59 ± 7.25a \| 265.61 ± 3.81b \| \| MR2 \| 1.56 ± 0.05a \| 2.39 ± 0.09 \| 1.75 ± 0.35ab \| 117.86 ± 5.35a \| 259.88 ± 2.45b \| 1.64 ± 0.05a \| 2.39 ± 0.09b \| 2.80 ± 0.34 \| 116.67 ± 8.97ab \| 269.43 ± 5.96b \| \| p-value \| 0.001 \| 0.203 \| 0.029 \| 0.001 \| <0.001 \| 0.001 \| 0.035 \| 0.598 \| 0.036 \| <0.001 \| ## Serum stress indexes and pro-inflammation factors As showen in Table 8, the LPS group had a higher serum LPS, CORT, IL-1β, IL-6, and TNF-αcontent, but lower levels of IL-10 than the CON group at 3 h and 8 h after LPS stimulation ($p \leq 0.05$). Meanwhile, MR groups (MR1 and MR2) significantly decreased LPS, CORT, IL-1β, IL-6, and TNF-α, but increased IL-10 contents compared with the LPS group at both time points ($p \leq 0.05$). TABLE 8 \| Items 1 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 3 h after LPS stimulation.4 \| 3 h after LPS stimulation.5 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| 8 h after LPS stimulation.4 \| 8 h after LPS stimulation.5 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| LPS (ng/L) \| CORT (ng/L) \| IL-1β (ng/L) \| IL-6 (ng/L) \| TNF-α (ng/L) \| IL-10 (ng/L) \| LPS (ng/L) \| CORT (ng/L) \| IL-1β (ng/L) \| IL-6 (ng/L) \| TNF-α (ng/L) \| IL-10 (ng/L) \| \| CON \| 258.15 ± 2.71b \| 418.01 ± 6.26b \| 130.44 ± 2.46b \| 50.55 ± 0.98b \| 68.45 ± 1.55b \| 61.72 ± 1.12b \| 269.02 ± 6.05b \| 423.22 ± 5.08b \| 136.72 ± 3.22b \| 52.63 ± 0.77b \| 71.58 ± 1.35b \| 57.30 ± 0.66b \| \| LPS \| 292.29 ± 4.82a \| 466.36 ± 10.37a \| 145.42 ± 2.82a \| 59.59 ± 0.97a \| 78.73 ± 1.79a \| 48.71 ± 1.36c \| 305.89 ± 2.57a \| 484.38 ± 7.79a \| 155.28 ± 1.16a \| 63.60 ± 1.79a \| 81.98 ± 0.61a \| 45.50 ± 1.99c \| \| MR1 \| 244.60 ± 4.60b \| 395.64 ± 8.47bc \| 126.19 ± 2.04b \| 47.88 ± 0.95bc \| 67.72 ± 1.50b \| 62.22 ± 1.16ab \| 253.24 ± 4.91b \| 413.72 ± 6.54b \| 132.31 ± 2.76b \| 49.87 ± 0.57bc \| 67.57 ± 1.19bc \| 61.90 ± 1.02ab \| \| MR2 \| 224.89 ± 3.78c \| 372.46 ± 9.43c \| 121.17 ± 2.75b \| 44.58 ± 0.91c \| 64.68 ± 1.08b \| 67.18 ± 1.52a \| 253.72 ± 3.58b \| 384.88 ± 6.11c \| 131.65 ± 2.63b \| 48.95 ± 0.70c \| 64.81 ± 1.33c \| 63.62 ± 1.28a \| \| p-value \| <0.001 \| <0.001 \| <0.001 \| <0.001 \| <0.001 \| <0.001 \| <0.001 \| <0.001 \| 0.001 \| <0.001 \| <0.001 \| <0.001 \| ## mRNA expression of antioxidant genes in the liver At 3 h after injection, the level of CAT gene expression in the LPS group significantly decreased ($p \leq 0.05$). However, the MR1 group showed significantly increased expression of Nrf2, CAT, and GSH-Px genes ($p \leq 0.05$). In addition, the MR2 group showed a higher level of CAT gene expression ($p \leq 0.05$). At 8 h, the LPS group showed no significantly affected gene expression levels ($p \leq 0.05$), but the MR1 group showed significantly increased expression of Nrf2, Keap1, SOD, and GSH-Px ($p \leq 0.05$). Meanwhile, the MR2 group had higher expression of Keap1, SOD, and GSH-Px ($p \leq 0.05$) (Table 9). TABLE 9 \| Unnamed: 0 \| 3 h after LPS stimulation \| 3 h after LPS stimulation.1 \| 3 h after LPS stimulation.2 \| 3 h after LPS stimulation.3 \| 3 h after LPS stimulation.4 \| 8 h after LPS stimulation \| 8 h after LPS stimulation.1 \| 8 h after LPS stimulation.2 \| 8 h after LPS stimulation.3 \| 8 h after LPS stimulation.4 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Items 1 \| Nrf2 \| Keap1 \| CAT \| SOD \| GSH-Px \| Nrf2 \| Keap1 \| CAT \| SOD \| GSH-Px \| \| CON \| 0.97 ±0.12ab \| 1.01 ± 0.07 \| 1.00 ± 0.03a \| 1.07 ± 0.08b \| 1.05 ± 0.13b \| 1.15 ± 0.18b \| 1.24 ± 0.29b \| 1.06 ± 0.14 \| 1.03 ± 0.13ab \| 1.02 ± 0.08b \| \| LPS \| 0.73 ± 0.14b \| 0.85 ± 0.07 \| 0.34 ± 0.02c \| 1.59 ± 0.02ab \| 0.70 ± 0.09b \| 0.85 ± 0.06b \| 1.15 ± 0.12b \| 0.82 ± 0.05 \| 0.80 ± 0.03b \| 0.66 ± 0.02b \| \| MR1 \| 1.18 ± 0.24a \| 0.83 ± 0.04 \| 0.60 ± 0.05b \| 2.16 ± 0.21a \| 1.77 ± 0.18a \| 3.32 ± 0.50a \| 3.35 ± 0.50a \| 0.75 ± 0.03 \| 0.75 ± 0.03a \| 1.95 ± 0.31a \| \| MR2 \| 0.71 ± 0.09b \| 0.78 ± 0.08 \| 0.60 ± 0.05b \| 2.09 ± 0.18a \| 0.99 ± 0.15b \| 2.04 ± 0.12b \| 4.38 ± 0.69a \| 0.87 ± 0.08 \| 1.31 ± 0.05a \| 1.66 ± 0.12a \| \| p-value \| 0.085 \| 0.173 \| <0.001 \| 0.006 \| <0.001 \| <0.001 \| 0.004 \| 0.138 \| <0.001 \| <0.001 \| ## Discussion Lipopolysaccharide-induced immune stress that leads to a decrease in the growth performance of broilers, mainly because the body’s protein anabolism is weakened, catabolism is enhanced, and the nutrients originally used for growth are transferred to resist the inflammatory reaction (Jiang et al., 2010; Yang et al., 2011). In the present study, LPS decreased the weight gain of broilers during the challenge, possibly due to a stress-related reduction in feed intake. The MR1 diet alleviated the negative effect of the LPS challenge on the FCR of broilers. This suggests that MR exerts a protective effect on the growth and development of broilers under stress. In addition, there was no significant difference between the growth performance of the MR groups and the control group, indicating that the growth performance of broilers would not be reduced by properly restricting of the dietary methionine levels correctly. Previous studies also have reported that appropriate MR does not lead to weight loss (Thivat et al., 2009). The thymus is the site of differentiation, development, and maturation of T-cells, and thymus enlargement is conducive to the enhancement of the cellular immune response in the body (Cooper et al., 2006). Our results revealed that the thymus index of the LPS group was decreased at 3 h, indicating that LPS could reduce cellular immune function in broilers. However, there was no difference in the thymus index between the LPS and control groups at 8 h, probably because the stimulation of LPS on the thymus was acute and subjects could recover quickly. The increase in thymus weight of LPS-challenged broilers on MR1 and MR2 diets indicated that the function of T lymphocytes and cellular immunity was enhanced. This might be attributed to the anti-inflammatory effects of MR. Several studies have reported that LPS injection causes liver damage and dysfunction in broilers (Morris et al., 2014; Jangra et al., 2020; Mei et al., 2020). Our pathological observation and scoring of the liver at two different time intervals showed that LPS could cause liver damage, while it could be alleviated by both types of MR diets. In addition, AST and ALT activities in serum are used clinical indicators of liver injury (Senior, 2012). Our results showed that at 8 h, the serum AST and ALT activities of broilers in the LPS group were higher than those in the CON group, which may be because of the time taken for transaminase in liver cells to enter the blood. However, the MR1 diet significantly decreased the AST elevation. Consistent with our results, Zhang et al. [ 2020] reported that endotoxins caused a significant increase in AST and ALT activities. The decrease in serum AST activity suggests that the MR1 diet could reduce LPS-induced liver injury. These results indicate that MR can protect the integrity of the broiler liver, and this protective effect may be attributed to the liver being the main organ for methionine metabolism. MR increases cystathionine β-synthase (CBS) and cystathionine γ cleaving enzyme (CSE) content in the liver, which promotes the synthesis of glutathione (GSH) and hydrogen sulphide (H2S) (Aggrey et al., 2018). Glutathione and H2S play antioxidant roles, thus protecting the liver from oxidative damage (Mou et al., 2020; Elwan et al., 2021). Oxidative stress originates from metabolic disorders involving reactive oxygen species (ROS)/reactive nitrogen species production and antioxidant production (Surai et al., 2019). Endotoxin stimulation has been reported to increase ROS production, but it decreases the activity of antioxidant enzymes (Zheng et al., 2016; Ji et al., 2021). In birds, antioxidant enzymes, including SOD, CAT, and GSH-Px, play an important role in controlling the negative effects of oxidative stress and preventing further immunopathological damage to host tissues (Zheng et al., 2020). Consistent with previous reports, our data showed that CAT and T-AOC activities were inhibited in the LPS challenge group, but the MDA content in the liver was increased. Malondialdehyde is the final product of the peroxidation reaction of free radicals on lipids and the most common marker of oxidative stress (Kamboh et al., 2016). Interestingly, dietary MR tended to improve CAT activity and total antioxidant capacity and decrease MDA content. In addition, SOD and GSH-Px activities were higher in the MR groups than those in the control group. This indicates that a methionine-restricted diet can improve LPS-induced oxidative stress in the liver, which mainly occurring at 8 h after LPS stimulation. Previous studies have also showed that MR significantly decreases the levels of ROS and MDA, and increases the levels of GSH-Px and T-AOC in the liver of high-fat-fed mice, possibly because MR can reduce the production of mitochondrial ROS and increase the endogenous H2S sulfide to alleviate oxidative stress (Maddineni et al., 2013; Ying et al., 2015). However, different results were observed for the serum. T-AOC, and SOD and GSH-Px activities were only reduced 3 h after LPS injection. After 8 h, besides the similar changes in T-AOC and GSH-Px, the MDA content in the serum was increased. This may be why MDA, as an end product of lipid peroxidation, takes time to be produced after LPS stimulation (Balabanlı and Balaban, 2015). In addition, compared with the liver, the antioxidant enzyme activity in the serum was altered 3 h after LPS stimulation. This may be because LPS is absorbed into the blood by the abdominal vein after intraperitoneal injection, which first affects the antioxidant enzyme activity in the blood, and then causes oxidative damage to the liver when the LPS in the blood is absorbed by the tissues (Suriguga et al., 2020). However, dietary methionine restriction alleviated the LPS-induced decline in T-AOC at both time intervals, suggesting that MR alleviates LPS-induced oxidative stress in serum. At 3 h after stimulation, the MR1 diet alleviated the decrease in SOD and CAT activities caused by LPS, while the MR2 diet mainly alleviated SOD activity. After 8 h, both MR diets alleviated the increase in MDA content induced by LPS, and the MR1 diet improved the activity of SOD. Wu et al. ( 2020b) also showed that MR could increase serum T-AOC levels and SOD activity and reduce serum MDA levels in mice fed a high-fat diet. These results indicate that a methionine-restricted diet may increase the antioxidant capacity of broilers by increasing the activity of antioxidant enzymes, possibly owing the activation of antioxidant pathway gene expression. Lipopolysaccharide causes oxidative stress and promotes inflammatory reactions (Rosadini and Kagan, 2017). It has been shown to increase the secretion of proinflammatory cytokines and inhibit animal growth (Takahashi, 2012; Tan et al., 2014; Li et al., 2015). Our results revealed that LPS stimulation significantly increased serum inflammatory cytokines TNF-α, IL-1β, and IL-6, and reduced the concentration of anti-inflammatory cytokine IL-10 at both time intervals. This indicates that the inflammation model was successfully established. However, MR diets significantly reduced the concentration of pro-inflammatory cytokines and increased the concentration of IL-10. A moderate reduction in dietary methionine level may alleviate LPS-induced inflammation. Zhang et al. [ 2019] discovered that MR significantly reduced plasma LPS and LBP concentrations in mice fed a high-fat diet and reduced the mRNA expression of the ileum genes TNF-αand IL-6, indicating that MR can reduce the inflammatory response by limiting the expression of inflammatory factors. In addition, Wu et al. ( 2020b) showed that MR could improve the intestinal flora, reduce the serum LPS concentration, and enhance the intestinal barrier of mice by upregulating single-chain fatty acid-producing bacteria and downregulating inflammatory LPS-producing bacteria. Our research also showed that MR could reduce the LPS concentration in the serum during the study period, and strengthen the intestinal immune barrier which might inhibit inflammation. Both oxidative stress and inflammation are considered to be the key factors of acute bacterial disease, and they promote and amplify each other. For example, the pro-inflammatory master gene NF-κB is a redox sensitive transcription factor. Conversely, NF-κB activates the genes of NADPH oxidase and COX2 (Arioz et al., 2019; Yu et al., 2022). The inhibitory effect of MR on inflammation and its direct antioxidant effect can break this vicious cycle. Nrf2 is a major regulator of cell defense against oxidative stress and is activated to promote the expression of antioxidant genes; Keap1, as an adaptor subunit of Cullin 3-based E3 ubiquitin ligase, regulates Nrf2 activity (Sajadimajd and Khazaei, 2018). The results showed that MR1 increased the gene expression of Nrf2 at 3 h and 8 h, while MR1 and MR2 increased the gene expression of Keap1 at 8 h alone. Previous studies have shown that during MR, the activity of the methionine metabolizing enzyme CBS and CSE was upregulated, which led to an increase in H2S and other downstream effectors (such as GSH and related compounds) (McIsaac et al., 2012; Petti et al., 2012). It has been observed that H2S modifies proteins post-translationally through S-sulfhydrination, transforming the—SH of cysteine into—SH, thus regulating protein activity. S-thinning of Kelch-like ECH-related protein 1 (Keap1) can activate Nrf2 (Yang et al., 2013). Methionine restriction may directly activate Nrf2 (Lewis et al., 2010). The increase in Nrf2 activity and the accompanying enhanced gene expression of antioxidant response elements will improve the antioxidant and detoxification capacities of the body. Our results suggest that the MR1 diet activated the Nrf2 pathway, and increased the gene expression of SOD, CAT, and GSH-PX at 3 h and 8 h after LPS injection. However, compared with MR1, the activation of the antioxidant pathway rate with the MR2 diet was slower until 8 h after injection. ## Conclusion In conclusion, appropriately reducing methionine levels in the diet ameliorated LPS-induced immune stress and liver damage by inhibiting inflammation and oxidative stress. In addition, reducing methionine in the diet by $0.1\%$ and $0.2\%$ had a similar effect on relieving stress when injecting LPS at different times; however, MR1 activated the Nrf2 pathway earlier than MR2. The results confirmed the immune regulation, anti-oxidation, and anti-inflammatory effects of two different levels of MR diets on LPS-challenged broilers under high stocking density, and provided a scientific basis for the actual production of broilers. ## Data availability statement The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author. ## Ethics statement The animal study was reviewed and approved by the Animal Health and Care Committee of the Shanxi Agricultural University. ## Author contributions JL and ZM conceived the project, designed and supervised the experiments. XP and YD designed the experiments. XP, XX, HC, and YW performed all of the experiments. XP, ZM, YS and LY analyzed the data. XP wrote the manuscript with input from all the authors. MH, YS, and JY revised the manuscript. All authors contributed to the article reading and approved the final manuscript. ## Conflict of interest Author LY is employed by New Hope Liuhe Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Aggrey S. E., Gonzalez-Ceron F., Rekaya R., Mercier Y.. Gene expression differences in the methionine remethylation and transsulphuration pathways under methionine restriction and recovery with D,L-methionine or D,L-HMTBA in meat-type chickens. J. Anim. Physiol. Anim. Nutr. Berl. (2018) 102 e468-e475. DOI: 10.1111/jpn.12779 2. Arioz B. I., Tastan B., Tarakcioglu E., Tufekci K. U., Olcum M., Ersoy N.. Melatonin attenuates LPS-induced acute depressive-like behaviors and microglial NLRP3 inflammasome activation through the SIRT1/nrf2 pathway. Front. Immunol. (2019) 10 1511. DOI: 10.3389/fimmu.2019.01511 3. Averós X., Estevez I.. Meta-analysis of the effects of intensive rearing environments on the performance and welfare of broiler chickens. Poult. Sci. (2018) 97 3767-3785. DOI: 10.3382/ps/pey243 4. Balabanlı B., Balaban T.. Investigation into the effects of boron on liver tissue protein carbonyl, MDA, and glutathione levels in endotoxemia. Biol. Trace Elem. Res. (2015) 167 259-263. DOI: 10.1007/s12011-015-0301-z 5. Brown-Borg H. M., Buffenstein R.. Cutting back on the essentials: Can manipulating intake of specific amino acids modulate health and lifespan?. Ageing Res. Rev. (2017) 39 87-95. DOI: 10.1016/j.arr.2016.08.007 6. Bryant C. E., Spring D. R., Gangloff M., Gay N. J.. The molecular basis of the host response to lipopolysaccharide. Nat. Rev. Microbiol. (2010) 8 8-14. DOI: 10.1038/nrmicro2266 7. Cooper M. D., Peterson R. D., South M. A., Good R. A.. The functions of the thymus system and the bursa system in the chicken. J. Immunol. (2006) 176 6370-6404. PMID: 16709795 8. Deng J., Zhang J., Chang Y., Wang S., Shi M., Miao Z.. Effects of Chinese yam polysaccharides on the immune function and serum biochemical indexes of broilers. Front. Vet. Sci. (2022) 9 1013888. DOI: 10.3389/fvets.2022.1013888 9. Duan J., Xiang L., Yang Z., Chen L., Gu J., Lu K.. Methionine restriction prevents lipopolysaccharide-induced acute lung injury via modulating CSE/H2S pathway. Nutrients (2022) 14 322. DOI: 10.3390/nu14020322 10. Elwan H., Xie C., Miao L. P., Dong X., Zou X. T., Mohany M.. Methionine alleviates aflatoxinb1-induced broiler chicks embryotoxicity through inhibition of caspase-dependent apoptosis and enhancement of cellular antioxidant status. Poult. Sci. (2021) 100 101103. DOI: 10.1016/j.psj.2021.101103 11. Fang H., Stone K. P., Wanders D., Forney L. A., Gettys T. W.. The origins, evolution, and future of dietary methionine restriction. Annu. Rev. Nutr. (2022) 42 201-226. DOI: 10.1146/annurev-nutr-062320-111849 12. Goo D., Kim J. H., Park G. H., Delos Reyes J. B., Kil D. Y.. Effect of heat stress and stocking density onGrowth performance, breast meat Quality,and intestinal barrier function in broiler chickens. Anim. (Basel) (2019) 9 107. DOI: 10.3390/ani9030107 13. Jangra A., Rajput P., Dwivedi D. K., Lahkar M.. Amelioration of repeated restraint stress-induced behavioral deficits and hippocampal anomalies with taurine treatment in mice. Neurochem. Res. (2020) 45 731-740. DOI: 10.1007/s11064-019-02945-8 14. Ji S. Y., Cha H. J., Molagoda I. M. N., Kim M. Y., Kim S. Y., Hwangbo H.. Suppression of lipopolysaccharide-induced inflammatory and oxidative response by 5-aminolevulinic acid in RAW 264.7 macrophages and zebrafish larvae. Biomol. Ther. Seoul. (2021) 29 685-696. DOI: 10.4062/biomolther.2021.030 15. Jiang Z., Schatzmayr G., Mohnl M., Applegate T. J.. Net effect of an acute phase response-partial alleviation with probiotic supplementation. Poult. Sci. (2010) 89 28-33. DOI: 10.3382/ps.2009-00464 16. Jin S. W., Zhang L., Lian Q. Q., Yao S. L., Wu P., Zhou X. Y.. Close functional coupling between Ca. Mediat. Inflamm. (2006) 2006 36192. DOI: 10.1155/mi/2006/36192 17. Kamboh A. A., Hang S. Q., Khan M. A., Zhu W. Y.. Animal (2016) 10 1619-1625. DOI: 10.1017/S1751731116000562 18. Leshchinsky T. V., Klasing K. C.. Divergence of the inflammatory response in two types of chickens. Dev. Comp. Immunol. (2001) 25 629-638. DOI: 10.1016/S0145-305X(01)00023-4 19. Lewis K. N., Mele J., Hayes J. D., Buffenstein R.. Nrf2, a guardian of healthspan and gatekeeper of species longevity. Integr. Comp. Biol. (2010) 50 829-843. DOI: 10.1093/icb/icq034 20. Li M., Chen L., Zhao Y., Sun H., Zhao L.. Research on the mechanism of HRP relieving IPEC-J2 cells immunological stress based on transcriptome sequencing analysis. Front. Nutr. (2022) 9 944390. DOI: 10.3389/fnut.2022.944390 21. Li Y., Zhang H., Chen Y. P., Yang M. X., Zhang L. L., Lu Z. X.. Bacillus amyloliquefaciens supplementation alleviates immunological stress in lipopolysaccharide-challenged broilers at early age. Poult. Sci. (2015) 94 1504-1511. DOI: 10.3382/ps/pev124 22. Livak K. J., Schmittgen T. D.. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods (2001) 25 402-408. DOI: 10.1006/meth.2001.1262 23. Ma B., Xing T., Li J., Zhang L., Jiang Y., Gao F.. Chronic heat stress causes liver damage via endoplasmic reticulum stress-induced apoptosis in broilers. Poult. Sci. (2022) 101 102063. DOI: 10.1016/j.psj.2022.102063 24. Maddineni S., Nichenametla S., Sinha R., Wilson R. P., Richie J. P.. Methionine restriction affects oxidative stress and glutathione-related redox pathways in the rat. Exp. Biol. Med. (Maywood) (2013) 238 392-399. DOI: 10.1177/1535370213477988 25. McIsaac R. S., Petti A. A., Bussemaker H. J., Botstein D.. Perturbation-based analysis and modeling of combinatorial regulation in the yeast sulfur assimilation pathway. Mol. Biol. Cell (2012) 23 2993-3007. DOI: 10.1091/mbc.E12-03-0232 26. Mei W., Hao Y., Xie H., Ni Y., Zhao R.. Hepatic inflammatory response to exogenous LPS challenge is exacerbated in broilers with fatty liver disease. Anim. (Basel) (2020) 10 514. DOI: 10.3390/ani10030514 27. Morris A., Shanmugasundaram R., Lilburn M. S., Selvaraj R. K.. 25-Hydroxycholecalciferol supplementation improves growth performance and decreases inflammation during an experimental lipopolysaccharide injection1 1Author disclosures: This research was partially supported by Dutch State Mines (DSM, Heerlen, The Netherlands) and Hatch grant awarded to RKS. Poult. Sci. (2014) 93 1951-1956. DOI: 10.3382/ps.2014-03939 28. Mou Y., Wen S., Li Y. X., Gao X. X., Zhang X., Jiang Z. Y.. Recent progress in Keap1-Nrf2 protein-protein interaction inhibitors. Eur. J. Med. Chem. (2020) 202 112532. DOI: 10.1016/j.ejmech.2020.112532 29. Petti A. A., McIsaac R. S., Ho-Shing O., Bussemaker H. J., Botstein D.. Combinatorial control of diverse metabolic and physiological functions by transcriptional regulators of the yeast sulfur assimilation pathway. Mol. Biol. Cell (2012) 23 3008-3024. DOI: 10.1091/mbc.E12-03-0233 30. Rosadini C. V., Kagan J. C.. Early innate immune responses to bacterial LPS. Curr. Opin. Immunol. (2017) 44 14-19. DOI: 10.1016/j.coi.2016.10.005 31. Sajadimajd S., Khazaei M.. Oxidative stress and cancer: The role of Nrf2. Curr. Cancer Drug Targets (2018) 18 538-557. DOI: 10.2174/1568009617666171002144228 32. Senior J. R.. Alanine aminotransferase: A clinical and regulatory tool for detecting liver injury–past, present, and future. Clin. Pharmacol. Ther. (2012) 92 332-339. DOI: 10.1038/clpt.2012.108 33. Siegmund B., Lear-Kaul K. C., Faggioni R., Fantuzzi G.. Leptin deficiency, not obesity, protects mice from Con A-induced hepatitis. Eur. J. Immunol. (2002) 32 552-560. DOI: 10.1002/1521-4141(200202)32:2<552:Aid-immu552>3.0.Co;2-h 34. Surai P. F., Kochish V. I., Fisinin M. T.. Antioxidant defence systems and oxidative stress in poultry biology: An update. Antioxidants (Basel) (2019) 8 235. DOI: 10.3390/antiox8070235 35. Suriguga S., Luangmonkong T., Mutsaers H. A. M., Groothuis G. M. M., Olinga P.. Host microbiota dictates the proinflammatory impact of LPS in the murine liver. Toxicol Vitro (2020) 67 104920. DOI: 10.1016/j.tiv.2020.104920 36. Takahashi K.. Effect of a cultures of. J. Poult. Sci. (2012) 49 94-100. DOI: 10.2141/jpsa.011079 37. Takahashi K., Takimoto T., Sato K., Akiba Y.. Effect of dietary supplementation of astaxanthin from Phaffia rhodozyma on lipopolysaccharide-induced early inflammatory responses in male broiler chickens (Gallus gallus) fed a corn-enriched diet. Anim. Sci. J. (2011) 82 753-758. DOI: 10.1111/j.1740-0929.2011.00898.x 38. Tan J., Liu S., Guo Y., Applegate T. J., Eicher S. D.. Dietary L-arginine supplementation attenuates lipopolysaccharide-induced inflammatory response in broiler chickens. Br. J. Nutr. (2014) 111 1394-1404. DOI: 10.1017/s0007114513003863 39. Thivat E., Farges M. C., Bacin F., D'Incan M., Mouret-Reynier M. A., Cellarier E.. Phase II trial of the association of a methionine-free diet with cystemustine therapy in melanoma and glioma. Anticancer Res. (2009) 29 5235-5240. PMID: 20044642 40. Wu G., Han L., Shi Y., Feng C., Yan B., Sun J.. Effect of different levels of dietary methionine restriction on relieving oxidative stress and behavioral deficits in middle-aged mice fed low-medium-or high-fat diet. J. Funct. Foods (2020a) 65 103782. DOI: 10.1016/j.jff.2020.103782 41. Wu G., Shi Y., Han L., Feng C., Ge Y., Yu Y.. Dietary methionine restriction ameliorated fat accumulation, systemic inflammation, and increased energy metabolism by altering gut microbiota in middle-aged mice administered different fat diets. J. Agric. Food Chem. (2020b) 68 7745-7756. DOI: 10.1021/acs.jafc.0c02965 42. Xie M., Hou S. S., Huang W., Fan H. P.. Effect of excess methionine and methionine hydroxy analogue on growth performance and plasma homocysteine of growing Pekin ducks. Poult. Sci. (2007) 86 1995-1999. DOI: 10.1093/ps/86.9.1995 43. Yang G., Zhao K., Ju Y., Mani S., Cao Q., Puukila S.. Hydrogen sulfide protects against cellular senescence via S-sulfhydration of Keap1 and activation of Nrf2. Antioxid. Redox Signal (2013) 18 1906-1919. DOI: 10.1089/ars.2012.4645 44. Yang X. J., Li W. L., Feng Y., Yao J. H.. Effects of immune stress on growth performance, immunity, and cecal microflora in chickens. Poult. Sci. (2011) 90 2740-2746. DOI: 10.3382/ps.2011-01591 45. Ying Y., Yun J., Guoyao W., Kaiji S., Zhaolai D., Zhenlong W.. Dietary L-methionine restriction decreases oxidative stress in porcine liver mitochondria. Exp. Gerontol. (2015) 65 35-41. DOI: 10.1016/j.exger.2015.03.004 46. Yu W.. Modulation of tight junction protein expression in chicken intestine by methionine substitude and the mechanism (2013) 47. Yu Z., Zhao L., Zhao J.-L., Xu W., Guo Z., Zhang A.-Z.. Dietary Taraxacum mongolicum polysaccharide ameliorates the growth, immune response, and antioxidant status in association with NF-κB, Nrf2 and TOR in Jian carp (. Aquaculture (2022) 547 737522. DOI: 10.1016/j.aquaculture.2021.737522 48. Zhang Y., Jia H., Jin Y., Liu N., Chen J., Yang Y.. Glycine attenuates LPS-induced apoptosis and inflammatory cell infiltration in mouse liver. J. Nutr. (2020) 150 1116-1125. DOI: 10.1093/jn/nxaa036 49. Zhang Y., Yang Y., Wang Y.. Effect of methionine restriction on gut redox status, inflammation and microbiota in high-fat diet-fed mice. Food Sci. (2019) 40 99-106. DOI: 10.7506/spkx1002-6630-20171226-333 50. Zheng X. C., Wu Q. J., Song Z. H., Zhang H., Zhang J. F., Zhang L. L.. Effects of Oridonin on growth performance and oxidative stress in broilers challenged with lipopolysaccharide. Poult. Sci. (2016) 95 2281-2289. DOI: 10.3382/ps/pew161 51. Zheng Y. W., Zhang J. Y., Zhou H. B., Guo Y. P., Ma Q. G., Ji C.. Effects of dietary pyrroloquinoline quinone disodium supplementation on inflammatory responses, oxidative stress, and intestinal morphology in broiler chickens challenged with lipopolysaccharide. Poult. Sci. (2020) 99 5389-5398. DOI: 10.1016/j.psj.2020.08.007
--- title: Occlusion preconditioned mice are resilient to hypobaric hypoxia-induced myocarditis and arrhythmias due to enhanced immunomodulation, metabolic homeostasis, and antioxidants defense authors: - Gabriel Komla Adzika - Richard Mprah - Ruqayya Rizvi - Adebayo Oluwafemi Adekunle - Marie Louise Ndzie Noah - Prosperl Ivette Wowui - Seyram Yao Adzraku - Joseph Adu-Amankwaah - Fengli Wang - Yuwen Lin - Lu Fu - Xiaomei Liu - Jie Xiang - Hong Sun journal: Frontiers in Immunology year: 2023 pmcid: PMC9975755 doi: 10.3389/fimmu.2023.1124649 license: CC BY 4.0 --- # Occlusion preconditioned mice are resilient to hypobaric hypoxia-induced myocarditis and arrhythmias due to enhanced immunomodulation, metabolic homeostasis, and antioxidants defense ## Abstract ### Background Sea-level residents experience altitude sickness when they hike or visit altitudes above ~2,500 m due to the hypobaric hypoxia (HH) conditions at such places. HH has been shown to drive cardiac inflammation in both ventricles by inducing maladaptive metabolic reprogramming of macrophages, which evokes aggravated proinflammatory responses, promoting myocarditis, fibrotic remodeling, arrhythmias, heart failure, and sudden deaths. The use of salidroside or altitude preconditioning (AP) before visiting high altitudes has been extensively shown to exert cardioprotective effects. Even so, both therapeutic interventions have geographical limitations and/or are inaccessible/unavailable to the majority of the population as drawbacks. Meanwhile, occlusion preconditioning (OP) has been extensively demonstrated to prevent hypoxia-induced cardiomyocyte damage by triggering endogenous cardioprotective cascades to mitigate myocardial damage. Herein, with the notion that OP can be conveniently applied anywhere, we sought to explore it as an alternative therapeutic intervention for preventing HH-induced myocarditis, remodeling, and arrhythmias. ### Methods OP intervention (6 cycles of 5 min occlusion with 200 mmHg for 5 min and 5 min reperfusion at 0 mmHg – applying to alternate hindlimb daily for 7 consecutive days) was performed, and its impact on cardiac electric activity, immunoregulation, myocardial remodeling, metabolic homeostasis, oxidative stress responses, and behavioral outcomes were assessed before and after exposure to HH in mice. In humans, before and after the application of OP intervention (6 cycles of 5 min occlusion with $130\%$ of systolic pressure and 5 min reperfusion at 0 mmHg – applying to alternate upper limb daily for 6 consecutive days), all subjects were assessed by cardiopulmonary exercise testing (CPET). ### Results Comparing the outcomes of OP to AP intervention, we observed that similar to the latter, OP preserved cardiac electric activity, mitigated maladaptive myocardial remodeling, induced adaptive immunomodulation and metabolic homeostasis in the heart, enhanced antioxidant defenses, and conferred resistance against HH-induce anxiety-related behavior. Additionally, OP enhanced respiratory and oxygen-carrying capacity, metabolic homeostasis, and endurance in humans. ### Conclusions Overall, these findings demonstrate that OP is a potent alternative therapeutic intervention for preventing hypoxia-induced myocarditis, cardiac remodeling, arrhythmias, and cardiometabolic disorders and could potentially ameliorate the progression of other inflammatory, metabolic, and oxidative stress-related diseases. ## Introduction Sea-level residents suffer from altitude sickness when they hike or visit altitudes above ~2,500 m due to the hypobaric hypoxia (HH) conditions at such places. Altitude sickness typically presents clinical manifestations such as shortness of breath, headache, dizziness, tiredness, mental confusion, and loss of appetite [1]. Meanwhile, recent studies have shown that besides the aforementioned symptoms, individuals experiencing altitude sickness have underlying myocarditis and arrhythmias that were either induced or aggravated by HH [2, 3]. Evidently, HH has been shown to drive cardiac inflammation in both ventricles by inducing maladaptive metabolic reprogramming of macrophages which evokes hypersecretion of the proinflammatory mediator – inducible nitric oxide synthase (iNOS) and cytokines (C-Reactive Proteins, Interleukin (IL)-1β and IL-18). HH-induced hyperactive proinflammatory responses expedite adverse cardiac remodeling by activating and sustaining fibrosis cascades, ultimately resulting in heart failure and sudden cardiac death [4, 5]. Therapeutic approaches developed against altitude sickness over the years have mainly been preventive interventions targeted at circumventing or mitigating the adverse outcomes of HH exposure. Notably, the use of salidroside (a phenylethanoid glycoside found in Rhodiola genus plants) and altitude preconditioning (AP) (as known as intermittent HH preconditioning) prior to visiting high altitudes have been extensively shown to exert cardioprotective effects [6, 7]. The efficacies of salidroside and AP interventions have been attributed to their abilities to decrease reactive oxygen species (ROS), induce adaptive regulation of antioxidants and anti-inflammatory-related pathways as well as enhance tissue oxygenation to prevent necrosis and apoptosis of cardiomyocytes (6–9). However, the availability of Rhodiola plants or salidroside is geographically limited to Europe, North America, and low-Arctic to high-temperature regions of Asia [10]. Similarly, hypoxia chambers for AP are inaccessible/unavailable to the majority of the population, and the intervention cannot be applied at one’s convenience before hiking or visiting high altitudes. Meanwhile, remote ischemic preconditioning [hereafter referred to as occlusion preconditioning (OP)] has been extensively demonstrated to prevent hypoxia-induced cardiomyocyte damage by triggering endogenous cardioprotective cascade (11–13). These generally positive outcomes of OP have encouraged its application in clinical trials and settings to reduce the severity of ischemic injuries and myocardial damage, even though the underlying mechanisms of the intervention are still being elucidated. Here, with the notion that OP can be conveniently applied anywhere, we sought to explore it as an alternative therapeutic intervention for preventing HH-induced myocarditis and cardiac arrhythmia. Herein, we demonstrate the cardioprotective potentials of OP in HH by comparing its impact on cardiac electric activity, hypertrophy and injury, immunoregulation, oxidative stress responses, and behavioral outcomes, with AP’s in HH. Also, we showed that OP enhances respiratory and oxygen-carrying capacity in humans. In addition, numerous studies have shown that the β2-adrenergic receptor (β2AR) confers cardioprotection in stressful conditions, including hypoxia [14, 15]. Hence, we utilized β2AR-knockout (β2AR-KO) mice and uncovered that β2AR is involved in mediating OP-induced cardioprotection in HH. These findings illustrate OP as a potent alternative therapeutic intervention for preventing hypoxia-induced myocarditis and as well suggest its potential to ameliorate other oxidative stress-related diseases. ## Experimental animal model protocols Eight- to twelve-week-old wild-type (Adrb2+/+) and β2AR-knockout (Adrb2-/-) FVB male mice were used in this study. The mice were kept and fed in a hypoxia chamber (Guizhou Fenglei Aviation Machinery Co., Ltd., Guizhou, China: FLYDWC50-IIA), and hypobaric hypoxia (HH) was induced by increasing altitude to 3000 m for 10 min, then to 4500 m for 10 min, followed by 5500 m for 20 min before finally increasing to 6000 m altitude for 7 days. Mice in the control group were kept and fed in a normobaric normoxia (NN) environment at sea level (with ambient oxygen percentage) for 7 days. To explore the therapeutic potentials of limb occlusion ischemic preconditioning, hair was removed from mice’s hindlimbs. Limb occlusion preconditioning (OP) was performed by applying a 200 mmHg pressure tourniquet for 5 min and allowing 5 min reperfusion at 0 mmHg. Six cycles of OP were performed daily on alternate hindlimbs for 7 days. Next, the mice were randomized into two groups; the first group, (OP) mice, were sacrificed, and the second group was exposed to HH stepwise as previously described for 7 days. The latter group was designated as OP prior to HH exposure (OPHH) (Supplementary Figure 1A). Additionally, we sought to compare the experimental outcomes from OP and OPHH with altitude preconditioning (AP) prior to HH exposure (APHH) models; hence, AP was done by exposing wild-type FVB to HH at 3500 altitudes for 30 min daily, for 7 days. Afterward, the mice were randomized into two groups; the first group (AP) mice were sacrificed, and the second group (APHH) mice were exposed to HH in a stepwise manner as previously described for 7 days (Supplementary Figure 1B). At the end of all experimental models, electrocardiography (EKG) data acquisitions were performed with PowerLab (ADInstruments, North America), and the mice were euthanatized by cervical dislocation. Hearts were excised, quickly wet-weighed for morphometric analysis, and processed for further investigations. The performed experiments were approved by the Experimental Animal Centre of Xuzhou Medical University and the Animal Ethics Committee of the Medical University (permit number: xz11- 12541) and conform to the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH Publication, 8th Edition, 2011). ## Electrocardiography Electrocardiography (EKG) data acquisitions were performed with the 3-lead monopolar needle electrode from PowerLab systems (ADInstruments, North America), as previously described [16]. ## Enzyme-linked immunosorbent assay (ELISA) Myocardia lysates were used to examine the concentration of proinflammatory (iNOS, IL-1β, and IL-18) and anti-inflammatory biomarkers (Arg-1, IL-10, and TGF-β) and cardiac hypertrophy/injury markers (ANP and BNP). Sera were used to assess cardiac troponin I (cTnI) and C-reactive protein (CRP) concentrations. IL-1β (JL18442; Jianglai Bio. Tech), IL-18 (JL20253; Jianglai Bio. Tech.), iNOS (JL20675; Jianglai Bio. Tech.), IL-10 (JL20242; Jianglai Bio. Tech.), TGF-β (JL13959; Jianglai Bio. Tech.), Arg-1 (JL13668; Jianglai Bio. Tech.), ANP (JL20612; Jianglai Bio. Tech.), BNP (JL12884; Jianglai Bio. Tech.), CRP (JL13196; Jianglai Bio. Tech.) and cTnI (JL31923; Jianglai Bio. Tech.) ELISAs were done in triplicates and as per the manufacturer’s instructions. ## NAD/NADH content assay Using equal weights (0.1 g) of myocardia, the coenzyme I NAD/NADH contents were assessed using assay kits (BC0310; Solarbio) and following the manufacturer’s instructions. ## Total antioxidant capacity assay Equal weights (0.1 g) of myocardia were used to evaluate the total antioxidant capacity (T-AOC). Assay kits (BC1310; Solarbio) were used according to the manufacturer’s instructions. ## Open field test (OFT) Locomotor activity and exploratory and anxiety-related behavior of the mice were examined before and after preconditioning or exposure to HH or NN by using the OFT apparatus. Briefly, the apparatus consists of a squared box (50cm x 50 cm) with its base divided into 9 squares; 1 central (zone C), 4 corners (zone B), and 4 peripheries (zone A). Before the initial and subsequent tests, fecal pellets or urine were cleaned, and the chamber was wiped-dry with $95\%$ ethanol to remove any clues and scent left by the last tested mouse. The mice were individually placed in zone C and left undisturbed to explore for 5 min while their locomotion activities were tracked with video tracking software (ANY-maze version 7.00). ## Elevated plus maze (EPM) Utilizing the EPM apparatus, anxiety-related behaviors were examined before and after preconditioning or exposure to HH or NN. In brief, the EPM consists of a plus (+) shaped apparatus with a central point, two opposite arms enclosed, and the other opposite arms opened. Before the initial and subsequent tests, fecal pellets or urine were cleaned, and the arms were wiped-dry with $95\%$ ethanol to remove any clues and scent left by the last tested mouse. During testing, the plus maze was elevated ~ 1 m from the floor, and each mouse was placed at the central point of the open and closed arms, with their head facing the open arm. The mice were allowed to explore the maze for 5 min while their locomotion activities and entries into either arm were tracked and recorded by video tracking software (ANY-maze version 7.00). ## Myocardial macrophage isolations Mice were euthanized by cervical dislocation; hearts were exposed and perfused-blanch with iced-cold PBS through the right and left ventricle by using a 5 mL syringe and 25 G needle. Hearts were then transferred in 12-well plates containing 1 mg/mL collagenase IV (Gibco™: 17104019) in 3 mL Hanks’ Balanced Salt Solution (HBSS) kept on ice and minced with sterile scissors. Minced myocardia were digested for 45 min at 37°C on a shaker (50 rpm). Next, the plates were vortexed, kept on ice, and new Pasteur pipettes were used to dissociate cells mechanically. Obtained suspensions were filtered through 35 μm strainers into 15 ml tubes containing 10 ml cold HBSS, centrifuged at 1500 rpm for 5 min, and the supernatants were discarded. Red blood cells in the pellet were hemolyzed with ACK buffer (Gibco™: A1049201) and washed twice with PBS. Myocardial macrophage phenotypes were identified and sorted with FACS (BD FACSAria™ III) after resuspension and incubation with Fc Blocker (Invitrogen; 14-9161-73; 1:100), PE-Cy5 anti-CD45 (BD Pharmingen™; 553082; 1:100), APC anti-F$\frac{4}{80}$ (BioLegend; 123116; 1:100), FITC anti-CD11b (BioLegend; 101206; 1:100), PerCP anti-CD86 (BioLegend; 105028; 1:100) and PE anti-CD206 (BioLegend; 141706; 1:100). ## Wheat germ agglutinin (WGA) staining Cryopreserved heart sections were fixed with $4\%$ formaldehyde for 30 min at room temperature (RT), washed thrice with PBS, and primed with HBSS for 15 min. Next, without permeabilization, the myocardial sections were incubated with WGA staining (Thermo Fisher Scientific; W11261) in the dark for 10 min at RT – followed by three times wash with PBS and DAPI counterstaining. Imaging was done at X60 magnification, and ImageJ (1.52a version; National Institute of Health USA) was used to assess cardiomyocyte surface area. ## Masson’s trichrome staining Myocardial sections were trichrome stained according to the manufacturer’s (Solarbio; G1340) instructions. Microscopy was done at X40 magnification, and collagen volume fractions (CVF) were analyzed with ImageJ. ## Immunohistochemical (IHC) staining CD86 (Abcam; ab53004; 1:1000) and CD206 (Abcam; ab8918; 1:1000) IHC staining were done as previously described [17], but with few optimizations. Briefly, frozen sections were used; hence, the antigen retrieval step in the described experiment was skipped, and myocardial sections were fixed with $4\%$ formaldehyde for 15 min prior to staining. Infiltrated CD86+ and CD206+ macrophages were observed at X40 magnification and quantified with ImageJ. ## Oil Red O (ORO) staining To investigate the metabolic state of the hearts, lipid depositions in myocardia were assessed by performing ORO staining described by the manufacturer (Solarbio; G1261). Lipid depositions were observed at X40 magnification and quantified with ImageJ. ## Periodic Acid Schiff (PAS) staining To examine the hearts’ metabolic state, glycogen and other polysaccharide contents of the myocardia were assessed by performing PAS staining described by the manufacturer (Solarbio; G1281). PAS-positive areas were observed at X40 magnification and quantified with ImageJ. ## Western blot Hearts were washed with cold PBS, homogenized, and cocktails of RIPA buffer, protease, and phosphatase inhibitor (ratio 100:1:1) were added to extract proteins. Protein sample concentrations were normalized, electrophoresed on 10-$12\%$ gels, and transferred onto 0.45 μm PVDF membrane (Millipore Immobilon®-P; IPVH08100). Membranes were blocked with $2\%$ BSA in TBST, and proteins of interest were blotted with the following antibodies: anti-HIF-1α (Proteintech; 20960-1-AP; 1:1000), anti-HIF-2α (Abcam; ab199; 1:1000), anti-Nrf2 (Proteintech; 16396-1-AP; 1:1000), anti-β2AR (Abcam; ab182136; 1:1000), anti-Scarb3 (Abcam; ab133625; 1:1000), anti-Slc2a1 (Abcam; ab115730; 1:1000), anti-GATA4 (Abcam; ab84593; 1:1000), GAPDH (Proteintech; 10494-1-AP; 1:1000) and HRP-conjugated Goat Anti-Rabbit IgG(H+L) (Proteintech; SA00001-2; 1:1000). Membranes were imaged using enhanced chemiluminescence (Tanon, China). ## Quantitative RT-PCR mRNAs were isolated from myocardial macrophages with TRIzol™ Reagent (Invitrogen™; 15596026), cDNAs synthesized using a reverse transcription kit (FSQ107; Toyobo), and qPCR analysis was conducted by utilizing SYBR Green Master Mix (Q111-02; Vazyme) according to manufacturer instructions. The assessed macrophage metabolic genes (Gcdh, Adcd1, Acaa2, Decr1, Hsd17b4, Hadha, Cpt2, Etfb, Echdc2, Scarb3, mTOR, Slc2a1, Hk2, Ldha, Aldoc, Fbp1, Pgm2, Gpi1, Pgk1, and Pfkfb3) and their respective primer sequences are tabulated here (Supplementary Table 1). GAPDH was utilized as the housekeeping gene, and mRNAs fold changes were computed by the 2−ΔΔCt method. ## Cardiopulmonary exercise test in humans Before and after the application of OP intervention (6 cycles of 5 min occlusion with $130\%$ of systolic pressure and reperfusion at 0 mmHg – alternating upper limb daily for 6 consecutive days), all human subjects were assessed by cardiopulmonary exercise testing (CPET) (Vyaire Medical; Vyntus® CPX) on a bicycle ergometer (Stex Fitness; S25U) using the ramp 10 Watts protocol (10 W increment in workload per 1 min). Data analysis included the following physiological indexes; heart rate (HR), systolic (Psys) and diastolic (Pdia) pressures, expiratory reserve, inhalation and exhalation vital capacities, minute ventilation (V’E), carbon dioxide output (V’CO2), oxygen uptake (V’O2), oxygen pulse (V’O2/HR), metabolic equivalent of task (MET) and respiratory exchange ratio (RER). Additionally, oxygen saturation (SPO2) was measured with an ear sensor probe (Integrated Nonin™). ## Statistics All results in this study are presented as mean ± standard error of the mean. Statistical analyses were done using GraphPad Prism (Software version 8.0.2). Unpaired t-test was used for comparing two groups, one-way ANOVA was used for comparing three or more groups, and two-way ANOVA was used for grouped data statistical analyses. P-values less than 0.05 were deemed statistically significant. ## OP preserves cardiac electric activity during hypobaric hypoxia Electrocardiograms (EKG) of mice post-HH exposure showed overt distortions in their cardiac electric activity (cEA) in comparison with the normobaric normoxia (NN) mice (control group). A detailed look at the EKG parameters revealed HH mice had modest increments in their heart rates (HRs) but with significant prolongations of QT, QTc, JT, and Tpeak-Tend intervals, and ST height, P duration, and R and T amplitudes (Figures 1A–G; Supplementary Table 2). Taken together, the aforementioned EKG alterations indicate that HH mice had severe arrhythmias resulting from chronic exposure to hypoxia. To determine the efficacy of OP’s impact on OPHH mice, we employed AP and APHH mice groups for comparison. The EKG data indicated that, similar to AP, OP prevents disruption of cEA with initial slight elevations of HRs and its normalization when exposed to HH. However, we uncovered that OP preserved cEA better than AP because unlike in OPHH mice – QT, QTc, JT, and Tpeak-Tend intervals, and ST height remained significantly increased in APHH compared to NN mice. Figure 1: OP preserves cardiac electric activity during hypobaric hypoxia. (A) Representative electrocardiography (EKG) of Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP) and Occlusion Preconditioned before HH exposure (OPHH) mice. P wave: atrial depolarization; Q wave: Interventricular septum depolarization; R wave: Ventricular depolarization; S wave: Purkinje fibres depolarization; J wave: Early ventricular repolarization; T wave: End of ventricular repolarization. (B-G) Graphical presentation of EKG parameters including; Heart Rate (HR), QT Interval, corrected QT Interval (QTc), JT Interval, Tpeak Tend Interval, and ST Height. (n= 5-9 mice per experimental group). $p<0.05, $$p<0.01, $$$p<0.001 HH vs NN; &p<0.05 APHH vs AP; #p<0.05 OPHH vs OP; p<0.05, *p<0.01. Data are expressed as mean ± SEM. Data were analyzed using one-way ANOVA, followed by Tukey’s post hoc analysis. ## OP mitigates myocardial hypertrophy, injury, and fibrosis during hypobaric hypoxia Mice exposed to HH without any preconditioning exhibited significant body weight (BW) loss and increased heart weight/body weight ratio (Figures 2A, B), depicting cardiac hypertrophy. Next, cardiomyocyte surface area, atrial natriuretic peptide (ANP), brain natriuretic peptide (BNP), and GATA4 expressions were ascertained to validate the incidence of cardiac hypertrophy (Figures 2C–H). These indexes and biomarkers were substantially increased in HH mice. Also, the extent of ANP, BNP, and GATA4 upregulation in HH heart revealed the incidence of cardiac injury. Compared with HH and APHH mice, OPHH mice showed the least weight loss and moderate increases in the prior mentioned cardiac hypertrophy and cardiac injury indexes. In addition, utilizing trichrome staining, we observed massive fibrosis in HH hearts (Figures 2I, J). Meanwhile, like APHH, OPHH mice had only modest collagen depositions with no significant differences in comparison to the NN mice. These findings indicated that employing the OP intervention before exposure to chronic HH confers cardioprotection by mitigating excessive myocyte hypertrophy, injury, and myocardial fibrosis. Figure 2: OP mitigates myocardial hypertrophy, injury, and fibrosis during hypobaric hypoxia. (A) Graphical representation of Body Weight (BW) trends of 14 days period by Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP) and Occlusion Preconditioned before HH exposure (OPHH) mice (n= 7-15 mice per experimental group). (B) Graphical presentation of Heart Weight (HW)/BW ratio (n= 5-10 mice per experimental group). (C-H) Indexes for cardiac hypertrophy assessment, including; Representative Wheat germ agglutinin (WGA) staining and Graphical presentation of Cardiomyocyte surface area (n=8-12 cells per section per 4-6 heart per group), Atrial natriuretic peptide (ANP) and Brain natriuretic peptide (BNP) concentrations (n=6-8 hearts per group), Representative Immunoblotting of GATA4 and its Graph plot (n= 3 hearts per group). ELISA were performed in triplicates. Immunoblots were performed in triplicates, and each blot band in the representative blot is an independent biological sample. (I, J) Representative Masson’s trichome staining and Graphical presentation of collagen volume fraction (CVF) showing the extent of fibrosis among experimental groups (n= 3-6 sections per 4,5 hearts per group). $p<0.05, $$$p<0.001 HH vs NN; &p<0.05, &&&p<0.001 APHH vs AP; ##p<0.01, ###p<0.001 OPHH vs OP; p<0.05, p<0.01, p<0.001. Data are expressed as mean ± SEM. Data were analyzed using one-way anova, followed by Tukey’s post hoc analysis. ## OP induces adaptive immunomodulation and metabolic homeostasis during hypobaric hypoxia To understand how severe hypoxia affects immunoregulation in the myocardia, we investigated the phenotype of macrophages infiltrating the heart after chronic exposure to HH. We observed a substantial influx of CD86+ (proinflammatory) macrophages into the myocardial of HH mice, while the CD206+ (anti-inflammatory) populations were repressed (Figures 3A, B; Supplementary Figures 2A-C). Contrarily, CD206+ macrophages outnumbered the CD86+ cells when AP and OP interventions were applied prior to HH exposure. Remarkably, the degree of CD86+ macrophage infiltrations across all the groups corresponded to sera levels of the damage-associated molecular pattern (DAMP) – cardiac troponin I (cTnI) (Figure 3C). Further investigations assessed the concentrations of inflammatory mediators and cytokines in the hearts. The proinflammatory response mediator – iNOS, was overtly upregulated in HH mice but modestly in AP and OP hearts and without any significant alterations in APHH and OPHH hearts; meanwhile, the contrast was observed for the anti-inflammatory mediator, arginase (Arg)-1 (Figures 3D, E). Similarly, we found that proinflammatory cytokines (IL-1β and IL-18) were significantly upregulated in HH but only moderately in APHH and OPHH mice hearts (Figures 3F, G). Additionally, the systemic inflammatory mark (CRP) was prominently upregulation in HH but not in APHH and OPHH mice (Supplementary Figure 2D). However, like Arg-1, the anti-inflammatory cytokines (IL-10 and TGF-β) secretions were repressed in HH but modestly increased in APHH and OPHH hearts (Figures 3H, I). These findings demonstrated that just like AP, the OP intervention circumvents HH-induced myocarditis by minimizing cardiomyocyte necrosis and DAMP secretions – ultimately preventing the induction of hyperactive proinflammatory responses. Figure 3: OP induces adaptive immunomodulation and metabolic homeostasis during hypobaric hypoxia. (A, B) Representative flow cytometry of myocardial macrophages gated on CD45+CD11b+F4/80 and Graphical plots of CD86+ and CD206+. Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP), and Occlusion Preconditioned before HH exposure (OPHH) mice hearts (n=4 hearts per group). $p<0.05 vs NNCD86+; ##p<0.01, ###p<0.001 vs HHCD206+ (C) Graphical presentation of sera cardiac troponin I (cTnI) concentrations. (D, E) Inflammatory mediators; Inducible nitric oxide synthase (iNOS) and Arginase-1 (Arg-1) concentrations assessed by ELISA using myocardia lysates. (F-I) Inflammatory cytokines; Interleukin (IL)-1β, IL-18, IL-10, and transforming growth factor (TGF)-β concentrations assessed by ELISA using myocardia lysates. All ELISA were performed in triplicates (n= 5-8 mice per group). (J-L) Representative Oil Red O (ORO) and Periodic Acid Schiff (PAS) staining of myocardial sections and their respective graphical presentations showing lipid and glycogen depositions percentages (n=4-6 sections per 4-6 mice per group). Yellow outlined boxes are original myocardial portions and red outline boxes are their zoomed-in (5x) inserts to show positive stained area (indicated with black arrows). (M-O) Representative Immunoblotting of Scarb3 and Slc2a1 and their respective Graphical plots; each blot band in the representative blot is an independent biological sample (n= 3 hearts per group). $$p<0.01, $$$p<0.001 HH vs NN; &&p<0.01 APHH vs AP; p<0.05, p<0.01, p<0.001. Data are expressed as mean ± SEM. Data were analyzed using one-way ANOVA, followed by Tukey’s post hoc analysis. Furthermore, besides cardiac fibroblast activation, hypoxia-induced glycolysis shift orchestrates immune cells reprogramming toward proinflammatory phenotypes (18–20). Hence, the metabolic states of the infiltrated CD45+F$\frac{4}{80}$+CD11b+ cells and the entire myocardial were assessed. We observed that lipid metabolism-related gene expressions (Gcdh, Adcd1, Acaa2, Decr1, Hsd17b4, Hadha, Cpt2, Etfb, Echdc2, and Scarb3) in the macrophages isolated from HH hearts were mostly downregulated. In contrast, glycolysis-related genes (Slc2a1, Hk2, Ldha, Aldoc, Fbp1, Pgm2, Gpi1, Pgk1, and Pfkfb3) and the cellular metabolic regulator – mechanistic target of rapamycin (mTOR) mRNA levels were upregulated. Intriguingly, we found only modest downregulation of the lipid metabolism-related genes and slight increases in mRNA levels of the glycolysis-related and mTOR genes in CD45+F$\frac{4}{80}$+CD11b+ cells obtained from APHH and OPHH hearts (Supplementary Figure 2E). Consistently, Oil red O and PAS staining showed HH mice myocardial had abundant lipid depositions compared to the NN mice, while glycogen and other polysaccharide contents were substantially depleted. Meanwhile, OPHH mice myocardial revealed a balance constitution of lipid metabolism and glycolysis substrates, as similarly observed in APHH hearts (Figures 3J–L). These phenomena were validated by immunoblotting myocardial protein lysate, which revealed that HH hearts had decreased expression of the fatty acid transporter (Scarb3). In contrast, the glucose transporter protein (Slc2a1) was increased – indicating a glycolysis shift in HH hearts but not entirely in APHH and OPHH hearts (Figures 3M–O). Overall, these findings indicate that the OP intervention exerts immunomodulation by adaptively regulating metabolic shifts to prevent glycolytic reprogramming of macrophages towards proinflammatory phenotypes during HH. ## OP induces adaptive modulation of oxidative stress responses Redox homeostasis (balance between ROS and antioxidants) signaling are major alterations occurring during chronic hypoxia [7]. Hence, to elucidate the underlying mechanisms employed by the OP intervention during HH, we investigated its impact on oxidative stress regulators – hypoxia-inducible factors (HIF-1α and HIF-2α) and antioxidant response element-dependent genes regulator, nuclear factor erythroid 2–related factor 2 (Nrf2). Compared to the control group (NN), we observed significant decreases of HIF-1α expression in HH mice hearts and further sharp declines in the protein levels when AP or OP interventions were applied before HH exposure. Conversely, we observed that HIF-2α and Nrf2 expressions were reduced substantially in HH but modestly upregulated in APHH and OPHH mice hearts (Figures 4A–D). Nicotinamide adenine dinucleotide (NAD+) and NADH ratio are good predictors of the redox homeostasis state [21]; as such, NAD+ and NADH contents in the heart were assessed. The outcomes demonstrated that AP and OP increased NAD+/NADH ratio modestly compared to NN. Secondly, HH hearts had a ~$60\%$ decrease in NAD+ and ~$35\%$ increase in NADH contents, thereby decreasing the NAD+/NADH ratio compared with NN, APHH, and OPHH hearts (Figures 4E–G). This indicated increased ROS with deficient antioxidant defenses in HH hearts but not in APHH and OPHH hearts which employed AP and OP interventions, respectively. The total antioxidant capacity assays (T-AOC) performed with myocardial lysates validated the prior statement. T-AOC of HH hearts reduced significantly but remained unaltered in APHH and OPHH, compared to NN hearts (Figure 4H). Thus, similar to AP, the OP intervention reinforces antioxidant responses to confer protection against hypoxia-induced oxidative stress damage – improving survival rates of OPHH compared to HH mice (Figure 4I). Figure 4: OP induces adaptive modulation of oxidative stress responses. (A-D) Representative Immunoblotting of Hypoxia-inducible factors (HIF)-1α, HIF-2α, and nuclear factor erythroid 2–related factor 2 (Nrf2), and their respective Graphical plots; each blot band in the representative blot is an independent biological sample (n= 4 hearts per group). Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP), and Occlusion Preconditioned before HH exposure (OPHH) mice hearts. (E-H) Antioxidants state indexes; Graphical plots of the concentrations of redox cofactors, Nicotinamide adenine dinucleotide (NAD)+hydrogen (NADH) and their ratio, as well as the Total antioxidant capacity (T-AOC) of the myocardia (n=8 hearts per group). (I) Graphical plot of survival data in Kaplan-Meier estimator (n=18 mice per group). $p<0.05, $$p<0.01, $$$p<0.001 HH vs NN; &p<0.05 APHH vs AP; #p<0.05 OPHH vs OP; p<0.01, *p<0.001. Data are expressed as mean ± SEM. Data were analyzed using one-way ANOVA, followed by Tukey’s post hoc analysis. Survival curves were analyzed with the Kaplan-Meier estimator. ## OPHH mice are resilient to HH-associated maladaptive behavioral outcomes Since OP intervention prevented myocarditis and distortion of cEA in OPHH mice, we next tested whether it influenced behavioral outcomes. We observed that HH mice exhibited the most dullness among the experimental groups. Utilizing the open field test (OFT) to validate our observation, it was determined that the HH mice had the most decreases in locomotive function – with the least total distance moved, average velocity, and mobile duration compared to NN, APHH, and OPHH mice. Also, HH mice exhibited significant immobile duration in the OFT, but this was not the case for APHH and OPHH mice (Figures 5A–F). Thus, similar to AP in APHH, the OP intervention resulted in only modest reductions of locomotive functions in OPHH mice. Figure 5: OPHH mice are resilient to HH-associated maladaptive behavioral outcomes. (A-H) Locomotive and exploratory behavioral assessment indexes from Open field test (OFT), including; Representative plots of Total distance moved and its graphical presentation, Representative plots of Average velocity, Mobile duration, and Immobile duration, and their graphical presentations and Representative heat-maps of Relative zone entries and its graphical presentation (n=6-10 mice per group). Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP), and Occlusion Preconditioned before HH exposure (OPHH) mice. (I, J) Anxiety-related and exploratory behavioral assessment indexes from Elevated plus maze (EPM), including; Representative heat-maps of Relative arm entries and its graphical presentation (n=6-10 mice per group). $p<0.05 HH vs NN; p<0.05. Data are expressed as mean ± SEM. Data were analyzed using one-way ANOVA, followed by Tukey’s post hoc analysis.* In addition, most HH mice were observed to restrict their movement to the peripheries (zone A) and corners (zone B) but avoided the central (zone C) of the OFT apparatus, thereby, had the least relative entries into zone C (Figures 5G, H) – a phenomenon shown to depict an increase in anxiety-related behaviors [22]. Intriguingly, neither APHH nor OPHH mice exhibited the same exploratory behavior as HH mice, although they were all exposed to chronic hypoxia. The elevated plus maze (EPM) was used to confirm increased anxiety-related behaviors in HH mice. Most HH mice refrained from exploring the EPM apparatus’s open arms and mostly limited their movement to within the closed arm; hence, they had the least relative open arm entries (Figures 5I, J). Conversely, APHH and OPHH mice still demonstrated exploratory patterns in both open and closed arms similar to NN and the respective preconditioning groups. Taken together, the findings from OFT and EPM indicated that chronic hypoxia exposure increased susceptibility to developing anxiety-related behavior, as reported previously [22]. Meanwhile, like AP to APHH, OP intervention makes OPHH mice resilient to hypoxia-induced negative behavioral outcomes. ## OP enhances respiratory and oxygen-carrying capacity in humans Still, in attempts to elucidate the underlying mechanism employed by the OP intervention, we assessed its impact on respiratory and oxygen-carrying capacity in humans. Cardiopulmonary exercise tests were performed on the healthy human subjects before (BOP) and after (AOP) the application of OP for 6 consecutive days. We found that HRs and diastolic and systolic blood pressures were slightly decreased in AOP compared to BOP (Figures 6A–C). Also, the inhalation and exhalation vital capacities were increased substantially while expiratory reserve volume elevated modestly in AOP compared to BOP (Figures 6D–F). The changes observed in the aforementioned indexes suggest that OP induces adaptive respiratory response and lowers the risk of pulmonary injuries and complications. Additionally, we found that after performing OP, minute ventilation (V’E), carbon dioxide output (V’CO2), oxygen uptake (V’O2), oxygen saturation (SpO2), oxygen pulse (V’O2/HR), and the metabolic equivalent of task (MET) were all significantly improved during the maximal workload (Max Watts) phase of exercising with the cycle ergometer (Figures 6G–L). Thus, OP enhanced the oxygen-carrying capacity and endurance in humans – an adaptation shown to mitigate the deleterious effects of severe hypoxia [23]. Lastly, we observed that in AOP, the respiratory exchange ratio (RER) was lowered and took more time to reach ≥1.00 compared to BOP (Figure 6M). Consolidating our observation in the mice myocardia and macrophages, the lowered RER in AOP showed that OP induces mechanisms that mitigate the extent of metabolic shift to glycolysis. Figure 6: OP enhances respiratory and oxygen-carrying capacity in humans. (A-M) Cardiopulmonary Exercise Test (CPET) indexes for respiratory and oxygen carrying capacity in humans (n=14 human volunteers), Before occlusion preconditioning (BOP) and After occlusion preconditioning (AOP), including; Graphical plots of Heart Rates (HR), Systolic blood pressures (Psys), Diastolic blood pressures (Pdia), Vital capacity of inhalation (IN), Vital capacity of exhalation(EX), Expiratory reserve volume, Minute ventilation (V’E), Carbon dioxide output (V’CO2), Oxygen uptake (V’O2), Oxygen saturation (SpO2), Oxygen pulse (V’O2/HR), Metabolic equivalent of task (MET) and Respiratory exchange ratio (RER). Rest, Warm-up, Second ventilation threshold (VT2), Maximal workload (Max Watts) and Recovery are timepoints of interest during the CPET. p<0.05, *p<0.01 AOP vs BOP. Data are expressed as mean ± SEM. Data were analyzed using an unpaired t-test for comparing two groups and two-way ANOVA for grouped analysis. ## β2AR is implicated in OP-induced adaptive responses against hypobaric hypoxia The pleiotropic nature of β2AR makes it an essential mediator for most adaptive responses to stressful conditions in the heart and immunoregulation [14]. Concordant with previous reports, we found β2AR overexpressed in HH compared to NN [24]. Meanwhile, AP, APHH, OP, and OPHH hearts had modest upregulation of β2AR (Supplementary Figure 3A, B). By utilizing β2AR knockout (Adrb2-/-) mice, we explored the β2AR’s involvement in HH-induced myocarditis and arrhythmias and OP-induced cardioprotective against HH. Consistently, the OP intervention prevented significant body weight loss due to hypoxia in the OPHHAdrb2+/+ mice; however, the contrast was observed in the OPHHAdrb2-/- mice (Figure 7A). While the HRs remained similar in OPHHAdrb2+/+ and OPHHAdrb2-/-, the EKG revealed arrhythmias in the latter (Figure 7B; Supplementary Figure 3C). We found that despite the occurrence of other forms of arrhythmias, the deletion of β2AR (in HHAdrb2-/-) circumvented the long QT, QTc, JT, and Tpeak-Tend intervals observed in HHAdrb2+/+. Intriguingly, this phenomenon was reverted when OP was applied to Adrb2-/- mice prior to HH exposure. The arrhythmias worsened in OPHHAdrb2-/- as QT, QTc, JT, and Tpeak-Tend intervals prolongations and distortion of other EKG indexes were aggravated, compared to OPHHAdrb2+/+ (Figures 7C, D; Supplementary Figure 3D, E and Supplementary Table 3). These observed outcomes show that β2AR is involved in the OP-induced signaling cascades to preserve cEA during hypoxia. Figure 7: β2AR is implicated in OP-induced adaptive responses against hypobaric hypoxia (A) Graphical presentation of Body weight (BW) alteration trends among Wild type (Adrb2+/+) and β2AR knockout (Adrb2-/-) mice in experiment groups; Normobaric Normoxia (NN), Hypobaric Hypoxia (HH), Altitude Preconditioned (AP), Altitude Preconditioned before HH exposure (APHH), Occlusion Preconditioned (OP) and Occlusion Preconditioned before HH exposure (OPHH) (n=5-15 mice per group). (B-D) Graphical presentation of electrocardiogram (EKG) indexes, including; Heart Rate, QT Interval, and JT interval (n=5-9 mice per group). (E-G) Inflammatory biomarker; C-reactive protein, Inducible nitric oxide synthase (iNOS), and Arginase-1 (Arg-1) concentrations assessed by ELISA. Assays were performed in triplicates (n=4 mice per group). (H-K) Representative Immunoblotting of Hypoxia-inducible factors (HIF)-1α, HIF-2α, and nuclear factor erythroid 2–related factor 2 (Nrf2), and their respective Graphical plots; each blot band in the representative blot is an independent biological sample (n= 3 hearts per group). (L) Graphical plots of the concentrations of Total antioxidant capacity (T-AOC) of myocardia (n=4 hearts per group).$p<0.05, $$p<0.01, $$$p<0.001 vs NNAdrb2+/+; &p<0.05, &&p<0.01, &&&p<0.001 vs HHAdrb2+/+; ##p<0.01 vs OPAdrb2+/+; ¥p<0.05, ¥¥p<0.01, ¥¥¥p<0.001 vs OPHHAdrb2+/+;!!!p<0.001 vs NNAdrb2-/-; %p<0.001 vs HHAdrb2-/-; £p<0.05, ££p<0.01, £££p<0.001 vs OPAdrb2-/-. Data are expressed as mean ± SEM. Data were analyzed using two-way ANOVA. Also, checking inflammatory markers, we determined the role of β2AR in OP-induced immunomodulation. At the baseline, proinflammatory responses (CRP and iNOS) were found further aggravated in HHAdrb2-/- than in HHAdrb2+/+ mice. While the OP intervention significantly mitigated CRP and iNOS upregulations in OPHHAdrb2+/+, this phenomenon was abolished by β2AR obliteration in OPHHAdrb2-/- mice (Figures 7E, F). Additionally, we found Arg-1 expression sustained in OPHHAdrb2+/+ but downregulated in the OPHHAdrb2-/- (Figure 7G). Overall, the loss of adaptive immunoregulation in OPHHAdrb2-/- compared to OPHHAdrb2+/+ mice suggests that β2AR-mediated signaling cascades are implicated in OP-induced immunomodulatory mechanisms. Next, as oxidative stress-responsive genes were adaptively modulated in the wild-type OPHH mice heart, we examined whether β2AR participated in HIF-1α, HIF-2α, and Nrf2 expressions regulation. We observed that, unlike HIF-2α and Nrf2, HIF-1α was stabilized in HHAdrb2-/- and also refractory to OP in Adrb2-/-, but not in the Adrb2+/+ hearts – suggesting that β2AR mediates the destabilization of HIF-1α (Figures 7H–K). Conversely, HIF-2α and Nrf2 expressions were sustained in OPHHAdrb2+/+ but not in OPHHAdrb2-/- hearts, indicating that β2AR scaffolded the stabilization of these proteins. Further, we investigated the impact of these oxidative stress-responsive proteins alterations on the antioxidant capacity and found that obliteration of β2AR (in OPHHAdrb2-/-) weakened the cardioprotective antioxidant defense mechanisms induced by OP against hypobaric hypoxia (Figure 7L). Thus, these findings taken together shows that β2AR participates in multiple adaptive responses induced by OP to circumvent the adverse outcomes of chronic HH exposure. ## Discussions The fear of experiencing altitude sickness deters civilian workers, hikers, tourists, and even defense personnel from going to places 2500 m above sea level for occupational or leisure purposes. Also, the possibility of HH at such altitudes to induce or aggravate myocarditis, arrhythmia, and ultimately heart failure due to adverse cardiac remodeling has made clinicians to advised those prone to cardiovascular complications and physiologically unprepared individuals against visiting high altitudes (25–27). Studies over the years have made attempts at elucidating the underlying pathomechanisms of altitude sickness and have generally demonstrated that the clinical manifestations observed are due to HH-induced maladaptive oxidative stress responses (imbalance between ROS and antioxidants), metabolic dysregulation and dampened anti-inflammatory defenses [7, 28, 29]. Currently, AP and salidroside are the primary preventive therapeutic interventions being employed to circumvent or mitigate the adverse effects of HH exposure. Even so, inaccessibility to hypoxic chambers and unavailability of salidroside due to geographical limitations are the respective drawbacks of these interventions. Our study aimed to explore OP – which is potent in preventing hypoxia-induced damages (11–13), as an alternative therapeutic intervention for HH-induced myocarditis and arrhythmias. To ascertain the efficacy of OP intervention (in OPHH), its impacts on EKG, cardiac architecture, immunomodulation, oxidative stress regulation, and behavior outcomes were compared with APHH. Concordant with previous studies [30, 31], we found that HH-induced tachycardia and prolongations of QT, QTc, JT, and Tpeak-Tend intervals, and ST height, P duration, and R and T amplitudes. Reportedly, these observations are because at HH; there is an increase in sympathetic activity, which triggers prolongation of repolarization, resulting in arrhythmia, heart failure, and sudden death [30]. However, AP has been demonstrated to prevent significant disruption of cEA [32]; consistently, our findings showed similar outcomes. Intriguingly, we observed that OPHH mice had fewer alterations in the EKG indexes than APHH mice compared to the NN mice. This led to the conclusion that OP preserved cEA modestly better than AP during HH exposure. Furthermore, unlike in HH mice, we observed that BW losses were only modest, and the extent of cardiomyocyte hypertrophies, injury, and fibrosis were mitigated in APHH and OPHH hearts. Lippl et al. and others have similarly shown that at high altitudes, there is a loss of appetite hence the excessive BW [33, 34]. Also, to compensate for oxygen demand, hypertrophy cascades are induced, resulting in excessive enlargement of cardiomyocytes and their apoptosis/necrosis, which in turn drives proinflammatory and fibrotic responses [33, 35]. Interestingly, both AP and OP have been shown to lessen adverse cardiac remodeling during hypoxic or ischemic events, just as we observed – and it has also been suggested that both interventions might have similar underlying mechanisms [32, 36]. While OP’s cardiac cardioprotection has been demonstrated mainly against ischemia/reperfusion injury, we show here for the first time that OP intervention is potent against HH-induced myocardial hypertrophy, injury, and fibrosis. Myocarditis scaffolded by unresolved proinflammatory responses drives the maladaptive remodeling of the heart in HH [5]; hence we investigated OP’s effect on immunomodulation. Typically observed at injured tissues or inflamed sites [37], we found massive infiltrations of proinflammatory (CD86+) macrophages in HH myocardia, whiles the reparative (CD206+) macrophage populations were significantly less. Also, inflammatory cytokines concentrations were altered in a similar fashion in the HH myocardia. In contrast, employing the OP intervention facilitated anti-inflammatory defenses while minimizing the proinflammatory responses in OPHH, as AP did in APHH. While we demonstrated OP’s immunoregulation in OPHH; Gorjipour et al. ’s earlier works had shown similar observations where the OP intervention enhanced the elevation of IL-10 while downregulating the circulation IL-8 to confer cardioprotection in coronary artery bypass graft surgery [38]. The metabolic state of inflammatory cells crucially influences their immune responses and functions [39]; as such, we sought to ascertain the metabolic state of infiltrated macrophages and the entire myocardia in further investigations. Interestingly, we found that glycolysis-related genes had increased ~3 folds while that lipid metabolism-related genes were downregulated in macrophages isolated from HH hearts. These findings are consistent with metabolic shifts, which facilitate biased reprogramming of macrophages toward proinflammatory phenotypes [19, 39]. Similarly, we observed that the entire HH myocardia had increased lipid deposition while glycogen and other polysaccharide contents were substantially depleted – all of which are consistent with glycolysis shift [40]. Contrarily, OP intervention modulated metabolic homeostasis by preventing complete glycolysis shift [41], thereby impeding the reprogramming of macrophages towards proinflammatory phenotypes in OPHH, as similarly done by AP in APHH. Also, disruption in redox homeostasis is a cofactor in HH-induced cardiac dysfunction [7], and our findings consolidated this fact, as HIF-1α, HIF-2α and Nrf2 expression were declining in HH mice hearts. Even so, it was observed that by employing OP intervention, HIF-2α and Nrf2 but not HIF-1α expressions were rescued and sustained in OPHH mice hearts. Similar outcomes were found in APHH mice hearts. These findings indicated that, like AP, OP stimulates adaptive oxidative stress regulation by reinforcing antioxidant responses. Consistently, OP has been shown to improve antioxidant defenses by enhancing NAD+ levels, which directly promotes Nrf2 antioxidant activities [42, 43]. Next, we investigated OP’s impact on the NAD+/NADH ratio. We found that the OP intervention had increased the NAD+/NADH ratio and prevented significant decreases in OPHH hearts, which contributed to sustaining their T-AOC like in APHH hearts, while we observed declines in HH hearts. In line with our finding, Morris-Blanco et al. previously demonstrated that OP did increase NAD+/NADH ratio via protein kinase C epsilon (PKCε) in neuronal-glial [42], hence it is suggestive that similar mechanisms might be involved here. Overall, compared to HH, survival rates improved in OPHH and APHH. Cardiovascular events promote anxiety-related behavior and vice versa; as such, it has become imperative to assess the behavioral outcomes of interventions targeted at improving cardiac health [44, 45]. Typically, HH has been shown decrease locomotive functions while increasing the levels of anxiety and depression in humans [1]. These were confirmed in the HH mice model as they had decreased total distance moved, average velocity, and mobile duration and mostly refrained from the central zone and open arms in the OFT and EPM apparatuses, respectively, during their locomotive and exploratory activities. Contrarily, like APHH mice, OPHH mice combed throughout zones and arms of the OFT and EPM apparatuses – indicating that similar to AP [22], the OP intervention makes mice resilient to HH-induced anxiety-related behavior and decreased locomotive functions. To have prevented HH-induced cardiac remodeling, it was hypothesized that the OP intervention must have induced mechanisms to facilitate adequate tissue oxygenation. Surprisingly and in validating our hypothesis, CPET parameters in AOP showed enhanced respiratory and oxygen-carrying capacity and endurance in humans as V’E, V’CO2, V’O2, SpO2, V’O2/HR, and MET were all substantially improved at Max Watts (Maximal workload). Concordantly, it has been shown that OP mitigates declines in regional oxygenation to confer cardioprotection when exposed to HH [46]. Also, OP modestly delayed the time for RER=1 in AOP compared to BOP; hence, indicating that the intervention induces mechanisms that mitigate the extent of metabolic shift to glycolysis, just as shown earlier and reported by others [41]. Lastly, we had previously demonstrated the adaptive roles of β2AR in cardioprotection and immunoregulation during stressful conditions [14]; hence we sought to investigate its implication in OP-induced cardioprotective against HH. We observed that HH mice characterized with arrhythmias had β2AR expressions drastically increased in their hearts. Consistent with our observation, Lang et al. demonstrated that the overexpression of β2AR significantly increases the predisposition to the occurrence of arrhythmias [47]. Surprisingly, β2AR deletion prevented long QT, QTc, JT, and Tpeak-Tend intervals in HHAdrb2-/-, mimicking the effect of β-blockers in preventing long QT syndrome [48]. Meanwhile, most of the adaptive responses we observed in OPHHAdrb2+/+ mice were abolished in OPHHAdrb2-/- mice as their BW losses were substantial, arrhythmia worsened, proinflammatory responses heightened against anti-inflammatory responses, and antioxidant defenses declined significantly. Consistent with our observations, carvedilol (a β-blockers) was shown to have abolished the cardioprotection conferred by OP during cardiac surgery [49]. In conclusion, by preserving cEA, mitigating cardiac remodeling, facilitating adaptive immunomodulation and oxidative stress responses, sustaining homeostasis in metabolic shifts, causing resilient against anxiety-related behaviors, and enhancing respiratory and oxygen-carrying capacity, OP demonstrates as a potent alternative therapeutic intervention for preventing HH-induced adverse effects on cardiac and overall health. Even so, OP is not recommended as an intervention for individuals on β-blocker medications prior to their visit to high altitudes/HH environments, as these medications blunt the cardioprotection conferred by the intervention and conversing aggravates myocarditis and arrhythmias. Lastly, immunoregulatory and metabolism homeostasis induced by OP is suggestive of its potential to ameliorate the progression of other inflammatory, metabolic and oxidative stress-related diseases. ## Data availability statement The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors. ## Ethics statement The studies involving human participants were reviewed and approved by Ethics Committee of the Xuzhou Medical University. The patients/participants provided their written informed consent to participate in this study. The animal study was reviewed and approved by Experimental Animal Centre of Xuzhou Medical University and the Animal Ethics Committee of the Xuzhou Medical University. ## Author contributions GKA and HS conceived and designed the project. GA, RM, RR, AA, MN, and PW performed animal models. GA, RM, RR, AA, and MN performed most experiments and analyzed data. GA, RR, AA, MN, PW, SA and JA-A performed assays. Cardiopulmonary Exercise Test was facilitated by JX, and FW, GA, LF, and JX conducted examinations and analyses. GA, SA, and XL performed flow cytometry experiments and analysis. GA, YL, and RM conducted behavioral experiments and analyses. GA wrote the manuscript with input from all authors and was approved by JX and HS. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1124649/full#supplementary-material ## References 1. Neuhaus C, Hinkelbein J. Cognitive responses to hypobaric hypoxia: Implications for aviation training. Psychol Res Behav Manag (2014) 7 297-302. DOI: 10.2147/prbm.s51844 2. Faulhaber M, Flatz M, Gatterer H, Schobersberger W, Burtscher M. Prevalence of cardiovascular diseases among alpine skiers and hikers in the Austrian Alps. High Alt Med Biol (2007) 8. DOI: 10.1089/ham.2007.1005 3. Mrakic-Sposta S, Gussoni M. Effects of acute and sub-acute hypobaric hypoxia on oxidative stress: a field study in the Alps. Eur J Appl Physiol (2021) 121 297-306. DOI: 10.1007/s00421-020-04527-x 4. Abe H, Semba H, Takeda N. The roles of hypoxia signaling in the pathogenesis of cardiovascular diseases. J Atheroscler Thromb (2017) 24. DOI: 10.5551/jat.RV17009 5. Chouvarine P, Legchenko E, Geldner J, Riehle C, Hansmann G. Hypoxia drives cardiac miRNAs and inflammation in the right and left ventricle. J Mol Med (Berl) (2019) 97. DOI: 10.1007/s00109-019-01817-6 6. Wang N, Song J, Zhou G, Li W, Ma H. Mechanism of salidroside relieving the acute hypoxia-induced myocardial injury through the PI3K/Akt pathway. Saudi J Biol Sci (2020) 27. DOI: 10.1016/j.sjbs.2020.04.035 7. González-Candia A, Candia AA, Paz A, Mobarec F, Urbina-Varela R, Campo AD. Cardioprotective antioxidant and anti-inflammatory mechanisms induced by intermittent hypobaric hypoxia. Antioxidants (Basel) (2022) 11 1043. DOI: 10.3390/antiox11061043 8. Chen CY, Hou CW, Bernard JR, Chen CC, Hung TC, Cheng LL. Rhodiola crenulata- and cordyceps sinensis-based supplement boosts aerobic exercise performance after short-term high altitude training. High Alt Med Biol (2014) 15. DOI: 10.1089/ham.2013.1114 9. Xiong Y, Wang Y, Xiong Y. Protective effect of salidroside on hypoxia-related liver oxidative stress and inflammation. Food Sci Nutr (2021) 9. DOI: 10.1002/fsn3.2459 10. Kubentayev SA, Zhumagul MZ. Current state of populations of rhodiola rosea l. (Crassulaceae) in East Kazakhstan. Bot Stud (2021) 62 19. DOI: 10.1186/s40529-021-00327-4 11. Murry CE, Jennings RB, Reimer KA. Preconditioning with ischemia: A delay of lethal cell injury in ischemic myocardium. Circulation (1986) 74. DOI: 10.1161/01.cir.74.5.1124 12. Hausenloy DJ, Yellon DM. Remote ischaemic preconditioning: underlying mechanisms and clinical application. Cardiovasc Res (2008) 79. DOI: 10.1093/cvr/cvn114 13. Lau JK, Roy P, Javadzadegan A, Moshfegh A, Fearon WF, Ng M. Remote ischemic preconditioning acutely improves coronary microcirculatory function. J Am Heart Assoc (2018) 7 e009058. DOI: 10.1161/jaha.118.009058 14. Hou H, Adzika GK, Wu Q, Ma T, Ma Y, Geng J. Estrogen attenuates chronic stress-induced cardiomyopathy by adaptively regulating macrophage polarizations. Front Cell Dev Biol (2021) 9. DOI: 10.3389/fcell.2021.737003 15. Song Y, Xu C, Liu J, Li Y, Wang H, Shan D. Heterodimerization with 5-HT(2B)R is indispensable for β(2)AR-mediated cardioprotection. Circ Res (2021) 128. DOI: 10.1161/circresaha.120.317011 16. Doggett TM, Tur JJ, Alves NG, Yuan SY, Tipparaju SM, Breslin JW. Assessment of cardiovascular function and microvascular permeability in a conscious rat model of alcohol intoxication combined with hemorrhagic shock and resuscitation. Methods Mol Biol (2018) 1717 61-81. DOI: 10.1007/978-1-4939-7526-6_6 17. Zhang L, Li C, Yang L, Adzika GK, Machuki JO, Shi M. Estrogen protects vasomotor functions in rats during catecholamine stress. Front Cardiovasc Med (2021) 8. DOI: 10.3389/fcvm.2021.679240 18. Jing C, Castro-Dopico T. Macrophage metabolic reprogramming presents a therapeutic target in lupus nephritis. Proc Natl Acad Sci USA (2020) 117. DOI: 10.1073/pnas.2000943117 19. Sun L, Yang X, Yuan Z, Wang H. Metabolic reprogramming in immune response and tissue inflammation. Arterioscler Thromb Vasc Biol (2020) 40 1990-2001. DOI: 10.1161/atvbaha.120.314037 20. Chen ZT, Gao QY, Wu MX, Wang M, Sun RL, Jiang Y. Glycolysis inhibition alleviates cardiac fibrosis after myocardial infarction by suppressing cardiac fibroblast activation. Front Cardiovasc Med (2021) 8. DOI: 10.3389/fcvm.2021.701745 21. Zhang Z, Xu HN, Li S, Antonio D, Chellappa K, Davis JG. Rapamycin maintains NAD(+)/NADH redox homeostasis in muscle cells. Aging (Albany NY) (2020) 12. DOI: 10.18632/aging.103954 22. Kushwah N, Jain V, Deep S, Prasad D, Singh SB, Khan N. Neuroprotective role of intermittent hypobaric hypoxia in unpredictable chronic mild stress induced depression in rats. PloS One (2016) 11. DOI: 10.1371/journal.pone.0149309 23. Zhang J, Wu Y, Peng XY, Li QH, Xiang XM, Zhu Y. The protective effect of a novel cross-linked hemoglobin-based oxygen carrier on hypoxia injury of acute mountain sickness in rabbits and goats. Front Physiol (2021) 12. DOI: 10.3389/fphys.2021.690190 24. Cheong HI, Asosingh K, Stephens OR, Queisser KA, Xu W, Willard B. Hypoxia sensing through β-adrenergic receptors. JCI Insight (2016) 1. DOI: 10.1172/jci.insight.90240 25. Bärtsch P, Gibbs JS. Effect of altitude on the heart and the lungs. Circulation (2007) 116. DOI: 10.1161/circulationaha.106.650796 26. Messerli-Burgy N, Meyer K, Steptoe A, Laederach-Hofmann K. Autonomic and cardiovascular effects of acute high altitude exposure after myocardial infarction and in normal volunteers. Circ J (2009) 73. DOI: 10.1253/circj.cj-09-0004 27. Louis Hofstetter US, Stefano F. Going to high altitude with heart disease. Cardiovasc Med (2017) 20 87-95. DOI: 10.4414/cvm.2017.00478 28. Pham K, Parikh K, Heinrich EC. Hypoxia and inflammation: Insights from high-altitude physiology. Front Physiol (2021) 12. DOI: 10.3389/fphys.2021.676782 29. Xu J, Chen WJ, Wang Z, Xin MY, Gao SH, Liu WJ. Profiles of transcriptome and metabolic pathways after hypobaric hypoxia exposure. Proteome Sci (2022) 20 16. DOI: 10.1186/s12953-022-00198-y 30. Roche F, Reynaud C, Pichot V, Duverney D, Costes F, Garet M. Effect of acute hypoxia on QT rate dependence and corrected QT interval in healthy subjects. Am J Cardiol (2003) 91. DOI: 10.1016/s0002-9149(03)00040-7 31. Mikołajczak K, Czerwińska K, Pilecki W, Poręba R, Gać P. The impact of temporary stay at high altitude on the circulatory system. J Clin Med (2021) 10 1622. DOI: 10.3390/jcm10081622 32. Estrada JA, Williams AG, Sun J, Gonzalez L, Downey HF, Caffrey JL. δ-opioid receptor (DOR) signaling and reactive oxygen species (ROS) mediate intermittent hypoxia induced protection of canine myocardium. Basic Res Cardiol (2016) 111 17. DOI: 10.1007/s00395-016-0538-5 33. Chouabe C, Ricci E, Amsellem J, Blaineau S, Dalmaz Y, Favier R. Effects of aging on the cardiac remodeling induced by chronic high-altitude hypoxia in rat. Am J Physiol Heart Circ Physiol (2004) 287. DOI: 10.1152/ajpheart.00199.2004 34. Lippl FJ, Neubauer S, Schipfer S, Lichter N, Tufman A, Otto B. Hypobaric hypoxia causes body weight reduction in obese subjects. Obes (Silver Spring) (2010) 18. DOI: 10.1038/oby.2009.509 35. Pena E, Brito J. Oxidative stress, kinase activity and inflammatory implications in right ventricular hypertrophy and heart failure under hypobaric hypoxia. Int J Mol Sci (2020) 21. DOI: 10.3390/ijms21176421 36. Surendra H, Diaz RJ, Harvey K, Tropak M, Callahan J, Hinek A. Interaction of δ and κ opioid receptors with adenosine A1 receptors mediates cardioprotection by remote ischemic preconditioning. J Mol Cell Cardiol (2013) 60. DOI: 10.1016/j.yjmcc.2013.04.010 37. Krzyszczyk P, Schloss R, Palmer A, Berthiaume F. The role of macrophages in acute and chronic wound healing and interventions to promote pro-wound healing phenotypes. Front Physiol (2018) 9. DOI: 10.3389/fphys.2018.00419 38. Gorjipour F, Saeedzadeh T, Toloueitabar Y, Kachoueian N, Bahlouli Ghashghaei S, Mortazian M. Remote ischemic preconditioning effects on inflammatory markers and myocardial protection in coronary artery bypass graft surgery. Perfusion (2022) 37 56-61. DOI: 10.1177/0267659120979293 39. Shirai T, Nazarewicz RR, Wallis BB, Yanes RE, Watanabe R, Hilhorst M. The glycolytic enzyme PKM2 bridges metabolic and inflammatory dysfunction in coronary artery disease. J Exp Med (2016) 213. DOI: 10.1084/jem.20150900 40. Yan J, Young ME, Cui L, Lopaschuk GD, Liao R, Tian R. Increased glucose uptake and oxidation in mouse hearts prevent high fatty acid oxidation but cause cardiac dysfunction in diet-induced obesity. Circulation (2009) 119. DOI: 10.1161/circulationaha.108.832915 41. Geng J, Zhang Y, Li S, Li S, Wang J, Wang H. Metabolomic profiling reveals that reprogramming of cerebral glucose metabolism is involved in ischemic preconditioning-induced neuroprotection in a rodent model of ischemic stroke. J Proteome Res (2019) 18 57-68. DOI: 10.1021/acs.jproteome.8b00339 42. Morris-Blanco KC, Cohan CH, Neumann JT, Sick TJ, Perez-Pinzon MA. Protein kinase c epsilon regulates mitochondrial pools of nampt and NAD following resveratrol and ischemic preconditioning in the rat cortex. J Cereb Blood Flow Metab (2014) 34. DOI: 10.1038/jcbfm.2014.51 43. Xie N, Zhang L, Gao W, Huang C. NAD(+) metabolism: Pathophysiologic mechanisms and therapeutic potential. Signal Transduct Target Ther (2020) 5 227. DOI: 10.1038/s41392-020-00311-7 44. Abed MA, Kloub MI, Moser DK. Anxiety and adverse health outcomes among cardiac patients: a biobehavioral model. J Cardiovasc Nurs (2014) 29. DOI: 10.1097/JCN.0b013e318292b235 45. Pino EC, Zuo Y, Borba CP, Henderson DC, Kalesan B. Clinical depression and anxiety among ST-elevation myocardial infarction hospitalizations: Results from nationwide inpatient sample 2004-2013. Psychiatry Res (2018) 266 291-300. DOI: 10.1016/j.psychres.2018.03.025 46. Zhong Z, Dong H, Wu Y, Zhou S, Li H, Huang P. Remote ischemic preconditioning enhances aerobic performance by accelerating regional oxygenation and improving cardiac function during acute hypobaric hypoxia exposure. Front Physiol (2022) 13. DOI: 10.3389/fphys.2022.950086 47. Lang D, Holzem K, Kang C, Xiao M, Hwang HJ, Ewald GA. Arrhythmogenic remodeling of β2 versus β1 adrenergic signaling in the human failing heart. Circ Arrhythm Electrophysiol (2015) 8. DOI: 10.1161/circep.114.002065 48. Vincent GM, Schwartz PJ, Denjoy I, Swan H, Bithell C, Spazzolini C. High efficacy of beta-blockers in long-QT syndrome type 1: contribution of noncompliance and QT-prolonging drugs to the occurrence of beta-blocker treatment "failures". Circulation (2009) 119. DOI: 10.1161/circulationaha.108.772533 49. Cho YJ, Nam K, Kim TK, Choi SW, Kim SJ, Hausenloy DJ. Sevoflurane, propofol and carvedilol block myocardial protection by limb remote ischemic preconditioning. Int J Mol Sci (2019) 20 269. DOI: 10.3390/ijms20020269
--- title: 'Differences of gut microbiota and behavioral symptoms between two subgroups of autistic children based on γδT cells-derived IFN-γ Levels: A preliminary study' authors: - Xin-Jie Xu - Ji-Dong Lang - Jun Yang - Bo Long - Xu-Dong Liu - Xiao-Feng Zeng - Geng Tian - Xin You journal: Frontiers in Immunology year: 2023 pmcid: PMC9975759 doi: 10.3389/fimmu.2023.1100816 license: CC BY 4.0 --- # Differences of gut microbiota and behavioral symptoms between two subgroups of autistic children based on γδT cells-derived IFN-γ Levels: A preliminary study ## Abstract ### Background Autism Spectrum Disorders (ASD) are defined as a group of pervasive neurodevelopmental disorders, and the heterogeneity in the symptomology and etiology of ASD has long been recognized. Altered immune function and gut microbiota have been found in ASD populations. Immune dysfunction has been hypothesized to involve in the pathophysiology of a subtype of ASD. ### Methods A cohort of 105 ASD children were recruited and grouped based on IFN-γ levels derived from ex vivo stimulated γδT cells. Fecal samples were collected and analyzed with a metagenomic approach. Comparison of autistic symptoms and gut microbiota composition was made between subgroups. Enriched KEGG orthologues markers and pathogen-host interactions based on metagenome were also analyzed to reveal the differences in functional features. ### Results The autistic behavioral symptoms were more severe for children in the IFN-γ-high group, especially in the body and object use, social and self-help, and expressive language performance domains. LEfSe analysis of gut microbiota revealed an overrepresentation of Selenomonadales, Negatiyicutes, Veillonellaceae and Verrucomicrobiaceae and underrepresentation of Bacteroides xylanisolvens and Bifidobacterium longum in children with higher IFN-γ level. Decreased metabolism function of carbohydrate, amino acid and lipid in gut microbiota were found in the IFN-γ-high group. Additional functional profiles analyses revealed significant differences in the abundances of genes encoding carbohydrate-active enzymes between the two groups. And enriched phenotypes related to infection and gastroenteritis and underrepresentation of one gut–brain module associated with histamine degradation were also found in the IFN-γ-High group. Results of multivariate analyses revealed relatively good separation between the two groups. ### Conclusions Levels of IFN-γ derived from γδT cell could serve as one of the potential candidate biomarkers to subtype ASD individuals to reduce the heterogeneity associated with ASD and produce subgroups which are more likely to share a more similar phenotype and etiology. A better understanding of the associations among immune function, gut microbiota composition and metabolism abnormalities in ASD would facilitate the development of individualized biomedical treatment for this complex neurodevelopmental disorder. ## Introduction Autism spectrum disorders (ASDs) are defined as a group of pervasive neurodevelopmental disorders characterized by persistent deficits in social communication and social interaction, plus restricted, repetitive patterns of behavior, interests, or activities [1]. The exact etiology of ASD is still unknown and accumulated results of recent studies suggest that instead of a single causative factor, ASD may be caused by the combined effects and interplay between genetic heritability and complex environmental risk factors (2–5). The heterogeneity in the symptomology and etiology of ASD has long been recognized by clinicians and researchers (6–12). A deep insight into this heterogeneity and using appropriate strategy to identify ASD subtypes are crucial, since different subtypes may result from different pathophysiology and respond differently to certain therapies (6, 9, 11–13). Previous subtyping methods mainly used behavioral characteristics and intellectual functioning as indicators (6, 10, 14–16). In recent years, subtyping ASD individuals according to their co-occurring medical disorders or associated physiological abnormalities has also emerged and got accumulated promising results (12, 17–20). It has been demonstrated that certain immune-mediated conditions (such as allergies and some autoimmune diseases) were more prevalent in ASD subjects (21–24). Immune dysfunction has been hypothesized to involve in the pathophysiology of a subtype of ASD [12, 20, 25]. Compared to behavioral symptoms, immune abnormalities are more objective since they can be measured using clinical and laboratory characteristic, thus would be a potential ideal subtyping indicator [25]. Gamma delta (γδ) T cells play important roles in inflammatory and autoimmune diseases [26]. They add to the imbalanced pro- and anti-inflammatory reactions and recruit other immune cells such as macrophages [27]. IFN-γ is one of the major cytokines produced by γδ T cells. Results from several previous studies revealed elevated IFN-γ levels in different tissues in some but not all ASD subjects (28–31), indicating that IFN-γ might participate in the immune dysfunction associated with certain subtype of ASD. The crosstalk between gut microbiota and host immune function has been increasingly recognized in recent years (32–35). Gut microbiota can interact with immune cells and modulate the function of immune system, and inflammation, which is caused by abnormal immune responses, can also influence the composition of the gut microbiome [32, 34]. Moreover, differences in the intestinal microbial community have also been found in children with ASD as compared to neurotypicals (17, 36–39). Since gut microbes can communicate with the host brain through multiple ways, including neuroactive compounds, toxin metabolites and immune modulation, it is suggested that alterations of gut-immune-brain axis play critical roles in ASD (38, 40–43). A better understanding of the association and comprehensive interactions among gut microbiota composition, immune characterization and behavioral symptoms in ASD, and subtyping this heterogenous disorder based on objective immunological characteristics into more homogeneous subgroups will not only provide useful information on the biological mechanisms underlying the pathogenesis of ASD, but also facilitate the development of individualized therapy strategies for ASD population [9, 18, 20]. In the present study, we recruited a cohort of 105 ASD children, and levels of IFN-γ derived from γδ T cells were measured to cluster children into subgroups. Comparison of autistic symptoms and gut microbiota composition was made between subgroups. Enriched KEGG orthologues markers and pathogen-host interactions based on metagenome were also analyzed to reveal the differences in functional features. Results of this study will provide additional evidence to support the association of gut microbiota alterations and immune dysfunction in ASD and suggest that IFN-γ could serve as a potential candidate biomarker to subtype ASD individuals into subgroups which tend to share a more similar phenotype and etiology. ## Participants The study was approved by the Institutional Review Board of Peking Union Medical College Hospital (IRB #ZS-824). Autistic children were consecutively recruited from the Herun Clinic in Beijing, China. The inclusion criteria for autistic children were [1]: Being diagnosed with autism which was confirmed by experienced psychiatrists according to the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM-V, 2013) criteria [2]. Free of antibiotic treatment, prebiotics and probiotics for at least 4 weeks before sample collection [3]. The children’s primary caregivers had good reading and comprehension skills and were able to fill in the relevant assessment scales [4]. The children’s parents or legal guardians volunteered to participate in this study and signed the informed consent. Autistic children with symptoms of other comorbid neurological or psychiatric disorders as confirmed by experienced clinicians or psychiatrists were excluded from the study. Detailed information on the purposes and procedures of the study were explained to the children’s parents or legal guardians. Written forms of full informed consent were obtained before involving the children in the study. ## Assessment of autistic symptoms The following scales were used to assess autistic symptoms in children: ## Detection of IFN-γ expression in γδ T cells Two milliliters of fasting venous blood samples were collected into chilled heparin tubes by trained nurses between 8:00 and 10:30 a.m. The samples were then diluted with equal volume of PBS, and the peripheral blood mononuclear cells (PBMCs) were separated by Ficoll (Tianjin Hao Yang Biological Technology Company) centrifugation. Isolated PBMCs were cultured in 24 well plates with complete medium (RPMI 1640 medium (Hyclone) with fetal bovine serum ($10\%$, Gibco), penicillin and streptomycin (100 u/ml)) and stimulated with 50ng/ml phorbol 12-myristate 13-acetate (PMA) (50ng/ml, Sigma) and 1µg/ml lonomycin (1ug/ml, Sigma) overnight. Brefeldin A (BFA) (Golgiplug, BD) was added in one hour after adding PMA and Ionomycin to block the secretion of the cytokines. The stimulated PBMCs were washed twice with PBS, centrifuged and then resuspended. Subsequently, CD3-PE conjugated antibody (BD Pharmingen), TCRγδ-FITC conjugated antibody (Biolegend) was added to the cells. After 30 minutes of incubation at 4°C avoiding light, the cells were washed twice with PBS. The cells were then permeabilized for staining of intercellular cytokine with Cytoperm/Cytofix Fixation/Permeabilization Kit (BD). Subsequently, cells were incubated with APC-conjugated IFN-γ antibody (BD Pharmingen) for an hour. Then the cells were washed and resuspended with PBS, followed by flow cytometry assessment. Flow cytometry was performed on BD Accuri C6 flow cytometer, and the following data analysis was conducted with CFlow Plus 1.0.164.15. ## Fecal sample collection and DNA extraction Children’s fresh fecal samples were obtained at home or Herun Clinic, immediately transferred into 1.5 ml sterile Eppendorf tubes (Axygen), and frozen into dry ice. All samples were stored at -80°C until analysis. DNA was extracted from fecal samples using the MO-BIO PowerSoil DNA Isolation Mini-Kit (Carlsbad) according to the protocol. DNA quality was assessed and controlled using gel electrophoresis. ## Metagenomic library construction and sequencing The sequencing library construction and template preparation was performed using the NEBNext UltraTM DNA Library Prep Kit (New England Biolabs) following manufacturer’s instructions (input DNA >100 ng). Each sample was barcoded and equal quantities of barcoded libraries were used for sequencing. The quality and quantity of the libraries were assessed using the Agilent 2100 High Sensitivity DNA Kit (Agilent Technologies) and the ABI 7500 Real Time PCR System (Applied Biosystems) before Illumina sequencing. Illumina HiSeq 2500 and Hiseq X Ten sequencing systems were used for paired-end 150bp sequencing. Data with adaptor contamination and low-quality reads were discarded from the raw data. We acquired ~223Gb high-quality data for the 38 samples with an average of ~5.9Gb per sample. ## Data analysis Taxonomic assignment of the main bacteria and the relative species abundances were calculated using MetaPhlAn (version 1.7.7) [47]. Biodiversity of the samples was processed with Vegan (version 2.4-6) in R package. The top 100 most abundance clades in each sample were selected to calculate the “Bray-Curtis” distance and the similarity between samples (Figure S3). The linear discriminant effect size (LEfSe) analysis was performed to find features (taxa) differentially represented between groups [48]. The Short Oligonucleotide Analysis Package(SOAP2, version 2.21) [49] was used to do the alignment and retain the unique mapped reads to do the downstream analysis. The relative abundance of these (super)contigs or genomes was calculated based on the number of aligned reads normalized by the (super)contig’s or genome’s size. The integrated non-redundant gene catalog database about the human gut microbiome was used to do the function analysis [50] with the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, the Carbohydrate-Active Enzymes database (CAZy), the Pathogen–Host Interactions database (PHI-base) and the Gut-Brain Modules (GBMs) as described in previously published articles (51–53). Other statistical analyses were performed using Statistical Package for the Social Science version 19.0 (SPSS Inc., Chicago, Illinois) and GraphPad Prism version 5.0 (GraphPad Software Inc., San Diego, CA). Continuous data were checked for normal distribution using the Shapiro-Wilk test first. Unpaired t test or non-parametric test (for those data that were not normally distributed) was used for comparison between groups. The Spearman or Pearson correlation test was applied to explore the correlation among autistic symptoms, gut microbiota, and functional categorization. The principal component analysis (PCA), orthogonal partial least-squares discriminate analysis (OPLS-DA) and the multivariate receiver operator curve (ROC) analysis were carried out using the methods as described in the protocol [54]. For all tests, a value of $p \leq 0.05$ (two-tailed) was considered statistically significant. False discovery rates (FDR) were controlled at 0.05 for multiple testing using the Benjamini and Hochberg method. ## Characteristics of the enrolled participants A total of 105 individuals met the inclusion criteria were recruited, and the top and bottom quarter of the participants were selected for questionnaires and fecal microbiota analyses based on their IFN-γ level derived from γδT cells. Since some of the participants were outpatients who can only spare a little time with our team, and some of the children may not defecate within these few days, not all of them have time to complete the behavioral symptoms assessment or have their fecal sample successfully collected. Those who completed all the questionnaires and fecal sample collection for metagenomic array constitute the final study samples in the present study. Demographics of the participants were summarized in Table 1. The two groups were well matched for chronological age and sex composition. The incidences of gastrointestinal symptoms such as diarrhea and constipation showed no statistical difference between groups. As expected, levels of IFN-γ derived from γδT cell were much higher in autistic children in the IFN-γ-High group as compared to the IFN-γ-Low group. Table 1 \| Unnamed: 0 \| IFN-γ-Low \| IFN-γ-High \| p \| \| --- \| --- \| --- \| --- \| \| n \| 17 \| 21 \| \| \| Age(y) [mean ± SEM(range)] \| 4.78 ± 0.36 (3.15-8.47) \| 4.74 ± 0.36 (3.02-8.54) \| 0.949 \| \| Gender \| Gender \| Gender \| Gender \| \| Male (%) \| 14 (82.35%) \| 19 (90.48%) \| 0.640 \| \| Female (%) \| 3 (17.65%) \| 2 (9.52%) \| 0.640 \| \| GI symptoms \| GI symptoms \| GI symptoms \| GI symptoms \| \| Diarrhea (%) \| 1 (5.88%) \| 3 (14.29%) \| 0.613 \| \| Constipation (%) \| 8 (47.06%) \| 6 (28.57%) \| 0.318 \| \| Diarrhea & Constipation# (%) \| 3 (17.65%) \| 5 (23.81%) \| 0.697 \| \| IFN-γ [mean ± SEM(range)] \| 0.80 ± 0.10 (0.09-1.32) \| 5.73 ± 0.48 (3.83-11.88) \| <0.001 \| ## Differences in the severity of autistic behavioral symptoms between groups Preliminary analysis of IFN-γ levels vs ASD severity indicated in recent clinical records (graded as mild, moderate or severe) suggests a rather skewed distribution (Figure S1). Significant differences in ABC metrics for severity of autistic behavioral symptoms were found between the IFN-γ-High and IFN-γ-Low groups. Children in the IFN-γ-High group had significantly higher ABC total scores (Median: 72.0, interquartile range (IQR): 59.5~84.0) than those in the IFN-γ-Low group (48.0, IQR 45.0~67.0, $p \leq 0.01$) (Figure 1A). Figure 1: Differences of ABC total and subscales scores between the IFN-γ-Low and IFN-γ-High groups. (A) ABC total scores; (B) Scores of ABC subscale III: Body and object use; (C) Scores of ABC subscale V: Social and self-help. The horizontal line and the box indicate the median and the interquartile range (IQR), and the whisker spans the minimum to maximum. p<0.05, ** p<0.01.* The post hoc analyses were conducted on the subscales scores (Table 2), and scores of two subscales in ABC (Body and object use, Social and self-help) demonstrated statistical differences between the two groups (Figures 1B, C), indicating that the related symptoms of those children in the IFN-γ-High group were much severe than those in the IFN-γ-Low groups. Table 2 \| Unnamed: 0 \| Group \| Group.1 \| Z \| p \| \| --- \| --- \| --- \| --- \| --- \| \| \| IFN-r-Low \| IFN-r- High \| Z \| p \| \| ABC subscales \| ABC subscales \| ABC subscales \| ABC subscales \| ABC subscales \| \| I. Sensory \| 12.0(7.0,18.0) \| 16.0(10.0,18.0) \| 1.263 \| 0.207 \| \| II. Relating \| 11.0(8.0,16.5) \| 16.0(11.0,19.0) \| 1.503 \| 0.133 \| \| III. Body and object use \| 8.0(4.0,10.0) \| 13.0(6.5,20.0) \| 2.061 \| 0.039* \| \| IV. Language \| 12.0(9.0,15.0) \| 14.0(10.0,19.0) \| 0.943 \| 0.346 \| \| V. Social and self-help \| 11.0(8.5,14.0) \| 14.0(11.0,17.0) \| 2.179 \| 0.029* \| \| ATEC subscales \| ATEC subscales \| ATEC subscales \| ATEC subscales \| ATEC subscales \| \| I. Speech/Language/Communication \| 12.0(6.0,15.5) \| 20.5(12.0,23.5) \| 2.261 \| 0.024* \| \| II. Sociability \| 22.0(18.5,30.0) \| 23.0(18.0,28.5) \| 0.28 \| 0.78 \| \| III. Sensory/Cognitive Awareness \| 19.0(13.5,25.0) \| 20.0(17.0,26.0) \| 0.883 \| 0.377 \| \| IV. Health/Physical/Behavior \| 28.0(19.5,30.5) \| 23.0(20.0,36.5) \| 0.353 \| 0.724 \| There was no statistical difference in ATEC total scores between groups (88.5, IQR 70.5~104.8 in IFN-γ-High group vs. 82.0, IQR 61.5~93.5 in IFN-γ-Low group, $p \leq 0.05$). However, the post hoc analyses of the subscales scores (Table 2) showed that scores of the Speech/Language/Communication subscale in ATEC demonstrated statistical difference (Figure 2A), indicating the language function was much more impaired in the IFN-γ-High group. Figure 2: Comparison of language function scores between the IFN-γ-Low and IFN-γ-High groups. (A) Scores of ATEC subscale I: Speech/Language/Communication; (B) Scores of CLSQ expressive language performance; (C) Scores of CLSQ receptive language performance. The horizontal line and the box indicate the median and the interquartile range (IQR), and the whisker spans the minimum to maximum. p<0.05.* In order to further evaluate the children’s expressive and receptive language performance respectively, the Clinical Language Status Questionnaire (CLSQ) was applied. As is shown in Figure 2B, children in the IFN-γ-High group got lower expressive language performance scores (5, IQR1~6) than those in the IFN-γ-Low group (7, IQR 5~8, $p \leq 0.05$). However, scores of receptive language performance showed no difference between the two groups (Figure 2C). ## Differences in the fecal microbiota composition between groups There was no significant difference in the alpha-diversity of the fecal microbiota between the two groups (Figure S2). The Bray–Curtis dissimilarity revealed no significant difference between the two groups (Figure S3, PERMANOVA, r 2 = 0.0276, $$p \leq 0.398$$). The LEfSe method was used to determine the taxa at different taxonomic levels which were enriched in the IFN-γ-High and IFN-γ-Low groups (Figure 3). Results of the LEfSe analysis revealed underrepresentation of Bacteroides xylanisolvens and Bifidobacterium longum in the IFN-γ-High group ($p \leq 0.01$, Wilcoxon rank-sum test; LDA>3.0). Overrepresentation of several phylotypes were also found in the IFN-γ-High group, and Selenomonadales, Negatiyicutes and Veillonellaceae were the top 3 enriched phylotypes with LDA<-4.0. Figure 3: Cladograms generated by LEfSe and LDA scores for bacterial taxa differentially abundant between groups. (A) Cladograms indicating differences in the bacterial taxa between the IFN-γ-Low and IFN-γ-High group. Green and red nodes indicate taxa that were enriched in the IFN-γ-Low group and the IFN-γ-High group, respectively. (B) LDA scores for differentially abundant bacterial taxa. Only taxa having a p<0.01 and LDA>2 are shown. Positive LDA scores indicate the taxa enriched in the IFN-γ-Low (IFN-γ-L) group (green), while negative LDA scores indicate the taxa enriched in the IFN-γ-High (IFN-γ-H) group (red), respectively. ## Differences in functional profiles from the metagenomic data between groups From the metagenomic data, the KEGG orthologues markers that were different between the IFN-γ-Low and IFN-γ-High groups were analyzed. The relative abundance of the KEGG orthologues markers related to amino acid metabolism, carbohydrate metabolism and lipid metabolism were found to be decreased in the IFN-γ-High group as compared to the IFN-γ-Low group with values of $p \leq 0.05$ (Figure 4). And the reported differences remained significant after applying the FDR. Figure 4: Differences of enriched KEGG orthologues markers between the IFN-γ-Low and IFN-γ-High groups. The values of the points represent the relative abundances of the KEGG orthologues markers related to (A) carbohydrate metabolism, (B) amino acid metabolism and (C) lipid metabolism. The horizontal line and the box indicate the median and the interquartile range (IQR), and the whisker spans the minimum to maximum. p<0.05.* In order to further explore the possible mechanisms underlying the differences relating carbohydrate metabolism between groups, the abundances of genes encoding carbohydrate-active enzymes (CAZymes) in the fecal microbiome were quantified. CAZymes were annotated by their family as defined in the CAZy database. Significant differences after applying the FDR in the abundances of genes encoding six CAZymes families were found between the two groups. For children in the IFN-γ-High group, the relative abundances of genes encoding GlycosylTransferase Family 56 (GT56), Polysaccharide Lyase Family 13 (PL13) and Polysaccharide Lyase Family 8 (PL8) was lower, while the relative abundances of genes encoding Carbohydrate Esterase Family 10 (CE10), Glycoside Hydrolase Family 95 (GH95) and GlycosylTransferase Family 28 (GT28) was higher as compared to those in the IFN-γ-Low group (Figure 5). Figure 5: Differences of the abundances of genes encoding carbohydrate-active enzymes (CAZymes) in the fecal microbiome between the IFN-γ-Low and IFN-γ-High groups. The values of the points represent the relative abundances of genes encoding (A) GlycosylTransferase Family 56 (GT56), (B) Polysaccharide Lyase Family 13 (PL13) and (C) Polysaccharide Lyase Family 8 (PL8), (D) Carbohydrate Esterase Family 10 (CE10), (E) Glycoside Hydrolase Family 95 (GH95) and (F) GlycosylTransferase Family 28 (GT28). The horizontal line and the box indicate the median and the interquartile range (IQR), and the whisker spans the minimum to maximum. p<0.05, ** p<0.01.* The PHI-base phenotypes related to infection (Pathogen gene: purT, Host species: Homo sapiens) and gastroenteritis (Pathogen gene: flhF, Host species: Homo sapiens) were found to be significantly enriched in the IFN-γ-High group (Figures 6A, B). Additionally, underrepresentation of one gut–brain module (MGB010) associated with histamine degradation was also found in the IFN-γ-High group (Figure 6C). Figure 6: Differences of the abundances of genes related to pathogen-host interactions and gut–brain modules in the fecal microbiome between the IFN-γ-Low and IFN-γ-High groups. The values of the points represent the relative abundances of genes related to (A) infection (Pathogen gene: purT, Host species: Homo sapiens), (B) gastroenteritis (Pathogen gene: flhF, Host species: Homo sapiens) and (C) gut–brain module MGB010 associated with histamine degradation. The horizontal line and the box indicate the median and the interquartile range (IQR), and the whisker spans the minimum to maximum. p<0.05, ** p<0.01.* ## The correlation analysis The Spearman correlation analysis was applied to explore the relationship among IFN-γ, autistic behavioral symptoms, gut microbiota and functional modules which were significant in univariate analysis. As is shown in the matrix in Figure 7, the relative abundance of Bacteroides xylanisolvens, which was enriched in the IFN-γ-Low group, was negatively correlated with IFN-γ level (rho=-0.434, $p \leq 0.05$), ABC total and subscales (Body and object use, Social and self-help) scores (all $p \leq 0.05$), and the relative abundance of GlycosylTransferase Family 28 (GT28) (rho=-0.535, $p \leq 0.05$), and positively correlated with CLSQ expressive language performance scores (rho=0.353, $p \leq 0.01$). Additionally, the relative abundance of GlycosylTransferase Family 28 (GT28) was negatively correlated with ABC language scores (rho=-0.326, $p \leq 0.05$), while the relative abundance of Polysaccharide Lyase Family 8 (PL8) was positively correlated with ABC language score (rho=0.303, $p \leq 0.05$). Among these above correlations, only the correlation between the relative abundance of Bacteroides xylanisolvens and CLSQ expressive language performance scores remained significant after applying the FDR. Figure 7: The Spearman correlation matrix among IFN-γ, autistic behavioral symptoms, gut microbiota and functional modules which were significant in autistic children. Color intensity reflects Spearman correlation coefficient. p<0.05, p<0.01, * p<0.001.* ## Multivariate analysis and potential discriminating features analysis Since univariate approaches ignore the correlations among variables as demonstrated in Figure 7, multivariate analyses were applied because these methods simultaneously take all variables into consideration. The principal component analysis (PCA) scores plot revealed that samples in the IFN-γ-Low group were more concentrated as compared to the scattered pattern of the IFN-γ-High group (Figure 8A). And as is shown in Figure 8B, the scores plot constructed using orthogonal partial least-squares discriminate analysis (OPLS-DA) revealed relatively good separation between the IFN-γ-Low and IFN-γ-High groups (Q2 = 0.469, $p \leq 0.05$; R2Y=0.726, $p \leq 0.05$). Figure 8: Scores plot of multivariate analysis and potential discriminating features analysis using random forests algorithm. (A) Principal component analysis (PCA) scores plot and (B) Orthogonal partial least-squares discriminate analysis (OPLS-DA) scores plot of ASD children in the IFN-γ-Low (green) and IFN-γ-High (red) groups. Each point represents the score of a single individual. The shaded areas indicate the 95% confidence ellipse regions for each group. (C) ROC curves from different multivariate models using different number of features. (D) The top 10 significant discriminating features ranked based on their frequencies of being selected during cross validation. The algorithm of the random forests was used to perform potential discriminating features analysis. The Monte-Carlo cross validation (MCCV) was applied to identify models with good performance. In each MCCV, two thirds ($\frac{2}{3}$) of the samples are used to evaluate the feature importance. The top 2, 3, 5, 10… important features are then used to build classification models which is validated on the $\frac{1}{3}$ the samples that were left out. The procedure were repeated multiple times to calculate the performance and confidence interval (CI) of each model. Based on the cross validation, the multivariate models using 10 variables achieved an AUC of 0.835 (Figure 8C). The top 10 significant discriminating features ranked based on their frequencies of being selected during cross validation are listed in Figure 8D. ## Discussion In the present study, a cohort of 105 ASD children were recruited and ranked based on their IFN-γ levels derived from γδT cells. The top $25\%$ and bottom $25\%$ of the participants were selected which constituted the final two groups, respectively. Our results demonstrated that autistic behavioral symptoms of children in the IFN-γ-high group were more severe, especially in the body and object use, social and self-help, and expressive language performance domains. The LEfSe analysis of gut microbiota revealed some bacterial taxa differentially abundant between groups. Decreased metabolism function of carbohydrate, amino acid and lipid in gut microbiota were found in the IFN-γ-high group. Additional functional profiles analyses also revealed significant differences in the abundances of genes encoding carbohydrate-active enzymes between groups. And enriched phenotypes related to infection and gastroenteritis and underrepresentation of one gut–brain module associated with histamine degradation were also found in the IFN-γ-High group. Results of multivariate analyses revealed relatively good separation between the two groups and suggest that IFN-γ could serve as a potential candidate biomarker to subtype ASD individuals into more homogeneous subtypes. Currently, the diagnosis of ASD is still made mainly based on behavioral symptoms [1]. And the concept of spectrum suggests that individuals with ASD may present with diverse sets of symptoms that vary widely from one individual to another [1, 55]. The symptom diversity may be caused by many different factors, and this heterogeneity brings about great difficulty for researchers to elucidate the anticipated etiology or risk factors for ASD, because it would not be expected that a same etiological factor would explain two vastly different phenotypes [56, 57]. It has now been well recognized that researchers should subtype these individuals within the spectrum to reduce the diversity and use a more homogeneous subtype to study the biological mechanism and explore effective treatment strategies [58]. Previous subtyping strategies are mostly defined by some particular symptom characteristics, such as social behavior or language ability [6, 7, 9, 10]. Another feasible approach is using biomedical features to stratify samples to reduce heterogeneity and produce subgroups which are more likely to share a more similar phenotype and etiology [11, 12, 59]. Compared to the behavioral symptom characteristics, using biomedical features as subtyping indicators has some advantages, because they are more objective and easily to measure, more directly to indicate the possible mechanisms underlying the associated heterogeneity, and could also provide useful information to further explore the potential targets to facilitate the development of individualized biomedical therapy strategies for certain ASD subtypes. Immunological involvement in the pathophysiology of certain subtypes of ASD has long been hypothesized and accumulated results from both clinical and animal research have identified the associations between immunologic function abnormalities and ASD (12, 21–24, 60–62). Moreover, clinical trials using immune-modulating or anti-inflammatory drugs in individuals with ASD also yield promising results, and the treatment responses were especially better for those with immunological or gastrointestinal disturbances (12, 63–66). Results of these previous studies suggest that biological characteristics relating to immune function may serve as potential biomarkers to reduce the heterogeneity in ASD and to improve the prediction of response to certain biomedical treatments [12]. In the present study, we choose IFN-γ derived from γδT cells as subtyping biomarkers, because γδT cell intrinsically combines innate immunity and adaptive immunity and plays important roles in inflammatory and autoimmune diseases, which were found to be more prevalent in ASD individuals (21–24, 67). And results of previous studies also indicated IFN-γ might play a role in the progression and exacerbation of autistic symptoms (28–31). Changes of INF-γ levels have been found in blood samples and brain tissues of ASD subjects, and animal studies also confirmed upregulation of INF-γ in animals with autistic-like behaviors [31, 68, 69]. It has been demonstrated that plasma levels of INF-γ correlated positively with plasma nitric oxide measures in ASD group and the higher NO production in ASD children may be secondary to IFN-γ mediated up-regulation of the inducible nitric oxide synthase (iNOS) [70]. INF-γ may interact with gut microbiota and PBMCs taken from ASD subjects produced elevated levels of IFN-γ against common dietary proteins [71]. High levels of INF-γ were also associated with a reduction in glucocorticoid receptor [72], which might result in excessive circulation of glucocorticoid, and the excessive glucocorticoid are well-known as neurotoxins [73]. Although the direct solid evidence is still lacking, these findings support the hypothesis that INF-γ may play a role in the pathologic mechanism of ASD. However, results of the INF-γ levels in ASD from the previous studies were not always consistent. Both higher and lower levels of INF-γ have been found in blood samples and PBMCs of ASD [see the summarized results in the excellent systematic reviews [74, 75]]. In our clinical practice, we also find that a great heterogenicity exists in the INF-γ levels in ASD. Levels of INF-γ are very high in a portion of ASD children, and their behavioral symptoms seems to be different from other ASD children. So, we hypothesized that within the heterogeneous broad spectrum of ASD, those ASD children with high INF-γ levels may represent a subgroup whose autistic symptoms and gut microbiota composition may be different from others. And the results of the present study turned out to support our initial hypothesis. When comparing the autistic behavioral symptoms between the two subgroups selected based on levels of IFN-γ derived from γδT cells, children with higher levels of IFN-γ got significantly higher scores in ABC, especially for the body and object use subscale and the social and self-help subscale. Additionally, children in this group also got higher scores in the speech/language/communication subscale in ATEC. Since there were some discrepancies as assessed by ABC and ATEC questionnaires in the language domain, and the expressive and receptive language abilities were evaluated with different weights but calculated as a whole in these two questionnaires [44, 45], the CLSQ was used to further assess children’s expressive and receptive language performance respectively [46]. The results revealed that only the expressive language performance was significantly impaired in the IFN-γ-High group. All these results suggest that autistic behavioral symptoms were different between the IFN-γ-High and IFN-γ-Low groups, and children with higher levels of IFN-γ may suffer from more severe symptoms of ASD. Since there exist intense interactions between gut microbiota and immune function, and alterations of gut-immune-brain axis has been suggested to act critical roles in the pathogenesis of ASD (32–35, 41, 42), differences in gut microbiota composition between the two groups were also analyzed. The most significant characteristic difference between the two subgroups is that Negativicutes, Selenomonadales and Veillonellaceae were more enriched in the IFN-γ-High group, with the LDA score less than -4. Different abundances of these three bacterial taxa were also found between autistic and neurotypical subjects in several other independent studies [23, 76, 77]. Indeed, the family Veillonellaceae belongs to the order Selenomonadales within the class Negativicutes, and they are all members of the phylum Firmicutes [78]. Species of Firmicutes could upregulate IFN-γ production and significant increased ratio of Firmicutes/Bacteroidetes has been reported associated with not only with ASD [77, 79], but also with other conditions that were found to be more prevalent within ASD subjects, such as obesity and diabetes [24, 80, 81]. The relative abundances of the species of Akkermansia muciniphila, Pyramidobacter piscolens, and Anaerotruncus colihominis were also found to be more enriched in the faces of ASD children in the IFN-γ-High group. Akkermansia muciniphila is a mucin-degrading bacterium, which has been suggested to play a role in inflammation and gut permeability [82, 83]. Lower relative abundances of Akkermansia muciniphila has been found in feces of autistic children, which might reflect an indirect evidence of a thinner gastrointestinal mucus barrier in ASD children [83]. Interestingly, there are also studies that found Akkermansia was present at higher relative abundances in feces of ASD individuals [76, 84], or even at very high levels (up to $59\%$) in several autistic individuals [23]. Results from these previous studies suggest that great diversity in the abundances of Akkermansia muciniphila may exist among different ASD individuals. In the present study, we found that within the spectrum of autism, there do exist significant differences in the relative abundances of Akkermansia muciniphila between ASD children in the IFN-γ-High and the IFN-γ-Low groups. Pyramidobacter piscolens is one of the members of the phylum Synergistetes. It was first isolated from human oral cavity [85] and is related to oral dysbiosis, which may result in periodontal diseases and abscess [86]. Oral dysbiosis and these oral health conditions are also found to be more common in ASD children [87]. Further studies revealed that Pyramidobacter piscolens could also be cultured from small intestine abscess, and it is now considered that Pyramidobacter piscolens is part of the commensal human microflora which plays a functional role but may also act as opportunistic pathogens [88]. Additionally, Pyramidobacter piscolens is one of the core species which can regulate lipid deposition [89] and may influence blood glucose metabolism [90]. Anaerotruncus colihominis belongs to phylum Firmicutes. It is a short-chain fatty acids (SCFA) producing species which is presumed to be anti-inflammatory and is related to autoimmunity (91–94). The abundance of Anaerotruncus colihominis was found to be negatively associated with cognitive function scores in patients with Alzheimer’s disease [95]. Significant lower abundances of Anaerotruncus colihominis has also been found in patients with rheumatoid arthritis (RA) [94], and a number of clinical and basic studies have demonstrated roles of IFN-γ in the pathogenesis of RA [96]. It has also been reported that Anaerotruncus colihominis possesses the ability to produce acetic and butyric acids [91], which could have a role in regulating gut epithelial barrier function and play possible roles in ASD [97, 98]. Although both of the Pyramidobacter piscolens and Anaerotruncus colihominis play functional roles in the metabolism of bioactive compounds which is perturbed in ASD, studies of the direct roles of the two species in the pathophysiology of ASD is rare. Our results demonstrated that there were significant differences in the relative abundances of Pyramidobacter piscolens and Anaerotruncus colihominis between ASD children in the IFN-γ-High and the IFN-γ-Low groups, and the biological significance of these findings warrants further research. The results of the LEfSe analysis also revealed underrepresentation of Bacteroides xylanisolvens and Bifidobacterium longum in the IFN-γ-High group. These two bacterial species are both non-pathogenic and process many probiotic qualities (99–101). Bacteroides xylanisolvens belongs to the second most abundant genus Bacteroides in the human intestine and they can break down many sugars including dietary fiber and xylan [102, 103]. It has been demonstrated that some strains of Bacteroides could modulate the function of innate immune system [104] and have the potential to relieve some behavioral and physiological abnormalities associated with ASD [105, 106]. Bifidobacterium longum is considered to be one of the earliest colonizers of the gastrointestinal tract in infants [107]. The domination of Bifidobacterium in infant’s gastrointestinal tract could hinder pathogenic organisms’ colonization through antimicrobial activity and competitive exclusion manners [108]. Bifidobacterium longum could also serve as a scavenger because it metabolizes a large variety of substrates including bile salts, human milk oligosaccharide and some other complex oligosaccharides [107, 109, 110]. The efficacy of Bifidobacterium longum in regulating immune (including its ability to suppress the expression of IFN-γ in vivo) and central nervous system functions and alleviating psychiatric disorder-related behaviors including ASD and obsessive-compulsive disorder has also been demonstrated [100, 101, 111]. It is worth mentioning that Bacteroides and Bifidobacterium species were also found to be depleted in ASD children in other independent cohort studies [76, 83, 112, 113]. Associations between gut microbiota and ASD certainly warrant further studies to elucidate a causation role in the pathogenesis of ASD. However, the consistency of these results across different ethnic groups using different sequencing and assay methods, together with their efficacy in alleviating autistic symptoms, strongly suggest that the loss of representation of these bacterial taxa is very robust and may be tightly associated with the pathophysiology of ASD. For the predicted KEGG pathway analysis results, we found that the IFN-γ-High group was less enriched in pathways related to amino acid metabolism, carbohydrate metabolism and lipid metabolism. As key partners involved in the maintenance of human physiology and health, gut microbes influence greatly on host metabolism and help balance important vital functions such as food digestion and nutrient bioavailability for the host [114]. The relatively depleted pathway orthologues markers related to metabolism of amino acid, lipid and carbohydrate in the IFN-γ-High group suggested that children in this subgroup may have higher risks of suffering from more sever metabolic dysfunction. Indeed, a great quantity of work has shown that children with ASD have perturbed metabolism as compared to neurotypical children (112, 115–123). For the amino acid metabolism, altered amino acid profile has been found in blood plasma [116, 117], urine [118, 119] and fecal [112] samples collected from ASD individuals. And it was postulated that gut microbial metabolism of phenylalanine and tyrosine may be involved in the pathogenesis of autism [120]. Impaired carbohydrate digestion [121] and lipid metabolism [122] were also found in ASD individuals. The abundance of affected bacterial phylotypes in the intestines or duodenum of ASD individuals was found to be associated with expression levels of disaccharidases and transporters, which is important for carbohydrate digestion and transport [121, 123]. Since Bacteroides spp. and Bifidobacterium spp. are specialized as primary and secondary degraders in the metabolism of complex carbohydrates [124], the depleted species of Bacteroides and Bifidobacterium in the IFN-γ-High group may impact the carbohydrate metabolism capability. Additionally, as is demonstrated in this study, the abundances of genes encoding six families of carbohydrate-active enzymes in the fecal microbiome were significantly different between the IFN-γ-High and IFN-γ-Low groups, this may partly explain the possible mechanisms underlying the differences relating carbohydrate metabolism between the two groups. Furthermore, some bacterial species possess the ability to ferment dietary carbohydrates into the production of short chain fatty acids (SCFAs) [125]. SCFAs can readily cross the gut–blood and blood–brain barriers and induce widespread effects on gut and brain via impact on epithelial barrier integrity, neurotransmitter synthesis and immune modulation (126–128). Since some of the metabolites such as Omega-6 (n-6) and Omega-3 (n-3) polyunsaturated fatty acids (PUFA) are essential nutrients for brain development and function, these metabolic alterations may be associated with the severity of autistic symptom [122, 129, 130]. All these results further support the notion that ASD is a pervasive developmental disorder with multisystem dysfunction and metabolic disturbance. Functional profiles analyses from the metagenomic data in our study also revealed that the abundances of genes related to infection (Pathogen gene: purT, Host species: Homo sapiens) and gastroenteritis (Pathogen gene: flhF, Host species: Homo sapiens) were significantly enriched in the IFN-γ-High group. Gastrointestinal disorders are one of the most common medical conditions that are comorbid with ASD, and these comorbidities can cause greater severity in autistic symptoms [131]. The results from our study further suggest that children in the IFN-γ-High group may suffer from higher incidence or severity of infection and gastroenteritis, but these results need to be further validated with medical examination. Another significant difference is the underrepresentation of the gut–brain module (MGB010) associated with histamine degradation in the IFN-γ-High group. Altered expression of histamine signaling genes has been found in ASD populations [132], and antagonism of histamine receptors could reduce autistic behavioral symptoms in ASD individuals and several relevant animal models (132–135). Moreover, histamine receptor antagonists can suppress IFN-γ production [136], while IFN-γ can also modulate histamine-induced IL-6 and IL-8 production [137]. Our data suggests the histamine degradation capability in fecal microbiota were much more impaired in children with higher levels of IFN-γ, and the decreased capability of histamine degradation may partly affect the autistic behavioral symptoms in ASD children. For the correlation analysis, the relative abundance of Bacteroides xylanisolvens showed most significant relationships with not only several autistic behavioral characteristics, but also with one of the carbohydrate-active enzymes families (GT28). Additionally, Bacteroides xylanisolvens was the top discriminating features in the multivariate models using the random forests algorithm, suggesting its importance in the separation of the two groups. Our results of the multivariate analysis indicate that although both of the two groups are within the spectrum, they can be separated using the IFN-γ as indicator to obtain subtypes with more similar features. Based on the cross validation, the ROC curves built using 10 variables achieved an AUC of 0.835. In this study, the ROC curves were generated by MCCV using balanced sub-sampling. In each MCCV, two thirds ($\frac{2}{3}$) of the samples were used to evaluate the feature importance. The top 2, 3, 5, 10… important features were then used to build classification models which were validated on the $\frac{1}{3}$ samples that were left out [54]. Since more variables consistently leads to better prediction, and due to the relatively small sample size in this study, there exists a risk of overfitting. Therefore, it is important to evaluate the models with a large number of samples to estimate their generalizability with high confidence. Since INF-γ level varies widely within the heterogeneous broad spectrum of ASD, and as is shown in our study, the behavioral symptoms, gut microbiota composition and some metabolic features of ASD children in the INF-γ-High group were different from those in the INF-γ-Low group, utilization of this information to segregate ASD children into different subgroups will greatly facilitate the pathophysiology study of a more homogeneous clusters of ASD in the future. Additionally, although there still lack of solid evidence, several anti-inflammatory compounds (such as Palmitoylethanolamide, celecoxib, flavonoid luteolin) have been studied to investigate their effect as an adjunctive therapy in improving behavioral symptoms in autistic individuals [64, 138, 139]. Since INF-γ is an important pro-inflammatory cytokine involved in ASD but varies widely within the heterogeneous spectrum, we believe that using these anti-inflammatory drugs in ASD subgroup with high INF-γ levels will yield more promising and consistent results. As a preliminary study, there are several limitations ought to be mentioned. Firstly, only children with ASD were enrolled in this study, lacking typically developing children as controls, and the comparison of these obtained results with a control group can be informative. Secondly, the sample size in this study is relatively small, which may decrease the statistical power, and there exists a risk of overfitting for the MCCV model. Results of this study need to be validated in an independent larger cohort. Thirdly, additional risk factors (such as having a close relative with ASD, very low birth weight, and complications at delivery) were not collected from the participants in this study. The results should be interpreted with caution due to the observational nature of the present study. Also, consistent with the sex ratio of ASD, participants were mostly males, which limited the analyses of sex differences. Finally, only IFN-γ was measured in this study without testing other cytokines such as interleukin and TNF, which limits the ability to explore the full picture of immunological profiles and characteristics in ASD children. Results of this study demonstrate only associations but not causations. Further studies are warranted to reveal the cause–effect relationships among IFN-γ levels, gut microbiota composition and autistic behavioral symptoms. Despite of these limitations and the preliminary nature of this study, our results suggest that levels of IFN-γ derived from γδT cell could serve as one of the potential candidate biomarkers to subtype ASD individuals to reduce heterogeneity and produce subgroups which are more likely to share a more similar phenotype and etiology. Our results also further support the notion that there exits comprehensive and complex interaction among gut microbiota, immune function and autistic phenotypes. And a better understanding of the associations between immune function and gut microbiota composition as well as metabolism abnormalities in ASD would provide us deep insights into the pathogenesis of ASD and give us important clues to facilitate the development of systemic biomedical treatment for this complex neurodevelopmental disorder. ## Data availability statement The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/, PRJNA530620. ## Ethics statement The studies involving human participants were reviewed and approved by the Institutional Review Board of Peking Union Medical College Hospital. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin. ## Author contributions X-JX, XY, X-FZ, and GT, conceptualization. X-JX, J-DL, and JY, methodology. J-DL and BL, validation. X-JX, J-DL, JY, and X-DL, formal analysis. X-JX, JY, and XY, investigation. X-JX and JY, writing—original draft preparation. X-JX, J-DL, and XY, writing—review and editing. X-JX and J-DL, visualization. XY and X-FZ, supervision. X-JX and XY, project administration and funding acquisition. All authors contributed to the article and approved the submitted version. ## Conflict of interest Authors J-DL and GT were employed by the company Geneis Beijing Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1100816/full#supplementary-material ## References 1. 1 American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 5th ed. Arlington, VA: American Psychiatric Association (2013). doi: 10.1176/appi.books.9780890425596.744053. Diagnostic and statistical manual of mental disorders (2013). DOI: 10.1176/appi.books.9780890425596.744053 2. Hallmayer J, Cleveland S, Torres A, Phillips J, Cohen B, Torigoe T. Genetic heritability and shared environmental factors among twin pairs with autism. Arch Gen Psychiatry (2011) 68. DOI: 10.1001/archgenpsychiatry.2011.76 3. Sandin S, Lichtenstein P, Kuja-Halkola R, Larsson H, Hultman CM, Reichenberg A. The familial risk of autism. JAMA (2014) 311. DOI: 10.1001/jama.2014.4144 4. Xu XJ, Shou XJ, Li J, Jia MX, Zhang JS, Guo Y. Mothers of autistic children: lower plasma levels of oxytocin and arg-vasopressin and a higher level of testosterone. PloS One (2013) 8. DOI: 10.1371/journal.pone.0074849 5. Xu XJ, Zhang HF, Shou XJ, Li J, Jing WL, Zhou Y. Prenatal hyperandrogenic environment induced autistic-like behavior in rat offspring. Physiol Behav (2015) 138 13-20. DOI: 10.1016/j.physbeh.2014.09.014 6. Wing L, Gould J. Severe impairments of social interaction and associated abnormalities in children: Epidemiology and classification. J Autism Dev Disord (1979) 9 11-29. DOI: 10.1007/BF01531288 7. O’Brien SK. The validity and reliability of the wing subgroups questionnaire. J Autism Dev Disord (1996) 26. DOI: 10.1007/BF02172477 8. Rubeis SD, He X, Goldberg AP, Poultney CS, Samocha K, Cicek AE. Synaptic, transcriptional and chromatin genes disrupted in autism. Nature (2014) 515. DOI: 10.1038/nature13772 9. Meng FC, Xu XJ, Song TJ, Shou XJ, Wang XL, Han SP. Development of an autism subtyping questionnaire based on social behaviors. Neurosci Bull (2018) 34 789-800. DOI: 10.1007/s12264-018-0224-8 10. Tager-Flusberg H. Defining language impairments in a subgroup of children with autism spectrum disorder. Sci China Life Sci (2015) 58. DOI: 10.1007/s11427-012-4297-8 11. Bernier R, Golzio C, Xiong B, Stessman HA, Coe BP, Penn O. Disruptive CHD8 mutations define a subtype of autism early in development. Cell (2014) 158. DOI: 10.1016/j.cell.2014.06.017 12. McDougle CJ, Landino SM, Vahabzadeh A, O’Rourke J, Zurcher NR, Finger BC. Toward an immune-mediated subtype of autism spectrum disorder. Brain Res (2015) 1617 72-92. DOI: 10.1016/j.brainres.2014.09.048 13. Zhang R, Jia MX, Zhang JS, Xu XJ, Shou XJ, Zhang XT. Transcutaneous electrical acupoint stimulation in children with autism and its impact on plasma levels of arginine-vasopressin and oxytocin: A prospective single-blinded controlled study. Res Dev Disabil (2012) 33. DOI: 10.1016/j.ridd.2012.02.001 14. Rapin I, Dunn MA, Allen DA, Stevens MC, Fein D. Subtypes of language disorders in school-age children with autism. Dev Neuropsychol (2009) 34 66-84. DOI: 10.1080/87565640802564648 15. Beglinger LJ, Smith TH. A review of subtyping in autism and proposed dimensional classification model. J Autism Dev Disord (2001) 31. DOI: 10.1023/a:1010616719877 16. Volkmar FR, Cohen DJ, Bregman JD, Hooks MY, Stevenson JM. An examination of social typologies in autism. J Am Acad Child Adolesc Psychiatry (1989) 28. DOI: 10.1097/00004583-198901000-00015 17. Wang M, Zhou J, He F, Cai C, Wang H, Wang Y. Alteration of gut microbiota-associated epitopes in children with autism spectrum disorders. Brain Behav Immun (2019) 75. DOI: 10.1016/j.bbi.2018.10.006 18. Tye C, Runicles AK, Whitehouse AJO, Alvares GA. Characterizing the interplay between autism spectrum disorder and comorbid medical conditions: An integrative review. Front Psychiatry (2019) 9. DOI: 10.3389/fpsyt.2018.00751 19. Yasuhara A. Correlation between EEG abnormalities and symptoms of autism spectrum disorder (ASD). Brain Dev (2010) 32. DOI: 10.1016/j.braindev.2010.08.010 20. McDougle CJ. Another step toward defining an immune-mediated subtype of autism spectrum disorder. JAMA Netw Open (2018) 1. DOI: 10.1001/jamanetworkopen.2018.0280 21. Zerbo O, Leong A, Barcellos L, Bernal P, Fireman B, Croen LA. Immune mediated conditions in autism spectrum disorders. Brain Behav Immun (2015) 46. DOI: 10.1016/j.bbi.2015.02.001 22. Mazefsky CA, Schreiber DR, Olino TM, Minshew NJ. The association between emotional and behavioral problems and gastrointestinal symptoms among children with high-functioning autism. Autism (2014) 18 493-501. DOI: 10.1177/1362361313485164 23. Kang DW, Park JG, Ilhan ZE, Wallstrom G, LaBaer J, Adams JB. Reduced incidence of prevotella and other fermenters in intestinal microflora of autistic children. PloS One (2013) 8. DOI: 10.1371/journal.pone.0068322 24. Croen LA, Zerbo O, Qian Y, Massolo ML, Rich S, Sidney S. The health status of adults on the autism spectrum. Autism (2015) 19. DOI: 10.1177/1362361315577517 25. Jyonouchi H, Geng L, Davidow AL. Cytokine profiles by peripheral blood monocytes are associated with changes in behavioral symptoms following immune insults in a subset of ASD subjects: An inflammatory subtype. J Neuroinflamm (2014) 11. DOI: 10.1186/s12974-014-0187-2 26. Gober HJ, Kistowska M, Angman L, Jenö P, Mori L, Libero GD. Human T cell receptor γδ cells recognize endogenous mevalonate metabolites in tumor cells. J Exp Med (2003) 197. DOI: 10.1084/jem.20021500 27. Hu C, Qian L, Miao Y, Huang Q, Miao P, Wang P. Antigen-presenting effects of effector memory Vγ9VδT cells in rheumatoid arthritis. Cell Mol Immunol (2012) 9. DOI: 10.1038/cmi.2011.50 28. Ahmad SF, Nadeem A, Ansari MA, Bakheet SA, Al-Ayadhi LY, Attia SM. Upregulation of IL-9 and JAK-STAT signaling pathway in children with autism. Prog Neuropsychopharmacol Biol Psychiatry (2017) 79. DOI: 10.1016/j.pnpbp.2017.08.002 29. Li X, Chauhan A, Sheikh AM, Patil S, Chauhan V, Li XM. Elevated immune response in the brain of autistic patients. J Neuroimmunol (2009) 207. DOI: 10.1016/j.jneuroim.2008.12.002 30. Singh VK. Plasma increase of interleukin-12 and interferon-gamma. pathological significance in autism. J Neuroimmunol (1996) 66. DOI: 10.1016/0165-5728(96)00014-8 31. Xu N, Li X, Zhong Y. Inflammatory cytokines: Potential biomarkers of immunologic dysfunction in autism spectrum disorders. Mediators Inflammation (2015) 2015. DOI: 10.1155/2015/531518 32. D’Amelio P, Sassi F. Gut microbiota, immune system, and bone. Calcif Tissue Int (2018) 102. DOI: 10.1007/s00223-017-0331-y 33. Burcelin R. Gut microbiota and immune crosstalk in metabolic disease. Mol Metab (2016) 5. DOI: 10.1016/j.molmet.2016.05.016 34. Shi N, Li N, Duan X, Niu H. Interaction between the gut microbiome and mucosal immune system. Mil Med Res (2017) 4 14. DOI: 10.1186/s40779-017-0122-9 35. Pickard JM, Zeng MY, Caruso R, Núñez G. Gut microbiota: Role in pathogen colonization, immune responses, and inflammatory disease. Immunol Rev (2017) 279 70-89. DOI: 10.1111/imr.12567 36. Rose DR, Yang H, Serena G, Sturgeon C, Ma B, Careaga M. Differential immune responses and microbiota profiles in children with autism spectrum disorders and co-morbid gastrointestinal symptoms. Brain Behav Immun (2018) 70. DOI: 10.1016/j.bbi.2018.03.025 37. Vuong HE, Hsiao EY. Emerging roles for the gut microbiome in autism spectrum disorder. Biol Psychiatry (2017) 81. DOI: 10.1016/j.biopsych.2016.08.024 38. Li Q, Han Y, Dy ABC, Hagerman RJ. The gut microbiota and autism spectrum disorders. Front Cell Neurosci (2017) 11. DOI: 10.3389/fncel.2017.00120 39. Coretti L, Paparo L, Riccio MP, Amato F, Cuomo M, Natale A. Gut microbiota features in young children with autism spectrum disorders. Front Microbiol (2018) 9. DOI: 10.3389/fmicb.2018.03146 40. Luna RA, Savidge TC, Williams KC. The brain-gut-microbiome axis: What role does it play in autism spectrum disorder. Curr Dev Disord Rep (2016) 3 75-81. DOI: 10.1007/s40474-016-0077-7 41. Kelly JR, Minuto C, Cryan JF, Clarke G, Dinan TG. Cross talk: The microbiota and neurodevelopmental disorders. Front Neurosci (2017) 11. DOI: 10.3389/fnins.2017.00490 42. Ghaisas S, Maher J, Kanthasamy A. Gut microbiome in health and disease: Linking the microbiome-gut-brain axis and environmental factors in the pathogenesis of systemic and neurodegenerative diseases. Pharmacol Ther (2016) 158 52-62. DOI: 10.1016/j.pharmthera.2015.11.012 43. Dinan TG, Cryan JF. The microbiome-gut-brain axis in health and disease. Gastroenterol Clin North Am (2017) 46 77-89. DOI: 10.1016/j.gtc.2016.09.007 44. Rellini E, Tortolani D, Trillo S, Carbone S, Montecchi F. Childhood autism rating scale (CARS) and autism behavior checklist (ABC) correspondence and conflicts with DSM-IV criteria in diagnosis of autism. J Autism Dev Disord (2004) 34. DOI: 10.1007/s10803-004-5290-2 45. Geier DA, Kern JK, Geier MR. A comparison of the autism treatment evaluation checklist (ATEC) and the childhood autism rating scale (CARS) for the quantitative evaluation of autism. J Ment Health Res Intellect Disabil (2013) 6. DOI: 10.1080/19315864.2012.681340 46. Duffy FH, Shankardass A, McAnulty GB, Eksioglu YZ, Coulter D, Rotenberg A. Corticosteroid therapy in regressive autism: A retrospective study of effects on the frequency modulated auditory evoked response (FMAER), language, and behavior. BMC Neurol (2014) 14. DOI: 10.1186/1471-2377-14-70 47. Segata N, Waldron L, Ballarini A, Narasimhan V, Jousson O, Huttenhower C. Metagenomic microbial community profiling using unique clade-specific marker genes. Nat Methods (2012) 9. DOI: 10.1038/nmeth.2066 48. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS. Metagenomic biomarker discovery and explanation. Genome Biol (2011) 12 R60. DOI: 10.1186/gb-2011-12-6-r60 49. Li R, Yu C, Li Y, Lam TW, Yiu SM, Kristiansen K. SOAP2: An improved ultrafast tool for short read alignment. Bioinformatics (2009) 25. DOI: 10.1093/bioinformatics/btp336 50. Li J, Jia H, Cai X, Zhong H, Feng Q, Sunagawa S. An integrated catalog of reference genes in the human gut microbiome. Nat Biotechnol (2014) 32. DOI: 10.1038/nbt.2942 51. Drula E, Garron ML, Dogan S, Lombard V, Henrissat B, Terrapon N. The carbohydrate-active enzyme database: functions and literature. Nucleic Acids Res (2022) 50. DOI: 10.1093/nar/gkab1045 52. Urban M, Cuzick A, Seager J, Wood V, Rutherford K, Venkatesh SY. PHI-base: the pathogen-host interactions database. Nucleic Acids Res (2020) 48. DOI: 10.1093/nar/gkz904 53. Valles-Colomer M, Falony G, Darzi Y, Tigchelaar EF, Wang J, Tito RY. The neuroactive potential of the human gut microbiota in quality of life and depression. Nat Microbiol (2019) 4. DOI: 10.1038/s41564-018-0337-x 54. Chong J, Wishart DS, Xia J. Using MetaboAnalyst 4.0 for comprehensive and integrative metabolomics data analysis. Curr Protoc Bioinf (2019) 68. DOI: 10.1002/cpbi.86 55. Masi A, DeMayo MM, Glozier N, Guastella AJ. An overview of autism spectrum disorder, heterogeneity and treatment options. Neurosci Bull (2017) 33. DOI: 10.1007/s12264-017-0100-y 56. Tordjman S, Cohen D, Coulon N, Anderson GM, Botbol M, Canitano R. Reframing autism as a behavioral syndrome and not a specific mental disorder: Implications of genetic and phenotypic heterogeneity. Neurosci Biobehav Rev (2017) 80 210. DOI: 10.1016/j.neubiorev.2017.01.030 57. An JY, Claudianos C. Genetic heterogeneity in autism: From single gene to a pathway perspective. Neurosci Biobehav Rev (2016) 68. DOI: 10.1016/j.neubiorev.2016.06.013 58. Newschaffer CJ, Fallin D, Lee NL. Heritable and nonheritable risk factors for autism spectrum disorders. Epidemiol Rev (2002) 24. DOI: 10.1093/epirev/mxf010 59. Hong RP, Hou YY, Xu XJ, Lang JD, Jin YF, Zeng XF. The difference of gut microbiota and their correlations with urinary organic acids between autistic children with and without atopic dermatitis. Front Cell Infect Microbiol (2022) 12. DOI: 10.3389/fcimb.2022.886196 60. Warren RP, Margaretten NC, Pace NC, Foster A. Immune abnormalities in patients with autism. J Autism Dev Disord (1986) 16. DOI: 10.1007/BF01531729 61. Wu S, Ding Y, Wu F, Li R, Xie G, Hou J. Family history of autoimmune diseases is associated with an increased risk of autism in children: A systematic review and meta-analysis. Neurosci Biobehav Rev (2015) 55. DOI: 10.1016/j.neubiorev.2015.05.004 62. Garay PA, Hsiao EY, Patterson PH, McAllister AK. Maternal immune activation causes age- and region-specific changes in brain cytokines in offspring throughout development. Brain Behav Immun (2013) 31 54-68. DOI: 10.1016/j.bbi.2012.07.008 63. Schneider CK, Melmed RD, Barstow LE, Enriquez FJ, Ranger-Moore J, Ostrem JA. Oral human immunoglobulin for children with autism and gastrointestinal dysfunction: A prospective, open-label study. J Autism Dev Disord (2006) 36. DOI: 10.1007/s10803-006-0141-y 64. Asadabadi M, Mohammadi MR, Ghanizadeh A, Modabbernia A, Ashrafi M, Hassanzadeh E. Celecoxib as adjunctive treatment to risperidone in children with autistic disorder: A randomized, double-blind, placebo-controlled trial. Psychopharmacol (Berl) (2013) 225. DOI: 10.1007/s00213-012-2796-8 65. Akhondzadeh S, Fallah J, Mohammadi MR, Imani R, Mohammadi M, Salehi B. Double-blind placebo-controlled trial of pentoxifylline added to risperidone: Effects on aberrant behavior in children with autism. Prog Neuropsychopharmacol Biol Psychiatry (2010) 34. DOI: 10.1016/j.pnpbp.2009.09.012 66. Boris M, Kaiser CC, Goldblatt A, Elice MW, Edelson SM, Adams JB. Effect of pioglitazone treatment on behavioral symptoms in autistic children. J Neuroinflamm (2007) 4. DOI: 10.1186/1742-2094-4-3 67. Melandri D, Zlatareva I, Chaleil RAG, Dart RJ, Chancellor A, Nussbaumer O. The γδTCR combines innate immunity with adaptive immunity by utilizing spatially distinct regions for agonist selection and antigen responsiveness. Nat Immunol (2018) 19. DOI: 10.1038/s41590-018-0253-5 68. Alfawaz HA, Bhat RS, Al-Ayadhi L, El-Ansary AK. Protective and restorative potency of vitamin d on persistent biochemical autistic features induced in propionic acid-intoxicated rat pups. BMC Complement Altern Med (2014) 14. DOI: 10.1186/1472-6882-14-416 69. Zhang Y, Gao D, Kluetzman K, Mendoza A, Bolivar VJ, Reilly A. The maternal autoimmune environment affects the social behavior of offspring. J Neuroimmunol (2013) 258 51-60. DOI: 10.1016/j.jneuroim.2013.02.019 70. Sweeten TL, Posey DJ, Shankar S, McDougle CJ. High nitric oxide production in autistic disorder: a possible role for interferon-gamma. Biol Psychiatry (2004) 55. DOI: 10.1016/j.biopsych.2003.09.001 71. Jyonouchi H, Sun S, Itokazu N. Innate immunity associated with inflammatory responses and cytokine production against common dietary proteins in patients with autism spectrum disorder. Neuropsychobiology (2002) 46 76-84. DOI: 10.1159/000065416 72. Patel N, Crider A, Pandya CD, Ahmed AO, Pillai A. Altered mRNA levels of glucocorticoid receptor, mineralocorticoid receptor, and Co-chaperones (FKBP5 and PTGES3) in the middle frontal gyrus of autism spectrum disorder subjects. Mol Neurobiol (2016) 53. DOI: 10.1007/s12035-015-9178-2 73. Sapolsky RM. The physiological relevance of glucocorticoid endangerment of the hippocampus. Ann N Y Acad Sci (1994) 746 294-307. DOI: 10.1111/j.1749-6632.1994.tb39247.x 74. Saghazadeh A, Ataeinia B, Keynejad K, Abdolalizadeh A, Hirbod-Mobarakeh A, Rezaei N. A meta-analysis of pro-inflammatory cytokines in autism spectrum disorders: Effects of age, gender, and latitude. J Psychiatr Res (2019) 115 90-102. DOI: 10.1016/j.jpsychires.2019.05.019 75. Nour-Eldine W, Ltaief SM, Abdul Manaph NP, Al-Shammari AR. In search of immune cellular sources of abnormal cytokines in the blood in autism spectrum disorder: A systematic review of case-control studies. Front Immunol (2022) 13. DOI: 10.3389/fimmu.2022.950275 76. Finegold SM, Dowd SE, Gontcharova V, Liu C, Henley KE, Wolcott RD. Pyrosequencing study of fecal microflora of autistic and control children. Anaerobe (2010) 16. DOI: 10.1016/j.anaerobe.2010.06.008 77. Strati F, Cavalieri D, Albanese D, De Felice C, Donati C, Hayek J. New evidences on the altered gut microbiota in autism spectrum disorders. Microbiome (2017) 5. DOI: 10.1186/s40168-017-0242-1 78. Sutcliffe IC. A phylum level perspective on bacterial cell envelope architecture. Trends Microbiol (2010) 18. DOI: 10.1016/j.tim.2010.06.005 79. Tomova A, Husarova V, Lakatosova S, Bakos J, Vlkova B, Babinska K. Gastrointestinal microbiota in children with autism in Slovakia. Physiol Behav (2015) 138. DOI: 10.1016/j.physbeh.2014.10.033 80. Koliada K, Syzenko G, Moseiko V, Budovska L, Puchkov K, Perederiy V. Association between body mass index and Firmicutes/Bacteroidetes ratio in an adult Ukrainian population. BMC Microbiol (2017) 17 120. DOI: 10.1186/s12866-017-1027-1 81. Riva A, Borgo F, Lassandro C, Verduci E, Morace G, Borghi E. Pediatric obesity is associated with an altered gut microbiota and discordant shifts in firmicutes populations. Environ Microbiol (2017) 19 95-105. DOI: 10.1111/1462-2920.13463 82. Collado MC, Derrien M, Isolauri E, de Vos WM, Salminen S. Intestinal integrity and akkermansia muciniphila, a mucin-degrading member of the intestinal microbiota present in infants, adults, and the elderly. Appl Environ Microbiol (2007) 73. DOI: 10.1128/AEM.01477-07 83. Wang L, Christophersen CT, Sorich MJ, Gerber JP, Angley MT, Conlon MA. Low relative abundances of the mucolytic bacterium akkermansia muciniphila and bifidobacterium spp. in feces of children with autism. Appl Environ Microbiol (2011) 77. DOI: 10.1128/AEM.05212-11 84. Inoue R, Sakaue Y, Sawai C, Sawai T, Ozeki M, Romero-Pérez GA. A preliminary investigation on the relationship between gut microbiota and gene expressions in peripheral mononuclear cells of infants with autism spectrum disorders. Biosci Biotechnol Biochem (2016) 80. DOI: 10.1080/09168451.2016.1222267 85. Downes J, Vartoukian SR, Dewhirst FE, Izard J, Chen T, Yu WH. Pyramidobacter piscolens gen. nov., sp. nov., a member of the phylum 'Synergistetes' isolated from the human oral cavity. Int J Syst Evol Microbiol (2009) 59. DOI: 10.1099/ijs.0.000364-0 86. Deng ZL, Szafrański SP, Jarek M, Bhuju S, Wagner-Döbler I. Dysbiosis in chronic periodontitis: Key microbial players and interactions with the human host. Sci Rep (2017) 7 3703. DOI: 10.1038/s41598-017-03804-8 87. Ferrazzano GF, Salerno C, Bravaccio C, Ingenito A, Sangianantoni G, Cantile T. Autism spectrum disorders and oral health status: review of the literature. Eur J Paediatr Dent (2020) 21 9-12. DOI: 10.23804/ejpd.2020.21.01.02 88. Marchandin H, Damay A, Roudière L, Teyssier C, Zorgniotti I, Dechaud H. Phylogeny, diversity and host specialization in the phylum synergistetes with emphasis on strains and clones of human origin. Res Microbiol (2010) 161 91-100. DOI: 10.1016/j.resmic.2009.12.008 89. Xie C, Teng J, Wang X, Xu B, Niu Y, Ma L. Multi-omics analysis reveals gut microbiota-induced intramuscular fat deposition. Anim Nutr (2021) 9 84-99. DOI: 10.1016/j.aninu.2021.10.010 90. Li G, Yin P, Chu S, Gao W, Cui S, Guo S. Correlation analysis between GDM and gut microbial composition in late pregnancy. J Diabetes Res (2021) 2021. DOI: 10.1155/2021/8892849 91. Lawson PA, Song Y, Liu C, Molitoris DR, Vaisanen ML, Collins MD. Anaerotruncus colihominis gen. nov., sp. nov., from human faeces. Int J Syst Evol Microbiol (2004) 54. DOI: 10.1099/ijs.0.02653-0 92. Yao Y, Yan L, Chen H, Wu N, Wang W, Wang D. Cyclocarya paliurus polysaccharides alleviate type 2 diabetic symptoms by modulating gut microbiota and short-chain fatty acids. Phytomedicine (2020) 77. DOI: 10.1016/j.phymed.2020.153268 93. Satokari R, Fuentes S, Mattila E, Jalanka J, de Vos WM, Arkkila P. Fecal transplantation treatment of antibiotic-induced, noninfectious colitis and long-term microbiota follow-up. Case Rep Med (2014) 2014. DOI: 10.1155/2014/913867 94. Lee JY, Mannaa M, Kim Y, Kim J, Kim GT, Seo YS. Comparative analysis of fecal microbiota composition between rheumatoid arthritis and osteoarthritis patients. Genes (Basel) (2019) 10. DOI: 10.3390/genes10100748 95. Jeong S, Huang LK, Tsai MJ, Liao YT, Lin YS, Hu CJ. Cognitive function associated with gut microbial abundance in sucrose and s-Adenosyl-L-Methionine (SAMe) metabolic pathways. J Alzheimers Dis (2022) 87. DOI: 10.3233/JAD-215090 96. Kato M. New insights into IFN-γ in rheumatoid arthritis: role in the era of JAK inhibitors. Immunol Med (2020) 43. DOI: 10.1080/25785826.2020.1751908 97. Wang L, Christophersen CT, Sorich MJ, Gerber JP, Angley MT, Conlon MA. Elevated fecal short chain fatty acid and ammonia concentrations in children with autism spectrum disorder. Dig Dis Sci (2012) 57. DOI: 10.1007/s10620-012-2167-7 98. Kau AL, Ahern PP, Griffin NW, Goodman AL, Gordon JI. Human nutrition, the gut microbiome and the immune system. Nature (2011) 474. DOI: 10.1038/nature10213 99. Ulsemer P, Toutounian K, Kressel G, Schmidt J, Karsten U, Hahn A. Safety and tolerance of bacteroides xylanisolvens DSM 23964 in healthy adults. Benef Microbes (2012) 3 99-111. DOI: 10.3920/BM2011.0051 100. de Vrese M, Winkler P, Rautenberg P, Harder T, Noah C, Laue C. Effect of lactobacillus gasseri PA 16/8, bifidobacterium longum SP 07/3, b. bifidum MF 20/5 on common cold episodes: A double blind, randomized, controlled trial. Clin Nutr (2005) 24. DOI: 10.1016/j.clnu.2005.02.006 101. Wang H, Lee IS, Braun C, Enck P. Effect of probiotics on central nervous system functions in animals and humans: A systematic review. J Neurogastroenterol Motil (2016) 22 589-605. DOI: 10.5056/jnm16018 102. Mirande C, Kadlecikova E, Matulova M, Capek P, Bernalier-Donadille A, Forano E. Dietary fibre degradation and fermentation by two xylanolytic bacteria bacteroides xylanisolvens XB1AT and roseburia intestinalis XB6B4 from the human intestine. J Appl Microbiol (2010) 109. DOI: 10.1111/j.1365-2672.2010.04671.x 103. Chassard C, Delmas E, Lawson PA, Bernalier-Donadille A. Bacteroides xylanisolvens sp. nov., a xylandegrading bacterium isolated from human faeces. Int J Syst Evol Microbiol (2008) 58. DOI: 10.1099/ijs.0.65504-0 104. Thaiss CA, Levy M, Suez J, Elinav E. The interplay between the innate immune system and the microbiota. Curr Opin Immunol (2014) 26. DOI: 10.1016/j.coi.2013.10.016 105. Hsiao EY, McBride SW, Hsien S, Sharon G, Hyde ER, McCue T. Microbiota modulate behavioral and physiological abnormalities associated with neurodevelopmental disorders. Cell (2013) 155. DOI: 10.1016/j.cell.2013.11.024 106. Gilbert JA, Krajmalnik-Brown R, Porazinska DL, Weiss SJ, Knight R. Toward effective probiotics for autism and other neurodevelopmental disorders. Cell (2013) 155. DOI: 10.1016/j.cell.2013.11.035 107. Schell MA, Karmirantzou M, Snel B, Vilanova D, Berger B, Pessi G. The genome sequence of bifidobacterium longum reflects its adaptation to the human gastrointestinal tract. Proc Natl Acad Sci U.S.A. (2002) 99. DOI: 10.1073/pnas.212527599 108. Lievin V, Peiffer I, Hudault S, Rochat F, Brassart D, Neeser JR. Bifidobacterium strains from resident infant human gastrointestinal microflora exert antimicrobial activity. Gut (2000) 47. DOI: 10.1136/gut.47.5.646 109. Tanaka H, Hashiba H, Kok J, Mierau I. Bile salt hydrolase of bifidobacterium longum - biochemical and genetic characterization. Appl Environ Microbiol (2000) 66. DOI: 10.1128/AEM.66.6.2502-2512.2000 110. Sela DA, Chapman J, Adeuya A, Kim JH, Chen F, Whitehead TR. The genome sequence of bifidobacterium longum subsp. infantis reveals adaptations for milk utilization within the infant microbiome. Proc Natl Acad Sci U.S.A. (2008) 105. DOI: 10.1073/pnas.0809584105 111. Ma X, Shin YJ, Jang HM, Joo MK, Yoo JW, Kim DH. Lactobacillus rhamnosus and bifidobacterium longum alleviate colitis and cognitive impairment in mice by regulating IFN-γ to IL-10 and TNF-α to IL-10 expression ratios. Sci Rep (2021) 11 20659. DOI: 10.1038/s41598-021-00096-x 112. De Angelis M, Piccolo M, Vannini L, Siragusa S, De Giacomo A, Serrazzanetti DI. Fecal microbiota and metabolome of children with autism and pervasive developmental disorder not otherwise specified. PloS One (2013) 8. DOI: 10.1371/journal.pone.0076993 113. Adams JB, Johansen LJ, Powell LD, Quig D, Rubin RA. Gastrointestinal flora and gastrointestinal status in children with autism - comparisons to typical children and correlation with autism severity. BMC Gastroenterol (2011) 11. DOI: 10.1186/1471-230X-11-22 114. Cani PD. Interactions between gut microbes and host cells control gut barrier and metabolism. Int J Obes Suppl (2016) 6. DOI: 10.1038/ijosup.2016.6 115. Mussap M, Noto A, Fanos V. Metabolomics of autism spectrum disorders: early insights regarding mammalian-microbial cometabolites. Expert Rev Mol Diagn (2016) 16. DOI: 10.1080/14737159.2016.1202765 116. Bugajska J, Berska J, Wojtyto T, Bik-Multanowski M, Sztefko K. The amino acid profile in blood plasma of young boys with autism. Psychiatr Pol (2017) 51. DOI: 10.12740/PP/65046 117. Bala KA, Doğan M, Mutluer T, Kaba S, Aslan O, Balahoroğlu R. Plasma amino acid profile in autism spectrum disorder (ASD). Eur Rev Med Pharmacol Sci (2016) 20 118. Xu XJ, Cai XE, Meng FC, Song TJ, Wang XX, Wei YZ. Comparison of the metabolic profiles in the plasma and urine samples between autistic and typically developing boys: a preliminary study. Front Psychiatry (2021) 12. DOI: 10.3389/fpsyt.2021.657105 119. Diémé B, Mavel S, Blasco H, Tripi G, Bonnet-Brilhault F, Malvy J. Metabolomics study of urine in autism spectrum disorders using a multiplatform analytical methodology. J Proteome Res (2015) 14. DOI: 10.1021/acs.jproteome.5b00699 120. Clayton TA. Metabolic differences underlying two distinct rat urinary phenotypes, a suggested role for gut microbial metabolism of phenylalanine and a possible connection to autism. FEBS Lett (2012) 586. DOI: 10.1016/j.febslet.2012.01.049 121. Williams BL, Hornig M, Buie T, Bauman ML, Cho Paik M, Wick I. Impaired carbohydrate digestion and transport and mucosal dysbiosis in the intestines of children with autism and gastrointestinal disturbances. PloS One (2011) 6. DOI: 10.1371/journal.pone.0024585 122. Brigandi SA, Shao H, Qian SY, Shen Y, Wu BL, Kang JX. Autistic children exhibit decreased levels of essential fatty acids in red blood cells. Int J Mol Sci (2015) 16. DOI: 10.3390/ijms160510061 123. Kushak RI, Winter HS, Buie TM, Cox SB, Phillips CD, Ward NL. Analysis of the duodenal microbiome in autistic individuals: association with carbohydrate digestion. J Pediatr Gastroenterol Nutr (2017) 64. DOI: 10.1097/MPG.0000000000001458 124. Fernandez-Julia PJ, Munoz-Munoz J, van Sinderen D. A comprehensive review on the impact of β-glucan metabolism by bacteroides and bifidobacterium species as members of the gut microbiota. Int J Biol Macromol (2021) 181. DOI: 10.1016/j.ijbiomac.2021.04.069 125. Wong JM, de Souza R, Kendall CW, Emam A, Jenkins DJ. Colonic health: Fermentation and short chain fatty acids. J Clin Gastroenterol (2006) 40. DOI: 10.1097/00004836-200603000-00015 126. Vijay N, Morris ME. Role of monocarboxylate transporters in drug delivery to the brain. Curr Pharm Des (2014) 20. DOI: 10.2174/13816128113199990462 127. Morris G, Berk M, Carvalho A, Caso JR, Sanz Y, Walder K. The role of the microbial metabolites including tryptophan catabolites and short chain fatty acids in the pathophysiology of immune-inflammatory and neuroimmune disease. Mol Neurobiol (2017) 54. DOI: 10.1007/s12035-016-0004-2 128. MacFabe DF. Short-chain fatty acid fermentation products of the gut microbiome: implications in autism spectrum disorders. Microb Ecol Health Dis (2012) 23 19260. DOI: 10.3402/mehd.v23i0.19260 129. Innis SM. The role of dietary n-6 and n-3 fatty acids in the developing brain. Dev Neurosci (2000) 22. DOI: 10.1159/000017478 130. Richardson AJ. Long-chain polyunsaturated fatty acids in childhood developmental and psychiatric disorders. Lipids (2004) 39. DOI: 10.1007/s11745-004-1350-z 131. Madra M, Ringel R, Margolis KG. Gastrointestinal issues and autism spectrum disorder. Child Adolesc Psychiatr Clin N Am (2020) 29. DOI: 10.1016/j.chc.2020.02.005 132. Wright C, Shin JH, Rajpurohit A, Deep-Soboslay A, Collado-Torres L, Brandon NJ. Altered expression of histamine signaling genes in autism spectrum disorder. Transl Psychiatry (2017) 7 e1126. DOI: 10.1038/tp.2017.87 133. Linday LA, Tsiouris JA, Cohen IL, Shindledecker R, DeCresce R. Famotidine treatment of children with autistic spectrum disorders: pilot research using single subject research design. J Neural Transm (Vienna) (2001) 108 593-611. DOI: 10.1007/s007020170059 134. Molenhuis RT, Hutten L, Kas MJH. Histamine H3 receptor antagonism modulates autism-like hyperactivity but not repetitive behaviors in BTBR T+Itpr3tf/J inbred mice. Pharmacol Biochem Behav (2022) 212. DOI: 10.1016/j.pbb.2021.173304 135. Eissa N, Azimullah S, Jayaprakash P, Jayaraj RL, Reiner D, Ojha SK. The dual-active histamine H3 receptor antagonist and acetylcholine esterase inhibitor E100 alleviates autistic-like behaviors and oxidative stress in valproic acid induced autism in mice. Int J Mol Sci (2020) 21. DOI: 10.3390/ijms21113996 136. Kamei M, Otani Y, Hayashi H, Nakamura T, Yanai K, Furuta K. Suppression of IFN-γ production in murine splenocytes by histamine receptor antagonists. Int J Mol Sci (2018) 19. DOI: 10.3390/ijms19124083 137. Kohda F, Koga T, Uchi H, Urabe K, Furue M. Histamine-induced IL-6 and IL-8 production are differentially modulated by IFN-gamma and IL-4 in human keratinocytes. J Dermatol Sci (2002) 28 34-41. DOI: 10.1016/s0923-1811(01)00147-5 138. Khalaj M, Saghazadeh A, Shirazi E, Shalbafan MR, Alavi K, Shooshtari MH. Palmitoylethanolamide as adjunctive therapy for autism: Efficacy and safety results from a randomized controlled trial. J Psychiatr Res (2018) 103. DOI: 10.1016/j.jpsychires.2018.04.022 139. Taliou A, Zintzaras E, Lykouras L, Francis K. An open-label pilot study of a formulation containing the anti-inflammatory flavonoid luteolin and its effects on behavior in children with autism spectrum disorders. Clin Ther (2013) 35 592-602. DOI: 10.1016/j.clinthera.2013.04.006
--- title: Assessment of existing anthropometric indices for screening sarcopenic obesity in older adults authors: - Jin Eui Kim - Jimi Choi - Miji Kim - Chang Won Won journal: The British Journal of Nutrition year: 2023 pmcid: PMC9975784 doi: 10.1017/S0007114522001817 license: CC BY 4.0 --- # Assessment of existing anthropometric indices for screening sarcopenic obesity in older adults ## Body Sarcopenic obesity is the coexistence of sarcopenia and obesity, which is characterised by age-related changes in body composition, decreased muscle mass, increased fat mass and decreased muscle strength and physical performance[1]. It is a common problem in older adults that results in physical disability as well as increased cardiovascular diseases (CVD) morbidity and mortality(2–4). While earlier diagnosis and treatment is imperative, sarcopenic obesity lacks a standardised diagnostic criterion and has traditionally been defined as low muscle mass and high-fat mass[5,6]. Dual-energy X-ray absorptiometry (DXA), MRI and CT are the most precise and accurate methods for measuring body composition; however, these methods have several limitations in clinical practice. They are expensive and time consuming; furthermore, CT is potentially hazardous because of radiation exposure[7,8]. More importantly, such equipment lacks portability, which makes it difficult to utilise them in extramural care settings. Therefore, it is crucial to develop a method to diagnose sarcopenic obesity in diverse settings while keeping in mind factors such as cost-effectiveness, time taken and safety. Anthropometry is a simple and practical method to estimate body composition[9], and there are several anthropometric indices that have been used widely in epidemiologic studies. The global standard to define obesity is based on body mass index (BMI); waist circumference (WC) and waist-to-height ratio (WHtR) were developed for indicators of abdominal obesity. Those anthropometric indices are strongly associated with the onset of CVD(10–13), and CVD is significantly associated with sarcopenic obesity. This may imply the possibility of anthropometric indices for screening sarcopenic obesity. However, BMI has shown a J-shaped relationship with CVD mortality, which resulted in the obesity paradox(14–16), possibly because of its inability to discriminate muscle mass from fat mass[17,18]. In other words, BMI cannot reflect low muscle mass and high fat mass simultaneously. After the studies revealed that abdominal fat, especially visceral fat, is strong predictor of all-cause and CVD mortality[19,20], WC and WHtR were proposed in order to overcome the limitation BMI. WC and WHtR had shown a significantly higher association with CVD mortality compared with BMI; however, similar obesity paradox phenomenon was observed[21]. In addition, a recent study suggested a high correlation among WC, WHtR and BMI[22], which represents that high WC and WHtR may also be due to high muscle mass, not solely by high-fat mass. Therefore, the adequacy of anthropometric indices as independent indicators of sarcopenic obesity is still not clear and has to be tested before clinical application. At the same time, a development of new anthropometric indices that can reciprocally reflect muscle mass and fat mass is needed. In 2018, a new anthropometric index called the weight-adjusted waist index (WWI), which standardised WC for weight, was developed to overcome the shortcomings of existing anthropometric indices[23]. The study showed that WWI had a relatively consistent and linear relationship with both CVD morbidity and mortality. More recently, WWI was shown to discriminate muscle mass from fat mass as it showed a negative association with muscle mass and a positive association with fat mass in older adults[24], suggesting WWI as a possible indicator of sarcopenic obesity. Along with the change in muscle mass, other components that are proposed to diagnose sarcopenia are muscle strength and physical performance(25–28); however, not only the association of WWI with muscle strength and physical performance has not yet been determined but also the studies regarding existing anthropometric indices and physical function are limited[29,30]. In this study, we aimed to analyse the association of different anthropometric indices, including BMI, WC, WHtR and WWI, with sarcopenic obesity to compare their feasibility for screening sarcopenic obesity in community-dwelling older adults. ## Abstract Sarcopenic obesity is defined as the presence of high fat mass and low muscle mass combined with low physical function, and it is closely related with the onset of cardiovasular diseases (CVD). The existing anthropometric indices, which are being utilised in clinical practice as predictors of CVD, may also be used to screen sarcopenic obesity, but their feasibility remained unknown. Using cross-sectional data of 2031 participants aged 70–84 years (mean age, 75·9 ± 3·9 years; 49·2 % women) from the Korean Frailty and Aging Cohort Study, we analysed the association of anthropometric indices, including body mass index (BMI), waist circumference (WC), waist-to-height ratio (WHtR) and weight-adjusted waist index (WWI) with sarcopenic obesity. Body composition was measured using dual-energy X-ray absorptiometry. Higher WWI, WHtR and WC quartiles were associated with higher risk of sarcopenic obesity; the odds ratio (OR) of sarcopenic obesity were highest in the fourth quartile of the WWI (OR: 10·99, 95 % CI: 4·92–24·85, P for trend < 0·001). WWI provided the best diagnostic power for sarcopenic obesity in men (area under the receiver operating characteristic curve: 0·781, 95 % CI: 0·751–0·837). No anthropometric indices were significantly associated with sarcopenic obesity in women. WWI was the only index that was negatively correlated with physical function in both men and women. WWI showed the strongest association with sarcopenic obesity, defined by high fat mass and low muscle mass combined with low physical function only in older men. No anthropometric indices were associated with sarcopenic obesity in older women. ## Study population The Korean Frailty and Aging Cohort Study (KFACS) is a nationwide multicentre cohort study that was primarily designed to assess the frailty status of community-dwelling older adults in South Korea. The participants were sex- and age-stratified community residents recruited from urban and rural areas around ten centres who were ambulatory with or without walking aids. The age ratio was 6:5:4 for age 70–74, 75–79 and 80–84 years, respectively, and the sex ratio was 1:1. Followed by the suggestion from the frailty consensus[31], the starting age of the KFACS was set from 70. The participants over 85 years were excluded due to their difficulty of centre visits and follow-up surveys; the advanced age over 85 years also had a higher probability of interrupting the identification of physical frailty-associated risk factors. Overall, the inclusion criteria of the participant were age 70–84 years, currently living in the community, having no problem with communication and no prior diagnosis of dementia. The baseline study comprised face-to-face interviews, health examinations and laboratory tests with a total of 3013 participants. Among the total participants, 2403 underwent body composition measurement with DXA in eight university hospitals and 610 with bioelectrical impedance analysis in two community centres. For this study, those who underwent bioelectrical impedance analysis were excluded because of possible systematic bias between DXA and bioelectrical impedance analysis[32]. The final analysis included 2031 participants after excluding 321 participants who had artificial joints, pins, plates or other types of metal objects in any part of their bodies, as metal implants could have affected the measurement accuracy of appendicular skeletal muscle mass (ASM) or percentage of body fat[33], and fifty-one participants who had missing data for the diagnostic criteria of sarcopenic obesity. The details of the KFACS protocol have been described previously[34]. ## Ethics The KFACS protocol was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects/patients were approved by the Institutional Review Board of the Clinical Research Ethics Committee of the Kyung Hee University Medical Center. Written informed consent was obtained from all participants (IRB number: 2015-12-103). This study was approved as an exempt from the Institutional Review Board review (IRB number: 2021-02-021). ## Anthropometric measurements The height, weight and WC of all the participants were recorded. The height and WC were measured to the nearest 0·1 cm, and weight was measured to the nearest 0·1 kg. BMI was calculated as weight (kg) divided by the square of the height (m2). WC (cm) was measured at the midpoint between the lower end of the last rib and the upper ridge of the iliac crest. WHtR was calculated as WC (cm) divided by height (cm)[35], and WWI was calculated as WC (cm) divided by the square root of the weight (√kg). Details regarding the derivation of WWI have been described in a previous study[23]. ## Body composition measurements ASM and percentage of body fat were measured using DXA (Lunar, GE Healthcare, Madison, WI; Hologic DXA, Hologic Inc.). The participants were asked to remove all metal accessories before the scan and lie in a supine position on the scanner table with their limbs placed parallel to their bodies, according to the manufacturer’s protocol. ASM was calculated as the sum of the lean masses of the right and left arms and legs. ASM index was defined as ASM/height2 (kg/m2)[6]. Our laboratory assessment of forty volunteers with repositioning between scans demonstrated that the coefficients of variation for whole-body composition were < 2·5 %. ## Muscle strength Muscle strength was evaluated by grip strength and measured using a digital handgrip dynamometer (T.K.K.5401; Takei Scientific Instruments Co. Ltd). The participants were asked to stand upright, place their shoulder in a neutral position with both arms fully extended and hold the dynamometer for 3 s with maximum strength. The strength was measured twice for each hand at 3-min intervals. The best records for each hand were rounded to the nearest 0·1 kg[36]. ## Physical performance Physical performance was evaluated by 4-m usual gait speed, the five-times sit-to-stand test and the Short Physical Performance Battery (SPPB). The 4-m usual gait speed was measured using an automatic gait speed meter (Dynamicphysiology), with acceleration and deceleration phases of 1·5 m each[37]. The participants performed two trials with their usual walking paces, and the average rounded to the nearest 0·01 m/s was taken for the analysis. The five-times sit-to-stand test was conducted by measuring the time it took for the participants to stand five times from a sitting position as quickly as possible from a straight-backed armchair without using their arms[38]. The SPPB consists of the 4-m usual gait speed test measures, five-times sit-to-stand test measures and three standing balance measures[38]. In the standing balance test, the participants were first asked to stand with their feet placed together as close as possible, then in a semi-tandem position, and finally in a tandem position for 10 s. Each item of the SPPB was scored on a scale of 0–4 based on the normative scores obtained from the Established Population for Epidemiologic Studies of the Elderly, which makes the total possible score between 0 and 12[39]. ## Definitions of sarcopenic obesity While there is no global consensus to define obesity by the percentage of body fat[40], the commonly used definition of obesity for older population was suggested in the previous research of New Mexico Aging Process Study. It defined obesity as the percentage of body fat greater than 60th percentile in the study population resulting in cut-off values of ≥ 28 % for men and ≥40 % for women[5,41]. Followed by the standard of the New Mexico Aging Process Study, we defined obesity as a high total fat mass greater than 60th percentile of our study population according to the percentage of body fat. The resulting cut-off values were ≥ 28·2 % for men and ≥ 38·8 % for women, which correspond to those of the New Mexico Aging Process Study. Three different definitions of sarcopenia were established based on the Asian Working Group for Sarcopenia 2019 consensus[25], which were as follows: [1] low muscle mass, [2] low muscle mass with low muscle strength and/or slow gait speed and [3] low muscle mass with low muscle strength and/or physical performance (slow gait speed, poor performance in the five-times sit-to-stand test and/or low SPPB score). By combining the definitions of obesity and sarcopenia, we established the following three sets of diagnostic criteria for sarcopenic obesity: Criterion 1: High fat mass + low muscle mass Criterion 2: High fat mass + low muscle mass + low muscle strength and/or slow gait speed Criterion 3: High fat mass + low muscle mass + low muscle strength and/or physical performance Low muscle mass was defined as an ASM/height2 value of < 7·00 kg/m2 for men and < 5·40 kg/m2 for women; low muscle strength was defined as a grip strength of < 28 kg for men and < 18 kg for women; the cut-off scores for the 4-m usual gait speed, five-times sit-to-stand test and SPPB for low physical performance were < 1·0 m/s, ≥ 12 s and ≤ 9, respectively, for both sexes[25]. ## Statistical analyses Data are presented as mean ± standard deviation (sd) for continuous variables and as numbers (percentages) for categorical variables. Continuous variables with skewed distributions are reported as median (interquartile range). To assess the differences in characteristics between the sexes, the means or medians of the two groups were compared using the Student’s t test or Mann–Whitney U-test, respectively. The percentages of categorical variables were compared using the χ 2 or Fisher’s exact test, as appropriate. As there was a significant sex-specific difference in the association between the anthropometric indices and sarcopenic obesity in the exploratory data analysis, all the analyses were stratified by sex. The association between the anthropometric indices and sarcopenic obesity was evaluated in the unadjusted and age-adjusted model using binary logistic regression. The results were reported as odds ratio (OR) according to the quartiles of each anthropometric index and corresponding 95 % CI to compare the strengths of the associations of the indices measured on different scales. The OR per sd were calculated using a multiple logistic regression model, and the predicted probability calculated from this model was used to evaluate the discriminative ability of the indices for sarcopenic obesity by analysing the receiver operating characteristic curve and the area under the receiver operating characteristic curve (AUC) (95 % CI). The AUC of the anthropometric indices were compared using DeLong’s method[42]. The correlation between the anthropometric indices and continuous components of sarcopenic obesity was evaluated using Pearson’s or Spearman’s correlation analysis, according to the distribution of variables. Statistical significance was set at a $P \leq 0$·05. All statistical analyses were performed using SAS software (version 9.4; SAS Institute Inc.) and R (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria). This study is a secondary analysis; the sample size was determined by recruitment in the KFACS[34] and satisfied the rule of ten events per variable in logistic regression analysis[43]. ## Results The characteristics of the study participants are listed in Table 1. This study included 1032 men and 999 women. Men were significantly older (76·4 ± 3·9 years v. 75·4 ± 3·9 years), had a higher WC (88·4 ± 8·3 cm v. 86·0 ± 8·2 cm), a lower BMI (23·9 ± 2·8 v. 24·4 ± 2·8), WHtR (0·54 ± 0·05 cm v. 0·57 ± 0·05 cm) and WWI (11·0 ± 0·6 cm/√kg v. 11·5 ± 0·7 cm/√kg) compared with women. Men were also more likely to have higher incidence rates of myocardial infarction, cerebrovascular disease, diabetes mellitus and chronic obstructive pulmonary disease than women. On the other hand, women were more likely to have higher incidence rates of hypertension, dyslipidaemia, osteoarthritis, rheumatoid arthritis, osteoporosis and asthma than men. The prevalence of sarcopenic obesity was higher in men than in women according to criteria 1 (21·8 % v. 14·3 %), 2 (10·5 % v. 7·0 %) and 3 (12·7 % v. 10·3 %); the difference between men and women was not significant in criteria 3 ($$P \leq 0$$·093). The percentage of body fat was lower in men than in women (26·3 ± 6·0 % v. 36·8 ± 5·9 %). Although men had higher ASM (19·2 ± 2·7 kg v. 13·5 ± 1·8 kg) and ASM/height2 (7·0 ± 0·8 kg/m2 v. 5·8 ± 0·7 kg/m2) than women, the proportion of low muscle mass was also higher in men than in women (48·3 % v. 27·1 %). The difference between the sexes in the proportion of low muscle strength was not significant (21·7 % in men and 19·0 % in women, $$P \leq 0$$·133). The proportion of slow gait speed (23·3 % v. 33·1 %) and low physical performance (39·5 % v. 53·9 %) was lower in men than in women. Table 1.Characteristics of the study participants(Mean values and standard deviations)TotalMenWomenMeanMeanMeanVariables(n 2031) sd (n 1032) sd (n 999) sd P Age, years75·93·976·43·975·43·9< 0·001Height, cm158·58·4164·95·7151·95·2< 0·001Weight, kg60·89·365·08·956·57·7< 0·001BMI, kg/m2 24·12·823·92·824·42·8< 0·001WC, cm87·28·388·48·386·08·2< 0·001WHtR, cm0·550·050·540·050·570·05< 0·001WWI, cm/√kg11·20·711·00·611·50·7< 0·001Percentage of body fat31·57·926·36·036·85·9< 0·001SBP, mmHg131·115·6131·115·3131·115·90·962DBP, mmHg77·49·178·09·176·99·20·009 n % n % n %Comorbidity Hypertension112255·354653·057657·70·033 Dyslipidaemia65432·624924·440541·2< 0·001 Myocardial infarction512·5393·8121·2< 0·001 Heart failure110·660·650·50·822 Angina1135·6626·0515·20·399 Peripheral artery disease170·8111·160·60·258 CVD894·4605·8292·90·001 Diabetes mellitus43221·324423·718818·90·008 Osteoarthritis38419·010610·327828·0< 0·001 Rheumatoid arthritis381·980·8303·0< 0·001 Osteoporosis25912·9262·523323·8< 0·001 Asthma653·2252·4404·00·043 COPD231·1201·930·30·001Mean sd Mean sd Mean sd Laboratory results Creatinine, mg/dl0·850·280·980·300·720·16< 0·001 eGFR, ml/min/1·73 m2 81·219·079·019·383·418·6< 0·001 FBS Median979996< 0·001 IQR90–11091–11289–108 Total cholesterol, mg/dl175·235·7168·634·7182·135·5< 0·001Triglyceride Median105100110< 0·001 IQR79–14275–13783–146 HDL-cholesterol, mg/dl53·014·250·814·355·213·8< 0·001 LDL-cholesterol, mg/dl108·933·0104·531·7113·433·9< 0·001 HbA1c, %6·00·86·00·86·00·80·158Skeletal muscle mass ASM by DXA, kg16·43·719·22·713·51·8< 0·001 ASM/height2, kg/m2 6·41·07·00·85·80·7< 0·001 Low muscle mass n 769498271< 0·001 %37·948·327·1Maximum grip strength, kg27·07·432·55·721·33·9< 0·001 Low muscle strength n 4142241900·133 %20·421·719·0Physical performance 4-m usual gait speed, m/s1·10·31·20·31·10·2< 0·001 n % n % n % 4-m usual gait speed < 1·0 m/s57128·124023·333133·1< 0·001 Five-times sit-to-stand test Median10·510·011·2< 0·001 IQR8·7–12·98·4–12·39·2–13·7 Five-times sit-to-stand test ≥ 12 s68833·928727·840140·1< 0·001 SPPB score, median (IQR) Median111211< 0·001 IQR10–1211–1210–12 SPPB score ≤ 9 points29414·51029·919219·2< 0·001 Low physical performance94646·640839·553853·9< 0·001Sarcopenic obesity Criterion 136818·122521·814314·3< 0·001 Criterion 21788·810810·5707·00·006 Criterion 323411·513112·710310·30·093WC, waist circumference; WHtR, waist-to-height ratio; WWI, weight-adjusted waist index; SBP, systolic blood pressure; DBP, diastolic blood pressure; CVD, cardiovascular diseases; COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; FBS, fetal bovine serum; ASM, appendicular skeletal muscle mass; DXA, dual-energy X-ray absorptiometry; SPPB, short physical performance battery; IQR, interquartile range. Criterion 1: high-fat mass + low muscle mass; Criterion 2: high-fat mass + low muscle mass + low muscle strength and/or slow gait speed and Criterion 3: high-fat mass + low muscle mass + low muscle strength and/or low physical performance. High-fat mass: body fat percentage of ≥ 28·2 % for men and ≥ 38·8 % for women; low muscle mass: ASM/height2 of < 7·00 kg/m2 for men and < 5·40 kg/m2 for women; low muscle strength: grip strength of < 28 kg for men and < 18 kg for women; slow gait speed: 4-m usual gait speed of < 1·0 m/s and low physical performance: five-times sit-to-stand test score of ≥ 12 s, 4-m usual gait speed of < 1·0 m/s and/or SPPB score of ≤ 9.Variables are expressed as means ± standard deviation for continuous variables and as n (%) for categorical variables. Continuous variables with skewed distributions were reported as median (IQR). P values were obtained using the χ 2 test, Fisher’s exact test or Student’s t test, as appropriate. ## Association between anthropometric indices and sarcopenic obesity The age-adjusted prevalence of sarcopenic obesity was high in men with higher WWI and WHtR for all the three diagnostic criteria, but not in women with higher WWI and WHtR (Fig. 1). Followed by the result, we analysed the age-adjusted OR of sarcopenic obesity according to the quartiles of the anthropometric indices as illustrated in Table 2. In men, higher WWI and WHtR quartiles were associated with higher risk of sarcopenic obesity for all the three diagnostic criteria, with the OR being the highest in the fourth quartiles of the WWI (OR: 5·84, 95 % CI: 3·51, 9·70, P for trend < 0·001 in criterion 1; OR: 14·64, 95 % CI: 5·20, 41·25 in criterion 2, P for trend < 0·001; and OR: 10·99, 95 % CI: 4·92, 24·85, P for trend < 0·001 in criterion 3) and WHtR (OR: 3·94, 95 % CI: 2·41, 6·45, P for trend < 0·001 in criterion 1; OR: 5·73, 95 % CI: 2·72–12·05, P for trend < 0·001 in criterion 2; and OR: 5·99, 95 % CI: 3·05, 11·79, P for trend < 0·001 in criterion 3). The OR of sarcopenic obesity were higher in the fourth quartiles of the WWI than in those of the WHtR, especially based on criteria 2 and 3, which included muscle strength and/or physical performance as diagnostic components of sarcopenic obesity. WC also showed the highest OR in the fourth quartiles on criteria 2 (OR: 3·61, 95 % CI: 1·86, 7·01, P for trend < 0·001) and 3 (OR: 4·32, 95 % CI: 2·31, 8·10, P for trend < 0·001), but their values were lower compared with WWI and WHtR. Meanwhile, no anthropometric indices were significantly associated with sarcopenic obesity in women. Similar results were observed in the unadjusted model in both men and women (online Supplementary Table S1). Fig. 1.Age-adjusted prevalence of sarcopenic obesity according to the quartiles of the anthropometric indices. WWI, weight-adjusted waist index; WC, waist circumference; WHtR, waist-to-height ratio. Criterion 1: high fat mass + low muscle mass; Criterion 2: high fat mass + low muscle mass + low muscle strength and/or slow gait speed; Criterion 3: high fat mass + low muscle mass + low muscle strength and/or low physical performance. High fat mass: body fat percentage of ≥ 28·2 % for men and ≥ 38·8 % for women; low muscle mass: appendicular skeletal muscle mass/height2 of < 7·00 kg/m2 for men and < 5·40 kg/m2 for women; low muscle strength: grip strength of < 28 kg for men and < 18 kg for women; slow gait speed: 4-m usual gait speed of < 1·0 m/s and low physical performance: five-times sit-to-stand test score of ≥ 12 s, 4-m usual gait speed of < 1·0 m/s and/or short physical performance battery score of ≤ 9. Table 2.Age-adjusted odds ratios of sarcopenic obesity according to the quartiles of the anthropometric indices(Odd ratio and 95 % confidence intervals)Criterion 1 P Criterion 2 P Criterion 3 P Quartile groupRange n %Age-adjusted OR95 % CI n %Age-adjusted OR95 % CI n %Age-adjusted OR95 % CIWWI Men 1st9·1–10·6228·51 (Reference)41·61 (Reference)72·71 (Reference) 2nd10·6–11·04517·42·251·31, 3·880·003155·83·831·25, 11·760·019197·42·801·16, 6·810·023 3rd11·0–11·46324·43·341·98, 5·63< 0·0013212·48·102·81, 23·35< 0·0013714·35·532·41, 12·70< 0·001 4th11·4–12·99536·85·843·51, 9·70< 0·0015722·114·645·20, 41·25< 0·0016826·410·994·92, 24·85< 0·001 P for trend < 0·001< 0·001< 0·001 Women 1st9·5–10·93514·11 (Reference)197·61 (Reference)2610·41 (Reference) 2nd10·6–11·53413·60·950·57, 1·57)0·830114·40·530·24, 1·140·103187·20·640·34, 1·200·163 3rd11·5–11·93714·81·030·62, 1·690·922176·80·780·39, 1·550·4822811·20·990·56, 1·760·980 4th11·9–14·03714·80·950·57, 1·580·849239·20·860·45, 1·650·6563112·40·940·53, 1·650·820 P for trend 0·9380·9550·811 P for interaction < 0·001< 0·001< 0·001BMI Men 1st14·5–22·02810·91 (Reference)176·61 (Reference)187·01 (Reference) 2nd22·0–23·97428·73·492·16, 5·63< 0·0013112·02·171·16, 4·070·0163915·12·611·44, 4·730·002 3rd23·9–25·77629·53·612·24, 5·83< 0·0013212·42·261·21, 4·220·0114115·92·771·54, 5·01< 0·001 4th25·7–33·14718·21·941·17, 3·210·0112810·91·991·05, 3·770·0353312·82·191·19, 4·030·012 P for trend 0·0290·0500·021 Women 1st16·0–22·6249·61 (Reference)135·21 (Reference)197·61 (Reference) 2nd22·6–24·45220·82·561·52, 4·32< 0·001218·41·870·90, 3·870·0913313·22·011·10, 3·670·023 3rd24·4–26·24819·22·331·37, 3·940·002249·62·241·10, 4·570·0263614·42·271·25, 4·110·007 4th26·2–33·2197·60·790·42, 1·480·456124·80·960·43, 2·180·931156·00·800·40, 1·630·539 P for trend 0·4770·8770·769 P for interaction 0·1660·4170·141WC Men 1st60·0–83·0238·91 (Reference)135·01 (Reference)145·41 (Reference) 2nd83·1–88·55722·12·991·77, 5·03< 0·001218·11·820·88, 3·750·105259·71·991·01, 3·950·048 3rd88·6–93·97529·14·332·61, 7·19< 0·0013513·63·231·65, 6·32< 0·0014316·73·731·97, 7·05< 0·001 4th94·0–115·07027·13·912·34, 6·51< 0·0013915·13·611·86, 7·01< 0·0014919·04·322·31, 8·10< 0·001 P for trend < 0·001< 0·001< 0·001 Women 1st59·5–80·03212·91 (Reference)187·21 (Reference)239·21 (Reference) 2nd80·1–86·04216·81·410·85, 2·320·180176·80·990·49, 1·990·9752510·01·140·63, 2·090·665 3rd86·1–91·54016·01·300·78, 2·140·313208·01·140·58, 2·230·7023112·41·410·79, 2·520·239 4th91·6–111·62911·60·880·51, 1·510·642156·00·770·38, 1·580·478249·61·000·55, 1·840·995 P for trend 0·6050·5980·811 P for interaction 0·0010·0150·007WHtR Men 1st0·35–0·51259·71 (Reference)93·51 (Reference)114·31 (Reference) 2nd0·51–0·545420·92·521·51, 4·21< 0·001218·12·591·16, 5·820·021259·72·521·21, 5·260·014 3rd0·54–0·576826·43·322·02, 5·47< 0·0013212·43·911·81, 8·43< 0·0013915·13·991·98, 8·01< 0·001 4th0·57–0·727830·23·942·41, 6·45< 0·0014617·85·732·72, 12·05< 0·0015621·75·993·05, 11·79< 0·001 P for trend < 0·001< 0·001< 0·001 Women 1st0·41–0·532911·61 (Reference)166·41 (Reference)228·81 (Reference) 2nd0·53–0·574216·81·530·92, 2·550·102156·00·940·45, 1·960·8662510·01·150·63, 2·120·647 3rd0·57–0·604216·81·510·90, 2·510·116197·61·150·57, 2·320·6923112·41·420·79, 2·540·239 4th0·60–0·763012·00·950·55, 1·640·847208·01·010·51, 2·020·9742510·00·960·52, 1·770·897 P for trend 0·8230·8430·943 P for interaction 0·0010·0100·001WWI, weight-adjusted waist index; WC, waist circumference; WHtR, waist-to-height ratio. Criterion 1: high fat mass + low muscle mass; Criterion 2: high fat mass + low muscle mass + low muscle strength and/or slow gait speed and Criterion 3: high fat mass + low muscle mass + low muscle strength and/or low physical performance. High fat mass: body fat percentage of ≥ 28·2 % for men and ≥ 38·8 % for women; low muscle mass: appendicular skeletal muscle mass/height2 of < 7·00 kg/m2 for men and < 5·40 kg/m2 for women; low muscle strength: grip strength of < 28 kg for men and < 18 kg for women; slow gait speed: 4-m usual gait speed of < 1·0 m/s and low physical performance: five-times sit-to-stand test score of ≥ 12 s, 4-m usual gait speed of < 1·0 m/s and/or short physical performance battery score of ≤ 9. P values were obtained using a binary logistic regression model. We also analysed the association of the anthropometric indices with sarcopenic obesity by OR per sd increase (online Supplementary Table S2). The highest OR were identified in WWI in men, whereas no anthropometric indices were significantly associated with sarcopenic obesity in women. We observed an independent association between the anthropometric indices and sarcopenia and obesity (online Supplementary Table S3) and found that WWI was the only index that was positively correlated with sarcopenia in men; thus, the coexistence of sarcopenia and obesity led to an increased association with WWI. However, this association was not observed in women due to the lack of a correlation between WWI and sarcopenia. ## The discriminative ability of the anthropometric indices for predicting sarcopenic obesity The discriminative ability of the anthropometric indices for predicting sarcopenic obesity was determined using the age-adjusted receiver operating characteristic curves for men and women (Fig. 2), and the AUC of the anthropometric indices for the diagnosis of sarcopenic obesity were obtained (online Supplementary Table S4). Although the values were relatively modest, the highest AUC were observed for the WWI (AUC: 0·692, 95 % CI: 0·653, 0·731 in criterion 1; AUC: 0·799, 95 % CI: 0·755, 0·842 in criterion 2 and AUC: 0·781, 95 % CI: 0·738, 0·824 in criterion 3) in men; the WWI had the best diagnostic power for sarcopenic obesity followed by WHtR, WC and BMI according to all the criteria (all P for difference in AUC < 0·05). In women, the diagnostic power of anthropometric indices could not be determined because the difference between the AUC of anthropometric indices was not significant according to all the criteria. We obtained similar results for both men and women in the unadjusted model (online Supplementary Fig. S1 and Supplementary Table S5). Fig. 2.Age-adjusted ROC curves for sarcopenic obesity according to the anthropometric indices. ROC, receiver operating characteristic; WWI, weight-adjusted waist index; WC, waist circumference; WHtR, waist-to-height ratio. Criterion 1: high-fat mass + low muscle mass; Criterion 2: high-fat mass + low muscle mass + low muscle strength and/or slow gait speed; Criterion 3: high-fat mass + low muscle mass + low muscle strength and/or low physical performance. High-fat mass: body fat percentage of ≥ 28·2 % for men and ≥ 38·8 % for women; low muscle mass: appendicular skeletal muscle mass/height2 of < 7·00 kg/m2 for men and < 5·40 kg/m2 for women; low muscle strength: grip strength of < 28 kg for men and < 18 kg for women; slow gait speed: 4-m usual gait speed of < 1·0 m/s and low physical performance: five-times sit-to-stand test score of ≥ 12 s, 4-m usual gait speed of < 1·0 m/s and/or short physical performance battery score of ≤ 9. The optimal cut-off values, sensitivity and specificity of the anthropometric indices for the diagnosis of sarcopenic obesity in men were also reported, according to Youden’s index (online Supplementary Table S6). ## Correlation between the anthropometric indices and components of sarcopenic obesity The correlation between the anthropometric indices and each diagnostic component of sarcopenic obesity is shown in Table 3. All anthropometric indices were positively correlated with the percentage of body fat in both men and women. WC showed the strongest correlation ($r = 0$·695, $P \leq 0$·001) in men followed by WHtR ($r = 0$·689, $P \leq 0$·001), BMI ($r = 0$·652, $P \leq 0$·001) and WWI ($r = 0$·480, $P \leq 0$·001). In women, BMI showed the strongest correlation with the percentage of body fat ($r = 0$·662, $P \leq 0$·001) followed by WC ($r = 0$·542, $P \leq 0$·001), WHtR ($r = 0$·499, $P \leq 0$·001) and WWI ($r = 0$·146, $P \leq 0$·001). WWI had the lowest correlation coefficient for percentage of body fat in both men and women. However, WWI was the only index that showed a negative correlation with ASM/height2 (r = –0·073, $$P \leq 0$$·020) in men, which confirmed that the WWI discriminated muscle mass from fat mass. Other indices did not discriminate muscle mass from fat mass in men; BMI ($r = 0$·602, $P \leq 0$·001), WC ($r = 0$·351, $P \leq 0$·001) and WHtR ($r = 0$·339, $P \leq 0$·001) all showed positive correlation with ASM/height2. In women, all the anthropometric indices did not discriminate muscle mass from fat mass; they all showed positive correlation with ASM/height2 (BMI ($r = 0$·483, $P \leq 0$·001); WC ($r = 0$·348, $P \leq 0$·001); WHtR ($r = 0$·358, $P \leq 0$·001) and WWI ($r = 0$·102, $$P \leq 0$$·001)). Table 3.Correlation between the anthropometric indices and diagnostic components of sarcopenic obesity(Coefficients and 95 % confidence intervals)WWIBMIWCWHtR r 95 % CI P r 95 % CI P r 95 % CI P r 95 % CI P Men n 1032 Percentage of body fat* 0·4800·431, 0·526< 0·0010·6520·616, 0·686< 0·0010·6950·662, 0·725< 0·0010·6890·656, 0·720< 0·001 ASM/height2* –0·073–0·133, –0·0120·0200·6020·562, 0·639< 0·0010·3510·297, 0·404< 0·0010·3390·284, 0·392< 0·001 Maximum grip strength* –0·245–0·302, –0·187< 0·0010·1820·123, 0·241< 0·0010·1050·044, 0·1650·001–0·026–0·086, 0·0350·411 4-m usual gait speed* –0·165–0·224, –0·105< 0·001–0·009–0·070, 0·0520·773–0·058–0·119, 0·0030·061–0·101–0·161, –0·0400·001 Five-times-sit-to-stand test† 0·1780·118, 0·236< 0·001–0·031–0·092, 0·0300·3220·0760·015, 0·1360·0150·0970·036, 0·1570·002 SPPB score† –0·138–0·197, –0·077< 0·0010·0770·016, 0·1370·014–0·027–0·088, 0·0340·392–0·048–0·109, 0·0130·124Women n 999 Percentage of body fat* 0·1460·085, 0·206< 0·0010·6620·626, 0·696< 0·0010·5420·497, 0·585< 0·0010·4990·451, 0·544< 0·001 ASM/height2* 0·1020·041, 0·1630·0010·4830·434, 0·529< 0·0010·3480·292, 0·401< 0·0010·3580·303, 0·411< 0·001 Maximum grip strength* –0·214–0·272, –0·154< 0·0010·0700·008, 0·1310·0280·036–0·027, 0·0970·262–0·096–0·157, –0·0340·002 4-m usual gait speed* –0·213–0·272, –0·153< 0·001–0·076–0·137, –0·0140·016–0·111–0·172, –0·049< 0·001–0·184–0·243, –0·123< 0·001 Five-times-sit-to-stand test† 0·1310·070, 0·192< 0·0010·1080·050, 0·1690·0010·1640·103, 0·223< 0·0010·1550·094, 0·215< 0·001 SPPB score† –0·166–0·225, –0·105< 0·001–0·084–0·145, –0·0220·008–0·148–0·209, –0·087< 0·001–0·158–0·218, –0·097< 0·001WWI, weight-adjusted waist index; WC, waist circumference; WHtR, waist-to-height ratio; ASM, appendicular skeletal muscle mass; SPPB, short physical performance battery.Pearson’s correlation coefficient.†Spearman’s correlation coefficient. The overall correlations between anthropometric indices and physical function measurements were weak; yet, WWI was the only index that showed a significant inverse relationship with physical function measurements in both men and women. Higher WWI was associated with lower maximum grip strength (r = −0·245, $P \leq 0$·001 in men; r = −0·215, $P \leq 0$·001 in women), 4-m usual gait speed (r = −0·165, $P \leq 0$·001 in men; r = −0·213, $P \leq 0$·001 in women) and SPPB score (r = −0·138, $P \leq 0$·001 in men; r = −0·166, $P \leq 0$·001 in women) and a longer time for the five-times sit-to-stand test ($r = 0$·178, $P \leq 0$·001 in men; $r = 0$·131, $P \leq 0$·001 in women). Such consistency was not observed in BMI and WC. WHtR showed inverse relationship with physical function measurements, similar to WWI; however, in men, the correlations of WHtR with maximum grip strength (r = −0·026, $$P \leq 0$$·411) and SPPB score (r = −0·048, $$P \leq 0$$·124) were not statistically significant. ## Discussion This is the first study that analysed the association of different anthropometric indices, including BMI, WC, WHtR and WWI, with sarcopenic obesity to test their feasibility as screening tool for sarcopenic obesity. In this study, WWI showed the strongest association with sarcopenic obesity in men and was the best screening tool compared with WHtR, WC and BMI. There was no statistical significance between all four anthropometric indices and sarcopenic obesity in women. Our findings also reported that WWI was the only index that discriminated muscle mass and fat mass in men, while all the anthropometric indices did not in women. Furthermore, WWI was the only index that showed significant inverse association with physical function in both men and women. Taken altogether, WWI has potential to be a simple screening tool for sarcopenic obesity in older men. In our study, BMI reported similar positive correlations with fat mass and muscle mass in both sexes; we confirmed that BMI’s inability to discriminate fat mass and muscle mass makes it inappropriate to screen sarcopenic obesity. WC is a measure of abdominal obesity that is highly associated with visceral fat[44,45]. Although WC showed better association with sarcopenic obesity along with reduced positive correlation with muscle mass compared with BMI, it was still too far to say that WC discriminated muscle mass and fat mass. Nevertheless, BMI and WC were good obesity indicators in line with the previous studies[46,47]; WC reported the best correlation with body fat mass in men, while BMI was the best in women. Therefore, we confirmed that BMI and WC cannot solely be used for screening sarcopenic obesity; instead, they have to be combined with physical function indicators[48,49], which may also reflect low muscle mass. WHtR showed significantly better performance as an indicator of sarcopenic obesity in men compared with BMI and WC possibly due its better reflection of body fat distribution by standardising WC for height. However, it was not free from the influence of WC and similar limitations were observed; it still did not discriminate muscle mass and fat mass and was weakly correlated with physical function measurements, especially with muscle strength. Considering the significant association of WHtR and sarcopenic obesity based on all the three criteria in men, WHtR can be a useful indicator in men when combined with appropriate muscle strength indicator, such as grip strength. In 2021, the WWI was suggested as an indicator that can reflect high fat mass and low muscle mass simultaneously, although the association of WWI with muscle strength and physical performance was not identified[24]. Unlike WHtR, WWI standardised WC for weight only and differentiated the effect of height on the same waist[23]. In our study, WWI showed a relatively lower correlation with body fat mass compared with the other anthropometric indices for both sexes. However, in men, WWI reflected fat mass and muscle mass in the opposite direction and showed a significant inverse association with physical function. It was the only index that showed a significant positive association with sarcopenia defined as low muscle mass combined with low muscle strength and/or physical performance in men. Consequently, the coexistence of sarcopenia and obesity reported an increased association with the WWI, making it the best index to screen sarcopenic obesity in men compared with the other anthropometric indices. The overall correlations of WWI with muscle mass and each physical function measurement were not strong; still, it was the only index that reflected the components simultaneously while maintaining statistical significance, which led to a better association between WWI and sarcopenic obesity in men compared with the other indices. Meanwhile, WWI did not reflect muscle mass and fat mass in the opposite direction in women as it did in men; it showed the lowest correlation with fat mass compared with the other indices. As a result, WWI was not significantly associated with sarcopenic obesity in women even it showed a significant inverse association with physical function. The significant sex-specific difference was observed in WWI, WC and WHtR in the association with sarcopenic obesity. Previous studies have shown that men have significantly higher visceral fat and lower extremity fat than women[50,51], while women have higher subcutaneous fat and greater fat infiltration into lower extremity muscles than men[52,53]. The sex-specific difference of WWI, WC and WHtR may be attributed to the insufficient reflection of subcutaneous or intermuscular fat. In the previous Health, Aging, and Body Composition Study, a higher amount of subcutaneous fat in women’s lower extremity area was independently associated with slow gait speed[54]. We found similar sex-specific difference in mean gait speed in our study population as the proportion of slow gait speed was higher in women. This finding may reflect a higher amount of subcutaneous fat deposited in lower extremity in our women population. In the most recent Multi-Ethnic Study of Atherosclerosis that examined the association between WWI and abdominal fat and muscle mass by CT scans, WWI was not only positively correlated with visceral fat area but also with subcutaneous fat of abdominal area, while negatively correlated with abdominal muscle area[55]. Overall, previous findings suggest WC-driven anthropometric indices may not well reflect lower extremity fat that could have interrupted reciprocal assessment of percentage of body fat and ASM/height2 in older women in our study. We were not able to confirm this assumption due to the lack of relevant data; further research into the sex-specific differences is needed, especially regarding relationship between the anthropometric indices and distribution of body fat. In addition, a previous study of Korean community-dwelling older adults found that ASM/height2 was the most reliable index for sarcopenia in men in terms of predicting functional limitation, while ASM/weight was better in women[56]. In light of this, we may need a different approach; applying different muscle indices in our study population could have yielded different results. Although we applied ASM/height2 in our study according to the Asian Working Group for Sarcopenia 2019 consensus, studies regarding the application of ASM/weight would be intriguing, and further studies on an anthropometric index for women are warranted. Our study had some limitations. First, this study included only Korean participants; thus, our findings cannot be generalised to other populations. A multi-ethnic study is required to confirm our findings, especially focusing on physical function. Second, although we calculated the optimal cut-off values of anthropometric indices for the diagnosis of sarcopenic obesity, we could not validate these findings with other scientific evidence as the attempt to apply anthropometric indices to diagnose sarcopenic obesity has not been utilised broadly in the clinical field. It is necessary to establish an appropriate cut-off value for the diagnosis of sarcopenic obesity in clinical practice. Third, the age range of our study was set from 70 to 84 years. We were not able to provide a result for older adults aged 85 years or older; the recruitment of participants with higher ages can be more valuable for this study as the prevalence of sarcopenic obesity increases with age. Finally, this study had a cross-sectional design and causal relationships could not be established; longitudinal studies are required to identify the causal relationship between the anthropometric indices and sarcopenic obesity. In conclusion, WWI showed the strongest association with three different definitions of sarcopenic obesity and was the best screening tool compared with WHtR, WC and BMI in older men. No anthropometric indices were associated with any definition of sarcopenic obesity in older women. WWI was the only index that was negatively correlated with physical function in both sexes. It has the potential to be utilised as a simple screening tool in clinical practice; however, further research into the sex-specific differences is warranted. ## References 1. Batsis JA, Villareal DT. Sarcopenic obesity in older adults: aetiology, epidemiology and treatment strategies. Nat Rev Endocrinol* (2018) 14 513-537. PMID: 30065268 2. Rolland Y, Lauwers-Cances V, Cristini C. Difficulties with physical function associated with obesity, sarcopenia, and sarcopenic-obesity in community-dwelling elderly women: the EPIDOS (EPIDemiologie de l’OSteoporose) Study. Am J Clin Nutr (2009) 89 1895-1900. PMID: 19369381 3. Wannamethee SG, Shaper AG, Lennon L. Decreased muscle mass and increased central adiposity are independently related to mortality in older men. Am J Clin Nutr (2007) 86 1339-1346. PMID: 17991644 4. Zamboni M, Mazzali G, Fantin F. Sarcopenic obesity: a new category of obesity in the elderly. Nutr Metab Cardiovasc Dis (2008) 18 388-395. PMID: 18395429 5. Baumgartner RN. Body composition in healthy aging. Ann N Y Acad Sci (2000) 904 437-448. PMID: 10865787 6. Baumgartner RN, Koehler KM, Gallagher D. Epidemiology of sarcopenia among the elderly in New Mexico. Am J Epidemiol (1998) 147 755-763. PMID: 9554417 7. Buckinx F, Landi F, Cesari M. Pitfalls in the measurement of muscle mass: a need for a reference standard. J Cachexia Sarcopenia Muscle (2018) 9 269-278. PMID: 29349935 8. Choi KM. Sarcopenia and sarcopenic obesity. Korean J Intern Med (2016) 31 1054-1060. PMID: 27809450 9. Physical status: the use and interpretation of anthropometry. Report of a WHO Expert Committee. World Health Organ Tech Rep Ser (1995) 854 1-452. PMID: 8594834 10. Browning LM, Hsieh SD, Ashwell M. A systematic review of waist-to-height ratio as a screening tool for the prediction of cardiovascular disease and diabetes: 0·5 could be a suitable global boundary value. Nutr Res Rev (2010) 23 247-269. PMID: 20819243 11. Cornier MA, Despres JP, Davis N. Assessing adiposity: a scientific statement from the American Heart Association. Circulation (2011) 124 1996-2019. PMID: 21947291 12. Nguyen NT, Magno CP, Lane KT. Association of hypertension, diabetes, dyslipidemia, and metabolic syndrome with obesity: findings from the National Health and Nutrition Examination Survey, 1999 to 2004. J Am Coll Surg (2008) 207 928-934. PMID: 19183541 13. Wilson PW, D’Agostino RB, Sullivan L. Overweight and obesity as determinants of cardiovascular risk: the Framingham experience. Arch Intern Med (2002) 162 1867-1872. PMID: 12196085 14. Hainer V, Aldhoon-Hainerova I. Obesity paradox does exist. Diabetes Care (2013) 36 S276-S281. PMID: 23882059 15. Kim NH, Lee J, Kim TJ. Body Mass Index and Mortality in the general population and in subjects with chronic disease in Korea: a nationwide cohort study (2002–2010). PLOS ONE (2015) 10 e0139924. PMID: 26462235 16. Uretsky S, Messerli FH, Bangalore S. Obesity paradox in patients with hypertension and coronary artery disease. Am J Med (2007) 120 863-870. PMID: 17904457 17. Jackson AS, Stanforth PR, Gagnon J. The effect of sex, age and race on estimating percentage body fat from body mass index: the Heritage Family Study. Int J Obes Relat Metab Disord (2002) 26 789-796. PMID: 12037649 18. Okorodudu DO, Jumean MF, Montori VM. Diagnostic performance of body mass index to identify obesity as defined by body adiposity: a systematic review and meta-analysis. Int J Obes (2010) 34 791-799 19. Kuk JL, Katzmarzyk PT, Nichaman MZ. Visceral fat is an independent predictor of all-cause mortality in men. Obesity (2006) 14 336-341. PMID: 16571861 20. Zhang C, Rexrode KM, van Dam RM. Abdominal obesity and the risk of all-cause, cardiovascular, and cancer mortality: sixteen years of follow-up in US women. Circulation (2008) 117 1658-1667. PMID: 18362231 21. Song X, Jousilahti P, Stehouwer CDA. Cardiovascular and all-cause mortality in relation to various anthropometric measures of obesity in Europeans. Nutrition, Metab Cardiovasc Dis (2015) 25 295-304 22. Christakoudi S, Tsilidis KK, Muller DC. A Body Shape Index (ABSI) achieves better mortality risk stratification than alternative indices of abdominal obesity: results from a large European cohort. Sci Rep (2020) 10 14541. PMID: 32883969 23. Park Y, Kim NH, Kwon TY. A novel adiposity index as an integrated predictor of cardiometabolic disease morbidity and mortality. Sci Rep (2018) 8 16753. PMID: 30425288 24. Kim NH, Park Y, Kim NH. Weight-adjusted waist index reflects fat and muscle mass in the opposite direction in older adults. Age Ageing (2021) 50 780-786. PMID: 33035293 25. Chen LK, Woo J, Assantachai P. Asian working group for sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc (2020) 21 300-307.e302. PMID: 32033882 26. Cruz-Jentoft AJ, Bahat G, Bauer J. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing (2019) 48 16-31. PMID: 30312372 27. Fielding RA, Vellas B, Evans WJ. Sarcopenia: an undiagnosed condition in older adults. Current consensus definition: prevalence, etiology, and consequences. International working group on sarcopenia. J Am Med Dir Assoc (2011) 12 249-256. PMID: 21527165 28. Studenski SA, Peters KW, Alley DE. The FNIH sarcopenia project: rationale, study description, conference recommendations, and final estimates. J Gerontol A Biol Sci Med Sci (2014) 69 547-558. PMID: 24737557 29. Ling CHY, Meskers CGM, Maier AB. Can anthropometric measures be used as proxies for body composition and physical function in geriatric outpatients?. Arch Gerontol Geriatr (2021) 94 104379. PMID: 33610124 30. Woo J, Leung J, Kwok T. BMI, body composition, and physical functioning in older adults. Obesity (2007) 15 1886-1894. PMID: 17636108 31. Morley JE, Vellas B, van Kan GA. Frailty consensus: a call to action. J Am Med Dir Assoc (2013) 14 392-397. PMID: 23764209 32. Kim M, Shinkai S, Murayama H. Comparison of segmental multifrequency bioelectrical impedance analysis with dual-energy X-ray absorptiometry for the assessment of body composition in a community-dwelling older population. Geriatr Gerontol Int (2015) 15 1013-1022. PMID: 25345548 33. Giangregorio LM, ebber CE. Effects of metal implants on whole-body dual-energy x-ray absorptiometry measurements of bone mineral content and body composition. Can Assoc Radiol J (2003) 54 305-309. PMID: 14689806 34. Won CW, Lee S, Kim J. Korean frailty and aging cohort study (KFACS): cohort profile. BMJ Open (2020) 10 e035573 35. Ashwell M, Gunn P, Gibson S. Waist-to-height ratio is a better screening tool than waist circumference and BMI for adult cardiometabolic risk factors: systematic review and meta-analysis. Obes Rev (2012) 13 275-286. PMID: 22106927 36. Kim M, Won CW, Kim M. Muscular grip strength normative values for a Korean population from the Korea National Health and Nutrition Examination Survey, 2014–2015. PLoS One (2018) 13 e0201275. PMID: 30125289 37. Kim M, Won CW. Combinations of gait speed testing protocols (automatic. Arch Gerontol Geriatr (2019) 81 215-221. PMID: 30623866 38. Guralnik JM, Simonsick EM, Ferrucci L. A short physical performance battery assessing lower extremity function: association with self-reported disability and prediction of mortality and nursing home admission. J Gerontol (1994) 49 M85-M94. PMID: 8126356 39. Guralnik JM, Ferrucci L, Pieper CF. Lower extremity function and subsequent disability: consistency across studies, predictive models, and value of gait speed alone compared with the short physical performance battery. J Gerontol Ser A, Biol Sci Med Sci (2000) 55 M221-231. PMID: 10811152 40. Ho-Pham LT, Campbell LV, Nguyen TV. More on body fat cutoff points. Mayo Clin Proc (2011) 86 584. PMID: 21628621 41. Baumgartner RN, Wayne SJ, Waters DL. Sarcopenic obesity predicts instrumental activities of daily living disability in the elderly. Obes Res (2004) 12 1995-2004. PMID: 15687401 42. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biom (1988) 44 837-845 43. Peduzzi P, Concato J, Kemper E. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol (1996) 49 1373-1379. PMID: 8970487 44. Koster A, Leitzmann MF, Schatzkin A. Waist circumference and mortality. Am J Epidemiol (2008) 167 1465-1475. PMID: 18417494 45. Snijder MB, van Dam RM, Visser M. What aspects of body fat are particularly hazardous and how do we measure them?. Int J Epidemiol (2006) 35 83-92. PMID: 16339600 46. Flegal KM, Shepherd JA, Looker AC. Comparisons of percentage body fat, body mass index, waist circumference, and waist-stature ratio in adults. Am J Clin Nutr (2009) 89 500-508. PMID: 19116329 47. Wang C, Hou XH, Zhang ML. Comparison of body mass index with body fat percentage in the evaluation of obesity in Chinese. Biomed Environ Sci (2010) 23 173-179. PMID: 20708495 48. Schrager MA, Metter EJ, Simonsick E. Sarcopenic obesity and inflammation in the InCHIANTI study. J Appl Physiol (2007) 102 919-925. PMID: 17095641 49. Stenholm S, Alley D, Bandinelli S. The effect of obesity combined with low muscle strength on decline in mobility in older persons: results from the InCHIANTI study. Int J Obes (2009) 33 635-644 50. Ley CJ, Lees B, Stevenson JC. Sex- and menopause-associated changes in body-fat distribution. Am J Clin Nutr (1992) 55 950-954. PMID: 1570802 51. Schorr M, Dichtel LE, Gerweck AV. Sex differences in body composition and association with cardiometabolic risk. Biol Sex Differ (2018) 9 28. PMID: 29950175 52. Goodpaster BH, Carlson CL, Visser M. Attenuation of skeletal muscle and strength in the elderly: the Health ABC Study. J Appl Physiol (2001) 90 2157-2165. PMID: 11356778 53. Visser M, Goodpaster BH, Kritchevsky SB. Muscle mass, muscle strength, and muscle fat infiltration as predictors of incident mobility limitations in well-functioning older persons. J Gerontol A Biol Sci Med Sci (2005) 60 324-333. PMID: 15860469 54. Beavers KM, Beavers DP, Houston DK. Associations between body composition and gait-speed decline: results from the Health, Aging, and Body Composition study. Am J Clin Nutr (2013) 97 552-560. PMID: 23364001 55. Kim JY, Choi J, Vella CA. Associations between Weight-adjusted waist index and abdominal fat and muscle mass: multi-ethnic study of atherosclerosis. Diabetes Metab J (2022) 56. Kim JH, Choi SH, Lim S. Sarcopenia and obesity: gender-different relationship with functional limitation in older persons. J Korean Med Sci (2013) 28 1041-1047. PMID: 23853487
--- title: 'SATO (IDEAS expAnded wiTh BCIO): Workflow for designers of patient-centered mobile health behaviour change intervention applications' authors: - Aneta Lisowska - Szymon Wilk - Mor Peleg journal: Journal of Biomedical Informatics year: 2023 pmcid: PMC9975785 doi: 10.1016/j.jbi.2022.104276 license: CC BY 4.0 --- # SATO (IDEAS expAnded wiTh BCIO): Workflow for designers of patient-centered mobile health behaviour change intervention applications ## Abstract Designing effective theory-driven digital behaviour change interventions (DBCI) is a challenging task. To ease the design process, and assist with knowledge sharing and evaluation of the DBCI, we propose the SATO (IDEAS expAnded wiTh BCIO) design workflow based on the IDEAS (Integrate, Design, Assess, and Share) framework and aligned with the Behaviour Change Intervention Ontology (BCIO). BCIO is a structural representation of the knowledge in behaviour change domain supporting evaluation of behaviour change interventions (BCIs) but it is not straightforward to utilise it during DBCI design. IDEAS (Integrate, Design, Assess, and Share) framework guides multi-disciplinary teams through the mobile health (mHealth) application development life-cycle but it is not aligned with BCIO entities. SATO couples BCIO entities with workflow steps and extends IDEAS Integrate stage with consideration of customisation and personalisation. We provide a checklist of the activities that should be performed during intervention planning with concrete examples and a tutorial accompanied with case studies from the Cancer Better Life Experience (CAPABLE) European project. In the process of creating this workflow, we found the necessity to extend the BCIO to support the scenarios of multiple clinical goals in the same application. To ensure the SATO steps are easy to follow for the incomers to the field, we performed a preliminary evaluation of the workflow with two knowledge engineers, working on novel mHealth app design tasks. ## Graphical abstract ## Introduction The amount of literature on digital health interventions (DHI) [1] is vast [2], [3], with studies aiming to improve mental [4] and physical health [5]. Interventions cover applications from different stages of the health management cycle and target populations across different ages, social status and cultures. Some DHIs aim to support patients with adherence to pharmacological treatment, dose management and side effects reporting. Other DHIs aid patients with modulation of health risk behaviours such as: inactivity, poor nutrition, or substance abuse. The later are called digital behaviour change interventions (DBCI) and their design, or more specifically design of patient-centered mobile health (mHealth) behaviour change intervention (BCI) applications, is the focus of this paper. Michie and colleagues have been working for many years on creating a standardisation of the behaviour change domain. The evaluation of BCIs and comparison of theories, studies, and trials that incorporate behaviour change is not feasible without a standard vocabulary and ontology. Such a structure organising the basic techniques that are implemented in BCIs is needed in order to attribute the success of behavioural change to particular behavioural techniques. Michie’s group developed taxonomy of behavioural change techniques (BCT) [6] and the Behaviour Change Intervention Ontology (BCIO) [7]. BCIO is a very useful tool supporting evaluation of interventions’ effectiveness, but it is relatively new and there are only few examples available on how to utilise it in mHealth application development. The previously provided examples do not cover interventions with multiple wellbeing goals and target behaviours. We found that it is not trivial to apply BCIO to that setting. Therefore, in this work we extend BCIO to mHealth applications with multiple wellbeing goals1 and provide examples of how to use it during the application development process. Our work aims to facilitate researchers with utilising BCIO for the application design in the future and ultimately ease the interventions’ effectiveness evaluation. IDEAS (Integrate, Design, Assess, and Share) [8] is a framework and toolkit of strategies for the development of more effective digital interventions to change health behaviour, integrating methods from behavioural theory, design thinking, and intervention evaluation and dissemination into the full life-cycle of app development. It is popularly used in the development of DBCI and provides a simple and easy to follow checklist of activities that should be performed during DBCI development that might be more accessible to the incomers in the field than the potential more comprehensive design guide provided by Michieet al. [ 9]. However it was developed in 2016 prior to release of BCIO, therefore, it does not align with it. Rather than discarding the IDEAS framework we show that the Integrate stage could be aligned with BCIO (See Fig. 1). We also extend IDEAS with the tailoring step, which has been previously shown to be a crucial component of effective interventions and mentioned in the Michie et al. [ 10] recommendations. This paper intends to provide a detailed actionable systematic workflow for mobile behaviour change application design drawing from IDEAS [8] and in alignment with the BCIO [7] and behaviour change techniques [6]. The SATO (IDEAS expAnded wiTh BCIO) includes a design workflow and checklist (See Fig. 3). We provide examples taken from the Horizon 2020 project Cancer Patient Better Life Experience (CAPABLE, https://capable-project.eu/) [11], where we follow a multi-stakeholder and evidence-based iterative development cycle for a DBCI, in teams of informaticians, clinicians, patients, and engineers, and also utilise feasibility studies on synthetic and public datasets. The main contributions of this research are: [1] application and extension of the BCIO to multi-BCIs mHealth applications that span several clinical goals, [2] DBCI design workflow and checklist validated with examples from the CAPABLE project, [3] reusable design templates bundling together several behaviour change techniques, and [4] tutorial on how to apply SATO to the DBCI application design. Fig. 1SATO (IDEAS expAnded wiTh BCIO). We expand the Integrate stage of IDEAS and align it with BCIO. Terms from BCIO are marked in blue and our extensions to BCIO and IDEAS are marked in pink. ## Related work We review available DBCI design frameworks and BCIO upon which we base our proposed DBCI development workflow. ## DBCI design frameworks Michie et al. [ 10] provides recommendations for developing and evaluating DBCIs falling into six general themes: Achieving rapid and efficient development, Understanding and promoting engagement, Advancing models and theories, Evaluating effectiveness, Evaluating cost-effectiveness and Ensuring regulatory, ethical, and information governance. This is a useful starting point for the DBCI developers highlighting all aspects of intervention design, however it lacks demonstration of how the recommendations are applied to a concrete DBCI. Miller et al. created a framework for Analyzing and Measuring Usage and Engagement Data (AMUsED) in Digital Interventions [12] which focuses on analysis of participant’s effective engagement with the intervention. The accompanying checklist when used during the design stage may help identify which data need to be captured to assess the links between BCTs and behaviour. The authors provide case studies of applying the AMUsED framework to two web-based interventions and suggest that the framework is flexible to fit with interventions delivered with other digital technologies. Although the AMUsED framework is very helpful, especially at the intervention evaluation stages, a broader framework is needed to guide through the full digital behaviour change intervention design process, including content generation and tailoring. The IDEAS [8] framework guides multi-disciplinary teams through the full mHealth application development life-cycle, and provides a disciplined way to incrementally translate behavioural theories into highly relevant and practical interventions. Its ten steps are organised into a four-phase process: Integrate phase, including [1] empathise with target users, [2] specify target behaviour, [3] ground in behavioural theory; DEsign phase, including [4] ideate implementation strategies, [5] prototype potential products, [6] gather user feedback, [7] build a minimum viable product; Assess phase, including [8] pilot test to assess potential efficacy and usability, and [9] evaluation of the efficacy in an RCT; and Share phase, with [10] share intervention and findings. We previously extended IDEAS with an ontology [13] that structures the target behaviour change intervention as a class derived from the HL7 Fast Healthcare Interoperability Resources (FHIR) standard [14]. We also demonstrated application of the proposed extension to a case study taken from the CAPABLE project, that used Fogg’s Tiny Habits behavioural model [15] to improve the sleep of cancer patients via Tai Chi, delivered via an mHealth app. In another work [16], we extended IDEAS’ Ideate substep of the Design phase, by providing concrete backend architectural components and graphical user-interface designs that implemented behavioural interventions. ## BCIO Ontology is an organisation and representation of the entities in a domain according to their properties and relations to one another [17]. Unlike theories, which provide explanations, onotologies are structural representation of the knowledge [18]. Michie et al. [ 7] designed the Behaviour Change Intervention Ontology (BCIO) following the principles of the Open Biological and Biomedical Ontology (OBO) Foundry (https://obofoundry.org) by extending the Basic Formal Ontology [19]. There are six main classes in the BCIO. The [1] BCI Scenario class, in which a [2] BCI, developed for a [3] Context (i.e., target population and setting), is exposed to the population via an [4] Exposure (including the Engagement and Reach activity), and it uses a [5] Mechanism of Action to yield an [6] Outcome Behaviour, which is the intended new behaviour that should form a habit. The Scenario’s Outcome Behaviour can be estimated in a clinical study. The OWL implementation of the BCIO is a work in progress and currently, the Exposure class is not fully developed and only one of its two parts – the BCI Engagement – is defined. Therefore, when using the OWL ontology, we provide an example of Engagement and not of the entire Exposure with its Reach component (see Fig. 2). Fig. 2Introducing CAPABLE Project Case Study: “Fatigue reduction scenario” and BCIO terminology. Table 1BCIO [7] examples for the Fatigue Reduction Scenario. Fig. 3SATO workflow for development of DBCI apps. ## DBCI design steps An important contribution of this study is the formulation of the SATO (IDEAS expAnded wiTh BCIO) workflow for the development of DBCI apps that create desirable behaviour change. The workflow consist of four main steps (See Fig. 3) that facilitate designers in addressing the following four big questions: 1.What is the clinical problem?2.What changes in behaviour will support resolution of the problem?3.How can we facilitate patients with behaviour change?4.How can we adjust the support to meet the needs of individual patients? The SATO steps follow BCIO terminology to facilitate utilisation of BCIO in mHealth research and knowledge sharing. SATO is theory-independent, but via BCIO it can support mHealth app design, based on any behavioural theory that the development team selects. To ease understanding of the SATO workflow, we first explain each SATO step and later, in a blue box, provide concrete examples from the CAPABLE project. We consider the design of a mHealth application that provides multiple clinical goals and for each goal — multiple BCIs. Based on our experience of implementing several clinical goals and BCIs as part of the CAPABLE project, we suggest that each BCI Scenario has a single clinical goal but each app has multiple goals (e.g., to prevent stress, to treat fatigue). Our running example is a BCI Scenario in which the clinical goal “to-treat fatigue”. At the BCI Scenario level, the desired Intervention Outcome is no clinically significant fatigue, as measured by the standard Fatigue Severity Scale [20] and the BCI’s Behaviour Outcome is practice of Tai Chi (Fig. 4). Fig. 4BCI Scenario, BCI, and BCT examples from the CAPABLE project. ## [S1] Describe BCI Scenario — What is the clinical problem? Following the IDEAS framework [8], we recommend that the DBCI will be developed by a multi-stakeholder development team consisting of the clinicians whose patients will use the DBCI, the patients themselves, and the technical developers (engineers, informaticians), which would meet regularly during the DBCI development phase. The first step that this team does is defining clinical problem through answering the following questions: 1.Who are the users?2.What are their goals?3.How can we measure that they met the goals? ## [S1.1] Understand the Context Through patient interviews and creation of user personas [21] the DBCI design team captures the information about the BCI Scenario’s patient population and their setting (see Fig. 5). It is important to consider the intervention setting of target users because it determines the possible mode of delivery of the BCI and influences the barriers to engagement. BCIO specifies by both social and physical setting. Fig. 5Case Study: BCI Context, (a) BCI population (b) BCI setting including BCI social and BCI physical setting. Fig. 6Case Study: Clinical Goals in Capable Project [22], [23], [24], [25], [26], [27]. ## [S1.2] Identify Clinical Goal (extension of BCIO) A crucial step in developing a DBCI is to select the important clinical objective or goal of this DBCI. The clinical goal will impact the choice of intervention and evaluation metrics. To select the clinical goals of the DBCI for the target population, research should be done to establish what wellbeing dimensions [28] are the ones most impacted by the patients’ condition (see Fig. 6), fitting with IDEAS’ Empathise step of the Integrate phase, that integrates insights from users and theory. This step should be led by the clinical researchers and performed via a literature search supplemented by questionnaires applied to the target population. Goal ontologies can be used to standardise the specification of clinical goals. For example, in the Goal-based Comorbidity decision-support method [29], goals specification follows the goal ontology developed by Fox et al. [ 30] that includes a verb and a noun phrase (eg, manage hypertension, prevent cardiovascular disease, treat fatigue), the HL7 FHIR [14] Goal resource, and relationships from the National Drug File - Reference Terminology (NDF-RT) ontology [31], such as may-treat, may-prevent, and has_physiological_effect [increase/decrease State] e.g., Increase_Physical_Activity (NDF-RT Physiological Effect Goal). Alternatively, in the Asbru [27] clinical guideline formalism, process goals (eg, monitor blood pressure) or state goals (eg, normal blood pressure) can be specified as temporal patterns that are meant to be maintained, avoided or achieved (e.g., achieve systolic blood pressure <140 within 1 month of starting antihypertensive medication). As BCIO follows the Basic Formal Ontology (BFO) [19], may-treat and may-prevent goals are represented as functions, and state goals (representing combination of lag measures and restrictions on them) are represented as dispositions. Finally, an Intervention Evaluation Study that evaluates a BCI Scenario (e.g., Fatigue Reduction Scenario) has output of Evaluation Finding (e.g. fatigue as measured by FSS). Related to the BCI, a BCI Evaluation Study that evaluates a BCI (e.g., Tai Chi BCI) has output of Evaluation Finding (e.g. number of times the Tai Chi video was pressed)(see Fig. 7). Fig. 7Case Study: Goal Hierarchy. The names of ontologies relating to clinical goals are shown in parentheses. BCIO concepts are shown in square brackets. ## [S1.3] Define Lag Measures To evaluate intervention effectiveness in supporting patients with reaching their clinical goals, researchers utilise standard patient-reported outcome measures (PROMs) [32] at the commencement and termination of the intervention. Each PROM should be selected to assess the patients’ states in dimension relevant to their clinical goals. Changes in scores on PROMs do not occur rapidly and therefore they are called lag measures [33]. Note that for the BCI Scenario we measure the change in patients’ physiological/emotional state related to their clinical goal and the performance of the target behaviours is considered at the BCI level (see Table 1 and Fig. 8). Fig. 8Case Study: Examples of lag measures [20], [34], [35], [36]. ## [S2] Identify BCIs for the clinical goals — What changes in behaviour will support resolution of the problem? To refine the clinical goals and identify the interventions that can meet the goals, we can turn to clinical practice guidelines. Following the evidence-based medicine (EBM) movement, “clinical practice guidelines that we can trust” are defined as “statements that include recommendations intended to optimise patient care, that are informed by a systematic review of evidence and an assessment of the benefit and harms of alternative care options for a clinical condition” (clinical objective/goal) [37]. Clinical practice guidelines usually address a specific clinical condition. Unfortunately, most of the clinical guidelines refer to medication-based care options, and evidence for non-medication interventions is usually limited. However, non-pharmacological life-style, exercise, and psycho-behavioural interventions are a promising way to care for mental wellbeing, including for example, chronic pain [38] and fatigue [39], and include evidence grades [40] based on cohort studies, and in some cases on randomised controlled trials and meta-analyses, which provide a higher grade of evidence. Even though the non-pharmacological therapies are not widely spread there are some guidelines recommended by EBM such as ESMO fatigue guideline [39] and a back-pain guideline [38]. ## [S2.1] Search for evidence-based intervention Clinical goal(s) is an important extension of the BCIO and searching for evidence-based intervention options that meet it fits the “specify target behaviour” step of the Integrate phase of IDEAS. As mentioned above, clinical practice guidelines, and other clinical sources following the EBM pyramid, are the best resource to search for evidence-based interventions. We reused the BCIO’s BCI_Source property, which originally is a property of a BCT, and linked it to BCI to highlight that intervention choice should be supported by relevant evidence (See Table 1 and Fig. 9). Fig. 9Case Study: Evidence-based BCIs [41]. ## [S2.2] Define Lead Measures To ensure that the BCI effectively supports patients in reaching the target clinical goal, ideally we would be able to check if a patient is on track of reaching their goal, and if not, modify the intervention. However, daily assessment through PROM questionnaires is not feasible long term, especially given that previous studies found that frequent surveys were not perceived favourable by the study participants [42] and could negatively impact engagement with the intervention. Therefore, it is important to identify measures that are related to the outcome but can also be captured frequently and automatically. The examples are provided in Fig. 10. Fig. 10Case Study: Identifying and refining lead measures through pilot studies [43], [44]. ## [S2.3] Select BCI supporting technologies To monitor intervention adherence, it might be helpful to pair the DBCI app with a wearable device. The choice of the device will depend on the population, their clinical goals, the selected BCIs and the lead measures. When selecting a wearable device we suggest to consider not only type of captured data but also the frequency and to test the candidate devices early in the application development cycle (see Fig. 11). Fig. 11Case Study: Supporting technology selection and evaluation [45], [46], [47], [48]. ## [S3] Create BCI Content — How can we facilitate patients with behaviour change? Once the patient population, their context, clinical goals and evidence-based BCIs meeting the goals are identified from evidence-based sources, the development team use agile software development methods [21], including refinement of stereotypical persona and development of user stories, to create a shared understanding of the anticipated user experience with the DBCI. The goal of this step is to diagnose potential barriers to engagement with the intervention and determine which techniques could be used to overcome them. ## [S3.1] Create user stories The creation of user stories facilitate designers with identifying obstacles that patients may face with performing the target behaviour (see examples in Fig. 12). The user stories also form the starting point for app screen mockups, which are later further refined with users’ feedback. Fig. 12Case Study: User Stories. ## [S3.2] Find intervention protocols and source In further iterations of the Ideate step, the knowledge engineers in the team search for specific existing sources, i.e., implementations for the different interventions, in the form of narrative instructions or videos demonstrating behaviour (see examples referenced in Fig. 13). These should be accepted by the clinicians and patients of the multidisciplinary app-design team. Fig. 13Case Study CAPABLE Project: Protocols and Sources [42], [49].Fig. 14Case Stud: BCT Bundle templates. ## [S3.3] Select BCTs Abraham and Michie [6] created a taxonomy of BCTs, which initially included 26 distinct BCTs and later was extended to 93 BCTs [50]. The most recent list includes 74 BCTs and is accompanied by a tool which helps to explore the links between the BCTs and mechanism of action https://theoryandtechniquetool.humanbehaviourchange.org/tool [51]. We suggest that BCTs could be bundled together to create BCI content and applied both on clinical goal and at single BCI levels (see Fig. 14). The BCT bundles organised into GUI designs fit with Ideate and Prototype steps from Design stage of IDEAS framework respectively. ## [S4] Customise and Personalise — How can we adjust the support to meet the needs of each individual? Customisation (or customisability) refers to a creation of a predefined set of options during the design step. The options, for example may include multiple BCIs contributing to the same goal, different levels of activity difficulty for varying levels of patient skill or sets of motivational messages addressing varying patients beliefs and needs. Personalisation on the other hand, is a process of matching the best options to a given patient at run time. This can either be performed by the user or by an algorithms. Personalisation is one of the most commonly used techniques in mobile health interventions [52] and it plays important role in influencing their effectiveness [53]. Tong et al. [ 54] conducted a systematic review of personalised mobile BCIs and highlighted that personalisation might be applied to: BCI_Content (e.g., demonstration video), BCI_Mode_of_delivery (e.g., voice message, game, wearable), BCI_Dose (e.g., number of daily notifications), BCI_Schedule_of_delivery. This means that personalisation considers what information is presented, how, when, and how often. Automatic, data-driven personalisation depends also on the type and source of the collected data, the frequency of data collection, and the personalisation algorithm. These should be also already considered when defining a set of customisation options. Fig. 15Behaviour Change Theories and BCI Tailoring [55], [56], [57], [58], [59], [60], [61], [62]. ## [S4.1] Define a set of customisation options To maximise the impact of education information on patients’ outcome behaviour, it is important that patients perceive the information to be personally relevant. Ghalibaf et al. conducted a systematic review of computer-based health information tailoring and identified six dimensions according to which patients could be characterised, these are: [1] socio-demographic (e.g., age, level of education), [2] medical history (eg, comorbidities), [3] health state (e.g., disease severity), [4] psycho-behavioural determinants (e.g., attitude, self-efficacy) [5] knowledge level, and [6] history of interactions (e.g., visited pages) [63]. In practice, the majority of studies used three or fewer dimensions for user profiling with socio-demographic and psycho-behavioural features being the most popular. The choice of the user categorisation dimensions depends on the BCI Content. Some parts are static (i.e., selected once prior to interventions commencement) for example patient name that is used in reminders; other are dynamic (i.e., change depending on the user interaction with the application) for example the content and phrasing of notification (See Fig. 16). Fig. 16Case Study: Customisation. ## [S4.2] develop personalisation methods The goal of personalisation is to match the best available customisation option to the patient in order to maximise the probability that they perform the target behaviour. According to Fogg’s Behaviour Model (FBM) [64], three factors impact behaviour completion: motivation, ability and trigger. Tailoring of Notification Content and Education Content may increase patients’ motivation, Behaviour Lesson Dose (e.g. length of the thai chi lesson), the user’s ability to perform target behaviour, and schedule of prompt delivery the user’s responsiveness to the notification. The personalisation can be performed manually by the user or automatically by the system. The automatic personalisation algorithms are either knowledge-based or data-based. The former rely on human expertise and most commonly incorporate a set of rules that determine how the system should behave at various conditions. The later rely on data and might utilise machine learning (ML) models (see examples in Fig. 17, Fig. 18). Fig. 17Examples of data-based Content personalisation [65], [66].Fig. 18Case Study: Automatic Personalisation Methods [47], [67], [68]. ## Evaluation In the evaluation of the SATO workflow, we focused on: [1] demonstration of the utilisation of the workflow in the complex multi-BCIs mHealth app design, and on [2] assessment of the workflow’s clarity and its usefulness to the application design process. The former is achieved through CAPABLE app design (Section 4.1) and the latter is achieved through preliminary study with two knowledge engineers who were asked to design an application using the SATO workflow for a novel scenario (Section 4.2). ## SATO validation for multiple BCIs We evaluated the applicability of our workflow and checklist by considering multiple clinical goals/BCI Scenarios, multiple BCIs (Capsules), and BCT Bundles (see Fig. 4). We found that our methods supported design of application for all of the considered scenarios, BCIs and BCT bundles. In Table 2 we summarise goals and interventions for which we defined content as part of the CAPABLE app. In Section 3 and Table 1 we demonstrate examples from the fatigue reduction scenario; however the CAPABLE app targets a wider range of goals with seven different BCIs for which content was developed following the SATO workflow (See A Fig. A.20(b)). Table 2Goals and content defined in CAPABLE app. NamesTotal numberGoals BCI ScenariosFatigue Reduction Sleep Improvement Stress/Anxiety Reduction Mood Improvement Physical Activity Increase5BCIs (Capsules)Deep Breathing, Imagery Training, Tai Chi, Yoga, Garden Bowl, Gratitude Journal, Photo Collage7BCT BundlesEducation_Content Habit_Planning_Content Behaviour_Lesson_Content Review_Content Notification_Content5Table 3Evaluation results. ## SATO clarity evaluation To assess the clarity of the SATO workflow, we asked two knowledge engineers to use it to design a mobile behaviour change application. The engineers first participated in a short tutorial which walked them through the SATO workflow steps and provided examples of the step execution (the same as those presented in blue case study tables). Then they were given two papers [69], [70] serving as EBM sources and a very brief problem statement on the basis of which they were asked to complete each step in the checklist of the SATO workflow. The task was to design a mobile DBCI app for helping educators to prevent burnout (see B). The assessment criteria of the design were created prior to performing the task by the knowledge engineers. The evaluation criteria and results are presented in Table 3. The majority of the steps were performed very well; the steps which proved to be challenging are highlighted in grey and briefly discussed below. Participants struggled the most with understanding of the BCI scenario context. Both participants correctly identified the population, but one participant skipped the description of the setting and the other redefined the population in the description of setting. To clarify this step we added in Section 3.1.1 a suggestion to create user personas to guide definition of both population and their setting. Fig. A.19Mockup screens for education content at the application and individual BCI level. Mock-ups were created by Bitsens UAB, a partner of the CAPABLE Consortium. Fig. A.20(a) Mockup of a goal-setting screen in the physicians’ app that is used during shared decision-making to set up goals for the patient (agree on a behavioural contract and review behavioural goals); (b) Mockup of the patient app showing BCIs. ( c) Mockup of the goal-review screen with feedback on how many of the behaviour goals were achieved compared to the set target. Mock-ups were created by Bitsens UAB, a partner of the CAPABLE Consortium. The other steps which might have been challenging for the participants was selection of supporting technology. Although both participants identified required technology the justification they provided for their selection was brief and did not consider reducing burden of self reporting through automation. At this step it might be already helpful to consider engagement data and the AMUsED framework [12]. The BCI tailoring step might also have been not perfectly clear. Both participants identified customisation options related to the schedule of delivery and content, but only one participant included in her design a customisation option for dose and mode of delivery. The two participants also used different approaches when selecting a personalisation method; one participant selected machine learning for content recommendation and described the data required for training of the model, wheres the other participant described manual personalisation of the schedule of delivery by the user. To improve the understanding of tailoring in Section 3.4 we highlighted in yellow boxes examples from the literature of utilising behaviour change theories in customisation (Fig. 15) and examples of developed personalisation algorithms (Fig. 17). These examples become part of the SATO tutorial (the workflow itself has not been modified). ## Discussion We proposed SATO, a DBCI design workflow aligned with the BCIO and the IDEAS framework, which we evaluated as being comprehensive for designing several BCI Scenarios for an mHealth app for cancer patients, as part of the CAPABLE project. To our knowledge, this is a first DBCI workflow which considers multiple BCIs and uses goal hierarchies with defined evaluation metrics at each level. We also provided examples on utilising behaviour change theories, such as SCT and FBM, when considering customisation and personalisation of the BCI for maximum engagement and adherence. The workflow and accompanying checklist were shown to be easy to follow when evaluated with two knowledge engineers in a novel application design scenario. Although we presented the design steps consecutively, in practice they are iterative (see Fig. 3). For example, at the point of defining the target population and setting, we started creating the user stories for the entire app and later refined them for each BCI Scenario and BCI. Similarly, the lead measures were changed after testing the actual capabilities of the selected smartwatches. Moreover, the knowledge engineers on our team suggested to incorporate a wide range of BCTs; nevertheless some content elements (eg, providing feedback through progress visualisation) were not included in the final app. Specifically, the psychologists who were part of the multi-stakeholder development team raised concern that the ability of cancer patients to perform target behaviour might actually deteriorate with time due to the toxic effects of the cancer therapy and the course of illness; in such cases, visualising progress data might negatively impact their emotional well-being. This example highlights the need for iterative redesign with the end users being continuously kept in mind. In this context SATO aligns very closely with the Gather step of DEsign stage of IDEAS framework. The proposed SATO workflow focuses strongly on the early phase of DBCI development, therefore the Share step from IDEAS framework is not comprehensively addressed. We have however utilised BCIO and extended it to facilitate knowledge sharing. The clinical evaluation study of the CAPABLE app, developed following SATO, with the cancer patients has not yet commenced, hence we could not provide concrete examples of Evaluation Findings. We will address this limitation and also evaluate the suitability of our chosen lead measures in future work, when conducting the clinical study of the usage of the CAPABLE system by cancer patients. The IDEAS framework guides multi-disciplinary teams through mHealth apps development process. BCIO on the other hand captures knowledge in the behaviour change domain and introduces common language for the behaviour change theories aiming to support linking BCIO entities with evidence from behavioural studies. SATO builds on top of both and aims to translate knowledge represented within BCIO into the Integrate and DEsign stage of IDEAS. BCIO entities are directly coupled with the SATO workflow steps to facilitate mHealth app developers with utilising BCIO concepts. The SATO does not aim to replace IDEAS but rather complement its initial stages to ease future knowledge sharing and mHealth app evaluation in context of BCIs effectiveness. ## Conclusions We described a process of designing digital behaviour change intervention which incorporates a range of behaviour change techniques and addresses multiple clinical goals. The step-by-step SATO workflow that we created extends BCIO to support the scenarios with multiple intertwined and hierarchical goals and therefore could be used for design of any non-pharmacological digital behaviour change intervention. We aimed to keep the process simple and provide concrete examples of: technology-independent system captured lead measures, application modules bundling several BCTs, and customisation templates based on behaviour change theories which could be readily reused in other DBCIs. ## CRediT authorship contribution statement Aneta Lisowska: Conceptualization, Methodology, Formal analysis, Investigation, Writing – original draft. Szymon Wilk: Conceptualization, Methodology, Formal analysis, Investigation, Writing – review & editing. Mor Peleg: Conceptualization, Validation, Investigation, Writing – review & editing, Project administration. ## Declaration of Competing Interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Co-author Prof. Mor Peleg is* Editor-in-Chief of Journal of Biomedical Informatics. ## Mockups See Fig. A.19. ## Evaluation task Follow SATO workflow to design mobile behaviour change intervention application for burnout educators. Please use only the SATO workflow, provided burnout papers [69], [70] and https://theoryandtechniquetool.humanbehaviourchange.org/tool Provide a short answer for each check mark in the SATO workflow. Ideally build a similar graph to one shown in the tutorial (see slide 35). Create BCI content for only one of identified BCIs. We are interested to evaluate if the workflow steps are clear and easy to follow not in the full fetched application design. Write down all applications aspects that can be customised and personalised and provide concrete examples how. ## References 1. FDA. (2022.0) 2. Klasnja P., Pratt W.. Healthcare in the pocket: mapping the space of mobile-phone health interventions. J. Biomed. Inform. (2012.0) 45 184-198. PMID: 21925288 3. Silva B.M., Rodrigues J.J., de la Torre Díez I., López-Coronado M., Saleem K.. Mobile-health: A review of current state in 2015. J. Biomed. Inform. (2015.0) 56 265-272. PMID: 26071682 4. Jameel L., Valmaggia L., Barnes G., Cella M.. mHealth technology to assess, monitor and treat daily functioning difficulties in people with severe mental illness: A systematic review. J. Psychiatr. Res. (2021.0) 5. Emberson M.A., Lalande A., Wang D., McDonough D.J., Liu W., Gao Z.. Effectiveness of smartphone-based physical activity interventions on individuals’ health outcomes: A systematic review. BioMed Res. Int. (2021.0) 2021 6. Abraham C., Michie S.. A taxonomy of behavior change techniques used in interventions. Health Psychol. (2008.0) 27 379. PMID: 18624603 7. Michie S., West R., Finnerty A.N., Norris E., Wright A.J., Marques M.M., Johnston M., Kelly M.P., Thomas J., Hastings J.. Representation of behaviour change interventions and their evaluation: Development of the upper level of the behaviour change intervention ontology. Wellcome Open Res. (2020.0) 5 8. Mummah S.A., Robinson T.N., King A.C., Gardner C.D., Sutton S.. IDEAS (Integrate, Design, Assess, and Share): a framework and toolkit of strategies for the development of more effective digital interventions to change health behavior. J. Med. Internet Res. (2016.0) 18 9. Michie S., Atkins L., West R.. A Guide to Designing Interventions (2014.0) 1003-1010 10. Michie S., Yardley L., West R., Patrick K., Greaves F.. Developing and evaluating digital interventions to promote behavior change in health and health care: recommendations resulting from an international workshop. J. Med. Internet Res. (2017.0) 19 11. Parimbelli E., Gabetta M., Lanzola G., Polce F., Wilk S., Glasspool D., Kogan A., Leizer R., Gisko V., Veggiotti N.. International Conference on Artificial Intelligence in Medicine (2021.0) 298-303 12. Miller S., Ainsworth B., Yardley L., Milton A., Weal M., Smith P., Morrison L.. A framework for analyzing and measuring usage and engagement data (AMUsED) in digital interventions. J. Med. Internet Res. (2019.0) 21 13. Veggiotti N., Sacchi L., Peleg M.. AMIA Annual Symposium Proceedings, Vol. 2021 (2021.0) 1186 14. 14HL7. Fast healthcare interoperability resources release 3 specification2017https://www.hl7.org/fhir/overview.html. Last Accessed At: 2022-11-15. (2017.0) 15. Fogg B.J.. (2019.0) 16. Peleg M., Michalowski W., Wilk S., Parimbelli E., Bonaccio S., O’Sullivan D., Michalowski M., Quaglini S., Carrier M.. Ideating mobile health behavioral support for compliance to therapy for patients with chronic disease: a case study of atrial fibrillation management. J. Med. Syst. (2018.0) 42 1-15 17. Gruber T.R., Allen J.A., Fikes R., Sandewall E.. Principles of Knowledge Representation and Reasoning: Proceedings of the Second International Conference (1991.0) 601-602 18. Hastings J., Michie S., Johnston M.. Theory and ontology in behavioural science. Nat. Hum. Behav. (2020.0) 4 226. PMID: 32066979 19. Arp R., Smith B., Spear A.D.. (2015.0) 20. Krupp L.B., LaRocca N.G., Muir-Nash J., Steinberg A.D.. The fatigue severity scale: application to patients with multiple sclerosis and systemic lupus erythematosus. Arch. Neurol. (1989.0) 46 1121-1123. PMID: 2803071 21. Lucassen G., Dalpiaz F., van der Werf J.M.E., Brinkkemper S.. Improving agile requirements: the quality user story framework and tool. Requir. Eng. (2016.0) 21 383-403 22. Mystakidou K., Parpa E., Tsilika E., Pathiaki M., Patiraki E., Galanos A., Vlahos L.. Sleep quality in advanced cancer patients. J. Psychosom. Res. (2007.0) 62 527-533. PMID: 17467407 23. Hofman M., Ryan J.L., Figueroa-Moseley C.D., Jean-Pierre P., Morrow G.R.. Cancer-related fatigue: the scale of the problem. Oncologist (2007.0) 12 4-10. PMID: 17573451 24. Zabora J., BrintzenhofeSzoc K., Curbow B., Hooker C., Piantadosi S.. The prevalence of psychological distress by cancer site. Psycho-Oncol. J. Psychol. Soc. Behav. Dimens. Cancer (2001.0) 10 19-28 25. Spiegel D., Giese-Davis J.. Depression and cancer: mechanisms and disease progression. Biol. Psychiat. (2003.0) 54 269-282. PMID: 12893103 26. Littman A.J., Tang M.-T., Rossing M.A.. Longitudinal study of recreational physical activity in breast cancer survivors. J. Cancer Surviv. (2010.0) 4 119-127. PMID: 20180037 27. Shahar Y., Miksch S., Johnson P.. Conference on Artificial Intelligence in Medicine in Europe (1997.0) 51-61 28. Linton M.-J., Dieppe P., Medina-Lara A.. Review of 99 self-report measures for assessing well-being in adults: exploring dimensions of well-being and developments over time. BMJ Open (2016.0) 6 29. Kogan A., Peleg M., Tu S.W., Allon R., Khaitov N., Hochberg I.. Towards a goal-oriented methodology for clinical-guideline-based management recommendations for patients with multimorbidity: GoCom and its preliminary evaluation. J. Biomed. Inform. (2020.0) 112. DOI: 10.1007/978-3-031-09342-5_44 30. Fox J., Alabassi A., Patkar V., Rose T., Black E.. An ontological approach to modelling tasks and goals. Comput. Biol. Med. (2006.0) 36 837-856. PMID: 16169546 31. US Dept. of Veterans Affairs J.. (2015.0) 1-21 32. Cella D., Hahn E.A., Jensen S.E., Butt Z., Nowinski C.J., Rothrock N., Lohr K.N.. (2015.0) 33. McChesney C., Covey S., Huling J.. (2012.0) 34. Morin C.M., Belleville G., Bélanger L., Ivers H.. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. Sleep (2011.0) 34 601-608. PMID: 21532953 35. Williams N.. The GAD-7 questionnaire. Occup. Med. (2014.0) 64 224 36. Kroenke K., Spitzer R.L., Williams J.B.. The PHQ-9: validity of a brief depression severity measure. J. Gen. Intern. Med. (2001.0) 16 606-613. PMID: 11556941 37. Graham R., Mancher M., Wolman D., Greenfield S., Steinberg E.. (2011.0) 38. Chou R., Huffman L.H.. Nonpharmacologic therapies for acute and chronic low back pain: a review of the evidence for an American Pain Society/American College of Physicians clinical practice guideline. Ann. Int. Med. (2007.0) 147 492-504. PMID: 17909210 39. Fabi A., Bhargava R., Fatigoni S., Guglielmo M., Horneber M., Roila F., Weis J., Jordan K., Ripamonti C.. Cancer-related fatigue: ESMO Clinical Practice Guidelines for diagnosis and treatment. Ann. Oncol. (2020.0) 31 713-723. PMID: 32173483 40. Murad M., Asi N., Alsawas M., Alahdab F.. New evidence pyramid. BMJ Evidence Based Med. (2016.0) 125-127 41. Song S., Yu J., Ruan Y., Liu X., Xiu L., Yue X.. Ameliorative effects of Tai Chi on cancer-related fatigue: a meta-analysis of randomized controlled trials. Supportive Care Cancer (2018.0) 26 2091-2102 42. Amorim A.B., Pappas E., Simic M., Ferreira M.L., Jennings M., Tiedemann A., Carvalho-e Silva A.P., Caputo E., Kongsted A., Ferreira P.H.. Integrating Mobile-health, health coaching, and physical activity to reduce the burden of chronic low back pain trial (IMPACT): a pilot randomised controlled trial. BMC Musculoskelet. Disord. (2019.0) 20 1-14. PMID: 30611236 43. Cornet V.P., Holden R.J.. Systematic review of smartphone-based passive sensing for health and wellbeing. J. Biomed. Inform. (2018.0) 77 120-132. PMID: 29248628 44. Lisowska A., Lavy S., Wilk S., Peleg M.. AMIA Annual Symposium Proceedings (2022.0) 45. Ghosh S., Banerjee A., Ray N., Wood P.W., Boulanger P., Padwal R.. 2016 IEEE Healthcare Innovation Point-of-Care Technologies Conference (2016.0) 188-191 46. 46S. Saganowski, P. Kazienko, M. Dziezyc, P. Jakimów, J. Komoszynska, W. Michalska, A. Dutkowiak, A. Polak, A. Dziadek, M. Ujma, Consumer wearables and affective computing for wellbeing support, in: MobiQuitous 2020-17th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, 2020, pp. 482–487. 47. Lisowska A., Wilk S., Peleg M.. International Conference on Artificial Intelligence in Medicine (2021.0) 72-82 48. Yardley L., Spring B.J., Riper H., Morrison L.G., Crane D.H., Curtis K., Merchant G.C., Naughton F., Blandford A.. Understanding and promoting effective engagement with digital behavior change interventions. Am. J. Prev. Med. (2016.0) 51 833-842. PMID: 27745683 49. WebMDq L.. (2022.0) 50. Michie S., Richardson M., Johnston M., Abraham C., Francis J., Hardeman W., Eccles M.P., Cane J., Wood C.E.. The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: building an international consensus for the reporting of behavior change interventions. Ann. Behav. Med. (2013.0) 46 81-95. PMID: 23512568 51. Carey R.N., Connell L.E., Johnston M., Rothman A.J., De Bruin M., Kelly M.P., Michie S.. Behavior change techniques and their mechanisms of action: a synthesis of links described in published intervention literature. Ann. Behav. Med. (2019.0) 53 693-707. PMID: 30304386 52. Dugas M., Gao G., Agarwal R.. Unpacking mHealth interventions: a systematic review of behavior change techniques used in randomized controlled trials assessing mHealth effectiveness. Digit. Health (2020.0) 6 53. Gosetto L., Ehrler F., Falquet G.. Integrated Citizen Centered Digital Health and Social Care (2020.0) 77-81 54. Tong H.L., Quiroz J.C., Kocaballi A.B., Fat S.C.M., Dao K.P., Gehringer H., Chow C.K., Laranjo L.. Personalized mobile technologies for lifestyle behavior change: A systematic review, meta-analysis, and meta-regression. Prev. Med. (2021.0) 55. Cugelman B.. Gamification: what it is and why it matters to digital health behavior change developers. JMIR Serious Games (2013.0) 1 56. Bandura A., Walters R.H.. (1977.0) 57. Edney S., Plotnikoff R., Vandelanotte C., Olds T., De Bourdeaudhuij I., Ryan J., Maher C.. “Active Team” a social and gamified app-based physical activity intervention: randomised controlled trial study protocol. BMC Public Health (2017.0) 17 1-10. PMID: 28049454 58. Edney S., Ryan J.C., Olds T., Monroe C., Fraysse F., Vandelanotte C., Plotnikoff R., Curtis R., Maher C.. User engagement and attrition in an app-based physical activity intervention: secondary analysis of a randomized controlled trial. J. Med. Internet Res. (2019.0) 21 59. 59A. Lisowska, S. Lavy, S. Wilk, M. Peleg, Personality and habit formation: Is there a link?, in: Proceedings of the AIxIA 2021 SMARTERCARE Workshop, CEUR-WS, 2021, pp. 42–47. 60. 60M. Schmidt-Kraepelin, S. Thiebes, S. Stepanovic, T. Mettler, A. Sunyaev, Gamification in health behavior change support systems-A synthesis of unintended side effects, in: Proceedings of the 14th International Conference on Wirtschaftsinformatik, 2019, pp. 1032–1046. 61. 61G.F. Tondello, R.R. Wehbe, L. Diamond, M. Busch, A. Marczewski, L.E. Nacke, The gamification user types hexad scale, in: Proceedings of the 2016 Annual Symposium on Computer-Human Interaction in Play, 2016, pp. 229–243. 62. Baretta D., Bondaronek P., Direito A., Steca P.. Implementation of the goal-setting components in popular physical activity apps: Review and content analysis. Digit. Health (2019.0) 5 63. Ghalibaf A.K., Nazari E., Gholian-Aval M., Tara M.. Comprehensive overview of computer-based health information tailoring: a systematic scoping review. BMJ Open (2019.0) 9 64. 64B.J. Fogg, A behavior model for Persuasive design, in: Proceedings of the 4th International Conference on Persuasive Technology, 2009, pp. 1–7. 65. Sadasivam R.S., Borglund E.M., Adams R., Marlin B.M., Houston T.K.. Impact of a collective intelligence tailored messaging system on smoking cessation: the Perspect randomized experiment. J. Med. Internet Res. (2016.0) 18 66. Manuvinakurike R., Velicer W.F., Bickmore T.W.. Automated indexing of internet stories for health behavior change: weight loss attitude pilot study. J. Med. Internet Res. (2014.0) 16 67. Lisowska A., Wilk S., Peleg M.. 2021 IEEE 34th International Symposium on Computer-Based Medical Systems (2021.0) 137-141 68. 68A. Lisowska, S. Wilk, M. Peleg, From personalized timely notification to healthy habit formation: A feasibility study of reinforcement learning approaches on synthetic data, in: Proceedings of the AIxIA 2021 SMARTERCARE Workshop, CEUR-WS, 2021, pp. 7–18. 69. Copeland D.. Brief workplace interventions addressing burnout, compassion fatigue, and teamwork: A pilot study. Western J. Nurs. Res. (2021.0) 43 130-137 70. Jamaludin I.I., You H.W.. Burnout in relation to gender, teaching experience, and educational level among educators. Educ. Res. Int. (2019.0) 2019. DOI: 10.1155/2019/7349135
--- title: 'Of primary health care reforms and pandemic responses: understanding perspectives of health system actors in Kerala before and during COVID-19' authors: - Hari Sankar D - Jaison Joseph - Gloria Benny - Devaki Nambiar journal: BMC Primary Care year: 2023 pmcid: PMC9975828 doi: 10.1186/s12875-023-02000-0 license: CC BY 4.0 --- # Of primary health care reforms and pandemic responses: understanding perspectives of health system actors in Kerala before and during COVID-19 ## Abstract ### Background In 2016, the Government of the southern Indian state of Kerala launched the Aardram mission, a set of reforms in the state’s health sector with the support of Local Self Governments (LSG). Primary Health Centres (PHCs) were slated for transformation into Family Health Centres (FHCs), with extended hours of operation as well as improved quality and range of services. With the COVID-19 pandemic emerging soon after their introduction, we studied the outcomes of the transformation from PHC to FHC and how they related to primary healthcare service delivery during COVID-19. ### Methods A qualitative study was conducted using In-depth interviews with 80 health system actors (male $$n = 32$$, female $$n = 48$$) aged between 30–63 years in eight primary care facilities of four districts in Kerala from July to October 2021. Participants included LSG members, medical and public health staff, as well as community leaders. Questions about the need for primary healthcare reforms, their implementation, challenges, achievements, and the impact of COVID-19 on service delivery were asked. Written informed consent was obtained and interview transcripts – transliterated into English—were thematically analysed by a team of four researchers using ATLAS.ti 9 software. ### Results LSG members and health staff felt that the PHC was an institution that guarantees preventive, promotive, and curative care to the poorest section of society and can help in reducing the high cost of care. Post-transformation to FHCs, improved timings, additional human resources, new services, fully functioning laboratories, and well stocked pharmacies were observed and linked to improved service utilization and reduced cost of care. Challenges of geographical access remained, along with concerns about the lack of attention to public health functions, and sustainability in low-revenue LSGs. COVID-19 pandemic restrictions disrupted promotive services, awareness sessions and outreach activities; newly introduced services were stopped, and outpatient numbers were reduced drastically. Essential health delivery and COVID-19 management increased the workload of health workers and LSG members, as the emphasis was placed on managing the COVID-19 pandemic and delivering essential health services. ### Conclusion Most of the health system actors expressed their belief in and commitment to primary health care reforms and noted positive impacts on the clinical side with remaining challenges of access, outreach, and sustainability. COVID-19 reduced service coverage and utilisation, but motivated greater efforts on the part of both health workers and community representatives. Primary health care is a shared priority now, with a need for greater focus on systems strengthening, collaboration, and primary prevention. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12875-023-02000-0. ## Background The Astana declaration of 2022 reaffirmed Primary Health Care (PHC) as the cornerstone for achieving Universal Health Coverage (UHC) and investments in improving the PHC system as the most efficient and inclusive approach to attain health-related United Nations Sustainable Development Goal 3 (SDG) by 2030 [1, 2]. Countries across the globe stand at varying stages on the UHC path: a modelling study from 67 Lower and Middle-Income Countries (LMICs) estimated that current PHC spending has to be at least doubled to make needed improvements in their systems and ensure PHC services are universally accessible [3]. Countries have committed to UHC and agreed to monitor their progress towards attaining it, even as there exist some critiques of the global consensus around what is to be prioritised in PHC and UHC reform [4]. Indian health policy reflects PHC and UHC as priorities as well: in 2015, the Ayushman Bharat-Pradhan Mantri Jan Arogya Yojana (AB-PMJAY) was rolled out by the Government of India with a component to provide universal access to PHC services to all its people [5]. Public health is a state subject in India and several Indian states had – before this central government reform, but also following it – devised tailor-made reforms to improve PHC service delivery [6–8]. The Kerala model of development is acclaimed for the investments in education, healthcare social infrastructure which yielded improved health outcomes like high life expectancy, low infant mortality and low birth rate [9]. Investment in the health sector has traditionally remained a priority area for governments in Kerala since the formation of the state [10]. In 2016, The Government of Kerala, through the Aardram mission [11], introduced a series of reforms in the health sector of the state with the support of Local Self Governments (LSGs). Primary Health Centres were slated for transformation into Family Health Centres (FHCs), with extended hours of operation as well as improved quality and range of services [11]. Kerala had early on sought to operationalise the constitutional obligation to decentralise power; as early as 1995 the state transferred funds, functions and functionaries of several government institutions including health to LSGs, meaning that two decades on, Primary Health Centres (now FHCs) roughly catering to 30,000 population in Kerala are managed by LSGs along with the health department. A typical PHC in Kerala has a one Medical Officer (MO) who provides clinical services and also supervises the public health team in the PHC. Every PHC has five to six subcentres roughly catering five thousand population. The public health team consists of a Health Inspector (HI) who supervises Junior Health Inspectors (JHI) in implementing communicable and non-communicable disease control activities. The Public Health Nurse (PHN) supervises a team of five to six Junior Public Health Nurses (JPHN) to provide a range of services to the population in their catchment area; they are supported by community health workers (CHW) named ASHAs, each of whom is assigned about a thousand individuals to support and connect with the health system [12]. LSGs have a controlling stake in PHCs and support them with additional human resources, maintenance funds, medicines and consumables [13]. The FHC program upgraded erstwhile resourcing, increasing the resources to three MOs and four staff nurses in FHCs, as well as the introduction of newer services at the primary level like Chronic Obstructive Pulmonary Disease management (COPD), diabetic retinopathy and depression screening [12, 14, 15]. The FHC program in Kerala under the Aardram Mission was the largest investment the state had made in recent years to improve health care infrastructure, with over 5,289 posts of hospital workers created and doubling the plan investment in the health sector [16]. The FHC program implementation was steadily progressing in the state when the first case of COVID - 19 in India was reported in Kerala in January 2020. Kerala's state enforced lockdown measures were followed by the national lockdown. Kerala’s COVID - 19 management efforts during the first wave of the pandemic received global acclaim. While case tallies in the state remained high, case fatality was low throughout, with early and high uptake of vaccination as well [17]. Some have credited this response to the existing strengths of the health system, particularly at the primary care level [18]. It could be said that the test of any reform is a moment of crisis. Aggregate figures used for monitoring mask the more in-depth nuances, stories, lessons and challenges that undergird success and failure. We contend that even as tracking of PHC reforms has been essential in understanding the status, qualitative studies which capture implementer perspectives and user accounts of health programs are vital in informing the policymakers about the barriers and enablers of the program locally [19]. Data on experiences /expertise collected through personal accounts are essential in course correction of program execution and generating future implementation plans. Kerala’s public health system, which is heavily supported by LSGs, presents a scenario which has multiple stakeholders other than health department personnel involved with implementing health reforms. In the case of Kerala, there has been some work to monitor UHC-relevant PHC reforms [20, 21]. A 2021 health department report found that annual outpatient numbers in primary care institutions grew almost $24\%$ between 2019–20 and 2017–18, reaching 31.5 million by 2020 post health reform [22]. An early evaluation in a single FHC with health staff and patients reported increased patient friendliness and improved service delivery [23]. Studies which evaluate the primary health reforms in Kerala specifically from the perspective of health system actors – from implementers- is lacking. Considering the system changes, as well as the shocks to the system introduced by COVID - 19 as well as floods and other disasters preceding it, we carried out a study to understand the outcomes of the FHC program from the perspectives of key supply-side actors before and during the context of COVID-19. ## Methods This narrative research study constituted the qualitative component of a larger mixed methods health system research project in Kerala titled “Assessing equity of Universal Health Coverage in India: From data to decision-making using mixed methods” [24]. Study sites were identified through multistage random sampling; Kerala’s 14 districts were grouped into four categories using an index developed through principal component analysis of human development indicators from the National Family Health Survey (NFHS) Round 4 (2015–16) data [25]. One district was randomly chosen from each of the four groups, two primary care health facilities in each district were randomly selected from the district facility list. The details of sampling and selection criteria are detailed elsewhere [26]. We conducted In-Depth Interviews (IDI) with 80 health system actors across the eight primary care facilities/FHCs and four districts in Kerala from July to October 2021. We employed a purposive criterion sampling approach: study participants were selected from three categories from each study site using purposive criterion sampling where the intent was to include health system actors involved with FHC/PHC functioning from: i) Elected representatives of LSG (panchayat) ($$n = 17$$) who held positions of Panchayat President (PP), Vice president (VP) Health Standing Committee Chairperson (HSC) and ward members; ii)Health Care Providers (HCP) ($$n = 40$$) who were posted as Medical Officers (MO), Staff Nurses (SN), and field staff like HI JPHN, JHI, PHN, Palliative care Nurses (PN) in health facilities selected for study as well as iii) Community health workers (ASHA) and local community leaders($$n = 23$$) working in the catchment area of selected facilities. The study was approved by the Institutional Ethics Committee of The George Institute for Global Health (Project Number $\frac{05}{2019}$). A three-member research team with qualitative research training consisting of two male research fellows and one female research assistant carried out the fieldwork, supervised by a senior health systems researcher. Administrative approval was taken from the Department of Health and Family Welfare, Government of Kerala. The team met the District Medical Officers (DMO) of four districts, shared the departmental permissions and outlined the study objectives and methods in detail and requested permission. After receiving permission from the DMO, permission from the MOs of each of the eight facilities as well as Panchayat presidents was sought, using the same process. HCPs who belonged to the identified categories and who agreed to participate in the study were briefed about the objectives of the study. The pilot-test of the questionnaire was done with one MO and an HI from department of health, Kerala and an LSG member from Thiruvananthapuram district of Kerala. The feedback from piloting of questionnaire helped to refine the questionnaire, set the flow and pace of interview, remove redundancy, record interview time and improve the cue for interview questions. The semi structured questionnaire in English /Malayalam were shared with the participants whenever possible, before the interview. As the COVID -19 pandemic emerged as a major determinant in the study period specific questions were added to study its impact. The semi-structured questionnaire used in the study asked the health system actors their view on the need for primary health care reform, their role(s) in implementing this reform through the FHC upgradation process, challenges faced during implementation of the FHC program, outcomes of the program (including the impact of COVID-19 therein). Interview questions of the topic guide are attached in (Supplementary table 1) and were adapted from Mohindra and colleagues prior elicitation based qualitative research with populations facing disadvantage in Kerala [27]. HCPs were met in person at a time and place of their convenience for conducting the interviews. Medical officers and health inspectors were interviewed in their offices, while other HCPs were interviewed in common meeting room where privacy could be ensured (i.e. scheduling at a time when the room was not being used). Most of the IDIs were carried out in person, five HCPS who were unable to meet in person were interviewed through telephone. The interviews were conducted strictly adhering to COVID -19 protocols prescribed by the state government, i.e. interviewers and participants wore a mask all the time and physical distancing was maintained. The IDIs with LSG members were carried out first by meeting Panchayat Presidents and briefing them about the study. The elected LSG members were met with in person in the LSG office for conducting the IDIs. The interviews were conducted in official rooms of the Panchayat President, Vice President and Health Standing Committee members. A hard copy of the Participant Information Sheet (PIS) was handed to each participant for the in-person interviews and signed informed consent was taken for participating in the study and for recording the interviews. For interviews conducted online, PIS was sent over mail/Whatsapp and a signed soft copy was obtained. Before commencing the interview, the participants shared the duly signed consent form with the researchers. Interviews were conducted in Malayalam and lasted for 20–60 min. To obtain context and perspectives of health system actors in various capacities and geographies pertaining to each of the study sites across four districts the interviews with all the pre-set list of participants were completed even though achieving early data saturation was reached with some of the study topics. The response rate of the study was $96\%$, as three health system actors could not participate in the interview due to their busy schedules and after multiple failed attempts to schedule, we decided to remove them from the study. All the interviews were audio-recorded, and the recordings were secured in a password-protected database with researcher-only access. Detailed field notes of the interviews were written by the researchers to support the transcripts. Transcription was done by a professional transcription firm that transliterated the Malayalam interviews to transcripts in English. The quality checks of transcripts were done by the research team. Inductive analysis of data was done using ATLAS.ti 9. Four members (DN,HS, JJ and GB) created a basic coding framework and developed a codebook iteratively. The codes that emerged were discussed and conflicts resolved. Codes were finalised and then applied by all coders across the dataset. The codes used for the study are described in Additional file 1. All coders’ ATLAS.ti files were merged, codes indexed and charted with emerging themes being discussed in weekly meetings over several months. The codes were arranged into themes and consolidated into a narrative summary and laid out in the results section. ## Participant characteristics We conducted IDIs with 80 participants aged 30 to 63, of whom $60\%$ ($$n = 48$$) were women (see Table 1). Out of the total participants recruited for the study, $30\%$ were LSG members. The range of professional work experience ranged from first-time elected representatives who had been working for just six months to health care providers and community leaders with over thirty years of experience. Table 1Participant characteristicsCategoryDesignationProfessional experience range (in years)FemaleMaleTotalLSG members and community leadersPanchayat President6 months -5 years347Panchayat Vice-President5 years011Health Standing Committee Member6 months-10 years358Ward Member7 months011Community Leader10–30 years167Health Care ProvidersMedical Officer1–15 years538Health Inspector25–30 years156Public Health Nurse25–32 years404Junior Health Inspector14–26 years077Junior Public Health Nurse (JPHN)6–21 years11011Nursing Officer (NO)1–6 years303Palliative Nurse (PN)8 years101Community Health Worker (CHW)10–13 years16016Total Participants483280The years of experience served in the position was considered for the study. Most of the elected representatives had decades of political experience but years served in positions like president is indicated in the study Our analysis yielded four themes, namely: the need for PHC reforms, the content of reforms, challenges encountered, and finally, impact of reforms including in relation to COVID-19.i) The need for primary health care reforms Most participants felt that the PHC should be an institution that guarantees preventive, promotive, and curative care to the poorest section of society and can help in reducing the high cost of care. The recent reform to convert PHCs to FHCs was welcomed by many participants, as evidenced by a comment made by a frontline health worker in Kollam “It is the people's wellbeing that the government wants -… as a part of this, for the poor people, for people who are financially backwards to buy medicines, etc., these facilities are made.” ( CHW KLM) Other participants opined that the state bore responsibility for protecting the health and providing education to its people and that this commitment had been displayed by successive governments in Kerala. The present reform was seen as a continuation of this. A community leader from Thiruvananthapuram had this to say:If you look at the history of Kerala, irrespective of the ruling party- even if they differ in some manner- the state always had a strong public health policy, right from the first democratic government that came to power in 1957. Along with or instead of building multispecialty hospitals for curative medicine, a public health system was built in the state…, HIs and JPHN form a large network that works at the Panchayat level among the people….. through the “Aardram’ mission, the government has been able to increase public participation in the health sector. ( CL TVM) A conflicting opinion about the investment in primary care through the recent reform was made by a HCP who felt the existing primary care system was robust enough and rather than investing more in primary care that the focus should be on secondary care. In a state like Kerala, there is no need to invest a huge amount in primary healthcare. In our state, most people are aware of the importance of such things. Before the Aardram Mission, we had NCD programs. We covered the maximum number of people in that, and they used to buy medicines from the PHCs. There are not many remote areas in Kerala. Therefore, I don't think it is a necessity. What I feel is that, if we use half of this fund for adding more facilities to Taluk Hospitals and hospitals above that, we could utilise it better. ( MO KLM)ii) What happened as part of the reforms? Post transformation to an FHC, LSG members and HCPs reflected on significant, immediately visible changes like extended outpatient hours (9 am -6 pm) and additional doctors and nurses posted in FHCs. Participants mentioned new services introduced in the upgraded FHCs that linked earlier reforms like the generation and use of electronic health records (through the E-health program) [28] to provision of precheck services by staff nurses: the nurse would check patient vitals and enter these on the e-health platform in advance of the actual consultation with the MO). Participants also noted the introduction of speciality clinics: Chronic Obstructive Pulmonary Disease (COPD) management (SWAAS) clinic, depression screening clinic and screening for diabetic retinopathy. It was observed that reforms ensured patient amenities, fully stocked laboratories, and pharmacies in FHCs. A community leader commented about the improvements in medicine supply ensured by the health departmentYes. Changes are there. There aren't any issues in the case of medicines. Earlier, the allotted medicines would not be enough. So doctors will demand more medicines. And then we(Hospital management committee) had to arrange supplementary medicines. This is how we used to manage hospital-related activities. Now it's not needed.(CL KSD) Health care providers felt that infrastructural changes and better staffing had increased the efficiency of the facility, quality of care and improved data reporting standards. This in turn helped facilities heighten their ambition for and in many cases achieve national-level quality standards prescribed for PHCs in India. Staff reported contributing to these efforts and felt proud of being part of the system as reflected in the opinion of a staff nurse in Thiruvananthapuram: “So after Aardram Mission came a thing … the NQAS [National Quality Assurance Standards]. …… We document all of these, and all of this will be analysed by NQAS, I was able to contribute a lot to this as well. I consider myself to be lucky to have contributed to this.” ( JPHN TVM) HCPs and LSG members expressed the opinion that post-FHC transformation, the outpatient visit numbers had gone up, and the coverage achieved by the facility had also increased. Participants felt that the program had a significant impact by delivering high-quality service and reducing out of pocket expenditure. A medical officer from Kasargode described this with an example. “One benefit of this is that out-of-pocket expenditure is very less now. Earlier when we used to shut down at 12:30 PM, if someone cut their hands(injury) they had to rely on private hospitals. If they go to a private hospital, they need to register first, and then they need to pay a doctor's fee and on top of that, they need to pay for any additional expenses, like an injection. We are treating them for Rs. 5 ($0.6) by keeping the hospital open until 6 PM instead of leaving them to spend Rs. 500($6.3) in a private hospital, for a patient, it is Rs. 495 ($6.2) saved when they choose to come here. ( MO KSD) Another outcome of the reform many participants referred to was an increase in confidence among people and trust in services offered by the government. A medical officer in Alappuzha noted:More people are coming here now. Those who used to depend on private hospitals are gradually coming here. I cannot demand that everybody should visit here but if the ratio of people who used to rely on private and public clinics earlier were 60:40 respectively now it has come to 50:50 or the public sector is leading. More people are visiting the PHC for NCD treatments. ( MO ALP)iii) Challenges in implementing reforms Notwithstanding these early gains, participants also reported that the reforms introduced newer positions like staff nurses and increased the number of medical officers, but overlooked other staff positions like field health staff, cleaning staff and clerical staff. While improvements in FHCs were appreciated, basic infrastructural and accessibility issues of frontline institutions like subcentres remained, as the following quotes from Junior Public Health Nurses in Alappuzha and Thiruvananthapuram demonstrate: “The sub-centre is situated at the backside of the Panchayat building. We do face issues in terms of transportation facilities. People can visit the place but vehicles cannot enter the area….. Only patients who can walk come here for treatments.” ( JPHN ALP)Even though a lot of human resources have increased, those have only happened in the treatment section. There are still a lot of drawbacks on the preventive side. Concerning us, a JPHN or a JHI should be working for a population of 5,000. But currently, a JPHN or JHI is working with 12,000 to 15,000 people. There are a lot of drawbacks because of this, and there has been no growth in the preventive side as a result of Aardram. ( JPHN TVM) LSG leaders also reported the geographical location of the PHC site and lack of public transport challenged access to care for all sections of the population in the facility catchment area. They also expressed concern over sustained support for reforms (as many of them are bankrolled by LSG) as not every LSG in Kerala has strong own revenue streams: a Panchayat president from Thiruvananthapuram voiced this concern:The government has directed the Panchayat to appoint a doctor, a nurse and a pharmacist. Their salaries should be given using the Panchayat fund itself. Our Panchayat has a meagre own fund. We are holding on because we come under the government's general purpose. We do not receive even $50\%$ of the expenditure as revenue. In such a situation, we will not be able to provide the services of medical staff. This is a huge crisis. ( PP TVM) Other kinds of challenges were also reported concerning roles. For example, Medical Officers were designated as the implementing officer of PHC reforms, but this required them to lead health teams with subordinates that had far more experience. Establishing lines of accountability, then, could be challenging at times, an MO from Kasargod reflected on thisThe real issue for a Medical Officer is not about the administration, it is the age. We face a lot of seniority problems. After we join, we are going to lead a group of subordinates like Junior Health Inspectors or Health Inspectors who are …. 50 to 56 [years old]. … they may not be following protocol. They just assume that it is a new boy. ( MO KSD).iv) Impact of COVID -19 on primary health care reforms The restrictions introduced as part of the state’s COVID-19 response hampered reforms. Lockdown restrictions caused outpatient visit numbers to drop and field-level health activities to be held in abeyance, including school and institution-based activities. Even essential health services like immunisation, maternal and child health services, as well as Non-Communicable Disease (NCD) services, were briefly shut down. Attention was instead focused on testing and tracing among expatriates returning home as well as containment and vaccination activities. The COVID First-Line Treatment Centres (CFLTC) and Domiciliary Care Centres (DCC) started in LSGs were supported by PHC staff. This was seen by some HCPs as an unprecedented burden: “This level of strain is there, if you ask any JPHN who is about to retire, they will tell you that they never faced a period when there was this much risk and strain in their service” (JPHN TVM). Lockdown restrictions were eased in phases and PHC services were adjusted according to the restrictions at each time. Attempts were made to maintain essential primary health services virtually and to ensure delivery of medicines at home through field health staff and volunteers. LSGs and HCPs worked together to train volunteer Rapid Response Team members (RRT) who delivered essential medicine to people based on instructions from the HCPs along with the field health staff. Firstly, as I mentioned, NCD medicines were supplied to different patients' houses from each ward through ASHA workers (CHW), RRT members and other healthcare staff. They provided medicines periodically every month from hospitals. Then we made sure that all the training and meetings were conducted online and not in person. These are the main changes we adopted during this COVID-19 crisis. ( MO TVM) Following an initial period of disruption, therefore, attempts were made to re-activate the resources and procedures introduced by primary health reforms. ## Discussion This study sought to understand the implementors' perspectives on the recent primary health reforms in Kerala. We learned that HCPs, LSG members and community leaders saw value and appropriateness in investment in PHC services by the government. Participants reflected on the improvements brought in by the reforms through upgradation of infrastructure, human resource and quality of service delivery however the lack of focus on preventive component of PHC services remains a challenge. The emergence of COVID-19 pandemic disrupted the reforms but the adaptive measures by the system were introduced to ensure uninterrupted essential primary care health care delivery. The health system actors we interviewed—elected leaders of LSG and HCPs—perceived the primary care reforms as a step in the right direction and defined PHC services as people-centred with preventive and promotive care components and many considered healthcare provision as a responsibility of the government. Alignment between local political leaders, decision makers and health providers about the need and scope of services is crucial to the success of any program. The WHO operational framework for PHC care describes the importance of commitment by political leaders in implementing PHC as multisectoral coordination for improving social, economic, environmental and commercial determinants of health is possible only with committed political leadership [29]. LSG leadership in Kerala appears to have the capacity and resources to carry out this stewardship role. This understanding can be attributed to the already existing robust PHC network in the state and over twenty-five years of decentralized governance, such that LSG roles in the health sector are actually in place at the grass-root level [30]. A review on decentralization and its health system impact(s) stated that role clarity, knowledge of local context and local decision-making can be critical determinants to the success of a program [31]. As reflected by our participants, a significant characteristic of the reform was introducing speciality clinics in FHCs for COPD and depression screening. SWAAS clinic was the first program in India to address the burden of COPD and Asthma through primary health care. as of January 2021 the program is reported to have screened over 148,870 patients, diagnosed nearly nineteen thousand COPD cases, and provided free medicines including inhalers [32, 33]. The depression screening program in FHCs has, as of July 2022, screened over sixty five thousand people and diagnosed nearly twelve thousand cases in the state [34]. Beyond this, our study suggests that FHC program may have escalated quality improvement activities in government health facilities of the state: 85 out of 932 primary care facilities in the state were certified by NQAS in 2021 [35]; the numbers continue to grow. A study to identify barriers and enablers to NQAS certification of the facilities in Kerala reported that transformation to FHCs ensured fully stocked pharmacies, diagnostics, patient amenities and improved commitment of the staff which is consistent with the reflections of our study participants [36]. Another study conducted among government doctors in Malappuram district in 2014, before the implementation of the FHC program, reported short patient interaction time in OPD and administrative work as challenges of working in the government health system [37] In 2021, post-FHC transformation, the MOs interviewed in our study reported that with additional doctors posted, efficiency had increased as there could be the division of administrative and clinical duties. A follow-up study exploring this difference using matching indicators could shed more light on this. Trust in government machinery and elected representatives has traditionally been high in Kerala [38] and with the state government's efforts in managing the COVID-19 pandemic, the public trust has only further increased. Our findings are echoed in other studies that have also concluded that people feel more confident to use in government health system post Aardram Mission and the COVID-19 pandemic in Kerala, the COVID-19 pandemic had an impact on public trust in the government health system globally [39–42]. The central government model of primary care reform, the Health and Wellness Centre (HWC) program focuses on developing subcentres (each PHC has at least six subcentres where outreach health services like immunization, health education etc. are delivered) with Middle-level Health providers (a trained nurse or AYUSH doctor) [43]. In contrast, till 2020, the FHC model was focused on PHC transformation, with an emphasis on posting additional MBBS doctors. Disease surveillance, health promotion, and palliative home care are all managed by public health staff in an FHC, but in the reform, no additional field health staff positions were created or posted; this was in fact identified by HCPs as a challenge in implementing the FHC program. The design of the FHC program itself was aimed at improving infrastructure and posting additional doctors and nurses for medical management [44]. Our study found that field health workers were dissatisfied with the lack of focus on developing subcentre infrastructure and field health service delivery through the FHC program. Global evidence from LMICs suggests that field-level public health activities that improve population health and non-physician health workers are a critical component of primary health care delivery [45, 46]. The Kerala Department of Health and Family Welfare has now started to integrate the HWC program into FHC by recruiting staff nurses as MLHP and improving the infrastructure of one subcentre in each FHC [47]. This is likely to directly address the concerns of frontline workers; further study can reveal the acceptability of these reforms for both supply and demand side actors as well as their impact(s) on population health. LSG members and community leaders who participated in our study reflected that physical access to FHC remained a challenge for many in the community and this affects the utilization of services too. Kerala from as early as the nineties is reported to have a good network of roads and $78\%$ of villages have health facilities available within five kilometres. A study in Kasaragod - one of the underdeveloped districts in the states- also reported that the median distance to a public health facility was 6 km [48, 49]. Notwithstanding these improvements, transport and physical access were still seen as access barriers, particularly since already existing PHCs (which may have been hard to reach) were the ones chosen for FHC transformation: existing geographical challenges reportedly remained. The role of decentralization in helping Kerala in improving population health is well discussed and documented [50, 51]. The ability of the local population to participate in health planning and the autonomy of LSGs to design and fund locally relevant interventions has produced successful models like palliative care cascade in the state [52]. In 2016, when the Aardram mission was launched, LSGs were a very important partner in implementation, particularly given the role they played in co-funding human resources and infrastructure upgradation. Our study results indicate that going forward, there could be a challenge with this as many LSGs were facing fund shortages, especially post COVID - 19. This raises an issue with the sustainability of the FHC model in which staff salaries are met through LSG funding. The COVID - 19 pandemic and the restrictions placed as part of controlling the pandemic affected the PHC service delivery globally [53, 54] as in Kerala’s newly introduced FHC program. However, insofar as the FHC program itself was built on a strong edifice of reforms, backed by community participation and political will, the state had some early success in managing the pandemic [55]. Sustaining the reforms, managing (raised?) expectations of the public and continuing coordination and collaboration in non-emergency contexts while also expanding focus on the field and preventive care will be major areas of attention and concern going forward. Implementer perspectives are a useful way to understand supply side experiences and operational challenges related to the FHC program. Further research needs to be focused on the impact of disruptions in primary health service delivery during the COVID - 19 pandemic on population level service utilisation, perception of services, health outcomes, and knock on impacts (for example on livelihoods and household expenditure). There are likely impacts COVID - 19 has had on the design and functioning of the FHC program -whether short or long term, which also warrant examination. The progress made in improving FHC infrastructure and human resources is appreciable, but the study calls for policy action focusing on outreach health infrastructure and service delivery. A policy intervention for equipping the staff performing duties based on the original design principles of FHC is also required. ## Strengths and limitations Our study captured and summarised the opinions of different health system actors including health care staff, community members and elected representatives providing a comprehensive information about the rollout of primary care reform in Kerala and how did it function during the COVID - 19 pandemic. There are currently very few studies that studied about the Family Health Centre model of Kerala, particularly in the context of COVID - 19 and it is this gap that our analysis helps fill. Notwithstanding this needed focus on an implementer lens – this analysis lacked perspectives of communities. This is a limitation as user perspectives are crucial in understanding any program. These perspectives are in the process of being gathered in the next stage of our project. Moreover, the timing of our data collection was after the second wave of COVID - 19 in Kerala: health workers were overwhelmed with vaccination duties, this may have affected their opinion and reflections about the FHC program which started three years before in 2018 and many components of the program being on hold for long period after. Ongoing research with supply and demand side actors, through a prospective mixed methods program of health systems research could continue to shed light on this. ## Conclusion Supply-side actors involved with primary health care reforms in the southern Indian state of Kerala had clarity in the concept of what primary health care is and what their roles are. The FHC program improved infrastructure, but the creation of new posts as part of the program was skewed towards clinical roles. Physical access to facilities remained a barrier. COVID - 19 affected the implementation of the FHC program through essential PHC service delivery remains uninterrupted. The focus now should be to get the program back on track and think of a sustainable way ahead after a considerable time spent by the health system focusing exclusively on COVID - 19 management. ## Supplementary Information Additional file 1: Supplementary table 1. Interview questions in topic guide and codes used for analysis. ## References 1. 1.Declaration of Astana. Global conference on Primary Health Care. Astana, Kazakhstan. World Health Organization and UNICEF; 2018 Oct. Available from: https://www.who.int/docs/default-source/primary-health/declaration/gcphc-declaration.pdf. [Cited 4 Jun 2020]. 2. 2.UN General Assembly, Transforming our world : the 2030 Agenda for Sustainable Development, 21 October 2015, Available from: https://www.un.org/sustainabledevelopment/health/. [Cited 9 Jun 2022]. 3. Stenberg K, Hanssen O, Bertram M, Brindley C, Meshreky A, Barkley S. Guide posts for investment in primary health care and projected resource needs in 67 low-income and middle-income countries: a modelling study. Lancet Glob Health (2019.0) 7 e1500-e1510. DOI: 10.1016/S2214-109X(19)30416-4 4. Sanders D, Nandi S, Labonté R, Vance C, Damme WV. From primary health care to universal health coverage—one step forward and two steps back. The Lancet (2019.0) 394 619-621. DOI: 10.1016/S0140-6736(19)31831-8 5. 5.Ved RR, Gupta G, Singh S. India’s health and wellness centres: realizing universal health coverage through comprehensive primary health care. WHO South-East Asia J Public Health. 2019;8(1):18. 6. Lahariya C. Mohalla Clinics of Delhi, India: Could these become platform to strengthen primary healthcare?. J Fam Med Prim Care (2017.0) 6 1-10. DOI: 10.4103/jfmpc.jfmpc_29_17 7. 7.Muraleedharan V, Dash U, Vaishnavi S, M R, Hariharan M. Universal Health Coverage-Pilot in Tamil Nadu: Has it delivered what was expected? Centre for Technology and Policy Department of Humanities and Social Sciences, IIT Madras; 2018 Feb p. 66. Available from: http://www.nrhmtn.gov.in/adv/UHC%20Pilot%20Report_Tamilnadu_IITM_24Feb2018.pdf. [Cited 29 Apr 2019]. 8. Gambhir RS, Kumar R, Aggarwal A, Goel R, Anand S, Bhardwaj A. Primarycare teams and recent experiments towards population coverage in India. J Fam Med Prim Care (2018.0) 7 845-851. DOI: 10.4103/jfmpc.jfmpc_49_18 9. 9.Parayil G. The “Kerala model” of development: Development and sustainability in the Third World. Third World Q. 1996;17(5):941–58. 10. Kutty VR. Historical analysis of the development of health care facilities in Kerala State. India Health Policy Plan (2000.0) 15 103-109. DOI: 10.1093/heapol/15.1.103 11. 11.National Health Mission Kerala- Aardram . Available from: https://arogyakeralam.gov.in/2020/04/01/aardram/. [Cited 18 Jan 2023]. 12. 12.Krishnan GA, Nair AK. Primary Health-Care Innovations with Superior Allusion to Family Health Centers. Indian J Community Med. 2021;46(1):149-52. 10.4103/ijcm.IJCM_123_20. Epub 2021 Mar 1. 13. 13.Nair MS, Naidu VN Public Health Interventions By Local Governments In Kerala: An Effectiveness Analysis BMJ Global Health 2016;1:A19 14. 14.Government of Kerala. G. 0. (P) No.46/201 7/H&FWD.Transforming Primary Health Centre to Family Health Centre. 2017. Available from: http://dhs.kerala.gov.in/docs/transfer/Judgements/aardram_100817.pdf. [Cited 17 May 2019]. 15. Sivaprasad S, Netuveli G, Wittenberg R, Khobragade R, Sadanandan R, Gopal B. Complex interventions to implement a diabetic retinopathy care pathway in the public health system in Kerala: the Nayanamritham study protocol. BMJ Open (2021.0) 11 e040577. DOI: 10.1136/bmjopen-2020-040577 16. 16.Isaac TT, Sadanandan R. COVID-19, Public Health System and Local Governance in Kerala. 2020;55(21):7-8. Available from:https://www.epw.in/journal/2020/21/perspectives/covid-19-public-health-system-and-local-governance.html. [cited 2023 Feb 8] 17. Jaya AM, Harries AD, Rahman A, Khogali M, Chinnakali P, Gopalakrishnan LG. Epidemiology and Response to the COVID-19 Pandemic in Kerala, India, 2020–2021: A Cross-Sectional Study. Trop Med Infect Dis (2022.0) 7 105. DOI: 10.3390/tropicalmed7060105 18. Sadanandan R. Kerala’s response to COVID-19. Indian J Public Health (2020.0) 64 99. DOI: 10.4103/ijph.IJPH_459_20 19. Jordan Z, Lockwood C, Munn Z, Aromataris E. The updated Joanna Briggs Institute Model of Evidence-Based Healthcare. JBI Evid Implement (2019.0) 17 58-71 20. 20.Nambiar D, D HS, Negi J, Nair A, Sadanandan R. Monitoring Universal Health Coverage reforms in primary health care facilities: Creating a framework, selecting and field-testing indicators in Kerala, India. PLOS ONE. 2020;15(8):e0236169. 21. Nambiar D, Sankar H, Negi J, Nair A, Sadanandan R. Field-testing of primary health-care indicators, India. Bull World Health Organ. (2020.0) 98 747-53. DOI: 10.2471/BLT.19.249565 22. 22.Department of Health and Welfare, Kerala, Making the SDG a Reality [Internet]. Thiruvananthapuram: 2021 Feb [cited 2021 Aug 28]. Available from: https://keralahealthconference.in/e-Book.pdf. 23. Vijayan SM, Puliyakkadi S, Chalil S. Facilitators and barriers of service utilization: Perspectives of stakeholders in a family health center of central Kerala - A qualitative study. Indian J Public Health. (2021.0) 65 136. DOI: 10.4103/ijph.IJPH_995_20 24. 24.The George Institute for Global Health. Assessing equity of Universal Health Coverage in India: From data to decision-making using mixed methods. Available from: https://www.georgeinstitute.org/projects/assessing-equity-of-universal-health-coverage-in-india-from-data-to-decision-making-using. [Cited 9 Jan 2023]. 25. 25.International Institute of Population Science. NFHS Kerala Fact sheet 2015–16. Mumbai;. Available from: http://rchiips.org/NFHS/pdf/NFHS4/KL_FactSheet.pdf. [Cited 8 Apr 2020]. 26. Negi J, Sankar DH, Nair AB, Nambiar D. Intersecting sex-related inequalities in self-reported testing for and prevalence of Non-Communicable Disease (NCD) risk factors in Kerala. BMC Public Health (2022.0) 22 544. DOI: 10.1186/s12889-022-12956-w 27. Mohindra K, Narayana D, Harikrishnadas C, Anushreedha S, Haddad S. Paniya Voices: A Participatory Poverty and Health Assessment among a marginalized South Indian tribal population. BMC Public Health (2010.0) 10 149. DOI: 10.1186/1471-2458-10-149 28. 28.State Digital Health Mission, eHealth Kerala. Available from: https://dashboard.ehealth.kerala.gov.in/. [Cited 12 Oct 2021]. 29. 29.World Health Organization, Fund (UNICEF) UNC. Operational framework for primary health care: transforming vision into action. World Health Organization; 2020. 106. Available from: https://apps.who.int/iris/handle/10665/337641. [Cited 5 Oct 2021]. 30. 30.Kerala, India: Decentralized governance and community engagement strengthen primary care. PHCPI. 2015. Available from: https://improvingphc.org/promising-practices/kerala. [Cited 5 Jun 2020]. 31. Panda B, Thakur HP. Decentralization and health system performance – a focused review of dimensions, difficulties, and derivatives in India. BMC Health Serv Res (2016.0) 16 561. DOI: 10.1186/s12913-016-1784-9 32. 32.Gopakumar S, Valamparampil MJ, Manu MS, Nair S, Kamala R, Atulya AR, et al. The first state-level public health program for obstructive airway disease in India: An early field-level evaluation. J Fam Med Prim Care. 2020;9(9):4998–5003. 33. 33.DHS Kerala.NCD-Bulletin-January-2021.pdf. Available from: https://dhs.kerala.gov.in/wp-content/uploads/2021/01/NCD-BULLETIN-latest-January-2021.pdf. [Cited 20 Jan 2023]. 34. 34.NHM Kerala. Fact sheet June 2022. Thiruvananthapuram: Arogyakeralam; 2022. Available from: https://arogyakeralam.gov.in/wp-content/uploads/2020/03/NHM-Fact-sheet-june-2022_.pdf. [Cited 17 Jan 2023]. 35. 35.NHSRC. NQAS Certification progress in STATE & UT. Available from: http://qi.nhsrcindia.org/sites/default/files/NQAS%20Progress%20Presentation-Day%201.pdf. [Cited 10 Jun 2022]. 36. 36.Nair S, L S, A S, Mathew OP. Enablers and Barriers of NQAS accreditation of Hospitals in Kerala. Thiruvananthapuram: Population Research Centre Kerala; 2019. Available from: http://qi.nhsrcindia.org/sites/default/files/Report%20on%20Enablers%20and%20Barriers%20of%20NQAS%20accreditation%20of%20Hospitals_%20PRC%20Kerala.PDF. [Cited 10 Jun 202]. 37. Vallikunnu V, Kumar SG, Sarkar S, Kar SS, Harichandrakumar KT. A Qualitative Study on Working Experience of Rural Doctors in Malappuram District of Kerala. India J Fam Med Prim Care (2014.0) 3 141-145. DOI: 10.4103/2249-4863.137643 38. 38.Heller P. A virus, social democracy, and dividends for Kerala. The Hindu. 2020 Apr 17; Available from: https://www.thehindu.com/opinion/lead/a-virus-social-democracy-and-dividends-for-kerala/article31370554.ece. [Cited 20 Jan 2023]. 39. Israelsen S, Malji A. COVID-19 in India: A Comparative Analysis of the Kerala and Gujarat Development Models’ Initial Responses. Prog Dev Stud (2021.0) 21 397-418. DOI: 10.1177/14649934211030462 40. Jalan J, Sen A. Containing a pandemic with public actions and public trust: the Kerala story. Indian Econ Rev (2020.0) 55 105-124. DOI: 10.1007/s41775-020-00087-1 41. 41.Antinyan A, Bassetti T, Corazzini L, Pavesi F. Trust in the Health System and COVID-19 Treatment. Front Psychol. 2021;12:643758. Available from: https://www.frontiersin.org/articles/10.3389/fpsyg.2021.643758. [Cited 20 Jan 2023]. 42. Almutairi AF, BaniMustafa A, Alessa YM, Almutairi SB, Almaleh Y. Public Trust and Compliance with the Precautionary Measures Against COVID-19 Employed by Authorities in Saudi Arabia. Risk Manag Healthc Policy (2020.0) 8 753-760. DOI: 10.2147/RMHP.S257287 43. 43.Lahariya C. Health & Wellness Centers to Strengthen Primary Health Care in India: Concept, Progress and Ways Forward. Indian J Pediatr. 2020;1–14. 44. 44.The New Indian Express. Aardram Mission: 1,000 posts sanctioned- Thiruvananthapuram. Available from: https://www.newindianexpress.com/states/kerala/2019/jul/04/aardram-mission-1000-posts-sanctioned-1999098.html. [Cited 18 Sept 2020]. 45. Levesque JF, Breton M, Senn N, Levesque P, Bergeron P, Roy DA. The Interaction of Public Health and Primary Care: Functional Roles and Organizational Models that Bridge Individual and Population Perspectives. Public Health Rev (2013.0) 35 14. DOI: 10.1007/BF03391699 46. Dodd R, Palagyi A, Jan S, Abdel-All M, Nambiar D, Madhira P. Organisation of primary health care systems in low- and middle-income countries: review of evidence on what works and why in the Asia-Pacific region. BMJ Glob Health (2019.0) 4 e001487. DOI: 10.1136/bmjgh-2019-001487 47. 47.Staff reporter The Hindu. Minister opens 1,603 upgraded health centres. Thiruvananthapuram. 2021. Available from: https://www.thehindu.com/news/national/kerala/minister-opens-1603-upgraded-health-centres/article33853170.ece. [Cited 10 Jun 2022]. 48. Franke RW, Chasin BH. Kerala State, India: Radical Reform as Development. Int J Health Serv (1992.0) 22 139-156. DOI: 10.2190/HMXD-PNQF-2X2L-C8TR 49. 49.Bhagyalakshmi. Utilization of noncommunicable disease services provided by public health facilities in Kasaragod, Kerala.. Available from: https://www.amhsjournal.org/article.asp?issn=2321-4848;year=2019;volume=7;issue=1;spage=18;epage=24;aulast=Bhagyalakshmi. [Cited 10 Jun 2022]. 50. Elamon J, Franke RW, Ekbal B. Decentralization of Health Services: The Kerala People’s Campaign. Int J Health Serv (2004.0) 34 681-708. DOI: 10.2190/4L9M-8K7N-G6AC-WEHN 51. Babu GR. Democratic Decentralization and People’s Participation in Health Projects in Kerala. J Health Develop (2007.0) 3 12 52. Azeez EPA, Anbuselvi G. Is the Kerala Model of Community-Based Palliative Care Operations Sustainable? Evidence from the Field. Indian J Palliat Care (2021.0) 27 18-22. DOI: 10.4103/IJPC.IJPC_45_20 53. Rezapour R, Dorosti AA, Farahbakhsh M, Azami-aghdash S, Iranzad I. The impact of the Covid-19 pandemic on primary health care utilization: an experience from Iran. BMC Health Serv Res (2022.0) 22 404. DOI: 10.1186/s12913-022-07753-5 54. Zhang T, Shen X, Liu R, Zhao L, Wang D, Lambert H. The impact of COVID-19 on primary health care and antibiotic prescribing in rural China: qualitative study. BMC Health Serv Res (2021.0) 21 1048. DOI: 10.1186/s12913-021-07082-z 55. Chathukulam J, Tharamangalam J. The Kerala model in the time of COVID19: Rethinking state, society and democracy. World Dev (2021.0) 137 105207. DOI: 10.1016/j.worlddev.2020.105207
--- title: 'The effect of linagliptin on microalbuminuria in patients with diabetic nephropathy: a randomized, double blinded clinical trial' authors: - Mozhgan Karimifar - Jamileh Afsar - Massoud Amini - Firouzeh Moeinzadeh - Awat Feizi - Ashraf Aminorroaya journal: Scientific Reports year: 2023 pmcid: PMC9975829 doi: 10.1038/s41598-023-30643-7 license: CC BY 4.0 --- # The effect of linagliptin on microalbuminuria in patients with diabetic nephropathy: a randomized, double blinded clinical trial ## Abstract The aim of the present study was to investigate the effect of linagliptin on microalbuminuria in patients with diabetic nephropathy (DN). The present double-blind randomized placebo-controlled clinical trial was performed on 92 patients with DN who were divided into two groups. The intervention and control groups received linagliptin 5 mg and placebo for 24 weeks, respectively. Blood pressure, lipid profile, liver enzymes, fasting plasma glucose (FPG), and urine albumin-creatinine ratio (UACR) were assessed and recorded before, 12 weeks, and 24 weeks after the beginning of the intervention. The mean value of UACR decrease was significant over time in both groups, with higher decrease in linagliptin group, however, the differences between two groups were not, statistically significant ($P \leq 0.05$). However, the percentage of improvement in microalbuminuria (UACR < 30 mg/g) in the linagliptin group was significantly higher than that of the control group during 24 weeks of intervention ($68.3\%$ vs. $25\%$; P-value < 0.001). There was no statistically significant difference in the mean value of the UACR and other parameters between linagliptin treated and placebo treated patients with diabetic nephropathy. Further studies, with longer periods of follow-up are suggested to examine these patients’ renal outcomes. ## Introduction Diabetes is one of the most common metabolic diseases, and its prevalence is increasing in adults, especially in developing countries such as Iran1. Type 2 diabetes (T2D) has been recognized as one of the most significant risk factors for microvascular and macrovascular diseases2,3. Diabetic nephropathy can be regarded as one of the main chronic microvascular complications in patients with T2D, occurs in about $35\%$ of patients with diabetes, and is the most common cause of end-stage renal disease (ESRD) and death from cardiovascular diseases4. Diabetic nephropathy can eventually lead to chronic kidney disease (CKD) and ESRD; we also know that diabetic patients undergoing hemodialysis have more complications than non-diabetic patients undergoing hemodialysis do. The risk of cardiovascular complications is higher in diabetic patients with albuminuria. Hyperglycemia is the main pathogenesis of diabetic nephropathy. The exact mechanism of diabetic nephropathy is unknown, however, factors such as, angiotensin II, growth factors, endothelin, advanced glycation end products [AGEs]), glomerular hyper filtration or hyper perfusion lead to an increased glomerular capillary pressure and structural changes in the glomerulus5. Early diagnosis of microalbuminuria and its control by lowering plasma glucose and blood pressure and administering angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers can prevent the progression of diabetic nephropathy6. Dipeptidyl peptidase-4 (DPP-4) inhibitors, also known as gliptins, are a novel class of oral hypoglycemic agents used to treat T2D7,8. These medications work by increasing the active levels of incretin peptides such as glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide9. GLP-1 and other incretins increase the secretion of glucose-dependent insulin10. DPP-4, however, rapidly inactivates these peptides and reduces their effect on glucose balance. DPP-4 inhibitors increase the function of incretins by delaying their breakdown. Linagliptin is a new DPP-4 inhibitor that has been approved in 2011 as a hypoglycemic drug in the United States, Europe, and Japan11. Linagliptin is the only DPP-4 inhibitor that do not need dose adjustment in patient with reduced renal function. The molecular structure of this medication is based on xanthine, which is different from other DPP-4 inhibitors10. Because linagliptin has a long half-life (more than 184 h) and a stable inhibitory effect on DPP4, it can be administered once daily12,13. The results of previous studies indicated that linagliptin, as a monotherapy or in combination with other hypoglycemic drugs, had good safety and tolerability and improved glycemic index14–16. Some studies have reported that, it can improve the renal function, reduce oxidative stress, reduce glomerular sclerosis, and reduce albuminuria17–20. However, some other studies have not confirmed the effect of this drug on the reduction of albuminuria21. Therefore, considering the limited and contradictory clinical evidence reporting the therapeutic effects of DPP-4 inhibitors, especially linagliptin, on diabetic nephropathy in patients with T2D, this study aimed to investigate the effect of linagliptin on microalbuminuria as a key step in preventing the progression of diabetic nephropathy in patients with T2D. ## Design of the study and participants This double-blind, randomized, placebo-controlled clinical trial was performed on 92 patients with T2D and nephropathy that referred to the Isfahan Endocrine and Metabolism Research Center from November 2019 to April 2021. Inclusion criteria consisted of T2D patients age ≥ 18 years, and microalbuminuria (urine albumin-creatinine ratio (UACR) of 30–300 mg/g (in three urine samples collected consecutively over two weeks before the beginning of the study) with or without GFR reduction (less than 60)), glycated hemoglobin (HbA1c) level of 6.5–$10\%$ (48–86 mmol/mol), body mass index (BMI) of less than 40 kg/m2. Furthermore, patients who were taking short-acting insulins, rosiglitazone, pioglitazone, GLP-1 receptor analogues, sodium glucose co-transporter 2 inhibitors or anti-obesity drugs within three months before the beginning of the study, those with a history of myocardial infarction, stroke, or transient ischemic attack within 6 months of the beginning of the study, patients who had non-diabetic renal failure or urinary tract infection, or those who had received a kidney transplant were not included in the study. They were excluded the study, in the case of not cooperating, not attending in the follow-up sessions, or showing drug-induced complications. ## The process of implementing interventions and measuring research variables The study was approved by the Isfahan University of Medical Sciences ethics committee (Approval no. IR.MUI.MED.REC.1397.230), and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients. The study protocol was registered at irct.ir as IRCT20171030037093N11 (https://irct.ir/trial/39062). Ninety two eligible patients were selected using convenience sampling method. Then, these patients were divided into two groups using random allocation software. At the beginning of the study, patients’ demographic and clinical information including sex, age, BMI, waist circumference (WC), systolic blood pressure (SBP), diastolic blood pressure (DBP), comorbidities, duration of T2D, history of drug use (lipid-lowering drugs, hypoglycemic drugs, and antihypertensive drugs), biochemical parameters including urine creatinine (Urine Cr), urine albumin (Urine Alb), UACR, Hemoglobin A1c (HbA1c), fasting plasma glucose (FPG), serum creatinine and glomerular filtration rate (GFR) by MDRD formula, lipid profile including triglycerides (TG), high-density lipoprotein (HDL), low-density lipoprotein (LDL), and total cholesterol, and liver enzymes including alanine aminotransferase (ALT), aspartate aminotransferase (AST), and alkaline phosphatase (ALP) were recorded. For all patients, a routine diabetes treatment was prescribed according to the standard protocol of American Diabetes association (ADA) statement. In addition, 5 mg of linagliptin was administered to patients in the intervention group for 24 weeks while patients in the control group received placebo for 24 weeks. It should be mentioned that in order to comply with the double-blind condition, linagliptin and placebo which had been already prepared by Alhavi Pharmaceutical Company, located in Tehran, Iran, in the same shape, size, and color. Starch had been used to make placebo tablets. The prepared drugs had been coded, and provided to the researcher. Therefore, the researcher, patients, the information evaluator, and the statistical analyst had no knowledge of the type of the intervention performed in the two groups. Furthermore, all patients were requested to follow the healthy dietary patterns and proper physical activity in the treatment process to control these factors as much as possible and prevent the disruptive effect of patients’ eating habits and physical activity on the results of the study. Patients were evaluated in terms of the complications of the medication two weeks after the beginning of the intervention and then every 4 weeks. In addition, patients’ anthropometric, blood pressure, and biochemical factors were assessed 12 weeks and 24 weeks after the intervention. To perform accurate measurements and evaluations, data collection was performed by a single specialist technician, and all biochemical tests were performed only in the Laboratory of Isfahan Endocrine and Metabolism Research Center. ## Statistical analysis The collected data were analyzed by SPSS software (ver.26) (IBM SPSS Statistics for Windows, Armonk, NY: IBM Corp.). Quantitative and qualitative variables were reported as means ± standard deviation (SD) and number (percentage), respectively. Kolmogorov–Smirnov test and Q-Q diagram were used to check the normality of data distribution. Basic quantitative and qualitative variables of study participants were compared two groups using independent samples t-test and Chi-squared test, respectively. Moreover, one-way repeated measures analysis of variance (ANOVA) was used for intra-group and inter-group comparisons to evaluate the mean changes of quantitative variables over 24 weeks from the beginning of the intervention. Mauchly’s test was used to evaluate the sphericity hypothesis, and if it was not established, the multivariate analysis was used. We used Bonferroni post hoc test for doing pairwise comparisons between time points as well as for comparing two groups at each time point. The significance level of less than 0.05 was considered in all analyses. ## Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The study was approved by the ethics Committee of the Isfahan University of Medical Sciences (Approved code: IR.MUI.MED.REC.1397.230) and its clinical trial code is recorded (IRCT20171030037093N11). ## Informed consent Written informed consent was obtained from all patients for precipitation and registration. ## Results Among 213 eligible patients, 92 patients were finally enrolled (46 patients in linagliptin group and 46 patients in control group). Five patients from the linagliptin group (3 patients due to their confirmed COVID-19 infection and 2 patients due to their mild skin complications had withdrawal from the study) and 10 patients from the control group (4 patients due to their confirmed COVID-19 infection and 6 patients due to their non-attendance in subsequent follow-ups) were excluded from the study (Fig. 1). Finally, 41 patients in the linagliptin group and 36 patients in the control group continued the study until the end of the trial. Basal characteristics of both groups are mentioned in Table 1. There was no significant difference between the two groups in terms of age, sex, BMI, comorbidities, and used medications (Table 1).Figure 1Consort flow diagram for recruitment of patients. Table 1Patients’ basic characteristics in the two groups. CharacteristicsLinagliptin group ($$n = 41$$)Control group ($$n = 36$$)P-value*Age; year57.56 ± 6.5557.53 ± 5.770.981Sex Male11 ($26.8\%$)11 ($30.6\%$)0.718 Female30 ($73.2\%$)25 ($69.4\%$)BMI; kg/m230.39 ± 17.5529.76 ± 4.750.865Comorbidity None19 ($46.3\%$)12 ($33.4\%$)0.281 HTN21 ($51.2\%$)21 ($58.3\%$) Hypothyroidism1 ($2.5\%$)3 ($8.3\%$)Medication Anti-diabetic26 ($63.4\%$)23 ($63.9\%$) Basal insulin15 ($36.6\%$)13 ($36.1\%$)0.966 ACEI3 ($7.3\%$)5 ($13.9\%$) ARB28 ($68.3\%$)23 ($63.9\%$) B-blocker5 ($12.2\%$)3 ($8.3\%$) CCB2 ($4.9\%$)5 ($13.9\%$) Diuretic3 ($7.3\%$)3 ($8.3\%$) Statin35 ($85.4\%$)30 ($83.3\%$) Aspirin25 ($64.1\%$)19 ($52.8\%$)Data are shown as n (%) or mean ± SD for categorical and continuous variables, respectively. ACEI angiotensin-converting enzyme inhibitors, ARB angiotensin receptor blockers, CCB calcium channel blockers.Each patient might take more than one drug.Resulted from independent samples t-test for continuous and chi-squared test for categorical variables. The patients’ mean weight, WC, and BMI did not differ significantly between the two groups before, 12 weeks, and 24 weeks after the intervention (Table 2). The changes in anthropometric variables were not significant over time in any of the groups (PTime > 0.05) and between the two groups. Furthermore, mean values of anthropometric measures were not significantly different between the two groups in any follow up time points (Pgroup > 0.05). Moreover, the interactive effect of time and intervention was not significant (Ptimegroup > 0.05) (Table 2).Table 2Patients’ anthropometric parameters in two groups. VariablesBaseline12 weeks24 weeksRepeated measures ANOVAPtimePgroupPtimegroupWeight; kg Linagliptin group ($$n = 41$$)74.66 ± 13.6474.76 ± 13.9774.19 ± 13.440.2900.6670.302 Control group ($$n = 36$$)75.24 ± 11.3275.67 ± 10.9376.44 ± 10.780.447 P0.7510.7590.437WC; cm Linagliptin group ($$n = 41$$)98.80 ± 18.7498.86 ± 18.5298.16 ± 18.200.5010.9320.528 Control group ($$n = 36$$)98.80 ± 14.1499.50 ± 14.5196.23 ± 20.190.291 P0.8850.8720.719BMI; kg/m2 Linagliptin group ($$n = 41$$)30.39 ± 17.5527.95 ± 5.1230.72 ± 16.720.5330.9170.613 Control group ($$n = 36$$)29.76 ± 4.7530.03 ± 4.4729.90 ± 4.260.305 P0.8650.0710.780Data are shown as mean ± SD.WC waist circumference, BMI body mass index. P is obtained from an independent samples t-test conducted for comparing the mean of the variables between the two groups at each follow-up time point. Pgroup shows the overall difference between the two groups over the follow-up period. Ptime indicates the changes in the mean of the variables in the intervention and control groups over the follow-up period. Ptimegroup presents the interaction between the time and intervention. Pgroup, Ptime, and Ptimegroup are obtained from repeated measures ANOVA. The results of the repeated measures ANOVA revealed that although SBP (PTime = 0.021) and DBP (PTime = 0.010) decreased significantly over 24 weeks after the beginning of the intervention in the linagliptin group, however its changes were not significantly different between the two groups over time (Pgroup > 0.05). We did not observe significantly interactive effect of time and intervention Ptimegroup = 0.324) (Table 3).Table 3Parameters of diabetic patient's blood pressure, plasma glucose, lipid profile, and liver enzymes in linagliptin and placebo treated groups. VariablesBaseline12 weeks24 weeksRepeated measures ANOVAPtimePgroupPtimegroupSBP; mmHg Linagliptin group ($$n = 41$$)125.67 ± 8.51122.84 ± 7.41121.35 ± 6.840.021#0.1220. 324* Control group ($$n = 36$$)117.48 ± 21.59121.28 ± 6.10122.71 ± 7.980.195 P0.021#0.3370.438DBP; mmHg Linagliptin group ($$n = 41$$)77.32 ± 7.9175.27 ± 6.4572.16 ± 7.500.010#0.8260.273 Control group ($$n = 36$$)72.36 ± 10.3173.14 ± 6.3172.71 ± 6.680.962 P0.020#0.1620.743FPG; mg/dL Linagliptin group ($$n = 41$$)156.63 ± 39.46145.97 ± 33.97136.89 ± 31.48 < 0.001#0.8390.670 Control group ($$n = 36$$)153.94 ± 34.47144.03 ± 34.79138.83 ± 34.010.003# P0.7530.8110.803HbA1c; % Linagliptin group ($$n = 41$$)8.17 ± 1.027.88 ± 1.107.54 ± 1.03 < 0.001#0.8930.921 Control group ($$n = 36$$)8.22 ± 1.107.81 ± 1.217.54 ± 1.30 < 0.001# P0.8060.7990.999TG; mg/dL Linagliptin group ($$n = 41$$)195.05 ± 67.05180.24 ± 50.61162.97 ± 46.290.01#0.1410.346 Control group ($$n = 36$$)213.47 ± 96.23196.37 ± 80.93197.74 ± 85.010.309 P0.3280.3110.033#Cholesterol; mg/dL Linagliptin group ($$n = 41$$)175.75 ± 31.76168.75 ± 33.18166.97 ± 35.330.0820.5360.519 Control group ($$n = 36$$)178.41 ± 39.97173.51 ± 34.19175.54 ± 36.680.665 P0.7460.5510.316LDL; mg/dL Linagliptin group ($$n = 41$$)87.66 ± 24.0280.05 ± 26.1286.98 ± 24.020.0800.6320.548 Control group ($$n = 36$$)87.57 ± 26.5284.54 ± 19.4690.48 ± 22.810.285 P0.9880.4130.528HDL; mg/dL Linagliptin group ($$n = 41$$)45.22 ± 10.1544.81 ± 9.4543.73 ± 10.470.6900.7500.083 Control group ($$n = 36$$)41.64 ± 8.4545.57 ± 10.1144.51 ± 9.590.013# P0.1000.7420.743ALT; U/L Linagliptin group ($$n = 41$$)18.73 ± 9.3122.02 ± 10.9121.13 ± 11.030.080.5530.906 Control group ($$n = 36$$)20.64 ± 9.6223.40 ± 10.2221.97 ± 10.600.169 P0.3800.5840.744AST; U/L Linagliptin group ($$n = 41$$)21.48 ± 6.8621.19 ± 7.6021.78 ± 9.840.8410.6030.941 Control group ($$n = 36$$)22.66 ± 7.3122.34 ± 8.6822.63 ± 9.720.941 P0.4680.5500.715ALP; U/L Linagliptin group ($$n = 41$$)226.44 ± 57.10235.38 ± 50.68230.243 ± 65.300.7510.032#0.992 Control group ($$n = 36$$)205.97 ± 46.16211.23 ± 42.95205.653 ± 46.460.473 P0.0910.033#0.071SBP systolic blood pressure, DBP diastolic blood pressure, FPG fasting plasma glucose, HbA1C hemoglobin A1c, TG triglycerides, LDL low-density lipoprotein, HDL high-density lipoprotein, ALT alanine aminotransferase, AST aspartate aminotransferase, ALP alkaline phosphatase. P is obtained from an independent samples t-test conducted for comparing the mean of the variables between the two groups at each follow-up time point. Pgroup indicates the overall difference between the two groups over the follow-up period. Ptime shows the changes in the mean of the variables in the intervention and control groups separately over the follow-up period. Ptimegroup demonstrates the interaction between the time and intervention. Pgroup, Ptime, and Ptimegroup was obtained from repeated measures ANOVA.Significant values are given in bold.#Statistically significant results.If the baseline values had a significant difference between the two groups, it was considered as a confounder and it was adjusted in the repeated measures ANOVA. Although, the mean values of FPG, HbA1c, and TG (not significantly decreased in control group) in each of two groups decreased significantly over follow up time (Ptime < 0.05), however the mean change of these variables were not significantly different between two groups (Pgroup > 0.05). The interactive effect of time and intervention was not statistically significant (Ptimegroup > 0.05). Also, the mean value of FPG and HbA1c was not significantly different between two groups in none of follow up time point ($P \leq 0.05$) and mean value of TG at 24 weeks after intervention in linagliptin group was significantly lower than control group ($P \leq 0.05$) (Table 3). Other parameters of lipid profile and liver enzymes were not found to be significantly different in inter-group (Ptime > 0.05) and intra-group comparisons ($P \leq 0.05$). Twelve weeks after intervention, only the liver enzyme ALP in the linagliptin group with the mean of 235.38 ± 50.68 U/L was significantly higher than the control group with the mean of 211.23 ± 42.95 U/L. Therefore, the results of the repeated measures ANOVA indicated that the effect of intervention (Pgroup = 0.032) was significant in this variable, and the interactive effect of time and intervention (Ptimegroup > 0.05) was not statistically significant (Table 3). Finally, none of the parameters related to the renal function and microalbuminuria were significantly different between the two groups before, 12 weeks, and 24 weeks after the intervention ($P \leq 0.05$). However, GFR decrease and Cr increase were significant in the linagliptin group over time (Ptime < 0.05). But the mean change of these variables were not significantly different between two groups (Pgroup > 0.05) and the interactive effect of time and intervention (Ptimegroup > 0.05) also was not significant (Table 4).Table 4Parameters related to renal function and microalbuminuria in two groups. VariablesBaseline12 weeks24 weeksRepeated measures ANOVAPtimePgroupPtimegroupGFR Linagliptin group ($$n = 41$$)68.44 ± 11.7165.12 ± 11.1563.55 ± 9.910.034#0.3200.632 Control group ($$n = 36$$)65.22 ± 12.1662.32 ± 12.0762.51 ± 11.290.238 P0.2410.3090.679Cr; mg/dl Linagliptin group ($$n = 41$$)0.93 ±.0.140.97 ± 0.120.99 ± 0.130.035#0.7150.804 Control group ($$n = 36$$)1.00 ± 0.171.02 ± 0.191.03 ± 0.140.371 P0.0580.1780.196Urine Cr; g/dl Linagliptin group ($$n = 41$$)85.44 ± 34.2291.15 ± 30.9094.91 ± 47.040.6270.2270.129 Control group ($$n = 36$$)90.55 ± 37.8078.81 ± 36.7079.51 ± 37.810.101 P0.5350.1270.132Urine Alb; mg/dl Linagliptin group ($$n = 41$$)7.47 ± 4.493.98 ± 2.072.72 ± 2.72 < 0.001#0.9900.027 Control group ($$n = 36$$)6.52 ± 5.044.29 ± 3.163.45 ± 3.01 < 0.001# P0.3840.6160.285UACR; mg/g Linagliptin group ($$n = 41$$)86.59 ± 47.6845.93 ± 26.2731.29 ± 31.69< 0.001#0.778< 0.001 Control group ($$n = 36$$)74.02 ± 43.6454.93 ± 33.9645.49 ± 30.55< 0.001# P0.2340.2110.057Cr creatinine, GFR glomerular filtration rate test, UrineAlb urine albumin, UrineCr urine creatinine, UACR urine albumin-creatinine ratio. P is obtained from an independent samples t-test conducted for comparing the mean of the variables between the two groups at each follow-up time point. Pgroup indicates the overall difference between the two groups over the follow-up period. Ptime demonstrates the changes in the mean of the variables in the intervention and control groups over the follow-up period. Ptimegroup indicates the interaction between the time and intervention. Pgroup, Ptime, and Ptimegroup are obtained from repeated measures ANOVA.Significant values are given in bold.#Statistically significant results. In addition, although the decrease in urine albumin and UACR was significant in both groups over time (Ptime < 0.05), however, the mean change of these variables was not significantly different between two groups (Pgroup > 0.05). The interactive effect of time and intervention was significant in these two variables (Ptimegroup < 0.05); (Table 4). Mean values of the renal function and microalbuminuria were not significantly different in each follow up time points ($P \leq 0.05$). The percentage of improvement in microalbuminuria (UACR < 30) in the linagliptin group was significantly higher than that of the control group during 24 weeks of intervention ($68.3\%$ vs. $25\%$; P-value < 0.001) (Fig. 2).Figure 2Percentage of improvement of microalbuminuria (UACR < 30 mg/dl) in the two groups. It should be noted that linagliptin adverse effects such as hypoglycemia or acute pancreatitis were not observed. ## Discussion Since there were several studies on the effectiveness or ineffectiveness of linagliptin on albumin excretion in urine21,24, we decided to investigate this issue in a controlled study. In our study, linagliptin did not cause a significant decrease in urinary albumin excretion in diabetic patients compared to the control group. The results of a recent human study also indicated that linagliptin had no significant effect on reducing albuminuria. The duration of this study was 24 weeks like our study. These researchers believe that a longer-term treatment is needed to determine the renal effects of this drug21. In our study, according to the interactive effect of time and intervention, and the greater improvement of UACR (< 30 mg/g), in the linagliptin group ($68.3\%$) than in the control group ($25\%$) (Fig. 2), it seems that we need a longer intervention to investigate the effect of linagliptin on albuminuria. Interestingly, other studies have reported the non-albuminuric protective effects of linagliptin22,23. Therefore, more research is required to elucidate the renal biology and pathophysiology of DPP-4 in this regard. On the other hand, FPG and HbA1C had a significant decrease in both linagliptin and control groups. A study, conducted by Ito et al. indicated that the mean level of HbA1C in the two groups with and without receiving linagliptin was not significantly different24. In another study, the levels of HbA1C were not significantly different between the intervention and placebo groups25. In contrast with the results of our study, Groop et al. research revealed that the linagliptin significantly improved the glycemic control in patients with type-2 DM21. However, the effect of other factors such as participants' degree of adherence to the diabetic diet and their level of physical activity cannot be ignored as they may play a role in controlling plasma glucose. The results of the present study revealed that triglyceride were significantly reduced in each group over 24 weeks from the beginning of the intervention, however the decreases were not significantly different between two groups. Howevere, the results of Monami et al. study showed that DPP-4 inhibitors had a reducing effect on triglycerides26. According to a meta-analysis, a reducing effect of combination therapy of DPP-4 inhibitors and metformin on triglycerides and cholesterol was reported27. In the present study, cholesterol change was not significantly different between the two groups. In addition to our study, among liver enzymes, ALP was significantly higher in the linagliptin group as compared with the control group 12 weeks after the intervention. In this respect, the results of the repeated measures ANOVA indicated the significant effect of the intervention. It is important to note that the increase in ALP was not significant between two groups 24 weeks after the beginning of the intervention. The reason for the transient increase in alkaline phosphatase in the twelfth week of the study in the linagliptin group may be due to the effect of linagliptin on bone metabolism or indicate hepatic cholestasis. As we had not measured GGT (Gama-Glutamyl Transferase), we cannot determine the alkaline phosphataseʼ origin. However, the effects of DPP-4 inhibitors, including linagliptin, on bone metabolism are still unknown. One study by Kanda et al. showed a protective effect of linagliptin on the bones of diabetic rats28. However, as the increase in ALP was not significant between the two groups 24 weeks after the beginning of the intervention, it seems that the increased in ALP is not an important issue. The results of the evaluation of renal function factors as well as microalbuminuria also showed that although there was a significant decrease in GFR and a significant increase in Cr in the linagliptin group, the effect of the intervention and the interactive effect of time and intervention were still not significant. Moreover the rate of decrease in GFR and increase in creatinine was not more than $30\%$ and the effect of the intervention was not significant, it is necessary to examine this finding with more participants and longer follow-up period. In this regard, Nishida et al. conducted a study to investigate the effects of DPP-4 inhibition on diabetic patients and noted a negative brief effect on creatinine23. However, a study by Rosenstock et al. found that linagliptin was safe for the kidneys12. It should be noted that this study was also associated with some limitations and strengths. The investigation area of the present study, which was the evaluation of the effect of linagliptin on microalbuminuria in patients with diabetic nephropathy, can be considered novel as few studies have been performed with this aim. The mentioned point can be regarded as one of the strong points of this study. However, this study was only been performed on patients with diabetic nephropathy (microalbuminuria with mild reduced GFR), so it is imprecise whether the findings can be generalized to patients with more advanced diabetic kidney diseases. In addition, the small number of participants and the short follow-up period can be regarded as another limitation of this study. ## Conclusion Linagliptin did not cause a significant decrease in urinary albumin excretion in diabetic patients with nephropathy compared to the control group, however, the percentage of improvement in microalbuminuria (UACR < 30 mg/g) in the linagliptin group was significantly higher than that of the control group during 24 weeks of intervention. We recommend more studies with longer periods of follow-up to examine the renal outcome of linagliptin in patients with diabetic nephropathy. ## References 1. Guariguata L, Whiting DR, Hambleton I, Beagley J, Linnenkamp U, Shaw JE. Global estimates of diabetes prevalence for 2013 and projections for 2035. Diabetes Res. Clin. Pract. (2014) 103 137-149. DOI: 10.1016/j.diabres.2013.11.002 2. Sarwar N, Gao P, Seshasai SR, Gobin R, Kaptoge S, Di Angelantonio E. Diabetes mellitus, fasting blood glucose concentration, and risk of vascular disease: A collaborative meta-analysis of 102 prospective studies. Lancet (London, England). (2010) 375 2215-2222. DOI: 10.1016/S0140-6736(10)60484-9 3. Mirahmadizadeh A, Delam H, Seif M, Banihashemi SA, Tabatabaee H. Factors affecting insulin compliance in patients with type 2 diabetes in South Iran, 2017: We are faced with insulin phobia. Iran. J. Med. Sci. (2019) 44 204-213. PMID: 31182886 4. Hadjadj S, Cariou B, Fumeron F, Gand E, Charpentier G, Roussel R. Death, end-stage renal disease and renal function decline in patients with diabetic nephropathy in French cohorts of type 1 and type 2 diabetes. Diabetologia (2016) 59 208-216. DOI: 10.1007/s00125-015-3785-3 5. Powers AC, Stafford JM, Rickels MR, Loscalzo J, Fauci A, Kasper D, Hauser S, Longo D, Jameson JL. Diabetes mellitus: Complications. Harrison's principles of internal medicine (2022) 6. Papadopoulou-Marketou N, Chrousos GP, Kanaka-Gantenbein C. Diabetic nephropathy in type 1 diabetes: A review of early natural history, pathogenesis, and diagnosis. Diab. Metab. Res. Rev. (2017) 33 1. DOI: 10.1002/dmrr.2841 7. Ahrén B. Gut peptides and type 2 diabetes mellitus treatment. Curr. Diab. Rep. (2003) 3 365-372. DOI: 10.1007/s11892-003-0079-9 8. Deacon CF. Dipeptidyl peptidase 4 inhibitors in the treatment of type 2 diabetes mellitus. Nat. Rev. Endocrinol. (2020) 16 642-653. DOI: 10.1038/s41574-020-0399-8 9. Cheang JY, Moyle PM. Glucagon-like peptide-1 (GLP-1)-based therapeutics: Current status and future opportunities beyond type 2 diabetes. ChemMedChem (2018) 13 662-671. DOI: 10.1002/cmdc.201700781 10. Glotfelty EJ, Olson L, Karlsson TE, Li Y, Greig NH. Glucagon-like peptide-1 (GLP-1)-based receptor agonists as a treatment for Parkinson's disease. Expert Opin. Investig. Drugs (2020) 29 595-602. DOI: 10.1080/13543784.2020.1764534 11. Johansen OE, Neubacher D, von Eynatten M, Patel S, Woerle HJ. Cardiovascular safety with linagliptin in patients with type 2 diabetes mellitus: A pre-specified, prospective, and adjudicated meta-analysis of a phase 3 programme. Cardiovasc. Diabetol. (2012) 11 3. DOI: 10.1186/1475-2840-11-3 12. Rosenstock J, Perkovic V, Johansen OE, Cooper ME, Kahn SE, Marx N. Effect of linagliptin vs placebo on major cardiovascular events in adults with type 2 diabetes and high cardiovascular and renal risk: The Carmelina randomized clinical trial. JAMA (2019) 321 69-79. DOI: 10.1001/jama.2018.18269 13. Deacon CF, Holst JJ. Linagliptin, a xanthine-based dipeptidyl peptidase-4 inhibitor with an unusual profile for the treatment of type 2 diabetes. Expert Opin. Investig. Drugs (2010) 19 133-140. DOI: 10.1517/13543780903463862 14. Thrasher J, Daniels K, Patel S, Whetteckey J, Woerle HJ. Efficacy and safety of linagliptin in Black/African American patients with type 2 diabetes: A 6-month, randomized, double-blind, placebo-controlled study. Endocrine Pract. (2014) 20 412-420. DOI: 10.4158/EP13365.OR 15. Inagaki N, Watada H, Murai M, Kagimura T, Gong Y, Patel S. Linagliptin provides effective, well-tolerated add-on therapy to pre-existing oral antidiabetic therapy over 1 year in Japanese patients with type 2 diabetes. Diabetes Obes. Metab. (2013) 15 833-843. DOI: 10.1111/dom.12110 16. Bajaj M, Gilman R, Patel S, Kempthorne-Rawson J, Lewis-D'Agostino D, Woerle HJ. Linagliptin improved glycaemic control without weight gain or hypoglycaemia in patients with type 2 diabetes inadequately controlled by a combination of metformin and pioglitazone: a 24-week randomized, double-blind study. Diabetic Med. J. Br. Diab. Assoc. (2014) 31 1505-1514. DOI: 10.1111/dme.12495 17. Mega C, de Lemos ET, Vala H, Fernandes R, Oliveira J, Mascarenhas-Melo F. Diabetic nephropathy amelioration by a low-dose sitagliptin in an animal model of type 2 diabetes (Zucker diabetic fatty rat). Exp. Diabetes Res. (2011) 2011 162092. DOI: 10.1155/2011/162092 18. Liu WJ, Xie SH, Liu YN, Kim W, Jin HY, Park SK. Dipeptidyl peptidase IV inhibitor attenuates kidney injury in streptozotocin-induced diabetic rats. J. Pharmacol. Exp. Ther. (2012) 340 248-255. DOI: 10.1124/jpet.111.186866 19. Panchapakesan U, Mather A, Pollock C. Role of GLP-1 and DPP-4 in diabetic nephropathy and cardiovascular disease. Clin. Sci. (2013) 124 17-26. DOI: 10.1042/CS20120167 20. Alter ML, Ott IM, von Websky K, Tsuprykov O, Sharkovska Y, Krause-Relle K. DPP-4 inhibition on top of angiotensin receptor blockade offers a new therapeutic approach for diabetic nephropathy. Kidney Blood Press. Res. (2012) 36 119-130. DOI: 10.1159/000341487 21. Groop PH, Cooper ME, Perkovic V, Hocher B, Kanasaki K, Haneda M. Linagliptin and its effects on hyperglycaemia and albuminuria in patients with type 2 diabetes and renal dysfunction: The randomized MARLINA-T2D trial. Diabetes Obes. Metab. (2017) 19 1610-1619. DOI: 10.1111/dom.13041 22. Fadini GP, Bonora BM, Cappellari R, Menegazzo L, Vedovato M, Iori E. Acute effects of linagliptin on progenitor cells, monocyte phenotypes, and soluble mediators in type 2 diabetes. J. Clin. Endocrinol. Metab. (2016) 101 748-756. DOI: 10.1210/jc.2015-3716 23. Nishida Y, Takahashi Y, Tezuka K, Akimoto H, Nakayama T, Asai S. Comparative effect of dipeptidyl-peptidase 4 inhibitors on laboratory parameters in patients with diabetes mellitus. BMC Pharmacol. Toxicol. (2020) 21 1-12. DOI: 10.1186/s40360-020-00407-4 24. Ito H, Abe M, Antoku S, Omoto T, Shinozaki M, Nishio S. Comparison of the antidiabetic effects of linagliptin among groups with a normal renal function and a mild or severe renal impairment–retrospective observation study of Japanese patients with type 2 diabetes mellitus. Expert Opin. Pharmacother. (2015) 16 289-296. DOI: 10.1517/14656566.2015.995091 25. Moeinzadeh F, Iraj B, Mortazavi M, Ramezani P. The renoprotective effect of linagliptin in type 2 diabetic patients with severely increased albuminuria. Iran. J. Kidney Dis. (2021) 15 344-350. PMID: 34582369 26. Monami M, Vitale V, Ambrosio ML, Bartoli N, Toffanello G, Ragghianti B. Effects on lipid profile of dipeptidyl peptidase 4 inhibitors, pioglitazone, acarbose, and sulfonylureas: Meta-analysis of placebo-controlled trials. Adv. Ther. (2012) 29 736-746. DOI: 10.1007/s12325-012-0045-5 27. Dai X, Wang H, Jing Z, Fu P. The effect of a dual combination of noninsulin antidiabetic drugs on lipids: A systematic review and network meta-analysis. Curr. Med. Res. Opin. (2014) 30 1777-1786. DOI: 10.1185/03007995.2014.921608 28. Kanda J, Furukawa M, Izumo N, Shimakura T, Yamamoto N, Takahashi HE. Effects of the linagliptin, dipeptidyl peptidase-4 inhibitor, on bone fragility induced by type 2 diabetes mellitus in obese mice. Drug Discov. Ther. (2020) 14 218-225. DOI: 10.5582/ddt.2020.03073
--- title: 'Multiparametric Tissue Characterization Utilizing the Cellular Metallome and Immuno-Mass Spectrometry Imaging' authors: - Martin Schaier - Sarah Theiner - Dina Baier - Gabriel Braun - Walter Berger - Gunda Koellensperger journal: JACS Au year: 2023 pmcid: PMC9975846 doi: 10.1021/jacsau.2c00571 license: CC BY 4.0 --- # Multiparametric Tissue Characterization Utilizing the Cellular Metallome and Immuno-Mass Spectrometry Imaging ## Abstract In this study, we present a workflow that enables spatial single-cell metallomics in tissue decoding the cellular heterogeneity. Low-dispersion laser ablation in combination with inductively coupled plasma time-of-flight mass spectrometry (LA-ICP-TOFMS) provides mapping of endogenous elements with cellular resolution at unprecedented speed. Capturing the heterogeneity of the cellular population by metals only is of limited use as the cell type, functionality, and cell state remain elusive. Therefore, we expanded the toolbox of single-cell metallomics by integrating the concepts of imaging mass cytometry (IMC). This multiparametric assay successfully utilizes metal-labeled antibodies for cellular tissue profiling. One important challenge is the need to preserve the original metallome in the sample upon immunostaining. Therefore, we studied the impact of extensive labeling on the obtained endogenous cellular ionome data by quantifying elemental levels in consecutive tissue sections (with and without immunostaining) and correlating elements with structural markers and histological features. Our experiments showed that the elemental tissue distribution remained intact for selected elements such as sodium, phosphorus, and iron, while absolute quantification was precluded. We hypothesize that this integrated assay not only advances single-cell metallomics (enabling to link metal accumulation to multi-dimensional characterization of cells/cell populations), but in turn also enhances selectivity in IMC, as in selected cases, labeling strategies can be validated by elemental data. We showcase the power of this integrated single-cell toolbox using an in vivo tumor model in mice and provide mapping of the sodium and iron homeostasis as linked to different cell types and function in mouse organs (such as spleen, kidney, and liver). Phosphorus distribution maps added structural information, paralleled by the DNA intercalator visualizing the cellular nuclei. Overall, iron imaging was the most relevant addition to IMC. In tumor samples, for example, iron-rich regions correlated with high proliferation and/or located blood vessels, which are key for potential drug delivery. ## Introduction Studying the metallome in biological samples at the cellular level by different imaging techniques has become important due to the crucial role of endogenous metals in metal homeostasis and as a consequence, in the context of diseases.1 While bulk elements (e.g., Na, K, and Mg) are essential for structure and information transfer in the body, trace metals (e.g., Fe, Cu, and Zn) form metalloproteins with catalytic function, like the superoxide dismutase (SOD), which is involved in the removal of free radicals.2 Even minor changes in metal homeostasis can indicate disease development, with the well-known examples of Menkes or Wilson’s disease, where copper transport is disturbed by genetic defects.3 The former is characterized by a copper deficiency, leading to progressive neurodegeneration, while in the latter, an excess of copper can cause cellular damage.4,5 Laser ablation–inductively coupled plasma mass spectrometry (LA-ICPMS) has become an established technique for multi-element mapping of biological samples, providing high sensitivity, high sample throughput, and spatial resolutions in the low μm range.6−8 Bioimaging applications by LA-ICPMS are primarily dominated by mapping of the metallome in various types of tissue samples (e.g., in the context of neurodegenerative diseases) and by studying the uptake of metallodrugs and nanoparticles in biological systems. The newest low-dispersion LA setups have enabled (sub-)cellular imaging (with spot sizes down to 1 μm) and the analysis of single cells at pixel acquisition rates of >200 Hz.9−11 These technological advancements of LA-ICPMS techniques have evolved to the concept of imaging mass cytometry (IMC),12 where phenotyping of single cells is performed in tissue samples with a spot size of 1 μm and pixel acquisition rates of 200 Hz (400 Hz with the latest instrumental generation). Highly multiplexed immunohistochemistry studies can be performed using antibodies that are labeled with different metal tags followed by LA-ICP-TOFMS detection. Metal tags include MaxPar metal-conjugated reagents, MeCATs (metal-coded affinity tags), single-atom chelates, Au/Ag nanoparticles, fluorescent Au nanoclusters, and quantum dots.13−18 In pioneering studies by Wang et al.19 and Giesen et al. ,20 highly multiplexed imaging of epidermal growth factor receptor 2 and 32 proteins was performed in human breast tissue sections. The CyTOF equipment designed for IMC analysis has been targeted toward the clinical field, with a growing field of applications in cancer research, immune-profiling, neurodegenerative diseases, and so forth.21−26 The latest generation of CyTOF instrumentation provides the detection of m/$z = 75$–209, as it was specifically designed for lanthanide detection. As a drawback, the analysis of elements from the lower mass range with biological key functions is not possible with this kind of instrumentation. There are currently different ICP-TOFMS systems on the market that provide the capability to measure m/$z = 14$–25627−29 and, therefore, imaging of endogenous elements at the cellular level.6 Several LA-ICPMS studies employed quadrupole-based ICP–MS systems, limiting the number of elements that can be measured simultaneously. Up to now, only a few imaging studies have addressed the combined detection of endogenous elements and metal-conjugated antibodies by LA-ICP-QMS. The assessment of co-localization of proteins with endogenous metals offers the possibility for studying the interactions between proteins and metal cofactors. In this regard, LA-ICP-QMS studies examined the relationship between nanoparticle/metal-tagged tyrosine hydroxylase (which served as proxy for dopamine) and iron levels in a mouse model of Parkinson’s disease.16,30,31 Co-localization of high Fe and dopamine levels was observed in the substantia nigra pars compacta and in the hypothalamus, in which dysfunction is associated with non-motor symptoms of Parkinson’s disease.16,31 With regard to sample preparation protocols for tissue imaging by LA-ICPMS, cryo-sectioning and formalin fixation followed by paraffin embedding (FFPE) or resin embedding are the most common ones. The effect of these methods on the spatial distribution of elements and their quantities in tissues was already investigated by LA-ICPMS measurements.32−34 A significant influence of the FFPE approach was observed for alkaline elements, while transition metals were less affected by the sample preparation steps. Especially, for calcium and zinc, contaminations introduced during the embedding process could be identified.34 Currently, there is no systematic study available for assessing the background levels of endogenous elements after multi-step immunostaining protocols. Therefore, this work evaluates the influence of immunostaining on the concentration and distribution of these elements for different tissue types and sample preparation protocols. For this purpose, different tissue sections (cryo-sections and FFPE sections) of mouse organs and tumor were compared by LA-ICP-TOFMS analysis before and after immunostaining. Elemental quantification was carried out using gelatin-based micro-droplet standards.35,36 Co-localization with hematoxylin and eosin stains as well as metal-conjugated antibodies was used to further validate the results. Based on these validations, selected elements with biological key functions (Na, P, and Fe) were visualized in different biological applications together with metal-conjugated antibodies to achieve highly multiplexed imaging with cellular resolution. The goal of this combined approach was to reveal different cell types and structural features within biological tissues and to relate them to the elemental homeostasis. In particular, there is great potential for iron, as it is an integral component of many organ functions. By comparing LA-ICP-TOFMS results with histological sections, the added value of this technology for answering biological questions will be demonstrated. ## Impact of Sample Preparation on the Endogenous Elemental Distribution Formalin fixation and paraffin embedding has emerged as gold standard for histological evaluations in clinics, as it preserves essential tissue structures and proteins, making it the ideal method for long-term storage. Tissue sections that are prepared by FFPE protocols undergo extensive washing and solvent treatment, which can significantly alter the qualitative and quantitative distributions of endogenous elements such as iron, copper, and zinc.34 An alternative is cryo-sectioning, where the tissue is directly frozen in an embedding matrix without any further steps prior to sectioning. Due to the reduced number of sample preparation steps, it is the preferred method for the LA-ICPMS analysis of intrinsic elements in biological samples. Sample preparation involving tissue embedding and/or immunostaining, where the tissue is exposed to multiple chemicals and washing steps can contribute to potential elemental contaminations and wash-out effects. Therefore, we systematically evaluated the effect of different sample preparation protocols, followed by immunostaining on elements intrinsically present in biological samples. Absolute quantification was achieved by multi-level matrix-matching calibrations established by gelatin micro-droplets, where low fg/pixel concentration levels mark the lower limits of quantification. Typically, the endogenous tissue/cellular concentration of, for example, sodium, magnesium, phosphorus iron, copper, and zinc, fall within the working range of the quantification method. Quantitative LA-ICP-TOFMS analysis of consecutive tumor tissue sections (cryo-sections and FFPE sections) revealed significant alterations of elemental levels for Mg, K, Ca, Cu, and Zn and, to a lesser extent, for Na, P, and Fe upon immuno-labeling (Figure S1, Tables S1 and S2). The observed impact was comparable for cryo-sections and FFPE sections, with the exception of Na and Fe. While Na and Fe levels decreased in cryo-sections upon labeling, the same procedure resulted in increased levels in FFPE tissue. P showed a consistent loss of up to $40\%$ for both sample preparation types after the immunostaining. However, no significant changes in qualitative elemental distributions were observed for these three elements. Highly elevated Cu concentrations were observed after the labeling procedure, most likely resulting from impurities of the TBS wash solution and due to high physiological Cu levels of BSA used during the blocking step. For Zn, a signal loss of almost $90\%$ was observed in the different tissue samples after the immunostaining approach. For both elements, the qualitative distribution pattern also changed, which makes their detection in stained tissue sections unreliable. The labeling procedure also had a significant impact on the signal intensities of Mg, K, and Ca, with a signal loss of up to $99\%$, resulting in signals under the limit of detection for these elements after labeling. Overall, it was concluded that accurate absolute quantification of endogenous elements present in biological samples is precluded, when immunostaining protocols are applied. This applies to FFPE and cryo-sections, despite the fact that for cryo-sectioning, the number of sample preparation steps is reduced to a minimum. Indeed, sections without antibody labeling would still be required to obtain absolute quantitative results on endogenous elements. We focused our investigation on FFPE tissue due to the following reasons: (i) the cell morphology is better preserved in FFPE tissue sections (Figure S2), an important factor for single-cell analysis and cell segmentation; (ii) cryo-sections tend to show more cutting artefacts than FFPE sections, which can result in tissue folding and cell overlap (Figure S2); (iii) only a fraction of commercially available metal-conjugated antibodies can be used on cryo-sections. ## Co-localization with Histological Features and Metal-Conjugated Antibodies With regard to the qualitative distribution of endogenous elements after multi-step sample preparation protocols, an orthogonal method such as the microscopic evaluation of a histological stain of an adjacent section is required to obtain reliable information. It has to be considered that consecutive sections are not identical and that changes in the tissue structure can occur, specifically at the single-cell level. Co-localization of the endogenous elemental pattern with distinct histological features and/or with tissue structures/cell types as visualized by metal-conjugated antibodies enables the use of endogenous elements as an additional layer of information in LA-ICP-TOFMS images. As a first step of validation, the similarity of the iron signal intensity maps of consecutive tumor tissue sections (Figure 1) was evaluated using the Structural Similarity Index (SSIM), which compares image parameters such as luminance, contrast, and structure.37 A score of 0.82 was obtained (Figure S3), which indicates strong similarity (1 → very strong, −1 → very weak), especially considering that the sections were consecutive and already showed structural differences during histological evaluation (Figure 1). The correlation matrix showed significant co-localization of iron with vimentin (a marker for mesenchymal cells including fibroblasts and endothelial cells of blood vessels) and also to a lower extent with α-SMA (myofibroblasts) and collagen, which are all integral parts of the connective tissue (Figure 2). Hardly any correlation was observed for pan-keratin, which marks the epithelial cells in the tumor. Furthermore, by selecting different regions of interest (ROIs), it could be determined that iron showed the highest correlation with vimentin within the tumor, while in the outer regions of the tumor tissue, iron showed the highest correlation with α-SMA and collagen (Figure S4). The iron distribution can therefore be assigned to biological characteristics of the tumor microenvironment both visually (via an H&E stain and SSIM) and statistically (via correlation with metal-conjugated antibodies). Figure 1: Signal intensity maps of 56Fe+ in mouse tumor of consecutive FFPE sections (A) before and (C) after the metal-conjugated antibody staining procedure, as determined by LA-ICP-TOFMS analysis. (B) H&E-stained tumor tissue of an adjacent FFPE section for microscopic evaluation. Figure 2: (A) FFPE section of a mouse tumor tissue stained with H&E. (B) Corresponding 56Fe+ signal intensity map of a consecutive tumor section, determined by LA-ICP-TOFMS analysis. (C) Co-localization of the iron signal was assessed using a correlation matrix with four different antibodies. The values show Pearson correlation coefficients (0.5–1: strong positive, 0.3–0.5: moderate positive, 0–0.3: weak positive). In a next step, selected applications of LA-ICP-TOFMS imaging in different mouse tissue samples will be presented, highlighting the combined analysis of elements with biological key functions and metal-conjugated antibodies for visualization of characteristic tissue structures and cell types. ## Spleen In the spleen, α-SMA and collagen enabled visualization of the splenic capsule (dense collagenous tissue with smooth muscle cells surrounding the spleen) and trabeculae, which are projections from the capsule into the parenchyma containing arteries and veins (Figure 3). The iron distribution as determined by LA-ICP-TOFMS imaging enabled to differentiate the white and red pulp and correlated with histological features, where high amounts of blood were present and circulating. High iron levels were detected in the red pulp, which is known to be responsible for blood filtering in the spleen, whereas a relatively low iron signal was found in the white pulp (Figure 3).38 The well-perfused red pulp also showed increased signal levels of KI-67, indicating high levels of proliferation (Figure S5). The highest iron content was observed in the marginal zone (interface between white and red pulp), which is known to exhibit a high blood circulation.39 The iron-rich marginal zone also showed higher intensities of CD19 (marker for B-cells) and Arginase-1 (M2 macrophages), while CD86 (M1 macrophages) was expressed to a higher extent in the other regions of the red pulp (Figure S5). These results are in good accordance with the literature, where a high number of B-cells and macrophages was reported in the marginal zone and the red pulps of the spleen, respectively.40−42 However, since the spleen was derived from an immunodeficient mouse model (CB-17/SCID), it has to be mentioned that the B- and T-cell levels were relatively low. Figure 3: (A) Signal intensity maps and (B) signal overlay of Collagen Type I (blue), α-SMA (yellow), and 56Fe+ (red) in the spleen of a mouse, with a ROI. (C) Corresponding H&E stain of a consecutive spleen section and (D) ROI with characteristic histological features. ## Liver Within the liver, the iron signal allowed for visualization of the blood flow, starting with low amounts of iron around the portal triad (bile duct, portal vein, and arteriole surrounded by a collagenous matrix) and increasing iron levels around the central veins (Figure 4). Phosphorus can be used to visualize the cell nuclei of individual hepatocytes with a similar signal as the iridium-based DNA intercalator, commonly used for IMC applications (Figure S6). The presence of the central veins in the liver was indicated by Collagen Type I, whereas α-SMA showed a thin layer in the inner side of the portal veins with high intensities around the hepatic artery (Figure S7).43 Interestingly, high abundance of E-Cadherin (cell–cell adhesion) was observed in regions with low iron content and seemed to connect the portal vein with the central vein. The apoptosis marker, Caspase 3, also showed a high intensity region in the center of the liver section, located directly on a vein. Figure 4: (A) Signal intensity maps and (B) signal overlay of 31P+ (blue), 56Fe+ (red), and Collagen Type I (yellow) in liver tissue of a mouse, with a ROI. (C) Corresponding H&E stain of a consecutive liver section and (D) ROI with characteristic histological features. ## Kidney High levels of sodium within the kidney were detected in the proximal convoluted tubules of the cortex region, whereas the proximal straight tubules inside the medulla showed lower sodium intensities (Figure 5). The highest sodium levels were found in the renal corpuscles, where filtering through the glomerular barrier takes place.44 Since sodium as an alkali metal is highly soluble and more easily affected by wash-out effects than, for example, P and Fe, it is particularly important to correlate its distribution with histological features and to set it in a biological context. In the investigated case, sodium hotspots could be directly correlated with the presence of renal corpuscles, as observed in the H&E stain of a consecutive kidney section (Figure 5D). Figure 5: (A) Signal intensity maps and (B) signal overlay of 23Na+ (blue), BCL-2 (yellow), and E-Cadherin (red) in mouse kidney, with a ROI. (C) Corresponding H&E stain of a consecutive kidney section and (D) ROI with characteristic histological features, which are marked in yellow. A higher magnification can be seen in Figure S8. The presence of proximal tubules was indicated by the BCL-2 marker, which plays a role in apoptosis regulation and was found to be strongly expressed in proximal convoluted tubules, lower in proximal straight tubules, and weakly in distal tubules (Figure 5).45 High sodium intensities were also found inside the collecting ducts of the medullary rays, which are involved in sodium homeostasis by regulating the amounts that get excreted in the urine.46 E-Cadherin proved to be more suited for the visualization of the kidney structure than the Ir-based DNA-intercalator since the renal cells contain multiple cell nuclei inside tubules (Figure S9). Using collagen and α-SMA, renal arteries could be visualized (Figure S10). Iron was found predominantly in the proximal tubules of the cortex, but also in medullary rays, which were enriched in pan-keratin. ## Lung The phosphorus signal provided an overview of the lung structure, showing cell nuclei and cytoplasm, while the iron signal allowed for visualization of the erythrocytes of the capillaries surrounding the alveoli (Figure 6). Blood vessels, terminal bronchiole, and the pulmonary artery were surrounded by α-SMA (Figure 6). Collagen was predominantly found in arteries and outer regions of the lung, while pan-keratin and cadherin were increased in the inner regions of the bronchiole and pulmonary artery (Figure S11). Individual cells within the capillaries showed proliferation (KI-67) with accumulation in one central region. As the organs were taken from a tumor-bearing mouse, this might indicate a metastatic event in the lung, which is further supported by the H&E stain (Figure S12A). The cell lump in this area showed low perfusion and the cell nuclei have a slightly different color. In addition to KI-67, increased intensities of CD44 (cell adhesion) and CD11b (innate immune cell marker) were also found, indicating tumor cell infiltration accompanied by an inflammatory event (Figure S12B). Figure 6: (A) Signal intensity maps and (B) signal overlay of 31P+ (blue), 56Fe+ (red), and α-SMA (yellow) in mouse lung, with a ROI. (B) Corresponding H&E stain of a consecutive lung section and (C) ROI with characteristic histological features. ## Tumor For the characterization of the tumor microenvironment, the iron distribution can serve as a useful tool for visualizing blood vessels, which provides valuable information about tissue perfusion and potentially drug delivery and penetration into tumor tissue. An HCT116 colon cancer tumor section obtained from a human xenograft grown from CB-17/SCID mice was analyzed (Figure 7), which exhibited a low vascular density. A heterogeneous iron distribution was observed with irregular branching of blood vessels emanating from the outer layers and large zones of ischemia and necrosis, which is typical for rapid tumor growth.47 Collagen made up most of the outer tumor layer and acted as scaffold for the cancer cells (Figure 7).48 In the presence of iron, α-SMA can be found, revealing the cancer associated fibroblasts inside the tumor tissue (Figure S13).49 The majority of epithelial cells (indicated by pan-keratin) showed proliferation, whereas a large zone of necrosis was indicated by the absence of KI-67 and E-cadherin (Figures 7 and S11). Furthermore, this was confirmed by the corresponding H&E stain.50 Most of the DNA damage (pH2AX) corresponded to dead cells in this zone, but some of the surrounding proliferating cells were also affected (Figure S13). Figure 7: (A) Signal intensity maps and (B) signal overlay of KI-67 (blue), 56Fe+ (red), and Collagen Type I (yellow) in an HCT116 colon cancer tumor section of a mouse, with a ROI. (C) Corresponding H&E stain of a consecutive tumor section and (D) ROI with characteristic histological features. ## Conclusions In this study, imaging mass cytometry was expanded toward simultaneous imaging of the cellular ionome. The results highlight the importance of evaluating multi-step sample preparation protocols for the analysis of endogenous elements in tissue samples by LA-ICPMS imaging. The elemental composition is already strongly influenced during tissue preparation (cryo-treatment, FFPE), an effect that is further enhanced by subsequent immunostaining procedures. While for selected elements such as Na, P, and Fe the qualitative tissue distribution remained intact after staining, quantification of endogenous elements was precluded. Tissue distributions of these elements with biological key functions were assessed at cellular resolution upon application of immunostaining procedures and allowed for linking ionome data to cell type/state and function. The added value of the validated wide mass range bioimaging strategy was emphasized using the prime example of preclinical in vivo models on cancer. Cell nuclei and parts of the cytoplasm could be visualized via the phosphorus signal, while sodium enabled the localization of renal corpuscles in the kidney and iron showed the red pulp with its marginal zones inside the spleen. This concept could be especially attractive in the context of disease progression, the evaluation of potential biomarkers, and the development of novel therapeutics. New embedding methods and sample preparation techniques with a reduced chemical background will be crucial to make full use of the potential of this promising technique. ## Chemicals and Reagents Ultrapure water (18.2 MΩ cm, ELGA Water purification system, Purelab Ultra MK 2, UK) was used for all dilutions and washing steps. A multi-element stock solution and single element standard solutions were purchased from Labkings (Hilversum, The Netherlands). Bovine serum albumin (lyophilized powder, BioReagent), Tris buffered saline (BioUltra), Triton X-100 (for molecular biology), m-xylene (anhydrous, ≥$99\%$), and ethanol (absolute, EMSURE) were purchased from Sigma-Aldrich (Steinheim, Germany). The target retrieval solution was bought from Agilent Technologies (Waldbronn, Germany). Paraformaldehyde aqueous solution (Electron Microscopy grade, $16\%$) was obtained from Science Services (Munich, Germany) in form of sealed ampoules, to ensure fresh solutions in each staining procedure. The metal-conjugated antibodies used in this study (Table S3) and the Intercalator-Ir (Cell-ID, 125 μM) were purchased from Fluidigm (San Francisco, CA, USA). LA-ICP-TOFMS measurements were carried out in an ISO class 7 clean room. All cell culture media and reagents were purchased from Sigma-Aldrich (Vienna, Austria) and all plastic dishes, plates, and flasks from StarLab (Hamburg, Germany) unless stated otherwise. ## Cell Culture The human colorectal cancer HCT116 cell line was kindly provided by Dr. Vogelstein from John Hopkins University, Baltimore. Cells were cultured in McCoy’s medium (M8403, Sigma-Aldrich, St. Louis, MO, USA) supplemented with $10\%$ fetal calf serum (FCS; PAA, Linz, Austria) and 2 mM glutamine (Sigma-Aldrich, St. Louis, MO, USA). All cultures were grown under standard cell culture conditions and checked for Mycoplasma contamination. ## Animal Experiments For in vivo experiments, 1 × 106 HCT116 cells were injected subcutaneously (s.c.) in serum-free RPMI-medium (R6504, Sigma-Aldrich, St. Louis, MO, USA) into the right flank of 11-week-old male CB-17/SCID mice. Animals were kept in a pathogen-free environment and handled in a laminar airflow cabinet. The experiments were performed according to the regulations of the Ethics Committee for the Care and Use of Laboratory Animals at the Medical University Vienna (proposal number BMWF-$\frac{66.009}{0140}$-II/3b/2011), the U.S. Public Health Service Policy on Human Care and Use of Laboratory Animals, as well as the United Kingdom Coordinating Committee on Cancer Prevention Research’s Guidelines for the Welfare of Animals in Experimental Neoplasia. Tumors were palpable on day 7 following s.c. injection. Animals were controlled for symptoms of distress daily, and tumor size was assessed regularly by caliper measurement. Tumor volume was calculated using the formula (length × width$\frac{2}{2}$). On day 17, mice were sacrificed. Tumors and organs were formalin-fixed in $4\%$ formaldehyde for 24 h (Carl Roth, #P087.3) and paraffin-embedded using a KOS machine (Milestone Medical, Sorisole, Italy). ## Histological Evaluations For histological evaluation, embedded tumors and organs were cut into three consecutive 4 μm thick sections per set. Every first and third section was used for LA-ICP-TOFMS analysis. The second, middle section was used for H/E staining (Figure S14). Tissue was deparaffinized, rehydrated, and stained with hematoxylin/eosin (H/E) by routine procedures. ## Immunostaining of Cryo-Sections and FFPE Sections The FFPE tumor tissue sections were deparaffinized by heating the slides in an oven for 1–2 h at 60 °C, followed by incubation with fresh xylene for 20 min. Descending grades of alcohol (100–$70\%$ EtOH) were used for re-hydration. After washing the slides with ultrapure water, heat-induced antigen retrieval was performed at 96 °C for 30 min using an antigen retrieval solution (Tris–EDTA, pH = 9). The slides were carefully cooled down and washed with ultrapure water and TBS. Cryo-sections were first fixed for 30 min with $4\%$ PFA in TBS and then rinsed 3 times with TBS. Both types of sections were incubated with $3\%$ BSA in TBS for 45 min at RT to block unspecific binding sites. The sections were then incubated with a cocktail of metal-tagged antibodies in a hydration chamber overnight at 4 °C. A summary of the metal-conjugated antibodies employed in this study can be found in Table S3. The antibody solution was prepared by adding small amounts of each antibody (1:50–1:200 dilutions of the respective antibodies) to $0.5\%$ BSA in TBS. In order to avoid the formation of aggregates, the antibodies were centrifuged beforehand at 13,000 g for 2 min. For cell permeabilization, the slides were incubated in $0.2\%$ Triton X-100 and washed afterward with TBS. A Cell-ID Intercalator-Ir (125 μM, Fluidigm, San Francisco, CA, USA) was used to stain the tissue sections, by adding the solution (0.30 μM) on the sections and incubating them for 30 min at RT in a hydration chamber. Finally, the slides were repeatedly washed with ultrapure water and left to air-dry until LA-ICP-TOFMS analysis. ## Calibration Standards for LA-ICP-TOFMS Analysis Quantification was performed by LA-ICP-TOFMS using gelatin-microdroplets, as described previously.35 For this purpose, liquid multi-element standard solutions were prepared gravimetrically from commercial standard stock solutions in $1\%$ (v/v) HNO3. Gelatin stock solution ($10\%$, w/w) was added to reach a final concentration of $1\%$ (w/w) gelatin. The resulting solutions were transferred into wells of a 384 well plate, which serves as the sample source of a micro-spotter system. A CellenONE X1 micro-spotter (Cellenion, Lyon, France) was used to generate arrays of the gelatin micro-droplet standards onto glass slides with a droplet size of 400 ± 5 pL resulting in droplet sizes of around 100 μm in diameter. The size of the droplets was evaluated by the software of the instrument and was used for normalization to establish absolute elemental quantities within the droplets. The entire micro-droplets were quantitatively and selectively ablated, and multi-element analysis was performed by LA-ICP-TOFMS. ## LA-ICP-TOFMS Analysis An Iridia 193 nm laser ablation (LA) system (Teledyne Photon Machines, Bozeman, MT, USA) was coupled to an icpTOF 2R (TOFWERK AG, Thun, Switzerland) ICP-TOFMS instrument. The LA system was equipped with an ultrafast low-dispersion cell51 in a Cobalt ablation chamber and coupled with the aerosol rapid introduction system (ARIS) to the ICP-TOFMS. An Ar make-up gas flow (∼0.90 L min–1) was introduced through the low-dispersion mixing bulb of the ARIS into the He carrier gas flow (0.60 L min–1) before entering the plasma. Daily tuning of the LA and ICP-TOFMS settings was performed using NIST SRM612 glass certified reference material (National Institute for Standards and Technology, Gaithersburg, MD, USA). Optimization was based on high intensities for 24Mg+, 59Co+, 115In+, and 238U+, low oxide formation based on the 238U16O+/238U+ ratio (<$2\%$) and low elemental fractionation based on the 238U+/232Th+ ratio (∼1). Daily optimization included to aim at a low aerosol dispersion characterized by the pulse response duration for 238U+ based on the FW0.01 M criterion, that is, the full peak width of the 238U+ signal response obtained upon a single laser shot, at $1\%$ of the height of the maximum signal intensity. Laser ablation sampling was performed in fixed dosage mode 2, at a repetition rate of 200 Hz and using a 5 μm spot size (square) with an interspacing of 2.5 μm between the lines resulting in a pixel size of 2.5 μm × 2.5 μm. Selective ablation of the samples was achieved by selecting an energy density below the ablation threshold of glass and above the ablation threshold of the samples.52 Gelatin micro-droplets, organs, and tumor sections were removed quantitatively using a fluence of 0.60 and 0.80 J cm–2, respectively. The icpTOF 2R ICP-TOFMS instrument has a specified mass resolution (R = m/Δm) of 6000 (full width half-maximum definition) and allows for the analysis of ions from m/$z = 14$–256. The integration and read-out rate match the LA repetition rate. The instrument was equipped with a torch injector of 2.5 mm inner diameter and nickel sample and skimmer cones with a skimmer cone insert of 2.8 mm in diameter. A radio frequency power of 1440 W, an auxiliary Ar gas flow rate of ∼0.80 L min–1, and a plasma Ar gas flow rate of 14 L min–1 were used. For all measurements, the collision cell technology (CCT) mode was used, where the collision cell was pressurized with a mixture of H2/He gas [$93\%$ He (v/v), $7\%$ H2 (v/v)] with an optimized flow rate of 4.2 mL min–1. The following CCT parameters were used: CCT focus lens: −6.3 V, CCT entry lens: −150 V, CCT mass: 261 V, CCT bias: −1 V, CCT exit lens: −90 V. Instrumental parameters for LA-ICP-TOFMS measurements in CCT mode are summarized in Table S4. ## Data Acquisition and Processing of LA-ICP-TOFMS Data LA-ICP-TOFMS data were recorded using TofPilot 2.10.3.0 (TOFWERK AG, Thun, Switzerland) and saved in the open-source hierarchical data format (HDF5, www.hdfgroup.org). Post-acquisition data processing was performed with Tofware v3.2.2.1 (TOFWERK AG, Thun, Switzerland), which is used as an add-on on IgorPro (Wavemetric Inc., Oregon, USA). The data processing included [1] drift correction of the mass peak position in the spectra over time via time-dependent mass calibration, [2] determining the peak shape, and [3] fitting and subtracting the mass spectral baseline. Data was further processed with HDIP version 1.6.6.d44415 × 105 (Teledyne Photon Machines, Bozeman, MT, USA). An integrated script was used to automatically process the files generated by Tofware and to generate 2D elemental distribution maps. ## Statistical Methods Elemental channel correlations were calculated using Spearman’s rank correlation coefficient, while the SSIM was used to score image similarity. Correlation matrices were created using the Python programming language v3.9, specifically by use of the following libraries: NumPy v1.23.0; pandas v1.4.3; Seaborn v0.11.2; and scikit-learn v1.1.1. ## References 1. Becker J. S., Matusch A., Wu B.. Bioimaging mass spectrometry of trace elements - recent advance and applications of LA-ICP-MS: A review. Anal. Chim. Acta (2014) 835 1-18. DOI: 10.1016/j.aca.2014.04.048 2. Sabine Becker J.. Imaging of metals in biological tissue by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS): state of the art and future developments. J. Mass Spectrom. (2013) 48 255-268. DOI: 10.1002/jms.3168 3. Cox D. W.. Disorders of copper transport. Br. Med. Bull. (1999) 55 544-555. DOI: 10.1258/0007142991902619 4. Kodama H., Fujisawa C., Bhadhprasit W.. Inherited copper transport disorders: biochemical mechanisms, diagnosis, and treatment. Curr. Drug Metab. (2012) 13 237-250. DOI: 10.2174/138920012799320455 5. Tümer Z., Møller L. B.. Menkes disease. Eur. J. Hum. Genet. (2010) 18 511-518. PMID: 19888294 6. Theiner S., Schweikert A., Van Malderen S. J. M., Schoeberl A., Neumayer S., Jilma P., Peyrl A., Koellensperger G.. Laser Ablation-Inductively Coupled Plasma Time-of-Flight Mass Spectrometry Imaging of Trace Elements at the Single-Cell Level for Clinical Practice. Anal. Chem. (2019) 91 8207-8212. DOI: 10.1021/acs.analchem.9b00698 7. Schoeberl A., Gutmann M., Theiner S., Schaier M., Schweikert A., Berger W., Koellensperger G.. Cisplatin Uptake in Macrophage Subtypes at the Single-Cell Level by LA-ICP-TOFMS Imaging. Anal. Chem. (2021) 93 16456-16465. DOI: 10.1021/acs.analchem.1c03442 8. Voloaca O. M., Clench M. R., Koellensperger G., Cole L. M., Haywood-Small S. L., Theiner S.. Elemental Mapping of Human Malignant Mesothelioma Tissue Samples Using High-Speed LA-ICP-TOFMS Imaging. Anal. Chem. (2022) 94 2597-2606. DOI: 10.1021/acs.analchem.1c04857 9. Van Malderen S. J. M., Managh A. J., Sharp B. L., Vanhaecke F.. Recent developments in the design of rapid response cells for laser ablation-inductively coupled plasma-mass spectrometry and their impact on bioimaging applications. J. Anal. At. Spectrom. (2016) 31 423-439. DOI: 10.1039/c5ja00430f 10. Gundlach-Graham A., Günther D.. Toward faster and higher resolution LA-ICPMS imaging: on the co-evolution of LA cell design and ICPMS instrumentation. Anal. Bioanal. Chem. (2016) 408 2687-2695. DOI: 10.1007/s00216-015-9251-8 11. Van Malderen S. J. M., Van Acker T., Vanhaecke F.. Sub-micrometer Nanosecond LA-ICP-MS Imaging at Pixel Acquisition Rates above 250 Hz via a Low-Dispersion Setup. Anal. Chem. (2020) 92 5756-5764. DOI: 10.1021/acs.analchem.9b05056 12. Chang Q., Ornatsky O. I., Siddiqui I., Loboda A., Baranov V. I., Hedley D. W.. Imaging Mass Cytometry. Cytometry A (2017) 91 160-169. DOI: 10.1002/cyto.a.23053 13. Pisonero J., Bouzas-Ramos D., Traub H., Cappella B., Álvarez-Llamas C., Richter S., Mayo J. C., Costa-Fernandez J. M., Bordel N., Jakubowski N.. Critical evaluation of fast and highly resolved elemental distribution in single cells using LA-ICP-SFMS. J. Anal. At. Spectrom. (2019) 34 655-663. DOI: 10.1039/c8ja00096d 14. Giesen C., Mairinger T., Khoury L., Waentig L., Jakubowski N., Panne U.. Multiplexed Immunohistochemical Detection of Tumor Markers in Breast Cancer Tissue Using Laser Ablation Inductively Coupled Plasma Mass Spectrometry. Anal. Chem. (2011) 83 8177-8183. DOI: 10.1021/ac2016823 15. Valencia E., Fernández B., Cruz-Alonso M., García M., González-Iglesias H., Fernández-Abedul M. T., Pereiro R.. Imaging of proteins in biological tissues by fluorescence microscopy and laser ablation-ICP-MS using natural and isotopically enriched silver nanoclusters. J. Anal. At. Spectrom. (2020) 35 1868-1879. DOI: 10.1039/c9ja00365g 16. Paul B., Hare D. J., Bishop D. P., Paton C., Nguyen V. T., Cole N., Niedwiecki M. M., Andreozzi E., Vais A., Billings J. L., Bray L., Bush A. I., McColl G., Roberts B. R., Adlard P. A., Finkelstein D. I., Hellstrom J., Hergt J. M., Woodhead J. D., Doble P. A.. Visualising mouse neuroanatomy and function by metal distribution using laser ablation-inductively coupled plasma-mass spectrometry imaging. Chem. Sci. (2015) 6 5383-5393. DOI: 10.1039/c5sc02231b 17. Waentig L., Jakubowski N., Hardt S., Scheler C., Roos P. H., Linscheid M. W.. Comparison of different chelates for lanthanide labeling of antibodies and application in a Western blot immunoassay combined with detection by laser ablation (LA-)ICP-MS. J. Anal. At. Spectrom. (2012) 27 1311-1320. DOI: 10.1039/c2ja30068k 18. Bishop D. P., Cole N., Zhang T., Doble P. A., Hare D. J.. A guide to integrating immunohistochemistry and chemical imaging. Chem. Soc. Rev. (2018) 47 3770-3787. DOI: 10.1039/c7cs00610a 19. Wang H. A. O., Grolimund D., Giesen C., Borca C. N., Shaw-Stewart J. R. H., Bodenmiller B., Günther D.. Fast Chemical Imaging at High Spatial Resolution by Laser Ablation Inductively Coupled Plasma Mass Spectrometry. Anal. Chem. (2013) 85 10107-10116. DOI: 10.1021/ac400996x 20. Giesen C., Wang H. A. O., Schapiro D., Zivanovic N., Jacobs A., Hattendorf B., Schüffler P. J., Grolimund D., Buhmann J. M., Brandt S., Varga Z., Wild P. J., Günther D., Bodenmiller B.. Highly multiplexed imaging of tumor tissues with subcellular resolution by mass cytometry. Nat. Methods (2014) 11 417. DOI: 10.1038/nmeth.2869 21. Ijsselsteijn M. E., van der Breggen R., Farina Sarasqueta A., Koning F., de Miranda N. F. C. C.. A 40-Marker Panel for High Dimensional Characterization of Cancer Immune Microenvironments by Imaging Mass Cytometry. Front. Immunol. (2019) 10 2534. DOI: 10.3389/fimmu.2019.02534 22. van Maldegem F., Valand K., Cole M., Patel H., Angelova M., Rana S., Colliver E., Enfield K., Bah N., Kelly G., Tsang V. S. K., Mugarza E., Moore C., Hobson P., Levi D., Molina-Arcas M., Swanton C., Downward J.. Characterisation of tumour microenvironment remodelling following oncogene inhibition in preclinical studies with imaging mass cytometry. Nat. Commun. (2021) 12 5906. DOI: 10.1038/s41467-021-26214-x 23. Le Rochais M., Hemon P., Pers J.-O., Uguen A.. Application of High-Throughput Imaging Mass Cytometry Hyperion in Cancer Research. Front. Immunol. (2022) 13 859414. DOI: 10.3389/fimmu.2022.859414 24. Wang Y. J., Traum D., Schug J., Gao L., Liu C., Atkinson M. A., Powers A. C., Feldman M. D., Naji A., Chang K.-M., Kaestner K. H.. Multiplexed In Situ Imaging Mass Cytometry Analysis of the Human Endocrine Pancreas and Immune System in Type 1 Diabetes. Cell Metab. (2019) 29 769-783. DOI: 10.1016/j.cmet.2019.01.003 25. Damond N., Engler S., Zanotelli V. R. T., Schapiro D., Wasserfall C. H., Kusmartseva I., Nick H. S., Thorel F., Herrera P. L., Atkinson M. A., Bodenmiller B.. A Map of Human Type 1 Diabetes Progression by Imaging Mass Cytometry. Cell Metab. (2019) 29 755-768. DOI: 10.1016/j.cmet.2018.11.014 26. Chen C., McDonald D., Blain A., Sachdeva A., Bone L., Smith A. L. M., Warren C., Pickett S. J., Hudson G., Filby A., Vincent A. E., Turnbull D. M., Reeve A. K.. Imaging mass cytometry reveals generalised deficiency in OXPHOS complexes in Parkinson’s disease. NPJ Parkinsons Dis (2021) 7 39. DOI: 10.1038/s41531-021-00182-x 27. Hendriks L., Gundlach-Graham A., Hattendorf B., Günther D.. Characterization of a new ICP-TOFMS instrument with continuous and discrete introduction of solutions. J. Anal. At. Spectrom. (2017) 32 548-561. DOI: 10.1039/c6ja00400h 28. Burger M., Hendriks L., Kaeslin J., Gundlach-Graham A., Hattendorf B., Günther D.. Characterization of inductively coupled plasma time-of-flight mass spectrometry in combination with collision/reaction cell technology – insights from highly time-resolved measurements. J. Anal. At. Spectrom. (2019) 34 135-146. DOI: 10.1039/c8ja00275d 29. Tharaud M., Schlatt L., Shaw P., Benedetti M.. Nanoparticle identification using single particle ICP-ToF-MS acquisition coupled to cluster analysis. From engineered to natural nanoparticles. J. Anal. At. Spectrom. (2022) 37 2042-2052. DOI: 10.1039/d2ja00116k 30. Clases D., Gonzalez de Vega R., Funke S., Lockwood T. E., Westerhausen M. T., Taudte R. V., Adlard P. A., Doble P. A.. Matching sensitivity to abundance: high resolution immuno-mass spectrometry imaging of lanthanide labels and endogenous elements in the murine brain. J. Anal. At. Spectrom. (2020) 35 728-735. DOI: 10.1039/c9ja00405j 31. Hare D. J., Lei P., Ayton S., Roberts B. R., Grimm R., George J. L., Bishop D. P., Beavis A. D., Donovan S. J., McColl G., Volitakis I., Masters C. L., Adlard P. A., Cherny R. A., Bush A. I., Finkelstein D. I., Doble P. A.. An iron–dopamine index predicts risk of parkinsonian neurodegeneration in the substantia nigra pars compactaAn iron-dopamine index predicts risk of parkinsonian neurodegeneration in the substantia nigra pars compacta. Chem. Sci. (2014) 5 2160-2169. DOI: 10.1039/c3sc53461h 32. Pushie M. J., Hollings A., Reinhardt J., Webb S. M., Lam V., Takechi R., Mamo J. C., Paterson P. G., Kelly M. E., George G. N., Pickering I. J., Hackett M. J.. Sample preparation with sucrose cryoprotection dramatically alters Zn distribution in the rodent hippocampus, as revealed by elemental mapping. J. Anal. At. Spectrom. (2020) 35 2498-2508. DOI: 10.1039/d0ja00323a 33. Jagielska A., Ruszczyńska A., Bulska E., Wagner B.. The impact of sample preparation on the elemental composition of soft tissues assessed by laser ablation ICP-MS. J. Anal. At. Spectrom. (2020) 35 1340-1350. DOI: 10.1039/d0ja00115e 34. Bonta M., Török S., Hegedus B., Döme B., Limbeck A.. A comparison of sample preparation strategies for biological tissues and subsequent trace element analysis using LA-ICP-MS. Anal. Bioanal. Chem. (2017) 409 1805-1814. DOI: 10.1007/s00216-016-0124-6 35. Schweikert A., Theiner S., Wernitznig D., Schoeberl A., Schaier M., Neumayer S., Keppler B. K., Koellensperger G.. Micro-droplet-based calibration for quantitative elemental bioimaging by LA-ICPMS. Anal. Bioanal. Chem. (2022) 414 485-495. DOI: 10.1007/s00216-021-03357-w 36. Šala M., Šelih V. S., van Elteren J. T.. Gelatin gels as multi-element calibration standards in LA-ICP-MS bioimaging: fabrication of homogeneous standards and microhomogeneity testing. Analyst (2017) 142 3356-3359. PMID: 28832035 37. Wang W., Bovik A. C., Sheikh H. R., Simoncelli E. P.. Image quality assessment: from error visibility to structural similarity. IEEE Trans Image Process (2004) 13 600. DOI: 10.1109/tip.2003.819861 38. Brozman M., Jakubovský J.. The red pulp of the human spleen. Structural basis of blood filtration. Z. mikrosk.-anat. Forsch. (1989) 103 316-328. PMID: 2672659 39. Steiniger B., Bette M., Schwarzbach H.. The open microcirculation in human spleens: a three-dimensional approach. J. Histochem. Cytochem. (2011) 59 639-648. DOI: 10.1369/0022155411408315 40. Borges da Silva H., Fonseca R., Pereira R. M., Cassado A. d. A., Álvarez J. M., D’Império Lima M. R.. Splenic Macrophage Subsets and Their Function during Blood-Borne Infections. Front. Immunol. (2015) 6 480. DOI: 10.3389/fimmu.2015.00480 41. Doyon-Laliberté K., Aranguren M., Poudrier J., Roger M.. Marginal Zone B-Cell Populations and Their Regulatory Potential in the Context of HIV and Other Chronic Inflammatory Conditions. Int. J. Mol. Sci. (2022) 23 3372. DOI: 10.3390/ijms23063372 42. Cerutti A., Cols M., Puga I.. Marginal zone B cells: virtues of innate-like antibody-producing lymphocytes. Nat. Rev. Immunol. (2013) 13 118-132. DOI: 10.1038/nri3383 43. Masugi Y., Abe T., Tsujikawa H., Effendi K., Hashiguchi A., Abe M., Imai Y., Hino K., Hige S., Kawanaka M., Yamada G., Kage M., Korenaga M., Hiasa Y., Mizokami M., Sakamoto M.. Quantitative assessment of liver fibrosis reveals a nonlinear association with fibrosis stage in nonalcoholic fatty liver disease. Hepatol. Commun. (2018) 2 58-68. DOI: 10.1002/hep4.1121 44. Haraldsson B., Nyström J., Deen W. M.. Properties of the Glomerular Barrier and Mechanisms of Proteinuria. Physiol. Rev. (2008) 88 451-487. DOI: 10.1152/physrev.00055.2006 45. Song X. F., Ren H., Andreasen A., Thomsen J. S., Zhai X. Y.. Expression of Bcl-2 and Bax in mouse renal tubules during kidney development. PLoS One (2012) 7. DOI: 10.1371/journal.pone.0032771 46. Morla L., Doucet A., Lamouroux C., Crambert G., Edwards A.. The renal cortical collecting duct: a secreting epithelium?. J. Physiol. (2016) 594 5991-6008. DOI: 10.1113/jp272877 47. Nagy J. A., Chang S. H., Dvorak A. M., Dvorak H. F.. Why are tumour blood vessels abnormal and why is it important to know?. Br. J. Cancer (2009) 100 865-869. DOI: 10.1038/sj.bjc.6604929 48. Fang M., Yuan J., Peng C., Li Y.. Collagen as a double-edged sword in tumor progression. Tumor Biol (2014) 35 2871-2882. DOI: 10.1007/s13277-013-1511-7 49. Park J., Tallquist M. D., Vasan R. S., Sawyer D. B.. Cardiac Fibroblast. Encyclopedia of Cardiovascular Research and Medicine (2018) 420-433 50. Simiantonaki N., Kurzik-Dumke U., Karyofylli G., Jayasinghe C., Kirkpatrick C. J.. Loss of E-cadherin in the vicinity of necrosis in colorectal carcinomas: association with NFkappaB expression. Int. J. Oncol. (2007) 31 269-275. DOI: 10.3892/ijo.31.2.269 51. Van Acker T., Van Malderen S. J. M., Van Helden T., Stremtan C., Šala M., van Elteren J. T., Vanhaecke F.. Analytical figures of merit of a low-dispersion aerosol transport system for high-throughput LA-ICP-MS analysis. J. Anal. At. Spectrom. (2021) 36 1201-1209. DOI: 10.1039/d1ja00110h 52. Van Acker T., Van Malderen S. J. M., Colina-Vegas L., Ramachandran R. K., Vanhaecke F.. Selective ablation of biological tissue and single cells on a glass substrate by controlling the laser energy density of nanosecond 193 nm laser radiation. J. Anal. At. Spectrom. (2019) 34 1957-1964. DOI: 10.1039/c9ja00126c
--- title: 'Screen exposure time of children under 6 years old: a French cross-sectional survey in general practices in the Auvergne-Rhône-Alpes region' authors: - Mehtap Akbayin - Aurélien Mulliez - Frédéric Fortin - Mathilde Vicard Olagne - Catherine Laporte - Philippe Vorilhon journal: BMC Primary Care year: 2023 pmcid: PMC9975848 doi: 10.1186/s12875-023-02009-5 license: CC BY 4.0 --- # Screen exposure time of children under 6 years old: a French cross-sectional survey in general practices in the Auvergne-Rhône-Alpes region ## Abstract ### Background The advent of miniature, easy-to-use and accessible multimedia products is leading to screen exposure that begins in early childhood. Overexposure in preschool may lead to adverse effects. The main objective of this study was to determine the average daily time (ADT) spent by children under 6 years of age, followed in general practice, in front of television or interactive screens. ### Methods A cross-sectional survey was conducted in the Auvergne-Rhône-Alpes region among randomly selected General Practitioners (GPs). The average daily screen time (ADST), regardless of the type of device (TVs, computers, tablets, smartphones, video game consoles), of the included children aged 0 to 2 years and 2 to 5 years was calculated from a self-questionnaire completed by the parents. A multivariate Poisson regression model was performed to analyse daily screen time, adjusted by factors selected on their clinical relevance and statistical significance. ### Results The 26 participating GPs included 486 parents. They reported an ADST of 26 (± 44) minutes on weekdays and 30 (± 46) minutes on weekends for children under 2 years of age. For children over 2 years of age, the ADST was 66 (± 82) minutes on weekdays and 103 (±91) minutes on weekends. There was an association between the children’s average screen time and certain sociodemographic and environmental factors. Children whose parents had higher levels of education, those living in a family without TV screens or those who were well informed about the possible adverse health consequences of overuse of screens had lower average screen time. On the other hand, children of parents who spent more than 2 hours a day in front of screens, were more exposed. ### Conclusions In our survey, the ADST of children under 6 years of age followed in general practice was higher than the current recommendations. GPs can warn parents of preschool children of the effects of overexposure to screens, particularly parents of at-risk children. ## Background Screens (televisions, computers, smartphones, tablets, video game consoles) are becoming an increasingly important part of children’s lives, starting at an early age. In the United States, before the COVID-19 pandemic, children under 2 years of age spent an average of 49 minutes per day in front of a screen and the average was 2 hours 30 minutes for children between 2 and 4 years of age [1]. Television and online videos were viewed the most by children under 6 years old, with $70\%$ of parents believing that they were beneficial to the children [1]. In France, studies have also confirmed this exposure from an early age [2, 3]. The effects of overexposure to screens during this critical period of brain development are beginning to be understood [4]. Numerous studies have demonstrated the harmful effects of screens on the cognitive development of children [5–9] and on their academic success [10, 11]. An association between time spent in front of mobile screens and behavioural difficulties (attention disorders, hyperactivity) has been shown in preschool children [12]. It has also been shown that exposure to screens favours sedentary behaviour and therefore obesity [7, 9, 13–16]. It also impacts sleep quality [7, 9, 17–19], language development [5–7, 9] and vision [20, 21]. Based on these findings, professional recommendations advise against exposure before the age of 18–24 months and limiting it to less than 1 h per day between the ages of 2 and 5 years, giving preference to parental guidance associated with high quality and interactive programs [9, 22–24]. Primary care practitioners, paediatricians and general practitioners (GPs) can relay this advice. Few studies have been conducted on the subject in general practice. However, assessing family habits can enable GPs to provide prevention messages to parents [25]. Parental education is a modifiable factor, and interventions for parents to promote early childhood development have been shown to be effective [26]. The primary objective of this work was to determine the daily time spent by a child under 6 years of age in front of a television or interactive screens. The secondary objectives were to assess screen use patterns in this age group and to identify factors associated with increased screen time. ## Design We conducted a descriptive cross-sectional survey of parents of children under 6 years of age in the Auvergne-Rhône-Alpes region in January 2019. This region, the second largest in France in terms of population size, is quite representative of the whole France considering that it contains urban areas with high population density and rural areas. ## Study population General practices were drawn at random from the list of GPs of the Auvergne-Rhône-Alpes Regional Health Agency, and parents of children who received medical care from the selected GPs were requested to participate to the study. Random selection was made using Stata software. The objectives and modalities of the study were presented to the GPs by telephone call before obtaining their oral agreement to participate. No response after three telephone calls was considered a refusal. Thirty questionnaires per GP were distributed. They were offered to parents by the medical secretaries or left directly in the waiting room. A collection box was made available for the participants to deposit the completed questionnaires. An information note was given to each parent, explaining the objectives of the study, the conditions of anonymization and the confidentiality of the data collected. The GPs participating in the study were contacted twice during the 4 weeks of inclusion. All parents with at least one child under 6 years of age who agreed to participate in the survey were eligible for inclusion. Parents with several children under the age of 6 were asked to complete the questionnaire for the youngest child. Exclusion criteria were parents of children older than 6 years or those with difficulties with the French language. ## Ethics consideration Ethical approval for the survey was obtained on the 5th of June 2019 from the Ethics Committee of the ‘Collège National des Généralistes Enseignants’ (CNGE) under number 16051998. The study was performed according to good research practices and the Declaration of Helsinki. Participants received an explanatory letter and an anonymous questionnaire by post. Written Informed consent was obtained from the participants. ## Questionnaire We designed a questionnaire with 32 questions divided into 3 parts. It was based on the latest position statements of the French Paediatric Society (SFP) [24]. To evaluate its comprehension and acceptability, the questionnaire was pretested with 30 parents with different sociodemographic characteristics. It was composed of the following 3 parts:The first part included sociodemographic data about the parents: age, sex, family situation, level of education, socio-professional category, and place of residence. Two questions asked the parents about their child’s daily screen time during the week and on the weekends. The second part was concerned with screen equipment in the child’s home, including in their bedroom. TV screens were distinguished from other multimedia screens (smartphones, computers, video game consoles, tablets).The third part included questions about the child: month and year of birth, sex, the presence of older siblings, and the number of children in the household. Two questions were designed to estimate the child’s average daily screen time (ADST), both during the week and on the weekends. The parents were also asked to identify the total amount of screen time on the day before the survey to obtain greater precision and to check the consistency of the responses. Data on the children’s exposure times and parental guidance were collected. A 6-point scale was used for the responses (never, rarely, sometimes, regularly, often, always). Each of the quantifying adverbs was associated with a temporality for a better objectivity of the answers (never, 1 to 2 times per month, 1 to 2 times per week, 3 to 4 times per week, 5 to 6 times per week, every day). Finally, information that assessed the parents’ knowledge of the effects of screen misuse, the establishment of family rules and their interest in discussing the subject with a health professional was collected. ## Endpoint The primary endpoint was the child’s ADST regardless of the type of device (TVs, computers, tablets, smartphones, video game consoles), reported by the parents in minutes and distinguishing weekdays from weekends. ## Sample size The objective was to include 500 questionnaires. This sample size was considered to be large enough to have representative data and accurate estimations. We assumed that for a 4-week inclusion period, approximately 20 to 30 questionnaires would be collected per GP. Thus, we needed the participation of at least 25 GPs to reach 500 questionnaires. Of the 6201 general practitioners in the Auvergne-Rhône-Alpes region, 150 were selected after randomisation. In case of refusal of a general practitioner drawn at random, this one would be replaced by a doctor of the same gender (the next one in the list) in order to keep a gender balance. ## Statistical analysis The study sample is described by numbers and percentages for categorical data and by means ± standard deviations and medians and interquartile ranges for continuous data. Daily screen time was graphically analysed, and normality was assessed using Shapiro–Wilk’s test. Daily screen time (weekdays and weekend days) was analysed using the Mann–Whitney test for two group comparisons and the Kruskal–Wallis test for 3 (or more) groups. Relationships between daily screen time and continuous data were analysed using Spearman’s correlation coefficient. In order to perform a multivariate Poisson regression of daily screen time, we first transformed screen time into event of exposure, considering 1 h of screen exposure as one event. Thus, children with no screen exposure were considered having 0 event, children with exposure between 1 minute and 60 minutes were considered having 1 event, children with exposure between 61 minutes and 120 minutes were considered having 2 events and so on. Then, a multivariate Poisson regression model was performed considering those screen time events as dependant variable and adjusted for factors selected according to their clinical relevance or statistical significance ($p \leq 0.15$) in univariate analysis. Results are shown as incidence rate ratios and their $95\%$ confidence interval. Analyses were performed using Stata (version 15, StataCorp, College Station). ## Description of the study population A total of 26 GPs agreed to participate in the study. Their characteristics are described in Table 1. Among the 780 questionnaires distributed, 531 were collected, and 486 were analysed. Thirty-six patients who did not meet the inclusion criteria and 9 who partially completed the study were excluded (Fig. 1). A total of $81.2\%$ ($$n = 394$$) of the questionnaires were completed by mothers; the mean age of the respondents was 34 years (± 5.1). Among the children, $46.1\%$ ($$n = 224$$) were girls; the mean age was 3.7 years (±1.5) (Table 2). A total of 449 households ($93.4\%$) reported owning at least one television set (Fig. 2). In addition, 431 families ($93.7\%$) reported owning at least one smartphone, 393 reported owning at least one computer ($84.2\%$) and 208 reported owning a tablet ($44.4\%$).Table 1Characteristics of participating general practitionersVariablesSample $$n = 26$$Overall [2018] $$n = 7419$$P-valueSex, n (%) Female13 [50]3340 (45.0)0.61 Male13 [50]4079 (55.0)Age, mean (sd)45.6 ± 1352.0 (± NA)NCNumber of years of practice14.2 ± 14.017.5 (± NA)NCLocation, n (%), ($$n = 7174$$) Rural6 (23.1)1038 (14.5)0.21 Urban20 (76.9)6136 (85.5) N missing0245Type of practice Individual practice, n (%)12 (46.1)4664 (62.9)0.08 Group practice, n (%)14 (53.8)2755 (37.1)Other GP teacher, n (%)7 (26.9)1590 (21.4)0.50Health insurance data, as of 31 January 2018NA Not available, NC Not computableFig. 1Flow-chart of the participantsTable 2Characteristics of the study population and daily weekdays screen timeSample characteristicsN = 486Weekdays screen time (minutes) Median [IQR]P valueChildren, $$n = 486$$ Age, mean ± standard deviation (sd)3.7 ± 1.5 ≤ 2 years, n (%)75 (15.4)0 [0–30]< 0.001 2–6 years, n (%)411 (84.6)50 [20–90]Gender Female, n (%)224 (46.1)30 [5–60]0.02 Male, n (%)262 (53.9)48 [20–90]Older siblings Yes n (%)262 (54.1)43 [15–90]0.18 No, n (%)222 (45.9)30 [10–60]TV in bedroom Yes n (%)27 (5.6)120 [60–150]< 0.001 No, n (%)456 (94.4)30 [10–60]Parents, $$n = 486$$ Age, mean ± sd33.9 ± 5.1 Number of children, mean ± sd2.01 ± 0.81 1, n (%)125 (25.7)30 [5–60]0.077 2, n (%)261 (53.7)30 [15–90] 3, n (%)77 (15.5)60 [30–60] ≥ 4, n (%)23 (4.7)60 [20–120] Gender Female, n (%)394 (81.2)30 [12–60]0.83 Male, n (%)91 (18.8)45 [15–66] Family situation In couple, n (%)438 (90.3)30 [10–60]< 0.001 Alone, n (%)47 (9.7)60 [30–120] Residence Urban, n (%)265 (54.5)30 [10–90]0.81 Rural, n (%)221 (45.5)30 [20–60] Socio-professional category Farmers, n (%)2 (0.4)40 [20–60]< 0.001 Craftsmen, merchants, business managers, n (%)25 (5.2)20 [0–60] Executives and intellectual professions, n (%)104 (21.6)30 [2–60] Intermediate professions, n (%)39 (8.1)30 [5–60] Employees, n (%)205 (42.5)50 [20–90] Workers, n (%)23 (4.8)60 [30–120] Retired, n (%)2 (0.4)165 [150–180] No activty, n (%)82 [17]60 [30–120] Education level Primary School24 (5.0)60 [45–120]< 0.001 Secondary School158 (32.8)60 [30–120] Bachelor’s degree or above299 (62.230 [5–60]Medians expressed in minutesFig. 2Distribution of households by type and frequency of existing screen equipment ## Average Daily Screen Time (ADST) The ADST for children under 2 years of age on weekdays was 26 minutes (±44.0), with a median IQR of 0 [0–30] (see Table 2 and Fig. 3) and that on weekends was 30 minutes (±46.9), with a median IQR of 0 [0–50]. The ADST of children older than 2 years on weekdays was 66 minutes (±82.1), with a median IQR of 50 [20–90] and that on weekends was 103 minutes (±91.3), with a median IQR of 90 [45–120]. Both daily weekday and weekend average screen times reported by the parents were significantly higher for boys than for girls (weekday with a median IQR of 48 [20–90], weekend days with a median IQR of 30 [5–60], $$P \leq 0.02$$, respectively) (Fig. 3). The ADST increased significantly with the age of the child ($p \leq 0.001$).Fig. 3Children daily weekdays screen time according to their characteristics According to the parents, 316 children ($66\%$) had watched a TV and/or multimedia screen the day before the questionnaire. The previous day’s average screen time was 43 minutes (±57) across all age groups. Children’s ADSTs increased concomitantly with parents’ ADSTs, both on weekdays (correlation coefficient $r = 0.31$, $p \leq 0.001$) and weekends ($r = 0.37$, $p \leq 0.001$). ## Children’s screen use habits According to the parents, 456 children ($94.4\%$) did not have a screen in their room, and 443 children ($91.5\%$) never used a multimedia screen in their room. Children who had a TV in their room had higher weekday and weekend ADSTs than those who did not (124 ± 91.8, median 120 [60–150] vs. 56.0 ± 76.6, median 30 [10–60], $P \leq 0.001$ on weekdays and 187 ± 131, median 150 [120–240] vs. 86.0 ± 83.7, median 60 [30–120], $P \leq 0.001$ on weekends, respectively) (see Table 2 and Fig. 3). One hundred and two parents ($25.4\%$) said they never discussed the content of the program their child watched, and for 72 ($14.8\%$) of them, the child was regularly or always alone in front of a screen. Concerning TV, 119 parents ($24.7\%$) estimated that their child watched TV during the week before school, and 62 parents ($12.7\%$) watched TV in the evening before bedtime. Furthermore, 53 children ($10.9\%$) were regularly present when their parents watched TV, and 45 children ($9.2\%$) were always present. In 63 households ($12.9\%$), the television was always on during meals. Finally, $78.8\%$ of the parents in our study wanted to discuss screen recommendations with their general practitioner. ## Factors associated with increased screen time – multivariate analysis The multivariate Poisson regression model adjusted for clinically relevant criteria showed that, compared to Secondary School parents, the higher the parents’ level of education, the less the children watched a screen (IRR = 0.79, $95\%$CI [0.66–0.94] (Tables 3 and 4 and Fig. 4). Compared to one TV at home (reference), on weekdays, children with no TV were less exposed (IRR = 0.67, $95\%$CI [0.45–1.01], $$p \leq 0.053$$), than children with three or more television screen (IRR = 1.38, $95\%$CI [0.99–1.92], $$p \leq 0.059$$). As well on weekdays as on weekends, when parent’s daily screen time was > 2 hours, children exposure increased, with IRR = 1.34, $95\%$CI [1.13–1.59], $$p \leq 0.001.$$ Compared to parents who were not aware of any harmful effects of screen overexposure considered as reference, the children of parents who were aware of at least 3 adverse effects had lower ADSTs during the weekdays and weekends. Finally, children from single-parent families tend to be more exposed than others on weekdays with IRR = 1,23, $95\%$ CI [0.9–1.65], $$p \leq 0.162$$ and weekends with IRR = 1.17, $95\%$ CI [0.9–1.47], $$p \leq 0.165$$ and weekends. Table 3Multivariate Poisson regression of weekdays exposureIRR$95\%$ CIP valueFamily situation Maried or Couple (Ref.) Single1.23[0.9–1.65]0.162Education level Primary School1.09[0.8–1.57]0.633 Secondary School (Ref.) Bachelor’s degree or above0.79[0.7–0.94]0.009TEM parents Parent’s screen time ≤ 2 hours (Ref.) Parent’s screen time > 2 hours1.34[1.1–1.59]0.001Number of TV at home 00.67[0.5–1.01]0.053 1 (Ref.) 21.17[0.9–1.47]0.173 3+1.38[1–1.92]0.059Number of nefast effects 0 (Ref.) 11.02[0.8–1.29]0.881 20.88[0.7–1.11]0.286 3+0.71[0.6–0.88]0.002Residence Urban (Ref.) Rural0.9[0.8–1.05]0.185Child age ≤ 2 years old (Ref.) > 2 years old1.95[1.5–2.61]< 0.001Child gender Boy (Ref.) Girl0.92[0.8–1.05]0.218Table 4Multivariate Poisson regression of weekend exposureIRR$95\%$ CIP valueFamily situation Maried or Couple (Ref.) Single1.17[0.9–1.47]0.165Education level Primary School1.22[1–1.56]0.103 Secondary School (Ref.) Bachelor’s degree or above0.85[0.8–0.97]0.016TEM parents Parent’s screen time ≤ 2 hours (Ref.) Parent’s screen time > 2 hours1.46[1.3–1.64]0Number of TV at home 00.81[0.6–1.11]0.184 1 (Ref.) 21.14[0.9–1.38]0.177 3+1.16[0.9–1.45]0.205Number of nefast effects 0 (Ref.) 10.86[0.7–1.01]0.068 20.87[0.7–1.02]0.084 3+0.78[0.6–0.95]0.013Residence Urban (Ref.) Rural1.05[0.9–1.18]0.365Child age ≤ 2 years old (Ref.) > 2 years old2.6[1.9–3.5]< 0.001Child gender Boy (Ref.) Girl0.85[0.8–0.96]0.009Fig. 4Factors associated with weekdays and weekend days screen time, according to multivariate Poisson regression ## Main results Our study which focuses on parents of children followed in general practice, highlights that children of parents with Bachelor’s degree or above education level were less exposed to screens. On weekdays, those without TV at home had less screen time. On the other hand, the children of parents who were well informed about the possible adverse health consequences of overuse of screens had lower average screen time. On weekend days, the ADST of girls was less important than boys. The ADST tended to increase among children from single-parent families on weekdays and weekend days. Finally, there was no difference by rural or urban location. ## Comparison with literature data We did not find similar studies among a population of patients followed in general practice. Several studies in primary care involved patients followed in paediatric clinics or offices [2, 27]. In our study, both weekday and weekend ADSTs were lower than those reported in American data* from 2020 [1]. Indeed, before the COVID-19 pandemic and the implementation of the first containment, American children under 2 years of age spent an average of 49 minutes per day in front of screens. Those aged 2 to 4 years had a daily screen time of 150 minutes per day. It should be explain by the fact that our data predate the American data and that all studies tend to show a steady increase in screen time among children [12, 14, 28, 29]. Our results for children under 2 years of age are also lower than those of an Australian study published in 2016 that revealed a daily screen time of more than 2 hours for $40\%$ of 18-month-olds [30]. This difference might be caused by our age range including children younger than 18 months with less screen time. Our study took place just before the Covid 19 pandemic. It can be assumed that the Covid pandemic changed children’s screen use behaviours, at least for older children. During the first wave of the pandemic, some studies assessed screen time among preschoolers. In an international study, parents reported an average increase of nearly 1 h of screen time per day in 3- to 7-year-old children [31]. This increase was largely due to their use for entertainment purposes. Another study, by Fitzpatrick et al., found an increase in screen time specially before bedtime [32]. In this study, children’s age and parental use of multimedia screens were factors associated with increased screen time, but teleworking parents were less likely to have overexposed children. Not surprisingly, older children with online schooling requirements spent more time in front of a screen at first containment. In an other international survey in children under the age of three, this increase in screen time was also confirmed during the first lockdown [33]. Our study also found that some environmental factors may influence children’s screen time. In similar studies, the number of TV screens in the home is associated with children’s screen time [34, 35]. In the same way, living in a single-parent family is a contributing factor to increased screen time. However, although there is an upward trend in our study, the association is not significant. One of the explanations for this is that screens can be a means of distracting or calming children when the parent is engaged in certain tasks, which would seem to be even more valid in a single-parent family [28, 36]. In studies, children’s screen time is commonly associated with parental screen time [29, 35]. Lauricella et al. found that the amount of time parents used a multimedia technology (computers, tablets, smartphones) was associated with the amount of time the child used that same technology [37]. In other words, parental behaviours and habits with respect to multimedia screens strongly influence those of the children. Birkin et al. and Matta et al. have established that the regulation of their own use of new technologies by parents allows to avoid the reproduction of harmful behaviours by their children. This can be done by setting up parental rules [35, 38]. Parental education level is associated with children’s screen time [14, 38]. Atkin et al. found that children of mothers with low levels of education were more likely to exceed 2 h/day of screen time [14] and for Kiliç et al. the frequency of tablet use and ownership among children was inversely related to maternal education and household income [28]. But for Paudel et al. review, the association between educational status and children’s mobile screen media use is not really demonstrated [39]. Some studies showed that children’s mobile screen use increases when parents perceive beneficial effects and educational value [37]. Similarly, negative parental beliefs about screen-based mobile media are associated with decreased screen use [40]. In addition, a French study conducted in 2019 showed that parents were mainly misinformed about the risk of obesity [25]. Finally, our data demonstrate that the more knowledge parents had about the harmful effects associated with overuse of screens, the more children’s screen time decreased. This finding is particularly interesting because there are relatively few data in the literature on this aspect. ## Strengths and limitations of the study The number of parents included reached 500, giving enough statistical power to the study. The final sample of GPs was representative of the region’s GPs for sex, age, and practice location. However, the results must be interpreted considering a number of limitations. The survey was systematically proposed to parents of children under 6 years of age, but this did not exclude a possible selection bias by the secretaries or GPs offering the questionnaires. Furthermore, parents who were involved and aware of this subject could easily agree to answer the questionnaire. Another selection bias was related to the nonparticipation of populations with a language barrier. The effect of non-participating population is difficult to interpret regarding the data available in the literature. This population may include people with low education level, and/or migrants. We could make the assumption that this exclusion underestimates the screen time of our population. Indeed, in her study about children under 3 years of age, Duch showed a positive correlation between screen time and ethnic minority status [41]. This, together with the low rate of participating GPs, limits the extrapolation of the results to the entire French paediatric population. The questionnaires were anonymously completed by the patients before or after the consultation and placed in a collection box. This method allowed the parents to be as honest as possible in their answers, but as this was a declarative survey, it does not exclude a possible social desirability bias, which could lead to a minimisation of the ADST. For our primary endpoint, we used the parents’ global memory to determine children’s ADSTs on weekdays and weekends. A memory bias is possible because of the difficulty of providing synthetic and global data for a usual practice. This bias can be balanced by the fact that the parents’ assessments of the previous day’s screen time across all age groups was lower than their overall estimates of daily weekday and weekend average screen times. In addition, our study did not assess daily exposure times for every type of screen. ## Perspectives on care An Australian study found that health-related habits in families crystallise most easily in early childhood [42]. Thus, educational measures regarding sensible screen use should be implemented in early childhood to promote appropriate use. In our study, $45\%$ of the parents had established rules for the use of screens, and the children of parents who were aware of several harmful effects had less screen time. If we add that nearly $80\%$ of the parents would like more information on the subject, these data should encourage primary care practitioners, especially GPs, to screen children at risk of overexposure and to sensitize their families to implement preventive measures. ## Conclusion Our study of parents of preschool children followed in general practice showed early exposure to screens with daily average screen times beyond the professional recommendations. Intervention by GPs is legitimate to assess screen use and identify children at risk of overuse. Communication with families can provide the explanations that parents often expect and develop prevention advice. ## References 1. Rideout VJ, Robb MB. The common sense census: media use by kids age zero to eight (2020) 2. Assathiany R, Guery E, Caron FM, Cheymol J, Picherot G, Foucaud P. Children and screens: a survey by French pediatricians. Arch Pediatr (2018) 25 84-88. DOI: 10.1016/j.arcped.2017.11.001 3. Gassama M, Bernard J, Dargent-Molina P, Charles M-A. Activités physiques et usage des écrans à l’âge de 2 ans chez les enfants de la cohorte Elfe (2018) 4. Black MM, Walker SP, Fernald LCH, Andersen CT. Advancing early childhood development: from science to scale 1. Lancet. (2017) 389 77-90. DOI: 10.1016/S0140-6736(16)31389-7 5. Harlé B, Desmurget M. Effets de l’exposition chronique aux écrans sur le développement cognitif de l’enfant. Arch Pédiatr (2012) 19 772-776. DOI: 10.1016/j.arcped.2012.04.003 6. Tomopoulos S, Dreyer BP, Berkule S, Fierman AH, Brockmeyer C, Mendelsohn AL. Infant media exposure and toddler development. Arch Pediatr Adolesc Med (2010) 164 1105-1111. DOI: 10.1001/archpediatrics.2010.235 7. Radesky JS, Christakis DA. Increased screen time: implications for early childhood development and behavior. Pediatr Clin N Am (2016) 63 827-839. DOI: 10.1016/j.pcl.2016.06.006 8. Yang X, Chen Z, Wang Z, Zhu L. The relations between television exposure and executive function in Chinese preschoolers: the moderated role of parental mediation behaviors. Front Psychol (2017) 8 1833. DOI: 10.3389/fpsyg.2017.01833 9. 9.Société canadienne de pédiatrieLe temps d’écran et les jeunes enfants : promouvoir la santé et le développement dans un monde numérique2017. Le temps d’écran et les jeunes enfants : promouvoir la santé et le développement dans un monde numérique (2017) 10. Hancox RJ, Milne BJ, Poulton R. Association of Television Viewing during Childhood with Poor Educational Achievement. Arch Pediatr Adolesc Med (2005) 159 614-618. DOI: 10.1001/archpedi.159.7.614 11. Poulain T, Peschel T, Vogel M, Jurkutat A, Kiess W. Cross-sectional and longitudinal associations of screen time and physical activity with school performance at different types of secondary school. BMC Public Health (2018) 18 563. DOI: 10.1186/s12889-018-5489-3 12. Poulain T, Vogel M, Neef M, Abicht F, Hilbert A, Genuneit J. Reciprocal associations between electronic media use and behavioral difficulties in preschoolers. Int J Environ Res Public Health (2018) 15 814. DOI: 10.3390/ijerph15040814 13. Colley RC, Garriguet D, Adamo KB, Carson V, Janssen I, Timmons BW. Physical activity and sedentary behavior during the early years in Canada: a cross-sectional study. Int J Behav Nutr Phys Act (2013) 10 54. DOI: 10.1186/1479-5868-10-54 14. Atkin AJ, Sharp SJ, Corder K, van Sluijs EMF. International Children’s Accelerometry database (ICAD) collaborators. Prevalence and correlates of screen time in youth: an international perspective. Am J Prev Med (2014) 47 803-807. DOI: 10.1016/j.amepre.2014.07.043 15. Hingle M, Kunkel D. Childhood obesity and the media. Pediatr Clin N Am (2012) 59 677-692. DOI: 10.1016/j.pcl.2012.03.021 16. Veldhuis L, Vogel I, Renders CM, van Rossem L, Oenema A, HiraSing RA. Behavioral risk factors for overweight in early childhood; the ‘be active, eat right’ study. Int J Behav Nutr Phys Act (2012) 9 74. DOI: 10.1186/1479-5868-9-74 17. Cain N, Gradisar M. Electronic media use and sleep in school-aged children and adolescents: a review. Sleep Med (2010) 11 735-742. DOI: 10.1016/j.sleep.2010.02.006 18. Carter B, Rees P, Hale L, Bhattacharjee D, Paradkar M. A meta-analysis of the effect of media devices on sleep outcomes. JAMA Pediatr (2016) 170 1202-1208. DOI: 10.1001/jamapediatrics.2016.2341 19. Brockmann PE, Diaz B, Damiani F, Villarroel L, Núñez F, Bruni O. Impact of television on the quality of sleep in preschool children. Sleep Med (2016) 20 140-144. DOI: 10.1016/j.sleep.2015.06.005 20. Chatard H. Toxicité des écrans sur la vision : quelles sont les conséquences de l’utilisation excessive des écrans chez l’enfant et le jeune adulte ?. Rev Francophone d’Orthop (2017) 10 129-131. DOI: 10.1016/j.rfo.2017.10.003 21. Lee HS, Park SW, Heo H. Acute acquired comitant esotropia related to excessive smartphone use. BMC Ophthalmol (2016) 16 37. DOI: 10.1186/s12886-016-0213-5 22. Hill D, Ameenuddin N, Reid Chassiakos Y, Cross C, Hutchinson J. Media and Young Minds. Pediatrics (2016) 138 e20162591. DOI: 10.1542/peds.2016-2591 23. Bozzola E, Spina G, Ruggiero M, Memo L, Agostiniani R. Media devices in pre-school children: the recommendations of the Italian pediatric society. Ital J Pediatr (2018) 44 69. DOI: 10.1186/s13052-018-0508-7 24. Picherot G, Cheymol J, Assathiany R, Barthet-Derrien M-S, Bidet-Emeriau M. Children and screens: Groupe de Pédiatrie Générale (Société française de pédiatrie) guidelines for pediatricians and families. Arch Pediatr (2018) 25 170-174. DOI: 10.1016/j.arcped.2017.12.014 25. Delmestre J, Skendi M, De Oliveira A, Coupaud E, Chastang J, Lindivat A. Exposition aux écrans des jeunes enfants : connaissance des parents et inégalités sociales. Exercer. (2021) N°173 203-209. DOI: 10.56746/EXERCER.2022.173.203 26. Jeong J, Franchett EE, Ramos de Oliveira CV, Rehmani K, Yousafzai AK. Parenting interventions to promote early child development in the first three years of life: a global systematic review and meta-analysis. PLoS Med (2021) 18 e1003602. DOI: 10.1371/journal.pmed.1003602 27. Byrne R, Terranova CO, Trost SG. Measurement of screen time among young children aged 0–6 years: a systematic review. Obes Rev (2021) 22 e13260. DOI: 10.1111/obr.13260 28. Kılıç AO, Sari E, Yucel H, Oğuz MM, Polat E, Acoglu EA. Exposure to and use of mobile devices in children aged 1–60 months. Eur J Pediatr (2019) 178 221-227. DOI: 10.1007/s00431-018-3284-x 29. Goh SN, Teh LH, Tay WR, Anantharaman S, van Dam RM, Tan CS. Sociodemographic, home environment and parental influences on total and device-specific screen viewing in children aged 2 years and below: an observational study. BMJ Open (2016) 6 e009113. DOI: 10.1136/bmjopen-2015-009113 30. Chandra M, Jalaludin B, Woolfenden S, Descallar J. Screen time of infants in Sydney, Australia: a birth cohort study. BMJ Open (2016) 6 e012342. DOI: 10.1136/bmjopen-2016-012342 31. Ribner AD, Coulanges L, Friedman S, Libertus ME, Hughes C, Foley S. Screen time in the coronavirus 2019 era: international trends of increasing use among 3- to 7-year-old children. J Pediatr (2021) 239 59-66.e1. DOI: 10.1016/j.jpeds.2021.08.068 32. Fitzpatrick C, Almeida ML, Harvey E, Garon-Carrier G, Berrigan F, Asbridge M. An examination of bedtime media and excessive screen time by Canadian preschoolers during the COVID-19 pandemic. BMC Pediatr (2022) 22 212. DOI: 10.1186/s12887-022-03280-8 33. Bergmann C, Dimitrova N, Alaslani K, Almohammadi A, Alroqi H, Aussems S. Young children’s screen time during the first COVID-19 lockdown in 12 countries. Sci Rep (2022) 12 2015. DOI: 10.1038/s41598-022-05840-5 34. Jago R, Stamatakis E, Gama A, Carvalhal IM, Nogueira H, Rosado V. Parent and child screen-viewing time and home media environment. Am J Prev Med (2012) 43 150-158. DOI: 10.1016/j.amepre.2012.04.012 35. Birken CS, Maguire J, Mekky M, Manlhiot C, Beck CE, Jacobson S. Parental factors associated with screen time in pre-school children in primary-care practice: a TARGet kids! Study. Public Health Nutr (2011) 14 2134-2138. DOI: 10.1017/S1368980011000516 36. De Decker E, De Craemer M, De Bourdeaudhuij I, Wijndaele K, Duvinage K. Influencing factors of screen time in preschool children: an exploration of parents’ perceptions through focus groups in six European countries. Obes Rev (2012) 13 75-78. DOI: 10.1111/j.1467-789X.2011.00961.x 37. Lauricella AR, Wartella E, Rideout VJ. Young children’s screen time: the complex role of parent and child factors. J Appl Dev Psychol (2015) 36 11-17. DOI: 10.1016/j.appdev.2014.12.001 38. Määttä S, Kaukonen R, Vepsäläinen H, Lehto E, Ylönen A, Ray C. The mediating role of the home environment in relation to parental educational level and preschool children’s screen time: a cross-sectional study. BMC Public Health (2017) 17 688. DOI: 10.1186/s12889-017-4694-9 39. Paudel S, Jancey J, Subedi N, Leavy J. Correlates of mobile screen media use among children aged 0–8: a systematic review. BMJ Open (2017) 7 e014585. DOI: 10.1136/bmjopen-2016-014585 40. Nikken P, Schols M. How and why parents guide the media use of young children. J Child Fam Stud (2015) 24 3423-3435. DOI: 10.1007/s10826-015-0144-4 41. Duch H, Fisher HM, Ensari I, Harrington A. Screen time use in children under 3 years old: a systematic review of correlates. Int J Behav Nutr Phys Act (2013) 10 102. DOI: 10.1186/1479-5868-10-102 42. Hamilton K, Spinks T, White KM, Kavanagh DJ, Walsh AM. A psychosocial analysis of parents’ decisions for limiting their young child’s screen time: an examination of attitudes, social norms and roles, and control perceptions. Br J Health Psychol (2016) 21 285-301. DOI: 10.1111/bjhp.12168
--- title: Excess healthcare costs of mental disorders in children, adolescents and young adults in the Basque population registry adjusted for socioeconomic status and sex authors: - Igor Larrañaga - Oliver Ibarrondo - Lorea Mar-Barrutia - Myriam Soto-Gordoa - Javier Mar journal: 'Cost Effectiveness and Resource Allocation : C/E' year: 2023 pmcid: PMC9975849 doi: 10.1186/s12962-023-00428-w license: CC BY 4.0 --- # Excess healthcare costs of mental disorders in children, adolescents and young adults in the Basque population registry adjusted for socioeconomic status and sex ## Abstract ### Background Mental illnesses account for a considerable proportion of the global burden of disease. Economic evaluation of public policies and interventions aimed at mental health is crucial to inform decisions and improve the provision of healthcare services, but experts highlight that nowadays the cost implications of mental illness are not properly quantified. The objective was to measure the costs of excess use of all healthcare services by 1- to 30-year-olds in the Basque population as a function of whether or not they had a mental disorder diagnosis. ### Methods A real-world data study was used to identify diagnoses of mental disorders and to measure resource use in the Basque Health Service Registry in 2018. Diagnoses were aggregated into eight diagnostic clusters: anxiety, attention deficit hyperactivity disorder, conduct disorders, mood disorders, substance use, psychosis and personality disorders, eating disorders, and self-harm. We calculated the costs incurred by each individual by multiplying the resource use by the unit costs. Annual costs for each cluster were compared with those for individuals with no diagnosed mental disorders through entropy balancing and two-part models which adjusted for socioeconomic status (SES). ### Results Of the 609,381 individuals included, 96,671 ($15.9\%$) had ≥ 1 mental disorder diagnosis. The annual cost per person was two-fold higher in the group diagnosed with mental disorders (€699.7) than that with no diagnoses (€274.6). For all clusters, annual excess costs associated with mental disorders were significant. The adjustment also evidenced a social gradient in healthcare costs, individuals with lower SES consuming more resources than those with medium and higher SES across all clusters. Nonetheless, the effect of being diagnosed with a mental disorder had a greater impact on the mean and excess costs than SES. ### Conclusions Results were consistent in showing that young people with mental disorders place a greater burden on healthcare services. Excess costs were higher for severe mental disorders like self-harm and psychoses, and lower SES individuals incurred, overall, more than twice the costs per person with no diagnoses. A socioeconomic gradient was notable, excess costs being higher in low SES individuals than those with a high-to-medium SES. Differences by sex were also statistically significant but their sizes were smaller than those related to SES. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12962-023-00428-w. ## Background Mental illnesses account for a considerable proportion ($10\%$) of the global burden of disease [1, 2]. The literature suggests that preventive interventions at early age are key to tackle adverse conditions experienced during childhood and adolescence and contribute to better levels of health in adulthood [3]. Therefore, economic evaluation of public policies and interventions aimed to reduce the burden is crucial to inform decisions about what is the best use of the limited resources available in order to maximize the health benefits [4]. Nonetheless, the cost of mental disorders is a poorly understood driver of decision-making about which interventions should be implemented in mental health [5, 6]. More economic evaluations in the field of mental illness have started to be conducted [7–10], but their limited use in decision-making contrasts with the importance placed on this type of research in the incorporation of preventive treatments and interventions in cancer and cardiovascular diseases [11–13]. Moreover, experts highlight that the cost implications are not adequately measured and large evidence gaps still exist regarding the economic case for mental health care [14], including inequalities by gender and socioeconomic status (SES) [6]. In this context, data are needed on excess healthcare costs associated with mental disorders for various purposes, such as conducting economic evaluations and measuring the burden [15]. To measure disease-specific burden, average costs should be disaggregated by the type of mental disorder and compared with those for similar populations without such disorders [15–18]. That is, there is a need to know not only the average cost but also the incremental cost in relation to the population without mental health diagnoses [17, 18]. In this sense, if a new intervention can modify the costs or benefits in health associated to a given mental disorder, the burden of the new scenario proposed by the intervention can be obtained and compared with the current scenario in order to guide decision-making [5, 6]. Currently, the relatively few data available are based on surveys collecting self-reported data from samples of patients with diagnoses of specific mental disorders [8, 17–19]. In this field, there is a lack of real-world data (RWD) studies, despite such research having been recognized by experts as a key source of information for understanding disease-specific resource use [20]. As RWD provide information on individual resource use for an entire population, analysis of these types of data makes it possible to measure population costs [21]. In turn, having population data disaggregated at the individual level, accurate unit costs can be provided to be used to estimate the economic burden of health disorders and to carry out subsequent cost analysis of interventions [3]. RWD also help explore differences in health, since groups that make greater use of health resources are generally those that have poorer health status [22, 23]. The monitoring of health disparities allows us to measure progress toward achieving health equity and social justice [24]. As people diagnosed with a mental disorder tend to use health services more than the general population, health service use may reveal trends in disparities in mental health [25, 26]. Besides mental disorders, inequities in resource use can also be associated with SES and mental disorders in populations that are strongly determined by socioeconomic characteristics [23, 27, 28]. Moreover, due to a greater vulnerability to environmental stress in the early stages of life, social determinants have more impact on children, adolescents and young adults [29, 30]. Therefore, knowing the joint impact of mental disorders and SES on healthcare costs would help to assess both, their burden and their relationship with social determinants [31]. Given all this, the objective of this study was to measure the excess use of healthcare resources and healthcare costs of people between 1 and 30 years of age in the Basque population, adjusted for SES and sex, as a function of whether or not they are diagnosed with some type of mental disorder. ## Methods A retrospective observational study was conducted to identify diagnoses of mental disorders and to measure resource use based on data from the Basque Health Service. The Basque Country is an industrialized northern region of Spain with a population of 2.2 million. In Spain, powers for managing health services are decentralised to the regions, and the health system recognises a universal right to healthcare under a Beveridge model. The Basque Health Service provides comprehensive healthcare to the entire Basque population. The protocol of the study was approved by the Clinical Research Ethics Committee of the Basque Country (number PI2019078). The Basque Health Service registry contains information on all psychiatric and somatic inpatient and outpatient encounters (admissions and consultations), primary care contacts and emergency room visits. Diagnoses are recorded using codes from the ninth and tenth revisions of the International Classifications of Diseases (ICD-9 and ICD-10). In the study, the definition of lifetime prevalence of Kessler et al. was applied, who estimated it as the proportion of respondents who had ever been diagnosed with a given disorder up to their age at interview [32]. Based on this prevalence-based approach, we calculated, first, the resources used (primary care, mental health centres, hospitals and pharmacy) and, second, the corresponding direct costs. By merging diagnoses and resource use in the population registry, we obtained individual data for the whole population disaggregated by clusters of mental disorders. The study population was all individuals who, as of 31 December 2018, were between 1 and 30 years old and registered with the Basque Health Service. Among this population, patients who had been diagnosed with any mental disorder at any point in their lifetime were identified by checking all lifetime episodes of primary care and hospital care. Diagnoses were aggregated into eight diagnostic clusters: anxiety (anxiety + acute stress reactions + adaptation reactions), attention deficit hyperactivity disorder (ADHD), conduct disorders, mood disorders (depression + bipolar disorder), substance use, psychosis and personality disorders, eating disorders, and self-harm. In addition, as patients from private practice seek drug reimbursement through the public system, we searched for individuals who had any relevant chronic prescriptions through Anatomical Therapeutic Chemical (ATC) codes for antidepressants (N06A group) or antipsychotics (N05A group) in individuals without a mental disorder diagnosis in the public health service records to include them in the clusters of mood disorders and psychosis respectively. In the identification process, we used the ICD-9, ICD-10 and ATC classification system codes (listed in Additional file 1: Table S1). The variables included in the study were: age, sex and income level based on drug co-payment, and diagnosis cluster. In addition, all the resource use of the target population was extracted for the year 2018. That way, the resource use profile of the general population was estimated. Data collected in primary care included all contacts with nurses and general practitioners at healthcare centres, at home or by telephone. For hospital care, we took into account all contacts with outpatient clinics, as well as with emergency and inpatient services. All the drugs prescribed to individuals were also considered. The information about the unit costs of different healthcare resources for 2018 in euros (EUR, €) was obtained from the accounting system of the Basque Health Service (Additional file 1: Table S2) and included all types of healthcare resources [salaries, diagnosis costs (Lab, Rx), equipment, investments, infrastructure (heating, electricity, cleaning services, etc.) and pharmacy]. To assess SES, we considered drug co-payment categories which are established based on household income (Additional file 1: Table S3). The contribution levels for the co-payment of medicines in the Spanish Health System were established in 2012 based on three criteria: income, age and degree of illness. Children, adolescents and young adults were assigned the most disadvantaged SES level (low SES) if the head of their household was exempt from co-payment or was retired, the most advantaged SES level (high SES) if the head of the household had an annual income from paid work equal or higher than €18,000, and otherwise, to a third category (medium SES), for heads of household with annual incomes from paid work lower than €18,000 [27]. We calculated the costs incurred by each individual multiplying the resource use by the unit costs. As individual data were available, the cost per patient was disaggregated into primary care, hospital care and pharmacy costs for each diagnostic cluster. ## Statistical analysis In the initial step, univariate statistical analysis was performed to compare the sociodemographic features of individuals with and without mental disorders. Fisher’s exact test was used for categorical variables with two categories and expected values less than or equal to 5, and otherwise a chi-square test. In the case of age, since it is a continuous variable with a normal distribution, the comparison of means was carried out using Student’s t test. In a second step, each diagnostic cluster was compared with the population with no diagnosed mental disorders to measure the excess cost using statistical models. In total, 10 independent statistical models were created: one for each diagnostic cluster, one for having two diagnoses or more, and one for having any mental disorder. Before developing the models, the data were pre-processed. When using nonrandomized studies to estimate costs, it must be taken into account that selection is influenced by individual characteristics. Since initial characteristics were likely to be different in the groups with and without mental disorders, they were balanced to ensure that they were comparable in terms of initial characteristics (age group, sex and SES) and independent of background characteristics. For that purpose, an entropy balancing technique was used to adjust the covariate distribution of the group with no diagnosed mental disorders by reweighting. This technique is based on a maximum entropy reweighting scheme and allows the pre-processing of data in observational studies with binary variables of interest [18, 33]. The technique reweighted the data from no diagnosed mental disorders units to match a set of moments that was computed from the data from the group with mental disorders. Hence, the covariate distribution obtained was more similar to that in the group with mental disorders. In that way, the covariate distributions in the reweighted data satisfied the balance conditions specified by the research team and the resulting weights were used to carry out an analysis comparing the two groups, where confounding factors between them were removed. As ten different models were built to analyse excess costs, covariate distribution adjustment using entropy balancing was also performed once per model. To measure the excess costs, it would not have been appropriate to use ordinary least squares regression models [34], since the costs did not follow a normal distribution and a substantial number of individuals had zero costs. Therefore, to obtain the excess costs for each mental health cluster and adjusted for the selected characteristics, regression analysis was performed using two-part models [18, 34, 35]. In the first part, the adjusted probability, p(x), that the cost was higher than zero was fitted with a logit regression model. In the second part, generalized linear models with a log link function and gamma distribution were used to calculate the mean cost values of the population with costs greater than zero. An advantage of two-part models is that their results are easy to interpret since they estimate the magnitude of the differences between groups (in our case, in costs) and not only if the compared means differ. As the estimated costs depend on the combination of the different covariates, they produce the average costs for each cluster and the adjusted excess costs. All the statistical analyses were carried out using R (version 3.3.2) and Stata (version 14) statistical programs with a significance level of $95\%$. Specifically, the initial univariate statistical analysis was performed with R, which is free, while the entropy balancing and two-part models were performed with Stata, to take advantage of dedicated packages available, namely, ebalance and twopm, respectively [36, 37]. ## Results The total population in the age range between 1 and 30 years included in the Basque Health Service database contained 609,381 individuals, of which 96,671 ($15.9\%$) had been diagnosed with at least one mental disorder at some point in their lifetime (Table 1). The SES distribution confirmed the social gradient in mental disorders, the prevalence rising from low SES to medium and high SES groups (also in Table 1). The lifetime prevalence of mental disorders and the use of resources in the population studied disaggregated by diagnostic cluster are presented in Table 2 and Additional file 1: Table S4 disaggregated by cluster. They show that the gradient according to SES is a pattern repeated in all the diagnostic clusters. Anxiety was the most prevalent type of mental disorder, diagnoses in this cluster being recorded in $6.6\%$ of the population under 30 years. Individuals with more than one diagnosis appear in various clusters. Among the entire population with mental disorders, $20\%$ had two or more diagnoses. Notably, Table 2 shows the greater use of healthcare resources by people with diagnosed mental disorders. Their rate of admission to psychiatric wards was higher ($0.7\%$) than that in people with no diagnoses ($0.0\%$). But notably their rate of admission to general wards was also nearly two-fold higher ($5.4\%$ versus $2.8\%$ in the population with no diagnosed mental disorders). Individuals with mental disorders also incurred noticeably higher annual drug prescription costs (€115.6 versus €33.7).Table 1Characteristics of the study population between 1 and 30 years of age as of 31 December 2018 with and without a diagnosed mental disorder (individuals 1–30 years of age, Basque Health Service registry 31 December 2018)Study populationDiagnosed mental disorderp-valueNoYesN%N%N%PatientsTotal609,381512,71084.196,67115.9Age (years)Mean15.5714.6920.23 < 0.001a0–12246,27540.4231,07993.815,1966.2 < 0.001b13–18123,10920.2100,58281.722,52718.318–24114,71018.887,91876.626,79223.425–30125,28720.693,13174.332,15625.7SexFemale296,55648.7251,82584.944,73115.1 < 0.001cMale312,82551.3260,88583.451,94016.6Socioeconomic statusLow47,4167.836,94577.910,47122.1 < 0.001bMedium312,13551.2254,30681.557,82918.5High249,83041.0221,45988.628,37111.4aCalculated using Student’s t test for continuous variablesbCalculated using chi-square testscCalculated using Fisher’s exact testTable 2Percentage of use of each resource over 1 year by the prevalence of mental disorders in the Basque population under 30 yearsPopulation sizePrimary care (%)Outpatient servicesEmergency services (%)Inpatient servicesIntensive care (%)Total (%)Psychiatry (%)Total (%)Psychiatry (%)General population609,38163.034.83.326.43.20.10.2Population with no diagnosed mental disorders512,710 ($84.1\%$)60.531.81.025.32.80.00.2Population with ≥ 1 diagnosed mental disorder96,671 ($15.9\%$)76.550.615.332.55.40.70.2Substance use19,507 ($3.2\%$)80.946.711.738.29.42.00.4Anxiety40,523 ($6.6\%$)83.352.514.936.36.51.00.3Mood disorders8,613 ($1.4\%$)81.069.742.038.111.76.10.5Psychosis and personality disorders5.745 ($0.9\%$)77.769.146.041.413.58.10.7Attention deficit hyperactivity disorder16,986 ($2.8\%$)71.651.118.427.93.90.60.2Conduct disorders26,415 ($4.3\%$)71.653.620.933.65.01.20.2Eating disorders4,629 ($0.8\%$)77.455.118.633.36.61.80.3Self-harm664 ($0.1\%$)85.572.752.359.027.120.32.12 or more diagnoses19,567 ($3.2\%$)83.561.929.641.39.63.30.4 The annual costs per person disaggregated by diagnostic group and cost component are listed in Table 3 and Additional file 1: Table S5. Hospital costs represented three quarters of the total cost. As patients with more than one diagnosis may be included in various clusters, the overall mean does not match the weighted average of the clusters. The total healthcare cost per person in the diagnosed group (€699.7) was more than twice that in the group with no diagnoses (€274.6). The clusters that consumed the most resources were self-harm, with mean costs of €4543.7, followed by psychosis and personality disorders with costs of €2359.8 and mood disorders with costs of €1874.7.Table 3Mean direct healthcare costs per person in € [2018] disaggregated by diagnostic group and cost componentTotal costsPrimary care costsNon-psychiatric hospital care costsPsychiatric hospital care costsDrug prescription costsGeneral population$342.044.813.1\%$$220.164.4\%$$30.48.9\%$$46.713.6\%$Population with no diagnosed mental disorder$274.637.113.5\%$$199.972.8\%$$3.91.4\%$$33.712.3\%$Population with ≥ 1 diagnosed mental disorder$699.785.712.2\%$$326.946.7\%$$171.524.5\%$$115.616.5\%$Substance use$1012.7107.410.6\%$$420.441.5\%$$363.835.9\%$$121.212.0\%$Anxiety$813.0116.414.3\%$$377.546.4\%$$204.925.2\%$$114.114.0\%$Mood disorders$1874.7115.26.1\%$$487.826.0\%$1,$157.261.7\%$$114.66.1\%$Psychosis and personality disorders2,$359.8102.44.3\%$$549.523.3\%$1,$591.067.4\%$$116.84.9\%$Attention deficit hyperactivity disorder$619.563.810.3\%$$255.141.2\%$$183.329.6\%$$117.318.9\%$Conduct disorders$778.364.48.3\%$$305.739.3\%$$294.137.8\%$$114.214.7\%$Eating disorders1,$070.887.08.1\%$$424.639.7\%$$443.341.4\%$$115.910.8\%$Self-harm4,$543.7156.43.4\%$$931.420.5\%$3,$338.873.5\%$$117.12.6\%$2 or more diagnoses1,$335.2121.19.1\%$$465.734.9\%$$633.847.5\%$$114.58.6\%$ For all the models developed, the balance achieved by entropy balancing (Additional file 1: Tables S6–15), the two-part models with their parameters (Additional file 1: Tables S16–25) and the results on mean and excess cost of the combined statistical analysis are provided in the supplementary material (Additional file 1: Tables S26–35). To summarise our results here, we present the mean and excess costs per patient by diagnostic cluster and disaggregated by SES in Table 4, sex in Table 5 and age-group in Table 6. For all clusters, annual excess costs in the groups of patients with mental disorders were more than double those in the groups with no diagnosed mental disorders. Tables 5, 6 also show the differences in adjusted costs by sex and age group. For all clusters, annual excess costs were higher in women than in men. Disaggregation by age group did not render a fully consistent pattern, but in general, younger age groups incurred lower excess costs. Table 4Mean and excess cost per patient in € [2018] of direct healthcare costs disaggregated by socioeconomic status and diagnostic groupMean cost (€)aExcess cost (€)p-valueaWith no mental disorders ($$n = 512$$,710)With mental disorder(s)Any mental disorder($$n = 96$$,671)High242610368 < 0.001Medium270684414 < 0.001Low4041007603 < 0.001Substance use($$n = 19$$,507)High215762547 < 0.001Medium258924666 < 0.001Low55619521395 < 0.001Anxiety($$n = 40$$,523)High251696445 < 0.001Medium280785505 < 0.001Low4481235787 < 0.001Mood disorders($$n = 8613$$)High25115571306 < 0.001Medium27317001428 < 0.001Low46628552389 < 0.001Psychosis and personality disorders($$n = 5$.745$)High23718751638 < 0.001Medium26220971835 < 0.001Low43434112977 < 0.001Attention deficit hyperactivity disorder($$n = 16$$,986)High228568339 < 0.001Medium249617368 < 0.001Low328803475 < 0.001Conduct disorder($$n = 26$$,415)High246687441 < 0.001Medium265744479 < 0.001Low3781047668 < 0.001Eating disorder($$n = 4629$$)High260894634 < 0.001Medium2951018723 < 0.001Low48616421156 < 0.001Self-harm($$n = 664$$)High33946924353 < 0.001Medium25834863228 < 0.001Low51468166302 < 0.0012 or more diagnoses($$n = 19$$,567)High2451142897 < 0.001Medium2621232970 < 0.001Low44520551610 < 0.001aCalculated using two-part models and groups were adjusted by age group, sex and SES using entropy balancingTable 5Mean and excess cost per patient in € [2018] of direct healthcare costs disaggregated by sex and diagnostic groupMean cost (€)aExcess cost (€)p-valueaWith no mental disorders ($$n = 512$$,710)With mental disorder(s)Any mental disorder($$n = 96$$,671)Female310770460 < 0.001Male246633387 < 0.001Substance use($$n = 19$$,507)Female3231118795 < 0.001Male247904657 < 0.001Anxiety($$n = 40$$,523)Female321880560 < 0.001Male245704459 < 0.001Mood disorders($$n = 8613$$)Female31719231606 < 0.001Male27717631485 < 0.001Psychosis and personality disorders($$n = 5$.745$)Female33425842250 < 0.001Male26821671899 < 0.001Attention deficit hyperactivity disorder($$n = 16$$,986)Female277677400 < 0.001Male240598358 < 0.001Conduct disorder($$n = 26$$,415)Female292808515 < 0.001Male263741478 < 0.001Eating disorder($$n = 4629$$)Female3261111785 < 0.001Male245859614 < 0.001Self-harm($$n = 664$$)Female35347274374 < 0.001Male26536363371 < 0.0012 or more diagnoses($$n = 19$$,567)Female31714461129 < 0.001Male2551226971 < 0.001aCalculated using two-part models and groups were adjusted by age group, sex and SES using entropy balancingTable 6Excess cost per patient in € [2018] of direct healthcare costs disaggregated by age group and diagnostic groupMean cost (€)Excess cost (€)p-valueaWithout mental disorder ($$n = 512$$,710)With mental disorderAny mental disorder($$n = 96$$,671)1–12287702415 < 0.00113–18269675406 < 0.00119–24255648393 < 0.00125–30293751457 < 0.001Substance use($$n = 19$$,507)1–123641.222858 < 0.00113–183261.130804 < 0.00119–24270955685 < 0.00125–302801.004724 < 0.001Anxiety($$n = 40$$,523)1–12276745469 < 0.00113–18306848543 < 0.00119–24269747478 < 0.00125–30301847546 < 0.001Mood disorders($$n = 8613$$)1–122511.4921.240 < 0.00113–183322.0361.705 < 0.00119–242771.7261.450 < 0.00125–303081.9401.632 < 0.001Psychosis and personality disorders($$n = 5$.745$)1–122471.8691.623 < 0.00113–182962.3242.028 < 0.00119–242892.3062.017 < 0.00125–303062.4722.166 < 0.001Attention deficit hyperactivity disorder($$n = 16$$,986)1–12281671390 < 0.00113–18253623370 < 0.00119–24227571345 < 0.00125–30257659403 < 0.001Conduct disorder($$n = 26$$,415)1–12262716454 < 0.00113–18270755485 < 0.00119–24267757490 < 0.00125–30332947616 < 0.001Eating disorder($$n = 4629$$)1–12272918646 < 0.00113–183311.137806 < 0.00119–24267920653 < 0.00125–303371.164827 < 0.001Self-harm($$n = 664$$)1–121632.0581.895 < 0.00113–183785.0404.661 < 0.00119–242553.4293.175 < 0.00125–303665.0274.661 < 0.0012 or more diagnoses($$n = 19$$,567)1–122811.253972 < 0.00113–183031.3971.094 < 0.00119–242621.228966 < 0.00125–302931.3881.096 < 0.001aCalculated using two-part models and groups were adjusted by age group, sex and SES using entropy balancing ## Discussion To our knowledge, this is the first study showing individual excess costs of persons with mental diagnoses and adjusted for SES and sex covering a general population of 609,381 individuals younger than 30 years old. Children, adolescents and young adults diagnosed with mental disorders used health services more and this implied a high excess cost, the annual cost per diagnosed person being, overall, more than twice the cost per person with no diagnoses. A socioeconomic gradient was notable, excess costs being higher in individuals with low SES than those with high-to-medium SES. The low SES category ($7.8\%$) grouped the adolescents and young people in households with no income with those whose health cardholder was on benefits and exempt from payment or retired regardless of their income (i.e., with an income lower or higher than €18,000). The rationale for this can be seen in Additional file 1: Table S3 which shows that adolescents and young people depending on a retired cardholder had a higher prevalence of mental disorders ($24.2\%$ and $20.5\%$) and consistent with SES relying not only on income but also on family structure. Differences by sex were also statistically significant but their sizes were smaller than those related to SES. Our cluster-disaggregated prevalence results for 18-year-olds are consistent with those described in Denmark in a population registry base study [38]. The healthcare costs were comprehensive as they included hospital care, primary care and pharmacy. Roughly three quarters of the costs per patient were hospital-related costs, which included those for emergency services and specialized outpatient clinics as well as hospital ward admissions. In another registry-based study, Christensen et al. estimated the total healthcare cost of all persons living in Denmark with a diagnosis of mental disorder [39]. When comparing with their mean annual healthcare costs, as would be expected for a country with lower salaries, our annual costs were in a lower range, but the ratio between the annual healthcare costs in diagnosed and non-diagnosed individuals was roughly three in both studies. When analysing the results on annual excess healthcare costs, they also found that schizophrenia and drug use disorders incurred the highest ones. However, the different age range of the two populations hampered the comparison with our results as we limited our study to individuals from 1 to 30 years and the somatic burden is much higher in older cohorts [40]. On the other hand, the mean total costs were within the range of the real per capita health spending by age group in Spain estimated by top-down methods and the estimated annual costs were also quite similar in both studies [41]. The excess costs were important in all three cost components, differences in hospital costs being greater in absolute terms, but the relative difference in pharmacy was also considerable. Drug prescription costs were 3.4 times higher in the group with mental diagnoses, revealing the use of psychoactive drugs in all age groups under 30 years. Two diagnosis clusters generated the highest costs per individual, self-harm with costs of €4543.7 and psychosis and personality disorders with costs of €2359.8. After statistical adjustment using the two-part models, they continued to be the clusters with the highest average and highest excess costs. When disaggregating by SES, the social gradient is reflected in the statistical models and the top figure of €6,302 was obtained for the self-harm cluster in the low SES group. The differences by SES are striking in all the clusters and especially between, on the one hand, low SES, and, on the other, medium and high SES categories (Table 2 and Additional file 1: Table S4) [42]. At this point it is important to remark that, as long as universal coverage is provided, any citizen has guaranteed access to health services. Nevertheless, it is possible that differences found may undervalue the whole reality when looking to the literature. Findings that children from low SES families respond more strongly to cost sharing policies such as co-payments [43], acting as a barrier when seeking healthcare assistance, suggest that there can be an underestimation in this group. Moreover, social and cultural factors like stigma and negative perceptions surrounding mental illness can also influence the use of healthcare system, especially conditioning the access of the most vulnerable groups [44]. Therefore, actual differences between SES groups could even increase. When looking over the effect of sex, the total spending by females is greater than by males in coherence with the literature [45, 46]. In the same way, the analyses revealed that, in terms of excess cost, women’s also had higher numbers in each diagnostic cluster. The higher total and excess costs found in females can be explained because women tend to use more the healthcare services in general [45, 46]. There are differences in cluster prevalence by sex, but they did not bias the excess cost calculation thanks to the adjustment achieved with the two-part models. It is noteworthy that individuals with mental disorders incurred higher costs not only for mental healthcare but also for somatic healthcare. A similar pattern of use has been found elsewhere among under 18-year-olds diagnosed with a mental disorder [47]. In adults, the higher resource use has been partially attributed to chronic comorbidities [48], but specific explanations are required for young people with very few chronic physical conditions. As suggested by the literature, a possible justification can be that the presence of a mental disorder was associated with an increased risk of subsequent medical conditions [49]. Different studies also indicate that parental coping with a mental illness is related to the mental health of their children [50–52], as well as with the increase in their healthcare services use [53, 54]. However, it must be taken into account that people with mental health disorders are a heterogeneous group with different health and social needs, where the drivers of their higher resource use are likely to be multifactorial [48]. It must also be considered that healthcare use and cost estimates in adolescents and young adults may be underestimated, as long as practitioners can be reluctant to diagnose certain disorders, especially more severe ones, until the patient reaches an older age [55, 56]. Therefore, initiatives should be developed to improve early recognition and mental health support for young people, seeking both to improve their care and potentially reduce inappropriate care and costs [57]. The availability of data on the excess costs of mental disorders opens an opportunity for undertaking studies on the effectiveness and cost-effectiveness of preventive interventions in adolescents [3]. In particular, reducing the incidence of self-harm, psychoses and personality disorders and mood disorders should be considered a public health priority, because these disorders are associated with disability, and also have serious economic consequences. The relevance of these findings is underlined by the effect of the coronavirus disease 2019 pandemic on the mental health of adolescents and has major implications for prevention planning [58, 59]. Preventive interventions for self-harm and suicide must be included in the guidelines to safeguard the mental health of adolescents and young adults affected by the pandemic and the measures restricting social mobility, with a focus on measures to mitigate anxiety, depression, and stigma, among other conditions. ## Limitations and strengths Our study was carried out from the perspective of the health system and therefore our data lack the weight of other categories such as social, judicial and educational costs. We acknowledge that a fully comprehensive approach to estimating the burden of mental disorders must incorporate a societal perspective by covering all cost categories assessed in top-down cost-of-illness studies such as data on crime, accidents and social care [17, 60]. Moreover, informal costs due to caregivers’ time should be accepted as part of the economic burden of mental disorders but so far these key components are not recorded in registries [61]. Our figures for excess costs would have been even higher if those cost categories had been measured [18]. Further, while the economic impact of informal care is important, so is the suffering and loss of quality of life of siblings who endure the care of children, adolescents and young adults with mental disorders [61]. Wittenberg et al. described this situation highlighting “health as a family affair” [61]. Another limitation of the study was the lack of validation of the diagnoses. As in other observational studies, the cohort effect may bias the results [38]. Our dataset is based on the integration of information on all the diagnoses of individuals recorded in the electronic health record of the public health service in their contacts with primary, hospital inpatient, emergency and outpatient care. This approach yielded consistent results in the diagnosis of dementia in various European countries [62]. In the Basque Country, nearly universal health coverage is provided, but in the age range studied, $20\%$ of the population also have private insurance. This cost component is absent in our database and therefore its size was not considered. The percentage of high-income individuals with double coverage (public and private) is greater in high SES people, and they may opt to use private rather than public providers, and hence, the differences by SES may be biased [63]. The lack of adjustment for comorbidities was also a potential limitation. Nonetheless, in these early stages of life, social determinants have a greater impact on health than physical chronic conditions [29]. Finally, another limitation of the study was the definition of the different clusters of mental disorders. Our approach to classifying mental disorders roughly followed the categories defined by Dalsgaard et al. for the same purpose also using ICD-10 codes and a population registry [38]. In contrast, self-reported symptoms in surveys are converted into codes from the successive versions of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV), to estimate prevalence indicators [32]. Besides its limitations, our study has important strengths. As the data were derived from a population-based registry covering the whole population, it provides a recent and comprehensive estimate of the direct medical costs for a population of more than half a million individuals under 30 years of age. Furthermore, the registry contained the contacts with healthcare providers in all the public settings in the Basque Country. A similar approach has been successfully applied in Denmark to estimate incidence rates of mental disorders [38, 64]. The joint use of Basque administrative and clinical databases allowed us to obtain population-based cost estimates across the entire healthcare system avoiding the selection bias associated with small samples from psychiatric settings. Another strength is the availability of all contacts to measure the resource use at the patient level directly, instead of relying on patient-reported healthcare use or top-down approaches [7, 16, 17, 65]. The current availability of information from electronic health records enables the undertaking of observational studies based on RWD that allow the measurement of actual resource use and costs. Nevertheless, the high external validity of these types of study may be weakened given their non-random design, where the baseline characteristics of the groups to be compared can differ due to selection bias. To overcome this issue, pre-processing techniques like entropy balancing or propensity score matching are crucial to adjust the covariate distribution of the control group by the reweighting or discarding of units [33, 66, 67]. Such techniques make the distribution more similar to the one in the comparison group. In this case, entropy balancing was used to carry out this task. In contrast to other pre-processing methods, this technique tackles the adjustment problem backwards and estimates the set of weights that satisfies the balance constraints that involve the first, second and higher moments of the covariate distributions as well as interactions. Because of that, a high degree of covariate balance can be obtained. Moreover, since entropy balancing weights show smooth variation across units, its appeal lies in its capacity to optimize the balance in the covariate distribution while retaining the maximum amount of information. Finally, compared to other techniques like propensity score matching, it can be faster computationally speaking, it being possible to obtain the weights within a few seconds even in large databases. ## Conclusions This study provides estimates of the excess economic costs of mental disorders for the first time in the Spanish population between 1 and 30 years of age based on a general population registry. Results are consistent in showing that young people with mental disorders place a greater burden on healthcare providers compared to population without mental disorders, and that the costs are especially high for severe mental disorders like self-harm and psychoses. Additionally, the results on excess healthcare costs obtained may facilitate future economic evaluations of interventions targeting adolescents and young adults, supporting decision-making in order to improve the provision of mental healthcare services. ## Supplementary Information Additional file 1: Healthcare Costs of Mental Disorders in Children, Adolescents and Young Adults in the Basque Population Registry Adjusted for Socioeconomic Status and Sex. Additional file 1 contains information about the ICD and ATC codes used to define each mental disorder category, the unit costs used, the distribution of prevalence of mental disorders by co-payment categories, the cost per patient of direct healthcare costs disaggregated by age and diagnostic group, the covariate balance achieved by entropy balancing, the two-part model parameters for each mental disorder category, and the mean and excess cost per patient of direct healthcare costs for each mental disorder category disaggregated by sex, age group and SES. ## References 1. Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease study 2019. Lancet (2020) 396 1204-22. DOI: 10.1016/S0140-6736(20)30925-9 2. Whiteford HA, Ferrari AJ, Degenhardt L, Feigin V, Vos T. The global burden of mental, neurological and substance use disorders: an analysis from the Global Burden of Disease study 2010. PLoS ONE (2015) 10 e0116820. DOI: 10.1371/journal.pone.0116820 3. Las Hayas C, Izco-Basurko I, Fullaondo A, Gabrielli S, Zwiefka A, Hjemdal O. UPRIGHT, a resilience-based intervention to promote mental well-being in schools: study rationale and methodology for a European randomized controlled trial. BMC Public Health (2019) 19 1413. DOI: 10.1186/s12889-019-7759-0 4. Neumann PJ, Ganiats TG, Russell LB, Sanders GD, Siegel JE. Cost-effectiveness in health and medicine (2016) 5. Knapp M. Economic evaluations and interventions for children and adolescents with mental health problems. J Child Psychol Psychiatry (1997) 38 3-25. DOI: 10.1111/j.1469-7610.1997.tb01503.x 6. Knapp M, Wong G. Economics and mental health: the current scenario. World Psychiatry (2020) 19 3-14. DOI: 10.1002/wps.20692 7. Shearer J, Byford S. The basics of economic evaluation in mental healthcare. BJPsych Adv (2015) 21 345-353. DOI: 10.1192/apt.bp.114.013003 8. Smit F, Cuijpers P, Oostenbrink J, Batelaan N, de Graaf R, Beekman A. Costs of nine common mental disorders: implications for curative and preventive psychiatry. J Ment Health Policy Econ (2006) 9 193-200. PMID: 17200596 9. Le LK-D, Esturas AC, Mihalopoulos C, Chiotelis O, Bucholc J, Chatterton ML. Cost-effectiveness evidence of mental health prevention and promotion interventions: a systematic review of economic evaluations. PLOS Med (2021) 18 e1003606. DOI: 10.1371/journal.pmed.1003606 10. Mihalopoulos C, Chatterton ML. Economic evaluations of interventions designed to prevent mental disorders: a systematic review. Early Interv Psychiatry (2015) 9 85-92. DOI: 10.1111/eip.12156 11. Arrospide A, Rue M, van Ravesteyn NT, Comas M, Soto-Gordoa M, Sarriugarte G. Economic evaluation of the breast cancer screening programme in the Basque Country: retrospective cost-effectiveness and budget impact analysis. BMC Cancer (2016) 16 344. DOI: 10.1186/s12885-016-2386-y 12. Soto-Gordoa M, Arrospide A, Moreno-Izco F, Martínez-Lage P, Castilla I, Mar J. Projecting burden of dementia in Spain, 2010–2050: impact of modifying risk factors. J Alzheimers Dis (2015) 48 721-730. DOI: 10.3233/JAD-150233 13. Arrospide A, Ibarrondo O, Castilla I, Larrañaga I, Mar J. Development and validation of a discrete event simulation model to evaluate the cardiovascular impact of population policies for obesity. Med Decis Mak: Int J Soc Med Decis Mak (2021). DOI: 10.1177/0272989X211032964 14. Caldwell DM, Davies SR, Thorn JC, Palmer JC, Caro P, Hetrick SE. School-based interventions to prevent anxiety, depression and conduct disorder in children and young people: a systematic review and network meta-analysis. Public Health Res (2021) 9 1-284. DOI: 10.3310/phr09080 15. Drummond MF, Sculpher MJ, Torrance GW, O’Brien BJ, Stoddart GL. Methods for the economic evaluation of health care programmes (2005) 16. Weatherly H, Drummond M, Claxton K, Cookson R, Ferguson B, Godfrey C. Methods for assessing the cost-effectiveness of public health interventions: key challenges and recommendations. Health Policy (2009) 93 85-92. DOI: 10.1016/j.healthpol.2009.07.012 17. Łaszewska A, Wancata J, Jahn R, Simon J. The excess economic burden of mental disorders: findings from a cross-sectional prevalence survey in Austria. Eur J Health Econ (2020) 21 1075-1089. DOI: 10.1007/s10198-020-01200-0 18. Dams J, Buchholz A, Kraus L, Reimer J, Scherbaum N, Konnopka A. Excess costs of alcohol-dependent patients in German psychiatric care compared with matched non-alcohol-dependent individuals from the general population: a secondary analysis of two datasets. BMJ Open (2018) 8 e020563. DOI: 10.1136/bmjopen-2017-020563 19. Parés-Badell O, Barbaglia G, Jerinic P, Gustavsson A, Salvador-Carulla L, Alonso J. Cost of disorders of the brain in Spain. PLoS ONE (2014) 9 e105471. DOI: 10.1371/journal.pone.0105471 20. Garrison LP, Neumann PJ, Erickson P, Marshall D, Mullins CD. Using real-world data for coverage and payment decisions: the ISPOR real-world data task force report. Value Health (2007) 10 326-335. DOI: 10.1111/j.1524-4733.2007.00186.x 21. Ibarrondo O, Lizeaga G, Martínez-Llorente JM, Larrañaga I, Soto-Gordoa M, Álvarez-López I. Health care costs of breast, prostate, colorectal and lung cancer care by clinical stage and cost component. Gac Sanit (2022) 36 246-252. DOI: 10.1016/j.gaceta.2020.12.035 22. Soto-Gordoa M, Arrospide A, Millán E, Calvo M, Igartua JI, Esnaola S. Gender and socioeconomic inequalities in the implementation of the Basque programme for multimorbid patients. Eur J Public Health (2019) 29 681-686. DOI: 10.1093/eurpub/ckz071 23. Mar J, Arrospide A, Soto-Gordoa M, Iruin Á, Tainta M, Gabilondo A. Dementia-related neuropsychiatric symptoms: inequalities in pharmacological treatment and institutionalization. Neuropsychiatr Dis Treat (2019) 15 2027-2034. DOI: 10.2147/NDT.S209008 24. Braveman P. What are health disparities and health equity? We need to be clear. Public Health Rep (2014) 129 5-8. DOI: 10.1177/00333549141291S203 25. Koopmans GT, Donker MCH, Rutten FHH. Common mental disorders and use of general health services: a review of the literature on population-based studies. Acta Psychiatr Scand (2005) 111 341-350. DOI: 10.1111/j.1600-0447.2005.00496.x 26. Tusa N, Koponen H, Kautiainen H, Korniloff K, Raatikainen I, Elfving P. The profiles of health care utilization among a non-depressed population and patients with depressive symptoms with and without clinical depression. Scand J Prim Health Care (2019) 37 312-318. DOI: 10.1080/02813432.2019.1639904 27. Oliver-Parra A, Dalmau-Bueno A, Ruiz-Muñoz D, García-Altés A. Relationship between parents’ mental disorders and socioeconomic status and offspring’s psychopathology: a cross-sectional study. PLoS ONE (2020) 15 e0240681. DOI: 10.1371/journal.pone.0240681 28. Lund C, Brooke-Sumner C, Baingana F, Baron EC, Breuer E, Chandra P. Social determinants of mental disorders and the sustainable development goals: a systematic review of reviews. Lancet Psychiatry (2018) 5 357-369. DOI: 10.1016/S2215-0366(18)30060-9 29. Reiss F, Meyrose A-K, Otto C, Lampert T, Klasen F, Ravens-Sieberer U. Socioeconomic status, stressful life situations and mental health problems in children and adolescents: results of the German BELLA cohort-study. PLoS ONE (2019) 14 e0213700. DOI: 10.1371/journal.pone.0213700 30. Appleyard K, Egeland B, van Dulmen MHM, Sroufe LA. When more is not better: the role of cumulative risk in child behavior outcomes. J Child Psychol Psychiatry (2005) 46 235-245. DOI: 10.1111/j.1469-7610.2004.00351.x 31. Marmot M. Social determinants of health inequalities. Lancet (2005) 365 1099-1104. DOI: 10.1016/S0140-6736(05)71146-6 32. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry (2005) 62 593-602. DOI: 10.1001/archpsyc.62.6.593 33. Hainmueller J. Entropy balancing for causal effects: a multivariate reweighting method to produce balanced samples in observational studies. Polit Anal (2012) 20 25-46. DOI: 10.1093/pan/mpr025 34. Glick HA, Doshi JA, Sonnad SS, Polsky D. Economic evaluation in clinical trials (2014) 35. Dams J, König H-H, Bleibler F, Hoyer J, Wiltink J, Beutel ME. Excess costs of social anxiety disorder in Germany. J Affect Disord (2017) 213 23-29. DOI: 10.1016/j.jad.2017.01.041 36. Hainmueller J, Xu Y. ebalance: a Stata package for entropy balancing. J Stat Softw (2013) 54 1-18. DOI: 10.18637/jss.v054.i07 37. Belloti F. twopm: two-part models. Stata J (2015) 15 3-20. DOI: 10.1177/1536867X1501500102 38. Dalsgaard S, Thorsteinsson E, Trabjerg BB, Schullehner J, Plana-Ripoll O, Brikell I. Incidence rates and cumulative incidences of the full spectrum of diagnosed mental disorders in childhood and adolescence. JAMA Psychiat (2020) 77 155-164. DOI: 10.1001/jamapsychiatry.2019.3523 39. Christensen MK, McGrath JJ, Momen NC, Whiteford HA, Weye N, Agerbo E. The cost of mental disorders in Denmark: a register-based study. Npj Mental Health Res (2022) 1 1. DOI: 10.1038/s44184-022-00001-y 40. Weye N, Momen NC, Whiteford HA, Christensen MK, Iburg KM, Santomauro DF. The contribution of general medical conditions to the non-fatal burden of mental disorders: register-based cohort study in Denmark. BJPsych Open (2022) 8 e180. DOI: 10.1192/bjo.2022.583 41. Blanco Moreno Á, Urbanos Garrido R, Thuissard Vasallo IJ. Real per capita health spending by age and sex in Spain (1998–2008): changes and effects on public healthcare expenditure projections. Gac Sanit (2013) 27 220-225. DOI: 10.1016/j.gaceta.2012.09.004 42. Mar J, Larrañaga I, Ibarrondo O, González-Pinto A, las Hayas C, Fullaondo A. Socioeconomic and gender inequalities in mental disorders among adolescents and young adults. J Psychiatry Ment Health (2022). DOI: 10.1016/j.rpsm.2022.07.001 43. Nilsson A, Paul A. Patient cost-sharing, socioeconomic status, and children’s health care utilization 2018. J Health Econ (2018) 59 109-124. DOI: 10.1016/j.jhealeco.2018.03.006 44. Gulliver A, Griffiths K, Christensen H. Perceived barriers and facilitators to mental health help-seeking in young people: a systematic review. BMC Psychiatry (2010) 10 113. DOI: 10.1186/1471-244X-10-113 45. Owens GM. Gender differences in health care expenditures, resource utilization, and quality of care. J Manag Care Pharm (2008) 14 2-6. PMID: 18439060 46. Cylus J, Hartman M, Washington B, Andrews K, Catlin A. Pronounced gender and age differences are evident in personal health care spending per person. Health Aff (2011) 30 153-160. DOI: 10.1377/hlthaff.2010.0216 47. Jacobson AM, Goldberg ID, Burns BJ, Hoeper EW, Hankin JR, Hewitt K. Diagnosed mental disorder in children and use of health services in four organized health care settings. Am J Psychiatry (1980) 137 559-565. DOI: 10.1176/ajp.137.5.559 48. Sporinova B, Manns B, Tonelli M, Hemmelgarn B, MacMaster F, Mitchell N. Association of mental health disorders with health care utilization and costs among adults with chronic disease. JAMA Netw Open (2019) 2 e199910. DOI: 10.1001/jamanetworkopen.2019.9910 49. Momen NC, Plana-Ripoll O, Agerbo E, Benros ME, Børglum AD, Christensen MK. Association between mental disorders and subsequent medical conditions. N Engl J Med (2020) 382 1721-1731. DOI: 10.1056/NEJMoa1915784 50. Pierce M, Hope HF, Kolade A, Gellatly J, Osam CS, Perchard R. Effects of parental mental illness on children’s physical health: systematic review and meta-analysis. Br J Psychiatry (2020) 217 354-363. DOI: 10.1192/bjp.2019.216 51. Mattejat F, Remschmidt H. The children of mentally ill parents. Dtsch Arztebl Int (2008) 105 413-418. PMID: 19626164 52. Sell M, Radicke A, Adema B, Daubmann A, Kilian R, Stiawa M. Parents with mental illness: parental coping behavior and its association with children’s mental health. Front Psychiatry (2021) 12 737861. DOI: 10.3389/fpsyt.2021.737861 53. Heuckendorff S, Johansen MN, Johnsen SP, Overgaard C, Fonager K. Parental mental health conditions and use of healthcare services in children the first year of life- a register-based, nationwide study. BMC Public Health (2021) 21 557. DOI: 10.1186/s12889-021-10625-y 54. Waldmann T, Stiawa M, Dinc Ü, Saglam G, Busmann M, Daubmann A. Costs of health and social services use in children of parents with mental illness. Child Adolesc Psychiatry Ment Health (2021) 15 10. DOI: 10.1186/s13034-021-00360-y 55. Baumann AE. Stigmatization, social distance and exclusion because of mental illness: the individual with mental illness as a “stranger”. Int Rev Psychiatry (2007) 19 131-135. DOI: 10.1080/09540260701278739 56. Thornicroft G, Rose D, Kassam A. Discrimination in health care against people with mental illness. Int Rev Psychiatry (2007) 19 113-122. DOI: 10.1080/09540260701278937 57. Saunders NR, Gandhi S, Chen S, Vigod S, Fung K, De Souza C. Health care use and costs of children, adolescents, and young adults with somatic symptom and related disorders. JAMA Netw Open (2020) 3 e2011295. DOI: 10.1001/jamanetworkopen.2020.11295 58. Ougrin D. Debate: emergency mental health presentations of young people during the COVID-19 lockdown. Child Adolesc Ment Health (2020) 25 171-172. DOI: 10.1111/camh.12411 59. Kumar A, Nayar KR. COVID 19 and its mental health consequences. J Ment Health (2021) 30 1-2. DOI: 10.1080/09638237.2020.1757052 60. Scott S, Knapp M, Henderson J, Maughan B. Financial cost of social exclusion: follow up study of antisocial children into adulthood. BMJ (2001) 323 191. DOI: 10.1136/bmj.323.7306.191 61. Wittenberg E, Prosser LA. Health as a family affair. N Engl J Med (2016) 374 1804-1806. DOI: 10.1056/NEJMp1604456 62. Mar J, Gorostiza A, Arrospide A, Larrañaga I, Alberdi A, Cernuda C. Estimation of the epidemiology of dementia and associated neuropsychiatric symptoms by applying machine learning to real-world data. Rev Psiquiatr Salud Ment (2022) 15 167-175. DOI: 10.1016/j.rpsm.2021.03.001 63. Cantarero-Prieto D, Pascual-Sáez M, Gonzalez-Prieto N. Effect of having private health insurance on the use of health care services: the case of Spain. BMC Health Serv Res (2017) 17 716. DOI: 10.1186/s12913-017-2667-4 64. Steinhausen H-C, Jakobsen H. Incidence rates of treated mental disorders in childhood and adolescence in a complete nationwide birth cohort. J Clin Psychiatry (2019) 80 17m12012. DOI: 10.4088/JCP.17m12012 65. Oliva-Moreno J, López-Bastida J, Montejo-González AL, Osuna-Guerrero R, Duque-González B. The socioeconomic costs of mental illness in Spain. Eur J Health Econ (2009) 10 361-369. DOI: 10.1007/s10198-008-0135-0 66. Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivar Behav Res (2011) 46 399-424. DOI: 10.1080/00273171.2011.568786 67. Diamond A, Sekhon JS. Genetic matching for estimating causal effects: a general multivariate matching method for achieving balance in observational studies. Rev Econ Stat (2012) 95 932-945. DOI: 10.1162/REST_a_00318
--- title: 'Body composition and physical fitness in adults born small for gestational age at term: a prospective cohort study' authors: - Maria Matre - Cathrin Vano Mehl - Silje Dahl Benum - Laura Jussinniemi - Eero Kajantie - Kari Anne I. Evensen journal: Scientific Reports year: 2023 pmcid: PMC9975870 doi: 10.1038/s41598-023-30371-y license: CC BY 4.0 --- # Body composition and physical fitness in adults born small for gestational age at term: a prospective cohort study ## Abstract There is lack of research on body composition and physical fitness in individuals born small for gestational age (SGA) at term entering mid-adulthood. We aimed to investigate these outcomes in adults born SGA at term. This population-based cohort study included 46 adults born SGA with birth weight < 10th percentile at term (gestational age ≥ 37 weeks) (22 women, 24 men) and 61 adults born at term with birth weight ≥ 10th percentile (35 women, 26 men) at 32 years. Body composition was examined anthropometrically and by 8-polar bioelectrical impedance analysis (Seca® mBCA 515). Fitness was measured by maximal isometric grip strength by a Jamar hand dynamometer, 40-s modified push-up test and 4-min submaximal step test. Participants born SGA were shorter than controls, but other anthropometric measures did not differ between the groups. Men born SGA had 4.8 kg lower grip strength in both dominant ($95\%$ CI 0.6 to 9.0) and non-dominant ($95\%$ CI 0.4 to 9.2) hand compared with controls. Grip strength differences were partly mediated by height. In conclusion, body composition and physical fitness were similar in adults born SGA and non-SGA at term. Our finding of reduced grip strength in men born SGA may warrant further investigation. ## Introduction Individuals born small for gestational age (SGA) are often defined as having a birth weight below the 10th percentile for their gestational age1. SGA is the most frequently used indicator of intrauterine growth restriction, which is a state where the foetus does not reach its genetic growth potential2. Males may be more susceptible to growth restriction in utero than females3. Being born SGA involves an extra vulnerability for later diseases, such as metabolic syndrome and cardiovascular diseases4,5. The prevalence of SGA depends on the birth weight standards used. When standards that are based on actual birth weights in the given population are used, $10\%$ of term-born infants are by definition born SGA. In low- and middle-income countries the prevalence is around $20\%$ according to Intergrowth standards6. Hence, the high prevalence represents a major concern for public health. Most infants born SGA show spontaneous catch-up growth by two years of age, however approximately $10\%$ do not, and persistent short stature is therefore one of the most common complications after being born SGA4. During childhood, studies have reported that those born SGA remain shorter and thinner with lower body mass index (BMI) than controls7–9, while there are reports of both less body fat7 and higher central adiposity8. In adulthood, studies have found individuals born SGA to be shorter10–12 and lighter11,12 than controls, but their BMI did not differ at 1811 or 22 years of age12. At 26 years of age, we did not find group differences in body composition, but women born SGA displayed lower lean mass than controls13. Another study indicated progression of adiposity from 22 to 30 years, as adults born SGA had more body fat, and their waist circumference increased, but there was no interaction effect with sex12. Also, low birth weight has been linked to reduced muscle mass and reduced muscle strength in childhood and adulthood14. A recent meta-analysis found a positive association between SGA and overweight and/or obesity15, whereas there were inconsistent findings of this association by sex15. Muscular and cardiorespiratory fitness are two important health-related components of physical fitness16, that both have been reported to be associated with all-cause mortality, as well as non-communicable diseases17,18. Grip strength, a simple and widely used measure of muscular strength, has proven to be particularly relevant as it is a strong predictor of future physical function, morbidity, and mortality18–20. Several studies have found associations of birth weight with muscular and cardiorespiratory fitness in adulthood21–23. However, these studies have either included adults born preterm or not specified the gestational age of their participants, making it difficult to distinguish if the result is due to factors related to preterm birth or low birth weight. Three relatively small studies found no difference in cardiorespiratory fitness, measured by maximal oxygen uptake (VO2max), between term-born young adults with a birth weight at or below the 10th percentile and a control group11,24,25. On the other hand, two recent Swedish registry studies that included a large number of 18-year-old men born at term, found strong associations of birth weight with grip strength and cardiorespiratory fitness26,27. There is lack of research on body composition and physical fitness in individuals born SGA entering mid-adulthood, an age when the prevalence of many non-communicable diseases starts to increase28. The aim of this study was to examine whether body composition and physical fitness differed between adults born SGA and non-SGA at term. We hypothesised that adults born SGA at term would display a less favourable body composition and lower level of fitness than the term-born control group. ## Study design This study is a part of the NTNU Low Birth Weight in a Lifetime Perspective study. The present study included two groups of adults born in 1986–1988; one group born SGA at term, and one group born non-SGA at term with birth weight ≥ 10th percentile, which serves as a control group. The participants took part in a larger data collection at 32 years of age. In addition to physical fitness tests, examinations included anthropometric measurements, examination of lung function, visual function as well as fine and gross motor function. Assessments were carried out from September 2019 to October 2020. ## Participants Participants were initially included in a multicentre study investigating the aetiology and consequences of intrauterine growth restriction29,30. Pregnant women living in the Trondheim region were enrolled before week 20 of pregnancy based on referral from general practitioners and obstetricians. Women were eligible if they had a singleton pregnancy and had been pregnant one or two times before ($$n = 1249$$). A $10\%$ random sample of these women were selected to serve as a control group ($$n = 132$$), using a sealed envelope method. A group of women at high risk of giving birth to an SGA infant were selected for follow-up if they had one or more defined risk criteria for SGA birth; a previous low birth weight child, low pre-pregnancy weight (< 50 kg), previous perinatal death, presence of chronic maternal disease (chronic renal disease, essential hypertension, or heart disease), or maternal cigarette smoking at conception ($$n = 390$$). Women in the control group and the high-risk group were thoroughly followed during pregnancy and their infants were examined at birth. The rest of the women ($$n = 727$$) were not followed during pregnancy (Fig. 1).Figure 1Flow of participants. SGA small for gestational age. At birth, all SGA infants born to mothers in either group were included in the SGA group (Fig. 1). An infant was defined as being born SGA if the birth weight was < 10th percentile for gestational age (GA), corrected for sex and parity, according to a reference standard using data from the Norwegian Medical Birth Registry29. Non-SGA infants born to mothers in the random sample were included in the control group. They were born with a birth weight ≥ 10th percentile. GA was based on the first day of the mother’s last menstrual period if this was accurately recalled ± 3 days. Ultrasound based GA was used if the last menstrual period was not recalled, or if there was a discrepancy of more than 14 days. Both groups were born at term (GA ≥ 37 weeks)29,30. The total sample included 104 participants born SGA and 120 controls (Fig. 1). Three individuals born SGA and two controls were excluded due to death, congenital syndrome/anomaly, or multimorbidity. Of the eligible, 15 individuals born SGA and 14 controls were not invited because they were living abroad, had no contact information or had previously refused to participate. Thus, a total of 190 were invited to the present study, 86 in the SGA group and 104 in the control group. Of these, 30 individuals born SGA and 36 controls did not consent to participate. Furthermore, 10 individuals born SGA and seven controls were not assessed clinically. Thus, 46 participants born SGA and 61 controls were assessed clinically, corresponding to $56.3\%$ of the invited. ## Non-participants There were no significant differences between participants and those who did not consent or were not assessed clinically regarding sex, gestational age, birth weight, head circumference, body length, ponderal index, maternal age at child’s birth or parental socioeconomic status (SES) in either group (data not shown). From the 26-year follow-up data were available on height, weight, BMI, waist and hip circumference, skinfold thickness and body composition measured by dual-energy x-ray absorptiometry (DXA). In the SGA group there were no differences, but in the control group, participants weighed 8.9 ($95\%$ CI 1.0 to 16.8) kg less than those who did not consent or were not assessed clinically. ## Background characteristics At birth, the infants in both groups were weighed to the nearest 10 g on a standard scale, and crown-heel length was measured with both legs extended to the nearest mm30. Ponderal index (g/cm3) was calculated based on these measurements. Parental socioeconomic status (SES) was calculated when participants attended the 14-year follow-up, supplemented for two participants at the 19-year follow-up, according to Hollingshead’s Two Factor Index of Social Position31, based on the parents’ education and occupation. This gives a social class rating from 1 (lowest) to 5 (highest). Educational attainment at the 32-year follow-up was collected by self-report and classified according to the International Standard Classification of Education (ISCED) levels 1 through 8. These were recoded into three categories: Lower secondary education or lower (ISCED levels 1–2) as no more than 10th class level, intermediate education (ISCED levels 3–5) as 11th–14th class level, and lower tertiary education or higher (ISCED levels 6–8) as a bachelor’s degree or higher. ## Outcome measures Assessments were carried out at NTNU/St. Olavs Hospital in Trondheim, Norway. A brief medical interview was conducted prior to examination, including whether the participant was pregnant, had a musculoskeletal diagnosis or other conditions affecting physical functioning. If the participant had a condition that made them unable to perform a physical test or that could be worsened by testing, they did not perform that particular test. All examinations were carried out by experienced and specially trained examiners, blinded to birth weight group. Anthropometric measurements were performed by a nurse and physical fitness tests by two physiotherapists and a medical research student. The examinations were carried out in the same order for each participant. At follow-up, the participants’ height, waist and hip circumference were measured to the nearest mm. Waist circumference was measured at the mid-point between the lowest rib and the crista iliaca, and hip circumference at the maximal circumference over the buttocks. Weight was measured by bioelectric impedance analysis using a Seca medical Body Composition Analyzer (Seca® mBCA 515) with a 100 g accuracy. Body mass index (BMI, kg/m2) and waist-to-hip ratio (waist circumference/hip circumference) was calculated. Bioelectrical impedance analysis measures included percent body fat, fat mass, fat free mass, skeletal muscle mass, total body water and extracellular water using the Seca 115 analytics software (Seca GmbH, Hamburg, Germany). Muscular fitness was measured by the maximal isometric grip strength of the hands and forearm muscles. A Jamar (Smith and Nephew, Memphis, TN) hand dynamometer was used. The dynamometer has 5 handle positions; position 3 and 4 were used for women and men, respectively. The participants were seated during the test, with shoulder abducted, a 90° angle in the elbow and a neutral position in the wrist, without support of the forearm32. Measurement was repeated three times in both dominant and non-dominant hand with 30 s recovery in between each attempt. Grip strength was measured in kg force and the maximal grip strength of the three measurements for each hand was used in the analysis. One participant in the control group could not perform the grip strength test with the dominant hand due to a hand fracture. The 40-s modified push-up test measures the muscular strength and endurance capacity of the upper body33 and is modified to improve standardisation. The participants started laying prone on a mat with their hands close to the shoulders and feet hip-width apart with their toes on the mat33. Before every push-up they had to clasp hands behind their back before pushing themselves to a straight leg push-up. In the top position they had to touch either of their hands with the other hand before returning to the push-up position and returning to the down-position. The number of correctly performed push-ups in 40 s were registered. One participant in the control group could not perform the push-up test due to a hand fracture. The Åstrand-Ryhming step test is a 4-min submaximal step-test that measures cardiorespiratory fitness34. The participants stepped on and off the step for four minutes paced by a metronome set to 46 beats per minute (i.e., 23 times up on the step/min). The height of the step was adapted to sex: 33 cm for women and 40 cm for men. Heart rate was observed during the test using a heart rate monitor (Firstbeat Technologies Oy) and recorded after 4 min of stepping and after being seated for 2 min. Two participants born SGA were not able to complete the test and were excluded from the analysis. ## Statistical analysis The analyses were conducted in SPSS version 27 (IBM Statistics). A p-value of less than 0.05 was considered statistically significant. Background characteristics were examined using Student’s t-test for continuous data, Exact Mann–Whitney U test for ordinal data and Pearson’s Chi square test for dichotomous variables. Group differences in outcome measures were analysed using independent samples t-test. The assumption of normally distributed variables was checked by visual inspection of histogram, boxplot, and Q–Q-plots of standardised residuals. As physical fitness differs between women and men35,36, we performed separate analyses by sex. Differences in physical fitness between groups were adjusted for height as a potential mediating factor in a univariate general linear model, since height has been consistently correlated with both being born SGA4,12,37 and physical fitness in previous literature22. To investigate whether physical conditions affected the results, sensitivity analyses were performed by excluding participants who were pregnant, had a musculoskeletal diagnosis or other conditions affecting physical functioning, as reported by the participants in the brief medical interview. A priori power calculations suggested, based on previous follow-up numbers in the SGA ($$n = 64$$) and control group ($$n = 81$$)38, that we would have the power to detect differences of 0.48 SD units with an alpha-level of 0.05 and a power of $80\%$, and 0.67 SD units with an alpha-level of 0.01 and desired power of $90\%$. ## Ethics The study was approved by the Regional Committee for Medical and Health Research Ethics in Central Norway [23879]. Written informed consent was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations. The data was pseudonymised and stored securely on a remote server with a two-step identifier. All methods were non-invasive and entailed low risk for injury or adverse events. An appointed doctor was medically responsible during data collection. Participants in need of health services were referred as appropriate. ## Results Participants’ background characteristics are shown in Table 1. There were 22 ($47.8\%$) women in the SGA group and 35 ($57.4\%$) in the control group ($$p \leq 0.327$$). Educational attainment did not differ between the groups (Table 1).Table 1Background characteristics of adults born small for gestational age (SGA) and non-SGA (control) at term. SGA ($$n = 46$$)Control ($$n = 61$$)p-valueMean(SD)Mean(SD)Gestational age (weeks)39.6(1.2)39.8(1.2)0.351Birth weight (g)2918[216]3686[467] < 0.001Birth weight SD score-1.5(0.4)0.2(0.9) < 0.001Head circumference (cm)a33.9(1.1)35.4(1.2) < 0.001Length (cm)b48.6(2.0)51.1(1.9) < 0.001Ponderal index (g/cm3)b2.5(0.2)2.8(0.3) < 0.001Maternal age (years)28.3(3.5)30.7(4.4)0.002Parental SES, n (%)c SES class 11(2.6)1(2.0) SES class 28(20.5)7(13.7) SES class 37(17.9)13(25.5)0.518 SES class 414(35.9)14(27.5) SES class 59(23.1)16(31.4)Age at follow-up (years)32.5(0.6)32.6(0.5)0.690n(%)n(%)Women22(47.8)35(57.4)0.327Education at follow-up Lower secondary or lower (ISCED 1–2)1(2.2)0[0] Intermediate (ISCED 3–5)19(41.3)22(36.1)0.444 Lower tertiary or higher (ISCED 6–8)26(56.5)39(63.9)ISCED International Standard Classification of Education, SD standard deviation, SES socioeconomic status (1–5, where 5 is highest), SGA small for gestational age.aData missing for five participants born SGA and four controls.bData missing for five participants born SGA and three controls.cData missing for seven participants born SGA and 10 controls.p-values based on Student’s t-test for continuous data and Exact Mann–Whitney U test for ordinal data (i.e., SES and ISCED). At follow-up, mean height was significantly lower in the SGA group compared with the control group. The other anthropometric measurements and bioelectrical impedance analysis measures did not differ between the groups (Table 2).Table 2Anthropometric measures and bioelectrical impedance analysis of adults born small for gestational age (SGA) and non-SGA (control) at term. SGA ($$n = 46$$)Control ($$n = 61$$)Mean difference($95\%$ CI)Mean(SD)Mean(SD)Height (cm)170.8(9.4)174.6(9.8)− 3.8(− 7.5 to − 0.1)Weight (kg)74.3(16.7)76.5(16.0)− 2.1(− 8.5 to 4.2)Body mass index (kg/m2)25.4(4.9)25.0(4.5)0.4(− 1.4 to 2.2)Waist circumference (cm)85.8(13.7)84.7(11.4)1.1(− 3.7 to 5.9)Hip circumference (cm)100.7(7.7)102.0(8.5)− 1.2(− 4.4 to 1.9)Waist-to-hip ratio0.85(0.08)0.83(0.06)0.02(− 0.01 to 0.05)Fat (%)a27.1(8.7)26.6(8.5)0.6(− 2.8 to 4.0)Fat mass (kg)a20.8(10.4)20.7(9.6)0.1(− 3.9 to 4.0)Fat free mass (kg)a54.0(11.3)55.7(11.7)− 1.6(− 6.2 to 2.9)Muscle mass (kg)a26.2(6.6)26.9(6.7)− 0.7(− 3.3 to 1.9)Total body water (kg)a39.7(8.3)41.1(8.5)− 1.3(− 4.7 to 2.0)CI confidence interval, SD standard deviation, SGA small for gestational age.aData missing for two participants born SGA and three controls due to pregnancy, uncertainty about pregnancy and a nerve stimulator implant to treat chronic pain. The results of the physical fitness tests are shown in Table 3. There were no group differences in grip strength, modified push-up test or step test results. Mean differences ranged from 0.08 SD units for the step test to 0.32 SD units for the modified push-up test. Separate analyses by sex showed that men in the SGA group had significantly lower grip strength in both hands compared with men in the control group. The mean difference was − 4.8 kg ($95\%$ CI − 0.6 to − 9.0 dominant hand, $95\%$ CI − 0.4 to − 9.2 non-dominant hand, p for interaction SGA × sex = 0.112 and 0.162, respectively).Table 3Physical fitness of adults born small for gestational age (SGA) and non-SGA (control) at term. SGA ($$n = 46$$)Control ($$n = 61$$)Mean difference($95\%$ CI)Mean(SD)Mean(SD)Grip strength, dominant hand (kg)37.0(8.2)38.5(10.3)− 1.5(− 5.1 to 2.0) Womena31.0(5.0)31.8(5.5)− 0.8(− 3.7 to 2.1) Men42.5(6.4)47.3(8.1)− 4.8(− 9.0 to − 0.6)Grip strength, non-dominant hand (kg)34.2(8.4)35.7(10.2) − 1.6(− 5.2 to 2.1) Women28.1(4.9)29.2(5.5)− 1.1(− 4.0 to 1.7) Men39.7(7.0)44.5(8.4)− 4.8(− 9.2 to − 0.4)Number of push-ups in 40 s11.1(4.8)9.6(4.1)1.5(− 0.2 to 3.2) Womena9.5(4.4)8.1(4.1)1.4(− 0.9 to 3.8) Men12.5(4.7)11.7(3.0)0.8(− 1.4 to 3.1)Heart rate after 4 min step test155.3(18.7)154.5(19.4)1.5(− 5.9 to 8.9) Women150.8(17.1)150.4(18.9)0.4(− 9.6 to 10.4) Menb159.9(19.5)160.0(19.1)− 0.1(− 11.3 to 11.2)CI confidence interval, SD standard deviation, SGA small for gestational age.aData missing for one control due to a hand fracture.bData missing for two participants born SGA who could not complete the test. Adults born SGA were shorter than controls (Table 2). Height was associated with grip strength ($r = 0.722$, $p \leq 0.001$ dominant hand, $r = 0.711$, $p \leq 0.001$ non-dominant hand). When we adjusted for height, the difference in maximal grip strength among men decreased to − 2.8 kg ($95\%$ CI − 1.7 to 7.3 dominant hand, $95\%$ CI − 2.0 to 7.6 non-dominant hand). Results were unchanged regarding anthropometric measures and body composition when we performed sensitivity analyses by excluding eight participants born SGA and five controls who were pregnant, had musculoskeletal diagnoses or other conditions affecting physical functioning. However, the SGA group performed 2.4 ($95\%$ CI 0.7 to 4.1) more push-ups than the control group. ## Discussion In this study we found no differences in body composition or physical fitness between adults born SGA and the control group, measured by grip strength, a 40-s modified push-up test and a 4-min submaximal step test. However, men in the SGA group had significantly lower grip strength in both the dominant and non-dominant hand compared with men in the control group. A strength of this study includes the prospective population-based design, where participants were recruited and followed from mid-pregnancy. At birth, SGA was defined as birth weight below the 10th percentile. This may also comprise individuals who are genetically small and not necessarily growth restricted. Additionally, the control group may comprise individuals who are growth restricted, but still have a birth weight above the 10th percentile. This could possibly contribute to smaller differences between the groups in this study. Nevertheless, the 10th percentile is a common cut-off used to identify SGA individuals6. At the 32-year follow-up, $56.3\%$ of the invited were assessed clinically. This low participation rate can partly be explained by the data collection being carried out during the Covid-19 pandemic. Even though follow-up rates of 50–$80\%$ participation have been suggested to be acceptable in cohort studies39, individuals performing worse may have a stronger tendency to drop out39,40. This could have led to a selection bias toward physically fit participants. However, there were few differences in background variables between participants and non-participants, and assessment of physical fitness was only a part of a larger follow-up examination. Thus, it seems unlikely that the results were affected by selection bias. Nevertheless, the loss to follow-up limits the sample size and hence gave wider confidence intervals than would be expected with a larger sample size. Another strength is the use of objective measurement tools to assess body composition and physical fitness, as self-reports may be biased by over- or underestimation41. Assessments were carried out in the same order for all participants by trained examiners blinded to birth weight groups. Bioelectrical impedance analysis by the Seca® mBCA 515 has shown to agree well with the accurate and precise DXA method42–44, which is considered the reference measurement for differentiating lean and fat tissues. Both grip strength measured by a dynamometer and the modified push-up test are reported to be valid instruments for assessing muscular fitness33,45. A limitation of the study was the measurement of cardiorespiratory fitness by a submaximal test with heart rate as the outcome, as heart rate is largely individual46, and a maximal exercise test measuring maximal oxygen uptake would evaluate cardiorespiratory fitness more accurately16. However, a submaximal test was considered more feasible in this study, as it is less time consuming and more comfortable for the participants. In this study, participants born SGA were shorter than controls, which we have previously documented in adolescence and young adulthood11,13,47. These findings are in line with other studies of children, adolescents, and adults7,10,48,49. However, evidence regarding overweight and adiposity is conflicting. We did not find differences in BMI, waist circumference or waist-to-hip ratio between the groups, consistent with our previous report of similar body composition of participants born SGA and controls at 26 years of age, measured by DXA13. Other studies of term-born adults with a birth weight < 10th percentile have reported both similar BMI and waist-to-hip ratio50, lower weight and reduced lean body mass51, and a higher percentage body fat12 and total abdominal fat mass50 compared with a control group. However, these studies used different birth weight percentiles to define the control group, which may explain some of the discrepancy. Our hypothesis that adults born SGA would display a lower fitness level than their peers was not confirmed in this study. However, men born SGA had approximately 5 kg lower grip strength than men in the control group in the unadjusted analyses. This result must be interpreted with caution, as the $95\%$ CI was rather wide. The difference in grip strength is consistent with the recent Swedish study that found strong associations between birth weight in men born at term and grip strength at 18 years of age27. However, in that study a one SD lower birth weight was associated with 1.8 kg lower grip strength. In the present study, mean difference in birth weight SD score was 1.44, corresponding to a 2.6 kg difference. This is consistent with what we observed in men and would also be included in the CI we observed among women. Further, our finding is also in accordance with other studies that have found strong associations between lower birth weight and reduced grip strength in adulthood, regardless of gestational age at birth21. The reduced grip strength found for men born SGA in this study could indicate increased risk of negative health outcomes, as increased hazard ratio of all-cause mortality ranging from 1.0820 to 1.1619, and for cardiovascular mortality of 1.1719, have been reported for every 5 kg reduction in grip strength. The differences in grip strength among men only may be related to motor development, as associations between motor development and grip strength in adulthood have been documented22. Growing up, boys in the general population are reported to have motor problems more often than girls52. In the SGA population, several studies also show that boys are more vulnerable to growth restriction in utero than girls, possibly because of a higher growth velocity3. Thus, boys and men born SGA may be more susceptible to unfavourable development outcomes. In support of this, we have previously reported reduced manual dexterity at 14 years of age in boys born SGA, and not girls47. This may be related to the findings of reduced grip strength in the present study. When we adjusted for height, the difference in grip strength was reduced and no longer significant, indicating that the difference was partly mediated through a lower height in men born SGA. This is in accordance with previous research reporting that height is associated with grip strength22. However, when grip strength is used as a predictor of mortality and functional capacity it is not adjusted for height19,20,53. Additionally, even when adjusted for height, the grip strength of men born SGA was lower than the 59 kg normative value for men of the same age in Norway35,54. This underlines the relevance of the lower grip strength finding in our study. Contrary to our hypothesis, we did not find any differences between the groups in the push-up test or step test. In a sensitivity analysis excluding participants with conditions affecting physical functioning, the adults born SGA even performed more push-ups than the controls. In a study of 287,000 male military conscripts Ahlquist et al.26 reported that among term-born men, each unit decrease in birth weight z-score was associated with reduced cardiorespiratory fitness of 7.9 W in maximal workload, corresponding to approximately 0.2 SD in that population. That study did not compare men born SGA with men born non-SGA and used a different proxy for cardiorespiratory fitness than we did. However, our confidence intervals among men ranged from less than − 0.5 SD to more than + 0.5 SD, and we may not have had adequate power to observe an association that was observed in the paper of Ahlquist et al.26. Ridgway et al.22 reported a weak association between lower birth weight and lower aerobic fitness, however, the sample also included late preterm born individuals. Our results are consistent with our previous findings at 18 years of age11 and with two other small studies of Danish men that found no differences in VO2max between those with birth weight ≤ 10th percentile and a control group at 19 and 24 years of age24,25. Thus, it seems unlikely that adults born SGA have any moderate or large deficit in cardiorespiratory fitness, but we cannot exclude a weak association with lower birth weight. Overall, the lack of differences between adults born SGA and non-SGA controls in this study is promising with regards to future health. However, reduced grip strength is an established predictor of future physical function, morbidity and mortality19,20,53, and is shown to track through life53. It is therefore worrying that men born SGA already at 32 years of age had reduced grip strength compared with men in the control group. Consequently, promoting a physically active lifestyle in men born SGA may be advantageous. A physically active lifestyle has been shown to form early in life55, indicating that promotion of physical activity, especially concerning activities that enhance muscular strength, should be a focus from childhood. ## Conclusion Overall, we found no differences in body composition or physical fitness between adults born SGA and non-SGA at term. However, men born SGA had lower grip strength than men born non-SGA. There are few studies concerning physical fitness in individuals born SGA at term entering mid-adulthood. Further research is therefore needed to determine whether adults born SGA have lower physical fitness than their non-SGA peers. ## References 1. 1.World Health OrganizationPhysical Status: The Use and Interpretation of Anthropometry. Report of a WHO Expert Committee1995WHO. Physical Status: The Use and Interpretation of Anthropometry. Report of a WHO Expert Committee (1995.0) 2. Romo A, Carceller R, Tobajas J. Intrauterine growth retardation (IUGR): Epidemiology and etiology. Pediatr. Endocrinol. Rev. (2009.0) 6 332-336. PMID: 19404231 3. DiPietro JA, Voegtline KM. The gestational foundation of sex differences in development and vulnerability. Neuroscience (2017.0) 342 4-20. DOI: 10.1016/j.neuroscience.2015.07.068 4. Saenger P, Czernichow P, Hughes I, Reiter EO. Small for gestational age: Short stature and beyond. Endocr. Rev. (2007.0) 28 219-251. DOI: 10.1210/er.2006-0039 5. Li X. Association of low birth weight with cardiometabolic diseases in Swedish twins: A population-based cohort study. BMJ Open (2021.0) 11 e048030. DOI: 10.1136/bmjopen-2020-048030 6. Lee AC. Estimates of burden and consequences of infants born small for gestational age in low and middle income countries with INTERGROWTH-21(st) standard: Analysis of CHERG datasets. BMJ (2017.0) 358 j3677. DOI: 10.1136/bmj.j3677 7. Kramer MS. Is restricted fetal growth associated with later adiposity? Observational analysis of a randomized trial. Am. J. Clin. Nutr. (2014.0) 100 176-181. DOI: 10.3945/ajcn.113.079590 8. Biosca M. Central adiposity in children born small and large for gestational age. Nutr. Hosp. (2011.0) 26 971-976. DOI: 10.1590/S0212-16112011000500008 9. Hediger ML. Muscularity and fatness of infants and young children born small- or large-for-gestational-age. Pediatrics (1998.0) 102 E60. DOI: 10.1542/peds.102.5.e60 10. Strauss RS. Adult functional outcome of those born small for gestational age: Twenty-six-year follow-up of the 1970 British Birth Cohort. JAMA (2000.0) 283 625-632. DOI: 10.1001/jama.283.5.625 11. Evensen KAI. Effects of preterm birth and fetal growth retardation on cardiovascular risk factors in young adulthood. Early Hum. Dev. (2009.0) 85 239-245. DOI: 10.1016/j.earlhumdev.2008.10.008 12. Meas T, Deghmoun S, Armoogum P, Alberti C, Levy-Marchal C. Consequences of being born small for gestational age on body composition: An 8-year follow-up study. J. Clin. Endocrinol. Metab. (2008.0) 93 3804-3809. DOI: 10.1210/jc.2008-0488 13. Balasuriya CND. Metabolic outcomes in adults born preterm with very low birthweight or small for gestational age at term: A cohort study. J. Clin. Endocrinol. Metab. (2018.0) 103 4437-4446. DOI: 10.1210/jc.2018-00464 14. Isganaitis E. Developmental programming of body composition: Update on evidence and mechanisms. Curr. Diab. Rep. (2019.0) 19 60. DOI: 10.1007/s11892-019-1170-1 15. de Mendonca E, de Lima Macena M, Bueno NB, de Oliveira ACM, Mello CS. Premature birth, low birth weight, small for gestational age and chronic non-communicable diseases in adult life: A systematic review with meta-analysis. Early Hum. Dev. (2020.0) 149 105154. DOI: 10.1016/j.earlhumdev.2020.105154 16. 16.American College of Sports MedicineACSM's Health-Related Physical Fitness Assessment Manual20185Wolters Kluwer. ACSM's Health-Related Physical Fitness Assessment Manual (2018.0) 17. Kodama S. Cardiorespiratory fitness as a quantitative predictor of all-cause mortality and cardiovascular events in healthy men and women: A meta-analysis. JAMA (2009.0) 301 2024-2035. DOI: 10.1001/jama.2009.681 18. Garcia-Hermoso A. Muscular strength as a predictor of all-cause mortality in an apparently healthy population: A systematic review and meta-analysis of data from approximately 2 million men and women. Arch. Phys. Med. Rehabil. (2018.0) 99 2100-2113. DOI: 10.1016/j.apmr.2018.01.008 19. Leong DP. Prognostic value of grip strength: Findings from the prospective urban rural epidemiology (PURE) study. Lancet (2015.0) 386 266-273. DOI: 10.1016/S0140-6736(14)62000-6 20. Kim Y. Independent and joint associations of grip strength and adiposity with all-cause and cardiovascular disease mortality in 403,199 adults: The UK Biobank study. Am. J. Clin. Nutr. (2017.0) 106 773-782. DOI: 10.3945/ajcn.117.156851 21. Dodds R. Birth weight and muscle strength: A systematic review and meta-analysis. J. Nutr. Health Aging (2012.0) 16 609-615. DOI: 10.1007/s12603-012-0053-9 22. Ridgway CL. Birth size, infant weight gain, and motor development influence adult physical performance. Med. Sci. Sports Exerc. (2009.0) 41 1212-1221. DOI: 10.1249/MSS.0b013e31819794ab 23. Tikanmäki M. Physical fitness in young adults born preterm. Pediatrics (2016.0) 137 1289. DOI: 10.1542/peds.2015-1289 24. Brons C. Mitochondrial function in skeletal muscle is normal and unrelated to insulin action in young men born with low birth weight. J. Clin. Endocrinol. Metab. (2008.0) 93 3885-3892. DOI: 10.1210/jc.2008-0630 25. Jensen CB, Storgaard H, Madsbad S, Richter EA, Vaag AA. Altered skeletal muscle fiber composition and size precede whole-body insulin resistance in young men with low birth weight. J. Clin. Endocrinol. Metab. (2007.0) 92 1530-1534. DOI: 10.1210/jc.2006-2360 26. Ahlqvist VH. Birth Weight and cardiorespiratory fitness among young men born at term: The role of genetic and environmental factors. J. Am. Heart Assoc. (2020.0) 9 e014290. DOI: 10.1161/JAHA.119.014290 27. Ahlqvist VH. Birth weight and grip strength in young Swedish males: A longitudinal matched sibling analysis and across all body mass index ranges. Sci. Rep. (2019.0) 9 9719. DOI: 10.1038/s41598-019-46200-0 28. 28.World Health Organization. Noncommunicable Diseases. https://www.who.int/news-room/fact-sheets/detail/noncommunicable-diseases (2021). 29. Bakketeig LS. Pre-pregnancy risk factors of small-for-gestational age births among parous women in Scandinavia. Acta Obstet. Gynecol. Scand. (1993.0) 72 273-279. DOI: 10.3109/00016349309068037 30. Vik T. Body proportions and early neonatal morbidity in small-for-gestational-age infants of successive births. Acta Obstet. Gynecol. Scand. Suppl. (1997.0) 165 76-81. PMID: 9219462 31. Hollingshead AB. Two-Factor Index of Social Position (1957.0) 32. Tolonen H. EHES Manual. Part B. Fieldwork Procedures (2016.0) 33. Suni JH. Health-related fitness test battery for adults: Aspects of reliability. Arch. Phys. Med. Rehabil. (1996.0) 77 399-405. DOI: 10.1016/s0003-9993(96)90092-1 34. Astrand PO, Ryhming I. A nomogram for calculation of aerobic capacity (physical fitness) from pulse rate during sub-maximal work. J. Appl. Physiol. (1954.0) 7 218-221. DOI: 10.1152/jappl.1954.7.2.218 35. Kjaer IG, Torstveit MK, Kolle E, Hansen BH, Anderssen SA. Normative values for musculoskeletal- and neuromotor fitness in apparently healthy Norwegian adults and the association with obesity: A cross-sectional study. BMC Sports Sci. Med. Rehabil. (2016.0) 8 37. DOI: 10.1186/s13102-016-0059-4 36. Loe H, Rognmo O, Saltin B, Wisloff U. Aerobic capacity reference data in 3816 healthy men and women 20–90 years. PLoS ONE (2013.0) 8 e64319. DOI: 10.1371/journal.pone.0064319 37. Berle JO, Mykletun A, Daltveit AK, Rasmussen S, Dahl AA. Outcomes in adulthood for children with foetal growth retardation. A linkage study from the Nord–Trondelag Health Study (HUNT) and the Medical Birth Registry of Norway. Acta Psychiatr. Scand. (2006.0) 113 501-509. DOI: 10.1111/j.1600-0447.2005.00704.x 38. Lærum AM. Psychiatric disorders and general functioning in low birth weight adults: A longitudinal study. Pediatrics (2017.0) 139 e20162135. DOI: 10.1542/peds.2016-2135 39. Fewtrell MS. How much loss to follow-up is acceptable in long-term randomised trials and prospective studies?. Arch. Dis. Child. (2008.0) 93 458-461. DOI: 10.1136/adc.2007.127316 40. Knudsen AK, Hotopf M, Skogen JC, Overland S, Mykletun A. The health status of nonparticipants in a population-based health study: The Hordaland Health Study. Am. J. Epidemiol. (2010.0) 172 1306-1314. DOI: 10.1093/aje/kwq257 41. Obling KH. Association between self-reported and objectively measured physical fitness level in a middle-aged population in primary care. Prev. Med. Rep. (2015.0) 2 462-466. DOI: 10.1016/j.pmedr.2015.05.010 42. Lahav Y, Goldstein N, Gepner Y. Comparison of body composition assessment across body mass index categories by two multifrequency bioelectrical impedance analysis devices and dual-energy X-ray absorptiometry in clinical settings. Eur. J. Clin. Nutr. (2021.0). DOI: 10.1038/s41430-020-00839-5 43. Day K. Comparison of a bioelectrical impedance device against the reference method dual energy X-ray absorptiometry and anthropometry for the evaluation of body composition in adults. Nutrients (2018.0) 10 1469. DOI: 10.3390/nu10101469 44. Bosy-Westphal A. Quantification of whole-body and segmental skeletal muscle mass using phase-sensitive 8-electrode medical bioelectrical impedance devices. Eur. J. Clin. Nutr. (2017.0) 71 1061-1067. DOI: 10.1038/ejcn.2017.27 45. Bobos P, Nazari G, Lu Z, MacDermid JC. Measurement properties of the hand grip strength assessment: A systematic review with meta-analysis. Arch. Phys. Med. Rehabil. (2020.0) 101 553-565. DOI: 10.1016/j.apmr.2019.10.183 46. Kokkinos P. Physical fitness evaluation. Am. J. Lifestyle Med. (2014.0) 9 308-317. DOI: 10.1177/1559827613520128 47. Evensen KAI. Motor skills in adolescents with low birth weight. Arch. Dis. Childhood Fetal Neonatal Ed. (2004.0) 89 F451-F455. DOI: 10.1136/adc.2003.037788 48. Karlberg J, Albertsson-Wikland K. Growth in full-term small-for-gestational-age infants: from birth to final height. Pediatr. Res. (1995.0) 38 733-739. DOI: 10.1203/00006450-199511000-00017 49. Kensara OA. Fetal programming of body composition: Relation between birth weight and body composition measured with dual-energy X-ray absorptiometry and anthropometric methods in older Englishmen. Am. J. Clin. Nutr. (2005.0) 82 980-987. DOI: 10.1093/ajcn/82.5.980 50. Rasmussen EL. Altered fat tissue distribution in young adult men who had low birth weight. Diabetes Care (2005.0) 28 151-153. DOI: 10.2337/diacare.28.1.151 51. Thomas N. Born with low birth weight in rural Southern India: What are the metabolic consequences 20 years later?. Eur. J. Endocrinol. (2012.0) 166 647-655. DOI: 10.1530/EJE-11-0870 52. Blank R. International clinical practice recommendations on the definition, diagnosis, assessment, intervention, and psychosocial aspects of developmental coordination disorder. Dev. Med. Child Neurol. (2019.0) 61 242-285. DOI: 10.1111/dmcn.14132 53. Rantanen T. Midlife hand grip strength as a predictor of old age disability. JAMA (1999.0) 281 558-560. DOI: 10.1001/jama.281.6.558 54. Andersen SA. Physical Fitness Among Adults and Elderly in Norway. Results from an Assessment in 2009–2010 (Fysisk form blant voksne og eldre i Norge. Resultater fra en kartlegging i 2009–2010) (2010.0) 55. Telama R. Tracking of physical activity from early childhood through youth into adulthood. Med. Sci. Sports Exerc. (2014.0) 46 955-962. DOI: 10.1249/MSS.0000000000000181
--- title: A comparative study on the Antihyperlipidemic and antibacterial potency of the shoot and flower extracts of Melastoma malabathricum Linn's authors: - Md. Abdul Kader - Md. Masuder Rahman - Shahin Mahmud - Md. Sharif Khan - Shamsunnahar Mukta - Fatama Tous Zohora journal: Clinical Phytoscience year: 2023 pmcid: PMC9975876 doi: 10.1186/s40816-023-00355-6 license: CC BY 4.0 --- # A comparative study on the Antihyperlipidemic and antibacterial potency of the shoot and flower extracts of Melastoma malabathricum Linn's ## Abstract ### Background Atherosclerosis is arteries’ thickening and stiffening condition manifested due to plaque formation by oxidized-LDL of abundant and deranged lipid metabolism. Traditionally, Melastoma malabathricum Linn (MM) leaves are used for anti-diabetics, abdominal problems, and high blood pressure. The current experiment unveils the potency of ethanol, acetone, and water MM extracts as antibacterial agents and alternative medicine during hyperlipidemic conditions. ### Methods A high cholesterol diet (HCD-2500 mg/kg) was provided with regular feeds for 3 weeks to induce hyperlipidemic mice. Afterward, comparing weight with Group-A (normal control), the hyperlipidemic mice were classified into five groups: Group-B (hyperlipidemic control), Group-C (MFA-500 mg/kg), Group-D (MSE-250 mg/kg), Group-E (MSE-500 mg/kg), and Group-F (ATOVAT-20 mg/kg). And the dosages were given orally for 28 days according to their body weight. Fasting blood was collected at the end of treatment, and serum was taken to test lipid profiling and liver enzymes. ### Results The body mass had waxed significantly ($P \leq 0.001$) in all the groups compared with Group-A. Subsequently, orally administered different doses where group-D and group-E demonstrated magnificent anti-hyperlipidemic potency ($P \leq 0.001$) compared with group-B. During treatment, rapid upward body mass was tardy in group-E ($P \leq 0.001$). However, the liver enzyme expression such as AST, ALT, and ALP was elevated ($P \leq 0.001$) in Group-F, they were significantly lessened ($P \leq 0.001$, $P \leq 0.01$) in Groups-C, D, and E, which indicates these extracts have significant anti-liver damaging potency. Alongside the antibacterial activity of MSE-1500 μg/disc, it exhibited the greatest (16.50 mm) zone of inhibition against Shigella dysenteriae. ### Conclusion However, in our current experiment, depending on the derived data, we can elicit that the Melastoma malabathricum shoot ethanolic (MSE) extract is a potential resource for developing alternative medicine to manage the hyperlipidemic condition. ## Introduction Hyperlipidemia is a chronic propulsive complication resulting from the aberrant metabolism of carbohydrates, proteins, and fats. It nestles as elevated levels of fat molecules such as Total cholesterol (TC), Triglyceride (TG), Low-density lipoprotein (LDL), and Very low-density lipoprotein (VLDL) in the circulatory system [1]. The fundamental cause of hyperlipidemia is lopsided energy existing in the bloodstream between calories consumed and calories expenditure which is bolstered by a sedentary lifestyle [2]. Alongside, the augmented blood cholesterol can be manifested due to genetic reasons [3]. The progressed hyperlipidemic condition is intertwined with several severe maladies in the body such as cardiovascular disease (CD) [4], diabetes mellitus type 2 (T2DM) [5], non-alcoholic fatty liver disease [6], kidney disease, reproductive disease as well as osteoarthritis [7]. In 2019 approximately 17.9 million people died from CD, representing $32\%$ of all global deaths in which $85\%$ had succumbed because of heart attack and stroke. In the impending days, experts are conjecturing that the prevalence of CD will be prodigious due to sedentary lifestyle during the covid-19 pandemic and infectious nature of coronavirus [8]. Cardiovascular diseases such as heart failure (FR), peripheral arterial disease (PAD), and coronary heart disease (CHD) evolve because of deposit fatty tissue (Plaques) inside the blood vessel called atherosclerosis [9]. As a result, the blood flow is averted to reach several organs such as the heart, brain, and extremities that belong to heart attack, stroke, and peripheral arterial disease. Atherosclerosis is inflammation triggered by the oxidized-LDL molecules propelled by inflammatory cytokines and biomarkers [10]. It has been recorded recently the production of reactive oxygen species (ROS) engendered as a byproduct of aerobic metabolism, drug, and toxins are entangled with many human diseases such as cardiovascular disease, cancer, Alzheimer’s disease (AD), aging, and atherosclerosis [11, 12]. The magnificent contribution of phytochemicals is combating several diseases such as cardiovascular [13], Cancer, Colorectal, diabetics, and Bacterial disease [14]. They also exhibit antioxidative (especially flavonoids) properties to mitigate the ROS to escape these detrimental diseases [15]. Where xenobiotics have a drastic impact on the liver and kidney during the treatment of diseases, the phytochemicals have impressive results without any side effects. This penchant propelled us to quest for anti-hyperlipidemic medicinal plants that produce several phytochemicals in leaves, flowers, roots, and fruits. These phyto-compounds act on various mechanisms to cure cancer, cardiovascular and bacterial disease [16]. During biotic and abiotic stresses, plants produce a plethora of secondary metabolites such as polyphenols, flavonoids, terpenoids, alkaloids, and plant sterols to defend themselves and provide unique bioactivity on humans health [17]. The Small shrub Melastoma malabathricum Linn (MM) belongs to the family of Melastomataceae commonly available in tropical and temperate Southeast Asian countries, locally known as Phutki to Bangladesh, India, and Senduduk to Malay [18]. Its leaves, shoots, barks, and roots are processed in various ways to treat various types of diseases such as high blood pressure, diabetes, dysentery, diarrhea, piles, leucorrhea, cancer, epilepsy, ulcers, gastric ulcers, skin diseases, arthritis, tenderness in the legs, bleeding, toothache, and smallpox from the traditional times [19]. The Melastoma malabathricum Linn (MM) leaves have traditionally been used against different diseases, but no comparative study of ethanol and acetone extract is available in the hyperlipidemic condition. Alongside, the relative study of ethanol and water extract against enlisted pathogenic bacteria in our current investigation was undocumented. That’s why to stand alternative medicine; an attempt was taken to unveil the unique potency of MM against the hyperlipidemic condition and bacterial infections. Moreover, scientific research indicates that similar species from different environmental and geographical locations significantly vary in their metabolites and biological activities [20]. Finally, we were engrained to manifest the concealed medicinal properties of Melastoma malabathricum Linn (MM) shoot and flower against the hyperlipidemic condition through a mice model. ## Plant materials The shoot and flower (purple-magenta petals) of Melastoma malabathricum were collected from the local area near Tetulia Sub-district of Panchagarh in Bangladesh and kept in a shading place. Subsequently, the plant is authenticated at the Department of Biotechnology and Genetic Engineering, Mawlana Bhashani Science and Technology University (MBSTU), Santosh, Tangail-1992, Bangladesh. ## Preparation of extracts The shoots were first appropriately washed with clean water to remove any adhering dirt. The lush flowers were also collected, and both of the plant materials had been kept in a shady ambiance. Then, completely dried, both plant materials were ground into a coarse powder by a grinding machine and stored in an airtight container. The crude powder of Melastoma malabathricum was dissolved in three solvents (1:5 g/ml), ethanol, acetone, and water, respectively, in different cylinders and enshrouded with aluminium foil for 7 days with occasional shaking and stirring to make the plant extract. Whether the shoot powder had given respectfully into absolute ethanol, water, and the flower powder had also dissolved only into acetone. Subsequently, each solution of plant powder was filtered through Whatman No.1 filter paper, and the filtrate solvents were completely evaporated under reduced pressure at 40 °C using a rotary evaporator. Subsequently, the derived plant extracts were tagged as MM shoot ethanol (MSE) extract, MM shoot water (MSW) extract, and MM flower acetone (MFA) extract. Finally, the plant extracts were kept in small sterile bottles under refrigerated conditions (4 °C) until used. ## Preparation of high-cholesterol diet (HCD) The 80 g cholesterol powder was mixed finely with 500 ml vanaspati ghee and edible coconut oil, resulting in a 160 mg/ml concentration, where the vanaspati ghee and palatable coconut oil ratio was 3:2 (v/v) [21]. The high-fat diet was supplied with daily grower feeds (Sonali grower feed in Bangladesh) during the induction of hyperlipidemic mice. ## Preparation of atorvastatin solution 5 mg tablets ($$n = 20$$) were obtained from a renowned dispensary at Tangail, Bangladesh, and finely pulverized before being dissolved in 1 ml distilled water to give a concentration of 5 μg/μl. An oral dose was chosen and provided a fixed time each day based on body mass. ## Preparation of carbimazole solution 5 mg tablets ($$n = 20$$) were collected from the same dispensary, ground and finely mixed in distilled water to a rising concentration of 5 μg/μl. They were administered orally every day during the HCD mediated induction of hyperlipidemic mice. ## Animals handling Abiding the ethical rules, the female albino 30 mice (~ 25 g) were imported from the international Center for diarrhoeal disease research, Bangladesh (ICDDR, B) and taken care of in well-ventilated mice laboratory at the Department of Biotechnology and Genetic Engineering, MBSTU. The age of the mice was 2 months, and they were given 5 days to acclimatize with this milieu, where they were exposed to 12 hours of light and 12 hours of darkness. They were provided grower feeds with clean tap water once daily (8.00 am - 9.00 am). ## Induction of hyperlipidemia The experiment was conducted on two-month-old female mice, where all of the groups were indiscriminately given the daily grower feeds and potable water. But, except for group A, the rest of the groups went through a high cholesterol diet (HCD) with 1.35 mg/kg of carbimazole for 3 weeks to induce hyperlipidemic mice. The hyperlipidemic condition of the HCD mice had confirmed taking weight compared with the standard control that was prominent after 3 weeks. ## Experimental design Thirty albino female mice had been categorized into six groups in which each of the groups had five mice (~ 25 g) to commence the experiment with potable water to drink. Group A: (Normal Control): Solely daily grower feed was provided. Group B: (Hyperlipidemic control): High cholesterol diet (HCD) 2500 mg/kg and 1.35 mg/kg of carbimazole were given with neither drug nor extract. Group C: Provided HCD 2500 mg/kg, 1.35 mg/kg of carbimazole, 500 mg/kg MFA dose. Group D: Served HCD 2500 mg/kg, 1.35 mg/kg of carbimazole, 250 mg/kg MSE dose. Group E: Served HCD 2500 mg/kg, 1.35 mg/kg of carbimazole, 500 mg/kg MSE dose. Group F: (Positive control): Received HCD 2500 mg/kg, 1.35 mg/kg of carbimazole and 20 mg/kg of atorvastatin. The treatment of this experiment was persistent for 28 days, where three groups received different plant extracts with different concentrations, and the impacts of these doses had evaluated by quantifying the liver markers (AST, ALT, and ALP) and performing histopathology. ## Biochemical assay and histopathology Given 28 days of treatments, all groups were anesthetized with chloroform and immediately sacrificed to collect the blood from the heart using the syringe. Subsequently, blood was kept in the Eppendorf tube and placed on an icebox to be clotted, and some organs (Heart, Liver, Kidney) were preserved in $10\%$ formaldehyde for histopathology analysis [22]. The clotted blood was centrifuged at 4000 rpm at 4 °C for 5 min to obtain serum which is then used to identify cardiovascular and liver-damaging markers using the “Human diagnostics worldwide kit” according to manufacturer’s instructions where Bioanalyzer (Humolyzer 3000, Germany) was utilized. Among the markers, the magnitude of LDL and VLDL had been calculated based on Friedewald et al. formula [23]: (a) VLDL = TG/5 (mg/dl) (b) {LDL = TC – (HDL + TG)} (mg/dl). The histopathology was performed amid glass slide preparation, where organs were dehydrated by passing through a graded series of alcohol and embedded in paraffin blocks to prepare 5 mm sections using a semi-automated rotary microtome. These slides were stained using hematoxylin and eosin [24]. ## Gram-negative bacteria The pathogenicity of gram-negative bacteria is immensely harmful to human health and other animals. The prevalence of diarrhea due to Shigella and E.coli in Bangladesh is prominent [25, 26]; that’s why some strains of these species had taken to bring out the antibacterial potentiality of Melastoma malabathricum using MSE and MSW Extract. The name of these gram-negative bacterial strains is Enterotoxigenic E.coli (ETEC), Enteropathogenic E.coli (EPEC), Shigella boydii (SB), Shigella flexneri (SF), Shigella sonnei (SS), Shigella dysenteriae (SD). ## Disk diffusion method The disk diffusion method was used to ferret out the antibacterial potentiality of MSE and MSW extract using Mueller Hinton Agar (MHA). First, the young culture of all the above strains was prepared by transferring 15 μl suspension culture to Mueller Hinton Broth (MHB) for 2.30 h of incubation at 37 °C. Afterward, 100 μl of young culture were soaked in an MHA plate and impregnated with sterile filter paper disks (6 mm in diameter). Subsequently, different concentrations of MSE and MSW extracts were dissolved in $5\%$ dimethyl sulfoxide (DMSO), and 15 μl of each extract was placed on the paper disks (600 μg/disc, 900 μg/disc, 1200 μg/disc, 1500 μg/disc) and standard antibiotics (Chloramphenicol 30 μg/disc) as well. Finally, all cultured plates were incubated at 37 °C overnight to give proper ambiance to grow bacterial culture. During that moment, contingent on phytochemicals concentration, the growth of bacterial culture was obviated around these disks that exhibited a transparent zone called the zone of inhibition. The zone of inhibition is measured with mm units. ## Statistical analysis All of the results in our current experiment were manifested as mean ± standard error mean (SEM), the analysis of variance (ANOVA) was utilized to demonstrate significant differences among the independent groups (IBM SPSS Statistics, Version 25). The values were significantly considered when the p-value was < 0.05. ## Acute toxicity study The pre-existing research on MM ethanolic extract has unveiled that these medicinal plants haven’t had any acute toxicity impact in the animal model. The experimental dose was 2000 mg/kg, and 5000 mg/kg continued for experimenting. No noticeable change occurred in skin and behavior like diarrhea, sleeping, salivation, and tremors [27]. We conducted our experiment with two plant extracts at a dose of 250 mg/kg and 500 mg/kg for 28 days; no mice demise, phenotypic and behavioral changes occurred due to extract toxicity. This result was also substantiated further by histopathology analysis of the liver, kidney, heart. ## The consequence of MM plant extracts on liver In our current experiment, the expression of liver markers was elevated significantly ($P \leq 0.001$) in the HCD induced hyperlipidemic mice compared with the normal control group. At the end of the treatment, the liver enzyme expression was tremendously reduced ($P \leq 0.001$, $P \leq 0.01$) in groups C, D, and E using fixed doses, compared with group B (Table 1). The magnitude of reduction of the liver enzyme expression (ALP, ALT, AST) comparing with hyperlipidemic control group B, were for the doses (a) MFA-500 mg/kg: about $7\%$, $24\%$, $17\%$ (b) MSE-250 mg/kg: about $5\%$, $15\%$, $8\%$ and (c) MSE-500 mg/kg: about $30\%$, $33\%$, $24\%$ and, in the group C, D and E, respectively. On the other hand, group F, treated with atorvastatin (Drug) 20 mg/kg, demonstrated a significantly higher ($P \leq 0.001$) level of AST, ALT, and ALP expression than group B (Table 1).Table 1Relation between different conc. of MM plant extracts and drug to liversLiver markers(U/L)Group A(Normal control)Group B(HYPER control)Group C(MFA-500mg/kg)Group D(MSE-250 mg/kg)Group E(MSE-500 mg/kg)Group F(ATOVAT-20 mg/kg)ALP132.2 ± 1.50198.8 ± 1.16a185.4 ± 1.81d189.4 ± 1.72e139.4 ± 1.86d260.4 ± 1.44dALT48 ± 1.58102.4 ± 1.21a79.2 ± 1.32d87.2 ± 1.50e69.2 ± 1.59d168.6 ± 1.57dAST189 ± 1.58275.2 ± 2.03a228.6 ± 1.63d258 ± 1.70d210.4 ± 1.94d388.8 ± 2d Outcomes of this experiment is expressed as mean ± standard error of mean (SEM) where $$n = 5$$; aP < 0.001; compared with normal control; dP < 0.001; eP < 0.01; compared with hyperlipidemic control; One-Way ANOVA: Dunnett’s multiple comparison test. Histopathology was done to substantiate the biochemical test of liver-damaging markers. The metabolic conversion of xenobiotics, phytochemicals, and deposited fats in the liver is converted into reactive intermediates such as electrophilic compounds or Reactive oxygen species (ROS), which can potentially transmute the structure and function of cellular macromolecules [28]. The group F treated with ATOVAT-20 mg/kg demonstrated scarring, which might be due to oxidative stress originating (ROS) from atorvastatin metabolism and amassed fats in the liver. In contrast, group C treated with MFA-500 mg/kg also didn’t restrain the deposition of fats significantly. Even though group D was administered MSE-250 mg/kg had a little bit of liver scarring and deposited fats, in group E, using MSE-500 mg/kg dose declined deposition of fats in the liver that was almost similar to normal control liver (Fig. 1).Fig. 1The Potentiality of Melastoma malabathricum extracts to protect the liver in HCD mice. a Normal group: Demonstrated no fat accumulation and hepatocytes enlargement (gray arrow). b Hyperlipidemic control: Exhibited prominently amassed micro fatty layers (yellow allow), enlarged hepatocytes (gray arrow), and cellular damaging (blue arrow) surrounding the central vein (CV) (green arrow) but also noticed the presence of inflammatory cells in the CV (red arrow). c MFA-500 mg/kg: Deposition of fat droplets among the intercellular space (yellow arrow), inflammatory cells infiltration (red allow), and, a little bit larger hepatocytes were also observed (gray arrow). d MSE-250 mg/kg: Fat granules hoarded among the intercellular space (yellow arrow) and hepatocyte enlargement (gray arrow) also occurred around CV. e MSE-500 mg/kg: Cellular infrastructure was almost the same as a normal control group, with no fat globules or fatty layer found in the liver. f ATOVAT-20 mg/kg: Seemingly no sign of deposited fats but a bit of cellular deformation (blue arrow) observed like as group D. (Pictures magnification 20X) ## The consequence of MM plant extracts on lipid profiling In the hyperlipidemia-induced mice, TG, TC, VLDL, and LDL level was significantly higher ($P \leq 0.001$), whereas HDL level was prominently lower ($P \leq 0.001$) compared with the normal group. During treatment, group C administered with MFA-500 mg/kg had shown trivial reduction ($P \leq 0.01$) of TG, TC, VLDL and LDL and elevated paltry ($p \leq 0.05$) HDL level; whereas group D, E and F with MSE-250 mg/kg, MSE-500 mg/kg and ATOVAT-20 mg/kg declined significantly ($P \leq 0.001$) and HDL level also augmented ($P \leq 0.001$; $P \leq 0.05$) compared with group B (Table 2).Table 2The antihyperlipidemic potentiality of MM plant extracts at different conc. in HCD miceLipid-molecules(mg/dl)Group A(Normal control)Group B(HYPER control)Group C(MFA-500mg/kg)Group D(MSE-250 mg/kg)Group E(MSE-500 mg/kg)Group F(ATOVAT-20 mg/kg)TG115 ± 1.58225.8 ± 1.80a218.2 ± 1.37e160.6 ± 1.72d131.8 ± 1.56d123 ± 1.41dTC50 ± 1.58110 ± 1.64a107.6 ± 1.63e78.2 ± 1.65d62.8 ± 1.36d55.8 ±.97dHDL23 ± 1.5810.4 ± 1.72a13.4 ± 1.29hn16.6 ± 1.43f22.4 ± 1.81d17.4 ± 1.36fVLDL23 ±.3245.1 ±.36a43.6 ±.27e32 ±.32d26.4 ±.33d24.6 ±.28dLDL4 ±.4754.4 ± 3a39.8 ± 8.95e29.5 ±.71d14 ± 3.45d13.8 ±.76d Outcomes of this experiment is expressed as mean ± standard error of mean (SEM) where $$n = 5$$; aP < 0.001; compared with normal control and dP < 0.001; eP < 0.01; fP < 0.05; hnp > 0.05 compared with hyperlipidemic control. One-Way ANOVA: Dunnett’s multiple comparison test. ## The consequence of MM extracts on the heart The result of lipid profiling is buttressed by histopathology analysis of the heart of each group of mice. Group B and C exposed cardiac muscle deformation, distorted intercalated disk, and significant hiatus between interstitial spaces due to deposition of fats (Yellow arrow), which have the strong potential to oxidize and trigger the immune response. Alongside, MSE-250 mg/kg dose-treated group D also manifested ample interstitial space and little fat deposition without any cardiac muscle contortion. In contrast, group E treated with MSE-500 mg/kg declined the deposition of fats and cardiac muscle distortion in the heart that was almost congruous with group A and F (Fig. 2).Fig. 2Effect of Melastoma malabathricum plant extracts at different conc. on the heart in HCD mice. a Normal group: Histopathology of the normal group’s heart demonstrated no changes among the intercalated disk (red arrow) and interstitial space (green arrow). b Hyperlipidemic control: Intercalated disk was contused (red arrow), interstitial space was augmented (green arrow), and cellular deformation was induced due to infiltration of fat molecules (yellow arrow). c MFA-500 mg/kg: Intercalated disk was deteriorated (red arrow), as well as the interstitial space was ameliorated (green arrow) endorsed by amassing fat globules (yellow arrow). d MSE-250 mg/kg: Histopathology of this group’s exhibited enhanced interstitial space without any cellular contortion. A little bit presence of fats was discerned in this analysis (yellow arrow). e MSE-500 mg/kg: This dose showed a stunning histopathology result which was almost similar to normal heart structure, minimized interstitial space without any cellular contusion. f ATOVAT-20 mg/kg: Atorvastatin drug control group’s histopathology appeared no cellular deformation that was almost normal but manifested with trivial interstitial space. ( Pictures magnification 20X) ## The consequence of MM extracts on the kidney The kidney histopathology demonstrated the presence of inflammatory cells, amassed fats molecules, and enlarged glomerulus in group B. This horrendous condition manifested maybe because of fat infiltration and subsequently oxidized LDL cholesterol. Group C, treated with MFA-500 mg/kg, exhibited the same expression. On the other hand, group D treated with MSE-250 mg/kg revealed attenuated expression of inflammatory cells, little deposition of fat molecules, and small size of glomerulus compared with hyperlipidemic control. But when the treatment was performed with the tested drug MSE-500 mg/kg, group E demonstrated appreciable results found no inflammatory cells, deposited fats globules, and standard glomerulus size almost resembled group A and F (Fig. 3).Fig. 3Effect of Melastoma malabathricum extracts at different conc. on the kidney in HCD mice. a Normal group: Histopathology demonstrated the average size of the glomerulus (blue arrow) without any accumulation of fatty layers or globules. b Hyperlipidemic control: Exacerbated and enlarged glomerulus (blue arrow) due to the deposition of fat molecules as well as the presence of inflammatory cells in the blood vessels (red arrow). c MFA-500 mg/kg: The glomerulus was a little bit larger (blue arrow), demonstrating the presence of inflammatory cells (red arrow) and enlarged fat globules (yellow arrow) after performing histopathology analysis. d MSE-250 mg/kg: A bit expanded glomerular size (blue arrow), some fat globules deposition (yellow arrow), and also appeared minute inflammatory cells in the blood vessel (red arrow). e MSE-500 mg/kg: The histopathology of the tested drug with this conc. Exhibited typical infrastructure of glomerulus and no prominent sign of fats deposition. f ATOVAT-20 mg/kg: Similar histopathology was observed compared to the normal group. ( Pictures magnification 20X) ## The trend of weight during treatment Three weeks later, the body weight in all groups (Provided HCD) had waxed significantly ($P \leq 0.001$) compared with the normal group, immediately given treatment with MSE-250 mg/kg and MSE-500 mg/kg in groups D and E, resulting abated the rapid upward bodyweight significantly ($P \leq 0.01$, $P \leq 0.001$) at the 6th and 7th weeks (Table 3). But group C treated with the MFA-500 mg/kg manifested a slight tendency ($P \leq 0.05$) to lose body mass compared with group B.Table 3The propensity of body weight during treatment with different MM extracts in hyperlipidemic miceWeeks (g)Group A(Normal control)Group B(HYPER control)Group C(MFA-500mg/kg)Group D(MSE-250 mg/kg)Group E(MSE-500 mg/kg)Group F(ATOVAT-20 mg/kg)Induction1day24.2 ±.0724.24 ±.0923.54 ±.0723.48 ±.0923.68 ±.0723.82 ±.051st26.6 ±.1727.7 ±.09a26.4 ±.09a26.5 ±.09a26.8 ±.04a26.7 ±.07a3rd28.8 ±.0533.5 ±.09a32.9 ±.04a31.6 ±.09a32.3 ±.12a31.2 ±.09aTreatment6th33.6 ±.6842.7 ±.66a39.4 ±.68f38.7 ±.75e37.4 ±.80d36.7 ±.71d7th34.9 ±.7044.3 ±.72a41.4 ±.65f39.9 ±.69d38.8 ±.69d38.3 ±.66d Results are expressed as mean ± standard error of mean (SEM); where $$n = 5$$; aP < 0.001; compared with normal control; dP < 0.001; eP < 0.01; fP < 0.05 compared with hyperlipidemic control. One-Way ANOVA: Dunnett’s multiple comparison test. ## The antibacterial potentiality of MM plant extract Different concentrations of MSE extract ($A = 600$ μg/disc, $B = 900$ μg/disc, $C = 1200$ μg/disc, $D = 1500$ μg/disc) had exhibited prominent zone of inhibition against S. dysenteriae, S. sonnei, S.flexneri strain, whereas MSW extract demonstrated a mediocre zone of inhibition. But both MSE and MSW extract concentrations revealed less potentiality to inhibit the growth of S. boydii, ETEC, and EPEC strains (Table 4).Table 4The zone of inhibition of different conc. of 2 MM extracts on some pathogenic bacteriaZone of Inhibition(mm in diameter)Name of the BacteriaMSEMSWNCCP[A][B][C][D][A][B][C][D][E][F]S. dysenteriae13 ± 114.75 ±.2515.50 ±.5016.50 ±.508.50 ±.509 ± 19 ± 19.50 ±.50026 ± 1S. sonnei9.50 ±.5010.50 ±.5010.50 ±.5011 ± 17 ± 18.50 ±.508.50 ±.508.50 ±.50024.50 ±.50S.flexneri11 ± 111.50 ±.5012.50 ±.5013.50 ±.506.50 ±.507.50 ±.507.50 ±.508.50 ±.50021.50 ±.50S. boydii7.50 ±.508.50 ±.509 ± 110.50 ±.506.50 ±.507.50 ±.507.50 ±.508.50 ±.50024.50 ±.50ETEC6.50 ±.508 ± 08.50 ±.509 ± 06.50 ±.507 ± 07.50 ±.508 ± 0024.50 ±.50EPEC6.50 ±.508 ± 08.50 ±.509 ± 06 ± 07 ± 07 ± 08 ± 00±16.50 ±.50 Results are expressed as mean ± standard error of mean (SEM) where $$n = 2$$; $A = 600$ μg/disc; $B = 900$ μg/disc; $C = 1200$ μg/disc; $D = 1500$ μg/disc; $E = 15$ μL/disc of $5\%$ DMSO; $F = 30$ μg/disc of Chloramphenicol (CP); Negative control (NC); MSE = Melastoma malabathricum shoot ethanol extract; MSW = Melastoma malabathricum shoot water extract. ## Discussion The Melastoma malabathricum has tremendous medicinal significance in several disease conditions like diarrhea, arthritis, gastric ulcer, skin disorder, cancer, diabetics, and high blood pressure because of possessing profuse phytochemicals [19]. The major concern of our study was to investigate the comparative anti-hyperlipidemic potency of MSE and MFA extract using two solvents. This approach has proceeded in hyperlipidemia-induced mice with a high cholesterol diet and carbimazole [29]. The HCD was prepared by combining miscellaneous fat-rich products like cholesterol, coconut oil, and natural Ghee [30]. On the other hand, Carbimazole (pro-drug) is an antagonist of the thyroid peroxidase enzyme; therefore, T3 and T4 thyroid hormones are reduced. Furthermore, it has been recorded that during hypothyroidism, the total blood cholesterol, VLDL, and TG level is ameliorated; that’s why this tactic had been taken to rapidly increase the blood total cholesterol level in the mice [31, 32]. The hyperlipidemic condition was induced by ingesting HCD and carbimazole to all groups with their daily feeds except the normal control according to their body mass. At every single week, the weight was recorded (Table 3), whereas the weight was voluminous compared with the normal control in the last week. It is also well defined that increasing cholesterol levels such as TG, TC, HDL, VLDL, and LDL propel body mass [33]. Since the discrepancy of body weight of these groups was ameliorated compared with normal control, the level of the lipid molecules was elevated in the blood that had been substantiated performing lipid profiling (Figs. 5 and 6). After absorption, the flush of lipid molecules in the blood was amassed in the liver, heart, and kidney that had manifested by histopathology analysis (Figs. 1, 2, and 3). The stockpiling of fat in these organs is baneful due to engender several maladies. For instance: The accumulation of fat globules in the liver instinctively can create scarring, fibrosis, hepatic insulin resistance, and oxidative stress resulting in inflammation due to activation of immune cells such as macrophages, neutrophils as well as inflammatory cytokines (IL-1α/β, TNF- α) [34]. In our experiment in group B, the liver condition was exacerbated due to the deposition of fat droplets that could produce oxidative stress and inflammatory cytokines, which demolished the surrounding cells resulting in the liver enzyme’s expression being augmented in the blood serum (Fig. 4). The drug control group demonstrated excessive AST, ALT, ALP levels than the other groups (Table 1); this extensive magnitude could be because of injured liver cells through ROS during xenobiotics (Atorvastatin) metabolism [28]. On the other hand, groups C, D, and E with the treatment of MM plant extract (MFA-500 mg/kg, MSE-250 mg/kg, MSE-500 mg/kg) exhibited less expression of these markers, when in group B, the manifestation of these enzymes was prodigious (Fig 4). In our current experiment, we observed that the MSE-500 extract had a magnificent effect to protect liver deposition of fats and scarring after 28 days of treatment in group E. In contrast, groups B, C, D, and F were present either fat deposition or scarring (Fig. 1).Fig. 4Comparative expression of liver markers at different conc. of MM extracts in HCD mice. The expression of these markers are mean ± standard error of mean (SEM); where $$n = 5$$; aP < 0.001; compared with normal control; dP < 0.001; eP < 0.01; compared with hyperlipidemic control In the circulatory system, the plethora of LDL cholesterol reaches sub-endothelial space through the surface adhesion molecules due to endothelial dysfunction and becomes so tendentious to oxidize LDL. Subsequently, the macrophage is activated to engulf the oxidized LDL and release pro-inflammatory cytokines, which engenders an inflammatory environment that contributes to plaque formation in the blood vessel for the impairment of blood flow [35, 36]. That’s why the presence of much LDL in the bloodstream is baneful for the body’s vascular function. Our current study has manifested the anti-hyperlipidemic efficiency of orally administrated different conc. of MM plant extract such as MFA-500 mg/kg exhibited trivial potency ($4\%$, $3\%$, $5\%$ and $28\%$) to subtract the lipid molecules (TG, TC, VLDL, LDL) respectively, but MSE-250 mg/kg lessened more ($28\%$, $30\%$, $29\%$ and $47\%$), whereas the MSE-500 mg/kg tested as alternative drug, had declined tremendously ($41\%$, $44\%$, $43\%$ and $74\%$) comparing with drug control (Figs. 5 and 6). Concomitantly, HDL level augmented ($30\%$, $60\%$ and $70\%$) using MFA-500 mg/kg, MSE-250 mg/kg and ATOVAT-20 mg/kg but administered MSE-500 mg/kg endorsed more to flush HLD up to $120\%$ (Table 2 and Fig. 5). Afterward, the histopathology analysis elicited that the MSE-500 mg/kg extract could effectively restrain fats deposition and keep normal cellular infrastructure in the heart (Fig. 2).Fig. 5Comparative expression of TG, TC and HDL at different conc. of MM extracts in HCD mice. The expression of these markers are mean ± standard error of mean (SEM); where $$n = 5$$; aP < 0.001; compared with normal control; dP < 0.001; eP < 0.01; fP < 0.05; hnp > 0.05 compared with hyperlipidemic controlsFig. 6Comparative expression of VLDL and LDL of MM extracts at different conc. in HCD mice. The expression of these markers are mean ± standard error of mean (SEM); where $$n = 5$$; aP < 0.001; compared with normal control; dP < 0.001; eP < 0.01; compared with hyperlipidemic control The kidney commonly excretes xenobiotics after metabolic biotransformation even though atorvastatin excretion is almost all biliary, less than $2\%$ eliminated through the kidney [37]. Phytochemicals are xenobiotics; that’s why both of them should have a significant impact on kidneys that had been manifested by kidney histopathology for all groups after 28 days of treatment. Our recent study revealed that in groups C and D using MFA-500 mg/kg and MSE-250 mg/kg, the kidney condition was least aggravated, such as inflammatory blood cells and deposition of fats compared with group B (Fig. 3). Whereas in groups E and F using MSE-500 mg/kg and ATOVAT-20 mg/kg, the presence of inflammatory blood vessels and fat deposition around the glomerulus disappeared, almost congruous with normal kidneys’ phenotype. Alongside the effect of MM extracts in blood cholesterols, livers, kidneys, and hearts in hyperlipidemic mice, the antibacterial activity was also be checked against some gram-negative pathogenic bacteria using four concentrations of MSE and MSW. The maximum zone of inhibition of MSE extract was 17 mm against S. dysenteriae, 14 mm against S. flexneri, and 12 mm against S. sonnei. Moreover, the maximum zone of inhibition of MSW extract was 10 mm against only S. dysenteriae (Table 4). In contrast, ETEC and EPEC strain demonstrated paltry potency with all concentrations of these two extracts. ## Conclusion In our current inquisition, the MSE-250 mg/kg dose demonstrated significant results in all cases. However, treatment with MSE-500 mg/kg was appealing and exposed its anti-hyperlipidemic potency by increasing HDL level. Moreover, this dose was liver and kidney-friendly by protecting inflammation and fat deposition, almost comparable to atorvastatin. But the stunning matter is that after 28 days of treatment, group F noticed a glimpse of liver scarring but didn’t exhibit anything like this in group E, which hints at using MSE-500 mg/kg as an alternative drug during the hyperlipidemic condition. On the contrary, the MFA-500 mg/kg didn’t exhibit any anti-hyperlipidemic potency after 28 days of treatment but showed a slightly amicable relationship with the liver. Alongside, the MSE extract at 1500 μg/disc exhibited a maximum zone of inhibition against most of the pathogenic strain than the MSW extract during the current experiment. ## Limitation The phytochemical screening and antioxidant properties of MM extracts didn’t carry out our fundamental research that would make it stronger and precise to demonstrate why extracts are friendly for liver, kidney, heart and which phytochemicals are responsible for antibacterial. ## References 1. Klop B, Elte JWF, Cabezas MC. Dyslipidemia in obesity: mechanisms and potential targets. Nutrients. (2013.0) 5 1218-1240. DOI: 10.3390/nu5041218 2. McKenzie HC. Equine Hyperlipidemias. Vet Clin North Am - Equine Pract (2011.0) 27 59-72. DOI: 10.1016/j.cveq.2010.12.008 3. Nelson RH. Hyperlipidemia as a risk factor for cardiovascular disease. Prim Care - Clin Off Pract (2013.0) 40 195-211. DOI: 10.1016/j.pop.2012.11.003 4. Artham SM, Lavie CJ, Milani RV, Ventura HO. The obesity paradox: impact of obesity on the prevalence and prognosis of cardiovascular diseases. Postgrad Med (2008.0) 120 34-41. DOI: 10.3810/pgm.2008.07.1788 5. O’Brien T, Nguyen TT, Zimmerman BR. Hyperlipidemia and diabetes mellitus. Mayo Clin Proc (1998.0) 73 969-976. DOI: 10.4065/73.10.969 6. Assy N, Kaita K, Mymin D, Levy C, Rosser B, Minuk G. Fatty infiltration of liver in hyperlipidemic patients. Dig Dis Sci (2000.0) 45 1929-1934. DOI: 10.1023/A:1005661516165 7. Sahadevan M, Kasiske BL. Hyperlipidemia in kidney disease: causes and consequences. Curr Opin Nephrol Hypertens (2002.0) 11 323-329. DOI: 10.1097/00041552-200205000-00009 8. Dhakal BP, Sweitzer NK, Indik JH, Acharya D, William P. SARS-CoV-2 infection and cardiovascular disease: COVID-19 heart. Hear Lung Circ (2020.0) 29 973-987. DOI: 10.1016/j.hlc.2020.05.101 9. Zhao Y, Xiang L, Liu Y, Niu M, Yuan J, Chen H. Atherosclerosis induced by a high-cholesterol and high-fat diet in the inbred strain of the Wuzhishan miniature pig. Anim Biotechnol (2018.0) 29 110-118. DOI: 10.1080/10495398.2017.1322974 10. Ridker PM. Hyperlipidemia as an instigator of inflammation: inaugurating new approaches to vascular prevention. J Am Heart Assoc (2012.0) 1 8-11. DOI: 10.1161/JAHA.112.000497 11. Landmesser U, Drexler H. Toward understanding of extracellular superoxide dismutase regulation in atherosclerosis: a novel role of uric acid?. Arterioscler Thromb Vasc Biol (2002.0) 22 1367-1368. DOI: 10.1161/01.ATV.0000027430.99956.4C 12. Singh U, Devaraj S, Jialal I, Vitamin E. Oxidative stress, and inflammation. Annu Rev Nutr (2005.0) 25 151-174. DOI: 10.1146/annurev.nutr.24.012003.132446 13. Vasanthi HR, ShriShriMal N, Das DK. Retraction notice: phytochemicals from plants to combat cardiovascular disease. Curr Med Chem (2012.0) 19 2242-2251. DOI: 10.2174/092986712800229078 14. Islam MD, Islam A, Tasnin N, Akter SF, Uddin MS. Exploration of antioxidant and anticancer activity of Stephania japonica leaves extract. J Pharm Res Int (2021.0) 33 256-272. DOI: 10.9734/jpri/2021/v33i28A31529 15. Barreca D, Gattuso G, Bellocco E, Calderaro A, Trombetta D, Smeriglio A. Flavanones: Citrus phytochemical with health-promoting properties. BioFactors (2017.0) 43 495-506. DOI: 10.1002/biof.1363 16. Ma L, Sun Z, Zeng Y, Luo M, Yang J. Molecular mechanism and health role of functional ingredients in blueberry for chronic disease in human beings. Int J Mol Sci (2018.0) 19 2785. DOI: 10.3390/ijms19092785 17. Leitzmann C. Characteristics and health benefits of phytochemicals. Forsch Komplementarmed (2016.0) 23 69-74 18. Kumar V, Ahmed D, Gupta PS, Anwar F, Mujeeb M. Anti-diabetic, anti-oxidant and anti-hyperlipidemic activities of Melastoma malabathricum Linn. Leaves in streptozotocin induced diabetic rats. BMC Complement Altern Med (2013.0) 13 1-19. DOI: 10.1186/1472-6882-13-222 19. Joffry SM, Yob NJ, Rofiee MS, MMRMM A, Suhaili Z, Othman F. Melastoma malabathricum ( L .) Smith Ethnomedicinal Uses , Chemical Constituents , and Pharmacological Properties : A Review. Evid Based Complement Alternat Med (2012.0) 2012 6-20. DOI: 10.1155/2012/258434 20. Sampaio BL, Edrada-Ebel R, Da Costa FB. Effect of the environment on the secondary metabolic profile of Tithonia diversifolia: a model for environmental metabolomics of plants. Sci Rep (2016.0) 6 1-11. DOI: 10.1038/srep29265 21. Shyamala MP, Venukumar MR, Latha MS. Antioxidant potential of the Syzygium aromaticum (Gaertn.) Linn.(cloves) in rats fed with high fat diet. Indian J Pharmacol (2003.0) 35 99-103 22. Werner M, Chott A, Fabiano A, Battifora H. Effect of formalin tissue fixation and processing on immunohistochemistry. Am J Surg Pathol (2000.0) 24 1016-1019. DOI: 10.1097/00000478-200007000-00014 23. 23.Friedewald WT, DSF RIL. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifugeitle. Clin chem. 18(6):499–502. 24. 24.Fischer AH, Jacobson KA, Rose J, Zeller R. Hematoxylin and eosin staining of tissue and cell sections. Cold Spring Harb Protoc. 2008;3(5):3–5.http://www.ihcworld.com/index.htm. 25. Khalil IA, Troeger C, Blacker BF, Rao PC, Brown A, Atherly DE. Morbidity and mortality due to shigella and enterotoxigenic Escherichia coli diarrhoea: the global burden of disease study 1990–2016. Lancet Infect Dis (2018.0) 18 1229-1240. DOI: 10.1016/S1473-3099(18)30475-4 26. Taniuchi M, Islam K, Sayeed MA, Platts-Mills JA, Islam MT, Khabir MIU. Etiology of diarrhea requiring hospitalization in Bangladesh by quantitative polymerase chain reaction, 2014-2018. Clin Infect Dis (2021.0) 73 E2493-9. DOI: 10.1093/cid/ciaa840 27. Alnajar ZAA, Abdulla MA, Ali HM, Alshawsh MA, Hadi AHA. Acute toxicity evaluation, antibacterial, antioxidant and immunomodulatory effects of Melastoma malabathricum. Molecules (2012.0) 17 3547-3559. DOI: 10.3390/molecules17033547 28. Gu X, Manautou JE. Molecular mechanisms underlying chemical liver injury. Expert Rev Mol Med (2012.0) 14 1-21 29. Kingsley UI, Steven OO, Agu CE, Orji OC, Chekwube BE, Nwosu TF. Anti-hyperlipidemic effect of crude methanolic extracts of Glycine max (soy bean) on high cholesterol diet-fed albino rats. Med Allied Sci (2017.0) 7 34-40 30. 30.Nwangwa EK, Ekhoye EI. Anti-Hyperlipidemic Activity of Aqueous Extract of Carica Papaya Seed in Albino Rats fed with High Fat Diet. Curr Trends Technol Sci. 2013;II(III):2279–0535. 31. Duntas LH. Thyroid disease and lipids. Thyroid (2002.0) 12 287-293. DOI: 10.1089/10507250252949405 32. Pucci E, Chiovato L, Pinchera A. Thyroid and lipid metabolism. Int J Obes (2000.0) 24 S109-S112. DOI: 10.1038/sj.ijo.0801292 33. Laurila A, Cole SP, Merat S, Obonyo M, Palinski W, Fierer J. High-fat, high-cholesterol diet increases the incidence of gastritis in LDL receptor-negative mice. Arterioscler Thromb Vasc Biol (2001.0) 21 991-996. DOI: 10.1161/01.ATV.21.6.991 34. Braunersreuther V, Viviani GL, Mach F, Montecucco F. Role of cytokines and chemokines in non-alcoholic fatty liver disease. World J Gastroenterol (2012.0) 18 727-735. DOI: 10.3748/wjg.v18.i8.727 35. Wang JC, Bennett M. Aging and atherosclerosis: mechanisms, functional consequences, and potential therapeutics for cellular senescence. Circ Res (2012.0) 111 245-259. DOI: 10.1161/CIRCRESAHA.111.261388 36. Frostegard J, Wu R, Giscombe R, Holm G, Lefvert AK, Nilsson J. Induction of T-cell activation by oxidized low density lipoprotein. Arterioscler Thromb (1992.0) 12 461-467. DOI: 10.1161/01.ATV.12.4.461 37. Lennernäs H. Clinical pharmacokinetics of atorvastatin. Clin Pharmacokinet (2003.0) 42 1141-1160. DOI: 10.2165/00003088-200342130-00005
--- title: Assessment of Sex Disparities in Nonacceptance of Statin Therapy and Low-Density Lipoprotein Cholesterol Levels Among Patients at High Cardiovascular Risk authors: - C. Justin Brown - Lee-Shing Chang - Naoshi Hosomura - Shervin Malmasi - Fritha Morrison - Maria Shubina - Zhou Lan - Alexander Turchin journal: JAMA Network Open year: 2023 pmcid: PMC9975905 doi: 10.1001/jamanetworkopen.2023.1047 license: CC BY 4.0 --- # Assessment of Sex Disparities in Nonacceptance of Statin Therapy and Low-Density Lipoprotein Cholesterol Levels Among Patients at High Cardiovascular Risk ## Key Points ### Question How are sex disparities in nonacceptance of statin therapy associated with control of low-density lipoprotein (LDL) cholesterol levels? ### Findings In this cohort study of 24 212 adults at high cardiovascular risk, patients who accepted a statin therapy recommendation by their clinicians achieved an LDL cholesterol level of less than 100 mg/dL in a median time of 1.5 years vs 4.4 years for patients who did not accept statin therapy. Women were significantly less likely than men to accept statin therapy recommendations and achieve an LDL cholesterol level of less than 100 mg/dL. ### Meaning This study suggests that patients who do not accept statin therapy have significantly higher LDL cholesterol levels; sex disparities in statin acceptance could be associated with cardiovascular risk for women. ## Abstract This cohort study assesses sex disparities in nonacceptance of statin therapy and assess their association with low-density lipoprotein (LDL) cholesterol control. ### Importance Many patients at high cardiovascular risk—women more commonly than men—are not receiving statins. Anecdotally, it is common for patients to not accept statin therapy recommendations by their clinicians. However, population-based data on nonacceptance of statin therapy by patients are lacking. ### Objectives To evaluate sex disparities in nonacceptance of statin therapy and assess their association with low-density lipoprotein (LDL) cholesterol control. ### Design, Setting, and Participants A retrospective cohort study was conducted from January 1, 2019, to December 31, 2022, of statin-naive patients with atherosclerotic cardiovascular disease, diabetes, or LDL cholesterol levels of 190 mg/dL (to convert to millimoles per liter, multiply by 0.0259) or more who were treated at Mass General Brigham between January 1, 2000, and December 31, 2018. ### Exposure Recommendation of statin therapy by the patient’s clinician, ascertained from the combination of electronic health record prescription data and natural language processing of electronic clinician notes. ### Main Outcomes and Measures Time to achieve an LDL cholesterol level of less than 100 mg/dL. ### Results Of 24 212 study patients (mean [SD] age, 58.8 [13.0] years; 12 294 women [$50.8\%$]), 5308 ($21.9\%$) did not accept the initial recommendation of statin therapy. Nonacceptance of statin therapy was more common among women than men ($24.1\%$ [2957 of 12 294] vs $19.7\%$ [2351 of 11 918]; $P \leq .001$) and was similarly higher in every subgroup in the analysis stratified by comorbidities. In multivariable analysis, female sex was associated with lower odds of statin therapy acceptance (0.82 [$95\%$ CI, 0.78-0.88]). Patients who did vs did not accept a statin therapy recommendation achieved an LDL cholesterol level of less than 100 mg/dL over a median of 1.5 years (IQR, 0.4-5.5 years) vs 4.4 years (IQR, 1.3-11.1 years) ($P \leq .001$). In a multivariable analysis adjusted for demographic characteristics and comorbidities, nonacceptance of statin therapy was associated with a longer time to achieve an LDL cholesterol level of less than 100 mg/dL (hazard ratio, 0.57 [$95\%$ CI, 0.55-0.60]). ### Conclusions and Relevance This cohort study suggests that nonacceptance of a statin therapy recommendation was common among patients at high cardiovascular risk and was particularly common among women. It was associated with significantly higher LDL cholesterol levels, potentially increasing the risk for cardiovascular events. Further research is needed to understand the reasons for nonacceptance of statin therapy by patients and to develop methods to ensure that all patients receive optimal therapy in accordance with their preferences and priorities. ## Introduction The benefits of statin therapy for patients at high cardiovascular risk are well established.1,2,3 However, many of these individuals are not being treated with statins.4,5,6 This lack of statin therapy is particularly pronounced among women.7,8,9,10,11 The reasons for the lack of statin therapy among patients at high cardiovascular risk and for the persistent sex disparity are not fully understood and are likely multifactorial.12,13,14,15 One possible reason for the lack of recommended therapy that has recently come to light is nonacceptance of clinicians’ treatment recommendations by patients.16,17 Anecdotal evidence and patient surveys suggest that this phenomenon is also associated with the lack of indicated statin therapy,15 but there is a dearth of population-based studies on the subject. The extent to which nonacceptance of statins by patients is associated with the sex disparities in statin therapy is unknown. A major reason for the paucity of research in this area is that information on nonacceptance of statin therapy recommendations by patients is not easily available. This information is typically not reflected in administrative data or structured data in the electronic health record (EHR). Instead, nonacceptance of statin therapy by patients is recorded primarily in narrative notes, requiring labor-intensive medical record reviews to study them. Over the last decade, however, technology for computational analysis of narrative electronic documents—natural language processing (NLP)—has become available, providing a powerful tool to investigate patient care processes that are documented only in narrative documents.18,19,20 This technology was used in the present study to test the hypothesis that nonacceptance by patients of a statin therapy recommendation is significantly associated with a lack of statin treatment among patients at high cardiovascular risk and specifically associated with sex disparities in statin therapy. ## Methods This study was approved by the Mass General Brigham institutional review board, and the requirement for informed consent was waived because of the low risk of adverse effects to study participants. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. ## Study Design This retrospective cohort study was conducted from January 1, 2019, to December 31, 2022, among patients at high cardiovascular risk with elevated low-density lipoprotein (LDL) cholesterol levels. The study investigated the association between a patient’s decision to accept a statin therapy recommendation and the time to achieve an LDL cholesterol level of less than 100 mg/dL (to convert to millimoles per liter, multiply by 0.0259) as well as patient characteristics associated with acceptance of statin therapy. ## Study Cohort Study patients included adults at high cardiovascular risk treated in practices affiliated with Mass General Brigham (a large integrated health care delivery network in Massachusetts founded by Brigham and Women’s Hospital and Massachusetts General Hospital) between January 1, 2000, and December 31, 2018. Patients were included if they fulfilled all of the following inclusion criteria: [1] older than 18 years; [2] diagnosis of atherosclerotic cardiovascular disease (ASCVD), diabetes, or an LDL cholesterol level of 190 mg/dL or more; [3] at least 2 primary care encounters prior to study entry; [4] no record of statin therapy prior to study entry; [5] an LDL cholesterol level of 100 mg/dL or more prior to study entry; and [6] at least 1 LDL cholesterol measurement obtained 12 months or more after study entry. Patients were excluded from analysis if they had missing demographic information or had a record of an adverse reaction to a statin prior to the first record of statin therapy. A total of 25 197 statin-naive adults at high cardiovascular risk followed up in primary care settings, with both elevated baseline LDL cholesterol and follow-up LDL cholesterol measurements recorded, were identified. After excluding patients who [1] had a record of an adverse effect to a statin prior to the first record of statin therapy or [2] were missing data on age, sex, or median income by zip code, 24 212 patients (mean [SD] age, 58.8 [13.0] years; 12 294 women [$50.8\%$]) were included in the study (Table 1; eFigure 1 in Supplement 1). A total of 4 575 430 EHR notes were analyzed with NLP to identify instances of statin nonacceptance by patients and their subsequent dates of study entry. Table 1. \| Characteristic \| Patients, No. (%) \| Patients, No. (%).1 \| P value \| \| --- \| --- \| --- \| --- \| \| Characteristic \| Female (n = 12 294) \| Male (n = 11 918) \| P value \| \| Age, mean (SD), y \| 60.7 (12.9) \| 56.8 (13.0) \| <.001 \| \| Study entry year, mean (SD)a \| 9.4 (4.8) \| 9.7 (4.7) \| <.001 \| \| Baseline LDL cholesterol, mean (SD), mg/dL \| 149.0 (37.9) \| 143.3 (35.5) \| <.001 \| \| Income (in the $10 000s), mean (SD)b \| 6.9 (2.5) \| 7.2 (2.7) \| <.001 \| \| Charlson Comorbidity Index, mean (SD) \| 5.2 (3.8) \| 4.6 (4.0) \| <.001 \| \| No. of drug allergies, mean (SD) \| 2.8 (3.4) \| 1.3 (1.9) \| <.001 \| \| Days of follow-up, mean (SD) \| 2930.3 (1668.7) \| 2793.6 (1632.5) \| <.001 \| \| English speaking \| 10 321 (84.0) \| 10 480 (87.9) \| <.001 \| \| Married \| 5269 (42.9) \| 7435 (62.4) \| <.001 \| \| Government insurance \| 5557 (45.2) \| 4392 (36.9) \| <.001 \| \| Smoker \| 4195 (34.1) \| 4523 (38.0) \| <.001 \| \| Ezetimibe use \| 267 (2.2) \| 279 (2.3) \| .38 \| \| CAD \| 3989 (32.5) \| 4770 (40.0) \| <.001 \| \| CVA \| 1915 (15.6) \| 1667 (14.0) \| <.001 \| \| PVD \| 1796 (14.6) \| 1590 (13.3) \| .004 \| \| Diabetes \| 6206 (50.5) \| 5793 (48.6) \| .004 \| \| LDL cholesterol ≥190 mg/dL \| 4216 (34.3) \| 3340 (28.0) \| <.001 \| \| Family history of diabetes \| 3230 (26.3) \| 2637 (22.1) \| <.001 \| \| Family history of CVD \| 2446 (19.9) \| 1963 (16.5) \| <.001 \| \| Race and ethnicity \| \| \| \| \| Asian \| 539 (4.4) \| 499 (4.2) \| <.001 \| \| Black \| 1163 (9.5) \| 905 (7.6) \| <.001 \| \| Hispanic \| 770 (6.3) \| 675 (5.7) \| <.001 \| \| White \| 8758 (71.2) \| 8907 (74.7) \| <.001 \| \| Otherc \| 1064 (8.7) \| 932 (7.8) \| <.001 \| A total of 11 667 patients ($48.2\%$) had ASCVD, and the rest had either diabetes or severe hypercholesterolemia; 5584 patients ($23.1\%$) had multiple indications for statin therapy. Among study patients, 5308 ($21.9\%$) initially did not accept statin therapy, and 1457 ($6.0\%$) never initiated a statin during the follow-up period. Women were more likely than men to both not accept the initial statin therapy recommendation ($24.1\%$ [2957 of 12 294] vs $19.7\%$ [2351 of 11 918]; $P \leq .001$) and never initiate a statin during the study ($7.2\%$ [881 of 12 294] vs $4.8\%$ [576 of 11 918]; $P \leq .001$). Similar findings were observed in every subgroup in the analysis stratified by comorbidities (Figure 1). Ezetimibe use was uncommon among patients who did not accept statin therapy and was similar among women (44 of 2957 [$1.5\%$]) and men (34 of 2351 [$1.4\%$]) ($P \leq .99$). The median baseline LDL cholesterol level was 137 mg/dL. Over the mean (SD) follow-up of 7.9 (4.5) years, 18 796 patients ($77.6\%$) reached an LDL cholesterol level of less than 100 mg/dL after a median of 1.9 years (IQR, 0.5-6.8 years). Figure 1.: Sex Differences in Statin AcceptanceError bars indicate standard error. ASCVD indicates atherosclerotic cardiovascular disease; CAD, coronary artery disease; CVA, cerebrovascular accident; LDL, low-density lipoprotein; and PVD, peripheral vascular disease. ## Exposures A patient was entered into the study on the first date that statin therapy was recommended by a health care professional. This was defined as the earliest of [1] the first record of statin therapy (indicating initial acceptance of statin therapy by the patient) or [2] the first documented nonacceptance of statin therapy by the patient. Patients exited the study on the first of the following: [1] the date of death; [2] loss to follow-up as indicated by the absence of documentation from primary care, cardiology, or endocrinology clinicians for greater than 12 months; or [3] December 31, 2018. ## Outcome Measures Acceptance or nonacceptance by the patient of a statin therapy recommendation by the health care professional served as the primary independent variable. Acceptance of statin therapy was indicated by a statin medication record in structured EHR data. Nonacceptance of statin therapy by the patient was ascertained from clinician documentation in narrative EHR notes using NLP. Time to LDL cholesterol control—time from study entry to the first measurement of an LDL cholesterol level of less than 100 mg/dL—served as the primary outcome. Achievement of an LDL cholesterol level of less than 100 mg/dL within 12 months of study entry was analyzed as a secondary outcome. ## Data Acquisition Information on baseline patient characteristics and study outcomes was obtained from the EHR at Mass General Brigham. Information on participant race and ethnicity was based on self-report and was also obtained from the EHR at Mass General Brigham. The following race and ethnicity categories were obtained: Asian, Black, Hispanic, White, and other. The “other” category included American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, and unknown. Natural language processing tools to identify documented nonacceptance of statin therapy by patients were developed using the publicly available Canary platform, version 2.01.21,22 Canary allows users to create customized NLP tools by first defining groups of related words (word classes) that represent subconcepts of interest (eg, statins) and then defining how these word classes can come together (eg, nonacceptance of statin therapy) to describe the concept being sought. They were validated against manual annotations of 3999 randomly selected clinician notes by 2 independent reviewers whose ratings were subsequently reconciled. Additional details of development and validation of the NLP tool can be found in eFigure 2 and eAppendices 1-3 in Supplement 1. To identify patients who did not accept statin therapy, all EHR notes from clinicians in all specialties written between the date of the first eligible diagnosis and the earliest date of the first record of statin therapy, the first record of statin intolerance, or the date of study exit were analyzed. ## Statistical Analysis An individual patient served as the unit of analysis. Univariate analyses were conducted using the t test for continuous variables and the χ2 test for categorical variables. The log-rank test was used to compare time to LDL cholesterol control between patients who initially accepted statin therapy vs patients who did not accept statin therapy. A marginal Cox proportional hazards regression model was used to estimate the association between statin therapy acceptance vs nonacceptance and time to LDL cholesterol control while accounting for clustering within individual clinicians and adjusting for demographic confounders (age, sex, race and ethnicity, primary language, marital status, health insurance, and median income by zip code), the Charlson Comorbidity Index, indication for statin therapy (ASCVD, diabetes, or LDL cholesterol ≥190 mg/dL), family history of ASCVD or diabetes, ezetimibe use, year of the statin therapy recommendation, and the baseline LDL cholesterol level. We also conducted a sensitivity analysis (results in eAppendix 4 and eFigure 3 in Supplement 1) of how possible errors of the NLP algorithm could affect the results of this analysis. To study the factors associated with statin therapy acceptance, a multivariable logistic regression model was constructed to estimate the association between statin therapy acceptance and patient characteristics while accounting for clustering within individual clinicians. Interpretation of P values was adjusted for multiple hypothesis testing using the Simes-Hochberg method.23,24 Analyses were performed using SAS, version 9.4 (SAS Institute Inc) and R, version 4.2.0 (R Group for Statistical Computing) for Monte Carlo simulations. All P values were from 2-sided tests and results were deemed statistically significant at $P \leq .05.$ ## Natural Language Processing We first sought to construct and evaluate the effectiveness of the NLP algorithm to identify patients not accepting of statin therapy. In the course of the evaluation of the NLP algorithm, of the 3999 notes in the validation set, 40 contained documentation of a patient not accepting a statin. The Cohen κ coefficient for interrater agreement between the 2 annotators was 0.872 ($95\%$ CI, 0.800-0.945). The NLP tool used in the analysis achieved a sensitivity of $87.5\%$ ($95\%$ CI, $73.2\%$-$95.8\%$), a specificity of $99.8\%$ ($95\%$ CI, $99.5\%$-$99.9\%$), and a positive predictive value of $77.8\%$ ($95\%$ CI, $65.1\%$-$86.8\%$). ## Nonacceptance of Statin Therapy and LDL Cholesterol Control Patients who accepted a statin therapy recommendation achieved an LDL cholesterol level of less than 100 mg/dL after a median of 1.5 years (IQR, 0.4-5.5 years) compared with 4.4 years (IQR, 1.3-11.1 years) (Figure 2) for patients who did not ($P \leq .001$). In a multivariable analysis adjusted for patients’ demographic characteristics and comorbidities, nonacceptance of statin therapy was associated with longer time to achieve LDL cholesterol control (hazard ratio [HR], 0.57 [$95\%$ CI, 0.55-0.60]; $P \leq .001$) (Table 2). Female sex was an independent risk factor associated with a longer time to achieve LDL cholesterol control (HR, 0.84 [$95\%$ CI, 0.81-0.87]; $P \leq .001$). A higher baseline LDL cholesterol level was also associated with a longer time to achieve LDL cholesterol control, while a higher Charlson Comorbidity Index, a history of diabetes or stroke, and a later study entry year were associated with a shorter time to achieve LDL cholesterol control. Figure 2.: Statin Acceptance and Time to Low-Density Lipoprotein (LDL) Cholesterol ControlThe last $5\%$ of the population in each group were not included in the plot due to low participant numbers and resulting wide $95\%$ CIs. To convert LDL cholesterol to millimoles per liter, multiply by 0.0259. TABLE_PLACEHOLDER:Table 2. In a secondary analysis, $42.1\%$ (7955 of 18 904) of patients who did accept statin therapy vs $21.0\%$ (1114 of 5308) of patients who did not accept statin therapy achieved an LDL cholesterol level of less than 100 mg/dL within 12 months ($P \leq .001$). In multivariable analysis, nonacceptance of statin therapy was associated with an odds ratio (OR) for achieving LDL cholesterol control within 12 months of 0.32 ($95\%$ CI, 0.30-0.35) ($P \leq .001$); findings of a secondary analysis using LDL decrease by $50\%$ as the outcomes were similar (eAppendix 5 in Supplement 1). Women (independently from other characteristics) were less likely to achieve LDL cholesterol control within 12 months (OR, 0.85 [$95\%$ CI, 0.80-0.90]; $P \leq .001$). ## Factors Associated With Statin Therapy Acceptance In multivariable analysis (Figure 3), women were less likely to agree to take a statin when first recommended by a clinician (OR, 0.82 [$95\%$ CI, 0.78-0.88]; $P \leq .001$). Presence of cardiovascular disease, higher baseline LDL cholesterol, and history of smoking were associated with a greater probability of statin acceptance. All of these associations were statistically significant after adjustment for multiple hypothesis testing. Figure 3.: Patient Characteristics and Acceptance of Statin Therapy RecommendationA multivariable logistic regression model that included all variables in the figure was constructed to estimate the association between statin therapy acceptance and patient characteristics while accounting for clustering within individual clinicians. To convert low-density lipoprotein (LDL) cholesterol to millimoles per liter, multiply by 0.0259. CCI indicates Charlson Comorbidity Index; CAD, coronary artery disease; CVA, cerebrovascular accident; CVD, cardiovascular disease; OR, odds ratio; and PVD, peripheral vascular disease. ## Discussion In this large, population-based cohort study, we found that many patients with unequivocal indications for cholesterol lowering did not accept statin therapy recommended to them by their health care professionals. These patients were subsequently less likely to achieve LDL cholesterol control within 1 year and took significantly longer to reach it compared with patients who initiated statins. Women had lower rates of statin acceptance than men, potentially contributing to the known sex disparities in LDL cholesterol control. These findings shed light on an important and previously unexamined aspect of prevention of cardiovascular disease. Nonacceptance of statin therapy recommendations is in many ways distinct from the more widely studied phenomenon of statin nonadherence. Unlike medication nonadherence, which is often due to high medication costs or adverse reactions the patient developed,25,26,27 nonacceptance of treatment recommendations takes place before the patient has had any direct experience with the medication; the reasons for it are therefore likely to be different. Many measures of the quality of cardiovascular care may interpret patient nonacceptance of statin therapy as “nonprescriptions” due to lack of appropriate action by the health care professional.28,29 This study demonstrates that patients are active agents in their care, and their preferences and priorities should be carefully taken into account when making treatment recommendations.30,31 Debates continue about indications for statin therapy for patients with low cardiovascular risk or for the primary prevention population.32,33,34 However, the present study identified high rates of nonacceptance of statin therapy among patients at high cardiovascular risk: those with existing ASCVD, diabetes, or an LDL cholesterol level of 190 mg/dL or more. These are vulnerable individuals for whom evidence-based cholesterol-lowering therapy could significantly lower the incidence of cardiovascular events and related morbidity and mortality. Therefore, the findings of this study have significant implications for public health as we continue to strive to decrease the risks of ASCVD—the number one cause of death in the US and worldwide.35,36 Acceptance of a statin therapy recommendation was found to be associated with achievement of LDL cholesterol control. This association is not as self-evident as it may seem because patients’ cholesterol levels may have been associated with many other factors. On the one hand, a patient’s initial acceptance of statin therapy does not necessarily guarantee continuous use of the statin, or even any use at all, because patients can often be nonadherent to statins.37,38 On the other hand, a patient’s initial nonacceptance does not necessarily indicate that the patient will never take a statin—nearly two-thirds of patients who initially did not accept a statin did eventually start taking statins, which may have occurred shortly after their initial nonacceptance. Finally, both groups of patients may have also elected to use other nonstatin cholesterol-lowering medications, such as ezetimibe, or implemented lifestyle changes that may lead to lower cholesterol levels. However, the association of statin nonacceptance with LDL cholesterol levels was marked despite these potential considerations. The present study identified significant sex disparities in the acceptance of statin therapy. Women were more than $20\%$ more likely than men not to accept their clinician’s initial statin therapy recommendation; similar findings were observed among patients with known coronary artery disease and all other comorbidity subgroups. This disparity increased as time went on; over the entire course of the study, women were $50\%$ more likely to never initiate statins. Multiple previous studies have reported lower rates of cholesterol control among women compared with men.39,40,41 These differences have been explained in part by sex disparities in the rates of adherence and adverse effects to statins.13,42,43,44 The present study suggests that disparities in nonacceptance of a statin therapy recommendation are another important factor; further research is needed to assess why women at high cardiovascular risk are less likely to accept their clinicians’ recommendation of statin therapy. A distinct aspect of this study was the use of novel NLP technology to identify patients who did not accept a statin therapy recommendation by their clinician. Nonacceptance of statin therapy by patients is typically documented only in narrative clinician notes because EHR systems do not usually have checkboxes, drop-down lists, or other structured data elements where it could be recorded. Data analysis for the study involved over 4 million electronic clinician notes, rendering traditional manual medical record review infeasible. On the other hand, the validated NLP tool developed for the purpose of this study processed this massive amount of data within days. This study therefore highlights the potential for the use of artificial intelligence technology in combination with vast data sets to make novel research questions accessible for investigation for the first time. ## Strengths and Limitations The present study has several strengths. To our knowledge, this is the first population-based study that examined nonacceptance of statin therapy recommendations by patients. It included a large population of patients treated in primary care settings, similar to most patients with hypercholesterolemia in the US. The study focused on patients with unequivocal indications for statin therapy, making its findings critical for the prevention of cardiovascular events among this high-risk population. Finally, the use of artificial intelligence NLP technology allowed for a unique viewpoint into a previously minimally explored aspect of the treatment of hypercholesterolemia. Study findings should also be interpreted in light of its limitations. Its observational nature does not allow for the identification of causal relationships. Some of the statin nonacceptance may not have been documented in clinician notes, and other instances may not have been detected by the NLP algorithm. Validation of the NLP algorithm was not stratified by biological sex, and it is possible that the accuracy of the algorithm is different for female vs male patients. Only the first statin therapy recommendation for each patient was examined; future studies should address whether similar findings hold for multiple recommendations. Specific reasons for statin nonacceptance, discussion of lifestyle changes, and association of clinician characteristics were not examined and should be investigated in future research. Although a broad selection of potential confounding variables was included in the multivariable analysis, other unknown confounders not accounted for may have influenced the association between statin acceptance and LDL cholesterol control. In addition, because our study cohort included only patients who were known to have obtained an LDL cholesterol level measurement after a statin recommendation, it may be biased by the exclusion of patients who did not return for follow-up care. Finally, the study population is composed of encounters within a single-center health care system; the patient population therefore may not be generalizable to other geographic or demographic cohorts. ## Conclusions In this cohort study, nonacceptance of a statin therapy recommendation was common among patients at high cardiovascular risk and was associated with higher LDL cholesterol levels, potentially translating into an increased incidence of cardiovascular events. Nonacceptance of statins was particularly prevalent among women, possibly contributing to the known sex disparities in treatment of high cholesterol. Further research is needed to identify the reasons why patients do not accept statin therapy recommendations and the reasons for the higher rates of this important clinical phenomenon among women. ## References 1. Mills EJ, O’Regan C, Eyawo O. Intensive statin therapy compared with moderate dosing for prevention of cardiovascular events: a meta-analysis of >40 000 patients. Eur Heart J (2011.0) 32 1409-1415. DOI: 10.1093/eurheartj/ehr035 2. Baigent C, Blackwell L, Emberson J. Efficacy and safety of more intensive lowering of LDL cholesterol: a meta-analysis of data from 170,000 participants in 26 randomised trials. Lancet (2010.0) 376 1670-1681. DOI: 10.1016/S0140-6736(10)61350-5 3. LaRosa JC, Grundy SM, Waters DD. Intensive lipid lowering with atorvastatin in patients with stable coronary disease. N Engl J Med (2005.0) 352 1425-1435. DOI: 10.1056/NEJMoa050461 4. Harrison TN, Scott RD, Cheetham TC. Trends in statin use 2009-2015 in a large integrated health system: pre- and post-2013 ACC/AHA guideline on treatment of blood cholesterol. Cardiovasc Drugs Ther (2018.0) 32 397-404. DOI: 10.1007/s10557-018-6810-1 5. Chen G, Farris MS, Cowling T. Treatment and low-density lipoprotein cholesterol management in patients diagnosed with clinical atherosclerotic cardiovascular disease in Alberta. Can J Cardiol (2019.0) 35 884-891. DOI: 10.1016/j.cjca.2019.04.008 6. Reynolds K, Mues KE, Harrison TN. Trends in statin utilization among adults with severe peripheral artery disease including critical limb ischemia in an integrated healthcare delivery system. Vasc Med (2020.0) 25 3-12. DOI: 10.1177/1358863X19871100 7. Gouni-Berthold I, Berthold HK, Mantzoros CS, Böhm M, Krone W. Sex disparities in the treatment and control of cardiovascular risk factors in type 2 diabetes. Diabetes Care (2008.0) 31 1389-1391. DOI: 10.2337/dc08-0194 8. Chen CY, Chuang SY, Fang CC. Gender disparities in optimal lipid control among patients with coronary artery disease. J Atheroscler Thromb (2014.0) 21 S20-S28. DOI: 10.5551/jat.21_Sup.1-S20 9. Driscoll A, Beauchamp A, Lyubomirsky G, Demos L, McNeil J, Tonkin A. Suboptimal management of cardiovascular risk factors in coronary heart disease patients in primary care occurs particularly in women. Intern Med J (2011.0) 41 730-736. DOI: 10.1111/j.1445-5994.2011.02534.x 10. Naicker K, Liddy C, Singh J, Taljaard M, Hogg W. Quality of cardiovascular disease care in Ontario’s primary care practices: a cross sectional study examining differences in guideline adherence by patient sex. BMC Fam Pract (2014.0) 15 123. DOI: 10.1186/1471-2296-15-123 11. Virani SS, Woodard LD, Ramsey DJ. Gender disparities in evidence-based statin therapy in patients with cardiovascular disease. Am J Cardiol (2015.0) 115 21-26. DOI: 10.1016/j.amjcard.2014.09.041 12. Karalis DG, Wild RA, Maki KC. Gender differences in side effects and attitudes regarding statin use in the Understanding Statin Use in America and Gaps in Patient Education (USAGE) study. J Clin Lipidol (2016.0) 10 833-841. DOI: 10.1016/j.jacl.2016.02.016 13. Zhang H, Plutzky J, Shubina M, Turchin A. Drivers of the sex disparity in statin therapy in patients with coronary artery disease: a cohort study. PLoS One (2016.0) 11. DOI: 10.1371/journal.pone.0155228 14. Zhang H, Plutzky J, Shubina M, Turchin A. Risk factors for lack of statin therapy in patients with diabetes and coronary artery disease. J Clin Lipidol (2016.0) 10 1406-1413. DOI: 10.1016/j.jacl.2016.09.010 15. Bradley CK, Wang TY, Li S. Patient-reported reasons for declining or discontinuing statin therapy: insights from the PALM registry. J Am Heart Assoc (2019.0) 8. DOI: 10.1161/JAHA.118.011765 16. Hosomura N, Malmasi S, Timerman D. Decline of insulin therapy and delays in insulin initiation in people with uncontrolled diabetes mellitus. Diabet Med (2017.0) 34 1599-1602. DOI: 10.1111/dme.13454 17. Turchin A, Hosomura N, Zhang H, Malmasi S, Shubina M. Predictors and consequences of declining insulin therapy by individuals with type 2 diabetes. Diabet Med (2020.0) 37 814-821. DOI: 10.1111/dme.14260 18. Meystre SM, Savova GK, Kipper-Schuler KC, Hurdle JF. Extracting information from textual documents in the electronic health record: a review of recent research. Yearb Med Inform (2008.0) 128-144. PMID: 18660887 19. Nadkarni PM, Ohno-Machado L, Chapman WW. Natural language processing: an introduction. J Am Med Inform Assoc (2011.0) 18 544-551. DOI: 10.1136/amiajnl-2011-000464 20. Ohno-Machado L. Realizing the full potential of electronic health records: the role of natural language processing. J Am Med Inform Assoc (2011.0) 18 539. DOI: 10.1136/amiajnl-2011-000501 21. Malmasi S, Sandor NL, Hosomura N, Goldberg M, Skentzos S, Turchin A. Canary: an NLP platform for clinicians and researchers. Appl Clin Inform (2017.0) 8 447-453. DOI: 10.4338/ACI-2017-01-IE-0018 22. 22Canary. Canary version 3.1.1. Accessed January 23, 2023. https://canary.bwh.harvard.edu/ 23. Simes RJ. An improved Bonferroni procedure for multiple tests of significance. Biometrika (1986.0) 73 751-754. DOI: 10.1093/biomet/73.3.751 24. Hochberg Y. A sharper Bonferroni procedure for multiple tests of significance. Biometrika (1988.0) 75 800-802. DOI: 10.1093/biomet/75.4.800 25. Kang H, Lobo JM, Kim S, Sohn MW. Cost-related medication non-adherence among U.S. adults with diabetes. Diabetes Res Clin Pract (2018.0) 143 24-33. DOI: 10.1016/j.diabres.2018.06.016 26. Wei MY, Ito MK, Cohen JD, Brinton EA, Jacobson TA. Predictors of statin adherence, switching, and discontinuation in the USAGE survey: understanding the use of statins in America and gaps in patient education. J Clin Lipidol (2013.0) 7 472-483. DOI: 10.1016/j.jacl.2013.03.001 27. Morrison FJ, Zhang H, Skentzos S, Shubina M, Bentley-Lewis R, Turchin A. Reasons for discontinuation of lipid-lowering medications in patients with chronic kidney disease. Cardiorenal Med (2014.0) 4 225-233. DOI: 10.1159/000368914 28. Sawano M, Kohsaka S, Abe T. Patterns of statin non-prescription in patients with established coronary artery disease: a report from a contemporary multicenter Japanese PCI registry. PLoS One (2017.0) 12. DOI: 10.1371/journal.pone.0182687 29. Canavero I, Cavallini A, Perrone P. Clinical factors associated with statins prescription in acute ischemic stroke patients: findings from the Lombardia Stroke Registry. BMC Neurol (2014.0) 14 53. DOI: 10.1186/1471-2377-14-53 30. Elwyn G, Frosch D, Thomson R. Shared decision making: a model for clinical practice. J Gen Intern Med (2012.0) 27 1361-1367. DOI: 10.1007/s11606-012-2077-6 31. Barry MJ, Edgman-Levitan S. Shared decision making—pinnacle of patient-centered care. N Engl J Med (2012.0) 366 780-781. DOI: 10.1056/NEJMp1109283 32. Taylor F, Ward K, Moore TH. Statins for the primary prevention of cardiovascular disease. Cochrane Database Syst Rev (2011.0). DOI: 10.1002/14651858.CD004816.pub4 33. Byrne P, Cullinan J, Smith SM. Statins for primary prevention of cardiovascular disease. BMJ (2019.0) 367 l5674. DOI: 10.1136/bmj.l5674 34. Redberg RF, Katz MH. Statins for primary prevention: the debate is intense, but the data are weak. JAMA (2016.0) 316 1979-1981. DOI: 10.1001/jama.2016.15085 35. Ahmad FB, Anderson RN. The leading causes of death in the US for 2020. JAMA (2021.0) 325 1829-1830. DOI: 10.1001/jama.2021.5469 36. 36World Health Organization. The top 10 causes of death. Updated December 9, 2020. Accessed June 17, 2022. https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death 37. Bates TR, Connaughton VM, Watts GF. Non-adherence to statin therapy: a major challenge for preventive cardiology. Expert Opin Pharmacother (2009.0) 10 2973-2985. DOI: 10.1517/14656560903376186 38. Phan K, Gomez YH, Elbaz L, Daskalopoulou SS. Statin treatment non-adherence and discontinuation: clinical implications and potential solutions. Curr Pharm Des (2014.0) 20 6314-6324. DOI: 10.2174/1381612820666140620162629 39. Chou AF, Scholle SH, Weisman CS, Bierman AS, Correa-de-Araujo R, Mosca L. Gender disparities in the quality of cardiovascular disease care in private managed care plans. Womens Health Issues (2007.0) 17 120-130. DOI: 10.1016/j.whi.2007.03.002 40. Bird CE, Manocchia M, Tomblin B. Mapping the gaps: gender differences in preventive cardiovascular care among managed care members in four metropolitan areas. Womens Health Issues (2018.0) 28 446-455. DOI: 10.1016/j.whi.2018.04.008 41. Vynckier P, Ferrannini G, Rydén L. Gender gap in risk factor control of coronary patients far from closing: results from the European Society of Cardiology EUROASPIRE V registry. Eur J Prev Cardiol (2022.0) 29 344-351. DOI: 10.1093/eurjpc/zwaa144 42. Mahtta D, Ahmed ST, Ramsey DJ. Statin prescription rates, adherence, and associated clinical outcomes among women with PAD and ICVD. Cardiovasc Drugs Ther (2020.0) 34 745-754. DOI: 10.1007/s10557-020-07057-y 43. Lee MT, Mahtta D, Ramsey DJ. Sex-related disparities in cardiovascular health care among patients with premature atherosclerotic cardiovascular disease. JAMA Cardiol (2021.0) 6 782-790. DOI: 10.1001/jamacardio.2021.0683 44. Zhang H, Plutzky J, Skentzos S. Discontinuation of statins in routine care settings: a cohort study. Ann Intern Med (2013.0) 158 526-534. DOI: 10.7326/0003-4819-158-7-201304020-00004
--- title: 'Multiple Automated Health Literacy Assessments of Written Health Information: Development of the SHeLL (Sydney Health Literacy Lab) Health Literacy Editor v1' journal: JMIR Formative Research year: 2023 pmcid: PMC9975914 doi: 10.2196/40645 license: CC BY 4.0 --- # Multiple Automated Health Literacy Assessments of Written Health Information: Development of the SHeLL (Sydney Health Literacy Lab) Health Literacy Editor v1 ## Abstract Producing health information that people can easily understand is challenging and time-consuming. Existing guidance is often subjective and lacks specificity. With advances in software that reads and analyzes text, there is an opportunity to develop tools that provide objective, specific, and automated guidance on the complexity of health information. This paper outlines the development of the SHeLL (Sydney Health Literacy Lab) Health Literacy Editor, an automated tool to facilitate the implementation of health literacy guidelines for the production of easy-to-read written health information. Target users were any person or organization that develops consumer-facing education materials, with or without prior experience with health literacy concepts. Anticipated users included health professionals, staff, and government and nongovernment agencies. To develop this tool, existing health literacy and relevant writing guidelines were collated. Items amenable to programmable automated assessment were incorporated into the Editor. A set of natural language processing methods were also adapted for use in the SHeLL Editor, though the approach was primarily procedural (rule-based). As a result of this process, the Editor comprises 6 assessments: readability (school grade reading score calculated using the Simple Measure of Gobbledygook (SMOG)), complex language (percentage of the text that contains public health thesaurus entries, words that are uncommon in English, or acronyms), passive voice, text structure (eg, use of long paragraphs), lexical density and diversity, and person-centered language. These are presented as global scores, with additional, more specific feedback flagged in the text itself. Feedback is provided in real-time so that users can iteratively revise and improve the text. The design also includes a “text preparation” mode, which allows users to quickly make adjustments to ensure accurate calculation of readability. A hierarchy of assessments also helps users prioritize the most important feedback. Lastly, the Editor has a function that exports the analysis and revised text. The SHeLL Health Literacy Editor is a new tool that can help improve the quality and safety of written health information. It provides objective, immediate feedback on a range of factors, complementing readability with other less widely used but important objective assessments such as complex and person-centered language. It can be used as a scalable intervention to support the uptake of health literacy guidelines by health services and providers of health information. This early prototype can be further refined by expanding the thesaurus and leveraging new machine learning methods for assessing the complexity of the written text. User-testing with health professionals is needed before evaluating the Editor’s ability to improve the health literacy of written health information and evaluating its implementation into existing Australian health services. ## Introduction Health literacy describes a person’s capacity to access, understand, appraise, and use information and services to promote and maintain good health [1]. National and international policies increasingly recognize disparities in health literacy as a critical source of health inequality. This was demonstrated most recently by the World Health Organization, which positioned health literacy as one of the 3 key pillars needed to achieve the Sustainable Development Goals [2]. A recent review demonstrated that internationally, policies concerning health literacy consistently argue that providing health information that all people can easily access and understand is fundamental to addressing health literacy [3]. However, integration of easy-to-understand health information into routine practice rarely happens, despite being a relatively simple and low-cost strategy. For example, less than $1\%$ of web-based Australian health information is estimated to meet recommended grade reading levels [4]. This issue persists even in the face of a global pandemic, where timely dissemination of understandable information is extremely important. Analysis of international COVID-19 materials from governments and official sources indicates that, on average, these are written above the recommended Grade 8 level for the general population, making them unsuitable for people with low health literacy [5]. Several well-established health literacy guidelines provide advice about how to structure, write, and visually present health information, for example, the Universal Precautions Toolkit [6] and the Patient Education Materials Assessment Tool (PEMAT) [7]. One of the most widely used health literacy guidelines recommends writing health information at or below a Grade 8 reading level in countries such as Australia [4,8] or Grades 5 to 6 in the United States [9]. This is a useful standard because it is specific, objective, replicable, and readily available from web-based readability calculators. However, the concept of a grade reading score is narrow in scope, and many of the underlying formulas assess language complexity primarily in terms of syllable counts, the lengths of words, and the lengths of sentences [10]. Additional guidelines advise on other aspects of the text, including how common the words are, sentence structure, grammar, overall structure, and the flow of ideas in the text. The importance of these other criteria is supported by recent machine learning algorithms that predict the accessibility of health information. For example, these studies have shown that text features such as familiarity with the text’s vocabulary and cohesion across sentences may also play an important role [11-16]. However, to date, many of the health literacy guidelines do not afford the same level of specificity and objectivity as those for grade reading scores. This is illustrated through the example of a PEMAT item that advises the use of common, everyday language. Though this guideline is valuable, it can be challenging to implement as there are no detailed instructions on how to assess or act on this criterion, notably which words are considered “everyday” and what an acceptable number of uncommon words for a given text length might be [7]. With recent advances in computer science, web-based software may be able to address this issue. For example, Ondov and colleagues [17] recently identified 45 papers investigating simplification for biomedical texts, including 32 tools or methods. Of these, 22 tools took a procedural (rules-based) approach; 10 primarily used a machine learning approach, that is, through natural language processing. This is a rapidly developing area of research. The authors noted that, though machine learning approaches provide more sophisticated output than traditional grade reading scores, the quality of these models is currently constrained by the training data sets that are available. In contrast, the authors argue that procedural approaches, though likely to provide less tailored feedback, have the benefit of being more predictable. Regardless, few of these projects have resulted in tools that can be easily accessed and used by health services staff. There are some existing web-based platforms that provide detailed feedback on general writing style. For example, the Hemingway App [18], Grammarly [19], StyleWriter [20], and VisibleThread [21] are web-based tools that variously provide feedback on aspects of the text such as readability (including long words and long sentences), unnecessary adverbs, passive voice, formality, tone, and engagement. However, only some of these provide specific suggested alternative phrasing to reduce the complexity of health information, with none specifically addressing health and medical jargon, such as terms identified by the Centers for Disease Control and Prevention’s Everyday Words for Public Health Communication [22]. Further, none have been specifically designed for health contexts or with health literacy guidelines in mind. Leroy and colleagues [23] have developed a promising tool for the simplification of medical information; however, the tool will require further testing to ensure it aligns with health literacy principles and establish whether it can be used effectively by health information developers. In 2020, our team developed a web-based platform to broaden the range of automated assessments available to people developing patient-facing health information in a manner that is easy for health staff to understand and use. This paper outlines the development of the “SHeLL (Sydney Health Literacy Lab) Health Literacy Editor,” including the rationale and operationalization for each of the included objective assessments. ## Objectives of the SHeLL Health Literacy Editor The SHeLL Health Literacy Editor aimed to assist Australian health information providers to develop health education materials for patients or community members (herein referred to as “consumers”) that adhere to health literacy guidelines to improve the quality, safety, and ease of reading of written health information. This would be achieved by developing objective and programmable health literacy assessments informed by existing health literacy guidelines (“inputs”) and established objective assessments from other fields, for example, linguistics. Where possible, other strategies to promote good health literacy practice were also incorporated, for instance, by providing immediate feedback on specific words, phrases, or sentences in addition to whole-text assessments such as readability. ## Target Users We identified our target users as any person or organization that develops consumer-facing health information materials. We do not anticipate that users necessarily have prior experience with health literacy concepts, but would expect users to speak and understand English and to have sufficient skills to write health information in English. Users may include health professionals, staff, and government and nongovernment agencies. ## Inputs Existing guidelines related to health literacy informed the selection of assessments in the SHeLL Health Literacy Editor (Table 1). Items from these guidelines were incorporated into the Editor if they were amenable to automated assessment, for instance, through calculations involving counts of the numbers of words and sentences, string (character) searches, or identification of grammar. Table 1 \| Guideline or resourcea \| Description and scope \| Items amenable to incorporation into the SHeLL Health Literacy Editor \| \| --- \| --- \| --- \| \| Universal Precautions Toolkit [9] \| A suite of 21 tools to promote health literacy in 4 domains: spoken communication, written communication, self-management and empowerment, and supportive systems \| Tool #11 (Assess, select, and create easy-to-understand materials): specific relevant recommendation is to write at the 5th or 6th grade reading level \| \| Patient Education Materials Assessment Tool [7] \| Subjectively rated tool to assess the understandability and actionability of health information. For printed materials, this includes 17 items that assess understandability and 7 items that assess actionability. \| Item 3: The material uses common, everyday language;Item 4: Medical terms are used only to familiarize the audience with the terms;Item 5: The material uses the active voice;Item 8: The material breaks or “chunks” information into short sections \| \| Centers for Disease Control and Prevention Clear Communication Index [24] \| 20-item tool to improve public communication adherence to plain language guidelines and support the implementation of US health literacy policies \| Item 6: Use active voiceItem 7: Use words the primary audience usesItem 8: Chunk information \| \| Evaluative Linguistic Framework [25] \| Framework for assessing patient information leaflets, based on linguistic theory. Items include consideration of organization and structure, metadiscourse, headings, technicality of vocabulary, lexical density, the relationship between reader and writer, and format \| Technicality of vocabulary;Lexical density \| \| Plain Language [26] \| Guidelines for preparing texts to meet US plain language standards, including text grade reading level, organization, and word choice. \| Use simple words and phrases (for words that can be identified using a string search); avoid noun strings; avoid jargon; minimize abbreviations; use active voice; write short paragraphs; write short sentences \| \| Health Literacy Online [27] \| Guidelines for web-based health information, including writing actionable content, displaying content clearly, organizing content and simplifying navigation, engaging users, and user testing \| 2.6 (Write in plain language) \| \| Everyday words for public health communication [22] \| A thesaurus containing simpler alternatives to public health jargon \| All entries \| \| Simply Put: Writing and Design Tips [28] \| Guidelines for preparing easy-to-understand information, including the written text, visual aspects, and testing with consumers \| “Use everyday words,” “Keep sentences short,” “spell out acronyms,” “Use active verbs” \| \| Suitability Assessment of Materials [29] \| Subjectively rated tool to assess the suitability of health-related information for adults, including content, literacy demand, graphics, layout, learning stimulation and motivation, and cultural appropriateness \| Literacy demand (Score of 5th Grade reading level or lower=superior; 6th-8th Grade=adequate; 9th Grade or above=not suitable). \| \| Person-centered language [30-35] \| Various language position statements from Australian peak bodies outlining preferred language for a given health condition \| Words or phrases that could be identified using a string search \| \| Question Understanding Aid [36] \| Web-based tool to assess the comprehensibility of survey questions and response options. \| Unfamiliar technical term, complex syntax, working memory overload \| ## Functionality Considerations As far as possible, the SHeLL Health Literacy Editor was designed to provide automated, immediate, and objective feedback on written health text. This was facilitated by incorporating software that can process and analyze English-language text called spaCy [38]. SpaCy breaks down text into sentences and words. It then uses rule-based methods and trained models to identify grammatical information about each word. This information includes the word’s part of speech (eg, whether it is a noun, preposition, or verb), lemma (base word form, eg, “write” is the lemma for the word “written”), and whether the word is a named entity (eg, John, Canada, Monday). ## Overview Based on the above inputs (Table 1), we identified 6 assessments that could be implemented for real-time use while editing a document on a web-based interface: readability, complex language, passive voice, text structure, lexical density or diversity, and person-centered language. For each of these assessments, we describe the rationale for its inclusion below. We implemented 4 features to assist with usability: a “text preparation” mode; ordering the assessments by importance (a hierarchy of assessments); functions to export the revised text to a Word document; and exporting a summary of assessments as a PDF. In addition, where possible, user instructions and feedback have been framed to set clear expectations about the intended use of the Editor and its assessments. ## Readability Readability estimates how difficult a text is to read, often presented in the form of a “Grade Reading Score” [9]. Grade reading scores are identified as a useful tool in many health literacy guidelines [8,9,39] and are widely used in health literacy research (see, eg, [5]). A variety of readability formulas are used to assess health information [40,41]. We identified the Simple Measure of Gobbledygook (SMOG) [42] as the most appropriate readability formula for the SHeLL Health Literacy Editor. It is the only readability formula for which the grade reading score assumes the reader has a complete comprehension of the text [40]. For example, the SMOG assumes that a Grade 8 reader would score $100\%$ on a multiple-choice comprehension test for a text written at a Grade 8 reading level. By comparison, the Flesch Reading Ease assumes that Grade 8 readers would correctly answer $75\%$ on a multiple-choice comprehension test for the same text [43]. The Flesch Kincaid, another widely used readability formula, assumes $35\%$ comprehension based on a cloze test rather than multiple choice questions [44]. As such, the SMOG provides a more conservative estimate of the grade reading score than other common readability formulas [40,44,45]. Other studies have demonstrated that SMOG assessments are also more consistent across random sampling within a text and are less sensitive to differences in formatting [40]. A target of a Grade 8 reading score or lower was selected to match Australian recommendations [8]. The SHeLL Health Literacy Editor provides an overall Grade Reading Score based on the SMOG formula, rounded to the nearest whole number (Figure 2). The SMOG formula estimates the Grade Reading Score based on the proportion of words in each sentence that are multisyllabic (>2 syllables). The Editor counts the number of syllables using an open-source English language dictionary that provides syllable counts for over 115,000 words [52]. If a given word is not listed in the dictionary, the syllable count is estimated from the patterns of vowels and consonants. To ensure the accuracy of the SMOG score presented to users, the automated calculation was compared to manually calculated scores using prose text that did not contain ambiguous syllable counts (eg, numbers and acronyms that can be pronounced as individual letters or as a single word, eg, “WHO” for World Health Organization). To assist users, the SHeLL Health Literacy Editor flags words in the text that are contributing to a higher SMOG calculation (ie, words that are >2 syllables). The Editor also flags sentences longer than 20 words. This sentence length was selected on the basis of other health literacy recommendations [27]. Figure 2: Screenshot of the SHeLL (Sydney Health Literacy Lab) Health Literacy Editor v1, full-text editor pane. ## Complex Language All health literacy guidelines emphasize the need to use simple, everyday language and minimize medical jargon (Table 1). In some instances, medical terminology may be required and should be defined and explained in simpler words. Similarly, acronyms are also often considered technical terms that should be defined in the first instance [28]. ## Passive Voice Using active voice is a key recommendation to improve how easy health information is to understand and act upon [7]. The passive voice refers to a grammatical construction that emphasizes the recipient of an action (eg, “the blood test was ordered by the doctor”), whereas the active voice places an emphasis on the entity carrying out the action (“the doctor ordered the blood test”). The SHeLL Health Literacy Editor identifies patterns of the verb “to be” (eg, “is,” “were”) and a past participle (eg, “delivered,” “given”) that indicate passive voice. Users can read a brief description of the passive voice, including worked examples that change passive voice constructions into the active voice. ## Text Structure The structure of paragraphs and sentences was identified as a factor relevant to text complexity by several guidelines (Table 1). For example, Health Literacy Online recommends keeping paragraphs to 3 lines or less [27]. Similarly, the US Plain Language guidelines recommend that paragraphs be between 3 and 8 sentences long and no more than 150 words [26]. The Plain Language guidelines also advise against “sentences loaded with dependent clauses and exceptions” [26]. An example is depicted in Panel A1 of Figure 1, in which 3 dependent clauses are underlined and numbered. The text can be restructured to improve clarity by reducing the number of dependent clauses and replacing words that indicate exceptions (Figure 1, Panel A2). Lastly, the Question Understanding Aid’s “Working Memory Overload” assessment (Table 1; [36]) advises against double-barreled phrasing and convoluted questions. For example, “Do you think that diet and exercise are effective for managing diabetes and cardiovascular disease?” is a double-barreled question. Responses could variously refer to diet, exercise, or both types of interventions and may relate to diabetes, cardiovascular disease, or both conditions. Figure 1: Illustrative examples of text structure (Panels A1 and A2) and lexical density (Panels B1 and B2). Simpler alternatives are shown in Panels A2 and B2. These examples are intended to illustrate differences in text structure and lexical density, respectively. Texts A2 and B2 may benefit from further simplification, for example, using dot points for each step in A2 and using simpler words in B2. The SHeLL Health Literacy Editor provides guidance on paragraph length by flagging paragraphs that are longer than 8 sentences or more than 150 words. This criterion also aligns with recommendations from the US Plain Language guidelines [26]. The Editor identifies complex questions as those consisting of at least 12 words and more than 2 conjunctions (“for,” “and,” “nor,” “but,” “or,” “yet,” and “so”), based on the Question Understanding Aid’s “Working Memory Overload” assessment (Table 1; [36]). In doing so, this flag aims to identify potential instances of double-barreled or convoluted questions but should only be considered a proxy for complex questions. ## Lexical Density and Diversity Lexical density is a component of the Evaluative Linguistic Framework (Table 1; [25]). However, lexical density and diversity have not been extensively studied in health contexts despite being common computational linguistic assessments [14,46]. Lexical density measures the ratio of words in a text that are “content words” versus “function words.” Content words tell us what a text is about (nouns, adjectives, most verbs, and most adverbs). Function words are those that carry grammatical meaning. A text with higher lexical density, therefore, conveys meaning more concisely. For example, compare the sentences in Panels B1 and B2 of Figure 1. The sentence in Panel B1 has a higher lexical density, with a ratio of 5 content words:1 function word, compared to the sentence in Panel B2 (6 content words:4 function words). Conceptually, it may be beneficial for health information materials to have a lower lexical density, as this style of writing is more indicative of spoken English (usually with a lexical density score between 1.5 and 2) than written English (usually between 3 and 6) [47]. This aligns with health literacy guidelines that recommend writing with a “conversational tone” [26,27]. Lexical diversity measures the proportion of words in a text that are unique. Higher lexical diversity indicates that a text has a larger vocabulary [48]. A text with higher lexical diversity may use more words for the same concept, for example, “cancer,” “carcinoma,” and “neoplasm.” A text with low lexical diversity may simply refer to “cancer.” The SHeLL Health Literacy Editor uses information about the part of speech to determine whether a word fulfills a function or content role. Prepositions (eg, in, on), pronouns (eg, she, them), determiners (eg, the, a), conjunctions (eg, and, that), and auxiliaries (eg, is, got, do) are categorized as function words; all other parts of speech are categorized as content words. The ratio of content words to function words per clause is then calculated [47]. The SHeLL Health Literacy Editor computes an unstandardized and standardized assessment of lexical diversity. The unstandardized assessment, or “type-token ratio,” is the ratio of unique words to total words. The type-token ratio is correlated with text length [14]. The Measure of Lexical Textual Diversity [54] is a standardized type-token ratio that adjusts for text length by averaging the type-token ratio across sequential strings of words within the text. Measure of Lexical Textual Diversity is more stable across texts of different lengths [54,55]. ## Person-Centered Language It is widely recommended that health information adopt a person-centered approach to health services [49,50]. Language can have a lasting impact on how people understand their condition, their treatment, and their place in the community. Person-centered language seeks to reduce blame, stigma, and judgment and encourage accuracy, autonomy, respect, and inclusion [51]. The SHeLL Health Literacy Editor draws on peak-body guidance for person-centered language across several conditions: diabetes, dementia, chronic pain, cancer, and mental health (including language that aligns with trauma-informed care) [30-35]. As language guidelines become available for other health conditions, these can be incorporated into the Editor. This feature flags sections of text that contain easily identifiable examples of language that are not person-centered; for example, rather than “sufferer,” guidelines recommend referring to “a person living with X condition.” Of note, this feature is not comprehensive, as some aspects of person-centered guidelines require the writer to consider aspects that are broader than individual words or phrases that can be identified using a string search function. ## Complex Language (Vocabulary) We identified several resources that provide simpler alternatives to complex language, including the Centers for Disease Control and Prevention’s Everyday Words for Public Health Communication, which was developed specifically to address health literacy needs in health communication [22]. Thesaurus entries from these resources were collated into a database listing the word, relevant string searches, and an accompanying thesaurus entry containing possible alternatives. Users can access thesaurus entries by hovering over a word (Figure 2). Users can enter up to 5 words that will be excluded from the complex language assessment if they believe readers will be familiar with the terms. This feature affords flexibility to the user while also seeking to discourage users from exempting all jargon from the complex language assessment. The maximum number of excluded words will be further refined as user feedback is gathered. In addition, the Editor identifies words that are uncommon in the English language based on word frequencies in a database of more than 270 million words from diverse English-language sources (learner materials, fiction, journals and magazines, nonfiction, radio, spoken English, documents, and TV) [53]. The database was specifically designed to identify words that would be most useful to people learning English as a second language. For example, its authors claim that the most frequent 2800 words provide learners with $90\%$ coverage for general English texts [53]. This assessment also uses spaCy’s trained named entity recognition model to prevent named entities such as companies, locations, organizations, languages, countries, and periods of time from being flagged as uncommon. Acronyms were identified as a series of at least 2 capital letters, or capital letters with a period in between. Lowercase letters were allowable as this is common practice in health (eg, SHeLL for Sydney Health Literacy Lab). An overall “text complexity” score is calculated from the proportion of words flagged with any of the 3 complex language assessments (“thesaurus,” “acronyms,” or “uncommon words”). No targets were available as this is a new objective assessment. ## Text Preparation Preparing a text for readability assessment is an important aspect of calculating a grade reading score. However, this preparation can be cumbersome when text (eg, headings) must be removed or altered for assessment purposes but is ultimately included in the document. To reduce this burden, the SHeLL Health Literacy Editor allows users to indicate which segments of text to exclude from the readability calculation without having to edit the text itself. Common text preparation decisions are set as a default setting [41]. For example, by default, the Editor does not count short bullet points (less than 4 words), headings that are less than 4 words, or URLs. Bullet points are considered a “sentence” even if there is no full stop at the end. ## Hierarchy of Assessments The SHeLL Health Literacy Editor flags sections of the text using opaque rectangular boxes (“highlights”) of different colors (Figure 1). Each color represents a different assessment. Assessments that are higher priorities overlay those that are lower priorities. This hierarchy prioritizes guidance for complex language, followed by passive voice, readability, and complex structure. Users can toggle assessments on or off to view overlapping highlights. To avoid overwhelming new users, only the 3 highest-ranked assessments are active by default: complex language, passive voice, and readability. ## Export and Summary Features Users can export a copy of the text as a Word document or as a “summary file” that provides all objective assessments and information about text preparation decisions, including the maximum of 5 words excluded from the complex language assessments. ## Setting Expectations for Intended Use We anticipate that users may need guidance to correctly interpret Editor feedback. For example, there is a risk that users may feel the need to remove all highlights from the text for the simplification task to be considered “complete.” To mitigate frustration and set realistic expectations, the Editor’s prompts and instructions emphasize that there are likely to be some highlighted words and, rather, to aim to make the text as simple as possible (eg, Aim for Grade 8 or lower). ## Summary The SHeLL Health Literacy Editor is an urgently needed, innovative tool to support the timely development of health-literate written health information. It objectively assesses the readability, complex language, passive voice, text structure, lexical density and diversity, and person-centered language. By explicitly aligning features with existing health literacy guidelines, the tool provides health information developers with a unique and targeted tool to improve the quality and safety of health information. The fact that assessments are provided in real-time supports iterative revisions. to reduce text complexity. A key strength of the SHeLL Health Literacy Editor is that it complements the widely and almost exclusively used readability score with other relevant assessments, including those specific to health. Other strengths include its capacity to improve the efficiency of preparing texts for readability analyses through the text preparation function; its capacity to build workforce skills in applying health literacy principles; and its feasibility for scaling up across an organization or jurisdiction given the minimal cost and resources involved. We have also completed extensive user testing of the SHeLL Health Literacy Editor with health staff, which is reported separately (Ayre et al, unpublished data). User-testing sought to evaluate and improve acceptability and usability, help prioritize additional features, and identify training needs. It is important to emphasize that the SHeLL Health Literacy Editor does not replace more comprehensive health literacy guidelines. For example, the PEMAT also provides guidance on actionability and visual elements. We envisage its scope as assisting people to develop simpler text to convey health information. A few specific aspects of the written text are also outside its scope. For example, strategies for communicating risk accurately and without bias [56] and guidelines about written text that operate beyond the level of the sentence (eg, outlining the text’s purpose and logical sequence of information) are also largely outside the current scope of the Editor, though they could be considered in future iterations. We envision that the SHeLL Health Literacy Editor would be used in the early stages of resource development. Involving consumers is critical to developing accessible and understandable health information resources [57]. However, obtaining consumer feedback is resource-intensive. The Editor will facilitate an efficient and scalable process in which health literacy principles are applied as much as possible to a text prior to consumer involvement. The Editor may also improve translation efforts by ensuring that the parent text is expressed simply prior to translation. ## Future Directions There are many avenues for further research involving the Editor. We intend to evaluate the Editor’s ability to improve the health literacy of written health information and evaluate its implementation into existing Australian health services. This evaluation could also investigate the relative importance of each of the Editor’s assessments and establish appropriate objective health literacy benchmarks that would complement existing subjective health literacy guidelines. Currently, the Editor’s features take a primarily procedural (rules-based) approach. In future iterations, increased use of machine learning approaches could enhance the Editor’s features. For example, the Editor could highlight sentences containing many dependent clauses and give specific advice about how to simplify these sentence structures. As another example, the value of the thesaurus function is largely driven by the number, quality, and relevance of the thesaurus entries. This could be further enhanced by leveraging large existing (manually developed) medical dictionaries and by incorporating machine learning methods that have “mined” pairs of jargon and lay terms using multiple corpora [17,58-60]. The uncommon language feature may be further improved by using the “SciSpaCy” variant that has been adapted to biomedical texts, as this may result in improved identification of medically named entities. Beyond the structure and content of individual sentences and words, newer approaches have the advantage of assessing the text more holistically, assessing high-level features such as cohesion and coherence [11-16]. The Editor could also help users identify whether jargon or acronyms are defined the first time they are used, and potentially incorporate this assessment into the text complexity score. Further work is also needed to establish how these newer assessments relate to the understanding of health information in health literacy priority populations, and to establish how information about coherence and cohesion can be effectively conveyed to users of the tool who are developing health information. Lastly, these assessments are often implied but not explicit in health literacy guidelines, and this additional research could ultimately help refine health literacy guidelines and improve their evidence base. ## Conclusions The SHeLL Health Literacy Editor provides health services and health information providers with an innovative new tool to improve written health information. The Editor provides objective, immediate feedback on a range of factors, complementing readability with other less widely used and objective assessments such as complex language. The Editor presents health services with a scalable and accessible intervention to address health literacy that staff developing written health information in different settings can easily use. This early prototype has several avenues through which the Editor can be further refined, including expanding the thesaurus and leveraging new machine learning algorithms for assessing the complexity of written text and suggesting alternative phrasing. Ultimately, these efforts seek to build capacity for health information developers to understand health literacy principles and then apply them effectively to educational materials. This systems-based approach has the potential to substantially improve the health literacy environment in our communities. ## References 1. Nutbeam D, Muscat DM. Health promotion glossary 2021. Health Promot Int (2021.0) 36 1578-1598. DOI: 10.1093/heapro/daaa157 2. Shanghai declaration on promoting health in the 2030 agenda for sustainable development. Health Promot Int (2017.0) 32 7-8. DOI: 10.1093/heapro/daw103 3. Trezona A, Rowlands G, Nutbeam D. Progress in implementing national policies and strategies for health literacy: what have we learned so far?. Int J Environ Res Public Health (2018.0) 15 1554. DOI: 10.3390/ijerph15071554 4. Cheng C, Dunn M. Health literacy and the internet: a study on the readability of Australian online health information. Aust N Z J Public Health (2015.0) 39 309-314. DOI: 10.1111/1753-6405.12341 5. Mac OA, Muscat DM, Ayre J, Patel P, McCaffery KJ. The readability of official public health information on COVID-19. Med J Aust (2021.0) 215 373-375. DOI: 10.5694/mja2.51282 6. DeWalt DA, Broucksou KA, Hawk V, Brach C, Hink A, Rudd R, Callahan L. Developing and testing the health literacy universal precautions toolkit. Nurs Outlook (2011.0) 59 85-94. DOI: 10.1016/j.outlook.2010.12.002 7. Shoemaker SJ, Wolf MS, Brach C. Development of the patient education materials assessment tool (PERMAT): a new measure of understandability and actionability for print and audiovisual patient information. Patient Educ Couns (2014.0) 96 395-403. DOI: 10.1016/j.pec.2014.05.027 8. Health literacy. Government of South Australia (2013.0) 9. Brega A, Barnard J, Mabachi N, Weiss B, DeWalt D, Brach C, Cifuentes M, Albright K, West D. Health literacy universal precautions toolkit, 2nd edition. Agency for Healthcare Research and Quality's (AHRQ) (2015.0) 10. Meade CD, Smith CF. Readability formulas: cautions and criteria. Patient Educ Couns (1991.0) 17 153-158. DOI: 10.1016/0738-3991(91)90017-y 11. Xie W, Ji C, Hao T, Chow CY. Predicting the easiness and complexity of english health materials for international tertiary students with linguistically enhanced machine learning algorithms: development and validation study. JMIR Med Inform (2021.0) 9 e25110. DOI: 10.2196/25110 12. Ji M, Liu Y, Zhao M, Lyu Z, Zhang B, Luo X, Li Y, Zhong Y. Use of machine learning algorithms to predict the understandability of health education materials: development and evaluation study. JMIR Med Inform (2021.0) 9 e28413. DOI: 10.2196/28413 13. Ji M, Liu Y, Hao T. Predicting health material accessibility: development of machine learning algorithms. JMIR Med Inform (2021.0) 9 e29175. DOI: 10.2196/29175 14. McNamara DS, Graesser AC, McCarthy PM, Cai Z. Automated Evaluation of Text and Discourse with Coh-Metrix (2014.0) 0521137292 15. Crossley SA, Skalicky S, Dascalu M, McNamara DS, Kyle K. Predicting text comprehension, processing, and familiarity in adult readers: new approaches to readability formulas. Discourse Process (2017.0) 54 340-359. DOI: 10.1080/0163853x.2017.1296264 16. Martinc M, Pollak S, Robnik-Šikonja M. Supervised and unsupervised neural approaches to text readability. Comput Linguist (2021.0) 47 141-179. DOI: 10.1162/coli_a_00398 17. Ondov B, Attal K, Demner-Fushman D. A survey of automated methods for biomedical text simplification. J Am Med Inform Assoc (2022.0) 29 1976-1988. DOI: 10.1093/jamia/ocac149 18. 18Hemingway Editor20212021-06-01 https://hemingwayapp.com/. Hemingway Editor (2021.0) 19. 19Grammarly20212021-06-01 https://app.grammarly.com/. Grammarly (2021.0) 20. 20StyleWriter Version 4, 20222022-12-02 https://www.stylewriter-usa.com/. StyleWriter Version 4, 2022 21. 21VisibleThread, The Language Analysis Platform That Means Business20222022-12-02 https://www.visiblethread.com/. VisibleThread, The Language Analysis Platform That Means Business (2022.0) 22. Everyday words for public health communication. Centers for Disease Control and Prevention (2016.0) 23. Leroy G, Kauchak D, Haeger D, Spegman D. Evaluation of an online text simplification editor using manual and automated metrics for perceived and actual text difficulty. JAMIA Open (2022.0) 5 ooac044. DOI: 10.1093/jamiaopen/ooac044 24. Baur C, Prue C. The CDC clear communication index is a new evidence-based tool to prepare and review health information. Health Promot Pract (2014.0) 15 629-637. DOI: 10.1177/1524839914538969 25. Clerehan R, Buchbinder R, Moodie J. A linguistic framework for assessing the quality of written patient information: its use in assessing methotrexate information for rheumatoid arthritis. Health Educ Res (2005.0) 20 334-344. DOI: 10.1093/her/cyg123 26. 26Plain Language Action and Information NetworkFederal Plain Language Guidelines, March 20112018-12-12 https://www.plainlanguage.gov/media/FederalPLGuidelines.pdf. Federal Plain Language Guidelines, March 2011 27. Health literacy online: a guide to simplifying the user experience. U.S. Department of Health and Human Services, Office of Disease Prevention and Health Promotion (2015.0) 28. 28National Adult Literacy AgencySimply Put: Writing and Design Tips2011IrelandALA, Dublin. Simply Put: Writing and Design Tips (2011.0) 29. Doak CC, Doak LG, Root JH. Teaching patients with low literacy skills. Am J Nurs (1996.0) 96 16M. DOI: 10.1097/00000446-199612000-00022 30. Position statement: a new language for diabetes: improving communications with and about people with diabetes. Diabetes Australia (2016.0) 31. Talking about pain: language guidelines for chronic pain, 2019. PainAustralia 32. Writing about cancer guidelines. Cancer Institute NSW (2021.0) 33. Recovery oriented language guide - second edition revised. Mental Health Coordinating Council (2018.0) 34. Communicating about mental ill-health: language. Mindframe (2021.0) 35. Dementia language guidelines. Dementia Australia (2021.0) 36. Graesser AC, Wiemer-Hastings K, Kreuz R, Wiemer-Hastings P, Marquis K. QUAID: a questionnaire evaluation aid for survey methodologists. Behav Res Methods Instrum Comput (2000.0) 32 254-262. DOI: 10.3758/bf03207792 37. Charnock D, Shepperd S, Needham G, Gann R. DISCERN: an instrument for judging the quality of written consumer health information on treatment choices. J Epidemiol Community Health (1999.0) 53 105-111. DOI: 10.1136/jech.53.2.105 38. Honnibal I, Montani I. spaCy 2: natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing. Sentometrics Research (2017.0) 39. NSW health: health literacy framework. Clinical Excellence Commission (2019.0) 40. Wang LW, Miller MJ, Schmitt MR, Wen FK. Assessing readability formula differences with written health information materials: application, results, and recommendations. Res Social Adm Pharm (2013.0) 9 503-516. DOI: 10.1016/j.sapharm.2012.05.009 41. Beaunoyer E, Arsenault M, Lomanowska AM, Guitton MJ. Understanding online health information: evaluation, tools, and strategies. Patient Educ Couns (2017.0) 100 183-189. DOI: 10.1016/j.pec.2016.08.028 42. Mc Laughlin GH. SMOG grading-a new readability formula. J Read (1969.0) 12 639-646 43. Flesch R. A new readability yardstick. J Appl Psychol (1948.0) 32 221-233. DOI: 10.1037/h0057532 44. Kincaid JP, Fishburne RP, Rogers RL, Chissom BS. Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel. Institute for Simulation and Training (1975.0) 45. Grabeel KL, Russomanno J, Oelschlegel S, Tester E, Heidel RE. Computerized versus hand-scored health literacy tools: a comparison of simple measure of gobbledygook (SMOG) and Flesch-Kincaid in printed patient education materials. J Med Libr Assoc (2018.0) 106 38-45. DOI: 10.5195/jmla.2018.262 46. Schillinger D, Balyan R, Crossley SA, McNamara DS, Liu JY, Karter AJ. Employing computational linguistics techniques to identify limited patient health literacy: findings from the ECLIPPSE study. Health Serv Res (2021.0) 56 132-144. DOI: 10.1111/1475-6773.13560 47. Halliday MAK. Spoken and Written Language (1989.0) 48. Biber D. Variation across Speech and Writing (1988.0) 49. 49National Safety and Quality Health Service Standards (second edition)2017Sydney, AustraliaAustralian Commission on Safety and Quality in Health Care. National Safety and Quality Health Service Standards (second edition) (2017.0) 50. Personalised care and support planning handbook: the journey to person-centred care. NHS England 51. Santana MJ, Manalili K, Jolley RJ, Zelinsky S, Quan H, Lu M. How to practice person-centred care: a conceptual framework. Health Expect (2018.0) 21 429-440. DOI: 10.1111/hex.12640 52. 52Carnegie Mellon University Pronouncing Dictionary, version 0.7b20142021-05-31 http://www.speech.cs.cmu.edu/cgi-bin/cmudict. Carnegie Mellon University Pronouncing Dictionary, version 0.7b (2014.0) 53. Browne C. The new general service list: celebrating 60 years of vocabulary learning. The Language Teacher (2013.0) 37 13-16 54. McCarthy PM, Jarvis S. MTLD, vocd-D, and HD-D: a validation study of sophisticated approaches to lexical diversity assessment. Behav Res Methods (2010.0) 42 381-392. DOI: 10.3758/BRM.42.2.381 55. Zenker F, Kyle K. Investigating minimum text lengths for lexical diversity indices. Assess Writ (2021.0) 47 100505. DOI: 10.1016/j.asw.2020.100505 56. Bonner C, Trevena LJ, Gaissmaier W, Han PKJ, Okan Y, Ozanne E, Peters E, Timmermans D, Zikmund-Fisher BJ. Current best practice for presenting probabilities in patient decision aids: fundamental principles. Med Decis Making (2021.0) 41 821-833. DOI: 10.1177/0272989X21996328 57. Mastroianni F, Chen YC, Vellar L, Cvejic E, Smith JK, McCaffery KJ, Muscat DM. Implementation of an organisation-wide health literacy approach to improve the understandability and actionability of patient information and education materials: a pre-post effectiveness study. Patient Educ Couns (2019.0) 102 1656-1661. DOI: 10.1016/j.pec.2019.03.022 58. Elhadad M, Sutaria K. Mining a lexicon of technical terms and lay equivalents. (2007.0). DOI: 10.3115/1572392.1572402 59. Kloehn N, Leroy G, Kauchak D, Gu Y, Colina S, Yuan NP, Revere D. Improving consumer understanding of medical text: development and validation of a new subsimplify algorithm to automatically generate term explanations in English and Spanish. J Med Internet Res (2018.0) 20 e10779. DOI: 10.2196/10779 60. Leroy G, Endicott JE, Mouradi O, Kauchak D, Just ML. Improving perceived and actual text difficulty for health information consumers using semi-automated methods. AMIA Annu Symp Proc (2012.0) 2012 522-31. PMID: 23304324
--- title: Clinicians’ and Patients’ Perspectives on Hypertension Care in a Racially and Ethnically Diverse Population in Primary Care authors: - Julie C. Lauffenburger - Renee A. Barlev - Rasha Khatib - Nicole Glowacki - Alvia Siddiqi - Marlon E. Everett - Michelle A. Albert - Punam A. Keller - Lipika Samal - Kaitlin Hanken - Ellen S. Sears - Nancy Haff - Niteesh K. Choudhry journal: JAMA Network Open year: 2023 pmcid: PMC9975920 doi: 10.1001/jamanetworkopen.2023.0977 license: CC BY 4.0 --- # Clinicians’ and Patients’ Perspectives on Hypertension Care in a Racially and Ethnically Diverse Population in Primary Care ## Key Points ### Question What are patients’ and clinicians’ perspectives on hypertension control and management in racially and ethnically diverse populations in primary care? ### Findings In this qualitative study of 15 patients and 15 clinicians, the participants felt that self-management (in particular, lifestyle modifications) should receive more attention; for example, they expressed difficulty with self-management activities, especially lifestyle modifications, and some hesitancy about and variation in intensifying medications and recommendations for follow-up care and patient self-monitoring. Current health information technology tools for hypertension were also thought to be limited. ### Meaning The results of this study suggest that, as current care models for hypertension are thought to be insufficient, more attention may need to be paid to ways to support treatment intensification and self-management, particularly through asynchronous interventions. ## Abstract This qualitative study of 15 patients and 15 clinicians examinged the barriers and facilitators to hypertension control within a racially and ethnically diverse health care system. ### Importance Hypertension control remains suboptimal, particularly for Black and Hispanic or Latino patients. A need exists to improve hypertension management and design effective strategies to efficiently improve the quality of care in primary care, especially for these at-risk populations. Few studies have specifically explored perspectives on blood pressure management by primary care providers (PCPs) and patients. ### Objective To examine clinician and patient perspectives on barriers and facilitators to hypertension control within a racially and ethnically diverse health care system. ### Design, Setting, and Participants This qualitative study was conducted in a large urban US health care system from October 1, 2020, to March 31, 2021, among patients with a diagnosis of hypertension from a racially and ethnically diverse population, for a range of hypertension medication use hypertension control, as well as practicing PCPs. Analysis was conducted between June 2021 and February 2022 using immersion-crystallization methods. ### Main Outcomes and Measures Perspectives on managing blood pressure, including medication adherence and lifestyle, considerations for intensification, and experiences and gaps in using health information technology tools for hypertension, were explored using semistructured qualitative interviews. These cycles of review were continued until all data were examined and meaningful patterns were identified. ### Results Interviews were conducted with 30 participants: 15 patients (mean [SD] age, 58.6 [16.2] years; 10 women [$67\%$] and 9 Black patients [$60\%$]) and 15 clinicians (14 PCPs and 1 medical assistant; 8 women [$53\%$]). Eleven patients ($73\%$) had suboptimally controlled blood pressure. Participants reported a wide range of experiences with hypertension care, even within the same clinics and health care system. Five themes relevant to managing hypertension for racially and ethnically diverse patient populations in primary care were identified: [1] difficulty with self-management activities, especially lifestyle modifications; [2] hesitancy intensifying medications by both clinicians and patients; [3] varying the timing and follow-up after changes in medication; [4] variation in blood pressure self-monitoring recommendations and uptake; and [5] limited specific functionality of current health information technology tools. ### Conclusions and Relevance In this qualitative study of the views of PCPs and patients on hypertension control, the participants felt that more focus should be placed on lifestyle modifications than medications for hypertension, particularly for patients from racial and ethnic minority groups. Participants also expressed concerns about the existing functionality of health information technology tools to support increasingly asynchronous hypertension care. More intentional ways of supporting treatment intensification, self-care, and follow-up care are needed to improve hypertension management for racially and ethnically diverse populations in primary care. ## Introduction More than 100 million individuals in the US are estimated to have hypertension, and the rates of control remain suboptimal.1 Control of hypertension among Black and Hispanic or Latino patients lags even further behind.2 These disparities are problematic because of differences in consequences3; an increase of 10 mm Hg in systolic blood pressure is associated with an $8\%$ increase in stroke risk among White patients but a $24\%$ increase among Black patients.4 Of the many factors associated with hypertension, suboptimal adherence to antihypertensive medications and lifestyle modifications, such as diet and exercise, are thought to be central,5 and clinical inertia, where clinicians do not intensify treatments when indicated, occurs in up to half of treatment episodes.6,7 Suboptimal adherence and clinical inertia are more common among patients from racial and ethnic minority groups.5,6,8 Furthermore, effective blood pressure management relies on monitoring response to therapy, but gaining access to accurate blood pressure values is challenging; Black and Hispanic or Latino patients are less likely than White patients to receive follow-up care or use home self-monitoring devices.2,9 Current care models for managing hypertension appear to be insufficient.10 Given the increasing prevalence of hypertension,11 the shortage of primary care providers (PCPs),12 and the increasing interest in asynchronous care or care outside of offices (associated in part with the COVID-19 pandemic),10 there is a need to design effective strategies to efficiently improve the quality of hypertension care. To our knowledge, few studies have explored perspectives on what would be specifically helpful to improve hypertension care, especially among racially and ethnically diverse populations in the US. We sought to examine the barriers and facilitators to efficient and effective hypertension care using in-depth qualitative interviews with patients and PCPs. We also sought to explore specific perspectives on current hypertension health information technology tools to recommend interventions that may overcome barriers. ## Methods This study, conducted between October 1, 2020, and March 31, 2021, followed the Consolidated Criteria for Reporting Qualitative Research (COREQ) guidelines.13 Patient and clinician participants provided verbal consent to participate and for use and recording of their interviews for research purposes, including publication. This study was approved by the Mass General Brigham institutional review board. Participants were recruited from Advocate Aurora Health, a large integrated health care delivery network in Illinois and Wisconsin serving a racially and ethnically diverse population. We used electronic health records (EHRs) to identify patients meeting the following eligibility criteria: [1] aged 18 years or older, [2] hypertension diagnosis, and [3] PCP visit at a study clinic in the prior 2 years. We focused on 2 Chicago clinics that serve a racially and ethnically diverse patient population to avoid contamination with a subsequent trial.14 Clinicians excluded patients if they thought they could not participate for cognitive reasons. Eligible patients were invited by telephone and an EHR-linked patient portal to participate. Potentially eligible clinicians included family practice and internal medicine primary care physicians, PCP-designated nurse practitioners, and medical assistants, who were contacted by email. ## Interviews To elicit personal accounts about hypertension management, we used individual, semistructured qualitative interviews. The lead author (J.C.L.), who is experienced in qualitative methods, drafted comprehensive guides that were designed to build empathy and that were iteratively refined by other coinvestigators (R.A.B., R.K., N.G., A.S., M.A.A., P.A.K., L.S., K.H., N.H., and N.K.C.) with expertise in primary care, behavioral science, disparities, and cardiology. Constructs from the Consolidated Framework for Implementation Research informed the development of the interview guides. We pilot tested and finalized guides with nonparticipant volunteers. The patient guide focused on coping with (ie, managing) high blood pressure, including adherence to medication and lifestyle modification, perspectives of monitoring tools, experiences accessing resources and tools, and interactions with clinicians about hypertension (eTable 1 in Supplement 1). The clinician guide focused on current practice and the challenges of managing patients with hypertension, considerations for initiating and intensifying medications, experiences using the EHR and other tools for hypertension, and suggestions for improvement (eTable 2 in Supplement 1). Due to the COVID-19 pandemic, interviews were conducted virtually using the Zoom video and audio platform and audio recorded. Interviews were conducted by a trained moderator and pharmacist (J.C.L.) in English. Several strategies were used to minimize misperceptions, including emphasizing that they were not part of the medical team. Before the interview, verbal consent was attained from clinicians and patients for permission for audio recordings. Clinicians were asked brief questions about race and ethnicity, sex, and training; information about race and ethnicity, age, sex, and blood pressure medications and values was obtained from the EHR for patients. The interviewer followed the guide but modified questions and asked follow-up questions depending on the participant’s responses. Sequential interviews were conducted until reaching thematic saturation.15 Each interview lasted 20 to 60 minutes (mean [SD] duration, 38 [14.1] minutes). Patients were offered $50 as compensation; clinicians’ clinics received $150. ## Analysis Analysis was conducted between June 2021 and February 2022. Interviews were transcribed verbatim and checked for accuracy. Two investigators (J.C.L. and R.A.B.) annotated a transcript selection independently and devised preliminary codes; after discussion, these codes were revised and agreed on. Then, each transcript was analyzed by the same 2 investigators using immersion and crystallization methods.16,17 We continued to review cycles until all data were examined and meaningful patterns emerged. Preliminary coding revealed themes around [1] access to ambulatory self-monitoring, [2] individual ways of intensifying treatments and monitoring, [3] difficulty with diet and exercise, and [4] the limited functionality of EHR hypertension tools. Dedoose, version 8.3.47b (SocioCultural Research Consultants) was used for storage and handling and analysis. ## Results We conducted 30 interviews with 15 patients (mean [SD] age, 58.6 [16.2] years; 10 women [$67\%$] and 9 Black patients [$60\%$]) and 15 clinicians (14 PCPs, and 1 medical assistant; 8 women [$53\%$]) (Table 1). Fourteen patients ($93\%$) were prescribed an antihypertensive medication; 11 ($73\%$) had suboptimally controlled blood pressure (ie, latest EHR-recorded blood pressure ≥$\frac{140}{90}$). Table 1. \| Characteristic \| No. (%) \| \| --- \| --- \| \| Clinicians \| Clinicians \| \| No. \| 15 \| \| Clinical role \| \| \| Physician \| 13 (87) \| \| Medical assistant \| 1 (7) \| \| Nurse practitioner \| 1 (7) \| \| Gender \| \| \| Female \| 8 (53) \| \| Male \| 7 (47) \| \| Race and ethnicitya \| \| \| Asian \| 2 (13) \| \| Black \| 2 (13) \| \| Hispanic or Latino \| 2 (13) \| \| White \| 7 (47) \| \| Patients \| Patients \| \| No. \| 15 \| \| Age, y \| \| \| 30-49 \| 4 (27) \| \| 50-64 \| 5 (33) \| \| ≥65 \| 6 (40) \| \| Gender \| \| \| Female \| 10 (67) \| \| Male \| 5 (33) \| \| Race and ethnicity \| \| \| Asian \| 1 (7) \| \| Black \| 9 (60) \| \| Hispanic or Latino \| 1 (7) \| \| White \| 5 (33) \| \| Latest blood pressure, mm Hg \| \| \| ≥140/90 \| 11 (73) \| Patients reflected on how they contextualize their diagnosis and integrate hypertension care into their routines and their relationship with PCPs. Clinicians described how they manage patients’ hypertension, communicate with patients, and use the EHR. We identified 5 themes (Table 2). Each theme is presented with representative quotations; other quotations are in eTable 3 in Supplement 1. Table 2. \| Theme \| Key takeaways \| \| --- \| --- \| \| Difficulty with self-management activities, especially lifestyle modifications \| Patients report more difficulty with diet and physical activity, exacerbated by COVID-19 pandemicClinicians also believe that adherence to medication is less of a challenge in their population \| \| Hesitancy intensifying medications by both clinicians and patients \| Thought to be more difficult to intensify medications than to initiate treatmentTend to individualize treatment choicesManagement typically done by primary care provider \| \| Varying timing and follow-up after changes in blood pressure medication \| Patterns of follow-up range reasonably widely even by clinicians within the same clinicPatient portal sometimes used to support follow-up \| \| Variation in blood pressure self-monitoring recommendations and uptake \| No consistent ways of recording or sharing of out-of-office blood pressure valuesInsurance key barrier to accessSize of blood pressure cuff represents an important physical barrier \| \| Limited specific functionality of current health information technology tools \| Interest in better way to observe longitudinal trends in hypertension care (eg, values and medications)Largely use self-created patient education materials because of lack of preferred electronic health record–embedded ones \| ## Difficulty With Self-management Activities, Especially Lifestyle Modifications Patients acknowledged that maintaining a healthy lifestyle is very difficult; even among those who used strategies to adhere to a healthy lifestyle, adherence to those strategies was imperfect. Clinicians noted specific challenges that some patients face when making lifestyle decisions, including easy access to healthy food and exercise locations; structural barriers were exacerbated by the COVID-19 pandemic: Patient: “Food and diet is a hard thing to do, especially with my kids—’cause if you change one person, you have to change everybody.” Clinician: “I see people sometimes working 2 jobs trying to make ends meet, and so they find themselves sleeping maybe 4 [or] 5 hours. Then they have no time, which would help them with their blood pressure.” Clinicians and patients alike generally believed that medication adherence was less difficult than lifestyle change. This belief stemmed from the fact that most antihypertensive medications are inexpensive and thus access would be easy. Clinicians described patient confusion about which lifestyle choices are truly healthy, recognizing that there may also be ranges of recommendations for patients, while medication-taking was deemed more straightforward. Some patients also reported clear medication-taking routines: Patient: “I just sit the pill bottle on the table, so as I’m getting ready for work, I make sure I take my medicine.” Clinician: “We’re an inner-city practice, and so cost is always important, but honestly, all those meds are generic. Really, the biggest barriers are, for my patients, lifestyle, and they don’t feel sick. In the south side of Chicago, I feel like that’s the biggest barrier because there’s pharmacies all over the place.” Regardless, for some patients, taking medications was their least favorite part of their hypertension care, even though it may not be as difficult for them: Patient: “I’m one of those people, don’t just give me pills. Give me some things to do. I don’t want to take all those pills if I don’t have to.” Patient: “I hate taking medicine. I travel a lot with friends, and it’s so funny because we go to breakfast, and we’re a bunch of old ladies pulling out these little pill bottles—I hate it.” ## Hesitancy Intensifying Medications by Both Clinicians and Patients Another common topic was intensifying antihypertensive medications. Patients often preferred sticking with lifestyle modifications and/or 1 medication, in the hopes of achieving blood pressure control. Clinicians felt that it was frequently more difficult to have conversations about additional medication than starting medication initially, in part because they feared pushback or losing engagement altogether: Patient: “He’s [doctor] been tryin’ to get me to change that medication for some time. I just was really tryin’ to get off of it before it’s time to go to another one.” Clinician: “It depends on age. If they’re young or elderly, it’s, ‘I don’t want be on a medicine.’ If in-between, sometimes there’s more toleration, but we’ll try something, they’ll get scared, won’t like a side effect, and then they’ll give up and I’ll lose them to follow-up.” When the choice to intensify treatment is made, clinicians provided several rationales when choosing treatments, although they relied largely on national guidelines, with some variations based on the perceived strength of evidence. Clinicians also described how they personalize their prescribing, focusing mainly on comorbidities or adverse effects when prescribing rather than factors such as sex or race and ethnicity: Clinician: “I don’t go with that whole Black man, White woman thing. For elderly, I’m going to use the same medication but lower doses. I try to avoid diuretics in older women, so they don’t go leaking urine all over.” Clinician: “You want to make sure that you’re optimizing someone’s therapy for that person. What medication you choose definitely matters depending on race, ethnicity, comorbidities. It’s never the same for everyone.” ## Varying Timing and Follow-up After Changes in Blood Pressure Medication Many patients expressed a preference for clinic-based interactions for hypertension, worrying that their connection with clinicians is otherwise insufficient. Some clinicians described specific approaches for follow-up, while others’ decisions were based on their perception of patients’ likelihood of following that plan. Follow-up patterns varied even by clinicians within the same clinic: Patient: “I had one virtual meeting, and after that, I said, ‘Hey, doc. I don’t like this. Next visit, I want to see you in person.’ Because he’s sitting right there. I can look him in the eyes. He can check my heartbeat and all that.” Clinician: “Some of it depends if it is a patient that is going to monitor their blood pressure at home. I’m usually aiming for like, 10 to 20 different readings, however they want to do that. If they don’t use the portal, I often have my nurse or medical assistant follow up in 2 to 3 weeks to review their blood pressures.” Clinicians had a modestly favorable view of the patient portal (ie, EHR-linked communication platform) to support follow-up but recognized that patient perceptions about and willingness to use the portal varied widely. Patients generally had strong reactions to the portal; some loved it, while others thought it was too complicated: Patient: “The portal is really cool. You can schedule virtual visits from there too. If it’s an emergency, like with my blood pressure, I call that number and she calls right back.” Patient: “I don’t like it. I’m not into all that computer, electronic stuff. I’m an old-school guy, and I’m not up-to-date on all this computer, electronic, media stuff now.” Clinician: “Usually, I’d have them come back in their usual 3-month visit. If they have their own cuff and since it’s COVID, I’ll have them check their blood pressures over the next month and send me stuff over the portal.” ## Variation in Blood Pressure Self-monitoring Recommendations and Uptake Patients and clinicians alike described variation in how often they used or recommended, respectively, home self-monitoring. Clinicians described varying success in obtaining values, including requesting patients to send them via the EHR portal, email, or telephone or requesting in-person blood pressure checks. Several clinicians recognized increased reliance on patient-reported data during the COVID-19 pandemic but found it difficult to direct care due to ranges in how data are collected. A common barrier was ensuring consistent or easy access to blood pressure monitors, particularly because clinicians had little knowledge about whether insurance would cover them. Clinicians expressed great interest in better guidance: Patient: “I would if I had extra money. I know I need to do it. It’s just, when can I spare that extra money?” Clinician: “So often, we ask patients to monitor at home, and you get a small chunk who will get a cuff and report data back. There is some confusion about the appropriate cuff. As a physician, I will, in our EHR, order a cuff, and I think, done. Right? Cuff’s been ordered. Well, was there coverage? There’s a lot of challenges I don’t even recognize about how one secures a cuff and if they secure an appropriate one. When patients come back and actually did A, B, and C, I think ‘Wow, you did?’” Physical barriers, such as blood pressure cuff size, were also repeatedly mentioned by patients and clinicians as something confusing. Standard cuffs were often too small for overweight patients. Patients also reported often preferring wrist cuffs, but clinicians vastly preferred arm cuffs, citing increased accuracy and data quality: Patient: “I don’t really check it at home because it’s really hard to get a cuff. They don’t fit when I order a small cuff, and I need a bigger cuff. I do take my blood pressure, but I don’t think it’s always accurate.” Clinician: “There’s such variability in the quality of cuffs available to the public, and some people buy a cuff that goes around your wrist, and I don’t trust those.” ## Limited Specific Functionality of Current Health Information Technology Tools Most clinicians did not describe any specific health information technology tools for hypertension that they routinely used or liked, such as those available in the EHR system. Of the range of suggestions, the most common were an interest in better tracking and observing longitudinal trends in care (eg, systolic blood pressure values and medications) embedded in the HER: Clinician: “I still have everyone giving me little pieces of paper I try to copy and get scanned. Then it goes into the media file, but I have to search. So, a clearer way to trend stuff, patients being able to input info easily that show me in a graph.” Regarding mobile health (mHealth) apps or other patient-directed education, clinicians expressed concerns about the quality of information and rarely recommended patient-facing apps. Instead, they preferred to provide written educational material, such as background information about hypertension to add to EHR-based after-visit summaries, but they expressed uncertainty about whether patients used the material because it was either too voluminous or untailored. Consequently, PCPs often described using self-created materials. Patients wanted better information and reported minimal mHealth use: Patient: “I wonder if information can be given to us as patients to help us to control our blood pressure and maybe eventually wean ourselves off medication. Something more proactive than writing a prescription.” Clinician: “I would love a little more streamlined, ‘This is the highest level of evidence’ summary sheet for sharing patient information.” ## Discussion In this qualitative study of patients and clinicians in a health system with a racially and ethnically diverse patient population, participants described a range of perspectives and experiences with hypertension care in primary care. Supporting patients with lifestyle modifications, blood pressure self-monitoring, and treatment intensification appeared to be key areas for potential improvement. Participants also expressed concerns about the functionality of health information technology to support asynchronous hypertension care, suggesting the need to design more effective strategies to efficiently improve quality within current constraints. In the US, most studies have focused on hypertension beliefs among racially and ethnically diverse patient populations rather than specific management practices, but they have elicited similar themes about wishing to avoid medication and preferring PCPs who focus on lifestyle modifications.18,19 Prior research also suggests that many patients, particularly in Black communities, expect hypertension to be cured.20 Other non-US studies have identified similar themes. Interviews in the Netherlands observed that patients did not prioritize hypertension and were interested in deprescribing antihypertensives.21 Clinicians in Australia indicated hesitancy about intensifying medications, especially for older adults.22 Patients with hypertension in Canada expressed similar interest in lifestyle options rather than medications.23 To our knowledge, few studies have focused on clinician and patient barriers to efficient care. These findings hold important implications for the design of interventions to improve blood pressure management. First, patients and clinicians believed that they are more challenged by lifestyle modifications compared with medication taking, although evidence suggests that adherence to antihypertensive medications is still suboptimal.24,25 Many noted that some lifestyle challenges were more prominent in Black and Hispanic or Latino communities. Thus, to be most acceptable to patients, adherence interventions may need to incorporate lifestyle recommendations to improve management.26,27 Second, blood pressure self-monitoring remains a source of confusion. Increasing interest in home-based hypertension strategies is underscored by inherent challenges in office-based hypertension management.10 Some effective options include interventions combining home blood pressure monitoring with medication titration protocols, although these interventions have not been extensively studied in racially and ethnically diverse communities.28 To enact these strategies, our study highlights some additional logistical challenges, including streamlining access to blood pressure cuffs, ensuring that PCPs and patients can obtain the most appropriate cuff for arm size, and value integration. Care should also be taken to ensure that devices are affordable, even though costs have been decreasing in recent years. Rates of blood pressure self-monitoring may also be lower in Black and Hispanic or Latino populations, due to differences in both how often self-monitoring is recommended by clinicians and how often it is done by patients.29 Third, gaps in treatment intensification were repeatedly described. Systematic medication titration protocols may help address this issue, but extensive study in multiethnic populations remains limited.30 Thus, interventions addressing concerns around treatment intensification, especially when increasing doses or adding new medications, seem necessary to address hesitancy. One avenue could be improving how EHR systems present longitudinal trends to illustrate need.31 Finally, gaps may also exist between the availability of patient-facing health information technology and uptake. Emerging literature remains mixed about mHealth apps and text messaging and lowering blood pressure32; these interventions appear most effective with 2-way communication with health care professionals.33,34 They have also been minimally studied in diverse populations.35 Our study suggests an implementation gap between their availability and their uptake, warranting further study. ## Limitations This study has several limitations. First, it was conducted in 2 clinics and with only 30 participants (although saturation was reached), which may limit generalizability, and further evaluation may be necessary. Second, while interviews were conducted by an external interviewer, some response bias may still be possible in response to the questions, especially because the interviewer was a pharmacist. Third, clinicians could exclude patients from interviews but focused on excluding those with cognitive difficulty. Fourth, interviews were conducted during the COVID-19 pandemic but should be generalizable to ongoing combinations of in-person and asynchronous care. ## Conclusions Participants in this qualitative study had heterogeneous experiences with hypertension care, even within the same clinic, and felt that more focus should be on lifestyle modifications. More intentional and tailored ways of supporting treatment intensification, self-care, and follow-up care may be needed to improve hypertension management in primary care practices serving diverse populations. ## References 1. Whelton PK, Carey RM, Aronow WS. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. J Am Coll Cardiol (2018.0) 71 e127-e248. DOI: 10.1016/j.jacc.2017.11.006 2. Gu A, Yue Y, Desai RP, Argulian E. Racial and ethnic differences in antihypertensive medication use and blood pressure control among US adults with hypertension: the National Health and Nutrition Examination Survey, 2003 to 2012. Circ Cardiovasc Qual Outcomes (2017.0) 10. DOI: 10.1161/CIRCOUTCOMES.116.003166 3. Khatib R, Glowacki N, Lauffenburger J, Siddiqi A. Race/ethnic differences in atherosclerotic cardiovascular disease risk factors among patients with hypertension: analysis from 143 primary care clinics. Am J Hypertens (2021.0) 34 948-955. DOI: 10.1093/ajh/hpab053 4. Howard G, Lackland DT, Kleindorfer DO. Racial differences in the impact of elevated systolic blood pressure on stroke risk. JAMA Intern Med (2013.0) 173 46-51. DOI: 10.1001/2013.jamainternmed.857 5. Ferdinand KC, Yadav K, Nasser SA. Disparities in hypertension and cardiovascular disease in Blacks: the critical role of medication adherence. J Clin Hypertens (Greenwich) (2017.0) 19 1015-1024. DOI: 10.1111/jch.13089 6. Milman T, Joundi RA, Alotaibi NM, Saposnik G. Clinical inertia in the pharmacological management of hypertension: a systematic review and meta-analysis. Medicine (Baltimore) (2018.0) 97. DOI: 10.1097/MD.0000000000011121 7. Chiu N, Lauffenburger JC, Franklin JM, Choudhry NK. Prevalence, predictors, and outcomes of both true- and pseudo-resistant hypertension in the Action to Control Cardiovascular Risk in Diabetes Trial: a cohort study. Hypertens Res (2021.0) 44 1471-1482. DOI: 10.1038/s41440-021-00739-6 8. Lauffenburger JC, Robinson JG, Oramasionwu C, Fang G. Racial/ethnic and gender gaps in the use of and adherence to evidence-based preventive therapies among elderly Medicare Part D beneficiaries after acute myocardial infarction. Circulation (2014.0) 129 754-763. DOI: 10.1161/CIRCULATIONAHA.113.002658 9. Manze M, Rose AJ, Orner MB, Berlowitz DR, Kressin NR. Understanding racial disparities in treatment intensification for hypertension management. J Gen Intern Med (2010.0) 25 819-825. DOI: 10.1007/s11606-010-1342-9 10. Haff N, Choudhry NK. Helping patients manage their own blood pressures: a strategy to address hypertension control in the United States. Circ Cardiovasc Qual Outcomes (2021.0) 14. DOI: 10.1161/CIRCOUTCOMES.121.008220 11. Brown AGM, Houser RF, Mattei J, Mozaffarian D, Lichtenstein AH, Folta SC. Hypertension among US-born and foreign-born non-Hispanic Blacks: National Health and Nutrition Examination Survey 2003-2014 data. J Hypertens (2017.0) 35 2380-2387. DOI: 10.1097/HJH.0000000000001489 12. 12American Association of Medical Colleges. The Complexities of Physician Supply and Demand: Projections From 2019 to 2034. Published June 2021. Accessed December 22, 2022. https://www.aamc.org/media/54681/download 13. O’Brien BC, Harris IB, Beckman TJ, Reed DA, Cook DA. Standards for reporting qualitative research: a synthesis of recommendations. Acad Med (2014.0) 89 1245-1251. DOI: 10.1097/ACM.0000000000000388 14. Lauffenburger JC, Khatib R, Siddiqi A. Reducing Ethnic and Racial Disparities by Improving Undertreatment, Control, and Engagement in Blood Pressure Management With Health Information Technology (REDUCE-BP) hybrid effectiveness-implementation pragmatic trial: rationale and design. Am Heart J (2023.0) 255 12-21. DOI: 10.1016/j.ahj.2022.10.003 15. Saunders B, Sim J, Kingstone T. Saturation in qualitative research: exploring its conceptualization and operationalization. Qual Quant (2018.0) 52 1893-1907. DOI: 10.1007/s11135-017-0574-8 16. Reis S, Biderman A, Mitki R, Borkan JM. Secrets in primary care: a qualitative exploration and conceptual model. J Gen Intern Med (2007.0) 22 1246-1253. DOI: 10.1007/s11606-007-0186-4 17. Borkan JM. Doing Qualitative Research (1999.0) 18. Buckley L, Labonville S, Barr J. A systematic review of beliefs about hypertension and its treatment among African Americans. Curr Hypertens Rep (2016.0) 18 52. DOI: 10.1007/s11906-016-0662-5 19. Kronish IM, Leventhal H, Horowitz CR. Understanding minority patients’ beliefs about hypertension to reduce gaps in communication between patients and clinicians. J Clin Hypertens (Greenwich) (2012.0) 14 38-44. DOI: 10.1111/j.1751-7176.2011.00558.x 20. Ogedegbe G, Mancuso CA, Allegrante JP. Expectations of blood pressure management in hypertensive African-American patients: a qualitative study. J Natl Med Assoc (2004.0) 96 442-449. PMID: 15101664 21. van Bussel E, Reurich L, Pols J, Richard E, Moll van Charante E, Ligthart S. Hypertension management: experiences, wishes and concerns among older people—a qualitative study. BMJ Open (2019.0) 9. DOI: 10.1136/bmjopen-2019-030742 22. Jansen J, McKinn S, Bonner C. General practitioners’ decision making about primary prevention of cardiovascular disease in older adults: a qualitative study. PLoS One (2017.0) 12. DOI: 10.1371/journal.pone.0170228 23. Jolles EP, Padwal RS, Clark AM, Braam B. A qualitative study of patient perspectives about hypertension. ISRN Hypertension. (2013.0) 2013. DOI: 10.5402/2013/671691 24. Abegaz TM, Shehab A, Gebreyohannes EA, Bhagavathula AS, Elnour AA. Nonadherence to antihypertensive drugs: a systematic review and meta-analysis. Medicine (Baltimore) (2017.0) 96. DOI: 10.1097/MD.0000000000005641 25. Choudhry NK, Kronish IM, Vongpatanasin W. Medication adherence and blood pressure control: a scientific statement from the American Heart Association. Hypertension (2022.0) 79 e1-e14. DOI: 10.1161/HYP.0000000000000203 26. Fletcher BR, Hartmann-Boyce J, Hinton L, McManus RJ. The effect of self-monitoring of blood pressure on medication adherence and lifestyle factors: a systematic review and meta-analysis. Am J Hypertens (2015.0) 28 1209-1221. DOI: 10.1093/ajh/hpv008 27. Choudhry NK, Isaac T, Lauffenburger JC. Effect of a remotely delivered tailored multicomponent approach to enhance medication taking for patients with hyperlipidemia, hypertension, and diabetes: the STIC2IT cluster randomized clinical trial. JAMA Intern Med (2018.0) 178 1182-1189. DOI: 10.1001/jamainternmed.2018.3189 28. McManus RJ, Little P, Stuart B. Home and Online Management and Evaluation of Blood Pressure (HOME BP) using a digital intervention in poorly controlled hypertension: randomised controlled trial. BMJ (2021.0) 372 m4858. DOI: 10.1136/bmj.m4858 29. Yi SS, Tabaei BP, Angell SY. Self-blood pressure monitoring in an urban, ethnically diverse population: a randomized clinical trial utilizing the electronic health record. Circ Cardiovasc Qual Outcomes (2015.0) 8 138-145. DOI: 10.1161/CIRCOUTCOMES.114.000950 30. Mills KT, Obst KM, Shen W. Comparative effectiveness of implementation strategies for blood pressure control in hypertensive patients: a systematic review and meta-analysis. Ann Intern Med (2018.0) 168 110-120. DOI: 10.7326/M17-1805 31. Samal L, Wright A, Wong BT, Linder JA, Bates DW. Leveraging electronic health records to support chronic disease management: the need for temporal data views. Inform Prim Care (2011.0) 19 65-74. DOI: 10.14236/jhi.v19i2.797 32. Morawski K, Ghazinouri R, Krumme A. Association of a smartphone application with medication adherence and blood pressure control: the MedISAFE-BP randomized clinical trial. JAMA Intern Med (2018.0) 178 802-809. DOI: 10.1001/jamainternmed.2018.0447 33. Lauffenburger JC, Choudhry NK. Text messaging and patient engagement in an increasingly mobile world. Circulation (2016.0) 133 555-556. DOI: 10.1161/CIRCULATIONAHA.116.021182 34. Alessa T, Hawley M, de Witte L. Identification of the most suitable app to support the self-management of hypertension: systematic selection approach and qualitative study. JMIR Mhealth Uhealth (2021.0) 9. DOI: 10.2196/29207 35. Alessa T, Hawley MS, Hock ES, de Witte L. Smartphone apps to support self-management of hypertension: review and content analysis. JMIR Mhealth Uhealth (2019.0) 7. DOI: 10.2196/13645
--- title: Gender, Racial, and Ethnic and Inequities in Receipt of Multiple National Institutes of Health Research Project Grants authors: - Mytien Nguyen - Sarwat I. Chaudhry - Mayur M. Desai - Kafui Dzirasa - Jose E. Cavazos - Dowin Boatright journal: JAMA Network Open year: 2023 pmcid: PMC9975935 doi: 10.1001/jamanetworkopen.2023.0855 license: CC BY 4.0 --- # Gender, Racial, and Ethnic and Inequities in Receipt of Multiple National Institutes of Health Research Project Grants ## Key Points ### Question What is the gender, racial, and ethnic diversity of elite National Institutes of Health investigators from 1991 to 2020? ### Findings In this cross-sectional study, while the number of principal investigators holding 3 or more research project grants increased 3-fold between 1991 and 2020, female and Black principal investigators were significantly underrepresented in this group, even after adjusting for career stage and degree. ### Meaning These results suggest that there is a growing funding gap among National Institutes of Health investigators, along with a persistent gender, race, and ethnic inequity among an elite class of SPIs. Consideration of the persistent gender, racial, and ethnic gaps in this elite class of investigators. ## Abstract This cross-sectional study of National Institutes of Health (NIH) grants examines trends in gender, racial, and ethnic diversity among principal investigators, with a particular focus on investigators holding 3 or more research project grants. ### Importance Diversity in the biomedical research workforce is essential for addressing complex health problems. Female investigators and investigators from underrepresented ethnic and racial groups generate novel, impactful, and innovative research, yet they are significantly underrepresented among National Institutes of Health (NIH) investigators. ### Objective To examine the gender, ethnic, and racial distribution of super NIH investigators who received 3 or more concurrent NIH grants. ### Design, Setting, and Participants This cross-sectional study included a national cohort of NIH-funded principal investigators (PIs) from the NIH Information for Management, Planning, Analysis, and Coordination (IMPAC II) database from 1991 to 2020. ### Exposures Self-identified gender, race and ethnicity, annual number of NIH grant receipt, career stage, and highest degree. ### Main Outcomes and Measures Distribution of investigators receiving 3 or more research project grants, referred to as super principal investigators (SPIs), by gender, race, and ethnicity. ### Results Among 33 896 investigators in fiscal year 2020, 7478 ($22.01\%$) identified as Asian, 623 ($1.8\%$) as Black, 1624 ($4.8\%$) as Hispanic, and 22 107 ($65.2\%$) as White; 21 936 ($61.7\%$) identified as men; and 8695 ($35.3\%$) were early-stage investigators. Between 1991 and 2020, the proportion of SPIs increased 3-fold from 704 ($3.7\%$) to 3942 ($11.3\%$). However, SPI status was unequal across gender, ethnic, and racial groups. Women and Black PIs were significantly underrepresented among SPIs, even after adjusting for career stage and degree, and were $34\%$ and $40\%$ less likely than their male and White colleagues, respectively, to be an SPI. Black women PIs were the least likely to be represented among SPIs and were $71\%$ less likely to attain SPI status than White men PIs (adjusted odds ratio, 0.29; $95\%$ CI, 0.21-0.41). ### Conclusions and Relevance In this cross-sectional study of a national cohort of NIH-funded investigators, the gender, ethnic, and racial gaps in receipt of multiple research project grants among NIH investigators was clearly apparent and warrants further investigation and interventions. ## Introduction Despite the benefits of diversity in scientific innovation, the distribution of National Institute of Health (NIH) funding has been historically disparate,1 with significant gender, racial, and ethnic inequalities in both NIH funding and success rate.2,3,4,5,6 In response, the NIH Working Group to the Advisory Committee and Director has made efforts in recent years to improve equity in NIH funding, leading to modest improvement in gender, racial, and ethnic representation among NIH investigators.1 However, little is known about gender, racial, and ethnic composition of principal investigators (PIs) who receive multiple NIH grants. Although holding 1 research project grant is indicative of career success,7 academic institutions are increasingly prioritizing the recruitment and retention of principal investigators who hold multiple research project grants.8 A faculty member’s overall portfolio of research project grants may influence key institutional decisions regarding recruitment, salary, tenure, promotion, and resource allocations, as well as national policy decisions on research funding.8,9 Despite the significant power and resources held by investigators with multiple, simultaneous research grants (hereafter referred to as super principal investigators [SPI]), gender, racial, and ethnic composition of SPIs is currently unknown. To evaluate gender, racial, and ethnic diversity of SPIs, we examined the distribution of SPIs over time using a national database of NIH investigators from 1991 to 2020. We also examined the likelihood of being an SPI by intersectional gender and racial or ethnic minority identity. ## Data Source Data were obtained for principal investigator–specific research project grants from the NIH Information for Management, Planning, Analysis, and Coordination (IMPAC II) database for fiscal years 1985 to 2020.1 The IMPAC II is a database maintained by the NIH and is provided with limited use. Research project grants included grants with the following activity codes: DP1, DP2, DP3, DP4, DP5, P01, PN1, PM1, R00, R01, R03, R15, R21, R22, R23, R29, R33, R34, R35, R36, R37, R61, R50, R55, R56, RC1, RC2, RC3, RC4, RF1, RL1, RL2, RL9, RM1, UA5, UC1, UC2, UC3, UC4, UC7, UF1, UG3, UH2, UH3, UH5, UM1, UM2, U01, U19, and U34. Grants awarded under the American Recovery and Reinvestment Act of 2009 (ARRA) and supplemental COVID-19 appropriations were excluded. Funding dollars were adjusted for inflation to 2019 US dollars using the Biomedical Research and Development Price Index. Patients or the public were not involved in the design, conduct, reporting, or dissemination plans of our research. Our analyses were conducted according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines and were deemed exempt by the Yale University institutional review board. ## Demographic Variables PIs’ gender, race, and ethnicity were self-reported by the faculty applying for NIH grant funding. PIs with unknown or withheld gender identity (14 291 [$1.8\%$]) were excluded from regression analyses. We examined trends in research project grants from 1991 to 2020 due to a significant portion of missing racial and ethnic data on investigators prior to 1991. Racial categories included American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or other Pacific Islander, White, more than 1 race, unknown, or withheld. Ethnicity categories included Hispanic, Not Hispanic, unknown, or withheld. Racial and ethnic identities were combined into the following categories: Asian, Black, Hispanic, White, and other (which included American Indian or Alaska Native, Native Hawaiian or other Pacific Islander, more than 1 race, unknown, or withheld). PIs who reported Hispanic ethnicity were categorized as Hispanic regardless of racial identity. ## Defining SPIs We defined SPIs as principal investigators holding 3 or more concurrent active research project grants in a given fiscal year (approximately the top $10\%$ of all NIH principal investigators in 2020). Unadjusted rates of being an SPI were determined by calculating the proportion of each gender, ethnic and racial, or intersectional group that were SPIs vs non-SPIs. The IMPAC II data set comprise both small and large dollars research project grants. To determine the robustness of utilizing number of grants to define SPI, we compared median and interquartile range (IQR) of total funding dollars received among SPIs and non-SPIs across demographic groups for fiscal year 2020. ## Statistical Analysis Nonparametric t tests and Kruskal-Wallis tests with posthoc Dunn correction for multiple comparisons were used to determine significance between median funding dollar amounts across groups. Multivariable logistic regression was used to determine the relative odds of women and investigators from underrepresented racial and ethnic groups of being an SPI compared with men and White investigators, respectively. Covariates include PI’s highest degree and career stage. Degree was defined as MD, MD/PhD, PhD, or other degrees. PI’s career stage was approximated using investigator’s age, categorized as early (age under 46 years), middle (age 46 to 58 years), and late (age above 58 years), as described previously.1 Finally, we included 3 time periods that delineate significant changes in the NIH budget: 1991-1998 (phase 1) before the first budget increase, 1999-2014 (phase 2) between the first and second budget increase, and 2015-2020 (phase 3) after the second budget increase, and tested the interaction between phase and gender, ethnic, and racial identities to determine whether the relative odds of being an SPI for disadvantaged groups (eg, women, Black, and Hispanic) have changed over time. Adjusted percentage of SPI investigators within each combined subgroup of gender, ethnic, and racial identity was determined from the fully adjusted logistic models. Statistical tests were 2-sided with type 1 error rate of 0.05. All analyses were performed using Stata version 16.1 (Stata Inc). ## Trends in NIH SPIs In fiscal year 1991, among 18 820 investigators, 1187 ($6.4\%$) identified as Asian, 100 ($0.5\%$) as Black, 320 ($1.7\%$) as Hispanic, and 14 630 ($77.7\%$) as White; 13 821 ($80.0\%$) identified as men; and 9124 ($52.8\%$) were early-stage investigators. In comparison, in fiscal year 2020, among 33 896 investigators, 7478 ($22.1\%$) identified as Asian, 623 ($1.8\%$) as Black, 1624 ($4.8\%$) as Hispanic, and 22 107 ($65.2\%$) as White; 21 936 ($61.7\%$) identified as men; and 8695 ($35.3\%$) were early-stage investigators (Table). Table. \| Characteristics \| Investigators, No. (%) \| Investigators, No. (%).1 \| Investigators, No. (%).2 \| P value \| \| --- \| --- \| --- \| --- \| --- \| \| Characteristics \| All (N = 33 896) \| Non-SPI (n = 29 989) \| SPI (n = 3907) \| P value \| \| Ethnicity and race \| \| \| \| \| \| Asian \| 7478 (22.1) \| 6523 (21.8) \| 955 (24.4) \| <.001 \| \| Black \| 623 (1.8) \| 588 (2.0) \| 35 (0.9) \| <.001 \| \| Hispanic \| 1624 (4.8) \| 1465 (4.9) \| 159 (4.1) \| <.001 \| \| Othera \| 2064 (6.1) \| 1861 (6.2) \| 203 (5.2) \| <.001 \| \| White \| 22 107 (65.2) \| 19 552 (65.2) \| 2555 (65.4) \| <.001 \| \| Gender \| \| \| \| \| \| Men \| 21 936 (64.7) \| 19 068 (63.6) \| 2868 (73.4) \| <.001 \| \| Women \| 11 960 (35.3) \| 10 921 (36.4) \| 1039 (26.6) \| <.001 \| \| Degree \| \| \| \| \| \| PhD \| 24 371 (71.9) \| 21 728 (72.5) \| 2643 (67.6) \| <.001 \| \| MD/PhD \| 3599 (10.6) \| 2999 (10.0) \| 600 (15.4) \| <.001 \| \| MD \| 5274 (15.6) \| 4623 (15.4) \| 651 (16.7) \| <.001 \| \| Other \| 652 (1.9) \| 639 (2.1) \| 13 (0.3) \| <.001 \| \| Career stage \| \| \| \| \| \| Early (<46 y) \| 10 386 (30.6) \| 9546 (31.8) \| 840 (21.5) \| <.001 \| \| Middle (46-58 y) \| 8395 (24.8) \| 7294 (24.3) \| 1101 (28.2) \| <.001 \| \| Late (>58 y) \| 12 969 (38.3) \| 11 228 (37.4) \| 1741 (44.6) \| <.001 \| \| Unknown \| 2146 (6.3) \| 1921 (6.4) \| 225 (5.8) \| <.001 \| \| Combined identity subgroups \| Combined identity subgroups \| Combined identity subgroups \| Combined identity subgroups \| Combined identity subgroups \| \| Asian \| \| \| \| <.001 \| \| Men \| 5088 (15.0) \| 4356 (14.5) \| 732 (18.7) \| <.001 \| \| Women \| 2390 (7.1) \| 2167 (7.2) \| 223 (5.7) \| <.001 \| \| Black \| \| \| \| <.001 \| \| Men \| 334 (1.0) \| 311 (1.0) \| 23 (0.6) \| <.001 \| \| Women \| 289 (0.9) \| 277 (0.9) \| 12 (0.3) \| <.001 \| \| Hispanic \| \| \| \| <.001 \| \| Men \| 1021 (3.0) \| 902 (3.0) \| 119 (3.0) \| <.001 \| \| Women \| 603 (1.8) \| 563 (1.9) \| 40 (1.0) \| <.001 \| \| Othera \| \| \| \| <.001 \| \| Men \| 1423 (4.2) \| 1268 (4.2) \| 155 (4.0) \| <.001 \| \| Women \| 641 (1.9) \| 593 (2.0) \| 48 (1.2) \| <.001 \| \| White \| \| \| \| <.001 \| \| Men \| 14 070 (41.5) \| 12 231 (40.8) \| 1839 (47.1) \| <.001 \| \| Women \| 8037 (23.7) \| 7321 (24.4) \| 716 (18.3) \| <.001 \| From 1991 to 2020, the total number of NIH PIs increased 1.8-fold from 18 820 to 34 936 (eFigure 1 in Supplement 1), which corresponds to a spending increase from $11.9 billion to $22.4 billion. Notably, there were 2 years in which the NIH budget increased substantially, in 1998 and 2015. We next examined trends in PIs holding multiple grants by summarizing the proportion of all PIs who held multiple active grants in a given fiscal year at 4 levels: PIs who held 2 or more active grants, 3 or more grants, 4 or more grants, or 5 or more grants (Figure 1). Since 1991, the proportion of PIs who held multiple grants has increased overall, with major inflection points occurring when there was an increase in the NIH budget in 1998 and 2015). Between 1991 and 2020, the percentage of PIs with 2 or more research project grants increased by $60\%$ (from $21.2\%$ to $33.2\%$), while the percentage of PIs with 3 or more research project grants increased 3-fold (from $3.7\%$ to $11.3\%$). The percentage of PIs with 4 or more research project grants increased more than 6-fold (from $0.6\%$ to $3.8\%$), and the percentage of PIs with 5 or more research project grants increased 10-fold (from $0.1\%$ to $1.2\%$). Figure 1.: Proportion of Principal Investigators (PIs) With Active Concurrent Research Project Grants From 1991-2020Super principal investigators (SPIs) included all PIs with 3 or more research project grants (approximately the top 10% of PIs in 2020). The number of PIs with 3 or more research project grants grew at a rate that outpaced the baseline increase in NIH investigators (3.0-fold vs 1.8-fold; $P \leq .001$). This SPI cohort of PIs with 3 or more concurrent active research project grants represented $10\%$ of all NIH-funded investigators in the past 5 years. Funding allocation to SPIs increased more than 2-fold from $12.7\%$ in 1991 to $28.0\%$ in 2020 (eFigure 2A in Supplement 1). In 2020, SPIs, who comprise $11.3\%$ of all PIs, received $28.0\%$ of federal NIH research funding, with a median (IQR) annual total research funding of $1.42 million ($1.08-$2.05 million) per PI compared with $0.38 million ($0.25-$0.62 million) per PI among non-SPI ($P \leq .001$) (eFigure 2B in Supplement 1). To determine the robustness of our findings, we performed a sensitivity analysis using total grant dollars across gender and ethnic and racial identities. Across all gender, ethnic, and racial groups, SPIs received a significant higher median (IQR) annual research project grant funding compared with non-SPIs (men: non-SPIs, $0.39 million [$0.26-$0.62 million] vs SPIs, $1.30 million [$0.95-$1.73 million]; women: non-SPIs, $0.38 million [$0.24-$0.61 million] vs SPIs, $1.30 million [$0.94-$1.74 million]; $P \leq .001$) (eFigure 3A and 3B in Supplement 1). There was no difference in annual median research funding dollars for men and women SPIs ($1.30 million for both men and women SPIs; $P \leq .99$), or across ethnic and racial groups ($1.30 million [$0.95-$1.77 million] for White SPIs, $1.20 million [$0.92-$1.61 million] for Asian SPIs, $1.20 million [$0.89-$1.70 million] for Hispanic SPIs, and $1.50 million [$1.00-$1.84 million] for Black SPIs; Dunn-corrected $P \leq .05$ for all comparisons) (eFigure 1B, eFigure 3 in Supplement 1). ## Representation of Women and Black NIH Investigators Among SPIs Next, we examined gender composition of SPIs (Figure 2A and Figure 2B). In 1991, $2.1\%$ of women and $4.4\%$ of men were SPIs. By 2020, these numbers increased to $8.7\%$ and $13.1\%$ for women and men, respectively (Figure 2A; eTable 1 in Supplement 1). After adjusting for degree and career stage, women PIs had $40\%$, $38\%$, and $34\%$ lower odds than men to attain SPI status in phases 1, 2, and 3, respectively (Figure 2B; eTable 1 in Supplement 1). While the interaction analysis between time phase and gender revealed that the relative disadvantage of women attaining SPI status has diminished over time ($$P \leq .003$$), women continue to have significantly lower odds of being an SPI than men. Figure 2.: Gender, Ethnic, and Racial Diversity Among SPIsError bars indicate 95% CIs. In panel B, odds ratios adjusted for career stage (early, middle, and late) and degree; in panel D, odds adjusted for career stage and degree. There was similar inequity in ethnic and racial representation among SPIs. In 1992, $4.1\%$, $4.0\%$, $4.4\%$, and $1.0\%$ of White, Asian, Hispanic, and Black PIs, respectively, were SPIs (Figure 2C). Although these proportions increased for all ethnic and racial groups over time, the proportion of Black PIs having SPI status remained significantly lower than White PIs in phase 3 ($5.6\%$ Black vs $11.5\%$ White; $P \leq .001$). Interaction analysis between time phase and ethnic and racial identity indicated that the relative odds of being an SPI for PIs of color (eg, Asian, Hispanic, and Black) significantly changed over time ($P \leq .001$). In phase 1, after adjusting for degree and career stage, compared with White PIs, Asian PIs were as likely to be an SPI (adjusted odds ratio [aOR], 1.03; $95\%$ CI, 0.94-1.13), and Black and Hispanic PIs were less likely to be an SPI (Black: aOR, 0.28; $95\%$ CI, 0.16-0.47; Hispanic: aOR, 0.79; $95\%$ CI, 0.66-0.95) (Figure 2D; eTable 1 in Supplement 1). By phase 3, Asian PIs were significantly more likely than White PIs to be an SPI (aOR, 1.07; $95\%$ CI, 1.03-1.11), while Hispanic PIs were as likely as White PIs to be an SPI (aOR, 0.92; $95\%$ CI, 0.84-1.00). Although the likelihood of Black PIs having SPI status increased over time, Black PIs were still half as likely as White PIs to be SPIs in phase 3 (aOR, 0.51; $95\%$ CI, 0.42-0.61) (Figure 2D; eTable 1 in Supplement 1). ## Black Women PIs Least Likely to Be SPIs We examined the intersections between gender and ethnic and racial identity among SPIs over time. Between 1991 and 2020, the proportion of SPIs among White, Asian, and Hispanic men increased at a higher rate compared with Black men and all women (eFigure 4 in Supplement 1). In 2020, while $13.1\%$ of White men PIs were SPIs, only $6.8\%$ and $4.1\%$ of Black men and women PIs were SPIs, respectively ($P \leq .001$). In phase 1, after adjusting for degree and career stage, compared with White men, Asian and Hispanic men were as likely to be an SPI (Asian men PIs: aOR, 1.09; $95\%$ CI, 0.99-1.20; Hispanic men PIs: aOR, 0.84; $95\%$ CI, 0.70-1.02) while Black men were significantly less likely to be an SPI (aOR, 0.33; $95\%$ CI, 0.19-0.58) (eFigure 4, eTable 2 in Supplement 1). Compared with White men PIs, all women PIs across ethnic and racial groups were less likely to be an SPI in phase 1, with Black women PIs having the largest disadvantage (White women PIs: aOR, 0.63; $95\%$ CI, 0.58-0.68; Asian women PIs: aOR, 0.42; $95\%$ CI, 0.32-0.55; Hispanic women PIs: 0.26; $95\%$ CI, 0.13-0.52; Black women PIs: 0.05; $95\%$ CI, 0-0.39) (eFigure 4, eTable 2 in Supplement 1). By phase 3, compared with White men, Asian men were more likely (aOR, 1.08; $95\%$ CI, 1.03-1.13), and Hispanic men were as likely (aOR, 0.95; $95\%$ CI, 0.86-1.04) to be an SPI, while Black men remained less likely to be an SPI (aOR, 0.55; $95\%$ CI, 0.44-0.67) (Figure 3). Compared with White men, White, Asian, Hispanic and Black women were $33\%$, $29\%$, $43\%$, and $71\%$ less likely to be an SPI in phase 3. Although interaction analysis revealed that the relative odds of being an SPI improved over time for Black men and for women across all ethnic and racial groups ($P \leq .001$), these groups remained significantly less likely than White men to be an SPI in phase 3 (Figure 3). Remarkably, even though the relative odds of Black women being an SPI improved after the first NIH budget increase (phase 1: aOR, 0.05; $95\%$ CI, 0.00-0.39; phase 2: aOR, 0.34; $95\%$ CI, 0.26-0.44), these odds have not changed after the most recent NIH budget increase (aOR, 0.29; $95\%$ CI, 0.21-0.41) (Figure 3). Figure 3.: Proportion of SPIs by Gender, Ethnic, and Racial Intersectional SubgroupsError bars indicates 95% CIs. Odds ratios were adjusted for career stage and degree. ## Discussion In this study, we found that an elite class of principal investigators who held 3 or more grants grew in number over the past 30 years. The proportion of SPIs among all PIs increased 3-fold from $3.7\%$ in 1991 to $11.3\%$ in 2020. Moreover, the compositional diversity of SPIs was not equitable across gender and ethnic and racial groups. Even after adjusting for career stage and degree, women and Black PIs were significantly less likely to have SPI status compared with White PIs. Black women were most disparately underrepresented among SPIs, with White men PIs being more than 3-fold more likely to be an SPI compared with Black women. The rise in the percentage of SPIs among all investigators and the concurrent ethnic and racial disparity among SPIs is concerning. Despite evidence of diminishing returns on investment for PIs receiving greater than $600 000 per year in funding,10,11 data suggest that NIH dollars are increasingly concentrated among a small proportion of investigators. This led NIH leadership to consider capping NIH funding to 3 R01-equivalent grants per investigator,12 although this policy was never implemented across NIH institutes and centers. Given the well-documented benefits of diversity among investigative teams, ethnic and racial disparities among PIs and SPIs could limit scientific impact and innovation,13,14,15,16 posing a substantial threat to the success of the US biomedical research enterprise. While the cause of the gender, ethnic, and racial gap in SPI status reported in this study is likely multifactorial, disparities in mentorship available to Black and women faculty may contribute to this gap.17 Mentorship not only guides early career faculty on a path to success but also exposes faculty to a network of peers that will facilitate collaborations and support.18,19,20 Black and women scientists are less likely than White and men scientists to be mentored by high impact senior mentors,21 and therefore less likely to acquire the scientific network, tacit knowledge, and sponsorship that are inherently required for securing grants. Furthermore, even when mentored by senior faculty, bias and racism may affect the relationship that Black and women faculty have with their mentors, resulting in negative mentoring that harms women and faculty of color.22,23,24 Patterns of grant submission may also influence the disparities in SPI status by gender, race, and ethnicity described in this study. Higher frequency grant submission and resubmission have been linked to funding success,25 and prior studies have reported that women, in aggregate, and Black faculty submit fewer grants than their counterparts.26,27 Investment in both early and mid-career meaningful mentorship initiatives for Black and women faculty will be essential to improve funding longevity and reduce the inequitable ethnic and racial distribution of NIH funding allocated to first-time PIs and among more established SPIs.28 Such programs may include expansion of diversity supplements for early-career faculty, developing mentoring networks for female and Black faculty,29,30,31 and incentivizing diverse team-based science through additional emphasis in program grants and multiple principal investigator awards. While programs to support grant submission and resubmission are critical, this intervention will likely remain insufficient to address the disparities reported in this study as prior research has shown that Black faculty receive lower scores and are less likely to be funded after grant resubmission than their White counterparts, even after controlling for training record, prior award, and publication history.26,27 Therefore, women and faculty from underrepresented ethnic and racial groups face disparity at 2 levels—initial application32 and reapplication26,27—suggesting a worrying trend of inequitable resource allocation at all career stages. These data suggest that structural interventions may be necessary to address bias in grant assessment, such as diversifying members of NIH study sections and program staff,33 which in 2021 was $33.8\%$ female, $2.3\%$ Black, and $4.5\%$ Hispanic.34 In addition to increasing diversity in study sections, the NIH could consider promoting bias training and education among reviewers. Despite historical struggles with diversity among PIs, the NIH has introduced several interventions to address funding disparities in recent years. In 2021, the NIH established the UNITE initiative to address systemic racism in the NIH and biomedical researcher workforce.35 The committees operating under the UNITE initiative have made considerable efforts to increase workforce diversity and reduce disparities in research funding, such as increasing funding to NIH institutes that receive a higher percentage of applications from female and faculty from underrepresented ethnic and racial groups (eg, National Institute on Minority Health and Health Disparities), increasing support to minority-serving institutions, and implementation of the Faculty Institutional Recruitment for Sustainable Transformation (FIRST) funding opportunity to support the recruitment of diverse faculty cohorts.36 Moving forward, the UNITE initiative and the NIH could consider structural changes in the grant review process to assess the ethnic and racial diversity of investigators listed in grant applications. Diverse teams are more innovative and produce higher quality research than homogenous teams,13,15,37 and including an investigator team diversity score could represent an evidence-based metric to promote high-impact science. Additionally, the NIH could incorporate an assessment of the institutional climate of equity and inclusion as a component of a grant application’s scoring criteria. Measures of equity and inclusion could include the institution’s compositional faculty diversity and equity in promotion and salary. Other structural reforms may include changes to timing of funding deadlines. Request for Application (RFA) and Funding Opportunity Announcement (FOA) for NIH grants are often released shortly before the submission deadline. Women, Hispanic, and Black scientists are less likely to have access to strong research networks and mentorship22,23 and are frequently overtasked with unpaid and unrewarded administrative duties that can be detrimental to their research and career success,38,39 including meeting grant submission deadlines. Providing more time between FOA and RFA release and submission deadline would allow women and faculty from underrepresented ethnic and racial groups time and resources to build and submit their grant applications. ## Limitations Our study had several limitations. This study examined the likelihood of being an SPI given that an investigator had already received NIH funding and does not account for submission behavior differences across demographic groups nor the environmental support (such as research funding available at the institution). These factors may affect the gender, racial, and ethnic disparities described and are important to examine to inform future policies and interventions. Furthermore, our study included funding to contact principal investigators and does not include delineation of whether a grant has multiple PIs. The multiple PI approach is critical for team-based science and can play an important role to improve diversity, as well as mentoring for young women and underrepresented faculty. In addition, a small percentage of PIs withheld their ethnic or racial identity and some ethnic and racial groups, such as Alaska Native, Native American, Hawaiian Native, and Pacific Islanders, were too small for analysis. The small number of Indigenous investigators reflected systemic marginalization across biomedicine and society. More attention should be paid to promote and enhance biomedical research funding to Indigenous investigators, as well as researchers with other marginalized identities such as socioeconomic disadvantage and faculty with disability.40 Lastly, this study focused on NIH-funded investigators and have limited generalizability to other federal and nonfederal funding agencies, such as the National Science Foundation,41 for which further study is needed. ## Conclusions In this cross-sectional study of NIH investigators from 1991 to 2020, we found a growing gap among NIH investigators that created a cohort of highly funded NIH investigators. Importantly, there were persistent gender, ethnic, and racial inequities among this elite class of SPIs. As the NIH develops critical initiatives and reforms to promote equity among its investigators, consideration of the persistent gender and ethnic and racial gaps in this elite class and the influence they have is critical for meaningful reform. ## References 1. Lauer MS, Roychowdhury D. Inequalities in the distribution of National Institutes of Health research project grant funding. eLife (2021.0). DOI: 10.7554/eLife.71712 2. Andriole DA, Yan Y, Jeffe DB. Mediators of racial/ethnic disparities in mentored K award receipt among U.S. medical school graduates. Acad Med (2017.0) 92 1440-1448. DOI: 10.1097/ACM.0000000000001871 3. Ginther DK, Kahn S, Schaffer WT. Gender, race/ethnicity, and National Institutes of Health R01 research awards: is there evidence of a double bind for women of color?. Acad Med (2016.0) 91 1098-1107. DOI: 10.1097/ACM.0000000000001278 4. Ley TJ, Hamilton BH. Sociology—the gender gap in NIH grant applications. Science (2008.0) 322 1472-1474. DOI: 10.1126/science.1165878 5. Stevens KR, Masters KS, Imoukhuede PI. Fund Black scientists. Cell (2021.0) 184 561-565. DOI: 10.1016/j.cell.2021.01.011 6. Ward HB, Levin FR, Greenfield SF. Disparities in gender and race among physician-scientists: a call to action and strategic recommendations. Acad Med (2022.0) 97 487-491. PMID: 34192723 7. Jacob BA, Lefgren L. The impact of research grant funding on scientific productivity. J Public Econ (2011.0) 95 1168-1177. DOI: 10.1016/j.jpubeco.2011.05.005 8. Mbuagbaw L, Anderson LN, Lokker C, Thabane L. Advice for junior faculty regarding academic promotion: what not to worry about, and what to worry about. J Multidiscip Healthc (2020.0) 13 117-122. DOI: 10.2147/JMDH.S240056 9. Peifer M. Call to restore NIH’s cap on grant funding. Science (2017.0) 357 364. DOI: 10.1126/science.aao2443 10. Lauer M, Roychowdhury D, Patel K, Walsh R, Pearson K. Marginal returns and levels of research grant support among scientists supported by the National Institutes of Health.. bioRxiv (2017.0). DOI: 10.1101/142554 11. Lorsch JR. Maximizing the return on taxpayers’ investments in fundamental biomedical research. Mol Biol Cell (2015.0) 26 1578-1582. DOI: 10.1091/mbc.E14-06-1163 12. Lauer M. (2017.0) 13. Freeman RB, Huang W. Collaboration: strength in diversity. Nature (2014.0) 513 305. DOI: 10.1038/513305a 14. Valantine HA, Lund PK, Gammie AE. From the NIH: a systems approach to increasing the diversity of the biomedical research workforce. CBE Life Sci Educ (2016.0) 15 fe4. DOI: 10.1187/cbe.16-03-0138 15. Nielsen MW, Bloch CW, Schiebinger L. Making gender diversity work for scientific discovery and innovation. Nature Human Behaviour (2018.0) 2 726-734. DOI: 10.1038/s41562-018-0433-1 16. Hofstra B, Kulkarni VV, Munoz-Najar Galvez S, He B, Jurafsky D, McFarland DA. The diversity-innovation paradox in science. Proc Natl Acad Sci U S A (2020.0) 117 9284-9291. DOI: 10.1073/pnas.1915378117 17. Yu H, Willis KA, Litovitz A. The effect of mentee and mentor gender on scientific productivity of applicants for NIH training fellowships.. bioRxiv (2021.0). DOI: 10.1101/2021.02.02.429450 18. Vishwanatha J, Pfund C, Ofili E, Okuyemi K. NIH’s mentoring makes progress. Science (2016.0) 354 840-841. DOI: 10.1126/science.aal1898 19. Sood A, Tigges B, Helitzer D. Mentoring early-career faculty researchers is important—but first “train the trainer.”. Academic Medicine (2016.0) 91 1598-1600. DOI: 10.1097/ACM.0000000000001264 20. Beech BM, Bruce MA, Thorpe RJ, Heitman E, Griffith DM, Norris KC. Theory-informed research training and mentoring of underrepresented early-career faculty at teaching-intensive institutions: the obesity health disparities PRIDE program. Ethn Dis (2018.0) 28 115-122. DOI: 10.18865/ed.28.2.115 21. Bova B. Mentoring revisited: the Black woman's experience. Mentoring & Tutoring (2000.0) 8 5-16. DOI: 10.1080/713685511 22. Cole ER, McGowan BL, Zerquera DD. First-year faculty of color: narratives about entering the academy. Equity & Excellence in Education (2017.0) 50 1-12. DOI: 10.1080/10665684.2016.1262300 23. Eby LT, McManus SE, Simon SA, Russell JEA. The protege's perspective regarding negative mentoring experiences: the development of a taxonomy. J Vocational Behavior (2000.0) 57 1-21. DOI: 10.1006/jvbe.1999.1726 24. Davis TM, Jones MK, Settles IH, Russell PG. Barriers to the successful mentoring of faculty of color. J Career Dev (2021.0) 49. DOI: 10.1177/08948453211013375 25. Haggerty PA, Fenton MJ. Outcomes of early NIH-funded investigators: experience of the National Institute of Allergy and Infectious Diseases. PLoS One (2018.0) 13. DOI: 10.1371/journal.pone.0199648 26. Hechtman LA, Moore NP, Schulkey CE. NIH funding longevity by gender. Proc Natl Acad Sci U S A (2018.0) 115 7943-7948. DOI: 10.1073/pnas.1800615115 27. Ginther Donna K, Schaffer Walter T, Schnell J. Race, ethnicity, and NIH research awards. Science (2011.0) 333 1015-1019. DOI: 10.1126/science.1196783 28. Enders FT, Golembiewski EH, Orellana MA. Changing the face of academic medicine: an equity action plan for institutions. J Clin Transl Sci (2022.0) 6. DOI: 10.1017/cts.2022.408 29. DeCastro R, Sambuco D, Ubel PA, Stewart A, Jagsi R. Mentor networks in academic medicine: moving beyond a dyadic conception of mentoring for junior faculty researchers. Acad Med (2013.0) 88 488-496. DOI: 10.1097/ACM.0b013e318285d302 30. Hill KA, Desai MM, Chaudhry SI, Nguyen M, Boatright D. NIH diversity supplement awards by year and administering institute. JAMA (2021.0) 326 2427-2429. DOI: 10.1001/jama.2021.19360 31. Hill KA, Desai MM, Chaudhry SI, Fancher T. National Institutes of Health diversity supplement awards by medical school. J Gen Intern Med (2022.0). DOI: 10.1007/s11606-022-07849-y 32. Taffe MA, Gilpin NW. Racial inequity in grant funding from the US National Institutes of Health. eLife (2021.0). DOI: 10.7554/eLife.65697 33. Volerman A, Arora VM, Cursio JF, Wei H, Press VG. Representation of women on National Institutes of Health study sections. JAMA Netw Open (2021.0) 4. DOI: 10.1001/jamanetworkopen.2020.37346 34. 34National Institutes of Health. CSR Data & Evaluations. Center for Scientific Review. Accessed June 24, 2022. https://public.csr.nih.gov/AboutCSR/Evaluations#reviewer_demographics 35. 35National Institutes of Health. UNITE—Milestones & Progress. Accessed June 24, 2022. https://www.nih.gov/ending-structural-racism/unite-milestones-progress 36. Bernard MA, Johnson AC, Hopkins-Laboy T, Tabak LA. The US National Institutes of Health approach to inclusive excellence. Nature Medicine (2021.0) 27 1861-1864. DOI: 10.1038/s41591-021-01532-1 37. Adams J. Collaborations: the fourth age of research. Nature (2013.0) 497 557-560. DOI: 10.1038/497557a 38. Trejo J. The burden of service for faculty of color to achieve diversity and inclusion: the minority tax. Mol Biol Cell (2020.0) 31 2752-2754. DOI: 10.1091/mbc.E20-08-0567 39. Williamson T, Goodwin CR, Ubel PA. Minority tax reform—avoiding overtaxing minorities when we need them most. N Engl J Med (2021.0) 384 1877-1879. DOI: 10.1056/NEJMp2100179 40. Swenor BK, Munoz B, Meeks LM. A decade of decline: grant funding for researchers with disabilities 2008 to 2018. PLoS One (2020.0) 15. DOI: 10.1371/journal.pone.0228686 41. Chen CY, Kahanamoku SS, Tripati A. Systemic racial disparities in funding rates at the National Science Foundation. eLife (2022.0). DOI: 10.7554/eLife.83071
--- title: Metabolomic analysis-identified 2-hydroxybutyric acid might be a key metabolite of severe preeclampsia authors: - Fang Wang - Lili Xu - Mingming Qi - Huimin Lai - Fanhua Zeng - Furong Liang - Qing Wen - Xihua Ma - Chan Zhang - Kaili Xie journal: Open Life Sciences year: 2023 pmcid: PMC9975955 doi: 10.1515/biol-2022-0572 license: CC BY 4.0 --- # Metabolomic analysis-identified 2-hydroxybutyric acid might be a key metabolite of severe preeclampsia ## Abstract This study set out to determine the key metabolite changes underlying the pathophysiology of severe preeclampsia (PE) using metabolic analysis. We collected sera from 10 patients with severe PE and from 10 healthy pregnant women of the same trimester and analyzed them using liquid chromatography mass spectrometry. A total of 3,138 differential metabolites were screened, resulting in the identification of 124 differential metabolites. Kyoto encyclopedia of genes and genomes pathway analysis revealed that they were mainly enriched in the following metabolic pathways: central carbon metabolism in cancer; protein digestion and absorption; aminoacyl-transfer RNA biosynthesis; mineral absorption; alanine, aspartate, and glutamate metabolism; and prostate cancer. After analysis of 124 differential metabolites, 2-hydroxybutyric acid was found to be the most critical differential metabolite, and its use allowed the differentiation of women with severe PE from healthy pregnant women. In summary, our analysis revealed that 2-hydroxybutyric acid is a potential key metabolite for distinguishing severe PE from healthy controls and is also a marker for the early diagnosis of severe PE, thus allowing early intervention. ## Introduction Preeclampsia (PE) is defined as a new onset of hypertension during pregnancy after the 20th week [1], and its global prevalence is $8\%$ [2]. Worldwide, PE and eclampsia are major causes of maternal and infant mortality [3]. In developing countries, it occurs at a rate of 1.8–$16.7\%$ [4] and causes 40–$60\%$ of maternal deaths [4]. Its rates have remained unchanged for decades, but the rates of severe PE have increased over recent decades [5]. Eclampsia can occur in patients with severe PE leading to symptoms of the nervous system [6] or hemolysis, elevated liver enzymes, low platelet count (HELLP) syndrome [7]. Depending on the clinical characteristics of a patient, PE can be classified as mild or severe [8]. Furthermore, it can be classified according to its time of clinical manifestation as “early-onset PE” (EOPE) in cases occurring before 34 weeks of pregnancy, or as “late-onset PE” (LOPE) in cases occurring after 34 weeks of pregnancy [9]. EOPE and LOPE might be more useful subclassifications [10]. The majority of the affected women suffers from PE at the late preterm or term stage, but about $12\%$ suffers from PE that begins early (before 34 weeks of pregnancy) [11]. EOPE is the result of placental defects and deficiency of trophoblast invasion and normal spiral artery remodeling; LOPE, however, may result from interactions between the normal senescence of the placenta and a maternal genetic history of cardiovascular disease [12,13]. Blood pressure (BP) control is crucial during PE to prevent systemic complications [14]. Therefore, early diagnosis and early intervention in PE are particularly important. There are many established risk factors for PE, such as nulliparity, advanced maternal age, overweight or obesity, chronic hypertension, diabetes, previous PE, family history of PE, and multiple pregnancy [15]; however, the exact causes of PE/eclampsia remain unclear [16]. Further research is needed concerning the cellular and molecular mechanisms of PE to improve the treatment of PE patients [17]. The study of metabonomics and metabolomics involves the use of accurate metabonomic (and/or metabolomic) analyses of metabolic changes occurring in cells, tissues, and whole organisms [18]. It is part of the “omics cascade” together with genomics, transcriptomics, and proteomics [19] and one of the many “-omics” technologies that are currently being developed [20]. Metabolomics, or metabonomics, primarily involves the elucidation of the end products in a specific organism or a cell [21], and it is the “ultimate” tool in the “omics chain,” as it is the closest to the phenotype [22]. It can be divided into two categories: untargeted and targeted [23]. While the former also known as discovery metabolomics, which is a global analysis of different metabolomics between control and experimental groups, the latter focuses on the analysis of specific metabolic clusters associated with certain metabolic pathways [20]. 2-Hydroxybutyric acid is elevated in many diseases and has some diagnostic values. In cancer, its levels have been elevated in a mouse model of colon carcinogenesis induced by azoxymethane/dextran sodium sulfate [24]; furthermore, nuclear lactate dehydrogenase A induces its production from reactive oxygen species and promotes human papilloma virus-induced cervical tumor growth [25]; moreover, patients in the initial diagnostic stage of acute myeloid leukemia can be identified by 2-hydroxybutyric acid [26]. In pneumonia diseases, compared to healthy controls, 2-hydroxybutyric acid was found to be enriched in COVID-19 patients and COVID-like patients and remained at higher levels after discharge [27]; meanwhile, elevated dehydrogenase can be an independent prognostic factor for death in hospitalized COVID-19 patients [28]; it also has a diagnostic value in community-acquired pneumonia [29]. 2-Hydroxybutyric acid has been more deeply studied in diabetes than in any other disease, and high levels of plasma are a good predictor of type 2 diabetes [30]. Furthermore, it can be used in the following circumstances: as a biomarker of insulin resistance; for disease tracking throughout the treatment of insulin resistance [31]; as a predictive marker for impaired glucose tolerance without the need for a glucose tolerance test [32]; along with branched-chain amino acids, it can predict worsening glycemic control in adolescents [33]; it has high values in the sera of patients with isolated postchallenge diabetes compared to normal subjects [34]; and its levels decrease significantly after laparoscopic sleeve gastrectomy in morbidly obese patients [35]. It also have some diagnostic values in other diseases: it is significantly higher in the blood of pregnant women carrying trisomy 21 fetuses than in healthy pregnant women [36]; urinary 2-hydroxybutyric acid predicts the development of acute kidney injury in presurgical samples [37]; and its levels are elevated in patients with a major depressive disorder [38]. In this study, we collected sera from 10 patients with severe PE and 10 healthy pregnant women of the same trimester and analyzed them using liquid chromatograph mass spectrometry (LC-MS), with a view to identifying the key metabolites of the former pathogenesis and providing new indicators for early diagnosis. ## Study population All samples were obtained from the Obstetrics Department of Zhuzhou Central Hospital and were divided into a normal control group (10 normotensive pregnant women) and a severe PE group ($$n = 10$$). The basic diagnostic criteria for PE are as follows: BP ≥$\frac{140}{90}$ mmHg, and urine protein ≥0.3 mg/24 h. Severe PE is diagnosed on the basis of the diagnostic criteria for PE with any of the following conditions present: systolic BP ≥160 mmHg, or diastolic BP ≥110 mmHg, or other manifestations of a multisystem disorder (e.g., severe proteinuria, thrombocytopenia, impaired liver function, severe persistent right upper quadrant or epigastric pain, renal insufficiency, pulmonary edema, or new-onset headache). ## Sample processing A peripheral blood sample (10 mL) was taken from each participant. All samples were centrifuged at 2,000 rpm for 10 min at room temperature using a centrifuge. Afterward, the supernatant was stored in a refrigerator at −80°C. ## Spectroscopy All samples were thawed at 4°C (insufficient samples were reduced to an equal scale); 100 µL of each sample was transferred into 2 mL centrifuge tubes (samples with a sample size of <50 µL were extracted by half of the experimental system, but the resolution system remained unchanged); 400 µL of methanol (−20°C) was added to each tube and vortexed for 60 s; the mixture was centrifuged at 4°C for 10 min at 12,000 rpm, and then the supernatant was transferred from each sample into another 2 mL centrifuge tube. Samples were concentrated to dry in a vacuum and subsequently dissolved with 150 µL 2-chlorobenzalanine (4 ppm) $80\%$ methanol solution, and the supernatant was filtered through a 0.22 µm membrane to obtain the prepared samples for gas chromatography mass spectrometry (GC-MS). For quality control (QC), 20 µL subsamples were taken (QC samples were used to monitor deviations of the analytical results from these pool mixtures and compare them to the errors caused by the analytical instrument itself). The remainder of the samples were used for LC-MS detection ## Chromatography and mass spectrometry conditions Chromatographic separation was performed with an ACQUITY UPLC® HSS T3 (150 mm × 2.1 mm, 1.8 µm, Waters) column maintained at 40°C. The temperature of the autosampler was 8°C. Gradient elution of analytes was carried out with $0.1\%$ formic acid in water and $0.1\%$ formic acid in acetonitrile or 5 mM ammonium formate in water and acetonitrile at a flow rate of 0.25 mL/min. Injection of 2 μL of each sample was performed after equilibration. An increasing linear gradient of solvent B (v/v) was used as follows: 0–1 min, $2\%$ B/D; 1–9 min, 2–$50\%$ B/D; 9–12 min, 50–$98\%$ B/D; 12–13.5 min, $98\%$ B/D; 13.5–14 min, 98–$2\%$ B/D; 14–20 min, $2\%$ D-positive model (14–17 min, $2\%$ B-negative model). The electrospray ionization multistage mass spectrometry experiments were used with a spray voltage of 3.5 and −2.5 kV in positive and negative modes, respectively. Sheath gas and auxiliary gas were set at 30 and 10 arbitrary units, respectively, while the capillary temperature was 325°C. The Orbitrap analyzer scanned over a mass range of m/z 81–1,000 for a full scan at a mass resolution of 70,000. Data-dependent acquisition MS/MS experiments were performed with a higher energy collisional dissociation scan. The normalized collision energy was 30 eV. Dynamic exclusion was implemented to remove some unnecessary information in MS/MS spectra. ## Multivariate statistical analysis The data were analyzed using SIMCA-P (v13.0) [35] software and the R language ropls [39] package. The main methods of analysis included principal component analysis (PCA), partial least squares-discriminant analysis (PLS-DA), and orthogonal partial least squares discriminant analysis (OPLS-DA) [40]. Unsupervised analysis (e.g., PCA) does not ignore within-group errors, eliminates random errors that are not relevant to the purpose of the study, focuses too much on details, and neglects the overall picture and patterns, and is ultimately detrimental to the detection of between-group differences and differential compounds. In such cases, it is necessary to use prior knowledge of the sample to further focus the data analysis on the aspect being studied, using a supervised analysis such as PLS-DA. OPLS-DA, another commonly used method in metabolomics data analysis, is an extension of PLS-DA. Compared to the PLS-DA, this method can effectively reduce the complexity of the model and enhance the explanatory power of the model without reducing the predictive power, thus maximizing the differences between groups. ## Differential metabolite screening Metabolites are screened to identify differential metabolites (biomarkers); the relevant conditions are as follows: p-value ≤0.05 + VIP (variable importance for the projection) ≥1. ## Identification of metabolites Metabolite identification was first confirmed on the basis of precise molecular weights (molecular weight error <30 ppm), followed by confirmation of annotation against the Metlin (http://metlin.scripps.edu) and MoNA (https://mona.fiehnlab.ucdavis.edu//) databases based on MS/MS fragmentation patterns to identify the final metabolites. ## Network analysis Metscape [41], a Cytoscape plug-in (v.3.9.0) [42], was used for the metabolic network analysis and data visualization. By building an association-based network, we found that 2-hydroxybutyric acid was important within this metabolic network (Figure 4). Furthermore, the results of the differential metabolite analysis showed that 2-hydroxybutyric acid differed significantly between preeclamptic and normal pregnant women (VIP = 1.670, p-value <0.05, false discovery rate (FDR) = 0.006, log FC = 1.016). Meanwhile, the results of analysis of variance for l-threonine and 5,6-dihydro-5-fluorouracil were VIP = 1.695, p-value <0.05, FDR = 0.022, log FC = 1.036; and VIP = 1.654, p-value <0.05, FDR = 0.006, log FC = 1.020, respectively. In the ROC plot (Figure 5), there was an area under the curve (AUC) of 0.99, indicating a high level of accuracy (high accuracy = AUC >0.9). Figure 4: Network analysis of identified differential metabolites. 2-Hydroxybutyric acid was identified as the key metabolite in a network analysis based on VIP and FDR values. Figure 5: ROC plot of 2-hydroxybutyric acid and the concentrations of 2-hydroxybutyric acid in PE group and normal group. (a) There was an AUC of 0.99. (b) The concentrations of 2-hydroxybutyric acid is 47559849.097 ± 14839964.376 in the PE group and 23509331.240 ± 4802006.390 in the normal group. ## Kyoto encyclopedia of genes and genomes (KEGG) analysis MetPA is part of metaboanalyst (www.metaboanalyst.ca) and is based on the KEGG metabolic pathway. The MetPA database identifies possible bioturbated metabolic pathways through metabolic pathway enrichment and topology analysis, and thus analyzes the metabolic pathways of metabolites. The MetPA database allows the analysis of metabolic pathways associated with two sets of differential metabolites, using a hypergeometric test as the data analysis algorithm, and relative-betweeness centrality for pathway topology. ## Subject characteristics Ten healthy pregnant women of the same trimester and 10 pregnant women with severe PE were included in this study. There were no significant differences between the two groups when subject characteristics such as age (31.6 ± 5.77) or week of gestation (32.1 ± 1.72 W) were taken into account. However, there were significant differences in BP, gestational week at termination, neonatal Apgar score, and neonatal weight. Table 1 shows the clinical features of the 10 severe PE cases. Table 1 \| NO \| Age (years) \| Gestational weeks \| Systolic pressure (mmHg) \| Edema \| 24 h urine protein (g/L) \| Albumin \| Ultrasound of pleural and peritoneal effusion \| Time to terminate pregnancy \| Neonatal Apgar Scores \| Neonatal weight (g) \| Echocard-iography \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| 1 \| 25 \| 33 + 3 \| 116–151 \| Normal \| 11.120 \| 22.6 g/L \| Normal \| 34 weeks \| 8–10 \| 1600 \| Normal \| \| 2 \| 32 \| 29 + 5 \| 109–161 \| (+) \| 0.603 \| Normal \| Normal \| 30 + 6 \| 9–10 \| 1270 \| Normal \| \| 3* \| 31 \| 31 + 5 \| 100–139 \| (+) \| 0.378 \| 25.2 g/L \| Normal \| 35 + 1 \| 9–10 \| 2,160 and 2,420 (twins) \| Normal \| \| 4 \| 29 \| 32 + 3 \| 120–160 \| Normal \| 1.229 \| Normal \| Normal \| Leave hospital \| Leave hospital \| Leave hospital \| Normal \| \| 5 \| 33 \| 31 + 2 \| 150–175 \| Normal \| Leave hospital \| Leave hospital \| Normal \| Leave hospital \| Leave hospital \| Leave hospital \| Normal \| \| 6 \| 45 \| 32 + 6 \| 107–166 \| Normal \| 5.292 \| 24 g/L \| Normal \| 32 + 6 \| 7–10 \| 1250 \| Normal \| \| 7 \| 31 \| 29 \| 130–170 \| Normal \| 6.134 \| Normal \| Normal \| 29 + 2 \| 0 \| 1050 \| ＜5 mL \| \| 8 \| 24 \| 32 + 6 \| 122–161 \| (++++) \| 4.772 \| 27.2 g/L \| Pleural effusion \| 33 + 2 \| 9–10 \| 1640 \| Normal \| \| 9 \| 34 \| 33 + 3 \| 110–147 \| (++) \| 0.454 \| 25.1 g/L \| Pleural effusion \| 33 + 3 \| 8–10 \| 1,720 and 1,790 (twins) \| Normal \| \| 10 \| 34 \| 34 + 3 \| 118–175 \| (+++) \| 1.742 \| 26.5 g/L \| Pleural effusion \| 34 + 6 \| 8–10 \| 1,920 and 1,890 (twins) \| Normal \| ## Multivariate analysis In this study, serum samples from the 10 healthy pregnant women and 10 severe PE patients were analyzed with a metabolomics approach, using GC-MS followed by the multivariate data analysis by PCA, PLS-DA, and OPLS-DA (Table 2). Initial unsupervised PCA (Figure 1a) and supervised PCA (Figure 1b and c) showed a clear separation of metabolites between the normal and severe EOPE groups. ## Identified metabolites In the GC-MS analysis, 3138 differential metabolites were screened and 124 metabolites (Figure 2) were eventually identified. Of the 124 identified differential metabolites, 45 downregulated products and 79 upregulated products were included. The classification of the identified differential metabolites are summarized in Figure 3; these were mainly located on metabolism of amino acids, carbohydrates, cofactors and vitamins, lipids, nucleotides, and peptides. Figure 2: Heatmap with all the significant metabolites. Figure 3: Classification of identified differential metabolites. The differential metabolites located on metabolism of amino acids, carbohydrates, cofactors and vitamins, lipids, nucleotides, and peptides. ## KEGG analysis The final results of the differential metabolite KEGG pathway enrichment analysis are shown in Figure 6. Significantly enriched pathways included central carbon metabolism in cancer, protein digestion and absorption, aminoacyl-transfer RNA (tRNA) biosynthesis, mineral absorption, alanine, aspartate and glutamate metabolism, and prostate cancer (FDR <0.05). Figure 6: KEGG pathway analysis of identified differential metabolites. Significantly enriched pathways include central carbon metabolism in cancer (37 differential metabolites); protein digestion and absorption (47 differential metabolites); aminoacyl-tRNA biosynthesis (52 differential metabolites); mineral absorption (29 differential metabolites); alanine, aspartate, and glutamate metabolism (28 differential metabolites); prostate cancer (11 differential metabolites) (FDR <0.05). The size of the circle represents the number of differential metabolites that are enriched in this pathway. −log(p): Negative values for the natural logarithm of the p-value. Impact: Metabolic pathway impact values. ## Discussion In this study, we analyzed the metabolites in the sera of 10 severe PE cases and 10 healthy pregnant women. A total of 3,138 differential metabolites were screened, resulting in the identification of 124 differential metabolites. After analysis of 124 differential metabolites, 2-hydroxybutyric acid was found to be the more critical differential metabolite; its presence clearly distinguished between severe PE and healthy pregnant women. The analysis of the KEGG pathway revealed that the metabolites were mainly enriched in the following metabolic pathways: central carbon metabolism in cancer; protein digestion and absorption; aminoacyl-tRNA biosynthesis; mineral absorption; alanine, aspartate, and glutamate metabolism; prostate cancer. 2-Hydroxybutyric acid is predominantly produced during the metabolism of l-threonine or the synthesis of glutathione and may be elevated by oxidative stress or the detoxification of exogenous substances in the liver [43]. It has been previously demonstrated that 2-hydroxybutyric acid, as a component of a metabolite-only model, can predict the EOPE [44]. Our study also found that this metabolite played an important role in severe PE; however, we used a different approach. Two of the KEGG enrichment analyses differential metabolites that we studied were associated with cancer. The other four KEGG pathways were all found to be associated with severe PE. The differential mRNAs between the preeclamptic and normal groups were also found to be enriched in the protein digestion and absorption pathways [45]. Thus, previous studies have found that l-arginine supplementation can be used to treat individuals with PE [46]. Harville et al. demonstrated that aminoacyl-tRNA biosynthesis is associated with hypertensive disorders of pregnancy [47]. Furthermore, the relationship between minerals and PE has been relatively well studied. Serum selenium levels have been shown to be associated with PE in several studies [48–51]; however, amniotic fluid selenium status has been shown to be uncorrelated with PE or preterm delivery [52,53]. On the other hand, Enebe et al. found that low levels of antioxidant trace elements, for example, selenium, copper, and magnesium, can promote the incidence of PE [54], and other studies have found that mineral and vitamin supplementation can reduce the incidence [55,56]. These findings illustrate the impact of mineral absorption pathways on PE. To our knowledge, there is no literature addressing the relationship between the alanine, aspartate, and glutamate metabolism pathway and PE. However, a proportion of patients with PE do have abnormal liver function. In summary, our analysis revealed that 2-hydroxybutyric acid might be a key metabolite for distinguishing severe PE from normal controls and is potentially a marker for the early diagnosis of severe PE, thus providing a basis for early detection and intervention. ## References 1. Deer E, Jones J, Cornelius DC, Comley K, Herrock O, Campbell N. Progesterone induced blocking factor reduces hypertension and placental mitochondrial dysfunction in response to sFlt-1 during pregnancy. Cells (2021 Oct 20) 10 2817. DOI: 10.3390/cells10112817 2. Tianthong W, Phupong V. Serum hypoxia-inducible factor-1α and uterine artery Doppler ultrasound during the first trimester for prediction of preeclampsia. Sci Rep (2021 Mar 23) 11 6674. DOI: 10.1038/s41598-021-86073-w 3. Daniels K, Lewin S. Translating research into maternal health care policy: a qualitative case study of the use of evidence in policies for the treatment of eclampsia and pre-eclampsia in South Africa. Health Res Policy Syst (2008 Dec 17) 6 12. DOI: 10.1186/1478-4505-6-12 4. Birhanu MY, Temesgen H, Demeke G, Assemie MA, Alamneh AA, Desta M. Incidence and predictors of pre-eclampsia among pregnant women attending antenatal care at debre markos referral hospital, North West Ethiopia: Prospective cohort study. Int J Womens Health (2020 Nov 10) 12 1013-21. DOI: 10.2147/IJWH.S265643 5. Eddy AC, Bidwell GL, George EM. Pro-angiogenic therapeutics for preeclampsia. Biol Sex Differ (2018 Aug 25) 9 36. DOI: 10.1186/s13293-018-0195-5 6. Ge J, Wang J, Xue D, Zhu Z, Chen Z, Li X. Why does a high-fat diet induce preeclampsia-like symptoms in pregnant rats. Neural Regen Res (2013 Jul 15) 8 1872-80. DOI: 10.3969/j.issn.1673-5374.2013.20.006 7. Savage AR, Hoho L. Knowledge of pre-eclampsia in women living in Makole Ward, Dodoma, Tanzania. Afr Health Sci (2016 Jun) 16 412-9. DOI: 10.4314/ahs.v16i2.9 8. Tang Y, Ye W, Liu X, Lv Y, Yao C, Wei J. VEGF and sFLT-1 in serum of PIH patients and effects on the foetus. Exp Ther Med (2019 Mar) 17 2123-8. DOI: 10.3892/etm.2019.7184 9. Tarca AL, Romero R, Benshalom-Tirosh N, Than NG, Gudicha DW, Done B. The prediction of early preeclampsia: Results from a longitudinal proteomics study. PLoS One (2019 Jun 4) 14 e0217273. DOI: 10.1371/journal.pone.0217273 10. Blair JD, Yuen RK, Lim BK, McFadden DE, von Dadelszen P, Robinson WP. Widespread DNA hypomethylation at gene enhancer regions in placentas associated with early-onset pre-eclampsia. Mol Hum Reprod (2013 Oct) 19 697-708. DOI: 10.1093/molehr/gat044 11. Wen YH, Yang HI, Chou HC, Chen CY, Hsieh WS, Tsou KI. Association of maternal preeclampsia with neonatal respiratory distress syndrome in very-low-birth-weight infants. Sci Rep (2019 Sep 13) 9 13212. DOI: 10.1038/s41598-019-49561-8 12. Park HJ, Kim SH, Jung YW, Shim SS, Kim JY, Cho YK. Screening models using multiple markers for early detection of late-onset preeclampsia in low-risk pregnancy. BMC Pregnancy Childbirth (2014 Jan 20) 14 35. DOI: 10.1186/1471-2393-14-35 13. Morikawa M, Kato-Hirayama E, Mayama M, Saito Y, Nakagawa K, Umazume T. Glycemic control and fetal growth of women with diabetes mellitus and subsequent hypertensive disorders of pregnancy. PLoS One (2020 Mar 16) 15 e0230488. DOI: 10.1371/journal.pone.0230488 14. Vizitiu R, Krauss-Grignard M, Garcia V, Valentin L, Samain E, Diemunsch P. Urapidyl for hypertension control in severe pre-eclampsia: a comparative study with nicardipine. Crit Care (2010) 14 P136. DOI: 10.1186/cc8368 15. Mrema D, Lie RT, Østbye T, Mahande MJ, Daltveit AK. The association between pre pregnancy body mass index and risk of preeclampsia: a registry based study from Tanzania. BMC Pregnancy Childbirth (2018 Feb 21) 18 56. DOI: 10.1186/s12884-018-1687-3 16. Grum T, Seifu A, Abay M, Angesom T, Tsegay L. Determinants of pre-eclampsia/Eclampsia among women attending delivery Services in Selected Public Hospitals of Addis Ababa, Ethiopia: a case control study. BMC Pregnancy Childbirth (2017 Sep 15) 17 307. DOI: 10.1186/s12884-017-1507-1 17. Roland CS, Hu J, Ren CE, Chen H, Li J, Varvoutis MS. Morphological changes of placental syncytium and their implications for the pathogenesis of preeclampsia. Cell Mol Life Sci (2016 Jan) 73 365-76. DOI: 10.1007/s00018-015-2069-x 18. Wang J, Zhou L, Lei H, Hao F, Liu X, Wang Y. Simultaneous quantification of amino metabolites in multiple metabolic pathways using ultra-high performance liquid chromatography with tandem-mass spectrometry. Sci Rep (2017 May 3) 7 1423. DOI: 10.1038/s41598-017-01435-7 19. Cruickshank-Quinn C, Zheng LK, Quinn K, Bowler R, Reisdorph R, Reisdorph N. Impact of blood collection tubes and sample handling time on serum and plasma metabolome and lipidome. Metabolites (2018 Dec 4) 8 88. DOI: 10.3390/metabo8040088 20. Liang Y, Ke X, Xiao Z, Zhang Y, Chen Y, Li Y. Untargeted metabolomic profiling using UHPLC-QTOF/MS reveals metabolic alterations associated with autism. Biomed Res Int (2020 Sep 11) 2020 6105608. DOI: 10.1155/2020/6105608 21. Liu M, Tang L, Liu X, Fang J, Zhan H, Wu H. An evidence-based review of related metabolites and metabolic network research on cerebral ischemia. Oxid Med Cell Longev (2016) 2016 9162074. DOI: 10.1155/2016/9162074 22. Utermann C, Parrot D, Breusing C, Stuckas H, Staufenberger T, Blümel M. Combined genotyping, microbial diversity and metabolite profiling studies on farmed Mytilus spp. from Kiel Fjord. Sci Rep (2018 May 22) 8 7983. DOI: 10.1038/s41598-018-26177-y 23. Wang D, Shi C, Ge ZH, Wei YX, Liu TT, Wang Y. Study of the mechanism of action of Guanxin Shutong capsules in the treatment of coronary heart disease based on metabolomics. Front Pharmacol (2021 Mar 25) 12 650438. DOI: 10.3389/fphar.2021.650438 24. Lin R, Piao M, Song Y, Liu C. Quercetin suppresses AOM/DSS-induced colon carcinogenesis through its anti-inflammation effects in mice. J Immunol Res (2020 May 21) 2020 9242601. DOI: 10.1155/2020/9242601 25. Liu Y, Guo JZ, Liu Y, Wang K, Ding W, Wang H. Nuclear lactate dehydrogenase A senses ROS to produce α-hydroxybutyrate for HPV-induced cervical tumor growth. Nat Commun (2018 Oct 24) 9 4429. DOI: 10.1038/s41467-018-06841-7 26. Kim HK, Son SY, Oh JS, Song YN, Byun JM, Koh Y. Metabolic profiling during acute myeloid leukemia progression using paired clinical bone marrow serum samples. Metabolites (2021 Aug 31) 11 586. DOI: 10.3390/metabo11090586 27. Bruzzone C, Bizkarguenaga M, Gil-Redondo R, Diercks T, Arana E, García de Vicuña A. SARS-CoV-2 infection dysregulates the metabolomic and lipidomic profiles of serum. iScience (2020 Oct 23) 23 101645. DOI: 10.1016/j.isci.2020.101645 28. Liu Z, Li J, Li M, Chen S, Gao R, Zeng G. Elevated α-hydroxybutyrate dehydrogenase as an independent prognostic factor for mortality in hospitalized patients with COVID-19. ESC Heart Fail (2021 Feb) 8 644-51. DOI: 10.1002/ehf2.13151 29. Zhou B, Lou B, Liu J, She J. Serum metabolite profiles as potential biochemical markers in young adults with community-acquired pneumonia cured by moxifloxacin therapy. Sci Rep (2020 Mar 10) 10 4436. DOI: 10.1038/s41598-020-61290-x 30. Li X, Xu Z, Lu X, Yang X, Yin P, Kong H. Comprehensive two-dimensional gas chromatography/time-of-flight mass spectrometry for metabonomics: Biomarker discovery for diabetes mellitus. Anal Chim Acta (2009 Feb 9) 633 257-62. DOI: 10.1016/j.aca.2008.11.058 31. Sousa AP, Cunha DM, Franco C, Teixeira C, Gojon F, Baylina P. Which role plays 2-hydroxybutyric acid on insulin resistance?. Metabolites (2021 Dec 3) 11 835. DOI: 10.3390/metabo11120835 32. Cobb J, Eckhart A, Motsinger-Reif A, Carr B, Groop L, Ferrannini E. α-Hydroxybutyric acid is a selective metabolite biomarker of impaired glucose tolerance. Diabetes Care (2016 Jun) 39 988-95. DOI: 10.2337/dc15-2752 33. Tricò D, Prinsen H, Giannini C, de Graaf R, Juchem C, Li F. Elevated α-hydroxybutyrate and branched-chain amino acid levels predict deterioration of glycemic control in adolescents. J Clin Endocrinol Metab (2017 Jul 1) 102 2473-81. DOI: 10.1210/jc.2017-00475 34. Chou J, Liu R, Yu J, Liu X, Zhao X, Li Y. Fasting serum α-hydroxybutyrate and pyroglutamic acid as important metabolites for detecting isolated post-challenge diabetes based on organic acid profiles. J Chromatogr B Anal Technol Biomed Life Sci (2018 Nov 15) 1100–1101 6-16. DOI: 10.1016/j.jchromb.2018.09.004 35. Yoshida N, Kitahama S, Yamashita T, Hirono Y, Tabata T, Saito Y. Metabolic alterations in plasma after laparoscopic sleeve gastrectomy. J Diabetes Investig (2021 Jan) 12 123-9. DOI: 10.1111/jdi.13328 36. Nemutlu E, Orgul G, Recber T, Aydin E, Ozkan E, Turgal M. Metabolic infrastructure of pregnant women with trisomy 21 fetuses; Metabolomic analysis. Z Geburtshilfe Neonatol (2019 Oct) 223 297-303. DOI: 10.1055/a-0877-7869 37. Martin-Lorenzo M, Ramos-Barron A, Gutierrez-Garcia P, Martin-Blazquez A, Santiago-Hernandez A, Rodrigo Calabia E. Urinary spermidine predicts and associates with in-hospital acute kidney injury after cardiac surgery. Antioxid (Basel) (2021 Jun 2) 10 896. DOI: 10.3390/antiox10060896 38. Pu J, Liu Y, Zhang H, Tian L, Gui S, Yu Y. An integrated meta-analysis of peripheral blood metabolites and biological functions in major depressive disorder. Mol Psychiatry (2021 Aug) 26 4265-76. DOI: 10.1038/s41380-020-0645-4 39. Boulesteix AL, Strimmer K. Partial least squares: a versatile tool for the analysis of high-dimensional genomic data. Brief Bioinforma (2007) 8 32-44 40. Thevenot EA, Roux A, Xu Y, Ezan E, Junot C. Analysis of the human adult urinary metabolome variations with age, body mass index and gender by implementing a comprehensive workflow for univariate and OPLS statistical analyses. J Proteome Res (2015) 14 3322-35. PMID: 26088811 41. Karnovsky A, Weymouth T, Hull T, Tarcea VG, Scardoni G, Laudanna C. Metscape 2 bioinformatics tool for the analysis and visualization of metabolomics and gene expression data. Bioinformatics (2012 Feb 1) 28 373-80. DOI: 10.1093/bioinformatics/btr661 42. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res (2003 Nov) 13 2498-504. DOI: 10.1101/gr.1239303 43. Shi D, Yan R, Lv L, Jiang H, Lu Y, Sheng J. The serum metabolome of COVID-19 patients is distinctive and predictive. Metabolism (2021 May) 118 154739. DOI: 10.1016/j.metabol.2021.154739 44. Bahado-Singh RO, Syngelaki A, Akolekar R, Mandal R, Bjondahl TC, Han B. Validation of metabolomic models for prediction of early-onset preeclampsia. Am J Obstet Gynecol (2015 Oct) 213 530.e1-530.e10. DOI: 10.1016/j.ajog.2015.06.044 45. He J, Liu K, Hou X, Lu J. Identification and validation of key non-coding RNAs and mRNAs using co-expression network analysis in pre-eclampsia. Med (Baltim) (2021 Apr 9) 100 e25294. DOI: 10.1097/MD.0000000000025294 46. Wu G, Meininger CJ, McNeal CJ, Bazer FW, Rhoads JM. Role of L-arginine in nitric oxide synthesis and health in humans. Adv Exp Med Biol (2021) 1332 167-87. DOI: 10.1007/978-3-030-74180-8_10 47. Harville EW, Li YY, Pan K, McRitchie S, Pathmasiri W, Sumner S. Untargeted analysis of first trimester serum to reveal biomarkers of pregnancy complications: a case-control discovery phase study. Sci Rep (2021) 11 3468. DOI: 10.1038/s41598-021-82804-1 48. Ghaemi SZ, Forouhari S, Dabbaghmanesh MH, Sayadi M, Bakhshayeshkaram M, Vaziri F. A prospective study of selenium concentration and risk of preeclampsia in pregnant Iranian women: a nested case-control study. Biol Trace Elem Res (2013 May) 152 174-9. DOI: 10.1007/s12011-013-9614-y 49. Haque MM, Moghal MM, Sarwar MS, Anonna SN, Akter M, Karmakar P. Low serum selenium concentration is associated with preeclampsia in pregnant women from Bangladesh. J Trace Elem Med Biol (2016 Jan) 33 21-5. DOI: 10.1016/j.jtemb.2015.08.002 50. Maleki A, Fard MK, Zadeh DH, Mamegani MA, Abasaizadeh S, Mazloomzadeh S. The relationship between plasma level of Se and preeclampsia. Hypertens Pregnancy (2011) 30 180-7. DOI: 10.3109/10641950903322931 51. Eze SC, Ododo NA, Ugwu EO, Enebe JT, Onyegbule OA, Eze IO. Serum selenium levels of pre-eclamptic and normal pregnant women in Nigeria: A comparative study. PLoS One (2020 Aug 27) 15 e0238263. DOI: 10.1371/journal.pone.0238263 52. Roy AC, Ratnam SS, Karunanithy R. Amniotic fluid selenium status in pre-eclampsia. Gynecol Obstet Invest (1989) 28 161-2. DOI: 10.1159/000293557 53. Ogrizek Pelkič K, Sobočan M, Takač I. Low selenium levels in amniotic fluid correlate with small-for-gestational age newborns. Nutrients (2020 Oct 5) 12 3046. DOI: 10.3390/nu12103046 54. Enebe JT, Dim CC, Ugwu EO, Enebe NO, Meka IA, Obioha KC. Serum antioxidant micronutrient levels in pre-eclamptic pregnant women in Enugu, south-East Nigeria: a comparative cross-sectional analytical study. BMC Pregnancy Childbirth (2020 Jul 6) 20 392. DOI: 10.1186/s12884-020-03081-w 55. Hovdenak N, Haram K. Influence of mineral and vitamin supplements on pregnancy outcome. Eur J Obstet Gynecol Reprod Biol (2012 Oct) 164 127-32. DOI: 10.1016/j.ejogrb.2012.06.020 56. Omotayo MO, Dickin KL, O’Brien KO, Neufeld LM, De Regil LM, Stoltzfus RJ. Calcium supplementation to prevent preeclampsia: Translating guidelines into practice in low-income countries. Adv Nutr (2016 Mar 15) 7 275-8. DOI: 10.3945/an.115.010736
--- title: 'General practice management of COPD patients following acute exacerbations: a qualitative study' authors: - Bianca Perera journal: The British Journal of General Practice year: 2023 pmcid: PMC9975965 doi: 10.3399/BJGP.2022.0342 license: CC BY 4.0 --- # General practice management of COPD patients following acute exacerbations: a qualitative study ## Abstract ### Background Exacerbations are the strongest risk factor for future exacerbations for patients living with chronic obstructive pulmonary disease (COPD). The period immediately following exacerbation is a high-risk period for recurrence and hospital admission, and is a critical time to intervene. GPs are ideally positioned to deliver this care. ### Aim To explore perceptions of GPs regarding the care of patients following exacerbations of COPD and to identify factors affecting the provision of evidence-based care. ### Design and setting A descriptive qualitative study was undertaken involving semi-structured, in-depth interviews with Australian GPs who volunteered to participate following a national survey of general practice care for COPD patients following exacerbations. ### Method Interviews were conducted via the Zoom video conference platform, which were audio-recorded and transcribed verbatim. QSR NVivo was used to support data management, coding, and inductive thematic analysis. ### Results Eighteen GPs completed interviews. Six key themes were identified: 1) GPs’ perceptions and knowledge in the management of COPD patients following exacerbation and admission to hospital; 2) pharmacological management; 3) consultation time; 4) communication between healthcare professionals; 5) access to other health services; and 6) patient compliance. ### Conclusion Delivery of post-exacerbation care to COPD patients is affected by GPs, patients, and health service-related factors. The care of COPD patients may be further improved by supporting GPs to overcome identified barriers. ## INTRODUCTION Acute exacerbations of COPD (AECOPDs) are important events that involve sustained worsening of symptoms that are beyond normal day-to-day variations and necessitate change in regular medication.1–3 Exacerbations are problematic as they are associated with rapid loss of lung function, increased healthcare expenditure related to hospital admissions, and reduced survival.4,5 Exacerbation prevention is therefore a fundamental priority of good COPD management.6 Of concern, exacerbations are the strongest risk factor for future exacerbations and become more frequent as the disease progresses.7 Recovery from exacerbations is frequently incomplete and may be associated with increased airway inflammation, suggesting it is important to monitor patients after AECOPD.8,9 More than $50\%$ of hospitalised patients are estimated to readmit within 12 months, with the highest risk being the 3 months immediately following discharge.10 Therefore, the period immediately after an exacerbation is a critical time to intervene to prevent future exacerbations. Several hospital-based care models, including clinical pathways and discharge care bundles, have been proposed to improve readmission risk in people with COPD following hospitalisation.11–13 An important issue that is not addressed in the literature is the effective transition of care from acute to community-based primary care. In Australia, where the current study was conducted, the majority of COPD care is delivered by GPs. They can deliver comprehensive post-exacerbation care for COPD patients, provide continuity of care, coordination, and integration of care with other health professionals and services. Evidence-based guidelines, such as Australia and New Zealand’s guidelines for the management of COPD (COPD-X), clearly describe what needs to be actioned following an AECOPD.14 Although exacerbations are one of the leading causes of preventable hospital admissions, little evaluation of the provision of care and application of evidence-based guidelines has been undertaken in general practice settings. ## Study design A qualitative descriptive study was undertaken as part of a larger mixed-methods project of GP management of COPD exacerbations in Australia. The scope of the interviews was informed by the COPD-X guidelines and responses to a nation-wide survey of GPs regarding general practice care of COPD patients following exacerbations. ## Recruitment Australian GPs who completed the national survey regarding GP care for people following acute exacerbations were invited to participate. This involved purposive distribution of surveys to practising GPs across all Australian states and territories who provided care for patients following AECOPDs in the previous 12 months. At the end of the survey, a convenience sample was established from participants who self-nominated to undertake a subsequent qualitative interview to further explore their personal experiences of care provision for this patient group. ## Data collection In-depth interviews were conducted via the Zoom video conference platform between April and October 2021 by the lead author, who is a practising GP and PhD candidate. Interviews were conducted at a time convenient to the participants. A semi-structured interview guide was developed following the approach of Minichiello et al.15 This approach involved a recursive model of interviewing, using a combination of closed and open-ended questions to explore GP perceptions on post-exacerbation care of hospitalised COPD patients and factors impacting the provision of evidence-based care in line with guideline recommendations (Box 1). Throughout, the interview participants were encouraged to share their perceptions and experiences of caring for patients following AECOPD. Digital recordings of interviews were transcribed verbatim by a professional transcriptionist service. Field notes and a reflective research journal were kept as part of the audit trail and to support the reflexive thematic analysis. After the first three interviews, transcripts and interviewer reflections were reviewed by the co-authors to refine the interview guide and provide feedback on interview technique. All participants were reimbursed with a A$250 gift voucher for associated lost clinical time and were given the opportunity to review their unedited transcripts to make clarifications or expand on what they had described in the interview. Repeat interviews were not needed but three participants returned their transcripts with minor changes. Recruitment ended after 18 interviews. At this point, no new concepts were emerging from interviews and the authors considered this to be a sufficient sample size to address the study aims. ## Data analysis QSR NVivo (Version 12) was used to support data management and coding. Transcripts were read multiple times by the first author prior to coding and thematic analysis using the 6-step approach to reflexive thematic analysis described by Braun et al 2019: familiarising with data, generating initial codes, searching for themes, reviewing themes, defining and naming themes, and producing the report.16 A preliminary coding scheme was developed by the first author based on an initial review of transcripts. The initial coding scheme was refined following further discussion, questioning, and probing by co-authors. A coding scheme was finalised and the complete set of transcripts was re-coded. A thematic map and charting were used to explore and understand relationships between concepts; they were also used to identify and clarify meaning-based patterns and features across the dataset. Several meetings were held to further refine and define candidate themes to clarify the scope and ‘core’ of each theme prior to final confirmation of themes. ## Participant characteristics Twenty-two GPs expressed interest to participate in a qualitative interview, of which one later declined and three did not respond to email or telephone communications. The 18 participants (6 male, 12 female) had varied experience in general practice ranging from 1.5 to 34 years and practised across a representative spread of Australian states and regions (Table 1). Mean interview duration was 32 minutes (range 22 to 46 minutes). Table 1. \| Variable \| Number of participants (%) \| \| --- \| --- \| \| Sex \| \| \| Male \| 6 (33.3) \| \| Female \| 12 (67.7) \| \| Years of experience \| \| \| 1–5 \| 7 (38.9) \| \| 6–20 \| 6 (33.3) \| \| >20 \| 5 (27.8) \| \| Location of the practice \| \| \| States \| \| \| Australian Capital Territory \| 1 (5.6) \| \| Northern Territory \| 1 (5.6) \| \| Queensland \| 3 (16.7) \| \| South Australia \| 1 (5.6) \| \| Tasmania \| 7 (38.8) \| \| Victoria \| 5 (27.7) \| \| Modified Monash Model1 \| \| \| Metropolitan \| 8 (44.4) \| \| Regional centres \| 3 (16.7) \| \| Rural towns \| 6 (33.3) \| \| Remote communities \| 1 (5.6) \| Thematic analysis of the interview data identified six key themes that impacted GP care for patients following hospitalisation for AECOPD (Figure 1). Figure 1.: Factors impacting GP post-exacerbation comprehensive care of COPD patients. COPD = chronic obstructive pulmonary disease. COPD-X = Australia and New Zealand’s guidelines for the management of COPD. HCPs = healthcare professionals. ## Theme 1: GPs’ priorities for Acute Exacerbations of COPD care were varied All participants considered it important for COPD patients to have a follow-up appointment with their GP after an acute exacerbation, but they differed in terms of perceived priorities and what should happen during this visit and the timing of when it should occur. Some GPs felt the follow-up visit should focus on the patients’ recovery and medication review: ‘It’s important to help them [patients] and see if they’re improving or not and to make sure people are taking the medications and puffers as prescribed. ’(GP 10, 10 years’ experience) Others showed a good understanding about exacerbations and considered prevention of exacerbations and readmissions as a priority: ‘… one exacerbation means, there’s a higher chance of getting another exacerbation. ’(GP 9, 8 years’ experience) ‘We have a good COPD management plan in place, including management of further exacerbations, as far as I’m aware it’s best practice to COPD.’(GP 12, 10 years’ experience) GPs also identified the follow-up visit after an AECOPD as an opportunity for the clinician to revisit their management and optimise care to prevent future exacerbations. Furthermore, with long-term doctor–patient relationships, GPs described they were well placed to deliver preventive care: ‘I think that often the hospital they’re trying to get the patient stable from that exacerbation but really our role is to be taken that as a risk factor for that patient and we know the patient the best, so then try and put in a plan in place to stop further exacerbations. ’(GP 11, 4 years’ experience) A small number of GPs in this sample reported using therapeutic guidelines to guide their management of COPD. Among GPs who were aware of Australia’s COPD-X guidelines, perceptions varied. Some felt they were too broad or that it was challenging to find information they needed; others did not find them to be a helpful reference, while others felt the COPD-X ‘concise’ guide18 was less complicated and easier to use. More experienced GPs described they were not averse to guidelines, but were more comfortable with their own way of practice. However, they recognised the need for updating their knowledge to ensure practice remained in line with guidelines. COPD action plans were seen as an important aspect of exacerbation management. Some felt they were very useful to help manage exacerbations and prevent hospital admissions: ‘Most of my patients are set up with a COPD action plan, I’ve never had a problem, once I’ve educated the patients with them following through, you make sure that the action plan is simple, straightforward, with my patients who’ve got flare ups of COPD, very rarely need to admit them. ’(GP 8, 26 years’ experience). In contrast, others viewed action plans more negatively, perceiving that older adults with COPD were not interested in ‘an extra piece of paper’. Some GPs reported that they give verbal advice rather than a written plan. However, GPs commented that participating in the present study had prompted them to consider using written COPD action plans more in their future clinical practice. Some participants felt barriers such as insufficient consultation time and health literacy restricted their implementation. ## Theme 2: challenged by pharmacological management of COPD patients Most GPs reported that they use a stepwise guide for pharmacological management: ‘… that’s the time I look at the guidelines basically. ’(GP 9, 8 years’ experience) Despite the usefulness of the stepwise approach of the Australian COPD-X guidelines, GPs expressed difficulty in selecting the most appropriate medication and inhaler/device as many inhaled medications are available under each category (for example, long-acting muscarinic antagonist, long-acting beta-agonist, and corticosteroids): ‘I don’t understand them, there are too many, I don’t know the difference between one and another. ’(GP 12, 10 years’ experience) Therefore, GPs restricted their selection of medications to the best known, or most familiar to them: ‘Very overwhelming. I wish there was just one of each, to be honest, I would really. I guess we’ll find our favourites and we just kind of stick with those. ’(GP 11, 4 years’ experience) Participants further reflected on their decision making in respect to the medications best suited for patients and the challenges they faced in making these decisions. Sometimes respiratory physicians commenced new medication, but GPs did not know the reason behind it, feeling it was confusing and complicated. As one GP explained: ‘Like half the time when they come back from respiratory specialist assessment, they’re on a medication I’ve never heard of. I find it’s complicated and I’m not sure what benefit they give … [or] necessarily better than the ones that I am familiar with. So, I find it confusing. ’(GP 12, 10 years’ experience) GPs also reported difficulties. They struggled to keep up with new developments in medication and devices used in COPD. Interviews revealed that GPs get to know about medication from advertisements in medical journals or educational sessions sponsored by pharmaceutical companies. They recognised the need to become more knowledgeable in pharmacological management but preferred to have succinct, evidence-based information of COPD medications independent of the pharmaceutical industry: ‘Keeping up to date with all the new puffers and devices, that’s certainly a challenge. ’(GP 15, 21 years’ experience) ‘I found it initially as registrar training very overwhelming, so many drug reps come with all their products and big charts, I just couldn’t get my head around what or which one were better ones. ’(GP 17, 5 years’ experience) ## Theme 3: maximising care within constrained consultation time Follow-up appointments for COPD exacerbations were mostly limited to a standard consultation and limited by the availability of the GP. This varied by practice size and its location. GPs commented on time as a barrier for providing comprehensive review and management advice within a standard consultation. For example, one GP described: ‘It’s so difficult with the amount of stuff that you have to get through in a 15-minute appointment. ’(GP 11, 4 years’ experience) COPD patients also commonly present with comorbidities, therefore the follow-up consultations were not always limited to COPD but also to manage other acute or chronic problems. Participants felt providing the best possible care at follow-up visit, within the time available for a standard consultation, was a challenge. Addressing their patients’ needs, values, and expectations was a central concern for GPs. One GP explained that: ‘… patients have an agenda and the doctors have an agenda, and often they don’t align. Patient has a list of problems and I have a list of COPD things I want to look through and I can’t do that [in the time].’(GP 11, 4 years’ experience) Many GPs welcomed the role of practice nurses to support the management of patients following AECOPD care, but also highlighted the variability of their role within Australian general practices. Practice nurses were mostly involved with vaccination and spirometry, and some made contributions to patient education, smoking cessation counselling, and checking inhaler technique, but this varied widely. ## Theme 4: timely communication between healthcare professionals was desired, but experiences were mixed. Hospital discharge summaries were identified by GPs as key communication tools. Most GPs were happy with the timeliness of electronically sent discharge summaries; however, they expressed mixed feelings regarding the content as it showed variation depending on where they worked and the hospital they dealt with. Few GPs were satisfied with the content: ‘It has improved tremendously over last two or three years, that’s because we get … hospital discharge summaries in our area, which has a message to the GP about what the follow-up plan is and what they want GPs to do, that’s actually very useful. ’(GP 15, 21 years’ experience) In contrast, other GPs identified serious concerns with the discharge summaries they had received. It was suggested that the hospital should arrange the follow-up appointment prior to the discharge of the patient: ‘There are lot of areas they could improve, so to be honest, I find medication aspect is almost useless. Not so much in terms of what happened during the admission, and one of my real frustrations is to follow up, follow-up of the results. ’(GP 1, >25 years’ experience) ‘I don’t think the patient generally is told to make an appointment with the GP and it would be really great if the hospital staff could help the patient make a follow-up appointment with the GP before they left the hospital, because it just feels like good handover. ’(GP10, 10 years’ experience). Throughout the interviews, GPs described their experiences of communication with outpatient hospital specialists as problematic. One GP described: ‘The hospital specialists, kind of revise and change things, sometimes without speaking to the community team or GP, then you don’t know, they haven’t spoken to us to understand the reasoning of why they’re on that medication, that’s quite frustrating. ’(GP 16, 10 years’ experience) Care providers often have their own disciplinary view of what the patient needs and how they manage the patient. GPs explained the importance of collaboration among healthcare professionals rather than working independently. Interviewees commented that they prefer to have proper feedback from other health professionals who were involved with patient care as it would help to deliver better personalised care for patients. For example, another GP explained: ‘The correspondence from outpatient pulmonary rehab, a pretty brief summary but I don’t think we get much that’s directly addressing that person as an individual. I prefer it be individualised; it’ll make a world of difference. ’(GP 15, 21 years’ experience) ## Theme 5: access to other healthcare services frustrated GPs’ ability to provide best practice care GPs shared their frustrations of not being able to provide the best practice of care for COPD patients and expressed their concern regarding accessibility of referral services. Less priority given to GP referrals in the public sector, costs involved in private programmes, and limited availability of programmes were identified as barriers to pulmonary rehabilitation: ‘In terms of the private allied health, that’s quite expensive. So even if people have a team care arrangement and things, you know, at least $50 out of pocket, usually a position that’s just completely out of reach for my patients. ’(GP 10, 10 years’ experience) Participants felt that many of these factors were capable of being addressed and, interestingly, felt that referrals for patients following AECOPD might be more appropriately organised by hospital staff before discharge. Most GP participants also described difficulty accessing outpatient respiratory specialists when required. This included accessibility and affordability issues related to limited availability of respiratory specialists, long waiting period in the public sector, and private sector costs: ‘It’s very long waiting period to see the specialist, there are only one or two private respiratory specialists here. We’ve got a poor socioeconomic status, so not a lot of people can see them privately. In public hospital, there’s long waiting period to get in. ’(GP 14, 12 years’ experience) ## Theme 6: patient compliance with care advice GPs felt responsible for motivating patients to quit smoking but found it challenging when patients continued to smoke after the provision of support to quit. GPs felt that responsibility for their health ultimately lay with the patients. Some participants described problems regarding discontinuity of care because of patients seeing multiple GPs (as they do not register at a single practice in Australia). This disruption of informational and management continuity was felt to impact on the provision of best care: ‘They’re like, I go to that doctor for that, and I come to you for this, that’s super frustrating.’ ( GP 12, 10 years’ experience) ‘I find it very, very difficult to chase information, I try to explain to patients with ten different doctors, you’re going to get ten different opinions, I’m not saying, and they all have, they have different merits. So, it’s really important to see one person, whoever you like, whoever you trust. ’(GP 17, 5 years’ experience) ## Summary This study provides detailed insights from the experiences of GPs responsible for the care of patients following AECOPD in Australia. Post-exacerbation care was influenced by several factors such as GPs’ knowledge and variation in care based on GPs perceived priorities, consultation time, confusion around pharmacological management, problems in communication between healthcare professionals, patient compliance, and difficulties accessing health services. Interviews also revealed key influences that drive these issues related to the GP, patient, and health services. One of the key concerns to emerge from interviews was uncertainty regarding pharmacological management of COPD. GPs expressed a desire for clearer information to guide inhaler therapy choice at each progression of stepwise management, particularly in light of frequent advances in clinical trials, device types, and combination therapy options. However, it was felt important that such information was received from trusted authorities independent of industry influence. Organisations such as National Prescribing Services have provided such a service in Australia.19 Teamwork and communication are crucial elements to deliver effective patient care; however, participants described rarely receiving sufficient clinical information from patients admitted to hospital or engaging in shared decision making with other healthcare professionals. Existing studies of GPs and specialists show areas of shared concerns and communication difficulties among both parties, suggesting a need for ongoing optimisation of feedback exchange.20,21 Involvement of GPs in discharge planning is advocated in COPD-X guidelines and clinical handover to community-based care is an essential standard for hospital staff in Australia’s National Safety and Quality Health Service Standards.14,22 Future audits of compliance with these requirements may be indicated to accurately identify the extent to which this presently occurs. GPs also described frustrations regarding insufficient time for post-exacerbation consultations to deliver comprehensive care. Initial consultations were typically limited to ‘level B’ (<20 minutes). An enabler proposed by interviewees was a Medicare rebate billing item number for multiple consults (for example, a ‘COPD cycle of care’) similar to prior items for asthma and diabetes.23,24 GPs also expressed a desire to strengthen the supportive role of nurses in clinical practice similar to successful models for diabetes educators in primary care for patients following AECOPD.25,26 It remains to be seen whether such initiatives would improve patient care or clinical outcomes. This study observed variability in the clinical priorities of interviewees that impacted on the application of COPD guidelines. This has also been observed in Australian tertiary hospitals.27 As with previous studies, practitioners’ clinical experience and perception were important, but patient needs and expectations were also perceived to influence application of clinical practice guidelines.28 Australian general practice is patient centred and places high value on patients’ narratives and shared decision making. This can challenge the delivery of evidence-based care.29,30 Although medical practitioners have a responsibility to maintain knowledge and skills, strategies may need to be considered to increase familiarity with guidelines. Interventions to support the implementation of guideline-based care need to be multifaceted and tailored specifically to the barriers unique to GPs’ health system(s) and funding models within which they operate.31 Guidelines are more likely to be used in clinical practice if they are simple, relevant to practice, and perceived as important.32 Specific software modules and educational visits have been shown to enhance understanding and implementing guidelines in general practice.33 ## Strengths and limitations The qualitative methodology used in this study permitted rich data collection of important insights into general practice care of patients following AECOPD. Participants self-selected (convenience sample) but no participants reported having special interests in respiratory diseases or COPD. This was expected, as recognition of ‘special interest pathways’ is uncommon among Australian GP professions, despite recent creation of such models.34 The sample comprised a broad representation of GPs with diverse characteristics such as experience, location, and country where they obtained their medical qualifications. The interviewer’s identity as a GP may also have contributed to the quality and richness of interview data as research has shown professionals interviewing each other (‘communication between equals’) can lead to ‘rich and intuitive responses’.35 However, possible ‘conceptual blindness’ also needs to be considered where interviewer’s feelings and opinion on the subject governs the dialogue.36 In this study, experienced qualitative researchers provided oversight of the analysis ensuring rigour and reflexivity. ## Comparison with existing literature Few participants discussed the role of vaccinations or non-pharmacological management options to potentially prevent future COPD exacerbations. This was interesting as data from the UK show such therapies are highly cost-effective.37 Hospital-initiated clinical pathways and AECOPD discharge ‘bundles’ involving delivery of multi-component guideline-based care elements (for example, inhaler technique review, facilitation of smoking cessation, COPD action plans, pulmonary rehabilitation referrals) are documented in the literature and practised in some countries.11–13 Such interventions are, however, rare in Australia. It is therefore unclear whether regional or healthcare organisational differences might explain such observations. Pulmonary rehabilitation is underutilised worldwide for many reasons.38 However, barriers identified in this study included referral processes (for example, long wait lists, low acceptance of GP referrals for public programmes), costs (for private programmes), and limited availability of local programmes. Socioeconomic deprivation, poor access to transport, and lack of perceived benefits further contribute to poor uptake rates in previous international studies.39–41 It is interesting that cost emerged as a barrier in this study, considering that many Australian programmes are provided via public health. This may be an artefact of sampling from higher socioeconomic settings; however, it may also suggest a lack of awareness of local public programme availability. Strategies to overcome barriers to undertaking pulmonary rehabilitation continue to require further exploration. Difficulties prioritising COPD care because of time constraints is a particular pressure point within Australia’s fee-for-service funding model. It has also been highlighted in a Scandinavian study of primary care health professionals.42 Time pressures result in clinicians switching reasoning towards intuitive decision-making strategies rather than structured approaches.43,44 This can be counteractive to efforts to enhance evidence-informed practice.44 Considering actual time involved and the complexity of the follow-up visit, although level C (long) consultations (>20 minutes, up to 40 minutes) might be more appropriate, evidence is limited to support an association between extending consultation time and quality of care in primary care.44,45 ## Implications for research and practice This study provides valuable insights into Australian GP care of patients following AECOPD. Several discrete issues affecting health service organisation and the delivery of evidence-based medicine were identified and, importantly, appear modifiable. This study highlights a necessity for GPs to maintain familiarity with evidence-based guidelines. This is supported by prior national and international studies with little apparent improvement over time.46–50 Additional support is likely indicated to ensure clinical practice is evidence informed. Much scope remains to explore ways to enhance communication between hospital and primary care settings, including potential engagement of GPs during times of hospital admissions.51 Use of GP-informed disease-specific templated clinical handovers and initiatives to strengthen patient and carer engagement regarding discharge planning could potentially further enhance post-exacerbation care. Such opportunities highlight an important ongoing need for future consumer-informed inquiry. ## Funding Royal Australian College of General Practitioners Foundation Family Medical Care Education and Research Grant 2020 (FMCER2020-02). ## Ethical approval The study was approved by the Monash University Human Research Ethics Committee. ( Ref. MUHREC 26571). ## Provenance Freely submitted; externally peer reviewed. ## Competing interests The authors have declared no competing interests. ## Discuss this article Contribute and read comments about this article: bjgp.org/letters ## References 1. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest (2000) 117 398S-401S. PMID: 10843984 2. Wedzicha JA, Miravitlles M, Hurst JR. Management of COPD exacerbations: a European Respiratory Society/American Thoracic Society guideline. Eur Respir J (2017) 49 1600791. PMID: 28298398 3. MacLeod M, Papi A, Contoli M. Chronic obstructive pulmonary disease exacerbation fundamentals: diagnosis, treatment, prevention and disease impact. Respirology (2021) 26 532-551. PMID: 33893708 4. Donaldson GC, Seemungal TAR, Bhowmik A, Wedzicha JA. Relationship between exacerbation frequency and lung function decline in chronic obstructive pulmonary disease. Thorax (2002) 57 847-852. PMID: 12324669 5. Anzueto A. Impact of exacerbations on COPD. Eur Respir Rev (2010) 19 113-118. PMID: 20956179 6. Vestbo J, Hurd SS, Agustí AG. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: GOLD executive summary. Am J Respir Crit Care Med (2013) 187 347-365. PMID: 22878278 7. Hurst JR, Vestbo J, Anzueto A. Susceptibility to exacerbation in chronic obstructive pulmonary disease. N Engl J Med (2010) 363 1128-1138. PMID: 20843247 8. Seemungal TA, Donaldson GC, Bhowmik A. Time course and recovery of exacerbations in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med (2000) 161 1608-1613. PMID: 10806163 9. Bhowmik A, Seemungal TA, Sapsford RJ, Wedzicha JA. Relation of sputum inflammatory markers to symptoms and lung function changes in COPD exacerbations. Thorax (2000) 55 114-120. PMID: 10639527 10. Geelhoed EA, Brameld KJ, Holman CDJ, Thompson PJ. Readmission and survival following hospitalization for chronic obstructive pulmonary disease: long-term trends. Intern Med J (2007) 37 87-94. PMID: 17229250 11. Bourbeau J, Julien M, Maltais F. Reduction of hospital utilization in patients with chronic obstructive pulmonary disease: a disease-specific self-management intervention. Arch Intern Med (2003) 163 585-591. PMID: 12622605 12. Plishka CT, Rotter T, Penz ED. Effects of clinical pathways for COPD on patient, professional, and systems outcomes: a systematic review. Chest (2019) 156 864-877. PMID: 31150639 13. Ospina MB, Mrklas K, Deuchar L. Effectiveness of discharge care bundles for patients with chronic obstructive pulmonary disease: a systematic review. A23 Understanding and reducing COPD rehospitalizations. Am J Respir Crit Care Med (2016) 193 A1107 14. Dabscheck E, George J, Hermann K. COPD-X Australian guidelines for the diagnosis and management of chronic obstructive pulmonary disease: 2022 update. Med J Aust (2022) 217 415-423. PMID: 36116098 15. Minichiello V, Aroni R, Hays TN. In-depth interviewing: principles, techniques, analysis (2008) 16. Braun V, Clarke V. Using thematic analysis in psychology. Qual Res Psychol (2006) 3 77-101 17. 17. Department of Health and Aged Care, Australian Government Modified Monash Model2021https://www.health.gov.au/health-topics/rural-health-workforce/classifications/mmm (accessed 1 Jan 2023).. Modified Monash Model (2021) 18. Dabscheck E, McDonald CF, Yang IA. Concise guidance for COPD. Respirology (2020) 25 1129-1132. PMID: 32856382 19. GP survey results: COPD medicines and inhalers. (2017) 20. Berendsen AJ, Kuiken A, Benneker WHGM. How do general practitioners and specialists value their mutual communication? A survey. BMC Health Serv Res (2009) 9 143. PMID: 19664238 21. Sampson R, Barbour R, Wilson P. The relationship between GPs and hospital consultants and the implications for patient care: a qualitative study. BMC Fam Pract (2016) 17 45. PMID: 27074867 22. Communication at clinical handover. (n.d) 23. Asthma cycle of care. (n.d) 24. Saunders M, Schattner P, Mathews M. Diabetes ’cycles of care’ in general practice — do government incentives help?. Aust Fam Physician (2008) 37 781-784. PMID: 18797538 25. Deeb A, Yousef H, Abdelrahman L. Implementation of a diabetes educator care model to reduce paediatric admission for diabetic ketoacidosis.. J Diabetes Res (2016) 2016 3917806. PMID: 27294150 26. Value of diabetes education. (n.d) 27. Cousins JL, Wood-Baker R, Wark PAB. Management of acute COPD exacerbations in Australia: do we follow the guidelines?. ERJ Open Res (2020) 6 00270-2019. PMID: 32337215 28. Arts DL, Voncken AG, Medlock S. Reasons for intentional guideline non-adherence: a systematic review. Int J Med Inform (2016) 89 55-62. PMID: 26980359 29. May C, Rapley T, Moreira T. Technogovernance: evidence, subjectivity, and the clinical encounter in primary care medicine. Soc Sci Med (2006) 62 1022-1030. PMID: 16162385 30. Hansen E, Walters J, Howes F. Whole person care, patient-centred care and clinical practice guidelines in general practice. Health Sociol Rev (2016) 25 157-170 31. Christl B, Lloyd J, Krastev Y. Preventing vascular disease: effective strategies for implementing guidelines in general practice. Aust Fam Physician (2011) 40 825-828. PMID: 22003490 32. Graham R, Mancher M, Wolman DM. Promoting adoption of clinical practice guidelines. Clinical practice guidelines we can trust (2011) 145-185 33. Hermens RP, Hak E, Hulscher ME. Adherence to guidelines on cervical cancer screening in general practice: programme elements of successful implementation. Br J Gen Pract (2001) 51 897-903. PMID: 11761203 34. 34. Royal Australian College of General Practitioners Specific interests groups2022https://www.racgp.org.au/getmedia/9528c7d3-6be1-4b2d-b2b3-429d7015bb04/RACGP-Specific-Interests-Groups-2022.pdf.aspx (accessed 1 Jan 2023).. Specific interests groups (2022) 35. Chew-Graham CA, May CR, Perry MS. Qualitative research and the problem of judgement: lessons from interviewing fellow professionals. Fam Pract (2002) 19 285-289. PMID: 11978720 36. Andersson SJ, Troein M, Lindberg G. Conceptions of depressive disorder and its treatment among 17 Swedish GPs. A qualitative interview study. Fam Pract (2001) 18 64-70. PMID: 11145631 37. 37. Primary Care Respiratory Society UK, British Thoracic Society IMPRESS Guide to the relative value of COPD interventions2012https://www.networks.nhs.uk/nhs-networks/impress-improving-and-integrating-respiratory/documents/IMPRESS%20COPD%20Relative%20Value%20Main%20Report.pdf (accessed 1 Jan 2023).. IMPRESS Guide to the relative value of COPD interventions (2012) 38. Vogiatzis I, Rochester CL, Spruit MA. Increasing implementation and delivery of pulmonary rehabilitation: key messages from the new ATS/ERS policy statement. Eur Respir J (2016) 47 1336-1341. PMID: 27132269 39. Steiner MC, Lowe D, Beckford K. Socioeconomic deprivation and the outcome of pulmonary rehabilitation in England and Wales. Thorax (2017) 72 530-537. PMID: 28077613 40. Harris D, Hayter M, Allender S. Improving the uptake of pulmonary rehabilitation in patients with COPD: qualitative study of experiences and attitudes. Br J Gen Pract (2008). DOI: 10.3399/bjgp08X342363 41. Keating A, Lee A, Holland AE. What prevents people with chronic obstructive pulmonary disease from attending pulmonary rehabilitation? A systematic review. Chron Respir Dis (2011) 8 89-99. PMID: 21596892 42. Sandelowsky H, Hylander I, Krakau I. Time pressured deprioritization of COPD in primary care: a qualitative study. Scand J Prim Health Care (2016) 34 55-65. PMID: 26849465 43. Dror IE, Busemeyer JR, Basola B. Decision making under time pressure: an independent test of sequential sampling models. Mem Cognit (1999) 27 713-725 44. Tsiga E, Panagopoulou E, Sevdalis N. The influence of time pressure on adherence to guidelines in primary care: an experimental study. BMJ Open (2013) 3 e002700 45. Wilson AD, Childs S. Effects of interventions aimed at changing the length of primary care physicians’ consultation. Cochrane Database Syst Rev (2006) CD003540. PMID: 16437458 46. Sehl J, O’Doherty J, O’Connor R. Adherence to COPD management guidelines in general practice? A review of the literature. Ir J Med Sci (2018) 187 403-407. PMID: 28735500 47. Lugtenberg M, Zegers-van Schaick JM, Westert GP, Burgers JS. Why don’t physicians adhere to guideline recommendations in practice? An analysis of barriers among Dutch general practitioners. Implement Sci (2009) 4 1-9. PMID: 19123945 48. Carlsen B, Glenton C, Pope C. Thou shalt versus thou shalt not: a meta-synthesis of GPs’ attitudes to clinical practice guidelines. Br J Gen Pract (2007). DOI: 10.3399/096016407782604820 49. Bereznicki B, Walters H, Walters J. Initial diagnosis and management of chronic obstructive pulmonary disease in Australia: views from the coal face. Intern Med J (2017) 47 807-813. PMID: 28276136 50. Harris MF, Lloyd J, Krastev Y. Routine use of clinical management guidelines in Australian general practice. Aust J Prim Health (2014) 20 41-46. PMID: 23075762 51. Scarfield P, Shepherd TD, Stapleton C. Improving the quality and content of discharge summaries on acute medicine wards: a quality improvement project. BMJ Open Qual (2022) 11 e001780
--- title: 'The association between history of prenatal loss and maternal psychological state in a subsequent pregnancy: an ecological momentary assessment (EMA) study' authors: - Claudia Lazarides - Nora K. Moog - Glenn Verner - Manuel C. Voelkle - Wolfgang Henrich - Christine M. Heim - Thorsten Braun - Pathik D. Wadhwa - Claudia Buss - Sonja Entringer journal: Psychological Medicine year: 2023 pmcid: PMC9975992 doi: 10.1017/S0033291721002221 license: CC BY 4.0 --- # The association between history of prenatal loss and maternal psychological state in a subsequent pregnancy: an ecological momentary assessment (EMA) study ## Abstract ### Background Prenatal loss which occurs in approximately $20\%$ of pregnancies represents a well-established risk factor for anxiety and affective disorders. In the current study, we examined whether a history of prenatal loss is associated with a subsequent pregnancy with maternal psychological state using ecological momentary assessment (EMA)-based measures of pregnancy-specific distress and mood in everyday life. ### Method This study was conducted in a cohort of $$n = 155$$ healthy pregnant women, of which $$n = 40$$ had a history of prenatal loss. An EMA protocol was used in early and late pregnancy to collect repeated measures of maternal stress and mood, on average eight times per day over a consecutive 4-day period. The association between a history of prenatal loss and psychological state was estimated using linear mixed models. ### Results Compared to women who had not experienced a prior prenatal loss, women with a history of prenatal loss reported higher levels of pregnancy-specific distress in early as well as late pregnancy and also were more nervous and tired. Furthermore, in the comparison group pregnancy-specific distress decreased and mood improved from early to late pregnancy, whereas these changes across pregnancy were not evident in women in the prenatal loss group. ### Conclusion Our findings suggest that prenatal loss in a prior pregnancy is associated with a subsequent pregnancy with significantly higher stress and impaired mood levels in everyday life across gestation. These findings have important implications for designing EMA-based ambulatory, personalized interventions to reduce stress during pregnancy in this high-risk group. ## Background The prevalence among women of childbearing age of prenatal loss, the loss of an unborn child during pregnancy through miscarriage or stillbirth, is substantial, with one out of five women experiencing a miscarriage (loss of an embryo or fetus before the 20th week of gestation), and one out of 160 women experiencing a stillbirth (loss of a fetus occurring after the 20th week of gestation and a weight above 500 g) (Blencowe et al., 2016; El Hachem et al., 2017; Lawn et al., 2016; MacDorman, Kirmeyer, & Wilson, 2012; Murphy et al., 2017; Price, 2006). The negative consequences on women's mental health of losing an unborn child have been reported in several studies: prenatal loss is related to a higher risk for psychiatric disorders such as post-traumatic stress disorder, anxiety disorders, and major depression (reviewed in Engelhard, van den Hout, & Arntz, 2001; Farren et al., 2016, 2018, 2020; Horesh, Nukrian, & Bialik, 2018; Hughes, Turton, & Evans, 1999; Jacob, Polly, Kalder, & Kostev, 2017; Turton, Evans, & Hughes, 2009). Over $80\%$ of women who experience prenatal loss become pregnant within the subsequent 12-month period (Regan et al., 2019; Sundermann, Hartmann, Jones, Torstenson, & Velez Edwards, 2017), and it is therefore likely that the negative effects of prenatal loss on maternal psychological well-being may extend to the subsequent pregnancy. Given the prominent role of maternal psychological state during pregnancy in many critical pregnancy, birth and offspring developmental and health outcomes (Bale et al., 2010; Buss, Entringer, & Wadhwa, 2012; Entringer, 2013; Entringer, Buss, & Wadhwa, 2012; Entringer, de Punder, Buss, & Wadhwa, 2018; Giannandrea, Cerulli, Anson, & Chaudron, 2013; Heim, Entringer, & Buss, 2019; Wadhwa, Entringer, Buss, & Lu, 2011), it is crucially important to determine the relationship of a previous prenatal loss on maternal psychological well-being during a subsequent pregnancy. Previous studies on the association between prenatal loss and maternal psychological state in a subsequent pregnancy have focused primarily on maternal anxiety and depression (Hughes et al., 1999; Hunter, Tussis, & MacBeth, 2017; Turton et al., 2009). These previous studies have several limitations. First, the majority of these studies have focused on clinical diagnoses of psychiatric disorders (Blackmore et al., 2011; Gong et al., 2013; Turton, Hughes, Evans, & Fainman, 2001), thereby precluding the ascertainment of whether this relationship is evident with variation in maternal psychosocial distress and affective state along a continuum, potentially below clinical thresholds. The clinical relevance of maternal psychological state in pregnancy is not restricted to psychopathology but is evident along a continuum (reviewed in Burgueno, Juarez, Genaro, & Tellechea, 2020; Graignic-Philippe, Dayan, Chokron, Jacquet, and Tordjman, 2014; Lautarescu, Craig, & Glover, 2020; Tarabulsy et al. 2014; Wadhwa et al. 2011; Walsh et al. 2019). Second, previous study on the association of prenatal loss with maternal psychological well-being has relied exclusively on the use of traditional, retrospective recall-based measures to characterize maternal psychological state (for a meta-analytic overview, refer Campbell-Jackson & Horsch, 2014; Hunter et al. 2017). Respondents are typically asked to rate how stressed, anxious, or depressed they have felt over the past week/month/since the beginning of their pregnancy. These traditional measures are prone to retrospective recall bias (Podsakoff, MacKenzie, & Podsakoff, 2012), thereby limiting their validity. In addition, most participants are asked to fill out the questionnaires in either a clinical or research laboratory setting, thereby potentially limiting their generalizability (ecological validity) to everyday real-life situations and circumstances. Third, the majority of previous studies have incorporated only one measurement time point, in either early or late pregnancy; only 4 of 19 previous studies have used a longitudinal study design (Hunfeld, Agterberg, Wladimiroff, & Passchier, 1996; Robertson-Blackmore et al., 2013; Tsartsara & Johnson, 2006; Woods-Giscombe, Lobel, & Crandell, 2010; refer recent meta-analysis by Hunter et al., 2017). It may be particularly important to assess maternal psychological state longitudinally across pregnancy because the association between history of prenatal loss and maternal psychological state may change across the course of gestation. Once the critical hallmark of 20th week of gestation is passed the risk for prenatal loss decreases significantly (ACOG, 2018; Ammon Avalos, Galindo, & Li, 2012; Mukherjee, Velez Edwards, Baird, Savitz, & Hartmann, 2013), potentially contributing to improvements in maternal well-being in the second half of pregnancy. This issue may have clinical relevance because studies of the effects of maternal stress during pregnancy have reported differential effects depending on the gestational time window of its occurrence (Buss et al., 2009, 2012a, 2012b; Davis, Head, Buss, & Sandman, 2017; Entringer et al., 2016). Fourth, several of the previous studies are limited in terms of study design, particularly the lack of appropriate comparison groups. For example, some studies have included in the comparison group a combination of women who were pregnant for the first time and also women who were pregnant at least one time before the current (index) pregnancy, whereas the group of women with a history of prenatal loss include, obviously, only multigravida (women who were pregnant at least once before), and these studies did not adjust for gravida or parity status (Abbaspoor, Razmju, & Hekmat, 2016; Bicking Kinsey, Baptiste-Roberts, Zhu, & Kjerulff, 2015; Cumming et al., 2007; Farren et al., 2016, 2018; Volgsten, Jansson, Svanberg, Darj, & Stavreus-Evers, 2018). Because the event/experience of a prior pregnancy might be associated with biological and psychological changes (Armstrong, Hutti, & Myers, 2009), this could potentially confound the association between history of prenatal loss and psychological state in a subsequent pregnancy. We note that in the current study we addressed this issue by including parity status as a covariate in all analysis. In addition, we conducted a sensitivity analysis by examining the effect of prenatal loss on our study outcomes in the study's subpopulation of multigravid women. Fifth, several studies have failed to account for other important potential confounders associated with both risk for prenatal loss and impairments in mental well-being, such as sociodemographic factors (e.g. income, maternal age) and obstetric characteristics (e.g. obstetric risk factors; Blackmore et al., 2011) EMA methods can address several of these above-discussed limitations by employing repeated real-time measurements of psychological states in participants' natural daily environments, thereby minimize biases associated with retrospective recall measures to provide more accurate and ecologically valid measures of psychological/behavioral states (Smyth & Stone, 2003). Thus, the aim of the current study was to examine the association of history of prenatal loss with assessments during a subsequent pregnancy in early as well as late gestation of maternal psychological state (maternal momentary pregnancy-specific distress and mood) using EMA methods. ## Participants The study was conducted at the Institute of Medical Psychology and the Department of Obstetrics at the Charité Universtitaetsmedizin Berlin, Germany. Women with a singleton, intrauterine pregnancy were recruited prior to 16 weeks gestation. Exclusion criteria were twin pregnancies, uterine, placental/cord anomalies, fetal congenital malformations, and systemic corticosteroid intake. The study protocol included two study visits at the laboratory during early (T1: 12–16 weeks gestation) and late pregnancy (T2: 30–34 weeks gestation), followed by a 4-day EMA period, as described below. The Charité Institutional Review Board approved the study, and all participants provided written, informed consent. The characteristics of the study participants are presented in Table 1. Miscarriage was defined as the loss of an unborn child during a recognized pregnancy before the 20th week of gestation, and if gestational age (GA) is not available, fetal weight equal or below 500 g (Farquharson, Jauniaux, Exalto, & Pregnancy, 2005). Prenatal losses occurring after the 20th week of gestation and a weight above 500 g were termed stillbirths (Tavares Da Silva et al., 2016). In total, $25.8\%$ of the participating women reported the experience of prenatal loss, either miscarriage or stillbirth, in a previous pregnancy ($$n = 40$$). $5.8\%$ reported two prenatal losses, and $1.9\%$ reported more than two prenatal losses ($$n = 3$$) in their reproductive history. The prevalence of obstetric complications during pregnancy was low in our study population ($5.8\%$). Table 1.Maternal sociodemographic and obstetric characteristicsMaternal characteristicsN = 155GA at assessment (weeks gestation, M ± s.d.)T1 – early gestation14.4 ± 1.5T2 – late gestation32.5 ± 1.3Maternal age (years, M ± s.d.)32.5 ± 4.7Highest degree (n, %)No degree14 (9.0)Technical or vocational training44 (28.3)Bachelor degree23 (14.8)Master degree54 (34.8)Ph. D.20 (12.9)Total monthly net household income [M ± s.d., in Euro, categories belowa n (%)]3994 ± 2411<125011 (7.1)1250–17496 (3.9)1750–224910 (6.5)2250–299923 (14.9)3000–399937 (24.0)4000–499926 (16.9)>500041 (26.6)Country of birth (n, %)Germany127 (82.5)Obstetric characteristicsHistory of prenatal loss (n, %) – yes40 (25.8) 0115 (74.2) 128 (18.1) 29 (5.8) >23 (1.9)Parity Nulliparous95 (61.2) Multiparous60 (38.7)Gravidity 176 (49.0) 244 (28.4) 317 (11.0) ⩾418 (11.6)Obstetric risk factors (n, %)9 (5.8) Gestational diabetes4 (2.6) Hypertension2 (1.3) Preterm labor4 (2.6)Note: Due to rounding, some totals may not correspond with the sum of the separate figures.aincome ranges based on KIGGs study's sociodemographic index (Lampert, Hoebel, & Kuntz, 2018). ## Maternal characteristics At each visit, trained study personnel conducted structured interviews to obtain information on sociodemographic characteristics and reproductive history (e.g. gravidity, parity, number of previous pregnancy losses), and estimated date of conception. GA at visit was computed based on early ultrasound measurements. Data on obstetric risk factors were abstracted from the medical record. ## EMA-based measures of maternal psychological state We assessed momentary maternal pregnancy-specific distress and mood using an EMA protocol for ambulatory, real-time measurement of affective states. The EMA protocol was delivered through the mobile phone application movisensXS (movisensXS; movisens, 2020). The 4-day EMA protocol spanned two consecutive weekdays and a weekend (Thursday–Sunday, or Saturday–Tuesday). Participants were provided a smartphone with an electronic diary application. Throughout the EMA period, participants were prompted on average eight times per day (prompts were between 30 and 90 min apart between the hours of 8AM and 8PM). Pregnancy-specific distress (PSD_mean; Rini, Dunkel-Schetter, Wadhwa, & Sandman, 1999) was assessed by inquiring about the woman's feelings (happiness and ambivalence) about being pregnant, her concerns about the baby's health, bodily discomfort due to pregnancy-related changes, and concerns about giving birth. Women rated these items on a Likert scale ranging from 0 to 5 (‘not at all’ to ‘completely’). Participants' ratings at each prompt were aggregated across all items and average scores were computed. The measure of pregnancy-specific stress was chosen for this study because measures that assess distress in a specific area of life may better reflect individual responses to these conditions than global stress questionnaires (Bussières et al., 2015; Stanton, Lobel, Sears, & DeLuca, 2002). This pregnancy-specific distress measure has previously been linked to pregnancy and offspring outcomes (e.g. Buss, Davis, Muftuler, Head, & Sandman, 2010; Glynn, Schetter, Hobel, & Sandman, 2008). Maternal momentary mood was measured by the multidimensional mood questionnaire (MDBF) developed for daily diary research and validated for EMA studies (Courvoisier, Eid, & Lischetzke, 2012; Courvoisier, Eid, Lischetzke, & Schreiber, 2010; Hinz, Daig, Petrowski, & Brahler, 2012; Steyer, Schwenkmezger, Notz, & Eid, 1994). Participants rated their momentary mood along three dimensions: valence (good–bad mood, GB), arousal (calmness–nervousness, CN), and tiredness (alertness–tiredness, AT) on 12 items, four items for each dimension, with a balanced number of negatively worded and positively worded items. Items were rated on a 6-point Likert scale ranging from 0 to 5 (‘not at all’ to ‘completely’). Positively worded items were reversed before aggregating answers to derive an average score for each dimension (good–bad mood: GB_mean, calm–nervous: CN_mean, alert–tired: AT_mean) for each prompt. The derived original scores were reversed to ease the interpretation of the results such that higher average scores for each dimension indicate an unfavorable affective state (bad mood, nervous, tired), and lower average scores indicate favorable affective states (good mood, calm, alert). ## Statistical analysis We performed all statistical analyses in R version 3.5.1 (R Development Core Team, 2018). The R-package nlme version 3.1-137 was used for linear mixed model analyses (Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2018). ## Variance decomposition of momentary measures We used linear mixed(-effect) models (LMMs; cf. multilevel models) to identify the proportion of variance at the different levels of the data (momentary, day, stage of pregnancy and participant level; Snijders & Bosker, 2012) with regard to pregnancy-specific distress and the three mood dimensions (valence, arousal, and tiredness). The applied analytical procedure has been described in detail elsewhere (Lazarides et al., 2020). In four separate 4-level random intercept LMMs, the percentage of total variance for pregnancy-specific distress, valence, arousal, and tiredness was computed at the level of the momentary measurements (level 1), days (level 2), stages of pregnancy (level 3), and participants (level 4). To account for the unequal spacing of the auto-correlated measurements across a day a continuous time-autoregressive covariance structure of order one was specified using time since wake in minutes (Goldstein, Healy, & Rasbash, 1994; Jones, 1993; Littell, Milliken, Stroup, Wolfinger, & Oliver, 2006). Restricted maximum likelihood was used for parameter estimation (for R code, see online Supplement A1). ## Linear mixed models EMA-based measures. To examine the effect of history of prenatal loss on pregnancy-specific distress and momentary maternal mood along the three dimensions valence, arousal, and tiredness, four separate 4-level LMMs were fitted to the nested data with the same random effect structure as described for the variance decomposition of EMA-based measures. We used the same continuous-autoregressive covariance structure to account for unequal temporal spacing of the momentary measurements. The exemplary R code is provided in online Supplement A2. Prenatal loss status (history of prenatal loss yes/no) was used as a dichotomous predictor. Relevant covariates (described below) were included as fixed effects in all models. Moderation by stage of pregnancy. To explore how potential differences in psychological state between women with and without a history of prenatal loss may change with advancing gestation, we included the interaction term between stage of pregnancy (i.e. visit number, T1 and T2) and prenatal loss status in the linear mixed models described above (R code, see online Supplement A3). Sensitivity analysis. The control group included women who were pregnant for the first time and women who were pregnant at least one time before the current pregnancy, whereas the group of women with a history of prenatal loss included only multigravida (multigravida = women that were pregnant at least once before). Because the experience of prior pregnancy may be associated with biological and psychological changes (Armstrong et al., 2009), this could introduce heterogeneity in the control group and limit the validity of reported differences in psychological well-being between women with and without prenatal loss. We therefore conducted a sensitivity analysis by testing our hypothesis in only multi-gravid women. ## Covariates All analyses were adjusted for the effects of potential confounders that have previously been associated with the risk for prenatal loss and impaired psychological well-being, including maternal age, parity (not included as a covariate in the sensitivity analyses), obstetric risk factors, and income (Magnus, Wilcox, Morken, Weinberg, & Haberg, 2019). The following covariates were included as fixed effects in all described LMMs (R code, see online Supplement A1): stage of pregnancy (early – T1 v. late pregnancy – T2), maternal age at first visit, income, parity category (0 – nulliparous, 1 – multiparous; not included in sensitivity analyses described above), obstetric risk factor (presence of any of the following conditions during the current pregnancy: preeclampsia, hypertension, gestational diabetes coded with ‘1’, no obstetric risk factors present coded with ‘0’). ## Compliance and handling of missing data Given the EMA protocol, each time a participant refrained from answering a prompt, declined to answer, ignored a prompt or did not conclude the entire survey, the smartphone application recorded a missing value. To assess compliance, we calculated the percent of missing prompts of the total number of prompts. In the statistical analyses, missing data were accounted for by use of full information restricted maximum likelihood estimation (Little & Rubin, 2002; Raudenbush & Bryk, 2002). Thus, LMMs make use of all available data. ## Compliance Compliance with the EMA protocol (number of missing prompts relative to the total number of prompts) was $86.3\%$, which is above the recommended $80\%$ for EMA-studies, and also above average compliance of 75–$78\%$ previously reported in two meta-analyses comprised of 168 EMA studies (Jones et al., 2019; Wen, Schneider, Stone, & Spruijt-Metz, 2017). ## Variance decomposition The variance decomposition indicates the amount of the total variation in pregnancy-specific distress, valence, arousal, and tiredness that is derived from the different levels of the data (i.e. variation between individuals as well as within individuals, across the stages of pregnancy, across a day, and across moments; de Haan-Rietdijk, Kuppens, and Hamaker, 2016; Schmiedek, Lovden, and Lindenberger, 2013). Based on the LMM, pregnancy-specific distress scores varied largely between individuals ($68.4\%$) and to a lesser extent from moment to moment ($12.1\%$), and from early to late pregnancy ($16.9\%$), as well as to a small degree from day to day ($5.1\%$; summary of results is given in online Supplementary Table S1, detailed results in online Supplementary Table S2). For the mood scales, GB_mean, CN_mean, and AT_mean, predominantly showed variation at the momentary level (48.8–$54.3\%$) and between individuals (27.6–$38.9\%$) rather than from day to day (7.1–$10.2\%$) or from early to late pregnancy (3.7–$5.9\%$; summary of results is given in online Supplementary Table S1, detailed results in online Supplementary Tables S3–S5). Intraclass correlation coefficients reflected this pattern of variation (online Supplementary Table S1). ## Descriptive statistics Summary statistics for pregnancy-specific distress and for each MDBF scale (valence, arousal, tiredness) are displayed separately for each time point (stage of pregnancy) in Table 2. Pregnancy-specific distress decreased slightly from early to late pregnancy in the whole sample. In general, mood improved from early to late gestation, as suggested by a decrease in mood valence (GB_mean), arousal (CN_mean) and level of tiredness (AT_mean) in the whole sample. The observed average and variation of mood scores are comparable with published norms for women in reproductive age (Hinz et al., 2012; Steyer et al., 1994; Steyer, Schwenkmezger, Notz, & Eid, 1997). Table 2.Summary statistics on valence (good – bad mood: GB_mean), arousal (calm – nervous: CN_mean), tiredness (alert – tired: AT_mean), and pregnancy-specific distress (PSD_mean), separately for each time point, T1 and T2T1 ($$n = 155$$)T2 ($$n = 104$$)ParameterMs.d. Ms.d. PSD_mean1.240.8911.190.917GB_mean1.410.9571.220.920CN_mean1.670.9471.490.941AT_mean2.111.042.011.16M, Mean; s.d., Standard deviation; N, Sample size at measurement time point. ## History of prenatal loss and EMA measures of psychological state during pregnancy An overview of the results of the linear mixed-effects model analysis is displayed in Table 3. More detailed results for each outcome are presented in online Supplementary Tables S6–S9. Table 3.Results of linear mixed models predicting EMA-based pregnancy-specific distress (PSD_mean), and affective states (valence, GB_mean; arousal, CN_mean, tiredness, AT_mean) by stage of pregnancy and prenatal loss statusFixed effectsB (s.e.)$95\%$ CI for BpPSD_meanIntercept1.803 (0.471)0.879–2.727<0.001*Prenatal loss0.465 (0.159)0.151–0.7790.004Stage of pregnancy<−0.001 (0.054)−0.106 to 0.1060.997GB_meanIntercept1.264 (0.347)0.584–1.944<0.001*Prenatal loss0.213 (0.117)−0.018 to 4.4460.070.Stage of pregnancy−0.110 (0.035)−0.180 to −0.0400.003CN_meanIntercept1.661 (0.367)0.943–2.380<0.001**Prenatal loss0.247 (0.124)0.003–0.4910.047Stage of pregnancy−0.118 (0.039)−0.195 to −0.0420.003AT_meanIntercept2.323 (0.386)1.566–3.080<0.001Prenatal loss0.293 (0.130)0.036–0.5500.026 Stage of pregnancy−0.100 (0.049)−0.197 to −0.0040.042 Note: Significance codes: $p \leq 0.01$ ‘ ’, $p \leq 0.10$ ‘.’, $p \leq 0.05$ ‘’, $p \leq 0.01$ ‘’, $p \leq 0.001$ ‘’. Results displayed for log-transformed cortisol. Transformation did not change magnitude, direction nor significance level of the reported effects. For fit indices see online Supplementary Table S10. ## Pregnancy-specific distress There was a significant effect of history of prenatal loss on pregnancy-specific distress: women with a history of prenatal loss reported significantly higher levels of pregnancy-specific distress assessed at a momentary level, in early as well as in late gestation ($B = 0.465$, $$p \leq 0.004$$, online Supplementary Table S6). On average, women with a history of prenatal loss reported 0.465-unit higher levels of pregnancy-specific distress on a scale from to those without a history of prenatal loss. ## Mood valence, arousal, and tiredness There was no significant main effect of prenatal loss status on mood valence (GB_mean) albeit mood was slightly impaired in women with a history of prenatal loss compared to those without across the course of pregnancy ($B = 0.214$, $$p \leq 0.070$$, online Supplementary Table S7). Arousal was positively associated with prenatal loss status (CN_mean: $B = 0.247$, $$p \leq 0.047$$, online Supplementary Table S8): across gestation, women with a history of prenatal loss showed increased arousal compared to women without prenatal loss, and were more tired (AT_mean: $B = 0.293$, $$p \leq 0.026$$, online Supplementary Table S9). Women with prenatal loss showed on average a 0.247-unit higher level of arousal and a 0.270-unit higher level of tiredness. ## Moderation of the association between history of prenatal loss and EMA measures of psychological state by stage of pregnancy There was no main effect of stage of pregnancy on levels of pregnancy-specific distress (PSD_mean: B = −0.0002, $$p \leq 0.997$$, online Supplementary Table S6). However, the moderation analysis revealed a significant positive interaction effect between stage of pregnancy and prenatal loss status (PSD_mean: $B = 0.314$, $$p \leq 0.012$$). As displayed in Fig. 1, pregnancy-specific distress decreased from early to late gestation in women without a history of prenatal loss, whereas it increased in women with a history of prenatal loss. Fig. 1.Grouped bar plot of EMA-based pregnancy-specific distress by history of prenatal loss in previous pregnancy and stage of pregnancy (mean ± 2 standard error bars). Across the whole sample, momentary maternal mood including valence (impaired mood), arousal, and tiredness decreased significantly across pregnancy (GB_mean: B = −0.110, $$p \leq 0.003$$; CN_mean: B = −0.118, $$p \leq 0.003$$; AT_mean: B = −0.100, $$p \leq 0.042$$; online Supplementary Tables S7–S9), indicating a general improvement of mood across the whole sample. We therefore conducted a moderation analysis to test if the effect of prenatal loss on mood dimensions was moderated by the stage of pregnancy. The moderation analysis revealed a significant interaction effect of stage of pregnancy and prenatal loss status on arousal (CN_mean: $B = 0.184$, $$p \leq 0.045$$, Fig. 2). Women without a history of prenatal loss reported lower levels of arousal in late compared to early gestation, while levels did not decrease in women with a history of prenatal loss. There was no significant interaction effect of stage of pregnancy and prenatal loss status on mood valence and tiredness (GB_mean: $B = 0.142$, $$p \leq 0.090$$; AT_mean: $B = 0.116$, $$p \leq 0.317$$). Fig. 2.Grouped bar plot displaying EMA-based arousal by history of prenatal loss in previous pregnancy and stage of pregnancy (mean ± 2 standard error bars). ## Sensitivity analysis When considering only women who were pregnant at least one time before the current pregnancy (40 women with and 39 without a history of prenatal loss), most of the previously reported effects remained unchanged in direction, magnitude, and significance level. Specifically, women with a history of prenatal loss reported higher levels of pregnancy-specific distress (PSD_mean: $B = 0.482$, $$p \leq 0.039$$), and were more nervous (CN_mean: $B = 0.359$, $$p \leq 0.037$$), more tired (AT_mean: $B = 0.494$, $$p \leq 0.003$$), across pregnancy compared to women without prenatal loss. Furthermore, there was a trend for an effect of prenatal loss status on mood valence when only including multigravid women. Women with a history of prenatal loss reported impaired mood compared to women without prenatal loss (GB_mean: $B = 0.294$, $$p \leq 0.081$$). ## Discussion To the best of our knowledge, this is the first study to use the EMA approach to comprehensively (i.e. serially, at a momentary level, across everyday life situations) quantify and compare stress and mood levels and trajectories in pregnancy in women with and without a prior history of prenatal loss. Our findings indicate that women with a prior history of prenatal loss experienced significantly more pregnancy-related stress and felt significantly more nervous and tired compared to those who have not previously experienced a prenatal loss. Moreover, our results suggest that these differences persisted and even grew or became amplified as pregnancy progressed. Maternal levels of stress and negative affect progressively decreased over the course of pregnancy in women without a history of prenatal loss, whereas they did not change or even increased in women with a prior history of prenatal loss. The magnitude of this observed difference is striking. Women in the prior prenatal loss group exhibited, on average, $38.6\%$ more pregnancy-specific stress, $18.3\%$ more arousal, and $15.5\%$ more exhaustion than those in the comparison group†1. The present study was not designed to address the clinical relevance of observed findings. Because the vast majority of studies of the effects of maternal stress in pregnancy have relied on the more traditional retrospective recall approach to quantify stress, as opposed to the EMA approach used in the current study, it is difficult to directly extrapolate clinical significance. We note, nevertheless, that several previous studies have reported that differences of comparable or even smaller magnitude in maternal stress during and/or across pregnancy have been independently associated with a range of adverse maternal, birth and child developmental and health outcomes, including premature birth, newborn and infant adiposity, neurodevelopmental deficits, and even cellular measures of aging (telomere length) (Buss et al., 2012b; Entringer et al., 2018; Gyllenhammer, Entringer, Buss, & Wadhwa, 2020; Lindsay, Buss, Wadhwa, & Entringer, 2018; Wadhwa et al., 2011). Based on this observation we submit it is likely that the magnitude of the observed difference in maternal stress in the current study may portend clinical significance. Our results are consistent with those of previous studies that find women report increased levels of post-traumatic stress, anxiety, and depression following pregnancy loss (Farren et al., 2016, 2020; Hughes et al., 1999; Turton et al., 2009). Across time, grief subsides and psychiatric disorders possibly remit, although, the emotional perturbations related to the experience of pregnancy loss remain (Kersting et al., 2007; Krosch & Shakespeare-Finch, 2017; Volgsten et al., 2018). A subsequent pregnancy has the potential to reactivate the affective memories associated with the past prenatal loss (Haas & Canli, 2008). The current study highlights the relevance of history of prenatal loss as a risk factor for increased prenatal stress in pregnancy and thus the potential negative consequences on pregnancy, birth and child development, and health outcomes. In the current study, we observed an overall improvement across pregnancy in psychological well-being. This observation is consistent with recent evidence from a clinical population of pregnant women, who reported a decrease in psychopathological symptoms from early to late pregnancy (Asselmann, Kunas, Wittchen, & Martini, 2020). This* general improvement of well-being and decrease in psychological stress may be associated with the attenuation of maternal biological stress responsivity across pregnancy (Entringer et al., 2010). However, in our study, women with a history of prenatal loss did not exhibit this decrease in stress and improvement in mood across pregnancy, which may be a consequence of their prior traumatic experience of losing a pregnancy and the resultant general feeling of uncontrollability. We have previously reported that the lack of stress attenuation across pregnancy is related to adverse pregnancy outcomes (Buss et al., 2009). We suggest that our study has several strengths. As of our knowledge, this is the first study to assess the effect of a history of prenatal loss on maternal psychological state in early and late pregnancy. We use EMA methods to assess maternal stress in women's everyday life in natural settings in contrast to previous research that exclusively relied on traditional retrospective questionnaires. Participants of the current study showed a high compliance with the EMA protocol. We assessed psychological state on a continuum, and relied on measures of mood and stress rather than focusing on clinical symptom categories or psychiatric diagnoses. We adjusted our analyses for the effect of important potential confounders, including sociodemographic factors, obstetric characteristics and number of previous pregnancies. By means of a sensitivity analysis within multigravida women only, we confirm the robustness of our results. Some limitations of the current study need to be acknowledged. First, data on psychological state was available only for the current pregnancy, measures of psychological state prior to the initial prenatal loss were not available. Stress has been discussed as a risk factor for prenatal loss. However, previous studies investigating whether maternal psychological stress predicts prenatal loss have produced mixed results (Klebanoff, Shiono, & Rhoads, 1990; Milad, Klock, Moses, & Chatterton, 1998; Nelson et al., 2003; Qu et al., 2017). All our analyses were adjusted for covariates potentially associated with both risk for prenatal loss and maternal psychological state, including maternal age, parity, obstetric risk, and household income. Second, we were unable to test the effect of the number of previous prenatal losses on psychological state during pregnancy due to the relatively small number of women who miscarried more than once. Previous large cohort studies suggest that with increasing number of prenatal losses women report even higher levels of depression and anxiety (Blackmore et al., 2011; Toffol, Koponen, & Partonen, 2013). We therefore submit that our findings may represent a conservative estimate of the true effect of history of prenatal loss on psychological state during pregnancy. Third, data on the length of inter-pregnancy intervals as well as on a whether or not women gave birth to a living child between the pregnancy loss and the current pregnancy were not available, and we were therefore unable to test the moderating effects of these variables. Previous studies suggest that the length of the inter-pregnancy interval does not affect the association between history of prenatal loss and depression and/or anxiety during a subsequent pregnancy or in the postpartum period (Gravensteen et al., 2018; Schetter, Saxbe, Cheadle, & Guardino, 2016). A large longitudinal cohort study reports a robust association between history of prenatal loss with increased levels of anxiety and depression during a subsequent pregnancy, which remained stable across the pre- and postnatal period of the index pregnancy, thereby indicating that the psychological impairment associated with previous prenatal loss might not attenuate significantly following the birth of a living child (Blackmore et al., 2011). Fourth, EMA studies assess the participants' psychological state repeatedly across multiple days raising the issue of measurement reactivity. However, previous EMA studies have provided no evidence for measurement reactivity with regard to mood, craving, body image, and suicidal thoughts (Coppersmith, 2020; De Vuyst, Dejonckheere, Van der Gucht, & Kuppens, 2019; Heron & Smyth, 2013; Hufford, Shields, Shiffman, Paty, & Balabanis, 2002; Rowan et al., 2007). The findings of our study suggest that women with a history of prenatal loss are at increased risk for experiencing higher levels of stress during pregnancy. Although, obstetric guidelines issued by the American College of Obstetricians and Gynecologists advice perinatal care providers to screen for postpartum depression, recommendations to not include screening for psychological stress (American Academy of Pediatrics & American College of Obstetricians and Gynecologists, 2017). Our results imply that pregnancy-specific distress might be a good screening tool for this purpose, since the effect on pregnancy loss on pregnancy specific distress in our study was substantial, and it primarily varied between individuals and may not have to be measured that frequently. The current study underscores the importance of using EMA methods in assessing stress and mood in the context of pregnancy, which then could be used to design personalized interventions to reduce maternal stress. EMA-based measures of psychological states can be used to estimate subject-specific ‘moments at risk’, such as the deviation from the individual average stress level in a given moment, that have a higher predictive value for maternal cortisol levels during pregnancy than traditional approaches (Lazarides et al., 2020). Future studies could test the efficacy of EMA-based targeted interventions in women with a history of prenatal loss. Given its substantial burden on maternal and offspring health, reducing stress during pregnancy in this high risk group could yield considerable public health benefit. ## Financial support This work was supported by European Research Council grants ERC-Stg 678073 and ERC-Stg 639766, and by NIH grants R01 HD-060628, R01 AG-050455, R01 HD-065825, UH3 OD-023349. ## Conflict of interest None. ## References 1. Abbaspoor Z., Razmju P. S., Hekmat K.. Relation between quality of life and mental health in pregnant women with prior pregnancy loss. The Journal of Obstetrics and Gynaecology Research (2016) 42 1290-1296. DOI: 10.1111/jog.13061 2. Early pregnancy loss. American College of Obstetricians and Gynecologists (2018) 132 3. Kilpatrick, S. J., Papile, L.-A. & Macones, G. A. (Eds.). American Academy of Pediatrics – Committee on Fetus and Newborn & American College of Obstetricians and Gynecologists – Committee on Obstretric Practice: Guidelines for PERINATAL CARE 2017. (2017) 4. Ammon Avalos L., Galindo C., Li D. K.. A systematic review to calculate background miscarriage rates using life table analysis. Birth Defects Research (2012) 94 417-423. DOI: 10.1002/bdra.23014 5. Armstrong D. S., Hutti M. H., Myers J.. The influence of prior perinatal loss on parents’ psychological distress after the birth of a subsequent healthy infant. Journal of Obstetric, Gynecologic, and Neonatal Nursing (2009) 38 654-666. DOI: 10.1111/j.1552-6909.2009.01069.x 6. Asselmann E., Kunas S. L., Wittchen H. U., Martini J.. Maternal personality, social support, and changes in depressive, anxiety, and stress symptoms during pregnancy and after delivery: A prospective-longitudinal study. PLoS One (2020) 15 e0237609. DOI: 10.1371/journal.pone.0237609 7. Bale T. L., Baram T. Z., Brown A. S., Goldstein J. M., Insel T. R., McCarthy M. M., Nemeroff C. B.. Early life programming and neurodevelopmental disorders. Biological Psychiatry (2010) 68 314-319. DOI: 10.1016/j.biopsych.2010.05.028 8. Bicking Kinsey C., Baptiste-Roberts K., Zhu J., Kjerulff K. H.. Effect of previous miscarriage on depressive symptoms during subsequent pregnancy and postpartum in the first baby study. Maternal and Child Health Journal (2015) 19 391-400. DOI: 10.1007/s10995-014-1521-0 9. Blackmore E. R., Cote-Arsenault D., Tang W., Glover V., Evans J., Golding J., O'Connor T. G.. Previous prenatal loss as a predictor of perinatal depression and anxiety. The British Journal of Psychiatry (2011) 198 373-378. DOI: 10.1192/bjp.bp.110.083105 10. Blencowe H., Cousens S., Jassir F. B., Say L., Chou D., Mathers C., Hogan D.. National, regional, and worldwide estimates of stillbirth rates in 2015, with trends from 2000: A systematic analysis. The Lancet Global Health (2016) 4 e98-e108. DOI: 10.1016/s2214-109x(15)00275-2 11. Burgueno A. L., Juarez Y. R., Genaro A. M., Tellechea M. L.. Systematic review and meta-analysis on the relationship between prenatal stress and metabolic syndrome intermediate phenotypes. International Journal of Obesity (2020) 44 1-12. DOI: 10.1038/s41366-019-0423-z 12. Buss C., Davis E. P., Muftuler L. T., Head K., Sandman C. A.. High pregnancy anxiety during mid-gestation is associated with decreased gray matter density in 6–9-year-old children. Psychoneuroendocrinology (2010) 35 141-153. DOI: 10.1016/j.psyneuen.2009.07.010 13. Buss C., Davis E. P., Shahbaba B., Pruessner J. C., Head K., Sandman C. A.. Maternal cortisol over the course of pregnancy and subsequent child amygdala and hippocampus volumes and affective problems. Proceedings of the National Academy of Sciences of the USA (2012a) 109 E1312-E1319. DOI: 10.1073/pnas.1201295109 14. Buss C., Entringer S., Reyes J. F., Chicz-DeMet A., Sandman C. A., Waffarn F., Wadhwa P. D.. The maternal cortisol awakening response in human pregnancy is associated with the length of gestation. American Journal of Obstetrics and Gynecology (2009) 201 398 e391-398. DOI: 10.1016/j.ajog.2009.06.063 15. Buss C., Entringer S., Wadhwa P. D.. Fetal programming of brain development: Intrauterine stress and susceptibility to psychopathology. Science Signaling (2012b) 5 pt7. DOI: 10.1126/scisignal.2003406 16. Bussières E.-L., Tarabulsy G. M., Pearson J., Tessier R., Forest J.-C., Giguère Y.. Maternal prenatal stress and infant birth weight and gestational age: A meta-analysis of prospective studies. Developmental Review (2015) 36 179-199. DOI: 10.1016/j.dr.2015.04.001 17. Campbell-Jackson L., Horsch A.. The psychological impact of stillbirth on women: A systematic review. (2014) 22 237-256. DOI: 10.2190/IL.22.3.d 18. Coppersmith D. D. L.. Frequent assessment of suicidal thinking does not increase suicidal thinking: Evidence from a high-resolution real-time monitoring study. Preprint from PsyArXiv (2020). DOI: 10.31234/osf.io/6bh82 19. Courvoisier D. S., Eid M., Lischetzke T.. Compliance to a cell phone-based ecological momentary assessment study: The effect of time and personality characteristics. Psychological Assessment (2012) 24 713-720. DOI: 10.1037/aO026733 20. Courvoisier D. S., Eid M., Lischetzke T., Schreiber W. H.. Psychometric properties of a computerized mobile phone method for assessing mood in daily life. Emotion (Washington, D.C.) (2010) 10 115-124. DOI: 10.1037/a0017813 21. Cumming G. P., Klein S., Bolsover D., Lee A. J., Alexander D. A., Maclean M., Jurgens J. D.. The emotional burden of miscarriage for women and their partners: Trajectories of anxiety and depression over 13 months. BJOG An International Journal of Obstetrics and Gynaecology (2007) 114 1138-1145. DOI: 10.1111/j.1471-0528.2007.01452.x 22. Davis E. P., Head K., Buss C., Sandman C. A.. Prenatal maternal cortisol concentrations predict neurodevelopment in middle childhood. Psychoneuroendocrinology (2017) 75 56-63. DOI: 10.1016/j.psyneuen.2016.10.005 23. de Haan-Rietdijk S., Kuppens P., Hamaker E. L.. What's in a day? A guide to decomposing the variance in intensive longitudinal data. Frontiers in Psychology (2016) 7 891. DOI: 10.3389/fpsyg.2016.00891 24. De Vuyst H. J., Dejonckheere E., Van der Gucht K., Kuppens P.. Does repeatedly reporting positive or negative emotions in daily life have an impact on the level of emotional experiences and depressive symptoms over time?. PLoS One (2019) 14 e0219121. DOI: 10.1371/journal.pone.0219121 25. El Hachem H., Crepaux V., May-Panloup P., Descamps P., Legendre G., & Bouet P.-E.. Recurrent pregnancy loss: Current perspectives. International Journal of Women's Health (2017) 9 331-345. DOI: 10.2147/ijwh.S100817 26. Engelhard I. M., van den Hout M. A., Arntz A.. Posttraumatic stress disorder after pregnancy loss. General Hospital Psychiatry (2001) 23 62-66. DOI: 10.1016/s0163-8343(01)00124-4 27. Entringer S.. Impact of stress and stress physiology during pregnancy on child metabolic function and obesity risk. Current Opinion in Clinical Nutrition and Metabolic Care (2013) 16 320-327. DOI: 10.1097/MCO.0b013e32835e8d80 28. Entringer S., Buss C., Rasmussen J. M., Lindsay K., Gillen D. L., Cooper D. M., Wadhwa P. D.. Maternal cortisol during pregnancy and infant adiposity: A prospective investigation. The Journal of Clinical Endocrinology & Metabolism (2016). DOI: 10.1210/jc.2016-3025 29. Entringer S., Buss C., Shirtcliff E. A., Cammack A. L., Yim I. S., Chicz-DeMet A., Sandman C. A.. Attenuation of maternal psychophysiological stress responses and the maternal cortisol awakening response over the course of human pregnancy. Stress (Amsterdam, Netherlands) (2010) 13 258-268. DOI: 10.3109/10253890903349501 30. Entringer S., Buss C., Wadhwa P.. Prenatal stress, telomere biology, and fetal programming of health and disease risk. Science Signaling (2012) 5 pt12. DOI: 10.1126/scisignal.2003580 31. Entringer S., de Punder K., Buss C., Wadhwa P. D.. The fetal programming of telomere biology hypothesis: An update. Philosophical Transactions of the Royal Society B: Biological Sciences (2018) 373. DOI: 10.1098/rstb.2017.0151 32. Farquharson R. G., Jauniaux E., Exalto N., Pregnancy E. S. I. G. f. E.. Updated and revised nomenclature for description of early pregnancy events. Human Reproduction (2005) 20 3008-3011. DOI: 10.1093/humrep/dei167 33. Farren J., Jalmbrant M., Ameye L., Joash K., Mitchell-Jones N., Tapp S., Timmerman D.. Post-traumatic stress, anxiety and depression following miscarriage or ectopic pregnancy: A prospective cohort study. British Medical Journal Open (2016) 6 e011864. DOI: 10.1136/bmjopen-2016-011864 34. Farren J., Jalmbrant M., Falconieri N., Mitchell-Jones N., Bobdiwala S., Al-Memar M., Tapp S.. Posttraumatic stress, anxiety and depression following miscarriage and ectopic pregnancy: A multicenter, prospective, cohort study. American Journal of Obstetrics & Gynecology (2020) 222 367 e361-367 e322. DOI: 10.1016/j.ajog.2019.10.102 35. Farren J., Mitchell-Jones N., Verbakel J. Y., Timmerman D., Jalmbrant M., Bourne T.. The psychological impact of early pregnancy loss. Human Reproduction Update (2018) 24 731-749. DOI: 10.1093/humupd/dmy025 36. Giannandrea S. A., Cerulli C., Anson E., Chaudron L. H.. Increased risk for postpartum psychiatric disorders among women with past pregnancy loss. Journal of Women's Health (2013) 22 760-768. DOI: 10.1089/jwh.2012.4011 37. Glynn L. M., Schetter C. D., Hobel C. J., Sandman C. A.. Pattern of perceived stress and anxiety in pregnancy predicts preterm birth. Health Psychology (2008) 27 43-51. DOI: 10.1037/0278-6133.27.1.43 38. Goldstein H., Healy M. J., Rasbash J.. Multilevel time series models with applications to repeated measures data. Statistics in Medicine (1994) 13 1643-1655. DOI: 10.1002/sim.4780131605 39. Gong X., Hao J., Tao F., Zhang J., Wang H., Xu R.. Pregnancy loss and anxiety and depression during subsequent pregnancies: Data from the C-ABC study. European Journal of Obstetrics & Gynecology and Reproductive Biology (2013) 166 30-36. DOI: 10.1016/j.ejogrb.2012.09.024 40. Graignic-Philippe R., Dayan J., Chokron S., Jacquet A. Y., Tordjman S.. Effects of prenatal stress on fetal and child development: A critical literature review. Neuroscience and Biobehavioral Reviews (2014) 43 137-162. DOI: 10.1016/j.neubiorev.2014.03.022 41. Gravensteen I. K., Jacobsen E. M., Sandset P. M., Helgadottir L. B., Radestad I., Sandvik L., Ekeberg O.. Anxiety, depression and relationship satisfaction in the pregnancy following stillbirth and after the birth of a live-born baby: A prospective study. BMC Pregnancy and Childbirth (2018) 18 41. DOI: 10.1186/s12884-018-1666-8 42. Gyllenhammer L. E., Entringer S., Buss C., Wadhwa P. D.. Developmental programming of mitochondrial biology: A conceptual framework and review. Philosophical Transactions of the Royal Society B: Biological Sciences (2020) 287 20192713. DOI: 10.1098/rspb.2019.2713 43. Haas B. W., Canli T.. Emotional memory function, personality structure and psychopathology: A neural system approach to the identification of vulnerability markers. Brain Research Reviews (2008) 58 71-84. DOI: 10.1016/j.brainresrev.2007.10.014 44. Heim C. M., Entringer S., Buss C.. Translating basic research knowledge on the biological embedding of early-life stress into novel approaches for the developmental programming of lifelong health. Psychoneuroendocrinology (2019) 105 123-137. DOI: 10.1016/j.psyneuen.2018.12.011 45. Heron K. E., Smyth J. M.. Is intensive measurement of body image reactive? A two-study evaluation using ecological momentary assessment suggests not. Body Image (2013) 10 35-44. DOI: 10.1016/j.bodyim.2012.08.006 46. Hinz A., Daig I., Petrowski K., Brahler E.. Mood in the German population: Norms of the multidimensional mood questionnaire MDBF. Psychotherapie Psychosomatik Medizinische Psychologie (2012) 62 52-57. DOI: 10.1055/s-0031-1297960 47. Horesh D., Nukrian M., Bialik Y.. To lose an unborn child: Post-traumatic stress disorder and major depressive disorder following pregnancy loss among Israeli women. General Hospital Psychiatry (2018) 53 95-100. DOI: 10.1016/j.genhosppsych.2018.02.003 48. Hufford M. R., Shields A. L., Shiffman S., Paty J., Balabanis M.. Reactivity to ecological momentary assessment: An example using undergraduate problem drinkers. Psychology of Addictive Behaviors (2002) 16 205-211. DOI: 10.1037/0893-164x.16.3.205 49. Hughes P. M., Turton P., Evans C. D. H.. Stillbirth as risk factor for depression and anxiety in the subsequent pregnancy: Cohort study. British Medical Journal (1999) 318 1721-1724. DOI: 10.1136/bmj.318.7200.1721 50. Hunfeld J. A., Agterberg G., Wladimiroff J. W., Passchier J.. Quality of life and anxiety in pregnancies after late pregnancy loss: A case-control study. Prenatal Diagnosis (1996) 16 783-790. DOI: 10.1002/(SICI)1097-0223(199609)16:9<783::AID-PD943>3.0.CO;2-7 51. Hunter A., Tussis L., MacBeth A.. The presence of anxiety, depression and stress in women and their partners during pregnancies following perinatal loss: A meta-analysis. Journal of Affective Disorders (2017) 223 153-164. DOI: 10.1016/j.jad.2017.07.004 52. Jacob L., Polly I., Kalder M., Kostev K.. Prevalence of depression, anxiety, and adjustment disorders in women with spontaneous abortion in Germany – A retrospective cohort study. Psychiatry Research (2017) 258 382-386. DOI: 10.1016/j.psychres.2017.08.064 53. Jones R. H.. Longitudinal data with serial correlation: A state-space approach (1993) 54. Jones A., Remmerswaal D., Verveer I., Robinson E., Franken I. H. A., Wen C. K. F., Field M.. Compliance with ecological momentary assessment protocols in substance users: A meta-analysis. Addiction (2019) 114 609-619. DOI: 10.1111/add.14503 55. Kersting A., Kroker K., Steinhard J., Ludorff K., Wesselmann U., Ohrmann P., Arolt V.. Complicated grief after traumatic loss: A 14-month follow up study. European Archives of Psychiatry and Clinical Neuroscience (2007) 257 437-443. DOI: 10.1007/s00406-007-0743-1 56. Klebanoff M. A., Shiono P. H., Rhoads G. G.. Outcomes of pregnancy in a national sample of resident physicians. New England Journal of Medicine (1990) 323 1040-1045. DOI: 10.1056/NEJM199010113231506 57. Krosch D. J., Shakespeare-Finch J.. Grief, traumatic stress, and posttraumatic growth in women who have experienced pregnancy loss. Psychological Trauma: Theory, Research, Practice, and Policy (2017) 9 425-433. DOI: 10.1037/tra0000183 58. Lampert T., Hoebel J., Kuntz B.. Socioeconomic status and subjective social status measurement in KiGGS wave 2. Journal of Health Monitoring (2018) 3. DOI: 10.17886/RKI-GBE-2018-033 59. Lautarescu A., Craig M. C., Glover V., Clow A., Smyth N.. Stress and brain health: Across the life course (2020) 150 17-40. DOI: 10.1016/bs.irn.2019.11.002 60. Lawn J. E., Blencowe H., Waiswa P., Amouzou A., Mathers C., Hogan D., Flenady V.. Stillbirths: Rates, risk factors, and acceleration towards 2030. The Lancet (2016) 387 587-603. DOI: 10.1016/s0140-6736(15)00837-5 61. Lazarides C., Ward E. B., Buss C., Chen W. P., Voelkle M. C., Gillen D. L., Wadhwa P. D.. Psychological stress and cortisol during pregnancy: An ecological momentary assessment (EMA)-Based within- and between-person analysis. Psychoneuroendocrinology (2020) 121 104848. DOI: 10.1016/j.psyneuen.2020.104848 62. Lindsay K. L., Buss C., Wadhwa P. D., Entringer S.. The interplay between nutrition and stress in pregnancy: Implications for fetal programming of brain development. Biological Psychiatry (2018). DOI: 10.1016/j.biopsych.2018.06.021 63. Littell R. C., Milliken G. A., Stroup W. W., Wolfinger R. D., Oliver S.. SAS For mixed models (2006) 64. Little R. J., Rubin D. B.. (2002) 200-220 65. MacDorman M. F., Kirmeyer S. E., Wilson E. C.. Fetal and perinatal mortality, United States, 2006. National Vital Statistics Reports (2012) 60 1-22 66. Magnus M. C., Wilcox A. J., Morken N. H., Weinberg C. R., Haberg S. E.. Role of maternal age and pregnancy history in risk of miscarriage: Prospective register based study. BMJ – British Medical Journal (2019) 364 l869. DOI: 10.1136/bmj.l869 67. Milad M. P., Klock S. C., Moses S., Chatterton R.. Stress and anxiety do not result in pregnancy wastage. Human Reproduction (1998) 13 68. movisens G.. (2020) 69. Mukherjee S., Velez Edwards D. R., Baird D. D., Savitz D. A., Hartmann K. E.. Risk of miscarriage among black women and white women in a U.S. Prospective Cohort Study. American Journal of Epidemiology (2013) 177 1271-1278. DOI: 10.1093/aje/kws393 70. Murphy S. L., Mathews T. J., Martin J. A.. Annual summary of vital statistics: 2013–2014. Pediatrics (2017) 139 71. Nelson D. B., Grisso J. A., Joffe M. M., Brensinger C., Shaw L., Datner E.. Does stress influence early pregnancy loss?. Annals of Epidemiology (2003) 13 72. Pinheiro J., Bates D., DebRoy S., Sarkar D., Core Team R. (2018) 73. Podsakoff P. M., MacKenzie S. B., Podsakoff N. P.. Sources of method bias in social science research and recommendations on how to control it. Annual Review of Psychology (2012) 63 539-569. DOI: 10.1146/annurev-psych-120710-100452 74. Price S. K.. Prevalence and correlates of pregnancy loss history in a national sample of children and families. Maternal and Child Health Journal (2006) 10 489-500. DOI: 10.1007/s10995-006-0123-x 75. Qu F., Wu Y., Zhu Y. H., Barry J., Ding T., Baio G., Muscat R.. The association between psychological stress and miscarriage: A systematic review and meta-analysis. Scientific Reports (2017) 7 1731. DOI: 10.1038/s41598-017-01792-3 76. Raudenbush S. W., Bryk A. S.. Hierarchical linear models: Applications and data analysis methods (2002) 1 77. R Development Core Team. (2018). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from https://www.R-project.org/.. (2018) 78. Regan A. K., Gissler M., Magnus M. C., Håberg S. E., Ball S., Malacova E., Nassar N.. Association between interpregnancy interval and adverse birth outcomes in women with a previous stillbirth: An international cohort study. The Lancet (2019) 393 1527-1535. DOI: 10.1016/s0140-6736(18)32266-9 79. Rini C. K., Dunkel-Schetter C., Wadhwa P. D., Sandman C. A.. Psychological adaptation and birth outcomes: The role of personal resources, stress, and sociocultural context in pregnancy. Health Psychology (1999) 18 333-345. DOI: 10.1037//0278-6133.18.4.333 80. Robertson-Blackmore E., Putnam F. W., Rubinow D. R., Matthieu M., Hunn J. E., Putnam K. T., Moynihan J. A.. Antecedent trauma exposure and risk of depression in the perinatal period. Journal of Clinical Psychiatry (2013) 74 e942-e948. DOI: 10.4088/JCP.13m08364 81. Rowan P. J., Cofta-Woerpel L., Mazas C. A., Vidrine J. I., Reitzel L. R., Cinciripini P. M., Wetter D. W.. Evaluating reactivity to ecological momentary assessment during smoking cessation. Experimental and Clinical Psychopharmacology (2007) 15 382-389. DOI: 10.1037/1064-1297.15.4.382 82. Schetter C. D., Saxbe D., Cheadle A., Guardino C.. Postpartum depressive symptoms following consecutive pregnancies: Stability, change, and mechanisms. Clinical Psychological Science (2016) 4 909-918. DOI: 10.1177/2167702616644894 83. Schmiedek F., Lovden M., Lindenberger U.. Keeping it steady: Older adults perform more consistently on cognitive tasks than younger adults. Psychological Science (2013) 24 1747-1754. DOI: 10.1177/0956797613479611 84. Smyth J. M., Stone A. A.. Ecological momentary assessment research in behaviroal medicine. Journal of Happiness Studies (2003) 4 35-52. DOI: 10.1023/A:1023657221954 85. Snijders T. A. B., Bosker R. J.. Multilevel analysis: An introduction to basic and advanced multilevel modeling (2012) 86. Stanton A. L., Lobel M., Sears S., DeLuca R. S.. Psychosocial aspects of selected issues in women's reproductive health: Current status and future directions. Journal of Consulting and Clinical Psychology (2002) 70 751-770. DOI: 10.1037/0022-006x.70.3.751 87. Steyer R., Schwenkmezger P., Notz P., Eid M.. (1994) 88. Steyer R., Schwenkmezger P., Notz P., Eid M.. (1997) 89. Sundermann A. C., Hartmann K. E., Jones S. H., Torstenson E. S., Velez Edwards D. R.. Interpregnancy interval after pregnancy loss and risk of repeat miscarriage. Obstetrics & Gynecology (2017) 130 1312-1318. DOI: 10.1097/AOG.0000000000002318 90. Tarabulsy G. M., Pearson J., Vaillancourt-Morel M. P., Bussieres E. L., Madigan S., Lemelin J. P., Duchesneau A. A.. Meta-Analytic findings of the relation between maternal prenatal stress and anxiety and child cognitive outcome. Journal of Developmental and Behavioral Pediatrics (2014) 35 38-43. DOI: 10.1097/Dbp.0000000000000003 91. Tavares Da Silva F., Gonik B., McMillan M., Keech C., Dellicour S., Bhange S., Tila M.. Stillbirth: Case definition and guidelines for data collection, analysis, and presentation of maternal immunization safety data. Vaccine (2016) 34 6057-6068. DOI: 10.1016/j.vaccine.2016.03.044 92. Toffol E., Koponen P., Partonen T.. Miscarriage and mental health: Results of two population-based studies. Psychiatry Research (2013) 205 151-158. DOI: 10.1016/j.psychres.2012.08.029 93. Tsartsara E., Johnson M. P.. The impact of miscarriage on women's pregnancy-specific anxiety and feelings of prenatal maternal-fetal attachment during the course of a subsequent pregnancy: An exploratory follow-up study. Journal of Psychosomatic Obstetrics & Gynecology (2006) 27 173-182. DOI: 10.1080/01674820600646198 94. Turton P., Evans C., Hughes P.. Long-term psychosocial sequelae of stillbirth: Phase II of a nested case-control cohort study. Archives of Women's Mental Health (2009) 12 35-41. DOI: 10.1007/s00737-008-0040-7 95. Turton P., Hughes P., Evans C. D., Fainman D.. Incidence, correlates and predictors of post-traumatic stress disorder in the pregnancy after stillbirth. British Journal of Psychiatry (2001) 178 556-560. DOI: 10.1192/bjp.178.6.556 96. Volgsten H., Jansson C., Svanberg A. S., Darj E., Stavreus-Evers A.. Longitudinal study of emotional experiences, grief and depressive symptoms in women and men after miscarriage. Midwifery (2018) 64 23-28. DOI: 10.1016/j.midw.2018.05.003 97. Wadhwa P. D., Entringer S., Buss C., Lu M. C.. The contribution of maternal stress to preterm birth: Issues and considerations. Clinics in Perinatology (2011) 38 351-384. DOI: 10.1016/j.clp.2011.06.007 98. Walsh K., McCormack C. A., Webster R., Pinto A., Lee S., Feng T., Krakovsky H. S.. Maternal prenatal stress phenotypes associate with fetal neurodevelopment and birth outcomes. Proceedings of the National Academy of Sciences of the USA (2019) 116 23996-24005. DOI: 10.1073/pnas.1905890116 99. Wen C. K. F., Schneider S., Stone A. A., Spruijt-Metz D.. Compliance with mobile ecological momentary assessment protocols in children and adolescents: A systematic review and meta-analysis. Journal of Internet Research (2017) 19 e132. DOI: 10.2196/jmir.6641 100. Woods-Giscombe C. L., Lobel M., Crandell J. L.. The impact of miscarriage and parity on patterns of maternal distress in pregnancy. Research in Nursing & Health (2010) 33 316-328. DOI: 10.1002/nur.20389
--- title: 'Association between risk of dementia and very late-onset schizophrenia-like psychosis: a Swedish population-based cohort study' authors: - J. Stafford - J. Dykxhoorn - A. Sommerlad - C. Dalman - J. B. Kirkbride - R. Howard journal: Psychological Medicine year: 2023 pmcid: PMC9975996 doi: 10.1017/S0033291721002099 license: CC BY 4.0 --- # Association between risk of dementia and very late-onset schizophrenia-like psychosis: a Swedish population-based cohort study ## Abstract ### Background Although the incidence of psychotic disorders among older people is substantial, little is known about the association with subsequent dementia. We aimed to examine the rate of dementia diagnosis in individuals with very late-onset schizophrenia-like psychosis (VLOSLP) compared to those without VLOSLP. ### Methods Using Swedish population register data, we established a cohort of 15 409 participants with VLOSLP matched by age and calendar period to 154 090 individuals without VLOSLP. Participants were born between 1920 and 1949 and followed from their date of first International Classification of Diseases [ICD], Revisions 8–10 (ICD-$\frac{8}{9}$/10) non-affective psychotic disorder diagnosis after age 60 years old (or the same date for matched participants) until the end of follow-up (30th December 2011), emigration, death, or first recorded ICD-$\frac{8}{9}$/10 dementia diagnosis. ### Results We found a substantially higher rate of dementia in individuals with VLOSLP [hazard ratio (HR): 4.22, $95\%$ confidence interval ($95\%$ CI) 4.05–4.41]. Median time-to-dementia-diagnosis was $75\%$ shorter in those with VLOSLP (time ratio: 0.25, $95\%$ CI 0.24–0.26). This association was strongest in the first year following VLOSLP diagnosis, and attenuated over time, although dementia rates remained higher in participants with VLOSLP for up to 20 years of follow-up. This association remained after accounting for potential misdiagnosis (2-year washout HR: 2.22, $95\%$ CI 2.10–2.36), ascertainment bias (HR: 2.89, $95\%$ CI 2.75–3.04), and differing mortality patterns between groups (subdistribution HR: 2.89, $95\%$ CI 2.77–3.03). ### Conclusions Our findings demonstrate that individuals with VLOSLP represent a high-risk group for subsequent dementia. This may be due to early prodromal changes for some individuals, highlighting the importance of ongoing symptom monitoring in people with VLOSLP. ## Background Although non-affective psychotic disorders typically have their first onset in adolescence or early adulthood (Kessler et al., 2007), recent population-based evidence suggests a second peak of incidence after 60 years old, particularly in women (Stafford, Howard, & Kirkbride, 2018; Stafford, Howard, Dalman, & Kirkbride, 2019). This is referred to as very late-onset schizophrenia-like psychosis (VLOSLP) in those aged over 60 years old (Howard, Rabins, Seeman, & Jeste, 2000). There is ongoing debate about the aetiology of VLOSLP and its relationship with neurodegeneration and dementia (Brodaty, Sachdev, Koschera, Monk, & Cullen, 2003; Vahia et al., 2010; Van Assche, Morrens, Luyten, Van de Ven, & Vandenbulcke, 2017). Understanding this issue is critical to providing insight into the aetiologies of VLOSLP and dementia, which could then inform clinical practice. Data on the association between VLOSLP and cognitive decline are, however, sparse, and most studies in this area have been limited by small, unrepresentative samples, and cross-sectional designs (Van Assche et al., 2017). Two longitudinal, population-based studies have focused on VLOSLP and dementia, reporting elevated risk of dementia in people with VLOSLP, but these studies had relatively short follow-up periods (maximum 7 years) (Kørner, Lopez, Andersen, & Kessing, 2008; Kørner, Lopez, Lauritzen, Andersen, & Kessing, 2009). This is problematic because dementia neuropathology may develop over decades (Bateman et al., 2012; Villemagne et al., 2013), with symptoms emerging up to 12 years before dementia diagnosis (Amieva et al., 2008), meaning that it is not possible to determine whether the observed association is because VLOSLP is a cause of dementia, or an early symptom. Long follow-up periods are required to fully characterise the relationship between VLOSLP and dementia and reduce the risk of protopathic bias. We examined the rate of subsequent dementia diagnosis in a large, Swedish, population-based cohort of individuals with VLOSLP and an age-matched comparison group without VLOSLP. We hypothesised that individuals with VLOSLP would have a higher rate of subsequent dementia diagnosis, and a shorter time-to-dementia-diagnosis than those without VLOSLP. We considered death as a potential competing risk, given higher mortality rates in those with psychotic disorders (Hayes, Marston, Walters, King, & Osborn, 2017), including VLOSLP (Talaslahti et al., 2015). However, we hypothesised that findings would not be explained by this competing risk or by ascertainment bias related to previous contact with health services in those with VLOSLP. Finally, given the lack of previous data in this area, we expected that the risk of dementia associated with VLOSLP would be similar across demographic subgroups, which we assessed by testing interactions between VLOSLP and sex, education level, and family liability for psychotic disorder. ## Study design and population We used Psychiatry Sweden data, a nationwide population register linkage for mental health research. We established a matched cohort design, including individuals living in Sweden born between 1920 and 1949 who were first diagnosed with an International Classification of Diseases (ICD; Eight, Ninth and Tenth Revision codes (WHO, 1992), online Supplementary Table S1) non-affective psychotic disorder in the NPR at age 60 years or older with no prior dementia diagnosis. We constructed an age-period-cohort-matched comparison group (matched within the same birth year), without a diagnosis of non-affective psychotic disorder in the NPR (10 matches per person with VLOSLP). Matched participants were required to be alive, living in Sweden, and without a dementia diagnosis on the date of their matched individual's VLOSLP diagnosis. The date of VLOSLP diagnosis, or that of an individual's matched participant, was used as the start of follow-up. Participants were followed until first recorded diagnosis with dementia in the NPR, death, emigration from Sweden, or the end of the follow-up period on 31 December 2011, whichever was earliest. ## Outcome Our main outcome was first dementia diagnosis recorded in the National Patient Register (NPR), which includes inpatient (1973–2011) and outpatient records (2001–2011) (ICD-$\frac{8}{9}$/10 codes listed in online Supplementary Table S2). ## Exposure The primary exposure was VLOSLP diagnosis (ICD codes listed in online Supplementary Table S1) in the NPR (earliest date of ascertainment: 1st January 1980). ## Covariates Data on age, sex, and region of birth were obtained from the Swedish Register of the Total Population. Region of birth was broadly categorised as follows: Sweden, Finland, other Nordic, other European, and other countries. Disposable income at age 60 years was obtained from the Longitudinal Integration Database for Health Insurance and Labour Market Studies (LISA) and grouped into quartiles based on all cohort members with disposable income from all sources (employment, welfare receipts, savings, investments). We obtained data on educational attainment from the LISA, grouped as pre-high school, high school, and post-high school. Familial liability of psychotic disorder was obtained by linking participants to their biological children using the Multigenerational Register and identifying whether any of these children had received a diagnosis of non-affective psychosis (as per codes in online Supplementary Table S1) in the NPR. ## Missing data Individuals with missing data, limited to disposable income and educational attainment, were excluded from the cohort prior to matching ($5.4\%$; online Supplementary Table S3). ## Statistical analysis First, we presented descriptive statistics of the cohort. Second, we used Cox regression to examine dementia rates in those with and without VLOSLP. We initially examined univariable associations between covariates and dementia diagnosis, assessing model fit using Akaike's information criterion (AIC), with lower scores indicating better fit. Next, we added variables with the lowest AIC values individually into a model including sex as an a priori confounder and the matching variable. We retained covariates in a forward-fitting model if they improved model fit, assessed via likelihood ratio test (LRT). We fitted and tested interactions (via LRT) between VLOSLP and sex, education level and family liability for psychotic disorder. We assessed the proportional hazards assumption using Schoenfeld residuals plots and tests. Third, we used accelerated failure time (AFT) models to estimate how much quicker those with VLOSLP were diagnosed with dementia relative to the non-VLOSLP group, expressed via time ratios. We compared the fit of different distributions for the AFT error term (including exponential, Weibull, log-normal or log-logistic) via AIC. We re-ran our crude and best-fitting model, as identified from Cox regression, in the AFT parameterization. Fourth, we assessed the potential impact of differential patterns of mortality on the association between VLOSLP and dementia using Fine and Gray competing risks regression (Fine & Gray, 1999). To examine the potential effect of misdiagnosis of dementia as VLOSLP, we conducted sensitivity analyses with washout periods, excluding those who were diagnosed with dementia within six months, one year, or two years from VLOSLP diagnosis, and their matched comparison participants, regardless of their outcome status. We conducted a second sensitivity analysis to investigate whether any observed association between VLOSLP and dementia could have been attributable to possible increased detection of dementia in the VLOSLP group. This may have arisen, if for example, they had more contact with health services as a result of VLOSLP than the matched comparison group, leading to more opportunities for detection of dementia. To test this, we divided the matched comparison group into those with and without a recorded inpatient or outpatient diagnosis for any condition in the year either side of cohort entry. If the rate of dementia was only higher in the VLOSLP group relative to the non-VLOSLP group without a recent diagnosis (but not relative to the non-VLOSLP group with a diagnosis during this time), this may suggest that findings were due to ascertainment bias. ## Cohort characteristics Our cohort consisted of 15 409 participants diagnosed with VLOSLP and 154 090 matched participants without VLOSLP. During follow-up, 13 610 ($8\%$) individuals were diagnosed with dementia (VLOSLP: $17.8\%$; non-VLOSLP: $7.1\%$ (χ2[1] = 2200, p = <0.001)). Median age-at-first-dementia-diagnosis was younger in the VLOSLP group (76 years, interquartile range (IQR): 72–81) than the non-VLOSLP group (82 years, IQR: 78–86; Mann−Whitney $z = 37.4$, p ⩽ 0.001). Compared with the non-VLOSLP group, participants with VLOSLP were more likely to be women ($60.6\%$ v. $53.8\%$), have a lower education level (pre-high school education: $59.1\%$ v. $53.2\%$), have a lower disposable income at age 60 years old (lowest quartile: $36.6\%$ v. $26.1\%$), have family liability for psychotic disorder ($5.3\%$ v. $2.5\%$), and were less likely to be Swedish-born ($85.5\%$ v. $90.9\%$) (all p ⩽ 0.001; Table 1). Table 1.Cohort characteristics in those with and without VLOSLPCharacteristicNo VLOSLP ($$n = 154$$ 090)VLOSLP ($$n = 15$$ 409)Median (IQR)Median (IQR)Mann−Whitney testAge-at-diagnosis with dementia [inter-quartile range (IQR)]82 (78–86)76 (72–81)p = <0.001N (%)N (%)χ2 testDementia groupχ2[1] = 2200.00, p = <0.001Dementia10 866 (7.05)2744 (17.81)No dementia143 224 (92.95)12 665 (82.19)Sexχ2[1] = 259.41, p = <0.001Men71 224 (46.22)6078 (39.44)Women82 866 (53.78)9331 (60.56)Educational attainmentχ2[2] = 239.24, p = <0.001Pre-high school81 994 (53.21)9111 (59.13)High school49 827 (32.34)4614 (29.94)Post-high school22 269 (14.45)1684 (10.93)Disposable income at age 60χ2[3] = 2100.00, $p \leq 0.001$Lowest [1]40 236 (26.11)5642 (36.61)236 892 (23.94)4935 (32.03)339 747 (25.79)3007 (19.51)Highest [4]37 215 (24.15)1825 (11.84)Family liability for non-affective psychotic disorderYes3878 (2.52)815 (5.29)χ2[1] = 399.95, $p \leq 0.001$No150 212 (97.48)14 594 (94.71)Region of birthSweden139 999 (90.86)13 173 (85.49)χ2[4] = 497.18, $p \leq 0.001$Other Nordic2573 (1.67)333 (2.16)Finland4819 (3.13)852 (5.53)Other European5321 (3.45)858 (5.57)Other1378 (0.89)193 (1.25) ## Association between VLOSLP and subsequent dementia In Cox regression models, compared with the non-VLOSLP group, we found a higher rate of dementia in participants with VLOSLP [fully adjusted hazard ratio (HR): 4.22, $95\%$ confidence interval ($95\%$CI) 4.05–4.41; Table 2], after adjustment for sex, education level, disposable income, region of birth and family liability for psychotic disorder. There was evidence of non-proportional hazards for the VLOSLP exposure, but not for other covariates (online Supplementary Table S4). Further exploration of this issue (Fig. 1) suggested that the dementia HR associated with VLOSLP was highest in the first year after VLOSLP diagnosis, although rates of dementia remained higher in the VLOSLP group for up to 20 years of follow-up. Fig. 1.Association between very late-onset schizophrenia-like psychosis and dementia during follow-up. Dementia HRs with $95\%$ confidence intervals. Table 2.Association between VLOSLP and incident dementiaVariableModel 1aModel 2b6-month washoutb N excluded: 8756 ($5\%$)1-year washoutb N excluded: 11 451 ($7\%$)2-year washoutb N excluded: 15 356 ($9\%$)N included (%)169 499 [100]169 499 [100]160 743 [95]158 048 [93]154 143 [91]Cox proportional hazards regression HRs for dementia by VLOSLP status ($95\%$ CI)VLOSLP (ref: no VLOSLP)4.21 (4.04–4.39)4.22 (4.05–4.41)3.12 (2.97–3.27)2.76 (2.62–2.91)2.22 (2.10–2.36)Fine and Gray regression subdistribution HRs for dementia by VLOSLP status ($95\%$ CI)VLOSLP (ref: no VLOSLP)2.90 (2.80–3.03)2.89 (2.77–3.03)2.07 (1.97–2.18)1.82 (1.73–1.92)1.45 (1.37–1.53)Weibull AFT model time ratios for dementia by VLOSLP status ($95\%$ CI) VLOSLP (ref: no VLOSLP)0.25 (0.24–0.27)0.25 (0.24–0.26)0.41 (0.40–0.43)0.46 (0.44–0.48)0.56 (0.54–0.58)aModel 1: Adjusted for matching variable.bModel 2: Adjusted for VLOSLP group, sex, education level, family liability for non-affective psychotic disorder, disposable income at age 60, region of birth, and matching variable. The association between VLOSLP and dementia varied by socio-demographic subgroup (online Supplementary Table S5), with evidence of effect modification between VLOSLP status and sex (LRT, p = <0.001), education level (LRT, p = <0.001), and family liability for psychotic disorder (LRT, $$p \leq 0.01$$), respectively. Thus, in the VLOSLP group, we observed a lower risk of dementia among women compared with men (HR: 0.86, $95\%$ CI 0.79–0.92), whereas this pattern was reversed in the comparison group (HR: 1.10, $95\%$ CI 1.05–1.14). In the comparison group, individuals with the highest educational attainment had a lower rate of dementia (HR: 0.93, $95\%$ CI 0.86–0.99), which was not found in the VLOSLP group (HR: 0.98, $95\%$ CI 0.85–1.12). By contrast, in the VLOSLP group, the rate of dementia was lower in those with the lowest educational attainment (HR: 0.80, $95\%$ CI 0.73–0.87), which was not observed in the comparison group (HR: 1.03, $95\%$ CI 0.99–1.08). Family liability for psychotic disorder was associated with a higher rate of dementia in the comparison group (HR: 1.21, $95\%$ CI 1.09–1.35), but not in the VLOSLP group (HR: 0.94, $95\%$ CI 0.80–1.11). Median time-to-dementia-diagnosis in those without VLOSLP was 9.1 years (IQR: 4.3–14.7), relative to 1.9 years (IQR: 0.4–6.0) in those with VLOSLP. In AFT models, a Weibull distribution for baseline survivorship provided best fit to the data (online Supplementary Table S6), and after full adjustment for covariates, a VLOSLP diagnosis was associated with a time ratio of 0.25 ($95\%$CI 0.24–0.26) (Table 2), indicating that time-to-dementia-diagnosis was $75\%$ shorter in this group than those without VLOSLP. ## Mortality as a competing risk Although mortality was higher in those with VLOSLP (fully adjusted HR: 2.85, $95\%$ CI 2.78–2.91; online Supplementary Table S7), the rate of dementia remained significantly higher in the VLOSLP group in a fully adjusted Fine and Gray regression, incorporating death as a competing risk [sub-distribution hazard ratio (SHR): 2.89, $95\%$ CI 2.77–3.03] (Table 2). ## Sensitivity analyses Using washout periods of 6-months, 1-year or 2 years to exclude participants whose dementia may have been misdiagnosed as VLOSLP (and their matched participants without VLOSLP), we found some attenuation in elevated dementia rates in the VLOSLP group, but these remained substantially higher than in the non-VLOSLP group in both fully adjusted Cox regression (2-year washout HR: 2.22, $95\%$ CI 2.10–2.36; Table 2) or competing risks regression (SHR: 1.45, $95\%$ CI 1.37–1.53; Table 2). We conducted a further sensitivity analysis to investigate whether higher rates of dementia in people with VLOSLP were explained by ascertainment bias arising from increased contact with the health system. The results showed that dementia rates were higher in the VLOSLP group relative to both the non-VLOSLP groups who had not received any diagnosis of any condition 12-months either side of cohort entry (HR: 4.90, $95\%$ CI 4.69–5.13), and those who had received a diagnosis during this time (indicative of health service contact) (HR: 2.89, $95\%$ CI 2.75–3.04;Table 3). These results suggest that higher rates of dementia in the VLOSLP group cannot be fully explained by a greater probability of detection of dementia due to greater contact with services. Table 3.Association of VLOSLP with incident dementia: sensitivity analyses to take into account differences in detection by VLOSLP status due to previous contact with health servicesFully adjusted Cox proportional hazards regressionVariableHR ($95\%$CI)6-month washout HR ($95\%$CI) N excluded: 8756 ($5\%$)1-year washout HR ($95\%$CI) N excluded: 11 451 ($7\%$)2-year washout HR ($95\%$CI) N excluded: 15 356 ($9\%$)N included (%)169 499 [100]160 743 [95]158 048 [93]154 143 [91]VLOSLP (ref: comparison group, no diagnosis near date of entry)4.90 (4.69–5.13)3.63 (3.45–3.82)3.23 (3.06–3.41)2.61 (2.46–2.77)VLOSLP (ref: comparison group, any diagnosis near date of entry)a2.89 (2.75–3.04)2.09 (1.98–2.21)1.84 (1.73–1.95)1.47 (1.38–1.56)Fully adjusted Fine and Gray competing risks regressionVariableSHR ($95\%$ CI)6-month washout SHR ($95\%$CI) N excluded: 8756 ($5\%$)1-year washout SHR ($95\%$CI) N excluded: 11 451 ($7\%$)2-year washout SHR ($95\%$CI) N excluded: 15 356 ($9\%$)N included (%)169 499 [100]160 743 [95]158 048 [93]154 143 [91]VLOSLP (ref: comparison group, no diagnosis near date of entry)3.08 (2.94–3.22)2.21 (2.09–2.32)1.94 (1.84–2.05)1.54 (1.46–1.64)VLOSLP (ref: comparison group, any diagnosis near date of entry)a2.52 (2.39–2.66)1.79 (1.69–1.90)1.56 (1.47–1.66)1.24 (1.16–1.32)Abbreviations: HR, hazard ratio; SHR, subdistribution hazard ratio; VLOSLP, very late-onset schizophrenia-like psychosis. All analyses are adjusted for: VLOSLP group, sex, education level, family liability for non-affective psychotic disorder, disposable income at age 60, region of birth and matching variable.aAny hospital diagnosis on the year either side of entry into the study (except psychotic disorders or dementia). To examine the joint effects of possible misdiagnosis and ascertainment bias on our results we re-ran these sensitivity analyses together (Table 3). The rate of dementia remained higher in the VLOSLP group relative to both comparison groups (v. participants in the non-VLOSLP group without diagnosis, with a 2-year washout HR: 2.61, $95\%$ CI 2.46–2.77; v. participants in the non-VLOSLP group with diagnosis, 2-year washout HR: 1.47, $95\%$ CI 1.38–1.56); similar patterns were present in the competing risk regression model (Table 3). ## Summary of findings We found a substantially higher rate of subsequent dementia diagnosis among individuals with VLOSLP, who were diagnosed with dementia $75\%$ more quickly than age-period-cohort matched comparison participants without VLOSLP. The relationship between VLOSLP and dementia persisted in our most conservative sensitivity analyses, suggesting that these findings are unlikely to be fully explained by the competing risk of mortality, possible misdiagnosed dementia, or ascertainment bias. Although the association between VLOSLP and dementia was strongest in the first year after VLOSLP diagnosis, the rate of dementia remained higher among the VLOSLP group over up to 20 years of follow-up. Taken together, our findings are consistent with the possibilities that VLOSLP is a prodromal feature of dementia, or that VLOSLP independently confers risk of later dementia via other mechanisms. ## Strengths and limitations To our knowledge, this is the largest study to date to examine rates of dementia in those with VLOSLP, using a comprehensive, nationwide cohort of people aged over 60 years old, and with the longest follow-up duration (up to 30 years). No previous studies have had sufficiently large samples and long enough follow-up periods to allow adequate detection of incident dementia in those with VLOSLP and characterise the consistency of associations over time. We were also able to consider other potential explanations for findings, including misdiagnosis, ascertainment bias and competing risks. Our use of age-period-cohort matched comparison participants implicitly controlled for several unobserved effects which may have otherwise arisen as a function of these variables, including changes in diagnostic systems over time or recording of dementia in clinical care. We acknowledge several limitations. While the specificity of dementia diagnoses in the Swedish registers is high, sensitivity is relatively low, with around half of dementia cases in the population not recorded in the registers (Rizzuto et al., 2018). This will have under-estimated the true rate of dementia in the population in this study. However, we would not have expected this bias to have acted differentially by VLOSLP status, except in relation to ascertainment bias as highlighted above, which we examined via sensitivity analyses. Although our sensitivity analysis took into account differences in the number of health service contacts between those with and without VLOSLP, we were unable to fully account for differences in the nature of contacts. It remains possible that our findings partly reflect greater involvement of mental health specialists and increased focus on identifying signs of cognitive decline among the VLOSLP group during follow-up. A validation study reported an average delay of 5.5 years in recording of dementia diagnoses in the registers (Rizzuto et al., 2018) which may have compromised the precision of estimates of time between VLOSLP and dementia diagnoses. Misclassification between dementia subtypes is common in the NPR (Rizzuto et al., 2018), hence we were not able to examine whether the association varied according to different diagnostic subtypes. Further, misdiagnosis between schizophrenia and dementia remains a possibility. Pre-existing cognitive deficits are common in schizophrenia (Bora, 2015) and could contribute to misdiagnosis of dementia in some cases. Conversely, early signs of dementia could be mistaken for cognitive deficits and negative symptoms in schizophrenia. Previous studies have shown that comorbidities are often under-detected in people with schizophrenia (Roberts, Roalfe, Wilson, & Lester, 2006; Smith, Langan, Mclean, Guthrie, & Mercer, 2013), possibly due to diagnostic overshadowing, whereby comorbidities are misattributed to psychotic symptoms (Viron & Stern, 2010). Although we conducted a sensitivity analysis to mitigate against misdiagnosis, and we would expect clinicians to be aware of the complexities of dementia diagnoses in the context of psychotic disorders, we cannot fully exclude the possibility of misdiagnosis in either direction. While we adjusted for length of education, we were unable to obtain more detailed information on educational attainment or cognitive functioning from the Swedish registers, which would have allowed greater insight into baseline cognitive function and reserve. Finally, although we excluded potential participants diagnosed with non-affective psychotic disorders in the registers before 60 years old, register data on psychiatric disorders were only available since 1973 meaning that we were unable to identify participants in either the VLOSLP or non-VLOSLP group who developed a diagnosable psychotic disorder before this date. This may have led us to include some participants with adult-onset psychosis in both the VLOSLP and non-VLOSLP groups. We would expect a higher proportion of these participants to be in the VLOSLP group, hence the accelerated rate of dementia in this group may partly reflect longer-standing psychiatric morbidity; previous studies have shown that adult-onset psychotic disorders are also associated with an increased risk of dementia (Cai & Huang, 2018; Ribe et al., 2015). ## Meaning of findings Our finding of an elevated rate of dementia diagnosis in individuals with VLOSLP corresponds with an earlier Danish study which found an elevated rate of dementia among people with late-onset schizophrenia (rate ratio (RR): 3.47, $95\%$ CI 2.19–5.5) and VLOSLP (RR: 3.15, $9\%$ CI 1.93–5.14) compared to osteo-arthritis patients (Kørner et al., 2009), and a recent cohort study conducted in Israel also demonstrating an association between VLOSLP and dementia (HR: 2.67, $95\%$ CI 1.82–3.91) (Kodesh et al., 2020). However, both studies were limited by follow-up periods shorter than 5 years. Previous studies have also demonstrated associations between dementia and younger-onset psychotic disorders (Cai & Huang, 2018). A Danish register study found that schizophrenia was associated with more than 2-fold increased rates of dementia (incidence rate ratio: 2.13, $95\%$ CI 2.00–2.27) (Ribe et al., 2015). In a cohort study of men in Western Australia, Almeida et al. [ 2018] found an association between dementia and VLOSLP (HR: 2.22, $95\%$ CI 1.74–2.84), and younger-onset psychotic disorders (<65 years HR: 2.73, $95\%$ CI 2.34–3.18). We have extended previous findings by quantifying time-to-dementia-diagnosis, estimated to be $75\%$ sooner in those with VLOSLP; using longer follow-up to clarify the consistency and persistence of the association; accounting for several potential sources of bias; and examining differences by socio-demographic subgroup. Our results highlight the importance of monitoring cognition and function in patients with VLOSLP, particularly in the first few years following diagnosis. These findings have clinical implications for treatment planning and may warrant reflection in clinical guidelines (Mueller, Thompson, Harwood, Bagshaw, & Burns, 2017). We observed differential patterns of association between dementia and educational attainment in VLOSLP and non-VLOSLP groups. In the non-VLOSLP group, higher education was associated with lower dementia risk, consistent with evidence regarding cognitive reserve (Sharp & Gatz, 2011). In contrast, for those with VLOSLP, lower educational attainment was associated with a reduced rate of dementia. While we are unsure of the mechanism leading to this finding, one possibility is that individuals with VLOSLP from a lower education group may be less likely to contact services or to have dementia symptoms detected. The rate of dementia diagnosis was particularly elevated for the VLOSLP group in the first year after VLOSLP diagnosis. As evidenced by sensitivity analyses, this may partly reflect misdiagnosis, and ascertainment bias due to increased contact with health services. However, importantly, the rate of dementia remained higher among the VLOSLP group for up to 20 years of follow-up. This finding is consistent with several explanations. One possibility is that psychotic disorders, including VLOSLP, could increase risk for dementia via factors such as poor physical health (Bushe & Holt, 2004; Crump, Winkleby, Sundquist, & Sundquist, 2013; Hennekens, Hennekens, Hollar, Casey, & Raton, 2005; Osborn et al., 2008), and associated health behaviours including smoking, poor diet and reduced physical activity (McCreadie, 2002, 2003). In addition, cognitive impairment, a core component of schizophrenia (Bora, 2015), could increase risk for dementia via reduced cognitive or brain reserve (Barnett, Salmond, Jones, & Sahakian, 2006), whereby those with a lower level of baseline cognitive functioning may require less neuropathology before meeting the clinical threshold for dementia diagnosis (Stern, 2006). Although these pathways may be more plausible in relation to chronic schizophrenia, they could also apply to those with VLOSLP, some of whom may have had longstanding subthreshold psychotic symptoms or schizotypal traits prior to VLOSLP diagnosis (Kay & Roth, 1961). In addition, stress and shared personality factors, including high levels of neuroticism, could partially account for these associations, having been identified as predictors of both schizophrenia (Howes et al., 2004; Lonnqvist et al., 2009; Van Os & Jones, 2001) and dementia (Johansson et al., 2010; Sindi et al., 2017; Terracciano et al., 2021). The relationship between VLOSLP and dementia could also reflect shared genetic vulnerabilities (Lyketsos & Peters, 2015), although a recent study using data from the English Longitudinal Study of Ageing found that, among community-dwelling adults aged >50 years, polygenic score for schizophrenia was associated with cognitive impairment at baseline, but not cognitive decline over 10 years of follow-up (Kępińska et al., 2020). Conversely, it is possible that VLOSLP symptoms could represent an early marker of Alzheimer's disease neuropathology, given that amyloid-β has been found to accumulate in the brain over several decades before dementia onset (Villemagne et al., 2013), and these neuropathological changes may lead to emergence of cognitive and non-cognitive symptoms of dementia prior to diagnostic threshold being reached. In line with this, the concept of mild behavioural impairment posits that late-onset neuropsychiatric symptoms, including apathy, emotion dysregulation, reduced impulse control, agitation, social inappropriateness, and psychotic symptoms, reflect possible early markers of preclinical dementia neuropathology (Ismail et al., 2017). Further, depression has consistently been found to be associated with subsequent dementia (Diniz, Butters, Albert, Dew, & Reynolds, 2013; Ownby, Crocco, Acevedo, John, & Loewenstein, 2007), and several cohort studies have demonstrated stronger associations between late-onset depression and dementia, relative to early or mid-life depression (Heser et al., 2013; Karlsson et al., 2015; Li et al., 2011). A recent cohort study with 28 years of follow-up demonstrated that the association between dementia and depression symptoms became apparent 11 years before dementia diagnosis, potentially reflecting prodromal dementia, albeit more than a decade before diagnosis (Singh-Manoux et al., 2017). It is clear that psychiatric symptoms with a late age-at-onset, including psychosis, are a marker of increased risk of dementia during the pre-cognitive impairment phase, and this group may be a priority group for future early intervention research in dementia (Sperling et al., 2011). Furthermore, while antipsychotic medication is effective for psychosis symptoms in people with VLOSLP and well-tolerated (Howard et al., 2018), further attention may be required on the long-term effects to guide antipsychotic use, considering the potential for adverse effects of antipsychotic use in dementia (Wang et al., 2005). ## Conclusions In this nationally representative sample, we found a strong and persistent association between very late-onset psychotic disorders and dementia. Our findings indicate that further investigation of the potential pathways between psychosis and dementia is warranted to fully understand and characterise how psychiatric morbidities accumulate together to perpetuate disparities in mental health disorders across the life course, and to inform future management approaches. ## Financial support JBK was supported by a Sir Henry Dale Fellowship (Wellcome Trust/Royal Society grant number: 101272/Z/13/Z); AS was supported by the UCL/Wellcome Trust Institutional Strategic Support Fund (grant number: 204841/Z/16/Z); JS was supported by a Postdoctoral Bridging Fellowship from the National Institute for Health Research, University College London Hospital, Biomedical Research Centre; JS/JD/AS/JBK/RH were supported by the National Institute for Health Research, University College London Hospital, Biomedical Research Centre; CD was supported by the Swedish Research Council (grant number: 523-2010-1052). Sponsors did not contribute to study design, data analysis, interpretation of data, writing of the report, or the decision to submit the article for publication. ## Conflict of interest The authors have declared that there are no conflicts of interest in relation to the subject of this study. ## References 1. Almeida O. P., Ford A. H., Hankey G. J., Yeap B. B., Golledge J., Flicker L.. Risk of dementia associated with psychotic disorders in later life: The Health in Men Study (HIMS). Psychological Medicine (2018) 49 232-242. PMID: 29564993 2. Amieva H., Le Goff M., Millet X., Orgogozo J. M., Pérès K., Barberger-Gateau P., Dartigues J. F.. Prodromal Alzheimer's disease: Successive emergence of the clinical symptoms. Annals of Neurology (2008) 64 492-498. PMID: 19067364 3. Barnett J. H., Salmond C. H., Jones P. B., Sahakian B. J.. Cognitive reserve in neuropsychiatry. Psychological Medicine (2006) 36 1053-1064. PMID: 16854246 4. Bateman R. J., Xiong C., Benzinger T. L. S., Fagan A. M., Goate A., Fox N. C., Morris J. C.. Clinical and biomarker changes in dominantly inherited Alzheimer's disease. New England Journal of Medicine (2012) 367 795-804. PMID: 22784036 5. Bora E.. Neurodevelopmental origin of cognitive impairment in schizophrenia. Psychological Medicine (2015) 45 1-9. PMID: 25065902 6. Brodaty H., Sachdev P., Koschera A., Monk D., Cullen B.. Long-term outcome of late-onset schizophrenia: 5-year follow-up study. The British Journal of Psychiatry (2003) 183 213-219. PMID: 12948993 7. Bushe C., Holt R.. Prevalence of diabetes and impaired glucose tolerance in patients with schizophrenia. British Journal of Psychiatry (2004) 184 s67-s71 8. Cai L., Huang J.. Schizophrenia and risk of dementia: A meta-analysis study. Neuropsychiatric Disease and Treatment (2018) 14 2047-2055. PMID: 30147318 9. Crump C., Winkleby M. A., Sundquist K., Sundquist J.. Comorbidities and mortality in persons with schizophrenia: A Swedish national cohort study. American Journal of Psychiatry (2013) 170 324-333. PMID: 23318474 10. Diniz B. S., Butters M. A., Albert S. M., Dew M. A., Reynolds C. F.. Late-life depression and risk of vascular dementia and Alzheimer's disease: Systematic review and meta-analysis of community-based cohort studies. The British Journal of Psychiatry: The Journal of Mental Science (2013) 202 329-335. PMID: 23637108 11. Fine J. P., Gray R. J.. A proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association (1999) 94 496-509 12. Hayes J. F., Marston L., Walters K., King M. B., Osborn D. P. J.. Mortality gap for people with bipolar disorder and schizophrenia: UK-based cohort study 2000–2014. The British Journal of Psychiatry (2017) 211 175-181. PMID: 28684403 13. Hennekens C. H., Hennekens A. R., Hollar D., Casey D. E., Raton B.. Schizophrenia and increased risks of cardiovascular disease. American Heart Journal (2005) 150 1115-1121. PMID: 16338246 14. Heser K., Tebarth F., Wiese B., Eisele M., Bickel H., Köhler M., Wagner M.. Age of major depression onset, depressive symptoms, and risk for subsequent dementia: Results of the German study on Ageing, Cognition, and Dementia in Primary Care Patients (AgeCoDe). Psychological Medicine (2013) 43 1597-1610. PMID: 23137390 15. Howard R., Cort E., Bradley R., Harper E., Kelly L., Bentham P., Gray R.. Antipsychotic treatment of very late-onset schizophrenia-like psychosis (ATLAS): A randomised, controlled, double-blind trial. The Lancet Psychiatry (2018) 5 553-563. PMID: 29880238 16. Howard R., Rabins P., Seeman M., Jeste D.. Late-onset schizophrenia and very-late-onset schizophrenia-like psychosis: An international consensus. American Journal of Psychiatry (2000) 157 172-178. PMID: 10671383 17. Howes O. D., McDonald C., Cannon M., Arseneault L., Boydell J., Murray R. M.. Pathways to schizophrenia: The impact of environmental factors. International Journal of Neuropsychopharmacology (2004) 7 S7-S13. PMID: 14972079 18. Ismail Z., Agüera-ortiz L., Brodaty H., Cieslak A., Cummings J., Fischer C. E., Lyketsos C.. The Mild Behavioral Impairment Checklist (MBI-C): A rating scale for neuropsychiatric symptoms in pre-dementia populations. Journal of Alzheimer's Disease: JAD (2017) 56 929-938. PMID: 28059789 19. Johansson L., Guo X., Waern M., Östling SÖ, Gustafson D., Bengtsson C., Skoog I.. Midlife psychological stress and risk of dementia: A 35-year longitudinal population study. Brain (2010) 133 2217-2224. PMID: 20488887 20. Karlsson I. K., Bennet A. M., Ploner A., Andersson T. M.-L., Reynolds C. A., Gatz M., Pedersen N. L.. Apolipoprotein E epsilon4 genotype and the temporal relationship between depression and dementia. Neurobiology of Aging (2015) 36 1751-1756. PMID: 25670333 21. Kay D. W. K., Roth M.. Environmental and hereditary factors in the schizophrenias of old age (“late paraphrenia”) and their bearing on the general problem of causation in schizophrenia. Journal of Mental Science (1961) 107 649-686. PMID: 13752013 22. Kępińska A. P., MacCabe J. H., Cadar D., Steptoe A., Murray R. M., Ajnakina O.. Schizophrenia polygenic risk predicts general cognitive deficit but not cognitive decline in healthy older adults. Translational Psychiatry (2020) 10 1-9. PMID: 32066695 23. Kessler C. R., Amminger G. P., Aguilar-Gaxiola S., Alonso J., Lee S., Ustun B.. Age of onset of mental disorders: A review of recent literature. Current Opinion in Psychiatry (2007) 20 359-364. PMID: 17551351 24. Kodesh A., Goldberg Y., Rotstein A., Weinstein G., Reichenberg A., Sandin S., Levine S. Z.. Risk of dementia and death in very-late-onset schizophrenia-like psychosis: A national cohort study. Schizophrenia Research (2020) 223 220-226. PMID: 32807646 25. Kørner A., Lopez L., Andersen P. K., Kessing L. V.. Delusional disorder in old age and the risk of developing dementia – a nationwide register-based study. Aging and Mental Health (2008) 12 625-629. PMID: 18855178 26. Kørner A., Lopez A. G., Lauritzen L., Andersen P. K., Kessing L.. Late and very-late first-contact schizophrenia and the risk of dementia − a nationwide register based study. International Journal of Geriatric Psychiatry (2009) 24 61-67. PMID: 18561206 27. Li G., Wang L. Y., Shofer J. B., Thompson M. L., Peskind E. R., McCormick W., Larson E. B.. Temporal relationship between depression and dementia: Findings from a large community-based 15-year follow-up study. Archives of General Psychiatry (2011) 68 970-977. PMID: 21893662 28. Lonnqvist J., Verkasalo M., Haukka J., Nyman K., Tiihonen J., Laaksonen I., Henriksson M.. Premorbid personality factors in schizophrenia and bipolar disorder: Results from a large cohort study of male conscripts. Journal of Abnormal Psychology (2009) 118 418-423. PMID: 19413416 29. Lyketsos C. G., Peters M. E.. Dementia in patients with schizophrenia. Evidence for heterogeneity. JAMA Psychiatry (2015) 72 1075-1076. PMID: 26444878 30. McCreadie R. G.. Use of drugs, alcohol and tobacco by people with schizophrenia: Case-control study. British Journal of Psychiatry (2002) 181 321-325 31. McCreadie R. G.. Diet, smoking and cardiovascular risk in people with schizophrenia. British Journal of Psychiatry (2003) 183 534-539 32. Mueller C., Thompson A., Harwood D., Bagshaw P., Burns A.. Mental health in older people A practice primer (2017) 33. Osborn D. P. J., Wright C. A., Levy G., King M. B., Deo R., Nazareth I.. Relative risk of diabetes, dyslipidaemia, hypertension and the metabolic syndrome in people with severe mental illnesses: Systematic review and metaanalysis. BMC Psychiatry (2008) 8 1-14. PMID: 18173833 34. Ownby R. L., Crocco E., Acevedo A., John V., Loewenstein D.. Depression and risk for Alzheimer disease: Systematic review, meta-analysis and metaregression analysis. Archives of General Psychiatry (2007) 63 530-538 35. Ribe A. R., Laursen T. M., Charles M., Katon W., Fenger-grøn M., Davydow D., Vestergaard M.. Long-term risk of dementia in persons with schizophrenia: A Danish population-based cohort study. JAMA Psychiatry (2015) 72 1095-1101. PMID: 26444987 36. Rizzuto D., Feldman A. L., Karlsson I. K., Dahl Aslan A. K., Gatz M., Pedersen N. L.. Detection of dementia cases in two Swedish health registers: A validation study. Journal of Alzheimer's Disease (2018) 61 1301-1310 37. Roberts L., Roalfe A., Wilson S., Lester H.. Physical health care of patients with schizophrenia in primary care: A comparative study. Family Practice Advance Access (2006) 24 34-40 38. Sharp E. S., Gatz M.. The relationship between education and dementia an updated systematic review. Alzheimer Disease & Associated Disorders (2011) 25 289-304. PMID: 21750453 39. Sindi S., Hagman G., Håkansson K., Kulmala J., Nilsen C., Kåreholt I., Kivipelto M.. Midlife work-related stress increases dementia risk in later life: The CAIDE 30-year study. Journals of Gerontology - Series B Psychological Sciences and Social Sciences (2017) 72 1044-1053. PMID: 27059705 40. Singh-Manoux A., Dugravot A., Fournier A., Abell J., Ebmeier K., Kivimäki M., Sabia S.. Trajectories of depressive symptoms before diagnosis of dementia: A 28-year follow-up study. JAMA Psychiatry (2017) 74 712-718. PMID: 28514478 41. Smith D. J., Langan J., Mclean G., Guthrie B., Mercer S. W.. Schizophrenia is associated with excess multiple physical-health comorbidities but low levels of recorded cardiovascular disease in primary care: Cross-sectional study. BMJ Open (2013) 3 e002808 42. Sperling R. A., Aisen P. S., Beckett L. A., Bennett D. A., Craft S., Fagan A. M., Phelps C. H.. Toward defining the preclinical stages of Alzheimer's disease: Recommendations from the national institute on aging-Alzheimer's association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimer's & Dementia (2011) 7 280-292 43. Stafford J., Howard R., Dalman C., Kirkbride J. B.. The incidence of nonaffective, nonorganic psychotic disorders in older people: A population-based cohort study of 3 million people in Sweden. Schizophrenia Bulletin (2019) 45 1152-1160. PMID: 30339239 44. Stafford J., Howard R., Kirkbride J. B.. The incidence of very late-onset psychotic disorders: A systematic review and meta-analysis, 1960–2016. Psychological Medicine (2018) 48 1775-1786. PMID: 29198197 45. Stern Y.. Cognitive reserve and Alzheimer disease. Alzheimer Disease & Associated Disorders (2006) 20 69-74 46. Talaslahti T., Alanen H. M., Hakko H., Isohanni M., Häkkinen U., Leinonen E.. Patients with very-late-onset schizophrenia-like psychosis have higher mortality rates than elderly patients with earlier onset schizophrenia. International Journal of Geriatric Psychiatry (2015) 30 453-459. PMID: 24990229 47. Terracciano A., Aschwanden D., Passamonti L., Toschi N., Stephan Y., Luchetti M., Súilleabh S. O.. Is neuroticism differentially associated with risk of Alzheimer's disease, vascular dementia, and frontotemporal dementia?. Journal of Psychiatric Research (2021) 138 34-40. PMID: 33819874 48. Vahia I., Palmer B. W., Depp C., Fellows I., Golshan S., Kraemer H. C., Jeste D. V.. Is late-onset schizophrenia a subtype of schizophrenia?. Acta Psychiatrica Scandinavica (2010) 122 414-426. PMID: 20199491 49. Van Assche L., Morrens M., Luyten P., Van de Ven L., Vandenbulcke M.. The neuropsychology and neurobiology of late-onset schizophrenia and very-late-onset schizophrenia-like psychosis: A critical review. Neuroscience and Biobehavioral Reviews (2017) 83 604-621. PMID: 28867652 50. Van Os J., Jones P. B.. Neuroticism as a risk factor for schizophrenia. Psychological Medicine (2001) 31 1129-1134. PMID: 11513380 51. Villemagne V. L., Burnham S., Bourgeat P., Brown B., Ellis K. A., Salvado O., Masters C. L.. Amyloid. The Lancet Neurology (2013) 12 357-367. PMID: 23477989 52. Viron M. J., Stern T. A.. The impact of serious mental illness on health and healthcare. Psychosomatics (2010) 51 458-465. PMID: 21051676 53. Wang P. S., Schneeweiss S., Avorn J., Fischer M. A., Mogun H., Solomon D. H., Brookhart M. A.. Risk of death in elderly users of conventional vs. atypical antipsychotic medications. New England Journal of Medicine (2005) 353 2335-2341. PMID: 16319382 54. World Health Organization (1992). The ICD-10 classification of mental and behavioural disorders: clinical descriptions and diagnostic guidelines. Geneva: World Health Organization.. The ICD-10 classification of mental and behavioural disorders: clinical descriptions and diagnostic guidelines (1992)
--- title: 'Longitudinal association between cardiovascular risk factors and depression in young people: a systematic review and meta-analysis of cohort studies' authors: - Anna B. Chaplin - Natasha F. Daniels - Diana Ples - Rebecca Z. Anderson - Amy Gregory-Jones - Peter B. Jones - Golam M. Khandaker journal: Psychological Medicine year: 2023 pmcid: PMC9975997 doi: 10.1017/S0033291721002488 license: CC BY 4.0 --- # Longitudinal association between cardiovascular risk factors and depression in young people: a systematic review and meta-analysis of cohort studies ## Abstract ### Background Depression is a common and serious mental illness that begins early in life. An association between cardiovascular disease (CVD) and subsequent depression is clear in adults. We examined associations between individual CVD risk factors and depression in young people. ### Methods We searched MEDLINE, EMBASE, and PsycINFO databases from inception to 1 January 2020. We extracted data from cohort studies assessing the longitudinal association between CVD risk factors [body mass index (BMI), smoking, systolic blood pressure (SBP), total cholesterol, high-density lipoprotein] and depression, measured using a validated tool in individuals with mean age of 24 years or younger. Random effect meta-analysis was used to combine effect estimates from individual studies, including odds ratio (OR) for depression and standardised mean difference for depressive symptoms. ### Results Based on meta-analysis of seven studies, comprising 15 753 participants, high BMI was associated with subsequent depression [pooled OR 1.61; $95\%$ confidence interval (CI) 1.21–2.14; I2 = $31\%$]. Based on meta-analysis of eight studies, comprising 30 539 participants, smoking was associated with subsequent depression (pooled OR 1.73; $95\%$ CI 1.36–2.20; I2 = $74\%$). Low, but not high, SBP was associated with an increased risk of depression (pooled OR 3.32; $95\%$ CI 1.68–6.55; I2 = $0\%$), although this was based on a small pooled high-risk sample of 893 participants. Generalisability may be limited as most studies were based in North America or Europe. ### Conclusions Targeting childhood/adolescent smoking and obesity may be important for the prevention of both CVD and depression across the lifespan. Further research on other CVD risk factors including blood pressure and cholesterol in young people is required. ## Introduction Depression is a common and serious mental illness with a lifetime risk of 10–$20\%$ (Hasin et al., 2018). The majority of depression cases are established by age 24 (Kessler et al., 2010) and this condition is the leading cause of disability among children and young people (Polanczyk, Salum, Sugaya, Caye, & Rohde, 2015). Following an initial depressive episode, the risk of recurrence is $60\%$ (Holtzheimer & Mayberg, 2011). Therefore, early-onset depression is associated with a longer period of risk for relapse as well as poor long-term outcomes (Holtzheimer & Mayberg, 2011; Kessler, 2012). A better understanding of the aetiology of depression in young people is required to develop effective strategies for prevention and treatment (Niarchou, Zammit, & Lewis, 2015). Cardiovascular disease (CVD) is a leading cause of health-related disability worldwide (Roth et al., 2018). There is evidence for bidirectional associations between CVD and depression in adults (Hiles et al., 2015; Inouye et al., 2018; Khandaker et al., 2019; Smolderen et al., 2017). A substantial body of literature suggests that depression is a key risk factor for CVD in adults and that it may predict poor outcomes following a cardiac event (Barefoot & Schroll, 1996; Hiles et al., 2015; Inouye et al., 2018; Khandaker et al., 2019; Lippi, Montagnana, Favaloro, & Franchini, 2009; Van der Kooy et al., 2007). CVD is also associated with subsequent depression in adults (Choi, Kim, Marti, & Chen, 2014; Hare, Toukhsati, Johansson, & Jaarsma, 2014; Kendler, Gardner, Fiske, & Gatz, 2009; Lippi et al., 2009). However, studies of CVD risk and subsequent depression in young people are relatively less common. A clearer understanding of the association between CVD risk factors and depression in young people is required. Early detection and management of CVD risk factors may reduce risks for both CVD and depression subsequently during the life-course. The World Health Organization defines young people as individuals aged 24 years or younger (WHO Study Group of Young People, 1986). Existing studies of CVD risk and depression in young people have often focused on individual risk factors such as body mass index (BMI) or smoking. In the past decade, a number of systematic reviews have highlighted an association between obesity in young people and depression across the lifespan (Hoare, Skouteris, Fuller-Tyszkiewicz, Millar, & Allender, 2014; Mannan, Mamun, Doi, & Clavarino, 2016; Mühlig, Antel, Föcker, & Hebebrand, 2016; Sutaria, Devakumar, Yasuda, Das, & Saxena, 2019). However, none of these studies specifically examined depression risk in young people. A recent systematic review of individuals aged 14–35 reported that childhood obesity is associated with approximately $50\%$ increased risk of depression (Sutaria et al., 2019). Another review reported that smoking in early life is associated with $73\%$ increased risk of depression in young people (Chaiton, Cohen, O'Loughlin, & Rehm, 2009). According to the Framingham study, other established CVD risk factors for adults include systolic blood pressure (SBP), total cholesterol, and high-density lipoprotein (HDL), in addition to smoking and BMI (Wilson et al., 1998; Wilson, Castelli, & Kannel, 1987). These CVD risk factors are all potentially modifiable and may be important in the aetiology and prevention of depression. CVD risk factors are increasingly being examined in young people; thus, a systematic review is required to summarise these findings. We conducted a systematic review and meta-analysis of existing studies to quantify the longitudinal association of five key CVD risk factors (BMI, smoking, SBP, total cholesterol, and HDL) and depression in young people. These CVD risk factors were chosen for a number of reasons: (i) they are part of the Framingham Cardiovascular Risk Score for adults; (ii) they are potentially modifiable; and (iii) they remain relevant in the context of young people. Our outcome was depression (binary or continuous) assessed using a validated tool. We also performed a number of sensitivity analyses, for example by excluding studies that only looked at one gender or excluding studies based on quality assessment. ## Search strategy and study selection This study has been performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Details of the protocol were prospectively registered on PROSPERO (see https://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42020172460). MEDLINE, EMBASE, and PsycINFO databases were searched to identify all relevant studies of the association between CVD risk factors and depression from database inception to 1 January 2020. The following keywords were used: ‘(cohort OR longitudinal OR prospective OR retrospective OR follow up stud) AND depress AND (adolescen* OR young person OR young people OR child* OR infant OR early adult OR youth* OR teen) AND ((cardiovascular AND risk) OR total cholesterol OR high density lipoprotein OR hdl OR smok OR bmi OR body mass index OR adiposity OR waist circumference OR body fat distribution OR skinfold thickness OR lipid accumulation product OR systolic blood pressure OR systolic bp OR sbp)’. See online Supplementary materials for the full search strategy. No language restriction was applied. The electronic search was complemented by hand-searching the reference lists of included studies. All titles and abstracts were examined to retrieve potentially relevant studies. ABC, NFD, AGJ, DP, and RZA applied inclusion/exclusion criteria and selected the final studies for this review. ## Selection criteria We included studies that: (i) had a longitudinal population-based cohort design (prospective or retrospective); (ii) included participants with a mean age of 24 years old or younger at follow-up; (iii) had at least one of the five CVD risk factors (BMI, smoking, SBP, total cholesterol, HDL) as the exposure at baseline; (iv) used a validated tool to measure depression (binary outcome or symptom score) at follow-up; and (v) reported effect estimate(s) for the association between CVD risk and subsequent depression. Studies were excluded if they did not have an unexposed group for a particular risk factor (e.g. experimental smoking used as the comparison group rather than no smoking), had depression as the exposure and the CVD risk factor as the outcome, or measured depression comorbid with another mental illness such as bipolar disorder or anxiety. ## Data extraction Data extraction was performed independently by ABC, NFD and DP, and disagreements were resolved by consensus. For each included study, we extracted the following data: (i) details of the cohort (country, name/setting, design, sample size, and follow-up length); (ii) assessment of exposure and outcome; (iii) age and sex of the included participants; and (iv) results of analysis (number of participants exposed at baseline, number of participants with depression at follow-up, adjusted/unadjusted effect estimates). When studies reported various methods for assessing the exposure or repeated measures of the exposure, we used the most comprehensive measure. For example, one study measured BMI eight times from birth to age 12 years (Wang, Leung, & Schooling, 2014). We chose the age 7 measure as the earliest age where BMI may be an appropriate measure of central adiposity to maximise the length of follow-up. In cases where there was more than one published report from the same population, we included the study with the larger sample size (Bares, 2014; Duncan & Rees, 2005; Goodman & Capitman, 2000; Munafò, Hitsman, Rende, Metcalfe, & Niaura, 2008). Some studies reported results where depression at baseline was adjusted for as well as analysis where baseline depression cases were removed. In such cases, we only included results where baseline depression was excluded to minimise reverse causality. ## Data synthesis and meta-analysis We performed separate meta-analyses for BMI, smoking, and SBP. Results from studies were pooled using the inverse variance method meaning that studies with larger sample sizes were given greater weight (Higgins et al., 2021; Schwarzer, 2007). Results for other CVD risk factors were summarised using a narrative review. Meta-analyses were performed separately for studies that reported beta estimates (continuous depressive symptoms outcome) or odds ratio (OR) (binary depression outcome). We used random-effect meta-analysis, which is appropriate where there is heterogeneity between the studies. Heterogeneity between studies was examined using the I2 statistic. We assessed publication bias by visual inspection of funnel plots, and by Egger's regression test for funnel plot asymmetry (mixed-effects meta-regression model). We considered a p value of <0.05 to indicate the existence of publication bias. We assessed study quality using the Newcastle-Ottawa Scale for cohort studies (Stang, 2010). We repeated analyses with only good/fair quality studies. We also repeated analyses after excluding studies with only female or male participants. Meta-analyses were carried out using the meta package version 4.11 in R version 3.6.1 (Schwarzer, 2007). ## Results Electronic search identified 6616 studies. After removing duplicates, 4821 studies remained. After title and abstract screening, 197 ($4.1\%$) potentially eligible studies were identified, of which 29 met the inclusion criteria and were included in the review (Albers & Biener, 2002; Bares, 2014; Beal, Negriff, Dorn, Pabst, & Schulenberg, 2014; Boutelle, Hannan, Fulkerson, Crow, & Stice, 2010; Chaiton, Cohen, Rehm, Abdulle, & O'Loughlin, 2015; Chang et al., 2017; Choi, Patten, Christian Gillin, Kaplan, & Pierce, 1997; Clark et al., 2007; Duncan & Rees, 2005; Eitle & Eitle, 2018; Frisco, Houle, & Lippert, 2013; Gage et al., 2015; Gomes et al., 2019; Goodman & Whitaker, 2002; Hammerton, Harold, Thapar, & Thapar, 2013; Hammerton, Thapar, & Thapar, 2014; Marmorstein, Iacono, & Legrand, 2014; Monshouwer et al., 2012; Perry et al., 2020; Piumatti, 2018; Pryor et al., 2016; Raffetti, Donato, Forsell, & Galanti, 2019; Ranjit et al., 2019a, b; Rhew et al., 2008; Roberts & Duong, 2013; Rubio, Kraemer, Farrell, & Day, 2008; Wang et al., 2014; Zhang, Woud, Becker, & Margraf, 2018). See Fig. 1 for study selection and Table 1 for characteristics of included studies. Fig. 1.PRISMA flow diagram for study selection. Table 1.Characteristics of studies included in systematic reviewStudyCountryFemale, No. (%) Baseline age range/mean (years)Minimum follow-up (years)Assessment of exposureExposed at baseline (%) or mean (s.d.)Assessment of depressionDepression at follow-up (%)Mean (s.d.) depressive symptom score at follow-upBMIBoutelle et al. [ 2010]USA495 [100]a13.51Obese v. healthy weight$10.7\%$Structured interview (K-SADS)1.2–3.51.3–1.4 (0.4)Chang et al. [ 2017]Taiwan969 (51.2)14.73Obese v. healthy weight$3.5\%$ (F); $3.4\%$ (M)Self-report (CES-DC)–1.8 (0.5) (F); 1.5 (0.4) (M)Clark et al. [ 2007]UK776 (51.3)11–142Obese v. healthy weight$19.6\%$Self-report (SMFQ)15.1–Eitle and Eitle [2018]USA308 (47.0) (AI); 3403 (50.0) (W)12–181Obese v. not obese$19.0\%$ (AI); $11.0\%$ (W)Self-report (CES-D)–12.4 (0.6) (AI); 10.1 (0.2) (W)Frisco et al. [ 2013]USA5243 [100]15.74Consistently obese v. healthy weight$9.8\%$Self-report (CES-D)7.0–Gomes et al. [ 2019]BrazilNM184Obese v. not obese$6.7\%$Structured interview (MINI-5)2.9–Goodman and Whitaker [2002]USA4718 (48.6)<201Obese v. not obese$9.7\%$Self-report (CES-D)8.9–Hammerton et al. [ 2014]bUKEPAD: 165 (57.1);ALSPAC: 2533 (52.1)EPAD: 9–17; ALSPAC: 11–14EPAD: 1.4; ALSPAC: 2Standardised scoreEPAD: $17.4\%$; ALSPAC: $23.7\%$EPAD: Semi-structured interview (CAPA); ALSPAC: Structured interview (DAWBA)EPAD: 8.3; ALSPAC: 1.6EPAD: 2.2 (NM) (F); 1.6 (NM) (M);ALSPAC: NMMarmorstein et al. [ 2014]USANM11–146Obese v. not obese$14.7\%$ (F); $10.7\%$ (M)Structured interview (SCID)12.3 (F); 7.6 (M)–Monshouwer et al. [ 2012]Netherlands820 (53.3)10–124Obese v. healthy weight$14.1\%$Structured interview (WHO CIDI-3)5.6–Perry et al. [ 2020]bUK1655 (51.6)99Raw score17.5 (2.5)Self-report (CIS-R)7.03.1 (NM)Piumatti [2018]bItaly178 (78.1)21.41Raw score21.1 (2.7)Self-report (PHQ-2)–1.1 (0.6)Pryor et al. [ 2016]Canada661 (54.1)6–121Early-onset v. never overweight$11.0\%$Self-report (Kovacs CDI)–NMRhew et al. [ 2008]USA206 (46.2)121Overweight v. healthy weight$22.4\%$ (F); $30.3\%$ (M)Structured interview (MFQ)–NMRoberts and Duong [2013]USA2040 (48.9) a11–171Obese v. healthy weight$19.7\%$Structured interview (DISC IV-Y)1.7–Wang et al. [ 2014]bHong Kong2793 (48.2)77Standardised score−0.03 (1.1) (F); 0.3 (1.3) (M)Self-report (PHQ-9)–4.0 (NM)Zhang et al. [ 2018]Germany1196 [100]211.4Overweight v. healthy weight$7.2\%$Structured interview (Diagnostic Interview for Mental Disorders Research Version)6.5–SmokingAlbers and Biener [2002]USA259 (49.6) a12–154Ever v. never smoker$33.9\%$Self-report (Kandel & Davies)13.2–Bares [2014]USA2486 (53.0) a16.65.5Unit increase in cigarette use$31.3\%$Self-report (CES-D)–4.9 (3.9)Beal et al. [ 2014]USA262 [100]11–202Ever v. never smoker2.9 (3.1)Self-report (CDI/BDI)–46.3 (10.8)Chaiton et al. [ 2015]bCanada416 (49.7)12–135Smoking initiation v. no initiation$47.4\%$Self-report (Mellinger scale)23.8–Choi et al. [ 1997]USA3215 (46.8)12–184Established v. never smoker$17.3\%$Self-report (Kandel & Davies)11.5–Clark et al. [ 2007]UK776 (51.3)11–142Tried/regular v. never smoker$35.2\%$Self-report (SMFQ)15.1–Duncan and Rees [2005]USA6748 (51.6)11–211Smoker v. non-smoker$25.8\%$ (F); $25.3\%$ (M)Self-report (CES-D)27.0 (F); 21.4 (M)13.0 (8.7) (F);10.8 (7.5) (M)Gage et al. [ 2015]UKNM162Unit increase in cigarette use$9.8\%$Self-report (CIS-R)7.2–Raffetti et al. [ 2019]Sweden1636 (51.2)13–141Current v. non-smoker$2.0\%$Self-report (CES-DC)8.315.7 (NM)Ranjit et al. ( 2019a)bFinland2358 (50.3) a143Regular v. never smoker$9.1\%$Self-report (GBI)5.1 (4.9)–Ranjit et al. ( 2019b)bFinland2174 (51.8) a17.55Ever v. never smoker$25.3\%$Self-report (GBI)4.5 (4.7)–Rubio et al. [ 2008]USA278 [100]230.25Smoker v. non-smoker$72.0\%$Self-report (CES-D)85.0–Piumatti [2018]Italy178 (78.1)21.41Daily v. non-smokers$13.6\%$Self-report (PHQ-2)–1.1 (0.6)Zhang et al. [ 2018]Germany1196 [100]211.4Current v. non-smoker$23.6\%$Structured interview (Diagnostic Interview for Mental Disorders Research Version)6.5–Systolic blood pressureHammerton et al. [ 2013]UKEPAD: 164 (58.4); ALSPAC: 2516 (52.1)EPAD: 9–17; ALSPAC: 11–14EPAD: 2.5; ALSPAC: 3Standardised scoreEPAD: 117.3 (13.2); ALSPAC: 111.0 (9.5)EPAD: Semi-structured interview (CAPA); ALSPAC: Structured interview (DAWBA)EPAD: 8.5; ALSPAC: 1.6–s.d., standard deviation; F, female; M, male; AI, American Indian; W, White; EPAD, Early Prediction of Adolescent Depression study; ALSPAC, Avon Longitudinal Study of Parents and Children; NS, not significant; NM, not mentioned; K-SADS, Kiddie Schedule for Affective Disorders and Schizophrenia; SMFQ, Short Mood and Feelings Questionnaire; MINI-5, Mini International Neuropsychiatric Interview Version Five; DSM, Diagnostic and Statistical Manual of Mental Disorders; CES-D, Centre for Epidemiological Studies Depression; CAPA, Child and Adolescent Psychiatric Assessment (Child Version); DAWBA, Development and Wellbeing Assessment (Child Version); SCID, Structured Clinical Interview for DSM-III-R; CIS-R, Clinical Interview Schedule Revised; ICD-10, International Statistical Classification of Diseases and Related Health Problems Tenth Revision; DISC IV-Y, Diagnostic Interview Schedule for Children for direct administration to children or adolescents; WHO CIDI-3, World Health Organisation Composite International Diagnostic Interview Version Three; CDI, Children's Depression Inventory; PHQ, Patient Health Questionnaire; MFQ, Mood and Feelings Questionnaire; GBI, General Behaviour Inventory; BDI, Beck's Depression Inventory.aBaseline sample only.bNot included in meta-analysis. The four studies not included in BMI meta-analysis were excluded because they measured BMI as a continuous variable. The three studies not included in smoking meta-analysis were excluded because they did not report effect estimates comparable with the other studies. All studies were prospective in design, except one which used a retrospective measure of depression at follow-up (Monshouwer et al., 2012). The majority of studies ($55.2\%$) were rated as ‘good’ quality using the Newcastle-Ottawa Scale (online Supplementary Tables S1 and S2). Sex, age, parental education, race/ethnicity, baseline depression, and alcohol use were the most commonly used confounders (online Supplementary Table S3). Based on data availability, meta-analysis for BMI included 13 studies (Boutelle et al., 2010; Chang et al., 2017; Clark et al., 2007; Eitle & Eitle, 2018; Frisco et al., 2013; Gomes et al., 2019; Goodman & Whitaker, 2002; Marmorstein et al., 2014; Monshouwer et al., 2012; Pryor et al., 2016; Rhew et al., 2008; Roberts & Duong, 2013; Zhang et al., 2018), and that for smoking included 11 studies (Albers & Biener, 2002; Bares, 2014; Beal et al., 2014; Chaiton et al., 2015; Choi et al., 1997; Clark et al., 2007; Duncan & Rees, 2005; Gage et al., 2015; Piumatti, 2018; Raffetti et al., 2019; Ranjit et al., 2019a, b; Rubio et al., 2008; Zhang et al., 2018) (online Supplementary Table S2). Meta-analysis for SBP included one study comprising two separate samples (Hammerton et al., 2013). We found no studies of total cholesterol or HDL that met our inclusion criteria. ## Longitudinal association between high BMI at baseline and risk of depression at follow-up Based on seven studies reporting an adjusted OR, comprising a total of 15 753 participants, the pooled OR for depression at follow-up associated with high BMI (>25) at baseline was 1.61 ($95\%$ confidence interval (CI) 1.21–2.14) (Fig. 2). There was limited evidence of heterogeneity between studies (I2 = $31\%$; $95\%$ CI 0–$71\%$; Cochran's $Q = 8.7$; $$p \leq 0.19$$). Separate meta-analysis of unadjusted effect estimates showed lower pooled results (online Supplementary Fig. S1). Fig. 2.Meta-analysis of longitudinal association between high BMI at baseline and subsequent depression in young people. ## Longitudinal association between smoking at baseline and risk of depression at follow-up Based on eight studies reporting an adjusted OR, comprising a total of 30 539 participants, the pooled OR for depression at follow-up associated with smoking at baseline was 1.73 ($95\%$ CI 1.36–2.20) (Fig. 3). There was evidence of heterogeneity between studies (I2 = $74\%$; $95\%$ CI 52–$86\%$; Cochran's $Q = 35.3$; $p \leq 0.01$). Separate meta-analysis of unadjusted effect estimates showed higher pooled results (online Supplementary Fig. S2). Fig. 3.Meta-analysis of longitudinal association between smoking at baseline and subsequent depression in young people. ## Longitudinal association between SBP at baseline and risk of depression at follow-up One study examined associations of both low and high SBP with depression in two separate samples comprising a total of 5111 participants. Meta-analysis of these studies suggest depression at follow-up is associated with low SBP at baseline (OR 3.32; $95\%$ CI 1.68–6.55), but not with high SBP at baseline (OR 0.82; $95\%$ CI 0.55–1.22) (Fig. 4). There was some evidence of heterogeneity for high SBP (I2 = $66\%$; $95\%$ CI 0–$92\%$; Cochran's $Q = 3.0$; $$p \leq 0.08$$) and little heterogeneity for low SBP (I2 = $0\%$; $95\%$ CI 0–$0\%$; Cochran's $Q = 0.04$; $$p \leq 0.84$$). Fig. 4.Meta-analysis of longitudinal association between SBP at baseline and subsequent depression in young people. SBP, systolic blood pressure; EPAD, Early Prediction of Adolescent Depression study; ALSPAC, Avon Longitudinal Study of Parents and Children. * Subset of ALSPAC participants with mothers with recurrent depression. †EPAD participants have mother/father with recurrent depression. ## Longitudinal association between high BMI, smoking at baseline and depressive symptoms at follow-up Based on five studies reporting an adjusted beta estimate, comprising a total of 11 516 participants, the standardised mean difference (SMD) for an increase in depressive symptoms at follow-up associated with high BMI at baseline was 0.05 ($95\%$ CI –0.08–0.18) (online Supplementary Fig. S3). There was evidence of heterogeneity between studies (I2 = $71\%$; $95\%$ CI 34–$88\%$; Cochran's $Q = 17.5$; $p \leq 0.01$). Based on five studies reporting an adjusted beta estimate, comprising a total of 21 490 participants, the SMD for an increase in depressive symptoms at follow-up associated with smoking at baseline was 0.37 ($95\%$ CI 0.10–0.64) (online Supplementary Fig. S3). There was evidence of heterogeneity between studies (I2 = $89\%$; $95\%$ CI 78–$94\%$; Cochran's $Q = 44.8$; $p \leq 0.01$). ## Results for sensitivity analysis After excluding three studies with only female participants (Boutelle et al., 2010; Frisco et al., 2013; Zhang et al., 2018), the adjusted pooled OR for depression at follow-up for high BMI at baseline was 1.43 ($95\%$ CI 0.94–2.18) (Fig. 2). There was some heterogeneity between studies (I2 = $47\%$; $95\%$ CI 0–$83\%$; Cochran's $Q = 5.7$; $$p \leq 0.13$$). After excluding four studies with only female or male participants (Choi et al., 1997; Duncan & Rees, 2005; Rubio et al., 2008; Zhang et al., 2018), the pooled adjusted OR for depression at follow-up for smoking at baseline was 1.23 ($95\%$ CI 1.02–1.49) (Fig. 3). There was little heterogeneity (I2 = $11\%$; $95\%$ CI 0–$86\%$; Cochran's $Q = 3.4$; $$p \leq 0.34$$). After excluding one study based on quality assessment (Duncan & Rees, 2005), the pooled adjusted OR for depression at follow-up for smoking at baseline was 1.48 ($95\%$ CI 1.21–1.80) (online Supplementary Fig. S4). There was some heterogeneity (I2 = $38\%$; $95\%$ CI 0–$73\%$; Cochran's $Q = 11.3$; $$p \leq 0.12$$). For BMI, we did not exclude any studies based on quality assessment. See online Supplementary Figs S5 and S6 for sensitivity analyses where the outcome of interest was depressive symptoms at follow-up. ## Publication bias Based on Egger's test and funnel plots, evidence for publication bias was not evident for studies reporting the adjusted association between high BMI and depression (Egger's test: $$p \leq 0.17$$; online Supplementary Fig. S7), smoking and depression (Egger's test: $$p \leq 0.80$$; online Supplementary Fig. S8), or high BMI and depressive symptoms (Egger's test: $$p \leq 0.35$$) (online Supplementary Fig. S9). Evidence for publication bias was present for studies reporting the adjusted association between smoking and depressive symptoms (Egger's test: $$p \leq 0.01$$) (online Supplementary Fig. S10). ## Discussion Depression and CVD are associated with each other in mid to late adulthood. Although depression is known to arise commonly in young people, the timing of the association with CVD risk is unclear and common risk factors for the two conditions raise the prospect of joint prevention. To the best of our knowledge, this is the first systematic review to consider the association between various CVD risk factors and subsequent depression in young people. We report four key findings: (i) BMI and smoking are the most well-studied risk factors for depression in this age group; (ii) both BMI and smoking at baseline are longitudinally associated with subsequent depression; (iii) smoking but not BMI is prospectively associated with depressive symptoms; and (iv) currently there is limited data on longitudinal associations of high SBP and cholesterol with subsequent depression in young people, which should be examined in future. Our results suggest that obesity could be an important risk factor for depression in young people. The pooled OR of 1.61 for the association of high BMI and depression is remarkably similar to previous studies in adults. Previous meta-analyses have reported ORs of 1.51 and 1.70 for the prospective association between childhood high BMI and adult depression (Luppino et al., 2010; Sutaria et al., 2019). Another meta-analysis reported that obese adolescents had an $40\%$ increased risk of experiencing depression as adults (Mannan et al., 2016). A recent Mendelian randomisation study using data from 812 000 adult participants also found that fat mass could be a causal factor for depression (Speed, Jefsen, Børglum, Speed, & Østergaard, 2019). We did not find an association between high BMI and depressive symptoms score, indicating a possibly non-linear association between BMI and depression whereby association is restricted to only those with more severe symptoms. Our findings also suggest that smoking could be a risk factor for depression in young people. The pooled OR of 1.73 for the association between smoking and depression is consistent with a previous meta-analysis in adults, which reported an OR of 1.62 (Luger, Suls, & Vander Weg, 2014). Similarly, a meta-analysis of nine cross-sectional and longitudinal studies found that adolescents exposed to second-hand smoking had increased odds of depression (Han, Liu, Gong, Ye, & Zhou, 2019). However, current evidence from observational cohort studies and genetic Mendelian randomisation studies reports mixed findings regarding the association between smoking and depression, with some reporting an association (Wootton et al., 2020) and others no association (Bjørngaard et al., 2013; Taylor et al., 2014). Therefore, residual confounding or reverse causality remain viable explanations for the observed association between smoking and depression. Further longitudinal studies and genetic Mendelian randomisation studies are required to investigate this issue. A number of potential mechanisms may be involved in the association of high BMI and subsequent depression, including low-grade systemic inflammation, hypothalamic–pituitary–adrenal axis (HPA) axis dysregulation, insulin resistance, and psychological distress. Inflammation is evident in around $25\%$ of individuals with depression (Osimo, Baxter, Lewis, Jones, & Khandaker, 2019) and atypical depression is associated with inflammation and metabolic dysregulation (Lamers et al., 2013, 2020). Adipose tissue also contains abundant inflammatory cytokines that are involved in fat metabolism (Heredia, Gómez-Martínez, & Marcos, 2012). Similarly, the role of insulin in regulating adipocyte function contributes to the close link between insulin resistance and obesity (Kahn & Flier, 2000). Furthermore, melancholic depression, higher levels of abdominal fat, HPA axis hyperactivity, and cortisol dysregulation are inter-related (Incollingo Rodriguez et al., 2015; Lamers et al., 2013, 2020). Adverse childhood experiences are also one of the most robust risk factors for depression (Gardner, Thomas, & Erskine, 2019), and are associated with increased risk for obesity and metabolic dysregulation (Farr et al., 2015). Low-grade systemic inflammation may be important to the association between smoking and depression (Berk et al., 2013). Studies are required to investigate the complex mechanisms that may underlie the association between high BMI and depression. Low SBP, but not high SBP, appeared to be associated with risk for depression in young people. However, our meta-analysis was based on only two cohorts at high risk for depression and further study is required. Low SBP appears to be associated with depression in young people at high-risk of depression but not in the general population (Hammerton et al., 2013). Low SBP has also been associated with depression in cross-sectional and longitudinal studies of middle-aged and elderly adults (Hildrum et al., 2007; Huang, Su, Jiang, & Zhu, 2020). Conversely, higher SBP has been prospectively associated with fewer depressive symptoms in older adults with CVD risk factors (Herrmann-Lingen et al., 2018). In adult populations, the relationship between SBP and depression may be independent of a range of lifestyle factors, age, and sex (Hildrum et al., 2007; Huang et al., 2020). The reason why low SBP has a potentially causal role in depression remains unclear. Neurons controlling blood pressure could be implicated in the association between SBP and depression. Neuropeptide Y, for example, reduces blood pressure and is involved in stress responses that have been linked to increased risk for depression, such as the HPA axis (Hildrum et al., 2007; Juruena, Bocharova, Agustini, & Young, 2018). Further research is required to understand the relationship between SBP and depression in young people, including potential mechanisms. We found no studies assessing the association between either total cholesterol or HDL on risk for subsequent depression in young people. A meta-analysis of 30 cross-sectional studies reported that higher total cholesterol was associated with lower levels of depression in adults (Shin, Suls, & Martin, 2008). Evidence from adults indicates both higher and lower HDL to be associated with increased risk for depression in adults. In a meta-analysis of 16 cross-sectional studies, high HDL was related to higher levels of depression, especially in women (Shin et al., 2008). Conversely, a meta-analysis of 11 case-control studies reported that lower HDL levels may be associated with first-episode major depressive disorder in adults (Wei et al., 2020). Given that abnormal HDL, LDL and triglyceride levels are increasingly common in adolescents [Centers for Disease Control and Prevention (CDC), 2010], effort should be made to study potential effects on mental health as well as physical health. Strengths of this study include the systematic literature search which identified a large number of relevant studies comprising a total of 93 021 participants. We included studies considering the effect of various CVD risk factors on either binary or continuous measures of depression/depressive symptoms. We assessed the studies using the validated Newcastle-Ottawa Scale as well as conducting sensitivity analyses to examine the robustness of our findings. However, this study is not without limitations. First, the majority of studies came from North America and Europe, limiting the generalisability of the results to other parts of the world. The number of studies in each of the meta-analyses was also relatively small, which resulted in wide CIs for the pooled effect estimates, and reduced the statistical power to detect publication bias. There was a considerable amount of heterogeneity between studies, particularly studies of depressive symptoms. Sensitivity analyses revealed that sex explained heterogeneity in some of the meta-analyses. However, stratifying by sex decreased the sample size, and consequently, statistical power to detect an association. In future, studies with larger samples are required. Finally, the possibility of residual confounding by unidentified factors remains high so we are cautious in terms of any conclusion regarding causality. Although studies included in this meta-analysis controlled for various potential confounding effects, other factors may also explain these associations. Further research is needed to examine whether observed associations are likely to be causal. Since randomised controlled trials are neither feasible nor ethical for some of the exposures under investigation (e.g. smoking, obesity), genetic approaches to dealing with residual confounding, such as Mendelian randomisation, would be particularly useful. In summary, we present evidence for a longitudinal association between CVD risk factors, namely high BMI and smoking, in childhood/adolescence and subsequent depression in young people. These risk factors could be important targets for the prevention of depression and CVD in young people and subsequently during the life course. Further study is needed to understand potential mechanisms for these associations as well as the relationship between other CVD risk factors, notably blood pressure and cholesterol and depression risk in young people. ## Financial support ABC is supported by the National Institute for Health Research (NIHR) CLAHRC RCF (MRR73-1795-00000) (https://www.nihr.ac.uk/), NIHR ARC East of England, the MQ: Transforming Mental Health (Data Science Award; grant code: MQDS$\frac{17}{40}$) (https://www.mqmentalhealth.org/home/). GMK acknowledges funding support from the Wellcome Trust (grant code: 201486/Z/16/Z) (https://wellcome.org/), MQ as above, the Medical Research Council (grant code: MC_PC_17213 and MR/S$\frac{037675}{1}$) (https://mrc.ukri.org/), and the BMA Foundation (J Moulton grant 2019) (http://www.bmafoundationmr.org.uk/). PBJ acknowledges funding from MQ, NIHR ARC East of England and the Medical Research Council, as above, and from NIHR PGfAR 0616-20003. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Registration number: CRD42020172460. ## Conflict of interest None. ## References 1. Albers A. B., Biener L.. The role of smoking and rebelliousness in the development of depressive symptoms among a cohort of Massachusetts adolescents. Preventive Medicine (2002) 34 625-631. DOI: 10.1006/pmed.2002.1029 2. Barefoot J. C., Schroll M.. Symptoms of depression, acute myocardial infarction, and total mortality in a community sample. Circulation (1996) 93 1976-1980. DOI: 10.1161/01.CIR.93.11.1976 3. Bares C. B.. Gender, depressive symptoms and daily cigarette use. Journal of Dual Diagnosis (2014) 10 187-196. DOI: 10.1080/15504263.2014.961852 4. Beal S. J., Negriff S., Dorn L. D., Pabst S., Schulenberg J.. Longitudinal associations between smoking and depressive symptoms among adolescent girls. Prevention Science (2014) 15 506-515. DOI: 10.1007/s11121-013-0402-x 5. Berk M., Williams L. J., Jacka F. N., O'Neil A., Pasco J. A., Moylan S., Maes M.. So depression is an inflammatory disease, but where does the inflammation come from?. BMC Medicine (2013) 11 200. DOI: 10.1186/1741-7015-11-200 6. Bjørngaard J. H., Gunnell D., Elvestad M. B., Davey Smith G., Skorpen F., Krokan H., Romundstad P.. The causal role of smoking in anxiety and depression: A Mendelian randomization analysis of the HUNT study. Psychological Medicine (2013) 43 711-719. DOI: 10.1017/S0033291712001274 7. Boutelle K. N., Hannan P., Fulkerson J. A., Crow S. J., Stice E.. Obesity as a prospective predictor of depression in adolescent females. Health Psychology (2010) 29 293-298. DOI: 10.1037/a0018645 8. Prevalence of abnormal lipid levels among youths – United States, 1999–2006. MMWR. Morbidity and Mortality Weekly Report (2010) 59 29-33. PMID: 20094024 9. Chaiton M. O., Cohen J. E., O'Loughlin J., Rehm J.. A systematic review of longitudinal studies on the association between depression and smoking in adolescents. BMC Public Health (2009) 9 356. DOI: 10.1186/1471-2458-9-356 10. Chaiton M., Cohen J. E., Rehm J., Abdulle M., O'Loughlin J.. Confounders or intermediate variables? Testing mechanisms for the relationship between depression and smoking in a longitudinal cohort study. Addictive Behaviors (2015) 42 154-161. DOI: 10.1016/j.addbeh.2014.11.026 11. Chang L.-Y., Chang H.-Y., Wu W.-C., Lin L. N., Wu C.-C., Yen L.-L.. Body mass index and depressive symptoms in adolescents in Taiwan: Testing mediation effects of peer victimization and sleep problems. International Journal of Obesity (2017) 41 1510-1517. DOI: 10.1038/ijo.2017.111 12. Choi N. G., Kim J., Marti C. N., Chen G. J.. Late-life depression and cardiovascular disease burden: Examination of reciprocal relationship. The American Journal of Geriatric Psychiatry (2014) 22 1522-1529. DOI: 10.1016/j.jagp.2014.04.004 13. Choi W. S., Patten C. A., Christian Gillin J., Kaplan R. M., Pierce J. P.. Cigarette smoking predicts development of depressive symptoms among U.S. adolescents. Annals of Behavioral Medicine (1997) 19 42-50. DOI: 10.1007/BF02883426 14. Clark C., Haines M. M., Head J., Klineberg E., Arephin M., Viner R., Stansfeld S. A.. Psychological symptoms and physical health and health behaviours in adolescents: A prospective 2-year study in East London. Addiction (2007) 102 126-135. DOI: 10.1111/j.1360-0443.2006.01621.x 15. Duncan B., Rees D. I.. Effect of smoking on depressive symptomatology: A reexamination of data from the national longitudinal study of adolescent health. American Journal of Epidemiology (2005) 162 461-470. DOI: 10.1093/aje/kwi219 16. Eitle D., Eitle T. M.. Obesity, overweightness, and depressive symptomology among American Indian youth. Journal of Racial and Ethnic Health Disparities (2018) 5 1305-1314. DOI: 10.1007/s40615-018-0479-9 17. Farr O. M., Ko B.-J., Joung K. E., Zaichenko L., Usher N., Tsoukas M., Mantzoros C. S.. Posttraumatic stress disorder, alone or additively with early life adversity, is associated with obesity and cardiometabolic risk. Nutrition, Metabolism and Cardiovascular Diseases (2015) 25 479-488. DOI: 10.1016/j.numecd.2015.01.007 18. Frisco M. L., Houle J. N., Lippert A. M.. Weight change and depression among US young women during the transition to adulthood. American Journal of Epidemiology (2013) 178 22-30. DOI: 10.1093/aje/kws462 19. Gage S. H., Hickman M., Heron J., Munafò M. R., Lewis G., Macleod J., Zammit S.. Associations of Cannabis and cigarette use with depression and anxiety at age 18: Findings from the Avon longitudinal study of parents and children. PLoS One (2015) 10 e0122896. DOI: 10.1371/journal.pone.0122896 20. Gardner M. J., Thomas H. J., Erskine H. E.. The association between five forms of child maltreatment and depressive and anxiety disorders: A systematic review and meta-analysis. Child Abuse & Neglect (2019) 96 104082. DOI: 10.1016/j.chiabu.2019.104082 21. Gomes A. P., Soares A. L. G., Menezes A. M. B., Assunção M. C., Wehrmeister F. C., Howe L. D., Gonçalves H.. Adiposity, depression and anxiety: Interrelationship and possible mediators. Revista de Saúde Pública (2019) 53 103. DOI: 10.11606/S1518-8787.2019053001119 22. Goodman E., Capitman J.. Depressive symptoms and cigarette smoking among teens. Pediatrics (2000) 106 748-755. DOI: 10.1542/peds.106.4.748 23. Goodman E., Whitaker R. C.. A prospective study of the role of depression in the development and persistence of adolescent obesity. Pediatrics (2002) 110 497-504. PMID: 12205250 24. Hammerton G., Harold G., Thapar A., Thapar A.. Depression and blood pressure in high-risk children and adolescents: An investigation using two longitudinal cohorts. BMJ Open (2013) 3 e003206. DOI: 10.1136/bmjopen-2013-003206 25. Hammerton G., Thapar A., Thapar A. K.. Association between obesity and depressive disorder in adolescents at high risk for depression. International Journal of Obesity (2014) 38 513-519. DOI: 10.1038/ijo.2013.133 26. Han C., Liu Y., Gong X., Ye X., Zhou J.. Relationship between second-hand smoke exposure and depressive symptoms: A systematic review and dose–response meta-analysis. International Journal of Environmental Research and Public Health (2019) 16 1356. DOI: 10.3390/ijerph16081356 27. Hare D. L., Toukhsati S. R., Johansson P., Jaarsma T.. Depression and cardiovascular disease: A clinical review. European Heart Journal (2014) 35 1365-1372. DOI: 10.1093/eurheartj/eht462 28. Hasin D. S., Sarvet A. L., Meyers J. L., Saha T. D., Ruan W. J., Stohl M., Grant B. F.. Epidemiology of adult DSM-5 major depressive disorder and its specifiers in the United States. JAMA Psychiatry (2018) 75 336. DOI: 10.1001/jamapsychiatry.2017.4602 29. Heredia F. P. D., Gómez-Martínez S., Marcos A.. Obesity, inflammation and the immune system. Proceedings of the Nutrition Society (2012) 71 332-338. DOI: 10.1017/S0029665112000092 30. Herrmann-Lingen C., Meyer T., Bosbach A., Chavanon M.-L., Hassoun L., Edelmann F., Wachter R.. Cross-sectional and longitudinal associations of systolic blood pressure with quality of life and depressive mood in older adults with cardiovascular risk factors: Results from the observational DIAST-CHF study. Psychosomatic Medicine (2018) 80 468-474. DOI: 10.1097/PSY.0000000000000591 31. Higgins J., Thomas J., Chandler J., Cumpston M., Li T., Page M., Welch V.. (2021) 32. Hildrum B., Mykletun A., Stordal E., Bjelland I., Dahl A. A., Holmen J.. Association of low blood pressure with anxiety and depression: The Nord-Trøndelag Health Study. Journal of Epidemiology and Community Health (2007) 61 53-58. DOI: 10.1136/jech.2005.044966 33. Hiles S. A., Baker A. L., de Malmanche T., McEvoy M., Boyle M., Attia J.. The role of inflammatory markers in explaining the association between depression and cardiovascular hospitalisations. Journal of Behavioral Medicine (2015) 38 609-619. DOI: 10.1007/s10865-015-9637-2 34. Hoare E., Skouteris H., Fuller-Tyszkiewicz M., Millar L., Allender S.. Associations between obesogenic risk factors and depression among adolescents: A systematic review. Obesity Reviews (2014) 15 40-51. DOI: 10.1111/obr.12069 35. Holtzheimer P. E., Mayberg H. S.. Stuck in a rut: Rethinking depression and its treatment. Trends in Neurosciences (2011) 34 1-9. DOI: 10.1016/j.tins.2010.10.004 36. Huang Y., Su Y., Jiang Y., Zhu M.. Sex differences in the associations between blood pressure and anxiety and depression scores in a middle-aged and elderly population: The Irish longitudinal study on ageing (TILDA). Journal of Affective Disorders (2020) 274 118-125. DOI: 10.1016/j.jad.2020.05.133 37. Incollingo Rodriguez A. C., Epel E. S., White M. L., Standen E. C., Seckl J. R., Tomiyama A. J.. Hypothalamic-pituitary-adrenal axis dysregulation and cortisol activity in obesity: A systematic review. Psychoneuroendocrinology (2015) 62 301-318. DOI: 10.1016/j.psyneuen.2015.08.014 38. Inouye M., Abraham G., Nelson C. P., Wood A. M., Sweeting M. J., Dudbridge F., Samani N. J.. Genomic risk prediction of coronary artery disease in 480000 adults. Journal of the American College of Cardiology (2018) 72 1883-1893. DOI: 10.1016/j.jacc.2018.07.079 39. Juruena M. F., Bocharova M., Agustini B., Young A. H.. Atypical depression and non-atypical depression: Is HPA axis function a biomarker? A systematic review. Journal of Affective Disorders (2018) 233 45-67. DOI: 10.1016/j.jad.2017.09.052 40. Kahn B. B., Flier J. S.. Obesity and insulin resistance. The Journal of Clinical Investigation (2000) 106 473-481. DOI: 10.1172/JCI10842 41. Kendler K. S., Gardner C. O., Fiske A., Gatz M.. Major depression and coronary artery disease in the Swedish Twin Registry. Archives of General Psychiatry (2009) 66 857-863. DOI: 10.1001/archgenpsychiatry.2009.94 42. Kessler R. C.. The costs of depression. The Psychiatric Clinics of North America (2012) 35 1-14. DOI: 10.1016/j.psc.2011.11.005 43. Kessler R. C., McLaughlin K. A., Green J. G., Gruber M. J., Sampson N. A., Zaslavsky A. M., Williams D. R.. Childhood adversities and adult psychopathology in the WHO World Mental Health Surveys. The British Journal of Psychiatry (2010) 197 378-385. DOI: 10.1192/bjp.bp.110.080499 44. Khandaker G. M., Zuber V., Rees J. M. B., Carvalho L., Mason A. M., Foley C. N., Burgess S.. Shared mechanisms between coronary heart disease and depression: Findings from a large UK general population-based cohort. Molecular Psychiatry (2019) 25 1477-1486. DOI: 10.1038/s41380-019-0395-3 45. Lamers F., Milaneschi Y., Vinkers C. H., Schoevers R. A., Giltay E. J., Penninx B. W. J. H.. Depression profilers and immuno-metabolic dysregulation: Longitudinal results from the NESDA study. Brain, Behavior, and Immunity (2020) 88 174-183. DOI: 10.1016/j.bbi.2020.04.002 46. Lamers F., Vogelzangs N., Merikangas K. R., de Jonge P., Beekman A. T. F., Penninx B. W. J. H.. Evidence for a differential role of HPA-axis function, inflammation and metabolic syndrome in melancholic versus atypical depression. Molecular Psychiatry (2013) 18 692-699. DOI: 10.1038/mp.2012.144 47. Lippi G., Montagnana M., Favaloro E. J., Franchini M.. Mental depression and cardiovascular disease: A multifaceted, bidirectional association. Seminars in Thrombosis and Hemostasis (2009) 35 325-336. DOI: 10.1055/s-0029-1222611 48. Luger T. M., Suls J., Vander Weg M. W.. How robust is the association between smoking and depression in adults? A meta-analysis using linear mixed-effects models. Addictive Behaviors (2014) 39 1418-1429. DOI: 10.1016/j.addbeh.2014.05.011 49. Luppino F. S., Wit L. M. D., Bouvy P. F., Stijnen T., Cuijpers P., Penninx B. W. J. H., Zitman F. G.. Overweight, obesity, and depression: A systematic review and meta-analysis of longitudinal studies. Archives of General Psychiatry (2010) 67 220-229. DOI: 10.1001/archgenpsychiatry.2010.2 50. Mannan M., Mamun A., Doi S., Clavarino A.. Prospective associations between depression and obesity for adolescent males and females: A systematic review and meta-analysis of longitudinal studies. PLoS One (2016) 11 e0157240. DOI: 10.1371/journal.pone.0157240 51. Marmorstein N. R., Iacono W. G., Legrand L.. Obesity and depression in adolescence and beyond: Reciprocal risks. International Journal of Obesity (2014) 38 906-911. DOI: 10.1038/ijo.2014.19 52. Monshouwer K., Smit F., Ruiter M., Ormel H., Verhulst F., Vollebergh W., Oldehinkel T.. Identifying target groups for the prevention of depression in early adolescence: The TRAILS study. Journal of Affective Disorders (2012) 138 287-294. DOI: 10.1016/j.jad.2012.01.026 53. Mühlig Y., Antel J., Föcker M., Hebebrand J.. Are bidirectional associations of obesity and depression already apparent in childhood and adolescence as based on high-quality studies? A systematic review. Obesity Reviews (2016) 17 235-249. DOI: 10.1111/obr.12357 54. Munafò M. R., Hitsman B., Rende R., Metcalfe C., Niaura R.. Effects of progression to cigarette smoking on depressed mood in adolescents: Evidence from the national longitudinal study of adolescent health. Addiction (2008) 103 162-171. DOI: 10.1111/j.1360-0443.2007.02052.x 55. Niarchou M., Zammit S., Lewis G.. The avon longitudinal study of parents and children (ALSPAC) birth cohort as a resource for studying psychopathology in childhood and adolescence: A summary of findings for depression and psychosis. Social Psychiatry and Psychiatric Epidemiology (2015) 50 1017-1027. DOI: 10.1007/s00127-015-1072-8 56. Osimo E. F., Baxter L. J., Lewis G., Jones P. B., Khandaker G. M.. Prevalence of low-grade inflammation in depression: A systematic review and meta-analysis of CRP levels. Psychological Medicine (2019) 49 1958-1970. DOI: 10.1017/S0033291719001454 57. Perry B. I., Khandaker G. M., Marwaha S., Thompson A., Zammit S., Singh S. P., Upthegrove R.. Insulin resistance and obesity, and their association with depression in relatively young people: Findings from a large UK birth cohort. Psychological Medicine (2020) 50 556-565. DOI: 10.1017/S0033291719000308 58. Piumatti G.. Motivation, health-related lifestyles and depression among university students: A longitudinal analysis. Psychiatry Research (2018) 260 412-417. DOI: 10.1016/j.psychres.2017.12.009 59. Polanczyk G. V., Salum G. A., Sugaya L. S., Caye A., Rohde L. A.. Annual research review: A meta-analysis of the worldwide prevalence of mental disorders in children and adolescents. Journal of Child Psychology and Psychiatry, and Allied Disciplines (2015) 56 345-365. DOI: 10.1111/jcpp.12381 60. Pryor L., Brendgen M., Boivin M., Dubois L., Japel C., Falissard B., Côté S. M.. Overweight during childhood and internalizing symptoms in early adolescence: The mediating role of peer victimization and the desire to be thinner. Journal of Affective Disorders (2016) 202 203-209. DOI: 10.1016/j.jad.2016.05.022 61. Raffetti E., Donato F., Forsell Y., Galanti M. R.. Longitudinal association between tobacco use and the onset of depressive symptoms among Swedish adolescents: The Kupol cohort study. European Child & Adolescent Psychiatry (2019) 28 695-704. DOI: 10.1007/s00787-018-1237-6 62. Ranjit A., Buchwald J., Latvala A., Heikkilä K., Tuulio-Henriksson A., Rose R. J., Korhonen T.. Predictive association of smoking with depressive symptoms: A longitudinal study of adolescent twins. Prevention Science (2019a) 20 1021-1030. DOI: 10.1007/s11121-019-01020-6 63. Ranjit A., Korhonen T., Buchwald J., Heikkilä K., Tuulio-Henriksson A., Rose R. J., Latvala A.. Testing the reciprocal association between smoking and depressive symptoms from adolescence to adulthood: A longitudinal twin study. Drug and Alcohol Dependence (2019b) 200 64-70. DOI: 10.1016/j.drugalcdep.2019.03.012 64. Rhew I. C., Richardson L. P., Lymp J., McTiernan A., McCauley E., Stoep A. V.. Measurement matters in the association between early adolescent depressive symptoms and body mass index. General Hospital Psychiatry (2008) 30 458-466. DOI: 10.1016/j.genhosppsych.2008.06.008 65. Roberts R. E., Duong H. T.. Obese youths are not more likely to become depressed, but depressed youths are more likely to become obese. Psychological Medicine (2013) 43 2143-2151. DOI: 10.1017/S0033291712002991 66. Roth G. A., Abate D., Abate K. H., Abay S. M., Abbafati C., Abbasi N., Murray C. J. L.. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: A systematic analysis for the Global Burden of Disease Study 2017. The Lancet (2018) 392 1736-1788. DOI: 10.1016/S0140-6736(18)32203-7 67. Rubio D. M., Kraemer K. L., Farrell M. H., Day N. L.. Factors associated with alcohol use, depression, and their co-occurrence during pregnancy. Alcoholism: Clinical and Experimental Research (2008) 32 1543-1551. DOI: 10.1111/j.1530-0277.2008.00705.x 68. Schwarzer G.. Meta: An R package for meta-analysis. R News (2007) 7 40-45 69. Shin J. Y., Suls J., Martin R.. Are cholesterol and depression inversely related? A meta-analysis of the association between two cardiac risk factors. Annals of Behavioral Medicine (2008) 36 33-43. DOI: 10.1007/s12160-008-9045-8 70. Smolderen K. G., Spertus J. A., Gosch K., Dreyer R. P., D'Onofrio G., Lichtman J. H., Krumholz H. M.. Depression treatment and health Status outcomes in young patients with acute myocardial infarction: Insights from the VIRGO study (variation in recovery: Role of gender on outcomes of young AMI patients). Circulation (2017) 135 1762-1764. DOI: 10.1161/CIRCULATIONAHA.116.027042 71. Speed M. S., Jefsen O. H., Børglum A. D., Speed D., Østergaard S. D.. Investigating the association between body fat and depression via mendelian randomization. Translational Psychiatry (2019) 9 1-9. DOI: 10.1038/s41398-019-0516-4 72. Stang A.. Critical evaluation of the Newcastle-Ottawa scale for the assessment of the quality of nonrandomized studies in meta-analyses. European Journal of Epidemiology (2010) 25 603-605. PMID: 20652370 73. Sutaria S., Devakumar D., Yasuda S. S., Das S., Saxena S.. Is obesity associated with depression in children? Systematic review and meta-analysis. Archives of Disease in Childhood (2019) 104 64-74. DOI: 10.1136/archdischild-2017-314608 74. Taylor A. E., Fluharty M. E., Bjørngaard J. H., Gabrielsen M. E., Skorpen F., Marioni R. E., Munafò M. R.. Investigating the possible causal association of smoking with depression and anxiety using Mendelian randomisation meta-analysis: The CARTA consortium. BMJ Open (2014) 4 e006141. DOI: 10.1136/bmjopen-2014-006141 75. Van der Kooy K., van Hout H., Marwijk H., Marten H., Stehouwer C., Beekman A.. Depression and the risk for cardiovascular diseases: Systematic review and meta analysis. International Journal of Geriatric Psychiatry (2007) 22 613-626. DOI: 10.1002/gps.1723 76. Wang H., Leung G. M., Schooling C. M.. Life course adiposity and adolescent depressive symptoms among Hong Kong adolescents. Journal of Adolescent Health (2014) 55 408-414. DOI: 10.1016/j.jadohealth.2014.03.009 77. Wei Y.-G., Cai D.-B., Liu J., Liu R.-X., Wang S.-B., Tang Y.-Q., Wang F.. Cholesterol and triglyceride levels in first-episode patients with major depressive disorder: A meta-analysis of case-control studies. Journal of Affective Disorders (2020) 266 465-472. DOI: 10.1016/j.jad.2020.01.114 78. WHO Study Group of Young People (1986). Young people's health - a challenge for society: Report of a WHO study group on young people and ‘health for all by the year 2000’. Geneva: World Health Organization. Retrieved from https://apps.who.int/iris/handle/10665/41720.. Young people's health - a challenge for society: Report of a WHO study group on young people and ‘health for all by the year 2000’ (1986) 79. Wilson P. W. F., Castelli W. P., Kannel W. B.. Coronary risk prediction in adults (The Framingham Heart Study). The American Journal of Cardiology (1987) 59 G91-G94. DOI: 10.1016/0002-9149(87)90165-2 80. Wilson P. W., D'Agostino R. B., Levy D., Belanger A. M., Silbershatz H., Kannel W. B.. Prediction of coronary heart disease using risk factor categories. Circulation (1998) 97 1837-1847. DOI: 10.1161/01.cir.97.18.1837 81. Wootton R. E., Richmond R. C., Stuijfzand B. G., Lawn R. B., Sallis H. M., Taylor G. M. J., Munafò M. R.. Evidence for causal effects of lifetime smoking on risk for depression and schizophrenia: A Mendelian randomisation study. Psychological Medicine (2020) 50 2435-2443. DOI: 10.1017/S0033291719002678 82. Zhang X. C., Woud M. L., Becker E. S., Margraf J.. Do health-related factors predict major depression? A longitudinal epidemiologic study. Clinical Psychology & Psychotherapy (2018) 25 378-387. DOI: 10.1002/cpp.2171
--- title: 'Repetitive transcranial magnetic stimulation treatment of major depressive disorder and comorbid chronic pain: response rates and neurophysiologic biomarkers' authors: - Juliana Corlier - Reza Tadayonnejad - Andrew C Wilson - Jonathan C Lee - Katharine G Marder - Nathaniel D Ginder - Scott A Wilke - Jennifer Levitt - David Krantz - Andrew F Leuchter journal: Psychological Medicine year: 2023 pmcid: PMC9976020 doi: 10.1017/S0033291721002178 license: CC BY 4.0 --- # Repetitive transcranial magnetic stimulation treatment of major depressive disorder and comorbid chronic pain: response rates and neurophysiologic biomarkers ## Abstract ### Background Major depressive disorder (MDD) and chronic pain are highly comorbid, and pain symptoms are associated with a poorer response to antidepressant medication treatment. It is unclear whether comorbid pain also is associated with a poorer response to treatment with repetitive transcranial magnetic stimulation (rTMS). ### Methods 162 MDD subjects received 30 sessions of 10 Hz rTMS treatment administered to the left dorsolateral prefrontal cortex (DLPFC) with depression and pain symptoms measured before and after treatment. For a subset of 96 patients, a resting-state electroencephalogram (EEG) was recorded at baseline. Clinical outcome was compared between subjects with and without comorbid pain, and the relationships among outcome, pain severity, individual peak alpha frequency (PAF), and PAF phase-coherence in the EEG were examined. ### Results $64.8\%$ of all subjects reported pain, and both depressive and pain symptoms were significantly reduced after rTMS treatment, irrespective of age or gender. Patients with severe pain were $27\%$ less likely to respond to MDD treatment than pain-free individuals. PAF was positively associated with pain severity. PAF phase-coherence in the somatosensory and default mode networks was significantly lower for MDD subjects with pain who failed to respond to MDD treatment. ### Conclusions Pain symptoms improved after rTMS to left DLPFC in MDD irrespective of age or gender, although the presence of chronic pain symptoms reduced the likelihood of treatment response. Individual PAF and baseline phase-coherence in the sensorimotor and midline regions may represent predictors of rTMS treatment outcome in comorbid pain and MDD. ## Introduction Major depressive disorder (MDD) and chronic pain are highly comorbid, especially in female patients (Bair et al., 2004). This comorbidity significantly decreases the quality of life of patients and represents one of the largest socioeconomic burdens worldwide (Leuchter et al., 2010; Walker, Kavelaars, Heijnen, & Dantzer, 2014). Pain symptoms can either precede or follow the onset of MDD (Bair, Robinson, Katon, & Kroenke, 2003; Chang et al., 2015; Gerrits, Van Oppen, Van Marwijk, Penninx, & Van Der Horst, 2014), suggesting a bidirectional relationship between MDD and chronic pain. Several studies have suggested that affective disorders and chronic pain have an overlapping pathophysiology and may share similar circuit mechanisms (Bair et al., 2003; Taylor, Becker, Schweinhardt, & Cahill, 2016). While both depression and pain symptoms can be alleviated by antidepressant medications (Gracely, Ceko, & Bushnell, 2012; Maletic & Raison, 2009), this comorbidity has generally been associated with greater resistance to pharmacological treatment (Bair et al., 2004, 2003; Gerrits et al., 2012; Leuchter et al., 2010; Von Korff & Simon, 1996). In particular, pain severity has been reported to be a strong predictor of poorer antidepressant medication treatment outcome and health-related quality of life (Bair et al., 2004). Better recognition, assessment, and treatment of comorbid pain may thus enhance the outcome of antidepressant therapy. Repetitive transcranial magnetic stimulation (rTMS) administered to the left dorsolateral prefrontal cortex (DLPFC) is an effective treatment for pharmaco-resistant depression (George et al., 2010; Janicak et al., 2013). While the mechanism of action (MOA) of this neuromodulation technique is not yet fully understood, there is evidence to suggest that the therapeutic effect of rTMS arises through the resetting of resting-state functional networks beyond the stimulation site (Corlier et al., 2019a, b; Fox, Halko, Eldaief, & Pascual-Leone, 2012; Leuchter, Hunter, Krantz, & Cook, 2015; To, De Ridder, Hart, & Vanneste, 2018). rTMS also appears to be efficacious for the treatment of other neuropsychiatric disorders including post-traumatic stress disorder, obsessive-compulsive disorder (Tadayonnejad et al., 2020), generalized anxiety disorder, bipolar depression, tinnitus, neurodegenerative disorders (Carpenter et al., 2018; Heath, Taylor, & McNerney, 2018; Lefaucheur et al., 2014, 2020; Soleimani, Jalali, & Hasandokht, 2016), and pain syndromes such as neuropathic pain, headache, fibromyalgia, and complex regional pain syndrome (Altas, Askin, Beşiroğlu, & Tosun, 2019; Galhardoni et al., 2015; Goudra et al., 2017; Hou, Wang, & Kang, 2016; Hsu, Daskalakis, & Blumberger, 2018; Knijnik et al., 2016; Saltychev & Laimi, 2017; Short et al., 2011). Most studies of rTMS on pain have targeted the primary motor cortex, but stimulation to left DLPFC has also successfully reduced pain symptoms even in non-MDD populations (Galhardoni et al., 2015; Johnson, Summers, & Pridmore, 2006; Lefaucheur et al., 2014; Short et al., 2011). One previous report demonstrated significant improvement in both mood and pain symptoms solely with 10 Hz rTMS treatment applied to left DLPFC (Phillips, Burr, & Dunner, 2018). However, it remains unclear whether this effect was gender-specific and whether the presence and severity of pain symptoms was associated with inferior rTMS clinical outcome, as is the case with pharmacological treatment. Identifying neurophysiological biomarkers of chronic pain would aid in the development of an rTMS protocol targeting this comorbidity. Such measures would allow the assessment of target engagement and could serve as early predictors of treatment outcome. Alpha band oscillations have previously been identified as a robust electroencephalographic (EEG) marker of chronic pain and might represent a possible biomarker for rTMS. For example, higher alpha power has been observed within the dynamic pain connectome in subjects with neuropathic pain, rheumatoid arthritis, and jaw pain (Kim & Davis, 2020; Kisler et al., 2020; Meneses et al., 2016; Wang et al., 2019), while decreased alpha oscillations have been reported in tonic pain (cf. Ploner, Sorg, & Gross, 2017). Lower peak alpha frequency [PAF, also called individual alpha frequency (IAF)] (Grandy et al., 2013; Petrosino, Zandvakili, Carpenter, & Philip, 2018) also has been observed in neuropathic pain and fibromyalgia (Kim et al., 2019). Sensorimotor PAF has been reported as a reliable biomarker of subjective pain intensity and pain sensitivity, with slower PAF possibly reflecting pre-disease pain sensitivity (Babiloni et al., 2006; Furman et al., 2018, 2019, 2020) or marking a ‘chronification’ process of pain (de Vries et al., 2013). Targeted modulation of alpha activity through visual stimulation, as well as transcranial alternating or direct current stimulation, also has been associated with chronic pain relief (Ahn, Prim, Alexander, McCulloch, & Fröhlich, 2019; Arendsen, Hugh-Jones, & Lloyd, 2018; Lopez-Diaz et al., 2021). Recent evidence indicates that alpha synchrony between relevant networks enables feedforward/feedback processing and flexible routing of information in the integration of sensory and contextual processes (Kim & Davis, 2020; Kisler et al., 2020; Ploner et al., 2017). The examination of PAF/IAF and synchrony in the nociceptive network may therefore serve as an effective biomarker of target engagement and clinical response for rTMS treatment of chronic pain and depression. Given the prevalence of comorbid chronic pain and MDD, and in light of their overlapping pathophysiology, it is of crucial importance to elucidate novel targets and integrated interventions to treat comorbid pain and depression, rather than targeting pain and depressive symptoms separately (Walker et al., 2014). The primary objectives of this study were thus to: [1] evaluate the effect of pain comorbidity on rTMS treatment response for MDD; [2] examine gender differences in rTMS treatment outcome; and [3] examine IAF and alpha band network synchrony as potential biomarkers for rTMS treatment outcome for comorbid MDD and chronic pain. ## Subjects Subjects were 162 outpatients with a primary diagnosis of MDD (Mini International Diagnostic Interview, MINI) (Sheehan et al., 1998) referred for treatment in the TMS Clinical and Research Service at UCLA. The research protocol was approved by the UCLA IRB and all subjects provided informed consent prior to research procedures. Subjects presented with at least moderately severe depressive symptoms based upon a 17-item Hamilton Depression Rating Scale Score (HAM-D17, minimal score >17) (Hamilton, 1960) and had failed to enter remission after at least three adequate antidepressant trials. Subjects were allowed to continue receiving psychotropic medication concurrent with rTMS and underwent standard safety screening and medical clearance before receiving rTMS treatment. The entire sample of 162 MDD subjects who completed a 30-session rTMS course for depression was categorized into subgroups as follows: Group [1] $\frac{105}{162}$ who reported pretreatment comorbid pain; Group [2] $\frac{57}{162}$ with no comorbid pain; Group (1A) $\frac{46}{105}$ comorbid subjects with a spontaneous EEG recorded prior to the first session; Group (1B) $\frac{59}{105}$ without EEG; Group (2A) $\frac{51}{57}$ pain-free subjects with a baseline EEG recording; Group (2B) $\frac{6}{57}$ pain-free subjects without EEG (see also Fig. 1). All procedures in this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975 as revised in 2008. Fig. 1.Breakdown of total sample into subgroups. The total sample of $$n = 162$$ was split into Groups 1 and 2 based on the presence of pain symptoms (cutoff: VAS>200). Each subgroup was further divided depending on whether subjects had a baseline EEG recording (Groups 1A and 2A) or not (Groups 1B and 2B). ## Clinical assessment Depressive symptoms were assessed at baseline and session 30 with the 30-item Inventory of Depressive Symptomatology Self Report version (IDS-SR) (Trivedi et al., 2004). The reliability of IDS-SR is demonstrated by its use as the primary outcome in STARD, the largest open-label, pragmatic trial for MDD (Sinyor, Schaffer, & Levitt, 2010). Chronic pain was assessed at baseline and treatment 30 using the eight-question Visual Analog Scale (VAS) for pain with total scores ranging from 0 to 800 (Chiarotto et al., 2019). Each question ranges [0–100] and assesses, during the last week, overall pain, along with separate items for headache, back, shoulder, stomach, and abdominal pain, interference with daily function, and nighttime awakenings due to pain. The VAS has been widely used in adult patients with various chronic pain conditions (Mori et al., 2010; Onesti et al., 2013) and has been established as a reliable measure in a review of over 850 studies (Karcioglu, Topacoglu, Dikme, & Dikme, 2018). Patients with different pain types were included in this sample, including fibromyalgia-like symptoms, neuropathic pain, chronic lower back pain, and complex regional pain syndrome, and the specific source of chronic pain for each individual was not formally diagnosed in the context of this study. We therefore were not able to perform separate analyses for each pain type. ## EEG acquisition Ninety-seven of 162 subjects had an EEG recording performed at baseline using the ANT Neuro TMS-compatible EEG system and EEGO v1.8 recording software (Advanced Neuro Technology (ANT); Enschede, the Netherlands). Electrodes were applied using the 64-electrode ‘WaveGuard’ cap with sintered Ag/AgCl electrodes positioned according to the Extended 10–20 System with EOG electrodes above and below the left eye. Data were recorded at a sampling rate of 2000 Hz using full-band EEG DC amplifiers without filters during data acquisition using a CPz electrode reference; data were converted to a common average reference offline for analysis. Impedances were kept below 10 kΩ. All subjects had at least 5 min of baseline resting-state EEG recording. ## rTMS procedures All TMS treatments were performed with either the Magstim Rapid 2 stimulator using a 70 mm coil (Magstim, Whitland, South Wales, UK), the Neuronetics Neurostar treatment system (Neuronetics, Malvern, PA, USA), or MagVenture (Alpharetta, GA, USA). Motor threshold (MT) determination was performed prior to the first treatment, with MT defined as the minimum stimulus intensity necessary to elicit an overt motor response in the right abductor pollicis brevis or first dorsal interosseus muscles for ⩾$50\%$ of applied stimuli. Following MT determination, treatments were performed with patients seated in a semi-reclined position using standard safety procedures and ear protection. All patients underwent treatment initially with 10 Hz stimulation to left DLPFC defined using the Beam F3 method (Beam, Borckardt, Reeves, & George, 2009), with a 40-pulse train and 26 s intertrain interval with a total of 3000 pulses per session (37.5 min duration). Clinicians adjusted stimulation intensity, coil angle, and number of pulses administered as needed to optimize tolerability. Treatment intensity was titrated to $120\%$ MT as tolerated with parameters modified according to a measurement-based care paradigm. Participants completed IDS-SR ratings every five treatments. If there was an absence of early clinical response to rTMS or worsening in anxiety or depressive symptoms, the treatment protocol could be complemented by intermittent theta-burst stimulation priming (iTBS) at left DLPFC (Lee et al., 2020) or by 1 Hz rTMS at the right DLPFC after the fifth session to optimize tolerability or augment clinical response (sequential bilateral treatment). In the total group of 162 subjects, 137 ($84.5\%$) received modified treatment parameters and the use of these parameters was included as a categorical covariate in the analysis of variance. ## Data analysis We performed a t test comparing pre- to post-rTMS IDS-SR scores of the entire sample to assess the therapeutic efficacy of rTMS treatment. To characterize pain comorbidity in the current sample, we calculated proportions of MDD patients with no, moderate, or severe pain using the VAS scale for pain based on the observed trimodal distribution of the pain scores (Fig. 2a). Following the trimodal distribution of these data, we grouped subjects into [1] no pain or no significant chronic pain (termed ‘no pain’ or ‘pain-free’ subjects), [2] moderate pain, or [3] severe pain. The cutoffs were as follows: no pain: VAS = [0–200]; moderate pain: VAS = [201–450]; and severe pain: VAS = [451–800]. The proportions of subjects in each category were calculated. Fig. 2.Pain prevalence and change with rTMS treatment. ( a) The reported pain severity followed a trimodal distribution, corresponding to three empirically defined groups with ‘No Pain’, ‘Moderate pain’, or ‘Severe pain’. ( b) Most MDD patients ($64.8\%$) presented with comorbid pain. ( c) Pain symptoms significantly decreased with rTMS treatment for depression after 15 and 30 sessions. ( d) There were no gender differences in reported pain severity or improvement. We then calculated Pearson's correlation coefficients to evaluate the link [1] between baseline depression and pain severity, [2] between final outcome in depressive and pain symptoms, as well as [3] between baseline pain and final depression outcomes. Because pain and somatic symptom ratings were included in the overall IDS score, we also recalculated the IDS after removing pain- and sleep-related items [1 to 4 (sleep-related) and 20, 25, 26, 30 (pain- and energy-related)], to test whether relationships between depression and pain were affected by the removal of pain and somatic items [this reduced measure referred to as IDS-SR[22]]. An analysis of variance (ANOVA) was conducted to examine the effect of rTMS treatment (factor Treatment #) on pain including the five following covariates: age, gender, TMS device (1–3), use of modified rTMS protocol (yes/no), concomitant use of psychotropic medication during rTMS treatment (yes/no). Additionally, we compared response and remission rates for depression among groups of varying pain levels with response defined as a ⩾$50\%$ decrease in IDS-SR score from baseline and remission was defined as an IDS-SR score of ⩽13. Clinical rTMS response and remission rates across the three groups were compared using a Kruskal-Wallis test. ## EEG analyses EEGs (conducted on Groups 1A and 2A) were preprocessed using the semi-automated FASTER toolbox (Nolan, Whelan, & Reilly, 2010), an ICA-based algorithm to remove eye movement, muscular, and line noise artifacts, followed by visual inspection to manually remove remaining artifacts. Individual PAF was estimated using two complementary approaches: [1] a ‘center of gravity’ (COG) method previously used in pain research by Furman and colleagues (Seminowicz et al., 2018), and [2] the IAF method used in MDD research by Corlier et al. ( 2019a, b). In both cases, the PAF was estimated based on the averaged power spectral density of postcentral channels Cz, C3, and C4 using a common average reference. We evaluated the relationship between IAF and pain severity by computing the Pearson's correlation coefficient between COG/IAF and the VAS baseline scores only for the patients with MDD and pain comorbidity (i.e. with a minimal VAS score >200). We then performed phase-coherence analysis using channels-of-interest (ChOI) corresponding to the surface areas above motor, somatosensory, anterior and posterior cingulate cortices, as well as the precuneus. The 14 ChOI included: Fz, FCz, Cz, CPz, Pz, POz, FC1, FC2, FC3, FC4, C1, C2, C3, and C4 (ChOI are marked by blue spheres in Fig. 3c/d) with 91 total possible pairwise connections. Phase-coherence was calculated using the weighted phase lag index (wPLI), which has the advantage of yielding a reliable estimate even in noisy conditions. This metric weighs the observed phase leads and lags by the magnitude of the imaginary component of the cross-spectrum, is less sensitive to volume conduction, and provides greater statistical power to detect changes in phase-synchronization (Vinck, Oostenveld, Van Wingerden, Battaglia, & Pennartz, 2011; Xing et al., 2016a, 2016b). WPLI was estimated based on 4 s segments of artifact-free EEG data using the fieldtip toolbox function ‘ft_connectivityanalysis’. Baseline levels of wPLI were compared between comorbid responders/remitters and non-responders/non-remitters using t tests. Benjamini-Hochberg False Discovery Rate (FDR) method was used for the correction of multiple comparisons. We chose the top 25 connections with the largest Cohen's D effect sizes to compare the average network connectivity between responders/remitters and non-responders/non-remitters for the comorbid and the pain-free groups. Fig. 3.Association between depressive and pain symptoms and clinical outcome. ( a) Left: Baseline pain severity was significantly associated with baseline depression severity. Center: Final pain outcome was significantly associated with final depression outcome; and Right: Baseline pain severity was significantly associated with final depression outcome. ( b) Clinical response rates for different pain groups. Probability of response to rTMS was smallest for the severe pain group. ( c) Clinical remission rates for different pain groups. Probability of remission with rTMS was smallest for the severe pain group. ## Results In total, $35.2\%$ of MDD subjects presented with no pain, $44.4\%$ of subjects reported moderate pain, and $20.4\%$ had severe pain symptoms, exhibiting a trimodal distribution of severity (Fig. 2a and b). Baseline and post-treatment depression scores were significantly higher in the comorbid than in the pain-free MDD individuals. There was no effect of age or gender on baseline pain severity (Table 1). On average, rTMS treatment significantly reduced depressive (two-tailed t test, $T = 14.0$, $p \leq 0.0001$) and pain symptoms (ANOVA, $F = 3.8$, $$p \leq 0.02$$), and posthoc t tests showed pain improvement by treatment 15 with further improvement seen at treatment 30 (T15, p15 = 0.0088, T30, p30 = 0.0067, Fig. 1c). There was no significant effect of gender ($F = 2.4$, $$p \leq 0.09$$, Fig. 1d), age ($F = 0.1$, $$p \leq 0.7$$), TMS device ($F = 0.1$, $$p \leq 0.9$$), use of psychotropic medication ($F = 0.7$, $$p \leq 0.4$$), or use of modified rTMS protocol ($F = 2.7$, $$p \leq 0.1$$) on outcome (Table 1). Table 1.Demographic and clinical information for all subjects and those with and without painAll subjects $$n = 162$$MDD + no pain (VAS ≤ 200) $$n = 57$$MDD + pain (VAS > 200) $$n = 105$$Test statisticp valueAge (years)49.4 ± 16.545.3 ± 14.544.6 ± 15.9T = −0.17p = 0.87Gender (% female)$$n = 90$$ $55.6\%$$$n = 29$$ $50.9\%$$$n = 61$$ $58.1\%$χ2 = 0.78p = 0.38Psychotropic medicationN = 141 $87\%$$$n = 50$$ $87.7\%$$$n = 91$$ $86.7\%$$F = 0.7$$p \leq 0.4$Baseline IDS-SR43.4 ± 11.738.5 ± 11.746.0 ± 10.8T = −4.1p < 0.001Tx 15 IDS-SR33.7 ± 13.529.1 ± 12.836.3 ± 13.2Final IDS-SR29.8 ± 14.924.7 ± 14.032.6 ± 14.7T = −3.2p = 0.0016Baseline VAS276.7 ± 175.579.5 ± 58.8383.8 ± 113.8T = 18.9p < 0.001Tx 15 VAS246.1 ± 160.8119.3 ± 89.3313.1 ± 149.6Final VAS238.1 ± 180.299.7 ± 131.3312.9 ± 157.8T = 7.3p < 0.001There was no age or gender difference between groups. Both initial depression severity and final depression score were significantly higher in the pain group. There was a significant positive correlation between the severity of pain and depression at baseline ($r = 0.43$, $$p \leq 1.56$$ × 10−08), between the final pain and depression outcome ($r = 0.53$, $$p \leq 9.59$$ × 10−11), and between baseline pain severity and final depression outcome ($r = 0.36$, $$p \leq 3.73$$ × 10−06). Response rates were significantly different across the three pain severity groups (χ2 = 8.5, $$p \leq 0.014$$), with $27\%$ lower rTMS response probability among patients with more severe pain compared to the pain-free group (rTMS response rates of 39, 22, and $12\%$ rates for no, moderate, and severe pain, respectively, Fig. 2c). Differences in remission rates among the three groups reached trend-level significance (χ2 = 5.26, $$p \leq 0.072$$, rTMS remission rates of 19, 11, and $3\%$ for no, moderate, or severe pain, respectively). The significant association between pain and depression symptoms persisted even after removing eight pain, sleep, and energy-related items from the IDS-SR ($r = 0.32$, $$p \leq 0.008$$ at baseline and $r = 0.35$, $$p \leq 0.003$$ at endpoint). EEG data in the comorbid patient subgroup revealed that a higher individual PAF was associated with more severe chronic pain complaints at baseline. This was confirmed for both COG and PAF measures (COG: $r = 0.41$, $$p \leq 0.00095$$; PAF: $r = 0.39$, $$p \leq 0.0047$$, Fig. 3a and b). Phase-coherence for the top 25 connections was significantly higher for comorbid responders/remitters than for non-responders/non-remitters at a level of $p \leq 0.05$ after FDR correction (Fig. 4b, Table 2). The average network connectivity across these 25 connections was also overall significantly different between responders/non-responders ($t = 3.4$, $$p \leq 0.001$$) and remitters/non-remitters ($t = 3.1$, $$p \leq 0.003$$; Fig. 4c and d, left panel). However, in the same network, there was no difference between pain-free responders/remitters and non-responders/non-remitters ($t = 1.7$, $$p \leq 0.1$$ and $t = 1.3$; $$p \leq 0.2$$, respectively; Fig. 4c and d, right panel). Fig. 4.Association between pain symptoms, rTMS response and EEG measures for comorbid subjects (VAS score>200). ( a) Left: Higher COG was associated with higher baseline VAS score. Right: Higher IAF was associated with higher baseline VAS score. ( b) Left: Phase-coherence was reduced at sensorimotor-midline locations for comorbid non-responders as compared to responders (axial and sagittal views); Right: There was no such difference between pain-free responders and non-responders. ( c) Left: The average phase-coherence of all connections was significantly reduced for comorbid non-responders compared to responders; Right: There was no such difference between pain-free responders and non-responders.(d) Left: The average phase-coherene of all connections was significantly reduced for comorbid non-remitters compared to remitters; Right: There was no such difference between pain-free remitters and non-remitters. Table 2.Listing of top 25 electrode pairs with largest Cohen's D effect sizes for differences in baseline phase-coherence between comorbid responders and non-responders to rTMS treatmentChan 1Chan 2wPLI responders mean ± s.d.wPLI non-responders mean ± s.d. T valueRaw p valueAdj p valueCohen's D1Fp1PO30.21 ± 0.120.08 ± 0.093.45030.00120.01781.22822Fp1O10.16 ± 0.10.06 ± 0.073.47880.00110.01781.19353Fp1F30.21 ± 0.120.09 ± 0.13.26690.00210.01781.16384Fp1P10.28 ± 0.20.09 ± 0.123.59268.21 × 10−040.01781.14445FpzF80.26 ± 0.170.1 ± 0.113.42860.00130.01781.13886FpzC30.25 ± 0.160.1 ± 0.13.40690.00140.01781.12727Fp1P20.27 ± 0.190.09 ± 0.113.50640.00110.01781.11258Fp1F40.24 ± 0.140.1 ± 0.113.1210.00320.01781.10349Fp1P60.27 ± 0.190.1 ± 0.123.35080.00170.01781.095710FpzCP10.25 ± 0.160.1 ± 0.113.19890.00260.01781.080411Fp1PO40.12 ± 0.080.05 ± 0.063.19960.00260.01781.063612FpzAF70.15 ± 0.110.06 ± 0.083.08460.00350.01781.051213Fp1T80.17 ± 0.10.07 ± 0.092.77550.00810.02621.050614Fp1CP20.22 ± 0.150.09 ± 0.093.2570.00220.01781.050215Fp1FC50.23 ± 0.140.1 ± 0.12.98360.00460.02111.041716Fp1CPz0.24 ± 0.180.09 ± 0.122.94610.00510.02120.987517Fp1CP10.23 ± 0.170.1 ± 0.13.14020.0030.01780.984118FpzP40.14 ± 0.10.06 ± 0.063.09190.00340.01780.974119FpzC40.21 ± 0.170.08 ± 0.093.10840.00330.01780.953120Fp1AF70.23 ± 0.170.09 ± 0.112.87680.00620.02350.950921Fp1CP60.22 ± 0.160.1 ± 0.12.85920.00650.02350.944422Fp1AF30.23 ± 0.160.09 ± 0.112.79690.00760.02570.943423FpzT70.16 ± 0.120.07 ± 0.082.86220.00640.02350.927724FpzFz0.23 ± 0.20.09 ± 0.13.04420.00390.01880.92125Fp1FCz0.13 ± 0.10.05 ± 0.043.28660.0020.01780.9114All connections were significantly different between the groups at the level of $p \leq 0.05$, FDR-corrected. ## Discussion More than half of the MDD subjects in our sample presented with moderate to severe chronic pain symptoms, independent of age or gender, and the severity of depressive and pain symptoms were highly correlated. rTMS treatment was efficacious in reducing both comorbid symptoms although the presence of pain was associated with a worse antidepressant rTMS outcome. EEG analysis revealed that individual PAF was positively associated with chronic pain severity while baseline phase-coherence along the midline and sensorimotor regions was significantly lower among non-responders/non-remitters than in responders/remitters with comorbid pain. Our finding that most MDD subjects report chronic pain symptoms that are associated with a poorer rTMS treatment outcome is consistent with the prior literature on medication treatment of MDD. The present findings confirm the high prevalence of this comorbidity and emphasize the need for a better understanding of the shared pathophysiology, as well as development of novel treatment strategies targeting both disorders (Bair et al., 2004). This study did not confirm previous reports of age- or gender-specific differences in pain prevalence and response (Phillips et al., 2018). Women and men in this study were equally likely to report and improve in pain regardless of age, although there was a trend-level effect for gender with females responding at a smaller degree. Failure to separate pain intensity and pain unpleasantness may underlie this discrepancy; future studies should attempt to more clearly distinguish these elements of pain. Baseline depressive and pain symptoms were positively correlated, and better depression outcomes correlated with greater pain improvement, even after removing the somatic symptoms from the IDS-SR. It is thus difficult to dissociate the pain response from the antidepressant effect of the rTMS treatment. Further studies are necessary to characterize the interaction of separate or common brain circuits that may be involved. These data suggest, however, that rTMS to left DLPFC may be a less effective treatment for depressed patients with comorbid pain than those without pain. This finding is consistent with prior studies that have reported a poorer outcome for MDD-pain comorbid patients when treated with medications (Bair et al., 2003; Gerrits et al., 2012; Leuchter et al., 2010). It is also in line with previous findings showing that higher baseline depression severity is associated with poorer clinical response to rTMS (Fitzgerald, Hoy, Anderson, & Daskalakis, 2016; Janicak et al., 2013; Krepel, Rush, Iseger, Sack, & Arns, 2019). We hypothesize, however, that the distinct MOA of rTMS may be better suited to target the shared neurophysiological pathways of depression and comorbid pain than antidepressant medication. The present results confirm that rTMS administered to left DLPFC does ameliorate comorbid pain symptoms in MDD. Combined with the broad prior treatment literature, however, these findings suggest that a multi-target rTMS approach for the combined treatment of MDD and chronic pain may therefore represent a promising alternative therapeutic approach. Future studies should systematically evaluate possible secondary targets for rTMS treatment of pain, such as the primary motor cortex, ventromedial prefrontal cortex, or the anterior cingulate cortex, which all have been linked to pain processing or pain relief after rTMS. Given that difficult-to-treat MDD patients with chronic pain are more likely to be administered opioids for pain relief, a novel rTMS therapeutic approach may help reduce opioid prescriptions and drug dependency for these challenging comorbid patients (Cahill & Taylor, 2017; Sullivan, Edlund, Steffick, & Unützer, 2005). These results also extend previous findings of an association between individual PAF and pain susceptibility to acute and prolonged experimentally induced pain (Furman et al., 2018; Seminowicz et al., 2018) and in chronic pain conditions (Babiloni et al., 2006; de Vries et al., 2013; Furman et al., 2019, 2020). Individual PAF represents a stable individual trait (Grandy et al., 2013; Petrosino et al., 2018) that multiple studies have shown to be associated with rTMS treatment outcome in MDD (Arns, Drinkenburg, Fitzgerald, & Kenemans, 2012; Corlier et al., 2019a, b). While our observation stands in contrast to previously reported alpha-band slowing in multiple sclerosis subjects with neuropathic pain (Kim et al., 2019), this discrepancy may be due to the different anatomical location of measured alpha rhythms, the specific type of pain, or the presence of MDD comorbidity. It is possible that MDD patients with higher PAF might have higher acute pain sensitivity prior to developing chronic pain, and that these patients may have a different intrinsic structure or function of brain circuits that conveys a higher likelihood of developing comorbid chronic pain. The present results suggest the possibility of neural circuitry common to both chronic pain and depression, and that certain neurophysiological features may define a biotype of depression that is both more susceptible to developing comorbid pain and more responsive to rTMS treatment. These findings also suggest that EEG biomarkers may help guide the search for an alternative rTMS approach for the treatment of MDD and pain comorbidity. We have identified a network including midline and sensorimotor regions that displays lower baseline phase-coherence for rTMS non-responders/non-remitters compared to responders/remitters among all patients with MDD and chronic pain comorbidity. This observation is consistent with the idea of network reorganization with chronic pain (Farmer, Baliki, & Apkarian, 2012). The topography of network should be further examined for its possible alignment with the dynamic pain connectome including the default mode network and/or salience network in chronic pain using source localization (Alshelh et al., 2018; Kucyi & Davis, 2017; Van Ettinger-Veenstra et al., 2019). Future studies also should examine whether the PAF or phase-coherence metrics in comorbid patients normalize with successful rTMS treatment. Two testable hypotheses for future studies are that: (a) a multi-target rTMS approach combining the established left DLPFC target for depression with a secondary pain target within the sensorimotor-midline network will enhance antidepressant response to rTMS, and (b) successful rTMS treatment for comorbid MDD and pain would re-establish non-comorbid levels of PAF and sensorimotor-midline functional connectivity. While the EEG findings among different pain types are beyond the scope of the present data, differences in PAF and network synchrony may be specific to certain chronic pain conditions and such possible differences should be systematically examined in future studies. ## Limitations The reported findings should be interpreted in the context of several limitations. First, these subjects were treated in a naturalistic setting with all individuals continuing their psychotropic medication during the rTMS course. Additionally, while all patients started with 10 Hz left DLPFC treatment, their stimulation protocol could be adjusted during the 30-session course according to a measurement-based care paradigm. While we did not detect a significant effect of medication or TMS parameters on outcome, it is possible that there may have been an interaction among types of concurrent medications and the rTMS effect on pain symptoms. Future trials should explicitly control for or exclude confounding medications. Second, we evaluated the severity of pain regardless of the specific etiology. Patients with fibromyalgia, neuropathic pain, chronic lower back pain, and complex regional pain syndrome were combined in this sample, despite their possibly different pathophysiological mechanisms. While rTMS has been previously shown to successfully improve various pain types, it remains unclear if the presence or direction of the association between pain symptoms and PAF or phase-coherence is clinically meaningful across different pain conditions. Follow-up research should systematically compare the effect of rTMS on different pain types. ## Conclusions We present preliminary evidence that MDD comorbidity with chronic pain results in lower rTMS response/remission rates for depression and propose that increased PAF and hypoconnectivity between sensorimotor and midline regions may predict this worse rTMS therapeutic prognosis in the comorbid population. Understanding the underlying circuitry and developing a more targeted treatment approach may help further develop a therapy for this common comorbidity. A multi-target rTMS approach represents a promising avenue for non-pharmacological treatment for comorbid MDD and chronic pain. ## Financial support This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. ## Conflict of interest Drs Corlier, Tadayonnejad, Marder, Ginder, Wilke, Levitt, and Krantz have no disclosures. Wilson has served as a consultant to HeartCloud, Inc. within the past 36 months. Dr Lee has received equipment in-kind support from Magventure Inc. Dr Leuchter discloses that within the past 36 months he has received research support from the National Institutes of Health, Department of Defense, CHDI Foundation, and NeuroSigma, Inc. He has served as a consultant to NeoSync, Inc., Ionis Pharmaceuticals, Inc., and ElMindA. He is Chief Scientific Officer of Brain Biomarker Analytics LLC (BBA). Dr Leuchter has an equity interest in BBA. ## References 1. Ahn S., Prim J. H., Alexander M. L., McCulloch K. L., Fröhlich F.. Identifying and engaging neuronal oscillations by transcranial alternating current stimulation in patients with chronic low back pain: A randomized, crossover, double-blind, sham-controlled pilot study. Journal of Pain* (2019) 20 277.e1-277.e11. DOI: 10.1016/j.jpain.2018.09.004 2. Alshelh Z., Marciszewski K. K., Akhter R., Di Pietro F., Mills E. P., Vickers E. R., Henderson L. A.. Disruption of default mode network dynamics in acute and chronic pain states. NeuroImage: Clinical (2018) 17 222-231. DOI: 10.1016/j.nicl.2017.10.019 3. Altas E. U., Askin A., Beşiroğlu L., Tosun A.. Is high-frequency repetitive transcranial magnetic stimulation of the left primary motor cortex superior to the stimulation of the left dorsolateral prefrontal cortex in fibromyalgia syndrome?. Somatosensory & Motor Research (2019) 36 56-62. DOI: 10.1080/08990220.2019.1587400 4. Arendsen L. J., Hugh-Jones S., Lloyd D. M.. Transcranial alternating current stimulation at alpha frequency reduces pain when the intensity of pain is uncertain. Journal of Pain (2018) 19 807-818. DOI: 10.1016/j.jpain.2018.02.014 5. Arns M., Drinkenburg W. H., Fitzgerald P. B., Kenemans J. L.. Neurophysiological predictors of non-response to rTMS in depression. Brain Stimulation (2012) 5 569-576. DOI: 10.1016/j.brs.2011.12.003 6. Babiloni C., Brancucci A., Del Percio C., Capotosto P., Arendt-Nielsen L., Chen A. C. N., Rossini P. M.. Anticipatory electroencephalography alpha rhythm predicts subjective perception of pain intensity. Journal of Pain (2006) 7 709-717. DOI: 10.1016/j.jpain.2006.03.005 7. Bair M. J., Robinson R. L., Eckert G. J., Stang P. E., Croghan T. W., Kroenke K.. Impact of pain on depression treatment response in primary care. Psychosomatic Medicine (2004) 66 17-22. DOI: 10.1097/01.PSY.0000106883.94059.C5 8. Bair M. J., Robinson R. L., Katon W., Kroenke K.. Depression and pain comorbidity: A literature review. Archives of Internal Medicine (2003) 163 2433-2445. DOI: 10.1001/archinte.163.20.2433 9. Beam W., Borckardt J. J., Reeves S. T., George M. S.. An efficient and accurate new method for locating the F3 position for prefrontal TMS applications. Brain Stimulation (2009) 2 50-54. DOI: 10.1016/j.brs.2008.09.006 10. Cahill C. M., Taylor A. M.. Neuroinflammation – a co-occurring phenomenon linking chronic pain and opioid dependence. Current Opinion in Behavioral Sciences (2017) 13 171-177. DOI: 10.1016/j.cobeha.2016.12.003 11. Carpenter L. L., Conelea C., Tyrka A. R., Welch E. S., Greenberg B. D., Price L. H., Philip N. S.. 5Hz Repetitive transcranial magnetic stimulation for posttraumatic stress disorder comorbid with major depressive disorder. Journal of Affective Disorders (2018) 235 414-420. DOI: 10.1016/j.jad.2018.04.009 12. Chang M. H., Hsu J. W., Huang K. L., Su T. P., Bai Y. M., Li C. T., Chen M. H.. Bidirectional association between depression and fibromyalgia syndrome: A nationwide longitudinal study. Journal of Pain (2015) 16 895-902. DOI: 10.1016/j.jpain.2015.06.004 13. Chiarotto A., Maxwell L. J., Ostelo R. W., Boers M., Tugwell P., Terwee C. B.. Measurement properties of visual analogue scale, numeric rating scale, and pain severity subscale of the brief pain inventory in patients with low back pain: A systematic review. Journal of Pain (2019) 20 245-263. DOI: 10.1016/j.jpain.2018.07.009 14. Corlier J., Carpenter L. L., Wilson A. C., Tirrell E., Gobin A. P., Kavanaugh B., Leuchter A. F.. The relationship between individual alpha peak frequency and clinical outcome with repetitive transcranial magnetic stimulation (rTMS) treatment of major depressive disorder (MDD). Brain Stimulation (2019a) 12 1572-1578. DOI: 10.1016/j.brs.2019.07.018 15. Corlier J., Wilson A., Hunter A. M., Vince-Cruz N., Krantz D., Levitt J., Leuchter A. F.. Changes in functional connectivity predict outcome of repetitive transcranial magnetic stimulation treatment of major depressive disorder. Cerebral Cortex (2019b) 29 4958-4967. DOI: 10.1093/cercor/bhz035 16. de Vries M., Wilder-Smith O. H. G., Jongsma M. L. A., van den Broeke E. N., Arns M., van Goor H., van Rijn C. M.. Altered resting state EEG in chronic pancreatitis patients: Toward a marker for chronic pain. Journal of Pain Research (2013) 6 815-824. DOI: 10.2147/JPR.S50919 17. Farmer M. A., Baliki M. N., Apkarian A. V.. A dynamic network perspective of chronic pain. Neuroscience Letters (2012) 520 197-203. DOI: 10.1016/j.neulet.2012.05.001 18. Fitzgerald P. B., Hoy K. E., Anderson R. J., Daskalakis Z. J.. A study of the pattern of response to rTMS treatment in depression. Depression and Anxiety (2016) 33 746-753. DOI: 10.1002/da.22503 19. Fox M. D., Halko M. A., Eldaief M. C., Pascual-Leone A.. Measuring and manipulating brain connectivity with resting state functional connectivity magnetic resonance imaging (fcMRI) and transcranial magnetic stimulation (TMS). Human Brain Mapping Journal (2012) 62 2232-2243 20. Furman A. J., Meeker T. J., Rietschel J. C., Yoo S., Muthulingam J., Prokhorenko M., Seminowicz D. A.. Cerebral peak alpha frequency predicts individual differences in pain sensitivity. NeuroImage (2018) 167 203-210. DOI: 10.1016/j.neuroimage.2017.11.042 21. Furman A. J., Prokhorenko M., Keaser M. L., Zhang J., Chen S., Mazaheri A., Seminowicz D. A.. Sensorimotor peak alpha frequency is a reliable biomarker of prolonged pain sensitivity. Cerebral Cortex (2020) 30 6069-6082. DOI: 10.1093/cercor/bhaa124 22. Furman A. J., Thapa T., Summers S. J., Cavaleri R., Fogarty J. S., Steiner G. Z., Seminowicz D. A.. Cerebral peak alpha frequency reflects average pain severity in a human model of sustained, musculoskeletal pain. Journal of Neurophysiology (2019) 122 1784-1793. DOI: 10.1152/jn.00279.2019 23. Galhardoni R., Correia G. S., Araujo H., Yeng L. T., Fernandes D. T., Kaziyama H. H., de Andrade D. C.. Repetitive transcranial magnetic stimulation in chronic pain: A review of the literature. Archives of Physical Medicine and Rehabilitation (2015) 96 S156-S172. DOI: 10.1016/j.apmr.2014.11.010 24. George M. S., Lisanby S. H., Avery D., McDonald W. M., Durkalski V., Pavlicova M., Sackeim H. A.. Daily left prefrontal transcranial magnetic stimulation therapy for major depressive disorder: A sham-controlled randomized trial. Archives of General Psychiatry (2010) 67 507-516. DOI: 10.1001/archgenpsychiatry.2010.46 25. Gerrits M. M. J. G., Van Oppen P., Van Marwijk H. W. J., Penninx B. W. J. H., Van Der Horst H. E.. Pain and the onset of depressive and anxiety disorders. Pain (2014) 155 53-59. DOI: 10.1016/j.pain.2013.09.005 26. Gerrits M. M. J. G., Vogelzangs N., Van Oppen P., Van Marwijk H. W. J., Van Der Horst H., Penninx B. W. J. H.. Impact of pain on the course of depressive and anxiety disorders. Pain (2012) 153 429-436. DOI: 10.1016/j.pain.2011.11.001 27. Goudra B., Shah D., Balu G., Gouda G., Balu A., Borle A., Singh P.. Repetitive transcranial magnetic stimulation in chronic pain: A meta-analysis. Anesthesia: Essays and Researches (2017) 11 751. DOI: 10.4103/aer.AER_10_17 28. Gracely R. H., Ceko M., Bushnell M. C.. Fibromyalgia and depression. Pain Research and Treatment (2012) 2012. DOI: 10.1155/2012/486590 29. Grandy T. H., Werkle-Bergner M., Chicherio C., Schmiedek F., Lövdén M., Lindenberger U.. Peak individual alpha frequency qualifies as a stable neurophysiological trait marker in healthy younger and older adults. Psychophysiology (2013) 50 570-582. DOI: 10.1111/psyp.12043 30. Hamilton M.. A rating scale for depression. Journal of Neurology, Neurosurgery and Psychiatry (1960) 23 56-62. DOI: 10.1136/jnnp.23.1.56 31. Heath A., Taylor J., McNerney M. W.. rTMS for the treatment of Alzheimer's disease: Where should we be stimulating?. Expert Review of Neurotherapeutics (2018) 18 903-905. DOI: 10.1080/14737175.2018.1538792 32. Hou W.-H., Wang T.-Y., Kang J.-H.. The effects of add-on non-invasive brain stimulation in fibromyalgia: A meta-analysis and meta-regression of randomized controlled trials. Rheumatology (2016) 55 1507-1517. DOI: 10.1093/rheumatology/kew205 33. Hsu J. H., Daskalakis Z. J., Blumberger D. M.. An update on repetitive transcranial magnetic stimulation for the treatment of co-morbid pain and depressive symptoms. Current Pain and Headache Reports (2018) 22 51. DOI: 10.1007/s11916-018-0703-7 34. Janicak P. G., Dunner D. L., Aaronson S. T., Carpenter L. L., Boyadjis T. A., Brock D. G., Demitrack M. A.. Transcranial magnetic stimulation (TMS) for major depression: A multisite, naturalistic, observational study of quality of life outcome measures in clinical practice. CNS Spectrums (2013) 18 322-332. DOI: 10.1017/S1092852913000357 35. Johnson S., Summers J., Pridmore S.. Changes to somatosensory detection and pain thresholds following high frequency repetitive TMS of the motor cortex in individuals suffering from chronic pain. Pain (2006) 123 187-192. DOI: 10.1016/j.pain.2006.02.030 36. Karcioglu O., Topacoglu H., Dikme O., Dikme O.. A systematic review of the pain scales in adults: Which to use?. American Journal of Emergency Medicine (2018) 36 707-714. DOI: 10.1016/j.ajem.2018.01.008 37. Kim J. A., Bosma R. L., Hemington K. S., Rogachov A., Osborne N. R., Cheng J. C., Davis K. D.. Neuropathic pain and pain interference are linked to alpha-band slowing and reduced beta-band magnetoencephalography activity within the dynamic pain connectome in patients with multiple sclerosis. Pain (2019) 160 187-197. DOI: 10.1097/j.pain.0000000000001391 38. Kim J. A., Davis K. D.. Neural oscillations: Understanding a neural code of pain. Neuroscientist (2020). DOI: 10.1177/1073858420958629 39. Kisler L. B., Kim J. A., Hemington K. S., Rogachov A., Cheng J. C., Bosma R. L., Davis K. D.. Abnormal alpha band power in the dynamic pain connectome is a marker of chronic pain with a neuropathic component. NeuroImage: Clinical (2020) 26. DOI: 10.1016/j.nicl.2020.102241 40. Knijnik L. M., Dussán-Sarria J. A., Rozisky J. R., Torres I. L. S., Brunoni A. R., Fregni F., Caumo W.. Repetitive transcranial magnetic stimulation for fibromyalgia: Systematic review and meta-analysis. Pain Practice (2016) 16 294-304. DOI: 10.1111/papr.12276 41. Krepel N., Rush A. J., Iseger T. A., Sack A. T., Arns M.. Can psychological features predict antidepressant response to rTMS? A discovery-replication approach. Psychological Medicine (2019) 50 264-272. DOI: 10.1017/S0033291718004191 42. Kucyi A., Davis K. D.. The neural code for pain: From single-cell electrophysiology to the dynamic pain connectome. Neuroscientist (2017) 23 397-414. DOI: 10.1177/1073858416667716 43. Lee J. C., Wilson A. C., Corlier J., Tadayonnejad R., Marder K. G., Pleman C. M., Leuchter A. F.. Strategies for augmentation of high-frequency left-sided repetitive transcranial magnetic stimulation treatment of major depressive disorder. Journal of Affective Disorders (2020) 277 964-969. DOI: 10.1016/j.jad.2020.09.011 44. Lefaucheur J. P., Aleman A., Baeken C., Benninger D. H., Brunelin J., Di Lazzaro V., Ziemann U.. Evidence-based guidelines on the therapeutic use of repetitive transcranial magnetic stimulation (rTMS): An update (2014–2018). Clinical Neurophysiology (2020) 131 474-528. DOI: 10.1016/j.clinph.2019.11.002 45. Lefaucheur J. P., André-Obadia N., Antal A., Ayache S. S., Baeken C., Benninger D. H., Garcia-Larrea L.. Evidence-based guidelines on the therapeutic use of repetitive transcranial magnetic stimulation (rTMS). Clinical Neurophysiology (2014) 125 2150-2206. DOI: 10.1016/j.clinph.2014.05.021 46. Leuchter A. F., Hunter A. M., Krantz D. E., Cook I. A.. Rhythms and blues: Modulation of oscillatory synchrony and the mechanism of action of antidepressant treatments. Annals of the New York Academy of Sciences (2015) 1344 78-91. DOI: 10.1111/nyas.12742 47. Leuchter A. F., Husain M. M., Cook I. A., Trivedi M. H., Wisniewski S. R., Gilmer W. S., Rush A. J.. *Painful physical symptoms and treatment outcome in major depressive disorder: A STARD (Sequenced Treatment Alternatives to Relieve Depression) report*. Psychological Medicine* (2010) 40 239-251. DOI: 10.1017/S0033291709006035 48. Lopez-Diaz K., Henshaw J., Casson A. J., Brown C. A., Taylor J. R., Trujillo-Barreto N. J., Sivan M.. Alpha entrainment drives pain relief using visual stimulation in a sample of chronic pain patients: A proof-of-concept controlled study. NeuroReport (2021) 32 394-398. DOI: 10.1097/WNR.0000000000001606 49. Maletic V., Raison C. L.. Neurobiology of depression, fibromyalgia and neuropathic pain. Frontiers in Bioscience (2009) 14 5291-5338. DOI: 10.2741/3598 50. Meneses F. M., Queirós F. C., Montoya P., Miranda J. G. V., Dubois-Mendes S. M., Sá K. N., Baptista A. F.. Patients with rheumatoid arthritis and chronic pain display enhanced alpha power density at rest. Frontiers in Human Neuroscience (2016) 10 11. DOI: 10.3389/fnhum.2016.00395 51. Mori F., Codecà C., Kusayanagi H., Monteleone F., Buttari F., Fiore S., Centonze D.. Effects of anodal transcranial direct current stimulation on chronic neuropathic pain in patients with multiple sclerosis. Journal of Pain (2010) 11 436-442. DOI: 10.1016/j.jpain.2009.08.011 52. Nolan H., Whelan R., Reilly R. B.. FASTER: Fully automated statistical thresholding for EEG artifact rejection. Journal of Neuroscience Methods (2010) 192 152-162. DOI: 10.1016/j.jneumeth.2010.07.015 53. Onesti E., Gabriele M., Cambieri C., Ceccanti M., Raccah R., Di Stefano G., Inghilleri M.. H-coil repetitive transcranial magnetic stimulation for pain relief in patients with diabetic neuropathy. European Journal of Pain (2013) 17 1347-1356. DOI: 10.1002/j.1532-2149.2013.00320.x 54. Petrosino N. J., Zandvakili A., Carpenter L. L., Philip N. S.. Pilot testing of peak alpha frequency stability during repetitive transcranial magnetic stimulation. Frontiers in Psychiatry (2018) 9. DOI: 10.3389/fpsyt.2018.00605 55. Phillips A. L., Burr R. L., Dunner D. L.. rTMS effects in patients with co-morbid somatic pain and depressive mood disorders. Journal of Affective Disorders (2018) 241 411-416. DOI: 10.1016/j.jad.2018.08.065 56. Ploner M., Sorg C., Gross J.. Brain rhythms of pain. Trends in Cognitive Sciences (2017) 21 100-110. DOI: 10.1016/j.tics.2016.12.001 57. Saltychev M., Laimi K.. Effectiveness of repetitive transcranial magnetic stimulation in patients with fibromyalgia. International Journal of Rehabilitation Research (2017) 40 11-18. DOI: 10.1097/MRR.0000000000000207 58. Seminowicz D. A., Thapa T., Furman A. J., Summers S. J., Cavaleri R., Fogarty J. S., Schabrun S. M.. (2018). DOI: 10.1101/278598 59. Sheehan D. V., Lecrubier Y., Sheehan K. H., Amorim P., Janavs J., Weiller E., Dunbar G. C.. The Mini-International Neuropsychiatric Interview (M.I.N.I.): The development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. Journal of Clinical Psychiatry (1998) 59 22-33 60. Short B. E., Borckardt J. J., Anderson B. S., Frohman H., Beam W., Reeves S. T., George M. S.. Ten sessions of adjunctive left prefrontal rTMS significantly reduces fibromyalgia pain: A randomized, controlled pilot study. Pain (2011) 152 2477-2484. DOI: 10.1016/j.pain.2011.05.033 61. Sinyor M., Schaffer A., Levitt A.. *The sequenced treatment alternatives to relieve depression (STARD) trial: A review*. Canadian Journal of Psychiatry* (2010) 55 126-135. DOI: 10.1177/070674371005500303 62. Soleimani R., Jalali M. M., Hasandokht T.. Therapeutic impact of repetitive transcranial magnetic stimulation (rTMS) on tinnitus: A systematic review and meta-analysis. European Archives of Oto-Rhino-Laryngology (2016) 273 1663-1675. DOI: 10.1007/s00405-015-3642-5 63. Sullivan M. D., Edlund M. J., Steffick D., Unützer J.. Regular use of prescribed opioids: Association with common psychiatric disorders. Pain (2005) 119 95-103. DOI: 10.1016/j.pain.2005.09.020 64. Tadayonnejad R., Wilson A. C., Corlier J., Lee J. C., Ginder N. D., Levitt J. G., Leuchter A. F.. Sequential multi-locus transcranial magnetic stimulation for treatment of obsessive-compulsive disorder with comorbid major depression: A case series. Brain Stimulation (2020) 13 1600-1602. DOI: 10.1016/j.brs.2020.10.003 65. Taylor A. M. W., Becker S., Schweinhardt P., Cahill C.. Mesolimbic dopamine signaling in acute and chronic pain: Implications for motivation, analgesia, and addiction. Pain (2016) 157 1194-1198. DOI: 10.1097/j.pain.0000000000000494 66. To W. T., De Ridder D., Hart J., Vanneste S.. Changing brain networks through non-invasive neuromodulation. Frontiers in Human Neuroscience (2018) 12 1-17. DOI: 10.3389/fnhum.2018.00128 67. Trivedi M. H., Rush A. J., Ibrahim H. M., Carmody T. J., Biggs M. M., Suppes T., Kashner T. M.. The inventory of depressive symptomatology, clinician rating (IDS-C) and self-report (IDS-SR), and the quick inventory of depressive symptomatology, clinician rating (QIDS-C) and self-report (QIDS-SR) in public sector patients with mood disorders: A psychometric evaluation. Psychological Medicine (2004) 34 73-82. PMID: 14971628 68. Van Ettinger-Veenstra H., Lundberg P., Alföldi P., Södermark M., Graven-Nielsen T., Sjörs A., Gerdle B.. Chronic widespread pain patients show disrupted cortical connectivity in default mode and salience networks, modulated by pain sensitivity. Journal of Pain Research (2019) 12 1743-1755. DOI: 10.2147/JPR.S189443 69. Vinck M., Oostenveld R., Van Wingerden M., Battaglia F., Pennartz C. M. A.. An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias. NeuroImage (2011) 55 1548-1565. DOI: 10.1016/j.neuroimage.2011.01.055 70. Von Korff M., Simon G.. The relationship between pain and depression. British Journal of Psychiatry (1996) 168 101-108. DOI: 10.1192/s0007125000298474 71. Walker A. K., Kavelaars A., Heijnen C. J., Dantzer R.. Neuroinflammation and comorbidity of pain and depression. Pharmacological Reviews (2014) 66 80-101. DOI: 10.1124/pr.113.008144 72. Wang W. E., Roy A., Misra G., Ho R. L. M., Ribeiro-Dasilva M. C., Fillingim R. B., Coombes S. A.. Altered neural oscillations within and between sensorimotor cortex and parietal cortex in chronic jaw pain. NeuroImage: Clinical (2019) 24. DOI: 10.1016/j.nicl.2019.101964 73. Xing M., Ajilore O., Wolfson O. E., Abbott C., Macnamara A., Tadayonnejad R., Leow A.. (2016a). DOI: 10.1007/978-3-319-47103-7_15 74. Xing M., Tadayonnejad R., Macnamara A., Ajilore O., Phan K. L., Klumpp H., Leow A.. (2016b). DOI: 10.1109/ISBI.2016.7493380
--- title: Assessment of serum biotin levels and its association with blood glucose in gestational diabetes mellitus authors: - N. Muthuraman - Reeta Vijayselvi - Yesudas Sudhakar P - Pamela Christudoss - Premila Abraham journal: 'European Journal of Obstetrics & Gynecology and Reproductive Biology: X' year: 2023 pmcid: PMC9976203 doi: 10.1016/j.eurox.2023.100181 license: CC BY 4.0 --- # Assessment of serum biotin levels and its association with blood glucose in gestational diabetes mellitus ## Abstract ### Aim The incidence of gestational diabetes mellitus is increasing worldwide. Biotin is shown to improve glycemic status in diabetes mellitus. We wanted to study whether there is a difference in biotin levels between mothers with and without gestational diabetes mellitus (GDM), association of biotin with blood glucose, and with the outcome of GDM. ### Methods We recruited 27 pregnant mothers with GDM and 27 pregnant mothers without GDM. We measured the biotin levels using enzyme linked immunosorbent assay (ELISA). We measured the blood glucose during OGTT and fasting insulin levels in the study participants. ### Results We found that biotin levels were slightly decreased in mothers with GDM [271 [250,335]] as compared to control mothers [309 [261,419]], though it was not statistically significant ($$p \leq 0.14$$). Blood glucose levels were found to be significantly higher in GDM mothers as compared to control mothers during fasting, 1 h and 2 h plasma sample obtained during OGTT. Biotin was not significantly associated with blood glucose in pregnant mothers. Logistic regression analysis showed that biotin (OR = 0.99, 95 % CI = 0.99–1.00) has no association with the outcome of GDM. ### Conclusion Ours is the first study to compare the biotin levels in GDM mothers and control mothers. We found that the biotin levels were not significantly altered in GDM mothers as compared to control mothers and biotin levels have no association with the outcome of GDM. ## Highlights •This is the first study to compare biotin levels in GDM mothers with control mothers.•Biotin levels were not significantly different in GDM mothers as compared to control mothers.•Biotin has no association with the outcome of GDM. ## Introduction Gestational diabetes mellitus (GDM) is defined by American Diabetes Association (ADA) as “diabetes diagnosed in the second or third trimester of pregnancy that was not clearly overt diabetes prior to gestation” [1]. The global prevalence of gestational diabetes mellitus is estimated to vary from 1 % to 30 % in different parts of the world [2]. Some of the known risk factors that can predispose the development of GDM in a pregnant mother are increased maternal age, obesity/overweight, family history of diabetes mellitus and cigarette smoking [3]. Insulin resistance and β-cell dysfunction are considered to be the main pathophysiology behind the development of GDM [4]. GDM can affect the outcomes of pregnancy by increasing the maternal and foetal complications. Short term complications of hyperglycaemia in GDM can lead to preeclampsia and increased propensity to undergo cesarean section in mothers. In infants born to mothers with GDM, complications such as macrosomia, shoulder dystocia, neonatal hypoglycemia and increased requirement for admission to intensive care units are commonly seen [5]. Meta-analysis results clearly shows that women who have GDM during pregnancy are at an increased risk of developing type 2 diabetes mellitus, as compared to the women who were normoglycemic during pregnancy [6]. Particularly women who were diagnosed to have GDM using International Association of Diabetes and Pregnancy Study Group (IADPSG) criteria are at higher risk of developing prediabetes and type 2 diabetes mellitus in the future [7]. Biotin is a water-soluble vitamin which acts as a coenzyme for carboxylase enzymes that is involved in the metabolism of carbohydrates and lipids [8]. Biotin is proposed as one of the natural supplements that can improve insulin sensitivity and glucose uptake in skeletal muscle [9]. Biotin is shown to have its hypoglycemic effect by increasing the expression of hepatic glucokinase gene and thereby effectively decreasing the blood glucose levels [10]. Trials using high dose biotin for type 1 diabetics have found promising results such as good glycemic control and better response to insulin insensitivity [11]. Also, there are animal studies which have shown that administration of high dose biotin can improve glucose tolerance in rats [12], [13]. Our pilot study showed that mega doses of biotin supplementation in diabetic rats lead to better pregnancy outcomes [14]. This prompted us to think whether biotin supplementation would be beneficial to GDM mothers in terms of reducing maternal and fetal complications. To pursue that, we thought that it would be relevant and informative to compare the biotin levels in mothers with and without GDM. We hypothesised that the biotin levels can be low in mothers with GDM as compared to the normoglycemic mothers and can be associated with the outcome of GDM. In this study we assessed the biotin levels in the pregnant mothers and looked for its association with blood glucose, insulin, and the outcome of GDM. ## Study participants This study was approved by the Institutional review board of Christian Medical College, Vellore. Pregnant woman in the gestational age of 24–28 weeks, who were referred for an oral glucose tolerance test (OGTT) from antenatal clinic in Obstetrics and Gynaecology – unit 4 of CMC Vellore, were recruited for the study after obtaining informed consent. The required sample size to show the difference in biotin levels between women with GDM and those without GDM was found to be 24 in each group, with 90 % power and alpha error of 5 %. Calculations were based on publication by Donald M. Mock et al. [ 15]. Only those who gave informed consent to participate in the study were included. Those who were unwilling to participate, known diabetic, on biotin supplements and with any pregnancy related complications were excluded from the study. Based on the one step (75 g of oral glucose) OGTT results, as per The International Association of the Diabetes and Pregnancy Study Groups (IADPSG) criteria (fasting blood glucose ≥92 mg/dL, or 1 h blood glucose ≥180 mg/dL, or 2 h blood glucose ≥153 mg/dL) they were grouped as pregnant women with GDM and pregnant women without GDM (control). ## Biotin estimation Biotin concentration was estimated in the fasting blood sample that was collected at the time of OGTT. We used IDK® Biotin ELISA K 8141 kit to estimate the biotin concentration in the serum sample. We followed the instructions given in the manual with inclusion of appropriate standards and control samples to carry out the assay. ## Blood glucose estimation Glucose levels were estimated in blood sample collected during fasting, 1 h and 2 h during OGTT. Glucose levels were estimated by glucose oxidase peroxidase method using Roche Cobas 8000c system in Clinical biochemistry department with strict adherence to quality control. ## Insulin levels Fasting blood sample collected at 24–28 weeks of gestation were used to measure insulin levels. Insulin levels were measured using Siemens CLIA, following the instructions laid out in the kit with strict adherence to quality control. Homeostasis model assessment of insulin resistance (HOMA-IR), was calculated using fasting insulin levels and fasting blood glucose level by appropriate formula as described earlier[16]. ## Statistical analysis Statistical analysis was done using R version 4.2.1. RStudio: Integrated Development Environment for R. RStudio, PBC, Boston, MA URL http://www.rstudio.com/. Normality of the data was tested using Shapiro-Wilk test. Wilcoxon Rank Sum test (Mann Whitney U test) was done to compare the median between two groups. Correlational analysis between the variables were done using Spearman correlation test. Simple logistic regression analysis was done to test the association of biotin with GDM outcome. ## Baseline characteristics of the study participants The baseline characteristics of the study participants are given in Table 1. The median age of pregnant mothers with GDM is slightly higher than the mothers without GDM and it approached close to statistical significance ($$p \leq 0.059$$, Fig. 1). As has been reported before the risk of gestational diabetes mellitus increases as the age during which the mother gets pregnant increases, which corroborates with our study’s finding.[17].Table 1Baseline characteristics of the study participants. Table 1VariableNControl, $$n = 271$$GDM, $$n = 271$$Age in years5426[24,29]29[26,32]Fasting blood glucose (mg/dL)5480 [76,86]89 [84,95]1 h blood glucose during OGTT (mg/dL)54135 [126, 152]178 [152, 188]2 h blood glucose during OGTT (mg/dL)54109 [106, 116]144 [124, 158]Biotin (ng/L)54309 [261, 419]271 [250, 335]Fasting Insulin (μU/mL)548.3 (5.5, 11.1)10.0 (5.1, 13.2)HOMA-IR541.70 (1.05, 2.25)2.10 (1.15, 3.15)1 Median (IQR)Fig. 1Comparison of median (IQR) age between GDM mothers and control using Wilcoxon Rank Sum test (Mann Whitney U test).Fig. 1 ## Blood glucose concentration All the study participants underwent oral glucose tolerance test during 24–28 weeks of gestation. We found that the fasting blood glucose levels obtained during OGTT were higher in the GDM mothers as compared to the mothers without GDM and it was found to be statistically significant ($p \leq 0.001$, Fig. 2A). During OGTT there was a statistically significant difference in the blood glucose levels measured at 1 h ($p \leq 0.001$, Fig. 2B) and 2 hours ($p \leq 0.001$, Fig. 2C) of post 75 g glucose intake between GDM mothers and mothers without GDM.Fig. 2Comparison of median (IQR) blood glucose value during fasting (2 A), at 1 h (2B) and 2 h (2 C) of OGTT between GDM mothers and control using Wilcoxon Rank Sum test (Mann Whitney U test).Fig. 2 ## Fasting insulin levels Serum insulin levels which were known to be elevated in early stages of insulin resistance and type 2 diabetes was measured in both the groups. We found a strong positive correlation between fasting blood glucose levels and fasting insulin levels in pregnant mothers, which was statistically significant ($R = 0.39$, $p \leq 0.01$, Fig. 3A). Fasting insulin levels were slightly higher in GDM mothers [10 (5.1,13.2)] as compared to mothers who didn’t have GDM [8.3 (5.5,11.1)], however, the difference was not statistically significant ($$p \leq 0.5$$, Fig. 3B). HOMA-IR value was found to be slightly higher in GDM mothers [2.10 (1.15, 3.15)] as compared to control mothers [1.70 (1.05, 2.25)] (Table 1). This is indicative of insulin resistance in the GDM mothers; however, the difference was not statistically significant ($$p \leq 0.2$$).Fig. 3Correlation analysis between insulin and fasting blood glucose (3 A) using Spearman method. Comparison of median (IQR) TNF alpha levels (3B) between GDM mothers and control using Wilcoxon Rank Sum test (Mann Whitney U test).Fig. 3 ## Biotin concentration in the fasting serum sample Biotin levels were measured in fasting blood samples of both GDM mothers and mothers without GDM collected during the time of OGTT. We found that the biotin levels were slightly decreased in the GDM mothers [271 [250,335]] as compared to the mothers without GDM [309 [261,419]]. The difference in biotin levels between the two groups was not statistically significant ($$p \leq 0.14$$, Fig. 4). We did correlational analysis for biotin with all the other study variables and found that biotin was not significantly associated with any of the studied variables (Table 2). We did a simple logistic regression analysis to see whether biotin has got any association with the outcome of gestational diabetes mellitus (Table 3). Our analysis showed that biotin levels in the serum has no association with GDM outcome (OR = 0.99, 95 % CI = 0.99–1.00).Fig. 4Comparison of median (IQR) biotin levels between GDM mothers and control using Wilcoxon Rank Sum test (Mann Whitney U test).Fig. 4Table 2Correlational analysis of biotin with other variables using Spearman method. Table 2VariablesBiotinRp valueInsulin-0.0450.75Fasting blood glucose-0.190.171 hr blood glucose during OGTT-0.20.142 hrs blood glucose during OGTT-0.130.33Table 3Simple logistic regression analysis to see the association of biotin with GDM.Table 3GDMPredictorsOdds RatiosCIp(Intercept)6.070.75 – 60.070.103Biotin ng L0.990.99 – 1.000.095Observations54R2 Tjur0.054 ## Discussion A recent meta-analysis showed that for every one year increase in maternal age from 18, the risk of GDM increases by 7.90 % in the pregnant mothers [18], which is reflected in our study as shown by the increased median age in pregnant mothers with GDM as compared to the pregnant mothers without GDM. We used the one step procedure and IADPSG criteria, which diagnoses GDM when there is violation of any one parameter (fasting, 1 hr or 2hrs blood glucose during OGTT) resulting in high false positivity, however very effective in identifying individuals who are more prone to develop type 2 diabetes in the future [19]. Our results demonstrate that glucose levels measured at all time points such as fasting, 1 h and 2 hrs during OGTT was significantly higher in mothers with GDM as compared to mothers without GDM even though the individual participants rarely violated all three parameters. Insulin resistance which is defined as decreased biological response to insulin is common in prediabetic state and type 2 diabetes mellitus [20]. Fasting insulin and HOMA-IR which are considered as markers of insulin resistance [21] are slightly elevated in GDM mothers as compared to controls, though it was not statistically significant in our study. Studies have shown that increased HOMA-IR in early pregnancy can be used as a predictive marker of gestational diabetes mellitus and were found to be higher in GDM mothers as compared to control mothers [22], [23]. Biotin levels were found to be lower in individuals with type 2 diabetes and administration of biotin was shown to lower blood glucose level in these individuals without altering the insulin levels [24]. A recent systematic review and meta-analysis on biotin showed that biotin can reduce fasting blood glucose, total cholesterol and triglyceride levels in individuals with type 2 diabetes mellitus [25]. In animal models biotin administration improved the glycaemic status of rats induced with diabetes mellitus [13], [26]. In our previous study we found that mega doses of biotin supplementation improved the pregnancy outcomes in streptozotocin induced gestational diabetes in rats [14]. Biotin levels assessed in serum and urinary excretion of biotin metabolites showed that biotin levels were reduced in pregnant mothers as compared to non-pregnant women [15]. Our study shows that biotin levels were not significantly different between GDM mothers and controls, even though it was slightly higher in control mothers. Our analysis reveals that biotin has no association with blood glucose, insulin, and the outcome of GDM in mothers. Small sample size is the limitation of our study. Including more participants in the study with analysis of biotin levels in different time points in pregnancy may give us directions towards the utility of biotin in gestational diabetes mellitus. ## Funding agency This study was funded by the Fluid research grant from Christian Medical College, Vellore, India IRB Min No. 11866 obtained by Muthuraman N. ## CRediT authorship contribution statement PA and MN designed the study. MN collected the sample, did the ELISA assays, analysed, and interpreted the results, wrote the first draft of the manuscript, and revised the edits in the draft. RV recruited the patients, gave critical inputs in the design of the study, edited the manuscript, and approved it. YSP did the biochemical assays, edited the manuscript, and approved it. PC supervised the assays, edited the manuscript, and approved it. PA interpreted the results, helped in drafting the manuscript, edited the manuscript and approved it. ## Conflict of Interest The authors declare no conflict of interest for this work. ## References 1. Professional Practice Committee. 2. Classification and diagnosis of diabetes: standards of medical care in diabetes—2022. diabetes care. 45(Supplement_1) (2021.0) S17-S38 2. McIntyre H.D., Catalano P., Zhang C., Desoye G., Mathiesen E.R., Damm P.. Gestational diabetes mellitus. Nat Rev Dis Prim (2019.0) 5 1-19. PMID: 30617281 3. Solomon C.G., Willett W.C., Carey V.J., Rich-Edwards J., Hunter D.J., Colditz G.A.. A prospective study of pregravid determinants of gestational diabetes mellitus. JAMA (1997.0) 278 1078-1083. PMID: 9315766 4. Catalano P.M., Huston L., Amini S.B., Kalhan S.C.. Longitudinal changes in glucose metabolism during pregnancy in obese women with normal glucose tolerance and gestational diabetes mellitus. Am J Obstet Gynecol (1999.0) 180 903-916. PMID: 10203659 5. Metzger B.E., Lowe L.P., Dyer A.R., Trimble E.R., Chaovarindr U.. Hyperglycemia and adverse pregnancy outcomes. N Engl J Med (2008.0) 358 1991-2002. PMID: 18463375 6. Bellamy L., Casas J.P., Hingorani A.D., Williams D.. Type 2 diabetes mellitus after gestational diabetes: a systematic review and meta-analysis. Lancet Lond Engl (2009.0) 373 1773-1779 7. Juan J., Sun Y., Wei Y., Wang S., Song G., Yan J.. Progression to type 2 diabetes mellitus after gestational diabetes mellitus diagnosed by IADPSG criteria: Systematic review and meta-analysis. Front Endocrinol (2022.0) 13 1012244 8. McMahon R.J.. Biotin in metabolism and molecular biology. Annu Rev Nutr (2002.0) 22 221-239. PMID: 12055344 9. 9Sa K., S.Y., Ds N., Aj DN. Natural supplements for improving insulin sensitivity and glucose uptake in skeletal muscle. Front Biosci Elite Ed [Internet]. 2015 Jan 1 [cited 2022 Nov 19];7(1). Available from: https://pubmed.ncbi.nlm.nih.gov/25553366/. 10. Dakshinamurti K., Cheah-Tan C.. Biotin-mediated synthesis of hepatic glucokinase in the rat. Arch Biochem Biophys (1968.0) 127 17-21. PMID: 5681418 11. 11Mf M. In type 1 diabetics, high-dose biotin may compensate for low hepatic insulin exposure, promoting a more normal expression of glycolytic and gluconeogenic enyzymes and thereby aiding glycemic control. Med Hypotheses [Internet]. 2016 Oct [cited 2022 Nov 19];95. Available from: https://pubmed.ncbi.nlm.nih.gov/27692165/. 12. 12H Z, K O, H S, Y F. Biotin administration improves the impaired glucose tolerance of streptozotocin-induced diabetic Wistar rats. J Nutr Sci Vitaminol (Tokyo) [Internet]. 1997 Jun [cited 2022 Nov 19];43(3). Available from: https://pubmed.ncbi.nlm.nih.gov/9268917/. 13. 13H Z, K O, M M, M I, M K, Y F. A high biotin diet improves the impaired glucose tolerance of long-term spontaneously hyperglycemic rats with non-insulin-dependent diabetes mellitus. J Nutr Sci Vitaminol (Tokyo) [Internet]. 1996 Dec [cited 2022 Nov 19];42(6). Available from: https://pubmed.ncbi.nlm.nih.gov/9089478/. 14. 14Abraham P., Ramamoorthy H., Francis D.V., Rabi S. Beneficial Effects of Megadoses of Biotin in Streptozotocin-induced Gestational Diabetes Mellitus in Rats – A Preliminary Study [Internet]. 2017 [cited 2022 Nov 23]. Available from: https://www.semanticscholar.org/paper/Beneficial-Effects-of-Megadoses-of-Biotin-in-in-%E2%80%93-A-Abraham-Ramamoorthy/e8cee7d81427a9ee0bf66a74d75ccd57c6b92c3f. 15. Mock D.M., Stadler D.D., Stratton S.L., Mock N.I.. Biotin status assessed longitudinally in pregnant women. J Nutr (1997.0) 127 710-716. PMID: 9164991 16. Matthews D.R., Hosker J.P., Rudenski A.S., Naylor B.A., Treacher D.F., Turner R.C.. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia (1985.0) 28 412-419. PMID: 3899825 17. Lao T.T., Ho L.F., Chan B.C.P., Leung W.C.. Maternal age and prevalence of gestational diabetes mellitus. Diabetes Care (2006.0) 29 948-949. PMID: 16567851 18. Li Y., Ren X., He L., Li J., Zhang S., Chen W.. Maternal age and the risk of gestational diabetes mellitus: A systematic review and meta-analysis of over 120 million participants. Diabetes Res Clin Pr (2020.0) 162 19. Rani P.R., Begum J.. Screening and diagnosis of gestational diabetes mellitus. Where Do We Stand J Clin Diagn Res JCDR (2016.0) 10 20. Himsworth H.P.. Diabetes mellitus: its differentiation into insulin-sensitive and insulin-insensitive types. Lancet (1936.0) 227 127-130 21. Alebić M.Š., Bulum T., Stojanović N., Duvnjak L.. Definition of insulin resistance using the homeostasis model assessment (HOMA-IR) in IVF patients diagnosed with polycystic ovary syndrome (PCOS) according to the Rotterdam criteria. Endocrine (2014.0) 47 625-630. PMID: 24522614 22. Duo Y., Song S., Zhang Y., Qiao X., Xu J., Zhang J.. Predictability of HOMA-IR for gestational diabetes mellitus in early pregnancy based on different first trimester BMI values. J Pers Med (2022.0) 13 60. PMID: 36675721 23. Song S., Zhang Y., Qiao X., Duo Y., Xu J., Peng Z.. HOMA-IR as a risk factor of gestational diabetes mellitus and a novel simple surrogate index in early pregnancy. Int J Gynaecol Obstet Organ Int Fed Gynaecol Obstet (2022.0) 157 694-701 24. Maebashi M., Makino Y., Furukawa Y., Ohinata K., Kimura S., Sato T.. Therapeutic evaluation of the effect of biotin on hyperglycemia in patients with non-insulin dependent diabetes mellitus. J Clin Biochem Nutr (1993.0) 14 211-218 25. 25Zhang Y., Ding Y., Fan Y., Xu Y., Lu Y., Zhai L., et al. Influence of biotin intervention on glycemic control and lipid profile in patients with type 2 diabetes mellitus: A systematic review and meta-analysis. Front Nutr [Internet]. 2022 [cited 2022 Nov 24];9. Available from: https://www.frontiersin.org/articles/10.3389/fnut.2022.1046800. 26. Reddi A., DeAngelis B., Frank O., Lasker N., Baker H.. Biotin supplementation improves glucose and insulin tolerances in genetically diabetic KK mice. Life Sci (1988.0) 42 1323-1330. PMID: 3280936
--- title: Umbilical cord mesenchymal stromal cells transplantation delays the onset of hyperglycemia in the RIP-B7.1 mouse model of experimental autoimmune diabetes through multiple immunosuppressive and anti-inflammatory responses authors: - C. C. Lachaud - N. Cobo-Vuilleumier - E. Fuente-Martin - I. Diaz - E. Andreu - G. M. Cahuana - J. R. Tejedo - A. Hmadcha - B. R. Gauthier - B. Soria journal: Frontiers in Cell and Developmental Biology year: 2023 pmcid: PMC9976335 doi: 10.3389/fcell.2023.1089817 license: CC BY 4.0 --- # Umbilical cord mesenchymal stromal cells transplantation delays the onset of hyperglycemia in the RIP-B7.1 mouse model of experimental autoimmune diabetes through multiple immunosuppressive and anti-inflammatory responses ## Abstract Type 1 diabetes mellitus (T1DM) is an autoimmune disorder specifically targeting pancreatic islet beta cells. Despite many efforts focused on identifying new therapies able to counteract this autoimmune attack and/or stimulate beta cells regeneration, TD1M remains without effective clinical treatments providing no clear advantages over the conventional treatment with insulin. We previously postulated that both the inflammatory and immune responses and beta cell survival/regeneration must be simultaneously targeted to blunt the progression of disease. Umbilical cord-derived mesenchymal stromal cells (UC-MSC) exhibit anti-inflammatory, trophic, immunomodulatory and regenerative properties and have shown some beneficial yet controversial effects in clinical trials for T1DM. In order to clarify conflicting results, we herein dissected the cellular and molecular events derived from UC-MSC intraperitoneal administration (i.p.) in the RIP-B7.1 mouse model of experimental autoimmune diabetes. Intraperitoneal (i.p.) transplantation of heterologous mouse UC-MSC delayed the onset of diabetes in RIP-B7.1 mice. Importantly, UC-MSC i. p. transplantation led to a strong peritoneal recruitment of myeloid-derived suppressor cells (MDSC) followed by multiple T-, B- and myeloid cells immunosuppressive responses in peritoneal fluid cells, spleen, pancreatic lymph nodes and the pancreas, which displayed significantly reduced insulitis and pancreatic infiltration of T and B Cells and pro-inflammatory macrophages. Altogether, these results suggest that UC-MSC i. p. transplantation can block or delay the development of hyperglycemia through suppression of inflammation and the immune attack. ## Introduction Type I diabetes Mellitus (T1DM) is an autoimmune disorder targeting the destruction of pancreatic insulin-producing beta cells, ultimately leading to hyperglycemia and the requirement of insulin therapy for survival. Latest findings indicate that T1DM originates from an early low chronic inflammation and disease of beta cells followed by the collapse of immune tolerance to pancreatic beta cell self-antigens such as insulin in individuals, a process which is facilitated by genetic predisposition and environmental factors (Pociot and Lernmark 2016; Antonela et al. 2017; Jerram et al. 2017; Norris et al. 2020; Roep et al. 2021; von Scholten et al. 2021). Although autoreactive CD45+, CD8+, CD3+ and CD4+ T Cells are the main cellular effectors of beta cells destruction (Willcox et al. 2009; Sarikonda et al. 2014), increasing evidence also demonstrated the critical participation of other immune cell types, mainly CD20+ B lymphocytes and pro-inflammatory CD68+ macrophages in the initiation and progression of the disease (Gaglia et al. 2011; Morgan et al. 2014; Campbell-Thompson et al. 2016; Christie 2016). Despite advances in our understanding of T1DM progression that has highlighted a relevant number of targets that upon modulation could either prevent or revert hyperglycemia in the Non-Obese Diabetic (NOD) mouse model of T1DM, translation to human patients has been mitigated by poor clinical trial outcomes (Thayer et al. 2010; Warshauer et al. 2020; von Scholten et al. 2021). As such, trials using anti-CD3 or anti-CD20 monoclonal antibodies to inhibit the development and mobilization of either T- or B- cells, led to a mild and transient improvement in C-peptide responses but failed to impede the recurrence of β-cell autoimmunity (Bluestone et al. 2010; Aronson et al. 2014; Vudattu and Herold 2014). A more recent clinical trial using Verapamil, a calcium channel blocker that improves beta cell survival was also shown to improve C-peptide response after a mixed meal highlighting the importance of targeting the metabolic damage to the beta cell mass in addition to the immune cells (Ovalle et al. 2018). The shortfalls of these “targeted monotherapies” have been a matter of debate and challenge the strategy of whether blocking a specific immune cell type or molecular mechanism is sufficient enough to attenuate the ongoing autoimmune attack and preserve the beta cell mass in T1DM (Cobo-Vuilleumier et al. 2018; Cobo-Vuilleumier and Gauthier 2020). Advanced therapies that simultaneously target multiple key immune events while establishing a local pancreatic anti-inflammatory and regenerative milieu have come into the limelight as a powerful alternative approach for the treatment of T1DM (Canibano-Hernandez et al. 2018; Warshauer et al. 2020). As such, cell-based immunotherapies using ex vivo expanded autologous tolerogenic dendritic cells or regulatory T Cells (Tregs) were shown to convey pleiotropic positive effects in the treatment of different human autoimmune disorders, including T1DM (Marek-Trzonkowska et al. 2012; Phillips et al. 2017; Stojanovic et al. 2017). Nonetheless, the isolation, differentiation and expansion of such immune cells are tedious and raise concerns on the feasibility of this approach for therapeutics. Alternatively, mesenchymal stromal/stem cells (MSC) have gained increasing interest due to their immunomodulatory and regenerative properties as well as ease of isolation/purification from various adult tissues (Franquesa et al. 2012; Shen et al. 2021). As of 2022, approximately 1500 clinical trials using MSC were registered at www.clinical trials.gov for indications ranging from neurological disease to diabetes. Of particular appeal are umbilical cord MSC (UC-MSC) that, in addition to possess immunosuppressive and anti-inflammatory properties, also display low immunogenicity offering the prospect of allogeneic transplantation (Can et al. 2017). To date, several clinical trials using UC-MSC infusion into newly onset T1DM patients have been yet completed but generated conflicting outcomes. On one hand, two independent studies reported no improvement in beta cell mass or C-peptide preservation whereas Tregs were increased in one study but not in the other subsequent to a 1 and 2 year follow-up of a single UC-MSC infusion (Haller et al. 2011; Giannopoulou et al. 2014). On the other hand, two other studies claimed that both C-peptide and HbA1c levels were improved after UC-MSC infusion resulting in reduced daily insulin dosage for patients at more than 1 year post-infusion (Hu et al. 2013; Lu et al. 2021). A common denominator to all studies was the absence of adverse side effects related to infusion (Haller et al. 2011; Hu et al. 2013; Giannopoulou et al. 2014; Lu et al. 2021). Given the high yield of healthy MSC that can be harvested from the umbilical cord combined with their immunomodulatory and regenerative properties, we sought to dissect the cellular and molecular events derived from UC-MSC intraperitoneal (i.p.) transplantation in the RIP-B7.1 mouse model of experimental autoimmune diabetes (EAD) in order to resolve pending discrepancies hindering further clinical studies. Of special relevance, transgene expression of the co-stimulatory B7.1 molecule in β-cells has shown to represent a highly reproducible EAD induction model (Rajasalu et al. 2004; Rajasalu et al. 2010; Cobo-Vuilleumier et al. 2018), by inducing a rapid and consistent insulitis mimicking closely the aggressive immune attack occurring in younger children with type 1 diabetes (Craig et al. 2019). Overall, we show that a single i. p. injection of UC-MSC could delay the onset of insulitis and hyperglycemia in RIP-B7.1 mice. Mechanistically, we show that UC-MSC i. p. transplantation drastically modified the composition and phenotype of peritoneal fluid cells, by mobilizing myeloid-derived suppressor cells (MDSC), T Cells and regulatory T Cells (Tregs). Interestingly, similar regulatory cells responses were further detected in diverse degrees into the spleen and pancreatic lymph nodes, in a timely orchestrated process correlating ultimately with a significantly lower pancreatic infiltration of T and B Cells as well as pro-inflammatory macrophages. ## Phenotypic characterization of umbilical cord mesenchymal stromal cells Consistent with mesenchymal stromal/stem cells (MSC) characteristics, umbilical cord MSC (UC-MSC) derived from E17 pregnant FVB mice displayed the typical fibroblastic morphology of MSC and accordingly expressed pericytes (α-SMA, desmin, PDGR-β and NG2) and mouse MSC (CD29, Sca-1 and CD44) markers (Supplementary Figures S1A–C). Consistent with their stemness properties, UC-MSC were also shown to possess mesodermal differentiation capacity as evidenced by their ability to acquire adipocytes, chondrocytes and osteocytes characteristics upon culture into specific lineage inductive media (Supplementary Figures S1D–F). ## Intraperitoneal UC-MSC transplantation delays EAD onset, reduces insulitis and normalizes the plasmatic cytokines profile in immunized-RIP-B7.1 mice Increasing amounts of UC-MSC were intraperitoneally (i.p.) transplanted into either immunized (IMM) or not (CT) RIP-B7.1 mice and glycemia was monitored up to 8-weeks post treatment (Supplementary Figure S2). A single i. p. dose of 5 × 105 UC-MSC provided the most effective reduction of hyperglycemia as compared to either a single or two doses of 2 × 105 UC-MSC (Supplementary Figure S2A). Long-term follow-up experiments indicated that a single i. p. dose of 5 × 105 UC-MSC provided a delay of approximately 7 weeks in the EAD onset of IMM-RIP-B7.1 mice, after when $100\%$ of them became hyperglycemic at 13-14 weeks post-transplantation (Figures 1A, B). Control RIPB7.1 mice (not immunized) which were transplanted with 5 × 105 UC-MSC didn´t show any obvious alterations in terms of overall health status, glycemia (Figure 1A) or weight (data not shown) by comparison to their non-transplanted counterparts. FIGURE 1: UC-MSC transplantation delay the onset of experimental autoimmune diabetes (EAD) in immunized RIP-B7.1 mice, reducing insulitis and normalizing plasmatic cytokine profile during the protective phase. (A) Summary measurements of non-fasting blood glucose in control (CT) and immunized (IMM) RIP-B7.1 mice transplanted with vehicle or 5 × × 105 UC-MSC, recorded until 14 weeks post transplantation. (B) Summary percentages of diabetic RIP-B7.1 mice (≥250 mg/dL) into different experimental groups. (A, B) Values are means ± s. e.m. of n = 18 mice for CT and CT+5 × 105 UC-MSC, n = 28 for IMN and n = 31 for IMM+5 × 105 UC-MSC. (C) Representative hematoxylin and eosin (H&E) histological staining of pancreatic sections from CT and IMM-RIP-B7.1 mice at 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC (40X magnification). (D) Insulitis scores as a grade of 0–4 according to percentage of infiltrated islets. (n = 4 mice per group). (E) Pancreatic sections from CT and IMM RIP-B7.1 mice at 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC were co-immunostained for insulin (green) and glucagon (red). Nuclei are stained with DAPI (blue). Representative single-channel fluorescence images are shown individually and merged. 40X magnification. (F) Plasmatic cytokines profile from CT and IMM RIP-B7.1 mice at 7 weeks after vehicle or 5 × 105 UC-MSC transplantation. Values are mean ± SE cytokine concentration [pg/mL]. CT, n = 8; CT + UC-MSC, n = 8; IMM, n = 10; IMM + UC-MSC, n = 10. (D, F) , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. To understand how UC-MSC transplantation protected temporally IMM-RIP-B7.1 mice against the programmed EAD, we first analyzed the presence of immune cells infiltration within islets (insulitis) at 7 days and 7 weeks post-transplantation (Figure 1C). A single dose of 5 × 105 UC-MSC produced the best protection against the development of the highest grades of insulitis in IMM-RIP-B7.1 mice (Figure 1D), and preserved islets integrity and normal α- and β-cell content (Figure 1E). Correlating with these results, IMM-RIP-B7.1 mice transplanted with UC-MSC had reduced plasmatic concentrations of the pro-inflammatory cytokines IL-6, IFN-γ and TNF-α as compared to vehicle-treated IMM-RIP-B7.1 mice (Figure 1F). ## Biodistribution of transplanted UC-MSC In vivo imaging system (IVIS) was used to track Dir+-labelled UC-MSC after i. p. transplantation (Figures 2A, B). Dir+ signals rapidly disseminated through the peritoneal cavity and incorporated to the different visceral organs tested, including the mesentery, gonadal fat depots and liver at 24 h post-transplantation. Strong Dir+ signals detected in the stroma of fresh liver section clearly indicate an accumulation of phagocytized Dir+ UC-MSC-derived residues (Figure 2B). Analysis of the spleen and mesentery suggested that Dir+ UC-MSC had not migrated massively into their stroma since the majority of Dir+ signal was concentrated into their outermost mesothelium (Figure 2C). In line with this finding, pancreatic lymph nodes (PLN) were also found to lack detectable internal Dir+ signal, which was instead detected onto adipose tissue surrounding PLN (Figure 2C). Finally, some large Dir+ UC-MSC aggregates were found adhered onto the mesenteric mesothelium in close proximity to the pancreas in most UC-MSC transplanted RIP-B7.1 mice, which supports a local action of UC-MSC to the target pancreas (Figures 2D, E). However, most of the Dir+ UC-MSC aggregates initially detected as adhered onto the mesenteric surface at day 1 post-transplantation were by contrast undetectable after 7 days when only traces of Dir+ signals inside mesothelium cells which could correspond to macrophages (Figures 2E, F). Finally, at 7 days post-transplantation of CFSE+ UC-MSC, fluorescent cells consistent with UC-MSC or phagocytes bearing UC-MSC-derived CFSE+ residues were detected inside CD54hi milky spots in the mesenteric submesothelium (Figure 2G). FIGURE 2: Transplanted UC-MSC rapidly spread inside the peritoneal cavity and adhere to the mesothelium surface of several peritoneal organs. (A) Shows wide dissemination of 5 × 105 Dir+ UC-MSC through the peritoneal cavity at 6 h post-transplantation (arrow points to the injection site). (B) Shows comparative Dir+ fluorescence levels into mesentery, gonadal adipose tissue and liver section from vehicle or Dir+ UC-MSC transplanted IMM mice, at 24 h post-transplantation. (C) Upper and middle images show comparative Dir+ fluorescence into the separated splenic and mesenteric mesothelium (arrows), respectively. Lower image shows Dir+ fluorescence signal in pancreatic lymph nodes (PLN), which was externally restricted to peripheral adipose tissue (asterisks), but not to PLN (ellipse). (D) Whole-mount visualization of connected mesentery and pancreas at 24 h post-transplantation of Dir+ UC-MSC and after washing in PBS allows the visualization of small to large Dir+ UC-MSC aggregates adhered on their surface. (E) En face bright field picture of mesentery and pancreas surface at 24 h post-transplantation of Dir+ UC-MSC, showing small to large blue Dir+ UC-MSC aggregates (ellipses) adhered on top of the mesenteric adipose tissue mesothelium but not on top of the pancreatic mesothelium. (F) En face bright field picture of mesentery and pancreas surface at 7 days post-transplantation of Dir+ UC-MSC. At that time post-transplantation, Dir+ aggregates were not observable on top of the mesentery surface (left image). By contrast, many areas of the mesenteric adipose tissue displayed residual Dir+ dots in surface cells (enlarged spot). (G) En face immunofluorescence picture of the mesentery surface showing Dir+ cells inside a milky spot (depicted by CD54-PEhigh lymphocytes) which should likely correspond to macrophages with internalized Dir+ residues. ## UC-MSC are massively targeted by peritoneal macrophages To start deciphering the immune protection mechanisms triggered by UC-MSC i. p. transplantation, we first analyzed peritoneal fluid cells (PFC) from IMM mice transplanted or not with fluorescently labelled UC-MSC (Supplementary Figure S3). Peritoneal drainages collected at 24 h post-transplantation with CFSE+ UC-MSC helped the visualization of numerous small to large cellular aggregates, consisting of a core of CFSE+/CD11bneg UC-MSC surrounded by several layers of CFSE−/CD11bhi myeloid cells (Figure 3A). Examination of PFC at day 1 post-transplantation of fluorescently-labelled UC-MSC, indicated that part of CD11b(hi) large peritoneal macrophages presented fluorescent residues in their cytoplasm, indicating they are actively involed in UC-MSC phagocytosis (Figure 3B). Most UC-MSC/myeloid cells aggregates were collected as free-floating aggregates within peritoneal lavages, while by contrast only a minor portion of them were found firmly attached on top of mesenteric adipose depots (Figures 2D, E). FIGURE 3: Intraperitoneally transplanted UC-MSC are targeted by peritoneal macrophages in a process accompanied by a strong mobilization of neutrophils and myeloid-derived suppressor cells in the peritoneal fluid. (A) Left image, shows low 2X magnification fluorescence picture of a peritoneal drainage collected at 24 h post-transplantation of CFSE+ UC-MSC and showing presence of many small to very large CFSE+ UC-MSC clusters. Scale bar is 500 µm. Middle image, merged fluorescent and light photograph of CFSE+ UC-MSC aggregates showing a core of CFSE+ UC-MSC surrounded by several layers of CSFEneg external cells (arrow). Right image, CD11b-PE staining of peritoneal lavage cells helped to visualize how CFSE+ UC-MSC aggregates had an external layer of CD11bhigh/CSFEneg peritoneal macrophages. Scale bar is 100 µm. (B) Shows CD11b-PE stained peritoneal fluid cells (PFC) collected by from IMM-RIP-B7.1 mice at 24 h post-transplantation of CSFE+ UC-MSC, or UC-MSC bearing phagocytosed Fluospheres. Note how, UC-MSC-derived fluorescent residues are by majority internalized into large CD11bhigh peritoneal macrophages (arrows) but not into small CD11blow expressing myeloid cells (ellipse). (C) CD11b-PE staining of PFC collected at 24 h from immunized mice transplanted with vehicle (IMM) or UC-MSC (IMM + MSC). Note the presence of numerous CD11blow expressing myeloid cells (ellipse) among PFC from IMM + MSC mice. Their smaller size is distinguishable from CD11bhigh large peritoneal macrophages (arrow). (D, E) Flow cytometric quantification of CD11bmed/MHC-II-/med myeloid-derived suppressor cells within PFC. (D) CD11b and MHC-II (I/A-I/E) expression analysis onto PFC from CT or IMM RIP-B7.1 mice 24 h post-transplantation. UC-MSC transplantation induced mobilization of two myeloid subpopulations with distinct SSC-H/FSC-H characteristics and MHC-II/CD11b costaining patterns. Upper, SSC-H/FSC-H dot plots of gated PFC (blue line area). Note how cells display distinct coloration according to their CD11b and MCH-II expression patterns clustered into distinct regions (lower). PFC from CT + MSC and IMM + MSC mice contain display increased CD11bmed/MHC-IImed (R1) and CD11bmed/MHC-IIneg-low (R2) myeloid cells subsets, which are delimited by a dashed ellipse into SSC-H/FSC-H dot plots. (E) Quantification of CD11bmed/MHC-II-/med myeloid cells (R1+R2) percentages. (F) Quantification of CD11bmed/Ly6C− myeloid cells and CD11bmed/Ly6C+ myeloid-derived suppressor cells (MDSC) percentages. (E, F) Analysis performed onto PFC from CT and IMM RIP-B7.1 mice transplanted with vehicle or 5 × 105 UC-MSC, at time of 1 day, 7 days and 7 weeks post-transplantation. (G) Q-PCR analysis of macrophage inhibitory factor (MIF), interleukin-10 (IL-10) and arginase I (ARG1) mRNA expression in PFC collected at 1 and 7 days after UC-MSC transplantation. (E–G) Values are mean ± s. e.m of n ≥ 5 mice for each experimental group. Results show relative mRNA expression to CT mice (value set as 1). , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. NS is for not significant differences. ## UC-MSC transplantation triggers a rapid and massive recruitment of myeloid-derived suppressor cells (MDSC) in the peritoneal fluid Analysis of PFC collected from RIP-B7.1 mice at day 1 post-transplantation evidenced how many of them corresponded to small CD11bmed myeloid cells (Figure 3C), clearly distinguishable from large rounded CD11bhi peritoneal macrophages (Cassado Ados et al. 2015). Correlating with this finding, PFC from both CT- and IMM-RIP-B7.1 mice transplanted with UC-MSC were found to contain approximately $30\%$ of CD11bmed myeloid cells, consisting mostly of [CD11bmed/MHC-IIlow] cells and in lower extent of [CD11bmed/MHC-IIneg] cells (Figures 3D, E). Further analysis indicated that these infiltrated CD11bmed myeloid cells were also composed by both CD11bmed/Ly6C+ and CD11bmed/Ly6C− myeloid cells (Figure 3F), being Ly6C expressed onto mouse MDSC (Saiwai et al. 2013). A more detailed analysis of these Ly6C+ MDSC indicated that they corresponded to granulocytic [CD11bmed/LY6C+/SCC-Hhi] and monocytic [CD11bmed/LY6Chi/SCC-Hmed] MDSC (data not shown). Quantitative real-time PCR (qPCR) analyses revealed that PFC from transplanted IMM-RIP-B7.1 mice had significantly higher levels of the enzyme arginase I (ARG1) as compared to either CT or IMM mice (Figure 3G), which is highly expressed in MDSC (Gabrilovich and Nagaraj 2009) and acts as a major immunosuppression mechanism of T Cells responses through depletion of L-arginine (Bronte and Zanovello 2005; Munder 2009). Conversely, the proinflammatory cytokine macrophage inhibitory factor (MIF) which expression was upregulated in PFC from IMM-RIP-B7.1 mice was by contrast downregulated in IMM-RIP-B7.1 mice transplanted with UC-MSC (Figure 3G). ## Initial acute peritoneal immune response induced by UC-MSC transplantation is followed by a massive activation of peritoneal macrophages and mobilization of T cells in the peritoneal fluid Analysis of PFC collected from the different experimental groups could evidence how resident peritoneal macrophages initially detected as CD11bhi/MHC-IIlow cells at day 1 post-transplantation had further shifted their immunophenotype to activated CD11bhi/MHC-IIhigh macrophages at 7 days post-transplantation (Figures 4A, B). Relevantly, PFC collected in mice at 7 days post-transplantation were also found to contain higher percentages of CD4+ and CD8+ T Cells (Figures 4C, D), suggesting that an active interaction between peritoneal macrophages and T Cells was possibly occurring at that time post-transplantation. Interestingly, percentages of activated CD4+/CD25+ and CD8+/CD25+ T Cells were also increased at 7 days post-transplantation (Figure 4E) and correlated with a higher expression levels by PFC of transcripts encoding interleukin-4 (IL-4) and FoxP3 (Figure 4F), which could indicate that part of the increase in activated CD4+/CD25+ T Cells correspond to regulatory T Cells (Tregs). FIGURE 4: UC-MSC transplantation induces in a second step a strong activation of peritoneal macrophages and mobilization of T Cells and regulatory T Cells. (A, B) Peritoneal macrophages are massively activated at 7 days post-transplantation, but return to a lower activation state after 7 weeks post-transplantation. (A) Representative CD11b and MHC-II expression dot plots of PFC from 7 days transplanted CT or IMM RIP-B7.1 mice with vehicle or 5 × 105 UC-MSC. Note how CD11bhi/MHC-IIneg-low peritoneal macrophages (R3 region) in CT and IMM mice shifted to an activated CD11bhi/MHC-IIhi phenotype (R4 region) in CT + MSC and IMM + MSC mice. (B) Quantification of CD11bhi/MHC-IIhi activated macrophages percentages within PFC. (C–F) Peritoneal fluid contains increased T Cells and activated T Cells percentages and regulatory T Cells markers at 7 days post-transplantation. (C) Representative CD4 and CD25 expression dot plots and CD4-PE stainings of PFC from IMM and IMM + MSC mice at 7 days post-transplantation. (D) Quantification of total CD4+ and CD8+ T Cells percentages within PFC. (E) Flow cytometric quantification of CD4+/CD25+ and CD8+/CD25+ activated T Cells percentages within PFC. (B, D, E) PFC were collected from CT and IMM RIP-B7.1 mice, at 24 h, 7 days and 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC. (F) Q-PCR analysis of CD4, interleukin-4 (IL-4) and forkhead box P3 (FoxP3) mRNA expression in PFC collected from CT and immunized RIP-B7.1 mice transplanted with vehicle (IMM) or 5 × 105 UC-MSC (IMM + MSC) at day 1 and 7 post-transplantation. Results show relative mRNA expression to CT mice (value set as 1). (B, D, E F) Values are mean ± s. e.m of n ≥ 5 mice for each experimental group. , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. NS is for not significant differences. ## Spleens of UC-MSC transplanted mice display increased MDSC percentages and anti-inflammatory/immunosuppressive markers As previously observed in the peritoneal fluid, mice analyzed at 7 days post-transplantation also displayed increased percentages of CD11b+/Ly6C+ MDSC in their spleen (Figure 5A), consisting of granulocytic [CD11b+/Ly6Clow/med/SSC-Hhigh] and monocytic [CD11b+/Ly6Chi/SSC-Hmed] MDSC (Supplementary Figure S3). Correlating with these findings, splenocytes from UC-MSC transplanted IMM-RIP-B7.1 mice also up-expressed interleukin-10 (IL-10) and ARG1, two genes which are highly expressed by MDCS (Figure 5B). Relevantly, and as similarly observed in PFC, splenocytes from IMM-RIP-B7.1 mice also displayed significantly higher expression of MIF as compared to CT mice (Figure 5B), confirming thus that MIF up-expression likely plays a relevant role in the EAD of IMM-RIP-B7.1 mice. However, and by contrast to results initially observed in PFC (Figure 3G), no significant differences for MIF expression were observed in response to UC-MSC transplantation in IMM-RIP-B7.1 mice, even if the results at days 1 and 7 post-transplantation show a pattern indicating a possible reduction in its expression (Fig. 5 b). FIGURE 5: UC-MSC transplantation induces splenic anti-inflammatory and immunosuppressive responses in immunized RIP-B7.1 mice. (A) Quantification of splenic CD11b+/Ly6C− myeloid cells and CD11b+/Ly6C+ myeloid-derived suppressor cells (MDSC). Note how both CT and IMM mice at 7 days post-transplantation display increased splenic percentages of CD11b+/Ly6C+ MDSC. (B) Q-PCR quantification of MIF, IL-10 and ARG1 mRNA expression. (C–E) Splenocytes from UC-MSC transplanted IMM mice display increased percentages of total and activated CD4+ T Cells and higher expression of the associated regulatory T Cells markers IL-4 and forkhead box P3 (FoxP3), compared to CT mice. (C) Quantification of total splenic CD4+ and CD8+ T Cells percentages. (D) Quantification of total splenic activated CD4+/CD25+ and CD8+/CD25+ T Cells percentages. (A, C, D) Splenocytes were collected from CT and IMM RIP-B7.1 mice, at 24 h, 7 days and 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC. (E) Q-PCR quantification of interleukin-4 (IL-4) and FoxP3 mRNA expression. (B, E) Splenocytes collected from CT and immunized RIP-B7.1 mice transplanted with vehicle (IMM) or 5 × 105 UC-MSC (IMM + MSC) at day 1 and 7 post-transplantation. Results show relative mRNA expression to CT mice (value set as 1). (A–E) Values are mean ± s. e.m of n ≥ 5 mice for each experimental group. , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. NS is for not significant differences. ## UC-MSC transplantation transiently increases activated CD4+/CD25+ T cells and tregs markers in the spleen of immunized-RIP-B7.1 mice We next assessed whether UC-MSC transplantation alters splenic T Cells percentages (Figures 5C, D). Although no drastic changes were observed over time post-transplantation, UC-MSC transplanted IMM-RIP-B7.1 mice displayed a transient increase in total CD4+ T Cells and CD4+/CD25+ activated T Cells at day 1 post-transplantation (Figures 5C, D). Accordingly, splenocytes of UC-MSC transplanted IMM-RIP-B7.1 mice had higher transcript levels of the Treg markers IL-4 and FoxP3 (Figure 5E). ## UC-MSC transplantation moderately increases MDSC and activated T cells in pancreatic lymph nodes Although FC analysis indicated a transient increase of CD11b+/Ly6C+ MDSC in pancreatic lymph nodes (PLN) from IMM-RIP-B7.1 mice at day 1 post-transplantation (Figure 6A), no obvious differences for MIF, IL-10 or ARG1 mRNA expression were observed between experimental groups (Figure 6B). Furthermore, and as previously observed in splenocytes, PLN cells from day 1 transplanted IMM-RIP-B7.1 mice also displayed higher percentages of both CD4+ T Cells and CD4+/CD25+ activated T Cells when compared to CT mice (Figures 6C, D), but however did not show significant increases in IL-4 and Foxp3 mRNA expression compared to non-transplanted IMM-RIP-B7.1 mice (Figure 6E). FIGURE 6: Immunized RIP-B7.1 transplanted with UC-MSC display transiently increased MDSC, activated T Cells and regulatory T Cells markers in pancreatic lymph nodes. (A) Quantification of CD11b+/Ly6C− and CD11b+/Ly6C+ MDSC within pancreatic lymph nodes (PLN). Note how PLN from day 1-transplanted IMM mice display increased percentage of CD11b+/Ly6C+ MDSC. (B) Q-PCR analysis of MIF, IL-10 and ARG1 mRNA expression in PLN. Results show relative mRNA expression to CT mice (value set as 1). (C–E) PLN from day 1 UC-MSC transplanted IMM mice transiently display higher percentages of total and activated CD4+ T Cells, correlating with higher expression of the regulatory T Cells markers forkhead box P3 (FoxP3). (C) Summary flow cytometric quantification of total CD4+ and CD8+ T Cells in PLN. (D) Quantification of activated CD4+/CD25+ and CD8+/CD25+ T Cells in PLN. (A, C, D) Splenocytes were collected from CT and IMM RIP-B7.1 mice, at 24 h, 7 days and 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC. (E) Q-PCR analysis of interleukin-4 (IL-4) and FoxP3 mRNA expression in PLN. (B, E) PLN were collected from CT and immunized RIP-B7.1 mice transplanted with vehicle (IMM) or 5 × 105 UC-MSC (IMM + MSC) at day 1 and 7 post-transplantation. (A–D) Values are mean ± s. e.m of n ≥ 5 mice for each experimental group. , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. NS is for not significant differences. ## UC-MSC transplantation induces a transient increase of activated B cells in the spleen and pancreatic lymph nodes of immunized RIP-B7.1 mice Flow cytometry analysis revealed higher percentages of activated CD19+/CD25+ B Cells in the spleen and pancreatic lymph nodes of IMM-RIP-B7.1 mice, at 1 day after UC-MSC transplantation (Supplementary Figure S4F, G). Percentages of activated B Cells percentages among PFCs were in turn decreased in both CT- and IMM-RIP-B7.1 mice, at 7 days post-transplantation (Supplementary Figure S4E). ## UC-MSC transplantation reduces pancreatic infiltration of T and B cells and pro-inflammatory macrophages in immunized RIP-B7.1 mice We next analyzed leukocytes infiltration in the pancreas at day 1 and 7 as well as at 7 weeks post-transplantation and found that as early as 15 days post-immunization (corresponding to 7 days post-transplantation), IMM-RIP-B7.1 mice already displayed a significant increase of both CD4+ T- and CD19+ B Cells among pancreatic stromal cells (Supplementary Figure S5). At 8 weeks post-immunization (7 weeks post-transplantation), IMM-RIP-B7.1 mice which already had developed severe insulitis and hyperglycemia (Figure 1) displayed the highest increase of CD11b+ myeloid cells, CD4+ and CD8+ T Cells and CD19+ B Cells (Supplementary Figure S5). Establishment of total cell counts per pancreas at that time post-transplantation (Figure 7A) could evidence how UC-MSC-transplanted IMM-RIP-B7.1 mice displayed significantly lower numbers of infiltrated CD4+ and CD8+ T Cells ($p \leq 0.05$), CD19+ B Cells ($p \leq 0.05$) and CD11b+ myeloid cells (p ≤ 0.03) as compared to vehicle-treated IMM-RIP-B7.1 mice (Figure 7B). FIGURE 7: UC-MSC transplantation significantly reduces pancreatic leukocytes infiltration in IMM-RIP-B7.1 mice. (A) Total pancreatic stromal cells counts obtained from experimental groups at 7 weeks after UC-MSC transplantation. (B) Quantification of total numbers of CD4+ and CD8+ T Cells, CD19+ B Cells and CD11b+ myeloid cells per pancreas in CT or IMM RIP-B7.1 mice at 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC. (C) Quantification of [CD11b+;F4/80+;CD206-] and [CD11b+;F4/80+;CD206+] pancreatic macrophages subsets. (D) Quantification of [CD11b+;Ly6C−;MHC-II-] and [CD11b+;Ly6C−;MHC-II+] pancreatic myeloid cells subsets. (E) Quantification of [CD11b+;MHC-II+;GR1+] and [CD11b+;MHC-II-;GR1+] pancreatic myeloid cells subsets. (A–E) Analysis performed onto pancreatic stromal cells were collected from CT- or IMM-RIP-B7.1 mice at 7 weeks post-transplantation with vehicle or 5 × 105 UC-MSC. (F) Q-PCR analysis of MIF, IL-10 and ARG1 mRNA expression in pancreatic stromal cells from CT and immunized RIP-B7.1 mice transplanted with vehicle (IMM) or 5 × 105 UC-MSC (IMM + MSC) at day 1 and 7 post-transplantation. Results show relative mRNA expression to CT mice (value set as 1). (B–F) Values are mean ± s. e.m of n ≥ 5 mice for each experimental group. , p ≤ 0.05; , p ≤ 0.03; , p ≤ 0.01, one-way ANOVA. NS is for not significant differences. ## UC-MSC transplantation significantly reduces pancreatic infiltration of pro-inflammatory macrophages and MDSC in hyperglycemic immunized-RIP-B7.1 mice The immunophenotype of CD11b+ myeloid cells infiltrated in pancreases of hyperglycemic IMM-RIP-B7.1 mice was further characterized by analyzing co-expression of the myeloid-related markers F$\frac{4}{80}$, CD206, MHC-II, Ly6C and Gr1. At 8 weeks post-immunization (7 weeks post-transplantation), approximately $80\%$ of infiltrated myeloid cells into pancreases of IMM-RIP-B7.1 mice were identified as [CD11b+/F$\frac{4}{80}$+/CD206-] cells (Figure 7C), which are consistent with M1 pro-inflammatory macrophages (Antonios et al. 2013; Cassado Ados et al. 2015). Additionally, and in a similar way, near $90\%$ of pancreatic infiltrated myeloid cells corresponded to [CD11b+/Ly6C−/MHC-II+] cells (Figure 7D), an immunophenotype which is mostly consistent with activated macrophages (Gordon and Taylor 2005), but that also identifies dendritic cells subsets (Welzen-Coppens et al. 2012). Interestingly, islet-resident macrophages in NOD mice were also shown to be identified as MHC-IIhi cells (Ferris et al. 2017). In contrast, UC-MSC-transplanted IMM-RIP-B7.1 mice displayed a significantly reduced percentages of both [CD11b+/F$\frac{4}{80}$+/CD206-] and [CD11b+/Ly6C−/MHC-II+] inflammatory macrophages, which is in agreement with their reduced hyperglycemia as compared to non-transplanted IMM mice (Figures 7B, D). Of particular interest is also our finding that nearly $10\%$ of myeloid cells infiltrated into pancreases of IMM-RIP-B7.1 mice were characterized as [CD11b+/MHC-IIneg/low/Gr1+] cells (Figure 7E) an immunophenotype which is consistent with MDSC (Gabrilovich and Nagaraj 2009). Interestingly, our results indicated that percentages of [CD11b+/MHC-IIneg/low/Gr1+] MDSC were also significantly reduced in UC-MSC transplanted IMM-RIP-B7.1 mice to levels closer to those of RIP-B7.1 control mice (Figure 7E). However, and in contrast with these results, pancreatic stromal cells from IMM-RIP-B7.1 mice did not express higher levels of the immunosuppressive enzyme ARG1 which was instead expressed at higher levels in UC-MSC transplanted IMM-RIP-B7.1 mice (Figure 7F). ## Discussion Long-term follow-up experiments indicated that a single i. p. transplantation of 5 × 105 UC-MSC in immunized RIP-B7.1 mice could efficiently delay the onset of EAD through induction of multiple anti-inflammatory and immunosuppressive responses into the peritoneal cavity, spleen and pancreatic lymph nodes, which ultimately correlated with lower plasmatic proinflammatory cytokines concentrations and a reduced pancreatic leukocytes infiltration. The i. p. cells transplantation approach has gained increasing interest in recent years, principally as local approach to treat peritoneal fibrosis or inflammatory or/and autoimmune diseases affecting visceral organs (Bertram et al. 1999; Li et al. 2009; Wakabayashi et al. 2014; El-Hossary et al. 2016). Interestingly, a study reported the successful attenuation of streptozotocin-induced diabetes via systemic transplantation of human UC-MSC, while in contrast no effectiveness was obtained through the i. p. transplantation route (El-Hossary et al. 2016). Contrary to these outcomes, our results suggest that the i. p. transplantation of heterologous UC-MSC is an effective transplantation approach for T1DM treatment. The strong differences between both experimental mouse diabetes models-chemical versus immune response-may explain divergent results. Indeed, MSC transplantation for T1DM treatment is based on their potent anti-inflammatory and/or immunosuppressive properties. Despite initial experimental transplantation studies with allogeneic MSC suggested they were endowed with strong immune privilege through induction of host tolerance (Barry et al. 2005), more recent studies however indicated that major histocompatibility complex (MHC)-mismatched MSC are relatively much more immunogenic than initially described, inducing significant cell-mediated and humoral immune responses in vivo (for review see (Berglund et al. 2017). In line with this evidence, we report here that FVB-derived UC-MSC (H2q MHC haplotype) transplanted within RIP-B7.1 mice (H2b MHC haplotype) were massively targeted by CD11bhigh peritoneal macrophages, leading to the formation of peritoneal UC-MSC/macrophages aggregates, which were eliminated from the peritoneal cavity through phagocytosis. Despite their rapid elimination from the host, transplanted UC-MSC however profoundly modified the cellular composition and inflammatory state of resident peritoneal fluid immune cells, by inducing a rapid and massive peritoneal mobilization of CD11bmed/Ly6C+/hi MDSC, which play a critical role in the resolution of acute inflammation (Saiwai et al. 2013; Li et al. 2021). Additionally and correlating with these findings, UC-MSC transplantation induced a significant upregulation of arginase I (ARG1) expression in PFC, an enzyme with potent immunosuppressive activity and which is highly expressed by murine MDSC and anti-inflammatory M2 type macrophages during the resolution phase of inflammation (Gabrilovich and Nagaraj 2009; Nagaraj et al. 2009; Rath et al. 2014). Importantly, we also show here that UC-MSC transplantation blunted the expression of the pro-inflammatory marker macrophage inhibitory factor (MIF) in peritoneal fluid cells (PFC), presumably in response to peritoneal MDSC mobilization and induction of the anti-inflammatory response. Although the role of MIF in T1DM development has already been reported (Stosic-Grujicic et al. 2008; Sanchez-Zamora et al. 2016; Korf et al. 2017), none of them have reported an upregulation of MIF in peritoneal immune cells during early diabetes induction. Herein, our finding of upregulated MIF expression in PFC from RIP-B7.1 mice at 8 days of immunization could indicate that peritoneal myeloid cells play an early role in the programmed autoimmune attack, which is in line with previous studies indicating a potential role for resident peritoneal immune cells populations such macrophages and B1-cells in the development of T1DM (Kendall et al. 2004; Silveira and Grey 2006; Emani et al. 2015). The coincidence of a strong activation of CD11bhi peritoneal macrophages and increased T Cells percentages and Tregs markers expression into PFC of RIP-B7.1 mice at 1 week post-transplantation represents the second temporal wave of immunoregulatory events observed in the peritoneal cavity after the initial target of UC-MSC by peritoneal macrophages and mobilization of MDSC in the peritoneal fluid. It is likely that the mobilization of Tregs within the peritoneal cavity is triggered by the anti-inflammatory/immunosuppressive milieu and T Cells chemokines produced by UC-MSC and MDSC. Mobilization of Tregs in the peritoneal fluid of UC-MSC transplanted IMM-RIP-B7.1 mice likely contributes to EAD attenuation. The strong upregulation of the antigen-presenting molecules MHC-II onto CD11bhi peritoneal macrophages at 1 week post-transplantation is consistent with their acquisition of a classical activation phenotype as occurring in thioglycollate-elicited or lipopolysaccharide (LPS)-induced peritoneal macrophages (Kim et al. 2015; Pavlou et al. 2017). Such phenomenon could likely indicate that peritoneal macrophages have transitioned into efferocytic or post-phagocytic macrophages presenting UC-MSC-derived antigens to peritoneal fluid T Cells. In support of such premise, MHC-IIhi activated peritoneal macrophages displayed a moderate loss of CD11b expression (data not shown), a phenomenon which has been previously reported in post-efferocytic pro-resolving macrophages (Korns et al. 2011; Schif-Zuck et al. 2011). We also show that UC-MSC transplantation via i. p. also induced anti-inflammatory and immunosuppressive responses into secondary lymphoid organs (SLOs) as evidenced by increased percentages of CD11b+/Ly6C+ MDSC and activated T Cells and Tregs markers in the spleen and pancreatic lymph nodes of transplanted IMM-RIP-B7.1 mice. Of particular interest, splenocytes from early immunized RIP-B7.1 mice were also found to upregulate MIF, evidencing that EAD development is associated with a splenic inflammatory process, a premise consistent with previous reports indicating a role for MIF expression in the development of T1DM and as critical regulator of innate immunity (Calandra and Roger 2003; Stosic-Grujicic et al. 2008; Sanchez-Zamora et al. 2016). According to their immunoregulatory activity, UC-MSC transplantation led to the upregulation by splenocytes of the anti-inflammatory cytokine IL10 and immunosuppressive enzyme ARG1 whereas the proinflammatory cytokine IFN-γ was downregulated, thus indicating a shift from a pro-to an anti-inflammatory environment. The UC-MSC-mediated upregulation of IL-4 observed in the spleen, is also consistent with the creation of a milieu favorable to the stimulation of activated B Cells and generation of Tregs cells and T helper 2 cells (Cameron et al. 1997; Guo and Rothstein 2013; Wynn 2015; Yang et al. 2017). Interestingly, the spleen and pancreatic lymph nodes of early UC-MSC transplanted mice display a transient increase in the number of activated CD19+/CD25+ B Cells, which is the subset of B Cells containing regulatory B Cells (Bregs), (Kessel et al. 2012; de Andres et al. 2014; Hong et al. 2019; Ben Nasr et al. 2021). Interestingly, different studies already indicated how MSC transplantation induced increased Bregs, which exert immunosuppression through secretion of IL-10 (Kleffel et al. 2015; Cho et al. 2017; Luk et al. 2017; Phillips et al. 2017; Phillips et al. 2019; Qu et al. 2021). Therefore, the observed high upregulation of IL-10 expression in the spleen of UC-MSC transplanted IMM RIP-B7.1 mice is consistent with a higher number of Bregs in response to UC-MSC transplantation. Interestingly, UC-MSC transplantation in IMM-RIP-B7.1 mice also exhibited an increase percentage of CD11b+/Ly6C+ MDSC and CD25+/CD4+ activated T Cells in pancreatic lymph nodes (PLN), which together with increased IL-4 and FoxP3 mRNA expression could indicate an expansion of Tregs. We also show that UC-MSC transplantation significantly reduced pancreatic infiltration of CD4+ and CD8+ T, CD19+ B and CD11b+ myeloid cells in IMM-RIP-B7.1 mice. Myeloid cells infiltrated in the pancreas of IMM-RIP-B7.1 mice were accordingly identified as [F$\frac{4}{80}$+ CD206-] M1 pro-inflammatory macrophages, which are a major cellular effector in insulitis (Korf et al. 2017). Of special interest, we also found that hyperglycemic IMM-RIP-B7.1 mice also displayed increased pancreatic infiltration of [CD11b+ MHC-II- Gr1+] MDSC, which is very striking due to their opposite role with M1 pro-inflammatory macrophages and also since a previous study reported decreased MDSC in islets of NOD mice (Fu et al. 2012). Our results are however supported by other studies showing increased MDSC in the peripheral blood and secondary lymphoid organs of NOD and STZ mice models as well as in T1DM and T2DM patients (for review see (Wang et al. 2021). A higher number of MDSC during insulitis development is likely a compensatory mechanism to suppress diabetogenic T Cell proliferation. Overall, our study provides evidence of multiple immunomodulary responses induced by UC-MSC in the RIP-B7.1 rodent model of T1DM, inducing a strong delay of their EAD. Whether the treatment with a second dose of UC-MSC would be enough to postpone the resurgence of T1DM is also an issue that would require additional studies. Our results therefore support the approach of the i. p. transplantation of allogenic UC-MSC for the treatment of early-diagnosed human T1DM patients. ## Isolation and culture of mouse UC-MSC UC-MSC were isolated from E17 old FVB (H2q) mice fetuses. Guidelines for the animal research protocols were established and approved by the Animal Experimentation and Ethics Committee of CABIMER (CEEA-CABIMER). Umbilical cords (UC) were sequentially trypsinized into PBS +$2\%$ bovine serum albumin (BSA) containing first $0.25\%$ trypsin (Gibco) to detach outermost epithelial lining cells. Partially digested UC were then digested and then 2 mg/mL type Ia collagenase (Sigma) to obtain UC-MSC. Both digestions were done in a water bath (37°C, 15 min) under agitation. Umbilical cords stromal cells were seeded at 20.000 cells/cm2 into 140 mm Petri dishes (Nunc) and cultured into a DMEM, low glucose, GlutaMAX™, pyruvate medium (GIBCO) supplemented with $5\%$ FBS, $1\%$ antibiotics, 1X MEM non-essential amino acids (Gibco), 1X insulin transferrin selenium (ITS), 100 µm 2-mercaptoethanol, 10 ng/mL bFGF and 10 ng/mL EGF in an incubator under low oxygen pressure for 72 h ($5\%$ CO2, $3\%$ O2, 37°C). UC-MSC were harvested and frozen in multiple aliquots. Prior transplantation, UC-MSC were thawed and subcultured for 72 h in the same media. ## RIP-B7.1 mice Mice experimentations were approved by the CABIMER Animal Committee, and performed in accordance with the Spanish law on animal use RD $\frac{53}{2013.}$ Mice were housed under specific pathogen-free conditions in the animal facility of CABIMER. Experimental autoimmune diabetes (EAD) was induced in 8–9 weeks old RIP-B7.1 mice transgenic mice (C57BL/6 H-2b background) by injection of 50 μg preproinsulin (ppINS) expression plasmid (PlasmidFactory GmbH, Germany) into each anterior tibialis muscles (Cobo-Vuilleumier et al. 2018). One week later, mice were intraperitoneally injected with 300 µL of DMEM (vehicle) or DMEM containing 5 × 105 UC-MSC under anesthesia using the IVIS system. Mice were euthanized at 1 day, 1 and 7 weeks after transplantation. Peritoneal fluid cells, splenocytes, pancreatic lymph nodes cells and pancreatic stromal cells were collected for FC and quantitative PCR analysis. ## Glycemic records Circulating glucose levels were measured weekly from tail vein blood samples using an Optium Xceed glucometer (Abbott Scientifica SA, Barcelona, Spain). Two consecutive measurements of non-fasting blood glucose ≥250 mg/dL were considered to indicate overt diabetes. ## UC-MSC tracking and biodistribution analysis Cultured UC-MSC suspensions were fluorescently labelled for 10 min with 10 µM carboxyfluorescein succinimidyl ester (CFSE) or 320 μg/mL XenoLight DiR (PerkinElmer) in DMEM for 20 min for in vivo biodistribution analysis using IVIS® Spectrum in vivo imaging system (PerkinElmer). Fluorescently labelled UC-MSC cultures were extensively washed before harvesting with trypsin. Fluorescently labelled UC-MSC suspension were identically washed twice through centrifugation before transplantation. ## ELISA: Plasmatic cytokines measurements Blood samples were collected by cardiac puncture and centrifuged into Microvette CB 300 K2E, Sarstedt (Nümbrecht, Germany) with EDTA and plasma collected and aliquots frozen at −80°C. Cytokine levels were determined using the mouse V-PLEX ProInflammatory Panel 1 kit 10-Plex (Meso Scale Discovery, Rockville, USA) and data acquired on an MSD MESOTM QuickPlex SQ120. ## Pancreas histology and immunohistochemistry Pancreases sections were obtained as previously reported (Lorenzo et al. 2015). Antibodies used in this study are listed in Supplementary Tables S1, S2. Nuclei were counterstained with DAPI, and sections mounted with DAKO fluorescent mounting medium. Images were acquired using either Leica DM6000B or a Leica TCS SP5 confocal microscope. Insulitis was scored in H&E stained paraffin sections of pancreas, as previously described (Mellado-Gil et al. 2016). For insulitis score, a total of 3 mice for each experimental group were analyzed at 7 weeks after transplantation. Total number of islets analyzed per group was 132 for CT, 82 for CT+5.105 UC-MSC, 71 for IMM, 183 for IMM+2.105 UC-MSC, 177 for IMM+5.105 UC-MSC and 180 for IMM+2 doses of 2.105 UC-MSC. ## Peritoneal drainages, spleen, pancreatic lymph nodes and pancreas sampling Peritoneal drainages were collected with minor modifications as previously described (Ray and Dittel 2010). Briefly, skin of the abdominal wall was removed and the exposed abdominal wall gently cleaned with ethanol prior injecting 8-10 mL cold PBS solution containing $2\%$ BSA and 2.5 mM EDTA into the peritoneal cavity. After a quick gentle massage, mice were placed side down on top of a platform and a small incision in the abdominal wall was made to collect peritoneal lavages into non-adherent Petri dishes placed on ice. Spleen and pancreases were isolated and stored into cold PBS+$2\%$BSA+2.5mMEDTA solution on ice. Pancreatic lymph nodes were surgically removed and collected under magnification. Splenocytes and pancreatic lymph nodes cells were teased into single cell suspensions by gentle disruption between the frosted ends of slides using a cold PBS+$2\%$ BSA+2.5 mM EDTA solution and filtered using 100 μm cell strainer (BD Falcon). See also Supplementary Methods for more detailed information. ## Flow cytometry Aliquots of freshly isolated cells resuspended in cold PBS+$2\%$BSA+2.5mMEDTA solution were incubated with antibodies for 30 min on ice. Stained cells were then fixed with $4\%$ paraformaldehyde and then analyzed on a FACSCalibur flow cytometer (BD Biosciences, Madrid, Spain). Data were analyzed using Flowing Software version 2.5.1 (Turku Centre for Biotechnology, Univ. Turku, Finland). ## RNA isolation and quantitative PCR (QPCR) RNA isolation and QPCR was performed as previously reported (Lachaud et al., 2014). Briefly, total RNA content was extracted with Trizol reagent (Invitrogen), and total RNA was reverse-transcribed into cDNA by using MMLV reverse transcriptase (Promega, Madison, WI, United States). qPCR was performed using SYBR-Green and detected using an ABI Prism 7500 system (Applied Biosystems, Foster City, CA, United States). Quantification of the mRNA level of each gene was normalized to β-actin mRNA. Results show folds of mRNA expression of a given gene relative to its expression in control non-immunized RIP-B7.1 mice, which served as the calibrator sample (set as 1). Primers sequences are listed in Supplementary Table S1. Primers aliquots can be obtained upon request. ## Figures edition Graphs were created using GraphPad Prism 7.00. Figures were created using Adobe Photoshop. ## Statistical analysis Results are expressed as mean ± s. e.m. N, is the number of mice analyzed for each group, in each analysis is ≥5 for flow cytometry and is ≥4 for qPCR. Statistical analysis was performed using the GraphPad Prism 7.00 Software (GraphPad, La Jolla, United States). Statistical differences were estimated by one-way ANOVA or Student´s test, whichever was appropriate. ## Data availability statement The raw data supporting the conclusion of this article will be made available by the authors, without undue reservation. ## Ethics statement The animal study was reviewed and approved by the Animal Experimentation and Ethics Committee of CABIMER (CEEA-CABIMER). ## Author contributions CCL: Conception and design, collection and assembly of data, data analysis and interpretation, manuscript writing and manuscript editing. N-CV, EF-M, and IDC: Collection and assembly of data, data analysis and interpretation. EA, GMC, JTH and AH: Data analysis and interpretation, manuscript editing. BRG, Conception and design, data analysis and interpretation, manuscript writing and editing. BS: Conception and design, data analysis and interpretation, manuscript writing and editing and financial support. All authors have read and agreed to the published the final version of the manuscript. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2023.1089817/full#supplementary-material ## Abbreviations ARG1, Arginase 1; Breg, B regulatory cells; CFSE, Carboxyfluorescein succinimidyl ester; EAD, Experimental autoimmune diabetes; FC, Flow cytometry; H&E, Hematoxylin and eosin; i.p., Intraperitoneal; IL-10, Interleukin 10; IL-4, Interleukin 4; IMC, Immature myeloid cells; iNOS, Inducible nitric oxide synthase; MDSC, Myeloid-derived suppressor cells; MHC-II, Major histocompatibility complex 2; MIF, Macrophage inhibitory factor; NOD, Non-obese diabetic mice; PDGR-β, Platelet derived growth factor beta; PFC, Peritoneal fluid cells; PLN, Pancreatic lymph nodes; ppINS, Preproinsulin; RIP-B7.1, Rat insulin promoter-B7.1; SLOs, Secondary lymphoid organs; T1DM, Type 1 diabetes mellitus; Treg, Regulatory T Cells; UC-MSC, Umbilical cord mesenchymal stromal cells. ## References 1. Antonela B., Ivana G., Ivan K., Nikolina V., Vesna B. P., Marina P.. Environmental risk factors for type 1 diabetes mellitus development. Exp. Clin. Endocrinol. Diabetes (2017) 125 563-570. DOI: 10.1055/s-0043-109000 2. Antonios J. K., Yao Z., Li C., Rao A. J., Goodman S. B.. Macrophage polarization in response to wear particles. Cell. Mol. Immunol. (2013) 10 471-482. DOI: 10.1038/cmi.2013.39 3. Aronson R., Gottlieb P. A., Christiansen J. S., Donner T. W., Bosi E., Bode B. W.. Low-dose otelixizumab anti-CD3 monoclonal antibody DEFEND-1 study: Results of the randomized phase III study in recent-onset human type 1 diabetes. Diabetes Care (2014) 37 2746-2754. DOI: 10.2337/dc13-0327 4. Barry F. P., Murphy J. M., English K., Mahon B. P.. Immunogenicity of adult mesenchymal stem cells: Lessons from the fetal allograft. Stem Cells Dev. (2005) 14 252-265. DOI: 10.1089/scd.2005.14.252 5. Ben Nasr M., Usuelli V., Seelam A. J., D'Addio F., Abdi R., Markmann J. F.. Regulatory B cells in autoimmune diabetes. J. Immunol. (2021) 206 1117-1125. DOI: 10.4049/jimmunol.2001127 6. Berglund A. K., Fortier L. A., Antczak D. F., Schnabel L. V.. Immunoprivileged no more: Measuring the immunogenicity of allogeneic adult mesenchymal stem cells. Stem Cell. Res. Ther. (2017) 8 288. DOI: 10.1186/s13287-017-0742-8 7. Bertram P., Tietze L., Hoopmann M., Treutner K. H., Mittermayer C., Schumpelick V.. Intraperitoneal transplantation of isologous mesothelial cells for prevention of adhesions. Eur. J. Surg. (1999) 165 705-709. DOI: 10.1080/11024159950189780 8. Bluestone J. A., Herold K., Eisenbarth G.. Genetics, pathogenesis and clinical interventions in type 1 diabetes. Nature (2010) 464 1293-1300. DOI: 10.1038/nature08933 9. Bronte V., Zanovello P.. Regulation of immune responses by L-arginine metabolism. Nat. Rev. Immunol. (2005) 5 641-654. DOI: 10.1038/nri1668 10. Calandra T., Roger T.. Macrophage migration inhibitory factor: A regulator of innate immunity. Nat. Rev. Immunol. (2003) 3 791-800. DOI: 10.1038/nri1200 11. Cameron M. J., Arreaza G. A., Zucker P., Chensue S. W., Strieter R. M., Chakrabarti S.. IL-4 prevents insulitis and insulin-dependent diabetes mellitus in nonobese diabetic mice by potentiation of regulatory T helper-2 cell function. J. Immunol. (1997) 159 4686-4692. DOI: 10.4049/jimmunol.159.10.4686 12. Campbell-Thompson M., Fu A., Kaddis J. S., Wasserfall C., Schatz D. A., Pugliese A.. Insulitis and beta-cell mass in the natural history of type 1 diabetes. Diabetes (2016) 65 719-731. DOI: 10.2337/db15-0779 13. Can A., Celikkan F. T., Cinar O.. Umbilical cord mesenchymal stromal cell transplantations: A systemic analysis of clinical trials. Cytotherapy (2017) 19 1351-1382. DOI: 10.1016/j.jcyt.2017.08.004 14. Canibano-Hernandez A., Del Burgo L. S., Espona-Noguera A., Ciriza J., Pedraz J. L.. Current advanced therapy cell-based medicinal products for type-1-diabetes treatment. Int. J. Pharm. (2018) 543 107-120. DOI: 10.1016/j.ijpharm.2018.03.041 15. Cassado Ados A., D'Imperio Lima M. R., Bortoluci K. R.. Revisiting mouse peritoneal macrophages: Heterogeneity, development, and function. Front. Immunol. (2015) 6 225. DOI: 10.3389/fimmu.2015.00225 16. Cho K. A., Lee J. K., Kim Y. H., Park M., Woo S. Y., Ryu K. H.. Mesenchymal stem cells ameliorate B-cell-mediated immune responses and increase IL-10-expressing regulatory B cells in an EBI3-dependent manner. Cell. Mol. Immunol. (2017) 14 895-908. DOI: 10.1038/cmi.2016.59 17. Christie M. R.. Delving into the type 1 diabetic islet: Evidence that B-cell infiltration of islets is linked to local hyperimmunity and accelerated progression to disease. Diabetes (2016) 65 1146-1148. DOI: 10.2337/dbi16-0008 18. Cobo-Vuilleumier N., Gauthier B. R.. Time for a paradigm shift in treating type 1 diabetes mellitus: Coupling inflammation to islet regeneration. Metabolism (2020) 104 154137. DOI: 10.1016/j.metabol.2020.154137 19. Cobo-Vuilleumier N., Lorenzo P. I., Rodriguez N. G., Herrera Gomez I. G., Fuente-Martin E., Lopez-Noriega L.. LRH-1 agonism favours an immune-islet dialogue which protects against diabetes mellitus. Nat. Commun. (2018) 9 1488. DOI: 10.1038/s41467-018-03943-0 20. Craig M. E., Kim K. W., Isaacs S. R., Penno M. A., Hamilton-Williams E. E., Couper J. J.. Early-life factors contributing to type 1 diabetes. Diabetologia (2019) 62 1823-1834. DOI: 10.1007/s00125-019-4942-x 21. de Andres C., Tejera-Alhambra M., Alonso B., Valor L., Teijeiro R., Ramos-Medina R.. New regulatory CD19(+)CD25(+) B-cell subset in clinically isolated syndrome and multiple sclerosis relapse. Changes after glucocorticoids. J. Neuroimmunol. (2014) 270 37-44. DOI: 10.1016/j.jneuroim.2014.02.003 22. El-Hossary N., Hassanein H., El-Ghareeb A. W., Issa H.. Intravenous vs intraperitoneal transplantation of umbilical cord mesenchymal stem cells from Wharton's jelly in the treatment of streptozotocin-induced diabetic rats. Diabetes Res. Clin. Pract. (2016) 121 102-111. DOI: 10.1016/j.diabres.2016.09.008 23. Emani R., Alam C., Pekkala S., Zafar S., Emani M. R., Hanninen A.. Peritoneal cavity is a route for gut-derived microbial signals to promote autoimmunity in non-obese diabetic mice. Scand. J. Immunol. (2015) 81 102-109. DOI: 10.1111/sji.12253 24. Ferris S. T., Zakharov P. N., Wan X., Calderon B., Artyomov M. N., Unanue E. R.. The islet-resident macrophage is in an inflammatory state and senses microbial products in blood. J. Exp. Med. (2017) 214 2369-2385. DOI: 10.1084/jem.20170074 25. Franquesa M., Hoogduijn M. J., Bestard O., Grinyo J. M.. Immunomodulatory effect of mesenchymal stem cells on B cells. Front. Immunol. (2012) 3 212. DOI: 10.3389/fimmu.2012.00212 26. Fu W., Wojtkiewicz G., Weissleder R., Benoist C., Mathis D.. Early window of diabetes determinism in NOD mice, dependent on the complement receptor CRIg, identified by noninvasive imaging. Nat. Immunol. (2012) 13 361-368. DOI: 10.1038/ni.2233 27. Gabrilovich D. I., Nagaraj S.. Myeloid-derived suppressor cells as regulators of the immune system. Nat. Rev. Immunol. (2009) 9 162-174. DOI: 10.1038/nri2506 28. Gaglia J. L., Guimaraes A. R., Harisinghani M., Turvey S. E., Jackson R., Benoist C.. Noninvasive imaging of pancreatic islet inflammation in type 1A diabetes patients. J. Clin. Invest. (2011) 121 442-445. DOI: 10.1172/JCI44339 29. Giannopoulou E. Z., Puff R., Beyerlein A., von Luettichau I., Boerschmann H., Schatz D.. Effect of a single autologous cord blood infusion on beta-cell and immune function in children with new onset type 1 diabetes: A non-randomized, controlled trial. Pediatr. Diabetes (2014) 15 100-109. DOI: 10.1111/pedi.12072 30. Gordon S., Taylor P. R.. Monocyte and macrophage heterogeneity. Nat. Rev. Immunol. (2005) 5 953-964. DOI: 10.1038/nri1733 31. Guo B., Rothstein T. L.. IL-4 upregulates Igα and Igβ protein, resulting in augmented IgM maturation and B cell receptor-triggered B cell activation. J. Immunol. (2013) 191 670-677. DOI: 10.4049/jimmunol.1203211 32. Haller M. J., Wasserfall C. H., Hulme M. A., Cintron M., Brusko T. M., McGrail K. M.. Autologous umbilical cord blood transfusion in young children with type 1 diabetes fails to preserve C-peptide. Diabetes Care (2011) 34 2567-2569. DOI: 10.2337/dc11-1406 33. Hong M., Liao Y., Liang J., Chen X., Li S., Liu W.. Immunomodulation of human CD19(+)CD25(high) regulatory B cells via Th17/Foxp3 regulatory T cells and Th1/Th2 cytokines. Hum. Immunol. (2019) 80 863-870. DOI: 10.1016/j.humimm.2019.05.011 34. Hu J., Yu X., Wang Z., Wang F., Wang L., Gao H.. Long term effects of the implantation of Wharton's jelly-derived mesenchymal stem cells from the umbilical cord for newly-onset type 1 diabetes mellitus. Endocr. J. (2013) 60 347-357. DOI: 10.1507/endocrj.ej12-0343 35. Jerram S. T., Dang M. N., Leslie R. D.. The role of epigenetics in type 1 diabetes. Curr. Diab Rep. (2017) 17 89. DOI: 10.1007/s11892-017-0916-x 36. Kendall P. L., Woodward E. J., Hulbert C., Thomas J. W.. Peritoneal B cells govern the outcome of diabetes in non-obese diabetic mice. Eur. J. Immunol. (2004) 34 2387-2395. DOI: 10.1002/eji.200324744 37. Kessel A., Haj T., Peri R., Snir A., Melamed D., Sabo E.. Human CD19(+)CD25(high) B regulatory cells suppress proliferation of CD4(+) T cells and enhance Foxp3 and CTLA-4 expression in T-regulatory cells. Autoimmun. Rev. (2012) 11 670-677. DOI: 10.1016/j.autrev.2011.11.018 38. Kim M., Park S. Y., Jin M. L., Park G., Son H. J.. Cucurbitacin B inhibits immunomodulatory function and the inflammatory response in macrophages. Immunopharmacol. Immunotoxicol. (2015) 37 473-480. DOI: 10.3109/08923973.2015.1085065 39. Kleffel S., Vergani A., Tezza S., Ben Nasr M., Niewczas M. A., Wong S.. Interleukin-10+ regulatory B cells arise within antigen-experienced CD40+ B cells to maintain tolerance to islet autoantigens. Diabetes (2015) 64 158-171. DOI: 10.2337/db13-1639 40. Korf H., Breser L., Van Hoeck J., Godoy J., Cook D. P., Stijlemans B.. MIF inhibition interferes with the inflammatory and T cell-stimulatory capacity of NOD macrophages and delays autoimmune diabetes onset. PLoS One (2017) 12 e0187455. DOI: 10.1371/journal.pone.0187455 41. Korns D., Frasch S. C., Fernandez-Boyanapalli R., Henson P. M., Bratton D. L.. Modulation of macrophage efferocytosis in inflammation. Front. Immunol. (2011) 2 57. DOI: 10.3389/fimmu.2011.00057 42. Lachaud C. C., Lopez-Beas J., Soria B., Hmadcha A.. EGF-induced adipose tissue mesothelial cells undergo functional vascular smooth muscle differentiation. Cell. Death Dis. (2014) 5 e1304. DOI: 10.1038/cddis.2014.271 43. Li S., Sun Z., Lv G., Guo X., Zhang Y., Yu W.. Microencapsulated UCB cells repair hepatic injure by intraperitoneal transplantation. Cytotherapy (2009) 11 1032-1040. DOI: 10.3109/14653240903121278 44. Li X., Liu J., Xing Z., Tang J., Sun H., Zhang X.. Polymorphonuclear myeloid-derived suppressor cells link inflammation and damage response after trauma. J. Leukoc. Biol. (2021) 110 1143-1161. DOI: 10.1002/JLB.3MA0821-029R 45. Lorenzo P. I., Fuente-Martin E., Brun T., Cobo-Vuilleumier N., Jimenez-Moreno C. M., I G. H. G.. PAX4 defines an expandable beta-cell subpopulation in the adult pancreatic islet. Sci. Rep. (2015) 5 15672. DOI: 10.1038/srep15672 46. Lu J., Shen S. M., Ling Q., Wang B., Li L. R., Zhang W.. One repeated transplantation of allogeneic umbilical cord mesenchymal stromal cells in type 1 diabetes: An open parallel controlled clinical study. Stem Cell. Res. Ther. (2021) 12 340. DOI: 10.1186/s13287-021-02417-3 47. Luk F., Carreras-Planella L., Korevaar S. S., de Witte S. F. H., Borras F. E., Betjes M. G. H.. Inflammatory conditions dictate the effect of mesenchymal stem or stromal cells on B cell function. Front. Immunol. (2017) 8 1042. DOI: 10.3389/fimmu.2017.01042 48. Marek-Trzonkowska N., Mysliwiec M., Dobyszuk A., Grabowska M., Techmanska I., Juscinska J.. Administration of CD4+CD25highCD127-regulatory T cells preserves beta-cell function in type 1 diabetes in children. Diabetes Care (2012) 35 1817-1820. DOI: 10.2337/dc12-0038 49. Mellado-Gil J. M., Jimenez-Moreno C. M., Martin-Montalvo A., Alvarez-Mercado A. I., Fuente-Martin E., Cobo-Vuilleumier N.. PAX4 preserves endoplasmic reticulum integrity preventing beta cell degeneration in a mouse model of type 1 diabetes mellitus. Diabetologia (2016) 59 755-765. DOI: 10.1007/s00125-016-3864-0 50. Morgan N. G., Leete P., Foulis A. K., Richardson S. J.. Islet inflammation in human type 1 diabetes mellitus. IUBMB Life (2014) 66 723-734. DOI: 10.1002/iub.1330 51. Munder M.. Arginase: An emerging key player in the mammalian immune system. Br. J. Pharmacol. (2009) 158 638-651. DOI: 10.1111/j.1476-5381.2009.00291.x 52. Nagaraj S., Collazo M., Corzo C. A., Youn J. I., Ortiz M., Quiceno D.. Regulatory myeloid suppressor cells in health and disease. Cancer Res. (2009) 69 7503-7506. DOI: 10.1158/0008-5472.CAN-09-2152 53. Norris J. M., Johnson R. K., Stene L. C.. Type 1 diabetes-early life origins and changing epidemiology. Lancet Diabetes Endocrinol. (2020) 8 226-238. DOI: 10.1016/S2213-8587(19)30412-7 54. Ovalle F., Grimes T., Xu G., Patel A. J., Grayson T. B., Thielen L. A.. Verapamil and beta cell function in adults with recent-onset type 1 diabetes. Nat. Med. (2018) 24 1108-1112. DOI: 10.1038/s41591-018-0089-4 55. Pavlou S., Wang L., Xu H., Chen M.. Higher phagocytic activity of thioglycollate-elicited peritoneal macrophages is related to metabolic status of the cells. J. Inflamm. (Lond) (2017) 14 4. DOI: 10.1186/s12950-017-0151-x 56. Phillips B. E., Garciafigueroa Y., Trucco M., Giannoukakis N.. Clinical tolerogenic dendritic cells: Exploring therapeutic impact on human autoimmune disease. Front. Immunol. (2017) 8 1279. DOI: 10.3389/fimmu.2017.01279 57. Phillips B. E., Garciafigueroa Y., Engman C., Trucco M., Giannoukakis N.. Tolerogenic dendritic cells and T-regulatory cells at the clinical trials crossroad for the treatment of autoimmune disease; emphasis on type 1 diabetes therapy. Front. Immunol. (2019) 10 148. DOI: 10.3389/fimmu.2019.00148 58. Pociot F., Lernmark A.. Genetic risk factors for type 1 diabetes. Lancet (2016) 387 2331-2339. DOI: 10.1016/S0140-6736(16)30582-7 59. Qu Z., Lou Q., Cooper D. K. C., Pu Z., Lu Y., Chen J.. Potential roles of mesenchymal stromal cells in islet allo- and xenotransplantation for type 1 diabetes mellitus. Xenotransplantation (2021) 28 e12678. DOI: 10.1111/xen.12678 60. Rajasalu T., Barth C., Spyrantis A., Durinovic-Bello I., Uibo R., Schirmbeck R.. Experimental autoimmune diabetes: A new tool to study mechanisms and consequences of insulin-specific autoimmunity. Ann. N. Y. Acad. Sci. (2004) 1037 208-215. DOI: 10.1196/annals.1337.034 61. Rajasalu T., Brosi H., Schuster C., Spyrantis A., Boehm B. O., Chen L.. Deficiency in B7-H1 (PD-L1)/PD-1 coinhibition triggers pancreatic beta-cell destruction by insulin-specific, murine CD8 T-cells. Diabetes (2010) 59 1966-1973. DOI: 10.2337/db09-1135 62. Rath M., Muller I., Kropf P., Closs E. I., Munder M.. Metabolism via arginase or nitric oxide synthase: Two competing arginine pathways in macrophages. Front. Immunol. (2014) 5 532. DOI: 10.3389/fimmu.2014.00532 63. Ray A., Dittel B. N.. Isolation of mouse peritoneal cavity cells. J. Vis. Exp. (2010) 35 1488. DOI: 10.3791/1488 64. Roep B. O., Thomaidou S., van Tienhoven R., Zaldumbide A.. Type 1 diabetes mellitus as a disease of the beta-cell (do not blame the immune system?). Nat. Rev. Endocrinol. (2021) 17 150-161. DOI: 10.1038/s41574-020-00443-4 65. Saiwai H., Kumamaru H., Ohkawa Y., Kubota K., Kobayakawa K., Yamada H.. Ly6C+ Ly6G- Myeloid-derived suppressor cells play a critical role in the resolution of acute inflammation and the subsequent tissue repair process after spinal cord injury. J. Neurochem. (2013) 125 74-88. DOI: 10.1111/jnc.12135 66. Sanchez-Zamora Y. I., Juarez-Avelar I., Vazquez-Mendoza A., Hiriart M., Rodriguez-Sosa M.. Altered macrophage and dendritic cell response in mif-/- mice reveals a role of mif for inflammatory-Th1 response in type 1 diabetes. J. Diabetes Res. (2016) 2016 7053963. DOI: 10.1155/2016/7053963 67. Sarikonda G., Pettus J., Phatak S., Sachithanantham S., Miller J. F., Wesley J. D.. CD8 T-cell reactivity to islet antigens is unique to type 1 while CD4 T-cell reactivity exists in both type 1 and type 2 diabetes. J. Autoimmun. (2014) 50 77-82. DOI: 10.1016/j.jaut.2013.12.003 68. Schif-Zuck S., Gross N., Assi S., Rostoker R., Serhan C. N., Ariel A.. Saturated-efferocytosis generates pro-resolving CD11b low macrophages: Modulation by resolvins and glucocorticoids. Eur. J. Immunol. (2011) 41 366-379. DOI: 10.1002/eji.201040801 69. Shen Z., Huang W., Liu J., Tian J., Wang S., Rui K.. Effects of mesenchymal stem cell-derived exosomes on autoimmune diseases. Front. Immunol. (2021) 12 749192. DOI: 10.3389/fimmu.2021.749192 70. Silveira P. A., Grey S. T.. B cells in the spotlight: Innocent bystanders or major players in the pathogenesis of type 1 diabetes. Trends Endocrinol. Metab. (2006) 17 128-135. DOI: 10.1016/j.tem.2006.03.006 71. Stojanovic I., Dimitrijevic M., Vives-Pi M., Mansilla M. J., Pujol-Autonell I., Rodriguez-Fernandez S.. Cell-based tolerogenic therapy, experience from animal models of multiple sclerosis, type 1 diabetes and rheumatoid arthritis. Curr. Pharm. Des. (2017) 23 2623-2643. DOI: 10.2174/1381612823666170214120708 72. Stosic-Grujicic S., Stojanovic I., Maksimovic-Ivanic D., Momcilovic M., Popadic D., Harhaji L.. Macrophage migration inhibitory factor (MIF) is necessary for progression of autoimmune diabetes mellitus. J. Cell. Physiol. (2008) 215 665-675. DOI: 10.1002/jcp.21346 73. Thayer T. C., Wilson S. B., Mathews C. E.. Use of nonobese diabetic mice to understand human type 1 diabetes. Endocrinol. Metab. Clin. North Am. (2010) 39 541-561. DOI: 10.1016/j.ecl.2010.05.001 74. von Scholten B. J., Kreiner F. F., Gough S. C. L., von Herrath M.. Current and future therapies for type 1 diabetes. Diabetologia (2021) 64 1037-1048. DOI: 10.1007/s00125-021-05398-3 75. Vudattu N. K., Herold K. C.. Treatment of new onset type 1 diabetes with teplizumab: Successes and pitfalls in development. Expert Opin. Biol. Ther. (2014) 14 377-385. DOI: 10.1517/14712598.2014.881797 76. Wakabayashi K., Hamada C., Kanda R., Nakano T., Io H., Horikoshi S.. Adipose-derived mesenchymal stem cells transplantation facilitate experimental peritoneal fibrosis repair by suppressing epithelial-mesenchymal transition. J. Nephrol. (2014) 27 507-514. DOI: 10.1007/s40620-014-0133-5 77. Wang S., Tan Q., Hou Y., Dou H.. Emerging roles of myeloid-derived suppressor cells in diabetes. Front. Pharmacol. (2021) 12 798320. DOI: 10.3389/fphar.2021.798320 78. Warshauer J. T., Bluestone J. A., Anderson M. S.. New frontiers in the treatment of type 1 diabetes. Cell. Metab. (2020) 31 46-61. DOI: 10.1016/j.cmet.2019.11.017 79. Welzen-Coppens J. M., van Helden-Meeuwsen C. G., Drexhage H. A., Versnel M. A.. Abnormalities of dendritic cell precursors in the pancreas of the NOD mouse model of diabetes. Eur. J. Immunol. (2012) 42 186-194. DOI: 10.1002/eji.201141770 80. Willcox A., Richardson S. J., Bone A. J., Foulis A. K., Morgan N. G.. Analysis of islet inflammation in human type 1 diabetes. Clin. Exp. Immunol. (2009) 155 173-181. DOI: 10.1111/j.1365-2249.2008.03860.x 81. Wynn T. A.. Type 2 cytokines: Mechanisms and therapeutic strategies. Nat. Rev. Immunol. (2015) 15 271-282. DOI: 10.1038/nri3831 82. Yang W. C., Hwang Y. S., Chen Y. Y., Liu C. L., Shen C. N., Hong W. H.. Interleukin-4 supports the suppressive immune responses elicited by regulatory T cells. Front. Immunol. (2017) 8 1508. DOI: 10.3389/fimmu.2017.01508
--- title: Is vitamin B12 deficiency a risk factor for gastroparesis in patients with type 2 diabetes? authors: - Sally S. Ahmed - Hala A. Abd El-Hafez - Mohamed Mohsen - Azza A. El-Baiomy - Enas T. Elkhamisy - Mervat M. El-Eshmawy journal: Diabetology & Metabolic Syndrome year: 2023 pmcid: PMC9976380 doi: 10.1186/s13098-023-01005-0 license: CC BY 4.0 --- # Is vitamin B12 deficiency a risk factor for gastroparesis in patients with type 2 diabetes? ## Abstract ### Background Diabetic gastroparesis is a severe diabetic complication refers to delayed gastric emptying in the absence of mechanical obstruction of the stomach. Vitamin B12 affects the dynamics of autonomic nervous system and its deficits has been linked to cardiovascular autonomic neuropathy therefore, vitamin B12 deficiency was hypothesized to be implicated in the development of diabetic gastroparesis. This study was conducted to explore the possible association between vitamin B12 deficiency and gastroparesis in patients with type 2 diabetes (T2D). ### Methods A total of 100 T2D patients with diabetes duration > 10 years and 50 healthy controls matched for age and sex were recruited for this study. T2D patients were divided into 2 groups: patients with gastroparesis and patients without gastroparesis. The diagnosis of gastroparesis was based on Gastroparesis Cardinal Symptom Index (GCSI) Score ≥ 1.9 and ultrasonographic findings including gastric emptying ˂ $35.67\%$ and motility index ˂ 5.1. Anthropometric measurements, plasma glucose, glycosylated hemoglobin (HbA1c), lipids profile, vitamin B12 and transabdominal ultrasonography were assessed. ### Results The frequency of vitamin B12 deficiency in total patients with T2D was $35\%$ ($54.5\%$ in patients with gastroparesis vs. $11.1\%$ in patients without gastroparesis, $P \leq 0.$ 001). Vitamin B12 level was negatively correlated with GCSI Score whereas, it was positively correlated with gastric emptying and motility index. Vitamin B12 deficiency was an independent predictor for gastroparesis in patients with T2D; it predicts gastroparesis at a cut off value of 189.5 pmol/L with $69.1\%$ sensitivity and $64.4\%$ specificity, $$P \leq 0.002.$$ ### Conclusions Beside the known risk factors of diabetic gastroparesis, vitamin B12 deficiency is an independent predictor of diabetic gastroparesis in patients with T2D. ## Background Diabetic gastroparesis is a clinical syndrome characterized by delayed gastric emptying in the absence of mechanical obstruction of the stomach [1]. The characteristic symptoms of gastroparesis are early satiety, nausea, vomiting, bloating and upper abdominal pain [2] however, diabetic gastroparesis is often asymptomatic [3, 4]. Diabetic gastroparesis is not uncommon disease [5]; the reported prevalence from the tertiary referral centers is 10–$30\%$ of patients with type 2 diabetes (T2D) [6–8] whereas, the community prevalence using strict diagnostic criteria is $1.1\%$ of patients with T2D [9]. Autonomic neuropathy, enteropathy and hyperglycemia are the most frequently implicated risk factors of diabetic gastroparesis [2, 10, 11]. Delayed gastric emptying leads to poor glycemic control and increased risk of hypoglycemia [12]. Indeed, gastroparesis significantly impairs quality of life [9] and is associated with morbidity and mortality [5, 13]. Recently, vitamin B12 levels have been found to be inversely related to glucose intolerance [14]. Additionally, vitamin B12 affects the dynamics of autonomic nervous system [15] and its deficits has been linked to cardiovascular autonomic neuropathy [16]. Vitamin B12 deficiency may be also implicated in the development of diabetic gastroparesis however, this association has not been yet investigated. Therefore, the aim of the present study was to explore the possible association between vitamin B12 deficiency and diabetic gastroparesis in patients with T2D. ## Methods This study comprised 100 adult patients with T2D and 50 age- and sex-matched healthy controls. Patients with T2D were consecutively recruited from Diabetes Outpatient Clinic at Mansoura Specialized Medical Hospital, Mansoura University, Mansoura, Egypt. The inclusion criteria were patients with duration of T2D > 10 years and who had symptoms of gastroparesis [Gastroparesis Cardinal Symptom Index (GCSI) Score ≥ 1.9]. Patients with T2D were submitted for transabdominal ultrasonography accordingly, they divided into 2 groups: patients with gastroparesis ($$n = 55$$) and patients without gastroparesis ($$n = 45$$). Exclusion criteria were history of digestive tract surgery or prior gastric outlet obstruction, thyroid disease, liver & renal failure, neuropsychiatric disorder, connective tissue disorders, malignancies, pregnancy and participants taking vitamin B12 and alcohol. Drugs that could potentially interfere with gastrointestinal motility such as GLP-1 receptor agonists and the amylin analog, α glucosidase inhibitors, and opioid analgesic were also excluded. Healthy controls were recruited from the same geographic area with the same exclusion criteria. All participants were subjected to a thorough medical history and underwent a clinical examination. Anthropometric measurements including height, body weight, body mass index (BMI) (kg/m2), and waist circumference (WC) were obtained using standardized techniques. The diagnosis of diabetic gastroparesis was based on the symptom validated questionnaire GCSI Score ≥ 1.9 and ultrasonographic findings including gastric emptying ˂ $35.67\%$ and motility index ˂ 5.1. The cut-off point of gastric emptying and motility index were calculated from our study healthy controls as mean—2SD. The GCSI Score consists of 9 symptoms covering 3 areas; nausea/vomiting subscale (3 symptoms: nausea, vomiting and retching), postprandial fullness/early satiety subscale (4 symptoms: stomach fullness, early satiety, postprandial fullness and loss of appetite) and bloating subscale (2 symptoms: bloating and stomach distension). All symptoms are rated from 0 to 5 over the prior 2 weeks [no symptoms = 0, very mild = 1, mild = 2, moderate = 3, severe = 4, and very severe = 5]. GCSI Score was calculated as the average of the 3 symptom subscales [17]. The clinical severity of gastroparesis was graded on a scale originally proposed by Abell et al. [ 18]; grade 1: mild gastroparesis (symptoms are relatively easily controlled and weight and nutrition can be maintained with a regular diet); grade 2: compensated gastroparesis (symptoms are partially controlled with the use of daily medications and nutrition can be maintained with dietary adjustments); grade 3: gastroparesis with gastric failure (uncountable refractory symptoms with frequent hospitalizations and/or inability to maintain nutrition via an oral route). Vitamin B12 deficiency was defined as vitamin B12 levels below 125 pmol/L [16]. Peripheral neuropathy was diagnosed based on neuropathy disability and symptom scores [19, 20]. Diabetic nephropathy was diagnosed according to Umanath & Lewis [21]. Diabetic retinopathy was assessed through fundus examination. ## Laboratory assay Fasting plasma glucose (FPG) and 2-h post prandial plasma glucose (PPG) were measured by commercially available kit, Cobas (Integra-400) supplied by Roche Diagnostics (Mannheim, Germany). Glycated hemoglobin (HbA1c) was estimated as an index of metabolic control on a DCA 2000 analyzer, fast ion exchange resin (Roche Diagnostic, Germany. Total cholesterol (TC), triglycerides (TGs) and high density lipoprotein cholesterol (HDL-C) were measured by commercially available kits (Cobas Integra-400). Low density lipoprotein cholesterol (LDL-C) was calculated according to Friedewald et al. [ 22]. Complete metabolic panel including ESR, complete blood count, renal, liver and thyroid function tests were also assessed. Serum vitamin B12 level was assayed by ELISA technique supplied by Bioassay technology. ## Ultrasonography assessment of gastric motility After an overnight fasting, patients sat in a chair, leaned slightly backwards and drank 400 ml meat soup (54.8 kcal, 0.38 g protein and 0.25 g fat). An ultrasound probe was positioned vertically to permit simultaneous visualization of the gastric antrum, superior mesenteric artery and abdominal aorta for evaluation of the antral contractions Fig 1. The examination was conducted by GE LOGIQ E9 ultrasound machine with a 5 MHz convex probe [23]. The gastric emptying (%) was estimated as: ([antral area at 1 min] – [antral area at 15 min]/antral area at 1 min) × 100. The motility index was estimated by calculating the mean amplitude x frequency of contractions. The amplitude of antral contractions is the difference between the relaxed and contracted areas during a 3-min interval, divided by the relaxed area. The frequency of antral contractions is the number of contractions during a 3-min interval beginning 2 min after ingestion of soup. Ultrasonography was assessed by the same operator to minimize variation in the examination procedure. Fasting blood glucose was less than 275 mg/dl on the day of testing and patients had stopped prokinetic drugs 7 days before the procedure. Fig. 1Vertically oriented ultrasonography demonstrates the relaxed phase at 1 min (A) and the contracted phase (B) at 3 min after ingestion of the meat soup. SMA: superior mesenteric artery, AO: aorta ## Statistical analysis This study was a pilot study with an initial sample size of 20 T2D patients who were excluded from the full scale study. The final calculated sample size was 97. Data entry and analysis were done by the SPSS statistical package (version 22, Armonk, NY: IBM Corp). The data were expressed as mean ± SD for continuous data, number and percent for categorical data and median (minimum–maximum) for skewed data. Student’s t and Mann–Whitney U tests were used to compare the 2 studied groups for parametric and non-parametric data, respectively. The chi-square and Fischer exact tests were performed to compare 2 or more groups of qualitative variables. The correlations of GCSI Score, gastric emptying and motility index with all other studied variables were analyzed by the Pearson and Spearman correlations analysis. Binary stepwise logistic regression analysis was used to predict the independent variables of binary outcome; the significant predictors in the univariate analysis were entered into the regression model. Receiver operating characteristic (ROC) curve was done to detect the level of vitamin B12 associated with gastroparesis in patients with T2D. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under the curve (AUC) and $95\%$ CI were evaluated. P ≤ 0.05 was considered as significant. ## Results Patients with T2D had significantly higher BMI, WC, systolic & diastolic blood pressures (SBP & DBP), FPG, 2 h PPG, HbA1c, TC, TGs, LDL-C and lower HDL-C than did healthy controls. Among the studied participants with T2D, the median age of diabetes duration was 17 years, $63\%$ had hypertension, $32\%$ had retinopathy, $24\%$ had diabetic nephropathy and $43\%$ had neuropathy ($18\%$ moderate, $25\%$ severe). With regard to anti-diabetic medications, $40\%$ received insulin, $37\%$ received oral anti-diabetic drugs and $23\%$ received combined oral anti-diabetic drugs and insulin. Vitamin B12 was significantly lower in patients with T2D than in healthy controls, the frequency of vitamin B12 deficiency was $35\%$ Table 1.Table 1Baseline characteristics of the study subjectsCharacteristicsPatients with T2D ($$n = 100$$)Healthy control ($$n = 50$$)P-valueAge (years)53.78 ± 9.6553.36 ± 8.580.791Gender0.940 Women/Men$\frac{52}{4826}$/24Duration of diabetes (years)17 (11–22)––Medications–– Insulin alone40 ($40\%$) SU alone17 ($17\%$) SGLT2 inhibitors alone7 ($7\%$) TZD alone6 ($6\%$) Metformin alone7 ($7\%$) Insulin & metformin11 ($11\%$) Insulin & TZD5 ($5\%$) Insulin & SGLT2 inhibitors7 ($7\%$)BMI (Kg/m2)36.47 ± 5.1021.87 ± 2.01 < 0.0001WC (cm)106.95 ± 21.2179.32 ± 6.39 < 0.001SBP (mm Hg)146.08 ± 13.81109.52 ± 8.69 < 0.001DBP (mm Hg)92.84 ± 8.8971.92 ± 3.94 < 0.001History of hypertension n (%)63 ($63\%$)––Retinopathy n (%)32 ($32\%$)–– Proliferative retinopathy n (%)22 ($22\%$) Non proliferative retinopathy n (%)10 ($10\%$)Nephropathy n(%)24 ($24\%$)––Peripheral neuropathy n (%)43 ($43\%$)–– Moderate n (%)25 ($25\%$) Severe n (%)18 ($18\%$)FPG (mg/dl)160.84 ± 33.2684.44 ± 9.17 < 0.0012 h PPG (mg/dl)256.71 ± 52.97118.89 ± 11.85 < 0.001HbA1c (%)9 ± 1.565.91 ± 0.30 < 0.001TC (mg/dl)269.03 ± 49.50166.46 ± 14.29 < 0.001TGs (mg/dl)208.53 ± 46.1792.76 ± 16.88 < 0.001LDL-C (mg/dl)141.55 ± 32.7977.78 ± 10.56 < 0.001HDL-C (mg/dl)40.50 ± 5.4059.28 ± 3.34 < 0.001Vitamin B12 level (pmol/L)150 (47–569)309 [127- 654] < 0.001Vitamin B12 deficiency n (%)35 ($35\%$)00.001*P is significant if ≤ 0.05Data are expressed as means ± standard deviation, numbers, proportion or median (minimum–maximum), T2D type 2 diabetes, BMI body mass index, WC waist circumference, SU sulphonylurea, TZD thiazoldindiaone, SGLT2 sodium glucose transporter, SBP systolic blood pressure, DBP diastolic blood pressure, FPG fasting plasma glucose, 2 h PPG 2 h post prandial plasma glucose, HbA1c Hemoglobin A1c, TC total cholesterol, TGs triglycerides, LDL-C low density lipoprotein cholesterol, HDL-C high density lipoprotein cholesterol T2D patients with gastroparesis had significantly longer diabetes duration, higher BMI, WC, SBP, DBP, FPG, HbA1c, TC, TGs and LDL-C compared with those without gastroparesis. The frequency of hypertension, proliferative retinopathy, nephropathy and peripheral neuropathy were significantly higher in patients with gastroparesis than in those without gastroparesis. Vitamin B12 level was significantly lower in patients with gastroparesis than in those without gastroparesis. The frequency of vitamin B12 deficiency was $54.5\%$ in patients with gastroparesis and $11.1\%$ in patients without gastroparesis, $$P \leq 0.001.$$ The grade of gastroparesis in T2D patients was distributed as $78.2\%$ for grade 1, $21.8\%$ for grade 2 with no reported cases in grade 3. Gastric emptying and motility index were significantly lower in T2D patients with gastroparesis than in those without gastroparesis. With regard to GCSI Score, there was no significant difference between patients with and without gastroparesis, Table 2.Table 2Baseline characteristics of the studied T2D patients with and without gastroparesisCharacteristicsT2D patients with gastroparesis ($$n = 55$$)T2D patients without gastroparesis ($$n = 45$$)P-valueAge (years)54.14 ± 9.2753.33 ± 10.190.678Gender Women/Men$\frac{29}{2623}$/220.872 Duration of diabetes (years)17 (11–22)11 (11–13)0.001Medications Metformin10 ($18.2\%$)8 ($17.7\%$)0.872 Non-metformin45 ($81.8\%$)37 ($82.3\%$)0.639 BMI (Kg/m2)38.75 ± 5.0833.68 ± 3.560.001* WC (cm)117.59 ± 12.6192.33 ± 3.460.001* SBP (mm Hg)150.93 ± 11.26140.15 ± 14.420.001* DBP (mmHg)95.36 ± 9.1889.75 ± 7.550.001* History of HTN n (%)41 ($74.5\%$)22 ($48.9\%$)0.008Retinopathy n (%) Proliferative19 ($34.5\%$)3 ($6.7\%$)0.001 Non proliferative7 ($12.7\%$)3 ($6.7\%$)0.315 Nephropathy n (%)20 ($36.4\%$)4 ($8.9\%$)0.001* Peripheral neuropathy n (%)35 ($63.6\%$)8 ($17.8\%$)0.001* Moderate n (%)20 ($57.1\%$)5 ($62.5\%$)0.004* Severe n (%)15 ($42.9\%$)3 ($37.5\%$)0.008* FPG (mg/dl)167.98 ± 32.35152.11 ± 32.610.017* 2 h PPG (mg/dl)263.90 ± 57.76247.91 ± 45.560.134 HbA1c (%)9.85 ± 1.667.95 ± 0.870.001* TC (mg/dl)308.40 ± 25.33220.91 ± 20.500.001* TGs (mg/dl)218.89 ± 54.48195.86 ± 29.250.012* LDL-C (mg/dl)155.07 ± 34.18125.02 ± 21.860.001* HDL-C (mg/dl)40.01 ± 6.9341.08 ± 3.880.327 Vitamin B12 level (pmol/L)88 (47–567)230 (55–569)0.001* Vitamin B12 deficiency n (%)30 ($54.5\%$)5 ($11.1\%$)0.001Gastroparesis assessment parameters Grade of gastroparesis Grade 143 ($78.2\%$)–– Grade 212 ($21.8\%$) Grade 30GCSI Score2.6 (2–3)2.3 (2–3)0.703Gastric emptying (%)26.33 (19.6–29.9)58.34 (46.48–78.12)0.001Motility index3.54 (2.12–4.12)7.64 (6.35–9.56)0.001*P is significant if ≤ 0.05Data are expressed as means ± standard deviation, numbers, proportion or median (minimum–maximum), T2D type 2 diabetes, BMI body mass index, WC waist circumference, SBP systolic blood pressure, DBP diastolic blood pressure, HTN hypertension, FPG fasting plasma glucose, 2 h PPG 2 h post prandial plasma glucose, HbA1c Hemoglobin A1c, TC total cholesterol, TGs triglycerides, LDL-C low density lipoprotein cholesterol, HDL-C high density lipoprotein cholesterol, GCSI Gastroparesis Cardinal Symptom Index GCSI Score was positively correlated with female sex, duration of T2D, BMI, SBP, DBP, retinopathy, nephropathy, neuropathy, FPG, HbA1c and grade of gastroparesis. Gastric emptying and motility index were negatively correlated with female sex, duration of T2D, BMI, SPB, DBP, retinopathy, nephropathy, neuropathy, FPG, HbA1c, TC, TGs and grade of gastroparesis. Table 3 Vitamin B12 levels were negatively correlated with GCSI Score and positively correlated with gastric emptying and motility index Figs 2, 3 and 4.Table 3Correlation of GCSI Score, gastric emptying and motility index, with other variables in T2D patients with gastroparesisVariablesGCSI ScoreGastric emptyingMotility indexrP-valuerP-valuerP-valueAge/years0.1350.18− 0.1250.21− 0.2450.58Sex (female)0.5120.03− 0.6140.02− 0.5800.03T2D Duration0.6170.02− 0.7150.01− 0.8110.001BMI (kg/m2)0.5110.04− 0.5450.03− 0.4510.04WC (cm)0.3150.30− 0.2890.40− 0.5810.50SBP (mm Hg)0.6800.02− 0.6900.01− 0.5820.02DBP(mm Hg)0.5980.04− 0.6480.002− 0.5210.02HTN0.6810.02− 0.6920.01− 0.5830.02Medications0.1350.18− 0.1250.21− 0.2450.58Retinopathy0.7190.001− 0.6950.002− 0.7110.003Nephropathy0.7180.001− 0.6890.02− 0.7130.003Neuropathy0.6020.01− 0.6310.003− 0.6740.004FPG (mg/dl)0.5820.03− 0.6970.02− 0.5150.012 h PPG(mg/dl)0.1120.60− 0.2510.40− 0.2410.10HbA1c (%)0.7140.003− 0.8050.01− 0.7540.02TC (mg/dl)0.0360.79− 0.5130.04− 0.5710.02TGs (mg/dl)0.0310.82− 0.5800.03− 0.6970.02LDL-C (mg/dl)0.2580.50− 0.1980.60− 0.2150.70HDL-C (mg/dl)− 0.1120.600.2540.400.2410.10GCSI Score––− 0.5400.69− 0.0280.84Grade of gastroparesis0.6110.01− 0.6280.01− 0.6210.01*P is significant if ≤ 0.05T2D type 2 diabetes, GCSI Gastroparesis Cardinal Symptom Index, BMI body mass index, WC waist circumference, SBP systolic blood pressure, DBP diastolic blood pressure, HTN hypertension, FPG fasting plasma glucose, 2 h PPG 2 h post prandial plasma glucose, HbA1c Hemoglobin A1c, TC total cholesterol, TGs triglycerides, LDL-C low density lipoprotein cholesterol, HDL-C high density lipoprotein cholesterolFig. 2Correlation between vitamin B12 level and Gastroparesis Cardinal Symptom Index Score in T2D patients with gastroparesisFig. 3Correlation between vitamin B12 level and gastric emptying in T2D patients with gastroparesisFig. 4Correlation between vitamin B12 level and motility index in T2D patients with gastroparesis Female sex, duration of T2D, BMI, retinopathy, neuropathy, nephropathy, FPG, HbA1c, TC, TGs and vitamin B12 deficiency were significant positive predictors of gastroparesis in patients with T2D Table 4. The cutoff value of vitamin B12 level associated with gastroparesis was 189.5 pmol/I with $69.1\%$ sensitivity, $64.4\%$ specificity, $70.4\%$ PPV and $63\%$ NPV, AUC was 0.678, $95\%$ CI (0.586–0.788), $$P \leq 0.002$$ Fig 5.Table 4Logistic regression analysis with gastroparesis as the dependent variable in patients with T2DVariablesOR ($95\%$ CI)βP-valueSex (Female)1.25 (1.14–5.2)0.2510.02Duration of diabetes/years16.95 (1.34–214.36)0.2510.02BMI (kg/m2)2.80 (1.24–4.21)1.2500.03Retinopathy3.10 (1.5–7.89)0.1240.01Nephropathy2.95 (1.1–5.1)0.1140.02Peripheral neuropathy2.80 (1.24–4.21)0.2540.03FPG (mg/dl)1.06 (1.024–1.09)1.2500.01HbA1c %17.07 (4.88–59.72)3.254 < 0.001TC (mg/dl)1.09 (1.03–1.16)2.5800.003TGs (mg/dl)1.08 (1.02–1.14)2.0200.006Vitamin B12 deficiency8.50 (3.5–15.89)2.1400.002P is significant if ≤ 0.05OR Odds Ratio, CI confidence interval, T2D type 2 diabetes, BMI body mass index, FPG fasting plasma glucose, HbA1c Hemoglobin A1c, TC total cholesterol, TGs triglyceridesFig. 5ROC curve ## Discussion In the current study, the diagnosis of diabetic gastroparesis was based on GCSI Score and transabdominal ultrasonography. Although gastric scintigraphy is the gold standard for diagnosis of gastroparesis [24, 25], ultrasonography is a radiation-free, readily available and a valid reliable imaging approach [26–28]. With transabdominal ultrasonography, gastric emptying and motility index were negatively correlated with female sex, duration of T2D, BMI, SPB, DBP, retinopathy, neuropathy, nephropathy, FPG, HbA1c, TC, TGs and grade of gastroparesis whereas, gastric emptying and motility index were not significantly correlated with GCSI Score. In line, Sogabe et al. [ 29] showed that gastric emptying and motility index values were significantly correlated with FPG. Authors concluded that the achievement of glycemic control improves both of gastric motility and gastrointestinal symptoms in patients with diabetic gastroparesis. Consistent with our results, Steinsvik et al. [ 30] found no significant associations between symptoms of gastroparesis and measurements of ultrasonography in patients with diabetic gastroparesis. In contrast, Darwiche et al. [ 31] found no significant associations between gastric emptying and the duration of diabetes, HbA1c, age or BMI; these incompatible findings are probably due to their small sample size. In the present study, vitamin B12 level was significantly lower in patients with T2D than in healthy controls moreover, it was significantly lower in T2D patients with gastroparesis than in those without gastroparesis. Of interest, vitamin B12 was negatively correlated with GCSI Score whereas, it was positively correlated with gastric emptying and motility index. Additionally, vitamin B12 deficiency was an independent predictor for gastroparesis in patients with T2D. Vitamin B12 predicts gastroparesis at a cutoff value of 189.5 pmol/L with $69.1\%$ sensitivity, $64.4\%$ specificity, $70.4\%$ PPV and $63\%$ NPV, $$P \leq 0.002.$$ In the current study, the frequency of vitamin B12 deficiency in total patients with T2D was $35\%$ ($54.5\%$ in patients with gastroparesis and $11.1\%$ in patients without gastroparesis). However, our findings are much higher than estimates from a study conducted by Amjad et al. [ 32] where vitamin B12 deficiency was detected in $17.5\%$ of patients with gastroparesis either diabetic or non-diabetic. The definition of vitamin B12 deficiency varies between studies; the cutoff point used in this study was 125 pmol/L which is comparable with Hansen et al. [ 16] however, it is low compared with what is used in other studies [33, 34]. The variability of cutoff limit for vitamin B12 could be explained by heterogeneity in age and race in the study populations. Vitamin B12 deficiency is a major public health problem caused by age, consumption of vegetarian diets, malabsorption and drugs such as chronic use of omeprazole and metformin [35–38]. An adequate vitamin B12 is essential for the proper functioning of the nervous system through maintenance of the myelin nerve sheaths [39, 40] therefore, vitamin B12 deficiency induces neurological disorders such as peripheral neuropathy [41, 42]. Moreover, vitamin B12 is used in the treatment of peripheral neuropathy [43]. Furthermore, vitamin B12 affects the dynamics of autonomic nervous system [15] and a significant association between vitamin B12 deficiency and cardiovascular autonomic neuropathy has been recently reported [16]. In the light of these findings, we hypothesized that vitamin B12 deficiency may be also implicated in the development of diabetic gastroparesis. To our knowledge, this is the first study to indicate the independent association between vitamin B12 and diabetic gastroparesis. In our study participants, the independent predictors of gastroparesis other than vitamin B12 deficiency were female sex, duration of diabetes, BMI, diabetic microvasascular complications, FPG, HbA1c, TGs and TC which are consistent with the existing literature [44–46]. In the current study, we did not observe an association between metformin treatment and diabetic gastroparesis. The increased frequency of vitamin B12 deficiency among patients with T2D taking metformin has been previously reported [47, 48], however this association depends on both the dose and the duration of treatment [49, 50]. It is believed that metformin induces vitamin B12 deficiency 5–10 years after treatment initiation due to late depletion of body storages [50]. In our study population, the duration of treatment was less than 5 years which may explain our finding. Finally, in addition to sex, duration of diabetes, BMI, diabetic microvasascular complications, FPG, HbA1c, TGs and TC, vitamin B12 deficiency was an independent predictor of diabetic gastroparesis. From the previous discussion 2 raising questions arise which needs further studies. First, whether gastroparesis leads to vitamin B12 deficiency therefore, vitamin B12 may be a cause and a consequence of diabetic gastroparesis. Second, whether vitamin B12 supplementation can improve diabetic gastroparesis. ## In conclusion Beside the known risk factors of diabetic gastroparesis, vitamin B12 deficiency is an independent predictor of diabetic gastroparesis in patients with T2D. ## References 1. Parkman HP, Hasler WL, Fisher RS. American gastroenterological association medical position statement: diagnosis and treatment of gastroparesis. Gastroenterology* (2004) 127 1589-1591. DOI: 10.1053/j.gastro.2004.09.054 2. Camilleri M, Parkman HP, Shafi MA, Abell TL, Gerson L. Clinical guideline: management of gastroparesis. Am J Gastroenterol (2013) 108 18-37. DOI: 10.1038/ajg.2012.373 3. Camilleri M, Bharucha AE, Farrugia G. Epidemiology, mechanisms, and management of diabetic gastroparesis. Clin Gastroenterol Hepatol (2011) 9 5-12. DOI: 10.1016/j.cgh.2010.09.022 4. Bharucha AE, Kudva Y, Basu A, Camilleri M, Low PA, Vella A. Relationship between glycemic control and gastric emptying in poorly controlled type 2 diabetes. Clin Gastroenterol Hepatol (2015) 13 466-76.e1. DOI: 10.1016/j.cgh.2014.06.034 5. Jung HK, Choung RS, Locke GR, Schleck CD, Zinsmeister AR, Szarka LA. The incidence, prevalence, and outcomes of patients with gastroparesis in Olmsted County, Minnesota, from 1996 to 2006. Gastroenterology (2009) 136 1225-1233. DOI: 10.1053/j.gastro.2008.12.047 6. Horowitz M, Harding PE, Maddox AF, Wishart JM, Akkermans LM, Chatterton BE. Gastric and oesophageal emptying in patients with type 2 (non-insulin-dependent) diabetes mellitus. Diabetologia (1989) 32 151-159. DOI: 10.1007/BF00265086 7. Intagliata N, Koch KL. Gastroparesis in type 2 diabetes mellitus: prevalence, etiology, diagnosis, and treatment. Curr Gastroenterol Rep (2007) 9 270-279. DOI: 10.1007/s11894-007-0030-3 8. Koch KL, Calles-Escandón J. Diabetic gastroparesis. Gastroenterol Clin North Am (2015) 44 39-57. DOI: 10.1016/j.gtc.2014.11.005 9. Choung RS, Locke GR, Schleck CD, Zinsmeister AR, Melton LJ, Talley NJ. Risk of gastroparesis in subjects with type 1 and 2 diabetes in the general population. Am J Gastroenterol (2012) 107 82-88. DOI: 10.1038/ajg.2011.310 10. Bharucha AE, Camilleri M, Low PA, Zinsmeister AR. Autonomic dysfunction in gastrointestinal motility disorders. Gut (1993) 34 397-401. DOI: 10.1136/gut.34.3.397 11. Rayner CK, Samsom M, Jones KL, Horowitz M. Relationships of upper gastrointestinal motor and sensory function with glycemic control. Diabetes Care (2001) 24 371-381. DOI: 10.2337/diacare.24.2.371 12. Kashyap P, Farrugia G. Diabetic gastroparesis: what we have learned and had to unlearn in the past 5 years. Gut (2010) 59 1716-1726. DOI: 10.1136/gut.2009.199703 13. Hyett B, Martinez FJ, Gill BM, Mehra S, Lembo A, Kelly CP. Delayed radionucleotide gastric emptying studies predict morbidity in diabetics with symptoms of gastroparesis. Gastroenterology (2009) 137 445-452. DOI: 10.1053/j.gastro.2009.04.055 14. Jayashri R, Venkatesan U, Rohan M, Gokulakrishnan K, Shanthi Rani CS, Deepa M. Prevalence of vitamin B12 deficiency in South Indians with different grades of glucose tolerance. Acta Diabetol (2018) 55 1283-1293. DOI: 10.1007/s00592-018-1240-x 15. Karnaze DS, Carmel R. Neurologic and evoked potential abnormalities in subtle cobalamin deficiency states, including deficiency without anemia and with normal absorption of free cobalamin. Arch Neurol (1990) 47 1008-1012. DOI: 10.1001/archneur.1990.00530090082017 16. Hansen CS, Jensen JS, Ridderstråle M, Vistisen D, Jørgensen ME, Fleischer J. Vitamin B12 deficiency is associated with cardiovascular autonomic neuropathy in patients with type 2 diabetes. J Diabetes Compl (2017) 31 202-208. DOI: 10.1016/j.jdiacomp.2016.08.025 17. Revicki DA, Rentz AM, Dubois D, Kahrilas P, Stanghellini V, Talley NJ. Gastroparesis cardinal symptom index (GCSI): development and validation of a patient reported assessment of severity of gastroparesis symptoms. Qual Life Res (2004) 13 833-844. DOI: 10.1023/B:QURE.0000021689.86296.e4 18. Abell TL, Bernstein RK, Cutts T, Farrugia G, Forster J, Hasler WL. Treatment of gastroparesis: a multidisciplinary clinical review. Neurogastroenterol Motil (2006) 18 263-283. DOI: 10.1111/j.1365-2982.2006.00760.x 19. Kisozi T, Mutebi E, Kisekka M, Lhatoo S, Sajatovic M, Kaddumukasa M. Prevalence, severity and factors associated with peripheral neuropathy among newly diagnosed diabetic patients attending Mulago hospital: a cross-sectional study. Afr Health Sci (2017) 17 463-473. DOI: 10.4314/ahs.v17i2.21 20. Meijer JW, Smit AJ, Sonderen EV, Groothoff JW, Eisma WH, Links TP. Symptom scoring systems to diagnose distal polyneuropathy in diabetes: the diabetic neuropathy symptom score. Diabet Med (2002) 19 962-965. DOI: 10.1046/j.1464-5491.2002.00819.x 21. Umanath K, Lewis JB. Update on diabetic nephropathy: core curriculum 2018. Am J Kidney Dis (2018) 71 884-895. DOI: 10.1053/j.ajkd.2017.10.026 22. Friedewald WT, Levy RT, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol without the use of preparative ultracentrifuge. Clin Chem (1972) 18 499-502. DOI: 10.1093/clinchem/18.6.499 23. Fujimura J, Haruma K, Hata J, Yamanaka H, Sumii K, Kajiyama G. Quantitation of duodenogastric reflux and antral motility by color Doppler ultrasonography. Study in healthy volunteers and patients with gastric ulcer. Scand J Gastroenterol. (1994) 29 897-902. DOI: 10.3109/00365529409094860 24. Abell TL, Camilleri M, Donohoe K, Hasler WL, Lin HC, Maurer AH. Consensus recommendations for gastric emptying scintigraphy: a joint report of the American neurogastroenterology and motility society and the society of nuclear medicine. Am J Gastroenterol (2008) 103 753-763. DOI: 10.1111/j.1572-0241.2007.01636.x 25. Orthey P, Yu D, Van Natta ML, Ramsey FV, Diaz JR, Bennett PA. Intragastric meal distribution during gastric emptying scintigraphy for assessment of fundic accommodation: correlation with symptoms of gastroparesis. J Nucl Med (2018) 59 691-697. DOI: 10.2967/jnumed.117.197053 26. Irvine EJ, Tougas G, Lappalainen R, Bathurst NC. Reliability and interobserver variability of ultrasonographic measurement of gastric emptying rate. Dig Dis Sci (1993) 38 803-810. DOI: 10.1007/BF01295904 27. Szarka LA, Camilleri M. Gastric emptying. Clin Gastroenterol Hepatol (2009) 7 823-827. DOI: 10.1016/j.cgh.2009.04.011 28. Muresan C, Surdea Blaga T, Muresan L, Dumitrascu DL. Abdominal ultrasound for the evaluation of gastric emptying revisited. J Gastrointestin Liver Dis (2015) 24 329-338. DOI: 10.15403/jgld.2014.1121.243.mur 29. Sogabe M, Okahisa T, Tsujigami K, Okita Y, Hayashi H, Taniki T. Ultrasonographic assessment of gastric motility in diabetic gastroparesis before and after attaining glycemic control. J Gastroenterol (2005) 40 583-590. DOI: 10.1007/s00535-005-1592-1 30. Steinsvik EK, Sangnes DA, Søfteland E, Biermann M, Assmus J, Dimcevski G. Gastric function in diabetic gastroparesis assessed by ultrasound and scintigraphy. Neurogastroenterol Motil (2022) 34 e14235. DOI: 10.1111/nmo.14235 31. Darwiche G, Mohammed SK, Aldawi N, Skaria S, Tesfa Y. Gastroparesis among type 1 and type 2 diabetic patients in the United Arab emirates. J Diabetes Mellitus (2014) 4 96-106. DOI: 10.4236/jdm.2014.42016 32. Amjad W, Qureshi W, Singh RR, Richter S. Nutritional deficiencies and predictors of mortality in diabetic and nondiabetic gastroparesis. Ann Gastroenterol (2021) 34 788-795. PMID: 34815644 33. Hannibal L, Lysne V, Bjørke-Monsen AL, Behringer S, Grünert SC, Spiekerkoetter U. Biomarkers and algorithms for the diagnosis of vitamin B12 deficiency. Front Mol Biosci (2016) 3 27. DOI: 10.3389/fmolb.2016.00027 34. Aparicio-Ugarriza R, Palacios G, Alder M, González-Gross M. A review of the cut-off points for the diagnosis of vitamin B12 deficiency in the general population. CCLM (2015) 53 1149-1159. DOI: 10.1515/cclm-2014-0784 35. Vidal-Alaball J, Butler C, Cannings-John R, Goringe A, Hood K, McCaddon A. Oral vitamin B12 versus intramuscular vitamin B12 for vitamin B12 deficiency. Cochrane Database Syst Rev (2005) 20 CD004655 36. Surendran S, Adaikalakoteswari A, Saravanan P, Shatwaan IA, Lovegrove JA, Vimaleswaran KS. An update on vitamin B12-related gene polymorphisms and B12 status. Genes Nutr (2018) 13 2. DOI: 10.1186/s12263-018-0591-9 37. Ortiz-Guerrero G, Amador-Munoz D, Calderon-Ospina CA, LopezFuentes D, Nava Mesa MO. Proton pump inhibitors and dementia: physiopathological mechanisms and clinical consequences. Neural Plast (2018) 2018 5257285. DOI: 10.1155/2018/5257285 38. Kim J, Ahn CW, Fang S, Lee HS, Park JS. Association between metformin dose and vitamin B12 deficiency in patients with type 2 diabetes. Medicine (2019) 98 e17918. DOI: 10.1097/MD.0000000000017918 39. Reynolds E. Vitamin B12, folic acid, and the nervous system. Lancet Neurol (2006) 5 949-960. DOI: 10.1016/S1474-4422(06)70598-1 40. Calderon-Ospina CA, Nava-Mesa MO. B Vitamins in the nervous system: current knowledge of the biochemical modes of action and synergies of thiamine, pyridoxine, and cobalamin. CNS Neurosci Ther (2020) 26 5-13. DOI: 10.1111/cns.13207 41. McCombe PA, McLeod JG. The peripheral neuropathy of vitamin B12 deficiency. J Neurol Sci (1984) 66 117-126. DOI: 10.1016/0022-510X(84)90147-3 42. Healton EB, Savage DG, Brust JC, Garrett TJ, Lindenbaum J. Neurologic aspects of cobalamin deficiency. Medicine (1991) 70 229-245. DOI: 10.1097/00005792-199107000-00001 43. Hakim M, Kurniani N, Pinzon RT, Tugasworo D, Basuki M, Haddani H. Management of peripheral neuropathy symptoms with a fixed dose combination of high-dose vitamin B1, B6 and B12: a 12-week prospective non-interventional study in Indonesia. Asian J Med Sci (2018) 9 32-40. DOI: 10.3126/ajms.v9i1.18510 44. Parkman HP, Wilson LA, Farrugia G, Koch KL, Hasler WL, Nguyen LA. Delayed gastric emptying associates with diabetic complications in diabetic patients with symptoms of gastroparesis. Am J Gastroenterol (2019) 114 1778-94. DOI: 10.14309/ajg.0000000000000410 45. Almogbel RA, Alhussan FA, Alnasser SA, Algeffari MA. Prevalence and risk factors of gastroparesis-related symptoms among patients with type 2 diabetes. Int J Health Sci (Qassim) (2016) 10 397-404. PMID: 27610063 46. Adar T, Lysy J. Pseudodyslipidemia: are we over-treating dyslipidemia in diabetic patients with undiagnosed gastroparesis?. Endocrine (2014) 45 26-27. DOI: 10.1007/s12020-013-0064-2 47. Nervo M, Lubini A, Raimundo FV, Faulhaber GA, Leite C, Fischer LM. Vitamin B12 in metformin-treated diabetic patients: a cross-sectional study in Brazil. Rev Assoc Med Bras (2011) 57 46-49. DOI: 10.1016/S0104-4230(11)70015-X 48. de Groot-Kamphuis DM, van Dijk PR, Groenier KH, Houweling ST, Bilo HJ, Kleefstra N. Vitamin B12 deficiency and the lack of its consequences in type 2 diabetes patients using metformin. Neth J Med (2013) 71 386-390. PMID: 24038568 49. Ting RZ, Szeto CC, Chan MH, Ma KK, Chow KM. Risk factors of vitamin B (12) deficiency in patients receiving metformin. Arch Intern Med (2006) 166 1975-1979. DOI: 10.1001/archinte.166.18.1975 50. Andres E, Noel E, Goichot B. Metformin-associated vitamin B12 deficiency. Arch Intern Med (2002) 162 2251-2252. DOI: 10.1001/archinte.162.19.2251-a
--- title: 'Systemic inflammation indicators and risk of incident arrhythmias in 478,524 individuals: evidence from the UK Biobank cohort' authors: - Xiaorong Yang - Shaohua Zhao - Shaohua Wang - Xuelei Cao - Yue Xu - Meichen Yan - Mingmin Pang - Fan Yi - Hao Wang journal: BMC Medicine year: 2023 pmcid: PMC9976398 doi: 10.1186/s12916-023-02770-5 license: CC BY 4.0 --- # Systemic inflammation indicators and risk of incident arrhythmias in 478,524 individuals: evidence from the UK Biobank cohort ## Abstract ### Background The role of systemic inflammation in promoting cardiovascular diseases has attracted attention, but its correlation with various arrhythmias remains to be clarified. We aimed to comprehensively assess the association between various indicators of systemic inflammation and atrial fibrillation/flutter (AF), ventricular arrhythmia (VA), and bradyarrhythmia in the UK Biobank cohort. ### Methods After excluding ineligible participants, a total of 478,524 eligible individuals ($46.75\%$ male, aged 40–69 years) were enrolled in the study to assess the association between systemic inflammatory indicators and each type of arrhythmia. ### Results After covariates were fully adjusted, CRP levels were found to have an essentially linear positive correlation with the risk of various arrhythmias; neutrophil count, monocyte count, and NLR showed a non-linear positive correlation; and lymphocyte count, SII, PLR, and LMR showed a U-shaped association. VA showed the strongest association with systemic inflammation indicators, and it was followed sequentially by AF and bradyarrhythmia. ### Conclusions Multiple systemic inflammatory indicators showed strong associations with the onset of AF, VA, and bradyarrhythmia, of which the latter two have been rarely studied. Active systemic inflammation management might have favorable effects in reducing the arrhythmia burden and further randomized controlled studies are needed. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12916-023-02770-5. ## Background Arrhythmias are a global challenge to human health [1]. The prevalence of arrhythmias is estimated to be $1.5\%$ to $5\%$ in the general population [2] and increases rapidly with age [1–4]. Atrial fibrillation/flutter (AF), ventricular arrhythmia (VA), and bradyarrhythmia are the most common types of arrhythmias that cause serious adverse outcomes [5, 6]. It is important to assess for risk factors to reduce the likelihood of arrhythmias onset in the first place. Systemic inflammation, the result of the release of proinflammatory cytokines and chronic activation of the innate immune system, has been implicated in the development of some chronic diseases [7–11]. Although there is emerging evidence on the role of inflammatory dysregulation in AF [12–16], those studies usually used a single biomarker and showed inconsistent results. Furthermore, the relationship between inflammatory markers and VA/bradyarrhythmia is rarely reported. Large prospective studies that involve multitudinous inflammatory indicators and provide high-level evidence are needed to systematically explore the association between inflammation and different types of arrhythmias. The UK Biobank (UKB) is a prospective cohort containing in-depth health information. Using this large-scale database, we systematically explored the relationship between eight systemic inflammation indicators and the incidence of AF, VA, and bradyarrhythmia. ## Study population The UKB enrolled more than 500,000 participants aged 40 to 69 years in 2006–2010 [17]. In brief, the clinical, genetic, and biochemical data of participants were obtained through questionnaires, genotyping, sample assays, physical measures, and linked electronic health data. Health outcomes were tracked for all the participants through linkage to national e-health-related datasets. Ethical approval for the UKB study was obtained from the National Information Governance Board for Health and Social Care and the National Health Service North West Multicenter Research Ethics Committee. Participants provided their written informed consent at baseline. This study utilized UKB resources under application number 82232 and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. In the current analysis, we excluded participants who subsequently withdrew their consent ($$n = 158$$); those with AF, VA, or bradyarrhythmia at baseline ($$n = 11$$,773); and those for whom information on systemic inflammatory markers was missing ($$n = 11$$,959). As a result, the primary analysis included a final group of 478,524 participants (Fig. 1).Fig. 1The flow diagram of the UK biobank participants in this study ## Systemic inflammation indicators The quality check procedure for blood sample data carried out at UKB is available at https://biobank.ndph.ox.ac.uk/showcase/showcase/docs/biomarker_issues.pdf. The instrument reports 31 parameters, the details of which are available at https://biobank.ndph.ox.ac.uk/showcase/ukb/docs/haematology.pdf. We extracted baseline count data for neutrophils, monocytes, lymphocytes, and platelets. Based on the blood cells counts, we calculated the values of combined inflammation indicators, including the neutrophil-to-lymphocyte ratio (NLR, neutrophils/lymphocytes), lymphocyte-to-monocyte ratio (LMR, lymphocytes/monocytes), platelet-to-lymphocyte ratio (PLR, platelets/lymphocytes), and systemic immune-inflammation index (SII, neutrophils × platelets/lymphocytes), which have been demonstrated to predict inflammatory status under several conditions in previous studies [18–20]. Moreover, serum C-reactive protein (CRP) was included in the current study and was detected by immunoturbidimetric high-sensitivity assays on a Beckman Coulter AU5800. ## Outcome ascertainment The primary outcomes of interest were AF, VA, and bradyarrhythmia, according to follow-up data obtained up to January 2022. Incident AF, VA, and bradyarrhythmia were identified based on the International Classification of Diseases, Ninth Revision (ICD-9) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes available from inpatient and outpatient records and causes of death for a linked medical encounter. We also used the occurrence of a relevant operative procedure for each arrhythmias subtype. Detailed disease definitions are provided in Additional file 1: Table S1. ## Assessment of covariates Race and ethnicity were ascertained as basic demographic variables and classified according to self-reports from the interviews. For the education levels, the College or University degree was described as College degree; the A levels/AS levels or equivalent was described as High school graduate; the O levels/GCSEs or equivalent was described as Middle school graduate, and other types were described as None of the above. The Townsend deprivation index was obtained based on participants’ zip codes: higher values of the index indicated higher levels of deprivation. Body mass index (BMI) was derived from measurements of height and weight obtained during the initial assessment and was calculated by dividing weight (kilograms) by height (meters) squared. Physical activity was categorized into tertiles according to participants’ responses to the question on the number of days of moderate physical activity for over 10 min per week. Participants were classified as non-drinkers, light-to-moderate drinkers, or heavy drinkers based on their self-reported average daily alcohol consumption. The presence of hypertension, myocardial infarction, angina, stroke, diabetes, hyperlipidemia, diseases of blood and blood-forming organs (DBBF), chronic diseases involving the immune mechanism, and malignant neoplasms at baseline was determined based on self-reported diseases and the ICD-9 and ICD-10 codes from inpatient and/or outpatient visits before the date of attending the assessment center. The definitions of these diseases are provided in Additional file 1: Table S2. ## Statistical analysis Continuous variables are represented by mean and standard deviation, and categorical variables are represented by proportion. We used Cox proportional hazard ratio (HR) and $95\%$ confidence interval (CI) to assess the association between systemic inflammatory indicators and each type of arrhythmia. Three major models were fitted: model 1 included age (continuous variable), sex, and race as covariates; model 2 included additional potential confounders, such as education, Townsend deprivation index, smoking status, frequency of alcohol drinking, BMI, and physical activity; and model 3 further included related diseases at baseline, including hypertension, myocardial infarction, angina, stroke, diabetes, hyperlipidemia, DBBF, chronic diseases involving the immune mechanism, and malignant neoplasms (Table 1). With systemic inflammation indicators as a continuous exposure variable, we used restricted cubic splines with 5 knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles to assess the potential non-linear effect of systemic inflammation status on arrhythmias, after shrinking 1‰ outliers of systemic inflammation indicators. To easily compare the associations of various systemic inflammation indicators with the three arrhythmia subtypes for different models and conditions, the systemic inflammation indicators were further divided into seven categories, taking into account the normal reference range, data distribution, and easy-to-understand numbers. The following sensitivity analyses were performed: [1] Participants with events in the first 2 years of follow-up were excluded to mitigate any potential effects of reverse causality. [ 2] Participants with malignant neoplasms, DBBF, heart diseases, and chronic diseases involving the immune mechanism at baseline were excluded to mitigate any potential bias due to survivorship. All analyses were conducted using STATA version 15.1 (Stata Corporation, College Station, TX, USA). Two-tailed p-values less than 0.05 were considered to indicate significance. Table 1Demographic and clinical characteristics of participants in a study of arrhythmias in the UK BiobankCharacteristicTotalN=478,524None of arrhythmiasN=445,647Any arrhythmiasN=32,877Atrial fibrillation/flutterN=24,484Ventricular arrhythmiaN=3789BradyarrhythmiaN=10,527Age (years)56.4 ± 8.156.0 ± 8.161.7 ± 6.362.0 ± 6.060.3 ± 7.161.8 ± 6.3Female, n (%)261,564 (54.66)249,713 (56.03)11,851 (36.05)9139 (37.33)1155 (30.48)3237 (30.75)White race, n (%)450,543 (94.60)418,996 (94.46)31,547 (96.53)23,701 (97.34)3544 (94.13)9974 (95.44)Townsend deprivation index−1.3 ± 3.1−1.3 ± 3.1−1.2 ± 3.2−1.2 ± 3.2−0.8 ± 3.3−1.2 ± 3.1Educational level, n (%) College degree155,367 (32.86)147,043 (33.38)8324 (25.70)6113 (25.34)940 (25.26)2727 (26.29) High school graduate53,340 (11.28)50,480 (11.46)2860 (8.83)2119 (8.78)349 (9.38)863 (8.32) Middle school graduate101,327 (21.43)94,899 (21.55)6428 (19.85)4817 (19.97)726 (19.51)2049 (19.76) None of the above162,808 (34.43)148,033 (33.61)14,775 (45.62)11,074 (45.91)1707 (45.86)4732 (45.86)Body mass index (kg/m2)27.39 ± 4.827.3 ± 4.728.9 ± 5.329.1 ± 5.428.6 ± 5.228.8 ± 5.0Smoking status, n (%) Never261,551 (54.94)246,970 (55.69)14,581 (44.66)10,736 (44.17)1539 (40.90)4835 (46.27) Former164,013 (34.45)149,806 (33.78)14,207 (43.52)10,768 (44.30)1592 (42.31)4544 (43.49) Current50,525 (10.61)46,665 (10.52)3860 (11.82)2803 (11.53)632 (16.80)1070 (10.24)Daily alcohol, n (%) Daily or almost daily96,967 (20.31)89,261 (20.07)7706 (23.51)5903 (24.18)857 (22.68)2357 (22.46) Three or four times a week110,464 (23.14)103,313 (23.23)7151 (21.82)5318 (21.79)807 (21.36)2295 (21.87) Once or twice a week123,611 (25.89)115,826 (26.05)7785 (23.75)5832 (23.89)897 (23.74)2531 (24.11) Less once a week108,252 (22.67)101,181 (22.75)7071 (21.57)5153 (21.11)820 (21.70)2315 (22.06) Never38,175 (8.00)35,108 (7.89)3067 (9.36)2204 (9.03)397 (10.51)998 (9.51)Physical activity, n (%) Light160,870 (35.49)150,391 (35.59)10,479 (34.21)7780 (34.14)1249 (35.88)3280 (33.35) Moderate113,373 (25.01)105,933 (25.07)7440 (24.29)5576 (24.47)831 (23.87)2429 (24.70) High178,998 (39.49)166,290 (39.35)12,708 (41.49)9430 (41.39)1401 (40.25)4125 (41.95)Disease history at baseline Hypertension133,643 (27.93)117,606 (26.39)16,037 (48.78)12,162 (49.68)1824 (48.14)5217 (49.56) Heart attack10,038 (2.10)7597 (1.70)2441 (7.43)1692 (6.91)488 (12.88)916 (8.70) Angina17,719 (3.70)14,001 (3.14)3718 (11.31)2681 (10.95)533 (14.07)1388 (13.19) Stroke7360 (1.54)6071 (1.36)1289 (3.92)1001 (4.09)146 (3.85)407 (3.87) Diabetes25,316 (5.29)21,439 (4.81)3877 (11.79)2772 (11.32)548 (14.46)1453 (13.80) Hyperlipidemia64,747 (13.53)56,398 (12.66)8349 (25.40)6120 (25.00)1042 (27.51)2966 (28.19) Diseases of blood and blood-forming organs (DBBF)4320 (0.90)3846 (0.86)474 (1.44)375 (1.53)74 (1.95)109 (1.04) Chronic diseases involving the immune mechanism65,347 (13.66)60,361 (13.55)4986 (15.17)3749 (15.32)618 (16.31)1526 (14.50) Malignant neoplasms41,796 (8.74)37,867 (8.50)3929 (11.95)3046 (12.44)423 (11.17)1135 (10.79)Systemic inflammation indicators C-reactive protein (CRP, mg/L)2.59 ± 4.342.53 ± 4.253.30 ± 5.323.36 ± 5.373.58 ± 6.023.07 ± 4.80 Neutrophil count (10^9 cells/L)4.22 ± 1.424.20 ± 1.414.49 ± 1.524.49 ± 1.524.67 ± 1.664.45 ± 1.48 Monocyte count (10^9 cells/L)0.47 ± 0.270.47 ± 0.270.52 ± 0.300.52 ± 0.330.54 ± 0.220.52 ± 0.22 Lymphocyte count (10^9 cells/L)1.97 ± 1.151.97 ± 1.151.96 ± 1.191.96 ± 1.271.98 ± 1.041.97 ± 1.01 Systemic immune-inflammation index (SII)598.99 ± 367.44596.47 ± 355.89633.07 ± 496.89635.53 ± 525.87673.37 ± 762.95616.75 ± 391.01 Neutrophil-to-lymphocyte ratio (NLR)2.35 ± 1.242.34 ± 1.202.57 ± 1.762.59 ± 1.882.71 ± 2.902.53 ± 1.32 Platelet-to-lymphocyte ratio (PLR)142.07 ± 69.11142.23 ± 67.88139.96 ± 84.00140.18 ± 90.12142.69 ± 148.84137.89 ± 60.54 Lymphocyte-to-monocyte ratio (LMR)4.64 ± 4.264.67 ± 4.324.23 ± 3.174.21 ± 3.124.22 ± 4.044.21 ± 2.86Follow-up years (years)12.21 ± 2.3512.58 ± 1.757.17 ± 3.327.13 ± 3.367.24 ± 3.287.79 ± 3.06 ## Population characteristics The baseline characteristics and systemic inflammatory indicators of participants [478,524] are presented according to the presence of arrhythmias and the subtypes (Table 1). In general, the mean age of all the participants was 56.4 ± 8.1 years, and 261,564 ($46.75\%$) were male. Over a mean follow-up period of 12.2 years, 32,877 participants developed arrhythmias, including 24,484 cases of AF, 3789 cases of VA, and 10,527 cases of bradyarrhythmia. Compared with the 445,647 participants in the control group, the 32,877 participants in the incident arrhythmia group were more likely to be older, male, and less educated, and tended to have higher BMI, higher systolic blood pressure, a higher smoking rate, and a higher prevalence of comorbid diseases (Table 1). Considering the possible correlation between systemic inflammation levels and baseline characteristics, we have shown the baseline characteristics of patients for different levels of CRP (as an indicator of systemic inflammation). Participants with higher CRP levels were more likely to be older, less educated, and smokers, and tended to have a higher Townsend deprivation index, higher BMI, higher systolic and diastolic blood pressure, lower alcohol consumption, lower physical activity intensity, and higher prevalence of comorbid diseases (Additional file 1: Table S3). ## Atrial fibrillation/flutter There were 24,484 incident AF events across 5.88 million person-years of follow-up (incidence rate: 4.16 events per 1000 person-years, $95\%$ CI: 4.11–4.21). After adjusting for all potential confounding variables, the CRP levels were significantly and positively associated with the risk of incident AF (Fig. 2A). Compared with the reference population with a CRP of <0.5 mg/L, the risk of incident AF in the population with CRP >10 mg/L was 1.33 ($95\%$ CI: 1.24–1.43) (Table 2). Moreover, we found that the HR for incident AF increased significantly with an increase in the neutrophil count, monocyte count, and NLR, although a slight opposite trend (not statistically significant) was found at the low neutrophil count, monocyte count, and NLR, (Fig. 2B, C, and F; Table 2).Fig. 2Multivariable-adjusted association between different systematic information indicators and the risk of atrial fibrillation/flutter by restricted cubic spline regression. A C-reactive protein; B neutrophil count; C monocyte count; D lymphocyte count; E systemic immune-inflammation index (neutrophils × platelets/lymphocytes); F neutrophil-to-lymphocyte ratio (neutrophils/lymphocytes); G platelet-to-lymphocyte ratio (platelets/lymphocytes); H lymphocyte-to-monocyte ratio (lymphocytes/monocytes). HR, hazard ratio; CI, confidence intervalsTable 2The association between various systematic information indicators and three arrhythmia subtypesVariablesAtrial fibrillation/flutterVentricular arrhythmiaBradyarrhythmiaModel 1Model 2Model 3Model 1Model 2Model 3Model 1Model 2Model 3C-reactive protein (mg/L) <0.51.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) ( 0.5, 1.0)1.10 (1.05~1.16)0.99 (0.94~1.04)1.00 (0.95~1.05)1.10 (0.97~1.25)1.03 (0.90~1.17)1.06 (0.93~1.20)1.11 (1.04~1.20)1.02 (0.94~1.09)1.04 (0.97~1.13) (1.0, 2.0)1.21 (1.16~1.27)0.98 (0.94~1.03)1.00 (0.95~1.05)1.32 (1.17~1.48)1.15 (1.01~1.30)1.20 (1.06~1.36)1.16 ((1.08~1.24)0.98 (0.91~1.06)1.03 (0.96~1.11) (2.0, 3.0)1.37 (1.30~1.44)1.02 (0.96~1.08)1.04 (0.98~1.10)1.32 (1.16~1.51)1.06 (0.92~1.23)1.13 (0.98~1.30)1.26 (1.17~1.37)1.01 (0.93~1.10)1.07 (0.98~1.16) (3.0, 4.0)1.47 (1.39~1.56)1.04 (0.97~1.10)1.06 (0.99~1.13)1.54 (1.32~1.79)1.20 (1.02~1.42)1.28 (1.09~1.51)1.42 (1.30~1.55)1.08 (0.98~1.19)1.15 (1.05~1.27) (4.0, 10.0)1.72 (1.63~1.81)1.12 (1.06~1.19)1.14 (1.07~1.20)1.92 (1.69~2.18)1.37 (1.19~1.58)1.45 (1.26~1.67)1.54 (1.42~1.66)1.11 (1.02~1.21)1.18 (1.08~1.29) >=10.02.10 (1.97~2.23)1.33 (1.24~1.43)1.33 (1.24~1.43)2.55 (2.18~2.98)1.81 (1.53~2.16)1.87 (1.57~2.22)1.68 (1.52~1.86)1.23 (1.10~1.37)1.30 (1.16~1.45)Neutrophil count (10^9 cells/L) <2.01.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) ( 2.0, 3.0)0.97 (0.86~1.09)1.00 (0.88~1.13)1.02 (0.90~1.16)1.06 (0.77~1.45)1.07 (0.77~1.49)1.10 (0.79~1.53)0.85 (0.72~1.00)0.84 (0.70~1.00)0.85 (0.71~1.01) (3.0, 4.0)1.02 (0.91~1.14)0.98 (0.87~1.11)0.99 (0.88~1.11)1.21 (0.90~1.64)1.13 (0.82~1.56)1.14 (0.83~1.56)0.93 (0.80~1.10)0.89 (0.75~1.05)0.88 (0.74~1.04) (4.0, 5.0)1.14 (1.02~1.28)1.05 (0.93~1.18)1.03 (0.91~1.16)1.34 (0.99~1.81)1.19 (0.87~1.64)1.17 (0.85~1.60)1.02 (0.87~1.20)0.93 (0.79~1.10)0.90 (0.76~1.06) (5.0, 6.0)1.29 (1.15~1.45)1.12 (0.99~1.26)1.08 (0.95~1.22)1.82 (1.34~2.47)1.52 (1.10~2.09)1.43 (1.04~1.97)1.15 (0.98~1.36)1.01 (0.85~1.20)0.94 (0.79~1.11) (6.0, 7.5)1.53 (1.36~1.72)1.25 (1.10~1.41)1.17 (1.03~1.33)2.18 (1.59~2.97)1.73 (1.24~2.40)1.57 (1.13~2.18)1.28 (1.08~1.51)1.07 (0.90~1.28)0.98 (0.82~1.17) >=7.51.88 (1.65~2.13)1.49 (1.30~1.71)1.36 (1.19~1.55)3.35 (2.42~4.64)2.43 (1.72~3.43)2.13 (1.51~3.01)1.42 (1.17~1.71)1.19 (0.98~1.45)1.05 (0.86~1.28)Monocyte count (10^9 cells/L) <0.31.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) ( 0.3, 0.4)0.98 (0.92~1.04)1.00 (0.94~1.06)1.00 (0.94~1.06)0.91 (0.78~1.07)0.93 (0.79~1.09)0.93 (0.79~1.1)0.95 (0.86~1.04)0.97 (0.88~1.07)0.97 (0.88~1.07) (0.4, 0.5)1.03 (0.98~1.09)1.01 (0.95~1.07)1.00 (0.94~1.06)1.01 (0.87~1.17)1.03 (0.88~1.20)1.02 (0.87~1.19)0.98 (0.90~1.07)0.98 (0.90~1.08)0.97 (0.89~1.06) (0.5, 0.6)1.11 (1.05~1.18)1.04 (0.98~1.10)1.02 (0.96~1.08)1.17 (1.01~1.36)1.11 (0.95~1.30)1.09 (0.93~1.27)1.06 (0.97~1.16)1.01 (0.92~1.11)0.99 (0.90~1.08) (0.6, 0.7)1.24 (1.17~1.32)1.11 (1.04~1.19)1.08 (1.01~1.15)1.24 (1.06~1.45)1.11 (0.94~1.31)1.06 (0.90~1.25)1.15 (1.05~1.26)1.07 (0.97~1.18)1.02 (0.93~1.13) (0.7, 0.8)1.40 (1.31~1.50)1.20 (1.11~1.28)1.13 (1.05~1.22)1.57 (1.33~1.86)1.38 (1.15~1.65)1.27 (1.06~1.53)1.34 (1.21~1.48)1.19 (1.07~1.32)1.11 (1.00~1.24) >=0.81.58 (1.47~1.69)1.28 (1.19~1.38)1.18 (1.10~1.27)2.14 (1.81~2.52)1.79 (1.50~2.13)1.58 (1.33~1.89)1.3 (1.17~1.45)1.14 (1.02~1.28)1.03 (0.92~1.15)Lymphocyte count (10^9 cells/L) <0.81.55 (1.37~1.75)1.74 (1.53~1.98)1.56 (1.38~1.78)1.57 (1.17~2.12)1.59 (1.14~2.22)1.39 (0.99~1.95)1.34 (1.11~1.63)1.42 (1.15~1.74)1.33 (1.08~1.64) (0.8, 1.5)1.01 (0.98~1.05)1.15 (1.10~1.19)1.13 (1.09~1.18)1.04 (0.94~1.14)1.18 (1.07~1.31)1.16 (1.05~1.29)0.97 (0.91~1.02)1.04 (0.98~1.11)1.04 (0.98~1.10) (1.5, 2.0)0.99 (0.96~1.03)1.06 (1.02~1.1)1.05 (1.02~1.09)0.89 (0.82~0.98)0.97 (0.88~1.06)0.97 (0.88~1.06)0.99 (0.94~1.05)1.04 (0.98~1.09)1.04 (0.98~1.09) (2.0, 2.5)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) ( 2.5, 3.0)1.04 (0.99~1.09)0.98 (0.93~1.03)0.97 (0.92~1.02)1.09 (0.97~1.23)1.01 (0.90~1.15)0.99 (0.88~1.12)1.02 (0.95~1.10)0.96 (0.89~1.03)0.94 (0.87~1.01) (3.0, 4.0)1.17 (1.10~1.24)1.02 (0.95~1.08)0.99 (0.93~1.06)1.32 (1.14~1.53)1.13 (0.97~1.32)1.08 (0.92~1.26)1.32 (1.21~1.44)1.23 (1.12~1.35)1.18 (1.07~1.29) >=4.01.35 (1.18~1.53)1.17 (1.02~1.35)1.10 (0.96~1.27)1.66 (1.23~2.25)1.39 (1.00~1.93)1.28 (0.92~1.77)1.32 (1.08~1.61)1.17 (0.94~1.46)1.11 (0.89~1.38)Systemic immune-inflammation index (SII) <3001.08 (1.02~1.13)1.07 (1.01~1.13)1.07 (1.01~1.13)1.15 (1.00~1.31)1.15 (1.00~1.32)1.14 (0.99~1.31)1.11 (1.03~1.20)1.10 (1.02~1.19)1.11 (1.02~1.20) [300, 400]1.06 (1.01~1.10)1.05 (1.00~1.10)1.06 (1.01~1.11)1.13 (1.01~1.28)1.13 (1.00~1.27)1.13 (1.00~1.28)1.01 (0.94~1.08)1.00 (0.93~1.07)1.00 (0.93~1.08) [400, 500]1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) [ 500, 600]1.05 (1.01~1.10)1.04 (0.99~1.08)1.03 (0.98~1.08)1.16 (1.03~1.31)1.14 (1.01~1.29)1.14 (1.01~1.29)1.03 (0.96~1.10)1.02 (0.95~1.10)1.02 (0.95~1.09) [600, 800]1.11 (1.06~1.16)1.09 (1.04~1.13)1.07 (1.02~1.11)1.20 (1.07~1.34)1.18 (1.05~1.32)1.16 (1.03~1.30)1.08 (1.01~1.15)1.06 (1,00~1.14)1.05 (0.98~1.12) [800, 1500]1.22 (1.17~1.27)1.18 (1.13~1.23)1.13 (1.08~1.19)1.54 (1.38~1.72)1.46 (1.30~1.65)1.41 (1.25~1.58)1.15 (1.08~1.23)1.12 (1.04~1.20)1.08 (1.01~1.16) >=15001.69 (1.56~1.83)1.62 (1.49~1.77)1.48 (1.36~1.62)2.88 (2.41~3.44)2.65 (2.19~3.21)2.39 (1.98~2.9)1.39 (1.22~1.58)1.37 (1.19~1.58)1.29 (1.12~1.49)Neutrophil-to-lymphocyte ratio (NLR) <1.51.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) ( 1.5, 2.0)1.00 (0.96~1.04)1.00 (0.96~1.05)0.99 (0.95~1.04)1.04 (0.93~1.17)1.02 (0.91~1.15)1.01 (0.90~1.14)0.99 (0.93~1.06)0.99 (0.92~1.06)0.98 (0.91~1.05) (2.0, 2.5)1.07 (1.02~1.12)1.06 (1.02~1.11)1.04 (1.00~1.09)1.14 (1.02~1.28)1.13 (1.00~1.27)1.10 (0.98~1.24)1.04 (0.97~1.11)1.03 (0.97~1.11)1.01 (0.94~1.08) (2.5, 3.0)1.15 (1.10~1.21)1.13 (1.07~1.19)1.10 (1.04~1.15)1.18 (1.04~1.33)1.14 (1.00~1.30)1.10 (0.96~1.25)1.06 (0.99~1.14)1.05 (0.98~1.14)1.02 (0.94~1.10) (3.0, 3.5)1.22 (1.16~1.29)1.21 (1.14~1.28)1.16 (1.10~1.22)1.33 (1.16~1.53)1.29 (1.12~1.49)1.22 (1.06~1.41)1.18 (1.09~1.28)1.17 (1.08~1.28)1.12 (1.03~1.21) (3.5, 5.0)1.36 (1.29~1.44)1.36 (1.29~1.43)1.28 (1.21~1.35)1.69 (1.48~1.92)1.61 (1.41~1.85)1.49 (1.30~1.71)1.19 (1.10~1.29)1.17 (1.08~1.27)1.10 (1.01~1.20) >=5.01.64 (1.53~1.77)1.62 (1.50~1.75)1.46 (1.35~1.58)2.33 (1.97~2.75)2.18 (1.82~2.61)1.91 (1.59~2.28)1.38 (1.23~1.55)1.36 (1.20~1.53)1.23 (1.09~1.39)Platelet-to-lymphocyte ratio (PLR) <801.22 (1.16~1.28)1.11 (1.05~1.17)1.08 (1.02~1.14)1.40 (1.23~1.59)1.22 (1.07~1.40)1.15 (1.00~1.32)1.23 (1.14~1.33)1.14 (1.05~1.24)1.10 (1.01~1.19) [80, 100]1.09 (1.05~1.14)1.05 (1.00~1.10)1.04 (0.99~1.09)1.22 (1.08~1.36)1.12 (1.00~1.27)1.10 (0.97~1.24)1.05 (0.98~1.13)1.02 (0.95~1.09)1.01 (0.94~1.08) [100, 120]1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) [ 120, 150]0.99 (0.95~1.03)1.02 (0.98~1.06)1.02 (0.97~1.06)1.01 (0.91~1.12)1.04 (0.93~1.16)1.04 (0.94~1.16)0.97 (0.91~1.03)0.98 (0.92~1.04)0.98 (0.92~1.05) [150, 200]0.98 (0.94~1.02)1.04 (0.99~1.08)1.04 (0.99~1.08)0.94 (0.85~1.05)1.00 (0.9~1.12)1.01 (0.90~1.13)0.93 (0.87~0.99)0.97 (0.91~1.03)0.98 (0.92~1.04) [200, 250]0.98 (0.92~1.03)1.06 (0.99~1.12)1.04 (0.98~1.11)1.17 (1.02~1.35)1.24 (1.07~1.44)1.24 (1.07~1.44)0.96 (0.88~1.05)1.01 (0.92~1.10)1.01 (0.93~1.11) >=2501.12 (1.05~1.21)1.26 (1.17~1.35)1.19 (1.10~1.28)1.50 (1.27~1.77)1.56 (1.31~1.86)1.47 (1.24~1.76)1.06 (0.95~1.18)1.14 (1.01~1.27)1.11 (0.99~1.24)Lymphocyte-to-monocyte ratio (LMR) <2.51.47 (1.39~1.56)1.49 (1.40~1.58)1.39 (1.30~1.47)1.79 (1.54~2.07)1.82 (1.55~2.13)1.66 (1.42~1.95)1.25 (1.15~1.37)1.23 (1.12~1.35)1.16 (1.06~1.27) (2.5, 3.0)1.23 (1.16~1.31)1.24 (1.17~1.32)1.19 (1.12~1.27)1.54 (1.32~1.79)1.63 (1.39~1.91)1.55 (1.33~1.82)1.06 (0.97~1.15)1.04 (0.94~1.14)1.00 (0.91~1.10) (3.0, 4.0)1.14 (1.08~1.20)1.15 (1.09~1.21)1.12 (1.06~1.19)1.19 (1.04~1.37)1.27 (1.10~1.47)1.24 (1.07~1.43)1.05 (0.98~1.14)1.05 (0.97~1.14)1.03 (0.95~1.11) (4.0, 5.0)1.07 (1.01~1.12)1.08 (1.02~1.14)1.07 (1.01~1.13)1.15 (1.00~1.33)1.2 (1.03~1.39)1.19 (1.03~1.38)0.99 (0.91~1.07)0.98 (0.90~1.06)0.97 (0.89~1.06) (5.0, 6.0)1.04 (0.98~1.10)1.05 (0.99~1.11)1.04 (0.98~1.11)1.05 (0.90~1.22)1.07 (0.91~1.26)1.07 (0.91~1.27)0.98 (0.89~1.07)0.96 (0.87~1.05)0.96 (0.87~1.05) (6.0, 8.0)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.)1.00 (Ref.) >=8.01.05 (0.97~1.14)1.02 (0.94~1.11)1.02 (0.93~1.11)1.32 (1.08~1.61)1.28 (1.03~1.58)1.28 (1.03~1.58)1.08 (0.96~1.22)1.07 (0.94~1.21)1.07 (0.94~1.21)Ref ReferenceModel 1: Adjusted age, sex, and raceModel 2: Adjusted Model 1+townsend deprivation index, education, BMI, smoking status, frequency of alcohol drinking, and physical activityModel 3: Adjusted Model 2+hypertension, heart attack, angina, stroke, diabetes, hyperlipidemia, diseases of blood and blood-forming organs, chronic diseases involving the immune mechanism, and malignant neoplasms A U-shaped relationship was observed between SII, PLR, and the risk of AF (Fig. 2D, E, and G; Table 2). For example, in the case of PLR, the lowest HR for incident AF was found in the [100,120] group, which was set as the reference group, while the highest HR was in the ≥250 group, followed by the <80 group: 1.19 ($95\%$ CI: 1.10~1.28) for the ≥250 group and 1.08 ($95\%$ CI:1.02~1.14) for the <80 group (Table 2). A U-shaped relationship was also observed between lymphocyte count and LMR levels, and the incidence risk of AF, but the HRs were not significant for higher levels, with only a slightly increasing trend observed (Fig. 2D and H; Table 2). ## Ventricular arrhythmias There were 3789 VA events across 5.99 million person-years of follow-up (incidence rate: 0.63 events per 1000 person-years, $95\%$ CI: 0.61–0.65). Figure 3 and Table 2 show the association between systemic inflammation indicator levels and VA risk. The relationship between systemic inflammation indicator levels and VA risk was stronger than the relationship between systemic inflammation indicator levels and AF risk. Fig. 3Multivariable-adjusted association between different systematic information indicators and the risk of ventricular arrhythmias by restricted cubic spline regression. A C-reactive protein; B neutrophil count; C monocyte count; D lymphocyte count; E systemic immune-inflammation index (neutrophils × platelets/lymphocytes); F neutrophil-to-lymphocyte ratio (neutrophils/lymphocytes); G platelet-to-lymphocyte ratio (platelets/lymphocytes); H lymphocyte-to-monocyte ratio (lymphocytes/monocytes). HR, hazard ratio; CI, confidence intervals In general, the risk of incident VA increased monotonically with an increase in the CRP level and neutrophil count (fully adjusted model: HRs = 1.00, 1.06, 1.20, 1.13, 1.28, 1.45, and 1.87 for the increasing CRP groups; HR = 1.00, 1.10, 1.14, 1.17, 1.43, 1.57, and 2.13 for the neutrophil count groups; Figs. 3A and 2B; Table 2). For monocyte count, the HRs for incident VA remained at around 1 in the first five groups (<0.7 × 109 cells/L) and subsequently increased to 1.27 ($95\%$ CI: 1.06–1.53) in the (0.7, 0.8) ×109 cells/L group and 1.58 ($95\%$ CI: 1.33–1.89) in the ≥0.8 × 109 cells/L group (Table 2). A similar association was observed between NLR and the occurrence of VA. A U-shaped association was observed between lymphocyte count, SII, PLR, LMR level, and risk of incident VA (Fig. 3D, E, F, and H). For example, compared with individuals with lymphocyte count at the mid-level ((2.0, 2.5) ×109 cells/L) at baseline, individuals with the lowest (<0.8 × 109 cells/L) and highest (≥4.0 × 109 cells/L) lymphocyte levels had a 1.28 ($95\%$ CI: 0.92~1.77) and 1.39 ($95\%$ CI: 0.99~1.95) times higher chance of being diagnosed with VA during follow-up (Table 2). ## Bradyarrhythmia There were 10,527 incident bradyarrhythmia events across 5.95 million person-years of follow-up (incidence rate: 1.77 events per 1000 person-years, $95\%$ CI: 1.73–1.80). After fully adjusting for covariates, the association of systemic inflammation with the risk of bradyarrhythmia was moderate, and not as strong as its association with AF and VA risk. There was a significant positive correlation between CRP level and incident bradyarrhythmia (Fig. 4A and Table 2). Compared to participants with a lower CRP level (<0.5 mg/L), HR was 1.15 ($95\%$ CI: 1.05–1.27) for the (3.0, 4.0) mg/L group, 1.18 ($95\%$ CI: 1.08–1.29) for the (4.0, 10.0) mg/L group, and 1.3 ($95\%$ CI: 1.16–1.45) for the ≥10.0 mg/L group (Table 2). Overall, the HR for incident bradyarrhythmia tended to increase with an increase in the neutrophil count, monocyte count, and NLR level, after showing a slightly decreasing trend at low neutrophil, monocyte, and NLR levels (Fig. 4B, C, and F; Table 2). Moderate U-shaped correlations were observed between lymphocyte count, SII, PLR, and LMR levels, and HR for incident bradyarrhythmia (Fig. 4D, E, F, and H; Table 2).Fig. 4Multivariable-adjusted association between different systematic information indicators and the risk of bradyarrhythmia by restricted cubic spline regression. A C-reactive protein; B neutrophil count; C monocyte count; D lymphocyte count; E systemic immune-inflammation index (neutrophils × platelets/lymphocytes); F neutrophil-to-lymphocyte ratio (neutrophils/lymphocytes); G platelet-to-lymphocyte ratio (platelets/lymphocytes); H lymphocyte-to-monocyte ratio (lymphocytes/monocytes). HR, hazard ratio; CI, confidence intervals ## Subgroup analyses Malignant neoplasms, DBBF, and chronic diseases involving the immune mechanism are three major diseases that may have direct effects on blood cell counts and CRP levels. Heart disease, hypertension, and hyperlipidemia are diseases that may have an impact on the outcome of arrhythmias. Subgroup analysis was therefore performed for the above populations, except for the DBBF group, for which we performed a sensitivity analysis because of its small sample size. In order to assess potential effect modification by age at baseline and sex, we further conducted subgroup analysis by age groups at baseline (less than 60 years old vs. 60 years old or older) and sex. In all subgroups, the associations between systemic inflammation indicators and the risk of arrhythmia outcomes were similar to those in the overall population analysis described above, although in some cases the trends were not statistically significant (Additional file 1: Tables S4–7). The levels of CRP and several composite inflammatory indicators, such as SII, NLR, PLR, and LMR, were more closely associated with arrhythmia risk than single-cell count markers such as lymphocyte, neutrophil, and monocyte. Further, the associations between systemic inflammation indicators and VA and AF were stronger than the association between the systemic inflammation indicators and bradyarrhythmia. ## Sensitivity analysis Excluding arrhythmic events that occurred within the first 2 years of follow-up did not obviously alter the results (Additional file 1: Table S8). Excluding participants with malignant neoplasms, DBBF, chronic diseases involving the immune mechanism, and heart diseases, did not alter the results either (Additional file 1: Table S9). ## Discussion The present study has several noteworthy findings: First, with regard to the relationship between the levels of systemic inflammation indicators and the risk of various arrhythmias, CRP showed a linear positive correlation; monocyte count, neutrophil count, and NLR showed a nonlinear positive correlation; lymphocyte count, SII, PLR, and LMR showed a U-shaped association. Second, after fully adjusting for covariates, the above association still existed and was strongest for VA, followed by AF and bradyarrhythmia. Third, the above trends were further validated in different populations through subgroup analyses and sensitivity analyses. Although previous studies have shown that elevated CRP is associated with the occurrence and recurrence of AF [12, 21, 22] and an increased risk of malignant VA [23, 24] in populations with structural heart diseases, we validated the exact linear correlation between CRP and various arrhythmias in a notably larger population. Importantly, even after adjusting for all potential confounders, in various subgroup analyses and sensitivity analyses, this linear correlation was sustained and robust and was the strongest among the selected inflammation indicators. A recent study suggested that CRP not only is an inflammatory marker, but also promotes inflammation and subsequent myocardial fibrosis through the TLR4/NF-κB/TGF-β pathway [25]. Another study showed that CRP played a proarrhythmic role by directly affecting calcium homeostasis in cardiomyocytes [26]. Our findings corroborate previous studies and confirm that CRP is a key indicator of arrhythmia risk. The results of previous studies on the association between differential leukocyte counts and incident AF have been conflicting and not fully consistent with our findings. For example, one study reported a positive association between total leukocyte count and the risk of AF in 6315 individuals, but no association was observed between differential leukocyte counts and the risk of AF [15]. A community-based cohort study with a larger sample of 14,500 participants reported that the total leukocyte, neutrophil, and monocyte counts were positively associated with higher AF risk, while the lymphocyte counts were inversely associated [27]. This inverse association between lymphocyte counts and incident AF differs from the U-shaped association observed in our study, and it might be explained by the elimination of extreme values in their study. The potential associations between differential leukocyte counts and the risk of other types of arrhythmias have been poorly studied. Herein we confirmed the precise non-linear positive correlation between neutrophil and monocyte counts and a U-shaped association for lymphocyte counts in the case of various arrhythmias. This correlation was not only found for AF, but also found for VA and bradyarrhythmia. The mechanisms by which leukocytes contribute to arrhythmias are complicated. Inflammatory cells infiltrate the myocardium and release reactive oxygen species, cytokines, myeloperoxidase, and hydrolase, leading to irregular interstitial fibrosis that causes electrical and structural remodeling of atrial and ventricular tissue, and consequently, the development of AF and VA [28–31]. Active adhesion and recruitment of inflammatory cells were observed in the atrial tissue of AF patients, and the involved cells included neutrophils, lymphocytes, monocytes, macrophages, and granulocytes [32–34]. The overproduction of inflammatory cytokines by persistent host inflammatory response can also act on the sinoatrial node and cause bradyarrhythmia [35]. However, bradyarrhythmia is mainly caused by sinoatrial/atrioventricular node dysplasia and degeneration. VA/AF are mainly associated with remodeling and sympathetic activation. Inflammation plays an important role in remodeling and sympathetic excitation, and a relatively mild role in dysplasia and degeneration. This may be the mechanistic explanation for the weaker association between systemic inflammation and bradyarrhythmia than VA/AF. The U-shaped association observed between lymphocyte counts and the incidence of arrhythmias also deserves our attention, which may be explained by the physiological stress and inflammatory states under abnormal (both high and low) lymphocyte counts [36]. NLR, PLR, LMR, and SII are composite inflammatory markers derived from ratios of differential leukocyte counts and platelets. They are believed to better reflect the intensity of systemic inflammation and are potentially superior to simple WBC counts [18–20, 37–39]. Our findings confirm this viewpoint. In the subgroup and sensitivity analyses, composite inflammatory markers were more frequently significantly associated with arrhythmia risk than simple blood cell counts. Previous researches on the relationship between these composite markers and AF recurrence/onset have reported conflicting results [13, 14, 38, 40]. Furthermore, the potential relationship between composite markers and other types of arrhythmias remains unresolved and is rarely reported [41]. However, in this large cohort study, we have provided compelling results and demonstrated the exact association between these composite markers and different arrhythmias. Most of the correlation curves were U-shaped and can be explained by the original U-shaped curve that reflects simple lymphocyte. Considering the character as continuous variables of inflammation indicators, it is not easy to obtain the exact proportion of population at a clinically increase in arrhythmia risk due to inflammation. But this study provides the values of specific risk for arrhythmia onset and will be helpful to provide reference indicators for constructing predictive models for arrhythmia occurrence. Since inflammation plays a prominent role in the development of different arrhythmias, anti-inflammatory drugs are likely to improve cardiovascular outcomes. Colchicine, for example, has a variety of anti-inflammatory effects as a safe and well-tolerated treatment for gout. Several clinical trials have demonstrated the protective effect of colchicine in postoperative atrial fibrillation and in post-ablation atrial fibrillation [42, 43]. Anti-inflammatory therapy targeting the interleukin-1β innate immunity pathway with canakinumab led to a significantly lower rate of recurrent cardiovascular events than placebo [44] and its role in preventing arrhythmias is also worthy of expectation. Combined with our study findings, early intervention on the systemic inflammation may be a promising therapy to reduce the occurrence of arrhythmia. ## Strengths and limitations Our study has several unique advantages. First, the UKB is a large prospective cohort including diversified inflammatory indicators for over 500,000 individuals with over 12 years of follow-up. This makes the present study the largest analysis that provides the highest level of evidence for the association between systemic inflammation and various arrhythmias to date. Second, our investigation links systemic inflammation to three arrhythmia subtypes, namely, AF, VA, and bradyarrhythmia, the latter two of which have been poorly studied. Third, this study included comprehensive measures of systemic inflammation indicators, including CRP and differential leukocyte count, and composite measures such as SII, LMR, NLR, and PLR. There are also some limitations to this study. First, this is an observational study and, therefore, cannot prove a causal relationship between systemic inflammation and cardiac arrhythmias. Second, the inflammatory indicators and confounding variables were only assessed at the baseline, and relevant information was lacking during follow-up. That is, these values could have changed over time, but we were unable to document or assess this. Third, we used data from reports of hospitalizations and deaths to diagnose the incidence of arrhythmias, and this may have led to an underestimation of the true incidence, given the likelihood of subclinical onset of arrhythmias. However, to make the study more clinically relevant, we included a diagnosis of arrhythmias that may cause serious adverse outcomes which are not usually followed by subclinical episodes of arrhythmias. Fourth, although we carefully adjusted for various major confounders, biases resulting from unknown and unmeasured confounders may still exist. Finally, this group included individuals of European origin and was mainly a white British population; this limits the applicability of the findings to populations belonging to other races. ## Conclusion This large-scale prospective study demonstrates that systemic inflammation levels are significantly associated with the risk of cardiac arrhythmias, and the above association is strongest for VA, followed sequentially by AF and bradyarrhythmia. Given the high morbidity and mortality and potential reversibility of severe arrhythmias, early prevention is critical. This study helps to provide reference indicators for constructing predictive models for arrhythmia occurrence in the future. Furthermore, aggressive management of systemic inflammation might have favorable effects on reducing the arrhythmia burden, which need to be further confirmed by randomized controlled studies. ## Supplementary Information Additional file 1: Table S1. Arrhythmia Definitions using the UK biobank. Table S2. Definitions of diseases history in the UK Biobank. Table S3. Demographic and clinical characteristics of participants in a study of arrhythmias in the UK Biobank. Table S4. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes among participants with heart diseases, hypertension, or hyperlipidemia at baseline by the Cox proportional hazard model. Table S5. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes among participants with chronic diseases involving the immune mechanism, or malignant neoplasms at baseline by the Cox proportional hazard model. Table S6. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes stratified by age group at baseline using the Cox proportional hazard model. Table S7. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes stratified by sex using the Cox proportional hazard model. Table S8. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes excluding incident cases occurred in the first 2 years of follow-up by the Cox proportional hazard model. Table S9. The hazard risk (HR) with $95\%$ confidence intervals ($95\%$CI) between various systematic information indicators and three arrhythmia subtypes excluding participants with heart diseases, diseases of blood and blood-forming organs, chronic diseases involving the immune mechanism, or malignant neoplasms at baseline by the Cox proportional hazard model. ## References 1. Fitzmaurice C, Allen C, Barber RM, Barregard L, Bhutta ZA, Brenner H. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life-years for 32 cancer groups, 1990 to 2015: a systematic analysis for the global burden of disease study. JAMA Oncol (2017.0) 3 524-548. DOI: 10.1001/jamaoncol.2016.5688 2. Lakshminarayan K, Anderson DC, Herzog CA, Qureshi AI. Clinical epidemiology of atrial fibrillation and related cerebrovascular events in the United States. Neurologist (2008.0) 14 143-150. DOI: 10.1097/NRL.0b013e31815cffae 3. Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM. Global burden of cardiovascular diseases and risk factors, 1990-2019: update from the GBD 2019 study. J Am Coll Cardiol (2020.0) 76 2982-3021. DOI: 10.1016/j.jacc.2020.11.010 4. Dai H, Zhang Q, Much AA, Maor E, Segev A, Beinart R. Global, regional, and national prevalence, incidence, mortality, and risk factors for atrial fibrillation, 1990-2017: results from the global burden of disease study 2017. Eur Heart J Qual Care Clin Outcomes (2021.0) 7 574-582. DOI: 10.1093/ehjqcco/qcaa061 5. Lloyd-Jones DM, Wang TJ, Leip EP, Larson MG, Levy D, Vasan RS. Lifetime risk for development of atrial fibrillation: the Framingham heart study. Circulation (2004.0) 110 1042-1046. DOI: 10.1161/01.CIR.0000140263.20897.42 6. Tang DH, Gilligan AM, Romero K. Economic burden and disparities in healthcare resource use among adult patients with cardiac arrhythmia. Appl Health Econ Health Policy (2014.0) 12 59-71. DOI: 10.1007/s40258-013-0070-9 7. Carrizales-Sepúlveda EF, Ordaz-Farías A, Vera-Pineda R, Flores-Ramírez R. Periodontal disease, systemic inflammation and the risk of cardiovascular disease. Heart Lung Circ (2018.0) 27 1327-1334. DOI: 10.1016/j.hlc.2018.05.102 8. Deng T, Lyon CJ, Bergin S, Caligiuri MA, Hsueh WA. Obesity, inflammation, and cancer. Annu Rev Pathol (2016.0) 11 421-449. DOI: 10.1146/annurev-pathol-012615-044359 9. Kronsten VT, Tranah TH, Pariante C, Shawcross DL. Gut-derived systemic inflammation as a driver of depression in chronic liver disease. J Hepatol (2022.0) 76 665-680. DOI: 10.1016/j.jhep.2021.11.008 10. Frodermann V, Rohde D, Courties G, Severe N, Schloss MJ, Amatullah H. Exercise reduces inflammatory cell production and cardiovascular inflammation via instruction of hematopoietic progenitor cells. Nat Med (2019.0) 25 1761-1771. DOI: 10.1038/s41591-019-0633-x 11. Liberale L, Badimon L, Montecucco F, Luscher TF, Libby P, Camici GG. Inflammation, aging, and cardiovascular disease: JACC review topic of the week. J Am Coll Cardiol (2022.0) 79 837-847. DOI: 10.1016/j.jacc.2021.12.017 12. Wang Z, Wang B, Li X, Zhang S, Wu S, Xia Y. Metabolic syndrome, high-sensitivity C-reactive protein levels and the risk of new-onset atrial fibrillation: results from the Kailuan study. Nutr Metab Cardiovasc Dis (2021.0) 31 102-109. DOI: 10.1016/j.numecd.2020.06.026 13. Selcuk M, Cinar T, Saylik F, Dogan S, Selcuk I, Orhan AL. Predictive value of systemic immune inflammation index for postoperative atrial fibrillation in patients undergoing isolated coronary artery bypass grafting. Meden Med J (2021.0) 36 318-324. DOI: 10.4274/MMJ.galenos.2021.37998 14. Bağcı A, Aksoy F. Systemic immune-inflammation index predicts new-onset atrial fibrillation after ST elevation myocardial infarction. Biomark Med (2021.0) 15 731-739. DOI: 10.2217/bmm-2020-0838 15. Nyrnes A, Njølstad I, Mathiesen EB, Wilsgaard T, Hansen JB, Skjelbakken T. Inflammatory biomarkers as risk factors for future atrial fibrillation. An eleven-year follow-up of 6315 men and women: the Tromsø study. Gend Med (2012.0) 9 536-47.e2. DOI: 10.1016/j.genm.2012.09.001 16. Boyalla V, Harling L, Snell A, Kralj-Hans I, Barradas-Pires A, Haldar S. Biomarkers as predictors of recurrence of atrial fibrillation post ablation: an updated and expanded systematic review and meta-analysis. Clin Res Cardiol (2022.0) 111 680-691. DOI: 10.1007/s00392-021-01978-w 17. Rusk N, The UK. Biobank. Nat Methods (2018.0) 15 1001. DOI: 10.1038/s41592-018-0245-2 18. Guragac A, Demirer Z. The neutrophil-to-lymphocyte ratio in clinical practice. Can Urol Assoc J (2016.0) 10 141. DOI: 10.5489/cuaj.3587 19. Zhang YR, Wang JJ, Chen SF, Wang HF, Li YZ, Ou YN. Peripheral immunity is associated with the risk of incident dementia. Mol Psychiatry (2022.0) 27 1956-1962. DOI: 10.1038/s41380-022-01446-5 20. Mazza MG, Lucchi S, Tringali AGM, Rossetti A, Botti ER, Clerici M. Neutrophil/lymphocyte ratio and platelet/lymphocyte ratio in mood disorders: a meta-analysis. Prog Neuro-Psychopharmacol Biol Psychiatry (2018.0) 84 229-236. DOI: 10.1016/j.pnpbp.2018.03.012 21. Xie JY, Noeman M, Pimenta D, Little C. C-reactive protein as a predictor for developing post-operative atrial fibrillation. Europace (2021.0) 23 159. DOI: 10.1093/europace/euaa303 22. Aviles RJ, Martin DO, Apperson-Hansen C, Houghtaling PL, Rautaharju P, Kronmal RA. Inflammation as a risk factor for atrial fibrillation. Circulation (2003.0) 108 3006-3010. DOI: 10.1161/01.CIR.0000103131.70301.4F 23. Biasucci LM, Giubilato G, Biondi-Zoccai G, Sanna T, Liuzzo G, Piro M. C reactive protein is associated with malignant ventricular arrhythmias in patients with ischaemia with implantable cardioverter-defibrillator. Heart (2006.0) 92 1147-1148. DOI: 10.1136/hrt.2005.065771 24. Wang CG, Qin XC, Nie SP, Wang CM, Ai H, Que B. C-reactive protein as a predictor of malignant ventricular arrhythmias in non-ST elevation myocardial infarction. J Geriatr Cardiol (2019.0) 16 614-620. PMID: 31555329 25. 25.Sun W, Wu Y, Gao M, Tian Y, Qi P, Shen Y, et al. C-reactive protein promotes inflammation through TLR4/NF-κB/TGF-β pathway in HL-1 cells. Biosci Rep. 2019;39(8):BSR20190888. 10.1042/BSR20190888. 26. Xie Y, Li Q, Zhang HF, Huang TC, Yang Y, Lin YQ. Effect of C reactive protein on the sodium-calcium exchanger 1 in cardiomyocytes. Exp Ther Med (2021.0) 22 815. DOI: 10.3892/etm.2021.10247 27. Misialek JR, Bekwelem W, Chen LY, Loehr LR, Agarwal SK, Soliman EZ. Association of white blood cell count and differential with the incidence of atrial fibrillation: the atherosclerosis risk in communities (ARIC) study. PLoS One (2015.0) 10 e0136219. DOI: 10.1371/journal.pone.0136219 28. Cooper LT. Myocarditis. N Engl J Med (2009.0) 360 1526-1538. DOI: 10.1056/NEJMra0800028 29. Friedrichs K, Baldus S, Klinke A. Fibrosis in atrial fibrillation - role of reactive species and MPO. Front Physiol (2012.0) 3 214. DOI: 10.3389/fphys.2012.00214 30. Hulsmans M, Sam F, Nahrendorf M. Monocyte and macrophage contributions to cardiac remodeling. J Mol Ccell Cardiol (2016.0) 93 149-155. DOI: 10.1016/j.yjmcc.2015.11.015 31. Lindholm D, James SK, Gabrysch K, Storey RF, Himmelmann A, Cannon CP. Association of Multiple biomarkers with risk of all-cause and cause-specific mortality after acute coronary syndromes: a secondary analysis of the PLATO biomarker study. JAMA Cardiol (2018.0) 3 1160-1166. DOI: 10.1001/jamacardio.2018.3811 32. Hohmann C, Pfister R, Mollenhauer M, Adler C, Kozlowski J, Wodarz A. Inflammatory cell infiltration in left atrial appendageal tissues of patients with atrial fibrillation and sinus rhythm. Sci Rep (2020.0) 10 1685. DOI: 10.1038/s41598-020-58797-8 33. Smorodinova N, Bláha M, Melenovský V, Rozsívalová K, Přidal J, Ďurišová M. Analysis of immune cell populations in atrial myocardium of patients with atrial fibrillation or sinus rhythm. PLoS One (2017.0) 12 e0172691. DOI: 10.1371/journal.pone.0172691 34. Yamashita T, Sekiguchi A, Iwasaki YK, Date T, Sagara K, Tanabe H. Recruitment of immune cells across atrial endocardium in human atrial fibrillation. Circ J (2010.0) 74 262-270. DOI: 10.1253/circj.CJ-09-0644 35. Srinivasan A, Pansuriya T, Wilson B, Sarva ST, Jayaram T, Alim S. Systemic inflammation-related bradycardia in COVID-19. Case Rep Cardiol (2021.0) 2021 9986955. PMID: 34567809 36. 36.Orakpoghenor O, Avazi DO, Markus TP, Olaolu OS. Lymphocytes: A Brief Review. Sci J Immunol Immunother. 2019;3(1):004–8. https://www.sciresliterature.org/Immunology/SJII-ID17.pdf. 37. Nøst TH, Alcala K, Urbarova I, Byrne KS, Guida F, Sandanger TM. Systemic inflammation markers and cancer incidence in the UK Biobank. Eur J Epidemiol (2021.0) 36 841-848. DOI: 10.1007/s10654-021-00752-6 38. Weymann A, Ali-Hasan-Al-Saegh S, Sabashnikov A, Popov AF, Mirhosseini SJ, Liu T. Prediction of new-onset and recurrent atrial fibrillation by complete blood count tests: a comprehensive systematic review with meta-analysis. Med Sci Monit Basic Res (2017.0) 23 179-222. DOI: 10.12659/MSMBR.903320 39. de Jager CP, van Wijk PT, Mathoera RB, de Jongh-Leuvenink J, van der Poll T, Wever PC. Lymphocytopenia and neutrophil-lymphocyte count ratio predict bacteremia better than conventional infection markers in an emergency care unit. Critical care (London, England) (2010.0) 14 R192. DOI: 10.1186/cc9309 40. Berkovitch A, Younis A, Grossman Y, Segev S, Kivity S, Sidi Y. Relation of neutrophil to lymphocyte ratio to risk of incident atrial fibrillation. Am J Cardiol (2019.0) 123 396-401. DOI: 10.1016/j.amjcard.2018.10.036 41. Chatterjee S, Chandra P, Guha G, Kalra V, Chakraborty A, Frankel R. Pre-procedural elevated white blood cell count and neutrophil-lymphocyte (N/L) ratio are predictors of ventricular arrhythmias during percutaneous coronary intervention. Cardiovasc Hematol Disord Drug Targets (2011.0) 11 58-60. DOI: 10.2174/187152911798346981 42. Deftereos S, Giannopoulos G, Kossyvakis C, Efremidis M, Panagopoulou V, Kaoukis A. Colchicine for prevention of early atrial fibrillation recurrence after pulmonary vein isolation: a randomized controlled study. J Am Coll Cardiol (2012.0) 60 1790-1796. DOI: 10.1016/j.jacc.2012.07.031 43. Lee JZ, Singh N, Howe CL, Low SW, Huang JJ, Ortega G. Colchicine for prevention of post-operative atrial fibrillation: a meta-analysis. JACC Clin Electrophysiol (2016.0) 2 78-85. DOI: 10.1016/j.jacep.2015.09.016 44. Ridker PM, Everett BM, Thuren T, MacFadyen JG, Chang WH, Ballantyne C. Antiinflammatory therapy with canakinumab for atherosclerotic disease. N Engl J Med (2017.0) 377 1119-1131. DOI: 10.1056/NEJMoa1707914
--- title: ANGPTL2 binds MAG to efficiently enhance oligodendrocyte differentiation authors: - Lu Chen - Zhuo Yu - Li Xie - Xiaoxiao He - Xingmei Mu - Chiqi Chen - Wenqian Yang - Xiaoping Tong - Junling Liu - Zhengliang Gao - Suya Sun - NanJie Xu - Zhigang Lu - Junke Zheng - Yaping Zhang journal: Cell & Bioscience year: 2023 pmcid: PMC9976406 doi: 10.1186/s13578-023-00970-3 license: CC BY 4.0 --- # ANGPTL2 binds MAG to efficiently enhance oligodendrocyte differentiation ## Abstract ### Background Oligodendrocytes have robust regenerative ability and are key players in remyelination during physiological and pathophysiological states. However, the mechanisms of brain microenvironmental cue in regulation of the differentiation of oligodendrocytes still needs to be further investigated. ### Results We demonstrated that myelin-associated glycoprotein (MAG) was a novel receptor for angiopoietin-like protein 2 (ANGPTL2). The binding of ANGPTL2 to MAG efficiently promoted the differentiation of oligodendrocytes in vitro, as evaluated in an HCN cell line. Angptl2-null mice had a markedly impaired myelination capacity in the early stage of oligodendrocyte development. These mice had notably decreased remyelination capacities and enhanced motor disability in a cuprizone-induced demyelinating mouse model, which was similar to the Mag-null mice. The loss of remyelination ability in Angptl2-null/Mag-null mice was similar to the Angptl2-WT/Mag-null mice, which indicated that the ANGPTL2-mediated oligodendrocyte differentiation effect depended on the MAG receptor. ANGPTL2 bound MAG to enhance its phosphorylation level and recruit Fyn kinase, which increased Fyn phosphorylation levels, followed by the transactivation of myelin regulatory factor (MYRF). ### Conclusion Our study demonstrated an unexpected cross-talk between the environmental protein (ANGPTL2) and its surface receptor (MAG) in the regulation of oligodendrocyte differentiation, which may benefit the treatment of many demyelination disorders, including multiple sclerosis. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13578-023-00970-3. ## Background Oligodendrocytes are one of the major glial cell types in the central nervous system (CNS) and are essential for myelin formation during CNS development, adaptive myelination in adulthood and remyelination upon damage [1, 2]. However, oligodendrocytes are highly vulnerable and easily affected by trauma, ischemia and immune-mediated demyelinating diseases, such as multiple sclerosis (MS). MS is characterized by focal lymphocytic infiltration, an imbalance between demyelination and remyelination and progressive neurodegeneration [3]. Although much effort was exerted to identify the intrinsic or extrinsic factors involved in the regulation of myelination and remyelination during the activation, migration, proliferation and differentiation of oligodendrocyte progenitors (OPCs) [4, 5, 6, 7, 8, 9, 10], the efficient enhancement of remyelination to delay neurodegeneration remains challenging due to the lack of knowledge on the mechanisms of OPC differentiation into mature oligodendrocytes. For demyelination-related diseases, the implantation of OPCs into OPC-depleted regions and enhancement of their differentiation into mature and functional oligodendrocytes are two major strategies to promote the remyelination program [11, 12]. Many studies showed that increased numbers of OPCs and some premyelinating oligodendrocytes, but not mature functional oligodendrocytes, predominantly existed in the chronically demyelinated brain area of MS [13, 14, 15], which indicates that full differentiation of oligodendrocytes is critical for remyelination in MS. Although many potential regulators for oligodendrocyte differentiation were reported [16, 17, 18], most of them are intrinsic factors, such as transcription factors (OLIG1, OLIG2, MYRF and SOX10) [19, 20, 21, 22], epigenetic regulators (DNMT1 and DNMT3) [16, 18] and long noncoding RNAs lncOLs [17]. Recently, the importance of extrinsic factors contributing to oligodendrocyte differentiation and their capacity for myelination and remyelination is increasing recognized, such as autotaxin (ATX) [23, 24], Sema3A [25]. However, how extrinsic factors linking receptors to regulating oligodendrocyte maturation remains largely unknown. The secretory protein angiopoietin-like protein 2 (ANGPTL2) is a member of the angiopoietin-like family that plays vital roles in multiple physiological and pathological states, including angiogenesis [26, 27, 28], lipid metabolism [29], obesity [30], thrombosis [31], atherosclerosis [32] fibrosis [33] and tumor metastasis [34, 35] Our previous study demonstrated that ANGPTL2 was a ligand for the immune-inhibitory receptor human leukocyte immunoglobulin-like receptor B2 (LILRB2), maintained the stemness of hematopoietic stem cells (HSCs) and enhanced leukemogenic activities [36]. Notably, we screened a membrane protein library and identified another receptor, MAG, that bound to ANGPTL2 with high binding affinity (~ 10 nM). ANGPTL2-MAG-mediated signaling promoted the differentiation of oligodendrocytes in vitro and in vivo and enhanced remyelination in a cuprizone-induced demyelination murine model. ANGPTL2 enhanced the phosphorylation of MAG to recruit and activate Fyn-mediated signaling and increase the expression of myelination-related genes, such as the transcription factor, MYRF. These findings provide unique insight into the regulation of the differentiation of oligodendrocytes and a potential strategy for the treatment of demyelination diseases. ## Human MAG is a new receptor for ANGPTL2 We previously showed that ANGPTL2 bound LILRB2 to maintain HSC activities. We further screened other potential surface molecules that may be receptors for ANGPTL2 using flow cytometric analysis and a customized human cDNA library of membrane proteins (Additional file 1: Fig. S1A). Notably, MAG, which is a type I single-pass transmembrane glycoprotein expressed in oligodendrocytes and Schwann cells that plays important roles in myelination in the CNS and peripheral nervous system (PNS) [37, 38] as a ligand for NgR [39], PirB [40] and β1-integrin [41], or as a receptor for gangliosides [42], specifically bound ANGPTL2, but not other ANGPTL members (Fig. 1A–B, Additional file 1: Fig. S1B–C). Immunoprecipitation assays further showed that ANGPTL2 interacted with the extracellular domain (ECD) of MAG (Fig. 1C). A chimeric receptor assay [43] was established to determine the interaction between ANGPTL2 and MAG. The MAG ECD was fused with transmembrane/intracellular domains of activating paired immunoglobulin-like receptor b (PILRb) in this system, followed by infection into murine T-cell hybridoma cells with a nuclear factor of activated T cells (NFAT)/GFP reporter gene (Additional file 1: Fig. S1D). We observed that ANGPTL2 induced marked GFP expression in MAG reporter cells, as determined by flow cytometry (Fig. 1D–E).Fig. 1Human MAG is a new receptor for ANGPTL2. A Flow cytometric analysis of conditioned medium with/without ANGPTL2-Flag binding to 293T cells with transient transfection of plasmids of N1-hMAG-EGFP (human MAG, full-length) or N1-EGFP (vector). Percentage of binding/unbinding cells to ANGPTL2 in EGFP+ cells are shown. B Quantitative data of percentage of binding cells in Panel A ($$n = 3$$). C ANGPLT2 bound to the ECD of human MAG (human MAG-ECD-FC) in the conditioned medium (CM) of cotransfected 293T cells using co-immunoprecipitation. Tie-2-ECD-FC served as a negative control. D Representative flow cytometric analysis showing that purified ANGPTL2-Flag protein enhanced GFP expression in human MAG chimeric reporter cells (Human MAG). Control reporter cells (control) without human MAG-ECD were examined. E Quantitative data in Panel D ($$n = 3$$). F Binding kinetics of ANGPTL2-Flag to human MAG-ECD were measured using a biolayer interferometry instrument (Octed RED $\frac{96}{96}$S). G–H Flow cytometric plots for the binding of ANGPTL2-Flag to full-length, mutant IgG 3 domain or IgG 4 domain of human MAG expressed 293T cells. Percentage of binding/unbinding cells to ANGPTL2 in EGFP+ cells (G) and quantitative data are shown (H) ($$n = 3$$). (*** $p \leq 0.001$) We evaluated the binding affinity between ANGPTL2 and MAG using biolayer interferometry (Octet) and surface plasmon resonance (SPR), which showed that the dissociation constants were 9.34 nM (Fig. 1F) and 6.47 nM (Additional file 1: Fig. S1E), respectively. We previously showed that the HGYC motifs of LILRB2 were essential for its binding to ANGPTL2 [43]. We examined the potential binding motifs of the extracellular IgG domain of MAG and speculated that G301/Y303 and G389/Y341 in the third or fourth IgG domain were the key binding motifs for ANGPTL2. Mutagenesis analysis showed that G301D/Y303A MAG mutation in IgG3 and G389D/Y341A in IgG4 abolished binding to ANGPTL2 than the MAG control, as determined by flow cytometric analysis (Fig. 1G–H), which suggests that IgG3/IgG4 is critical for ANGPTL2-mediated binding/activation of MAG. ## Murine homologue of human MAG binds to ANGPTL2 We determined whether ANGPTL2 also bound to the murine homologue of human MAG. Notably, murine MAG exhibited $94\%$ identical sequence to human MAG, which indicates that this gene is very conserved between different species. Similar to human MAG, we found that ANGPTL2 bound murine MAG using flow cytometric analysis (Fig. 2A–B). Co-immunoprecipitation assays showed that ANGPTL2 directly interacted with murine MAG (Fig. 2C). We further constructed chimeric reporter cells with murine MAG ECD and revealed that ANGPTL2 efficiently induced GFP expression in reporter cells (Fig. 2D–E). Notably, the dissociation constant of ANGPTL2 to murine MAG was similar to human MAG as determined by Octet (16.4 nM, Fig. 2F) and SPR (3.08 nM, Additional file 1: Fig. S2A). These data clearly showed that MAG was the receptor for ANGPTL2 with high affinity. Fig. 2Murine homologue of human MAG binds to ANGPTL2. A Flow cytometric analysis of conditioned medium with/without ANGPTL2-Flag binding to 293 T cells with transient transfection of plasmids of N1-mMAG-EGFP (murine MAG, full-length) or N1-EGFP (vector). Percentage of binding/unbinding cells to ANGPTL2 in EGFP+ cells are shown. B Quantitative data* of percentage of binding cells in Panel A ($$n = 3$$). C ANGPLT2 bound to the ECD of murine MAG (murine MAG-ECD) in the conditioned medium (CM) of 293 T cotransfected cells using co-immunoprecipitation. Tie-2-ECD-FC served as a negative control. D Representative flow cytometric analysis showing that purified ANGPTL2-Flag protein enhanced GFP expression in murine MAG chimeric reporter cells (Murine MAG). Control reporter cells (control) without murine MAG-ECD were examined. E Quantitative data in Panel D ($$n = 3$$). F Binding kinetics of ANGPTL2-Flag to murine MAG-ECD were measured using a biolayer interferometry instrument (Octed RED $\frac{96}{96}$S). (*** $p \leq 0.001$) ## ANGPTL2 promotes oligodendrocyte differentiation Mag deletion disrupted and multiplied compact myelin lamella and increased periaxonal spacing in the murine CNS [44, 45]. Notably, the newly identified receptor MAG was highly expressed in oligodendrocytes, but not oligodendrocyte precursor cells (https://www.proteinatlas), which indicates that ANGPTL2-MAG mediated signaling may play important roles in oligodendrocyte maturation at late differentiation stage (but not in early precursor cell stage). To determine whether ANGPTL2-MAG-mediated signaling is required for the oligodendrocyte maturation, we first established an in vitro oligodendrocyte differentiation system using a rat hippocampus-derived adult neural progenitor (HCN) cell line, which efficiently differentiates into mature oligodendrocytes in the presence of IGF1 [46]. HCN cells are efficiently induced into mature oligodendrocytes with small and round somata and web-like branches in morphology. These cells were positive for several markers of mature oligodendrocytes, such as myelin basic protein (MBP) [47], myelin-associated glycoprotein (MAG) [48, 49] and myelin-oligodendrocyte glycoprotein (MOG) [47] (Additional file 1: Fig. S3A), which were increased at the mRNA and protein levels over time (Additional file 1: Fig. S3B–C). We further treated HCN cells with ANGPTL2 in the presence of IGF1 and demonstrated that more morphological oligodendrocytes with web-like branches were elicited compared to the control (Fig. 3A). Immunofluorescence staining also showed a notable increase in the number of mature MBP+ oligodendrocytes with more complicated web-like branches upon ANGPTL2 treatment ($20.15\%$ vs. $42.36\%$, Fig. 3B–C), which indicates an accelerated processes of oligodendrocyte maturation. Because we previously showed that ANGPTL2 bound to LILRB2 and its murine ortholog, paired Ig-like receptor (PirB), to maintain the stemness of HSCs [36], we further examined PirB expression during the differentiation from HCN cells in vitro. Notably, the Pirb mRNA level was almost undetectable, but the expression levels of Mag, Mbp, Mog and Angtpl2 were increased during the process of oligodendrocyte differentiation (Additional file 1: Fig. S3D–E), which suggested that MAG was a specific receptor for ANGPTL2 and ANGPTL2 might be secreted by oligodendrocytes to exert its autocrine effect in promoting oligodendrocyte differentiation. Meanwhile, the frequencies of PDGFa+Ki-67+ oligodendrocyte precursor cells were comparable to the control group, supporting the notion that ANGTPL2 had no effect on the proliferation and differentiation of precursor cells (Additional file 1: Fig. S3F–G).Fig. 3ANGPTL2 promotes oligodendrocyte differentiation. A Cell morphology of HCN cells after treatment with IGF1 for the differentiation to oligodendrocytes in vitro with/without ANGPTL2 at the indicated time points. B Immunofluorescence staining for MBP in HCN cells 72 h after induction in Panel A. C Quantitative data in Panel B are shown; A total of 22 or 23 sections were counted ($$n = 4$$). D Representative electron microscopy images in the optic nerve fiber and corpus callosum of the brains from Angptl2+/+ and Angptl2−/− mice on day 15; red arrow indicates the normal myelin, green arrow indicates uncompacted myelin lamella; yellow arrow indicates redundant compact myelin; purple indicates excess of cytoplasm in the periaxonal space. E Shown are the quantification data of abnormally myelinated axons including uncompacted myelin lamella, redundant compact myelin and excess of cytoplasm in the periaxonal space, in optic nerve fibers and corpus callosum. 100–410 axons in each mouse were counted ($$n = 3$$). F–I Quantification of the G ratios of the myelinated axons of the optic nerve fiber (F) or corpus callosum (H) in D. A total of 120 ~ 200 axons in each mouse were counted ($$n = 3$$). The scatter plots for the individual G-ratio values and axonal size distribution in the optic nerve fibers (G) or corpus callosum (I) in Panel D are shown. (*** $p \leq 0.001$) To further examine the role of ANGPTL2 in oligodendrocyte differentiation, we established Angptl2 knockout mice with deletion of exon 2 by breeding male Angptl2fl/fl; Stra8-cre mice (Angptl2 was specifically deleted in early-stage spermatogonia, spermatocytes and sperm) with wild-type (WT) female mice (Additional file 1: Fig. S3H). The genetic deletion of Angptl2 was further confirmed using PCR (Additional file 1: Fig. S3I). No detectable ANGPTL2 protein was observed in different mouse tissues of Angptl2-null (or Angptl2−/−) mice, including the heart, lung and brain, as determined by Western blot (Additional file 1: Fig. S3J). Notably, Angptl2-null mice had several defects in the myelination of axons, such as uncompacted myelin lamella, redundant compact myelin and excess of cytoplasm in the periaxonal space, in both optic nerves and corpus callosum as evaluated by using transmission electron microscopy (Fig. 3D–E). The thickness of myelin lamella in the optic nerve and corpus callosum from Angptl2-null mice was significantly thinner than WT mice at the early neonatal stage (Day 15, Fig. 3F–I), as indicated by an increased G-ratio, but not at the late adult stage (Day 35, Additional file 1: Fig. S3K–N). These results indicate that ANGPTL2 plays a vital role in myelin formation at the early stage of mouse development, which is consistent with the findings in Mag-null mice with delayed myelination [44, 45] and further suggest that ANGPTL2-MAG-mediated signaling is crucial for oligodendrocyte differentiation. ## ANGPTL2 promotes oligodendrocyte differentiation and remyelination under pathological conditions To further evaluate the role of ANGPTL2 in pathological states, we used a cuprizone-induced murine demyelination model [50] and examined the dynamic changes in myelination in the corpus callosum. The demyelination status in the corpus callosum was similar between WT and Angptl2-null mice 5 weeks after cuprizone diet treatment as determined by oil red O staining (Additional file 1: Fig. S4A–B). Immunohistochemical staining also showed similar decreased levels of several myelin related proteins, including MBP, MOG and MAG (Additional file 1: Fig. S4C–D). The expression level of a marker of astrocyte activation, GFAP, was also comparable between WT and Angptl2-null mice (Additional file 1: Fig. S4C–D), although it has been reported to be upregulated in brain due to the compensatory effect from demyelination after cuprizone treatment [51]. These data also indicated that ANGPTL2 had no effect on the process of demyelination. Notably, spontaneous remyelination in the corpus callosum in Angptl2−/− mice was much slower as evidenced by less myelin staining, while more nonspecific oil red O positive deposits (a phenomenon that oil red O deposits in damaged myelin region [52]), than the WT mice after the withdrawal of cuprizone for an additional 2 weeks, as measured by oil red O staining (Fig. 4A–B). Angptl2-null mice also had decreased expression levels for myelin related proteins MBP, MOG and MAG, but increased astrocyte marker of GFAP due to the compensatory effect, as determined by the immunohistochemical staining (Fig. 4C–D).Fig. 4ANGPTL2 promotes oligodendrocyte differentiation and remyelination under pathological conditions. A Representative images of the corpus callosum region stained with oil red O in Angptl2+/+ and Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks. Myelin was stained in red color. Meanwhile, more nonspecific oil red O positive deposits (a phenomenon that oil red O deposits in damaged myelin region) were observed in Angptl2−/− mice (arrows). B Quantitative data of the demyelinating areas in Panel A; six sections from each mouse were analyzed ($$n = 7$$). C Immunofluorescence staining for MBP, MOG, MAG and GFAP in the corpus callosum region from Angptl2+/+ and Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks. D Quantification of the fluorescence intensity of MBP, MOG, MAG and GFAP in the corpus callosum region in Panel C. Four sections from each mouse were analyzed ($$n = 6$$–7). E Representative electron microscopy images of the corpus callosum from Angptl2+/+ and Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks; red arrow indicates uncompacted myelin lamella; F Quantification of the G ratios of the remyelinated axons of the corpus callosum in Panel E. Approximately 70 axons were counted per mouse ($$n = 3$$). G The scatter plot for the individual G-ratio values and axonal size distribution of the corpus callosum in Panel E. H Motor coordination of Angptl2+/+ and Angptl2−/− mice after cuprizone (Cupz) treatment for five weeks and cuprizone withdrawal for two additional weeks in the rotarod test; 7–8 mice were used for each group. (* $p \leq 0.05$, $p \leq 0.$ 01, $p \leq 0.001$) Although very severe demyelination was observed using electron microscopy 5 weeks after cuprizone treatment (Additional file 1: Fig. S4E), the newly formed myelin in WT mice was more tightly wrapped and integrated than the Angptl2-null mice 2 weeks after cuprizone withdrawal (Fig. 4E). The myelin sheath in Angptl2-null mice was thinner than the WT mice, as evaluated by the G-ratio of the remyelinated axons (Fig. 4F–G). To test the difference in functional recovery after cuprizone treatment, a widely used assay for the assessment of motor coordination in rodents, the rotarod test, was used in WT and Angptl2-null mice to measure motor coordination and balance. The motor of WT and Angptl2-null mice began to decline compared to control mice fed normal chow 24 days after cuprizone treatment (Fig. 4H). However, Angptl2-null mice had a shorter latency to falling from the rotarod instrument than WT mice (160 s vs. 227.4 s at day 45, 185.2 s vs. 252.8 s at day 49, Fig. 4H) after cuprizone withdrawal, which indicated that ANGPTL2 promoted functional recovery from demyelination. ## ANGPTL2 promotes oligodendrocyte differentiation and remyelination via its receptor MAG MAG deficiency induces a delay in oligodendrocyte differentiation and myelination formation at an early age [53], which is consistent with the phenotype of Angptl2-null mice under physiological conditions. We further investigated whether MAG was the functional receptor for ANGPTL2 during the cuprizone-induced demyelination and remyelination. There was no significant difference in demyelination levels between WT and Mag-null mice as measured by oil red O staining (Additional file 1: Fig. S5A–B) and immunohistochemistry (Additional file 1: Fig. S5C). Similar to the findings in Angptl2-null mice, Mag-null mice showed a slower recovery from acute demyelination than WT littermates with less myelin staining (while more nonspecific oil red O positive deposits) and lower expression levels of MBP and MOG (but higher GFAP level) (Fig. 5A–D). The G-ratio in axons under electron microscopy also revealed that MAG promoted remyelination 2 weeks after withdrawal of cuprizone chow (Fig. 5E–G), but no difference was observed during the progression of demyelination between these two groups after 5 weeks of cuprizone chow treatment (Additional file 1: Fig. S5D). Mag-null mice also had a much shorter latency to falling from the rotarod instrument than WT mice (168 s vs. 225 s at day 45, 171.8 vs. 236.6 s at day 49, Fig. 5H) after the withdrawal of cuprizone chow. Fig. 5ANGPTL2 promotes remyelination via its receptor MAG. A Representative images of the corpus callosum region stained with oil red O in Mag+/+ and Mag−/− mice after the withdrawal of cuprizone for two additional weeks. Myelin was stained in red color. Meanwhile, more nonspecific oil red O positive deposits (a phenomenon that oil red O deposits in damaged myelin region) were observed in Angptl2−/− mice (arrows). B Quantification of the demyelinating areas in Panel A; six sections from each mouse were analyzed ($$n = 6$$). C Immunofluorescence images of MBP, MOG, MAG and GFAP staining in the corpus callosum region from Mag+/+ and Mag −/− mice after the withdrawal of cuprizone for two additional weeks. D Quantification of the fluorescence intensity of MBP, MOG, MAG and GFAP in the corpus callosum in Panel C. Four sections from each mouse were analyzed ($$n = 6$$). E Representative electron microscopy images of the corpus callosum from Mag+/+ and Mag−/− mice after the withdrawal of cuprizone for two additional weeks. Red arrow indicates uncompacted myelin lamella; F Quantification of the G ratios of the remyelinated axons of the corpus callosum in Panel E. Approximately 80 axons were counted per mouse ($$n = 3$$). G The scatter plot for the individual G-ratio values and axonal size distribution of the corpus callosum in Panel E. H Motor coordination for Mag+/+ and Mag−/− mice after cuprizone treatment for five weeks and cuprizone (Cupz) withdrawal for two additional weeks in the rotarod test; 19–22 mice were used for each group. I Representative images of the corpus callosum region stained with oil red O in Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks. J Quantification of the demyelinating areas in Panel I; six sections from each mouse were analyzed ($$n = 3$$). K Immunofluorescence images of MBP, MOG, MAG and GFAP staining in the corpus callosum region from Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks. L Quantification of the fluorescence intensity of MBP, MOG, MAG and GFAP in the corpus callosum in Panel K; four sections from each mouse were analyzed ($$n = 3$$). M Representative electron microscopy images of the corpus callosum from Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice after the withdrawal of cuprizone for two additional weeks. Red arrow indicates uncompacted myelin lamella. N Quantification of the G ratios of the remyelinated axons of the corpus callosum in Panel E. Approximately 70 axons were counted per mouse ($$n = 3$$). O The scatter plot for the individual G-ratio values and axonal size distribution of the corpus callosum in Panel M. P Motor coordination in Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice after cuprizone (Cupz) treatment for 5 weeks and withdrawal of cuprizone for two additional weeks in the rotarod test; 8–13 mice were used for each group. ( $p \leq 0.05$, $p \leq 0.$ 01, $p \leq 0.$ 001) We* generated Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− double mutant mice to test whether ANGPTL2 mediated enhanced myelination was dependent on the MAG receptor. Notably, Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice showed comparable remyelination capacities following cuprizone withdrawal as determined by oil red O staining (Fig. 5I–J), immunohistochemical staining (Fig. 5K–L) and electron microscopy (Fig. 5M–O). Similar demyelination levels in the corpus callosum were found between Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice 5 weeks after cuprizone chow treatment (Additional file 1: Fig. S5E–H). Notably, no significant differences in motor coordination or balance ability were observed between Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice in the cuprizone-induced mouse model (Fig. 5P), which suggested that ANGPTL2 fine-tuned oligodendrocyte differentiation by binding the MAG receptor to mediate downstream signaling under physiological and pathological situations. ## ANGPTL2-MAG induces Fyn-mediated signaling to enhance the differentiation of oligodendrocytes To elucidate the ANGPTL2-MAG-mediated downstream signaling that controls oligodendrocyte differentiation and myelination, RNA-Seq analyses (GSE199393) were performed using the postnatal brains of WT and Angptl2-null mice at day 15. GO analysis showed that several pathways involved in oligodendrocyte differentiation or myelination (such as myelin sheath and ensheathment of neurons) (Fig. 6A–B) and related candidate genes (Mbp, Mag, Pou3f1, Nab2, Nkx6-2, Myrf, Olig2, Fa2h, Sod1, Gal3st1, Lgi4, Fgfr3, Pllp, Kcnj10 and Trf) were notably decreased in Angptl2-null mice (Additional file 1: Fig. S6A). We further confirmed the mRNA expression levels of these candidate genes using quantitative RT-PCR and demonstrated that oligodendrocyte markers (Mbp, Mag), two key transcription factors (Nkx6-2, Myrf), metabolic regulators (Sod1, Gal3st1) and others (Kcnj10, Trf) were markedly downregulated (Fig. 6C). Because MYRF is critical for the differentiation and myelination of oligodendrocytes and enhance the expression of MAG, MBP and MOG [54, 55], we evaluated MYRF protein levels in the brains of Angptl2+/+, Angptl2−/−, Mag+/+, Mag−/−, Angptl2+/+ Mag−/− and Angptl2−/−Mag−/− mice. The results revealed that MYRF was downregulated in Angptl2−/− and Mag−/− mice at day 5 and day 15 compared to their littermates (Fig. 6D–E). However, there was no difference in MYRF protein levels between Angptl2+/+ Mag−/− and Angptl2−/−Mag−/− mice (Fig. 6F), which suggested that MYRF was the downstream candidate target of ANGPLT2-MAG signaling. Fig. 6ANGPTL2-MAG induces Fyn-mediated signaling to enhance the differentiation of oligodendrocytes. A Gene Ontology (GO) analysis of the downregulated differentially expressed genes (DEGs) in the brains of Angptl2+/+ and Angptl2−/− mice at day 15 as determined by RNA sequencing ($$n = 3$$). B Enrichment score plots from GSEA related to the GO signature for myelin sheath and ensheathment of neurons ($$n = 3$$). FDR, false discovery rate; NES, normalized enrichment score. C Relative mRNA levels of potential candidates related to myelination markers, transcription factors, metabolic regulators and other genes in the brain tissues of Angptl2+/+ and Angptl2−/− mice at day 15 as measured by quantitative RT-PCR ($$n = 3$$). D Immunoblot analysis of MYRF and ANGPTL2 protein levels in the brain tissues of Angptl2+/+ and Angptl2−/− mice at day 5, day 15 and day 35. Ratio of MYRF/β-actin was quantified and normalized against Angptl2+/+, respectively. One representative experiment is shown. E–F Immunoblot analysis of MYRF protein levels in the brain tissues of Mag+/+ and Mag−/− mice (E) or Mag−/−Angptl2+/+ and Mag−/−Angptl2−/− mice (F) at day 5, day 15 and day 35. Ratio of MYRF/β-actin was quantified and normalized against Angptl2+/+, respectively. One representative experiment is shown. G–H MAG directly interacted with FYN, as detected by forward (G) or reverse (H) co-immunoprecipitation assays. CMV-MAG (full-length)-FC and pLVX-FYN-strepII plasmids were used in this experiment. One representative experiment is shown. I RSC96 cells with ectopic expression of MAG (full-length)-FLAG and FYN-StrepII were treated with ANGPTL2 proteins, followed by co-immunoprecipitation analysis to evaluate the changes in tyrosine phosphorylation levels of MAG and FYN using 4G10 and p-SRC (Tyr416) antibodies, respectively. The levels of immunoprecipitated protein were quantified and normalized against the control group, respectively. One representative experiment is shown. J RSC96 cells overexpressing FYN-StrepII or MAG (full-length)-FC were subjected to immunoblot analysis to determine MYRF protein levels. Ratio of MYRF/β-actin was quantified and normalized against negative control (empty vector), respectively. One representative experiment is shown. K Western blot analysis of the protein levels of P-SRC (Tyr416), Fyn and MBP in HCN cells 72 h after induction with IGF1 (100 ng/ml), with/without ANGPTL2-Flag (2 μg/ml) and AZD0530 (2 μM) as indicated. Ratios of P-SRC (Tyr416)/β-actin, Fyn/β-actin, MYRF/β-actin and MBP/β-actin were quantified and normalized against the control treated with IGF1 alone, respectively. One representative experiment is shown. L Schematic diagram of the working model for the role of ANGPTL2-MAG in oligodendrocytes differentiation, myelination and differentiation. (*** $p \leq 0.$ 001) Fyn tyrosine kinase is a downstream signal of MAG and plays a critical role in oligodendrocyte differentiation and myelination [56]. However, the detailed mechanisms are not known. We demonstrated that MAG directly interacted with Fyn using a co-immunoprecipitation assay (Fig. 6G–H). MAG was overexpressed in an oligodendrocyte cell line, RSC96, and stimulated with ANGTPL2 to analyze the phosphorylation levels of MAG and Fyn, which are required for the initiation of downstream signaling cascades. Notably, MAG tyrosine phosphorylation levels increased markedly over time, as detected by 4G10 antibodies (Fig. 6I). The immunoprecipitated total and phosphorylation Fyn levels (p-Fyn was determined by the p-SRC416 antibody [57] due to lacking of its specific antibody) were also enhanced in the presence of ANGPTL2 (Fig. 6I). The overexpression of MAG and FYN upregulated the protein levels of MYRF (Fig. 6J). Using in vitro oligodendrocyte induction system with HCN cells, we found purified ANGTPTL2 protein significantly increased the Fyn phosphorylation level as detected by p-SRC (Tyr416) and the expression level of the downstream targets of MYRF and MBP, which can be abolished by adding the FYN inhibitor, AZD0530 (Fig. 6K, Additional file 1: Fig. S6E–F). Moreover, we also found that there were more mature oligodendrocytes with more web-like branches in morphology in the group supplemented with ANGPTL2 compared to the control group, which could be blocked by adding AZD0530 (Additional file 1: Fig. S6C–D). These results further suggested that ANGPTL2 promoted oligodendrocyte maturation mainly through Fyn-mediated signaling pathways. The phosphorylation levels of Fyn were markedly decreased in the brains of Angptl2-null mice on day 5 and day 15, but not day 35 (Additional file 1: Fig. S6B). The protein levels of MBP and MAG were also reduced in Angptl2-null mice at day 15, but not day 35 (Additional file 1: Fig. S6B), which suggested that ANGPTL2 deletion led to a transient delay in myelination. In summary, we demonstrated that ANGPTL2 bound MAG to enhance oligodendrocyte differentiation, myelination or remyelination capacity under physiological and pathological states, which was tightly regulated by the downstream Fyn/MYRF signaling (Fig. 6K). Our study provided a unique angle for understanding the oligodendrocyte differentiation and developing the potential strategy for the treatment of demyeliniation disorders in nerve systems. ## Discussion ANGPTLs (ANGPTL1-8) are secretory proteins with similar structures to angiopoietin proteins and have multiple functions in angiogenesis, tissue repair [26, 29, 58], lipid metabolism, inflammation [28, 59], cardiovascular diseases [60, 61] and cancer development [34, 62]. Unlike the angiopoietin proteins, ANGPTL2 do not bind Tie1 or Tie2 [26], which suggests that ANGPTL2 play different roles by binding unknown receptors. For example, Oike et al. demonstrated that ANGPTL2 bound integrin α5β1 on adipocytes, endothelial cells and cancer cells to promote cell motility via Rac-mediated signaling [29]. CD146 was identified as a novel ANGPLT2 receptor manipulating lipid metabolism and energy expenditure [63]. We showed that LILRB2 was the receptor for ANGPTL2 to sustain the stemness of HSCs and leukemia stem cells [36]. We further revealed that ANGPTL2 bound to MAG on oligodendrocytes to enhance their differentiation via Fyn/MYRF mediated signals. The current study identified MAG as another important receptor for ANGPTL2, and MAG was highly expressed on oligodendrocytes with a high binding affinity (KD ≈ 10 nM), which was comparable to the binding affinity between ANGPTL2 and LILRB2 (KD = 5.5 nM). Notably, LILRB2 was rarely expressed on oligodendrocytes (https://www.proteinatlas), which indicated that ANGPTL2 exerted its effects primarily via MAG-mediated pathways. These findings provide deep insight into the potential roles of ANGPTL2 in the regulation of oligodendrocyte differentiation and demyelinating disorders. However, the cell types in the CNS that primarily provide ANGPTL2 for oligodendrocyte differentiation are largely unknown. Although ANGPTL2 is primarily secreted by endothelial cells [64], macrophages [65] or adipocytes [29], public databases also indicate that oligodendrocyte precursor cells and oligodendrocytes have the highest expression levels of ANGTPL2 in the human brain (https://www.proteinatlas). However, the expression profile maybe different between human and mouse brains. Further efforts are required for the identification of the potential cell types (including oligodendrocyte precursor cells and oligodendrocytes) that contribute to ANGPTL2 levels in the brain for oligodendrocyte differentiation and myelination and the difference of ANGPTL2 expression profiles between human and mouse brains must be elucidated. We also found that relatively high ANGPTL2 protein level was still expressed with oligodendrocyte differentiation from HCN cells in vitro, which suggested that the cell-autonomous effect might play a role in oligodendrocyte differentiation. We currently are testing its potential autocrine effect by knocking down Angptl2 in HCN cells during in vitro differentiation. Meanwhile, other cell types in the brain may be alternative sources for ANGPTL2 to trigger MAG-mediated signaling and promote oligodendrocyte differentiation in a paracrine manner. Interestingly, it seemed that MAG was only expressed on the differentiated oligodendrocytes, but not oligodendrocyte precursor cells (Additional file 1: Fig. S3G–H), and ANGPTL2 had no effect on the proliferation and differentiation of oligodendrocyte precursor cells (Additional file 1: Fig. S3F–G). Meanwhile, ANGPTL2 was also highly expressed in oligodendrocyte precursor cells (Additional file 1: Fig. S3D–E), suggested that oligodendrocyte precursor cells might secreted ANGPTL2 to support oligodendrocyte maturation in a paracrine manner. However, due to the lack of the suitable ANGPTL2 antibody for immunohistochemistry, more efforts are required to delineate the expression patterns of ANGPTL2 in the mouse brain. Moreover, we found that both of ANGPTL2 and MYRF level were gradually declined with ages (Fig. 6D), which might suggest that MYRF level was controlled by ANGPTL2/MAG signaling in an ANGPTL2 dose dependent manner. We expect that ANGPTL2 may mainly promote oligodendrocyte differentiation at the early stage or the process of remyelination, when large amounts of mature oligodendrocytes are required for rapid myelination. However, it is very difficult to perform rescue experiment in neonatal mice because ANGPTL2 may be not effectively transferred or reached enough amount in specific microenviroment to support oligodendrocyte maturation after ICV injection. Similarly, IP injection may also result in the failure in delivering ANGPTL2 into specific microenviroment for oligodendrocyte maturation due to the existence of blood–brain barrier. Other ways to increase ANGPTL2 level in brains are required to further delineate the function of ANGPTL2 in oligodendrocyte maturation. The transient phenotype in developmental myelination in knockout animals indicates that ANGPTL2 is not fully necessary for full differentiation, and neither the maintenance of normal myelin in the adult animals. It will be also important to elucidate how ANGPTL2 affects oligodendrocyte differentiation at different stages and whether other ANGPTLs are required for differentiation. MAG acts as a glue or spacer for glia and axons [66, 67] and plays an important role in the maintenance of glia-axon communication. However, the roles and detailed mechanisms of MAG in myelin formation are controversial. Li et al. reported that MAG was not critical for myelin formation, but it was required for the maintenance of the cytoplasmic collar and periaxonal space of myelinated axons in adult mice [44]. However, Pernet et al. found a delayed oligodendrocyte differentiation and abnormal myelin structure in central nervous system of Mag-null mice within the first month after birth [53]. We showed that ANGPTL2-MAG signaling enhanced oligodendrocyte differentiation and myelination at the early stage and remyelination progression, under certain pathological stresses. These results strengthen the hypothesis that MAG is critical for the homeostasis of the oligodendrocyte pool and its myelination function. MAG has been reported to be a ligand for the Nogo receptor (NgR) and PirB, which mainly involves in the inhibition of axon elongation [39–41, 68]. For example, MAG and Nogo66 can compete for binding to NgR to inhibit neurite outgrowth [39, 68], PirB is another receptor for MAG, Nogo66 and OMgp to serve as the inhibitor in axonal regeneration [40]. MAG serves as a receptor for nerve cell surface gangliosides GD1a and GT1b to mediate nerve regeneration inhibition [42]. Interestingly, MAG also can form the complex with β1-integrin to mediate axonal growth cone repulsive response of hippocampal neurons independent of NgR through FAK activation [41]. However, except for its inhibitory effect on axon growth, whether MAG can serve as a receptor to mediate the downstream signaling to enhance the oligodendrocyte differentiation is not clear. Herein, we demonstrated that ANGPTL2 served as the potent ligand for MAG and enhanced oligodendrocyte differentiation by increasing the phosphorylation of MAG to recruit and activate the non-receptor tyrosine kinase Fyn to enhance the expression of the downstream key transcription factor MYRF and its targets of some myelin-related markers, such as MBP, MAG and MOG [54, 55]. These results provide a unique angle to understand the multifaceted functions of MAG via individual interactions with certain surface molecules. However, how MAG activates Fyn kinase and the downstream target MYRF, which further connect to the differentiation and myelination of oligodendrocytes, is largely unknown. ## Animals The Angptl2 knockout mice (Angptl2fl/fl) with insertion of loxp in flanks of exon 2 in C57BL/6 background were generated by Nanjing Mouse Model, Lt Corporation. The male Angptl2fl/fl; stra8-cre mice (Angptl2 is specifically deleted in early-stage spermatogonia, spermatocytes and sperm) were crossed with wild-type (WT) female mice to generate Angptl2 knockout mice (Angptl2−/−). The genetic deletion of Angptl2 was further confirmed by PCR. The Mag knockout mice (Mag−/−) in C57BL/6 background were purchased from Mutant Mouse Regional Resource Centers (MMRRC). Mag−/−Angptl2−/− double knockout mice and their littermates Mag−/−Angptl2+/+ mice were also bred for the related experiments. The Guideline for Animal Care at Shanghai Jiao Tong University School of Medicine approved all the animal experimental procedures. ## Murine cuprizone-induced demyelination model and rotarod test The 8–10 week old mice were fed with a standard rodent chow with $0.25\%$ (w/w) cuprizone (bis (cycloheanone) oxaldihydrazone; Sigma) for 5 weeks to induce the demyelination in mouse brains, followed by the switch to the normal chow for 2 additional weeks to allow the recovery from demyelination. The cuprizone-induced demyelinated or remyelinated mice were anesthetized, perfused with PBS, followed by fixation with $4\%$ PFA. The whole mouse brains were removed, post-fixed, sectioned at 30 μm using a vibratome, and subjected for the oil-red O staining, immunohistochemical analysis and electron microcopy analysis. In consideration of the cuprizone induced lesions are in the variability in lesion size and location, the corresponding section every ten serial coronal sections in the corpus callosum of mice was selected and 6–8 sections in one brain were stained for analysis. For the motor coordination assessment, the 8–10 week old mice first received a training of running on a rotating rod at an accelerating speed from 4 to 40 rotations per min for 300 s for 1 week (Harvard apparatus, UK). The mice that could still stay on the rotating cylinder at a speed of 4 rotations/min for 300 s were used for the following motor coordination analysis. The latency to fall off a rotating rod at a speed of 4 rotations/min and the body weight of each mouse were measured every 3 days during the 5 weeks’ cuprizone chow feeding and following 2 weeks’ normal chow feeding. ## Oil-red O staining and demyelination area scoring Oil-red O Assay Kit (BASO Life Technologies, BA4081) was used for the evaluation of myelination status according to the manufacturer’s instructions. Briefly, coronal brain slices were rehydrated with distilled water for 5 min, treated with $60\%$ isopropanol for 2 min and rinsed in oil red O for 10 min. Excess stain was removed by washing the slides with $60\%$ isopropanol followed by washing with distilled water. Slices were counterstained with hematoxylin to visualize nuclei. Images were taken using an Olympus IX71 inverted fluorescence microscope and quantitative analysis was performed using Olympus cellsens. Region of interest at corpus callosum was drawn using the “irregular AOI” tool and red areas were counted within the lesion areas using the “count and measure objects” tool. Percentage of the demyelination area was calculated by the ratio of the unstained area and total corpus callosum area. ## Immunofluorescence staining and immunohistochemistry For immunofluorescence staining, coronal brain slices were blocked with permeable buffer ($0.3\%$ Triton X-100 in PBS) containing $10\%$ donkey serum for an hour at room temperature and incubated with primary antibodies in permeable buffer containing $2\%$ donkey serum overnight at 4 °C. The slices were then washed three times with PBS-T ($0.1\%$ Tween 20 in PBS) for 10 min each time and incubated with Alexa Fluor secondary antibodies (Thermo Fisher) in the PBS for 2 h at room temperature. Nuclei were counterstained with DAPI (Invitrogen). For primary antibodies, rabbit anti-MBP (CST, Cat#78896), mouse anti-MAG (Abcam, ab89780), mouse anti-MOG polyclonal antibody (Beyotime, AF7488 and Santa Cruz Biotech, SC-166172), rabbit anti-GFAP and (CST, Cat#12389). Images were taken using an Olympus IX71 inverted fluorescence microscope, and quantitative image analysis was performed using Image J. In some cases, an adult hippocampus-derived neural progenitor cell line, HCN cells, was used for the analysis of ANGPTL2-mediated oligodendrocyte differentiation. HCN cells were plated on ornithine-pretreated glass, fixed with $4\%$ PFA for 10 min at room temperature, blocked with TBS containing $0.1\%$ Triton and $2\%$ BSA, and incubated with primary antibodies, including rabbit anti-MBP (CST, Cat#78896), mouse anti-MOG (Beyotime, AF7488 and Santa Cruz Biotech, SC-166172), mouse anti-MAG (Abcam, ab89780), mouse anti-PDGFRα (R&D, AF1062) and mouse anti-Ki-67-FITC (BD, Cat#612472) at 4 °C overnight. After washing with PBS for three times, HCN cells were incubated with the appropriated secondary antibody conjugated with Alexa Fluor 555 (Thermo Fisher) for one hour at room temperature. Nuclei were counterstained with DAPI (Invitrogen). ## Electron microscopy analysis Mice were perfused with the PBS buffer containing $2.5\%$ PFA and $2.5\%$ glutaraldehyde and mouse brains were isolated for the subsequent electron microscopy analysis. Corpus callosum were washed, fixed in $1\%$ osmium tetroxide, dehydrated in acetone and embedded in EPON. A 70-nm thinness sagittal section was cut with a diamond knife, mounted on copper slot grids precoated with Formvar and stained with uranyl acetate and lead citrate for the examination of myelination with Hitachi H-7650 transmission electron microscope. G-ratios were determined using Image J software. Approximately 120–200 axons in each mouse (three mice per group) were analyzed and the total numbers of axons counted were about 360–600 in each group. The significant difference was analyzed by Unpaired Student’s t test (two-tailed) as shown in Fig. 3F, H, and 200 ~ 500 remyelinated axons were calculated for each group in Fig. 4F, 5F, N. ## Cell culture 293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with $10\%$ FBS. An adult hippocampus-derived neural progenitor cell line, HCN cells [46, 69], were cultured in DMEM-F12 (Gibco) plus N2 supplement (Gibco, 17502-048), L-Glutamine (Gibco, A2916801), 20 ng/ml bFGF (Peprotech, AF-100-18B) and penicillin–streptomycin (P/S). To induce the differentiation of HCN cells to mature oligodendrocytes, HCN cells were carefully digested with $0.05\%$ trypsin, neutralized with defined trypsin inhibitor (Gibco, R-007-100), washed with PBS to get rid of all the remaining growth factor of bFGF, followed by culturing in the medium of DMEM-F12 plus 100 ng/ml IGF1 (Peprotech, 100-11), N2 supplement (Gibco, 17502-048), L-Glutamine (Gibco, A2916801) and P/S for additional 3 days. All the plates for the cell growth and differentiation of HCN cells were pretreated a with poly-ornithine (Sigma, P3655) and murine laminin (Invitrogen, 23017–015). ## Flow cytometric analysis for the ANGPTL2 binding activities to MAG CMV-ANGPTL2-Flag plasmid or other plasmids of ANGPTL members [36] were transfected into 293T cells and the conditioned medium was collected 48 h after transfection for the binding assay with MAG or the purification of ANGPTL2 using M2 resin (Sigma, A2220). Human or murine MAG protein fused with GFP were cloned in the CMV-N1-GFP plasmid and transfected into 293T cells. The 293T cells were collected 48 h after transfection and incubated with ANGPTL2-Flag condition medium or control medium at 4 °C for 1–2 h, followed by the incubation with the secondary antibody of anti-Flag-allophycocyanin (APC) (Biolegend, Cat# 637308) and propidium iodide (PI). The potential binding activities of ANGPTL2 to MAG were determined by flow cytometric analysis. ## Selecting surface molecules for ANGPTL2 The selecting surface molecules were mainly based on a flow cytometric assay. The membrane protein library contained individual expression plasmid without reporting tags, such as GFP or mCherry. To indicate the expression level of transfected surface molecules, the fluorescence indictor of XZ201-GFP plasmid was co-transfected with individual membrane plasmid. Individual membrane plasmid (0.25 µg) and XZ201-GFP plasmid were mixed at ratio of 1:1 and transiently transfected into 293T cells. The transfected 293T cells were collected 48 h later and incubated with conditional medium containing ANGPTL2-FLAG protein at 4 °C for 1–2 h, followed by the incubation with the secondary antibody of anti-Flag-allophycocyanin (APC) and propidium iodide (PI). The potential binding activities of ANGPTL2 to membrane protein were determined by flow cytometric analysis. To prepare the ligand of ANGPTL2, the CMV-ANGPTL2-Flag plasmid was used for the transfection into 293T cells and the conditioned medium was collected 48 h later as described previously. ## Chimeric receptor reporter assay Human or murine MAG chimeric receptor cells were established for the evaluation of the binding between ANGPTL2 and MAG according to the protocol previously described [43]. In brief, the ECDs of MAG and its mutants (IgG3 and IgG4) were fused to the intracellular domain PIRLb, which could further recruit adaptor DAP12 to transactivate NFAT-GFP expression upon binding to their potential ligands, such as ANGPTL2. In this assay, purified ANGPTL2 protein (0.02 mg/mL) was pre-coated on 96-well plate at 37 °C for 4–6 h, followed by culture with 4 × 104 MAG reporter cells or control reporter cells in each well. The percentage of GFP+ reporter cells that represented the activation by ANGPTL2 binding to MAG were measured by flow cytometry 24 h after culture. ## Bio-layer interferometry or surface plasmon resonance The binding affinity of human or murine MAG to ANGPTL2 was first determined by Octet RED96/Octet RED96e instrument (ForteBio). Condition medium containing human or murine MAG ECD fused with human Fc (MAG-hFC) was collected 48 h after transfection. The AHC biosensors were coated with MAG-hFc protein (condition medium containing human or mouse MAG-hFc protein was loaded on biosensors for 480 s/600 s), washed with kinetics buffer for 300 s/300 s before the determination of the association (300 s/90 s) and dissociation (800 s/90 s) constant upon binding to purified ANGPTL2 protein at indicated doses. Data were analyzed using ForteBio Data Analysis Software v9. Alternatively, Biacore T200 instrument (GE Healthcare) and CM5 sensor chips were used to determine the binding affinity of human or murine MAG-hFC to ANGPTL2 protein. Briefly, anti-hFc antibody (Sigma, Cat#12136) was pre-immobilized in parallel-flow channels of a CM5 sensor chip using the amine coupling kit (GE Healthcare). Human or murine MAG-hFc in the condition medium was injected into one of the channels and captured by the anti-hFc antibodies pre-coated on the CM5 sensor chip. To measure the binding affinity of human or murine MAG-hFc to ANGPTL2, indicated doses of purified ANGPTL2 protein were injected into the flow system. The binding affinity constants were analyzed with Biacore T200 evaluation software V3. ## RT-PCR The total RNA of HCN cells and their differentiated cells, or Angptl2+/+ and Angptl2−/− mouse brain tissues were extracted and used for real-time RT-PCR. The reactions were performed as previously described [70]. Briefly, 10 μL reactions with 2 × ABI SYBR® Green PCR master mix, primers and cDNA were used for the evaluation of indicated gene expression levels. The experiments were conducted in triplicate with Applied Biosystems 7900HT. The mRNA level was normalized to the level of β-actin RNA transcripts. The primer sequences for related genes were showed in Additional files 2. The unedited agarose gel figures were showed in Additional file 3. ## Co-immunoprecipitation Plasmids encoding human or murine MAG-ECD-Fc, Tie2-ECD-Fc, ANGPTL2-strep II and control vector were transiently co-transfected into 293T cells. The supernatant were collected 48 h after transfection and incubated with Protein A/G beads (Santa Cruz, sc2003) at 4 °C for 8 h, followed by washing with pre-chilled PBS with $0.1\%$ NP-40 for 5 times. In the co-immunoprecipitation experiment for FYN and MAG, 293 T cells were transiently co-transfected with MAG (full-length)-FC, FYN-strepII and control vectors, lysed and incubated with Protein A/G beads (Santa Cruz, sc2003) or anti-strepII (GenScript, Cat#A00626-40) at 4 °C overnight, followed by washing with pre-chilled PBS with $0.1\%$ NP-40 for 8 times. The immunoprecipitated proteins were examined by using the indicated antibodies for anti-strepII (GenScript, Cat#A00626-40) or anti-Fc (Sigma, Cat#12136) by western blot. The unedited agarose gel figures were showed in Additional file3 ## Western blot Brain tissues isolated from Angptl2−/−, Mag−/−, Mag−/−Angptl2+/+, Mag−/−Angptl2−/− and their controls were homogenized with RIPA lysis buffer (Beyotime, Cat#P0013C; supplemented with 1 mM PMSF, 2 mM sodium orthovanadate and protease inhibitor) at 4 °C for 30 min, and ultrasonicated for western blot analysis. Samples were separated by SDS-PAGE, transferred to nitrocellulose membranes and immunoblotted with primary antibodies as following: anti-MBP (CST, Cat#78896), anti-MAG (Abcam, ab89780), anti-MOG polyclonal antibody (Santa Cruz Biotech, SC-166172), anti-Fyn (Abcam, ab184276), anti-Phospho-Src family (Tyr416) (CST, Cat #2101), anti-MYRF (Abclonal, A16355), anti-ANGPTL2 (R&D, AF1444) and anti-β-Actin-pAb-HRP-DirecT (MBL, PM053-7). The unedited western blot gel figures were showed in Additional file 3. ## Statistical analysis GraphPad and SPSS software programs, version 19.0, were used for statistical analysis. The results are presented as mean ± SD. The data were analyzed by Student’s t test (two-tailed), one-way ANOVA with Tukey’s multiple comparison test, or two-way ANOVA with Sidak’s multiple comparison test according to the experimental design. All experiments were performed independently more than 3 times. For all experiments, p ≤ 0.05 was considered a significant difference ($p \leq 0.05$; $p \leq 0.01$; $p \leq 0.001$). ## Supplementary Information Additional file 1: Figure S1. Related to Fig. 1. Human MAG is a new receptor for ANGPTL2. Figure S2. related to Fig. 2. Murine homologue of human MAG binds to ANGPTL2. Figure S3. Related to Fig. 3. ANGPTL2 promotes oligodendrocyte differentiation. Figure S4. Related to Fig. 4. ANGPTL2 promotes oligodendrocyte differentiation and remyelination under pathological conditions. Figure S5. Related to Fig. 5. ANGPTL2 promotes remyelination via its receptor MAG. Figure S6. Related to Fig. 6. ANGPTL2-MAG induces Fyn-mediated signalling to enhance the differentiation of oligodendrocytes. Additional file 2: Table S1. Primer sequences for q-RT-PCR and genotyping. Additional file 3: Unedited images from gels and western blots. ## References 1. Young KM, Psachoulia K, Tripathi RB, Dunn SJ, Cossell L, Attwell D. Oligodendrocyte dynamics in the healthy adult CNS: evidence for myelin remodeling. Neuron* (2013.0) 77 873-85. DOI: 10.1016/j.neuron.2013.01.006 2. McKenzie IA, Ohayon D, Li H, de Faria JP, Emery B, Tohyama K. Motor skill learning requires active central myelination. Science (2014.0) 346 318-22. DOI: 10.1126/science.1254960 3. Reich DS, Lucchinetti CF, Calabresi PA. Multiple Sclerosis. N Engl J Med (2018.0) 378 169-80. DOI: 10.1056/NEJMra1401483 4. Swiss VA, Nguyen T, Dugas J, Ibrahim A, Barres B, Androulakis IP. Identification of a gene regulatory network necessary for the initiation of oligodendrocyte differentiation. PLoS One (2011.0) 6 e18088. DOI: 10.1371/journal.pone.0018088 5. Gaesser JM, Fyffe-Maricich SL. Intracellular signaling pathway regulation of myelination and remyelination in the CNS. Exp Neurol (2016.0) 283 501-11. DOI: 10.1016/j.expneurol.2016.03.008 6. Steelman AJ, Zhou Y, Koito H, Kim S, Payne HR, Lu QR. Activation of oligodendroglial Stat3 is required for efficient remyelination. Neurobiol Dis (2016.0) 91 336-46. DOI: 10.1016/j.nbd.2016.03.023 7. He D, Marie C, Zhao C, Kim B, Wang J, Deng Y. Chd7 cooperates with Sox10 and regulates the onset of CNS myelination and remyelination. Nat Neurosci (2016.0) 19 678-89. DOI: 10.1038/nn.4258 8. Jarjour AA, Boyd A, Dow LE, Holloway RK, Goebbels S, Humbert PO. The polarity protein scribble regulates myelination and remyelination in the central nervous system. PLoS Biol (2015.0) 13 e1002107. DOI: 10.1371/journal.pbio.1002107 9. Nakatani H, Martin E, Hassani H, Clavairoly A, Maire CL, Viadieu A. Ascl1/Mash1 promotes brain oligodendrogenesis during myelination and remyelination. J Neurosci (2013.0) 33 9752-68. DOI: 10.1523/JNEUROSCI.0805-13.2013 10. Qian Z, Li H, Yang H, Yang Q, Lu Z, Wang L. Osteocalcin attenuates oligodendrocyte differentiation and myelination via GPR37 signaling in the mouse brain. Sci Adv (2021.0) 7 eabi5811. DOI: 10.1126/sciadv.abi5811 11. Miller RH, Fyffe-Maricich SL. Restoring the balance between disease and repair in multiple sclerosis: insights from mouse models. Dis Model Mech (2010.0) 3 535-9. DOI: 10.1242/dmm.001958 12. Zhang H, Jarjour AA, Boyd A, Williams A. Central nervous system remyelination in culture–a tool for multiple sclerosis research. Exp Neurol (2011.0) 230 138-48. DOI: 10.1016/j.expneurol.2011.04.009 13. Fancy SP, Kotter MR, Harrington EP, Huang JK, Zhao C, Rowitch DH. Overcoming remyelination failure in multiple sclerosis and other myelin disorders. Exp Neurol (2010.0) 225 18-23. DOI: 10.1016/j.expneurol.2009.12.020 14. Chang A, Tourtellotte WW, Rudick R, Trapp BD. Premyelinating oligodendrocytes in chronic lesions of multiple sclerosis. N Engl J Med (2002.0) 346 165-73. DOI: 10.1056/NEJMoa010994 15. Kuhlmann T, Miron V, Cui Q, Wegner C, Antel J, Bruck W. Differentiation block of oligodendroglial progenitor cells as a cause for remyelination failure in chronic multiple sclerosis. Brain (2008.0) 131 1749-58. DOI: 10.1093/brain/awn096 16. Moyon S, Huynh JL, Dutta D, Zhang F, Ma D, Yoo S. Functional characterization of DNA methylation in the oligodendrocyte lineage. Cell Rep (2016.0) 15 748-60. DOI: 10.1016/j.celrep.2016.03.060 17. He D, Wang J, Lu Y, Deng Y, Zhao C, Xu L. lncRNA functional networks in oligodendrocytes reveal stage-specific myelination control by an lncOL1/Suz12 complex in the CNS. Neuron (2017.0) 93 362-78. DOI: 10.1016/j.neuron.2016.11.044 18. [18]Moyon S, Ma D, Huynh JL, Coutts DJC, Zhao C, Casaccia P, et al. Efficient Remyelination Requires DNA Methylation. eNeuro. 2017;4(2). 19. Emery B, Agalliu D, Cahoy JD, Watkins TA, Dugas JC, Mulinyawe SB. Myelin gene regulatory factor is a critical transcriptional regulator required for CNS myelination. Cell (2009.0) 138 172-85. DOI: 10.1016/j.cell.2009.04.031 20. Duncan GJ, Plemel JR, Assinck P, Manesh SB, Muir FGW, Hirata R. Myelin regulatory factor drives remyelination in multiple sclerosis. Acta Neuropathol (2017.0) 134 403-22. DOI: 10.1007/s00401-017-1741-7 21. Yu Y, Chen Y, Kim B, Wang H, Zhao C, He X. Olig2 targets chromatin remodelers to enhancers to initiate oligodendrocyte differentiation. Cell (2013.0) 152 248-61. DOI: 10.1016/j.cell.2012.12.006 22. Li H, Lu Y, Smith HK, Richardson WD. Olig1 and Sox10 interact synergistically to drive myelin basic protein transcription in oligodendrocytes. J Neurosci (2007.0) 27 14375-82. DOI: 10.1523/JNEUROSCI.4456-07.2007 23. Wheeler NA, Lister JA, Fuss B. The autotaxin-lysophosphatidic acid axis modulates histone acetylation and gene expression during oligodendrocyte differentiation. J Neurosci (2015.0) 35 11399-414. DOI: 10.1523/JNEUROSCI.0345-15.2015 24. Yuelling LM, Fuss B. Autotaxin (ATX): a multi-functional and multi-modular protein possessing enzymatic lysoPLD activity and matricellular properties. Biochim Biophys Acta (2008.0) 1781 525-30. DOI: 10.1016/j.bbalip.2008.04.009 25. Syed YA, Hand E, Mobius W, Zhao C, Hofer M, Nave KA. Inhibition of CNS remyelination by the presence of semaphorin 3A. J Neurosci. (2011.0) 31 3719-28. DOI: 10.1523/JNEUROSCI.4930-10.2011 26. Kim I, Moon SO, Koh KN, Kim H, Uhm CS, Kwak HJ. Molecular cloning, expression, and characterization of angiopoietin-related protein. Angiopoietin-related protein induces endothelial cell sprouting. J Biol Chem (1999.0) 274 26523-8. DOI: 10.1074/jbc.274.37.26523 27. Oike Y, Yasunaga K, Suda T. Angiopoietin-related/angiopoietin-like proteins regulate angiogenesis. Int J Hematol (2004.0) 80 21-8. DOI: 10.1532/IJH97.04034 28. Hato T, Tabata M, Oike Y. The role of angiopoietin-like proteins in angiogenesis and metabolism. Trends Cardiovasc Med (2008.0) 18 6-14. DOI: 10.1016/j.tcm.2007.10.003 29. Tabata M, Kadomatsu T, Fukuhara S, Miyata K, Ito Y, Endo M. Angiopoietin-like protein 2 promotes chronic adipose tissue inflammation and obesity-related systemic insulin resistance. Cell Metab (2009.0) 10 178-88. DOI: 10.1016/j.cmet.2009.08.003 30. Kadomatsu T, Tabata M, Oike Y. Angiopoietin-like proteins: emerging targets for treatment of obesity and related metabolic diseases. FEBS J (2011.0) 278 559-64. DOI: 10.1111/j.1742-4658.2010.07979.x 31. Fan X, Shi P, Dai J, Lu Y, Chen X, Liu X. Paired immunoglobulin-like receptor B regulates platelet activation. Blood (2014.0) 124 2421-30. DOI: 10.1182/blood-2014-03-557645 32. Horio E, Kadomatsu T, Miyata K, Arai Y, Hosokawa K, Doi Y. Role of endothelial cell-derived angptl2 in vascular inflammation leading to endothelial dysfunction and atherosclerosis progression. Arterioscler Thromb Vasc Biol (2014.0) 34 790-800. DOI: 10.1161/ATVBAHA.113.303116 33. Motokawa I, Endo M, Terada K, Horiguchi H, Miyata K, Kadomatsu T. Interstitial pneumonia induced by bleomycin treatment is exacerbated in Angptl2-deficient mice. Am J Physiol Lung Cell Mol Physiol (2016.0) 311 L704-L13. DOI: 10.1152/ajplung.00005.2016 34. Endo M, Nakano M, Kadomatsu T, Fukuhara S, Kuroda H, Mikami S. Tumor cell-derived angiopoietin-like protein ANGPTL2 is a critical driver of metastasis. Cancer Res (2012.0) 72 1784-94. DOI: 10.1158/0008-5472.CAN-11-3878 35. Odagiri H, Kadomatsu T, Endo M, Masuda T, Morioka MS, Fukuhara S. The secreted protein ANGPTL2 promotes metastasis of osteosarcoma cells through integrin alpha5beta1, p38 MAPK, and matrix metalloproteinases. Sci Signal (2014.0) 7 ra7. DOI: 10.1126/scisignal.2004612 36. Zheng J, Umikawa M, Cui C, Li J, Chen X, Zhang C. Inhibitory receptors bind ANGPTLs and support blood stem cells and leukaemia development. Nature (2012.0) 485 656-60. DOI: 10.1038/nature11095 37. Quarles RH. Myelin-associated glycoprotein (MAG): past, present and beyond. J Neurochem (2007.0) 100 1431-48. PMID: 17241126 38. Lopez PH. Role of myelin-associated glycoprotein (siglec-4a) in the nervous system. Adv Neurobiol (2014.0) 9 245-62. DOI: 10.1007/978-1-4939-1154-7_11 39. Domeniconi M, Cao Z, Spencer T, Sivasankaran R, Wang K, Nikulina E. Myelin-associated glycoprotein interacts with the Nogo66 receptor to inhibit neurite outgrowth. Neuron (2002.0) 35 283-90. DOI: 10.1016/S0896-6273(02)00770-5 40. Atwal JK, Pinkston-Gosse J, Syken J, Stawicki S, Wu Y, Shatz C. PirB is a functional receptor for myelin inhibitors of axonal regeneration. Science (2008.0) 322 967-70. DOI: 10.1126/science.1161151 41. Goh EL, Young JK, Kuwako K, Tessier-Lavigne M, He Z, Griffin JW. beta1-integrin mediates myelin-associated glycoprotein signaling in neuronal growth cones. Mol Brain (2008.0) 1 10. DOI: 10.1186/1756-6606-1-10 42. Vyas AA, Patel HV, Fromholt SE, Heffer-Lauc M, Vyas KA, Dang J. Gangliosides are functional nerve cell ligands for myelin-associated glycoprotein (MAG), an inhibitor of nerve regeneration. Proc Natl Acad Sci USA (2002.0) 99 8412-7. DOI: 10.1073/pnas.072211699 43. Deng M, Lu Z, Zheng J, Wan X, Chen X, Hirayasu K. A motif in LILRB2 critical for Angptl2 binding and activation. Blood (2014.0) 124 924-35. DOI: 10.1182/blood-2014-01-549162 44. Li C, Tropak MB, Gerlai R, Clapoff S, Abramow-Newerly W, Trapp B. Myelination in the absence of myelin-associated glycoprotein. Nature (1994.0) 369 747-50. DOI: 10.1038/369747a0 45. Montag D, Giese KP, Bartsch U, Martini R, Lang Y, Bluthmann H. Mice deficient for the myelin-associated glycoprotein show subtle abnormalities in myelin. Neuron (1994.0) 13 229-46. DOI: 10.1016/0896-6273(94)90472-3 46. Hsieh J, Aimone JB, Kaspar BK, Kuwabara T, Nakashima K, Gage FH. IGF-I instructs multipotent adult neural progenitor cells to become oligodendrocytes. J Cell Biol (2004.0) 164 111-22. DOI: 10.1083/jcb.200308101 47. Brunner C, Lassmann H, Waehneldt TV, Matthieu JM, Linington C. Differential ultrastructural localization of myelin basic protein, myelin/oligodendroglial glycoprotein, and 2',3'-cyclic nucleotide 3'-phosphodiesterase in the CNS of adult rats. J Neurochem (1989.0) 52 296-304. DOI: 10.1111/j.1471-4159.1989.tb10930.x 48. Linnington C, Webb M, Woodhams PL. A novel myelin-associated glycoprotein defined by a mouse monoclonal antibody. J Neuroimmunol (1984.0) 6 387-96. DOI: 10.1016/0165-5728(84)90064-X 49. Trapp BD. Myelin-associated glycoprotein. Location and potential functions. Ann NY Acad Sci (1990.0) 605 29-43. DOI: 10.1111/j.1749-6632.1990.tb42378.x 50. Matsushima GK, Morell P. The neurotoxicant, cuprizone, as a model to study demyelination and remyelination in the central nervous system. Brain Pathol (2001.0) 11 107-16. DOI: 10.1111/j.1750-3639.2001.tb00385.x 51. Steelman AJ, Thompson JP, Li J. Demyelination and remyelination in anatomically distinct regions of the corpus callosum following cuprizone intoxication. Neurosci Res (2012.0) 72 32-42. DOI: 10.1016/j.neures.2011.10.002 52. Tsiperson V, Li X, Schwartz GJ, Raine CS, Shafit-Zagardo B. GAS6 enhances repair following cuprizone-induced demyelination. PLoS One (2010.0) 5 e15748. DOI: 10.1371/journal.pone.0015748 53. Pernet V, Joly S, Christ F, Dimou L, Schwab ME. Nogo-A and myelin-associated glycoprotein differently regulate oligodendrocyte maturation and myelin formation. J Neurosci (2008.0) 28 7435-44. DOI: 10.1523/JNEUROSCI.0727-08.2008 54. Bujalka H, Koenning M, Jackson S, Perreau VM, Pope B, Hay CM. MYRF is a membrane-associated transcription factor that autoproteolytically cleaves to directly activate myelin genes. PLoS Biol (2013.0) 11 e1001625. DOI: 10.1371/journal.pbio.1001625 55. Koenning M, Jackson S, Hay CM, Faux C, Kilpatrick TJ, Willingham M. Myelin gene regulatory factor is required for maintenance of myelin and mature oligodendrocyte identity in the adult CNS. J Neurosci (2012.0) 32 12528-42. DOI: 10.1523/JNEUROSCI.1069-12.2012 56. Umemori H, Sato S, Yagi T, Aizawa S, Yamamoto T. Initial events of myelination involve Fyn tyrosine kinase signalling. Nature (1994.0) 367 572-6. DOI: 10.1038/367572a0 57. Matte A, Lupo F, Tibaldi E, Di Paolo ML, Federti E, Carpentieri A. Fyn specifically regulates the activity of red cell glucose-6-phosphate-dehydrogenase. Redox Biol (2020.0) 36 101639. DOI: 10.1016/j.redox.2020.101639 58. Kubota Y, Oike Y, Satoh S, Tabata Y, Niikura Y, Morisada T. Isolation and expression patterns of genes for three angiopoietin-like proteins, Angptl 1, 2 and 6 in zebrafish. Gene Expr Patterns (2005.0) 5 679-85. DOI: 10.1016/j.modgep.2005.02.006 59. Mattijssen F, Kersten S. Regulation of triglyceride metabolism by angiopoietin-like proteins. Biochim Biophys Acta (2012.0) 1821 782-9. DOI: 10.1016/j.bbalip.2011.10.010 60. Wang Z, Zheng H, Chen H, Lin X, Chen J, Wang L. Elevated serum angiopoietin-like protein 2 in patients with acute coronary syndrome. Arch Med Res (2015.0) 46 257-64. DOI: 10.1016/j.arcmed.2015.05.003 61. Aoi J, Endo M, Kadomatsu T, Miyata K, Ogata A, Horiguchi H. Angiopoietin-like protein 2 accelerates carcinogenesis by activating chronic inflammation and oxidative stress. Mol Cancer Res (2014.0) 12 239-49. DOI: 10.1158/1541-7786.MCR-13-0336 62. Toiyama Y, Tanaka K, Kitajima T, Shimura T, Kawamura M, Kawamoto A. Elevated serum angiopoietin-like protein 2 correlates with the metastatic properties of colorectal cancer: a serum biomarker for early diagnosis and recurrence. Clin Cancer Res (2014.0) 20 6175-86. DOI: 10.1158/1078-0432.CCR-14-0007 63. Wu Z, Liu J, Chen G, Du J, Cai H, Chen X. CD146 is a novel ANGPTL2 receptor that promotes obesity by manipulating lipid metabolism and energy expenditure. Adv Sci (Weinh) (2021.0) 8 2004032. DOI: 10.1002/advs.202004032 64. [64]Huang D, Sun G, Hao X, He X, Zheng Z, Chen C, et al. ANGPTL2-containing small extracellular vesicles from vascular endothelial cells accelerate leukemia progression. J Clin Invest. 2021;131(1). 65. Tazume H, Miyata K, Tian Z, Endo M, Horiguchi H, Takahashi O. Macrophage-derived angiopoietin-like protein 2 accelerates development of abdominal aortic aneurysm. Arterioscler Thromb Vasc Biol (2012.0) 32 1400-9. DOI: 10.1161/ATVBAHA.112.247866 66. Owens GC, Bunge RP. Schwann cells depleted of galactocerebroside express myelin-associated glycoprotein and initiate but do not continue the process of myelination. Glia (1990.0) 3 118-24. DOI: 10.1002/glia.440030205 67. Owens GC, Bunge RP. Schwann cells infected with a recombinant retrovirus expressing myelin-associated glycoprotein antisense RNA do not form myelin. Neuron (1991.0) 7 565-75. DOI: 10.1016/0896-6273(91)90369-B 68. Liu BP, Fournier A, GrandPre T, Strittmatter SM. Myelin-associated glycoprotein as a functional ligand for the Nogo-66 receptor. Science (2002.0) 297 1190-3. DOI: 10.1126/science.1073031 69. Gao Z, Ure K, Ables JL, Lagace DC, Nave KA, Goebbels S. Neurod1 is essential for the survival and maturation of adult-born neurons. Nat Neurosci (2009.0) 12 1090-2. DOI: 10.1038/nn.2385 70. Zhang Y, Xia F, Liu X, Yu Z, Xie L, Liu L. JAM3 maintains leukemia-initiating cell self-renewal through LRP5/AKT/beta-catenin/CCND1 signaling. J Clin Invest (2018.0) 128 1737-51. DOI: 10.1172/JCI93198
--- title: Identification of alternative splicing events related to fatty liver formation in duck using full-length transcripts authors: - Yiming Wang - Linfei Song - Mengfei Ning - Jiaxiang Hu - Han Cai - Weitao Song - Daoqing Gong - Long Liu - Jacqueline Smith - Huifang Li - Yinhua Huang journal: BMC Genomics year: 2023 pmcid: PMC9976415 doi: 10.1186/s12864-023-09160-4 license: CC BY 4.0 --- # Identification of alternative splicing events related to fatty liver formation in duck using full-length transcripts ## Abstract ### Background Non-alcoholic fatty liver disease (NAFLD) is one of most common diseases in the world. Recently, alternative splicing (AS) has been reported to play a key role in NAFLD processes in mammals. Ducks can quickly form fatty liver similar to human NAFLD after overfeeding and restore to normal liver in a short time, suggesting that ducks are an excellent model to unravel molecular mechanisms of lipid metabolism for NAFLD. However, how alternative splicing events (ASEs) affect the fatty liver process in ducks is still unclear. ### Results Here we identify 126,277 unique transcripts in liver tissue from an overfed duck (77,237 total transcripts) and its sibling control (69,618 total transcripts). We combined these full-length transcripts with Illumina RNA-seq data from five pairs of overfed ducks and control individuals. Full-length transcript sequencing provided us with structural information of transcripts and Illumina RNA-seq data reveals the expressional profile of each transcript. We found, among these unique transcripts, 30,618 were lncRNAs and 1,744 transcripts including 155 lncRNAs and 1,589 coding transcripts showed significantly differential expression in liver tissues between overfed ducks and control individuals. We also detected 27,317 ASEs and 142 of them showed significant relative abundance changes in ducks under different feeding conditions. Full-length transcript profiles together with Illumina RNA-seq data demonstrated that 10 genes involving in lipid metabolism had ASEs with significantly differential abundance in normally fed (control) and overfed ducks. Among these genes, protein products of five genes (CYP4F22, BTN, GSTA2, ADH5, and DHRS2 genes) were changed by ASEs. ### Conclusions This study presents an example of how to identify ASEs related to important biological processes, such as fatty liver formation, using full-length transcripts alongside Illumina RNA-seq data. Based on these data, we screened out ASEs of lipid-metabolism related genes which might respond to overfeeding. Our future ability to explore the function of genes showing AS differences between overfed ducks and their sibling controls, using genetic manipulations and co-evolutionary studies, will certainly extend our knowledge of genes related to the non-pathogenic fatty liver process. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12864-023-09160-4. ## Background In humans, non-alcoholic fatty liver disease (NAFLD) is one of the most common global diseases with the overall incidence being 46.9 cases per 1000 people per year in a recent survey [1]. NAFLD has clinical-pathological symptoms including isolated steatosis, non-alcoholic steatohepatitis (NASH) and liver fibrosis [2]. The possibility of NAFLD is higher in men and increases with advancing age. NAFLD in obese children and adolescents further develops into more serious diseases [3, 4]. Similarly, intake of energy-rich food induces fatty liver in ducks, which shows the same pathology with human NAFLD [5, 6]. However, ducks can recover from fatty liver quickly and protect their liver against pathological changes such as fibrosis and ultimately cirrhosis that frequently happen in human NAFLD [7–9]. Therefore, ducks provide a good model to unravel the molecular mechanisms underlying lipid metabolism and hepatic steatosis. Alternative splicing is one of the most important events that regulate function of proteins through generating different transcript isoforms. Alternative splicing has been reported in many bioprocesses including human ageing, human cancer and sex selection of birds [10–12]. Alternative splicing also plays an important role in NAFLD and dysregulation of alternative splicing contributes to development of NAFLD [13–15]. For example, DRAK2 inhibits SRPK1-mediated SRSF6 phosphorylation and leads to changes of SRSF6-associated alternative splicing of mitochondrial function-related genes to aggravate NALFD procedure [16]. To our knowledge, studies on fatty liver of ducks have been focused on genes related to lipid metabolism based on RNA-seq short read data or explored nutrition complement affecting fat deposition in duck liver. [ 17–20]. However, the role of alternative splicing is unclear in process of responding to fatty liver of duck. Here we performed full-length transcript sequence using a pair of sibling ducks, which were fed with high fat corn feed and commercial forage, respectively. We annotated transcripts and compared their expression in liver tissues of overfed and control ducks. This effort identified 27,317 ASEs, with 142 of them having significant frequency changes in liver tissues between overfed and control ducks. Moreover, we have identified five lipid metabolism related genes (CYP4F22, BTN, GSTA2, ADH5, and DHRS2 genes) from these 142 events. These observations revealed the probable ASEs regulating the formation and defense process of liver in avoiding pathogenic fatty liver disease. ## PacBio full-length high-coverage liver transcriptomic profile of overfed and control ducks We sequenced liver transcriptomes of a sib-pair ducks using the PacBio Sequel platform. This identified 172,671 and 185,070 full-length transcripts from 6,327,390 subreads of overfed duck and 6,716,303 subreads of control individuals, respectively. We then identified 77,237 and 69,618 unique transcripts from liver tissues of overfed and control ducks respectively, and merged these transcripts into a single data set containing 126,277 unique transcripts (Fig. 1a). Alignment of 77,237 and 69,618 liver transcripts of overfed and control ducks to our duck reference gene set showed that 9,554 and 9,515 genes were expressed, respectively. These numbers of expressed genes covered $82.05\%$ and $82.88\%$ genes detected by Illumina RNA-seq data in overfed and control ducks (unpublished data), suggesting that these two full-length transcriptomes were high coverage and provided a reasonable substrate for the analysis presented in this study. Sequence alignment of 126,277 unique transcripts to our new duck assembly (SKLA1.0, PRJNA792297) and gene reference set showed that $27.01\%$ of them were unique transcripts, while $72.99\%$ have different transcripts (Fig. 1b). The average number of exons was 6.4, the average length of these 126,277 unique transcripts 3,864 bp and 27,410 transcripts had more than 10 exons (Fig. 1c). When compared to our duck reference gene sets, a total of 81,246 transcripts were mapped to 10,888 genes and 45,031 transcripts were novel transcripts (Fig. 1d). Among these 45,031 novel transcripts, 30,512 were annotated as novel transcripts of lncRNAs and 14,519 were annotated as novel transcripts of coding genes. Moreover, 302 transcripts were intra-chromosomal fusion transcripts. These data suggested that our full-length transcriptome was a rich source of biological diversity. Fig. 1Transcript processing workflow and statistics. a Procedure of total transcripts access for ducks. b The ratio of transcripts with multiple isoforms and unique transcripts without other isoforms. c The number of transcript isoforms with different exon number. d Venn diagram showing common and unique transcripts with or without reference genes ## Comparison of transcript expression between overfed and control ducks We compared full-length liver transcriptomic profiles of the above sib-pair ducks. This effort found 56,659 transcripts uniquely presented in overfed ducks, 49,040 transcripts only presented in control ducks and 20,578 transcripts were observed in both overfed and control ducks (Fig. 2a). We further counted expression levels of transcript by TPM (Transcripts per kilobase of exon model divided by million mapped reads) and identified 1,744 transcripts of 1,282 genes showing significantly differential expression (DETs, p-value < 0.01) in liver tissues between overfed and control ducks (Additional file 1: Table S1). Among 1,744 DETs, 982 were upregulated and 762 were downregulated with p-value < 0.01 in overfed ducks when compared to those in control ducks (Fig. 2b). Using thresholds of \|log2FC\|> 1 (FC, Fold Change), we identified 683 being upregulated and 382 being downregulated in overfed ducks when compared to their sibling controls. Gene ontology (GO) analysis indicated that 1,282 genes presenting DETs were enriched in 45 biological functions, with 27 involved in fatty acid metabolic process (GO:0,006,631) with FDR < 0.05 (Fig. 2c). KEGG analysis demonstrated that 9 genes showing DETs were enriched in biosynthesis of unsaturated fatty acids, fatty acid metabolism and fatty acid elongation pathway (p-value < 0.01, Fig. 2d). FADS1 (Fatty Acid Desaturase 1) and FADS2 (Fatty Acid Desaturase 2) were previously reported to reduce lipid accumulation and influence the NAFLD process in mice [21–25]. Interestingly, we found that transcript isoforms of FADS1 (Fatty Acid Desaturase 1) TCONS_00055559 and FADS2 (Fatty Acid Desaturase 2) TCONS_00057710 were significantly upregulated in overfed ducks when compared to those in control ducks. These results reveal detailed information of expression profiles of FADS1 and FADS2 at the transcript-level and identify the main transcripts of FADS1 and FADS2 which might function in the formation of fatty liver in ducks to alleviate liver injury. Moreover, we compared reference transcripts to the above 1,744 differentially expressed transcripts to verify the confidence of detected transcripts. We found 893 of these DETs, including TCONS_00057710, were known transcripts of FADS1, while TCONS_00055559 was a novel transcript of FADS1. Aligning all four reference FADS1 protein sequences to TCONS_00055559 protein sequence, we found that TCONS_00055559 was a new recombination of FADS1 exons. This observation suggested that TCONS_00055559 was a new transcript of FADS1 in ducks (Additional file 2: Fig S1 and Additional file 3: Table S2).Fig. 2Analysis of differentially expressed transcripts. a Venn diagram of unique and common transcripts of overfed and control groups. b Volcano plot for differentially expressed transcripts (FC > 2, p-value < 0.01 in up class, FC < 0.5, p-value < 0.01 in down class). c GO enrichment analysis of genes with significantly differentially expressed transcripts. d KEGG enrichment analysis of genes with significantly differentially expressed transcripts ## Prediction of lncRNA and lncRNA-coding cis-acting pairs For the above 126,277 unique transcripts, 42,642 transcripts were predicted as non-coding sequences and 30,618 were annotated as lncRNA, including 155 DETs (Fig. 3a). We compared characteristics of lncRNA and protein-coding transcript isoforms. We found that lncRNA had lower mean expression level (TPM) than protein-coding transcripts did, in both control and overfed ducks (Fig. 3b). Among 30,618 lncRNA transcripts, 12,861 did not overlap with coding genes and 17,757 did overlap with coding-genes. Detailed transcript structure analysis indicated, among these lncRNA transcripts, a few ($2.13\%$) had more than three exons, a small percentage ($11.64\%$) had two or three exons, and many ($86.23\%$) had only one exon. This was different from the case of protein-coding transcripts, where most ($59.70\%$) had more than three exons, a few ($12.68\%$) had two or three exons and the remainder ($27.62\%$) had only one exon (Fig. 3c). Moreover, we calculated the correlation between 155 lncRNA DET and adjacent protein-coding transcripts with 10 liver tissue RNA-seq transcriptomes. This analysis identified 57 lncRNA-coding cis-acting pairs, including 34 lncRNA and 52 protein-coding transcripts from 32 genes with a Pearson correlation higher than 0.8. Amongst these pairs, four genes (ENPP1 (ectonucleotide pyrophosphatase 1), SERPINA1 (serpin family A member 1), MGAT2 (alpha-1,6-mannosyl-glycoprotein 2-beta-N-acetylglucosaminyltransferase) and SSU72 (RNA polymerase II CTD phosphatase) were reported to have close association with NAFLD process (Fig. 3d). Overexpression of ENPP1 in mice leads to insulin resistance and MGAT2 deficiency reduces lipid absorption and insulin resistance [26, 27]. SERPINA1 was associated with severity of NAFLD and SSU72 influenced NAFLD deterioration [28, 29]. It will therefore be of interest to study whether and how lncRNA interacts with these four genes to regulate the fatty liver process of ducks. Fig. 3Identification and characteristics of lncRNA. a Venn diagram of non-coding transcripts predicted by GeneMark, CPC, and CNCI. b Expression level of transcripts of coding genes and lncRNA in overfed and control groups. c The number of lncRNA with different exon number d Correlated cis-acting pairs of DETs from lncRNA and neighbouring coding genes within 10 kb ## Identification of ASEs using duck full-length transcripts Alternative splicing always requires the spliceosome, which catalyzes splicing reactions [30]. Under the function of the spliceosome, transcripts undergo one or more forms of alternative splicing. We counted ASEs which included skipped exon (SE), mutually exclusive exons (MX), alternative 5’splice site (A5), alternative 3’splice site (A3) and retained intron (RI) (Additional file 2: Fig S2a). Among the above 126,277 unique transcripts, we detected 27,317 ASEs in 5,665 genes, which included $18\%$ RI, $41\%$ SE, $19\%$ A3, $20\%$ A5 and $2\%$ MX (Additional file 2: Fig S2b). We aligned transcripts to our duck reference genome to define ASEs as known and novel classes. This found 26,979 ASEs in known reference genes and 338 ASEs in novel genes. We then compared ASE characteristics in liver transcriptomes of overfed and control ducks. This analysis detected 20,823 ASEs in liver of control and 26,228 in overfed ducks. Amongst these, the liver of control duck had a large proportion of RI events ($43\%$), a small proportion of SE ($20\%$), A3 ($18\%$) and A5 ($17\%$), and a few MX ($2\%$). This is similar to the case in overfed ducks, where there was $41\%$ RI, $20\%$ SE, $18\%$ A3, $19\%$ A5 and $2\%$ MX events (Fig. 4a). We then compared the number of transcript isoforms in liver transcriptomes and found that 9 genes (ACAT1 (acetyl-CoA acetyltransferase 1), ACSL1 (acyl-CoA synthetase long chain family member 1), CPT1A (carnitine palmitoyltransferase 1A), FADS2 (fatty acid desaturase 2), ACSL5 (acyl-CoA synthetase long chain family member 5), PTPLAD1 (3-hydroxyacyl-CoA dehydratase 3), FASN (fatty acid synthase), ACOX1 (acyl-CoA oxidase 1) and ACACA (acetyl-CoA carboxylase alpha)) had different numbers of protein sequences between overfed and control ducks (Fig. 4b). Among them, FASN was a key enzyme in fatty acid biosynthesis, bound to FMN (Flavine Mononucleotide) cofactor via its DUS (Dihydrouridine synthase) domain to produce reactive oxygen species (ROS) in NADPH-dependent oxidation [31]. Interestingly, we found that FASN had a transcript (‘overfed4’ in Fig. 4c and Additional File 3: Table S3) which contained the DUS domain and expressed in livers of five overfed ducks, but was not detected in these of five control ducks (Fig. 4c). These observations together with overfed ducks with fatty liver did not present inflammation and fibrosis (unpublished data) suggesting that ducks might relieve the oxidative damage of fatty liver through AS of these genes. Fig. 4Statistics of ASEs and analysis of ASEs in fatty acid metabolism genes. a The ratio of each ASEs class in overfed and control groups. b Transcript numbers of nine fatty acid related genes in overfed and control ducks. c Protein sequence alignment of FASN transcripts from overfed and control ducks ## Significantly changed ASEs involved in lipid metabolism Transcript isoforms can have similar or antagonistic functions. For example, MAVS (mitochondrial antiviral signaling protein), a regulator of antiviral innate immunity, expresses two transcript isoforms, where the miniMAVS antagonizes the full-length MAVS to induce interferon production [32]. Here we calculated the frequency changes of ASEs with expression profiles of transcript isoforms. We observed that five (CYP4F22, BTN, GSTA2, ADH5, and DHRS2) genes showed significant differential ASEs in liver transcriptomes of overfed ducks compared with control ducks (Additional file 2: Fig S3). CYP4F22 (cytochrome P450 family 4 subfamily F member 22) as a fatty acid ω-hydroxylase involved in lipid metabolism to maintain the skin barrier in mice [33]. Signal peptides carry information for protein secretion and play an important role in human diseases [34–36]. Interestingly, we found overfed ducks preferred to express a transcript isoform with a signal peptide in its N-terminal, while control ducks preferred to express a transcript isoform without N-terminal signal peptide (Fig. 5a and Additional file 2: Fig S4a). We then evaluated the impact on biological function of an amino acid indel in transcript isoforms using PROVEAN software (score < 2.5 indicates a harmful detrimental change) [37]. This suggested that deletion of 146 amino acids at the N-terminal in CYP4F22 transcript isoform TCONS_00116966 (with a score of -289.14), was suggested to be deleterious to CYP4F22 function in ducks. We performed the cross-species alignment of CYP4F22 proteins in six birds and found the N-terminal 24 amino acids showed low conservation, while the remainder of the sequences were relatively conserved in the N-terminal 200 amino acids of CYP4F22 proteins (Fig. 5b and Additional file 3: Table S4). Since the transcript isoform of CYP4F22 which missing the signal peptide are preferred in control ducks, we inferred that ASEs might regulate secretion or localization of CYP4F22 proteins by alternative splicing. We also noticed an A5 ASEs in a BTN gene, which had been reported to regulate milk-lipid secretion in mice [38]. The alternative transcript TCONS_00099256 encoded a 513aa (amino acid) longer protein and was expressed at lower levels, while TCONS_00099264 had a 209aa truncation of the cytoplasmic domain and was expressed more highly in overfed ducks when compared to that in control individuals (Fig. 5c and Additional file 2: Fig S4b). Furthermore, we found that the B30.2 domain was lost in the TCONS_00099264 encoding protein (Additional file 2: Fig S4c). ADH5 (alcohol dehydrogenase 5 class-3, also called ADH-3) has been shown to protect the liver from the damage of nonalcoholic hepatic steatosis in mice [39]. We found an ASE event in the ADH5 gene of ducks leading to a 122aa truncation in the N-terminal of the protein and having deleterious consequences on protein function (-459.358 Provean score) (Fig. 5d and Additional file 2: Fig S4d). Cross-species sequence alignment analysis showed ADH5 proteins were highly conserved (Fig. 5e). The frequency of this ASE event was lower in overfed ducks, thus overfed ducks had more full-length transcripts of ADH5 protein. Fig. 5Analysis of transcripts related to lipid metabolism. a Structural comparison of a truncated transcript (TCONS_00116966) and full-length CDS (TCONS_00117169) for the CYP4F22 gene in ducks (the top line is chromosome coordinates axis). b Multiple protein sequence alignment of CYP4F22 gene in ducks with five other birds. c Structural comparison of an A5 alternative transcript (TCONS_00099256) and full-length CDS for BTN gene in ducks d Structural comparison of a truncated transcript (TCONS_00049539) and full-length CDS for ADH5 gene in ducks. e Multiple protein sequence alignment of ADH5 gene in ducks with five other birds The GSTA2 gene functions in detoxification of electrophilic compounds such as H2O2 or other products of oxidative stress [40]. The relative abundance of an A3 event in GSTA2 was observed to be lower in control individuals (Additional file 2: Fig S3). However, no obvious change was identified during Phobius analysis of the alternative transcripts (TCONS_00036568) of the GSTA2 gene (Additional file 2: Fig S4e and Additional file 2: Fig S5a). The results from Provean show limited expected harm of the short protein sequence change, suggesting a relatively slight effect of the protein truncation (-13.715 score). Therefore, this A3 event has less influence on GSTA2 protein function. The DHRS2 gene localizes in the mitochondria and plays a role in oxidation–reduction processes [41]. An N-terminal truncation by an ASE event was found in the DHRS2 gene in ducks, with the frequency of this ASE event being lower in overfed ducks. Thus overfed ducks express more full-length transcript of the DHRS2 gene (Additional file 2: Fig S3). We showed the selective N-terminal truncation of predicted DHRS2 protein sequences (Additional file 2: Fig S4f). The alignment of DHRS2 protein in duck with another five species shows that the 1-100aa region has relatively higher conservation (Additional file 2: Fig S5c). Provean predicts the potentially harmful result of this truncation with a score of -312.288. Under the overfed condition, ducks reduced levels of the truncated protein and increased expression of the full-length protein of the DHRS2 gene. ## Discussion Ducks provide a good model for the study of fatty liver. Overfeeding of energy-rich food in ducks quickly induced non-pathogenic fatty liver. We performed full-length transcript sequencing of sibling ducks to acquire full-length transcript isoforms and to further detect ASEs. We identified 77,237 transcripts in liver from overfed ducks and 69,618 transcripts in control ducks. The expressional profile and structural information of full-length transcripts were used to evaluate the relative abundance of ASEs in duck livers under different feeding conditions. The enrichment of ASEs in lipid metabolism related genes indicates transcript-level changes under the overfeeding condition. Premature mRNA may produce different mature mRNA by ASEs. Our study provides us a group of differential ASEs between overfed and control ducks. ASEs with significantly differential abundance may reveal the regulation pattern of AS splicing involved in lipid metabolism. Signal peptides (16-30aa) are important in multiple fields such as protein secretion mechanisms and disease diagnosis [42]. CYP4F22 plays an important role in producing acylceramide, which is a key lipid of skin barrier in mice [43]. The loss of signal peptide of CYP4F22 protein indicates that ASEs may cause alternative protein localization to influence lipid metabolism of ducks. The BTN (Butyrophilin) gene family was identified in lactating mammary gland and associated with lipid secretion [38, 44]. B30.2 is a classical conserved domain of BTN genes, possessing multiple functions including resisting virus invasion, regulating T cell activity, and lipid secretion [45–47]. The identified transcript isoform (TCONS_00099264) of the BTN family gene have lost the B30.2 domain in our study. The presence or absence of the B30.2 domain in the identified BTN protein may change the binding ability of the BTN protein. The structural changes in BTN protein products suggest that the lack of conservation of this domain or functional region is also a mode of regulation of lipid metabolism in duck liver. Ducks may have unique mechanism to protect their liver from damage after lipid deposition and ASEs may play a key role in the protection process. Previous studies showed that oxidative stress induced by lipid accumulation was considered as one of the key factors for the exacerbation of NALFD [48, 49]. Glutathione (GSH) is a classical antioxidant substance, which can improve antioxidant defense ability. Increasing the level of glutathione is considered as one of the methods to treat NAFLD. Mice given glycine-based treatment recover from NAFLD, with glutathione accumulating in the process of treatment, indicating that glutathione can protect liver from NAFLD [50]. The ratio of GSH/GSSG (glutathione/oxidized glutathione) is a good marker for oxidative status of cells and high level of GSSG indicates the severe steatosis and oxidative stress in liver [51]. Studies have shown that the synthetic substrates (glycine and serine) of GSH were lower, and GSH level was decreased in NAFLD patients [52]. The concentration of GSSG in human was significantly increased and the GSH/GSSG ratio was lower with NAFLD [53]. The depletion of GSH means serious oxidative stress and probable injury in human liver. However, in waterfowl such as mule ducks, GSH is not depleted during the fatty liver period and the GSH/GSSG ratio is relatively higher compared with human [54]. The different dynamics of GSH compared with human might contribute to the non-pathogenic result of fatty liver in ducks, which was different from that of human NAFLD. Among genes detected with ASEs in duck, ADH5 protect glutathione from consumption of endogenous formaldehyde [55]. We found differential ASEs in the ADH5 gene of ducks, implying that this might regulate the ADH5 protein to resist the damage of oxidation. We also found alternative splicing in GSTA2 (glutathione S-transferase alpha 2) in ducks. GSTA2 functions in oxidative stress and protects cells from oxidation through combination with GSH [56]. These results suggested that alternative splicing may enhance antioxidant ability to avoid damage form fatty liver in ducks. Our studies on ASEs shed further light on regulation of lipid metabolism and GSH metabolism at the transcript level and provide us with evidence of the potential factors leading to differential fatty liver disease processes in humans. ## Conclusions Our study provides the full-length liver transcriptome of Pekin ducks to allow analysis of transcript structure. A total of 126,277 transcripts were generated and 27,317 ASEs identified, enabling us to further explore the events related to non-pathogenic fatty liver. ASEs of numerous genes involved in lipid metabolism were significantly changed by in ducks with fatty liver. Identified candidate genes GSTA2, ADH5 and DHRS2 are involved in oxidation resistance and ASEs might change their protein product to function in fatty liver process. The future challenge will be the functional validation of each transcript isoform involved in fatty liver in poultry and cross species experiments in mice. Taken together, our full-length transcriptome sequencing of overfed and control ducks enlightens us to the role of ASEs in the formation of and defense against fatty liver. ## Animal feeding Five sib-pairs of 11 week-old male ducks were reared at the Jiangsu Institute of Poultry Science, China and divided into two groups. The control group were fed with 180 g/d (gram/day) commercial feed to 14 weeks old. The overfed group were fed with 150 g corn twice a day on the first three days of the 12th week and increasing to 200 g twice a day on the last four days of the 12th week to adapt to the overfeeding condition. After the preparation of overfeeding at the 12th week, the overfed group was fed 150 g corn three time a day until 14 weeks old. After 14 weeks feeding, the sib-pair ducks were euthanized by electronarcosis and cervical dislocation and then liver tissues were collected. ## PacBio full-length transcriptome library preparation and sequencing Library construction was performed according to the PacBio official protocol of Huada Gene Co. Ltd. BGI (Beijing, China). Total RNA was extracted from liver tissue of a sib-pair ducks from control and overfeeding groups using Trizol reagent (ThermoFisher Scientific). After quality testing, RNA was reverse transcribed into cDNA by SMARTer™ PCR cDNA Synthesis Kit (Clontech, CA, USA). Full-length transcriptomic libraries were constructed to capture complete structure information. SMART primers were incorporated and PCR was performed for single stranded cDNA and double stranded cDNA in turn. Bluepippin (Sage Science, MA, USA) was used for cDNA library length classification and PCR amplification was performed again in a different cDNA library. Sequencing adaptors were linked to cDNA, and linear DNA without adaptor was removed. Finally, after quality tests using an Agilent 2100 (Agilent, CA, USA) and Qubit HS (Invitrogen, CA, USA), sequencing was carried out on the PacBio sequel platform (PacBio, CA, USA). ## Sequence processing Full-length transcriptome raw sequencing data was strictly processed in accordance with the PacBio official smrtlink_5.1.0 work flow (https://www.pacb.com/support/ software-downloads). With this pipeline, CCS (Circular consensus sequencing) reads were generated and classified into full-length non-chimeric (FLNC) and non-full-length reads. FLNC reads were then passed through ICE (Iterative Clustering for Error Correction) and input into ICE Partial and Quiver, together with non-full-length reads to acquire unpolished reads. Reads were then polished using RNA-seq data with LoRDEC software (version 0.6) with parameters -k 19, -s 3 [57]. The polished reads were mapped to our recently developed high quality reference genome, SKLA1.0 (NCBI BioProject accession number PRJNA792297) by minimap2 with parameter -ax splice -uf [58]. The redundant results of minimap2 were removed by cupcake (version 28.0.0) with parameter -c 0.85 -i 0.9 –dun-merge-5-shorter (https://github.com/Magdoll/cDNA_Cupcake). Transcripts from ducks under different feeding conditions were merged non-redundantly for subsequent analysis. Sqanti3 was used to evaluate and annotate the long-read transcriptome [59]. Sqanti3 transcript evaluation was performed using default parameters. Associated reference genes of each transcript and different types of splice junction were identified and classified by Sqanti3. Fusion transcript were also identified by Sqanti3 and the distance between transcript members in one fusion must be greater than 10,000 bp. ## LncRNA prediction CNCI (Coding-Non-Coding Index), CPC (Coding Potential Calculator) and GeneMark were used for transcript coding potential identification [60–62]. The non-coding transcripts identified by all three algorithms were filtered using thresholds of ORF < 100aa and transcript length > 200nt (nucleotide). ORF sequences were acquired from transdecoder (ver 5.5.0) (https://github.com/TransDecoder/Trans Decoder). Pfam domain and super family prediction was implemented and transcripts found by Pfam database were eliminated. A region 10,000 bp upstream and downstream of lncRNA in the DET set was regarded as the maximum cis-acting screening window, and coding genes within this range were inferred as cis-acting target genes. Pearson correlation was performed to test reliability of cis-acting pairs and the pairs within one gene range were excluded. ## Differential transcript analysis The transcript-level expression was calculated by the Kallisto software (version 0.48.0) with default parameters based on short reads and full-length transcript sequences [63]. Kallisto uses pseudoalignment framework and can quantify the expression of transcripts without additional alignment or reference genome. Transcript expression level in TPM (transcript per million) was used for significantly differentially expressed transcript screening through the sleuth R package with a threshold of p-value < 0.01 [64]. Differentially expressed transcripts were annotated by eggNOG webtools and GO enrichment analysis was performed by DAVID with FDR < 0.05 [65, 66]. KEGG enrichment analysis was performed by KOBAS online tools with p-value < 0.01 [67]. The data used for KEGG enrichment originates from KEGG pathway database (https://www.kegg.jp/kegg/pathway.html) [68]. ## Detection and analysis of alternative splicing events Alternative splicing (AS) event analysis was implemented by suppa2 software (version 2.3) with parameters: -e SE SS MX RI -f ioe [69]. The combination of identified ASEs and transcript-level expression were used to screen out significant ASEs by suppa2 with p-value < 0.05. The Phobius software (https://phobius.sbc.su.se/) was used to predict transmembrane topology and signal peptides. The NCBI CDD tool was used to predict conserved domains. The protein sequences of all transcripts of fatty acid related genes with ASEs were acquired from ORFfinder and the redundant protein sequences were removed. The predicted deleteriousness of protein sequences changes was evaluated by Provean [37]. Protein sequences were downloaded from the NCBI website and multiple sequence alignment was performed using the Prank tool (version 170,703). ## Supplementary Information Additional file 1.Additional file 2.Additional file 3. ## References 1. Riazi K, Azhari H, Charette JH, Underwood FE, King JA, Afshar EE. The prevalence and incidence of NAFLD worldwide: a systematic review and meta-analysis. Lancet Gastroenterol Hepatol (2022) 7 851-861. DOI: 10.1016/S2468-1253(22)00165-0 2. Yki-Järvinen H. Non-alcoholic fatty liver disease as a cause and a consequence of metabolic syndrome. Lancet Diabetes Endocrinol (2014) 2 901-910. DOI: 10.1016/S2213-8587(14)70032-4 3. Ayonrinde OT, Olynyk JK, Beilin LJ, Mori TA, Pennell CE, de Klerk N. Gender-specific differences in adipose distribution and adipocytokines influence adolescent nonalcoholic fatty liver disease. Hepatology (2011) 53 800-809. DOI: 10.1002/hep.24097 4. Berentzen TL, Gamborg M, Holst C, Sorensen TIA, Baker JL. Body mass index in childhood and adult risk of primary liver cancer. J Hepatol (2014) 60 325-330. DOI: 10.1016/j.jhep.2013.09.015 5. Alsanea S, Gao M, Liu D. Phloretin prevents high-fat diet-induced obesity and improves metabolic homeostasis. AAPS J (2017) 19 797-805. DOI: 10.1208/s12248-017-0053-0 6. Hermier D, Rousselotpailley D, Peresson R, Sellier N. Influence of orotic acid and estrogen on hepatic lipid storage and secretion in the goose susceptible to liver steatosis. Biochim Biophys Acta (1994) 1211 97-106. DOI: 10.1016/0005-2760(94)90143-0 7. Wei R, Han C, Deng D, Ye F, Gan X, Liu H. Research progress into the physiological changes in metabolic pathways in waterfowl with hepatic steatosis. Br Poult Sci (2021) 62 118-124. DOI: 10.1080/00071668.2020.1812527 8. Davail S, Guy G, Andre JM, Hermier D, Hoo-Paris R. Metabolism in two breeds of geese with moderate or large overfeeding induced liver-steatosis. Comp Biochem Physiol A Mol Integr Physiol (2000) 126 91-99. DOI: 10.1016/S1095-6433(00)00190-2 9. Hermier D, Guy G, Guillaumin S, Davail S, Andre JM, Hoo-Paris R. Differential channelling of liver lipids in relation to susceptibility to hepatic steatosis in two species of ducks. Comp Biochem Physiol B Biochem Mol Biol (2003) 135 663-675. DOI: 10.1016/S1096-4959(03)00146-5 10. Rogers TF, Palmer DH, Wright AE. Sex-specific selection drives the evolution of alternative splicing in birds. Mol Biol Evol (2021) 38 519-530. DOI: 10.1093/molbev/msaa242 11. Zhang Y, Qian J, Gu C, Yang Y. Alternative splicing and cancer: a systematic review. Signal Transduct Target Ther (2021) 6 807-820 12. Bhadra M, Howell P, Dutta S, Heintz C, Mair WB. Alternative splicing in aging and longevity. Hum Genet (2020) 139 357-369. DOI: 10.1007/s00439-019-02094-6 13. del Rio-Moreno M, Alors-Perez E, Gonzalez-Rubio S, Ferrin G, Reyes O, Rodriguez-Peralvarez M. Dysregulation of the splicing machinery is associated to the development of nonalcoholic fatty liver disease. J Clin Endocrinol Metab (2019) 104 3389-3402. DOI: 10.1210/jc.2019-00021 14. Wang H, Lekbaby B, Fares N, Augustin J, Attout T, Schnuriger A. Alteration of splicing factors' expression during liver disease progression: impact on hepatocellular carcinoma outcome. Hepatol Int (2019) 13 454-467. DOI: 10.1007/s12072-019-09950-7 15. Baralle M, Baralle FE. Alternative splicing and liver disease. Ann Hepatol (2021) 26 100534-100541. DOI: 10.1016/j.aohep.2021.100534 16. Li Y, Xu J, Lu Y, Bian H, Yang L, Wu H. DRAK2 aggravates nonalcoholic fatty liver disease progression through SRSF6-associated RNA alternative splicing. Cell Metab (2021) 33 2004-2020. DOI: 10.1016/j.cmet.2021.09.008 17. Zhang Y, Chang Y, Yang T, Wen M, Zhang Z, Liu G. The hepatoprotective effects of zinc glycine on liver injury in meat duck through alleviating hepatic lipid deposition and inflammation. Biol Trace Elem Res (2020) 195 569-578. DOI: 10.1007/s12011-019-01860-x 18. Herault F, Duby C, Baeza E, Diot C. Adipogenic genes expression in relation to hepatic steatosis in the liver of two duck species. Animal (2018) 12 2571-2577. DOI: 10.1017/S1751731118000897 19. Jin S, Yang L, Fan X, Wu M, Xu Y, Chen X. Effect of divergence in residual feed intake on expression of lipid metabolism-related genes in the liver of meat-type ducks. J Anim Sci (2019) 97 3947-3957. DOI: 10.1093/jas/skz241 20. Jiang Y, Xie M, Fan W, Xue J, Zhou Z, Tang J. Transcriptome analysis reveals differential expression of genes regulating hepatic triglyceride metabolism in pekin ducks during dietary threonine deficiency. Front Genet (2019) 10 710-724. DOI: 10.3389/fgene.2019.00710 21. Zheng J, Zhang J, Guo Y, Cui H, Lin A, Hu B. Improvement on metabolic syndrome in high fat diet-induced obese mice through modulation of gut microbiota by sangguayin decoction. J Ethnopharmacol (2020) 246 112225-112236. DOI: 10.1016/j.jep.2019.112225 22. Liu Y-j, Li H, Tian Y, Han J, Wang X-y, Li X-y. PCTR1 ameliorates lipopolysaccharide-induced acute inflammation and multiple organ damage via regulation of linoleic acid metabolism by promoting FADS1/FASDS2/ELOV2 expression and reducing PLA2 expression. Lab Invest. (2020) 100 904-915. DOI: 10.1038/s41374-020-0412-9 23. Hayashi Y, Shimamura A, Ishikawa T, Fujiwara Y, Ichi I. FADS2 inhibition in essential fatty acid deficiency induces hepatic lipid accumulation via impairment of very low-density lipoprotein (VLDL) secretion. Biochem Biophys Res Commun (2018) 496 549-555. DOI: 10.1016/j.bbrc.2018.01.064 24. Athinarayanan S, Fan Y-Y, Wang X, Callaway E, Cai D, Chalasani N. Fatty acid desaturase 1 influences hepatic lipid homeostasis by modulating the PPAR alpha-FGF21 Axis. Hepatol Commun (2021) 5 461-477. DOI: 10.1002/hep4.1629 25. Gromovsky AD, Schugar RC, Brown AL, Helsley RN, Burrows AC, Ferguson D. Δ-5 fatty acid desaturase FADS1 impacts metabolic disease by balancing proinflammatory and proresolving lipid mediators. Arterioscler Thromb Vasc Biol (2018) 38 218-231. DOI: 10.1161/ATVBAHA.117.309660 26. Pan W, Ciociola E, Saraf M, Tumurbaatar B, Tuvdendorj D, Prasad S. Metabolic consequences of ENPP1 overexpression in adipose tissue. Am J Physiol Endocrinol Metab (2011) 301 E901-E911. DOI: 10.1152/ajpendo.00087.2011 27. Tsuchida T, Fukuda S, Aoyama H, Taniuchi N, Ishihara T, Ohashi N. MGAT2 deficiency ameliorates high-fat diet-induced obesity and insulin resistance by inhibiting intestinal fat absorption in mice. Lipids Health Dis (2012) 11 75-84. DOI: 10.1186/1476-511X-11-75 28. Kim H-S, Yoon J-S, Jeon Y, Park E-J, Lee J-K, Chen S. Ssu72-HNF4 alpha signaling axis classify the transition from steatohepatitis to hepatocellular carcinoma. Cell Death Differ (2022) 29 600-613. DOI: 10.1038/s41418-021-00877-x 29. Semmler G, Balcar L, Oberkofler H, Zandanell S, Strasser M, Niederseer D. PNPLA3 and SERPINA1 variants are associated with severity of fatty liver disease at first referral to a tertiary center. J Pers Med. (2021) 11 165. DOI: 10.3390/jpm11030165 30. Wahl MC, Will CL, Luehrmann R. The spliceosome: design principles of a dynamic RNP machine. Cell (2009) 136 701-718. DOI: 10.1016/j.cell.2009.02.009 31. Chaiyen P, Fraaije MW, Mattevi A. The enigmatic reaction of flavins with oxygen. Trends Biochem Sci (2012) 37 373-380. DOI: 10.1016/j.tibs.2012.06.005 32. Brubaker SW, Gauthier AE, Mills EW, Ingolia NT, Kagan JC. A bicistronic MAVS transcript highlights a class of truncated variants in antiviral immunity. Cell (2014) 156 800-811. DOI: 10.1016/j.cell.2014.01.021 33. Miyamoto M, Itoh N, Sawai M, Sassa T, Kihara A. Severe skin permeability barrier dysfunction in knockout mice deficient in a fatty acid omega-hydroxylase crucial to acylceramide production. J Invest Dermatol (2020) 140 319-326. DOI: 10.1016/j.jid.2019.07.689 34. Datta R, Waheed A, Shah GN, Sly WS. Signal sequence mutation in autosomal dominant form of hypoparathyroidism induces apoptosis that is corrected by a chemical chaperone. Proc Natl Acad Sci U S A (2007) 104 19989-19994. DOI: 10.1073/pnas.0708725104 35. Bonfanti R, Colombo C, Nocerino V, Massa O, Lampasona V, Iafusco D. Insulin gene mutations as cause of diabetes in children negative for five type 1 diabetes autoantibodies. Diabetes Care (2009) 32 123-125. DOI: 10.2337/dc08-0783 36. Vezzoli V, Duminuco P, Vottero A, Kleinau G, Schuelein R, Minari R. A newvariant in signal peptide of the human luteinizing hormone receptor (LHCGR) affects receptor biogenesis causing leydig cell hypoplasia. Hum Mol Genet (2015) 24 6003-6012. DOI: 10.1093/hmg/ddv313 37. Choi Y, Chan AP. PROVEAN web server: a tool to predict the functional effect of amino acid substitutions and indels. Bioinformatics (2015) 31 2745-2747. DOI: 10.1093/bioinformatics/btv195 38. Ogg SL, Weldon AK, Dobbie L, Smith AJH, Mather IH. Expression of butyrophilin (Btn1a1) in lactating mammary gland is essential for the regulated secretion of milk-lipid droplets. Proc Natl Acad Sci U S A (2004) 101 10084-10089. DOI: 10.1073/pnas.0402930101 39. Goto M, Kitamura H, Alam MM, Ota N, Haseba T, Akimoto T. Alcohol dehydrogenase 3 contributes to the protection of liver from nonalcoholic steatohepatitis. Genes Cells (2015) 20 464-480. DOI: 10.1111/gtc.12237 40. Hayes JD, Flanagan JU, Jowsey IR. Glutathione transferases. Annu Rev Pharmacol Toxicol (2005) 45 51-88. DOI: 10.1146/annurev.pharmtox.45.120403.095857 41. Yuan X, Sun Y, Cheng Q, Hu K, Ye J, Zhao Y. Proteomic analysis to identify differentially expressed proteins between subjects with metabolic healthy obesity and non-alcoholic fatty liver disease. J Proteomics (2020) 221 103683-103692. DOI: 10.1016/j.jprot.2020.103683 42. Owji H, Nezafat N, Negandaripour M, Hajiebrahimi A, Ghasemi Y. A comprehensive review of signal peptides: structure, roles, and applications. Eur J Cell Biol (2018) 97 422-441. DOI: 10.1016/j.ejcb.2018.06.003 43. Ohno Y, Nakamichi S, Ohkuni A, Kamiyama N, Naoe A, Tsujimura H. Essential role of the cytochrome P450 CYP4F22 in the production of acylceramide, the key lipid for skin permeability barrier formation. Proc Natl Acad Sci U S A (2015) 112 7707-7712. DOI: 10.1073/pnas.1503491112 44. Franke WW, Heid HW, Grund C, Winter S, Freudenstein C, Schmid E. Antibodies to the major insoluble milk fat globule membrane-associated protein: specific location in apical regions of lactating epithelial cells. J Cell Biol (1981) 89 485-494. DOI: 10.1083/jcb.89.3.485 45. Jeong J, Rao AU, Xu J, Ogg SL, Hathout Y, Fenselau C. The PRY/SPRY/B30.2 domain of butyrophilin 1A1 (BTN1A1) binds to xanthine oxidoreductase: implications for the function of BTN1A1 in the mammary gland and other tissues. J Biol Chem. (2009) 284 22444-22456. DOI: 10.1074/jbc.M109.020446 46. Morger D, Zosel F, Buhlmann M, Zuger S, Mittelviefhaus M, Schuler B. The three-fold axis of the HIV-1 capsid lattice is the species-specific binding interface for TRIM5 alpha. J Virol (2018) 92 e01541-e1617. DOI: 10.1128/JVI.01541-17 47. Lu Y, Zhou T, Xu C, Wang R, Feng D, Li J. Occludin is a target of Src kinase and promotes lipid secretion by binding to BTN1a1 and XOR. PLoS Biol (2022) 20 e3001518-e3001540. DOI: 10.1371/journal.pbio.3001518 48. Carter-Kent C, Zein NN, Feldstein AE. Cytokines in the pathogenesis of fatty liver and disease progression to steatohepatitis: implications for treatment. Am J Gastroenterol (2008) 103 1036-1042. DOI: 10.1111/j.1572-0241.2007.01709.x 49. Reddy JK, Rao MS. Lipid metabolism and liver inflammation. II. fatty liver disease and fatty acid oxidation. Am J Physiol Gastrointest Liver Physiol. (2006) 290 852-858. DOI: 10.1152/ajpgi.00521.2005 50. Rom O, Liu Y, Liu Z, Zhao Y, Wu J, Ghrayeb A. Glycine-based treatment ameliorates NAFLD by modulating fatty acid oxidation, glutathione synthesis, and the gut microbiome. Sci Transl Med. (2020) 12 eaaz2841-eaaz2855. DOI: 10.1126/scitranslmed.aaz2841 51. Chen Y, Dong H, Thompson DC, Shertzer HG, Nebert DW, Vasiliou V. Glutathione defense mechanism in liver injury: Insights from animal models. Food Chem Toxicol (2013) 60 38-44. DOI: 10.1016/j.fct.2013.07.008 52. Mardinoglu A, Bjornson E, Zhang C, Klevstig M, Soderlund S, Stahlman M. Personal model-assisted identification of NAD(+) and glutathione metabolism as intervention target in NAFLD. Mol Syst Biol (2017) 13 916-932. DOI: 10.15252/msb.20167422 53. Hardwick RN, Fisher CD, Canet MJ, Lake AD, Cherrington NJ. Diversity in antioxidant response enzymes in progressive stages of human nonalcoholic fatty liver disease. Drug Metab Dispos (2010) 38 2293-2301. DOI: 10.1124/dmd.110.035006 54. Lo B, Marty-Gasset N, Manse H, Bannelier C, Bravo C, Domitile R. Cellular markers of mule duck livers after force-feeding. Poult Sci (2020) 99 3567-3573. DOI: 10.1016/j.psj.2020.03.048 55. Burgos-Barragan G, Wit N, Meiser J, Dingler FA, Pietzke M, Mulderrig L. Mammals divert endogenous genotoxic formaldehyde into one-carbon metabolism. Nature (2017) 548 549-554. DOI: 10.1038/nature23481 56. Ng KT-P, Yeung OW-H, Lam YF, Liu J, Liu H, Pang L. Glutathione S-transferase A2 promotes hepatocellular carcinoma recurrence after liver transplantation through modulating reactive oxygen species metabolism. Cell Death Discov. (2021) 7 188-201. DOI: 10.1038/s41420-021-00569-y 57. Salmela L, Rivals E. LoRDEC: accurate and efficient long read error correction. Bioinformatics (2014) 30 3506-3514. DOI: 10.1093/bioinformatics/btu538 58. Li H. New strategies to improve minimap2 alignment accuracy. Bioinformatics (2021) 37 4572-4574. DOI: 10.1093/bioinformatics/btab705 59. Tardaguila M, de la Fuente L, Marti C, Pereira C, Pardo-Palacios FJ, del Risco H. SQANTI: extensive characterization of long-read transcript sequences for quality control in full-length transcriptome identification and quantification (vol 28, pg 396, 2018). Genome Res (2018) 28 1096-1096. DOI: 10.1101/gr.239137.118 60. Bruna T, Lomsadze A, Borodovsky M. GeneMark-EP+: eukaryotic gene prediction with self-training in the space of genes and proteins. NAR Genom Bioinform. (2020) 2 lqaa026-lqaa026. DOI: 10.1093/nargab/lqaa026 61. Kong L, Zhang Y, Ye Z-Q, Liu X-Q, Zhao S-Q, Wei L. CPC: assess the protein-coding potential of transcripts using sequence features and support vector machine. Nucleic Acids Res (2007) 35 W345-W349. DOI: 10.1093/nar/gkm391 62. Sun L, Luo H, Bu D, Zhao G, Yu K, Zhang C. Utilizing sequence intrinsic composition to classify protein-coding and long non-coding transcripts. Nucleic Acids Res (2013) 41 e166-e173. DOI: 10.1093/nar/gkt646 63. Bray NL, Pimentel H, Melsted P, Pachter L. Near-optimal probabilistic RNA-seq quantification. Nat Biotechnol (2016) 34 525-527. DOI: 10.1038/nbt.3519 64. Pimentel H, Bray NL, Puente S, Melsted P, Pachter L. Differential analysis of RNA-seq incorporating quantification uncertainty. Nat Methods (2017) 14 687-690. DOI: 10.1038/nmeth.4324 65. Cantalapiedra CP, Hernandez-Plaza A, Letunic I, Bork P, Huerta-Cepas J. eggNOG-mapper v2: functional annotation, orthology assignments, and domain prediction at the metagenomic scale. Mol Biol Evol (2021) 38 5825-5829. DOI: 10.1093/molbev/msab293 66. Sherman BT, Hao M, Qiu J, Jiao X, Baseler MW, Lane HC. DAVID: a web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res (2022) 50 W216-W221. DOI: 10.1093/nar/gkac194 67. Bu D, Luo H, Huo P, Wang Z, Zhang S, He Z. KOBAS-i: intelligent prioritization and exploratory visualization of biological functions for gene enrichment analysis. Nucleic Acids Res (2021) 49 W317-W325. DOI: 10.1093/nar/gkab447 68. Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res (2023) 51 D587-D592. DOI: 10.1093/nar/gkac963 69. Trincado JL, Entizne JC, Hysenaj G, Singh B, Skalic M, Elliott DJ. SUPPA2: fast, accurate, and uncertainty-aware differential splicing analysis across multiple conditions. Genome Biol (2018) 19 40-50. DOI: 10.1186/s13059-018-1417-1
--- title: Investigation of bedaquiline resistance and genetic mutations in multi-drug resistant Mycobacterium tuberculosis clinical isolates in Chongqing, China authors: - Yan Hu - Jun Fan - Damin Zhu - Wenguo Liu - Feina Li - Tongxin Li - Huiwen Zheng journal: Annals of Clinical Microbiology and Antimicrobials year: 2023 pmcid: PMC9976417 doi: 10.1186/s12941-023-00568-0 license: CC BY 4.0 --- # Investigation of bedaquiline resistance and genetic mutations in multi-drug resistant Mycobacterium tuberculosis clinical isolates in Chongqing, China ## Abstract ### Background To investigate the prevalence and molecular characterization of bedaquiline resistance among MDR-TB isolates collected from Chongqing, China. ### Methods A total of 205 MDR-TB isolates were collected from Chongqing Tuberculosis Control Institute between March 2019 and June 2020. The MICs of BDQ were determined by microplate alamarblue assay. All strains were genotyped by melting curve spoligotyping, and were subjected to WGS. ### Results Among the 205 MDR isolates, the resistance rate of BDQ was $4.4\%$ ($\frac{9}{205}$). The 55 ($26.8\%$) were from male patients and 50 ($24.4\%$) were new cases. Furthermore, 81 ($39.5\%$) of these patients exhibited lung cavitation, 13 ($6.3\%$) patients afflicted with diabetes mellitus, and 170 ($82.9\%$) isolates belonged to Beijing family. However, the distribution of BDQ resistant isolates showed no significant difference among these characteristics. Of the 86 OFX resistant isolates, 8 isolates were XDR ($9.3\%$, $\frac{8}{86}$). Six BDQ resistant isolates ($66.7\%$, $\frac{6}{9}$) and two BDQ susceptible isolates ($1.0\%$, $\frac{2}{196}$) carried mutations in Rv0678. A total of 4 mutations types were identified in BDQ resistant isolates, including mutation in A152G ($50\%$, $\frac{3}{6}$), T56C ($16.7\%$, $\frac{1}{6}$), GA492 insertion ($16.7\%$, $\frac{1}{6}$), and A274 insertion ($16.7\%$, $\frac{1}{6}$). BDQ showed excellent activity against MDR-TB in Chongqing. ### Conclusions BDQ showed excellent activity against MDR-TB in Chongqing. The resistance rate of BDQ was not related to demographic and clinical characteristics. Mutations in Rv0678 gene were the major mechanism to BDQ resistance, with A152G as the most common mutation type. WGS has a good popularize value and application prospect in the rapid detection of BDQ resistance. ## Introduction Drug-resistant tuberculosis, especially multidrug-resistant tuberculosis (MDR-TB), remains a major threat to global TB control and prevention strategy. In 2020, an estimate of approximate 0.5 million rifampicin-resistant (RR-)/MDR-TB cases occurred globally, of which $78\%$ were MDR-TB [1]. The treatment of MDR-TB is challenging due to the lack of effective drugs, and the overall rate of treatment success is currently $57\%$, imposing a burden on health care resources [1]. Therefore, new and effective anti-TB drugs are urgently needed to improve the chemotherapy of MDR-TB [2]. Bedaquiline (BDQ), a novel oral diarylquinoline drug, had excellent efficacy against both drug susceptible and drug resistant MTB [3] and was recommended by WHO for the treatment of MDR [4]. However, BDQ resistance was also emerged with the introduction to the treatment regimens, and several mechanisms of BDQ resistance had been identified. Mutations in atpE gene, encoding subunit C of the ATP synthase, can prevent BDQ from binding to the C subunit, thus resulting in BDQ resistance [3]. Mutations in Rv0678 gene, coding for the repressor of MmpS5-MmpL5 efflux system, were associated with resistance to BDQ [5, 6]. Besides, mutations in gene encoding the uncharacterized transporter Rv1979c and the cytoplasmic peptidase PepQ were also reported to confer BDQ resistance [7–9]. Though BDQ has not been widely used in China, the primary drug resistance of BDQ has emerged [6]. Chongqing, the only municipal city in Southwest China with a high incidence of tuberculosis, will promote the use of BDQ in the treatment of MDR-TB. However, with little information about the prevalence of BDQ resistance in Chongqing, it is meaningful to investigate the prevalence and molecular characterization of BDQ resistance by whole genome sequencing (WGS) among MDR-TB isolates, which will improve the diagnosis and treatment of MDR patients. ## Bacterial strains A total of 205 MDR-TB isolates were collected from Chongqing Tuberculosis Control Institute between March 2019 and June 2020. All isolates were from patients with symptoms suggestive of active pulmonary TB, and the demographic and clinical characteristics were obtained. All isolates were subcultured on the Löwenstein–Jensen (L–J) medium for 4 weeks at 37 ℃. ## Conventional drug susceptibility testing Drug susceptibility was determined using the $1\%$ proportion method on L–J medium according to the guidelines of the WHO [10], with rifampin (RIF), 40 μg/ml; isoniazid (INH), 0.2 μg/ml; streptomycin (SM), 10 μg/ml; ethambutol (EMB), 2 μg/ml; capreomycin (CM), 40 μg/ml; kanamycin (KM), 30 μg/ml; ofloxacin (OFX), 2 μg/ml; amikacin (AM). The MDR-TB was defined as resistance to at least INH and RIF. Extensively drug-resistant tuberculosis (XDR-TB) isolates were defined as MDR-TB isolates with additional resistance to both OFX and KM [11]. ## Minimum inhibitory concentrations For MDR-TB identified by conventional drug susceptibility testing, the MICs of BDQ were determined using microplate alamarblue assay [12]. The breakpoint concentrations were defined as 0.25 μg/ml for BDQ according to the European Committee on Antimicrobial Susceptibility Testing (EUCAST) guidelines [13, 14]. Mycobacterium tuberculosis H37Rv (ATCC 27249) was used as the control strain. The MIC value was defined as the lowest concentration of antibiotic that inhibited visible growth of mycobacteria. MIC50 and MIC90 was defined as the concentration required to inhibit the growth of $50\%$ and $90\%$ of the strains, respectively. ## MeltPro assay Genomic DNA from MDR-TB isolates was extracted using the cetyltrimethylammonium bromide (CTAB) method. All strains were genotyped by melting curve spoligotyping performed in the SLAN-96S system (Hongshi, Shanghai, China) as previously described [15]. The results were automatically exported by the SLAN software (Zeesan, Xiamen, China), followed by comparing to the SITVIT database to identify the genotype. ## Whole genome sequencing The qualified DNA samples were sent to the Annoroad Gene Technology (Beijing, China) for whole genome sequencing (WGS) service based on Illumina Hiseq2500 sequencing platform. The sequencing reads were aligned to the H37Rv reference genome (NC_000962). ## Statistical analysis The person chi-square test or Fisher exact test was used to compare proportions or resistant rates. A $P \leq 0.05$ was considered statistically significant. All the statistical analyses were performed in the SPSS 20.0 (IBM Corp., Armonk, NY). ## BDQ MIC to MDR The distribution of MDR isolates at the MIC of BDQ was shown in Fig. 1. Among the 205 MDR isolates, the number of bacteria showing MIC > 0.25 μg/ml as determined by BDQ resistance was $4.4\%$ ($\frac{9}{205}$). The MIC50 and MIC90 values were 0.031 μg/ml and 0.125 μg/ml, respectively. Fig. 1Distribution of minimum inhibitory concentration (MIC, μg/ml) of BDQ for MDR ($$n = 205$$) ## Clinical data analysis of MDR isolates Demographic and clinical characteristics of MDR isolates patients were summarized in Table 1. For the 205 MDR patients, 55 ($26.8\%$) were from female patients, and there were 50 ($24.4\%$) new cases and 155 ($75.6\%$) re-treated cases. The resistance rate of BDQ ($4.4\%$, $\frac{9}{205}$) was lower than that of commonly used first- and second-line drugs with SM ($72.2\%$, $\frac{148}{205}$), EMB ($37.6\%$, $\frac{77}{205}$), OFX ($42.0\%$, $\frac{86}{205}$) and KM ($14.6\%$, $\frac{30}{205}$). Of the 9 BDQ resistant isolates, the proportion of ancient Beijing strains ($88.9\%$, $\frac{8}{9}$) was significantly higher than that of modern Beijing strains ($11.1\%$, $\frac{1}{9}$) ($P \leq 0.01$), and the number of OFX resistant isolates was significantly higher than that of OFX sensitive isolates ($P \leq 0.01$).Table 1Differences of characteristics between BDQR and BDQS MDR strainsCharacteristicsNo. (%) of isolates ($$n = 205$$)No. (%) of isolatesBDQR($$n = 9$$)BDQS($$n = 196$$)ORP($95\%$CI)Sex Female55(26.8)3(33.3)52(26.5)Ref. Male150(73.2)6(66.7)144(73.5)0.72(0.17–2.99)0.95Age (years) ≤ 4069(33.7)4(44.4)65(33.2)Ref. 41–5996(46.8)3(33.3)93(47.4)0.52(0.11–2.42)0.65 ≥ 6040(19.5)2(22.2)38(19.4)0.86(0.15–4.89)1.00Lineage Lineage 437(18.0)0(0.0)37(18.9)Ref. Lineage 2168(82.0)9(100.0)159(8.1)0.95(0.91–0.98)0.32Genotype Modern Beijing98(47.8)1(11.1)97(49.5)Ref. Ancient Beijing72(35.1)8(88.9)64(32.7)12.13(1.48–99.28) < 0.01 Non-Beijing35(17.1)0(0.0)35(17.8)1.01(0.99–1.03)1.00Treatment History New case50(24.4)3(33.3)47(24.0)Ref. Re-treated155(75.6)6(66.7)149(76.0)0.63(0.15–2.62)0.81Lung Cavitation No124(60.5)6(66.7)118(60.2)Ref Yes81(39.5)3(33.3)78(39.8)0.76(0.18–3.11)0.97Diabetes Mellitus No192(93.7)7(77.8)185(94.4)Ref. Yes13(6.3)2(22.2)11(5.6)4.81(0.89–25.90)0.10Previous exposure to None16(7.8)1(11.1)15(7.7)Ref. FL drugs124(60.5)5(55.6)119(60.7)0.63(0.07–5.76)0.52 FL and SL drugs65(31.7)3(33.3)62(31.6)0.73(0.07–7.48)1.00Resistance to SM148(72.2)7(77.8)141(71.9)4.81(0.89–25.91)0.10 EMB77(37.6)5(55.6)72(36.7)2.15(0.56–8.28)0.43 OFX86(42.0)8(88.9)78(39.8)12.10(1.48–98.68)< 0.01 KM30(14.6)1(11.1)29(14.8)0.72(0.09–5.97)1.00 ## MDR against BDQ in different resistance pattern The MIC of BDQ resistant isolates against SM, EMB, KM and OFX was shown in Table 2. The 31 isolates sensitive to SM, EMB, KM and OFX were all susceptibility to BDQ. As the number of drug resistance increases, the drug resistance rate of BDQ increased from 0 to $14.4\%$. Of the 86 OFX resistant isolates, 8 isolates were XDR with the resistance rate of $9.3\%$ ($\frac{8}{86}$). And the resistance rate of BDQ in OFX resistant isolates ($9.3\%$) was higher than that in SM resistant isolates ($4.7\%$), EMB resistant isolates ($6.5\%$), and KM resistant isolates ($3.3\%$). The resistance rate of BDQ in isolates resistant to any first and second line drug ($8.9\%$) was higher than that in isolates resistant to first line drugs ($7.7\%$) and second line drugs ($1.2\%$), respectively. Table 2MIC distribution of BDQ resistant isolates against SM, EMB, KM and OFXDrug resistance profileNo. of strainsNo. of strains with different MIC (μg/ml)No. (%) of BDQ resistant strains≤ 0.0080.0160.0310.0630.1250.250.51248All isolates2058287750267630009 (4.4)Fully susceptible isolates312613712000000[0]Resistant to one drug5645221572100001(1.8)Resistant to two drugs761113017113300003 (3.9)Resistant to three drugs351411951220004(11.4)Resistant to four drugs7021210100001(14.4)Resistant to SM1483185735235610007(4.7)Resistant to EMB77292817123410005(6.5)Resistant to OFX862132723112620008(9.3)Resistant to KM30277921100001(3.3)Resistant to SM and (or) EMB82483518133100001(1.2)Resistant to KM and (or) OFX13133410100001(7.7)Resistant to any first and second line drugs791112621103520007(8.9) ## WGS Identification of BDQ resistance-related mutations The BDQ-resistant mutants were performed by WGS in 205 MDR isolates (Table 3). No mutations within the atpE, pepQ, and Rv1979 gene were observed in 9 BDQ resistant isolates. Six BDQ resistant isolates ($66.7\%$, $\frac{6}{9}$) and two BDQ susceptible isolates ($1.0\%$, $\frac{2}{196}$) carried mutations in Rv0678, which has statistical significance. A total of 4 mutations types were identified in BDQ resistant isolates, including A152G mutation causing Gln51Arg amino acid change ($50\%$, $\frac{3}{6}$), T56C mutation causing Phe19Ser amino acid change ($16.7\%$, $\frac{1}{6}$), GA492 insertion ($16.7\%$, $\frac{1}{6}$), and A274 insertion ($16.7\%$, $\frac{1}{6}$). Besides, G307A causing Gly103Ser amino acid change and G184A causing Ala62Thr amino acid change in the Rv0678 gene were identified in BDQ sensitive isolates. The six BDQ resistant isolates with mutations in Rv0678 gene all belonged to ancient Beijing genotype, and were resistant to at least two drugs in Table 4. Both of the two BDQ susceptible isolates with mutations in Rv0678 gene, one non-Beijing and one mordern Beijing genotype, were resistant to SM.Table 3Mutation analysis of BDQ resistant genes among 205 MDR isolatesResistance patternIsolate numberGene mutation typeNo. of isolates (%)MIC of BDQ (μg/ml)atpERv0678pepQRv1979BDQ resistant isolates [9]22A050, 22A133, 22A148WTCAG152CGG Gln51ArgWTWT30.50022A118WTTTC56TCC Phe19SerWTWT10.50022A177WT492 position ins_GAWTWT11.00022A180WT274 position ins -AWTWT11.000Total6 (66.7)a0.500–1.000BDQ sensitive isolates [196]22A076WTGGC307AGC Gly103SerWTWT10.12522A079WTGCC184ACC Ala62ThrWTWT10.25022A128WTWTGCC411GCT Ala137AlaWT10.03122A174WTWTGAA1080GAT/Glu360AspWT10.01622A012, 22A025, 22A030, 22A032, 22A041WTWTWTGTT1276ATT/Val426lle50.031–0.06322A222WTWTWTC(−70)G10.06322A196WTWTWTGCG717GCA/Ala239Ala10.06322A005WTWTWTTCG785TTG/Ser262Leu10.01622A227WTWTWTGTC286ATC/Val96lle10.03122A204WTWTWTGCC449GTC/Ala150Val10.25022A039WTWTWTGTT155GGT/Val52Gly10.03122A016, 22A029, 22A201, 22A220, 22A223WTWTWTWT5≤ 0.008–0.03122A006, 22A009WTWTWTWT20.03122A044WTWTWTWT10.01622A206WTWTWTWT10.03122A028WTWTWTWT1≤ 0.00822A008WTWTWTWT10.031Total26 (13.3)b≤ 0.008–0.250Compared a to b: X2 = 14.795P < 0.001Table 4Drug resistance data of isolates with mutations in Rv0678 geneIsolate numberMIC of BDQ (μg/ml)Drug resistance profileGenotype22A0500.5SM + EMB + OFXAncient Beijing22A1180.5SM + OFXAncient Beijing22A1330.5SM + EMB + OFXAncient Beijing22A1480.5SM + EMB + OFX + KMAncient Beijing22A1771EMB + OFX + KMAncient Beijing22A1801SM + EMB + OFXAncient Beijing22A0760.125SMnon-Beijing22A0790.25SMMordern Beijing ## Genotypic predictions As shown in Table 5, the sensitivity of WGS prediction for BDQ resistance was $66.7\%$, the specificity was $99.0\%$, the positive predictive value was $75.0\%$, and the negative predictive value was $98.5\%$.Table 5WGS predictions versus DST phenotype for BDQWGSDST phenotype (n)TotalSensitivity (%)Specificity (%)PPV (%)NPV (%)KappaResistant[9]Sensitive[196]Mutation62866.7(30.9–91.0)99.0(96.0–99.8)75.0(35.6–95.5)98.5(95.3–99.6)0.693Non-mutaion3194197 ## Discussion Although BDQ has been proven to be highly effective in the treatment of MDR-TB [16], inadequate or incomplete use may lead to the emergence of resistant strains [17]. Unfortunately, few studies have explored the resistance status of MDR-TB against BDQ in Chongqing. Therefore, we performed drug susceptibility test and conducted sequence analyses of BDQ resistance genes for 205 MDR isolates. The resistance rate of MDR-TB to BDQ was $4.4\%$, lower than that of commonly used first- and second-line drugs, indicating that BDQ has strong activity against MDR isolates in Chongqing. Though the resistance rate lower than that reported in Shanxi ($5.56\%$) [15] and in national survey in China ($7.16\%$) [18], higher than reported in a retrospective cohort study in China ($2.2\%$) [19] and national drug resistance surveillance in 2015 ($1\%$) [20]. These inconsistent results may be attributed to the difference in the epidemic strains, medication background and the breakpoints used across studies. Given the cross resistance between BDQ and clofazimine, prior exposure to clofazimine could reduce the susceptibility to BDQ [21]. And the period from the start of treatment can also affect the BDQ MIC [22]. To our knowledge, all isolates were without documented prior use of BDQ, and $4.4\%$ MDR-TB strains resistant to BDQ suggesting that though BDQ showed excellent activity against MDR-TB, the emergence of BDQ resistant isolates may lead to the rapid loss of this valuable new drug. Therefore, it is necessary to dynamically monitor the BDQ resistance to optimize BDQ administration regimen, further to avoid the occurrence of acquired resistance, and maximize the effectiveness of new drugs, even in patients who have not been exposed to BDQ. The resistance rate of BDQ in isolates resistant to any first and second line drug ($8.9\%$) was higher than that in isolates resistant to first line drugs ($1.2\%$) and second line drugs ($7.7\%$), indicating that with the increase of drug resistance types and the complexity of resistant background, the BDQ resistance rate also increased. In addition, we found that the BDQ resistance rate in retreated patients ($66.7\%$) was higher than that of new patients ($33.3\%$), whether this attributed to the past medical history needs to be further studied. Of the 9 BDQ resistant isolates, the proportion of OFX resistant isolates ($\frac{8}{9}$) was significantly higher than that of OFX sensitive isolate ($\frac{1}{9}$), and the resistance rate of BDQ in OFX resistant isolates ($9.3\%$) was higher than that in SM resistant isolates ($4.7\%$), EMB resistant isolates ($6.5\%$), KM resistant isolates ($3.3\%$), suggesting isolates resistant to OFX were more likely to develop BDQ resistance, which was a risk factor of BDQ resistance. Since the development and approval of BDQ for clinical use, the number of BDQ resistant isolates associated with inadequate or incomplete treatment is steadily growing [22]. To investigate the potential mechanisms and genetic background of BDQ resistant isolates, we performed whole-genome sequencing. Though the fact that mutations in the atpE, pepQ, and Rv1979c gene confer bedaquiline resistance [3, 7, 8], no mutations were observed in this study. The $66.7\%$ ($\frac{6}{9}$) BDQ resistant isolates had variants in the Rv0678 gene, which was the main mechanism of primary BDQ resistance in Chongqing, and all belonged to low level resistance (0.5–1 μg/ml). The mutation loci in Rv0678 gene were scattered and the mutation types were complicated. Of the 6 isolates carrying Rv0678 mutations included two non-synonymous Single Nucleotide Polymorphisms SNPs and deletions, the most frequently variations were A152G ($50\%$), which has reported to be associated with BDQ resistance in MDR isolates [23]. Besides, the A274 insertion identified in the present study was found in clinical BDQ-resistant isolates [6]. However, there were three BDQ resistance isolates ($33.3\%$, $\frac{3}{9}$) without mutations, suggesting additional mechanisms must be involved in the resistance, such as other potential target and non-target resistance mechanisms, such as changes in cell wall permeability caused by transcriptional and protein levels and drug efflux pump structure. Two BDQ susceptible isolates with mutations in Rv0678 gene were in the critical concentration of BDQ resistance and a gradient below the critical concentration, which may be attributed to operational factors, such as result interpretation, bacteria activity, drug concentration or other inaccurate factors. Two pepQ mutant strains and 11 Rv1979 mutant strains were all sensitive to BDQ, which were not related to drug resistance. Moreover, the other two (Rv0678 T56C and GA492 insertion) were novel mutation types, which were not reported previously. Further analysis in expression levels of MmpS5 and MmpL5 efflux pump will contribute to illustrate the role of these novel mutations in BDQ resistance. The Beijing genotype was the predominant isolates in Chongqing with $47.8\%$ modern Beijing genotype and $35.1\%$ ancient Beijing genotype. However, the proportion of ancient Beijing strains ($88.9\%$, $\frac{8}{9}$) was significantly higher than that of modern Beijing strains ($11.1\%$, $\frac{1}{9}$) in BDQ resistant isolates, and $75\%$ ($\frac{6}{8}$) BDQ resistant isolates with Rv0678 mutation were ancient Beijing type, indicating ancient Beijing genotype was more prone to BDQ resistance and Rv0678 mutation. In this study, WGS for BDQ drug resistance was consistent with phenotypic drug susceptibility test. However, the relatively dispersed mutation loci of BDQ resistance associated genes may result in the presence of "false-susceptible" detected by PCR-sequencing of hot spots of current resistance-associated genes. Therefore, WGS can quickly and accurately determine the mutation loci, and has preferable specificity ($99\%$) in predicting BDQ resistance. But for the non-target resistance mechanism, the phenotypic drug sensitive test was superior to WGS. So, the phenotypic drug sensitive test together with WGS was helpful to early diagnosis and individualized treatment of drug-resistant tuberculosis, which has excellent application value in the rapid detection of BDQ resistance. ## Conclusions BDQ showed excellent activity against MDR-TB in Chongqing. The resistance rate of BDQ was not related to demographic and clinical characteristics. Mutations in Rv0678 gene were the major mechanism to BDQ resistance, with A152G as the most common mutation type. WGS has a good popularize value and application prospect in the rapid detection of Bdq resistance. ## References 1. 1.World Health OrganizationGlobal tuberculosis report 20202020GenevaWorld Health Organization. Global tuberculosis report 2020 (2020) 2. Yao C, Guo HP, Li Q, Zhang XX, Shang YY, Li TX, Wang YF, Xue ZT, Wang L, Li L. Prevalence of extensively drug-resistant tuberculosis in a Chinese multidrug-resistant TB cohort after redefinition. Antimicrob Resist Infect Control (2021) 10 126. DOI: 10.1186/s13756-021-00995-8 3. Andries K, Verhasselt P, Guillemont J, Göhlmann HWH, Neefs JM, Winkler H, Gestel JV, Timmerman P, Zhu M, Lee E. A diarylquinoline drug active on the ATP synthase of. Science (2005) 307 223-227. DOI: 10.1126/science.1106753 4. 4.World Health OrganizationConsolidated guidelines on drug-resistant tuberculosis treatment2019GenevaWorld Health Organization. Consolidated guidelines on drug-resistant tuberculosis treatment (2019) 5. Hartkoorn RC, Uplekar S, Cole ST. Cross-Resistance between clofazimine and bedaquiline through upregulation of MmpL5 in. Antimicrob Agents Chemother (2014) 58 2979-2981. DOI: 10.1128/AAC.00037-14 6. Villellas C, Coeck N, Meehan CJ, Lounis N, Jong BD, Rigouts L, Andries K. Unexpected high prevalence of resistance-associated Rv0678 variants in MDR-TB patients without documented prior use of clofazimine or bedaquiline. J Antimicrob Chemother (2017) 72 684-690. PMID: 28031270 7. Nieto Ramirez LM, Quintero Vargas K, Diaz G. Whole genome sequencing for the analysis of drug resistant strains of. Antibiotics. (2020) 9 133. DOI: 10.3390/antibiotics9030133 8. Almeida D, Ioerger T, Tyagi S, Li SY, Mdluli K, Andries K, Grosset J, Sacchettinie J, Nuermberger E. Mutations in pepQ confer low-level resistance to bedaquiline and clofazimine in. Antimicrob Agents Chemother (2016) 60 4590-9. DOI: 10.1128/AAC.00753-16 9. Degiacomi G, Sammartino JC, Sinigiani V, Marra P, Urbani A, Pasca MR. In vitro study of bedaquiline resistance in. Front Microbiol (2020) 11 559469. DOI: 10.3389/fmicb.2020.559469 10. Falzon D, Jaramillo E, Schunemann HJ, Arentz M, Bauer M, Bayona J, Blanc L, Caminero JA, Daley CL, Duncombe C. WHO guidelines for the programmatic management of drug-resistant tuberculosis: 2011 update. Eur Respir J (2011) 38 516-528. DOI: 10.1183/09031936.00073611 11. Zhao YL, Xu SF, Wang LX, Chin DP, Wang SF, Jiang GL, Xia H, Zhou Y, Li Q, Ou XC. National survey of drug-resistant tuberculosis in China. N Engl J Med (2012) 366 2161-2170. DOI: 10.1056/NEJMoa1108789 12. Xie LX, Wang XB, Zeng J, Zhou ML, Duan XK, Li QM, Zhang Z, Luo HP, Pang L, Li W. Proteome-wide lysine acetylation profiling of the human pathogen. Int J Biochem Cell Biol (2015) 59 193-202. DOI: 10.1016/j.biocel.2014.11.010 13. Cholo MC, Mothiba MT, Fourie B, Anderson R. Mechanisms of action and therapeutic efficacies of the lipophilic antimycobacterial agents clofazimine and bedaquiline. J Antimicrob Chemother (2017) 72 338-353. DOI: 10.1093/jac/dkw426 14. Dookie N, Rambaran S, Padayatchi N, Mahomed S, Naidoo K. Evolution of drug resistance in. J Antimicrob Chemother (2018) 73 1138-1151. DOI: 10.1093/jac/dkx506 15. Yang J, Pang Y, Zhang TH, Xian XP, Li Y, Wang R, Wang PT, Zhang M, Wang JY. Molecular characteristics and in vitro susceptibility to bedaquiline of. China Int J Infect Dis (2020) 99 163-170. DOI: 10.1016/j.ijid.2020.07.044 16. Ahmad N, Ahuja SD, Akkerman W, Alffenaar JWC, Andersonet LF, Baghaei P, Bang D, Barry PM, Bastos ML. Treatment correlates of successful outcomes in pulmonary multidrug-resistant tuberculosis: an individual patient data meta-analysis. Lancet (2018) 392 821-34. DOI: 10.1016/S0140-6736(18)31644-1 17. Khoshnood S, Goudarzi M, Taki E, Darbandi A, Kouhsari E, Heidary M, Motahar M, Moradi M, Bazyar H. Bedaquiline: current status and future perspectives. J Glob Antimicrob Resist (2021) 25 48-59. DOI: 10.1016/j.jgar.2021.02.017 18. Wang GR, Jiang GL, Jing W, Zong ZJ, Yu X, Chen XT, Li WM, Huang HR. Prevalence and molecular characterizations of seven additional drug resistance among multidrug-resistant tuberculosis in China: a subsequent study of a national survey. J Infect (2021) 82 371-377. DOI: 10.1016/j.jinf.2021.02.004 19. Huang HR, Ding N, Yang TT, Li CD, Jia XM, Wang GR, Zhong J, Zhang J, Jiang GL, Wang SQ. Cross-sectional whole-genome sequencing and epidemiological study of multidrug-resistant. Clin Infect Dis (2019) 69 405-413. DOI: 10.1093/cid/ciy883 20. He WC, Liu CF, Liu DX, Ma AJ, Song YM, He P, Bao JJ, Li YC, Zhao B, Fan JL. Prevalence of Mycobacterium tuberculosis resistant to bedaquiline and delamanid in China. J Glob Antimicrob Resist (2021) 26 241-248. DOI: 10.1016/j.jgar.2021.06.007 21. Liu YH, Gao MQ, Du J, Wang L, Gao JT, Shu W. Reduced Susceptibility of. Clin Infect Dis (2021) 73 e3391-e3397. DOI: 10.1093/cid/ciaa1002 22. Peretokina IV, Krylova LY, Antonova OV, Kholina MS, Kulagina EV, Nosova EY, Safonova SG, Borisov SE, Zimenkov DV. Reduced susceptibility and resistance to bedaquiline in clinical. J Infect (2020) 80 527-35. DOI: 10.1016/j.jinf.2020.01.007 23. Ismail N, Omar SV, Ismail NA, Peters RPH. Collated data of mutation frequencies and associated genetic variants of bedaquiline, clofazimine and linezolid resistance in. Data Brief (2018) 20 1975-1983. DOI: 10.1016/j.dib.2018.09.057
--- title: 'Clinical significance of obstructive sleep apnea in patients with acute coronary syndrome with or without prior stroke: a prospective cohort study' authors: - Bin Wang - Wen Hao - Jingyao Fan - Yan Yan - Wei Gong - Wen Zheng - Bin Que - Hui Ai - Xiao Wang - Shaoping Nie journal: European Journal of Medical Research year: 2023 pmcid: PMC9976418 doi: 10.1186/s40001-023-01071-0 license: CC BY 4.0 --- # Clinical significance of obstructive sleep apnea in patients with acute coronary syndrome with or without prior stroke: a prospective cohort study ## Abstract ### Background and objective Whether obstructive sleep apnea (OSA) is associated with worse prognosis in patients with acute coronary syndrome (ACS) with or without prior stroke remains unclear. We investigated the association of OSA with cardiovascular events in ACS patients with or without prior stroke. ### Methods Between June 2015 and January 2020, we prospectively recruited eligible ACS patients who underwent cardiorespiratory polygraphy during hospitalization. We defined OSA as an apnea hypopnea index (AHI) ≥ 15 events/hour. The primary composite end point was major adverse cardiovascular and cerebrovascular events (MACCEs), including cardiovascular death, myocardial infarction, stroke, ischemia-driven revascularization, or hospitalization for unstable angina or heart failure. ### Results Among 1927 patients enrolled, 207 patients had prior stroke ($10.7\%$) and 1014 had OSA ($52.6\%$). After a mean follow-up of 2.9 years, patients with stroke had significantly higher risk of MACCEs than those without stroke (hazard ratio [HR]:1.49; $95\%$ confidence interval [CI]: 1.12–1.98, $$P \leq 0.007$$). The multivariate analysis showed that patients with OSA had 2.0 times the risk of MACCEs in prior stroke group (41 events [$33.9\%$] vs 18 events [$20.9\%$]; HR:2.04, $95\%$ CI:1.13–3.69, $$P \leq 0.018$$), but not in non-prior stroke group (186 events [$20.8\%$] vs 144 events [17.4]; HR:1.21, $95\%$ CI 0.96–1.52, $$P \leq 0.10$$). No significant interaction was noted between prior stroke and OSA for MACCE (interaction $$P \leq 0.17$$). ### Conclusions Among ACS patients, the presence of OSA was associated with an increased risk of cardiovascular events in patients with prior stroke. Further trials exploring the efficacy of OSA treatment in high-risk patients with ACS and prior stroke are warranted. Trial registration Clinicaltrials.gov identifier NCT03362385. ### Supplementary Information The online version contains supplementary material available at 10.1186/s40001-023-01071-0. ## Background Obstructive sleep apnea (OSA) is a complex and heterogeneous common chronic disease characterized by repetitive episodes of upper airway collapse, affecting $40\%$–$60\%$ of patients with ACS [1, 2]. Current evidence has shown that OSA initiates and exacerbates coronary atherosclerosis and is closely related to poor outcomes in patients with ACS or cerebrovascular disease [1, 3–6]. However, several randomized controlled trials have indicated that treatment with continuous positive airway pressure (CPAP) is not associated with lower rates of recurrent cardiovascular events in patients with ACS [7–10]. It is well known that ACS and stroke share common pathogeneses, such as lipid disorders and arterial thrombosis. For that reason, ACS patients presenting with a history of stroke were not uncommon, constituted $8.25\%$ of patients, and presented a therapeutic conundrum [11]. Furthermore, patients with prior stroke had particularly poor outcomes compared with those without prior stroke, including a higher risk of death, MI, and stroke [11–15]. Preliminary evidence also suggests that OSA is an independent risk factor for stroke and is associated with recurrent ischemic stroke and worsening outcomes [16–18]. Although whether CPAP treatment reduces the risk of stroke in OSA patients remains controversial, patients’ adherent to CPAP therapy (> 4 h per day) may benefit [19]. In addition, it is unclear whether the effect of OSA on the prognosis of patients with ACS varies based on previous stroke. Considering the association between ACS, OSA and stroke, we hypothesized that OSA combined with prior stroke may have a synergistic deleterious effect that increases future cardiovascular risk. Hence, based on a large, prospective cohort study, we evaluated the impact of OSA on the risk of cardiovascular events in ACS patients with or without prior stroke. ## Study design and population The OSA-ACS project (NCT03362385) is a prospective, observational, single-center study designed to evaluate the association between OSA and cardiovascular outcomes among patients with ACS. The study design has been described previously [3, 20]. In the current study, we aimed to investigate the clinical importance of OSA for patients with ACS stratified by stroke history. In brief, from June 2015 to January 2020, in the Center for Coronary Artery Disease at Beijing Anzhen Hospital, Capital Medical University, ACS patients aged 18 years to 85 years were enrolled and underwent overnight sleep studies. The exclusion criteria included cardiac arrest or cardiogenic shock, malignancies, and failed sleep studies (those who failed to obtain adequate and satisfactory recordings). Next, patients with predominantly central sleep apnea (≥ $50\%$ central events and a central apnea–hypopnea index (AHI) ≥ 10/h), recording time < 180 min, and those receiving regular CPAP therapy (> 4 h/day and > 21 days/month) after discharge were excluded. The protocol was approved by the Ethics Committee of Beijing Anzhen Hospital, Capital Medical University [2,013,025]. All participants provided written informed consent, and the study was conducted according to the amended Declaration of Helsinki. Thirty patients were lost to follow-up and therefore excluded from the analysis. The study flowchart is presented in Fig. 1.Fig. 1Flowchart of the study. ACS, acute coronary syndrome; CPAP, continuous positive airway pressure; OSA, obstructive sleep apnea ## Overnight sleep study For eligible patients with a hospital stay of 24–72 h, an overnight sleep study was conducted using portable cardiorespiratory polygraphy (ApneaLink Air, ResMed, Australia), which has previously been validated [21, 22]. All sleep studies were double scored manually by independent sleep somnologists blinded to the clinical characteristics, and confirmed by a senior somnologist in cases of discrepancy. They reviewed all the data and checked whether there are missed or misjudged events. The following signals were recorded: nasal airflow, thoraco-abdominal movements, snoring episodes, pulse, and percutaneous oxygen saturation (SpO2). Apnea was defined as an absence of airflow for 10 s or more. Hypopnea was defined as an airflow reduction of $30\%$ for ≥ 10 s with a decrease in SpO2 > $4\%$. Oxygen desaturation was defined as a decrease in arterial oxygen saturation greater than $4\%$. The AHI was defined as the number of episodes of apnea hypopnea per hour of recording. The oxygen desaturation index (ODI) was calculated as the amount of time when the oxygen saturation dropped by ≥ $4\%$ from baseline per hour of sleep. Valid tests required a minimum of 3 h of satisfactory signal recording. In accordance with previous studies, recruited patients were categorized into OSA (AHI ≥ 15 events/h) and non-OSA (AHI < 15 events/h) groups [4, 23]. ## Procedures and management Clinical care for all patients was offered at the discretion of the attending clinician based on current guidelines [24, 25]. When clinically indicated, PCI with stenting or coronary artery bypass graft (CABG) was performed. Patients with OSA, particularly those with excessive daytime sleepiness, were referred to sleep centers for further evaluation and treatment. ## Follow-up and outcomes Following the sleep study, follow-ups were performed at one month, three months, and six months and then every six months afterwards. Clinical visits, medical chart reviews, and telephone calls by two investigators blinded to the patients' clinical information were used to collect clinical event data. At each visit, a composite of cardiovascular events was assessed, and events were then confirmed by source documentation and adjudicated by the clinical event committee. Major adverse cardiovascular and cerebrovascular events (MACCEs), defined as a composite of cardiovascular death, MI, stroke, ischemia-driven revascularization, or hospitalization for unstable angina or heart failure, were the primary outcomes of this study. Secondary cardiovascular endpoints included the individual components of the primary composite endpoint and other composites of cardiovascular events, including cardiovascular death, MI, or stroke. The endpoints were defined according to the Standards for Data Collection in Cardiovascular Trials Initiative [26]. ## Statistical analysis Data were compared using Student's t test or the Mann–Whitney U test for continuous variables expressed as means ± standard deviations or medians (interquartile ranges). Counts and proportions (%) for categorical variables were compared using Fisher’s exact test or χ2 statistics as appropriate. The Kaplan–Meier product limit method was used to calculate the survival rate as the period from the initial sleep study to any MACCEs, with the data censored at the last recorded follow-up. Cox regression analysis was used to investigate independent risk factors for endpoints, and adjusted hazard ratios (HR) with $95\%$ CI were calculated. Cox regression models were built using baseline variables that were considered clinically relevant or demonstrated a univariate relationship with the outcomes. The AHI was divided by AHI < 15, 15–30, > 30 to assess the relationship between OSA severity and the risk of cardiovascular events in patients with or without prior stroke. Multiplicative interaction terms were included in the fully adjusted models to evaluate if prior stroke modified the associations between OSA and risk of cardiovascular events. All analyses were conducted with SPSS V.26.0 (IBM SPSS, Armonk, New York, USA). A two-sided $P \leq 0.05$ was considered statistically significant. ## Baseline characteristics Between June 2015 and January 2020, a total of 2160 patients with ACS were recruited, of whom 2109 underwent successful cardio-respiratory polygraphy. Among them, 1,969 patients with OSA and follow-up were included. Among them, only 42 ($2.1\%$) of patients received regular CPAP therapy (> 4 h/day and > 21 days/month) and the rate was similar between those with and without prior stroke ($2.8\%$ vs. $2.1\%$, $$P \leq 0.47$$). A total of 1927 ACS patients met the initial eligibility criteria and 207 patients ($10.74\%$) had a stroke history. A previous stroke was associated with older age, a less proportion of males, higher systolic blood pressure, and a higher prevalence of established risk factors, such as diabetes, hypertension, and hyperlipidemia, a lower level of Min SpO2, Mean SpO2, and a higher level of T90 SpO2 < $90\%$. Previous MI, PCI, and CABG rates were similar between the two groups (Additional file 1: Table S1). The baseline characteristics of the OSA and non-OSA groups with or without prior stroke are shown in Table 1. Patients with OSA exhibited a higher body mass index (BMI) and neck and waist circumferences, irrespective of prior stroke. In the non-prior stroke group, patients with OSA were more males, more likely to have hypertension and prior PCI, had higher level of glycosylated hemoglobin and C-reactive protein, lower level of high-density lipoprotein, and lower left ventricular ejection fraction, and more likely to receive PCI and CABG procedures. In both the prior stroke group and the non-prior stroke group, other characteristics were generally well matched between OSA and non-OSA patients. Table 1Characteristics of the patients at baselineVariablesAll ($$n = 1927$$)Prior stroke ($$n = 207$$)Non-prior stroke ($$n = 1720$$)OSA ($$n = 121$$)Non-OSA($$n = 86$$)P-valueOSA ($$n = 893$$)Non-OSA($$n = 827$$)P valueDemographics Age, y56.4 ± 10.562.0 ± 8.961.6 ± 9.60.8055.8 ± 10.655.7 ± 10.30.78 Male1629 ($84.5\%$)100 ($82.6\%$)65 ($75.6\%$)0.21786 ($88.0\%$)678 ($82.0\%$) < 0.001 BMI, kg/m227.1 ± 3.627.9 ± 3.725.8 ± 3.2 < 0.00128.1 ± 3.526.0 ± 3.4 < 0.001 Neck, circumference, cm41 (38–43)41 (39 to 44)39 (37 to 41) < 0.00141.5 (39 to 44)40 (38 to 42) < 0.001 Waist, circumference, cm99 (93 to 105)101 (97 to 109)96 (90 to 100.3) < 0.001101 (95 to 108)97 (91 to 102) < 0.001 Systolic BP, mm Hg126 (117–138)127 (118 to 145)130 (120 to 142)0.47126 (117 to 138)126 (116 to 137)0.26 Diastolic BP, mm Hg76 (70–84)76 (70 to 83.5)75 (68.8 to 81.0)0.6478 (70 to 86)75 (69 to 83) < 0.001Medical history Diabetes mellitus609 ($31.6\%$)50 ($58.1\%$)36 ($41.9\%$)0.46262 ($29.3\%$)254 ($30.7\%$)0.53 Hypertension1247 (64.7)103 ($85.1\%$)68 ($79.1\%$)0.26588 ($65.8\%$)488 ($59.0\%$)0.003 Hyperlipidemia637 (33.1)52 ($43.0\%$)31 ($36.0\%$)0.32291 ($32.6\%$)263 ($31.8\%$)0.73 Prior MI316 (16.4)23 ($19.0\%$)14 ($16.3\%$)0.61154 ($17.2\%$)125 ($15.1\%$)0.23 Previous PCI399 (20.7)33 ($27.3\%$)16 ($18.6\%$)0.15201 ($22.5\%$)149 ($18.0\%$)0.021 Previous CABG29 (1.5)1 ($0.8\%$)1 ($1.2\%$)0.8117 ($1.9\%$)10 ($1.2\%$)0.25 Smoking0.730.08 No953 ($49.5\%$)76 ($62.8\%$)52 ($60.5\%$)442 ($49.5\%$)383 ($46.3\%$) Yes974 ($50.5\%$)45 ($37.2\%$)34 ($39.5\%$)451 ($50.5\%$)444 ($53.7\%$)Baseline tests Glucose, mmol/L5.98 (5.32 to 7.51)6.10 (5.47 to 8.24)6.10 (5.28 to 7.44)0.246.04 (5.33 to 7.61)5.90 (5.29 to 7.29)0.17 Hemoglobin A1C, %6.10 (5.60 to 7.00)6.30 (5.60 to 7.38)6.10 (5.70 to 7.00)0.356.10 (5.70 to 7.00)6.00 (5.60 to 6.90)0.028 Triglyceride, mmol/L1.51 (1.10 to 2.20)1.44 (1.07 to 1.90)1.38 (1.11 to 1.89)0.531.59 (1.13 to 2.31)1.47 (1.05 to 2.18)0.16 Total Cholesterol, mmol/L4.12 (3.47 to 4.91)3.93 (3.40 to 4.50)3.92 (3.46 to 4.80)0.474.18 (3.52 to 4.94)4.09 (3.41 to 4.96)0.35 HDL-C, mmol/L1.00 (0.86 to 1.16)0.98 (0.87 to 1.14)1.01 (0.84 to 1.17)0.800.98 (0.85 to 1.13)1.02 (0.87 to 1.18)0.003 LDL-C, mmol/L2.43 (1.90 to 3.09)2.34 (1.92 to 2.88)2.31 (1.90 to 3.00)0.722.46 (1.95 to 3.11)2.42 (1.83 to 3.11)0.17 Hs-CRP, mmol/L2.06 (0.82 to 6.13)2.35 (0.86 to 9.25)1.50 (0.65 to 3.98)0.242.53 (1.05 to 7.24)1.44 (0.61 to 4.69) < 0.001 LVEF, %61.0 (56.0 to 65.0)62.0 (55.5 to 65.5)58.0 (66.0 to 66.0)0.8660.0 (55.0 to 65.0)62.0 (57.0 to 66.0)0.010Diagnosis0.960.013 Unstable angina1132 ($58.7\%$)74 ($61.2\%$)54 ($62.8\%$)498 ($55.8\%$)506 ($61.2\%$) NSTEMI365 ($18.9\%$)23 ($19.0\%$)15 ($17.4\%$)168 ($18.8\%$)159 ($19.2\%$) STEMI430 ($22.3\%$)24 ($19.8\%$)17 ($19.8\%$)227 ($25.4\%$)162 ($19.6\%$)Procedures Vessels0.640.19 0170 ($8.8\%$)9 ($7.4\%$)7 ($7.8\%$)69 ($7.7\%$)85 ($10.3\%$) 1512 ($26.6\%$)21 ($17.4\%$)19 ($22.1\%$)240 ($26.9\%$)232 ($28.1\%$) ≥ 21245 ($64.6\%$)91 ($75.2\%$)60 ($69.8\%$)584 ($65.4\%$)510 ($61.7\%$) PCI1209 ($62.7\%$)72 ($59.5\%$)52 ($60.5\%$)0.89595 ($66.6\%$)490 ($59.3\%$)0.002 CABG130 ($6.7\%$)14 ($11.6\%$)10 ($11.6\%$)0.9945 ($5.0\%$)61 ($7.4\%$)0.044Medications on discharge Aspirin1877 ($97.4\%$)114 ($94.2\%$)81 ($94.2\%$)0.99873 ($97.8\%$)809 ($97.8\%$)0.93 P2Y12 inhibitors1768 ($91.7\%$)110 ($90.9\%$)80 ($93.0\%$)0.59828 ($92.7\%$)750 ($90.7\%$)0.13 β-blockers1488 ($77.2\%$)92 ($76.0\%$)52 ($60.5\%$)0.016707 ($79.2\%$)637 ($77.0\%$)0.28 ACEIs/ARBs1195 ($62.0\%$)82 ($67.8\%$)55 ($64.0\%$)0.57583 ($65.3\%$)475 ($57.4\%$)0.001 Statins1897 ($98.4\%$)119 ($98.3\%$)83 ($96.5\%$)0.40878 ($98.3\%$)817 ($98.8\%$)0.42The data is presented as mean ± SD, median (first quartile to third quartile), or n (%)ACEIs Angiotensin-Converting Enzyme Inhibitors, ARBs angiotensin-receptor blockers, BMI body mass index, BP blood pressure, CAD coronary artery disease, CABG coronary artery bypass grafting, LVEF left ventricular ejection fraction, MI myocardial infarction, OSA obstructive sleep apnea, PCI percutaneous coronary intervention ## Results of the sleep study The characteristics differed significantly between the two groups. The prevalence of OSA was $52.6\%$, and patients with OSA had a lower minimum oxygen saturation than those without OSA. Further information is shown in Table 2.Table 2Results of sleep studyVariablesAll ($$n = 1297$$)Prior stroke ($$n = 207$$)Non-prior stroke ($$n = 1720$$)P valueOSA ($$n = 893$$)Non-OSA($$n = 827$$)P valueOSA ($$n = 121$$)Non-OSA($$n = 86$$)AHI, events/h16.0 (8.0 to 30.0)29.6 (23.1 to 40.7)7.6 (4.2 to 10.2) < 0.00128.8 (20.6 to 42.307.7 (4.1 to 10.8) < 0.001ODI, events/h16.2 (8.8 to 28.6)27.3 (21.3 to 36.4)8.9 (5.9 to 11.5) < 0.00127.6 (20.1 to 40.0)8.5 (4.8 to 11.9) < 0.001Min SpO2, %85 (81 to 88)80 (76 to 85)87 (85 to 89) < 0.00183 (78 to 86)88 (84 to 90) < 0.001Mean SpO2, %94 (93 to 95)93 (92 to 94)94 (93 to 95) < 0.00193 (92 to 94)94 (93 to 95) < 0.001T90 SpO2 < $90\%$2.3 (0.4 to 10.0)7.7 (3.0 to 23.0)1.0 (0.2 to 5.2) < 0.0016.0 (2.0 to 15.0)0.5 (0.0 to 3.0) < 0.001Epworth Sleepiness Scale7.0 (4.0 to 11.0)8.6 (6.0 to 12.0)6.8 (3.0 to 11)0.0368.3 (5.0 to 12.0)7.0 (3.0–10.0) < 0.001The data is presented as median (first quartile to third quartile)AHI indicates apnea–hypopnea index, ODI oxygen desaturation index, OSA obstructive sleep apnea, SpO2 percutaneous oxygen saturation, T90 SpO2 percentage of total sleep time with saturation < $90\%$ ## Outcomes in the overall population according to OSA and prior stroke The mean median follow-up time was 2.9 years (1.5 to 3.6). Among patients with ACS, the presence of OSA was associated with a higher rate of MACCE compared with patients without OSA in the overall population (log-rank, $$P \leq 0.004$$). Kaplan–Meier analysis showed that the cumulative incidence of MACCEs was significantly higher in the prior stroke group than in the non-prior stroke group (log-rank, $P \leq 0.001$; Additional file 1: Fig. S1). After adjustment for the baseline risk of cardiovascular events, patients with prior stroke were strongly associated with a higher rate of MACCEs than patients without prior stroke (HR: 1.49, $95\%$ CI: 1.12–1.98, $$P \leq 0.007$$). Furthermore, the patients were classified into three different groups based on their AHI: no/mild OSA (AHI < 15), moderate OSA (15 ≤ AHI ≤ 30), and severe OSA (30 < AHI). The association between OSA and MACCE among patients with prior stroke remained significant in moderate and severe OSA compared with the no/mild OSA (Additional file 1: Fig. S2). ## Outcomes of OSA versus non-OSA patients stratified by prior stroke In the prior stroke group, Kaplan–Meier analysis showed that MACCEs were significantly more common in the OSA group than in the non-OSA group (log-rank, $$P \leq 0.019$$; Fig. 2A). In the non-prior stroke group, the incidence of MACCEs was also higher in the OSA group than in the non-OSA group (log-rank, $$P \leq 0.039$$; Fig. 2B). After adjustment for age, sex, body mass index, smoking status, hypertension, diabetes mellitus and dyslipidemia, OSA was associated with an increased risk of MACCEs in the prior stroke group (HR: 2.04, $95\%$ CI: 1.13–3.69, $$P \leq 0.018$$) but not in the non-prior stroke group (HR: 1.21, $95\%$ CI: 0.96–1.52, $$P \leq 0.10$$) (Table 3). No significant interaction was noted between prior stroke and OSA for MACCE (interaction $$P \leq 0.17$$).Fig. 2Kaplan–Meier curves for the analysis cardiovascular events in OSA versus non-OSA groups in prior stroke group and non-prior stroke group. Kaplan–Meier estimates hospitalization for MACCE (A, B) and composite of cardiovascular death, myocardial infarction, and stroke (C, D) between OSA and non-OSA groups in prior stroke group (A, C) and non-prior stroke group (B, D). MACCE, major adverse cardiovascular and cerebrovascular event. OSA, obstructive sleep apneaTable 3Cox regression analyses evaluating the association between OSA and risk of cardiovascular events by a history of strokeVariablesPrior stroke ($$n = 207$$)Non-prior stroke ($$n = 1720$$)Unadjusted HR ($95\%$ CI)P-valueAdjusted HRa ($95\%$ CI)P-valueUnadjusted HR ($95\%$ CI)P valueAdjusted HRa ($95\%$ CI)P valueMACCE1.93 (1.10 to 3.36)0.0212.04 (1.13 to 3.69)0.0181.26 (1.01 to 1.56)0.0401.21 (0.96 to 1.52)0.10Cardiovascular death1.15 (0.32 to 4.06)0.831.09 (0.28 to 4.23)0.901.23 (0.54 to 2.80)0.631.05 (0.44 to 2.49)0.91Myocardial infarction4.15 (0.48 to 35.6)0.196.61 (0.61 to 71.56)0.121.55 (0.85 to 2.84)0.151.44 (0.76 to 2.74)0.27Stroke3.22 (0.91 to 11.4)0.073.03 (0.82 to 11.25)0.100.83 (0.40 to 1.75)0.620.84 (0.39 to 1.83)0.67Ischemia-driven revascularization1.57 (0.63 to 3.89)0.331.62 (0.62 to 4.23)0.331.32 (0.94 to 1.85)0.111.27 (0.89 to 1.82)0.19Hospitalization for unstable angina1.37 (0.65 to 2.87)0.411.42 (0.65 to 3.11)0.381.25 (0.97 to 1.61)0.091.21 (0.92 to 1.58)0.17Hospitalization for heart failure1.16 (0.19 to 6.98)0.870.96 (0.14 to 6.80)0.970.95 (0.36 to 2.53)0.920.81 (0.29 to 2.26)0.69Composite of major cardiovascular eventsb2.73 (1.17 to 6.38)0.0202.84 (1.17 to 6.93)0.0221.29 (0.86 to 1.95)0.231.19 (0.77 to 1.84)0.44All-cause death0.67 (0.22 to 1.99)0.470.67 (0.21 to 2.17)0.501.01 (0.51 to 1.99)0.990.86 (0.42 to 1.75)0.68All repeat revascularization1.65 (0.74 to 3.65)0.221.87 (0.81 to 4.34)0.141.18 (0.89 to 1.55)0.261.11 (0.83 to 1.49)0.49MACCE, major adverse cardiovascular and cerebrovascular event including cardiovascular death, MI, stroke, ischemia-driven revascularization, or hospitalization for unstable angina or heart failure. OSA obstructive sleep apneaaModel adjusted for age, sex, BMI, smoker, hypertension, diabetes mellitus and dyslipidemiabComposite end point of major cardiovascular events included cardiovascular death, myocardial infarction, and stroke Among patients with prior stroke, Kaplan–Meier analysis showed that the cumulative incidence of composite events of cardiovascular death, MI, or stroke was significantly higher in the OSA group than in the non-OSA group (log-rank, $$P \leq 0.015$$; Fig. 2C) but not among patients with non-prior stroke (log-rank, $$P \leq 0.23$$; Fig. 2D). The fully adjusted multivariable Cox regression model analysis showed that the association between OSA and the outcomes remained statistically significant in the prior stroke group (HR: 2.84, $95\%$ CI: 1.17–6.93, $$P \leq 0.022$$) but not in the non-prior stroke group (HR: 1.19, $95\%$ CI: 0.78–1.84, $$P \leq 0.44$$). The other endpoints were not different and are listed in Table 3. The crude numbers of events are listed in Additional file 1: Table S2. We also performed additional subgroup analyses according to prior coronary artery disease, prior myocardial infarction, and prior cardiovascular diseases (myocardial infarction, history of revascularization, heart failure, or atrial fibrillation/flutter). Although differences were found between some subgroups, the association of OSA with MACCE was not modified by these confounding factors (interaction $P \leq 0.14$ for all) (Additional file 1: Table S3). Then, we calculated the MACCE rate between CPAP and non-CPAP groups in prior stroke ($33.3\%$ vs $33.9\%$, $$P \leq 0.99$$) and non-prior stroke groups ($19.4\%$ vs $20.8\%$, $$P \leq 0.99$$) and found no significant differences in both groups. ## Discussion To our knowledge, this is the first prospective study evaluating the prognostic value of OSA in ACS patients with or without prior stroke. After adjustment for potential confounders, compared with patients without OSA, patients with OSA had a 2.0-fold higher risk of MACCE in the prior stroke group, but not in the non-prior stroke group, although no significant interaction was noted between prior stroke and OSA for MACCE (interaction $$P \leq 0.17$$). The incidence of composite events of cardiovascular death, MI, or stroke was also significantly higher in the OSA versus non-OSA group among patients with prior stroke but not among those without prior stroke. Given that ACS and stroke have similar pathogeneses, such as atherosclerosis and thrombosis, it is not surprising that prior stroke plays an important role in determining ACS outcome [14, 27]. Studies have indicated that the percentage of patients with ACS and a prior stroke is approximately $10\%$ [28]. Our findings are consistent with recent data showing that $10.74\%$ of patients had a history of prior stroke, and previous stroke patients tended to be older, were more likely to be female, and were more likely to have diabetes, hypertension, and hyperlipidemia. Recently, a large study revealed that patients with prior stroke who undergo PCI have an increased risk of long-term cardiovascular and cerebrovascular complications, specifically recurrent strokes [13]. Similarly, we also found that prior stroke significantly predicted subsequent cardiovascular events in ACS patients. Atherosclerosis, comorbidities, cardiovascular risk factors, and low use of medical and invasive therapy might contribute to poor outcomes in patients with a history of stroke [15, 29]. Therefore, given the high incidence of prior stroke among ACS patients and its possible effect on prognosis, greater attention should be given to ACS patients with prior stroke. Current evidence indicates that OSA is closely related to poor outcomes in patients with ACS or cerebrovascular disease [1, 4, 5]. Our previous studies also showed that OSA was closely related to poor cardiovascular outcomes in ACS onset, especially regarding diabetes status [3, 20]. Emerging evidence has also demonstrated a close relationship between OSA and stroke. OSA is more prevalent following incident ischemic strokes since insults to the central nervous system result in changes in breathing patterns or could mask previously undiagnosed pre-stroke OSA in the poststroke period [30]. Moreover, several studies have demonstrated that OSA in poststroke patients is associated with an increased risk of recurrent ischemic stroke and worse outcomes [17, 18, 31]. Patients with untreated severe OSA are twice as likely to suffer an incident stroke, and this risk is especially relevant to younger and middle-aged patients, without a difference between men and women [32, 33]. In terms of mechanism, OSA may initiate and worsen atherosclerosis via the activation of oxidative stress, inflammation, the sympathetic nervous system, and metabolic abnormalities, eventually leading to high morbidity and mortality in cerebrovascular disease [34–36]. The ESADA Cohort demonstrated the role of OSA-related hypoxia in the risk of developing cardioembolic complications such as stroke [37]. Thus, OSA and stroke patients with ACS must be given extra attention, as they may exhibit a synergistic deleterious effect that increases future cardiovascular risk. The benefits of CPAP therapy are well recognized: it eliminates obstructive events during sleep and substantially improves the consequences, especially daytime sleepiness, neurocognitive deficits and driving performance. However, in randomized controlled trials, the use of CPAP was not associated with a reduction in cardiovascular outcomes among patients with ACS [7–10]. Given that the phenotype of patients is not homogeneous, the deleterious effects of OSA could be different depending on the specific subgroups of ACS patients [38]. Therefore, it is important to focus on identifying specific subgroups of patients with ACS and reevaluate the effect of OSA treatment on cardiovascular diseases [39]. In particular, there is evidence suggesting that CPAP may improve sleepiness, neurological recovery, and depressive symptoms post-stroke in stroke survivors with OSA [19]. In our study, patients with OSA and prior stroke were at the increased risk of incurring a MACCE, therefore representing a high-risk subset most likely to respond to the intervention. There is no significant interaction between prior stroke and OSA for the combined or individual cardiovascular events, possibly due to the lack of power as a result of the small proportion of prior stroke group in this cohort and small number of events in a relatively short follow-up duration. This finding invites us to consider the possibility that a deleterious OSA effect, which was not observed in the entire population that suffered an ACS, exists in this specific phenotype. In contrast, the ancillary study of the ISAACC study showed that OSA was associated with an increased risk of recurrent cardiovascular events in patients without previous heart disease [23]. The variability of results might be partly explained by racial differences and suggests potential heterogeneity of ACS phenotype. We recruited predominantly East Asian patients, which cannot be generalized to patients with other ethnic or racial backgrounds. Noteworthy, patients in our study had more traditional risk factors and more severe daytime sleepiness than those in the ISAACC study [23]. Additionally, patients with prior stroke represents a high-risk subgroup with more female and more than $30\%$ higher prevalence of all 3 traditional risk factors (hypertension, diabetes, hyperlipidemia), and had higher long-term events rate than those without prior stroke. Coexistence of these factors with OSA may generate synergistic effects and promote progression of lesions, thus increasing ischemic events in the long run [20, 40]. The premise of precision medicine is to use a variety of tools to differentiate an individual patient from other patients with similar clinical presentations and thus tailor treatments to that patient's particular needs. Hence, patients with ACS who previously had a stroke should be screened for OSA, and interventions may be necessary. Moreover, more OSA trials should be performed in this subgroup to determine the effects of OSA treatment. ## Limitations This study has several potential limitations. First, the prior stroke cohort included 207 patients, which may diminish the generalizability of the prognostic value. Second, the diagnosis of OSA based on portable sleep monitors may underestimate the severity of OSA. However, studies have shown that portable polygraphy can be used as an alternative to polysomnography for OSA diagnosis [41]. Third, patients self-reported their prior stroke history, which could result in some bias, and we could not confirm whether the strokes were hemorrhagic or ischemic. Therefore, the patients in our study received professional assistance to obtain admission information and conduct grouping, minimizing this bias. Finally, this study recruited predominantly East Asian patients, so it cannot be generalized to patients with other ethnic or racial backgrounds. ## Conclusions Among ACS patients, the presence of OSA was associated with an increased risk of cardiovascular events in patients with prior stroke. Further trials exploring the efficacy of OSA treatment in high-risk patients with ACS and prior stroke are warranted. ## Supplementary Information Additional file 1: Table S1. Baseline patient characteristics. Table S2. Crude Number of all Events by Prior Stroke Categories. Table S3. Association of OSA with risk of MACCE according to subgroups. Fig. S1. Kaplan-Meier curves for the analysis of cardiovascular events in prior stroke versus non-prior stroke group. Fig. S2. Kaplan-Meier curves for the analysis of cardiovascular events according to obstructive sleep apnea severity in prior stroke (A) versus non-prior stroke group (B). ## References 1. Lee C-H, Sethi R, Li R, Ho H-H, Hein T, Jim M-H, Loo G, Koo C-Y, Gao X-F, Chandra S. Obstructive sleep apnea and cardiovascular events after percutaneous coronary intervention. Circulation (2016) 133 2008-2017. DOI: 10.1161/CIRCULATIONAHA.115.019392 2. Johnson KG, Johnson DC. Frequency of sleep apnea in stroke and TIA patients: a meta-analysis. J Clin Sleep Med (2010) 6 131-137. DOI: 10.5664/jcsm.27760 3. Fan J, Wang X, Ma X, Somers VK, Nie S, Wei Y. Association of obstructive sleep apnea with cardiovascular outcomes in patients with acute coronary syndrome. J Am Heart Assoc (2019) 8 e010826. DOI: 10.1161/JAHA.118.010826 4. Yeghiazarians Y, Jneid H, Tietjens JR, Redline S, Brown DL, El-Sherif N, Mehra R, Bozkurt B, Ndumele CE, Somers VK. Obstructive sleep apnea and cardiovascular disease: a scientific statement from the American heart association. Circulation (2021) 144 e56-e67. DOI: 10.1161/CIR.0000000000000988 5. Drager LF, McEvoy RD, Barbe F, Lorenzi-Filho G, Redline S. Sleep apnea and cardiovascular disease: lessons from recent trials and need for team science. Circulation (2017) 136 1840-1850. DOI: 10.1161/CIRCULATIONAHA.117.029400 6. Martínez-García MA, Campos-Rodríguez F, Soler-Cataluña JJ, Catalán-Serra P, Román-Sánchez P, Montserrat JM. Increased incidence of nonfatal cardiovascular events in stroke patients with sleep apnoea: effect of CPAP treatment. Eur Respir J (2012) 39 906-912. DOI: 10.1183/09031936.00011311 7. Yu J, Zhou Z, McEvoy RD, Anderson CS, Rodgers A, Perkovic V, Neal B. Association of positive airway pressure with cardiovascular events and death in adults with sleep apnea: a systematic review and meta-analysis. JAMA (2017) 318 156-166. DOI: 10.1001/jama.2017.7967 8. Sánchez-de-la-Torre M, Sánchez-de-la-Torre A, Bertran S, Abad J, Duran-Cantolla J, Cabriada V, Mediano O, Masdeu MJ, Alonso ML, Masa JF. Effect of obstructive sleep apnoea and its treatment with continuous positive airway pressure on the prevalence of cardiovascular events in patients with acute coronary syndrome (ISAACC study): a randomised controlled trial. Lancet Respir Med (2020) 8 359-367. DOI: 10.1016/S2213-2600(19)30271-1 9. McEvoy RD, Antic NA, Heeley E, Luo Y, Ou Q, Zhang X, Mediano O, Chen R, Drager LF, Liu Z. CPAP for prevention of cardiovascular events in obstructive sleep apnea. N Engl J Med (2016) 375 919-931. DOI: 10.1056/NEJMoa1606599 10. Peker Y, Glantz H, Eulenburg C, Wegscheider K, Herlitz J, Thunström E. Effect of positive airway pressure on cardiovascular outcomes in coronary artery disease patients with nonsleepy obstructive sleep apnea. the RICCADSA randomized controlled trial. Am J Respir Crit Care Med (2016) 194 613-620. DOI: 10.1164/rccm.201601-0088OC 11. Mukherjee D, Eagle KA, Kline-Rogers E, Feldman LJ, Juliard J-M, Agnelli G, Budaj A, Avezum A, Allegrone J, FitzGerald G, Steg PG. Impact of prior peripheral arterial disease and stroke on outcomes of acute coronary syndromes and effect of evidence-based therapies (from the global registry of acute coronary events). Am J Cardiol (2007) 100 1-6. DOI: 10.1016/j.amjcard.2007.02.046 12. Ducrocq G, Amarenco P, Labreuche J, Alberts MJ, Mas J-L, Ohman EM, Goto S, Lavallée P, Bhatt DL, Steg PG. A history of stroke/transient ischemic attack indicates high risks of cardiovascular event and hemorrhagic stroke in patients with coronary artery disease. Circulation (2013) 127 730-738. DOI: 10.1161/CIRCULATIONAHA.112.141572 13. Xu J-J, Jia S-d, Zhu P, Jiang L, Jiang P, Song Y, Zhao X-Y, Li J-X, Chen J, Yang Y-J. Does prior stroke predict long-term recurrent stroke after percutaneous coronary intervention? five-year results from a large cohort study. Front Neurol (2021) 12 740136. DOI: 10.3389/fneur.2021.740136 14. Zhang D, Song X, Chen Y, Raposeiras-Roubín S, Abu-Assi E, Henriques JPS, D'Ascenzo F, Saucedo J, González-Juanatey JR, Wilton SB. Outcome of patients with prior stroke/transient ischemic attack and acute coronary syndromes. Angiology (2020) 71 324-332. DOI: 10.1177/0003319719889524 15. Sposato LA, Hilz MJ, Aspberg S, Murthy SB, Bahit MC, Hsieh C-Y, Sheppard MN, Scheitz JF. Post-stroke cardiovascular complications and neurogenic cardiac injury: JACC state-of-the-art review. J Am Coll Cardiol (2020) 76 2768-2785. DOI: 10.1016/j.jacc.2020.10.009 16. Yaggi HK, Concato J, Kernan WN, Lichtman JH, Brass LM, Mohsenin V. Obstructive sleep apnea as a risk factor for stroke and death. N Engl J Med (2005) 353 2034-2041. DOI: 10.1056/NEJMoa043104 17. Parra O, Sánchez-Armengol A, Bonnin M, Arboix A, Campos-Rodríguez F, Pérez-Ronchel J, Durán-Cantolla J, de la Torre G, González Marcos JR, de la Peña M. Early treatment of obstructive apnoea and stroke outcome: a randomised controlled trial. Eur Respir J (2011) 37 1128-1136. DOI: 10.1183/09031936.00034410 18. Brown DL, Shafie-Khorassani F, Kim S, Chervin RD, Case E, Morgenstern LB, Yadollahi A, Tower S, Lisabeth LD. Sleep-disordered breathing is associated with recurrent ischemic stroke. Stroke (2019) 50 571-576. DOI: 10.1161/STROKEAHA.118.023807 19. Bassetti CLA, Randerath W, Vignatelli L, Ferini-Strambi L, Brill A-K, Bonsignore MR, Grote L, Jennum P, Leys D, Minnerup J. EAN/ERS/ESO/ESRS statement on the impact of sleep disorders on risk and outcome of stroke. Eur Respir J (2020). DOI: 10.1183/13993003.01104-2019 20. Wang X, Fan J, Du Y, Ma C, Ma X, Nie S, Wei Y. Clinical significance of obstructive sleep apnea in patients with acute coronary syndrome in relation to diabetes status. BMJ Open Diabetes Res Care (2019) 7 e000737. DOI: 10.1136/bmjdrc-2019-000737 21. Xie J, Sert Kuniyoshi FH, Covassin N, Singh P, Gami AS, Wang S, Chahal CAA, Wei Y, Somers VK. Nocturnal hypoxemia due to obstructive sleep apnea is an independent predictor of poor prognosis after myocardial infarction. J Am Heart Assoc (2016). DOI: 10.1161/JAHA.115.003162 22. Kuniyoshi FHS, Garcia-Touchard A, Gami AS, Romero-Corral A, van der Walt C, Pusalavidyasagar S, Kara T, Caples SM, Pressman GS, Vasquez EC. Day-night variation of acute myocardial infarction in obstructive sleep apnea. J Am Coll Cardiol (2008) 52 343-346. DOI: 10.1016/j.jacc.2008.04.027 23. Zapater A, Sánchez-de-la-Torre M, Benítez ID, Targa A, Bertran S, Torres G, Aldomà A, De Batlle J, Abad J, Duran-Cantolla J. The effect of sleep apnea on cardiovascular events in different acute coronary syndrome phenotypes. Am J Respir Crit Care Med (2020) 202 1698-1706. DOI: 10.1164/rccm.202004-1127OC 24. Price S, Katz J, Kaufmann CC, Huber K. The year in cardiovascular medicine 2021: acute cardiovascular care and ischaemic heart disease. Eur Heart J (2022) 43 800-806. DOI: 10.1093/eurheartj/ehab908 25. Escaned J, Jaffer FA, Mehilli J, Mehran R. The year in cardiovascular medicine 2021: interventional cardiology. Eur Heart J (2022) 43 377-386. DOI: 10.1093/eurheartj/ehab884 26. Hicks KA, Tcheng JE, Bozkurt B, Chaitman BR, Cutlip DE, Farb A, Fonarow GC, Jacobs JP, Jaff MR, Lichtman JH. 2014 ACC/AHA key data elements and definitions for cardiovascular endpoint events in clinical trials: a report of the American college of cardiology/American heart association task force on clinical data standards (writing committee to develop cardiovascular endpoints data standards). J Am Coll Cardiol (2015) 66 403-469. DOI: 10.1016/j.jacc.2014.12.018 27. Cooper HA, Domanski MJ, Rosenberg Y, Norman J, Scott JH, Assmann SF, McKinlay SM, Hochman JS, Antman EM. Acute ST-segment elevation myocardial infarction and prior stroke: an analysis from the magnesium in coronaries (MAGIC) trial. Am Heart J (2004) 148 1012-1019. DOI: 10.1016/j.ahj.2004.02.017 28. Colantonio LD, Gamboa CM, Kleindorfer DO, Carson AP, Howard VJ, Muntner P, Cushman M, Howard G, Safford MM. Stroke symptoms and risk for incident coronary heart disease in the REasons for Geographic and racial differences in stroke (REGARDS) study. Int J Cardiol (2016) 220 122-128. DOI: 10.1016/j.ijcard.2016.06.030 29. Sposato LA, Fridman S, Whitehead SN, Lopes RD. Linking stroke-induced heart injury and neurogenic atrial fibrillation: a hypothesis to be proven. J Electrocardiol (2018). DOI: 10.1016/j.jelectrocard.2018.02.006 30. Koo DL, Nam H, Thomas RJ, Yun C-H. Sleep disturbances as a risk factor for stroke. J Stroke (2018) 20 12-32. DOI: 10.5853/jos.2017.02887 31. Javaheri S, Peker Y, Yaggi HK, Bassetti CLA. Obstructive sleep apnea and stroke: the mechanisms, the randomized trials, and the road ahead. Sleep Med Rev (2022) 61 101568. DOI: 10.1016/j.smrv.2021.101568 32. Chang C-C, Chuang H-C, Lin C-L, Sung F-C, Chang Y-J, Hsu CY, Chiang L-L. High incidence of stroke in young women with sleep apnea syndrome. Sleep Med (2014) 15 410-414. DOI: 10.1016/j.sleep.2013.12.011 33. Lamberts M, Nielsen OW, Lip GYH, Ruwald MH, Christiansen CB, Kristensen SL, Torp-Pedersen C, Hansen ML, Gislason GH. Cardiovascular risk in patients with sleep apnoea with or without continuous positive airway pressure therapy: follow-up of 4.5 million danish adults. J Int Med (2014) 276 659-666. DOI: 10.1111/joim.12302 34. Javaheri S, Barbe F, Campos-Rodriguez F, Dempsey JA, Khayat R, Javaheri S, Malhotra A, Martinez-Garcia MA, Mehra R, Pack AI. Sleep apnea: types, mechanisms, and clinical cardiovascular consequences. J Am Coll Cardiol (2017) 69 841-858. DOI: 10.1016/j.jacc.2016.11.069 35. Cowie MR. Sleep apnea: state of the art. Trends Cardiovasc Med (2017) 27 280-289. DOI: 10.1016/j.tcm.2016.12.005 36. Liu B, Li Y, Du J, She Q, Deng S. Epicardial adipose tissue in patients with obstructive sleep apnea: a systematic review and meta-analysis. Cardiovasc Innov Appl (2020) 000 81-88 37. Pengo MF, Faini A, Grote L, Ludka O, Joppa P, Pataka A, Dogas Z, Mihaicuta S, Hein H, Anttalainen U. Impact of sleep apnea on cardioembolic risk in patients with atrial fibrillation: data from the ESADA cohort. Stroke (2021) 52 712-715. DOI: 10.1161/STROKEAHA.120.030285 38. Martinez-Garcia MA, Campos-Rodriguez F, Gozal D. Obstructive sleep apnoea in acute coronary syndrome. Lancet Respir Med (2020) 8 e15. DOI: 10.1016/S2213-2600(20)30040-0 39. Peker Y. Obstructive sleep apnoea in acute coronary syndrome. Lancet Respir Med (2020) 8 e14. DOI: 10.1016/S2213-2600(20)30042-4 40. Wang X, Fan J, Guo R, Hao W, Gong W, Yan Y, Zheng W, Ai H, Que B, Hu D. Association of OSA with cardiovascular events in women and men with acute coronary syndrome. Eur Respir J (2022). DOI: 10.1183/13993003.01110-2022 41. Corral J, Sánchez-Quiroga M-Á, Carmona-Bernal C, Sánchez-Armengol Á, de la Torre AS, Durán-Cantolla J, Egea CJ, Salord N, Monasterio C, Terán J. Conventional Polysomnography Is Not Necessary for the Management of Most Patients with Suspected Obstructive Sleep Apnea. Noninferiority, Randomized Controlled Trial. Am J Respir Crit Care Med (2017) 196 1181-1190. DOI: 10.1164/rccm.201612-2497OC
--- title: 'Compliance of public health facilities with essential medicines and health supplies redistribution guidelines in Mbale district, Eastern Uganda: a mixed-methods study' authors: - Immaculate Kyalisiima - Freddy Eric Kitutu - Linda Gibson - Immaculate Akaso - Amos Ndaabe - Herbert Bush Aguma - David Musoke - Richard Odoi Adome - Paul Kutyabami journal: Journal of Pharmaceutical Policy and Practice year: 2023 pmcid: PMC9976424 doi: 10.1186/s40545-023-00545-0 license: CC BY 4.0 --- # Compliance of public health facilities with essential medicines and health supplies redistribution guidelines in Mbale district, Eastern Uganda: a mixed-methods study ## Abstract ### Introduction Redistribution of essential medicines and health supplies (EMHS) is a mechanism to address supply chain uncertainty by moving excess stock of health commodities from health facilities that are overstocked to health facilities with shortages, where it is most needed. It prevents the wastage of scarce resources and improves efficiency within a health supply chain system. Many public health facilities in Uganda experience stock-outs, overstocking, and expiry of essential medicines. This study assessed the compliance of public health facilities with the Uganda Ministry of Health redistribution strategy for EMHS in Mbale district, Eastern Uganda. ### Methods A mixed-methods study was conducted among 55 respondents at public health facility level and five key informants at the district level. Audio-recorded data were transcribed and coded to develop themes. Thematic analysis was performed using ATLAS.ti Version 8.5. Quantitative data were analysed using IBM SPSS Version 24.0. ### Results About a third ($33\%$) of the surveyed health facilities complied with EMHS redistribution guidelines. Respondents agreed that EMHS redistribution had helped reduce health commodity expiries and stock-outs in health facilities. Respondents who did not know about the timely release of funds for redistribution were $68\%$ less likely to comply, and those who said the guidelines were never shared were $88\%$ less likely to comply with the guidelines. ### Conclusions Compliance with the EMHS redistribution guidelines was low and associated with failure to share the guidelines with staff and inadequate awareness about release funds for EMHS redistribution. The district local government should allocate more funds to the EMHS redistribution. ## Background Redistribution of Essential Medicines and Health Supplies (EMHS) is an essential mechanism in supply chains that allows public health facilities to move unused stock of health commodities that has not expired to where it is required [1]. This mechanism is essential for public health facilities in Uganda because of the inequity in EMHS allocations by the National Medical Stores (NMS) [2]. NMS is the government's central medical store mandated to procure, warehouse and distribute EMHS to all public health facilities. EMHS are procured using government funds from taxes and donors. Public health facilities are classified into seven levels based on the catchment population and services offered, namely: national referral hospitals, regional referral hospitals, referral hospitals, general hospitals, health centres (HCs) level IV, HCs level III, and HCs level II [3]. Higher level health facilities, HCs level IV and Hospitals order EMHS from the central medical store, NMS, every 2 months based on their allocated budgets. Lower level health facilities, HCs, level II and level III are allocated uniform kits per district [4]. Pharmaceutical kits contain selected EMHS in predetermined quantities for supply to lower level health facilities [5, 6]. Uganda is a low-income country with a GDP of $45.557 billion [7]. According to WHO, in 2020, the country spent $3.96\%$ of its GDP on health [8]. Data on total pharmaceutical expenditure is not readily available [9]. The proportion of health expenditure allocated to medicines was estimated to be about $30\%$. Health expenditure is majorly from the government, with donors providing the largest share of funding. In addition, out-of-pocket payments and private insurance are significant sources of health care funding [10]. The annual budget for medicines in the public sector in FY $\frac{2020}{2021}$ was UGX446.4 billion, representing about $15\%$ of the health sector budget. The health budget as a proportion of the national budget was only $6.1\%$, falling short of the $15\%$ recommended by the Abuja Declaration [11]. The annual budget for EMHS is far below the estimated national need for medicines in public health facilities, where most patients seek medical care [12]. Moreover, facilities at the same level are allocated the same budgets, yet their patient loads differ widely [2, 13]. This results in varied availability of EMHS characterized by excess stock and stock-outs in health facilities [10]. Redistribution of EHMS promotes optimal use of available health commodities by providing a platform to move them from health facilities they are less needed or overstocked to health facilities in greater need with under-stocking. As such, it prevents EMHS shortages and mitigates wastage in health facilities. These EMHS are an integral part of health service delivery and should always be available if the Ministry of Health’s (MOH) targets for service delivery are to be met. However, many public health facilities in Uganda continue to report stockouts, expiries, and only about $40.6\%$ achievement of set targets in managing endemic diseases, such as malaria [4, 11, 14, 15]. The recent annual health sector performance report showed that only $43\%$ of the health facilities had $95\%$ availability of a basket of commodities over a 3-month period which is below the target of $75\%$ of the facilities [11]. It is also not uncommon to find health facilities overstocked with some medicines [2]. All this indicates sub-optimal health supply chain system performance and that redistribution of EMHS is not being done appropriately in these facilities to mitigate frequent stock-outs [9]. Maldistribution is attributed to weak supply chains resulting from defective inventory practices, such as poor stock monitoring, deficiency of knowledge of basic expiry prevention tools, low participation of clinicians, profit- and incentive-based quantification, and irrational use of medicines [16, 17]. The low availability of EMHS undermines the efforts to meet the targets of the MOH and the World Health Organization for equity and efficiency [16, 18]. Reduced availability results in reduced access to health care, increased out-of-pocket expenditure as patients must look for alternatives, medical errors, non-adherence to treatment, drug resistance, and increased morbidity and mortality [17, 19]. Supply chains of EMHS, therefore, need to be strengthened as they come with a lot of uncertainty and interruptions impacting performance and efficiency of the health system. Mechanisms, such as redistribution, when effectively implemented, optimise the use of available resources and improve the efficiency with which health systems work. In countries, such as the United States of America, redistribution of medicines has been utilized to reduce medicines wastage and leads to great cost avoidance [20]. Redistribution is important in bridging gaps in service delivery. However, redistribution can be challenging as it requires proper stock management and collaborative action among decentralised actors of the health supply chain to ensure efficient delivery of EMHS [21]. To streamline and harmonise redistribution, the MOH developed the National Redistribution Strategy for the Prevention of Expiries and Handling of Expired Medicines and Health Supplies (EMHS) in 2012 and revised it in 2018 [1, 22]. The strategy was to avoid EMHS wastage through reported paradoxes of excess stock in one facility, while another has a stock out within the same district. The stock of a specific medicine is considered to be excess when the months of stock are more than four on any specific day. However, on numerous occasions, health workers have been noted as reluctant to follow the guidelines to redistribute [1]. An unpublished report of a survey done in Uganda found compliance with the 2012 redistribution guidelines to be $39\%$ [23]. The survey, however, did not include any of the districts in the Bugisu sub-region. Mbale district is the regional capital of Bugisu and has experienced a decline in health services performance on the district league table of MOH. Therefore, this study assessed the compliance of health facilities within Mbale district with the redistribution guidelines and explored factors associated with non-compliance to the redistribution guidelines. The study findings are instrumental in identifying gaps contributing to compliance with the EMHS redistribution guidelines, which is critical for decision-making. ## Study design and setting We used a mixed-methods cross-sectional study design. Quantitative data was used to determine compliance and identify-associated factors. Qualitative data were collected simultaneously to supplement the factors associated with compliance. The data collected were from primary care facilities and administrative staff at district level in Mbale from June to August 2020. Mbale district is one of the districts found in the mid-eastern region of Eastern Uganda. It is divided into three constituencies: two counties (Bungokho south and north) and Mbale municipality, with its largest city being Mbale city. Mbale district has 36 public health facilities, of which 35 are primary care facilities, and one is a regional referral hospital that serves the eastern region [3]. The district had an estimated population projection of 586,300 people in 2020 [24]. The district was chosen for the study because of its huge decline in health sector performance from being in the top 10 districts in $\frac{2016}{2017}$ to 34th and 45th in $\frac{2017}{2018}$ and $\frac{2019}{2020}$, respectively [25–27]. ## Study population Our study units comprised primary care health facilities in Mbale district from which respondents were purposively selected. We included in the study all primary care facilities funded by the government. The selected facilities were from health centre levels II, III, and IV. HCs levels I, II, III, and IV are the only facilities in the healthcare system that provide primary care in Uganda [2, 28]. They deliver the first connection between the public and the formal health sector and focus mainly on infectious disease prevention and treatment services. Village Health Teams (VHTs) constitute health centre I and are attached to a nearby health facility [29]. HCs level IV are mandated to serve a target population of 100,000 people and have provisions for an operating theatre, inpatient, and laboratory services, and act as a referral facility for HCs level III in their jurisdiction. HCs level III have a target population of 20,000 people and have provisions for basic laboratory services, maternity care, and inpatient care. HCs level II, on the other hand, are lower level facilities and are mandated to serve a target population of about 5,000 people providing outpatient services and outreach programs only [28]. The health facilities are staffed differently based on their level of care, determined by the MOH. HCs level IV have between 10 to 21 staff, which include; 1 doctor, clinical officers, qualified nurses, midwives, and nursing aides. HCs level III have 10 to 6 staff, including clinical officers, qualified nurses, midwives, and nurse aides. HCs level II, as lower level facilities, have approximately four staff including a qualified nurse, midwife, and nurse aides [28]. Specific cadres are employed depending on the services to be offered across the different facility levels, although currently, staffing levels stand at only $45\%$, $20\%$, and $13\%$ of HCs levels II, III, and IV, respectively [11]. In addition to the staffing constraints, nationally, HCs levels IV, III and II received $2.6\%$, $6.3\%$, and $2.3\%$, respectively, of the National Medical Store (NMS) expenditure in the $\frac{2020}{2021}$ financial year [11]. Ordering of medicines follows a pull system at Hospitals and HCs level IV and a push or kit system at HCs level III and II [4]. The pull system requires each health facility to determine what and how much to order. On the other hand, the push system requires each health facility to receive predefined kits [5]. The health facility in charge approves bi-monthly orders for medicines at each facility before submission. A graduate pharmacist makes orders at the hospital level, dispensers with qualifications of diploma in pharmacy at HCs level IV, and stores in charge at HCs level III and II. Annual facility budgets are divided into six cycles to approximate the value of commodities ordered every cycle. The quantity to order each month is calculated using a formula incorporating the quantity on hand and maximum stock. When the total amount exceeds the cycle budget, vetting commodities is done using VEN classification to prioritize the vital items [30]. ## Study sample size We required a sample size of at least 30 health facilities according to the WHO guidelines on survey of health facilities [31]. The sample size for quantitative data was determined using Yamane’s proportionate method, since the district had a finite population size. This sampling method was used to combine responses into categories and sample size based on proportions [32]. The proportions of each of these facilities were used to determine the exact number of participating facilities. Once the number had been determined, these were then randomly sampled. The sample size was calculated using the formula below based on a finite number of facilities.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = \frac{N}{{1 + N\left(e \right)^{2} }}$$\end{document}n=N1+Ne2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = {{\left({35} \right)} \mathord{\left/ {\vphantom {{\left({35} \right)} {1 + 35\left({0.05} \right)^{2} }}} \right. \kern-0pt} {1 + 35\left({0.05} \right)^{2} }} = 33$$\end{document}$$n = 35$$/1+350.052=33where n is the sample size, N is the population size, and e is the margin of error. Proportions for each level of facility in the study area were (3 HCs level IV ($8.6\%$), 23 HCs level III ($65.7\%$), and 9 HCs level II ($25.7\%$)). Therefore, 3 HCs level IV, 22 HCs level III, and 8 HCs level II were sampled. The sample size of respondents at health facility level was calculated based on the assumption that a facility had one in charge and one store in charge. The total number of respondents was, therefore, 66. The District Health Officer (DHO) and four Medicines Management Supervisors (MMS) were purposively selected for key informant interviews. MMSs are health workers that provide supportive supervision to improve medicines management in public health facilities [33, 34]. ## Variables The outcome variable of the study was compliance with EMHS redistribution guidelines. This was a binary outcome, where a facility would be compliant or non-compliant. A facility was reported compliant if it scored equal to or above an arbitrary cutoff of $80\%$ on the observational checklist. The independent variables were categorical and included constructs of the EMHS redistribution guidelines, including the steps and triggers of redistribution. Others were the factors that affect stock levels and redistribution at health facilities, the availability of funds, bureaucratic processes, knowledge of staff, up-to-date stock cards, availability of EMHS, and communication channels. ## Data collection Three data collection tools were used in the study, including a questionnaire, a key informant guide, and an observational checklist. The tools were distributed among the respondents that had been purposively selected. The tools were administered by the principal investigator together with two research assistants. The assistants had a minimum qualification of a bachelor’s degree and were trained on data collection procedures before commencement of the exercise. A pilot study was conducted in $5\%$ of the calculated sample size before data collection to pre-test the data collection tools. Administrative clearance was obtained from the offices of the Chief Administrative Officer (CAO), the DHO, the town clerk, and the manager responsible for each health facility. On arrival at the facility or district office for the study, the researchers introduced themselves and stated their reason for the visit. The researchers found the selected respondents at their respective workstations between 8:30 am and 4:30 pm. They were given introductory letters, briefed about the study being conducted, given consent forms, and allowed to participate freely without coercion. Only respondents that gave informed consent were allowed to participate in the study. The interviews ran for about 30–45 min each, and these were supplemented with on-site observations using an observational checklist. ## Quantitative data collection Quantitative data was gathered using a questionnaire and checklist. The questionnaire was used to collect data from respondents at the health centres to determine their socio-demographic characteristics and possible factors associated with non-compliance. The questionnaire was administered by research assistants and had both open and closed-ended questions allowing the respondents to explain where necessary. The questions were formulated based on the activities involved in the process of redistribution as stated in the guidelines, and some questions were adapted from a similar questionnaire used to carry out a scoping study about compliance with redistribution guidelines [1, 23]. Questions asked included knowledge of the steps of redistribution and its financing, if the facility had excess stock, if the guidelines were available and if they had ever been trained. Probing for interesting responses and observation of nonverbal responses were also done. The observational checklist supplemented the responses and was guided by triggers, including; A facility has an excess of one EMHS, while another has a deficit, the facility's stock is expected to expire before being used, the facility has EMHS distributed to it in error, especially when a lower level facility received supplies meant for a higher level facility and facility has more EMHS of a short shelf life than what was forecasted for use. While using the checklist, we reviewed copies of the issue and requisition vouchers and stock records. We checked for occurrences of excess stock or deficits, availability of the guidelines, and the use of any communication system to alert departments. A percentage score was computed from the scores among the facilities that experienced triggers of redistribution to determine compliance. ## Selection of tracer commodities We used documents of six tracer items to investigate compliance, including Artemether/Lumefantrine tablets, isoniazid tablets, cotrimoxazole tablets, oxytocin injections, metformin tablets, and rapid diagnostic test kits for malaria. We selected two tracer commodities, Artemether/Lumefantrine and malaria rapid diagnostic test kits because of the high burden of malaria and the large number of people at risk of getting the disease [35, 36]. Isoniazid and cotrimoxazole were selected for their use among HIV patients for preventive treatment against tuberculosis and as prophylaxis against opportunistic infections. We also selected metformin because of the increasing burden of Diabetes Mellitus and oxytocin because of its importance in managing obstetric conditions [35, 37]. ## Qualitative data collection We used a key informant interview guide to conduct interviews with the DHO and the MMSs, who had the most information about the overall redistribution process in the district. These interviews increased our understanding of the findings as they had extended probing. Open-ended questions were used during the interviews, and the sessions were recorded using a mini digital voice recorder to reduce the likelihood of omitting relevant information. ## Data quality control and management To ensure the quality of data collected was assured, the tools used were pre-tested on two health facilities in a similar setting before their use in the field. The tools were then revised based on the feedback to ensure that these would be comprehended and appropriate responses obtained from the respondents. The research assistants were also trained before data collection to ensure that all relevant information was captured. They were trained on the study objectives, how to use the audio recorders and the ethics of working when interacting with the respondents. Completeness of data was ensured by checking filled tools in real time to identify any missing data. ## Data analysis Quantitative data was transferred from a Microsoft Excel 2016 spreadsheet (Microsoft Corporation, Washington, USA) and entered into Epi-Info version 3.5.1(CDC, Atlanta, Georgia). to be checked for consistency. Cleaned data was then exported to IBM SPSS Version 24.0 for analysis. Descriptive statistics were determined for the population and respondents from the univariate analysis. Test for association was done using logistics regression through multivariate analysis. Associations with the diverse factors were determined using odds ratios, p values, and a confidence interval of $95\%$. Qualitative data were analysed using ATLAS.ti version 24.0(Scientific Software Development GmbH, Berlin, Germany) Thematic analysis was utilized. Qualitative data collected from recordings of interviews were transcribed, and a codebook was created for the different variables. The coded qualitative data were then categorized and grouped into themes for analysis. Continuous theme searching and reviewing were done until no new codes were observed from the scripts. ## Background characteristics We visited a total of 33 public health facilities in Mbale district and had a response rate of $98\%$. We were only able to meet 55 respondents instead of the intended 66 respondents at the health facilities, because some health facility in-charges also served the role of the store in-charges. In addition, one store in-charge was on maternity leave and, therefore, could not be interviewed (Table 1).Table 1Background information of the respondents at the selected health facilitiesCharacteristicsFrequency (n)Percentage (%)Level of the health facility Health centre IV59.1 Health centre III3869.1 Health centre II1221.8Professional qualifications of the respondent Nursing cadre1934.5 Clinical Officer2545.5 Medical officer814.5 Other35.5Sex of the respondent Male2443.6 Female3156.4Years of services 5–101018.2 10–152749.1 > 151832.7Position of the respondent at the health facility Facility in-charge4174.5 Stores in-charge1125.5 The majority ($69.1\%$, $$n = 38$$) of respondents were from health centre III, and about half of the respondents ($45.5\%$, $$n = 25$$) were clinical officers by cadre. More than half ($56.4\%$, $$n = 31$$) of the respondents were female, and nearly half ($49.1\%$, $$n = 27$$) had been in services for 10–15 years. Most ($74.5\%$, $$n = 41$$) of the respondents were health facility in-charges. In addition, we interviewed five key informants, four of whom were male, three had worked for more than 15 years, and three were clinical officers by cadre. ## Compliance with the EMHS redistribution guidelines The majority ($67.3\%$, $$n = 37$$) of respondents noted that the health facility did not have guidelines on redistribution available, and only ($30.9\%$, $$n = 17$$) said the guidelines being used were up-to-date. Only ($29.1\%$, $$n = 16$$) respondents agreed that guidelines were always used, and the information was verified with copies of vouchers and stock cards filled during redistribution at the facility. In addition, these respondents also confirmed that guidelines were always used in case there was a trigger (excess stock) to redistribute. The respondents’ information was verified by reviewing signed copies of the health facility's HMIS 017 vouchers and stock records. Compliance of the health facilities was determined from the percentage scores of each facility that had excess stock. About one-third of health facilities ($33.3\%$) were found compliant. ## Triggers of redistribution Key informants agreed with the respondents at the facility that redistribution took place only when necessary. The respondents also noted that communicating excess stock at the health facility was done to the DHO through the health facility in charge. “In case of excess, I call the DHO, they have medicines management supervisors who inform us on what to do, or we take for redistribution. There are forms that we fill in all drugs in excess we take them to the district, and they also send messages using mTrac” HF-05 However, the key informants stated that sometimes the drugs were brought to the district stores when they were about to expire and would not serve the intended purpose. “Sometimes the drugs come when it's late; they calculate and find that some drugs may not be consumed.” KII-05 ## Following the redistribution steps in the guidelines When asked about the steps of redistribution, the responses were mixed as more than half 30 ($54.5\%$) did not know about the existence and the content of the guidelines., However, they knew there was a procedure to follow in cases of excess and deficits to avoid expiry. “The guidelines are not properly disseminated because, at first, they were between facilities to facilities. It was not properly emphasized, and not everyone perceives it seriously if not followed by mTrac.” HF-10 Even though most 39 ($70.9\%$) respondents thought that knowledge of the guidelines was not well-disseminated, all key informants did not seem to agree. "Every facility has a copy of the guidelines, but you know when you ask the staff, and you ask, they claim they don't have, but when you check, you find the books stacked in a corner full of dust” KII-01 All the key informants agreed that the redistribution procedures were clear and that facilities were making an effort. The process was long and required individuals to use their funds.“….the process is generally tedious and might even cost patients their lives.” KII-02“Those steps are manageable……due to the length of the procedure, it makes the guidelines a bit complicated for some people, and at the end of the day, you find in charges saying To Whom It May Concern the drugs are over so let's wait for the next consignment.” KII-01 ## Factors associated with compliance with redistribution guidelines Respondents who did not know if the money required for redistribution was released on time were $67.6\%$ ($$p \leq 0.007$$, $95\%$ CI 0.127–0.685) less likely to comply with the redistribution guidelines than those who said the money required for redistribution was never released on time. Respondents who said the guidelines are sometimes shared with the staff in other departments were $92\%$ ($$p \leq 0.037$$, $95\%$ CI 0.097–0.0264) less likely to comply with the redistribution guidelines than those who said the guidelines are always shared with the staff in other departments. Similarly, respondents who reported that the guidelines were never shared with staff in other departments were $88\%$ ($$p \leq 0.003$$ $95\%$ CI 0.097–0.0264) less likely to comply with the redistribution guidelines than those who said the guidelines were always shared with the staff in other departments (Table 2).Table 2Factors associated with compliance with the EMHS redistribution guidelinesVariablesOR ($95\%$ CI)p valueMoney is required to re-distribute released on time Never1.0 I don’t know0.324 (0.127–0.685)0.007Guidelines were shared with staff in other departments Always1.0 Sometimes0.080 (0.061–0.109)0.037 Never0.120 (0.097–0.0264)0.003 ## Transportation costs and delivery The biggest challenge faced during redistribution was the cost of transporting the medicines and supplies, leading to delays in identification and movement. Funds for redistribution were not readily available. “One of the key constraints is money. Most facilities don't have money when needed. The PHC and RBF money comes, but many activities need the money, and, in most cases, redistribution is not one of those activities.” KII-01“In cases where a facility is far away from the district, staff start grumbling asking why they should put in their money.” KII-01“Financially, there's a big challenge if the partners withdrew, we might have interruptions” KII-05 However, some respondents believed that if people responsible for resources at the facility used them appropriately, they would not have transportation challenges. “Some health facilities do not appreciate what they have. If I run out of stock and I need to get drugs from another facility when all the vouchers are signed, I hire a means of transport to get the drug to me because the facilities always have money” HF-10 ## Bureaucracy involved in the process of redistribution More than half, 29 ($52.7\%$) of the respondents, thought that the required signatures on the notification forms were too many, with some not being stationed at the facilities making the process long. In addition, respondents agreed that the process was too long. “People who sign are many…., sometimes the DHO is very busy to sign, someone can keep on coming until they give up” HF-02 ## Inconsistency in the delivery process Inconsistencies in the delivery of drugs and supplies and failure to inspect items contributed to non-compliance with the EMHS redistribution guidelines. Respondents stated that the NMS and Joint Medical Stores (JMS) continuously delivered what was not requested by the facilities. “The uncertainty of deliveries where someone thinks that if I give out this and this, they are not very sure that the next cycle they will receive that item from NMS so at times they end up not redistributing” KII-03 ## Knowledge gaps among the healthcare workers The respondents believe that the health workers were not well conversant with the stock management and the redistribution guidelines as many had not been trained on these, hindering the process. “Now, some staff do not know what to do with physical counts and do not check for expiry date. Sometimes physical counts are done and even include drugs that are expired” KII-02 The key informants also noted that on top of most facilities not having personnel trained to manage stores, the few who train are constantly transferred to carry out other roles.“… after we have mentored somebody for a month, you come back and find when a person has been transferred. There are frequent transfers, so you retrain.” KII-03"You train today, get an action point, and the person is transferred." KII-02 The key informants also noted that they trained, but the staff did not improve. “We have taken the health workers through the given guidelines, but they don’t want to understand” KII-04 ## Attitude of the health workers and facility person’s in-charge Most of the key informants and some respondents noted that the attitude of the staff towards stock ownership also affected redistribution as some regarded it as personal property, causing them to hold onto the excess items. Others noted that some health facilities did not take the responsibility to carry out proper stock-taking, making it hard to plan. "People forget that these are government drugs and then think that the drugs are the facility's drugs, so there is that rigidity in releasing the excess." KII-03"When you give some in-charges the guidelines, they just keep them in their bags, which creates a problem." KII-04 ## Individual responsibility The respondents also noted that some in-charges did not take the initiative to carry out necessary activities to prevent the accumulation of excess stock and expiries. They cited being overloaded with work as a hindrance to paying close attention to taking appropriate action on redistributable stock in the facility store. "Then appointed people may get overwhelmed with work and not get time to go to the store, and sometimes drugs might expire without their knowledge." KII-03 ## Discussion Our findings showed that compliance with the EMHS redistribution guidelines was at only $33.3\%$ and was associated with failure to disseminate guidelines and the lack of knowledge on the timely release of money for redistribution. This study is important, because no published literature on compliance with redistribution guidelines exists. The study findings are of major importance to districts in devising strategies to address stock-outs, overstocking and expiries of essential medicines in public health facilities. Only a third ($33.3\%$) of the health facilities complied with the EMHS redistribution guidelines. These findings are comparable to those from a nationwide scoping review done in Uganda by Sematiko, where compliance with the guidelines was as low as $39\%$ [23]. However, the findings differ from another study that found that $75.9\%$ of public health facilities in Uganda used redistribution to cope with stock-outs [4]. The difference in findings could be due to a social desirability bias leading to an overestimation of redistribution. Failure to comply with the guidelines undermines efforts by the government to fight expiries, stock-outs, and their undesirable consequences. Indeed, a study by Zakumumpa et al. showed redistribution as a strategy to fight chronic stock-outs among health facilities [14]. The low compliance in the present study implies that even when redistribution triggers are present, such as excess stock, while another has a deficit, facility in-charges and staff still hold back from following the redistribution guidelines. In this study, we found that respondents who reported that the guidelines were never shared with staff in other departments were $88\%$ less likely to comply. Failure to share guidelines limits the knowledge circulation about redistribution among staff. This finding corresponds with a study by Srivastava and Singh to identify the antecedents and consequences of integrated supply chain performance in healthcare systems. The study showed that knowledge sharing among staff impacted the performance of the supply chain [38]. Another study done in Iran on the role of information sharing on supply chains showed that information sharing positively affected the integration, efficiency and performance of drug supply chains [39]. Failure to share guidelines further led to staff knowledge gaps, as over half ($54.5\%$) of the staff did not know about the guidelines, even though key informants stated that they gave out guidelines and trained staff. The knowledge gap affects stock management practices and compliance with the guidelines. This finding is supported by a study done in Malawi to investigate the management of drug supplies for life-threatening diseases, which showed that training in the management of medicines affected their availability [40]. Gaps in knowledge affect the ability of staff to appropriately quantify and make projections when planning, further causing mal-distribution and expiries, as pointed out in a study done in Uganda [41]. In our study, the knowledge gap was further attributed to the constant transfer of staff and the recruitment of staff that are not qualified to carry out the relevant assigned roles in the management of EMHS. We found that this led to staff including expired medicines in the physical counts. In addition, we also found that compliance was associated with staff not knowing if money was released on time. The failure to share information on availability of funds among health facility staff means that few staff knew about the party meant to incur the cost of redistribution. This implies that staff will become more reluctant to redistribute as no money at the facilities is allocated, and they will rely on implementing partners. Previous studies showed that redistribution was expensive and required additional funding to be effected [2, 6]. In the guidelines, no budgetary allocations are made to cater for transportation. A study conducted in Elgeyo Marakwet County, Kenya, to assess factors affecting supply chain efficiency in public health facilities showed that healthcare workers agreed that transport coordination affected their supply distribution [42]. Timely access to transport facilitation for redistribution of medicines is, therefore, essential. We also found that redistribution was bureaucratic to avoid the occurrences of theft and to streamline the process. The lengthy procedures affected redistribution as the respondents felt that the approval process was tedious, discouraging many from initiating it. This finding also agrees with the study by Sematiko, where all the respondents agreed that the long bureaucratic process affected redistribution [23]. A study by Mikkelsen-Lopez et al. to determine the pattern of availability of Artemisinin Combination Therapy (ACT) and possible causes of stock-outs in public health facilities in Tanzania found that the long bureaucratic processes were a major factor in reducing availability of stock [43]. There is a need to minimize the approval process to ease redistribution of excess stock. We also found that inconsistencies in the delivery of EMHS at health facilities also affected compliance with the guidelines and increased practises like hoarding EMHS stock that may be required at another health facility due to the uncertainty created. Some health facilities intentionally did not declare excess stock, because they were unsure if they would get supply in the next cycle from NMS. In a study to understand the problems underlying drug shortages, distribution uncertainties were seen to contribute to more shortages as some facilities stocked items that were disproportionate to their demand [44]. Our findings align with a study done in Malawi to investigate the management of drug supplies for life-threatening diseases. In the study, deficient deliveries from regional medical stores and uneven distribution of drugs among health centres contributed to inadequate EMHS supplies [40]. The NMS should stick to delivery schedules of medicines to reduce uncertainty among health facilities. The study sought to learn from the health workers involved in redistribution of medicines. The study's strength is that it employed mixed methods, which enabled us to understand compliance with redistribution guidelines in-depth. The study also had limitations. First, some of the health workers approached had concerns about being reprimanded because of the information provided. Reassurance of the respondents’ confidentiality was done to mitigate this challenge. Second, during data collection, some health facilities had improperly filled stock cards and the requisition and issue vouchers which affected some of the data to support determination of the level of compliance. Third, some of the assessment of the items for measuring compliance was based on self-reports by the respondents, hence a possibility of social desirability bias. ## Conclusions Compliance with the EMHS redistribution guidelines was low. Several hindrances to compliance with redistribution guidelines were cited. Key hindrances were lack of guidelines, failure to share the guidelines with staff and not knowing about releasing money for redistribution. To improve compliance with redistribution guidelines, funds for transporting medicines should be budgeted for at the district. Health facility in-charges should sensitize staff working in stores about the redistribution process. The DHO should prioritize regular supervision of health facilities with an emphasis on preventing medicine stock-outs and overstocking. ## References 1. 1.MOH. Uganda National Redistribution Strategy for Prevention of Expiry and Handling of Expired Medicines and Health Supplies. Kampala-Uganda: Ministry of Health; 2018. https://www.health.go.ug/download-attachment/8aOxHkltUisF0GBDek7WMdsnlx9VkT4TEaHfdu2uVqU. 2. Kusemererwa D, Alban A, Obua OT, Trap B. An exploratory study on equity in funding allocation for essential medicines and health supplies in Uganda’s public sector. BMC Health Serv Res (2016.0) 16 453. DOI: 10.1186/s12913-016-1698-6 3. 3.MOH. National Health Facility Master Facility List 2018. Kampala, Uganda: Ministry of Health; 2018. http://library.health.go.ug/publications/health-facility-inventory/national-health-facility-master-facility-list-2018. 4. Lugada E, Ochola I, Kirunda A, Sembatya M, Mwebaze S, Olowo M. Health supply chain system in Uganda: assessment of status and of performance of health facilities. J Pharm Policy Pract (2022.0) 15 58. DOI: 10.1186/s40545-022-00452-w 5. 5.MSHMDS-3: managing access to medicines and health technologies2012ArlingtonKumarian Press. MDS-3: managing access to medicines and health technologies (2012.0) 6. 6.MOHAssessment of the essential medicines kit-based supply system in Uganda2011KampalaMinistry of Health. Assessment of the essential medicines kit-based supply system in Uganda (2011.0) 7. 7.UBOS. AGDP Publication Tables. Kampala-Uganda: Uganda Bureau of Statistics; 2022. 8. 8.Health Expenditure Profile Uganda. World Health Organization. https://apps.who.int/nha/database/country_profile/Index/en. 2020. Accessed 28 Dec 2022. 9. Lugada E, Komakech H, Ochola I, Mwebaze S, Olowo Oteba M, Okidi LD. Health supply chain system in Uganda: current issues, structure, performance, and implications for systems strengthening. J Pharm Policy Pract (2022.0) 15 14. DOI: 10.1186/s40545-022-00412-4 10. 10.MOH10-Year Roadmap for Government of Uganda’s Health Supply Chain Self-Reliance 2021/2022–2031/20322021KampalaMinistry of Health. 10-Year Roadmap for Government of Uganda’s Health Supply Chain Self-Reliance 2021/2022–2031/2032 (2021.0) 11. 11.MOHAnnual Health sector performance report 2020/20212021KampalaMinistry of Health. Annual Health sector performance report 2020/2021 (2021.0) 12. Mujasi PN, Puig-Junoy J. Predictors of primary health care pharmaceutical expenditure by districts in Uganda and implications for budget setting and allocation. BMC Health Serv Res (2015.0) 15 334. DOI: 10.1186/s12913-015-1002-1 13. Mujasi PN, Puig-Junoy J. What variables should be considered in allocating Primary health care Pharmaceutical budgets to districts in Uganda?. J Pharm Policy Pract (2015.0) 8 3. DOI: 10.1186/s40545-014-0023-1 14. Zakumumpa H, Kiweewa FM, Khuluza F, Kitutu FE. “The number of clients is increasing but the supplies are reducing”: provider strategies for responding to chronic antiretroviral (ARV) medicines stock-outs in resource-limited settings: a qualitative study from Uganda. BMC Health Serv Res (2019.0) 19 312. DOI: 10.1186/s12913-019-4137-7 15. 15.GHSC-PSMUganda National Supply Chain Assessment Report: capability and performance2018ArlingtonUSAID Global Health Supply Chain Program-Procurement and Supply Management. Uganda National Supply Chain Assessment Report: capability and performance (2018.0) 16. Nabukeera MS. Challenges and barriers to the health service delivery system in Uganda. IOSR J Nurs Health Sci (2016.0) 5 30-38 17. Ventola CL. The drug shortage crisis in the United States: causes, impact, and management strategies. P t (2011.0) 36 740-757. PMID: 22346307 18. 18.WHO. Medicines shortages: global approaches to addressing shortages of essential medicines in health systems. WHO Drug Inf. 2016;30(2):180–5. 19. White KG. UK interventions to control medicines wastage: a critical review. Int J Pharm Pract (2010.0) 18 131-140. DOI: 10.1211/ijpp.18.03.0002 20. Uba RO, Priftanji D, Hill JD. Redistribution of short-dated emergency medications as a cost avoidance and drug shortage management strategy. Am J Health Syst Pharm (2020.0) 77 1504-1509. DOI: 10.1093/ajhp/zxaa200 21. Gupta U, Ramesh A. Analyzing the barriers of health care supply Chain in India: the contribution and interaction of factors. Procedia Soc Behav Sci (2015.0) 189 217-228. DOI: 10.1016/j.sbspro.2015.03.217 22. 22.MOH. Uganda National Redistribution Strategy for Prevention of Expiry and Handling of Expired Medicines and Health Supplies. Kampala-Uganda: Ministry of Health; 2012. http://library.health.go.ug/publications/medicines/uganda-national-redistribution-strategy-prevention-expiry-and-handling. 23. 23.Sematiko GK. A Scoping Study for the Review and Update of the Uganda National Redistribution Strategy for Prevention of Expiry and Handling of Expired Medicines and Health Supplies. 2018. 24. 24.UBOS. Population By Sex For 146 districts Kampala, Uganda: Uganda Bureau of Statistics; 2022. 25. 25.MOH. Annual Health Sector Performance Report 2016/2017. Kampala-Uganda: Ministry of Health; 2017. 26. 26.MOH. Annual Health Sector Performance Report 2019/2020. Kampala-Uganda: Ministry of Health; 2020. 27. 27.MOH. Annual Health Sector Performance Report 2017/18. Kampala, Uganda: Ministry of Health; 2018. http://library.health.go.ug/publications/annual-quarterly-performance-reports/annual-health-sector-performance-report-201718. 28. 28.MOH. Uganda Hospital and Health Centre IV Census Survey. Kampala, Uganda: Ministry of Health; 2014. http://library.health.go.ug/publications/health-infrastructure/uganda-hospital-and-health-centre-iv-census-survey. 29. Dowhaniuk N. Exploring country-wide equitable government health care facility access in Uganda. Int J Equity Health (2021.0) 20 38. DOI: 10.1186/s12939-020-01371-5 30. 30.MOH. Uganda Essential Medicines and Health Supplies Management Manual. In: Medicines PSaN, editor. Kampala-Uganda: Ministry of Health; 2018. 31. 31.WHO. How to investigate drug use in health facilities: selected drug use indicators. Geneva: World Health Organization; 1993. 32. Yamane T. Statistics: an introductory analysis (1973.0) 33. Trap B, Musoke R, Kirunda A, Oteba MO, Embrey M, Ross-Degnan D. Article 2: longitudinal study assessing the one-year effects of supervision performance assessment and recognition strategy (SPARS) to improve medicines management in Uganda health facilities. J Pharm Policy Pract (2018.0) 11 15. DOI: 10.1186/s40545-018-0142-1 34. Kwesiga B, Wagner AK, Seru M, Ross-Degnan D, Trap B. Costs and effectiveness of the supervision, performance assessment and recognition (SPARS) strategy for medicines management in Uganda. J Pharm Policy Pract (2019.0) 12 36. DOI: 10.1186/s40545-019-0196-8 35. 35.MOH. Uganda Clinical Guidelines 2016. Kampala, Uganda: Ministry of Health; 2016. http://library.health.go.ug/publications/guidelines/uganda-clinical-guidelines-2016. 36. Yeka A, Gasasira A, Mpimbaza A, Achan J, Nankabirwa J, Nsobya S. Malaria in Uganda: challenges to control on the long road to elimination: I. Epidemiology and current control efforts. Acta Trop (2012.0) 121 184-195. DOI: 10.1016/j.actatropica.2011.03.004 37. Kibirige D, Lumu W, Jones AG, Smeeth L, Hattersley AT, Nyirenda MJ. Understanding the manifestation of diabetes in sub Saharan Africa to inform therapeutic approaches and preventive strategies: a narrative review. Clin Diabetes Endocrinol (2019.0) 5 2. DOI: 10.1186/s40842-019-0077-8 38. Srivastava S, Singh RK. Exploring integrated supply chain performance in healthcare: a service provider perspective. Benchmarking (2021.0) 28 106-130. DOI: 10.1108/BIJ-03-2020-0125 39. 39.Sheikhi M, Goodarzi M, Nowrozinejad A, Sheikhi S, editors. The Role of Information Sharing on Supply Chain. 2nd European Conference on Industrial Engineering and Operations Management (IEOM); 2018; Paris, France. 40. Lufesi NN, Andrew M, Aursnes I. Deficient supplies of drugs for life threatening diseases in an African community. BMC Health Serv Res (2007.0) 7 86. DOI: 10.1186/1472-6963-7-86 41. Nakyanzi JK, Kitutu FE, Oria H, Kamba PF. Expiry of medicines in supply outlets in Uganda. Bull World Health Organ (2010.0) 88 154-158. DOI: 10.2471/BLT.08.057471 42. Kanda MK, Iravo MA. Access Factors Affecting Supply Chain Efficiency of Medical Supplies in public Health Centres in Kenya: a Case Study of Public Health Centres in Elgeyo Marakwet Count. Int J Acad Res Account Finance Manag Sci (2015.0) 5 32-41 43. Mikkelsen-Lopez I, Shango W, Barrington J, Ziegler R, Smith T, deSavigny D. The challenge to avoid anti-malarial medicine stock-outs in an era of funding partners: the case of Tanzania. Malar J (2014.0) 13 181. DOI: 10.1186/1475-2875-13-181 44. Bogaert P, Bochenek T, Prokop A, Pilc A. A qualitative approach to a better understanding of the problems underlying drug shortages, as viewed from Belgian, French and the European Union’s Perspectives. PLoS ONE (2015.0) 10 e0125691. DOI: 10.1371/journal.pone.0125691
--- title: 'Mutation spectrum of Kallmann syndrome: identification of five novel mutations across ANOS1 and FGFR1' authors: - Guoming Chu - Pingping Li - Qian Zhao - Rong He - Yanyan Zhao journal: 'Reproductive Biology and Endocrinology : RB&E' year: 2023 pmcid: PMC9976430 doi: 10.1186/s12958-023-01074-w license: CC BY 4.0 --- # Mutation spectrum of Kallmann syndrome: identification of five novel mutations across ANOS1 and FGFR1 ## Abstract ### Background Kallmann syndrome (KS) is a common type of idiopathic hypogonadotropic hypogonadism. To date, more than 30 genes including ANOS1 and FGFR1 have been identified in different genetic models of KS without affirmatory genotype–phenotype correlation, and novel mutations have been found. ### Methods A total of 35 unrelated patients with clinical features of disorder of sex development were recruited. Custom-panel sequencing or whole-exome sequencing was performed to detect the pathogenic mutations. Sanger sequencing was performed to verify single-nucleotide variants. Copy number variation-sequencing (CNV-seq) was performed to determine CNVs. The pathogenicity of the identified variant was predicted in silico. mRNA transcript analysis and minigene reporter assay were performed to test the effect of the mutation on splicing. ### Results ANOS1 gene c.709 T > A and c.711 G > T were evaluated as pathogenic by several commonly used software, and c.1063-2 A > T was verified by transcriptional splicing assay. The c.1063-2 A > T mutation activated a cryptic splice acceptor site downstream of the original splice acceptor site and resulted in an aberrant splicing of the 24-basepair at the 5′ end of exon 8, yielding a new transcript with c.1063–1086 deletion. FRFR1 gene c.1835delA was assessed as pathogenic according to the ACMG guideline. The CNV of del[8](p12p11.22)chr8:g.36140000_38460000del was judged as pathogenic according to the ACMG & ClinGen technical standards. ### Conclusions Herein, we identified three novel ANOS1 mutations and two novel FGFR1 variations in Chinese KS families. In silico prediction and functional experiment evaluated the pathogenesis of ANOS1 mutations. FRFR1 c.1835delA mutation and del[8](p12p11.22)chr8:g.36140000_38460000del were assessed as pathogenic variations. Therefore, our study expands the spectrum of mutations associated with KS and provides diagnostic evidence for patients who carry the same mutation in the future. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12958-023-01074-w. ## Background Idiopathic hypogonadotropic hypogonadism (IHH) is a rare genetic disease characterized by delayed or complete lack of puberty, which is caused by hypothalamic-pituitary–gonadal (HPG) axis dysfunction [1]. IHH is clinically divided into Kallmann syndrome with anosmia/hyposmia (KS) and normosmic IHH (nIHH). KS accounts for $50\%$ of all IHH cases [2], and KS occurs in 1 per 30,000 men and 1 per 125,000 women [3]. KS is clinically heterogeneous, mainly manifesting as gonadal dysplasia and possibly accompanied by other congenital malformations, such as renal agenesis or hypoplasia, cleft lip/palate, dental agenesis, hearing impairment, bimanual synkinesis, or skeletal anomalies [1]. KS exhibits different inheritance patterns and genetic heterogeneity including X-linked recessive (ANOS1), autosomal recessive (PROK2 and PROKR2), and autosomal dominant (FGFR1, FGF8, and CHD7) genes [4, 5]. ANOS1, located at the Xp22.3 and also known as Kallmann syndrome 1 (KAL1), is the first gene found to be associated with the X-linked form of KS [6]. Anosmin-1 protein that is encoded by ANOS1 contains four domains: a N-terminal cysteine-rich region (CR domain), a whey acidic protein-like (WAP) domain, four fibronectin type III (FnIII) domains, and a histidine-rich C-terminal region [7]. Currently, more than 200 mutations in ANOS1 have been included in the Human Gene Mutation Database (HGMD), all of which spread on the whole gene, and the number of mutation is increasing, but no mutation hotspots have been found in the affected regions [8]. Fibroblast growth factor receptor 1 (FGFR1) is located at 8p11.23. FGFR1 is a member of the tyrosine kinase superfamily of receptors. Dodé et al. reported that loss-of-function mutations of FGFR1 underlay KS2 for the first time in 2003. KS2 is an autosomal dominant disease [9]. To date, more than 300 mutations in FGFR1 have been included in HGMD, all of which included missense mutation, nonsense mutation, splice mutation, and rarely deletions. FGFR1 signaling plays an important role in neuronal migration, differentiation, survival, as well as cell proliferation during embryonic development mainly via the PI3K/Akt and MAPK pathways [10]. In this study, we recruited 35 unrelated individuals with IHH manifestations and identified three novel mutations in ANOS1 and two novel mutations in FGFR1. Additionally, we verified the novel splice site mutation for transcription function; the c.1063-2 A > T mutation generated an abnormal transcript and translated into a truncated protein product. Thus far, more than 30 genes linked with IHH pathogenesis have been identified; however, the mutation spectrum is incomplete owing to the high genetic and clinical heterogeneity of KS [11]. Thus, our study expanded the mutation spectrum of KS and provided diagnostic evidence for patients with the same mutation in the future. ## Participants In total, 35 unrelated patients were recruited from Shengjing Hospital of China Medical University, all patients came for consultation because of disorder of sex development (DSD) (Table S1). In family 1, the proband (patient 1) manifested as small penis, cryptorchidism, and testicular dysplasia. In family 2, the proband (patient 2) phenotyped as small penis and testicular dysplasia. In family 3, the proband (patient 3) and his nephew showed cryptorchidism, left renal agenesis, and olfactory disorder. In family 4, the proband (patient 4) manifested as small penis, olfactory disorder, testicular dysplasia and polydactylism. In family 5, the proband (patient 5) phenotyped as cryptorchidism and testicular dysplasia. ## Genetic testing Genomic DNA was extracted from peripheral blood samples using a QIAamp DNA Blood Mini Kit (QIAGEN, Germany) according to the manufacturer’s instructions. Custom-panel sequencing (575 genes covering endocrine-related diseases) and whole-exome sequencing (WES) were performed by MyGenostics Inc. (Beijing, China) (Table S1). After sequencing, the raw data were saved as a FASTQ format. Both Illumina sequencing adapters and low quality reads (< 80 bp) were filtered by cutadaptor software (http://code.google.com/p/cutadapt/). The clean reads were mapped to the UCSC hg19 human reference genome using the parameter BWA of Sentieon software.(https://www.sentieon.com/). The duplicated reads were removed using the parameter driver of Sentieon software, and the parameter driver is used to correct the base, so that the quality value of the base in the reads of the final output BAM file can be closer to the real probability of mismatch with the reference genome, and the mapped reads were used for the detection of variation. The variants of SNP and InDel were detected by the parameter driver of Sentieon software. Then, the data would be transformed to VCF format. Variants were further annotated by ANNOVAR software (http://annovar.openbioinformatics.org/en/latest/). In the present study, three steps were used to select the potential pathogenic variants in downstream analysis: [1] variant reads should be more than 5, and variant ration should be no less than $30\%$; [2] the variants should be removed, when the highest minor allele frequency (MAF) in 1000 Genomes, EXAC, ESP6500, gnomAD, and dbSNP was more than $5\%$; [3] the synonymous mutations should be removed, when they were not in the HGMD database. After that, the rest mutations should be the potential pathogenic mutations for further analysis. The candidate causative gene mutations were verified through the Sanger sequencing. Target gene fragments including mutations were amplified using the PrimeSTAR® HS DNA Polymerase (#R010Q, Takara, Dalian, China). Sanger sequencing was performed on 3730 DNA analyzer (Applied Biosystems, USA) using BigDye™ Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, USA). Copy number variation-sequencing (CNV-seq) analysis was conducted using library construction kit (KR2000, Berry Genomics, Beijing, China) on the Illumina Nextseq CN500 platform. Sequencing data were analyzed by the data analysis system (Berry Genomics, Beijing, China). ## ANOS1 mRNA transcript analysis Total RNA was extracted from peripheral blood samples using a TransZol Up Plus RNA Kit (#ER501-01, TransGen Biotech Co., LTD, Dalian, Beijing, China) according to the manufacturer’s instructions. cDNA was synthesized from total RNA using PrimeScript™ RT reagent Kit with gDNA Eraser (#RR047A, Takara, Dalian, China). The cDNA products (NM_000216.4) were used as templates for PCR amplification using the following primers:F: 5′—CGAGTGGCTGCTGTGAATGTG—3′R: 5′—GCGAGTGGGTCGTCGTCTT—3′ ANOS1 mRNA transcripts were identified by Sanger sequencing. ## Minigene reporter assay The wild-type and variant amplicons encompassing intron 7, exon 8, and intron 8 of ANOS1 (NC_000023) were obtained from the genomic DNA of the father and proband, respectively, using the following primers:F: 5′—CCGCTCGAGGCAGTCAGGAGCCACCGC—3′R: 5′—CTAGCTAGCCTCTCCCTCCATTGTGCCTTG—3′ The PCR products were cloned into the pSPL3 vector, which was kindly provided by Professor Leping Shao, The Affiliated Qingdao Municipal Hospital of Qingdao University, China) by XhoI and NheI. Wild-type (pSPL3-ANOS1-WT) and variant (pSPL3-ANOS1-MUT) constructs were verified by Sanger sequencing (data not shown). Human embryo kidney cells (HEK293) were transfected with the plasmid of pSPL3-ANOS1-WT or pSPL3-ANOS1-MUT for 24 h using jetPEI®DNA transfection Reagent (Polyplus, France) according to the manufacturer’s protocol. cDNA was obtained from RNA extraction by reverse-transcription polymerase chain reaction (RT-PCR). The cDNA products were used as templates for PCR amplification using the following pSPL3 vector-specific primers [12]:F: 5′—TCTGAGTCACCTGGACAACC—3′R: 5′—ATCTCAGTGGTATTTGTGAGC—3′ Splicing transcripts were identified by Sanger sequencing. ## Pathogenicity analysis of the variation All mutations were assessed according to the American College of Medical Genetics and Genomics (ACMG) guideline. The pathogenicity of missense mutation was predicted using six tools: Revel (https://sites.google.com/site/revelgenomics), SIFT (http://sift.jcvi.org), PROVEAN (http://provean.jcvi.org), PolyPhen-2 (http://genetics.bwh.harvard.edu/pph2), MutationTaster (https://www.mutationtaster.org), M-CAP (http://bejerano.stanford.edu/mcap/), and GERP (http://mendel.stanford.edu/sidowlab/downloads/gerp/index.html). The evaluation of anosmin-1 amino acid conservation was performed by Ugene software [13] using the data from HomoloGene (ID:55,445). CNVs were evaluated according to the technical standards recommended by the ACMG and the Clinical Genome Resource (ClinGen). ## Monogene mutation screening and pedigree analysis Five cases out of 35 unrelated patients were detected positive for mutations by custom-panel sequencing or WES, and their family members were analyzed. In family 1, a hemizygote ANOS1 (the transcript is NM_000216, hg19) mutation c.1063-2 A > T was found in the proband (Fig. 1A, II1), and the results of pedigree verification confirmed that the mutation was inherited from the proband’s mother (I2).Fig. 1Pedigree and sequencing results of families 1–3. The pedigree showed core family members: square symbol represented males; circle symbol indicated females; Roman numerals represented generations; and Arabic numerals indicated the position of each individual within the family. A Pedigree and sequencing results of family 1. Sanger sequencing showed that the proband (II1) carried a hemizygous ANOS1 c.1063-2 A > T mutation, while the proband’s mother (I2) carried a heterozygous ANOS1 c.1063-2 A > T mutation. B Pedigree and sequencing results of family 2. NGS showed that the proband (II1) carried a hemizygous ANOS1 c.711 G > T mutation, while Sanger sequencing showed that the proband’s mother (I2) carried a heterozygous ANOS1 c.711 G > T mutation. C Pedigree and sequencing results of family 3. Sanger sequencing showed that the proband (II1) carried a hemizygous ANOS1 c.709 T > A mutation, the proband’s mother (I2) and sister (II2) carried a same heterozygous ANOS1 c.709 T > A mutation, and the proband’s nephew (III1) carried a hemizygote ANOS1 c.709 T > A mutation In family 2, a hemizygote ANOS1 c.711 G > T mutation was found in the proband (Fig. 1B, II1), and the results of the pedigree verification confirmed that the mutation was inherited from the proband’s mother (I2). In family 3, a hemizygote ANOS1 c.709 T > A mutation was found in the proband (Fig. 1C, II1). The results of the pedigree verification showed that the proband’s mother (I2) and sister (II2) carried heterozygous ANOS1 c.709 T > A mutation, and the proband’s nephew (III1) carried hemizygote ANOS1 c.709 T > A mutation. In family 4, a de novo heterozygous FGFR1 (the transcript is NM_023110, hg19) mutation c.1835delA was found in the proband (Fig. 2A, II1), and the results of the pedigree verification showed that the father and mother were normal (I1 and I2).Fig. 2Pedigree and sequencing results of families 4 and 5. A Pedigree and sequencing results of family 4. Sanger sequencing showed that the proband (II1) carried a heterozygous FGFR1 c.1835delA mutation, while the proband’s father (I1) and mother (I2) were normal. B Pedigree and CNV-seq results of family 5. CNV-seq results showed that the proband (II1) carried a 2.32 Mb deletion at 8p12-p11.22 (36,140,000–38,460,000), while the proband’s father (I1) and mother (I2) were normal In family 5, the custom-panel sequencing result showed a sporadic 1.68 Mb deletion at 8p11.23-p11.22, and the CNV-seq results showed a 2.32 Mb deletion at 8p12-p11.22 (36,140,000–38,460,000) in the proband (Fig. 2B, II1), while the father and mother were normal (I1 and I2). ## Verification of ANOS1 c.1063-2 A > T splicing mutation ex vivo and in vitro To clarify the splicing effect of ANOS1 c.1063-2 A > T mutation in family 1, cDNA sequence analysis was performed ex vivo. As shown in Fig. 3, c.1063-2 A > T mutation resulted in an aberrant splicing of 24-bp at the 5′ end of exon 8 and generated a new transcript with c.1063–1086 deletion; both of the normal and aberrant transcripts of ANOS1 were detected in the proband’s peripheral blood (Fig. 3A). In addition, few aberrant transcripts were found in the maternal peripheral blood (Fig. 3A).Fig. 3Results of the ANOS1 mRNA transcripts analysis. A Sanger sequencing results of the RT-PCR products showed existence of both the wild-type (WT) transcript and the mutant transcript in the peripheral blood. The mutant ANOS1 transcript was aberrant splicing with c.1063–1086 deletion. B Schematic structures of anosmin-1 proteins. The domains of anosmin-1 protein were retrieved from the Ensembl database (Transcript ID: ENST00000262648.8). The truncated protein exhibited a mutated FnIII.2 (ΔFnIII.2) domain with amino acid deletion at 355–362 Splicing experiment in vitro was performed by constructing a splicing reporter minigene vector. As shown in Fig. 4B, the RT-PCR product sequence of the mutation construct was shorter than that of the wild-type construct, and the sequencing analysis revealed that the c.1063-2 A > T mutation led to an aberrant splicing consistent with the validation ex vivo. As expected, a truncated protein with p.355–362 deletion was generated (Fig. 3B).Fig. 4Results of minigene splicing assay. A Schematic diagram of minigene construction. The asterisk indicates the location of the c.1063-2 A > T mutation. The splicing pattern of wild-type (WT, top) and mutant (MUT, bottom) was showed respectively. B Minigene RT-PCR product sequencing results. The WT minigene formed normal mRNA composed of exon A, exon 8 and exon B. The mutant minigene caused a splicing abnormality, resulting in the 24 bp nucleotides deletion of 5′ end of exon 8 (c.1063–1086 deletion) ## Pathogenicity analysis of ANOS1 c.709 T > A and c.711 G > T missense mutations ANOS1 c.711 G > T and c.709 T > A variations in family 2 and 3 were two missense mutations that led to different amino acid substitutions of p.W237R and p.W237C, respectively, at the same position of anosmin-1 protein. The two novel variants had not been presented in the normal population by searching databases (1000 Genomes, EXAC, ESP6500, gnomAD, and dbSNP) (Table S2), and the W239 of anosmin-1 is highly conserved in multiple species (Fig. S1). Both p.W237R and p.W237C were predicted to be damaging using multiple software programs, such as Revel, SIFT, PolyPhen-2, MutationTaster, M-CAP, GERP, and PROVEAN (Table S2). ## Pathogenicity analysis of FRFR1 mutations FRFR1 c.1835delA variation in family 4 was a frameshift mutation that resulted in conversion of glutamate-612 to glycine and generated a stop codon at amino acid 631 (p.Glu612Glyfs20). According to the ACMG guideline, FRFR1 c.1835delA was assessed as a pathogenicity mutation using the evidence of PVS1 (null variant), PS2 (de novo variant), and PM2 (absent from controls). FRFR1 p.Glu612Glyfs20 mutation could express a truncated protein lacking the second tyrosine kinase domain (TKD) (Fig. 5B).Fig. 5Chromosome 8p12-p11.22 gene map and the topological structure of FGFR1. A The orange box in the chromosome 8 diagram at the top indicated the region highlighted below. The OMIM morbid genes are listed at the bottom. B Membrane topology structure of full length FGFR1. Each circle represented a residue with the one-letter symbol. The different domains were labeled as indicated. Locations of pathogenic variations were colored as indicated. and the red box marked the amino acids that affected by FRFR1 c.1835delA mutation The CNV of seq[GRCh37]del[8](p12p11.22)chr8:g.36140000_38460000del in patient 5 contained 19 protein-coding genes, including 6 morbid genes of the Online Mendelian Inheritance in Man (OMIM) database (Fig. 5A). The CNV was judged as a pathogenic variation using the evidence of 2A (complete overlap of an established haploinsufficiency gene FGFR1 which contributes to KS2 [ClinGen ID: ISCA-32176], 1.00 point) according to the ACMG & ClinGen technical standards. ## Discussion IHH is an inheritable disorder with clinical and genetic heterogeneity. The major pathogenesis of IHH is failure to activate pulsatile secretion of gonadotropin-releasing hormone (GnRH) during puberty [14]. As the main category of IHH, KS exhibits the typical characteristics of gonadal dysplasia and anosmia [15]. Complex symptoms hinder clinicians from making an accurate diagnosis of KS, while ultrasonography, magnetic resonance imaging, and serum hormone level measurements are routine auxiliary diagnostic methods. To date, more than 20 pathogenic genes have been found to be associated with KS, including ANOS1, FGFR1, PROK2, PROKR2, NELF, KISSR1, CHD7, SEMA3A, and FGF8 [16]. Genetic analysis is the core diagnostic method of KS. Anosmin-1 encoded by ANOS1 plays important roles in substrate adhesion and cell migration of GnRH-1 neurons, axon outgrowth, and collateral formation, which is basically related to the central nervous system [17–19]. Herein, we reported three novel ANOS1 mutations. One of them, ANOS1 c.1063-2 A > T mutation in family 1 led to the production of a new transcript with c.1063–1086 deletion by the transcriptional splicing verification, and a truncated protein with p.355–362 deletion was predicted. It has been known that mutations that affect the splice donor and acceptor site (canonical GT-AG) are highly predictive of splicing defects, and always lead to an abnormal transcript with complete skipping of the downstream exon [20]. However, it is also possible that some splice acceptor site mutations disrupt the original splice acceptor site, and activate a cryptic splice acceptor site, resulting in an aberrant splicing with partial deletion of the downstream exon [21]. Therefore, the verification of a novel splice site variant is essential to clarify the pathogenic mechanism. In the present study, ANOS1 c.1063-2 A > T mutation broken the splice acceptor site AG at intron 7, substituted by TG, and activated a cryptic splice acceptor site AG in exon 8. Intriguingly, both normal and aberrant transcripts of ANOS1 were detected in the proband’s peripheral blood. Previous studies have reported that males carrying a hemizygous splicing mutation of genes located on chromosome X showed both normal and aberrant transcripts, such as c.5786 + 4 A > G in ATRX, c.1044 + 5 G > A in FOXP3, and c.463-6 T > G in OGT [22–24], so far the underlying mechanism remained unclear. We speculated that the introduced non-canonical splice site GT-TG caused by the c.1063-2 A > T mutation could produce the normal transcript as GT-TG splice site had been found in animal genomes [25]. Moreover, skewed X-chromosome inactivation (XCI) may be the reason of the few aberrant transcripts in the maternal peripheral blood. Polla et al. reported that the aberrant transcript of MED12 in the peripheral blood samples of those heterozygous women could not be detected because of the skewed XCI [26]. Additionally, we reported two novel missense mutations of p.W237C in family 2 and p.W237R in family 3, which changed the amino acid sequence at the same position of anosmin-1 protein to different amino acid substitutions. W237 was located within the FnIII.1 domain with a highly conservatism. Anosmin-1 is a component of the FGFR1 pathway [27]. WAP, FnIII.1 FnIII.3, and FnIII.4 domains of anosmin-1 were confirmed to interact with FGFR1 [28, 29], and relevant mutations such as N267K in FnIII.1, E514K and F517L in FnIII.3 were proved to influence their interaction [30]. p.W237C and p.W237R may also regulate the signal transduction of the FGFR1 pathway and lead to disease occurrence. Notably, the mutations carried by patients in families 2 and 3 led to different clinical manifestations. Both patients in family 3 showed unilateral deletion of the left kidney, while patients in family 2 had healthy kidney, which attracted our attention to study the relationship between anosmin-1 and kidney development. Approximately $10\%$ of males with ANOS1 pathogenic variants also showed unilateral renal agenesis [31]. Georgopoulos et al. proposed that the FnIII repeats of ANOS1 might be essential for normal renal development [32], but the specific mechanism has not been fully elucidated. Our findings may provide a new idea for the mechanism of anosmin-1 involvement in kidney development. FGFR1 is responsible for an autosomal dominant form of KS. A full-length FGFR1 protein consists of an extracellular region, composed of three immunoglobulin-like domains (IgI, IgII, and IgIII) responsible for the receptor’s affinity and specificity to its ligands, a single hydrophobic membrane-spanning segment, and two cytoplasmic TKDs with tyrosine kinase activity (Fig. 5B) [33]. The analysis of the total 198 FGFR1 missense mutations listed in the HGMD database showed no obvious characteristics. The overall mutation locations are relatively scattered, mainly in the IgG-like domains and intracellular TKDs, but rarely in the transmembrane region (Fig. 5B). Herein, we identified a novel frameshift FGFR1 mutation c.1835delA (p.Glu612Glyfs20) in family 4. Following stop codons mutations of FGFR1 including p.R661 and p.Q680* were considered as pathogenic mutations, with protein truncation and nonsense-mediated mRNA decay (NMD) being the underlying mechanisms [34, 35]. Similarly, FGFR1 p.Glu612Glyfs20 mutation was supposed to result in the synthesis of a truncated inactive receptor lacking an essential portion of the catalytic TKD. Alternatively, the FGFR1 mRNA containing the premature stop codon could be degraded through NMD, leading to haploinsufficiency. In addition, we reported a family (family 5) with a CNV covering the entire FGFR1. To our knowledge, the whole FGFR1 gene deletion was previously reported in only seven cases of KS [9, 33, 36, 37]. By searching the ClinVar database, only six cases with complete FGFR1 were included (VCV000147723, VCV000057114, VCV000150494, VCV000145616, VCV000060360, VCV000362858, VCV000282690, and VCV000395746). The ClinGen database revealed that the haploinsufficiency of FGFR1 contributed to KS2. The 2.32 Mb deletion at 8p12-p11.22 (36,140,000–38,460,000) in the proband of family 5 covered the whole FGFR1 gene. This rare CNV was judged as the pathogenic mutation of KS in this patient as a sporadic case, which is the first report of whole FGFR1 deletion in Chinese population. ## Conclusions In this study, we identified three novel pathogenic mutations of ANOS1 and two novel mutations of FGFR1 in five Chinese families. In silico prediction indicated that ANOS1 c.711 G > T and c.709 T > A were pathogenic mutations. Splicing experiments elucidated the pathogenesis of the ANOS1 c.1063-2 A > T mutation. FRFR1 c.1835delA mutation and del[8](p12p11.22)chr8:g.36140000_38460000del were assessed as pathogenic variations. Therefore, our study expands the spectrum of mutations associated with KS and provides diagnostic evidence for patients who carry the same mutation in the future. ## Supplementary Information Additional file 1: Fig. S1 Conservation analysis* of anosmin-1. Assessment of amino acid conservation of anosmin-1 using Ugene. W237 of anosmin-1 (indicated in a black box) is highly conserved across various species. Additional file 2: Table S1 The clinical and genetic data of patients. Additional file 3: Table S2 The results of population databases retrieval and pathogenicity prediction for ANOS1 gene c.709 T > A and c.711 G > T variation. ## References 1. Quinton R, Duke VM, Robertson A, Kirk JM, Matfin G, de Zoysa PA. Idiopathic gonadotrophin deficiency: genetic questions addressed through phenotypic characterization. Clin Endocrinol (Oxf) (2001.0) 55 163-174. DOI: 10.1046/j.1365-2265.2001.01277.x 2. Boehm U, Bouloux PM, Dattani MT, de Roux N, Dodé C, Dunkel L. Expert consensus document: European Consensus Statement on congenital hypogonadotropic hypogonadism–pathogenesis, diagnosis and treatment. Nat Rev Endocrinol (2015.0) 11 547-564. DOI: 10.1038/nrendo.2015.112 3. Laitinen EM, Vaaralahti K, Tommiska J, Eklund E, Tervaniemi M, Valanne L. Incidence, phenotypic features and molecular genetics of Kallmann syndrome in Finland. Orphanet J Rare Dis (2011.0) 6 41. DOI: 10.1186/1750-1172-6-41 4. Stamou MI, Georgopoulos NA. Kallmann syndrome: phenotype and genotype of hypogonadotropic hypogonadism. Metabolism (2018.0) 86 124-134. DOI: 10.1016/j.metabol.2017.10.012 5. 5.Jiang X, Li D, Gao Y, Zhang X, Wang X, Yang Y, et al. A novel splice site variant in ANOS1 gene leads to Kallmann syndrome in three siblings. Gene. 2020;726:144177. 10.1016/j.gene.2019.144177. 6. 6.Franco B, Guioli S, Pragliola A, Incerti B, Bardoni B, Tonlorenzi R, et al. A gene deleted in Kallmann’s syndrome shares homology with neural cell adhesion and axonal path-finding molecules. Nature. 1991;353:529–36. 10.1038/353529a0. 7. del Castillo I, Cohen-Salmon M, Blanchard S, Lutfalla G, Petit C. Structure of the X-linked Kallmann syndrome gene and its homologous pseudogene on the Y chromosome. Nat Genet (1992.0) 2 305-310. DOI: 10.1038/ng1292-305 8. Nie M, Xu H, Chen R, Mao J, Wang X, Xiong S. Analysis of genetic and clinical characteristics of a Chinese Kallmann syndrome cohort with ANOS1 mutations. Eur J Endocrinol (2017.0) 177 389-398. DOI: 10.1530/EJE-17-0335 9. Dodé C, Levilliers J, Dupont JM, De Paepe A, Le Dû N, Soussi-Yanicostas N. Loss-of-function mutations in FGFR1 cause autosomal dominant Kallmann syndrome. Nat Genet (2003.0) 33 463-465. DOI: 10.1038/ng1122 10. Böttcher RT, Niehrs C. Fibroblast growth factor signaling during early vertebrate development. Endocr Rev (2005.0) 26 63-77. DOI: 10.1210/er.2003-0040 11. Gach A, Pinkier I, Szarras-Czapnik M, Sakowicz A, Jakubowski L. Expanding the mutational spectrum of monogenic hypogonadotropic hypogonadism: novel mutations in ANOS1 and FGFR1 genes. Reprod Biol Endocrinol (2020.0) 18 8. DOI: 10.1186/s12958-020-0568-6 12. Zhang R, Chen Z, Song Q, Wang S, Liu Z, Zhao X. Identification of seven exonic variants in the SLC4A1, ATP6V1B1, and ATP6V0A4 genes that alter RNA splicing by minigene assay. Hum Mutat (2021.0) 42 1153-1164. DOI: 10.1002/humu.24246 13. 13.Okonechnikov K, Golosova O, Fursov M; UGENE team. Unipro UGENE: a unified bioinformatics toolkit. Bioinformatics. 2012;28:1166–7. 10.1093/bioinformatics/bts091. 14. Bianco SD, Kaiser UB. The genetic and molecular basis of idiopathic hypogonadotropic hypogonadism. Nat Rev Endocrinol (2009.0) 5 569-576. DOI: 10.1038/nrendo.2009.177 15. Sarfati J, Bouvattier C, Bry-Gauillard H, Cartes A, Bouligand J, Young J. Kallmann syndrome with FGFR1 and KAL1 mutations detected during fetal life. Orphanet J Rare Dis (2015.0) 10 71. DOI: 10.1186/s13023-015-0287-9 16. 16.Chen K, Wang H, Lai Y. Kallmann Syndrome Due to Heterozygous Mutation in SOX10 Coexisting With Waardenburg Syndrome Type II: Case Report and Review of Literature. Front Endocrinol (Lausanne). 2021;11:592831. 10.3389/fendo.2020.592831. 17. Hamada AJ, Esteves SC, Agarwal A. A comprehensive review of genetics and genetic testing in azoospermia. Clinics (Sao Paulo) (2013.0) 68 39-60. DOI: 10.6061/clinics/2013(sup01)06 18. 18.Cariboni A, Pimpinelli F, Colamarino S, Zaninetti R, Piccolella M, Rumio C, et al. The product of X-linked Kallmann’s syndrome gene (KAL1) affects the migratory activity of gonadotropin-releasing hormone (GnRH)-producing neurons. Hum Mol Genet. 2004;13:2781–91. 10.1093/hmg/ddh309. 19. Soussi-Yanicostas N, Faivre-Sarrailh C, Hardelin JP, Levilliers J, Rougon G, Petit C. Anosmin-1 underlying the X chromosome-linked Kallmann syndrome is an adhesion molecule that can modulate neurite growth in a cell-type specific manner. J Cell Sci (1998.0) 111 2953-2965. DOI: 10.1242/jcs.111.19.2953 20. 20.Saadah OI, Banaganapalli B, Kamal NM, Sahly AN, Alsufyani HA, Mohammed A, et al. Identification of a Rare Exon 19 Skipping Mutation in ALMS1 Gene in Alström Syndrome Patients From Two Unrelated Saudi Families. Front Pediatr. 2021;9:652011. 10.3389/fped.2021.652011. 21. 21.Thongnoppakhun A, Rungroj N, Wilairat P, Vareesangthip K, Sirinavin C, Yenchitsomanus PT. A novel splice-acceptor site mutation (IVS13-2A>T) of polycystic kidney disease 1 (PKD1) gene resulting in an RNA processing defect with a 74-nucleotide deletion in exon 14 of the mRNA transcript. Hum Mutat. 2000;15:115. 10.1002/(SICI)1098-1004(200001)15:1%3c115::AID-HUMU22%3e3.0.CO;2-Z. 22. 22.Cong Y, Wu J, Wang H, Wu K, Huang C, Yang X. Identification of a Hemizygous Novel Splicing Variant in ATRX Gene: A Case Report and Literature Review. Front Pediatr. 2022;10:834087. 10.3389/fped.2022.834087. 23. Hwang JL, Park SY, Ye H, Sanyoura M, Pastore AN, Carmody D. FOXP3 mutations causing early-onset insulin-requiring diabetes but without other features of immune dysregulation, polyendocrinopathy, enteropathy. X-linked syndrome Pediatr Diabetes (2018.0) 19 388-392. DOI: 10.1111/pedi.12612 24. Willems AP, Gundogdu M, Kempers MJE, Giltay JC, Pfundt R, Elferink M. Mutations in N-acetylglucosamine (O-GlcNAc) transferase in patients with X-linked intellectual disability. J Biol Chem (2017.0) 292 12621-12631. DOI: 10.1074/jbc.M117.790097 25. Frey K, Pucker B. Animal, Fungi, and Plant Genome Sequences Harbor Different Non-Canonical Splice Sites. Cells (2020.0) 9 458. DOI: 10.3390/cells9020458 26. Polla DL, Bhoj EJ, Verheij JBGM, Wassink-Ruiter JSK, Reis A, Deshpande C. De novo variants in MED12 cause X-linked syndromic neurodevelopmental disorders in 18 females. Genet Med (2021.0) 23 645-652. DOI: 10.1038/s41436-020-01040-6 27. Ayari B, Soussi-Yanicostas N. FGFR1 and anosmin-1 underlying genetically distinct forms of Kallmann syndrome are co-expressed and interact in olfactory bulbs. Dev Genes Evol (2007.0) 217 169-175. DOI: 10.1007/s00427-006-0125-0 28. Murcia-Belmonte V, Astillero-López V, Esteban PF. Anosmin 1 Interacts with the Prokineticin Receptor 2 In Vitro Indicating a Molecular Link Between Both Proteins in the Pathogenesis of Kallmann Syndrome. Protein Pept Lett (2016.0) 23 650-655. DOI: 10.2174/0929866523666160517123331 29. Murcia-Belmonte V, Esteban PF, García-González D, De Castro F. Biochemical dissection of Anosmin-1 interaction with FGFR1 and components of the extracellular matrix. J Neurochem (2010.0) 115 1256-1265. DOI: 10.1111/j.1471-4159.2010.07024.x 30. de Castro F, Esteban PF, Bribián A, Murcia-Belmonte V, García-González D, Clemente D. The adhesion molecule anosmin-1 in neurology: Kallmann syndrome and beyond. Adv Neurobiol (2014.0) 8 273-292. DOI: 10.1007/978-1-4614-8090-7_12 31. 31.Louden ED, Poch A, Kim HG, Ben-Mahmoud A, Kim SH, Layman LC. Genetics of hypogonadotropic Hypogonadism-Human and mouse genes, inheritance, oligogenicity, and genetic counseling. Mol Cell Endocrinol. 2021;534:111334. 10.1016/j.mce.2021.111334. 32. Georgopoulos NA, Koika V, Galli-Tsinopoulou A, Spiliotis BE, Adonakis G, Keramida MK. Renal dysgenesis and KAL1 gene defects in patients with sporadic Kallmann syndrome. Fertil Steril (2007.0) 88 1311-1317. DOI: 10.1016/j.fertnstert.2006.12.044 33. Sato N, Katsumata N, Kagami M, Hasegawa T, Hori N, Kawakita S. Clinical assessment and mutation analysis of Kallmann syndrome 1 (KAL1) and fibroblast growth factor receptor 1 (FGFR1, or KAL2) in five families and 18 sporadic patients. J Clin Endocrinol Metab (2004.0) 89 1079-1088. DOI: 10.1210/jc.2003-030476 34. Dodé C, Fouveaut C, Mortier G, Janssens S, Bertherat J, Mahoudeau J. Novel FGFR1 sequence variants in Kallmann syndrome, and genetic evidence that the FGFR1c isoform is required in olfactory bulb and palate morphogenesis. Hum Mutat (2007.0) 28 97-98. DOI: 10.1002/humu.9470 35. Pitteloud N, Acierno JS, Meysing A, Eliseenkova AV, Ma J, Ibrahimi OA. Mutations in fibroblast growth factor receptor 1 cause both Kallmann syndrome and normosmic idiopathic hypogonadotropic hypogonadism. Proc Natl Acad Sci U S A (2006.0) 103 6281-6286. DOI: 10.1073/pnas.0600962103 36. 36.Wang D, Lai P. Global retardation and hereditary spherocytosis associated with a novel deletion of chromosome 8p11.21 encompassing KAT6A and ANK1. Eur J Med Genet. 2020;63:104082. 10.1016/j.ejmg.2020.104082. 37. Trarbach EB, Teles MG, Costa EM, Abreu AP, Garmes HM, Guerra G. Screening of autosomal gene deletions in patients with hypogonadotropic hypogonadism using multiplex ligation-dependent probe amplification: detection of a hemizygosis for the fibroblast growth factor receptor 1. Clin Endocrinol (Oxf) (2010.0) 72 371-376. DOI: 10.1111/j.1365-2265.2009.03642.x
--- title: Brain fatty acid and transcriptome profiles of pig fed diets with different levels of soybean oil authors: - Bruna Pereira Martins da Silva - Simara Larissa Fanalli - Julia Dezen Gomes - Vivian Vezzoni de Almeida - Heidge Fukumasu - Felipe André Oliveira Freitas - Gabriel Costa Monteiro Moreira - Bárbara Silva-Vignato - James Mark Reecy - James Eugene Koltes - Dawn Koltes - Júlio Cesar de Carvalho Balieiro - Severino Matias de Alencar - Julia Pereira Martins da Silva - Luiz Lehmann Coutinho - Juliana Afonso - Luciana Correia de Almeida Regitano - Gerson Barreto Mourão - Albino Luchiari Filho - Aline Silva Mello Cesar journal: BMC Genomics year: 2023 pmcid: PMC9976441 doi: 10.1186/s12864-023-09188-6 license: CC BY 4.0 --- # Brain fatty acid and transcriptome profiles of pig fed diets with different levels of soybean oil ## Abstract ### Background The high similarity in anatomical and neurophysiological processes between pigs and humans make pigs an excellent model for metabolic diseases and neurological disorders. Lipids are essential for brain structure and function, and the polyunsaturated fatty acids (PUFA) have anti-inflammatory and positive effects against cognitive dysfunction in neurodegenerative diseases. Nutrigenomics studies involving pigs and fatty acids (FA) may help us in better understanding important biological processes. In this study, the main goal was to evaluate the effect of different levels of dietary soybean oil on the lipid profile and transcriptome in pigs’ brain tissue. ### Results Thirty-six male Large White pigs were used in a 98-day study using two experimental diets corn-soybean meal diet containing $1.5\%$ soybean oil (SOY1.5) and corn-soybean meal diet containing $3.0\%$ soybean oil (SOY3.0). No differences were found for the brain total lipid content and FA profile between the different levels of soybean oil. For differential expression analysis, using the DESeq2 statistical package, a total of 34 differentially expressed genes (DEG, FDR-corrected p-value < 0.05) were identified. Of these 34 DEG, 25 are known-genes, of which 11 were up-regulated (log2 fold change ranging from + 0.25 to + 2.93) and 14 were down-regulated (log2 fold change ranging from − 3.43 to -0.36) for the SOY1.5 group compared to SOY3.0. For the functional enrichment analysis performed using MetaCore with the 34 DEG, four pathway maps were identified (p-value < 0.05), related to the ALOX15B (log2 fold change − 1.489), CALB1 (log2 fold change − 3.431) and CAST (log2 fold change + 0.421) genes. A “calcium transport” network (p-value = 2.303e-2), related to the CAST and CALB1 genes, was also identified. ### Conclusion The results found in this study contribute to understanding the pathways and networks associated with processes involved in intracellular calcium, lipid metabolism, and oxidative processes in the brain tissue. Moreover, these results may help a better comprehension of the modulating effects of soybean oil and its FA composition on processes and diseases affecting the brain tissue. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12864-023-09188-6. ## Background The pigs (Sus scrofa) have global economic impact as it is the second most consumed meat-based protein source worldwide [1, 2]. Additionally, pigs are considered an animal model and have been used in research in the area of nutrigenomics and human metabolic diseases. Moreover, pigs can be used to understand neurodegenerative diseases due to similar of the brain anatomy, development, function, and neurophysiological process compared to the brains of small laboratory animals and humans [3–6]. The brain contains high lipid content, making up approximately $50\%$ of the brain’s dry weight, only lower than the adipose tissue [7]. Lipids are essential for brain structure and function, and the central nervous system is fundamental for the regulation of metabolism and lipid balance [8, 9]. In addition, some regions of the brain are capable to detect nutrients and hormones that regulate energy balance and feeding [8, 9]. A noteworthy factor is that the diet fed to the pigs can alter the lipid and fatty acids (FA) profiles of the tissues [10, 11]. Thus, soybean oil has been commonly used as part of the feed composition for growing-finishing pigs because it results in improved growth performance and beneficial effects to consumers [12]. In addition, soybean oil is high in polyunsaturated fatty acids (PUFA), being rich in linolenic acid (LA, C18:2 n-6), which is associated with the reduction of cardiovascular diseases and serum cholesterol [13]. Dietary derived FA, such as LA and alpha-linolenic acid (ALA, C18:3 n-3), act as precursors of PUFA like docosahexaenoic acid (DHA, C22:6 n-3) and arachidonic acid (AA, C20:4 n-6). Dietary supplementation of DHA may have potential neuroprotection effects against chronic and acute inflammation in the central nervous system, as well as slowing cognitive decline in Alzheimer’s disease [14]. PUFA and their metabolites act in the brain by activating receptors and cell signaling pathways. Additionally, they are responsible for modulating the system related to signaling lipids, present in phospholipids of the neuronal cell membrane, and for regulating synaptic function [15, 16]. While the roles of specific classes of FA in brain function are being elucidated, the understanding of the genes involved in the dietary modulation of FA in the brain is unclear and limited. Thus, the objective of this work was to determine if different levels of dietary soybean oil fed to male pigs would modify the lipid and transcriptome profile of the brain. ## Total lipid content and FA profile Table 1 shows the total lipid composition and FA profile of the brain tissue from pigs given diets with different levels of soybean oil (SOY1.5 vs. SOY3.0). No changes (p-value ≤ 0.05) were identified in the total lipid content and the FA profile between the treatments. Table 1Total lipid content and FA profile in brain tissue of pigs fed diets containing different levels of soybean oilFatty acid, %Dietary treatment1Pooled SEM2p-valueSOY1.5SOY3.0Total lipids9.92810.2920.1130.199Saturated fatty acid (SFA)Myristic acid (C14:0)0.5220.5210.0060.927Palmitic acid (C16:0)26.84827.0370.1890.709Stearic acid (C18:0)29.13128.3710.2080.110Monounsaturated fatty acid (MUFA)Palmitoleic acid (C16:1)0.4940.4620.0150.387Oleic acid (C18:1 n-9)30.07129.9550.1430.678Eicosenoic acid (C20:1 n-9)1.8971.8980.0240.967Polyunsaturated fatty acid (PUFA)Linoleic acid (C18:2 n-6)2.3212.3090.2620.984Alpha-linolenic acid (C18:3 n-3)ND3ND--Eicosapentaenoic acid (C20:5 n-3, EPA)0.1410.1350.0060.759Docosahexaenoic acid (C22:6 n-3, DHA)8.7818.9260.1510.620Total SFA56.58455.9250.2770.396Total MUFA32.49432.5010.1920.987Total PUFA10.85211.6850.2400.062Total n-3 PUFA48.7059.0140.1250.136Total n-6 PUFA51.8061.7680.0970.901PUFA:SFA ratio60.1920.2070.0050.134n-6:n-3 PUFA ratio70.2100.2310.0180.607Atherogenic index80.6640.6610.0080.9211Pigs ($$n = 36$$; 18 pigs/treatment) were fed either a corn-soybean meal diet containing $1.5\%$ soybean oil (SOY1.5) or diets containing with $3.0\%$ soybean oil (SOY3.0). Values represent the least square means. 2SEM = standard error of the least square means. 3ND = not detected. 4Total n-3 PUFA = {[C18:3 n-3] + [C20:5 n-3] + [C22:6 n-3]}. 5Total n-6 PUFA = C18:2 n-6. 6PUFA:SFA ratio = total PUFA/total SFA. 7Σ n-6/Σ n-3 PUFA ratio. 8Atherogenic index = (4 × [C14:0]) + (C16:0)/(total MUFA] + [total PUFA]), where brackets indicate concentrations [17]. ## RNA-Seq data and differentially expressed genes An average number of total reads per sample of 33.4 M and 32.9 M, was obtained for the SOY1.5 group, before and after quality control, respectively. For the SOY3.0 group, the average number of sequenced reads, before and after quality control, were 34.3 M and 33.9 M, respectively. Of the total reads obtained for both groups, after quality control, $95.02\%$ of them reads were mapped against the reference genome SScrofa11.1 (Additional file 1, Table S1). Differential analysis was performed comparing the level of gene expression between the groups, and a total of 22,931 genes were identified in the brain tissue. Of this 34 were DEG (FDR-corrected p-value < 0.05). Within the 34 DEG, 25 were known-genes, 11 being up-regulated (log2 fold change ranging from + 0.25 to + 2.93) and 14 being down-regulated (log2 fold change ranging from − 3.43 to -0.36) in the SOY1.5 compared to the SOY3.0. The genes with the most altered expression were CALB1 (log2 fold change − 3.43; FDR = 0.03) and VMO1 (log2 fold change + 2.93; FDR < 0.01). The list of expressed genes and DEG are demonstrated in Table S2. ## Functional enrichment analysis The MetaCore software was used to identify pathway maps from the list of 34 DEG (FDR < 0.05). Four pathway maps were identified (p-value < 0.05), related to the following genes: arachidonate 15-lipoxygenase type B (ALOX15B), calbidin-1 (CALB1), and calpastatin (CAST), as shown in Table 2. Table 2Pathway maps SOY1.51 vs. SOY3.02 in brain tissue of pigs fed diets containing different levels of soybean oilPathway mapp-valueDEG3log2 fold changeLinoleic acid metabolism1.970e-02 ALOX15B -1.489Prostaglandin-1 biosynthesis and metabolism3.597e-02 ALOX15B -1.489Renal secretion of inorganic electrolytes3.721e-02 CALB1 -3.431Immune response_IL-5 signaling via PI3K, MAPK, and NF-kB4.77e-02 CAST + 0.4211SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY3.0: corn-soybean meal diet containing $3.0\%$ soybean oil. 3DEG: Differentially expressed genes. The ALOX15B DEG, showing a down-regulation in the SOY1.5 group compared to SOY3.0 (log2 fold change − 1.489). The ALOX15B, participate in two of the four significant enriched pathway maps identified: “Linoleic acid metabolism” (p-value = 1.970e-2, Fig. 1), and “Prostaglandin-1 biosynthesis and metabolism” (p-value = 3.597e-2, Fig. 2). Fig. 1Linoleic acid metabolism in brain tissue of pigs fed diets containing different levels of soybean oil (SOY1.51 vs. SOY3.02). 1SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY1.5: corn-soybean meal diet containing $3.0\%$ soybean oil. The experimental data is represented by the thermometer-like figure on the map. The downward thermometer (blue) indicates down-regulation of the ALOX15B DEG (log2 fold change − 1.489) in the SOY1.5 group compared to SOY3.0. Network objects are represented by individual symbols. The green “T” icon shows which object is associated with the brain tissue. Interactions between objects are represented by arrows, mechanisms, and logical relationships. Further explanations are provided at https://portal.genego.com/legends/MetaCoreQuickReferenceGuide.pdf. Fig. 2Prostaglandin-1 biosynthesis and metabolism in brain tissue of pigs fed diets containing different levels of soybean oil (SOY1.51 vs. SOY3.02). 1SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY1.5: corn-soybean meal diet containing $3.0\%$ soybean oil. The experimental data is represented by the thermometer-like figure on the map. The downward thermometer (blue) indicates down-regulation of the ALOX15B DEG (log2 fold change − 1.489) in the SOY1.5 group compared to SOY3.0. Network objects are represented by individual symbols. The green “T” icon shows which object is associated with the brain tissue. Interactions between objects are represented by arrows, mechanisms, and logical relationships. Further explanations are provided at https://portal.genego.com/legends/MetaCoreQuickReferenceGuide.pdf. The CALB1 DEG, showing a down-regulation in the SOY1.5 group compared to SOY3.0 (log2 fold change − 3.431). The CALB1, participates in the enriched pathway map “Renal secretion of inorganic electrolytes” (p-value = 3.721e-2, Fig. 3). Fig. 3Renal secretion of inorganic electrolytes in brain tissue of pigs fed diets containing different levels of soybean oil (SOY1.51 vs. SOY3.02). 1SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY1.5: corn-soybean meal diet containing $3.0\%$ soybean oil. The experimental data is represented by the thermometer-like figure on the map. The downward thermometer (blue) indicates down-regulation of the CALB1 DEG (log2 fold change − 3.431) in the SOY1.5 group compared to SOY3.0. Network objects are represented by individual symbols. The green “T” icon shows which object is associated with the brain tissue. Interactions between objects are represented by arrows, mechanisms, and logical relationships. Further explanations are provided at https://portal.genego.com/legends/MetaCoreQuickReferenceGuide.pdf. The CAST DEG, showing an up-regulation in the SOY1.5 group compared to SOY3.0 (log2 fold change + 0.421). The CAST participates in the enriched pathway map “Immune response IL-5 signaling via PI3K, MAPK, and NF-kB” (p-value = 4.770e-2, Fig. 4). Fig. 4Immune response_IL-5_signaling in brain tissue of pigs fed diets containing different levels of soybean oil (SOY1.51 vs. SOY3.02). 1SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY1.5: corn-soybean meal diet containing $3.0\%$ soybean oil. The experimental data is represented by the thermometer-like figure on the map. The upward thermometer (red) indicates up-regulation of the CAST DEG (log2 fold change + 0.421) in the SOY1.5 group compared to SOY3.0. Network objects are represented by individual symbols. The green “T” icon shows which object is associated with the brain tissue. Interactions between objects are represented by arrows, mechanisms, and logical relationships. Further explanations are provided at https://portal.genego.com/legends/MetaCoreQuickReferenceGuide.pdf. To better understand the behavior of the genes and their interactions, process networks were additionally generated by using the MetaCore software. The “Calcium transport” process network (p-value = 2.303e-2), was the only network detected herein, containing the DEG CALB1 (log2 fold change − 3.431) and CAST (log2 fold change + 0.421) (Fig. 5). Fig. 5Calcium transport in brain tissue of pigs fed diets containing different levels of soybean oil (SOY1.51 vs. SOY3.02). 1SOY1.5: corn-soybean meal diet containing $1.5\%$ soybean oil. 2SOY1.5: corn-soybean meal diet containing $3.0\%$ soybean oil. The experimental data are represented by the intensity of the blue and red circles on the network. The blue circle indicates down-regulation of the CALB1 DEG, and the red circle indicates up-regulation of the CAST DEG SOY1.5 group compared to SOY3.0. Green arrows indicate positive interactions, red arrows indicate negative interactions, and gray arrows indicate unspecified interactions. Further explanations are provided at https://portal.genego.com/legends/MetaCoreQuickReferenceGuide.pdf. ## Discussion No changes were identified in the total lipid content and the FA profile between the treatments. The results found in the functional enrichment analysis, demonstrated that the use of different levels of soybean oil alters the transcriptomic profile of pig brain, affecting key processes for the well-functioning of this tissue. For the enriched pathways illustrated in Figs. 1 and 2, the ALOX15B participates in lipid oxidation and peroxidation reactions. According to Stelzer et al. [ 18], among the pathways associated with this gene there were “eicosanoid synthesis” and “arachidonic acid metabolism” and the related Gene Ontology (GO) annotations include “calcium ion binding” and “lipid binding”. Fanalli et al. [ 19] demonstrated that the addition of soybean oil in the diet of pigs in different proportions acted on the modulation of genes, pathway maps and networks associated with inflammation, immune response, oxidative stress, and neurodegenerative diseases, in muscle and liver. Lipoxygenases (LOX) are a family of enzymes responsible for the oxidation of lipids and the generation of a range of metabolites such as eicosanoids and PUFA-related compounds. These metabolites play diverse physiological and pathological roles in inflammatory, neurodegenerative, and cardiovascular diseases, as well as, in defence mechanisms [20, 21]. Lipoxygenases have also been reported in cell differentiation [22, 23], apoptosis [24], and play an important role in the immune response by helping to regulate cytokine secretion [25]. Among the LOX reported in mammals, the ALOX15 isoform may oxygenate complex lipid-protein assemblies found in biomembranes and lipoproteins [26]. The ALOX15 also binds to membranes, with intracellular calcium as a main cofactor for this interaction [27, 28]. It has been reported that ALOX15 is expressed at higher levels in human airway epithelial cells, in eosinophils and immature red blood cells [29]. Furthermore, according to van Leyen et al. [ 30] and Han et al. [ 31], expression and regulation of ALOX15 transcription also occurs in various areas of the brain, but at lower levels. In the study of Shalini et al. [ 32], a higher expression of ALOX15 mRNA was found in the prefrontal cortex. The main product of AA oxygenation by ALOX$\frac{15}{15}$B is 15-hydroxyeicosatetraenoic (15-HETE) [33]. The 15-HETE is considered an important precursor of specialized pro-resolving lipid mediators and is associated with pro- and anti-inflammatory effects [34, 35]. It has also been reported that 15-HETE is a ligand and activator of the peroxisome proliferator-activated receptor gamma (PPAR-γ), which at high concentrations may generate reactive oxygen species in cells [36, 37], and may induce the production of the pro-inflammatory cytokine Interleukin-12 (IL-12) [35, 38]. Zhan et al. [ 39] demonstrated that the application of flaxseed-enriched diet (rich in n-3 PUFA, similar to soybean oil), showed decreases in the expression of pro-inflammatory cytokine genes through activation of PPAR-γ in muscle, adipose tissue and spleen of growing-finishing barrows. Among the results of DHA oxidation by ALOX15, are the specialized pro-resolving lipid mediator D5, a mediator that may be associated in the resolution of inflammation and in the regulation of immune response [40]. Another important mediator related to the resolution of inflammation, reduction of leukocyte trafficking, and negative regulation of cytokine expression is neuroprotectin D1 (NPD1) [41, 42]. NPD1 is reported as an anti-inflammatory molecule, which acts in neuroplasticity and brain signaling, and when in altered conditions, may be found in neuroinflammatory disorders and chronic neurodegeneration [32]. In the study of Chaung et al. [ 43] dietary supplementation of phosphatidylserine and DHA improved antioxidant activity and cognitive function (spatial memory) in rat pups during brain development. Richter et al. [ 44] demonstrated that daily intake of soy-derived phosphatidylserine, had positive effects on cognitive function (learning and memory) in elderly people with impaired memory function. The ALOX15 was found to have increased expression in the brains of Alzheimer’s patients [26, 45, 46]. Praticò et al. [ 45], reported higher levels of $\frac{12}{15}$-LOX and its metabolites $\frac{12}{15}$(S)-HETE in the temporal and frontal brain regions of Alzheimer’s patients. It was further found in in vitro studies using neuronal cells with Alzheimer’s mutation, that $\frac{12}{15}$-LOX is associated with regulation of tau phosphorylation and Aβ plaque production. In addition, regulates synaptic pathology associated with behavioral deficiencies [47, 48]. Additionally, studies have shown that $\frac{12}{15}$-LOX is crucial in Parkinson’s disease [49–51]. According to the research of Li et al. [ 49] and Canals et al. [ 50], activation of these isoforms was associated with a decrease in glutathione concentration (a marker of Parkinson’s disease) in neurons, which may induce nitric oxide neurotoxicity and damage to dopaminergic neurons. The mechanism of action of $\frac{12}{15}$-LOX is still unclear. For example, inhibiting $\frac{12}{15}$-LOX has been shown to reduce reactive oxygen species-induced neuronal cell death [51]. Other studies found that $\frac{12}{15}$-LOX and its metabolites have both pro- and anti-inflammatory effects. This controversial nature of $\frac{12}{15}$-LOX has been reported to be dependent on the metabolites produced, the site of inflammation, and the levels of these metabolites produced [35]. The brain is a tissue that contains a wide range of metabolites and in distinct concentrations. Thus, due to the controversial nature of ALOX15B in metabolic and oxidative processes, further investigations are needed to understand the influence of the downregulation of this DEG in the SOY1.5 group that was found in our study. Further research is required to confirm the action of ALOX15B in the progression of neurodegenerative and inflammatory diseases. For the enriched pathway in Fig. 3, the CALB1 gene binds to intracellular calcium transported via the epithelial calcium channel and transports it across the cytosol toward the basolateral membrane [52]. As a protein-encoding gene that participates in calcium transport, GO annotations for the CALB1 are found to be related to “calcium ion binding” and “vitamin D binding” [18]. It is a highly conserved calcium-binding protein that belongs to a family of high-affinity calcium-binding proteins [53, 54]. Furthermore, studies have shown that the CALB1 gene is highly expressed in brain tissue and found in the majority of neuronal cells and that it is not vitamin D dependent [53–55]. Calcium is one of the most important signaling factors and acts to regulate several important cellular functions such as growth, differentiation, proliferation, cell survival and apoptosis, membrane excitability, and gene transcription. Calcium is also essential for maintaining normal brain function [56]. Thus, the dysregulation of calcium homeostasis and endoplasmic reticulum stress is associated with several pathological conditions such as Parkinson’s, Huntington’s, and Alzheimer’s diseases, and affects numerous signaling pathways [56, 57]. This pathogenic event may also cause amyloidogenesis, energy deficits in neurons, protein aggregation and oxidative stress, and changes in mitochondrial dysfunction, plasticity, and synaptic transmission [58]. Disturbed mitochondrial calcium regulation may also be associated with the link between neuronal dysfunction and disruption of the mitochondria-associated membrane (MAM) contact site of the endoplasmic reticulum and mitochondria, since calcium acts to modulate neurotransmitter release during the synapse [59]. This dysregulation of the MAM-mitochondria linkage dysfunction may also be associated with neurodegenerative diseases such as Alzheimer’s disease [59]. The MAMs are regions of the endoplasmic reticulum that mediate communication between the reticulum and the mitochondria [59, 60]. They are regions that are involved in calcium transport, are responsible for several lipid biosynthetic enzymatic activities, and are also a strategic site for lipid metabolism [59, 61, 62]. According to Vance [59], defects associated with these regions have been identified in neurodegenerative diseases and insulin resistance/type 2 diabetes. The CALB1 helps maintain calcium homeostasis, regulate intracellular calcium responses to physiological stimuli, and modulating synaptic transmission [54]. Another important role of CALB1, is its action in the prevention of neuronal death [54, 63]. The CALB1 also plays an important role in buffering cytosolic calcium and helps prevent lipid peroxidation, through its expression in pancreatic-β cells, by eliminating the production of lipid hydroperoxide, which is induced by proinflammatory cytokines [64]. There is evidence that CALB1 acts to protect neurons against calcium-mediated neurotoxicity and may be considered a cytochemical marker for neuronal plasticity [55]. Decreases in CALB1 expression/concentration in brain tissue has been associated with neurodegeneration in Alzheimer’s, Parkinson’s, and Huntington’s diseases [18, 65] and in ischemic injury studies [66, 67]. Lower CALB1 expression has also been associated with a higher rate of neuronal death [68]. Increased expression of CALB1, on the other hand, has been reported to induce neurite growth in dopaminergic neuronal cells, demonstrating its protective role, especially in neurological diseases, such as Parkinson’s disease [63, 69]. For Alzheimer’s disease, it has been reported that CALB1 has protective effects against the pro-apoptotic action of mutant presenilin 1 (PS-1), attenuating the increase in intracellular calcium and aiding in the prevention of impaired mitochondrial function [70]. PS-1 acts by sensitizing cells to apoptosis induced by Aβ peptide, which damages neurons through a mechanism involving disruption of calcium homeostasis and generation of oxidative stress [70]. Thus, regarding CALB1 down-expression in the SOY1.5 group, we observed that a lower percentage of soybean oil CALB1 gene is less expressed indicating a negative relationship with this diet and a positive relationship with the neurodegenerative processes. For the enriched pathway in Fig. 4, IL-5 activates and elevates the expression of CAST. The CAST binds to and inhibits calpain 1 (mu) in the presence of calcium, which activates and cleaves the apoptosis regulatory protein Bax. The Bax will act by preventing or reducing the frequency, rate, or extent of cell death by apoptotic process [71, 72]. The protein encoded by CAST is an endogenous calpain inhibitor and is also related to the proteolysis of amyloid precursor protein. Furthermore, this protein is thought to influence the expression levels of genes that encode structural or regulatory proteins “Neuroscience” and “neurodegenerative diseases” are two related pathways associated with this gene. Related GO annotations of CAST include “RNA binding” and “cysteine-type endopeptidase inhibitor activity” [18, 73]. The CAST is a cell-permeable peptide that acts as an endogenous inhibitor of calpain in the central nervous system [73, 74]. Calpains are cysteine proteases that are activated by calcium, that is, they are positively regulated by calcium and negatively regulated by CAST [75, 76]. These proteases, when in dysregulation of calcium homeostasis, have been implicated in neuronal cell dysfunction and death [76], as well as neurodegenerative diseases [77–79]. Calpains have several important roles such as differentiation, cell attachment motility, signal transduction covering cell signaling pathways, regulation of gene expression and membrane fusion [73, 75]. Furthermore, calpains are reported to play important roles in neuronal functions, implying that the activation of this protease needs to be under a rigid control, which is performed by CAST. Thus, the well-known calpain-calpastatin system may be an important target for therapeutic approaches related to neurodegenerative diseases [76]. According to Goll et al. [ 75], CAST is also involved in the regulation of kinases, receptors, and transcription factors. CAST expression has been shown to have a neuroprotective effect on cerebral ischemia [80]. In the study of Rao et al. [ 81], higher expression of CAST in JNPL3 (mutant tau P301L) mouse models was used to attenuate calpain expression, which has been linked to the development of tauopathy (neurotoxicity caused by tau protein) and neurodegeneration in Alzheimer’s disease. In an Amyotrophic Lateral Sclerosis mouse model, higher CAST expression was associated with neuroprotective effects. According to Rao et al. [ 82], the CAST gene reduces calpain activation, decreases abnormal cytoskeletal protein breakdown, increases survival time, inhibits tau production and CDK5 activation, and decreases SOD1. The calpain-calpastatin system is also reported in excitotoxicity, a pathological or neurodegenerative process that is initiated by overactivation of neurotransmitters such as glutamate. Excitotoxicity leads to increased cellular calcium levels, which causes activation of various proteases, including calpains [83]. Furthermore, missing CAST may impair early stages of neurogenesis [84]. Thus, we observed a higher expression of CAST in the SOY1.5 group, that suggests a positive relationship between the gene and the metabolic and oxidative processes found for this group. The identified network, along with the illustrated genes, corroborate the results found in the pathway maps, indicating that varying the amount of soybean oil in the diet of immunocastrated male pigs influences gene expression in brain tissue. Furthermore, the significance of the detected DEG and their association with intracellular calcium is noteworthy. This processes network (Calcium transport) and the genes enriched in this network corroborate the results found in the pathway maps, indicating that changing the level of soybean oil in pigs’ diet has an effect on gene expression. Therefore, the findings of our study point in a promising direction for furthering our understanding of the pathways and networks associated with calcium-dependent metabolic processes involved in lipid metabolism and oxidative processes. More research is needed to better understand the mechanisms by which dietary factors like FA may influence important physiological processes and gene expression in brain tissue. Understanding the mechanisms involved in calcium homeostasis and energy metabolism in the initiation and progression of neurodegenerative diseases and oxidative/inflammatory processes is extremely important. ## Conclusion This study showed that different levels of soybean oil in pig diets affect the transcriptomic profile but not the total lipid content or FA profile of brain tissue. The genes, pathways, and networks identified herein play important roles in lipid metabolism, immune response, and calcium transport. Furthermore, because pigs are model animals for human metabolic diseases, the DEG identified, as well as their action in brain tissue, demonstrate the importance of FA in metabolic and oxidative processes. Thus, the current study may help future research in the field of nutrigenomics and help to better understand how the diet, with the inclusion of soybean oil, may influence and modulate biological processes important for brain tissue. Further investigation is required to define what proportion of soybean oil helps in directing and modulating neuroprotection and reducing inflammation in brain tissue. ## Ethics Statement The procedures involving animals were evaluated and approved by the Ethics Committee for the Use of Animals (CEUA, number 2018-28, and protocol 2018.5.1787.11.6) of the Luiz de Queiroz College of Agriculture (ESALQ) – University of São Paulo (USP). All procedures followed the guidelines by the Brazilian Council of Animal Experimentation and the ethical principles in animal research, according to FASS [85], the Guide for the Care and Use of Agricultural Animals in Agricultural Research and Teaching. This study was carried out in compliance with the ARRIVE guidelines. ## Animals, experimental design, and diets Thirty-six immunocastrated male pigs, the offspring of three sires and thirty-two females of the Large White breed, were used for this study. Pigs were genotyped for the halothane mutation (RYR1 gene) and only homozygous halothane-negative (NN) were used [86]. The pigs had an average body weight of 28.44 ± 2.95 kg and an average age of 71 ± 1.8 days, and were randomly distributed to the treatments during the experimental period of 98 days. Two treatments were used, with six replicate pens per treatment, and three pigs per pen, totalizing 18 pigs per treatment. The pigs ad libitum access to feed and water throughout the experimental period, and each pen was equipped with a dry feeder and a nipple drinker. The immunocastration was performed by administering two doses of 2 ml of Vivax® (Pfizer Animal Health, Parkville, Australia) on day 56 (127 days of age) and day 70 (141 days of age), according to the manufacturer’s recommendations. The experimental diet consisted of a six-phase diet: Grower I - day 0 to 21; Grower II - day 21 to 42; Finisher I - day 42 to 56; Finisher II - day 56 to 63; Finisher III - day 63 to 70; and, Finisher IV - day 70 to 98. Dietary treatments consisted of corn-soybean meal diets either containing $1.5\%$ soybean oil (SOY1.5), a standard diet used in pig production, or containing $3.0\%$ soybean oil (SOY3.0). The diets were formulated to meet or exceed the nutritional requirements according to Rostagno et al. [ 87], and were provided as a meal form, without antibiotic growth promoters. The diets were formulated to have a similar metabolizable energy content (3.36 Mcal/kg). Details of the diets in this study are adapted and described in Tables S4–S6 [19, 88, 89]. The pigs were slaughtered with a final body weight of 133.9 ± 9.4 kg on day 98 of the experiment. Whole brains of the animals were collected and immediately frozen in liquid nitrogen, transported, and stored in a -80 °C freezer until total RNA extraction. The same portion of the middle region of the frontal lobe was delimited in all brain samples in order to obtain a sample as uniform as possible with the same proportion of white and gray matter and the layers. Complete procedures were described in Almeida et al. [ 88] and Silva et al. [ 90]. ## Total lipid content and FA profile analyses For the analysis of total lipid content, 5 g of brain samples were used (in duplicate), which were ground, packed in plastic bags and stored at 4ºC. The ground samples were dried in an oven with air circulation at 105 °C for 12 h. After drying, the samples were packed in filter paper cartridges and placed in a Soxhlet type extraction system. The extraction was conducted with hexane and occurred during six hours, according to the method described by AOAC [91]. The percentage of total lipid in the samples was obtained by the difference between the weight of the flask containing the extracted lipid and the empty flask (previously weighed, the flask was left in an oven at 105 °C for 2 h before each weighing) multiplied by 100. The FA profile was determined from the total lipid content using 10 g samples. The lipids were cold extracted using the method proposed by Bligh Dyer [92] and the methylation of the samples was performed according to Hartman e Lago [93], with adaptations based on AOCS [94] (method AM 5 − 04). The complete procedures were described by Silva et al. [ 90] and Almeida et al. [ 88]. Data were analyzed as a randomized complete block design using the MIXED procedure of SAS (SAS Inst. Inc., Cary, NC), with pen being considered as the experimental unit. The model included the random effects of pen and block and the fixed effects of soybean oil levels. Outliers were removed from the data sets and residuals were tested for a normal distribution using the Shapiro-Wilk test (UNIVARIATE procedure). Means were adjusted by using the LSMEANS statement. Differences were declared significant when p-value ≤ 0.05 based on the F-test. ## RNA extraction, library preparation, sequencing and data analysis For the total RNA extraction from the brain samples, we used the commercial kit for RNA extraction (RNeasy® Mini Kit, Qiagen) and the Trizol reagent (Invitrogen). The inclusion of first step using the Trizol, allowed for better phase separation and thus lipid removal as brain tissue has a large amount of lipids (~ $10\%$). Quality and concentration of total RNA was obtained by using the NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific) and Qubit® 2.0 Fluorometer. The RNA integrity was evaluated by using the Agilent 2100 Bioanalyzer (Agilent, Santa Clara-CA, USA). All samples presented an RNA Integrity Number (RIN) greater than or equal to 7.5 (Table S3). For library preparation, 2 µL of total RNA from each sample was used, according to the protocol described in the TruSeq RNA Sample Preparation kit v2 manual (Illumina, San Diego, CA). The average library size was estimated using the Agilent Bioanalyzer 2100 (Agilent, Santa Clara, CA, USA) and the libraries were quantified using quantitative PCR with the quantification kit, from the KAPA library (KAPA Biosystems, Foster City, CA, USA). TruSeq PE Cluster kit v3-cBot-HS (Illumina, San Diego, CA, USA) was used for the sequencing. The samples were pooled and sequenced by using the HiSeq 2500 equipment (Illumina, San Diego, CA, USA) with a TruSeq SBS Kit v3-HS (200 cycles), according to the manufacturer’s instructions. All sequencing steps were performed at the ESALQ-USP Animal Genomics Center, located in the Animal Biotechnology Laboratory of ESALQ-USP, Piracicaba, São Paulo, Brazil. For the steps of quality control, low complexity reads and adapters were removed using Trim Galore software (v.0.6.5). The minimum length of reads after removal was 70 bases, with Phred Score lower than 33. Quality control was done by using FastQC software (v.0.11.8) [http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/]. The reference genome used was the Sus Scrofa 11.1, available from Ensembl [http://www.ensembl.org/Sus_scrofa/Info/Index]. Alignment, mapping, and abundance (read counts) of mRNAs for all known-genes was performed using STAR software (v.2.7.6a) [95], and the gene expression levels were normalized using the counts scaled by total number of reads or counts per million (CPM). ## Identification of differentially expressed genes, and functional enrichment analysis The differentially expressed genes (DEG) between the SOY1.5 and SOY3.0 groups were identified by using the DESeq2 statistical package (R/Bioconductor) [http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html], using a multi-factor design [96]. Before statistical analysis, filtering criteria were used: (i) removal of genes with zero counts for all samples, that is, unexpressed genes, (ii) removal of genes with less than one read per sample on average were removed (very lowly expressed); (iii) removal of genes that were not present in at least $50\%$ of the samples were removed (rarely expressed). The model used, included treatments as the variable of interest and father as a fixed effect. Correction for multiple testing was performed, according to the False Discovery Rate (FDR) method [97], and the threshold value used for significance was FDR < 0.05. The enrichment analysis was performed using the MetaCore software (Clarivate Analytics, London, UK, v.21.4, build 70,700) [https://clarivate.com/products/metacore/]. The pathway maps were identified from the list of known-genes DEG obtained from SOY1.5 vs. SOY3.0 (FDR < 0.05) comparison. For annotation and functional enrichment, the Homo sapiens genome was used as background reference. Functional enrichment analysis to obtain comparative pathways and networks was performed, using the standard parameter. The filters for the metabolic maps of interest were used: energy metabolism, lipid metabolism, steroid metabolism, regulation of cellular processes (immune response, neurophysiological process, and oxidative stress), regulation of metabolism, mental disorders, nutritional and metabolic diseases, nervous system diseases, and tox processes. To understand the behavior of genes and their interactions, networks were created. ## Electronic supplementary material Below is the link to the electronic supplementary material. Additional file 1. Table S1. Total reads in brain samples of pigs fed diets containing different levels of soybean oil. Additional file 2. Table S2. Differentially expressed genes in brain tissue of pigs fed diets containing different levels of soybean oil. Additional file 3. Table S3. Quality, concentrattion and RNA integrity number of brain samples of pigs fed diets containing different levels of soybean oil. Additional file 4. Table S4. Composition of the experimental diets (as-fed basis). Table S5. Analyzed fatty acid profile of grower diets (as-fed basis). Table S6. Analyzed fatty acid profile of finisher diets (as-fed basis). ## References 1. 1.OECD-FAO. OECD-FAO Agricultural Outlook 2021–2030. OECD; 2021. 2. Pan Z, Yao Y, Yin H, Cai Z, Wang Y, Bai L. Pig genome functional annotation enhances the biological interpretation of complex traits and human disease. Nat Commun (2021.0) 12 5848. DOI: 10.1038/s41467-021-26153-7 3. Lind NM, Moustgaard A, Jelsing J, Vajta G, Cumming P, Hansen AK. The use of pigs in neuroscience: modeling brain disorders. Neurosci Biobehav Rev (2007.0) 31 728-51. DOI: 10.1016/j.neubiorev.2007.02.003 4. Lunney JK. Advances in Swine Biomedical Model Genomics. Int J Biol Sci (2007.0) 3 179-84. DOI: 10.7150/ijbs.3.179 5. 5.Dawson H, Chen C, Wang T, Urban J. Comparative Nutrigenomics Analysis of the Pig, Mouse and Human (P15-004-19).Curr Dev Nutr. 2019;3. 6. 6.Hoffe B, Holahan MR. The Use of Pigs as a Translational Model for Studying Neurodegenerative Diseases.Front Physiol. 2019;10. 7. Hsu M-C, Huang Y-S, Ouyang W-C. Beneficial effects of omega-3 fatty acid supplementation in schizophrenia: possible mechanisms. Lipids Health Dis (2020.0) 19 159. DOI: 10.1186/s12944-020-01337-0 8. Hamilton JA, Hillard CJ, Spector AA, Watkins PA. Brain uptake and utilization of fatty acids, lipids and lipoproteins: application to Neurological Disorders. J Mol Neurosci (2007.0) 33 2-11. DOI: 10.1007/s12031-007-0060-1 9. 9.Bruce KD, Zsombok A, Eckel RH. Lipid Processing in the Brain: A Key Regulator of Systemic Metabolism.Front Endocrinol (Lausanne). 2017;8 APR. 10. Lu P, Zhang LY, Yin JD, Everts AKR, Li DF. Effects of soybean oil and linseed oil on fatty acid compositions of muscle lipids and cooked pork flavour. Meat Sci (2008.0) 80 910-8. DOI: 10.1016/j.meatsci.2008.04.010 11. Alencar SAdaS, Kiefer C, Nascimento KMR, de Viana S, Corassa LH, de Gomes A. Dietary soybean oil modulates fatty acid composition of pork. Trop Anim Health Prod (2021.0) 53 357. DOI: 10.1007/s11250-021-02804-1 12. 12.Stein HH, Berger LL, Drackley JK, Fahey GC, Hernot DC, Parsons CM. Nutritional Properties and feeding values of soybeans and their coproducts. In: Soybeans. Elsevier; 2008. pp. 613–60. 13. 13.Fan L, Eskin NAM. The use of antioxidants in the preservation of edible oils. Handbook of Antioxidants for Food Preservation.Elsevier; 2015.pp. 373–88. 14. Sun GY, Simonyi A, Fritsche KL, Chuang DY, Hannink M, Gu Z. Docosahexaenoic acid (DHA): an essential nutrient and a nutraceutical for brain health and diseases. Prostaglandins Leukot Essent Fatty Acids (2018.0) 136 3-13. DOI: 10.1016/j.plefa.2017.03.006 15. Rapoport SI. Translational studies on regulation of brain docosahexaenoic acid (DHA) metabolism in vivo. Prostaglandins Leukot Essent Fatty Acids (2013.0) 88 79-85. DOI: 10.1016/j.plefa.2012.05.003 16. Bazinet RP, Layé S. Polyunsaturated fatty acids and their metabolites in brain function and disease. Nat Rev Neurosci (2014.0) 15 771-85. DOI: 10.1038/nrn3820 17. Ulbricht TLV, Southgate DAT. Coronary heart disease: seven dietary factors. The Lancet (1991.0) 338 985-92. DOI: 10.1016/0140-6736(91)91846-M 18. Stelzer G, Rosen N, Plaschkes I, Zimmerman S, Twik M, Fishilevich S. The GeneCards suite: from Gene Data Mining to Disease Genome sequence analyses. Curr Protoc Bioinformatics (2016.0) 54 1-33. DOI: 10.1002/cpbi.5 19. Fanalli SL, da Silva BPM, Gomes JD, de Almeida VV, Freitas FAO, Moreira GCM. Differential Gene expression Associated with soybean oil level in the Diet of Pigs. Animals (2022.0) 12 1632. DOI: 10.3390/ani12131632 20. Gertow K, Nobili E, Folkersen L, Newman JW, Pedersen TL, Ekstrand J. 12- and 15-lipoxygenases in human carotid atherosclerotic lesions: Associations with cerebrovascular symptoms. Atherosclerosis (2011.0) 215 411-6. DOI: 10.1016/j.atherosclerosis.2011.01.015 21. Karatas H, Cakir-Aktas C. 12/15 lipoxygenase as a therapeutic target in Brain Dosorders. Archives of Neuropsychiatry (2019.0) 56 288-91. PMID: 31903039 22. van Leyen K, Duvoisin RM, Engelhardt H, Wiedmann M. A function for lipoxygenase in programmed organelle degradation. Nature (1998.0) 395 392-5. DOI: 10.1038/26500 23. 23.Adel S, Karst F, González-Lafont À, Pekárová M, Saura P, Masgrau L et al. Evolutionary alteration of ALOX15 specificity optimizes the biosynthesis of antiinflammatory and proresolving lipoxins. Proceedings of the National Academy of Sciences. 2016;113:E4266-75. 24. Claria J. Regulation of cell proliferation and apoptosis by bioactive lipid mediators. Recent Pat Anticancer Drug Discov (2006.0) 1 369-82. DOI: 10.2174/157489206778776961 25. Danielsson KN, Rydberg EK, Ingelsten M, Akyürek LM, Jirholt P, Ullström C. 15-Lipoxygenase-2 expression in human macrophages induces chemokine secretion and T cell migration. Atherosclerosis (2008.0) 199 34-40. DOI: 10.1016/j.atherosclerosis.2007.10.027 26. Ivanov I, Kuhn H, Heydeck D. Structural and functional biology of arachidonic acid 15-lipoxygenase-1 (ALOX15). Gene (2015.0) 573 1-32. DOI: 10.1016/j.gene.2015.07.073 27. Watson A, Doherty FJ. Calcium promotes membrane association of reticulocyte 15-lipoxygenase. Biochem J (1994.0) 298 377-83. DOI: 10.1042/bj2980377 28. Brinckmann R, Schnurr K, Heydeck D, Rosenbach T, Kolde G, Kühn H. Membrane translocation of 15-Lipoxygenase in hematopoietic cells is calcium-dependent and activates the oxygenase activity of the enzyme. Blood (1998.0) 91 64-74. DOI: 10.1182/blood.V91.1.64 29. Nadel JA, Conrad DJ, Ueki IF, Schuster A, Sigal E. Immunocytochemical localization of arachidonate 15-lipoxygenase in erythrocytes, leukocytes, and airway cells. J Clin Invest (1991.0) 87 1139-45. DOI: 10.1172/JCI115110 30. van Leyen K, Kim HY, Lee S-R, Jin G, Arai K, Lo EH. Baicalein and 12/15-Lipoxygenase in the ischemic brain. Stroke (2006.0) 37 3014-8. DOI: 10.1161/01.STR.0000249004.25444.a5 31. Han J, Sun L, Xu Y, Liang H, Cheng Y. Activation of PPARγ by 12/15-lipoxygenase during cerebral ischemia-reperfusion injury. Int J Mol Med (2015.0) 35 195-201. DOI: 10.3892/ijmm.2014.1998 32. Shalini S-M, Ho CF-Y, Ng Y-K, Tong J-X, Ong E-S, Herr DR. Distribution of Alox15 in the rat brain and its role in prefrontal cortical resolvin D1 formation and spatial Working Memory. Mol Neurobiol (2018.0) 55 1537-50. DOI: 10.1007/s12035-017-0413-x 33. 33.Snodgrass RG, Brüne B. Regulation and Functions of 15-Lipoxygenases in Human Macrophages.Front Pharmacol. 2019;10. 34. Kutzner L, Goloshchapova K, Heydeck D, Stehling S, Kuhn H, Schebb NH. Mammalian ALOX15 orthologs exhibit pronounced dual positional specificity with docosahexaenoic acid. Biochim Biophys Acta (2017.0) 1862 666-75. DOI: 10.1016/j.bbalip.2017.04.001 35. Singh NK, Rao GN. Emerging role of 12/15-Lipoxygenase (ALOX15) in human pathologies. Prog Lipid Res (2019.0) 73 28-45. DOI: 10.1016/j.plipres.2018.11.001 36. Huang JT, Welch JS, Ricote M, Binder CJ, Willson TM, Kelly C. Interleukin-4-dependent production of PPAR-γ ligands in macrophages by 12/15-lipoxygenase. Nature (1999.0) 400 378-82. DOI: 10.1038/22572 37. Sun L, Xu Y-W, Han J, Liang H, Wang N, Cheng Y. 12/15-Lipoxygenase metabolites of arachidonic acid activate PPARγ: a possible neuroprotective effect in ischemic brain. J Lipid Res (2015.0) 56 502-14. DOI: 10.1194/jlr.M053058 38. Li J, Rao J, Liu Y, Cao Y, Zhang Y, Zhang Q. 15-Lipoxygenase promotes chronic Hypoxia–Induced Pulmonary artery inflammation via positive Interaction with Nuclear Factor-κB. Arterioscler Thromb Vasc Biol (2013.0) 33 971-9. DOI: 10.1161/ATVBAHA.113.301335 39. Zhan ZP, Huang FR, Luo J, Dai JJ, Yan XH, Peng J. Duration of feeding linseed diet influences expression of inflammation-related genes and growth performance of growing-finishing barrows. J Anim Sci (2009.0) 87 603-11. DOI: 10.2527/jas.2007-0177 40. Perry SC, Kalyanaraman C, Tourdot BE, Conrad WS, Akinkugbe O, Freedman JC. 15-Lipoxygenase-1 biosynthesis of 7S,14S-diHDHA implicates 15-lipoxygenase-2 in biosynthesis of resolvin D5. J Lipid Res (2020.0) 61 1087-103. DOI: 10.1194/jlr.RA120000777 41. Hong S, Gronert K, Devchand PR, Moussignac R-L, Serhan CN. Novel docosatrienes and 17S-Resolvins generated from Docosahexaenoic Acid in Murine Brain, Human Blood, and glial cells. J Biol Chem (2003.0) 278 14677-87. DOI: 10.1074/jbc.M300218200 42. Kohli P, Levy BD. Resolvins and protectins: mediating solutions to inflammation. Br J Pharmacol (2009.0) 158 960-71. DOI: 10.1111/j.1476-5381.2009.00290.x 43. Chaung H-C, Chang C-D, Chen P-H, Chang C-J, Liu S-H, Chen C-C. Docosahexaenoic acid and phosphatidylserine improves the antioxidant activities in vitro and in vivo and cognitive functions of the developing brain. Food Chem (2013.0) 138 342-7. DOI: 10.1016/j.foodchem.2012.10.082 44. Richter Y, Herzog Y, Lifshitz Y, Hayun R, Zchut S. The effect of soybean-derived phosphatidylserine on cognitive performance in elderly with subjective memory complaints: a pilot study. Clin Interv Aging (2013.0) 8 557. PMID: 23723695 45. Praticò D, Zhukareva V, Yao Y, Uryu K, Funk CD, Lawson JA. 12/15-Lipoxygenase is increased in Alzheimer’s Disease. Am J Pathol (2004.0) 164 1655-62. DOI: 10.1016/S0002-9440(10)63724-8 46. Yang H, Zhuo J-M, Chu J, Chinnici C, Praticò D. Amelioration of the Alzheimer’s Disease phenotype by absence of 12/15-Lipoxygenase. Biol Psychiatry (2010.0) 68 922-9. DOI: 10.1016/j.biopsych.2010.04.010 47. Succol F, Praticò D. A role for 12/15 lipoxygenase in the amyloid betha precursor protein metabolism. J Neurochem (2007.0) 103 380-7. PMID: 17877641 48. Joshi YB, Giannopoulos PF, Praticò D. The 12/15-lipoxygenase as an emerging therapeutic target for Alzheimer’s disease. Trends Pharmacol Sci (2015.0) 36 181-6. DOI: 10.1016/j.tips.2015.01.005 49. Li Y, Maher P, Schubert D. A role for 12-lipoxygenase in nerve cell death caused by glutathione depletion. Neuron (1997.0) 19 453-63. DOI: 10.1016/S0896-6273(00)80953-8 50. Canals S, Casarejos MJ, de Bernardo S, Rodríguez-Martín E, Mena MA. Nitric oxide triggers the toxicity due to glutathione depletion in midbrain cultures through 12-Lipoxygenase. J Biol Chem (2003.0) 278 21542-9. DOI: 10.1074/jbc.M213174200 51. Zhang Y. Peroxynitrite-Induced neuronal apoptosis is mediated by Intracellular Zinc Release and 12-Lipoxygenase activation. J Neurosci (2004.0) 24 10616-27. DOI: 10.1523/JNEUROSCI.2469-04.2004 52. Lambers TT, Mahieu F, Oancea E, Hoofd L, de Lange F, Mensenkamp AR. Calbindin-D28K dynamically controls TRPV5-mediated Ca2 + transport. EMBO J (2006.0) 25 2978-88. DOI: 10.1038/sj.emboj.7601186 53. Christakos S, Gabrielides C, Rhoten WB. Vitamin D-Dependent calcium binding proteins: Chemistry, distribution, functional considerations, and Molecular Biology. Endocr Rev (1989.0) 10 3-26. DOI: 10.1210/edrv-10-1-3 54. 54.Christakos S, Mady LJ, Dhawan P. Calbindin-D 28K and Calbindin-D 9K and the epithelial calcium channels TRPV5 and TRPV6. In: Vitamin D. Elsevier; 2018. pp. 343–59. 55. 55.Ferrante RJ, Kowall NW, Richardson EP. Neuronal Plasticity: A Potential Role in Spiny Striatal Neuron Degeneration in Huntington’s Disease. Trophic Regulation of the Basal Ganglia.Elsevier; 1994.pp. 465–77. 56. Genovese I, Giamogante F, Barazzuol L, Battista T, Fiorillo A, Vicario M. Sorcin is an early marker of neurodegeneration, Ca2 + dysregulation and endoplasmic reticulum stress associated to neurodegenerative diseases. Cell Death Dis (2020.0) 11 861. DOI: 10.1038/s41419-020-03063-y 57. Supnet C, Bezprozvanny I. Neuronal Calcium Signaling, mitochondrial dysfunction, and Alzheimer’s Disease. J Alzheimer’s Disease (2010.0) 20 487-98. DOI: 10.3233/JAD-2010-100306 58. Lindholm D, Wootz H, Korhonen L. ER stress and neurodegenerative diseases. Cell Death Differ (2006.0) 13 385-92. DOI: 10.1038/sj.cdd.4401778 59. Vance JE. MAM (mitochondria-associated membranes) in mammalian cells: lipids and beyond. Biochim Biophys Acta (2014.0) 1841 595-609. DOI: 10.1016/j.bbalip.2013.11.014 60. 60.Yang M, Li C, Yang S, Xiao Y, Xiong X, Chen W et al. Mitochondria-Associated ER Membranes – The Origin Site of Autophagy.Front Cell Dev Biol. 2020;8. 61. Arruda AP, Hotamisligil GS. Calcium homeostasis and organelle function in the pathogenesis of obesity and diabetes. Cell Metab (2015.0) 22 381-97. DOI: 10.1016/j.cmet.2015.06.010 62. Lee S, Min KT. The interface between ER and mitochondria: molecular compositions and functions. Mol Cells (2018.0) 41 1000-7. PMID: 30590907 63. Castrogiovanni P, Sanfilippo C, Imbesi R, Maugeri G, lo Furno D, Tibullo D. Brain CHID1 expression correlates with NRGN and CALB1 in healthy subjects and AD patients. Cells (2021.0) 10 882. DOI: 10.3390/cells10040882 64. Rabinovitch A, Suarez-Pinzon WL, Sooy K, Strynadka K, Christakos S. Expression of Calbindin-D 28k in a pancreatic islet β-Cell line protects against Cytokine-Induced apoptosis and necrosis. Endocrinology (2001.0) 142 3649-55. DOI: 10.1210/endo.142.8.8334 65. 65.Iacopino AM, Christakos S. Specific reduction of calcium-binding protein (28-kilodalton calbindin-D) gene expression in aging and neurodegenerative diseases. Proceedings of the National Academy of Sciences. 1990;87:4078–82. 66. Burke RE, Baimbridge KG. Relative loss of the striatal striosome compartment, defined by calbindin-D28k immunostaining, following developmental hypoxic-ischemic injury. Neuroscience (1993.0) 56 305-15. DOI: 10.1016/0306-4522(93)90333-B 67. Ahmadian SS, Rezvanian A, Peterson M, Weintraub S, Bigio EH, Mesulam M-M. Loss of calbindin-D 28K is associated with the full range of tangle pathology within basal forebrain cholinergic neurons in Alzheimer’s disease. Neurobiol Aging (2015.0) 36 3163-70. DOI: 10.1016/j.neurobiolaging.2015.09.001 68. Kook S-Y, Jeong H, Kang MJ, Park R, Shin HJ, Han S-H. Crucial role of calbindin-D28k in the pathogenesis of Alzheimer’s disease mouse model. Cell Death Differ (2014.0) 21 1575-87. DOI: 10.1038/cdd.2014.67 69. Yuan H-H, Chen R-J, Zhu Y-H, Peng C-L, Zhu X-R. The neuroprotective effect of overexpression of Calbindin-D28k in an animal model of Parkinson’s Disease. Mol Neurobiol (2013.0) 47 117-22. DOI: 10.1007/s12035-012-8332-3 70. 70.Guo Q, Christakos S, Robinson N, Mattson MP. Calbindin D28k blocks the proapoptotic actions of mutant presenilin 1: Reduced oxidative stress and preserved mitochondrial function. Proceedings of the National Academy of Sciences. 1998;95:3227–32. 71. Shen Z-J, Esnault S, Schinzel A, Borner C, Malter JS. The peptidyl-prolyl isomerase Pin1 facilitates cytokine-induced survival of eosinophils by suppressing bax activation. Nat Immunol (2009.0) 10 257-65. DOI: 10.1038/ni.1697 72. 72.Ilmarinen P, Moilanen E, Kankaanranta H. Regulation of Spontaneous Eosinophil Apoptosis—A Neglected Area of Importance.J Cell Death. 2014;7:JCD.S13588. 73. Li Y, Bondada V, Joshi A, Geddes JW. Calpain 1 and calpastatin expression is developmentally regulated in rat brain. Exp Neurol (2009.0) 220 316-9. DOI: 10.1016/j.expneurol.2009.09.004 74. Teng F, Yin Y, Guo J, Jiang M. Calpastatin peptide attenuates early brain injury following experimental subarachnoid hemorrhage. Exp Ther Med (2020.0) 19 2433-40. PMID: 32226486 75. Goll DE, Thompson VF, Li H, Wei W, Cong J. The calpain system. Physiol Rev (2003.0) 83 731-801. DOI: 10.1152/physrev.00029.2002 76. Stifanese R, Averna M, de Tullio R, Pedrazzi M, Beccaria F, Salamino F. Adaptive modifications in the Calpain/Calpastatin system in Brain cells after persistent alteration in Ca2 + homeostasis. J Biol Chem (2010.0) 285 631-43. DOI: 10.1074/jbc.M109.031674 77. Ray S. Currently evaluated calpain and caspase inhibitors for Neuroprotection in Experimental Brain Ischemia. Curr Med Chem (2006.0) 13 3425-40. DOI: 10.2174/092986706779010342 78. Vosler PS, Brennan CS, Chen J. Calpain-mediated signaling mechanisms in neuronal Injury and Neurodegeneration. Mol Neurobiol (2008.0) 38 78-100. DOI: 10.1007/s12035-008-8036-x 79. Jourdi H, Hamo L, Oka T, Seegan A, Baudry M. BDNF mediates the neuroprotective effects of positive AMPA receptor modulators against MPP+-induced toxicity in cultured hippocampal and mesencephalic slices. Neuropharmacology (2009.0) 56 876-85. DOI: 10.1016/j.neuropharm.2009.01.015 80. Rami A, Volkmann T, Agarwal R, Schoninger S, Nürnberger F, Saido TC. β2-Adrenergic receptor responsiveness of the calpain–calpastatin system and attenuation of neuronal death in rat hippocampus after transient global ischemia. Neurosci Res (2003.0) 47 373-82. DOI: 10.1016/j.neures.2003.07.002 81. Rao M, McBrayer MK, Campbell J, Kumar A, Hashim A, Sershen H. Specific calpain inhibition by calpastatin prevents Tauopathy and Neurodegeneration and restores normal lifespan in tau P301L mice. J Neurosci (2014.0) 34 9222-34. DOI: 10.1523/JNEUROSCI.1132-14.2014 82. Rao M, Campbell J, Palaniappan A, Kumar A, Nixon RA. Calpastatin inhibits motor neuron death and increases survival of hSOD1 G93A mice. J Neurochem (2016.0) 137 253-65. DOI: 10.1111/jnc.13536 83. Neumar RW, Meng FH, Mills AM, Xu YA, Zhang C, Welsh FA. Calpain activity in the rat brain after transient forebrain ischemia. Exp Neurol (2001.0) 170 27-35. DOI: 10.1006/exnr.2001.7708 84. 84.Machado VM, Morte MI, Carreira BP, Azevedo MM, Takano J, Iwata N et al. Involvement of calpains in adult neurogenesis: implications for stroke. Front Cell Neurosci. 2015;9 FEB. 85. 85.FASS. Guide for the care and use of agricultural animals in agricultural research and teaching. Champaingn, IL, US; 2010. 86. 86.Fujii J, Otsu K, Zorzato F, de Leon S, Khanna VK, Weiler JE et al. Identification of a Mutation in Porcine Ryanodine Receptor Associated with Malignant Hyperthermia. Science (1979). 1991;253:448–51. 87. 87.Rostagnolo HS, Albino LFT, Donzele JL, Gomes PC, de Oliveira RF, Lopes DC et al. Tabelas Brasileiras Para Aves e Suínos: Composição de Alimentos e Exigências Nutricionais. 3rd edition. Viçosa, MG: Universidade Federal de Viçosa; 2011. 88. Almeida VV, Silva JPM, Schinckel AP, Meira AN, Moreira GCM, Gomes JD. Effects of increasing dietary oil inclusion from different sources on growth performance, carcass and meat quality traits, and fatty acid profile in genetically lean immunocastrated male pigs. Livest Sci (2021.0) 248 104515. DOI: 10.1016/j.livsci.2021.104515 89. Fanalli SL, da Silva BPM, Gomes JD, Ciconello FN, de Almeida VV, Freitas FAO. Effect of dietary soybean oil inclusion on liver-related transcription factors in a pig model for metabolic diseases. Sci Rep (2022.0) 12 10318. DOI: 10.1038/s41598-022-14069-1 90. da Silva JPM, Costa KA, Almeida VV, Coutinho LL, da Silva BPM, Cesar ASM. Fatty acid profile in brain and hepatic tissues from pigs supplemented with canola oil. Revista Brasileira de Agrotecnologia (2021.0) 11 414-20. DOI: 10.18378/REBAGRO.V12I2.8736 91. 91.AOAC. Official methods of analysis of the Association os Official Analytical Chemists.Association of Official Analytical Chemists. 1995;2. 92. Bligh EG, Dyer WJ. A rapid method of total extration and purification. Can J Biochem Physiol (1959.0) 37 911-7. DOI: 10.1139/y59-099 93. Hartman L, Lago RC. Rapid preparation of fatty acid methyl esters from lipids. Lab Pract (1973.0) 22 475-81. PMID: 4727126 94. 94.AOCS. Official approved procedure am 5 – 04, Rapid determination of oil/fat utilizing high temperature solvent extraction. American Oil Chemists’ Society; 2005. 95. 95.Dobin A, Gingeras TR. Mapping RNA-seq Reads with STAR. Curr Protoc Bioinformatics. 2015;51:11.14.1-11.14.19. 96. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol (2014.0) 15 550. DOI: 10.1186/s13059-014-0550-8 97. Benjamini Y, Hochberg Y. Controlling the false Discovery rate: a practical and powerful Approach to multiple testing. J Roy Stat Soc: Ser B (Methodol) (1995.0) 57 289-300
--- title: Impact of ammonia levels on outcome in clinically stable outpatients with advanced chronic liver disease authors: - Lorenz Balcar - Julia Krawanja - Bernhard Scheiner - Rafael Paternostro - Benedikt Simbrunner - Georg Semmler - Mathias Jachs - Lukas Hartl - Albert Friedrich Stättermayer - Philipp Schwabl - Matthias Pinter - Thomas Szekeres - Michael Trauner - Thomas Reiberger - Mattias Mandorfer journal: JHEP Reports year: 2023 pmcid: PMC9976454 doi: 10.1016/j.jhepr.2023.100682 license: CC BY 4.0 --- # Impact of ammonia levels on outcome in clinically stable outpatients with advanced chronic liver disease ## Abstract ### Background & Aims Ammonia levels predicted hospitalisation in a recent landmark study not accounting for portal hypertension and systemic inflammation severity. We investigated (i) the prognostic value of venous ammonia levels (outcome cohort) for liver-related outcomes while accounting for these factors and (ii) its correlation with key disease-driving mechanisms (biomarker cohort). ### Methods (i) The outcome cohort included 549 clinically stable outpatients with evidence of advanced chronic liver disease. ( ii) The partly overlapping biomarker cohort comprised 193 individuals, recruited from the prospective Vienna Cirrhosis Study (VICIS: NCT03267615). ### Results (i) In the outcome cohort, ammonia increased across clinical stages as well as hepatic venous pressure gradient and United Network for Organ Sharing model for end-stage liver disease [2016] strata and were independently linked with diabetes. Ammonia was associated with liver-related death, even after multivariable adjustment (adjusted hazard ratio [aHR]: 1.05 [$95\%$ CI: 1.00–1.10]; $$p \leq 0.044$$). The recently proposed cut-off (≥1.4 × upper limit of normal) was independently predictive of hepatic decompensation (aHR: 2.08 [$95\%$ CI: 1.35–3.22]; $p \leq 0.001$), non-elective liver-related hospitalisation (aHR: 1.86 [$95\%$ CI: 1.17–2.95]; $$p \leq 0.008$$), and – in those with decompensated advanced chronic liver disease – acute-on-chronic liver failure (aHR: 1.71 [$95\%$ CI: 1.05–2.80]; $$p \leq 0.031$$). ( ii) Besides hepatic venous pressure gradient, venous ammonia was correlated with markers of endothelial dysfunction and liver fibrogenesis/matrix remodelling in the biomarker cohort. ### Conclusions Venous ammonia predicts hepatic decompensation, non-elective liver-related hospitalisation, acute-on-chronic liver failure, and liver-related death, independently of established prognostic indicators including C-reactive protein and hepatic venous pressure gradient. Although venous ammonia is linked with several key disease-driving mechanisms, its prognostic value is not explained by associated hepatic dysfunction, systemic inflammation, or portal hypertension severity, suggesting direct toxicity. ### Impact and implications A recent landmark study linked ammonia levels (a simple blood test) with hospitalisation/death in individuals with clinically stable cirrhosis. Our study extends the prognostic value of venous ammonia to other important liver-related complications. Although venous ammonia is linked with several key disease-driving mechanisms, they do not fully explain its prognostic value. This supports the concept of direct ammonia toxicity and ammonia-lowering drugs as disease-modifying treatment. ## Graphical abstract ## Highlights •Diabetes mellitus was independently linked to increased venous ammonia levels.•The prognostic performance of ammonia for liver-related outcomes is comparable to the UNOS MELD [2016] score and HVPG.•Ammonia predicts liver-related outcomes, independently of established prognostic indicators including CRP and HVPG.•The prognostic value of ammonia is linked with several key disease-driving mechanisms.•However, it is not explained by hepatic dysfunction, systemic inflammation, or portal hypertension severity, suggesting direct toxicity. ## Introduction Advanced chronic liver disease (ACLD) is a major source of morbidity and mortality worldwide, with non-alcoholic fatty liver disease (NAFLD) emerging as the predominant cause of ACLD in several regions.1,2 The first development of hepatic decompensation – most commonly ascites, although hepatic encephalopathy (HE) is the predominant first event in NAFLD2 – denotes a watershed moment in the natural history of ACLD, as it is accompanied by a substantial increase in long-term mortality.3 Those with decompensated ACLD (dACLD) are at risk of acute-on-chronic liver failure (ACLF), which is defined by extrahepatic organ dysfunction (including acute encephalopathy) and high short-term mortality.4 Portal hypertension, which is accompanied by portosystemic shunting, and bacterial translocation-induced systemic inflammation are considered as the main drivers of clinical deterioration.5 In people with ACLD, hyperammonaemia is driven by microbiome changes and ammonia overproduction,6,7 decreased hepatic/extrahepatic metabolic capacity (urea cycle and glutamine synthetase8,9), and portosystemic shunting via collaterals.10,11 The diagnostic utility of ammonia testing for HE is controversially discussed, as hyperammonaemia is not the only mechanism for HE development, and thus, people with HE may present with normal ammonia levels.10 Nevertheless, in those with altered mental status, values within the normal range may question the diagnosis of HE.12 Notably, experimental studies indicate ammonia toxicity beyond its role in HE (e.g. liver fibrogenesis and immune dysfunction13,14). In a recent landmark study, Tranah et al.15 evaluated the impact of ammonia on liver-related outcomes in clinically stable individuals with ACLD and values ≥1.4 × the upper limit of normal predicted liver-related events in stable outpatients with ACLD. However, it remains unclear whether this association is independent of portal hypertension/systemic inflammation, that is, well-established disease-driving mechanisms. Moreover, the findings of experimental studies, that is the link between hyperammonaemia and liver fibrogenesis, remains to be confirmed in humans to support the potential role of ammonia-lowering drugs as a disease-modifying treatment. The objectives of our study were (i) to externally validate and extend previous findings on the prognostic value of venous plasma ammonia levels in a large, well-characterised cohort, while accounting for portal hypertension and systemic inflammation severity and (ii) to investigate the relationship between venous plasma ammonia and biomarkers of other disease-driving mechanisms. ## Study design and participants We performed a retrospective, single-centre cohort study in individuals with ACLD who underwent hepatic venous pressure gradient (HVPG) measurement at the Vienna Hepatic Hemodynamic Lab (outcome cohort; Fig. S1). Those from the outcome cohort were included between Q$\frac{2}{04}$ and Q$\frac{4}{20.}$ Inclusion criteria were (i) liver stiffness measurement ≥10 kPa and/or HVPG ≥6 mmHg, and (ii) availability of venous plasma ammonia levels. Furthermore, individuals were excluded if any of the following criteria were present: those with a history of orthotopic liver transplantation, any active extrahepatic malignancy, non-parenchymal liver disease, non-elective hospitalisation as a result of a liver-related complication at HVPG measurement or within 28 days before HVPG measurement, unsuccessful/unreliable HVPG measurement, bacterial infection, or missing information on important laboratory parameters and/or clinical follow-up. Recruitment and follow-up over the study period are depicted in Fig. S2. In addition, we assessed biomarkers in a partly overlapping cohort of individuals ($$n = 193$$) from the prospective Vienna Cirrhosis Study (VICIS; NCT03267615; pathophysiology cohort) who were recruited between Q$\frac{1}{2017}$ and Q$\frac{3}{2022}$ (Fig. S1), applying similar inclusion and exclusion criteria (biomarker cohort). Overall, $$n = 82$$ participants are included in both cohorts ($15\%$ [$\frac{82}{549}$] vs. $42\%$ [$\frac{82}{193}$]). ## HVPG measurement Under local anaesthesia and ultrasound guidance, a catheter introducer sheath was inserted into the right internal jugular vein.16 Subsequently, a hepatic vein was cannulated and the free and wedged hepatic venous pressures were obtained at least as triplicate measurements by a balloon catheter,17 as recommended by Baveno VII.18 ## Measurement of biomarkers Routine laboratory tests, venous plasma ammonia, and biomarkers (von Willebrand factor [vWF], procalcitonin [PCT], IL-6, enhanced liver fibrosis [ELF®] test, copeptin, renin, and bile acids [BAs]) were performed by the ISO-certified Department of Laboratory Medicine of the Medical University of Vienna using commercially available methods that are applied in clinical routine and blood samples obtained via a central venous line (i.e. the side port of the catheter introducer sheath) at the time of HVPG measurement. Venous plasma ammonia was sampled and rapidly transported on ice to the central laboratory. In line with the previous landmark study,15 venous plasma ammonia levels were divided by the sex-specific upper limit of normal of our local laboratory (i.e. 60 mmol/L for males and 51 mmol/L for females). ## Clinical stages of ACLD, definition of hepatic decompensation and of ACLF Participants were classified according to recently defined prognostic/clinical stages. The definition was adapted from D’Amico et al.19 dACLD was defined by the presence or history of at least one decompensating event, that is ascites, variceal bleeding, or HE. ACLF was defined according to European Foundation for the Study of Chronic Liver Failure (EF-CLIF) criteria.20 ## Statistical analysis All statistical analyses were performed using IBM SPSS Statistics 27 (IBM, New York, NY, USA), R 4.1.2 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria), or GraphPad Prism 8 (GraphPad Software, San Diego, CA, USA). Categorical variables were reported as absolute (n) and relative frequencies (%), whereas continuous variables as mean ± SD or median (interquartile range [IQR]), as appropriate. Student’s t test was used for group comparisons of normally distributed variables and the Mann–Whitney U test for non-normally distributed variables. Group comparisons of categorical variables were performed using either Χ2 or Fisher’s exact test, as appropriate. Univariable and multivariable linear regression analyses were applied to evaluate factors associated with ammonia. Follow-up time was calculated as the time from HVPG measurement to the date of liver transplantation, death, or last follow-up at one of the hospitals of the Vienna hospital association by the reverse Kaplan–Meier method. Impact of venous plasma ammonia levels on liver-related outcomes was assessed using Cox regression and competing risk analyses considering the removal/suppression of the primary aetiological factor (as defined by Baveno VII,18 i.e. initiation of antiviral therapy/reported alcohol abstinence), liver transplantation, or non-liver-related death, as competing risks. Analyses were performed for hepatic decompensation/liver-related death, liver-related death, development of ACLF/requirement of liver transplantation/liver-related death, and non-elective liver-related hospitalisation/liver-related death as outcomes of interest. For the outcome development of ACLF/requirement of liver transplantation/liver-related death – in individuals who had already experienced hepatic decompensation at baseline (i.e. the main at-risk population) – removal/suppression of the primary aetiological factor or non-liver-related death were considered as competing risks. For competing risk regression analyses, Fine and Gray competing risks regression models (cmprsk: subdistribution analysis of competing risks, https://CRAN.R-project.org/package=cmprsk)21,22 were calculated. Univariable and multivariable Cox regression analyses were performed to evaluate parameters independently associated with the events of interest. In a first step, we included all parameters into univariable Cox regression models. Baseline characteristics which we considered of particular importance for the endpoint of interest (i.e. age, indicators of hepatic dysfunction, HVPG, and C-reactive protein [CRP]) were further included into two separate multivariable models. The Child–Turcotte–Pugh (CTP) and United Network for Organ Sharing (UNOS) model for end-stage liver disease (MELD) [2016] scores have significant overlap in terms of included variables. Therefore, we generated separate models with either CTP or UNOS MELD [2016] scores. Time-dependent area under the receiver operating characteristic curve (AUROC) analyses were performed and the R-package ‘timeROC’ was used to compare the prognostic performances for hepatic decompensation/liver-related death and liver-related death between established prognostic indicators (UNOS MELD [2016] score and HVPG) and ammonia (multiplicity-adjusted p values) over time. Spearman’s correlation analyses were conducted to investigate potential associations between ammonia and biomarkers in the biomarker cohort. A heatmap plot was used for graphical illustration of associations between ammonia and biomarkers. The level of significance was set at a two-sided p value of <0.05. ## Ethics This study has been conducted in accordance with the principles of the Declaration of Helsinki and its amendments and has been approved by the local ethics committee (EK$\frac{1531}{2022}$ and EK$\frac{1262}{2017}$), which waived the requirement of written informed consent for the retrospective analysis of the outcome cohort. All participants included in the prospective biomarker cohort (i.e. VICIS study) provided written informed consent for study participation. ## Study population of the outcome cohort Overall, 2,550 individuals underwent HVPG measurement within the study period (Fig. S1). After applying inclusion and exclusion criteria, 549 people were finally included into the outcome cohort. Mean age at HVPG measurement was 54 ± 12 years and most were male ($$n = 370$$, $67\%$; Table 1). Viral hepatitis was the most common aetiology of liver disease ($$n = 207$$, $38\%$), followed by alcohol-related liver disease (ARLD; $$n = 196$$, $36\%$), other aetiologies of ACLD ($$n = 89$$, $16\%$) and NAFLD ($$n = 57$$, $10\%$). Regarding portal hypertension severity, mean HVPG was 16 ± 7 mmHg and $65\%$ of individuals ($$n = 319$$) had varices, of whom 67 ($12\%$) had a history of variceal bleeding. Mean UNOS MELD [2016] was 12 ± 5, and mean CTP score was 7 ± 2 points. Most individuals were classified as CTP-A ($$n = 342$$, $62\%$), whereas $31\%$ of participants were classified as CTP-B ($$n = 168$$) and $7\%$ as CTP-C ($$n = 39$$). A total of 252 participants ($46\%$) had already experienced hepatic decompensation at study inclusion. Eight percent ($$n = 42$$) had a history of overt HE. Accordingly, $7\%$ ($$n = 38$$) were on lactulose, $5\%$ ($$n = 27$$) on rifaximin, and $10\%$ ($$n = 54$$) on oral l-ornithine O-aspartate at baseline. The median baseline ammonia level was 37.3 (IQR: 28.2–51.6) mmol/L and ammonia adjusted for the upper limit of normal (NH3-ULN) was 0.66 (IQR: 0.49–0.91).Table 1Detailed patient characteristics at the time of HVPG measurement of the outcome and the pathophysiology cohort. Patient characteristicsOutcome cohort, $$n = 549$$Pathophysiology cohort, $$n = 193$$p valueAge, years, mean ± SD54.4 ± 11.555.6 ± 12.40.208Sex, n (%) Male370 ($67\%$)131 ($68\%$)0.902 Female179 ($33\%$)62 ($32\%$)Aetiology, n (%) Viral207 ($38\%$)36 ($19\%$)<0.001 ARLD196 ($36\%$)82 ($43\%$) NAFLD57 ($10\%$)20 ($10\%$) Other89 ($16\%$)55 ($29\%$)Varices, n (%)319 ($65\%$)106 ($65\%$)1.000History of decompensation, n (%)252 ($46\%$)119 ($62\%$)<0.001History of variceal bleeding, n (%)67 ($12\%$)21 ($11\%$)0.625Ascites, n (%) None354 ($65\%$)105 ($54\%$)0.036 Mild160 ($29\%$)75 ($39\%$) Severe35 ($6\%$)13 ($7\%$)History of hepatic encephalopathy, n (%)42 ($8\%$)30 ($16\%$)0.545HVPG, mmHg, mean ± SD16 ± 715 ± 60.190 HVPG 0–5 mmHg, n (%)42 ($8\%$)—<0.001 HVPG 6–9 mmHg, n (%)83 ($15\%$)44 ($23\%$) HVPG 10-15 mmHg, n (%)137 ($25\%$)61 ($32\%$) HVPG ≥16 mmHg, n (%)287 ($52\%$)88 ($46\%$)UNOS MELD [2016], points, mean ± SD11.8 ± 4.511.4 ± 4.60.326CTP score, points, mean ± SD6.5 ± 1.76.6 ± 1.90.541 A, n (%)342 ($62\%$)117 ($61\%$)0.475 B, n (%)168 ($31\%$)57 ($30\%$) C, n (%)39 ($7\%$)19 ($9\%$)Laboratory parameters, median (IQR) or mean ± SD Platelet count, G/L107 (73–152)103 (70–142)0.248 Sodium, mmol/L138.0 ± 3.7138.1 ± 3.70.788 Creatinine, mg/dl0.7 (0.6–0.9)0.8 (0.6–0.9)0.401 Albumin, g/L36.5 ± 5.737.2 ± 5.50.136 Bilirubin, mg/dl1.0 (0.7–1.9)1.0 (0.6–1.8)0.638 INR1.3 ± 0.31.4 ± 0.30.438 AST, U/L48 (34–67)40 (29–55)<0.001 ALT, U/L35 (23–60)30 (22–42)<0.001 CRP, mg/L0.2 (0.1–0.6)0.2 (0.1–0.5)0.730 Ammonia, mmol/L37.3 (28.2–51.6)34.5 (26.0–47.4)0.061 NH3-ULN0.66 (0.49–0.91)0.58 (0.43–0.79)0.061Categorical variables were reported as absolute (n) and relative frequencies (%), whereas continuous variables as mean ± SD or median (IQR), as appropriate. Student’s t test was used for group comparisons of normally distributed variables and Mann–Whitney U test for non-normally distributed variables. Group comparisons of categorical variables were performed using either Chi-squared or Fisher’s exact test, as appropriate. Values of p in bold denote $p \leq 0.05.$ ALT, alanine transaminase; ARLD, alcohol-related liver disease; AST aspartate transaminase; CRP, C-reactive protein; CTP, Child–Turcotte–Pugh; HVPG, hepatic venous pressure gradient; INR, international normalised ratio; NAFLD, non-alcoholic fatty liver disease; NH3-ULN, ammonia level corrected to the upper limit of normal; UNOS MELD [2016] score, United Network for Organ Sharing model for end-stage liver disease [2016]. ## Clinical events during follow-up in the outcome cohort Participants were followed for a median of 41.0 ($95\%$ CI: 37.3–44.7) months. A total of 104 deaths ($19\%$) were considered liver-related, whereas 175 events ($32\%$) were captured for the combined endpoint hepatic decompensation/liver-related death. For the endpoint liver-related death, 199 competing risks ($36\%$) occurred, whereas for the combined endpoint (hepatic decompensation/liver-related death), 176 competing risks ($32\%$) were captured. Among decompensated individuals, 45 ($18\%$) developed ACLF. ## Ammonia levels increase with liver disease and portal hypertension severity in the outcome cohort NH3-ULN/NH3 consistently increased with liver disease/portal hypertension severity, as evaluated by the CTP score ($p \leq 0.001$) and UNOS MELD [2016] score ($p \leq 0.001$), as well as severity of portal hypertension ($p \leq 0.001$; Fig. 1; Table S1). Additionally, NH3-ULN also increased across clinical stages ($p \leq 0.001$).Fig. 1NH3-ULN across liver disease severity. Comparison of NH3 corrected to the upper limit of normal according to (A) CTP (B) UNOS MELD [2016] score and (C) HVPG strata as well as (D) clinical stages in the outcome cohort. NH3-ULN levels were reported as median (IQR) and compared with the Mann–Whitney U test. CS, clinical stages; CTP, CTP, Child–Turcotte–Pugh score; HVPG, hepatic venous pressure gradient; NH3-ULN, ammonia adjusted for the upper limit of normal; pc, probably compensated; ULN, upper limit of normal; UNOS MELD [2016] score, United Network for Organ Sharing Model of end-stage liver disease [2016] score. ## Univariable and multivariable analyses of factors associated with ammonia in the outcome cohort In univariable analyses, NH3-ULN was directly associated with severity of liver disease (CTP score: unstandardised regression coefficient [B]: 0.095 [$95\%$ CI: 0.070, 0.117]; $p \leq 0.001$, UNOS MELD [2016] score: B: 0.032 [$95\%$ CI: 0.023, 0.041]; $p \leq 0.001$), and portal hypertension severity (HVPG: B: 0.020 [$95\%$ CI: 0.014, 0.026]; $p \leq 0.001$; Table 2). Additionally, there was a positive association with systemic inflammation (CRP: B: 0.109 [$95\%$ CI: 0.026, 0.192]; $$p \leq 0.011$$), and with body mass index (BMI: B: 0.012 [$95\%$ CI: 0.004, 0.020]; $$p \leq 0.005$$), presence of varices (B: 0.128 [$95\%$ CI: 0.073, 0.183]; $p \leq 0.001$), diabetes (B: 0.130 [$95\%$ CI: 0.016, 0.243]; $$p \leq 0.025$$), and dACLD (B: 0.276 [$95\%$ CI: 0.194, 0.358]; $p \leq 0.001$). Finally, NH3-ULN was negatively associated with male sex (B: -0.142 [$95\%$ CI: -0.232, -0.052]; $$p \leq 0.002$$), arterial hypertension (B: -0.099 [$95\%$ CI: -0.185, -0.013]; $$p \leq 0.024$$), as well as serum sodium (B: -0.017 [$95\%$ CI: -0.028, -0.005]; $$p \leq 0.004$$) and albumin levels (B: -0.021 [$95\%$ CI: -0.029, -0.014]; $p \leq 0.001$).Table 2Simple and multiple linear regression analysis of factors associated with NH3corrected for the upper limit of normal including – among other parameters – either CTP score, as well as serum sodium and creatinine (model 1), or UNOS MELD [2016] score, clinical stage, and serum albumin (model 2) in the outcome cohort. Patient characteristicsUnivariableModel 1 (including CTP score, sodium, creatinine)Model 2 (including MELD and albumin)B$95\%$ CIp valueB$95\%$ CIp valueB$95\%$ CIp valueAge, year0.002-0.002, 0.0060.250——————Male sex-0.142-0.232, -0.0520.002-0.174-0.285, -0.0640.002-0.164-0.272, -0.0550.003BMI, kg/m20.0120.004, 0.0200.0050.0130.002, 0.0230.0190.0140.003, 0.0240.010Overweight†0.0850.000, 0.1710.049——————Obesity‡0.080-0.021, 0.1810.121——————Prediabetes§-0.012-0.127, 0.1030.839——————Diabetes#0.1300.016, 0.2430.0250.1230.004, 0.2410.0430.1340.016, 0.2520.026Arterial hypertension¶-0.099-0.185, -0.0130.024-0.092-0.202, 0.0180.099-0.073-0.185, 0.0380.197Hypertriglyceridemia††-0.099-0.244, 0.0470.184——————Hypercholesterolemia‡‡0.044-0.091, 0.1790.524——————HDL below threshold§§-0.016-0.111, 0.0790.744——————Statin use-0.100-0.251, 0.0510.192——————Hepatic steatosis##0.064-0.030, 0.1580.183——————ARLD vs. other aetiologies0.074-0.015, 0.1630.102——————Varices0.1280.073, 0.183<0.0010.0870.020, 0.1540.0110.0690.001, 0.1380.048CTP score, point0.0950.070, 0.117<0.0010.0870.050, 0.124<0.001———UNOS MELD [2016], point0.0320.023, 0.041<0.001———0.0220.008, 0.0360.002HVPG, mmHg0.0200.014, 0.026<0.0010.006-0.004, 0.0150.2340.004-0.006, 0.0130.417Decompensated vs. compensated ACLD0.2760.194, 0.358<0.001———0.1170.001, 0.2340.048Sodium, mmol/L-0.017-0.028, -0.0050.0040.009-0.007, 0.0240.266———Creatinine, mg/dl0.107-0.054, 0.2680.1920.2310.032, 0.4300.023———Albumin, g/L-0.021-0.029, -0.014<0.001———-0.004-0.015, 0.0070.519ALT, U/L0.0000.000, 0.0000.314——————CRP, mg/L0.1090.026, 0.1920.011-0.037-0.150, 0.0770.526-0.020-0.133, 0.0930.727Values of p in bold denote $p \leq 0.05.$†BMI ≥25 kg/m2.‡BMI ≥30 kg/m2.§Fasting blood glucose 100–125 mg/dl; HbA1c 5.7–$6.4\%$.#Fasting blood glucose >125 mg/dl, HbA1c ≥$6.5\%$, or antidiabetic medication.¶Blood pressure >$\frac{140}{90}$ mmHg, or antihypertensive medication.††Triglycerides >150 mg/dl.‡‡Total cholesterol >200 mg/dl.§§Value <35 mg/dl for males and <39 mg/dl for females.##Biopsy-proven, controlled attenuation parameter >248 dB/m, or diagnosed by ultrasound. ALT alanine transaminase; ARLD alcohol-related liver disease; CRP, C-reactive protein; CTP, Child–Turcotte–Pugh score; HVPG, hepatic venous pressure gradient; UNOS MELD [2016] score, United Network for Organ Sharing model for end-stage liver disease [2016] score. After multivariable adjustment for either CTP score, sodium, and creatinine (model 1) or UNOS MELD [2016] score, serum albumin, and dACLD (model 2), severity of liver disease (model 1: CTP score: B: 0.087 [$95\%$ CI: 0.050, 0.124]; $p \leq 0.001$; model 2: UNOS MELD [2016] score: B: 0.022 [$95\%$ CI: 0.008, 0.036]; $$p \leq 0.002$$), presence of dACLD (model 2: B: 0.117 [$95\%$ CI: 0.001, 0.234]; $$p \leq 0.048$$), BMI (model 1: B: 0.013 [$95\%$ CI: 0.002, 0.023]; $$p \leq 0.019$$; model 2: B: 0.014 [$95\%$ CI: 0.003, 0.024]; $$p \leq 0.010$$), male sex (model 1: B: -0.174 [$95\%$ CI: -0.285, -0.064]; $$p \leq 0.002$$; model 2: B: -0.164 [$95\%$ CI: -0.272, -0.055]; $$p \leq 0.003$$), diabetes (model 1: B: 0.123 [$95\%$ CI: 0.004, 0.241]; $$p \leq 0.043$$; model 2: B: 0.134 [$95\%$ CI: 0.016, 0.252]; $$p \leq 0.026$$), presence of varices (model 1: B: 0.087 [$95\%$ CI: 0.020, 0.154]; $$p \leq 0.011$$; model 2: B: 0.069 [$95\%$ CI: 0.001, 0.138]; $$p \leq 0.048$$), and creatinine levels (model 1: B: 0.231 [$95\%$ CI: 0.032, 0.430]; $$p \leq 0.023$$) were the only parameters with independent positive associations (Table 2). ## Impact of ammonia on liver-related outcomes in the outcome cohort NH3 not only increased with liver disease severity in cross-sectional analyses but was also longitudinally associated with liver-related death (HR: 1.08 $95\%$ CI: 1.05–1.12]; $p \leq 0.001$). Its independent prognostic value was confirmed in two multivariable models (model 1: adjusted HR [aHR]: 1.05 [$95\%$ CI: 1.00–1.12]; $$p \leq 0.044$$; model 2: aHR: 1.04 [$95\%$ CI: 1.00–1.08]; $$p \leq 0.049$$), which were adjusted for age, HVPG, and CRP as well as additionally CTP-score, sodium, and serum creatinine levels in model 1 and UNOS MELD [2016] score, decompensation status, and serum albumin levels in model 2 (Table 3).Table 3Uni- and multivariable Cox regression analyses of factors associated with liver-related death including – among other parameters – CTP score, serum sodium, and creatinine (model 1) or UNOS MELD [2016] score, clinical stage, and serum albumin (model 2) in the outcome cohort. Patient characteristicsUnivariableModel 1 (including CTP score, sodium, and creatinine)Model 2 (including MELD and albumin)HR ($95\%$ CI)p valueaHR ($95\%$ CI)p valueaHR ($95\%$ CI)p valueAge, year1.05 (1.03–1.07)<0.0011.05 (1.03–1.08)<0.0011.05 (1.03–1.07)<0.001HVPG, mmHg1.09 (1.06–1.13)<0.0011.03 (0.99–1.07)0.1241.03 (0.99–1.07)0.119CTP scoreA11——B3.05 (1.99–4.69)<0.0012.04 (1.23–3.38)0.006——C5.89 (3.40–10.21)<0.0013.55 (1.76–7.17)<0.001——UNOS MELD [2016] score, point1.11 (1.07–1.15)<0.001——1.04 (0.99–1.09)0.080Decompensated vs. compensated ACLD2.38 (1.60–3.54)<0.001——1.02 (0.63–1.66)0.934Sodium, mmol/L0.92 (0.88–0.96)<0.0010.99 (0.94–1.05)0.812——Creatinine, mg/dl1.91 (0.99–3.71)0.0550.72 (0.35–1.49)0.371——Albumin, g/L0.92 (0.89–0.94)<0.001——0.95 (0.92–0.99)0.003CRP, mg/L2.08 (1.59–2.71)<0.0011.36 (0.98–1.90)0.0701.25 (0.87–1.78)0.227NH3, μmol/L, per 101.08 (1.05–1.12)<0.0011.05 (1.00–1.10)0.0441.04 (1.00–1.08)0.049Concordance ± SE0.764 ± 0.0240.776 ± 0.023AIC1,084.9401,084.510Values of p in bold denote $p \leq 0.05.$ ACLD, advanced chronic liver disease; aHR, adjusted hazard ratio; AIC, Akaike information criterion; ARLD, alcohol-related liver disease; CRP, C-reactive protein; CTP, Child–Turcotte–Pugh score; HVPG, hepatic venous pressure gradient; NAFLD, non-alcoholic fatty liver disease; NH3-ULN ammonia adjusted for the upper limit of normal; UNOS MELD [2016] score, United Network for Organ Sharing model of end-stage liver disease [2016] score. Next, we evaluated the prognostic performance of the previously provided cut-off of 1.4 in our outcome cohort. Importantly, this cut-off was not only associated with liver-related death in competing risk regression analysis (Table S2), but also hepatic decompensation (Cox regression Table 4 and Table S3; competing risk regression Table S4), as well as liver-related hospitalisation (Cox regression Table S5; competing risk regression Table S6), and ACLF in dACLD (Cox regression Table S7; competing risk regression Table S8).Table 4Uni- and multivariable Cox regression analyses of factors associated with hepatic decompensation/liver-related death including – among other parameters – CTP score, serum sodium, and creatinine (model 1) or serum UNOS MELD [2016] score, clinical stage, and serum albumin (model 2) in the outcome cohort. Patient characteristicsUnivariableModel 1 (including CTP score, sodium, and creatinine)Model 2 (including MELD and albumin)HR ($95\%$ CI)p valueaHR ($95\%$ CI)p valueaHR ($95\%$ CI)p valueAge, year1.03 (1.02–1.05)<0.0011.02 (1.01–1.04)0.0071.02 (1.01–1.04)0.003HVPG, mmHg1.15 (1.12–1.17)<0.0011.10 (1.07–1.13)<0.0011.09 (1.06–1.12)<0.001CTP scoreA11——B4.62 (3.33–6.43)<0.0012.22 (1.52–3.25)<0.001——C6.91 (4.33–11.04)<0.0012.47 (1.38–4.40)0.002——UNOS MELD [2016] score, point1.12 (1.09–1.15)<0.001——0.99 (0.96–1.03)0.713Decompensated vs. compensated ACLD5.63 (3.96–8.00)<0.001——2.35 (1.58–3.49)<0.001Sodium, mmol/L0.90 (0.87–0.93)<0.0010.98 (0.94–1.03)0.443——Creatinine, mg/dl2.49 (1.52–4.06)<0.0011.38 (0.84–2.26)0.199——Albumin, g/L0.89 (0.86–0.91)<0.001——0.95 (0.92–0.98)<0.001CRP, mg/L2.47 (1.98–3.08)<0.0011.69 (1.28–2.23)<0.0011.64 (1.24–2.16)<0.001NH3-ULN ≥1.4 vs. <1.43.84 (2.58–5.71)<0.0012.08 (1.35–3.22)<0.0011.99 (1.31–3.02)0.001Concordance ± SE0.819 ± 0.0150.825 ± 0.014AIC1,819.5691,807.099Values of p in bold denote $p \leq 0.05.$ACLD, advanced chronic liver disease; aHR, adjusted hazard ratio; AIC, Akaike information criterion; ARLD, alcohol-related liver disease; CRP, C-reactive protein; CTP, Child–Turcotte–Pugh score; HVPG, hepatic venous pressure gradient; NAFLD, non-alcoholic fatty liver disease; NH3-ULN, ammonia adjusted for the upper limit of normal; UNOS MELD [2016] score, United Network for Organ Sharing Model of end-stage liver disease [2016] score. In addition, we compared the prognostic performance of NH3-ULN for hepatic decompensation/liver-related death and liver-related death to UNOS MELD [2016] score and HVPG in time-dependent AUROC analyses for the following time points: 12, 24, 36, 48, and 60 months. Regarding hepatic decompensation/liver-related death, HVPG showed significantly better discriminatory ability compared to NH3-ULN at 24 months (NH3-ULN vs. HVPG: $$p \leq 0.044$$ after accounting for multiplicity). In contrast, NH3-ULN was comparable to time-dependent AUROC of the UNOS MELD [2016] score (Fig. 2A). Importantly, time-dependent AUROC of NH3-ULN for liver-related death was comparable to those of UNOS MELD [2016]-score and HVPG at all tested time points (Fig. 2B).Fig. 2Prognostic implications of NH3-ULN compared to HVPG and UNOS MELD [2016].Univariable time-dependent area under the receiver operating curve (AUROC) analyses comparing the prognostic performances of UNOS MELD [2016], HVPG, and NH3-ULN for prognostication of (A) hepatic decompensation/liver-related death and (B) liver-related death in the outcome cohort. HVPG, hepatic venous pressure gradient; NH3-ULN, ammonia adjusted for the upper limit of normal; UNOS MELD [2016] score, United Network for Organ Sharing Model of end-stage liver disease [2016] score. Finally, stratifying the cohort according to NH3-ULN quartiles (q1: <0.49, q2: 0.49–0.66, q3: 0.66–0.91, q4: ≥0.91) identified patient groups with a distinct prognosis (subdistribution HR [SHR] $p \leq 0.001$; Fig. 3A and B). In line, the previously proposed cut-off identified individuals with particular poor outcomes in regard to liver-related death (SHR: 3.39 [$95\%$ CI: 2.08–5.53]; $p \leq 0.001$; Fig. 3C) as well as hepatic decompensation/liver-related death (SHR: 4.11 [$95\%$ CI: 2.74–6.16]; $p \leq 0.001$; Fig. 3D).Fig. 3Stratification of outcomes according to NH3-ULN.Cumulative incidence plots of remaining free of liver-related death (A,C) and hepatic decompensation/liver-related death (B,D) according to NH3-ULN quartiles (A,B) and previously published cut-off of 1.4 (C,D) in the outcome cohort. For these cumulative incidence plots, competing risk curves were depicted. In A and B, the subdistribution hazard ratios (SHR) were calculated. For C and D, individuals with <1.4 vs. ≥1.4 NH3-ULN were compared. NH3-ULN, ammonia adjusted for the upper limit of normal; q, quartile. ## Associations between ammonia and disease-driving mechanisms in the biomarker cohort Baseline characteristics of the biomarker cohort are provided in Table 1. Participants included in the biomarker cohort had less viral and more ARLD as underlying aetiology and were more often decompensated at baseline ($62\%$ vs. $46\%$; $p \leq 0.001$). Finally, we evaluated the correlation of ammonia with the severity of liver disease (UNOS MELD [2016] score and HVPG), serum BA levels, endothelial dysfunction (vWF), markers of systemic inflammation (CRP, PCT, and IL-6) as well as liver fibrogenesis/matrix remodelling (ELF-test), and markers of hyperdynamic circulation/systemic haemodynamic impairment (mean arterial pressure [MAP], copeptin, renin, and serum sodium). As demonstrated in Fig. 4, ammonia showed low correlations with UNOS MELD [2016] score, BAs, HVPG, vWF, and ELF-test, as well as associations with CRP, IL-6, and PCT, MAP, renin, and serum sodium in the biomarker cohort. Further results on the correlations of ammonia with these biomarkers among compensated and decompensated individuals are reported in the Supplementary material. Fig. 4Heatmaps of correlations of NH3and pathophysiological biomarkers in the biomarker cohort.∗$p \leq 0.05$; ∗∗$p \leq 0.001$; Spearman’s correlation analyses were conducted. ACLD, advanced chronic liver disease; BA, bile acid; CRP, C-reactive protein; ELF, enhanced liver fibrosis; HVPG, hepatic venous pressure gradient; IL-6, interleukin-6; MAP, mean arterial pressure; NH3, ammonia; PCT, procalcitonin; UNOS MELD [2016], United Network for Organ Sharing model for end-stage liver disease [2016]; vWF, von Willebrand factor antigen. ## Discussion Although routinely measured ammonia is of limited value for diagnosing HE, a recent landmark study highlighted its prognostic implications. In our study, venous ammonia increased across clinical stages of ACLD as well as with more severe hepatic dysfunction and portal hypertension. Importantly, time-dependent AUROC values of ammonia for liver-related death were similar to the laboratory-based composite score UNOS MELD [2016] and HVPG, which can only be measured invasively. Ammonia was not only independently associated with liver-related death – as shown previously – but also with other liver-related outcomes including ACLF, even after adjusting for liver disease, systemic inflammation, and portal hypertension severity. Notably, our findings support the use of the previously proposed cut-off of NH3-ULN of ≥1.4 for risk stratification. Finally, we have provided information on associated pathophysiologic mechanisms that may explain the prognostic value of ammonia, that is, liver fibrogenesis/matrix remodelling and endothelial dysfunction. Interestingly, diabetes was independently associated with venous plasma ammonia levels, even after adjusting for various co-factors (adjusted B: 0.134 [$95\%$ CI: 0.016, 0.252]; $$p \leq 0.026$$; Table 2). Diabetes may increase venous plasma ammonia levels by autonomic dysfunction, extended gastrointestinal transit times, and bacterial overgrowth as well as increased protein catabolism and accelerated muscle-breakdown.23 Even in earlier stages of fibrosis in individuals with NAFLD/metabolic-associated fatty liver disease (MAFLD), deficiencies in urea synthesis (in part caused by impaired liver-α-cell axis with glucagon resistance and impaired ureagenesis24), microglial activation, astrocyte swelling, and possibly even neurodegenerative changes and brain atrophy – all caused by elevated ammonia levels – have been reported.25,26 From a clinical perspective, Haj and Rockey27 have found that in individuals with cirrhosis hospitalised with HE, ammonia was often normal and did not impact treatment decisions, thereby arguing against the routine use of ammonia as either an initial diagnostic test or for guiding medical therapy. However, moving to risk stratification, ammonia is increasingly acknowledged as a prognostic biomarker. Vierling et al.28 have shown that ammonia was able to identify people at risk for HE-related events. Moreover, a recent multicentre study by Tranah et al.15 indicated that ammonia is predictive of hospitalisation/liver-related complications and mortality, showing a better prognostic performance than traditional scores. To corroborate the main finding of this study, we have externally validated the cut-off of ≥1.4 NH3-ULN. However, the authors did neither adjust their multivariable models for systemic inflammation nor portal hypertension severity, which has been accounted for in our work. Notably, ammonia levels varied significantly throughout the study centres, with only one centre reporting ammonia levels that were comparable to our cohort, which may be explained by differences in patient selection and characteristics. Interestingly, ammonia levels reported by Gairing et al.29 were similar to ours and only a minority of individuals had ammonia levels above the ULN ($13\%$ vs. $19\%$ in our cohort). In people admitted because of acute decompensation of cirrhosis, ammonia levels have been found to be independently associated with mortality.30 Our study extends these findings, as it demonstrated that stratifying individuals according to the proposed cut-off of ≥1.4 NH3-ULN identifies decompensated individuals at risk for ACLF, even if they are still outpatients/clinically stable. Therefore, it may provide the opportunity for the timely initiation of disease-modifying interventions that are currently under investigation (e.g. LIVERHOPE, NCT03150459). Hyperammonaemia in ACLD is driven by ammonia overproduction/altered microbiome in the intestinal tract,6,7 decreased metabolic capacity in the hepatocytes leading to a reduced ammonia metabolism in the urea cycle and via glutamine synthetase,8,9 and portal hypertension through the development of portosystemic shunting via collateral flow.10,11 Effects of hyperammonaemia include immune dysfunction and sarcopenia as well as direct negative implications on liver disease progression.10 Recent works on the impact of hyperammonaemia in animal and in-vitro studies on fibrosis demonstrated the induction of oxidative stress and apoptosis and the activation of hepatic stellate cells (HSCs).31,32 Intriguingly, we observed consistent (i.e. both in cACLD and dACLD) positive correlations between ammonia and the ELF-test, which has been shown to reflect fibrogenesis/extracellular matrix remodelling irrespective of the stage of ACLD and indicate HSC activation.33 Moreover, there were also positive correlations with systemic inflammation, vWF as a marker of endothelial dysfunction, and severity of hepatic dysfunction and portal hypertension. Finally, ammonia also correlated with BAs, which may be interpreted as a biomarker for portosystemic shunting,34 which has also been linked with disease progression.35,36 The main limitation of our study is its retrospective design. However, participants were thoroughly characterised at the time of HVPG measurement. For our multivariable models, we were unable to consider several potentially important prognostic indicators (e.g. sarcopaenia/frailty), as they have not been recorded systematically. Nevertheless, participants included in our study were extensively characterised in terms of portal hypertension severity, prognostic scores, and routine laboratory parameters including markers of systemic inflammation – importantly, all of these aspects have been considered in our analyses. Model selection was based on expert opinion/biological relevance. Applying backward elimination for variable selection yielded a ‘slimmer’ model for predictive purposes, that still included NH3. Furthermore, we cannot exclude that some hepatic decompensation events have been missed. However, we have thoroughly reviewed electronic health records of the Vienna hospital association and nationwide electronic health records. Moreover, we have also performed searches of the liver transplant database of our institution (i.e. the only transplant centre in eastern Austria) and examined the nationwide death registry. As complete information on (reason of) death is guaranteed by the latter measure, we included liver-related death in all composite endpoints to ensure the ascertainment of the most severe disease courses. We cannot rule out selection bias because we only included people undergoing HVPG measurement. However, haemodynamic evaluations are routinely performed for risk stratification and treatment monitoring purposes at our centre, and thus, we are confident that our study population is quite representative of clinically stable outpatients with ACLD treated at our centre. Finally, ammonia testing has several limitations. There is substantial laboratory variability37 and arterial ammonia might be preferred over venous ammonia.[38], [39], [40] However, it has been shown that venous ammonia closely correlates with arterial ammonia in people with cirrhosis41 and venous sampling substantially increases feasibility and therefore clinical utility in outpatients. We have provided a detailed description of the measurement of ammonia in the Materials and methods section of this study and are confident that our results are reliable, because preanalytical and analytical conditions were highly standardised. In conclusion, venous ammonia predicts hepatic decompensation, non-elective liver-related hospitalisation, ACLF, and liver-related death, independently of established prognostic indicators including CRP and HVPG. Although venous ammonia is linked with several key disease-driving mechanisms, its prognostic value is not explained by associated hepatic dysfunction, systemic inflammation, or portal hypertension severity, suggesting direct toxicity. ## Financial support The authors received no financial support to produce this manuscript. ## Authors’ contributions Concept of the study: LB, JK, MM. Data collection: LB, JK, RP, BSi, BSc, MM. Statistical analysis: LB, JK, MM. Drafting of the manuscript: LB, JK, MM. Revision for important intellectual content and approval of the final manuscript: all authors. ## Data availability statement The data that support the findings of this study are available from the corresponding author upon reasonable request. ## Conflicts of interest The authors have nothing to disclose regarding the work under consideration for publication. LB, JK, GS, MJ, LH, AFS, PS, and TS have nothing to disclose. The following authors disclose conflicts of interests outside the submitted work. RP received travel support from AbbVie, Gilead and Takeda. BSc received travel support from AbbVie, Ipsen and Gilead. BSi received travel support from AbbVie and Gilead. MP served as a speaker and/or consultant and/or advisory board member for Bayer, Bristol-Myers Squibb, Eisai, Ipsen, Lilly, MSD, and Roche, and received travel support from Bayer and Bristol-Myers Squibb. MT served as a speaker and/or consultant and/or advisory board member for Albireo, BiomX, Falk, Boehringer Ingelheim, Bristol-Myers Squibb, Falk, Genfit, Gilead, Hightide, Intercept, Janssen, MSD, Novartis, Phenex, Pliant, Regulus, Siemens and Shire, and received travel support from AbbVie, Falk, Gilead, and Intercept as well as grants/research support from Albireo, Alnylam, Cymabay, Falk, Gilead, Intercept, MSD, Takeda, and UltraGenyx. He is also co-inventor of patents on the medical use of 24-norursodeoxycholic acid. TR received grant support from AbbVie, Boehringer-Ingelheim, Gilead, Intercept, MSD, Myr Pharmaceuticals, Philips Healthcare, Pliant, Siemens, and W. L. Gore & Associates; speaking honoraria from AbbVie, Gilead, Gore, Intercept, Roche, and MSD; consulting/advisory board fees from AbbVie, Bayer, Boehringer-Ingelheim, Gilead, Intercept, MSD, and Siemens; and travel support from AbbVie, Boehringer-Ingelheim, Gilead, and Roche. MM served as a speaker and/or consultant and/or advisory board member for AbbVie, Collective Acumen, Gilead, Takeda, and W. L. Gore & Associates and received travel support from AbbVie and Gilead. Please refer to the accompanying ICMJE disclosure forms for further details. ## Supplementary data The following are the supplementary data to this article:Multimedia component 1Multimedia component 2Multimedia component 3Multimedia component 4 ## References 1. Eslam M., Sanyal A.J., George J.. MAFLD: a consensus-driven proposed nomenclature for metabolic associated fatty liver disease. Gastroenterology (2020) 158 1999-2014.e1. PMID: 32044314 2. Sanyal A.J., Van Natta M.L., Clark J., Neuschwander-Tetri B.A., Diehl A., Dasarathy S.. Prospective study of outcomes in adults with nonalcoholic fatty liver disease. N Engl J Med (2021) 385 1559-1569. PMID: 34670043 3. Mandorfer M., Simbrunner B.. Prevention of first decompensation in advanced chronic liver disease. Clin Liver Dis (2021) 25 291-310. PMID: 33838851 4. Trebicka J., Fernandez J., Papp M., Caraceni P., Laleman W., Gambino C.. The PREDICT study uncovers three clinical courses of acutely decompensated cirrhosis that have distinct pathophysiology. J Hepatol (2020) 73 842-854. PMID: 32673741 5. Costa D., Simbrunner B., Jachs M., Hartl L., Bauer D., Paternostro R.. Systemic inflammation increases across distinct stages of advanced chronic liver disease and correlates with decompensation and mortality. J Hepatol (2021) 74 819-828. PMID: 33075344 6. Rai R., Saraswat V.A., Dhiman R.K.. Gut microbiota: its role in hepatic encephalopathy. J Clin Exp Hepatol (2015) 5 S29-S36. PMID: 26041954 7. Romero-Gómez M., Ramos-Guerrero R., Grande L., de Terán L.C., Corpas R., Camacho I.. Intestinal glutaminase activity is increased in liver cirrhosis and correlates with minimal hepatic encephalopathy. J Hepatol (2004) 41 49-54. PMID: 15246207 8. Fleming K.E., Wanless I.R.. Glutamine synthetase expression in activated hepatocyte progenitor cells and loss of hepatocellular expression in congestion and cirrhosis. Liver Int (2013) 33 525-534. PMID: 23362937 9. Wierling C.. Bridging the gap between metabolic liver processes and functional tissue structure by integrated spatiotemporal modeling applied to hepatic ammonia detoxification. Hepatology (2014) 60 1823-1825. PMID: 25130974 10. Deutsch-Link S., Moon A.M.. The ongoing debate of serum ammonia levels in cirrhosis: the good, the bad, and the ugly. Am J Gastroenterol (2023) 118 10-13. PMID: 36001400 11. Tapper E.B., Jiang Z.G., Patwardhan V.R.. Refining the ammonia hypothesis: a physiology-driven approach to the treatment of hepatic encephalopathy. Mayo Clin Proc (2015) 90 646-658. PMID: 25865476 12. EASL Clinical Practice Guidelines on the management of hepatic encephalopathy. J Hepatol (2022) 77 807-824. PMID: 35724930 13. Angelova P.R., Kerbert A.J.C., Habtesion A., Hall A., Abramov A.Y., Jalan R.. Hyperammonaemia induces mitochondrial dysfunction and neuronal cell death. JHEP Rep (2022) 4 14. Hadjihambi A., Cudalbu C., Pierzchala K., Simicic D., Donnelly C., Konstantinou C.. Abnormal brain oxygen homeostasis in an animal model of liver disease. JHEP Rep (2022) 4 15. Tranah T.H., Ballester M.P., Carbonell-Asins J.A., Ampuero J., Alexandrino G., Caracostea A.. Plasma ammonia levels predict hospitalisation with liver-related complications and mortality in clinically stable outpatients with cirrhosis. J Hepatol (2022) 77 1554-1563. PMID: 35872326 16. Ferlitsch A., Bota S., Paternostro R., Reiberger T., Mandorfer M., Heinisch B.. Evaluation of a new balloon occlusion catheter specifically designed for measurement of hepatic venous pressure gradient. Liver Int (2015) 35 2115-2120. PMID: 25585656 17. Reiberger T., Schwabl P., Trauner M., Peck-Radosavljevic M., Mandorfer M.. Measurement of the hepatic venous pressure gradient and transjugular liver biopsy. J Vis Exp (2020) 160 18. de Franchis R., Bosch J., Garcia-Tsao G., Reiberger T., Ripoll C.. Baveno VII – renewing consensus in portal hypertension. J Hepatol (2022) 76 959-974. PMID: 35120736 19. D'Amico G., Morabito A., D'Amico M., Pasta L., Malizia G., Rebora P.. Clinical states of cirrhosis and competing risks. J Hepatol (2018) 68 563-576. PMID: 29111320 20. Moreau R., Jalan R., Gines P., Pavesi M., Angeli P., Cordoba J.. Acute-on-chronic liver failure is a distinct syndrome that develops in patients with acute decompensation of cirrhosis. Gastroenterology (2013) 144 1426-1437. PMID: 23474284 21. Fine J.P., Gray R.J.. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc (1999) 94 496-509 22. Therneau T., Crowson C., Atkinson E.. (2022) 23. Gairing S.J., Schleicher E.M., Labenz C.. Diabetes mellitus – risk factor and potential future target for hepatic encephalopathy in patients with liver cirrhosis?. Metab Brain Dis (2022). DOI: 10.1007/s11011-022-01068-4 24. Richter M.M., Galsgaard K.D., Elmelund E., Knop F.K., Suppli M.P., Holst J.J.. The liver-α-cell axis in health and in disease. Diabetes (2022) 71 1852-1861. PMID: 35657688 25. Kjærgaard K., Mikkelsen A.C.D., Wernberg C.W., Grønkjær L.L., Eriksen P.L., Damholdt M.F.. Cognitive dysfunction in non-alcoholic fatty liver disease-current knowledge, mechanisms and perspectives. J Clin Med (2021) 10 673. PMID: 33572481 26. Weinstein G., Zelber-Sagi S., Preis S.R., Beiser A.S., DeCarli C., Speliotes E.K.. Association of nonalcoholic fatty liver disease with lower brain volume in healthy middle-aged adults in the Framingham study. JAMA Neurol (2018) 75 97-104. PMID: 29159396 27. Haj M., Rockey D.C.. Ammonia Levels do not guide clinical management of patients with hepatic encephalopathy caused by cirrhosis. Am J Gastroenterol (2020) 115 723-728. PMID: 31658104 28. Vierling J.M., Mokhtarani M., Brown R.S., Mantry P., Rockey D.C., Ghabril M.. Fasting blood ammonia predicts risk and frequency of hepatic encephalopathy episodes in patients with cirrhosis. Clin Gastroenterol Hepatol (2016) 14 903-906.e1. PMID: 26707685 29. Gairing S.J., Kaps L., Schleicher E.M., Galle P.R., Labenz C.. Ammonia – an old friend with a new area of application. J Hepatol (2023) 78 e22-e23. PMID: 35988683 30. Shalimar M.F., Sheikh R.P., Mookerjee B., Agarwal S.K., Acharya R.. Prognostic role of ammonia in patients with cirrhosis. Hepatology (2019) 70 982-994. PMID: 30703853 31. De Chiara F., Thomsen K.L., Habtesion A., Jones H., Davies N., Gracia-Sancho J.. Ammonia scavenging prevents progression of fibrosis in experimental nonalcoholic fatty liver disease. Hepatology (2020) 71 874-892. PMID: 31378982 32. Jia B., Yu Z.J., Duan Z.F., Lü X.Q., Li J.J., Liu X.R.. Hyperammonaemia induces hepatic injury with alteration of gene expression profiles. Liver Int (2014) 34 748-758. PMID: 24134218 33. Simbrunner B., Villesen I.F., Königshofer P., Scheiner B., Bauer D., Paternostro R.. Systemic inflammation is linked to liver fibrogenesis in patients with advanced chronic liver disease. Liver Int (2022) 42 2501-2512. PMID: 35822301 34. Simbrunner B., Trauner M., Reiberger T.. Review article: therapeutic aspects of bile acid signalling in the gut-liver axis. Aliment Pharmacol Ther (2021) 54 1243-1262. PMID: 34555862 35. Praktiknjo M., Simón-Talero M., Römer J., Roccarina D., Martínez J., Lampichler K.. Total area of spontaneous portosystemic shunts independently predicts hepatic encephalopathy and mortality in liver cirrhosis. J Hepatol (2020) 72 1140-1150. PMID: 31954206 36. Simón-Talero M., Roccarina D., Martínez J., Lampichler K., Baiges A., Low G.. Association between portosystemic shunts and increased complications and mortality in patients with cirrhosis. Gastroenterology (2018) 154 1694-1705.e4. PMID: 29360462 37. Dam G., Ott P., Aagaard N.K., Vilstrup H.. Branched-chain amino acids and muscle ammonia detoxification in cirrhosis. Metab Brain Dis (2013) 28 217-220. PMID: 23315357 38. Brusilow S.W., Koehler R.C., Traystman R.J., Cooper A.J.. Astrocyte glutamine synthetase: importance in hyperammonemic syndromes and potential target for therapy. Neurotherapeutics (2010) 7 452-470. PMID: 20880508 39. da Fonseca-Wollheim F.. Preanalytical increase of ammonia in blood specimens from healthy subjects. Clin Chem (1990) 36 1483-1487. PMID: 2387046 40. Kramer L., Tribl B., Gendo A., Zauner C., Schneider B., Ferenci P.. Partial pressure of ammonia versus ammonia in hepatic encephalopathy. Hepatology (2000) 31 30-34. PMID: 10613724 41. Ong J.P., Aggarwal A., Krieger D., Easley K.A., Karafa M.T., Van Lente F.. Correlation between ammonia levels and the severity of hepatic encephalopathy. Am J Med (2003) 114 188-193. PMID: 12637132
--- title: Anti-Trypanosoma cruzi activity of Coptis rhizome extract and its constituents authors: - Yuki Tayama - Shusaku Mizukami - Kazufumi Toume - Katsuko Komatsu - Tetsuo Yanagi - Takeshi Nara - Paul Tieu - Nguyen Tien Huy - Shinjiro Hamano - Kenji Hirayama journal: Tropical Medicine and Health year: 2023 pmcid: PMC9976467 doi: 10.1186/s41182-023-00502-2 license: CC BY 4.0 --- # Anti-Trypanosoma cruzi activity of Coptis rhizome extract and its constituents ## Abstract ### Background Current therapeutic agents, including nifurtimox and benznidazole, are not sufficiently effective in the chronic phase of Trypanosoma cruzi infection and are accompanied by various side effects. In this study, 120 kinds of extracts from medicinal herbs used for Kampo formulations and 94 kinds of compounds isolated from medicinal herbs for Kampo formulations were screened for anti-T. cruzi activity in vitro and in vivo. ### Methods As an experimental method, a recombinant protozoan cloned strain expressing luciferase, namely Luc2-Tulahuen, was used in the experiments. The in vitro anti-T. cruzi activity on epimastigote, trypomastigote, and amastigote forms was assessed by measuring luminescence intensity after treatment with the Kampo extracts or compounds. In addition, the cytotoxicity of compounds was tested using mouse and human feeder cell lines. The in vivo anti-T. cruzi activity was measured by a murine acute infection model using intraperitoneal injection of trypomastigotes followed by live bioluminescence imaging. ### Results As a result, three protoberberine-type alkaloids, namely coptisine chloride, dehydrocorydaline nitrate, and palmatine chloride, showed strong anti-T. cruzi activities with low cytotoxicity. The IC50 values of these compounds differed depending on the side chain, and the most effective compound, coptisine chloride, showed a significant effect in the acute infection model. ### Conclusions For these reasons, coptisine chloride is a hit compound that can be a potential candidate for anti-Chagas disease drugs. In addition, it was expected that there would be room for further improvement by modifying the side chains of the basic skeleton. ### Supplementary Information The online version contains supplementary material available at 10.1186/s41182-023-00502-2. ## Background Chagas disease, caused by a protozoan parasite Trypanosoma cruzi (T. cruzi), is a debilitating illness that affects from 6 to 7 million [1, 2] people mostly in Latin America. The disease is now expanding to non-endemic areas due to human migration [3]. The disease can be categorized into two distinct phases namely: an acute phase and a chronic phase. Acute phase is defined by high parasitemia, fever and lymphadenopathy and is usually resolved within 4–8 weeks [4]. Chronic phase begins after the acute phase and stays asymptomatic for decades (indeterminate phase) until clinical manifestation of the disease developed [3]. About 30–$40\%$ of chronically infected individuals develop typical clinical complications which involve cardiac and or gastrointestinal lesions [5]. Chemotherapeutics that are currently used for treatment of the disease are nifurtimox and benznidazole that have been used since 1960 [6, 7]. These drugs have been reported to show limited therapeutic activity against the infection [1] and have low compliance among patients due to the toxic side effects [8]. Still there is no vaccine for Chagas disease currently [9]. Thus, the lack of an efficient drug treatment requires the development of new anti-T. cruzi compound that has improved tolerability, safety, lower toxicity and improved efficacy on both phases of the disease [10]. Traditional Chinese medicine has more than 2000-year history and its standardization has been developed over a long period of time [11]. However, Kampo is a traditional treatment system originated from Chinese medicine was developed in Japan [12, 13]. In general, Kampo herbs are supposed to have a rich resource of active components [14, 15]. In this study, 120 kinds of extracts from medicinal herbs used for Kampo formulations and 94 kinds of compounds isolated from medicinal herbs for Kampo formulations which have been well quality controlled and maintained by Japanese leading research institute, Institute of Natural Medicine, University of Toyama, were screened for their anti-T. cruzi activity in vitro. We have confirmed that the protoberberine-type alkaloids with isoquinoline skeleton, which are the major active components of several plant extracts, showed a significant anti-T. cruzi activity in vitro as well as in vivo. ## Kampo extracts and compounds The Kampo extracts and compounds library was provided by the Institute of Natural Medicine (The University of Toyama, Toyama, Japan). The library contains 120 kinds of extracts from medicinal herbs used for Kampo formulations and 94 kinds of compounds isolated from medicinal herbs for Kampo formulations (Additional file 1: Table S1, S2, Figs. 1, 2). Ultra-pure water generated by Milli-Q (Merck KGaA, Darmstadt, Germany) was the solvent for all herbal extracts and the concentration was adjusted by dry weight of the extract. Compounds were preserved at a concentration of 10 mM dissolved in dimethyl sulfoxide (DMSO; Wako Pure Chemicals Industrial Ltd, Japan). For more extensive experiments, coptisine chloride was purchased from Toronto Research Chemicals (Canada), and benznidazole was purchased from Sigma-Aldrich (USA). For in vivo administration solvent, 10 mg/ml solution in $7\%$ Tween-80 (Sigma-Aldrich, USA), $3\%$ ethanol (v/v) (Wako Pure Chemicals Industrial Ltd, Japan) and $90\%$ (v/v) Milli-Q water [16] was prepared. Fig. 1Screen of 120 kinds of extracts (20 μg/ml) and 94 kinds of compounds (20 μM) isolated from medicinal herbs by luminescence intensity to evaluate inhibitory effect. Data are presented as percent reduction in intensity. Experiments were done in triplicatesFig. 2Cytotoxicity screen of 120 kinds of extracts (20 μg/ml) and 94 kinds of compounds (20 μM) isolated from medicinal herbs with alamarBlue reagent. Data are presented as percent reduction in intensity, indicating cell viability. Experiments were done in triplicates ## Mammalian control and host cell lines Newborn mouse heart fibroblast cells (NMH cells) were obtained from Bio-Resource Center, Institute of Tropical Medicine (NEKKEN), Nagasaki University. HuH28 (derived from 37-year-old female) cells [17] were maintained at Chulabhorn International College of Medicine, Thammasat University, Thailand. ## Parasites We used a laboratory strain of T. cruzi which luciferase gene was transfected and integrated named Luc2-Tulahuen originally donated by Professor Takeshi Nara at Iryo Sosei University [18], and were obtained from the bio-resource center at the institute of tropical medicine (NEKKEN) funded by National Bio-resource Project (NBRP) Japan. Epimastigote form of T. cruzi was cultured in liver infusion tryptose (LIT) medium (Liver Infusion Broth (BD 226920), Tryptose (BD 211713), Becton Dickinson, USA) supplemented with $10\%$ fetal bovine serum (FBS) (Gibco, USA), washed human red blood cells (Japan Red Cross Blood Center) and 500 mg/ml of G418 (Thermo Fisher Scientific, USA) at 26 °C until parasites reach to logarithmic stage. T. cruzi epimastigotes partially transformed to metacyclic forms were transferred to confluent NMH cells maintained in minimal essential medium (MEM) (Wako Pure Chemicals Industrial Ltd, Japan) supplemented with $10\%$ newborn calf serum (Thermo Fisher Scientific, USA) at 37 °C and $5\%$ CO2 as described elsewhere [19]. After 48 h incubation, free parasites outside the host cells were washed out and fresh complete MEM was added. ## In vitro anti-T. cruzi (trypomastigotes and amastigotes) assay Luc2-Tulahuen trypomastigotes (3 × 104/well) and NMH cells (5 × 104/well) were mixed into a 96-well white plate. Test samples and controls were added and incubated with the cells for 72 h. NP-40 (Wako Pure Chemicals Industrial Ltd, Japan) was used as the positive control, and a solvent in which the drug was dissolved as a negative. After 72 h of incubation, 100 μl of luciferin solution containing $0.6\%$ NP-40 (Piccagene) (Toyo Ink Group, Japan) was added. Luminescence intensity was measured by a plate reader (ARVO MX 1420) (Measurement time: 1 s) [20]. ## In vitro anti-T. cruzi (intracellular amastigotes) assay The intracellular amastigotes assay was described by Alonso-Padilla et al. [ 21]. Briefly, 6 × 106 trypomastigotes and 3 × 106 host cells (NMH cells) are seeded in a 25 cm2 flask in serum-free media (MEM + $1\%$ newborn calf serum) to enhance the intracellular infection. After 24 h incubation, the cells were washed twice with phosphate buffered saline (PBS), then replaced by MEM + $10\%$ newborn calf serum and incubated for another 24 h. Infected cells were detached by trypsinization. Infected host fibroblast cells (5 × 104/well) were seeded into a 96-well plate. Test samples and the controls were added to the mixture and incubated for 72 h. Luminescence intensity was measured by a plate reader and the Piccagene (Measurement time: 1 s) [22, 23]. ## In vitro assay for anti-epimastigote form of T. cruzi activity Luc2-Tulahuen epimastigotes (2 × 105/well) were dispensed into 96-well plate. After 72 h incubation, luminescence intensity was measured by adding lysis buffer with luciferin as substrate (PicaGene Luminescence Kit, Fuji Film Wako chemicals, Japan) (Measurement time: 1 s) [24]. The IC50 was calculated using the following equation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{IC}}_{50}{=10}^{\mathrm{log}\left(\frac{\mathrm{A}}{\mathrm{B}}\right)\times \frac{50-C}{\mathrm{D}-C}+\mathrm{log}\left(\mathrm{B}\right)},$$\end{document}IC50=10logAB×50-CD-C+logB,where A is the lowest concentration at which the percentage inhibition exceeds $50\%$, B is the highest concentration at which percentage inhibition is less than $50\%$, and C and D are the percentage inhibition of the sample at concentrations B and A, respectively. ## Evaluation of cytotoxicity and measurement of 50% injury concentration NMH cells or HuH28 cells (1 × 104/well) of 100 μl aliquots were seeded on a 96-well black plate. Test samples were added and incubated for 72 h. NP-40 was added as a positive control ($100\%$ cytotoxicity) and a solvent in which the samples were dissolved as a negative control ($0\%$ cytotoxicity). Then 10 μl of alamarBlue reagent ($10\%$, Funakoshi Co., Tokyo, Japan) for mitochondria staining was added and incubated for another 4 h. The fluorescence intensity (544 nm/590 nm) was measured with a plate reader (measurement time: 0.1 s). The concentration required to reduce cell viability by $50\%$ (CC50) was calculated using the following equation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{CC}}_{50}{=10}^{\mathrm{log}\left(\frac{\mathrm{A}}{\mathrm{B}}\right)\times \frac{50-C}{\mathrm{D}-C}+\mathrm{log}\left(\mathrm{B}\right)},$$\end{document}CC50=10logAB×50-CD-C+logB,where A is the lowest concentration at which the cell viability exceeds $50\%$, B is the highest concentration at which cell viability is less than $50\%$, and C and D are the cell viability of the sample at concentrations B and A, respectively. ## Infection of mice Mice were housed and maintained in Nagasaki University Biomedical Research Center (12 h light/dark cycle). Female BALB/c mice from 6 to 10 weeks old (20–25 g) were used in all the experiments. Mice were obtained as wild type from SLC (Japan). In standard experiments, 5 × 103 in vitro tissue culture-derived trypomastigotes (TCTs) was intra-peritoneally (i.p.) inoculated into mouse [25–27]. The mice were handled according to the international guidelines and institutional guideline of Nagasaki University for the use and maintenance of experimental animals. Ethical approval for this study was obtained from the institutional ethical review board, Nagasaki University (approval number R12005). ## Bioluminescence imaging for efficacy evaluation Mice were intra-peritoneally injected with 150 mg/kg d-luciferin (SYD labs, USA), then anesthetized using $2.5\%$ (vol/vol) gaseous isoflurane in oxygen. To measure bioluminescence, mice were placed in an IVIS Lumina II system (Caliper Life Science, USA) and images were acquired 10 min after d-luciferin administration using LivingImage 4.3 (Caliper Life Sciences, USA). Exposure time was fixed as 5 min. Anesthesia was maintained throughout the imaging process through the nose cone. Whole body luminescence was determined by drawing the region of interest (ROI) and quantifying the bioluminescence expressed as total flux (photons/second; p/s) [16, 28]. Coptisine chloride was administered intra-peritoneally (i.p.) at a dose of 30 mg/kg twice a day. Benznidazole (100 mg/kg) as a positive control and the drug solvent as a negative control were administered orally (p.o.) once a day. Drug was administered for 5 consecutive days from day 4 to 8 of infection. Measurements with IVIS Lumina II were performed at five different times: on day 3 of infection (the day before drug treatment), on day 9 of infection (the day after drug treatment was completed), on day 14, 28 and 40 of infection. To maintain fairness, different experimenters performed drug treatments and luminescence intensity measurements. At day 29, BALB/c mice were injected with cyclophosphamide (Sigma-Aldrich, USA) (200 mg/kg) as an immunosuppressant by i.p. and were followed by a maximum of 3 doses at 3 day intervals [16, 29]. The use of immunosuppressant like cyclophosphamide could help enhance the visibility of the infected lesions even after 100 dpi. In this animal study, treatment was given when the luminescence intensity was high. In addition, an immunosuppressant was administered when the difference in luminescence intensity among the three groups became small due to natural immunity of the mice. ## Statistical analysis Data were tabulated on Microsoft Excel and statistically analyzed. For statistical comparisons, ANOVA analysis was used to determine the statistical significance of difference in values from the control groups. Data were expressed as the mean ± standard error, and the results were obtained from at least three independent experiments. A p-value < 0.05 was considered statistically significant. ## In vitro assay One hundred and twenty extracts and 94 compounds were screened for their cytotoxic effect on the trypomastigotes and amastigotes using a mixture culture experiment (Fig. 1), some sample were shown to exhibit more than $80\%$ reduction of parasite signals. Those positive samples were Phellodendron bark (the bark of Phellodendron amurense Ruprecht) and Coptis rhizome (the rhizome of Coptis japonica Makino, Coptis chinensis Franchet, Coptis deltoidea C.Y. Cheng et Hsiao or Coptis teeta Wallich) from the extracts, Alison B, Alkanin, Berberin chloride, Coptisine chloride, Dehydrocorydaline nitrate, Palmatine chloride, Shikonin and Timosaponin A-III from the compounds indicated by red colored in Fig. 1. Similar set of the Kampo library was applied to mouse fibroblast cells (NMH) and human bile duct carcinoma cell line (HuH28) to observe any cytotoxic effect on host cells as shown in Fig. 2. Within two extract samples (Phellodendron bark, Coptis rhizome) and 8 compounds samples (Alison B, Alkanin, Berberin chloride, Coptisine chloride, Dehydrocorydaline nitrate, Palmatine chloride, Shikonin, Timosaponin A-III) which showed more than $80\%$ inhibition in the T. cruzi mixture culture (Fig. 1), only three samples (Coptisine chloride, Dehydrocorydaline nitrate, Palmatine chloride) showed relatively lower cytotoxicity (Fig. 2). Therefore, we decided to select those three positive compounds (Coptisine chloride, Dehydrocorydaline nitrate, Palmatine chloride) for further analysis. Those three compounds shared the same isoquinoline skeleton namely coptisine chloride, dehydrocorydaline nitrate, and palmatine chloride. After the first screening as shown in Figs. 1 and 2 as a representative of three repeated experiments, we determined IC50 and CC50 values of the selected candidate compounds. As performed in the first in vitro screening, we used two T. cruzi-culture systems for IC50 as shown in Table 1. For this second screening, we added three more compounds, epiberberine chloride, berberrubine chloride, dl-tetrahydro coptisine to already selected compounds belonging to the same protoberberine-type alkaloids (Fig. 3, Table 1).Table 1IC50, and CC50, of different compounds on T. cruzi (Trypo: trypomastigote, Ama: amastigote, Epi: epimastigote) and mammalian cell (NMH, HuH28) linesCellsT. cruzi Trypo + AmaT. cruzi Intracell AmaT. cruzi EpiNMHHuH28AssayLuciferase activityAlamarBlue viabilityIC50 (μM)CC50 (μM)Coptisine chloride4.484.9815.1> 40> 100Dehydrocorydaline nitrate3.477.8917.3> 40> 100Palmatine chloride2.6617.422.3> 40> 100Epiberberine chloride> 40> 40NT> 40> 100Berberrubine chloride20.133.5NT> 40> 100dl-Tetrahydro coptisine> 40> 40NT> 40> 100Benznidazole2.784.913.78> 40> 100Each experiment was repeated three times and representative data were shownFig. 3Chemical structure of the investigated compounds The reduced luminescence intensity of luciferin-expressing protozoa suggested that coptisine chloride (IC50: 4.48 μM), dehydrocorydaline nitrate (IC50: 3.47 μM) and palmatine chloride (IC50: 2.66 μM) had an antiprotozoan effect (trypomastigotes in the medium or intracellular amastigotes). It was found to be as effective as benznidazole (IC50: 2.78 μM) in activity quantification (Additional file 1: Fig. S1, Table S1). Next, the effect on proliferative amastigotes in trypanosomes infected with NMH cells was estimated. Coptisine chloride (IC50: 4.98 μM) was the most effective as benznidazole (IC50: 4.91 μM) due to reduced luminescence intensity in luciferin-expressing protozoa (Additional file 1: Fig. S2, Table S1). On the other hand, coptisine chloride (IC50: 15.1 µM) was fourfold less effective as benznidazole (IC50: 3.78 µM) against epimastigotes (Additional file 1: Fig. S3, Table S1). Cytotoxicity to NMH cells or HuH28 cells was examined for those 6 compounds (Additional file 1: Figs S4, S5). These results indicate that there are differences in the effectiveness of the isoquinoline skeleton depending on the number of methoxy groups and side chains (Fig. 3, Table 1). ## In vivo assay After a 5-day treatment regimen with coptisine chloride (30 mg/kg bid i.p.) from day 4 to 8, the luminescence intensity was significantly reduced in the treated group compared with the untreated group. Though the luminescence intensity of the coptisine chloride group was slightly higher than that of the benznidazole group, still significant reduction was observed, as shown in Figs. 4 and 5. It was found that the luminescence intensity of the coptisine chloride-treated group was significantly reduced even after the immune suppression treatment (day 40) compared with the control group (Figs. 4, 5). Therefore, it was shown that coptisine chloride has a therapeutic effect on the acute phase in vivo model. Fig. 4Female BALB/c mice infected with 5 × 103 trypomastigotes were treated with coptisine chloride (30 mg/kg, 2 × i.p.); positive control is benznidazole (100 mg/kg, 1 × p.o.); negative control is the drug solvent. Bioluminescence imaging was obtained in an IVIS Lumina II systemFig. 5Luminescence intensity at different timepoints in the mice model infected with 5 × 103 trypomastigotes and treated with coptisine chloride (COP) (30 mg/kg, 2 × i.p.); positive control is benznidazole (BNZ) (100 mg/kg, 1 × p.o.); negative control is the drug solvent. Bioluminescence imaging was obtained in an IVIS Lumina II system. Data are presented as mean and standard deviation in intensity of three experiments ($p \leq 0.05$, $p \leq 0.01$, *$p \leq 0.005$ when compared with vehicle). In this experiment, we excluded mice with luminescence intensity less than 1 × 107 in pre-drug measurements (Day 3) ## Discussion Chagas disease is expanding to non-endemic areas and an effective treatment is needed. Herbal compounds from traditional medicine are promising candidates for drug development, and their efficacy was demonstrated in our study. The experimental system of mixing trypomastigotes and NMH cells confirmed the effects on trypomastigotes in the medium and amastigotes in the cells. The reduced luminescence intensity of luciferin-expressing protozoa suggested that coptisine (IC50: 4.48 μM), dehydrocorydaline (IC50: 3.47 μM) and palmatine (IC50: 2.66 μM) had an antiprotozoal effect. It was found to be as effective as benznidazole (IC50: 2.78 μM) in activity quantification (Table 1). The experiment mimics the acute phase of the disease with trypomastigotes in the blood and amastigotes in the cells [30, 32]. As a result, these compounds are thought to be effective in the acute phase. In addition, Phellodendron bark and Coptis rhizome showed more than $80\%$ inhibition in the T. cruzi mixture culture (Fig. 1). However, these extracts also showed cytotoxicity. These consist of berberine, coptisine and palmatine. The cytotoxicity of Phellodendron bark and Coptis rhizome may be attributed to berberine. However, the urgent requirement from the present clinical setting in the world is to get an effective curative drug for chronic phase infection [1–3]. Therefore, future work will focus on testing the Kampo compounds in the chronic stage of infection. The experimental system for intracellular amastigotes using NMH cells allowed us to confirm the effect on proliferative amastigotes in trypanosomes-infected NMH cells which mimics the chronic phase of the disease [31, 32]. Coptisine (IC50: 4.98 μM) showed comparable levels of IC50 as benznidazole (IC50: 4.91 μM) (Table 1). Although we have not yet examined in vivo chronic Chagas model, coptisine was revealed to be a possible candidate for an effective drug in the chronic phase. As shown in Table 1, other related protoberberine-type alkaloids, dehydrocorydaline and palmatine showed strong activity in the mixture experiment with trypomastigote, amastigote and NMH, but not against intracellular amastigotes. When those three active compounds were examined for their cytotoxicity on epimastigote, they showed almost equal effectiveness of IC50 around 15–20 μM, which were less effective compared with benznidazole (3.78 μM). Therefore, coptisine is specifically more active to amastigote than to epimastigote, whereas benznidazole shows both stages active. In addition, three active protoberberine-type alkaloids showed equal cytotoxicity on the T. cruzi epimastigote cells, but the effect on the intracellular amastigote significantly increased in only coptisine which might be due to the increased permeability through host cell membrane. Although comparison on the activities of tested compounds did not show clear structure–activity relationships, following trends were observed. The presence of methylenedioxy at rings A (at C-2 and 3) and D (C-9 and 10) and aromatization of ring C could be the key factor for the anti-Trypanosoma cruzi activity. However, the effects of the presence of methylenedioxy or methoxy groups of rings A/D were not clear. The presence of methyl group at C-13 may reduce the anti-Trypanosoma cruzi activity on the intracellular amastigote. The presence of a hydroxy group at C-9 may reduce the activity. To clarify the structure–activity relationships of the protoberberine-type alkaloids on anti-Trypanosoma cruzi activity, further studies using more diverse derivatives are required. In animal experiments, we confirmed the equivalent parasiticidal activity of coptisine as benznidazole as evidenced by the reduction in luminescence intensity shown in Figs. 4 and 5. In addition, the relapse detected on day 40 after administration of the immunosuppressant during a period between day 30 and 39 was smaller than that in the benznidazole treatment group. This suggests that coptisine was more effective to reduce the number of parasites compared with the current standard curative protocol of benznidazole in the acute model [28]. Taken together, coptisine is a novel anti-Trypanosoma cruzi compound with an equivalent effectiveness as benznidazole when examined in vitro acute and chronic and by in vivo acute model. Our results also indicate that berberine structure with an isoquinoline skeleton might be used for more effective compounds design by the modification of side chains. It has been reported that coptisine chloride or protoberberine-type alkaloids has various pharmaceutical effects such as anti-cancer [36, 37], anti-inflammation, and anti-diabetic as well as anti-infectious diseases [33–35, 38, 39]. Coptisine non-competitively inhibits DHODH in Plasmodium falciparum* and showed weak inhibitory activity against human DHODH [35]. T. cruzi DHODH, which catalyzes the production of orotate, and was demonstrated to be essential for T. cruzi survival, could be one of our coptisine’s potential targets [40–42]. It has been reported that coptisine chloride possesses anti-cancer activity through the inhibition of PI3K/Akt/mTOR signaling and subsequent mitochondrial ROS production in the hepatocellular cancer Hep3B cells [36, 37]. T. cruzi infection has been demonstrated to stimulate host PI3K signaling in human and mice macrophages that allow intracellular parasites to grow and survive [43–47]. These reports suggest that inhibition of PI3K/Akt/mTOR signaling by coptisine chloride may have affected the growth and survival of intracellular parasites. Coptisine chloride has also been reported to inhibit LPS-stimulated inflammation by blocking the activation of NF-κB and MAPK in macrophages [48–51]. In coptisine chloride therapy for colitis, coptisine chloride significant suppresses mRNA expression, releases of pro-inflammatory cytokines (TNF-α, IFN-γ, IL-1β, IL-6, IL-17) and enhances the mRNA expression level of IL-10, an anti-inflammatory cytokine [52, 53]. T. cruzi infection could stimulate both protective and pathogenic host immune responses through natural or adaptive immunological pathways [54–59]. Although we did not see any significant adverse events during in vivo experiment, previously reported immune inhibitory effects of coptisine should be carefully observed. When considering drug development, the target product profile (TPP) is important, and oral administration is preferable for aiming a therapeutic drug for chronic Chagas disease that may need at least 2 weeks regimen. However, it is known that the oral bioavailability of coptisine chloride is low [60]. It is necessary to investigate whether these problems can be solved by improving the side chain or enclosing the compound to improve the blood concentration. In the present study, we identified a series of protoberberine-type alkaloids as strong candidate anti-Trypanosoma cruzi medicine using in vitro and in vivo models. Among them, coptisine was expected to be the most effective candidate compound. The model mouse used this time has poor persistence of luminescence intensity, and it is considered difficult to measure in the chronic phase. However, it may be possible to measure the chronic phase using new recombinant protozoa. In this study, we found that compounds identified in vitro are also effective in vivo. Since chronic Chagas treatment is a typical neglected tropical disease unmet needs [61, 62], further development must be facilitated. ## Supplementary Information Additional file 1. Supplementary tables and figures. ## References 1. Chagas diseases in Latin America: an epidemiological update based on 2010 estimates. Wkly Epidemiol Rec (2015.0) 90 33-44. PMID: 25671846 2. 2.World Health Organization 2021, ‘Chagas disease (American trypanosomiasis)’, viewed 20 of January 2021, http://www.who.int/chagas/disease/en/. 3. Lewis MD, Fortes Francisco A, Taylor MC, Burrell-Saward H, McLatchie AP, Miles MA, Kelly JM. Bioluminescence imaging of chronic. Cell Microbiol (2014.0) 16 1285-1300. DOI: 10.1111/cmi.12297 4. Hyland KV, Asfaw SH, Olson CL, Daniels MD, Engmana DM. Bioluminescent imaging of. Int J Parasitol (2008.0) 38 1391-1400. DOI: 10.1016/j.ijpara.2008.04.002 5. Machado FS, Dutra WO, Esper L, Gollob KJ, Teixeira MM, Factor SM, Weiss LM, Nagajyothi F, Tanowitz HB, Garg NJ. Current understanding of immunity to. Semin Immunopathol (2012.0) 34 753-770. DOI: 10.1007/s00281-012-0351-7 6. Schmunis GA, Yadon ZE. Chagas disease: a Latin American health problem becoming a world health problem. Acta Trop (2010.0) 115 14-21. DOI: 10.1016/j.actatropica.2009.11.003 7. Moraes CB, Giardini MA, Kim H, Franco CH, Araujo-Junior AM, Schenkman S, Chatelain E, Freitas-Junior LH. Nitroheterocyclic compounds are more efficacious than CYP51 inhibitors against. Sci Rep (2014.0) 4 4703. DOI: 10.1038/srep04703 8. Palmeiro-Roldán R, Fonseca-Berzal C, Gómez-Barrio A, Arán VJ, Escario JA, Torrado-Durán S, Torrado-Santiago S. Development of novel benznidazole formulations: physicochemical characterization and in vivo evaluation on parasitemia reduction in Chagas disease. Int J Pharm (2014.0) 472 110-117. DOI: 10.1016/j.ijpharm.2014.06.015 9. Díaz-Chiguer DL, Márquez-Navarro A, Nogueda-Torres B, de la Luz G, León-Ávila J-V, Hernández-Campos A. In vitro and in vivo trypanocidal activity of some benzimidazole derivatives against two strains of. Acta trop (2012.0) 122 108-112. DOI: 10.1016/j.actatropica.2011.12.009 10. Lascano F, Bournissen FG, Altcheh J. Review of pharmacological options for the treatment of Chagas disease. Br J Clin Pharmacol (2022.0) 88 383-402. DOI: 10.1111/bcp.14700 11. Cheng JT. Drug therapy in Chinese traditional medicine. J Clin Pharmacol (2000.0) 40 445-450. DOI: 10.1177/00912700022009198 12. Hoffmann KM, Herbrechter R, Ziemba PM, Lepke P, Beltrán L, Hatt H, Werner M, Gisselmann G. Kampo medicine: evaluation of the pharmacological activity of 121 herbal drugs on GABAA and 5-HT3A receptors. Front Pharmacol (2016.0) 7 219. DOI: 10.3389/fphar.2016.00219 13. Chen X, Xiang L, Shi L, Li G, Yao H, Han J, Lin Y, Song J, Chen J. Identification of crude drugs in the Japanese pharmacopoeia using a DNA barcoding system. Sci Rep (2017.0) 7 42325. DOI: 10.1038/srep42325 14. Li F-S, Weng J-K. Demystifying traditional herbal medicine with modern approach. Nat Plants (2017.0) 3 17109. DOI: 10.1038/nplants.2017.109 15. Pan S-Y, Litscher G, Gao S-H, Zhou S-F, Zhi-Ling Yu, Chen H-Q, Zhang S-F, Tang M-K, Sun J-N, Ko K-M. Historical perspective of traditional indigenous medical practices: the current renaissance and conservation of herbal resources. Evid Based Complementary Altern Med (2014.0) 2014 525340. DOI: 10.1155/2014/525340 16. Francisco AF, Lewis MD, Jayawardhana S, Taylor MC, Chatelain E, Kelly JM. Limited ability of posaconazole to cure both acute and chronic. Antimicrob Agents Chemother (2015.0) 59 4653-4661. DOI: 10.1128/AAC.00520-15 17. Narahara C, Saeheng T, Chaijaroenkul W, Dumre SP, Na-Bangchang K, Karbwang J. β-Eudesmol induces the expression of apoptosis pathway proteins in cholangiocarcinoma cell lines. J Res Med Sci (2020.0) 25 7. DOI: 10.4103/jrms.JRMS_291_19 18. Hashimoto M, Morales J, Fukai Y, Suzuki S, Takamiya S, Tsubouchi A, Inoue S, Inoue M, Kita K, Harada S, Tanaka A, Aoki T, Nara T. Critical importance of the de novo pyrimidine biosynthesis pathway for. Biochem Biophys Res Commun (2012.0) 417 1002-1006. DOI: 10.1016/j.bbrc.2011.12.073 19. Tani O, Akutsu Y, Ito S, Suzuki T, Tateishi Y, Yamaguchi T, Niimi T, Namatame I, Chiba Y, Sakashita H, Kubota T, Yanagi T, Mizukami S, Hirayama K, Furukawa K, Yamasaki K. NMR biochemical assay for oxidosqualene cyclase: evaluation of inhibitor activities on. J Med Chem (2018.0) 61 5047-5053. DOI: 10.1021/acs.jmedchem.8b00484 20. Bettiol E, Samanovic M, Murkin AS, Raper J, Buckner F, Rodriguez A. Identification of three classes of heteroaromatic compounds with activity against intracellular. PloS Negl Trop Dis (2009.0) 3 e384. DOI: 10.1371/journal.pntd.0000384 21. Alonso-Padilla J, Cotillo I, Presa JL, Cantizani J, Peña I, Bardera AI, Martín JJ, Rodriguez A. Automated high-content assay for compounds selectively toxic to. PLoS Negl Trop Dis (2015.0) 9 e0003493. DOI: 10.1371/journal.pntd.0003493 22. De Rycker M, Thomas J, Riley J, Brough SJ, Miles TJ, Gray DW. Identification of trypanocidal activity for known clinical compounds using a new. PLoS Negl Trop Dis (2016.0) 10 e0004584. DOI: 10.1371/journal.pntd.0004584 23. Imanol Peña M, Manzano P, Cantizani J, Kessler A, Alonso-Padilla J, Bardera AI, Alvarez E, Colmenarejo G, Cotillo I, Roquero I, de Dios-Anton F, Barroso V, Rodriguez A, Gray DW, Navarro M, Kumar V, Sherstnev A, Drewry DH, Brown JR, Fiandor JM, Julio Martin J. New compound sets identified from high throughput phenotypic screening against three kinetoplastid parasites: an open resource. Sci Rep (2015.0) 5 8771. DOI: 10.1038/srep08771 24. Zuma AA, da Silva RB, Simon J, Garden SJ, de Souza W. In vitro study of the trypanocidal activity of anilinophenanthrolines against. Parasitol Int (2021.0) 83 102338. DOI: 10.1016/j.parint.2021.102338 25. Caldas S, Caldas IS, de Figueiredo DL, de Lima WG, de Paula OR, Cecílio AB, Ribeiro I, Talvani A, Bahia MT. Real-time PCR strategy for parasite quantification in blood and tissue samples of experimental. Acta Trop (2012.0) 123 170-177. DOI: 10.1016/j.actatropica.2012.05.002 26. Assíria Fontes Martins T, de Figueiredo Diniz L, Mazzeti AL, da Silva do Nascimento AF, Caldas S, Caldas IS, de Andrade IM, Ribeiro I, Bahia MT. Benznidazole/itraconazole combination treatment enhances anti-. PLoS ONE (2015.0) 10 e0128707. DOI: 10.1371/journal.pone.0128707 27. Novaes RD, Santos EC, Cupertino MC, Bastos DS, Oliveira JM, Carvalho TV, Neves MM, Oliveira LL, Talvani A. Parasitol Res (2015.0) 114 2873-2881. DOI: 10.1007/s00436-015-4488-x 28. Francisco AF, Jayawardhana S, Lewis MD, White KL, Shackleford DM, Chen G, Saunders J, Osuna-Cabello M, Read KD, Charman SA, Chatelain E, Kelly JM. Nitroheterocyclic drugs cure experimental. Sci Rep (2016.0) 6 35351. DOI: 10.1038/srep35351 29. Bahia MT, Nascimento AF, Mazzeti AL, Marques LF, Gonçalves KR, Mota LW, Diniz LD, Caldas IS, Talvani A, Shackleford DM, Koltun M, Saunders J, White KL, Scandale I, Charman SA, Chatelain E. Antitrypanosomal activity of fexinidazole metabolites, potential new drug candidates for Chagas disease. Antimicrob Agents Chemother (2014.0) 58 4362-4370. DOI: 10.1128/AAC.02754-13 30. Teixeira AR, Hecht MM, Guimaro MC, Sousa AO, Nitz N. Pathogenesis of Chagas’ disease: parasite persistence and autoimmunity. Clin Microbiol Rev (2011.0) 3 592-630. DOI: 10.1128/CMR.00063-10 31. Coura JR, Borges-Pereira J. Chagas disease: 100 years after its discovery. A systemic review. Acta Trop (2010.0) 115 5-13. DOI: 10.1016/j.actatropica.2010.03.008 32. 32.Kirchhoff L. Chagas Disease (American Trypanosomiasis) Clinical Presentation: History, Physical, Causes [Internet]. Emedicine.medscape.com. 2019 [cited 9 July 2021]. Available from: https://emedicine.medscape.com/article/214581-clinical. 33. 33.Zhang Z, Deng A, Yu J. Current research status on pharmacological activities of coptisine. China J Chin Mater Med. 2013:2750–2754. 34. Jiasi Wu, Luo Yu, Deng D, Siyu Su, Li S, Xiang Li, Yingfan Hu, Wang P, Meng X. Coptisine from Coptis chinensis exerts diverse beneficial properties: a concise review. J Cell Mol Med (2019.0) 12 7946-7960 35. Lang L, Hu Q, Wang J, Liu Z, Huang J, Lu W, Huang Y. Coptisine, a natural alkaloid from Coptidis Rhizoma, inhibits plasmodium falciparum dihydroorotate dehydrogenase. Chem Biol Drug Des (2018.0) 92 1324-1332. DOI: 10.1111/cbdd.13197 36. Kim SY, Hwangbo H, Kim MY, Ji SY, Lee H, Kim G-Y, Kwon C-Y, Leem SH, Hong SH, Cheong JH, Choi YH. Coptisine induces autophagic cell death through down-regulation of PI3K/Akt/mTOR signaling pathway and up-regulation of ROS-mediated mitochondrial dysfunction in hepatocellular carcinoma Hep3B cells. Arch Biochem Biophys (2021.0) 697 108688. DOI: 10.1016/j.abb.2020.108688 37. Kim SY, Hwangbo H, Lee H, Park C, Kim G-Y, Moon S-K, Yun SJ, Kim W-J, Cheong J, Choi YH. Induction of apoptosis by coptisine in Hep3B hepatocellular carcinoma cells through activation of the ROS-mediated JNK signaling pathway. Int J Mol Sci (2020.0) 21 5502. DOI: 10.3390/ijms21155502 38. Teklemichael AA, Mizukami S, Toume K, Mosaddeque F, Kamel MG, Kaneko O, Komatsu K, Karbwang J, Huy NT, Hirayama K. Anti-malarial activity of traditional Kampo medicine. Malar J (2020.0) 19 204. DOI: 10.1186/s12936-020-03273-x 39. Nonaka M, Murata Y, Takano RY, Han Y, Bin Kabir MH, Kentaro Kato K. Screening of a library of traditional Chinese medicines to identify anti-malarial compounds and extracts. Malar J (2018.0) 17 244. DOI: 10.1186/s12936-018-2392-4 40. Cheleski J, Rocha JR, Pinheiro MP, Wiggers HJ, da Silva ABF, Nonato MC, Montanari CA. Novel insights for dihydroorotate dehydrogenase class 1A inhibitors discovery. Eur J Med Chem (2010.0) 45 5899-5909. DOI: 10.1016/j.ejmech.2010.09.055 41. Munier-Lehmann H, Vidalain PO, Tangy F, Janin YL. On dihydroorotate dehydrogenases and their inhibitors and uses. J Med Chem (2013.0) 56 3148-3167. DOI: 10.1021/jm301848w 42. Inaoka DK, Iida M, Hashimoto S, Tabuchi T, Kuranaga T, Balogun EO, Honma T, Tanaka A, Harada S, Nara T, Kita K, Inoue M. Design and synthesis of potent substrate-based inhibitors of the. Bioorg Med Chem (2017.0) 25 1465-1470. DOI: 10.1016/j.bmc.2017.01.009 43. Chuenkova MV, PereiraPerrin M. Sci Signal. (2009.0) 2 ra74. DOI: 10.1126/scisignal.2000374 44. Wilkowsky SE, Barbieri MA, Stahl P, Isola EL. Exp Cell Res (2001.0) 264 211-218. DOI: 10.1006/excr.2000.5123 45. Todorov AG, Einicker-Lamas M, de Castro SL, Oliveira MM, Guilherme A. Activation of host cell phosphatidylinositol 3-kinases by. J Biol Chem (2000.0) 275 32182-32186. DOI: 10.1074/jbc.M909440199 46. López-Peláez M, Soria-Castro I, Boscá L, Fernández M, Alemany S. Cot/tpl2 activity is required for TLR-induced activation of the Akt p70 S6k pathway in macrophages: implications for NO synthase 2 expression. Eur J Immunol (2011.0) 41 1733-1741. DOI: 10.1002/eji.201041101 47. Silva MC, Davoli-Ferreira M, Medina TS, Sesti-Costa R, Silva GK, Lopes CD, Cardozo LE, Gava FN, Lyroni K, Dias FC, Frade AF, Baron M, Nakaya HI, Figueiredo F, Alves-Filho JC, Cunha FQ, Tsatsanis C, Chevillard C, Cunha-Neto E, Hirsch E, Silva JS, Cunha TM. Canonical PI3Kγ signaling in myeloid cells restricts. Nat Commun (2018.0) 9 1513. DOI: 10.1038/s41467-018-03986-3 48. Jiasi Wu, Zhang H, Boyang Hu, Yang L, Wang P, Wang F, Meng X. Coptisine from Coptis chinensis inhibits production of inflammatory mediators in lipopolysaccharide-stimulated RAW 264.7 murine macrophage cells. Eur J Pharmacol (2016.0) 780 106-114. DOI: 10.1016/j.ejphar.2016.03.037 49. Chen H-B, Luo C-D, Liang J-L, Zhang Z-B, Lin G-S, Jia-Zhen Wu, Li C-L, Tan L-H, Yang X-B, Zi-Ren Su, Xie J-H, Zeng H-F. Anti-inflammatory activity of coptisine free base in mice through inhibition of NF-κB and MAPK signaling pathways. Eur J Pharmacol (2017.0) 811 222-231. DOI: 10.1016/j.ejphar.2017.06.027 50. Zou Z-Y, Yin-Ran Hu, Ma H, Wang Y-Z, He K, Xia S, Hao Wu, Xue D-F, Li X-G, Ye X-L. Coptisine attenuates obesity-related inflammation through LPS/TLR-4-mediated signaling pathway in Syrian golden hamsters. Fitoterapia (2015.0) 105 139-146. DOI: 10.1016/j.fitote.2015.06.005 51. Zhou K, Hu LI, Liao W, Yin D, Rui F. Coptisine prevented IL-β-induced expression of inflammatory mediators in chondrocytes. Inflammation (2016.0) 39 1558-1565. DOI: 10.1007/s10753-016-0391-6 52. Ai G, Huang Z, Cheng J, Xie J, Zeng H, Liu Y, Li Y, Huang X, Chen J, Ziren Su. Gut microbiota-mediated transformation of coptisine into a novel metabolite 8-oxocoptisine: insight into its superior anti-colitis effect. Front Pharmacol (2021.0) 12 639020. DOI: 10.3389/fphar.2021.639020 53. Wang Y, Liu J, Huang Z, Li Y, Liang Y, Luo C, Ni C, Xie J, Ziren Su, Chen J, Li C. Coptisine ameliorates DSS-induced ulcerative colitis via improving intestinal barrier dysfunction and suppressing inflammatory response. Eur J Pharmacol (2021.0) 896 173912. DOI: 10.1016/j.ejphar.2021.173912 54. Campos MA, Closel M, Valente EP, Cardoso JE, Akira S, Alvarez-Leite JI, Ropert C, Gazzinelli RT. Impaired production of proinflammatory cytokines and host resistance to acute infection with. J Immunol (2004.0) 172 1711-1718. DOI: 10.4049/jimmunol.172.3.1711 55. Mukherjee S, Huang H, Petkova SB, Albanese C, Pestell RG, Braunstein VL, Christ GJ, Wittner M, Lisanti MP, Berman JW, Weiss LM, Tanowitz HB. Infect Immun (2004.0) 72 5274-5282. DOI: 10.1128/IAI.72.9.5274-5282.2004 56. Soto CD, Solana ME, Poncini CV, Pino-Martinez AM, Tekiel V, González-Cappa SM. Dendritic cells devoid of IL-10 induce protective immunity against the protozoan parasite. Vaccine (2010.0) 28 7407-7413. DOI: 10.1016/j.vaccine.2010.08.105 57. Suman S, Rachakonda G, Mandape SN, Sakhare SS, Villalta F, Pratap S, Lima MF, Nde PN. Phospho-proteomic analysis of primary human colon epithelial cells during the early. PLoS Negl Trop Dis (2018.0) 12 e0006792. DOI: 10.1371/journal.pntd.0006792 58. Wan X, Wen JJ, Koo SJ, Liang LY, Garg NJ. SIRT1-PGC1α-NFκB pathway of oxidative and inflammatory stress during. PLoS Pathog (2016.0) 12 e1005954. DOI: 10.1371/journal.ppat.1005954 59. Vásquez VC, Russomando G, Espínola EE, Sanchez Z, Mochizuki K, Roca Y, Revollo J, Guzman A, Quiroga B, Rios Morgan S, Vargas Ortiz R, Zambrana Ortega A, Espinoza E, Nishizawa JE, Kamel MG, Kikuchi M, Mizukami S, Na-Bangchang K, Tien Huy N, Hirayama K. IL-17A, a possible biomarker for the evaluation of treatment response in. PLoS Negl Trop Dis (2019.0) 13 e0007715. DOI: 10.1371/journal.pntd.0007715 60. Kheir MM, Wang Y, Hua L, Jun Hu, Li L, Lei F, Lijun Du. Acute toxicity of berberine and its correlation with the blood concentration in mice. Food Chem Toxicol (2010.0) 48 1105-1110. DOI: 10.1016/j.fct.2010.01.033 61. Francisco AF, Jayawardhana S, Olmo F, Lewis MD, Wilkinson SR, Taylor MC, Kelly JM. Challenges in Chagas disease drug development. Molecules (2020.0) 25 2799. DOI: 10.3390/molecules25122799 62. Beltran-Hortelano I, Alcolea V, Font M, Pérez-Silanes S. Examination of multiple. Bioorg Med Chem (2022.0) 58 116577. DOI: 10.1016/j.bmc.2021.116577
--- title: Circulating amino acid levels and colorectal cancer risk in the European Prospective Investigation into Cancer and Nutrition and UK Biobank cohorts authors: - Joseph A. Rothwell - Jelena Bešević - Niki Dimou - Marie Breeur - Neil Murphy - Mazda Jenab - Roland Wedekind - Vivian Viallon - Pietro Ferrari - David Achaintre - Audrey Gicquiau - Sabina Rinaldi - Augustin Scalbert - Inge Huybrechts - Cornelia Prehn - Jerzy Adamski - Amanda J. Cross - Hector Keun - Marc Chadeau-Hyam - Marie-Christine Boutron-Ruault - Kim Overvad - Christina C. Dahm - Therese Haugdahl Nøst - Torkjel M. Sandanger - Guri Skeie - Raul Zamora-Ros - Kostas K. Tsilidis - Fabian Eichelmann - Matthias B. Schulze - Bethany van Guelpen - Linda Vidman - Maria-José Sánchez - Pilar Amiano - Eva Ardanaz - Karl Smith-Byrne - Ruth Travis - Verena Katzke - Rudolf Kaaks - Jeroen W. G. Derksen - Sandra Colorado-Yohar - Rosario Tumino - Bas Bueno-de-Mesquita - Paolo Vineis - Domenico Palli - Fabrizio Pasanisi - Anne Kirstine Eriksen - Anne Tjønneland - Gianluca Severi - Marc J. Gunter journal: BMC Medicine year: 2023 pmcid: PMC9976469 doi: 10.1186/s12916-023-02739-4 license: CC BY 4.0 --- # Circulating amino acid levels and colorectal cancer risk in the European Prospective Investigation into Cancer and Nutrition and UK Biobank cohorts ## Abstract ### Background Amino acid metabolism is dysregulated in colorectal cancer patients; however, it is not clear whether pre-diagnostic levels of amino acids are associated with subsequent risk of colorectal cancer. We investigated circulating levels of amino acids in relation to colorectal cancer risk in the European Prospective Investigation into Cancer and Nutrition (EPIC) and UK Biobank cohorts. ### Methods Concentrations of 13-21 amino acids were determined in baseline fasting plasma or serum samples in 654 incident colorectal cancer cases and 654 matched controls in EPIC. Amino acids associated with colorectal cancer risk following adjustment for the false discovery rate (FDR) were then tested for associations in the UK Biobank, for which measurements of 9 amino acids were available in 111,323 participants, of which 1221 were incident colorectal cancer cases. ### Results Histidine levels were inversely associated with colorectal cancer risk in EPIC (odds ratio [OR] 0.80 per standard deviation [SD], $95\%$ confidence interval [CI] 0.69–0.92, FDR P-value=0.03) and in UK Biobank (HR 0.93 per SD, $95\%$ CI 0.87–0.99, P-value=0.03). Glutamine levels were borderline inversely associated with colorectal cancer risk in EPIC (OR 0.85 per SD, $95\%$ CI 0.75–0.97, FDR P-value=0.08) and similarly in UK Biobank (HR 0.95, $95\%$ CI 0.89–1.01, $$P \leq 0.09$$) In both cohorts, associations changed only minimally when cases diagnosed within 2 or 5 years of follow-up were excluded. ### Conclusions Higher circulating levels of histidine were associated with a lower risk of colorectal cancer in two large prospective cohorts. Further research to ascertain the role of histidine metabolism and potentially that of glutamine in colorectal cancer development is warranted. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12916-023-02739-4. ## Background Colorectal cancer is the third most common cancer globally, with around 1.9 million cases diagnosed in 2020, and the second most common cause of cancer-related death [1]. There is great potential to reduce this burden since most colorectal cancer cases are sporadic [2] and are associated with modifiable risk factors such as body fatness [3], alcohol intake [4], and diet [5]. Colorectal cancer development is also influenced by metabolic factors [6, 7]. For example, insulin and insulin-like growth factors are thought to play causal roles in colorectal tumorigenesis [8], likely through the promotion of cell proliferation and growth signaling pathways [9]. Broad metabolic dysfunction may lead to perturbed small-molecule metabolism, which in turn elicits bioactivity at the level of tissues and organs. Amino acids are among the most abundant circulating metabolites and serve as building blocks of proteins, precursors of many signaling molecules, and an important energy source via the citric acid cycle. Certain amino acids may also fuel cancer development [10], and marked changes in blood amino acid concentrations have been extensively observed in colorectal cancer patients [11]. For example, levels of amino acids such as glutamine, citrulline, alanine, and histidine have been inversely associated with advancing disease stage [12, 13], while valine and leucine were among the metabolites that distinguished colorectal cancer cases using a discovery-replication strategy [14]. Similarly, the concentrations of several blood amino acids distinguished early-stage colorectal cancer cases from controls in Japanese patients, most notably aspartic acid [15], as well as ornithine and lysine [16]. Glutamine was a notable discriminant in patients newly diagnosed with colorectal cancer compared to controls in a Chinese hospital-based study [17]. Overall, amino acid levels were generally inversely associated with prevalent colorectal neoplasia, suggesting a depletion of serological concentrations in cases compared to healthy individuals. Amino acid profiling could therefore potentially help identify early-stage disease [18], as well as providing insights into mechanisms of carcinogenesis. Despite these observations, few prospective studies have been conducted to test the hypothesis that pre-diagnostic amino acid concentrations are associated with colorectal cancer risk. Two such studies of nested case-control design that analyzed pre-diagnostic serum or plasma by untargeted metabolomics found limited dysregulation of lipophilic metabolites only [19, 20], while in a case-control study nested in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort that measured tryptophan and serotonin levels, tryptophan was inversely associated with colon cancer [21]. The aim of the current study was thus to test these associations in a larger and more comprehensive analysis. We first employed the EPIC nested case-control study as a discovery cohort, which measured between 13 and 21 amino acids in fasting plasma or serum in relation to colorectal cancer. In a replication step, we tested those amino acids associated with colorectal cancer risk in EPIC in the UK Biobank cohort, in which 9 overlapping compounds had been measured in over 111,000 participants. Together, the two cohorts allow for the largest and most detailed investigation of circulating amino acids and colorectal cancer risk performed to date. ## The EPIC cohort The EPIC cohort includes over 520,000 individuals who were recruited between 1992 and 2000 from 23 study centers across 10 European countries (Denmark, France, Germany, Greece, Italy, Norway, Spain, Sweden, the Netherlands, and the UK). Participants were 35-70 years of age at recruitment, and approximately $70\%$ of the cohort are women. The study design has been previously described [22, 23]. In brief, extensive questionnaire data on dietary and lifestyle variables were collected at baseline, and approximately $75\%$ of individuals provided non-fasting blood samples. Incident cases of colorectal cancer were identified through record linkage with regional cancer registries or via a combination of methods, such as the use of health insurance records, contacts with cancer and pathology registries, and active follow-up through participants and their next of kin. Colorectal cancer was defined using the tenth edition of the International Classification of Disease (ICD-10) and the second edition of the International Classification of Disease for Oncology (ICD-O-2). Proximal colon cancers included those found within the cecum, ascending colon, hepatic flexure, transverse colon, and splenic flexure (C18.0 and C18.2–18.5). Distal colon cancers included those found within the descending (C18.6) and sigmoid (C18.7) colon. Overlapping (C18.8) and unspecified (C18.9) lesions of the colon were classed as colon cancers only. Cancer of the rectum included cancers occurring at the recto-sigmoid junction (C19) and rectum (C20). The current study employed a fasted subset of EPIC data, obtained from two separate metabolomics studies on colorectal cancer, as a discovery cohort. Samples were analyzed using the Biocrates AbsoluteIDQTM p180 kit (467 cases and 467 matched controls) and the p150 kit (1141 cases and 1141 controls). Combining these studies and then excluding non-fasting participants resulted in a final combined sample of 654 fasted cases and 654 controls, of which 354 case-control pairs were analyzed using the p180 kit. Controls were selected using incidence density sampling from all cohort members who were alive and free of cancer (except non-melanoma skin cancer) at the time of diagnosis of the colorectal cancer cases. Controls were matched to cases on age at recruitment (within 6 months), sex, study center, follow-up time since blood collection, time of day at blood collection (within 4 h), and fasting status. Women were further matched on menopausal status (pre-, peri-, and post-menopausal) and, in pre-menopausal women, phase of menstrual cycle at blood collection. Approval for the study was obtained from the International Agency for Research on Cancer (IARC) and local center review boards. All participants provided written informed consent. ## The UK Biobank cohort The UK Biobank aims to investigate the genetic, lifestyle, and environmental causes of a range of diseases [24]. Between 2006 and 2010, 502,656 adults aged between 40 and 69 years (229,182 men and 273,474 women) who were registered with the UK National Health Service were recruited at 22 study assessment centers. Ethical approval was obtained from the North West Multicentre Research Ethics Committee, the National Information Governance Board for Health and Social Care in England and Wales, and the Community Health Index Advisory Group in Scotland. All participants provided written informed consent. The present study was undertaken under application number 25897. During the baseline recruitment visit, participants completed a self-administered questionnaire on socio-demographics (including age, sex, education, and Townsend deprivation score), health and medical history, lifestyle exposures (including smoking habits, dietary intakes, and alcohol consumption), early life exposures, and medication use. Physical measurements were taken, including weight, height, and waist circumference. Colorectal cancer cases were defined using the 10th Revision of the International Classification of Diseases (ICD-10). Colorectal cancers comprised those of the proximal colon (C18.0 and C18.2–18.5), distal colon (C18.6–C18.7), overlapping and unspecified lesions of the colon (C18.8–C18.9), and rectal cancers (C19–C20), as described above. Blood samples, with data on time since last meal, were collected from all participants at recruitment and additionally from around 20,000 participants who attended a repeat assessment visit between 2012 and 2013. The current study included all participants for whom metabolite profiling had been performed at the time of the study, and thus had available amino acid measurements. From our supplied dataset that contained observations for 502,524 participants, exclusions were made for voluntary withdrawal from the study ($$n = 36$$) and prevalent cancer at recruitment ($$n = 27$$,240). Of the remainder, plasma amino acid measurements were available for 111,323 participants, and these were included as the replication cohort (Fig. 1).Fig. 1Flow chart showing discovery and replication study design. CRC, colorectal cancer; EPIC, European Prospective Investigation into Nutrition and Cancer ## Laboratory methods In EPIC, targeted metabolomics profiling was performed at the International Agency for Research on Cancer (Biocrates AbsoluteIDQTM p180 kit) and the Helmholtz Centre in Munich (Biocrates AbsoluteIDQTM p150 kit). The samples were prepared as per the Biocrates kit instructions [25, 26]. Assay preparation steps were carried out on 96 well plates and a volume of 10 μL plasma was prepared. The p150 kit allows the quantification of up to 13 amino acids and the p180 kit up to 21 amino acids (Additional file 1: Supplemental Methods) [25, 27]. Liquid chromatography–mass spectrometry (LC-MS) was used to quantify the levels of the amino acids in accordance with the kit manufacturer’s instructions. All 21 amino acids included were fully quantified in μmol/L. The amino acids quantified were arginine, glutamine, glycine, histidine, methionine, ornithine, phenylalanine, proline, serine, threonine, tryptophan, tyrosine, and valine (p150 and p180 kits); and alanine, asparagine, aspartate, citrulline, glutamate, isoleucine, leucine, and lysine (p180 kit only). See Additional file 1: Supplemental Methods for full details of sample preparation. Coefficients of variation for amino acids are given in Table S1. Analysis of plasma from around 118,000 participants of the UK Biobank was performed using nucleic magnetic resonance (NMR) spectroscopy on the Nightingale metabolic biomarker platform (Nightingale Health Ltd, Finland), which comprises 249 metabolic measures, among which are concentrations of 9 amino acids. In brief, stored plasma samples prepared in 96-well plates were thawed, mixed gently, and centrifuged for 3 min at 3400 g to remove the precipitate. Aliquots of each sample were mixed with phosphate buffer, loaded onto a cooled sample changer, and analyzed by NMR spectroscopy. Metabolic biomarkers were identified and quantified from two separate spectra, a pre-saturated proton NMR spectrum, and a T2-relaxation-filtered spectrum. Six identical Bruker AVANCE IIIHD instruments were employed in parallel. The amino acids quantified were alanine, glutamine, glycine, histidine, isoleucine, leucine, valine, phenylalanine, and tyrosine. See Additional file 1: Supplemental Methods for further details. ## Statistical analysis Pearson correlations between non-log transformed amino acid concentrations were first calculated in the 654 EPIC fasted controls only and in UK Biobank participants with a fasted time of >4 h ($$n = 56$$,688). Hierarchical clustering of concentration profiles using Ward’s method was used to visualize and identify notable clusters of correlated metabolites. ## Analysis of discovery cohort In EPIC, case-control status was modeled using conditional logistic regression and odds ratios (OR) and $95\%$ confidence intervals (CI) estimated for each amino acid. Models were adjusted for an a priori determined set of potential confounders comprising smoking status (never, former, current, unknown), alcohol drinking history (never, former, current, lifetime, unknown), Cambridge physical activity index (inactive, moderately inactive, moderately active, active, unknown) and body mass index (BMI; <25, 25–30, and >30kg/m2), all at baseline. The false discovery rate (FDR) procedure was used to adjust P-values and an FDR P-value threshold of 0.05 was used for statistical significance. Continuous models per SD concentration and categorical models by quartile were fit for each amino acid. For the categorical models, inner quartile cut points were determined by the metabolite concentrations among control participants. To test for trends across categories, quartile medians were additionally modeled as continuous variables. ## Analysis of replication cohort Amino acids that were significantly associated with colorectal cancer per SD concentration in EPIC were carried forward for testing in the UK Biobank cohort. Here, time to colorectal cancer diagnosis was modeled using Cox proportional hazards regression and hazard ratios (HR) and $95\%$ CI estimated for each amino acid. Time at study entry was age at recruitment, while exit time was age at incident cancer diagnosis, death, or the last date at which follow-up was considered complete. Multivariable models were kept as similar as possible to those fit in EPIC and were adjusted for BMI category (<25, 25–30, >30 kg/m2), total physical activity (<10, 10–20, 20–40, 40–60, >60 metabolic equivalent of task [MET] h/week), alcohol consumption frequency (never, special occasions only, 1–3 times/month, 1–2 times per week, 3–4 times/week, daily or almost daily, unknown/prefer not to answer), smoking status (smoker, former smoker, never smoker), time since last meal (hours), and family history of colorectal cancer (yes/no). Stratification variables were age at recruitment in 5-year intervals, Townsend deprivation index quintiles, and assessment center region. A raw P-value threshold of 0.05 was used for statistical significance. ## Stratified and sensitivity analyses The above analysis was repeated but excluding individuals diagnosed within the first 2, 5, and 10 years of the study in EPIC, and within the first 2 and 5 years in the UK Biobank. Sex-stratified models were also performed for all amino acids measured in both cohorts and, in UK Biobank, amino acid models were conducted for colon and rectal subsites separately. Heterogeneity by sex and by tumor subsite was tested for by fitting models with and without interaction terms and comparing these by likelihood ratio test. As sensitivity analyses, models for glutamine and histidine only were repeated additionally adjusting for major sources of animal proteins (red and processed meat, poultry, fish, and dairy product intake), and amino acid models were repeated in EPIC only using non-fasted participants as well as fasted participants. Analyses were conducted either in the R open-source statistical programming language (version 3.6.3 on the RStudio environment) or STATA version 16.1 (StataCorp Inc). ## Results A median follow-up of 14.4 years was observed for the 654 colorectal cancer cases and 654 controls in EPIC while, during a median follow-up of 10.7 years in the UK Biobank, 1221 incident cases of colorectal cancer occurred among the 111,323 participants with available amino acid measurements. The EPIC and UK Biobank populations were of similar ages at baseline and at colorectal cancer diagnosis although, in EPIC, most participants ($77.4\%$) were from Italian or Spanish centers. Full baseline characteristics are shown in Table 1. Glutamine, alanine, and glycine were at the highest circulating concentrations overall, as quantified in EPIC (Fig. 2). Fasting concentrations of amino acids in cancer-free participants were almost always positively correlated, with the following correlated clusters noted in EPIC: glycine and serine; arginine, methionine, and tryptophan; valine, isoleucine, and leucine; and histidine and phenylalanine (Fig. 3). In the UK Biobank, valine, isoleucine, and leucine concentrations (branched-chain amino acids) were strongly intercorrelated. Table 1Baseline characteristics of participants, by cohortEPIC nested case-control study ($$n = 654$$ cases and 654 controls)UK Biobank Prospective cohort ($$n = 111$$,323 of which 1221 incident colorectal cancers)ControlsCasesNon-casesCasesSex Male288 (44.0)288 (44.0)59,295 (53.9)513 (42.0) Female366 (56.0)366 (56.0)50,807 (46.1)708 (58.0)Country France29 (4.4)29 (4.4)- Italy336 (41.4)336 (41.4)- Spain235 (36.0)235 (36.0)- UK23 (3.5)23 (3.5)110,1021221 Netherlands2 (0.3)2 (0.3)- Germany27 (4.1)27 (4.1)- Denmark2 (0.3)2 (0.3)-Age at blood collection (years) Mean54.7 ± 7.354.8 ± 7.356.3 ± 8.161.0 ± 6.6Follow-up time to diagnosis (years) Mean-8.9 ± 4.5-9.7 ± 2.0BMI (kg/m2) Mean (SD)26.7 ± 3.827.5 ± 4.427.4 ± 4.828.0 ± 4.8Waist circumference (cm) Mean (SD)88.6 ± 12.090.7 ± 13.690.3 ± 13.494.0 ± 14.1Height (cm) Mean (SD)163.1 ± 9.0163.7 ± 9.0168.5 ± 9.3169. 9 ± 9.1Smoking status Non or former smoker334 (51.1)308 (47.1)60,105 (54.6)549 (45.0) Never smoker176 (26.9)171 (26.1)37,735 (34.3)547 (44.8) Smoker141 (21.6)170 (26.0)11,694 (10.6)123 (10.1) Unknown3 (0.5)5 (0.8)568 (0.5)2 (0.2)Alcohol intake status Never drinker62 (9.5)69 (10.6)4837 (4.3)4837 (4.3) Former drinker54 (8.3)47 (7.2)3965 (3.6)3965 (3.6) Drinker at recruitment (current)509 (77.8)504 (77.1)102,251 (91.8)102,251 (91.8) Unknown29 (4.4)34 (5.2)274 (0.2)274 (0.2)Physical activity status Inactive214 (32.7)234 (35.8)31,763 ($28.8\%$)350 (28.7) Moderately inactive255 (39.0)252 (38.5)21,532 ($19.6\%$)246 (20.1) Moderately active105 (16.1)105 (16.1)35,798 ($32.5\%$)375 (30.7) Active79 (12.1)62 (9.5)17,086 ($15.5\%$)214 (17.5) Missing1 (0.2)1 (0.2)3923 ($3.6\%$)36 (2.9)Highest educational level None95 (14.5)98 (15.2)00 Primary school completed274 (42.0)261 (40.6)00 Technical/professional school84 (12.9)73 (11.4)31,067 (28.2)332 (27.2) Secondary school104 (15.9)125 (19.4)46,786 (42.0)468 (38.3) Higher education/university90 (13.8)79 (12.3)12,509 (11.4)141 (11.5) Not specified6 (0.9)7 (1.1)19,740 (17.7)280 (22.9)Oral contraceptive use in women Ever5 (1.4)5 (1.4)11,033 (10.0)119 (9.7) Never360 (98.4)361 (98.6)47,994 (43.6)391 (32.0) Unknown1 (0.0)0 (0.0)51,075 (46.4)711 (58.2)Oral hormone therapy use in women Ever31 (8.5)37 (10.1)36,987 (61.8)36,716 (33.3) Never334 (91.3)326 (89.1)22,515 (37.6)22,276 (20.2) Unknown1 (0.0)3 (0.01)307 (0.0)303 (0.3)Means and SD or frequency and percentage are shown unless stated otherwiseSD, standard deviation; BMI, body mass indexFig. 2Blood concentrations of amino acids as determined in fasted EPIC participants on the p150 or p180 Biocrates platform. Based on 654 and 354 cancer-free controls for p150 and p180 platforms, respectivelyFig. 3Fasting amino acid concentrations and their intercorrelations in EPIC cancer-free controls. Compounds are ordered by the hierarchical cluster as determined by Ward’s method. Squares represent groups of highly correlated compounds ## Associations of pre-diagnostic amino acid concentrations with colorectal cancer risk In the EPIC discovery phase, histidine concentrations were inversely associated with colorectal cancer risk (OR 0.80 per SD concentration, $95\%$ CI 0.69–0.92, FDR P-value = 0.03) (Table 2). A statistically significant trend was also observed by quartile of histidine concentration (P-trend = 0.002). Lysine was also inversely associated with colorectal cancer risk (OR 0.78 per SD concentration, $95\%$ CI 0.66–0.93, FDR P-value = 0.05), and glutamine was borderline inversely associated with risk (OR 0.85 per SD concentration, $95\%$ CI 0.75–0.97, FDR P-value = 0.08). For both lysine and glutamine, individuals in Q4 of concentrations had a lower risk compared to those in Q1, with an apparent decreasing trend across quartiles (P-trend for both amino acids = 0.01).Table 2Associations between concentrations of 21 plasma or serum amino acids and colorectal cancer risk in the EPIC nested case-control discovery and UK Biobank replication cohortsAmino acid (by decreasing blood concentration)Colorectal cancer cases aOR/HR per SD concentration ($95\%$ CI) bFDR P-valueOR/HR ($95\%$ CI) for Q2 bcOR/HR ($95\%$ CI) for Q3 bcOR/HR ($95\%$ CI) for Q4 bcP-trend dDiscovery cohort (EPIC nested case-control study) Glutamine6540.85 (0.75–0.97)0.080.84 (0.61–1.16)0.72 (0.52–1.01)0.65 (0.46–0.92)0.01 Alanine3541.04 (0.89–1.22)0.751.04 (0.68–1.64)1.06 (0.68–1.64)1.24 (0.79–1.95)0.47 Glycine6540.89 (0.75–1.05)0.450.99 (0.70–1.40)0.82 (0.54–1.24)0.73 (0.46–1.16)0.08 Valine6540.92 (0.78–1.08)0.470.98 (0.69–1.38)0.76 (0.52–1.11)0.91 (0.58–1.41)0.45 Lysine3540.78 (0.66–0.93)0.050.82 (0.53–1.28)0.83 (0.53–1.31)0.51 (0.31–0.83)0.008 Proline6541.06 (0.93–1.20)0.481.02 (0.74–1.42)1.15 (0.82–1.62)1.05 (0.72–1.52)0.73 Serine6540.89 (0.73–1.08)0.450.83 (0.60–1.16)0.73 (0.49–1.09)0.61 (0.37–1.01)0.05 Leucine3540.79 (0.64–0.98)0.150.99 (0.64–1.54)0.72 (0.44–1.21)0.75 (0.44–1.29)0.20 Threonine6540.96 (0.85–1.09)0.601.14 (0.83–1.56)0.85 (0.61–1.19)0.97 (0.68–1.37)0.50 Ornithine6540.88 (0.69–1.13)0.470.90 (0.63–1.28)0.63 (0.35–1.15)0.65 (0.33–1.28)0.73 Arginine6541.02 (0.83–1.24)0.871.18 (0.85–1.65)0.93 (0.59–1.45)0.98 (0.57–1.68)0.91 Tyrosine6540.90 (0.76–1.06)0.450.92 (0.67–1.27)0.89 (0.61–1.28)0.84 (0.54–1.32)0.41 Histidine6540.80 (0.69–0.92)0.030.79 (0.57–1.09)0.59 (0.41–0.84)0.61 (0.42–0.89)0.005 Tryptophan6540.85 (0.64–1.12)0.450.62 (0.43–0.88)0.53 (0.29–0.96)0.55 (0.27–1.13)0.07 Isoleucine3540.92 (0.75–1.14)0.640.93 (0.59–1.46)0.75 (0.45–1.23)0.85 (0.49–1.47)0.39 Phenylalanine3540.89 (0.77–1.03)0.451.24 (0.89–1.74)0.97 (0.68–1.39)0.86 (0.58–1.28)0.13 Glutamate3541.10 (0.85–1.43)0.641.71 (1.04–2.81)2.02 (1.16–3.52)1.82 (0.96–3.45)0.19 Asparagine3541.08 (0.91–1.28)0.581.10 (0.70–1.72)1.40 (0.89–2.20)0.93 (0.57–1.53)0.91 Methionine6540.88 (0.69–1.13)0.470.90 (0.63–1.28)0.63 (0.35–1.15)0.65 (0.33–1.28)0.20 Citrulline3541.15 (0.97–1.37)0.400.60 (0.37–0.97)1.06 (0.69–1.64)1.11 (0.69–1.78)0.36 Aspartate e3540.99 (0.81–1.21)0.91----Replication cohort (UK Biobank cohort) Glutamine12210.95 (0.89–1.01)0.090.89 (0.76–1.05)0.99 (0.84–1.16)0.85 (0.72–1.00)0.12 Histidine12210.93 (0.87–0.99)0.031.01 (0.86–1.17)0.76 (0.65–0.90)0.86 (0.73–1.01)0.008OR, odds ratio; HR, hazard ratio; FDR, false discovery rate; Q, quartile. Estimates for which $95\%$ CI do not include 1 are given in bold texta In EPIC, amino acids measured using the Biocrates AbsoluteIDQTM p180 kit were measured in 354 cases and 354 controls onlyb Multivariable models were adjusted for smoking status (never, former, and current smoker), alcohol use (never, former, only at recruitment, and lifetime drinker), physical activity at recruitment (inactive, moderately inactive, moderately active, active) and body mass index (<25, 25–30, and >30 kg/m2). In UK Biobank, categories differed slightly for total physical activity (<10, 10–20, 20–40, 40–60, >60 metabolic equivalent of task [MET] h/week), alcohol consumption frequency (never, special occasions only, 1–3 times/month, 1–2 times per week, 3–4 times/week, daily or almost daily, unknown/prefer not to answer), and were also adjusted for family history of colorectal cancer (yes/no) and time since last meal (hours)c Q1 of amino acid concentrations was considered the referent group. Inner quartile cut points were determined by metabolite concentrations in controls onlyd Calculated by the replacement of continuous data by the median values of concentration quartiles in modee The categorical analysis was not performed due to a high proportion of missing values Histidine and glutamine, but not lysine, were among the nine amino acids measured in UK Biobank and were thus carried forward to the replication stage. Histidine was also significantly inversely associated with colorectal risk in UK Biobank (HR 0.93 per SD concentration, $95\%$ CI 0.87–0.99, P-value = 0.03), with a significantly decreasing trend across quartiles of concentration. Glutamine was again borderline inversely associated with risk on a continuous scale (HR 0.95 per SD, $95\%$ CI 0.89–1.01 respectively, P-value = 0.09), and individuals in Q4 of concentrations were at a lower risk of colorectal cancer than those in Q1 (HR 0.85, $95\%$ CI 0.72–1.00). ## Analysis by follow-up time In EPIC, ORs for histidine and glutamine did not appreciably change when cases diagnosed within 2, 5, or 10 years were excluded (OR 0.82 per SD concentration, $95\%$ CI 0.67–1.01 and OR 0.82, $95\%$ CI 0.68–0.99 respectively for the two amino acids, exclusion of the first 10 years of follow-up) (Table 3). Similarly, minor changes in HR were observed for the exclusion of 2 and 5-year periods of follow-up for these amino acids in UK Biobank. Table 3Associations between concentrations of serum and plasma amino acids and colorectal cancer risk in the EPIC nested case-control discovery and UK Biobank replication cohorts by follow-up time to diagnosis, where availableAmino acid aOR/HR ($95\%$ CI) per SD concentration by follow-up time bFull follow-up>2 years>5 years>10 yearsEPIC nested case-control studyn = 654 cases, 654 controlsn = 601 cases, 601 controlsn = 505 cases, 505 controlsn = 289 cases, 289 controls Glutamine0.85 (0.75–0.97)0.85 (0.75–0.97)0.83 (0.72–0.96)0.82 (0.68–0.99) Glycine0.89 (0.75–1.05)0.92 (0.77–1.11)0.90 (0.73–1.10)0.92 (0.70–1.21) Valine0.92 (0.78–1.08)0.91 (0.77–1.08)0.93 (0.77–1.12)1.04 (0.82–1.33) Proline1.06 (0.93–1.20)1.09 (0.95–1.24)1.08 (0.94–1.25)0.96 (0.80–1.15) Serine0.89 (0.73–1.08)0.91 (0.74–1.11)0.85 (0.68–1.05)0.81 (0.61–1.06) Threonine0.96 (0.85–1.09)0.97 (0.85–1.10)0.93 (0.81–1.07)0.87 (0.72–1.04) Ornithine0.88 (0.69–1.13)0.98 (0.82–1.17)0.98 (0.80–1.20)1.03 (0.78–1.36) Arginine1.02 (0.83–1.24)1.04 (0.84–1.28)1.07 (0.85–1.35)1.02 (0.77–1.36) Tyrosine0.90 (0.76–1.06)0.89 (0.75–1.06)0.87 (0.72–1.05)0.90 (0.71–1.12) Histidine0.80 (0.69–0.92)0.81 (0.69–0.94)0.81 (0.69–0.96)0.82 (0.67–1.01) Tryptophan0.85 (0.64–1.12)0.80 (0.59–1.08)0.77 (0.55–1.06)0.77 (0.52–1.12) Phenylalanine0.89 (0.77–1.03)0.87 (0.75–1.02)0.84 (0.71–1.00)0.83 (0.67–1.03) Methionine0.88 (0.69–1.13)0.86 (0.67–1.12)0.84 (0.64–1.11)0.89 (0.65–1.22)UK Biobank cohort ($$n = 111$$,323)$$n = 1221$$ casesn = 1042 casesn = 696 cases Glutamine0.95 (0.89–1.01)0.94 (0.88–1.01)0.98 (0.90–1.06) Histidine0.93 (0.87–0.99)0.95 (0.88–1.01)0.94 (0.87–1.03)OR, odds ratio; Q, quartile. Estimates for which $95\%$ CI do not include 1 are given in bold texta In decreasing order of measured blood concentration. Only amino acids measured in all participants in EPIC were includedb Multivariable models were adjusted for smoking status (never, former, and current smoker), alcohol use (never, former, only at recruitment, and lifetime drinker), physical activity at recruitment (inactive, moderately inactive, moderately active, active) and body mass index (<25, 25–30, and >30 kg/m2).). In UK Biobank, categories differed slightly for total physical activity (<10, 10–20, 20–40, 40–60, >60 metabolic equivalent of task [MET] h/week), alcohol consumption frequency (never, special occasions only, 1–3 times/month, 1–2 times per week, 3–4 times/week, daily or almost daily, unknown/prefer not to answer), and were also adjusted for family history of colorectal cancer (yes/no) and time since last meal (hours) ## Stratified and sensitivity analysis In EPIC, most available colorectal samples were for cancers of the colon ($\frac{625}{654}$) and estimates for colon cancer mirrored those of colorectal cancer (Additional file 1: Table S2). Nevertheless, in the UK Biobank where $31.7\%$ of colorectal cases ($\frac{388}{1221}$) were rectal cancers, HRs for colorectal and colon cancers were also similar. Here, glutamine concentrations were similarly associated with risk of colorectal cancer and colon cancer only (HR 0.92 per SD concentration, $95\%$ CI 0.85–0.99), while no association was observed for rectal cancer (HR 1.02 per SD, $95\%$ CI 0.91-1.13). Heterogeneity between colon and rectal tumor subsites approached but did not reach statistical significance ($$P \leq 0.13$$). As regards histidine, hazard ratios were similar for colon cancer (HR 0.95, $95\%$ CI 0.88-1.02), rectal cancer (HR 0.89, $95\%$ CI 0.80–0.99), and colorectal cancer overall (HR 0.93, $95\%$ CI 0.87-0.99). Inverse associations for amino acids were more pronounced in men than in women (Additional file 1: Table S3), and heterogeneity by sex was observed for histidine in UK Biobank (P-heterogeneity = 0.02). Heterogeneity by sex was not observed for any other amino acid measured in UK Biobank or for any amino acid measured in EPIC. For the EPIC and UK Biobank participants included in the main study, adjustment for major sources of amino acid intake (red and processed meat, poultry, fish, eggs, and dairy products) did not change associations between circulating amino acids and colorectal cancer risk (Additional file 1: Table S4). Likewise, in sensitivity analyses including non-fasting as well as fasting participants in EPIC, associations did not change appreciably (Additional file 1: Table S5). ## Discussion In this analysis of pre-diagnostic circulating amino acid levels and colorectal cancer risk, histidine was found to be robustly inversely associated and glutamine borderline inversely associated with colorectal cancer risk via a discovery-replication strategy in two large prospective cohorts. In addition, odds ratios and hazards ratios for these amino acids were attenuated minimally by the exclusion of cases diagnosed within 10 years of follow-up. This study provides strong evidence that lower levels of histidine, and possibly glutamine, are associated with subsequent risk of colorectal cancer, even up to 10 years before a colorectal cancer diagnosis. Circulating levels of several amino acids have previously been found to be inversely associated with colorectal neoplasia, but in studies of cross-sectional design only. Glutamine, for example, was one of several amino acids found to be lower in colorectal cancer patients compared to healthy controls [28], while histidine was lower among stage IV colorectal cancer cases than stage I cases [12] and even inversely correlated with tumor stage [29]. Untargeted metabolomics studies using discovery and validation cohorts demonstrated leucine and the dipeptide glutamine-leucine to be among those metabolites that distinguished cases from controls [14, 30]. Nevertheless, few studies have analyzed pre-diagnostic samples to investigate whether amino acid dysregulation precedes tumorigenesis. Two other prospective case-control studies on colorectal cancer with some amino acid measurements also found no significant associations [19, 31]. Our study is therefore the first to observe inverse associations of amino acid levels with colorectal cancer risk in a prospective setting and in independent studies. Although the above evidence suggests that tumor energy requirements give rise to the depletion of circulating glutamine and histidine, the levels of these amino acids among colorectal cancer cases in our prospective cohorts were associated with colorectal cancer risk at least 10 years prior to a colorectal cancer diagnosis, suggesting that alterations in the metabolism of these compounds may either reflect etiological pathways associated with the development of disease or metabolic changes linked to early events in colorectal tumorigenesis. The most pronounced finding of the current study was a robust inverse association between circulating histidine and colorectal cancer risk in both EPIC and UK Biobank. Histidine, an essential amino acid derived from the diet, is converted by histidine decarboxylase to the biogenic amine histamine [32], a signaling molecule that mediates an acute inflammatory response by binding to specific receptors [33]. Histidine decarboxylase activity may be upregulated in tumor cells and is thought to accelerate cell proliferation and angiogenesis [34]. For instance, the enzyme was found to be more active in colon cancer cells, particularly metastatic tumor cells, than normal colonic cells [34]. Given that an inverse association between histidine and colorectal cancer risk was apparent as long as 10 years before diagnosis, perturbations to specific etiologic pathways may be hypothesized. Prior evidence suggests that higher histidine concentrations mitigate metabolic dysregulation; for example, dietary supplementation with histidine was found to improve insulin sensitivity, possibly via the suppression of pro-inflammatory cytokine expression, in women with metabolic syndrome [35]. These findings suggested that histidine may even hold potential as a therapeutic agent against metabolic disease. However, levels of histidine have also been found to be positively associated with breast cancer [36]. Therefore, additional laboratory studies are needed to elucidate the potential role of this amino acid in carcinogenesis. Glutamine was borderline inversely associated with colorectal cancer risk in both cohorts. Glutamine is among the most abundant small-molecule metabolites in circulation and plays a central role in amino acid metabolism. It is used by proliferating cancer cells as an energy source [37] and is likely an important substrate throughout colorectal tumorigenesis [38]. The reasons for lower circulating glutamine in individuals who went on to develop colorectal cancer compared to those who remained cancer-free are uncertain. Firstly, given the slow development of colorectal cancer, lowered glutamine may reflect the undetected presence of polyps or early cancerous lesions in cases at baseline [28]. Secondly, regardless of the presence of such lesions and even controlling for major risk factors, abnormally low glutamine levels may reflect cancer-promoting metabolism. For instance, as well as being directly related to the tumor stage, glutamine levels have been inversely associated with serum C-reactive protein and inflammatory cytokines [13, 39]. Also, lowered glutamine and the glutamine-glutamate ratio was reported to be associated with incident type II diabetes [40], an established risk factor for colorectal cancer [41]. Lowered glutamine may represent dysregulation of the glutamine-glutamate axis. Although our study measured glutamate in a subset of EPIC cases and controls only, some evidence for a positive association of glutamate with colorectal cancer risk was observed in the categorical analysis. Glutamine concentrations may influence multiple mechanisms related to cancer development which deserve further investigation in experimental models. The inverse associations observed between amino acid concentrations and colorectal cancer risk may also reflect cancer-promoting dysbiosis of the gut microbiota. Similar to the protection against colorectal cancer afforded by short-chain fatty acids produced from dietary fiber by microbiota via the mitigation of an inflammatory microenvironment [42], certain components of the gut microbiota may act upon amino acids in the lumen to influence inflammation and tumorigenesis [43],[44]. For example, the production of histidine decarboxylase by gut microbes has been suggested to decrease intestinal inflammation via the binding of histamine to the receptor HR2 in the gut lumen [45]. Also, specific components of the microbiota are also believed to mediate the relationship between branched-chain amino acids and insulin resistance [46]. It is therefore plausible that the associations of histidine and glutamine with colorectal cancer risk in the current study may reflect variability in gut microbial activity and its interaction with host metabolism. Further mechanistic research is needed to investigate links between histidine and glutamine metabolism, the gut microbiota and colorectal tumorigenesis. The main strengths of this study include the prospective design, the use of large-scale cohorts with extensive participant data, and robust amino acid measurements in participants of well-characterized fasting status. We excluded non-fasted participants from the outset in the EPIC discovery cohort to minimize the effects of recent dietary intake upon amino acid levels which could have complicated interpretation of the results. As an additional safeguard against this bias, we performed sensitivity analyses for fasting status in EPIC and for major dietary sources of amino acids in both cohorts, which did not appreciably attenuate risk estimates. This is consistent with a recent study in EPIC that found weak or no correlations between amino acid intake and their blood concentrations [47]. In terms of limitations, only 9 of the 21 amino acids were measured in all EPIC and UK Biobank samples, while only 13 were measured all EPIC samples, with limited statistical power for the remainder. It is plausible that levels of amino acids other than glutamine and histidine are associated with colorectal cancer and we note that HR or OR point estimates were lower than 1 for most compounds in both cohorts. With greater statistical power, particularly in UK Biobank, other amino acids would likely have been found inversely associated with colorectal cancer risk via the discovery-replication strategy. Also, measurements of amino acids were taken at the study baseline only, and the technical and biological reproducibility of measurement was therefore not accounted for. However, studies calculating intra-class correlation in blood samples suggest that polar metabolites such as amino acids are measured reproducibly, particularly in fasted participants [48]. Statistical power was limited for individual colorectal cancer subsites, particularly rectal cancer. Also, we were not able to consider amino acids in tissue samples, which may better represent the tumor microenvironment and provider deeper insight into the biological implications of our findings. ## Conclusions Circulating histidine levels were robustly inversely associated with colorectal cancer risk in two independent prospective cohorts with similar, albeit slightly weaker, evidence for glutamine. This knowledge should contribute to a better understanding of the underpinnings of colorectal cancer and metabolism and could potentially support new prevention or early detection strategies. Further research using experimental models to assess potential causality of the identified associations is now needed. ## Supplementary Information Additional file 1. Supplemental methods for laboratory analyses; Table S1, Reported coefficients of variation for 21 amino acids measured in the EPIC and UK Biobank cohorts; Table S2, Associations between 21 plasma or serum amino acids and colorectal, colon and rectal cancer risk in the EPIC and UK Biobank cohorts; Table S3, Associations between 21 plasma or serum amino acids and colorectal cancer risk in the EPIC and UK Biobank cohorts, by sex; Table S4, Associations between amino acids associated with colorectal cancer in either cohort in the main study, additionally adjusted for intakes of major sources of animal protein (red and processed meat, poultry, fish and dairy products) in the EPIC and UK Biobank cohorts; Table S5, Associations between concentrations of 21 plasma or serum amino acids and colorectal cancer risk in fasted participants only and all available participants in the EPIC nested case-control study. ## References 1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: Cancer J Clin (2021) 71 209-249. PMID: 33538338 2. Weitz J, Koch M, Debus J, Höhler T, Galle PR, Büchler MW. Colorectal cancer. Lancet (2005) 365 153-165. DOI: 10.1016/S0140-6736(05)17706-X 3. Abar L, Vieira AR, Aune D, Sobiecki JG, Vingeliene S, Polemiti E, Stevens C, Greenwood DC, Chan DSM, Schlesinger S. Height and body fatness and colorectal cancer risk: an update of the WCRF-AICR systematic review of published prospective studies. Eur J Nutr (2018) 57 1701-1720. DOI: 10.1007/s00394-017-1557-1 4. Park JY, Dahm CC, Keogh RH, Mitrou PN, Cairns BJ, Greenwood DC, Spencer EA, Fentiman IS, Shipley MJ, Brunner EJ. Alcohol intake and risk of colorectal cancer: results from the UK Dietary Cohort Consortium. Br J Cancer (2010) 103 747-756. DOI: 10.1038/sj.bjc.6605802 5. Tabung FK, Brown LS, Fung TT. Dietary Patterns and Colorectal Cancer Risk: a Review of 17 Years of Evidence (2000-2016). Curr Colorectal Cancer Rep (2017) 13 440-454. DOI: 10.1007/s11888-017-0390-5 6. Giovannucci E, Michaud D. The role of obesity and related metabolic disturbances in cancers of the colon, prostate, and pancreas. Gastroenterology (2007) 132 2208-2225. DOI: 10.1053/j.gastro.2007.03.050 7. Gunter MJ, Alhomoud S, Arnold M, Brenner H, Burn J, Casey G, Chan AT, Cross AJ, Giovannucci E, Hoover R. Meeting report from the joint IARC–NCI international cancer seminar series: a focus on colorectal cancer. Ann Oncol (2019) 30 510-519. DOI: 10.1093/annonc/mdz044 8. Murphy N, Song M, Papadimitriou N, Carreras-Torres R, Langenberg C, Martin RM. Associations Between Glycemic Traits and Colorectal Cancer: A Mendelian Randomization Analysis. J Natl Cancer Inst. (2022) 114 740-752. DOI: 10.1093/jnci/djac011 9. Giovannucci E, Harlan DM, Archer MC, Bergenstal RM, Gapstur SM, Habel LA, Pollak M, Regensteiner JG, Yee D. Diabetes and Cancer A consensus report. Diabetes Care (2010) 33 1674-1685. DOI: 10.2337/dc10-0666 10. Lieu EL, Nguyen T, Rhyne S, Kim J. Amino acids in cancer. Exp Mol Med (2020) 52 15-30. DOI: 10.1038/s12276-020-0375-3 11. Goveia J, Pircher A, Conradi L-C, Kalucka J, Lagani V, Dewerchin M, Eelen G, DeBerardinis RJ, Wilson ID, Carmeliet P. Meta-analysis of clinical metabolic profiling studies in cancer: challenges and opportunities. EMBO Mol Med (2016) 8 1134-1142. DOI: 10.15252/emmm.201606798 12. Geijsen A, van Roekel EH, van Duijnhoven FJB, Achaintre D, Bachleitner-Hofmann T, Baierl A, Bergmann MM, Boehm J, Bours MJL, Brenner H. Plasma metabolites associated with colorectal cancer stage: Findings from an international consortium. Int J Cancer (2019) 146 3256. DOI: 10.1002/ijc.32666 13. Sirniö P, Väyrynen JP, Klintrup K, Mäkelä J, Karhu T, Herzig K-H, Minkkinen I, Mäkinen MJ, Karttunen TJ, Tuomisto A. Alterations in serum amino-acid profile in the progression of colorectal cancer: associations with systemic inflammation, tumour stage and patient survival. Br J Cancer (2019) 120 238-246. DOI: 10.1038/s41416-018-0357-6 14. Geijsen A, Brezina S, Keski-Rahkonen P, Baierl A, Bachleitner-Hofmann T, Bergmann MM, Boehm J, Brenner H, Chang-Claude J, van Duijnhoven FJB. Plasma metabolites associated with colorectal cancer: A discovery-replication strategy. Int J Cancer (2019) 145 1221-1231. DOI: 10.1002/ijc.32146 15. Nishiumi S, Kobayashi T, Ikeda A, Yoshie T, Kibi M, Izumi Y. A Novel Serum Metabolomics-Based Diagnostic Approach for Colorectal Cancer. PLoS One. (2012) 7 110. DOI: 10.1371/journal.pone.0040459 16. Nishiumi S, Kobayashi T, Kawana S, Unno Y, Sakai T, Okamoto K, Yamada Y, Sudo K, Yamaji T, Saito Y. Investigations in the possibility of early detection of colorectal cancer by gas chromatography/triple-quadrupole mass spectrometry. Oncotarget (2017) 8 17115-17126. DOI: 10.18632/oncotarget.15081 17. Tan BB, Qiu YP, Zou X, Chen TL, Xie GX, Cheng Y, Dong TT, Zhao LJ, Feng B, Hu XF. Metabonomics Identifies Serum Metabolite Markers of Colorectal Cancer. J Proteome Res (2013) 12 3000-3009. DOI: 10.1021/pr400337b 18. Erben V, Bhardwaj M, Schrotz-King P, Brenner H. Metabolomics Biomarkers for Detection of Colorectal Neoplasms: A Systematic Review. Cancers (2018) 10 246. DOI: 10.3390/cancers10080246 19. McCullough ML, Hodge RA, Campbell PT, Stevens VL, Wang Y. Pre-Diagnostic Circulating Metabolites and Colorectal Cancer Risk in the Cancer Prevention Study-II Nutrition Cohort. Metabolites. (2021) 11 156. DOI: 10.3390/metabo11030156 20. Shu X, Xiang YB, Rothman N, Yu DX, Li HL, Yang G, Cai H, Ma X, Lan Q, Gao YT. Prospective study of blood metabolites associated with colorectal cancer risk. Int J Cancer (2018) 143 527-534. DOI: 10.1002/ijc.31341 21. Papadimitriou N, Gunter MJ, Murphy N, Gicquiau A, Achaintre D, Brezina S, Gumpenberger T, Baierl A, Ose J, Geijsen A. Circulating tryptophan metabolites and risk of colon cancer: Results from case-control and prospective cohort studies. Int J Cancer (2021) 149 1659-1669. DOI: 10.1002/ijc.33725 22. Riboli E, Hunt KJ, Slimani N, Ferrari P, Norat T, Fahey M, Charrondiere UR, Hemon B, Casagrande C, Vignat J. European Prospective Investigation into Cancer and Nutrition (EPIC): study populations and data collection. Public Health Nutr. (2002) 5 1113-1124. DOI: 10.1079/PHN2002394 23. Riboli E, Kaaks R. The EPIC Project: rationale and study design. European Prospective Investigation into Cancer and Nutrition. Int J Epidemiol (1997) 26 S6-14. DOI: 10.1093/ije/26.suppl_1.S6 24. Allen N, Sudlow C, Downey P, Peakman T, Danesh J, Elliott P, Gallacher J, Green J, Matthews P, Pell J. UK Biobank: Current status and what it means for epidemiology. Health Policy Technol (2012) 1 123-126. DOI: 10.1016/j.hlpt.2012.07.003 25. Romisch-Margl W, Prehn C, Bogumil R, Rohring C, Suhre K, Adamski J. Procedure for tissue sample preparation and metabolite extraction for high-throughput targeted metabolomics. Metabolomics (2012) 8 133-142. DOI: 10.1007/s11306-011-0293-4 26. Rothwell JA, Murphy N, Besevic J, Kliemann N, Jenab M, Ferrari P, Achaintre D, Gicquiau A, Vozar B, Scalbert A. Metabolic Signatures of Healthy Lifestyle Patterns and Colorectal Cancer Risk in a European Cohort. Clin Gastroenterol Hepatol (2022) 20 e1061-e1082. DOI: 10.1016/j.cgh.2020.11.045 27. Stepien M, Duarte-Salles T, Fedirko V, Floegel A, Barupal DK, Rinaldi S, Achaintre D, Assi N, Tjonneland A, Overvad K. Alteration of amino acid and biogenic amine metabolism in hepatobiliary cancers: Findings from a prospective cohort study. Int J Cancer (2016) 138 348-360. DOI: 10.1002/ijc.29718 28. Gu JP, Xiao YQ, Shu D, Liang XR, Hu XM, Xie YY. Metabolomics Analysis in Serum from Patients with Colorectal Polyp and Colorectal Cancer by H-1-NMR Spectrometry. Dis Markers. (2019) 2019 3491852. DOI: 10.1155/2019/3491852 29. Uchiyama K, Yagi N, Mizushima K, Higashimura Y, Hirai Y, Okayama T, Yoshida N, Katada K, Kamada K, Handa O. Serum metabolomics analysis for early detection of colorectal cancer. J Gastroenterol (2017) 52 677-694. DOI: 10.1007/s00535-016-1261-6 30. Li JK, Li J, Wang H, Qi LW, Zhu YM, Lai MD. Tyrosine and Glutamine-Leucine Are Metabolic Markers of Early-Stage Colorectal Cancers. Gastroenterology (2019) 157 257. DOI: 10.1053/j.gastro.2019.03.020 31. Kuhn T, Floegel A, Sookthai D, Johnson T, Rolle-Kampczyk U, Otto W, von Bergen M, Boeing H, Kaaks R. Higher plasma levels of lysophosphatidylcholine 18:0 are related to a lower risk of common cancers in a prospective metabolomics study. BMC Med (2016) 14 13. DOI: 10.1186/s12916-016-0552-3 32. Jutel M, Akdis M, Akdis CA. Histamine, histamine receptors and their role in immune pathology. Clin Exp Allergy (2009) 39 1786-1800. DOI: 10.1111/j.1365-2222.2009.03374.x 33. Smolinska S, Jutel M, Crameri R, O'Mahony L. Histamine and gut mucosal immune regulation. Allergy (2014) 69 273-281. DOI: 10.1111/all.12330 34. Cianchi F, Cortesini C, Schiavone N, Perna F, Magnelli L, Fanti E, Bani D, Messerini L, Fabbroni V, Perigli G. The role of cyclooxygenase-2 in mediating the effects of histamine on cell proliferation and vascular endothelial growth factor production in colorectal cancer. Clin Cancer Res (2005) 11 6807-6815. DOI: 10.1158/1078-0432.CCR-05-0675 35. Feng RN, Niu YC, Sun XW, Li Q, Zhao C, Wang C, Guo FC, Sun CH, Li Y. Histidine supplementation improves insulin resistance through suppressed inflammation in obese women with the metabolic syndrome: a randomised controlled trial. Diabetologia (2013) 56 985-994. DOI: 10.1007/s00125-013-2839-7 36. Jobard E, Dossus L, Baglietto L, Fornili M, Lecuyer L, Mancini FR, Gunter MJ, Tredan O, Boutron-Ruault MC, Elena-Herrmann B. Investigation of circulating metabolites associated with breast cancer risk by untargeted metabolomics: a case-control study nested within the French E3N cohort. Br J Cancer (2021) 124 1734-1743. DOI: 10.1038/s41416-021-01304-1 37. Cluntun AA, Lukey MJ, Cerione RA, Locasale JW. Glutamine Metabolism in Cancer: Understanding the Heterogeneity. Trends Cancer (2017) 3 169-180. DOI: 10.1016/j.trecan.2017.01.005 38. Zhao YQ, Zhao X, Chen V, Feng Y, Wang L, Croniger C, Conlon RA, Markowitz S, Fearon E, Puchowicz M. Colorectal cancers utilize glutamine as an anaplerotic substrate of the TCA cycle in vivo. Sci Rep (2019) 9 19180. DOI: 10.1038/s41598-019-55718-2 39. Ling HH, Pan Y-P, Fan C-W, Tseng W-K, Huang J-S, Wu T-H, Chou W-C, Wang C-H, Yeh K-Y, Chang P-H. Clinical Significance of Serum Glutamine Level in Patients with Colorectal Cancer. Nutrients (2019) 11 898. DOI: 10.3390/nu11040898 40. Liu XR, Zheng Y, Guasch-Ferre M, Ruiz-Canela M, Toledo E, Clish C, Liang LM, Razquin C, Corella D, Estruch R. High plasma glutamate and low glutamine-to-glutamate ratio are associated with type 2 diabetes: Case-cohort study within the PREDIMED trial. Nutr Metab Cardiovasc Dis (2019) 29 1040-1049. DOI: 10.1016/j.numecd.2019.06.005 41. Morze J, Wittenbecher C, Schwingshackl L, Danielewicz A, Rynkiewicz A, Hu FB, Guasch-Ferre M. Metabolomics and Type 2 Diabetes Risk: An Updated Systematic Review and Meta-analysis of Prospective Cohort Studies. Diabetes Care (2022) 45 1013-1024. DOI: 10.2337/dc21-1705 42. Watson KM, Gaulke CA, Tsikitis VL. Understanding the microbiome: a primer on the role of the microbiome in colorectal neoplasia. Ann Gastroenterol (2020) 33 223-236. PMID: 32382225 43. Mardinoglu A, Shoaie S, Bergentall M, Ghaffari P, Zhang C, Larsson E, Bäckhed F, Nielsen J. The gut microbiota modulates host amino acid and glutathione metabolism in mice. Mol Syst Biol (2015) 11 834. DOI: 10.15252/msb.20156487 44. Salahshouri P, Emadi-Baygi M, Jalili M, Khan FM, Wolkenhauer O, Salehzadeh-Yazdi A. A Metabolic Model of Intestinal Secretions: The Link between Human Microbiota and Colorectal Cancer Progression. Metabolites (2021) 11 456. DOI: 10.3390/metabo11070456 45. Gao C, Major A, Rendon D, Lugo M, Jackson V, Shi Z, Mori-Akiyama Y, Versalovic J. Histamine H2 Receptor-Mediated Suppression of Intestinal Inflammation by Probiotic Lactobacillus reuteri. MBIO (2015) 6 e01358. DOI: 10.1128/mBio.01358-15 46. Pedersen HK, Gudmundsdottir V, Nielsen HB, Hyotylainen T, Nielsen T, Jensen BAH, Forslund K, Hildebrand F, Prifti E, Falony G. Human gut microbes impact host serum metabolome and insulin sensitivity. Nature (2016) 535 376. DOI: 10.1038/nature18646 47. Iguacel I, Schmidt JA, Perez-Cornago A, Van Puyvelde H, Travis R, Stepien M, Scalbert A, Casagrande C, Weiderpass E, Riboli E. Associations between dietary amino acid intakes and blood concentration levels. Clin Nutr (2021) 40 3772-3779. DOI: 10.1016/j.clnu.2021.04.036 48. Breier M, Wahl S, Prehn C, Fugmann M, Ferrari U, Weise M, Banning F, Seissler J, Grallert H, Adamski J. Targeted Metabolomics Identifies Reliable and Stable Metabolites in Human Serum and Plasma Samples. PLoS One (2014) 9 e89728. DOI: 10.1371/journal.pone.0089728
--- title: Early exposure to infections increases the risk of allergic rhinitis—a systematic review and meta-analysis authors: - JunRong Chen - Xiaohua Liu - Zixin Liu - Yaqian Zhou - Li Xie - Jialin Zhang - Jin Tan - Yide Yang - Mei Tian - Yunpeng Dong - Jian Li journal: BMC Pediatrics year: 2023 pmcid: PMC9976500 doi: 10.1186/s12887-023-03870-0 license: CC BY 4.0 --- # Early exposure to infections increases the risk of allergic rhinitis—a systematic review and meta-analysis ## Abstract ### Objective The purpose of this study was to provide evidence for early life care by meta-analyzing the relationship between infection during pregnancy and up to 2 years of age and the risk of subsequent allergic rhinitis (AR). ### Methods Published studies up to April 2022 were systematically searched in PubMed, Embase, Web of Science, Cochrane Library, SinoMed, CNKI, Wanfang Database, and VIP. Literature screening, including quality assessment, was performed, and the effect values (OR, HR, RR) and $95\%$ confidence intervals ($95\%$ CI) of infection during pregnancy and up to 2 years of age and allergic rhinitis were extracted from each qualified study. ### Results In total, 5 studies with a sample size of 82,256 reported the relationship between infection during pregnancy and offspring AR. Meta-analysis showed that maternal infection during pregnancy was associated with an increased risk of childhood AR in offspring (OR = 1.34, $95\%$ CI: 1.08–1.67). Altogether, 13 studies with a sample size of 78,426 reported evidence of an association between infection within 2 years of age and subsequent AR in children. A pooled meta-analysis of all studies showed that early infection within 2 years of age was closely associated with childhood AR (OR = 1.25, $95\%$ CI: 1.12–1.40), especially upper respiratory tract infection (OR = 1.32, $95\%$ CI: 1.06–1.65) and gastrointestinal infections (OR = 1.37, $95\%$ CI: 1.01–1.86), but ear infection showed similar results in the cohort study (OR = 1.13, $95\%$ CI: 1.04–1.22). ### Conclusion Current evidence suggests that infection during pregnancy, early upper respiratory infection, gastrointestinal infections and ear infection within 2 years of age would increase the risk of AR in children. Therefore, the prevention of infection during pregnancy and in infancy and young children needs to be emphasized. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12887-023-03870-0. ## Introduction One of the most common chronic diseases, especially in children, is allergic rhinitis (AR). There is a chronic immune-mediated disease mediated by immunoglobulin E that occurs on the nasal mucosa after a specific individual is exposed to allergens, and symptoms include sneezing, watery mucus, nasal itching and congestion [1], which seriously affect children's quality of life, daily activities, sleep and learning [2]. According to a meta-analysis covering 102 countries, the worldwide prevalence of childhood AR is $12.66\%$ [3]. However, a comprehensive analysis of the relationship between infection and AR is lacking despite the fact that its etiology is unclear, it is currently believed to be closely related to the combination of genetic and environmental factors. In addition to genetic and epigenetic mechanisms, the living environment and gut microbiota also affect AR. In accordance with the Developmental Origins of Health and Diseases (DOHaD) theory, adverse exposures during early life may have an adverse impact on the development of programming and the occurrence of chronic diseases in late life. However, taking into account the hygiene hypothesis, early exposure to microorganisms may prevent the development of allergic diseases. The better your lifestyle, the less likely you are to encounter microorganisms, leading to a greater incidence of allergic diseases. The period of pregnancy is the most critical period for the development of the fetus, and mounting evidence has shown that maternal infection during pregnancy increases the risk of adverse perinatal outcomes and long-term health outcomes for offspring, such as low birth weight and mental illness [4–6]. The perturbed gut microbiota in the first 1000 days of life, from pregnancy to 2 years after birth, increases the risk for allergic disease and obesity in later life, highlighting the importance of understanding the relationships of perinatal factors with the establishment of diverse gut microbiota [7–10]. Although previous studies have shown that maternal infection during pregnancy can increase the risk of asthma and eczema in offspring [11], a meta-analysis by Van Meel et al. [ 12] also showed that early respiratory tract infection was associated with the development of asthma in school-age children, but the relationship between infection and AR has received less attention. Recently, an increasing number of studies have surveyed the relationship between early exposure to infections during pregnancy and within 2 years old and the risk of AR in late life. Recent studies have shown that the occurrence of AR is closely related to exposure to antibiotics and air pollution in early life [13, 14]. Of interest, there was a lack of uniformity in the research conclusions on the relationship between infections and later AR. McKeever [15] showed that early personal infections do not provide significant protection against allergic diseases, whereas Bremner found that early respiratory infections may increase the risk of later allergic rhinitis [16]. As early exposure to infection might be an understanding of the pathogenesis of AR, it is necessary to synthesize all available published literature on the relationship between infection during the early 1000 days of life and the risk of AR in late life in the form of a comprehensive meta-analysis. ## Data sources and search strategy The search strategies used the PICO principle to ensure that the retrieved journal-published literature was as comprehensive as possible. Search terms were including medical subject heading terms and text words related to subjects were developed in Pubmed and then adapted for Pubmed, Embase, Web of science, Cochrane, Sinomed, CNKI, Wanfang Database, and VIP from inception through April 30, 2022,the search terms were as follows:antenatal, prenatal, pregnancy, pregnant, perinatal, gestational, maternal, mother, newborn, infant, early life, toddler, febrile, infection, infestation, rhinitis, allergic, rhinitis, allergic, seasonal, allergic rhinitides, hayfever, hay fever. No publication, population, or language restrictions were applied, and attention was paid to checking the list of references on relevant topics. In addition, it was impossible to contact authors to have access to the full text or original data. Search terms and strategies are described in Additional file 1. ## Study selection Titles and abstracts of all potentially eligible articles retrieved from each database and managed in Endnote X9 software. The literature criteria employed for the meta-analysis included the following: [1]. Study population: children aged 0–18 years old. [ 2]. Study types: cohort studies, cross-sectional studies, and case–control studies. Exposure factors: infection within 2 years after birth or maternal infection during pregnancy. [ 4]. Outcome: clearly the offspring have AR, the relative risk (RR), hazard ratio (HR), or odds ratio (OR) and their confidence intervals can be obtained, or enough data to calculate them. The exclusion criteria were as follows: [1]. The full text or original data are not available. [ 2]. Publication languages other than Chinese or English. [ 3]. Duplicate publications. [ 4]. Reviews, systematic reviews or meta-analyses, conference abstracts, research protocols. [ 5]. Viral skin infections (only one study). [ 6]. Low-quality research. ## Quality assessment and data extraction The quality of the cohort studies and case–control studies was measured with the Newcastle–Ottawa Scale (NOS) [17], and cross-sectional studies were assessed using the Agency for Healthcare Research and Quality [18] (AHRQ). The NOS scale has a total score of 9 points, including three aspects: study population selection, comparability, and outcome, of which comparability can be scored up to 2 points, and studies with scores < 5 points are considered high risk of bias studies. The AHRQ scale has a total score of 11 points with 11 items, answering “yes” can be scored as 1 point, and answering “unclear or no” is scored as 0 points. The quality scale is divided into 0–3 points for low quality, 4–7 points for medium quality, and ≥ 8 points for high quality. In this study, the quality of literature assessed as moderate to high quality was included in the analysis (Score ≥ 6 points). The parameters and data were extracted using a standardized spreadsheet from each study, including author, year, study country, study type, study object, sample size, exposure factors, diagnostic method, and effect values. Two researchers independently screened the literature and extracted data according to the selection of the study. After each phase of the screening and data extraction process, any disagreements regarding records between two researchers were resolved by discussion or by consulting a third investigator if consensus could not be reached. ## Data synthesis and analysis RevMan 5.4 and Stata 16.0 software were used for statistical analysis, and the OR and $95\%$ confidence interval ($95\%$ CI) used in the meta-analysis P ≤ 0.05 was considered to be statistically significant. Statistical heterogeneity across studies was tested by the Q statistic and I2 value. If no significant heterogeneity was examined (I2 < $50\%$ and $P \leq 0.1$), pooled estimates were calculated using a fixed‐effects model; otherwise, a random‐effects model was adopted [19, 20]. Analysis of the risk relationship between different infections and AR in children. First, if a study reported different exposures;Second, if the study was stratified by exposure factors and study participants (number of infections, different periods of infection, and different ages of children) without providing overall estimates of infection and AR,the effect estimate and $95\%$ CI from the literature were combined at first and as the final extracted effect into meta-analysis. Moreover, a subgroup analysis was performed,and sensitivity analysis was estimated by omitting every study individually. If the heterogeneity among the studies decreases after excluding one study, it shows that this study is the cause of the heterogeneity. Publication bias was assessed using funnel plots and Begg’s test of bias, with $P \leq 0.05$ indicating significant publication bias. ## Study selection and characteristics In total, 3866 studies were identified according to our search strategies. A total of 249 duplicate studies were excluded, and 3547 studies were inconsistent with the research purpose after reviews of the titles and abstracts. Then, 70 studies were screened and required full-text reading for detailed evaluation. Finally, 19 studies were included in this meta-analysis after full text review. More specific screening details are shown in Fig. 1 and totally 51 references were excluded and excluded reasons are shown in the Additional file 2. Among these 19 studies, 14 studies with a total sample size of 78,426 reported evidence of infection within 2 years of age and subsequent AR in children, including 7 cohort studies [15, 21–26], 5 cross-sectional studies [27–31] and 2 case–control studies [16, 32]. Five studies with a total sample size of 82,256 reported evidence of infection during pregnancy and AR in offspring, including 3 cohort studies [33–35], 1 cross-sectional study [36] and 1 case–control study [37]. Except for the study of Thomson 2010 [21], the quality evaluation of all the included studies was of medium to high quality. This study used follow-up parental reports for infection exposure and AR outcome determinations. The following analysis did not include this evidence because the assessments were unreliable and the sample size was not large enough. Three of the remaining 18 studies extracted crude effect sizes [22, 27, 33], one was obtained by calculation [27], and all others extracted adjusted effect sizes. Tables 1, 2 and 3 summarize the specific information of the included studies and quality evaluation, respectively. Fig. 1Screening flow chartTable 1Characteristics of included studiesAuthor YearCountryExposure periodNAge (y)Exposure factorsOutcome evaluationMckeever 2002 [34]United KingdomDuring pregnancy292380–11①aHsieh 2016 [35]Taiwan, ChinaDuring Pregnancy422170–9②aIlli 2014 [33]GermanyDuring Pregnancy5260–5③cXu 1999 [36]FinlandDuring Pregnancy80887④cLiao 2015 [37]ChinaDuring Pregnancy21874–12⑤aBremner 2008 [16]United KingdomFirst year of life35492–5⑥⑦⑩Not describedDe 2005 [28]NetherlandsFrom birth to 2 years of age15558–13⑥bMckeever 2002 [15]United KingdomFirst year of life292380–8⑥⑦⑧bKang 2019 [32]ChinaFrom birth to 6 months old50012.8 ± 5.2, 13.7 ± 5.2⑥⑦bLin 2015 [23]Taiwan, ChinaFirst month of life164130–8⑩aRamsey 2007 [22]United StatesFirst year of life4400–7⑥⑨aKemp 2009 [29]AustraliaFirst month of life9138,16⑥aMacintyre 2010 [24]GermanyFrom birth to 2 years of age16906③b + SPT/IgEMai 2009 [25]SwedenFirst year of life33060–4、0–8⑥⑧cThomson 2010 [21]AustraliaFrom birth to 2 years of age4880–6⑥⑦⑧bKitsantas 2018 [30]United StatesFirst year of life14666⑥cPonsonby 1999 [31]AustraliaFrom birth to 85 days old8657⑥bStrachan 1996 [27]United KingdomFirst month of life179683–16⑥⑦cHarris 2007 [26]United KingdomFrom birth to 2 years of age5230–8⑥⑦⑨⑩b①: gastrointestinal, respiratory, conjunctivitis, otitis media, candida, bacterial,viral, ②: periodontitis and gingivitis, ③: colds, ④: febrile infections, ⑤: bacterial and viral, ⑥: Respiratory infection, ⑦: Gastrointestinal infection, ⑧: Otitis media infection, ⑨: Ear infection, ⑩: Urinary infectiona: Doctor's diagnosis, b: ISAAC:International study of asthma and allergies in childhood, c: Self-report, SPT:Skin prick testTable 2Quality assessment of included studies—cohort study and case–control studyAuthor YearStudy population selectionIntergroup comparabilityOutcome measurementTotal pointsMai 2009 [25]4127Lin 2015 [23]4228Thomson 2010 [21]2125Mckeever 2002 [15]4228Macintyre 2010 [24]4217Illi 2014 [33]4127Ramsey 2007 [22]3126Hsieh 2016 [35]4239Mckeever 2002 [34]4228Kang 2019 [32]4116Bremner 2008 [16]3227Liao 2015 [37]4116Harris 2007 [26]4228Table 3Quality assessment of included studies—cross-sectional studyAuthor Year1234567891011Total pointsDe 2005 [28]YNYNYNYYNYN6Kitsantas 2018 [30]YNYNYNNYYYN6Ponsonby 1999 [31]YYYNYNNNYYN6Kemp 2009 [29]YNYNYYYYNYN7Xu 1999 [36]YNYYYNNYNYN6Strachan 1996 [27]YNYNYNYYNYN6Y:Yes, N:No/Unclear; 1. Define the source of information (survey, record review); 2. List inclusion and exclusion criteria for exposed and unexposed subjects (cases and controls) or refer to previous publications; 3. Indicate time period used for identifying patients; 4. Indicate whether or not subjects were consecutive if not population-based; 5. Indicate if evaluators of subjective components of study were masked to other aspects of the status of the participants; 6. Describe any assessments undertaken for quality assurance purposes (eg, test/retest of primary outcome measurements; 7. Explain any patient exclusions from analysis; 8. Describe how confounding was assessed and/or controlled; 9. lf applicable, explain how missing data were handled in the analysis; 10. Summarize patient response rates and completeness of data collection; 11. Clarify what follow-up, if any, was expected and the percentage of patients for which incomplete data or follow-up was obtained ## Meta-analysis of infection during pregnancy and the risk of AR A total of 5 studies identified the relationship between maternal infection during pregnancy and the risk of AR in offspring. Maternal infection during pregnancy was associated with an increased risk of AR in offspring (OR = 1.34, $95\%$ CI: 1.08–1.67) in all 5 studies as a forest plots (Fig. 2). In performing subgroup analyses, infection during pregnancy was associated with a significantly increased risk of AR in offspring in 3 cohort studies (OR = 1.14, $95\%$ CI: 1.10–1.18) and in the literature that AR was diagnosed by a doctor (OR = 1.38, $95\%$ CI: 1.06–1.81), but there was a trend in 2 case–control studies (OR = 2.37, $95\%$ CI: 1.00–5.61) and in the literature that AR was diagnosed by self-reporting (OR = 1.38, $95\%$ CI: 1.06–1.81). The analysis results are as follows (Table 4).Fig. 2Meta-analysis of infection during pregnancy and the risk of ARTable 4Subgroup analysis of infection during pregnancy and the risk of AROR($95\%$CI)PHeterogeneityTotal studies1.34 (1.08–1.67)$0.00885\%$$P \leq 0.001$Study type Cohort study1.14(1.10–1.18)< $0.0010\%$0.42 Non-cohort study2.37 (1.00–5.65)$0.0584\%$$P \leq 0.001$Outcome evaluation Doctor's diagnosis1.38 (1.06–1.81)$0.0292\%$$P \leq 0.001$ Self-report1.27 (0.91–1.78)$0.160\%$0.34 ## Meta-analysis of infection within 2 years of age and the risk of AR These overall results showed that early infection within 2 years of age was closely associated with childhood AR (OR = 1.25, $95\%$ CI: 1.12–1.40), especially in cohort studies (OR = 1.17, $95\%$ CI: 1.06–1.29) and noncohort studies (OR = 1.51, $95\%$ CI: 1.13–2.02). Compared with 3 papers that confirmed AR by self-reporting, the results showed that early infection within 2 years of age was significantly related to AR in outcome evaluated by ISAAC (OR = 1.17, $95\%$ CI: 1.06–1.30) and in trend related to AR in outcome evaluated by doctor's diagnosis (OR = 2.05, $95\%$ CI: 0.96 -4.35) (Fig. 3).Fig. 3Meta-analysis of infection within 2 years of age and the risk of allergic rhinitis Early upper respiratory tract infection, bronchitis, lower respiratory tract infection, ear infection, gastrointestinal infection and neonatal urinary tract infection were reported in the included studies. Subtype of infection analysis showed that a history of upper respiratory tract infection was associated with an increased risk of AR in children (OR = 1.32, $95\%$ CI: 1.06–1.65). However, neither a history of bronchitis nor lower respiratory tract infection was related to AR in children. Ear infection was associated with an increased risk of AR in children in 4 cohort studies (OR = 1.13, $95\%$ CI: 1.04–1.22) but not in the case–control study (OR = 0.96, $95\%$ CI: 0.87–1.06). Gastrointestinal infection increased the risk of AR in children in the cohort study (OR = 1.20, $95\%$ CI: 1.05–1.37), but there was an obvious tendency in 3 noncohort studies (OR = 1.70, $95\%$ CI: 0.95–3.06, $$p \leq 0.08$$). Urinary tract infection was not associated with an increased risk of AR in children in any type of study. The specific results are shown in Table 5.Table 5Subgroup analysis of infection within 2 years of age and the risk of allergic rhinitisNoOR ($95\%$CI)PHeterogeneityAll studies131.25 (1.12–1.40)< 0.001I2 = $89\%$ $P \leq 0.001$ Cohort studies61.17 (1.06–1.29)0.001I2 = $80\%$ 0.002 Non-cohort studies71.51 (1.13–2.02)0.005I2 = $93\%$ $P \leq 0.001$Outcome evaluation Doctor's diagnosis32.05 (0.96 -4.35)0.06I2 = $95\%$ $P \leq 0.001$ Self-report31.29 (0.81–2.08)0.29I2 = $70\%$ 0.03 ISAAC51.17 (1.06–1.30)0.003I2 = $77\%$ 0.002 ISAAC + SPT11.08 (0.78–1.49)0.64NA NAUpper respiratory tract infection All studies71.32 (1.06–1.65)0.01I2 = $83\%$$P \leq 0.001$ Cohort study11.22 (1.05–1.41)0.008NANA Non-cohort studies61.37 (1.02–1.85)0.02I2 = $85\%$$P \leq 0.001$Bronchitis All studies30.98 (0.88–1.09)0.74I2 = $47\%$0.15 Cohort studies21.32 (0.87–2.00)0.97I2 = $41\%$0.19 Case control study10.96 (0.86–1.07)0.46NANALower respiratory tract All studies51.16 (0.79–1.70)0.46I2 = $75\%$0.003 Cohort studies21.03 (0.63–1.68)0.9I2 = $0\%$0.89 Non-cohort studies31.24 (0.71–2.16)0.45I2 = $87\%$0.003Ear infection All studies51.06 (1.00–1.13)0.06I2 = $39\%$0.16 Cohort studies41.13 (1.04–1.22)0.003I2 = $0\%$0.97 Case control study10.96 (0.87–1.06)0.42NANAGastrointestinal infection All studies41.37 (1.01–1.86)0.04I2 = $91\%$$P \leq 0.001$ Cohort study11.20 (1.05–1.37)0.007NANA Non-cohort studies31.70 (0.95–3.06)0.08I2 = $94\%$$P \leq 0.001$Urinary tract infection All studies21.16 (0.83–1.63)0.39I2 = $70\%$0.07 Cohort study11.32 (1.23–1.41)< 0.001NANA Case control study10.92 (0.63–1.35)0.67NANA ## Sensitivity analysis and publication bias test For maternal infection, the overall sensitivity analysis was carried out by removing the literature one by one, and it was found that the effects on the combined results were stable and reliable. However, when Liao 2015 [37] was removed, the remaining heterogeneity between studies became no longer significant (I2 = $0\%$, $$P \leq 0.39$$), and the results still showed that infection during pregnancy could increase the risk of developing AR in offspring (OR = 1.14, $95\%$ CI: 1.11–1.18). Funnel plots did not show the possibility of publication bias, and Begg’s test determined this outcome ($$P \leq 0.451$$, $P \leq 0.05$) (Fig. 4). Publication bias test was not performed in this analysis because fewer than 10 studies were included [38].Fig. 4Funnel plots For another infection in the first two years of life, based on the results of the above preliminary analysis, the sensitivity analysis was carried out by using the method of eliminating literature one by one. It can be seen that unstable results for the upper respiratory tract infection and greater heterogeneity between studies, when De 2005 [28] and Kang 2019 [32] studies were excluded, a significant decrease of the test for heterogeneity was observed (I2 = $0\%$,$$P \leq 0.02$$), the results of all remaining studies still show that upper respiratory tract infection was associated with increased risk of AR in children (OR = 1.11,$95\%$CI:1.02–1.22). Similarly, for the gastrointestinal tract, after excluding the study of Strachan 1996 [27], the results of the remaining three studies were (OR = 1.59, $95\%$ CI: 1.06–2.37). The sensitivity analysis results of other factors are stable and reliable. As before, no evidence of a significant publication bias was found, which was confirmed by the Begg’s test ($$P \leq 0.246$$, $P \leq 0.05$). However, asymmetry can be found in funnel plots (Fig. 4). In order to solve this problem, we estimated a sensitivity analysis using the trim and fill method (adding 2 new virtual studies) showed that the results was reliable (OR = 1.15,$95\%$CI:1.01–1.30). Thus, we found that the outcome was not affected by publication bias. ( Fig. 5).Fig. 5Filled funnel plot ## Discussion In the current study, we shed light on the link between infection and AR in children. The results of the stratified analysis found that maternal infections during pregnancy as well as early infections of the upper respiratory tract, gastrointestinal infections and ear infection within 2 years old increased the risk of AR in children. AR is more prevalent in children than any other chronic illness, which makes it a serious public health concern. Taking preventive measures requires a thorough understanding of the risk factors for allergic diseases in children. It is increasingly recognized that the development of fetal and infant allergies is influenced in large part by the early years of their life, and this cannot be ignored. During the past few years, the DOHaD theory has become a hotspot in the research field of allergic diseases. Maternal infection during pregnancy can increase the risk of asthma and eczema in offspring [11]. A meta-analysis [12] also showed that early respiratory tract infection within 2 years old was associated with the development of asthma in school-age children. It has been demonstrated in numerous studies that maternal adverse exposure during pregnancy, such as passive smoking [39], diet [40], psychological status [41], pregnancy complications [42], and antibiotic exposure during pregnancy [43], was associated with a greater likelihood of AR in offspring. Additionally, studies have provided compelling evidence that adverse exposures in the early postnatal period, such as antibiotic use [13], pet exposure [44] and air pollution [45], may increase the risk of a child developing allergies. This meta-analysis demonstrated that infection during the first 1000 days of life from pregnancy to 2 years of age could increase the subsequent childhood AR. To the best of our knowledge, some of these correlation results may be partly explained by the fact that infections can alter microbiome stability. It has been suggested that the balance of gut and lung microbes may play a role in allergic disease [46]. Studies have shown that AR patients have fewer gut microbes than healthy individuals [47]. As a result of these arguments, we may be able to conclude that the microbiota may play a role in allergic diseases. The microbial composition of the nasopharynx has been shown to influence airway sensitivity in a recent study [48]. Similarly, infection during pregnancy can lead to the presence of bacteria in the uterus or amniotic fluid, which can be transmitted to the fetus and then affect the gut flora of the fetus [49]. According to Gayen et al. [ 50], the control of inflammatory, immune, and respiratory processes was influenced by differential leukocyte gene expression in neonates who were exposed to fetal membrane infection during pregnancy. In animal studies, elevated IL-17A production has been associated with allergic airway inflammation in neonates infected with Streptococcus pneumoniae [51]. The latest study also showed that mice infected with Streptococcus pneumonia developed more pronounced airway responses and had a higher level of serum-specific IgE and Th2 cytokines in the lung. It has therefore been shown that early respiratory infection with Streptococcus pneumoniae can exacerbate later allergic airway inflammation and adult-associated asthma caused by house dust mites [52]. It is worth mentioning that research on the effects of antibiotic use on allergic diseases has gradually increased. It has been noted that both exposure to antibiotics and infections have been shown to be related to allergic diseases in the absence of mutual adjustment factors [31]. Slightly inconsistent with this opinion, Mai et al. [ 53] noted that early postnatal respiratory infection may confound the association between antibiotic use and allergic diseases. McKeever et al. [ 34] examined the interaction of infection and antibiotic use during pregnancy with allergic disease in offspring in two models simultaneously. According to their findings, infections are not associated with AR in offspring, although adjusting it did not notably affect the use of antibiotics, increasing the risk of allergic disease. However, Lin et al. 's study [54] found that both the initial infection and antibiotic use are independent risk factors for secondary atopic dermatitis in children. Thus, future research should examine whether infections and AR are related, whether the correlation could be confounded by antibiotic use, and whether antibiotics are related to these conditions. The strength of this study lies in the fact that a search of eight databases was conducted for this study, and the quality of the included literatures were rated as moderate to high. The shortcoming of this study: there is heterogeneity among evidence, and clinically, heterogeneity exists first, as an example, De et al. [ 28], was a cross-sectional study with a selected specific population aged 8–13 years, and a recall questionnaire provided the main evaluation method. Then, the large majority of reports were from populations of predominantly European countries, which would lack the accuracy of interpretation of the overall population. Second, methodological differences could also contribute to the divergence of results between studies, as different adjusted confounding factors among the studies, especially most studies did not examine the impact of antibiotic use following infection. In addition, with the exception of cohort studies, outcome measures obtained through self-report or information from parental interviews may be liable to recall bias. Furthermore, very few studies have been conducted on the impact of pregnancy infections and urinary tract infections on AR, and there was no clear explanation for pregnancy infection. In this analysis, the 5 studies addressed different types of infection, and more prospective cohort studies should ideally be designed with a greater focus on the confounding effects of infection and antibiotic use in the future. ## Conclusion In conclusion, our study identifies and characterizes that infection during pregnancy, early upper respiratory infection, gastrointestinal infections and ear infection within 2 years of age is associated with subsequent childhood AR and suggests that health care workers should strengthen health education in their communities. However, the confounding effects of antibiotics and infection remain to be studied in more detail. ## Supplementary Information Additional file 1. Searchteims and strategies. Additional file 2. Listof excluded studies with reasons for exclusion. ## References 1. Diagnosis and treatment in children with allergic rhinitis. Chin J Pract Pediatr (2019.0) 34 169-175 2. Blaiss MS, Hammerby E, Robinson S. The burden of allergic rhinitis and allergic rhinoconjunctivitis on adolescents: a literature review. Ann Allergy Asthma Immunol (2018.0) 121 43-52.e3. DOI: 10.1016/j.anai.2018.03.028 3. Pols DH, Wartna JB, Van Alphen EI. Interrelationships between atopic disorders in children: a meta-analysis based on ISAAC questionnaires. PLoS ONE (2015.0) 10 e0131869. DOI: 10.1371/journal.pone.0131869 4. Cohen R, Gutvirtz G, Wainstock T. Maternal urinary tract infection during pregnancy and long-term infectious morbidity of the offspring. Early Hum Dev (2019.0) 136 54-59. DOI: 10.1016/j.earlhumdev.2019.07.002 5. Huang J, Zheng L, Su Y. Effects of group B streptococcus infection on vaginal micro-ecology and pregnancy outcomes of pregnant women in late pregnancy. Eur J Obstet Gynecol Reprod Biol (2021.0) 267 274-279. DOI: 10.1016/j.ejogrb.2021.11.419 6. Zhan Y, Si M, Li M. The risk of Helicobacter pylori infection for adverse pregnancy outcomes: a systematic review and meta-analysis. Helicobacter (2019.0) 24 e12562. DOI: 10.1111/hel.12562 7. Butel MJ, Waligora-Dupriet AJ, Wydau-Dematteis S. The developing gut microbiota and its consequences for health. J Dev Orig Health Dis (2018.0) 9 590-597. DOI: 10.1017/S2040174418000119 8. Ahmadizar F, Vijverberg SJH, Arets HGM. Early-life antibiotic exposure increases the risk of developing allergic symptoms later in life: a meta-analysis. Allergy (2018.0) 73 971-986. DOI: 10.1111/all.13332 9. Aghaali M, Hashemi-Nazari SS. Association between early antibiotic exposure and risk of childhood weight gain and obesity: a systematic review and meta-analysis. J Pediatr Endocrinol Metab (2019.0) 32 439-445. DOI: 10.1515/jpem-2018-0437 10. Grech A, Collins CE, Holmes A. Maternal exposures and the infant gut microbiome: a systematic review with meta-analysis. Gut Microbes (2021.0) 13 1-30. DOI: 10.1080/19490976.2021.1897210 11. Zhu T, Zhang L, Qu Y. Meta-analysis of antenatal infection and risk of asthma and eczema. Medicine (Baltimore) (2016.0) 95 e4671. DOI: 10.1097/MD.0000000000004671 12. Van Meel ER, Mensink-Bout SM, Den Dekker HT. Early-life respiratory tract infections and the risk of school-age lower lung function and asthma: a meta-analysis of 150 000 European children. Eur Respir J (2022.0) 60 2102395. DOI: 10.1183/13993003.02395-2021 13. Ni J, Friedman H, Boyd BC. Early antibiotic exposure and development of asthma and allergic rhinitis in childhood. BMC Pediatr (2019.0) 19 225. DOI: 10.1186/s12887-019-1594-4 14. Lu C, Norbäck D, Li Y. Early-life exposure to air pollution and childhood allergic diseases: an update on the link and its implications. Expert Rev Clin Immunol (2020.0) 16 813-827. DOI: 10.1080/1744666X.2020.1804868 15. Mckeever TM, Lewis SA, Smith C. Early exposure to infections and antibiotics and the incidence of allergic disease: a birth cohort study with the west midlands general practice research database. J Allergy Clin Immunol (2002.0) 109 43-50. DOI: 10.1067/mai.2002.121016 16. Bremner SA, Carey IM, Dewilde S. Infections presenting for clinical care in early life and later risk of hay fever in two UK birth cohorts. Allergy (2008.0) 63 274-283. DOI: 10.1111/j.1398-9995.2007.01599.x 17. Stang A. Critical evaluation of the Newcastle-Ottawa scale for the assessment of the quality of nonrandomized studies in meta-analyses. Eur J Epidemiol (2010.0) 25 603-605. DOI: 10.1007/s10654-010-9491-z 18. Zeng XT, Liu H, Chen X. Meta-analysis series four: quality assessment tools for observational studies. Chin J Evid Based Cardiovasc Med (2012.0) 4 297-299 19. Cumpston M. Updated guidance for trusted systematic reviews: a new edition of the Cochrane handbook for systematic reviews of interventions. Cochrane Database Syst Rev (2019.0) 10 ED000142. PMID: 31643080 20. Lee SW, Koo MJ. PRISMA 2020 statement and guidelines for systematic review and meta-analysis articles, and their underlying mathematics: life cycle committee recommendations. Life Cycle (2022.0) 2 e9. DOI: 10.54724/lc.2022.e9 21. Thomson JA, Widjaja C, Darmaputra AAP. Early childhood infections and immunisation and the development of allergic disease in particular asthma in a high-risk cohort: A prospective study of allergy-prone children from birth to six years. Pediatr Allergy Immunol. (2010.0) 21 1076-85. DOI: 10.1111/j.1399-3038.2010.01018.x 22. Ramsey CD, Gold DR, Litonjua AA. Respiratory illnesses in early life and asthma and atopy in childhood. J Allergy Clin Immunol (2007.0) 119 150-156. DOI: 10.1016/j.jaci.2006.09.012 23. Lin CH, Lin WC, Wang YC. Association between neonatal urinary tract infection and risk of childhood allergic rhinitis. Medicine (Baltimore) (2015.0) 94 1. PMID: 26402832 24. Macintyre EA, Chen CM, Herbarth O. Early-life otitis media and incident atopic disease at school age in a birth cohort. Pediatr Infect Dis J (2010.0) 29 E96-E99. DOI: 10.1097/INF.0b013e3181fcd9e8 25. Mai X, Kull I, Bergström A. Antibiotic use in early life and the development of allergic disorders: Infection as the explanation. Allergy (2009.0) 64 21-22 26. Harris JM, Mills P, White C. Recorded infections and antibiotics in early life: associations with allergy in UK children and their parents. Thorax (2007.0) 62 631-637. DOI: 10.1136/thx.2006.072124 27. Strachan DP, Taylor EM, Carpenter RG. Family structure, neonatal infection, and hay fever in adolescence. Arch Dis Child (1996.0) 74 422-426. DOI: 10.1136/adc.74.5.422 28. De Meer G, Janssen NaH, Brunekreef B. Early childhood environment related to microbial exposure and the occurrence of atopic disease at school age. Allergy (2005.0) 60 619-25. DOI: 10.1111/j.1398-9995.2005.00746.x 29. Kemp A, Ponsonby AL, Dwyer T. The interaction between early life upper respiratory tract infection and birth during the pollen season on rye-sensitized hay fever and ryegrass sensitization–a birth cohort study. Pediatr Allergy Immunol (2009.0) 20 536-544. DOI: 10.1111/j.1399-3038.2008.00817.x 30. Kitsantas P, Nirmalraj L. Effects of respiratory syncytial virus infection in infancy on asthma and respiratory allergy in 6-year-old children. South Med J (2018.0) 111 698-702. DOI: 10.14423/SMJ.0000000000000881 31. Ponsonby AL, Couper D, Dwyer T. Relationship between early life respiratory illness, family size over time, and the development of asthma and hay fever: a seven year follow up study. Thorax (1999.0) 54 664-669. DOI: 10.1136/thx.54.8.664 32. Kang X, Tu H, Tian T. Home environment and diseases in early life are associated with allergic rhinitis. Int J Pediatr Otorhinolaryngol (2019.0) 118 47-52. DOI: 10.1016/j.ijporl.2018.12.015 33. Illi S, Weber J, Zutavern A. Perinatal influences on the development of asthma and atopy in childhood. Ann Allergy Asthma Immunol (2014.0) 112 132. DOI: 10.1016/j.anai.2013.11.019 34. Mckeever TM, Lewis SA, Smith C. The importance of prenatal exposures on the development of allergic disease: a birth cohort study using the West Midlands General Practice Database. Am J Respir Crit Care Med (2002.0) 166 827-832. DOI: 10.1164/rccm.200202-158OC 35. Hsieh VCR, Liu CC, Hsiao YC. Risk of allergic rhinitis, allergic conjunctivitis, and eczema in children born to mothers with gum Inflammation during pregnancy. PLoS One. (2016.0) 11 e0156185. DOI: 10.1371/journal.pone.0156185 36. Xu B, Järvelin MR, Pekkanen J. Prenatal factors and occurrence of rhinitis and eczema among offspring. Allergy (1999.0) 54 829-836. DOI: 10.1034/j.1398-9995.1999.00117.x 37. Liao SY, Ye YF, Ye JX. Maternal exposure factors related with children's allergic rhinitis during pregnancy. J Wenzhou Med Univ (2015.0) 7 493-496 38. 38.Higgins J. Cochrane handbook for systematic reviews of interventions. Version 5.1. 0 [updated March 2011]. The Cochrane Collaboration. https://www.cochrane-handbook.org. 2011. 39. Zhou Y, Chen J, Dong Y. Maternal tobacco exposure during pregnancy and allergic rhinitis in offspring: a systematic review and meta-analysis. Medicine (Baltimore) (2021.0) 100 e26986. DOI: 10.1097/MD.0000000000026986 40. Baïz N, Just J, Chastang J. Maternal diet before and during pregnancy and risk of asthma and allergic rhinitis in children. Allergy Asthma Clin Immunol (2019.0) 15 40. DOI: 10.1186/s13223-019-0353-2 41. Flanigan C, Sheikh A, Dunngalvin A. Prenatal maternal psychosocial stress and offspring's asthma and allergic disease: a systematic review and meta-analysis. Clin Exp Allergy (2018.0) 48 403-414. DOI: 10.1111/cea.13091 42. Yavuz ST, Siebert S, Akin O. Perinatal risk factors for asthma in children with allergic rhinitis and grass pollen sensitization. Allergy Asthma Proc (2018.0) 39 1-7. DOI: 10.2500/aap.2018.39.4122 43. Cait A, Wedel A, Arntz JL. Prenatal antibiotic exposure, asthma and the atopic march: a systematic review and meta-analysis. Allergy (2022.0) 77 3233. DOI: 10.1111/all.15404 44. Zhang Y, Tan M, Qian X. Interaction between early-life pet exposure and methylation pattern of ADAM33 on allergic rhinitis among children aged 3–6 years in China. Allergy Asthma Clin Immunol (2021.0) 17 44. DOI: 10.1186/s13223-021-00526-5 45. Lin L, Li T, Sun M. Effect of particulate matter exposure on the prevalence of allergic rhinitis in children: a systematic review and meta-analysis. Chemosphere (2021.0) 268 128841. DOI: 10.1016/j.chemosphere.2020.128841 46. Pascal M, Perez-Gordo M, Caballero T. Microbiome and Allergic Diseases. Front Immunol (2018.0) 9 1584. DOI: 10.3389/fimmu.2018.01584 47. Zhou MS, Zhang B, Gao ZL. Altered diversity and composition of gut microbiota in patients with allergic rhinitis. Microb Pathog (2021.0) 161 105272. DOI: 10.1016/j.micpath.2021.105272 48. Tsai MH, Shih HJ, Su KW. Nasopharyngeal microbial profiles associated with the risk of airway allergies in early childhood. J Microbiol Immunol Infect (2022.0) 55 777. DOI: 10.1016/j.jmii.2022.01.006 49. Penders J, Thijs C, Vink C. Factors influencing the composition of the intestinal microbiota in early infancy. Pediatrics (2006.0) 118 511-521. DOI: 10.1542/peds.2005-2824 50. Gayen Nee' Betal S, Murthy S, Favara M. Histological Chorioamnionitis induces differential gene expression in human cord blood mononuclear leukocytes from term neonates. Sci Rep (2019.0) 9 5862. DOI: 10.1038/s41598-019-42205-x 51. Yang B, Liu R, Yang T. Neonatal Streptococcus pneumoniae infection may aggravate adulthood allergic airways disease in association with IL-17A. PLoS ONE (2015.0) 10 e0123010. DOI: 10.1371/journal.pone.0123010 52. Peng D, Shi Y, Pang J. Early-life infection of the airways with Streptococcus pneumoniae exacerbates HDM-induced asthma in a murine model. Cell Immunol (2022.0) 376 104536. DOI: 10.1016/j.cellimm.2022.104536 53. Mai XM, Kull I, Wickman M. Antibiotic use in early life and development of allergic diseases: respiratory infection as the explanation. Clin Exp Allergy (2010.0) 40 1230-1237. DOI: 10.1111/j.1365-2222.2010.03532.x 54. Lin TL, Fan YH, Chang YL. Early-life infections in association with the development of atopic dermatitis in infancy and early childhood: a nationwide nested case-control study. J Eur Acad Dermatol Venereol (2022.0) 36 615-622. DOI: 10.1111/jdv.17908
--- title: 'Associations between parenthood and dementia in men and women: biology or confounding?' authors: - Saima Basit - Jan Wohlfahrt - Heather A. Boyd journal: BMC Neurology year: 2023 pmcid: PMC9976501 doi: 10.1186/s12883-023-03108-7 license: CC BY 4.0 --- # Associations between parenthood and dementia in men and women: biology or confounding? ## Abstract ### Background High parity and extremes of age at first birth have been linked with increased dementia risk in women, with exposure to pregnancy-associated physiological changes proposed as an explanation. However, confounding by socioeconomic and lifestyle factors could also produce such associations, whereby men would share similar patterns of association. We investigated whether these associations hold for both sexes. ### Methods In a cohort study including all women ($$n = 2$$,222,638) and men ($$n = 2$$,141,002) ≥ 40 years of age in 1994–2017 in Denmark, we used Cox regression to evaluate associations between number of children, age at first birth, and dementia risk separately for women and men. ### Results During follow-up, 81,413 women and 53,568 men (median age at diagnosis, 83.3 and 80.3 years, respectively) developed dementia. Compared with having one child, having two or more children was associated with modest decreases in overall dementia risk in both sexes (hazard ratio [HR] range 0.82–0.91, Pdifference men vs. women = 0.07). Although the associations between childlessness and overall dementia risk differed statistically for men and women, the association magnitudes differed only slightly (HRmen 1.04, $95\%$ confidence interval [CI] 1.01–1.06; HRwomen 0.99, $95\%$ CI 0.97–1.01; $$P \leq 0.002$$). Associations between age at becoming a parent and overall dementia were also similar for women and men, with the exception of older (≥ 40 years) first-time parents (HRmen 1.00, $95\%$ CI 0.96–1.05; HRwomen 0.92, $95\%$ CI 0.86–0.98; $$P \leq 0.01$$). With few exceptions, sub-analyses by dementia subtype and timing of onset also revealed similar patterns and effect magnitudes for women and men. ### Conclusions Associations between number of children, age at becoming a parent, and dementia risk were similar for both sexes. Lifestyle and socioeconomic factors are more likely to explain the observed associations than normal pregnancy-related physiological changes. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12883-023-03108-7. ## Background Incurable and untreatable, dementia is emerging as a leading cause of morbidity and mortality worldwide, with the economic and social costs of dementia care placing increasing burdens on society [1–4]. Because dementia may affect women disproportionately [5], interest in sex-specific risk factors for dementia has intensified [6, 7]. Since factors that affect brain health early in adulthood may be especially relevant, the potential impact of pregnancy on dementia risk in women is of particular interest. Pregnancy complications such as preeclampsia and stillbirth are associated with an increased risk of dementia [8–10], suggesting that vascular pathology during pregnancy might be an indicator of later dementia risk. Interestingly, studies have also linked childbirth in general and number of children with Alzheimer’s disease risk in women [11–14], and a recent study found that that having five or more completed pregnancies was associated with a substantial increase in the risk of dementia overall [15]. Repeated exposure to the routine physiological changes associated with pregnancy (including significant changes in endocrine, immunological, and cardiovascular function) has been proposed as an explanation for these observations. Interestingly, however, a new study recently contradicted the previous studies, finding instead a U-shaped relationship between parity and dementia, with similar associations in men and women [16]. The underlying biology of pregnancy has also been suggested to account for the increased risk of Alzheimer’s disease observed in women who are very young (< 20 years) or relatively old (≥ 40 years) when they deliver their first child [17] and the increased risk of dementia overall in women < 25 years of age at first delivery [16]. However, rather than reflecting a causal link between normal pregnancy-associated physiological changes and later dementia, previously observed associations between number of children, age at first birth, and increased dementia risk in women could largely result from uncontrolled confounding by socioeconomic and lifestyle factors. Factors such as education, income, health-seeking behavior, access to healthcare, social engagement in the community, nutrition, stress, smoking, and alcohol use are all associated with reproductive patterns. Uncontrolled confounding by social factors and large between-country differences in the distributions of these factors and in their associations with reproductive patterns might also explain the between-study variation that exists in observed association magnitudes, direction, and patterns. In a nationwide cohort study of more than 4 million persons, we examined associations between number of children and age at becoming a parent with dementia risk in both women and men. By examining the associations in both sexes under the same study conditions, we could evaluate how likely it was that observed associations could be explained by pregnancy-related factors rather than by uncontrolled confounding unrelated to the biology of pregnancy; if the observed associations were equally strong for men and women, an underlying mechanism based on female sex hormones would be unlikely. ## Data sources and cohort The Civil Registration System registers all Danish residents using unique personal identification numbers and updates information on demographics, kinship, and vital status daily [18]. The National Patient Register contains information on all hospital discharge diagnoses assigned since 1977 and all outpatient diagnoses assigned since 1995, registered using International Classification of Diseases (ICD) codes [19]. The Causes of Death Register holds information on the causes (underlying and contributing) of deaths in Denmark since 1970 [20]. The National Prescription Register contains individual-level information on all prescriptions filled in Denmark since 1994, recorded using Anatomic Therapeutic Chemical (ATC) codes [21]. Using information from the Civil Registration System, we constructed a study cohort consisting of all persons who were ≥ 40 years of age and resident in Denmark at some point between 1994 and 2017. Cohort members were followed from 40 years of age or 1 January 1994 (when ICD-10 codes were introduced in Denmark, allowing more reliable sub-classification of dementia), whichever came later, until the first of dementia, death, emigration, designated “missing” in the Civil Registration System, or 30 May 2017 (end of follow-up). Persons who died or emigrated before 1994 were not eligible to be included in the cohort. Persons with a diagnosis of dementia before the start of follow-up ($$n = 2839$$) were excluded. ## Exposure: number of children and age at becoming a parent We obtained information on number of live-born children and age at becoming a parent (also referred to as age at first birth) through the Civil Registration System. Number of children was treated as a time-dependent variable and increased with the birth of additional children during follow-up. ## Outcome: Dementia Dementia was defined as registration of a dementia code in the National Patient Register during follow-up and was classified as Alzheimer’s disease (ICD-10 codes F00.0-F00.9, G30.0, G30.1, G30.8, G30.9), vascular dementia (ICD-10 codes F01.0-F01.9), and other/unspecified dementia (ICD-10 code F02.0, F03.9). ( The ICD-8 codes 290.00, 290.10, 290.11, 290.19, and 299.99 were also used when excluding persons diagnosed with dementia before the start of follow-up). Dementia registered before 65 years of age was considered early-onset dementia; dementia diagnoses first registered ≥ 65 years were considered late-onset dementia. ## Covariates We adjusted all analyses for birth year (5-year intervals), age (time-dependent variable), and key comorbidities present at the start of follow-up or developing during the study period (time-dependent variables). Comorbidities were identified using the National Patient Register and the Causes of Death Register, based on the following ICD-8 and-10 codes: stroke, 433.xx, 436.xx, I63.0-I63.9; myocardial infarction, 410.xx, I21.0-I23.9; ischemic heart disease, 411.xx-414.xx, 420.xx-429.xx, I20.0-I20.9, I24.0-I25.9; heart failure, 427.09–427.19, 427.99, 428.99, 782.49, I50.0-I50.9; diabetes, 249.00–250.09, E10.0-E14.9; renal disease, 400.39, 403.99, 404.99, I12.0-I13.9, I15.0, I15.1, N18.0-N18.9. Hypertension was identified based on the filling of two prescriptions for anti-hypertensive medication (ATC codes C02-03, C07-09 registered in the National Prescription Register) within a 6-month period. ## Analyses Using Cox regression with age as the underlying time, we estimated hazard ratios for dementia overall and by subtype, comparing persons with no children and those with two or more children to persons with one child. Among persons with one or more children, we also analyzed the association between age at first birth and dementia. To evaluate whether there were sex-specific differences in the associations of interest, we included parental sex in our models and evaluated the interaction between sex and the reproductive history variables. When analyzing associations with dementia subtypes, we used competing risk methodology [22] to estimate separate hazard ratios for each dementia subtype. In these analyses, follow-up ended with a person’s first dementia diagnosis, regardless of subtype, but only cases of the specific dementia subtype being analyzed counted as an event. ( As in the overall analyses, follow-up also ended at death, emigration, registration as “missing”, or the end of the study period). To assess the importance of death as a competing risk in our main analyses, we also evaluated associations between number of children, age at first birth, and all-cause mortality in a Cox regression analysis with all-cause mortality as the outcome (identified using the Civil Registration System) and with censoring at the first dementia diagnosis. We assessed adherence to the proportional hazards assumption by examining the interaction between the exposure variables and age (< 65 vs ≥ 65 years). All analyses were performed using SAS statistical software, version 9.4 (SAS Institute, Inc., Cary, N.C.). ## Results We followed 2 222 638 women and 2 141 002 men for 32 656 082 and 30 583 988 person-years, respectively, with a median follow-up of 15.5 years (interquartile range [IQR] 7.7–23.4 years) for women and 14.8 years (IQR 7.3–23.4) for men. During follow-up, 134 981 persons developed dementia, of whom 81 413 ($60.3\%$) were women and 53 568 ($39.7\%$) were men. The initial dementia diagnosis was vascular dementia for 6790 women ($8.3\%$) and 6443 men ($12.0\%$), Alzheimer’s disease for 23 003 women ($28.3\%$) and 13 743 men ($25.7\%$), and other/unspecified for 51 620 women ($63.4\%$) and 33 382 men ($62.3\%$). The median age at diagnosis was 83.3 years (IQR 77.7–87.9 years) for women and 80.3 years (IQR 73.9–85.4 years) for men. Table 1 shows the characteristics of the study cohort at the start of follow-up. Table 1Characteristics at the start of follow-upa for a cohort of persons ≥ 40 years old followed in the period 1994–2017 in DenmarkCharacteristic at the start of follow-upWomen($$n = 2$$ 222 638)Men($$n = 2$$ 141 002)Numberb%Numberb%Number of children 0564 57325.40592 93927.69 1401 42118.06373 57017.45 2783 06935.23726 95633.95 3347 88515.65325 19715.19 492 1194.1489 2094.17 ≥ 533 5711.5133 1311.55Age (years) at birth of first child No children555 21124.98561 56226.23 < 20179 3328.0741 6201.94 20–24606 32627.28384 13917.94 25–29523 46323.55582 78327.22 30–34245 37411.04357 20816.68 35–3991 5644.12148 2426.92 ≥ 4021 3680.9665 4483.06Year of birth < 1920247 88111.15146 5306.84 1920–1929243 48510.95207 1439.68 1930–1939267 28612.03261 13612.20 1940–1949379 10917.06394 80418.44 1950–1959378 92417.05397 86418.58 1960–1969416 37618.73438 51320.48 ≥ 1970289 57713.03295 01213.78Age (years) at the start of follow-up < 451 121 35450.451 169 29054.61 45–49206 7259.30216 95810.13 50–54163 8167.37168 9107.89 55–59138 0386.21136 3136.37 60–64124 5925.61117 2225.48 65–69122 8105.531081 965.05 ≥ 70345 30315.54224 11310.47Heart disease Yes21 5430.9722 8571.07 No2 201 09599.032 118 14598.93Stroke Yes54 3142.4478 2813.66 No2 168 32497.562 062 72196.34Kidney disease Yes23230.1029760.14 No2 220 31599.902 138 02699.86Diabetes Yes35 2291.5938 6081.80 No2 187 40998.412 102 39498.20Hypertension Yes69 8873.1442 1421.97 No2 152 75196.862 098 86098.03aFollow-up began on 1 January 1994 or at 40 years of age, whichever came later. Accordingly, persons who began follow-up in 1994 represented a range of ages from 40 to 109 years. In contrast, persons who began follow-up at a later date were necessarily all 40 years of age at study entry, as they had aged into the cohort. This condition accounts for the overrepresentation of persons < 45 years of age at the start of follow-upbNote that these totals reflect the numbers at the start of follow-up. Because relatively young persons were overrepresented at this point, rates of heart disease, stroke, kidney disease, diabetes, and hypertension are lower than might be expected if the age distribution of the cohort at the start of follow-up reflected the age distribution in the general population. Since with the exception of birth year, age at birth of the first child, and age at the start of follow-up, all variables were time-dependent, variable distributions will have changed during the study period The overall pattern of association between number of children and overall dementia risk differed statistically for men and women ($$p \leq 0.009$$, Fig. 1; Supplementary Table 1), largely due to a small but influential difference between childless men and women. Compared with having one child, being childless was associated with a slight increase in overall dementia risk in men (hazard ratio [HR] 1.04, $95\%$ confidence interval [CI] 1.01–1.06) but was not associated with dementia risk in women (HR 0.99, $95\%$ CI 0.97–1.01) ($$p \leq 0.002$$ for difference). In contrast, patterns of association among persons with two or more children did not differ for men and women (comparison of the overall pattern of association in men and women with children: $$p \leq 0.07$$ for difference). Compared with having only one child, having two, three, four, or five or more children was associated with modest decreases in overall dementia risk in both men and women (hazard ratio [HR] range: 0.82–0.91) (Fig. 1; Supplementary Table 1).Fig. 1Associations between number of children and overall dementia for men and women, 1994–2017, Denmark. Hazard ratios with $95\%$ confidence intervals compare the risks of dementia among women (red) and men (blue) with different numbers of children. All hazard ratios are adjusted for birth year (5-year intervals), cardiovascular disease, stroke, hypertension, chronic kidney disease and diabetes; age was the underlying time scale in the Cox model Patterns of association by number of children were generally comparable for men and women for both early- ($$p \leq 0.24$$) and late-onset dementia ($$p \leq 0.20$$) (Supplementary Figure 1 and Supplementary Table 2). Sex-specific association magnitudes only differed for the association between childlessness and late-onset dementia (HRmen 1.01, $95\%$ CI 0.99–1.04; HRwomen 0.98, $95\%$ CI 0.97–1.00; $$p \leq 0.04$$) and even then, the difference in HRs was small. Similarly, there was little evidence of sex-specific differences in association magnitudes for persons with two or more children for any dementia subtypes (comparison of the overall pattern of association in men and women with children: $$p \leq 0.30$$, 0.46, and 0.19 for vascular dementia, Alzheimer’s disease, and other/unspecified dementia, respectively) (Supplementary Figure 2 and Supplementary Table 3). Associations between childlessness and both vascular dementia and Alzheimer’s disease were also of similar magnitude for men and women (vascular dementia: HRmen 1.00, $95\%$ CI 0.93–1.07, HRwomen 0.97, $95\%$ CI 0.91–1.04, $$p \leq 0.49$$; Alzheimer’s disease: HRmen 0.84, $95\%$ CI 0.79–0.88, HRwomen 0.89, $95\%$ CI 0.85–0.92, $$p \leq 0.08$$). However, childlessness was associated with a greater increase in the risk of other/unspecified dementia in men (HR 1.11, $95\%$ CI 1.08–1.14) than in women (HR 1.03, $95\%$ CI 1.01–1.05) ($p \leq 0.0001$). Similar patterns were observed when dementia subtype and timing of dementia onset were examined together (Supplementary Figure 3 and Supplementary Table 4). The overall pattern of association between age at first birth and overall dementia risk differed statistically for men and women ($$p \leq 0.02$$), owing largely to a modest difference in the associations observed for older (≥ 40 years) first-time parents (Fig. 2; Supplementary Table 5); older first-time mothers had a slightly reduced risk of dementia overall, compared with women who were 25–29 years of age at first birth (HR 0.92, $95\%$ CI 0.86–0.98), whereas no reduction in dementia risk was observed for older first-time fathers (HR 1.00, $95\%$ CI 0.96–1.05) ($$p \leq 0.01$$ for difference). Becoming a parent at a young age (< 20 years) was associated with a modest increase in overall dementia risk in both sexes (HRmen 1.18, $95\%$ CI 1.08–1.29, HRwomen 1.10, $95\%$ CI 1.06–1.15, $$p \leq 0.17$$) (Fig. 2).Fig. 2Associations between age at first birth and overall dementia for men and women, 1994–2017, Denmark. Hazard ratios with $95\%$ confidence intervals compare the risks of dementia among women (red) and men (blue) with different ages at first childbirth. All hazard ratios are adjusted for birth year (5-year intervals), cardiovascular disease, stroke, hypertension, chronic kidney disease and diabetes; age was the underlying time scale in the Cox model The strength of the associations between age at first birth and risk of early-onset dementia differed little for men and women, regardless of age category ($$p \leq 0.78$$ for overall difference, Supplementary Figure 4 and Supplementary Table 6). The overall pattern of association between age at first birth and risk of late-onset dementia was also similar for both sexes ($$p \leq 0.08$$, Supplementary Figure 4 and Supplementary Table 6), with the exception of a small difference in sex-specific association magnitudes for the oldest (≥ 40 years) first-time parents (HRmen 1.01, $95\%$ CI 0.96–1.06, HRwomen 0.92, $95\%$ CI 0.86–0.98, $$p \leq 0.01$$). There was little evidence of sex-specific differences in patterns of association between age at first birth and either vascular dementia ($$p \leq 0.98$$) or unspecified dementia ($$p \leq 0.18$$) (Supplementary Table 7). Overall, sex-specific patterns of association differed for Alzheimer’s disease ($$p \leq 0.02$$), but within age categories, any sex-specific differences were small (Supplementary Figure 5 and Supplementary Table 7). Examining dementia subtype and timing of onset together did not reveal evidence of meaningful differences in the associations between age at first birth and dementia risk in women and men (Supplementary Figure 6 and Supplementary Table 8). The pattern of associations for mortality were similar to the patterns we observed for dementia, although the associations were somewhat more pronounced for childless persons and persons who were young when they became parents (Supplementary Tables 9 and 10). The mortality results suggest that for these groups (childless persons and persons who became parents at a young age) in particular, the hazard ratios for dementia (the focus in this study) cannot necessarily be used to discuss cumulative risk ratios. ## Discussion In this population-based cohort study of more than 4 million persons, we found that, with few exceptions, patterns of association between number of children, age at becoming a parent, and dementia risk were the same for men and women; where sex-specific estimates did differ, the absolute differences in effect magnitude were small. The similarity of our findings in both sexes challenges previous suggestions that pregnancy-related hormonal changes or other normal pregnancy-associated physiological changes might explain observed associations between number of children and dementia risk in women. Rather than being evidence of a causal mechanism involving pregnancy-related hormonal factors, the observed associations between reproductive factors and dementia are more likely due to uncontrolled confounding by social factors that include educational attainment, income, employment, cohabitation status, social network, and lifestyle factors. In early adulthood, these factors may affect reproductive patterns and decisions to have (or not have) children, how many children to have, and when to have them; some are also associated with involuntary childlessness [23–26]. In turn, reproductive patterns and decisions can exert effects on later socioeconomic status, social behaviours, stress levels, and lifestyle. Since demographic, social, lifestyle, and health factors both early and later in life are also associated with later dementia risk [27–31], these factors are obvious potential confounders or mediators. For example, the observation that childlessness was associated with increased risks of early-onset dementia in both sexes might reflect differences in social contact and support experienced by persons with and without family networks [32, 33]; alternatively, persons in families with a history of dementia may choose not to have children so as not to pass on a genetic predisposition to dementia. Similarly, becoming a parent at a young age is associated with lower socioeconomic status [34], which is a known risk factor for dementia [27, 29, 31]. Even where results differed for men and women, confounding is a more plausible explanation than the biology of pregnancy; associations between social factors and reproductive decisions differ for men and women [23, 25, 26]. Although our results linking extremes of age at first birth with modestly increased dementia risks are consistent with the results of previous studies [16, 17], our finding of similar associations for men and women suggests uncontrolled confounding as a more likely explanation for the associations than female reproductive biology. With the exception of findings from a recent study based on data from the UK biobank, [16] our results for number of children differed markedly from those reported by others. While most previous studies reported that motherhood was associated with Alzheimer’s disease risk and that a woman’s risk appeared to increase with increasing numbers of children [13–15], both we found that being childless was associated with modest increases in dementia risk, whereas having two or more children either was not associated with dementia risk or was associated with modest risk reductions. We found no evidence of a trend in dementia risk with increasing numbers of children and certainly no evidence that having large numbers of children was harmful, contrary to the previous studies [14, 15]. Given our claim that associations between reproductive patterns and dementia risk are subject to substantial confounding by social factors, these between-study differences are not surprising. Existing studies come from various different countries, and there is considerable international variation (and temporal variation) in the distributions of potential confounding factors and the relationships these factors have with both reproductive patterns and dementia [25, 26, 35–37]. For example, in Denmark, which throughout the study period had universal education and healthcare, easy and universal access to contraception, and the right to legal abortion, one might expect that a large proportion of higher-order births were planned. Having many children would therefore not necessarily be associated with lower education or socioeconomic status, unplanned pregnancies, and other factors associated both directly and indirectly (e.g. via poor mental health, including depression [38]) with later dementia risk, as might be more likely in countries lacking such social “safety net” features. We contend that adequate adjustment for confounding by social factors would substantially change the observed associations between reproductive patterns and dementia risk in a way specific to the nation, region, population group, and time period under study. Although our results suggest that pregnancy per se – that is, the normal physiological changes associated with pregnancy – is unlikely to be associated with increased risk of dementia in women once social factors specific to the population under study are taken into consideration, there is no question that some pregnancy complications are associated with an increased risk of dementia. Recent studies have linked both hypertensive disorders of pregnancy and stillbirth with increased risks of dementia, vascular dementia in particular [8–10]. Furthermore, a genetic polymorphism that increases the risk of Alzheimer’s disease may also be associated with gestational diabetes [39]. A history of pregnancy complications may therefore provide a useful indicator of increased future dementia risk, helping to identify groups of potentially at-risk women who might benefit from additional clinical attention. While we posit that residual confounding, rather than the biology of pregnancy, is largely responsible for observed associations between number of children, age at first birth, and dementia, we were unable to adjust our results for social and lifestyle factors, as only limited information on such factors is available in the Danish registers. For example, information on smoking, alcohol use and body mass index are only available for women in the first trimester of pregnancy. However, a recent study of reproductive history and risk of dementia adjusted for smoking and other lifestyle factors, including age, Townsend index (a measure of material deprivation), ethnicity, smoking, systolic blood pressure, body mass index, diabetes, total cholesterol, and use of antihypertensive medications and lipid-lowering medications [16]. Even after these adjustments, the study found similar associations between number of children and risk of dementia in men and women, indicating that the influence of confounding by lifestyle factors on the association of interest does not differ meaningfully for men and women. Consequently, it is unlikely that differential residual confounding by these factors is hiding sex-specific differences in association that suggest a pregnancy-related link in women. Registration of dementia diagnoses is probably incomplete, or at least delayed, because general practitioners, who do not report to the National Patient Register, handle a certain proportion of milder cases of dementia [40]. However, a study of dementia diagnoses registered in the National Patient Register found that $88\%$ of persons with a registered dementia diagnosis did in fact have dementia according to their medical records, and registered diagnoses of Alzheimer’s disease, vascular dementia, and other/unspecified dementia agreed with the diagnosis noted in the medical record for $97\%$, $96\%$, and $81\%$ of patients, respectively [41]. Since neither diagnosis nor registration of dementia was likely to have depended on number of children or age at first birth, any misclassification of dementia status was likely to have been non-differential. Our study’s greatest strength was the inclusion of men, which provided insight into the probable relative importance of social and biological factors in explaining the observed associations. Moreover, the use of the entire Danish population as the study cohort minimized selection bias and provided enormous statistical power. The use of registers ensured the absence of recall bias in the ascertainment of both exposure and outcome. ## Conclusions Associations between number of children, age at becoming a parent, and dementia risk were generally the same for both sexes, regardless of dementia subtype and timing of onset. Our findings argue against the suggestion that the normal physiological changes associated with pregnancy influence dementia risk in women. Even in the few instances where association magnitudes differed for men and women, the observed associations are more likely to be explained by lifestyle and socioeconomic factors associated with reproductive history than by the biology of pregnancy. ## Supplementary Information Additional file 1: Supplementary Table 1. Hazard ratios for dementia overall by number of children in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 2. Hazard ratios for dementia overall by number of children and timing of dementia onset, in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 3. Hazard ratios for dementia subtypes by number of children in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 4. Hazard ratios for dementia subtypes by number of children and timing of dementia onset in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 5. Hazard ratios for dementia overall by age at first becoming a parent in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 6. Hazard ratios for dementia overall by age at first becoming a parent and timing of dementia onset in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 7. Hazard ratios for dementia subtypes by age at first becoming a parent in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Table 8. Hazard ratios for dementia subtypes by age at first becoming a parent and timing of onset of dementia in a cohort of individuals ≥40 years in the period 1994-2017 in Denmark. Supplementary Figure 1. Associations between number of children and dementia, by timing of dementia onset, in a cohort of individuals ≥40 years old in the period 1994-2017 in Denmark. Supplementary Figure 2. Associations between number of children and dementia, by dementia subtype, in a cohort of individuals ≥40 years old in the period 1994-2017 in Denmark. Supplementary Figure 3. Associations between number of children and dementia, by dementia subtype and timing of dementia onset, in a cohort of individuals ≥40 years old in the period 1994-2017 in Denmark. Supplementary Figure 4. Associations between age at first birth and overall dementia, by timing of dementia onset, in a cohort of individuals ≥40 years old with ≥1 childbirths in the period 1994-2017 in Denmark. Supplementary Figure 5. Associations between age at first birth and overall dementia, by dementia subtype, in a cohort of individuals ≥40 years old with ≥1 childbirths in the period 1994-2017 in Denmark. Supplementary Figure 6. Associations between age at first birth and overall dementia, by dementia subtype and timing of dementia onset, in a cohort of individuals ≥40 years old with ≥1 childbirths in the period 1994-2017 in Denmark. ## References 1. 1.Global Health EstimatesDeaths by cause, age, sex, by country and by region, 2000–20162016GenevaWorld Health Organization2018. Deaths by cause, age, sex, by country and by region, 2000–2016 (2016) 2018 2. Salloway S, Sperling R, Fox NC. Two phase 3 trials of bapineuzumab in mild-to-moderate Alzheimer’s disease. N Engl J Med (2014) 370 322-333. DOI: 10.1056/NEJMoa1304839 3. Olanrewaju O, Clare L, Barnes L. A multimodal approach to dementia prevention: a report from the Cambridge Institute of Public Health. Alzheimer’s Dement Transl Res Clin Interv (2015) 1 151-156. DOI: 10.1016/j.trci.2015.08.003 4. Wimo A, Guerchet M, Ali GC. The worldwide costs of dementia 2015 and comparisons with 2010. Alzheimers Dement (2017) 13 1-7. DOI: 10.1016/j.jalz.2016.07.150 5. Carter CL, Resnick EM, Mallampalli M, Kalbarczyk A. Sex and gender differences in alzheimer’s disease: recommendations for future research. J Women’s Health (2012) 21 1018-1023. DOI: 10.1089/jwh.2012.3789 6. Nebel RA, Aggarwal NT, Barnes LL. Understanding the impact of sex and gender in Alzheimer’s disease: A call to action. Alzheimers Dement (2018) 14 1171-1183. DOI: 10.1016/j.jalz.2018.04.008 7. Ronquillo JG, Baer MR, Lester WT. Sex-specific patterns and differences in dementia and Alzheimer’s disease using informatics approaches. J Women Aging (2016) 28 403-411. DOI: 10.1080/08952841.2015.1018038 8. Andolf EG, Sydsjö GC, Bladh MK. Hypertensive disorders in pregnancy and later dementia: a Swedish National Register Study. Acta Obstet Gynecol Scand (2017) 96 464-471. DOI: 10.1111/aogs.13096 9. Basit S, Wohlfahrt J, Boyd HA. Pre-eclampsia and risk of dementia later in life: nationwide cohort study. BMJ (2018) 363 k4109. DOI: 10.1136/bmj.k4109 10. Basit S, Wohlfahrt J, Boyd HA. Pregnancy loss and risk of later dementia: a nationwide cohort study, Denmark, 1977–2017. Alzheimers Dement (N Y) (2019) 5 146-153. DOI: 10.1016/j.trci.2019.02.006 11. Ptok U, Barkow K, Heun R. Fertility and number of children in patients with Alzheimer’s disease. Arch Womens Ment Health (2002) 5 83-86. DOI: 10.1007/s00737-002-0142-6 12. Beeri MS, Rapp M, Schmeidler J. Number of children is associated with neuropathology of Alzheimer’s disease in women. Neurobiol Aging (2009) 30 1184-1191. DOI: 10.1016/j.neurobiolaging.2007.11.011 13. Colucci M, Cammarata S, Assini A. The number of pregnancies is a risk factor for Alzheimer’s disease. Eur J Neurol (2006) 13 1374-1377. DOI: 10.1111/j.1468-1331.2006.01520.x 14. Jang H, Bae JB, Dardiotis E. Differential effects of completed and incomplete pregnancies on the risk of Alzheimer disease. Neurology (2018) 91 e643-e651. DOI: 10.1212/WNL.0000000000006000 15. Bae JB, Lipnicki DM, Han JW. Does parity matter in women's risk of dementia? A COSMIC collaboration cohort study. BMC Med (2020) 18 210. DOI: 10.1186/s12916-020-01671-1 16. Gong J, Harris K, Peters SAE, Woodward M. Reproductive factors and the risk of incident dementia: a cohort study of UK Biobank participants. PLoS Med (2022) 19 e1003955. DOI: 10.1371/journal.pmed.1003955 17. Rocca WA, Van Duijn CM, Clayton D. Maternal age and alzheimer’s disease: a collaborative re-analysis of case-control studies. Int J Epidemiol (1991) 20 S21-S27. DOI: 10.1093/ije/20.Supplement_2.S21 18. Schmidt M, Pedersen L, Sørensen HT. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol (2014) 29 541-549. DOI: 10.1007/s10654-014-9930-3 19. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scand J Public Health (2011) 39 30-33. DOI: 10.1177/1403494811401482 20. Helweg-Larsen K. The Danish Register of Causes of Death. Scand J Public Health (2011) 39 26-29. DOI: 10.1177/1403494811399958 21. Kildemoes HW, Sørensen HT, Hallas J. The Danish National Prescription Registry. Scand J Public Health (2011) 39 38-41. DOI: 10.1177/1403494810394717 22. Lau B, Cole SR, Gange SJ. Competing risk regression models for epidemiologic data. Am J Epidemiol (2009) 170 244-256. DOI: 10.1093/aje/kwp107 23. Tough S, Benzies K, Fraser-Lee N, Newburn-Cook C. Factors influencing childbearing decisions and knowledge of perinatal risks among Canadian men and women. Matern Child Health J (2007) 11 189-198. DOI: 10.1007/s10995-006-0156-1 24. Huber S, Bookstein FL, Fieder M. Socioeconomic status, education, and reproduction in modern women: an evolutionary perspective. Am J Hum Biol (2010) 22 578-587. DOI: 10.1002/ajhb.21048 25. Mills M, Rindfuss RR, McDonald P. Why do people postpone parenthood? Reasons and social policy incentives. Hum Reprod Update (2011) 17 848-60. DOI: 10.1093/humupd/dmr026 26. Hammarberg K, Collins V, Holden C. Men's knowledge, attitudes and behaviours relating to fertility. Hum Reprod Update (2017) 23 458-480. DOI: 10.1093/humupd/dmx005 27. Beydoun MA, Beydoun HA, Gamaldo AA. Epidemiologic studies of modifiable factors associated with cognition and dementia: systematic review and meta-analysis. BMC Public Health (2014) 14 643. DOI: 10.1186/1471-2458-14-643 28. Seifan A, Schelke M, Obeng-Aduasare Y. Early life epidemiology of Alzheimer's disease–a critical review. Neuroepidemiology (2015) 45 237-254. DOI: 10.1159/000439568 29. Marden JR, Tchetgen EJ, Kawachi I. Contribution of socioeconomic status at 3 life-course periods to late-life memory function and decline: early and late predictors of dementia risk. Am J Epidemiol (2017) 186 805-814. DOI: 10.1093/aje/kwx155 30. Sommerlad A, Ruegger J, Singh-Manoux A. Marriage and risk of dementia: systematic review and meta-analysis of observational studies. J Neurol Neurosurg Psychiatry (2018) 89 231-238. DOI: 10.1136/jnnp-2017-316274 31. Weng PH, Chen JH, Chiou JM. The effect of lifestyle on late-life cognitive change under different socioeconomic status. PLoS One (2018) 13 e0197676. DOI: 10.1371/journal.pone.0197676 32. Kuiper JS, Zuidersma M, Oude Voshaar RC. Social relationships and risk of dementia: a systematic review and meta-analysis of longitudinal cohort studies. Ageing Res Rev (2015) 22 39-57. DOI: 10.1016/j.arr.2015.04.006 33. Khondoker M, Rafnsson SB, Morris S. Positive and negative experiences of social support and risk of dementia in later life: an investigation using the English Longitudinal Study of Ageing. J Alzheimer’s Dis (2017) 58 99-108. DOI: 10.3233/JAD-161160 34. Coyne CA, D’Onofrio BM. Some (but not much) progress toward understanding teenage childbearing: a review of research from the past decade. Adv Child Dev Behav (2012) 42 113-152. DOI: 10.1016/B978-0-12-394388-0.00004-6 35. Rizzi L, Rosset I, Roriz-Cruz M. Global epidemiology of dementia: Alzheimer's and vascular types. Biomed Res Int (2014) 2014 908915. DOI: 10.1155/2014/908915 36. Alladi S, Hachinski V. World dementia: One approach does not fit all. Neurology (2018) 91 264-270. DOI: 10.1212/WNL.0000000000005941 37. Sharp ES, Gatz M. Relationship between education and dementia: an updated systematic review. Alzheimer Dis Assoc Disord (2011) 25 289-304. DOI: 10.1097/WAD.0b013e318211c83c 38. Herd P, Higgins J, Sicinski K. The implications of unintended pregnancies for mental health in later life. Am J Public Health (2016) 106 421-429. DOI: 10.2105/AJPH.2015.302973 39. Vacínová G, Vejražková D, Lukášová P. Associations of polymorphisms in the candidate genes for Alzheimer's disease BIN1, CLU, CR1 and PICALM with gestational diabetes and impaired glucose tolerance. Mol Biol Rep (2017) 44 227-231. DOI: 10.1007/s11033-017-4100-9 40. Phung TKT, Waltoft BL, Kessing LV. Time trend in diagnosing dementia in secondary care. Dement Geriatr Cogn Disord (2010) 29 146-153. DOI: 10.1159/000269933 41. Phung TKT, Andersen BB, Høgh P, Kessing LV. Validity of dementia diagnoses in the Danish hospital registers. Dement Geriatr Cogn Disord (2007) 24 220-228. DOI: 10.1159/000107084
--- title: Prevalence of QT prolongation and its risk factors in patients with type 2 diabetes authors: - Khaled Aburisheh - Mohammad F. AlKheraiji - Saleh I. Alwalan - Arthur C. Isnani - Mohamed Rafiullah - Muhammad Mujammami - Assim A. Alfadda journal: BMC Endocrine Disorders year: 2023 pmcid: PMC9976503 doi: 10.1186/s12902-022-01235-9 license: CC BY 4.0 --- # Prevalence of QT prolongation and its risk factors in patients with type 2 diabetes ## Abstract ### Background QT prolongation increases cardiovascular mortality in diabetes. The risk factors for QT prolongation vary across different studies. There is no data on the QT prolongation in patients with diabetes from the Arab region, where diabetes is highly prevalent. Here we aimed to assess the prevalence of QT prolongation and its associated risk factors in patients with type 2 diabetes from Saudi Arabia. ### Method This was a retrospective, cross-sectional, hospital-based file review study. Data were collected from the medical records of patients with type 2 diabetes aged above 14 years and underwent ECG examination, and laboratory investigations were done within one month of ECG. ### Results The study included 782 patients with a prevalence of QTc prolongation of $13\%$. Patients with prolonged QTc interval were characterized by older age, higher BMI, longer diabetes duration, lower total cholesterol and LDL-C, and more diabetic nephropathy, hypertension, and CVD cases. They were also more in insulin treatment, antihypertensive medications, loop diuretics, and potassium-sparring diuretics. Logistic regression analysis revealed the odds of prolonged QTc interval increased significantly with CVD (OR = 1.761, $95\%$ CI:1.021–3.036, $$p \leq 0.042$$), and usage of loop diuretics (OR = 2.245, $95\%$ CI:1.023–4.923, $$p \leq 0.044$$) after adjusting for age, gender, and duration of diabetes. ### Conclusion The risk factors associated with QTc prolongation in patients with type 2 diabetes are CVD, and loop diuretics. Age, BMI, and diabetes duration were more in people with QTc prolongation, whereas total cholesterol and LDL-C levels were lower. More patients had diabetic nephropathy, hypertension, and CVD with prolonged QTc. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12902-022-01235-9. ## Background Diabetes-related mortality is mainly attributed to its cardiovascular complications [1]. Timely detection and treatment are essential in averting the high mortality associated with cardiovascular complications of diabetes. QT prolongation in the electrocardiogram (ECG) is one of the most commonly seen disorders of the heart. It may lead to potentially dangerous cardiac arrhythmia such as Torsade de Pointes. The QT interval represents the total duration of the ventricular depolarization and repolarization. Since the QT interval is influenced by the changes in the heart rate, a corrected QT interval (QTc) is used clinically [2]. It is usually found in patients with certain congenital conditions but can also be caused by several comorbid conditions and medications. Prolongation of QT interval is observed in many chronic inflammatory conditions, including diabetes [3]. Even though the effects of QT prolongation have been milder in the general population, it may increase the risk of mortality in patients with preexisting cardiac abnormalities. QTc prolongation independently predicted the cardiovascular mortality in people with diabetes [4]. Cardiovascular disease being one of the frequent complications associated with diabetes, the effect of QT prolongation in such patients need attention. Patients with diabetes were reported to have a higher prevalence of QTc prolongation [5]. Patients with diabetes have a 2–10 times higher risk of sudden cardiac death than general population. Hyperglycemia and chronic changes in myocardium are the main factors behind the increased prevalence of QTc prolongation in patients with diabetes [6]. Hyperglycemia and coronary heart disease (CHD) were found to be the strong predictors of high QTc interval. In a population-based cross-sectional study, a high prevalence of QTc was observed in patients with diabetes. The CHD was independently associated with QTc interval prolongation even after adjustment for age and sex [7]. A recent study found that female gender and treatment with insulin sensitizers were the independent contributors for QTc interval prolongation [8]. QTc was found to independently predict all-cause mortality in a prospective cohort of people with type 2 diabetes [9]. Current HbA1c, long-term postprandial hyperglycemia, and higher glycemic variability in postprandial glucose levels were found to be strong independent risk factors of QTc prolongation [10]. Apart from hyperglycemia and cardiovascular disease (CVD), other diabetic complications and hypoglycemia were also found to be associated with QTc prolongation in patients with diabetes. Baseline QTc interval was significantly associated with the QTc prolongation during severe hypoglycemia in patients with type 2 diabetes [11]. QTc interval prolongation observed during hypoglycemia was independent of serum potassium levels [12]. Other metabolic diseases such as obesity, dyslipidemia, and hypertension were also found to be risk factors for QTc prolongation in patients with diabetes [13]. Although the association of QTc prolongation with various risk factors and diabetic complications has been reported in many studies, the results have mainly been inconsistent across studies. In addition, to our knowledge, there are no studies on the QTc prolongation in patients with diabetes from the Arab region, where diabetes is highly prevalent. Therefore, the present study is undertaken to assess the prevalence of QT prolongation and its associated risk factors in patients with type 2 diabetes from Saudi Arabia. ## Methods This was a retrospective, cross-sectional, hospital-based file review study. The study population included patients with type 2 diabetes mellitus aged above 14 years who underwent ECG examination and laboratory investigations done within one month of the ECG recording. Patients with type 1 diabetes, pregnant women, patients using medications affecting QT interval (Supplementary Table 1), and patients with abnormal serum potassium levels were excluded (normal range 3.5–5 mmol/l). We did not exclude patients with atrial fibrillation or other arrhythmias. Patient records were screened from September 2013 to August 2015 using a simple random selection technique. The QT interval is defined as the time between the start of the Q wave and the end of the T wave in the heart’s electrical cycle. QTc was calculated by a 12-lead ECG machine automatically with Bazett formula (Esaote Biomedica Archimede Series 4200, Genova, Italy). The upper limit of normal was kept at 440 ms for males and 460 ms for females [14]. The study was conducted in accordance with the relevant guidelines and regulations and the study protocol was reviewed and approved by the Institutional Review Board, College of Medicine, King Saud University, Riyadh, Saudi Arabia. The data was collected through file review based on a pre-designed data collection questionnaire. Medical records of all the patients were systematically reviewed to collect age, sex, body weight, height, body mass index (BMI), systolic and diastolic blood pressure (BP), heart rate, diabetes duration, presence of comorbidities such as hypertension and dyslipidemia, and presence of diabetic complications namely retinopathy, nephropathy, peripheral neuropathy, and vasculopathy. The laboratory data included fasting and two-hour postprandial blood glucose, glycated hemoglobin (HbA1c), and lipids (total cholesterol, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and triglycerides). The details of medications prescribed to the patients were also collected. Diabetic nephropathy was considered if the patient had microalbuminuria (defined as a urinary albumin-to-creatinine level of 30–299 mg/g Cr) or macroalbuminuria (defined as a urinary albumin-to-creatinine level of more than 300 mg/g Cr). Diabetic neuropathy, vasculopathy, and retinopathy were considered positive if it was documented in the patients’ record. Retinopathy was reported when patients had either non-proliferative diabetic retinopathy or proliferative diabetic retinopathy with or without macular edema. The presence of at least one definite microaneurysm in any field of fundoscopy was considered as the diabetic retinopathy as assessed by an ophthalmologist. Neuropathy was considered based on physicians’ notes on the basis of neuropathic symptoms and signs or objectively abnormal results including insensitivity to a 10 g monofilament and abnormal vibration perception threshold based on the biothesiometer and without other significant disease. ## Statistical analysis The data were analyzed using the Statistical Package for Social Sciences (SPSS) version 23.0 (IBM-SPSS, Armonk, New York, USA). Data are expressed as mean ± standard deviation for continuous variables and as number and percentage for categorical variables. Data were tested for normality of distribution. Significant differences in the mean were tested by the independent samples t-test. The chi-square test was used to determine whether there are statistically significant differences in the expected and observed frequencies in QTc prolongation (yes or no) according to gender, presence of comorbid conditions, family history of CVD, types of arrhythmia, PR interval, ST-segment results, and medications. The relationship between two continuous variables was tested using the Pearson’s correlation coefficient test. A multivariate logistic regression model was constructed for significant variables in the univariate and bivariate analyses associated with QTc prolongation. In the logistic regression model, QTc prolongation (dichotomized as prolonged and not prolonged) was the dependent variable, and significant covariates including BMI, LDL, total cholesterol, nephropathy, CVD, hypertension, use of insulin, loop diuretics, potassium-sparing diuretics and use of antiplatelet medication were included in the model as predictors for QTc prolongation. Logistic regression analysis for risk factors of prolonged QTc interval was done with adjustment for age, gender, and duration of diabetes. A p-value of ≤0.05 was considered statistically significant. ## Results Data from 782 patients were included in the analysis, with 680 patients ($87.0\%$) showing normal QTc intervals and 102 ($13.0\%$) prolonged QTc intervals. The mean age of the study population was 56.4 years, and the mean duration of diabetes was 14.3 years. Retinopathy was the most frequent diabetic complication in our study population ($41.2\%$), followed by neuropathy, nephropathy, and CVD. The majority of the included patients had dyslipidemia and hypertension (74.7 and $66.9\%$, respectively). Ninety-five percent of the patients were taking antidiabetic medications, with metformin being the most frequent antidiabetic treatment. The other frequently used drugs were meant for treating hypertension and dyslipidemia, followed by antiplatelet medications. The complete demographic, clinical characteristics, and medications of the study population are shown in Table 1.Table 1Demographic and clinical characteristics of the study populationParametersMean ± SD orN (%)Total number of patients782Age (years)56.4+ ± 11.8Sex Male399 ($51.0\%$) Female383 ($49.0\%$)Height (m)161.2 ± 9.2Body weight (kg)82.9 ± 15.9BMI (kg/m2)31.9 ± 5.9Duration of diabetes (years)14.3 ± 8.7HbA1c (%)8.7 ± 1.8Systolic BP (mmHg)134 ± 16.9Diastolic BP (mmHg)74.4 ± 10.2Heart rate (bpm)83 ± 13.3Fasting blood glucose (mm/L)9.2 ± 3.5Postprandial blood glucose (mmol/L)13.4 ± 5.1Total cholesterol (mmol/L)4.4 ± 1.0LDL-C (mmol/L)2.5 ± 0.9HDL-C (mmol/L)1.23 ± 0.5Triglycerides (mg/L)1.6 ± 0.9Retinopathy (%)322 ($41.2\%$)Neuropathy (%)273 ($34.9\%$)Nephropathy159 ($20.3\%$)CVA15 ($1.9\%$)CVD121 ($15.5\%$PVD27 ($3.5\%$)Hypertension523 ($66.9\%$)Hyperlipidemia584 ($74.7\%$)Sodium138.4 ± 3.1Potassium4.2 ± 0.4Antidiabetic medications743 ($95\%$) Metformin623 ($79.7\%$) Thiazolidinediones124 ($15.9\%$) Sulfonylureas279 ($35.7\%$) Meglitinides36 ($4.6\%$) DDP-4 inhibitors80 ($10.2\%$) Alpha-glucosidase inhibitors39 ($5\%$) Insulin322 ($41.2\%$)Antihypertension medications448 ($57.3\%$) ACE inhibitors114 ($14.6\%$) ARBs254 ($32.5\%$) Calcium channel blockers134 ($17.1\%$) Beta-blockers126 ($16.1\%$) Thiazide diuretics107 ($13.7\%$) Loop diuretics35 ($4.5\%$) Potassium-sparing diuretics4 ($0.5\%$)Lipid lowering medications512 ($65.5\%$) Statins494 ($63.2\%$) Fibrates25 ($3.2\%$) Ezetimibe5 ($0.6\%$) Anti-platelet medications501 ($64.1\%$) Prolonged QTc102 ($13\%$)ACE inhibitors Angiotensin-converting enzyme inhibitors, ARB Angiotensin receptor blocker, BMI Body mass index, BP Blood pressure, CVA Cerebrovascular accident, CVD Cardiovascular disease, DPP-4 inhibitors Dipeptidyl-peptidase 4 inhibitors, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, PVD Peripheral vascular disease. There was no significant difference in the gender distribution of the study population. The prevalence of QTc prolongation was not significantly different across gender ($$p \leq 0.135$$). Patients with prolonged QTc interval were significantly older ($$p \leq 0.001$$), had higher BMI ($$p \leq 0.030$$), longer diabetes duration ($$p \leq 0.050$$), and lower total cholesterol and LDL cholesterol levels ($$p \leq 0.019$$ and $$p \leq 0.017$$, respectively) compared to patients who had normal QTc interval. Diabetic nephropathy and CVD were the only diabetic complications that were significantly higher in patients with prolonged QTc intervals. Hypertension was more prevalent in patients with QTc prolongation. Comparing the diabetic medications, patients with prolonged QTc were significantly more insulin users. Moreover, antihypertensive and antiplatelet medications and diuretics (loop diuretics and potassium-sparing diuretics) were used more in patients with QTc prolongation (Table 2).Table 2Comparison of the patient characteristics between normal and prolonged QTc categoriesParametersQTc prolongedN = 102QTc normalN = 680p valuesAge (years)60.0 ± 11.055.8 ± 11.90.001Sex Male45 ($44.1\%$)354 ($52.1\%$)0.135 Female57 ($55.9\%$)326 ($47.9\%$)Height (m)161.1 ± 8.6161.3 ± 9.30.831Body weight (kg)85.5 ± 13.982.5 ± 16.10.070BMI (kg/m2)33.1 ± 5.431.8 ± 6.00.030Duration of diabetes (years)15.9 ± 9.014.1 ± 8.60.050HbA1c (%)8.6 ± 1.78.7 ± 1.80.565Systolic BP (mmHg)135.1 ± 17.4133.9 ± 16.90.509Diastolic BP (mmHg)73.6 ± 9.874.5 ± 10.30.434Heart rate (bpm)83.4 ± 14.982.9 ± 13.10.723Fasting blood glucose (mm/L)9.1 ± 3.59.2 ± 3.50.790Postprandial blood glucose (mmol/L)13.5 ± 4.413.4 ± 5.10.922Total cholesterol (mmol/L)4.2 ± 1.04.4 ± 1.00.019LDL-C (mmol/L)2.3 ± 0.82.5 ± 0.90.017HDL-C (mmol/L)1.23 ± 0.61.23 ± 0.50.928Triglycerides (mg/L)1.6 ± 0.81.7 ± 0.90.590Retinopathy (%)45 ($44.1\%$)277 ($40.7\%$)0.517Neuropathy (%)37 ($36.3\%$)236 ($34.7\%$)0.757Nephropathy29 ($28.4\%$)130 ($19.1\%$)0.029CVA2 ($2.0\%$)13 ($1.9\%$)0.973CVD29 ($28.4\%$)92 ($13.5\%$)< 0.001PVD2 ($2.0\%$)25 ($3.7\%$)0.376Hypertension78 ($76.5\%$)445 ($65.4\%$)0.027Hyperlipidemia81 ($79.4\%$)503 ($74.0\%$)0.239Sodium138.4 ± 3.1138.4 ± 3.10.799Potassium4.2 ± 0.44.2 ± 0.40.780Diabetes medications98 ($99.0\%$)645 ($99.8\%$)0.126 Sensitizers78 ($76.5\%$)545 ($80.1\%$)0.390 Thiazolidinediones12 ($11.8\%$)112 ($16.5\%$)0.225 Secretagogues32 ($21.4\%$)247 ($36.3\%$)0.330 Meglitinides3 ($2.9\%$)33 ($4.9\%$)0.390 DDP4 inhibitors17 ($13.7\%$)66 ($9.7\%$)0.212 Alphaglucosidase inhibitors7 ($6.9\%$)32 ($4.7\%$)0.351 Insulin53 ($52.0\%$)269 ($39.6\%$)0.018Anti-hypertension medications72 ($70.6\%$)376 ($55.3\%$)0.004 ACE inhibitors19 ($18.6\%$)959 ($14.0\%$)0.214 ARBs33 ($32.4\%$)221 ($32.5\%$)0.976 Calcium channel blockers24 ($23.5\%$)110 ($16.2\%$)0.066 Beta-blockers22 ($21.6\%$)104 ($15.3\%$)0.108 Thiazide diuretics15 ($14.7\%$)92 ($13.5\%$)0.747 Loop diuretics14 ($13.7\%$)21 ($3.1\%$)< 0.001* Potassium-sparing diuretics2 ($2.0\%$)2 ($0.3\%$)0.028Lipid lowering medications71 ($69.6\%$)441 ($64.9\%$)0.346 Statins71 ($69.6\%$)423 ($62.2\%$)0.148 Fibrates3 ($2.9\%$)22 ($3.2\%$)0.875 Ezetimibe1 ($1.0\%$)4 ($0.6\%$)0.643 Anti-platelet medication75 ($73.5\%$)426 ($62.6\%$)0.033ACE inhibitors Angiotensin-converting enzyme inhibitors, ARB Angiotensin receptor blocker, BMI Body mass index, BP Blood pressure, CVA Cerebrovascular accident, CVD Cardiovascular disease, DPP-4 inhibitors Dipeptidyl-peptidase 4 inhibitors, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, PVD Peripheral vascular diseasep ≤ 0.05 A logistic regression model was constructed with QTc prolongation as the dependent variable and significant factors including BMI, total cholesterol, LDL-C, nephropathy, CVD, hypertension, and the use of insulin, loop diuretics, potassium-sparing diuretics, and antiplatelet medication as independent factors with adjustment for age, gender, and duration of diabetes. The logistic regression analysis showed the following significant predictors for prolonged QTc interval; the odds of prolonged QTc interval increased significantly with CVD (OR = 1.761, $95\%$CI:1.021–3.036, $$p \leq 0.042$$), and usage of loop diuretics (OR = 2.245, $95\%$CI:1.023–4.923, $$p \leq 0.044$$) (Table 3).Table 3Logistic regression analysis for predictors of prolonged QTc intervalPredictorsBOR (ExpB)$95\%$ CIP value$Nephropathy0.2511.2850.766–2.1580.342CVD0.5561.7611.021–3.0360.042Hypertension−0.1900.8270.482–1.4190.490BMI0.0251.0250.988–1.0640.188LDL-C0.0161.0160.656–1.5740.944Total cholesterol−0.1630.8490.581–1.2410.398Insulin0.2711.3120.590–2.9170.506Anti-platelet medications0.2541.2890.789–2.1080.311Loop diuretics0.8092.2451.023–4.9230.044Potassium-sparing diuretics1.0472.8490.345–23.5180.331BMI Body mass index, CVD Cardiovascular disease, LDL-C Low-density lipoprotein cholesterol$Adjusted for age, gender, and duration of diabetesp ≤ 0.05 ## Discussion There is wide variability in the reported prevalence of QTc prolongation in the diabetic population ranging from 11.3 to $59.3\%$ [15, 16]. It might be a consequence of the heterogeneity in the definition of prolonged QTc and the differences in the mean age of the populations. Moreover, inaccurate identification of the beginning and the end of the QT interval by the observer or the software in the automatic analysis of ECG tracings could also be behind this variation. The prevalence of the prolonged QTc in our study was $13\%$. Even though a higher proportion of women showed QTc prolongation, it was not statistically significant. Female gender was reported to be an independent risk factor for QTc prolongation in patients with diabetes in a cross-sectional cohort study where a different criterion for prolonged QTc in females was not used [17]. Despite a higher cut-off to define the QTc prolongation in females in our study, the prevalence of QTc prolongation in females was only slightly higher. The results show that patients with QTc prolongation were older in our study. Similar results were seen in the diabetic and general population [16, 17]. Older people are more likely to have worsening diabetes and complications, especially CHD, which could predispose them to alterations in the QTc interval. Patients with increased QTc interval were found to have a higher BMI This is in confirmation with a previously reported study [17]. BMI is a known risk factor for cardiovascular diseases and is also associated with a poor prognosis of diabetes. However, in another study, there was no difference in the BMI between patients with prolonged QTc and normal QTc intervals [7]. Notably, the duration of diabetes was similar between the patient groups in these studies, whereas in our study, the patients with prolonged QTc had a slightly longer duration. The effect of glycemia on the QT prolongation has not been consistent. Ninkovic et al. reported that most of the glycemic parameters like fasting blood glucose, postprandial blood glucose, HbA1c, mean blood glucose, and mean glucose excursions were significantly higher in individuals with QTc prolongation [17]. In addition, mean blood glucose was also found to be predictive of prolonged QTc interval and QTc dispersion. However, in another study, no differences were found in the HbA1c between normal and prolonged QTc interval groups [7]. In Diabetes Cohort Study, the baseline blood glucose and HbA1c levels were similar between patients with prolonged QTc interval and normal QTc interval [9]. The present study found no difference in fasting and postprandial blood glucose and HbA1c levels between the study groups. The most common comorbidities of diabetes like hypertension and dyslipidemia were associated with a higher prevalence of QT prolongation in individuals with diabetes [7, 13]. Both systolic and diastolic pressures were found to be significantly higher in people with QTc prolongation. Our results show that cases of hypertension were significantly higher in individuals with prolonged QTc. The systolic and diastolic blood pressures were not different between the two groups. As the patients with QTc prolongation had significantly higher usage of antihypertensive medications, the observation of comparable blood pressure levels between the groups is understandable. Many studies suggest that the association of hypertension with left ventricular hypertrophy induced changes in the myocardium and sympathovagal imbalance as likely mechanisms behind the relationship of hypertension and QTc prolongation [7, 18]. Previous studies reported either higher cholesterol levels in patients with QTc prolongation or no difference [9, 13]. Interestingly, our results show that patients with QTc prolongation were characterized by lower total cholesterol and LDL-C levels. With the high prevalence of CVD in patients with prolonged QTc, those patients might likely have been on intensive statin therapy. However, it is not clear whether this would have been large enough to produce a significant difference in the lipid profile. Diabetic nephropathy and CVD are the complications that were present significantly higher in patients with prolonged QTc intervals in our study. Previous studies reported more patients with prolongation of QTc intervals had diabetic retinopathy, neuropathy, and nephropathy [17, 19]. The presence of microalbuminuria is known to prolong QTc interval in individuals with type 2 diabetes. However, the pathophysiology behind the coexistence of both conditions is not understood. Independent association of diabetic complications such as neuropathy and nephropathy with QTc prolongation was reported previously [19]. But, in our study, we found CVD, the only diabetic complication associated independently with prolonged QTc interval. Coronary ischemia, infarction, and other cardiac abnormalities can induce QTc prolongation [20]. Ischemia causes structural damage of the heart and is associated with hyperactivity of the sympathetic system, leading to an increase in the ventricular depolarization duration represented by the QT interval in the ECG [17]. In our study population, insulin treatment was higher among patients with QTc prolongation. A similar observation was found by Kobayashi et al. [ 19]. The influence of insulin on the QTc interval can be in many ways. Patients on insulin treatment are likely to be older with a longer duration of diabetes and uncontrolled diabetes. These factors are known to be associated with the prolongation of QTc. In addition, insulin is likely to cause QTc prolongation by altering the sympathetic activity and serum potassium concentration [21]. Insulin-induced hypoglycemia can also cause a prolongation of QTc intervals [22]. Despite being involved in multiple mechanisms, the use of insulin did not significantly increase the risk of QTc interval prolongation. We found loop diuretics as independent predictors of QTc prolongation. These drugs were reported to be predictors of QTc prolongation previously [23]. Usually, loop diuretics do not cause QTc interval prolongation, and their effect on QTc prolongation was independent of electrolyte changes [24]. But since the patients treated with loop diuretics were likely to have hypertension and cardiovascular disease, more patients treated with loop diuretics likely had QTc prolongation. Potassium-sparring diuretics were linked to QTc interval shortening, and it was dependent on the serum electrolytes levels [24]. Our study population had only a few patients using potassium-sparing diuretics, and we excluded patients with abnormal serum potassium levels. Therefore, the results need to be interpreted cautiously. Intake of antiplatelet medications was seen higher in patients with prolonged QTc intervals. Aspirin and clopidogrel were the antiplatelet medications used by our study patients. These drugs are usually prescribed to patients with diabetes who have additional risk factors for cardiovascular diseases or as a secondary prevention strategy in patients with preexisting cardiovascular disease. Our study population has significantly more patients having cardiovascular disease and prolonged QTc. Therefore, people with QTc prolongation are more likely to have had antiplatelet medications. In addition, clopidogrel is associated with prolongation of QTc interval [25]. Since fewer patients were on clopidogrel in our study, its influence on the overall outcome is likely to be negligible. The present study has identified the risk factors associated with QTc prolongation in patients with type 2 diabetes. CVD, and diuretic medications (loop diuretics) were significantly associated with prolonged QTc interval after adjusting age, gender, and duration of diabetes. The patient population with prolonged QTc interval was characterized by older age, higher BMI, longer duration of diabetes, lower total cholesterol and LDL-C, and more diabetic nephropathy, hypertension, and CVD cases. The number of patients with prolonged QTc was also more under insulin treatment, antihypertensive medications, and diuretics (loop diuretics). While most of these factors could be explained for their link with QTc prolongation, lower total cholesterol and LDL-C levels need further investigation. It is not clear whether it was an outcome of intensive statin therapy in patients with QTc prolongation. Our study has several limitations. This is a retrospective cross-sectional study; therefore, the findings are not suggestive of temporal or causal relationships between risk factors and QTc interval prolongation. Since our study is based on file documentation review, it could have led to biased data analysis. Further, we did not analyze the other QT parameters, such as QT dispersion which has been shown to have a predictive role in all-cause and cardiovascular mortality in type 2 diabetes. ## Conclusion The risk factors associated with QTc prolongation in patients with type 2 diabetes were found to be CVD, and diuretic medications (loop diuretics). Age, BMI, and duration of diabetes were more in people with QTc prolongation, whereas total cholesterol and LDL-C levels were lower. More patients had diabetic nephropathy, hypertension, and CVD in individuals with prolonged QTc. ## Supplementary Information Additional file 1: Supplementary Table 1. List of excluded medications known to affect the QTc interval. ## Authors’ information Not applicable. ## References 1. Li S, Wang J, Zhang B, Li X, Liu Y. Diabetes Mellitus and Cause-Specific Mortality: A Population-Based Study. Diabetes Metab J [Internet] (2019.0) 43 319-341. DOI: 10.4093/dmj.2018.0060 2. 2.Cubeddu LX. QT Prolongation and Fatal Arrhythmias: A Review of Clinical Implications and Effects of Drugs. Am J Ther [Internet]. 2003:10 Available from: https://journals.lww.com/americantherapeutics/Fulltext/2003/11000/QT_Prolongation_and_Fatal_Arrhythmias__A_Review_of.13.aspx. 3. Lazzerini PE, Capecchi PL, Laghi-Pasini F. Assessing QT interval in patients with autoimmune chronic inflammatory diseases: perils and pitfalls. Lupus Sci Med (2016.0) 3 e000189. DOI: 10.1136/lupus-2016-000189 4. Naas AA, Davidson NC, Thompson C, Cummings F, Ogston SA, Jung RT. QT and QTc dispersion are accurate predictors of cardiac death in newly diagnosed non-insulin dependent diabetes: cohort study. BMJ (1998.0) 316 745-746. DOI: 10.1136/bmj.316.7133.745 5. Lu Z, Lense L, Sharma M, Shah A, Luu Y, Cardinal L. Prevalence of QT prolongation and associated LVEF changes in diabetic patients over a four-year retrospective time period. J community Hosp Intern Med Perspect [Internet] (2017.0) 7 87-94. DOI: 10.1080/20009666.2017.1320203 6. 6.Taubel J, Pimenta D. Considering the risk of QTc prolongation in patients with diabetes mellitus. e-Journal Cardiol. Pract. 2022;22, No8, 06 Apr 2022. 7. Veglio M, Bruno G, Borra M, Macchia G, Bargero G, D’errico N. Prevalence of increased QT interval duration and dispersion in type 2 diabetic patients and its relationship with coronary heart disease: a population-based cohort. J Intern Med (2002.0) 251 317-324. DOI: 10.1046/j.1365-2796.2002.00955.x 8. Lu C-F, Ge X-Q, Wang Y, Su J-B, Wang X-Q, Zhang D-M. The relationship between adenosine deaminase and heart rate-corrected QT interval in type 2 diabetic patients. Endocr Connect (2021.0) 10 894-901. DOI: 10.1530/EC-21-0199 9. Cox AJ, Azeem A, Yeboah J, Soliman EZ, Aggarwal SR, Bertoni AG. Heart rate-corrected QT interval is an independent predictor of all-cause and cardiovascular mortality in individuals with type 2 diabetes: the Diabetes Heart Study. Diabetes Care (2014.0) 37 1454-1461. DOI: 10.2337/dc13-1257 10. Su J-B, Yang X-H, Zhang X-L, Cai H-L, Huang H-Y, Zhao L-H. The association of long-term glycaemic variability versus sustained chronic hyperglycaemia with heart rate-corrected QT interval in patients with type 2 diabetes. PLoS One (2017.0) 12 e0183055. DOI: 10.1371/journal.pone.0183055 11. Cha SA, Yun JS, Lim TS, Kang YG, Lee KM, Song KH. Baseline-Corrected QT (QTc) Interval Is Associated with Prolongation of QTc during Severe Hypoglycemia in Patients with Type 2 Diabetes Mellitus. Diabetes Metab J (2016.0) 40 463-472. DOI: 10.4093/dmj.2016.40.6.463 12. Beom JW, Kim JM, Chung EJ, Kim JY, Ko SY, Na SD. Corrected QT Interval Prolongation during Severe Hypoglycemia without Hypokalemia in Patients with Type 2 Diabetes. Diabetes Metab J (2013.0) 37 190-195. DOI: 10.4093/dmj.2013.37.3.190 13. Li X, Ren H, Xu Z, Liu Y, Yang X, Liu J. Prevalence and Risk Factors of Prolonged QTc Interval among Chinese Patients with Type 2 Diabetes. Exp Diabetes Res (2012.0) 2012 1-6. DOI: 10.1155/2012/234084 14. McGuinness S, McKee A, Sidebotham D, Sidebotham D, Mckee A, Gillham M. Chapter 8 - Monitoring [Internet]. Levy JHBT-CCC, editors (2007.0) 120-137 15. Vasheghani M, Sarvghadi F, Beyranvand MR, Emami H. The relationship between QT interval indices with cardiac autonomic neuropathy in diabetic patients: a case control study. Diabetol Metab Syndr (2020.0) 12 102. DOI: 10.1186/s13098-020-00609-0 16. Ma Q, Li Z, Guo X, Guo L, Yu S, Yang H. Prevalence and risk factors of prolonged corrected QT interval in general Chinese population. BMC Cardiovasc Disord [Internet] (2019.0) 19 276. DOI: 10.1186/s12872-019-1244-7 17. Ninkovic VM, Ninkovic SM, Miloradovic V, Stanojevic D, Babic M, Giga V. Prevalence and risk factors for prolonged QT interval and QT dispersion in patients with type 2 diabetes. Acta Diabetol (2016.0) 53 737-744. DOI: 10.1007/s00592-016-0864-y 18. Buja G, Miorelli M, Turrini P, Melacini P, Nava A. Comparison of QT dispersion in hypertrophic cardiomyopathy between patients with and without ventricular arrhythmias and sudden death. Am J Cardiol (1993.0) 72 973-976. DOI: 10.1016/0002-9149(93)91118-2 19. Kobayashi S, Nagao M, Asai A, Fukuda I, Oikawa S, Sugihara H. Severity and multiplicity of microvascular complications are associated with QT interval prolongation in patients with type 2 diabetes. J Diabetes Investig (2018.0) 9 946-951. DOI: 10.1111/jdi.12772 20. Rana BS, Band MM, Ogston S, Morris AD, Pringle SD, Struthers AD. Relation of QT interval dispersion to the number of different cardiac abnormalities in diabetes mellitus. Am J Cardiol (2002.0) 90 483-487. DOI: 10.1016/S0002-9149(02)02518-3 21. Gastaldelli A, Emdin M, Conforti F, Camastra S, Ferrannini E. Insulin prolongs the QTc interval in humans. Am J Physiol Regul Integr Comp Physiol (2000.0) 279 R2022-R2025. DOI: 10.1152/ajpregu.2000.279.6.R2022 22. Kacheva S, Karges B, Göller K, Marx N, Mischke K, Karges W. QT prolongation caused by insulin-induced hypoglycaemia - An interventional study in 119 individuals. Diabetes Res Clin Pract (2017.0) 123 165-172. DOI: 10.1016/j.diabres.2016.11.021 23. Heemskerk CPM, Pereboom M, van Stralen K, Berger FA, van den Bemt PMLA, Kuijper AFM. Risk factors for QTc interval prolongation. Eur J Clin Pharmacol [Internet] (2018.0) 74 183-191. DOI: 10.1007/s00228-017-2381-5 24. Snitker S, Doerfler RM, Soliman EZ, Deo R, St Peter WL, Kramlik S. Association of QT-Prolonging Medication Use in CKD with Electrocardiographic Manifestations. Clin J Am Soc Nephrol (2017.0) 12 1409-1417. DOI: 10.2215/CJN.12991216 25. 25.Fernandes FM, Silva EP, Martins RR, Oliveira AG. QTc interval prolongation in critically ill patients: Prevalence, risk factors and associated medications. PLoS One [Internet]. 2018;13:e0199028 Available from: https://pubmed.ncbi.nlm.nih.gov/29898002.
--- title: 'Racial/ethnic differences in social determinants of health and health outcomes among adolescents and youth ages 10–24 years old: a scoping review' authors: - Patricia Monroe - Jennifer A. Campbell - Melissa Harris - Leonard E. Egede journal: BMC Public Health year: 2023 pmcid: PMC9976510 doi: 10.1186/s12889-023-15274-x license: CC BY 4.0 --- # Racial/ethnic differences in social determinants of health and health outcomes among adolescents and youth ages 10–24 years old: a scoping review ## Abstract ### Introduction With the recent emergence of the Healthy People 2030 goals there is a need to understand the role of SDOH on health inequalities from an upstream perspective. This review summarizes the recent body of evidence on the impact of SDOH across adolescence and youth health outcomes by race/ethnicity using the Health People 2030 Framework. ### Methods A systematic, reproducible search was performed using PubMed, Academic Search Premier, PsychInfo, and ERIC. A total of 2078 articles were screened for inclusion. A total of 263 articles met inclusion criteria, resulting in 29 articles included for final synthesis. ### Results Across the 29 articles, 11 were cross-sectional, 16 were cohort, and 2 were experimental. Across SDOH categories (economic stability, education access and quality, health care access and quality, neighborhood and built environment, and social and community context), 1 study examined self-efficacy, 6 educational attainment, 10 behavior, 5 smoking, 11 alcohol use, 10 substance use, and 1 quality of life. The majority of outcomes represented in this search included health behaviors such as health risk behavior, smoking, alcohol use, and substance use. Across the 29 articles identified, significant differences existed across outcomes by race/ethnicity across SDOH factors, however magnitude of differences varied by SDOH category. ### Discussion SDOH differentially affect adolescents and youth across race/ethnicity. The lived adverse experiences, along with structural racism, increase the likelihood of adolescents and youth engaging in risky health behaviors and negatively influencing health outcomes during adolescence and youth. Research, public health initiatives, and policies integrating SDOH into interventions at early stage of life are needed to effectively reduce social and health inequalities at a population level. ## Introduction The impact of social determinants of health (SDOH) on the overall well-being of individuals and population health has been well established [1–4]. The World Health Organization describes SDOH as multiple factors impacting the health of individuals, including “structural determinants and conditions of daily living,” unequally around the world [5], which over the life course, contribute to the development of chronic illnesses, often among those most vulnerable [5–7]. The Healthy People 2020 Framework for SDOH established a foundation for examining the role of SDOH across health and well-being, fostering a substantial gain in knowledge regarding the contribution of SDOH on health and well-being of adults, including racial/ethnic disparities in SDOH [2, 4, 8–10]. With the recent emergence of the Healthy People 2030 goals to “create social, physical, and economic environments that promote attaining the full potential for health and well-being for all” there is a need to understand the role of SDOH on health inequalities from an upstream perspective, specifically the presence of SDOH in early life [11], that may contribute to adult morbidity. Adolescence as a phase of development begins at age 10 and transitions to youth at age 19 through age 24, according to the WHO [12]. This period of development presents a unique time of transition, with increased awareness of social processes and vital brain development [12–14]. This transition is marked by development of new behaviors, influenced by societal contexts, resulting in positive or adverse health outcomes and ultimately, inequalities as adolescents age and transition into adulthood [12, 14, 15]. Evidence shows that SDOH and social risk factors such as access to care, health insurance, food security, access to transportation, neighborhood deprivation, and economic disadvantage have been found to negatively impact outcomes for racial/ethnic minority adolescents [16–22], lending to racial/ethnic health disparities across health indices. For example, higher rates of obesity and behavior problems, poorer cardiovascular and oral health, and lower rates of health-related quality of life [16–19, 21, 22]. While the existing body of evidence provides identification of key SDOH factors that lend to health disparities across adolescent and youth outcomes, systematic evaluation of the existing evidence for how SDOH impact the health, well-being, and the development of adolescents and youth across racial/ethnic groups has been limited [15, 23]. To address SDOH factors that lend to health inequalities for adolescence and youth, an in depth understanding of how SDOH contribute to health outcomes in adolescence and youth across racial/ethnic groups must first be established. This scoping review therefore aims to evaluate and synthesize the existing evidence for the role of SDOH on adolescent and youth health outcomes within the United States, and to summarize similarities and differences found across race/ethnicity. Specifically, using the Healthy People 2030 Framework as a guiding framework, the literature was searched matching terms to the SDOH domains of economic stability, education access and quality, health care access and quality, neighborhood and built environment, and social and community context, to evaluate and summarize existing evidence for the role of SDOH on adolescent and youth health and well-being across a broad spectrum of health indicators. ## Information sources, eligibility criteria, and search PRISMA Guidelines were used for identifying, screening, and study selection for final synthesis. No protocol was prepared for this review. Articles were chosen based on eligibility criteria listed below, established a priori by the authors. A reproducible search strategy was used to identify articles investigating the impact of SDOH on the health outcomes of adolescents and youth 10–24 years of age, based on the WHO definition [12]. Four different databases were utilized to ensure the inclusion of a robust set of articles. Articles published between 2014 and 2021 were searched using PubMed, Academic Search Premier, PsychInfo, and ERIC. This date range was chosen a priori to maximize applicability and relevance of evidence. Medical Subject Heading (MeSH) terms representing SDOH based on the Healthy People 2030 Framework were used, see Table 1, along with additional inclusion criteria (listed below, and Table 2).Table 1Search termsMeSH Terms – SDOHMeSH Terms – OutcomesMeSH Terms – CharacteristicsSocial determinants of healthHealth social determinantHealth social determinantsSocioeconomic factorSocioeconomic statusSocioeconomic gradientSocioeconomic positionLow incomePovertyTraumaPsychological traumaStressSocial supportSocial disparitySocial environmentSocial exclusionSocial factorSocial gradientSocial positionSocial cohesionSelf-efficacyEducational achievementPsychological resilienceBehavior, health riskSmoking behaviorsAlcohol drinkingSubstance abuse detectionHealth related Quality of lifeRace factorsMinority healthHealth status disparitiesAdolescentaYoung adultaasearched as Boolean/phraseTable 2Inclusion, exclusion criteriaInclusionExclusion• At least one or more outcomes must be included as an outcome evaluated in study:◦ Self-efficacy◦ Educational attainment◦ Psychological resilience◦ Behavior, health risk◦ Smoking behavior◦ Alcohol use◦ Substance use◦ Quality of life• Racial/ethnic differences in outcomes must be presented• Type of study:◦ Cross-sectional◦ Cohort◦ Clinical trial◦ Quasi-experimental◦ Pre-post• English language• Disease specific focus• Population age:◦ Younger than 10 years◦ Older than 25 years• Type of study:◦ Systematic reviews◦ Meta-analysis◦ Scoping review• Protocol, design, or rationale papers Eligible articles were included based on the following inclusion criteria: 1) published in English, 2) based in the United States, 3) study design: cross-sectional, cohort, clinical trial, quasi-experimental, or pre-post study design, 4) outcomes demonstrated across at least two racial/ethnic groups showing outcomes were examined by race/ethnicity. Additionally, one or more of the following outcomes had to be included: 1) self-efficacy, 2) educational attainment, 3) psychological resilience, 4) health risk behavior, 5) smoking behavior, 6) alcohol use, 7) substance use, 8) quality of life. Health risk behavior, including high risk sexual behavior, delinquent behavior, and health promoting behavior including physical activity and dietary intake. Outcomes were chosen based on the evidence of health behaviors during adolescence and youth impacting morbidity and mortality in adulthood, as well as the potential impact of moderators, such as self-efficacy, educational attainment, and psychological resilience, throughout the life course on health outcomes [24–26]. ## Study selection and data collection Study selection was based on an initial title and abstract review by PM and MH. Studies were evaluated for inclusion using a checklist that included eligibility criteria. Studies not meeting eligibility criteria were excluded. After the title and abstract review, full text articles that met initial inclusion criteria were included for full text synthesis. Initial and full text review of studies were done separately with oversight provided by JAC and LEE. The checklist ensured consistent decision-making processes were followed for each paper reviewed. After full text synthesis by PM, MH and JAC, articles not meeting inclusion criteria were excluded with reasons. Please see Fig. 1 for PRISMA guidelines with details of studies excluded and retained at each phase. The articles included for data extraction are shown in Table 3. Data extraction included the study design, SDOH category, and outcomes assessed. Data quality was assessed using the JBI critical appraisal checklist [27]. JBI provides checklists by study design. This review used the appropriate checklist for the appropriate design in each paper. Final article decisions were made by PM, JAC, and LEE based on the checklists and included articles meeting all criteria to ensure quality across articles summarized in this review. Fig. 1PRISMA Flow DiagramTable 3Study design, race/ethnicity, and outcomesAuthor/YearStudy DesignRace/EthnicityOutcomesCross-sectionalCohortExperimentalAfrican AmericanHispanicNon-Hispanic WhiteAsian/Pacific IslanderAmerican IndianOther race/ethnicityMultiracialSelf-EfficacyEducational AttainmentPsychological ResilienceHealth Risk BehaviorSmokingAlcoholSubstance UseQOLAbram et al. 2017 [28]XXXXXXAssini-Meytin et al. 2019 [29]XXXXXXXBaiden et al. 2019 [30]XXXXXXXXBares et al. 2019 [31]XXXXXBenner & Wang, 2014 [32]XXXXXXXBenner & Wang, 2015 [33]XXXXXXXBohnert et al. 2017 [34]XXXXBrooks-Russell et al. 2019 [35]XXXXXXXXCage et al. 2018 [36]XXXXXXCamenga et al. 2018 [37]XXXXChambers et al. 2018 [38]XXXXXChampion et al. 2016 [39]XXXXXXXChong et al. 2019 [40]XXXXClark et al. 2015 [41]XXXXCombs et al. 2018 [42]XXXXXXXDocherty et al. 2018 [43]XXXXGerard & Booth 2015 [44]XXXXXXHai 2019 [45]XXXXHatchel & Marx 2018 [46]XXXXHussong et al. 2018 [47]XXXXXXKing 2017 [48]XXXXXXXKomro et al. 2016 [49]XXXXKucheva 2018 [50]XXXXLeventhal et al. 2018 [51]XXXXXXXXXXLindberg et al. 2019 [52]XXXXXXRespress et al. 2018 [53]XXXXXSanta Maria et al. 2018 [54]XXXXXXXXWade & Peralta 2017 [55]XXXXWallander et al. 2019 [22]XXXXXaQOL Quality of Life ## Study selection Figure 1 shows the PRISMA diagram with the results for study identification, screening, eligibility, and final synthesis. After searching PubMed, Academic Search Premier, PsychInfo, and ERIC, 2124 studies were identified. An additional 5 articles were found after completing a hand search. After duplicates were removed, 2078 articles remained for title and abstract screening using the inclusion criteria listed above. Articles that met inclusion criteria included 263, and an additional 234 were excluded with reasons (i.e. outside age limit, study outside of the United States, and not including at least two racial/ethnic groups). A total of 29 articles were included for final synthesis. ## Study characteristics and outcomes of studies The results of each study are shown in Tables 3 and 4. Table 3 summarizes results by study design and outcome. Across the 29 studies, 11 were cross-sectional [30, 31, 35, 40, 42, 46, 49, 52–55], 16 were cohort [22, 28, 29, 32, 33, 36–38, 41, 43–45, 47, 48, 50, 51], and 2 were experimental [34, 39]. There were no quasi-experimental study designs that met the eligibility criteria for inclusion in this review. Table 4Social Determinant of Health Categories and OutcomesAuthor/YearSocial Determinants CategoriesOutcomesEconomicSocial/CommunityNeighborhood/Built EnvironmentSelf-EfficacyEducational AttainmentPsychological ResilienceHealth Risk BehaviorSmokingAlcoholSubstance UseQOLaAbram et al. 2017 [28]XXXAssini-Meytin et al. 2019 [29]XXXXBaiden et al. 2019 [30]XXXBares et al. 2019 [31]XXBenner & Wang, 2014 [32]XXBenner & Wang, 2015 [33]XXXBohnert et al. 2017 [34]XXBrooks-Russell et al. 2019 [35]X XXCage et al. 2018 [36]XXCamenga et al. 2018 [37]XXChambers et al. 2018 [38]XXXChampion et al. 2016 [39]XXXXXChong et al. 2019 [40]XXClark et al. 2015 [41]XXCombs et al. 2018 [42]XXDocherty et al. 2018 [43]XXGerard & Booth 2015 [44]XXHai 2019 [45]XXHatchel & Marx 2018 [46]XXHussong et al. 2018 [47]XXXXKing 2017 [48]XXKomro et al. 2016 [49]XXKucheva 2018 [50]XXLeventhal et al. 2018 [51]XXXXLindberg et al. 2019 [52]XXRespress et al. 2018 [53]XXSanta Maria et al. 2018 [54]XXXWade & Peralta 2017 [55]XXWallender et al. 2019 [22]XXaQOL Quality of Life Table 4 summarizes results by the five SDOH categories and outcomes. Across the 29 studies, 1 study examined self-efficacy [40], 6 educational attainment [28, 29, 32, 36, 42, 50], 10 health risk behavior [30, 34, 38, 39, 43, 44, 47, 48, 52, 53], 5 smoking [35, 37, 39, 47, 51] 12 alcohol use [29, 30, 33, 35, 39, 41, 45, 47, 49, 51, 54, 55], 10 substance use [28, 29, 31, 33, 35, 38, 39, 46, 51, 54], and 1 quality of life [22]. There were no studies that included psychological resilience as an outcome. ## Social determinant of health The Healthy People 2030 Framework categorizes SDOH into five key areas: economic stability, education access and quality, social and community context, healthcare access and quality, and neighborhood and built environment. These five key areas span a variety of topics subsequently used to identify objectives and evidence-based strategies to address public health issues [56]. The majority of the articles included for final synthesis in this review included the SDOH within the social and community context. Out of the 29 articles included for final synthesis, a total of 18 articles were categorized within the social and community context [28, 30, 32, 33, 36, 38–41, 44–49, 51, 53, 55]. Four out the 29 articles were categorized as neighborhood and built environment [31, 35, 37, 50], and 7 as economic stability [22, 29, 34, 42, 43, 52, 54]. No articles included for final synthesis included health and health care and education categories of SDOH. ## Discussion This scoping review is one of the first to our knowledge to provide a summary of recent evidence on the role of SDOH across 9 health outcomes in US adolescence and youth aged 10–24 [12] by race and ethnicity. A reproducible search across four databases yielded 2124 articles of which 29 were included for final synthesis and extraction based on inclusion criteria. ## Economic stability Economic stability refers to employment, food insecurity, housing instability, and poverty [56]. Of the 29 total studies, 7 studies examined the role of economic stability [22, 29, 34, 42, 43, 52, 54]. Outcomes examined across all studies included educational attainment, health risk behavior, alcohol and substance use, education, and quality of life. All 7 studies demonstrated racial differences in health outcomes among adolescents who were found to be economically disadvantaged. For example, economically disadvantaged African American women who participated in community-based programming, significantly decreased sedentary time and increased physical activity, compared to Hispanic women of the same age group [34]. Lindberg et al. [ 52] found that African American men whose mothers did not have a college degree were more likely to engage in sexual activity prior to the age of 13, compared to any other racial/ethnic and maternal education combination. For adolescents and youth who were homeless, Santa Maria et al. [ 54] found increased use of alcohol, marijuana, synthetic marijuana, and stimulants for those living on the street compared to those who had unstable housing or living in a shelter. These findings varied by race/ethnicity. For example, non-Hispanic white adolescents and youth had the highest lifetime use of alcohol during adolescence, synthetic marijuana, stimulants, and opioids, with significant past month use of marijuana by Hispanic and “other” race/ethnicity adolescents and youth [54]. Additionally, for adolescents and youth who were homeless and who had higher rates of adverse childhood experiences, increases were found in past use of alcohol, synthetic marijuana, and opioids, though not significant for marijuana or stimulants [54]. For adolescents and youth who experienced foster care, racial and ethnic differences were identified for rates of early pregnancy or parenthood [42]. Specifically, American Indian women and men had higher rates of early parenthood compared to those who did not identify as American Indian. Similarly, Hispanic women had significantly higher rates of early pregnancy compared to non-Hispanic women, though Hispanic men demonstrated no significant differences [42]. When considering economically disadvantaged teen fathers, Assini-Meytin et al. [ 29] found that African American teen fathers had lower rates of substance and alcohol use in adolescence and youth compared to non-Hispanic white and Hispanic teen fathers. By young adulthood, a greater proportion of African American and Hispanic teen fathers had not completed high school compared to non-Hispanic white teen fathers, though the difference was not significant [29]. Wallander et al. [ 22] found racial and ethnic differences in health-related quality of life among non-Hispanic white, African American, and Hispanic adolescents and youth, especially within early adolescence, ages 10–13. Non-Hispanic white adolescents had consistently higher quality of life, with Hispanic adolescents reporting the lowest quality of life across three grade periods, 5th, 7th, and 10th [22]. However, when adjusting for SES, differences between non-Hispanic white adolescents and African American adolescents were no longer present, though differences between non-Hispanic white and Hispanic, and African American and Hispanic remained [22]. Docherty et al. [ 43] examined the role of economic disadvantage on the risk of gun-carrying between African American and non-Hispanic white adolescents and did not find any racial/ethnic differences. Findings showed that peer delinquency was a stronger predictor of gun carrying at higher levels of neighborhood disadvantage, with aggression as a stronger predictor at lower levels of disadvantage [43]. African American adolescents had higher rates of neighborhood disadvantage, with a stronger predictor of peer delinquency, compared to non-Hispanic white adolescents [43]. ## Social and community context Social and community context refers to civic participation, incarceration, discrimination, and social cohesion [56]. The majority of articles, 18, included in this review included the social and community context [28, 30, 32, 33, 36, 38–41, 44–49, 51, 53, 55]. Among these articles, 13 articles referred to social cohesion, 4 to discrimination, 1 to incarceration, and none to civic participation. The outcomes examined within the 18 articles included educational attainment, substance use, health risk behavior, alcohol use, self-efficacy, and smoking behavior. Overall, 13 of 18 articles found racial or ethnic differences in outcomes. For example, after juvenile detention, non-Hispanic white women were twice as likely to attain education compared to Hispanic or African American women [28]. Examining discrimination, adolescents and youth of color exhibited differing rates of negative health behavior related to alcohol, smoking, sexual risk behavior, and delinquent behavior when subjected to societal discrimination [51], discrimination at school [38, 53], and fear of police bias [55]. Both Leventhal et al. [ 51] and Respress et al. [ 53] found when either subjected to teacher discrimination [53] or having an increase in concern for societal discrimination [51], racial/ethnic minority adolescents and youth participated in smoking and risky sexual behavior at higher rates. Specifically, Leventhal et al. [ 51] found experiences of societal discrimination was associated with significantly more smoking days within the past-month for African American and Hispanic adolescents compared to other racial/ethnic groups. For students who identified as “other” race, teacher discrimination increased the likelihood for engaging in risky sexual behavior by nearly 2.2 times [53]. Chambers et al. [ 38] also found the more inclusive school environment, the less delinquent behavior, such as involvement in violence, was demonstrated by African American students compared to non-Hispanic white students. As the number of African American students and staff increased, and the perception of teacher discrimination decreased, the lower number of delinquent behaviors were demonstrated [38]. However, the greater amount of perceived peer inclusion, the rate of delinquent behavior increased for African American students compared to non-Hispanic white students [38]. Additionally, Wade & Peralta [55] found fear of race biased policing decreased odds of heavy episodic drinking among racial/ethnic minority adolescents. Finally, while discrimination was a risk factor for depression in Native American women compared to non-native women, it did not have a direct or indirect effect on alcohol use [49]. Overall, Komro et al. [ 49] found no significant differences in alcohol use for non-native and Native American women with similar predictive and protective factors, including alcohol access, parental communication, and best friend’s alcohol use. Related to discrimination, demographic marginalization within schools was found to impact racial/ethnic differences in outcomes. Demographic marginalization refers to the proportion of students with dissimilar backgrounds [33]. For adolescents experiencing racial/ethnic marginalization within schools, ability to experience school attachment was lower, leading to more depressive symptoms, ultimately leading to higher levels of alcohol or substance use [33]. Additionally, African American students who experienced only racial/ethnic marginalization or both racial/ethnic and SES marginalization were found to have lower school attachment and educational attainment compared to all other races/ethnicities [32]. Finally, social cohesion was the most common category within the social and community context. Types, intensity, and length of time of social cohesion factors were associated with adolescent health outcomes. Parenting style and background, determined by acceptance and control, were found to contribute to racial/ethnic differences in substance use [41]. Specifically, Clark et al. [ 41] found no significant differences for parenting style and not engaging in heavy episodic drinking (HED) between non-Hispanic white and African American adolescents. However, for adolescents who did report HED, permissive and authoritarian parenting were risk factors for African American adolescents. Authoritarian parenting style was in turn beneficial for African American adolescents who did not report HED at age 12 [41]. Overall, higher parental socio-economic status was protective for both racial groups, with access to alcohol in the home a greater risk for African Americans [41]. Religiosity was found to be a buffering effect to alcohol and binge drinking for non-Hispanic white adolescents, compared to non-White adolescents [45]. Social interactions, both positive and negative, among peers was shown to impact health outcomes for adolescents. Chong et al. [ 40] found that racial/ethnic minority adolescents with greater involvement in Gay-Straight Alliances had greater race-related self-efficacy, the ability to address diversity, compared to non-Hispanic white adolescents. For both non-Hispanic white adolescents and racial/ethnic minority adolescents, having close friends who identified as racial/ethnic minorities increased self-efficacy. Furthermore, participation in discussions related to racial issues increased racial self-efficacy for non-Hispanic white adolescents, but only increased for racial/ethnic minority adolescents if discussions were frequent [40]. Gerard & Booth [44] considered the impact of individual, family, and school variables on the involvement in aggressive or delinquent behavior by non-Hispanic white and all minority adolescents. School connectedness was found to have a significant relationship with behavior for non-Hispanic white adolescents, not minority adolescents [44]. Hatchel & Marx found that school belongingness served to significantly mediate the relationship between peer victimization and drug use, also noting that while non-White adolescents experienced greater levels of victimization, there was not higher engagement in drug use [46]. In addition, Hussong et al. [ 47] found no racial/ethnic differences when considering social integration and depressive symptoms on substance use across varying time-points in adolescents. For adolescents who experienced bullying, physical violence, or sexual violence, differing responses of health risk behavior were found across race/ethnicity. For example, Champion et al. [ 39] found Mexican-American women with a history of violence were three times more likely to report substance use compared to African American women with similar histories. While Baiden et al. [ 30] found African American adolescents experiencing bullying or personal violence had $33\%$ lower odds of suicidal ideation compared to non-Hispanic white adolescents when controlling for other demographic factors. However, when controlling for all predictors, these differences did not remain [30]. For adolescence and youth who have experienced abuse or neglect, racial/ethnic differences were found across outcomes. Cage et al. [ 36] found when considering race/ethnicity alone, there were no significant differences in educational attainment. However, when considering both race/ethnicity and gender, significant differences were found. Non-Hispanic white men and women, along with Hispanic women were over twice as likely to complete high school or obtain GED compared to African American men. Additionally, King [48] found Hispanic women had the highest rates of adolescent births compared to all other race/ethnicities, with non-Hispanic white and Asian-Pacific Islander with significantly lower birth rates. Type and occurrence of abuse, as well as time in foster care predicted the rates of early birth across all racial/ethnic groups [48]. For non-Hispanic white women, time in foster care and age of abuse were significant predictors, with reoccurrence and physical abuse significant predictors for African American and Hispanic adolescence respectively [48]. ## Neighborhood and built environment Neighborhood and built environment refer to access to foods that support healthy eating patterns, crime and violence, environmental conditions, and quality of housing [56]. Overall, 4 articles included in this review examined racial differences in neighborhood and built environment for adolescence [31, 35, 37, 50]. Three outcomes examined within these articles included substance use, smoking, and educational attainment. Differences in outcomes among racial/ethnically diverse adolescents were mixed. Bares et al. [ 31] found non-Hispanic white adolescents who lived on farms had higher rates of opioid use compared to African American and Hispanic adolescents who live on farms and across all races/ethnicities living in the country or city. Considering change in past 30-day prevalence of marijuana use after retail sales became legalized, Brooks-Russell et al. [ 35] did not find any significant change across all racial and ethnic adolescents. Camenga et al. [ 37] found similar results after exposure to e-cigarette advertising, no racial differences were found in e-cigarette use. Kucheva [50] found racial/ethnic differences when considering two different subsidized housing: public and privately managed [50]. For example, African American adolescent men in private subsidized housing and public subsidized housing were less likely to become teenage parents [50]. However, African American women were less likely to graduate high school if they lived within a privately managed subsidized housing, compared to non-Hispanic white men and women and African American men [50]. ## Limitations While this review provides a summary of recent evidence on racial/ethnic differences in SDOH and outcomes among adolescents and youth, there are several limitations that should be considered. First, the search for this review included only articles written in English, thus excluding articles that may have been relevant to understanding SDOH and adolescence published in another language. Second, studies that were disease specific or targeting a sub-population of adolescence and youth were not included, for example adolescents living with a pre-existing chronic or mental health condition. Therefore, SDOH that may contribute to specific disease occurrence or outcomes may vary from what has been presented in this summary. Finally, this review is considered narrative and cannot speak to any causal relationships. ## Implications for Public Health Education & Programming for adolescence and youth Review of the literature demonstrates the role of multiple factors on adolescent and youth health outcomes based on the Healthy People 2030 SDOH Framework. While significant differences in outcomes were found across race/ethnicity, intersectionality of adolescent and youth identities is a critically important influence to consider for future work [57–59]. Interactions of social and structural factors, often outside the control of adolescents and youth, create a multi-dimensional understanding of adolescent and youth health behavior [57]. This review demonstrates the limited number of studies focused on SDOH domains outside of the social and community context. Identifying the impact of barriers within each domain, especially the inequitable influence across adolescent and youth populations, is crucial to addressing positive health outcomes for health during adolescence and youth. These aspects of adolescents’ and youth lives, along with structural racism [60] and disadvantage [61–63] increase the likelihood of engaging in risky health behaviors and ultimately leading to negative health outcomes during adolescence and youth and future adulthood [61, 63]. Therefore, to effectively address health and well-being, researchers, practitioners, and public health educators should consider a multidimensional and structural lens [57] when studying and developing programming for adolescent youth health [64, 65]. Public health initiatives and policies should also address social inequities to limit the accumulation of disadvantage throughout the life course [63]. School based public health initiatives in areas of sex education, safe and supportive environments and school policy improvement have been successful in addressing health inequities among adolescents [66]. Current adolescent and youth surveillance systems focus on risk behavior and school policies and practices but are limited in the inclusion of SDOH. Review of the literature demonstrates the limited data on SDOH factors, especially education and health care, relative to adolescence and youth health outcomes. SDOH indicators should be included in public health surveillance of adolescents and youth. For example, additional data related to food insecurity, housing instability, discrimination, and crime and violence could provide needed context to effectively dismantle structural barriers to positive health outcomes for adolescent and youth. Policies and programs can be tailored to specific needs of addressing adolescent and youth health equity. In addition, collaboration among the community, public health organizations, healthcare institutions and school districts is essential in addressing this multifaceted issue [67]. Finally, future research should be innovative and interdisciplinary to capture the intersectional identities of adolescence [57], health behavior, and health outcomes. ## References 1. Ferrie JE, Shipley MJ, Stansfeld SA, Marmot MG. Effects of chronic job insecurity and change in job security on self reported health, minor psychiatric morbidity, physiological measures, and health related behaviours in British civil servants: the Whitehall II study. J Epidemiol Community Health (2002.0) 56 450-454. DOI: 10.1136/jech.56.6.450 2. Wilkinson RG, Marmot MM. Social determinants of health: the solid facts (2003.0) 3. Marmot MG, Wilkinson RG. Social determinants of health (2008.0) 4. Solar O, Irwin AA. A conceptual framework for action on the social determinants of health: social determinants of health discussion paper 2 (policy and practice) (2010.0) 5. 5.Commission on Social Determinants of HealthClosing the gap in a generation: health equity through action on the social determinants of health2008World Health Organization. Closing the gap in a generation: health equity through action on the social determinants of health (2008.0) 6. 6.Office of Disease Prevention and Health PromotionHealthy people 2020: An opportunity to address the societal determinants of health in the United States2010. Healthy people 2020: An opportunity to address the societal determinants of health in the United States (2010.0) 7. Wilkinson RG, Marmot MG. Social determinants of health: the solid facts (1998.0) 8. Braveman P, Gottlieb L. The social determinants of health: it’s time to consider the causes of the causes. Public Health Rep (2014.0) 129 Suppl 2 19-31. DOI: 10.1177/00333549141291S206 9. Norton JM, Moxey-Mims MM, Eggers PW. Social determinants of racial disparities in CKD. J Am Soc Nephrol (2016.0) 27 2576-2595. DOI: 10.1681/ASN.2016010027 10. Walker RJ, Strom Williams J, Egede LE. Influence of race, ethnicity and social determinants of health on diabetes outcomes. Am J Med Sci (2016.0) 351 366-373. DOI: 10.1016/j.amjms.2016.01.008 11. 11.Office of Disease Prevention and Health Promotion. Social determinants of health. Healthy People 2030: U. S. Department of Health and Human Services. https://health.gov/healthypeople/objectives-and-data/social-determinants-health 12. Health and health behavior among young people. Health Policy for Children and Adolescents, HEPCA Series 1 (2000.0) 1-132 13. Blakemore SJ, Mills KL. Is adolescence a sensitive period for sociocultural processing?. Annl Review of Psych (2014.0) 65 187-207. DOI: 10.1146/annurev-psych-010213-115202 14. Frech A. Healthy behavior trajectories between adolescence and Young adulthood. Adv Life Course Res (2012.0) 17 59-68. DOI: 10.1016/j.alcr.2012.01.003 15. Viner RM, Ozer EM, Denny S. Adolescence and the social determinants of health. Lancet (2012.0) 379 1641-1652. DOI: 10.1016/S0140-6736(12)60149-4 16. D’Agostino EM, Patel HH, Hansen E, Mathew MS, Messiah SE. Longitudinal effects of transportation vulnerability on the association between racial/ethnic segregation and adolescence cardiovascular health. J Racial Ethn Health Disparities (2021.0) 8 618-629. DOI: 10.1007/s40615-020-00821-8 17. Elster A, Jarosik J, VanGeest J, Fleming M. Racial and ethnic disparities in health care for adolescents: a systematic review of the literature. Arch Pediatr Adolesc Med (2003.0) 157 867-874. DOI: 10.1001/archpedi.157.9.867 18. Kim Y, Landgraf A, Colabianchi N. Living in high-SES neighborhoods is protective against obesity among higher-income children but not low-income children: results from the healthy communities study. J Urban Health (2020.0) 97 175-190. DOI: 10.1007/s11524-020-00427-9 19. Lau M, Lin H, Flores G. Racial/ethnic disparities in health and health care among U.S. adolescents. Health Serv Res (2012.0) 47 2031-2059. DOI: 10.1111/j.1475-6773.2012.01394.x 20. Vancea M, Utzet M. How unemployment and precarious employment affect the health of young people: a scoping study on social determinants. Scand J Public Health (2017.0) 45 73-84. DOI: 10.1177/1403494816679555 21. Vartanian TP, Houser L. The interactive role of SNAP participation and residential neighborhood in childhood obesity. J Children Pov (2020.0) 26 167-189. DOI: 10.1080/10796126.2020.1826246 22. Wallander JL, Fradkin C, Elliott MN. Racial/ethnic disparities in health-related quality of life and health status across pre-, early-, and mid-adolescence: a prospective cohort study. Qual Life Res (2019.0) 28 1761-1771. DOI: 10.1007/s11136-019-02157-1 23. Mohajer N, Earnest J. Widening the aim of health promotion to include the most disadvantaged: vulnerable adolescents and the social determinants of health. Health Educ Res (2010.0) 25 387-394. DOI: 10.1093/her/cyq016 24. Banspach S, Zaza S, Dittus P, Michael S, Brindis CD, Thorpe P. CDC grand rounds: adolescence—preparing for lifelong health and wellness. MMWR Morb Mort Wkly Report (2016.0) 2016 759-762. DOI: 10.15585/mmwr.mm6530a2 25. Egerter S, Braveman P, Sadegh-Nobari T, Grossman-Kahn R, Dekker M. Education and health. Exploring the social determinants of health, issue brief #5 (2011.0) 1-17 26. 26.Office of Disease Prevention and Health Promotion. Adolescent health. Healthy People 2030. U.S. Department of Health and Human Services. https://health.gov/healthypeople/objectives-and-data/browse-objectives/adolescents. 27. Tufanaru C, Mun Z, Aromataris E, Campbell J, Hopp L, Aromataris E, Munn Z. Systematic reviews of effectiveness. Joanna Briggs Institute Reviewer’s Manual (2017.0) 28. Abram KM, Azores-Gococo NM, Emanuel KM. Sex and racial/ethnic differences in positive outcomes in delinquent adolescence after detention: a 12-year longitudinal study. JAMA Pediatr (2017.0) 171 123-132. DOI: 10.1001/jamapediatrics.2016.3260 29. Assini-Meytin LC, Garza MA, Green KM. Racial and ethnic differences in teenage Fathers’ early risk factors and socioeconomic outcomes later in life. Child Adolescence Care Forum (2019.0) 48 361-376. DOI: 10.1007/s10566-018-9483-9 30. Baiden P, Mengo C, Boateng GO, Small E. Investigating the association between age at first alcohol use and suicidal ideation among high school students: evidence from the adolescence risk behavior surveillance system. J Affect Disord (2019.0) 242 60-67. DOI: 10.1016/j.jad.2018.08.078 31. Bares CB, Weaver A, Kelso MF. Adolescent opioid use: examining the intersection of multiple inequalities. J of Prev Inter Comm (2019.0) 47 295-309. DOI: 10.1080/10852352.2019.1617382 32. Benner AD, Wang Y. Demographic marginalization, social integration, and adolescents’ educational success. J of Adolescence Adol (2014.0) 43 1611-1627. DOI: 10.1007/s10964-014-0151-6 33. Benner AD, Wang Y. Adolescent substance use: the role of demographic marginalization and socioemotional distress. Dev Psychol (2015.0) 51 1086-1097. DOI: 10.1037/dev0000026 34. Bohnert AM, Bates CR, Heard AM. Improving urban minority Girls’ health via community summer programming. J Racial Ethn Health Disparities (2017.0) 4 1237-1245. DOI: 10.1007/s40615-016-0333-x 35. Brooks-Russell A, Ma M, Levinson AH. Adolescent marijuana use, marijuana-related perceptions, and use of other substances before and after initiation of retail marijuana sales in Colorado (2013-2015). Prev Sci (2019.0) 20 185-193. DOI: 10.1007/s11121-018-0933-2 36. Cage J, Corley NA, Harris LA. The educational attainment of maltreated adolescence involved with the child welfare system: exploring the intersection of race and gender. Chil & Adolescence Ser Rev (2018.0) 88 550-557. DOI: 10.1016/j.childadolescence.2018.04.006 37. Camenga D, Gutierrez KM, Kong G, Cavallo D, Simon P, Krishnan-Sarin S. E-cigarette advertising exposure in e-cigarette naïve adolescents and subsequent e-cigarette use: a longitudinal cohort study. Addict Behav (2018.0) 81 78-83. DOI: 10.1016/j.addbeh.2018.02.008 38. Chambers BD, Erausquin JT. Race, sex, and discrimination in school settings: a multilevel analysis of associations with delinquency. J Sch Health (2018.0) 88 159-166. DOI: 10.1111/josh.12589 39. Champion JD, Young C, Rew L. Substantiating the need for primary care-based sexual health promotion interventions for ethnic minority adolescent women experiencing health disparities. J Am Assoc Nurse Pract (2016.0) 28 487-492. DOI: 10.1002/2327-6924.12346 40. Chong ESK, Poteat VP, Yoshikawa H, Calzo JP. Fostering adolescence self-efficacy to address transgender and racial diversity issues: the role of gay-straight alliances. Sch Psychol Q (2019.0) 34 54-63. DOI: 10.1037/spq0000258 41. Clark TT, Yang C, McClernon FJ, Fuemmeler BF. Racial differences in parenting style typologies and heavy episodic drinking trajectories. Health Psychol (2015.0) 34 697-708. DOI: 10.1037/hea0000150 42. Combs KM, Begun S, Rinehart DJ, Taussig H. Pregnancy and childbearing among Young adults who experienced Foster Care. Child Maltreat (2018.0) 23 166-174. DOI: 10.1177/1077559517733816 43. Docherty M, Beardslee J, Mulvey E, Schubert C, Pardini D. Childhood risk factors associated with adolescent gun carrying among black and white males: an examination of self-protection, social influence, and antisocial propensity explanations. Law Hum Behav (2018.0) 42 110-118. DOI: 10.1037/lhb0000270 44. Gerard JM, Booth MZ. Family and school influences on adolescents’ adjustment: the moderating role of adolescence hopefulness and aspirations for the future. J of Adol (2015.0) 44 1-16. DOI: 10.1016/j.adolescence.2015.06.003 45. Hai AH. Are there gender, racial, or religious denominational differences in Religiosity’s effect on alcohol use and binge drinking among adolescence in the United States? A propensity score weighting approach. Subst Use Misuse (2019.0) 54 1096-1105. DOI: 10.1080/10826084.2018.1555598 46. Hatchel T, Marx R. Understanding Intersectionality and Resiliency among Transgender Adolescents: Exploring Pathways among Peer Victimization, School Belonging, and Drug Use. Int J Environ Res Public Health (2018.0) 15 1289. DOI: 10.3390/ijerph15061289 47. Hussong AM, Ennett ST, McNeish D. Teen social networks and depressive symptoms-substance use associations: developmental and demographic variation. J Stud Alcohol Drugs (2018.0) 79 770-780. DOI: 10.15288/jsad.2018.79.770 48. King B. First births to maltreated adolescent girls: differences associated with spending time in Foster Care. Child Maltreat (2017.0) 22 145-157. DOI: 10.1177/1077559517701856 49. Komro KA, Livingston MD, Garrett BA, Boyd ML. Similarities in the etiology of alcohol use among native American and non-native Young women. J Stud Alcohol Drugs (2016.0) 77 782-791. DOI: 10.15288/jsad.2016.77.782 50. Kucheva Y. Subsidized housing and the transition to adulthood. Demography (2018.0) 55 617-642. DOI: 10.1007/s13524-018-0656-9 51. Leventhal AM, Cho J, Andrabi N, Barrington-Trimis J. Association of Reported Concern about Increasing Societal Discrimination with Adverse Behavioral Health Outcomes in late adolescence. JAMA Pediatr (2018.0) 172 924-933. DOI: 10.1001/jamapediatrics.2018.2022 52. Lindberg LD, Maddow-Zimet I, Marcell AV. Prevalence of sexual initiation before age 13 years among male adolescents and Young adults in the United States. JAMA Pediatr (2019.0) 173 553-560. DOI: 10.1001/jamapediatrics.2019.0458 53. Respress BN, Amutah-Onukagha NN, Opara I. The effects of school-based discrimination on adolescents of color sexual health outcomes: a social determinants approach. Social Work Public Health (2018.0) 33 1-16. DOI: 10.1080/19371918.2017.1378953 54. Santa Maria DM, Narendorf SC, Cross MB. Prevalence and correlates of substance use in homeless adolescence and Young adults. J Addict Nurs (2018.0) 29 23-31. DOI: 10.1097/JAN.0000000000000206 55. Wade J, Peralta RL. Perceived racial discrimination, heavy episodic drinking, and alcohol abstinence among African American and white college students. J Ethn Subst Abus (2017.0) 16 165-180. DOI: 10.1080/15332640.2015.1113152 56. 56.Office of Disease Prevention and Health Promotion. Social determinants of health. Healthy People 2020. U.S. Department of Health and Human Services. https://www.healthypeople.gov/2020/topics-objectives/topic/social-determinants-of-health. 57. Lopez N, Gadsden VL. Health inequities, social determinants, and intersectionality. NAM Perspectives (2016.0) 58. Evans CR. Adding interactions to models of intersectional health inequalities: comparing multilevel and conventional methods. Soc Sci Med (2019.0) 221 95-105. DOI: 10.1016/j.socscimed.2018.11.036 59. Kern MR, Duinhof EL, Walsh SD. Intersectionality and adolescent mental well-being: a Cross-nationally comparative analysis of the interplay between immigration background, socioeconomic status and gender. J Adolesc Health (2020.0) 66 S12-S20. DOI: 10.1016/j.jadohealth.2020.02.013 60. 60.Trent M, Dooley DG, Dougé J. The impact of racism on child and adolescent health. Ped. 2019;144(2). 10.1542/peds.2019-1765. 61. Nurius PS, Prince DM, Rocha A. Cumulative disadvantage and adolescence well-being: a multi-domain examination with life course implications. Child Adolesc Social Work J (2015.0) 32 567-576. DOI: 10.1007/s10560-015-0396-2 62. Gustafsson PE, San Sebastian M, Janlert U, Theorell T, Westerlund H, Hammarström A. Life-course accumulation of neighborhood disadvantage and allostatic load: empirical integration of three social determinants of health frameworks. Am J Public Health (2014.0) 104 904-910. DOI: 10.2105/AJPH.2013.301707 63. Senter JP, Bucay-Harari L, Castillo-Salgado C. Using social indicators to describe neighborhood-level disparities in adolescent health in Baltimore City circa 2017. J Adolesc Health (2020.0) 67 270-277. DOI: 10.1016/j.jadohealth.2020.01.021 64. 64.Freudenberg N, Franzosa E, Chisholm J, Libman K. New approaches for moving upstream: how state and local health departments can transform practice to reduce health inequalities. Health Educ Behav. 2015;42(1_suppl):46S–56S. 10.1177/1090198114568304. 65. Turan JM, Elafros MA, Logie CH. Challenges and opportunities in examining and addressing intersectional stigma and health. BMC Med. (2019.0) 17 7. DOI: 10.1186/s12916-018-1246-9 66. 66.Centers for Disease Control and PreventionSuccess Stories2020. Success Stories (2020.0) 67. 67.Center for Disease Control and Prevention. Improving School-Based Health and Education Policies. www.cdc.gov/healthyadolescence/stories/pdf/DASH-Health-Education-Policy-Success-2018.pdf
--- title: 'The effect of metabolic dysfunction-associated fatty liver disease and diabetic kidney disease on the risk of hospitalization of heart failure in type 2 diabetes: a retrospective cohort study' authors: - Seung Eun Lee - Juhwan Yoo - Bong-Seong Kim - Han Seok Choi - Kyungdo Han - Kyoung-Ah Kim journal: Diabetology & Metabolic Syndrome year: 2023 pmcid: PMC9976518 doi: 10.1186/s13098-023-01006-z license: CC BY 4.0 --- # The effect of metabolic dysfunction-associated fatty liver disease and diabetic kidney disease on the risk of hospitalization of heart failure in type 2 diabetes: a retrospective cohort study ## Abstract ### Background Diabetes mellitus is a major risk factor for heart failure. A recent consensus statement recommended annual cardiac biomarker testing (e.g. natriuretic peptide or high-sensitivity cardiac troponin) for all patients with diabetes. We aimed to identify patients at a higher risk of hospitalization for heart failure among patients with type 2 diabetes to prioritize those who would require screening. ### Methods Overall, 1,189,113 patients who underwent two medical health checkup cycles (2009–2012 and 2011–2014) and had stable diabetic kidney disease (DKD) phenotype in the Korean National Health Insurance Service database were included in this study. After excluding those with concurrent proteinuria (PU) and reduced estimated glomerular filtration rate, three groups (no-DKD, PU+DKD, and PU−DKD) were identified. A fatty liver index of ≥ 60 was defined as metabolic dysfunction–associated fatty liver disease (MAFLD). Patients were followed up until December 2018 or until outcomes developed. The Cox proportional hazard model was used to compare the risk of hospitalization for heart failure across groups. ### Results During an average of 6.6 years of follow-up, 5781 patients developed hospitalization for heart failure. After adjusting for covariates, the risk of hospitalization for heart failure was highest in the PU+DKD group [HR 3.12, $95\%$ CI (2.75–3.55)], followed by the PU−DKD group [HR 1.85, $95\%$ CI (1.73–1.99)] using the no-DKD group as the reference category. The risk of hospitalization for heart failure was comparable regardless of MAFLD status in patients who already had DKD. However, in the no-DKD group, the risk of hospitalization for heart failure was 1.4 times higher in patients with MAFLD than in those without [HR 1.41, $95\%$ CI (1.31–1.52)]. ### Conclusions In lines with the international consensus statement, we suggest that annual cardiac biomarker testing should be conducted at least in patients with DKD and/or MAFLD. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13098-023-01006-z. ## Background Heart failure (HF) is a global health problem with a rising prevalence rate [1, 2] mostly due to prolonged human lifespan. Worldwide, HF affects approximately 26 million people [3] and the burden is estimated to increase continuously. Mortality and morbidity associated with HF are high, and the mortality rate of HF was reported to be approximately $10\%$ at 30 days, 20–$30\%$ at 1 year, and 45–$60\%$ over 5 years of follow-up [4]. HF is attributed to cumulative exposure to multiple risk factors; hence, early screening, detection, and correct modifiable risk factors are essential for reducing HF-related burden [2]. Type 2 diabetes is a well-known risk factor for HF [5]. Based on a report that shows many people with diabetes have subclinical structural heart disease, the American Diabetes Association (ADA) recommends measuring natriuretic peptide or high-sensitivity cardiac troponin in patients with diabetes on at least a yearly basis [6]. However, it may be impractical to perform biochemical tests for all patients with diabetes, and risk stratification with specific clinical recommendations should be provided. Chronic kidney disease (CKD) is a well-established risk factor for HF [7]. A growing body of evidence has recently shown the association between nonalcoholic fatty liver disease (NAFLD) and HF [8, 9]. In 2020, a new nomenclature for metabolic dysfunction–associated fatty liver disease (MAFLD) was proposed as a substitute for NAFLD [10], and this definition was endorsed by global multi-stakeholder [11]. It may be useful for identifying a greater number of individuals with metabolically complicated fatty liver and an increased risk for cardiovascular disease (CVD) [12, 13]. Several multivariable models have been used to predict the risk of HF in patients with diabetes [6, 14]. Although renal dysfunction was included as a risk factor for HF in reported models, none of them considered MAFLD. Therefore, in this study, we investigated whether the inclusion of diabetic kidney disease (DKD) and/or MAFLD improves the prediction of HF risk in patients with type 2 diabetes. ## Data source and study population This study used data from the Korean National Health Insurance Service (NHIS), which is the sole insurance provider for all Korean residents. The NHIS-established databases (DBs) included the qualification, treatment, and medical checkup DBs [15]. Briefly, the qualification DB included data on qualifications, including age, sex, location, and socioeconomic variables; the treatment DB contained payment data to the clinic upon treatment of the patients at the clinic; and the medical checkup DB comprised major results from medical checkups, behavior, and habitual data from the questionnaire. We used the qualification and medical checkup DBs to examine the baseline characteristics of the study population and the treatment DB to investigate the outcomes. Because we used previously collected, publicly available, de-identified data, ethical review by the Institutional Review Board and informed consent were exempted. Permission for the use of health check-up data was granted by the NHIS (NHIS-2021-1-634). ## Study design This study included 1,779,819 subjects with type 2 diabetes who underwent at least two general medical checkups in 2009–2012 and 2011–2014 (Additional file 1: Fig. S1A). The exclusion criteria were as follows: [1] individuals diagnosed with cancer ($$n = 68$$,282); [2] individuals diagnosed with thyrotoxicosis ($$n = 78$$,467); [3] individuals with renal diseases other than DKD ($$n = 135$$,698); [4] individuals with rheumatic mitral valve disease ($$n = 4695$$); [5] individuals with missing values ($$n = 48$$,959); and [6] those with an estimated glomerular filtration rate (eGFR) of less than 30 mL/min/1.73 m2 ($$n = 14$$,889). Additionally, patients with proteinuric DKD with reduced eGFR (< 60 mL/min/1.73 m2) at the first examination were excluded since the very high-risk Kidney Disease: Improving Global Outcomes (KDIGO) categories are well known for poor cardiovascular outcomes [16]. Subsequently, we only included patients with a stable DKD phenotype for over 2 years. Patients with a stable DKD phenotype for over 2 years were subclassified according to the presence or absence of MAFLD and were followed up until hospitalization for heart failure (HHF) or December 2018 (Additional file 1: Fig. S1B). ## Definitions of diabetes, DKD, and MAFLD Type 2 diabetes was defined as the presence of the diagnostic code (International Classification of Disease-Tenth Revision (ICD-10) code: E11–E14) and the prescription of relevant glucose-lowering drugs. When the participants did not meet the criteria above, they were defined as having type 2 diabetes if their fasting plasma glucose levels were ≥ 126 mg/dL during a medical checkup. The eGFR was determined using the equation from the Modification of Diet in Renal Disease study [17] and reduced eGFR was defined as values less than 60 mL/min/1.73 m2. Notably, positive proteinuria (PU) of ≥ 1 + was defined based on the urinary dipstick test. Additionally, the DKD phenotype was categorized into three distinct groups based on the eGFR levels (normal vs. reduced) and PU (negative vs. positive) as follows: group 1 (no-DKD), normal eGFR and negative PU; group 2 (PU+DKD), normal eGFR and positive PU; and group 3 (PU−DKD), reduced eGFR and negative PU. The fatty liver index (FLI) was used to identify patients with MAFLD [18]. According to the criteria [18], MAFLD was diagnosed regardless of having other etiologies such as alcohol-associated fatty liver disease and viral hepatitis. FLI was calculated using the following equation: ex/(1 + ex) × 100, $x = 0.953$ × log (triglyceride) + 0.139 × body mass index + 0.718 × log (gamma-glutamyl transferase (GGT)) + 0.053 × waist circumference − 15.745). Particularly, an FLI of ≥ 60 was defined as MAFLD [19]. ## Definitions of comorbidities Patients with HF were identified based on ICD-10 codes for heart failure (I50). Hypertension was indicated in patients according to the ICD-10 code for hypertension (I10–I13, I15) and the prescribed antihypertensive medications. Participants were also considered hypertensive if their systolic blood pressure was ≥ 140 mmHg and/or diastolic blood pressure was ≥ 90 mmHg during a general medical checkup. Moreover, patients with dyslipidemia were identified by the ICD-10 code for dyslipidemia (E78) with treatment undergone using lipid-lowering agents or a total cholesterol level ≥ 240 mg/dL during a medical checkup. Proliferative diabetic retinopathy (PDR) was established if the participants had two or more diagnoses of diabetic retinopathy (H360) and a procedure code for pan-retinal photocoagulation (S5160, S5161). ## Laboratory and clinical examination This study obtained laboratory results and clinical characteristics during the second examination. Body mass index was calculated as weight divided by height in meters squared (kg/m2). After overnight fasting, venous samples were used to evaluate fasting plasma glucose, total cholesterol, triglyceride, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, creatinine, aspartate aminotransferase (AST), alanine aminotransferase (ALT), GGT, and hemoglobin levels. Additionally, health-related lifestyles were evaluated using self-administered questionnaires categorized as current smokers or non-smokers, heavy drinkers (≥ 30 g/day of alcohol) or non-heavy drinkers, and participants with or without regular exercise. ## Outcome The primary outcome of this study was the hospitalization of patients for HF. Cases were defined as patients who were admitted to a hospital with a discharge diagnostic code for HF (I50). ## Statistical analysis Descriptive statistics were used to summarize baseline characteristics. Baseline characteristics across the groups were presented as numbers (percentages) for categorical variables and mean ± standard deviation for continuous variables. If the distribution of continuous variables was heavily skewed, the geometric mean was used. To analyze the differences in baseline characteristics between the groups, a one-way analysis of variance was used for continuous variables, and the chi-squared test was used for categorical variables. The cumulative incidence of HHF was calculated using Kaplan–Meier estimates, and we performed a log-rank test to analyze the differences in HHF risk across groups. The incidence of HHF was expressed as the number of events per 1000 person-years. Cox proportional hazards regression analysis was performed to assess the hazard ratio (HR) for HHF across the groups. Model 1 was unadjusted; Model 2 was adjusted for age and sex; and Model 3 was adjusted for smoking, drinking, and physical activity. Additionally, Model 4 was adjusted for comorbidities, including hypertension, dyslipidemia, atrial fibrillation, and ischemic heart disease. Finally, Model 5 was further adjusted for fasting glucose, diabetes duration, hemoglobin levels, and insulin usage. Subgroup analyses with tests for interaction were performed according to age group (< 65 vs. ≥ 65 years), sex, and the presence or absence of prevalent HF. Statistical analyses were conducted using the SAS software (version 9.4; SAS Institute, Cary, NC, USA). Statistical significance was set at $P \leq 0.05.$ ## Baseline characteristics of the study population Table 1 presents the baseline characteristics of the study population according to changes in the DKD phenotype. The prevalence of DKD phenotypes was $95.2\%$ (1,132,$\frac{531}{1}$,189,113) in the no-DKD group, $1.3\%$ (15,$\frac{619}{1}$,189,113) in the PU+DKD group, and $3.4\%$ (40,$\frac{963}{1}$,189,113) in the PU−DKD group. The prevalence of MAFLD was $25.1\%$ (298,522 of 1,189,113), and patients with MAFLD were younger, more obese, and more likely to be male than those without MAFLD. Additionally, they tended to be current smokers and heavy drinkers and did not exercise regularly. The prevalence of hypertension and dyslipidemia was higher in patients with MAFLD than in those without MAFLD. Despite the shorter duration of diabetes, patients with MAFLD showed higher fasting plasma glucose levels than those without MAFLD. Predictably, the AST, ALT, and GGT levels were higher in patients with MAFLD than in those without MAFLD.Table 1Baseline characteristics according to DKD/MAFLD phenotypeno-DKD($$n = 1$$,132,531)PU+DKD($$n = 15$$,619)PU−DKD($$n = 40$$,963)MAFLD−MAFLD+MAFLD−MAFLD+MAFLD−MAFLD+n848,716283,8159107651232,7688195Male507,170 (59.76)231,817 (81.68)6670 (73.24)5390 (82.77)12,627 (38.53)4475 (54.61)< 0.001Age (years)57.48 ± 11.8852.62 ± 11.360.14 ± 10.4553.79 ± 10.7570.62 ± 8.0967.96 ± 8.79< 0.001BMI (mg/k2)23.89 ± 2.6227.91 ± 3.2324 ± 2.6128.3 ± 3.6224.27 ± 2.6928.56 ± 3.19< 0.001WC (cm)82.17 ± 7.0992.86 ± 7.3883.72 ± 6.8194.17 ± 8.0583.98 ± 7.2895.46 ± 7.22< 0.001Current smoker198,817 (23.43)109,561 (38.6)2613 (28.69)2558 (39.28)2883 (8.8)1137 (13.87)< 0.001 Heavy drinker58,984 (6.95)54,637 (19.25)805 (8.84)1380 (21.19)599 (1.83)517 (6.31)< 0.001 Regular Exercise200,956 (23.68)55,512 (19.56)2138 (23.48)1259 (19.33)6648 (20.29)1508 (18.4)< 0.001Comorbidities Hypertension409,799 (48.28)169,157 (59.6)6818 (74.87)5118 (78.59)26,963 (82.28)7234 (88.27)< 0.001 Dyslipidemia322,792 (38.03)130,242 (45.89)4819 (52.92)3891 (59.75)18,350 [56]4939 (60.27)< 0.001 IHD126,334 (14.89)37,077 (13.06)1911 (20.98)1144 (17.57)10,708 (32.68)2689 (32.81)< 0.001 AF7394 (0.87)2295 (0.81)171 (1.88)106 (1.63)936 (2.86)273 (3.33)< 0.001 Stroke48,612 (5.73)11,547 (4.07)923 (10.14)394 (6.05)5466 (16.68)1199 (14.63)< 0.001 PAD147,245 (17.35)39,420 (13.89)2151 (23.62)1111 (17.06)10,198 (31.12)2422 (29.55)< 0.001 CVD256,905 (30.27)71,782 (25.29)3815 (41.89)2111 (32.42)18,938 (57.79)4561 (55.66)< 0.001 Heart failure18,840 (2.22)5978 (2.11)357 (3.92)201 (3.09)2917 (8.9)772 (9.42)< 0.001Severity of diabetes FPG ≥ 150 mg/dL175,770 (20.71)84,175 (29.66)3787 (41.58)3340 (51.29)6040 (18.43)2022 (24.67)< 0.001 FPG (mg/dL)131.11 ± 41.77141.98 ± 45.4154.7 ± 58164.19 ± 53.07127.41 ± 43.07136.17 ± 46.43< 0.001 DM ≥ 5 yrs297,221 (35.02)65,008 (22.91)5938 (65.2)2803 (43.04)20,628 (62.95)4365 (53.26)< 0.001 Insulin use59,635 (7.03)13,919 (4.9)1997 (21.93)835 (12.82)5445 (16.62)1256 (15.33)< 0.001 ≥ 2 oral GLD346,564 (40.83)102,721 (36.19)5862 (64.37)3615 (55.51)18,700 (57.07)4541 (55.41)< 0.001 PDR3751 (0.44)437 (0.15)346 (3.8)99 (1.52)359 (1.1)50 (0.61)< 0.001SBP (mmHg)126.3 ± 14.78131.15 ± 14.51132.64 ± 16.82136.69 ± 16.86129.64 ± 16.05132.05 ± 16.01< 0.001DBP (mmHg)77.37 ± 9.5981.7 ± 9.8779.46 ± 10.5384.18 ± 11.176.38 ± 10.1779 ± 10.17< 0.001eGFR (mL/min/1.73m2)91.69 ± 36.7492.26 ± 40.487.31 ± 35.0391.7 ± 46.950.58 ± 7.2950.78 ± 7.28< 0.001Non HDL-C (mg/dL)136.22 ± 38.54155.02 ± 43.5139.24 ± 43.11160.8 ± 51.24132.94 ± 40.33147.64 ± 57.7< 0.001AST (IU/L)23.75 (23.73–23.77)32.63 (32.57–32.69)23.57 (23.38–23.76)33.44 (33.01–33.87)23.14 (23.06–23.23)28.61 (28.32–28.9)< 0.001 ALT (IU/L)22.6 (22.58–22.63)37.86 (37.78–37.94)22.52 (22.29–22.75)36.85 (36.33–37.38)19.03 (18.94–19.13)27.6 (27.26–27.94)< 0.001γGTP (IU/L)27.78 (27.74–27.81)74.51 (74.3–74.72)31.27 (30.86–31.69)78.87 (77.37–80.39)22.79 (22.65–22.92)49.13 (48.33–49.94)< 0.001no-DKD: normal eGFR (eGFR ≥ 60) with negative PU; PU+DKD: normal eGFR with positive PU; PU−DKD: reduced eGFR (eGFR < 60) with negative PUValues are presented as mean ± standard deviation or number (%). Data for the parameters marked with an asterisk () are presented as the geometric mean and $95\%$ confidence intervalAF: atrial fibrillation; ALT: alanine aminotransferase; AST: aspartate aminotransferase; BMI: body mass index; CVD: cardiovascular disease; DBP: diastolic blood pressure; DKD: diabetic kidney disease; DM: diabetes mellitus; eGFR: estimated glomerular filtration rate; FPG: fasting plasma glucose; γGTP: gamma-glutamyl transferase; GLD: glucose-lowering drugs; HDL-C: high-density lipoprotein cholesterol; IHD: ischemic heart disease; MAFLD: metabolic dysfunction–associated fatty liver disease; PAD: peripheral artery disease; PDR: proliferative diabetic retinopathy; PU: proteinuria; RAS: renin-angiotensin system; SBP: systolic blood pressure; WC: waist circumference ## Risk of HHF according to DKD phenotype During a mean follow-up of 6.6 years, 5781 of 1,189,113 patients were hospitalized for HF. The incidence rate of HHF was highest in the PU−DKD group, followed by the PU+DKD and no-DKD groups (4.14, 2.64, and 0.60 per 1000 person-years among patients, respectively) (Additional file 1: Table S1). After age and sex adjustments, the risk of HHF was higher in the PU+DKD group than in the PU−DKD group (PU+DKD: HR = 4.25, $95\%$ CI 3.75–4.82; PU−DKD: HR = 2.46, $95\%$ CI 2.30–2.64 using a no-DKD group as the reference category). This difference remained consistent after adjusting for social factors (model 3) and comorbidities (model 4). Notably, after adjusting for factors associated with the severity of diabetes (model 5), the effect of DKD phenotypes on HHF persisted (PU+DKD: HR = 3.12, $95\%$ CI 2.75–3.55; PU−DKD: HR = 1.85, $95\%$ CI 1.73–1.99 using a no-DKD group as the reference category). ## Risk of HHF according to DKD/MAFLD phenotype The cumulative incidence of HHF was higher in the MAFLD− group than in the MAFLD+ group (log-rank test, $P \leq 0.001$) (Fig. 1). However, after age and sex adjustment, the risk of HHF was comparable to or higher in patients in the MAFLD+ group than in those in the MAFLD- group (Additional file 1: Table S2). In the final models adjusting for potential confounding factors, the risk of HHF was significantly higher in no-DKD patients with MAFLD than in those without MAFLD (Fig. 2). Contrarily, the risk of HHF was comparable regardless of MAFLD status between the PU+DKD and PU−DKD groups (Fig. 2).Fig. 1Cumulative incidence plot of hospitalization for heart failure according to the DKD/MAFLD phenotype. The black, blue, and red lines indicate the no-DKD, PU+DKD, and PU−DKD groups, respectively. The bold line indicates MAFLD and the dashed line indicates no MAFLDFig. 2HRs and $95\%$ CI of HHF according to DKD/MAFLD phenotype. DKD: diabetic kidney disease; HHF: hospitalization for heart failure; MAFLD: metabolic dysfunction–associated fatty liver disease; PU: proteinuria. no-DKD: normal eGFR (eGFR ≥ 60) with negative PU; PU+DKD: normal eGFR with positive PU; PU−DKD: reduced eGFR (eGFR < 60) with negative PU. HRs were adjusted for age, sex, smoking, alcohol, exercise, hypertension, dyslipidemia, atrial fibrillation, ischemic heart disease, fasting plasma glucose, diabetes duration, hemoglobin levels, and insulin usage ## Risk of HHF according to the FLI categories in the no-DKD group Previously, a lower FLI cutoff for diagnosing MAFLD has been suggested in the Korean population [20]. Consequently, we further analyzed the risk of HHF according to the three FLI categories (< 30 vs. 30–59 vs. ≥ 60). After adjustment for potential confounding factors, patients in the FLI > 60 and FLI 30–60 groups exhibited 1.5 times and 1.1 times, respectively, higher risk of HHF using an FLI < 30 group as reference (Additional file 1: Table S3). Additionally, subgroup analyses showed that the effect of FLI persisted regardless of age, sex, and the presence of previous HF (Fig. 3).Fig. 3Subgroup analyses among no-DKD patients stratified by age, sex, and previous HF. DKD: diabetic kidney disease; FLI: fatty liver index; HR: hazard ratio; HF: heart failure; IR: incidence rate. HRs were adjusted for age, sex, smoking, alcohol, exercise, hypertension, dyslipidemia, atrial fibrillation, ischemic heart disease, fasting plasma glucose, diabetes duration, hemoglobin levels, and insulin usage ## Discussion In this study, we observed that proteinuria and renal dysfunction were risk enhancers for HHF in patients with type 2 diabetes, which is consistent with previous studies [7]. In patients without proteinuria and reduced eGFR, MAFLD significantly increased the risk of HHF, suggesting that active diagnostic and interventional strategies should be provided for patients with diabetes, at least in those who concomitantly have DKD and/or MAFLD. Although the 2022 ADA consensus report on HF has mandated annual cardiac biomarker testing for all patients with diabetes [6], the prevalence of HF and the healthcare system in each country might influence guideline approval. For instance, Korea has a relatively low prevalence of HF ($1.53\%$) compared to Western countries (~ $2.2\%$) [21]. As a primary diagnosis, HF accounts for $0.78\%$ of all hospital admissions in Korea compared to $3.04\%$ in the USA [22]. Therefore, to identify patients at risk of HHF more precisely at the population level, we aimed to combine clinically available HHF risk enhancers, including fatty liver disease and CKD, to improve the implementation of guideline-directed medical therapy. Notably, growing evidence suggests that individuals with MAFLD are at a higher risk of CKD [23] or cardiovascular disease than those with NAFLD [24]. The prevalence of MAFLD in patients diagnosed with the FLI was reported as $28.4\%$ in a recent meta-analysis [25], and this is similar to the $25\%$ prevalence of MAFLD observed in our study. Interestingly, the prevalence of MAFLD was the highest in PU+DKD ($40\%$), followed by no-DKD ($25\%$) and PU−DKD ($20\%$). The highest MAFLD in PU+DKD is reminiscent of the severe insulin-resistant diabetes subtype, which is associated with an increased risk of fatty liver disease and macroalbuminuria [26]. Although the effects of NAFLD on CVD risk in patients with type 2 diabetes have been well established [27, 28], the association between NAFLD and incident HF has not been thoroughly explored. A meta-analysis showed that patients with NAFLD are $60\%$ more likely to develop HF [29]. Similarly, we found increased HHF risk in patients with type 2 diabetes and MAFLD compared with those with type 2 diabetes without MAFLD; however, this was only observed in the absence of DKD. In the no-DKD group, a higher FLI score was significantly associated with a higher risk of HHF, which is consistent across subgroups (men or women; age ≥ 65 or < 65 years; previous HF yes or no). CKD affects approximately $50\%$ of patients with type 2 diabetes globally [30]. In addition, patients with type 2 diabetes and CKD are more likely to have diabetes-related complications, including cardiovascular morbidity [31]. Consistent with previous results, the risk of HHF events increased significantly in patients with type 2 diabetes and CKD compared with those with type 2 diabetes without CKD. Of note, the risk of HHF was higher in PU+DKD than in PU−DKD group. Normoalbuminuric DKD has become a widely prevalent variant of renal impairment in diabetes. Women, older, and nonsmoking individuals with good glycemic control have a better chance of preserving normoalbuminuria, even in the case of declining renal function [32]. Normoalbuminuric DKD, despite of a more favorable option in terms of the risk of end-stage renal disease, was reported to associate with cardiovascular disease [33]. This unique group needs further clarification of its pathophysiology, and therapeutic targets since recent EMPA-KIDNEY outcomes also showed no benefit of adding sodium-glucose cotransporter 2 (SGLT2) inhibitors in this group [34]. As DKD is a risk factor for HF [6], we hypothesized that the presence of MAFLD would increase the risk of HF in this group [35]. The PU−DKD group showed a higher HHF risk when combined with MAFLD, although the difference was not statistically significant. Surprisingly, the PU+DKD group without MAFLD showed a higher HHF risk than those with MAFLD. Albuminuria has been associated with HF risk independent of eGFR [36]. Conversely, reduced eGFR was not significantly associated with incident heart failure at normal albuminuria levels [36]. Although MAFLD has potential mechanisms involved in HF risks, including low-grade inflammation [37], the direct effect of inflammation on the myocardium [38], and increased epicardial fat tissue [39, 40], proteinuria per se is a marker for generalized vascular endothelial dysfunction, which is likely to have a much stronger effect than that of fatty liver disease on HF [41]. Otherwise, insulin resistance might be a shared mechanism related to HF in patients with underlying DKD or MAFLD. Previously, Parente et al. reported that the waist-height ratio (WHR), a marker of central obesity [42], enhances the risk of HHF among patients with type 1 diabetes, regardless of proteinuria status [43] which is in contrast to our findings. There is a possibility that MAFLD did not affect patients with type 2 diabetes and DKD because they already had insulin resistance. However, it is expected that patients with type 1 diabetes have much less insulin resistance, which could explain the additive effect of WHR and DKD on the risk of HHF among patients with type 1 diabetes [42]. Of note, the effect of glucose-lowering drugs that modify the risk of HHF (e.g. SGLT2 inhibitors) were not calculated in this study. Because SGLT2 inhibitors are recommended to patients with DKD or HF due to its cardiorenal protective effects [44], DKD patients in our study might be exposed more to SGLT2 inhibitors than no-DKD patients that can lead to underestimate the risk of DKD on HHF. Interventional strategies for high-risk populations should also be considered. The Asia–Pacific Working Party on Nonalcoholic Fatty Liver Disease or the American Association for the Study of Liver Diseases recommends that pioglitazone be considered in patients with nonalcoholic steatohepatitis [45]. However, our data suggested that individuals with MAFLD or DKD should be cautious about initiating pioglitazone due to the possible risk of heart failure. In this regard, glucose-lowering drugs, including SGLT2 inhibitors or glucagon-like peptide 1 receptor agonists (GLP-1 RAs) that confer protection against major cardiovascular diseases, are promising for preventing HF in NAFLD or MAFLD [46]. To the best of our knowledge, this is the first study to examine the effect of DKD and/or MAFLD on HHF in patients with diabetes with a relatively mild or moderate risk of CVD. To overcome the evaluation of eGFR and proteinuria status at a single point in time, we only included subjects with a stable DKD status over a two-year interval. However, this study has some limitations. First, this study used claims data previously gathered for reimbursement purposes; thus, the diagnostic codes for certain patients might be incorrect. Second, another drawback is using a urinary dipstick test rather than a direct measure of urinary albumin excretion. The accuracy of the dipstick test may be affected by urine-specific gravity or pH [47]. In addition, it is not sensitive enough to detect microalbuminuria. Third, we could not adjust for several important variables (drugs affecting DKD or HF, laboratory tests for inflammation, and dietary habits except alcohol) and calculate indices such as AST to platelet ratio index or fibrosis-4 (FIB-4) index because of a lack of information. Fourth, we excluded the FLI score < 60 group as not having MAFLD, which might inadvertently categorize mild fatty liver disease as non-MAFLD. We also did not evaluate the advanced hepatic fibrosis or steatosis status. Recently, ADA guideline proposed algorithm using FIB-4 index for risk stratification in individuals with NAFLD or nonalcoholic steatohepatitis [48]. Further study should confirm the difference in fibrotic burden between MAFLD and its impact on HF outcomes. Finally, the clinical characteristics of patients with diabetes differ significantly across ethnic groups, and the results of this study cannot be directly applied to other ethnicities. ## Conclusion Our study’s results indicated that DKD and/or MAFLD increased the risk of HHF. In line with ADA’s HF guidelines, we suggest that annual cardiac biomarker testing should be conducted at least in patients with DKD or MAFLD. In addition, interventional strategies, including treatment with SGLT2 inhibitors and GLP-1 RA, should be considered to prevent HF and ultimately reduce HF-related morbidity and mortality. ## Supplementary Information Additional file 1: Fig. S1. Study population (A) and study design (B). Table S1. Incidence rate and risk of hospitalization for heart failure according to the DKD phenotype. Table S2. Incidence rate and risk of hospitalization for heart failure according to the DKD phenotype and MAFLD. Table S3. Incidence rate and risk of hospitalization for heart failure in the no-DKD group stratified by FLI. ## References 1. Park JH, Ha KH, Kim BY, Lee JH, Kim DJ. Trends in cardiovascular complications and mortality among patients with diabetes in South Korea. Diabetes Metab J (2021.0) 45 120-124. DOI: 10.4093/dmj.2020.0175 2. Jafari LA, Suen RM, Khan SS. Refocusing on the primary prevention of heart failure. Curr Treat Options Cardiovasc Med (2020.0) 22 13. DOI: 10.1007/s11936-020-00811-3 3. Savarese G, Lund LH. Global public health burden of heart failure. Card Fail Rev (2017.0) 3 7-11. DOI: 10.15420/cfr.2016:25:2 4. Levy D, Kenchaiah S, Larson MG, Benjamin EJ, Kupka MJ, Ho KK. Long-term trends in the incidence of and survival with heart failure. N Engl J Med (2002.0) 347 1397-1402. DOI: 10.1056/NEJMoa020265 5. Shah AD, Langenberg C, Rapsomaniki E, Denaxas S, Pujades-Rodriguez M, Gale CP. Type 2 diabetes and incidence of cardiovascular diseases: a cohort study in 1.9 million people. Lancet Diabetes Endocrinol (2015.0) 3 105-113. DOI: 10.1016/S2213-8587(14)70219-0 6. Pop-Busui R, Januzzi JL, Bruemmer D, Butalia S, Green JB, Horton WB. Heart failure: an underappreciated complication of diabetes. A consensus report of the American Diabetes Association. Diabetes Care (2022.0) 45 1670-1690. DOI: 10.2337/dci22-0014 7. Dhingra R, Gaziano JM, Djousse L. Chronic kidney disease and the risk of heart failure in men. Circ Heart Fail (2011.0) 4 138-144. DOI: 10.1161/CIRCHEARTFAILURE.109.899070 8. Park J, Kim G, Kim H, Lee J, Lee YB, Jin SM. The association of hepatic steatosis and fibrosis with heart failure and mortality. Cardiovasc Diabetol (2021.0) 20 197. DOI: 10.1186/s12933-021-01374-8 9. Lee H, Kim G, Choi YJ, Huh BW, Lee BW, Kang ES. Association between non-alcoholic steatohepatitis and left ventricular diastolic dysfunction in type 2 diabetes mellitus. Diabetes Metab J (2020.0) 44 267-276. DOI: 10.4093/dmj.2019.0001 10. Eslam M, Sanyal AJ, George J. MAFLD: a consensus-driven proposed nomenclature for metabolic associated fatty liver disease. Gastroenterology (2020.0) 158 1999-2014. DOI: 10.1053/j.gastro.2019.11.312 11. Mendez-Sanchez N, Bugianesi E, Gish RG, Lammert F, Tilg H, Nguyen MH. Global multi-stakeholder endorsement of the MAFLD definition. Lancet Gastroenterol Hepatol (2022.0) 7 388-390. DOI: 10.1016/S2468-1253(22)00062-0 12. Tsutsumi T, Eslam M, Kawaguchi T, Yamamura S, Kawaguchi A, Nakano D. MAFLD better predicts the progression of atherosclerotic cardiovascular risk than NAFLD: Generalized estimating equation approach. Hepatol Res (2021.0) 51 1115-1128. DOI: 10.1111/hepr.13685 13. Alharthi J, Gastaldelli A, Cua IH, Ghazinian H, Eslam M. Metabolic dysfunction-associated fatty liver disease: a year in review. Curr Opin Gastroenterol (2022.0) 38 251-260. DOI: 10.1097/MOG.0000000000000823 14. Berg DD, Wiviott SD, Scirica BM, Gurmu Y, Mosenzon O, Murphy SA. Heart failure risk stratification and efficacy of sodium-glucose cotransporter-2 inhibitors in patients with type 2 diabetes mellitus. Circulation (2019.0) 140 1569-1577. DOI: 10.1161/CIRCULATIONAHA.119.042685 15. 15.National Health Insurance Data Sharing Service: Korean National Health Insurance Service; 2021. https://nhiss.nhis.or.kr/bd/ab/bdaba000eng.do. 16. KDIGO 2020 clinical practice guideline for diabetes management in chronic kidney disease. Kidney Int (2020.0) 98 S1-S115. PMID: 32998798 17. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group. Ann Intern Med (1999.0) 130 461-470. DOI: 10.7326/0003-4819-130-6-199903160-00002 18. Eslam M, Newsome PN, Sarin SK, Anstee QM, Targher G, Romero-Gomez M. A new definition for metabolic dysfunction-associated fatty liver disease: an international expert consensus statement. J Hepatol (2020.0) 73 202-209. DOI: 10.1016/j.jhep.2020.03.039 19. Bedogni G, Bellentani S, Miglioli L, Masutti F, Passalacqua M, Castiglione A. The Fatty Liver Index: a simple and accurate predictor of hepatic steatosis in the general population. BMC Gastroenterol (2006.0) 6 33. DOI: 10.1186/1471-230X-6-33 20. Cho EJ, Jung GC, Kwak MS, Yang JI, Yim JY, Yu SJ. Fatty liver index for predicting nonalcoholic fatty liver disease in an asymptomatic Korean population. Diagnostics (2021.0) 11 2233. DOI: 10.3390/diagnostics11122233 21. Choi HM, Park MS, Youn JC. Update on heart failure management and future directions. Korean J Intern Med (2019.0) 34 11-43. DOI: 10.3904/kjim.2018.428 22. Ponikowski P, Anker SD, AlHabib KF, Cowie MR, Force TL, Hu S. Heart failure: preventing disease and death worldwide. ESC Heart Fail (2014.0) 1 4-25. DOI: 10.1002/ehf2.12005 23. Sun DQ, Jin Y, Wang TY, Zheng KI, Rios RS, Zhang HY. MAFLD and risk of CKD. Metabolism (2021.0) 115 154433. DOI: 10.1016/j.metabol.2020.154433 24. Lee H, Lee YH, Kim SU, Kim HC. Metabolic dysfunction-associated fatty liver disease and incident cardiovascular disease risk: a nationwide cohort study. Clin Gastroenterol Hepatol (2021.0) 19 2138-47.e10. DOI: 10.1016/j.cgh.2020.12.022 25. Chan KE, Koh TJL, Tang ASP, Quek J, Yong JN, Tay P. Global prevalence and clinical characteristics of metabolic-associated fatty liver disease: a meta-analysis and systematic review of 10 739 607 individuals. J Clin Endocrinol Metab (2022.0) 107 2691-2700. DOI: 10.1210/clinem/dgac321 26. Ahlqvist E, Storm P, Käräjämäki A, Martinell M, Dorkhan M, Carlsson A. Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables. Lancet Diabetes Endocrinol (2018.0) 6 361-369. DOI: 10.1016/S2213-8587(18)30051-2 27. Targher G, Lonardo A, Byrne CD. Nonalcoholic fatty liver disease and chronic vascular complications of diabetes mellitus. Nat Rev Endocrinol (2018.0) 14 99-114. DOI: 10.1038/nrendo.2017.173 28. Zhou YY, Zhou XD, Wu SJ, Hu XQ, Tang B, Poucke SV. Synergistic increase in cardiovascular risk in diabetes mellitus with nonalcoholic fatty liver disease: a meta-analysis. Eur J Gastroenterol Hepatol (2018.0) 30 631-636. DOI: 10.1097/MEG.0000000000001075 29. Salah HM, Pandey A, Van Spall HGC, Michos ED, McGarrah RW, Fudim M. Meta-analysis of nonalcoholic fatty liver disease and incident heart failure. Am J Cardiol (2022.0) 171 180-181. DOI: 10.1016/j.amjcard.2022.02.012 30. Thomas MC, Brownlee M, Susztak K, Sharma K, Jandeleit-Dahm KA, Zoungas S. Diabetic kidney disease. Nat Rev Dis Primers (2015.0) 1 15018. DOI: 10.1038/nrdp.2015.18 31. Chang YT, Wu JL, Hsu CC, Wang JD, Sung JM. Diabetes and end-stage renal disease synergistically contribute to increased incidence of cardiovascular events: a nationwide follow-up study during 1998–2009. Diabetes Care (2014.0) 37 277-285. DOI: 10.2337/dc13-0781 32. Thomas MC, Macisaac RJ, Jerums G, Weekes A, Moran J, Shaw JE. Nonalbuminuric renal impairment in type 2 diabetic patients and in the general population (national evaluation of the frequency of renal impairment cO-existing with NIDDM [NEFRON] 11). Diabetes Care (2009.0) 32 1497-1502. DOI: 10.2337/dc08-2186 33. Buyadaa O, Magliano DJ, Salim A, Koye DN, Shaw JE. Risk of rapid kidney function decline, all-cause mortality, and major cardiovascular events in nonalbuminuric chronic kidney disease in type 2 diabetes. Diabetes Care (2020.0) 43 122-129. DOI: 10.2337/dc19-1438 34. Herrington WG, Staplin N, Wanner C, Green JB, Hauske SJ. Empagliflozin in patients with chronic kidney disease. N Engl J Med (2023.0) 388 117-127. DOI: 10.1056/NEJMoa2204233 35. Mantovani A, Byrne CD, Benfari G, Bonapace S, Simon TG, Targher G. Risk of heart failure in patients with nonalcoholic fatty liver disease: JACC review topic of the week. J Am Coll Cardiol (2022.0) 79 180-191. DOI: 10.1016/j.jacc.2021.11.007 36. Blecker S, Matsushita K, Köttgen A, Loehr LR, Bertoni AG, Boulware LE. High-normal albuminuria and risk of heart failure in the community. Am J Kidney Dis (2011.0) 58 47-55. DOI: 10.1053/j.ajkd.2011.02.391 37. Anstee QM, Mantovani A, Tilg H, Targher G. Risk of cardiomyopathy and cardiac arrhythmias in patients with nonalcoholic fatty liver disease. Nat Rev Gastroenterol Hepatol (2018.0) 15 425-439. DOI: 10.1038/s41575-018-0010-0 38. Luo B, Li B, Wang W, Liu X, Xia Y, Zhang C. NLRP3 gene silencing ameliorates diabetic cardiomyopathy in a type 2 diabetes rat model. PLoS ONE (2014.0) 9 e104771. DOI: 10.1371/journal.pone.0104771 39. Emamat H, Tangestani H, Behrad Nasab M, Ghalandari H, Hekmatdoost A. The association between epicardial adipose tissue and non-alcoholic fatty liver disease: a systematic review of existing human studies. EXCLI J (2021.0) 20 1096-1105. PMID: 34345229 40. van Woerden G, Gorter TM, Westenbrink BD, Willems TP, van Veldhuisen DJ, Rienstra M. Epicardial fat in heart failure patients with mid-range and preserved ejection fraction. Eur J Heart Fail (2018.0) 20 1559-1566. DOI: 10.1002/ejhf.1283 41. Deckert T, Feldt-Rasmussen B, Borch-Johnsen K, Jensen T, Kofoed-Enevoldsen A. Albuminuria reflects widespread vascular damage. The Steno hypothesis. Diabetologia (1989.0) 32 219-226. DOI: 10.1007/BF00285287 42. Ashwell M, Hsieh SD. Six reasons why the waist-to-height ratio is a rapid and effective global indicator for health risks of obesity and how its use could simplify the international public health message on obesity. Int J Food Sci Nutr (2005.0) 56 303-307. DOI: 10.1080/09637480500195066 43. Parente EB, Harjutsalo V, Forsblom C, Groop PH. The impact of central obesity on the risk of hospitalization or death due to heart failure in type 1 diabetes: a 16-year cohort study. Cardiovasc Diabetol (2021.0) 20 153. DOI: 10.1186/s12933-021-01340-4 44. Fontes-Carvalho R, Santos-Ferreira D, Raz I, Marx N, Ruschitzka F, Cosentino F. Protective effects of SGLT-2 inhibitors across the cardiorenal continuum: two faces of the same coin. Eur J Prev Cardiol (2022.0) 29 1352-1360. DOI: 10.1093/eurjpc/zwab034 45. Ando Y, Jou JH. Nonalcoholic fatty liver disease and recent guideline updates. Clin Liver Dis (2021.0) 17 23-28. DOI: 10.1002/cld.1045 46. Yabiku K. Efficacy of sodium-glucose cotransporter 2 inhibitors in patients with concurrent type 2 diabetes mellitus and non-alcoholic steatohepatitis: a review of the evidence. Front Endocrinol (2021.0) 12 768850. DOI: 10.3389/fendo.2021.768850 47. White SL, Yu R, Craig JC, Polkinghorne KR, Atkins RC, Chadban SJ. Diagnostic accuracy of urine dipsticks for detection of albuminuria in the general community. Am J Kidney Dis (2011.0) 58 19-28. DOI: 10.1053/j.ajkd.2010.12.026 48. ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D. 4. Comprehensive medical evaluation and assessment of comorbidities: standards of care in diabetes-2023. Diabetes Care (2023.0) 46 S49-S67. DOI: 10.2337/dc23-S004
--- title: Expression pattern of miR-193a, miR122, miR155, miR-15a, and miR146a in peripheral blood mononuclear cells of children with obesity and their relation to some metabolic and inflammatory biomarkers authors: - Maryam Behrooz - Samaneh Hajjarzadeh - Houman Kahroba - Alireza Ostadrahimi - Milad Bastami journal: BMC Pediatrics year: 2023 pmcid: PMC9976520 doi: 10.1186/s12887-023-03867-9 license: CC BY 4.0 --- # Expression pattern of miR-193a, miR122, miR155, miR-15a, and miR146a in peripheral blood mononuclear cells of children with obesity and their relation to some metabolic and inflammatory biomarkers ## Abstract ### Background The widespread presence of childhood obesity has increased considerably over three decades. The present study was designed to investigate expression patterns of miR-146a, miR-155, miR-15a, miR-193a, and miR-122 in peripheral blood mononuclear cells (PBMCs) in children who are obese along with their association with metabolic and inflammatory biomarkers. ### Methods Ninety test subjects were admitted. The profile of blood pressure, resting energy expenditure (REE), anthropometric measures, body composition, dietary intakes, physical activity levels, insulin, and lipid profile, fasting blood glucose (FBG), high-sensitivity C-reactive protein (hs-CRP), and pubertal stage have been measured. Total RNA (including small RNAs) was extracted from PBMCs. The expression levels of miRNAs were measured by stem-loop RT-qPCR. ### Results The miR-155a expression level was significantly lower in obese children, children with high hs-CRP, and children with high-fat mass. Obese girls had significantly higher PBMC levels of miR-122. MiR-155a had a significant negative association with fasting insulin, HOMA-IR, and hs-CRP. There were significant positive associations between miR-193a and miR-122 expression levels and fasting insulin, HOMA-IR, and TG. MiR-15a was positively correlated with fasting insulin and HOMA-IR. Children with metabolic syndrome, insulin resistance, and high-fat mass had higher PBMC levels of miR-122 and miR-193a. Higher miR-193a and miR-122 levels were also detected in PBMCs of children with fast REE, compared to those with slow REE, and the subjects with high hs-CRP, respectively. ### Conclusion lower level of miR-155 expression in obese subjects and significant associations unfolds the need for more studies to detect the possible underlying mechanisms. ## Introduction Obesity is a by-product of interactions among many environmental and genetic factors [1]. The common occurrence of childhood and adolescent obesity has seen a dramatic increase over the past three decades, turning it into a global health issue [2–4]. Obesity is strongly associated with insulin resistance, chronic adipose tissue inflammation, and hyperlipidemia. Undeniably, childhood and adolescent obesity increases the risk of chronic disorders such as cardiovascular diseases, dyslipidemia, metabolic syndrome, and diabetes mellitus in adulthood [5, 6]. While some obese children are metabolically healthy (MHO), others show abnormalities ranging from mild chronic inflammation to insulin resistance [7, 8]. It is estimated about 20 to $30\%$ of obese children and adolescents are metabolically healthy [7, 9]. In most studies, metabolically healthy obesity has been suggested to be defined as obesity with no indication for associated metabolic disorders, such as dyslipidemia, metabolic syndrome. etc. [ 10]. Various hypotheses have been raised to explain the underlying mechanisms of this bipartite among obese children, either being metabolically healthy or suffering from obesity-related metabolic disorders, but more investigations are needed to achieve a consensus about the exact mechanism [11]. There is an urgent need for discovering biomarkers that help in the early detection and prognosis of high-risk individuals and the prevention of childhood obesity-related disorders [3]. MicroRNAs (i.e. miRNAs), recognized as promising biomarkers for various diseases [12]. are short non-coding molecules (about 22 nucleotides in length) that play crucial roles in regulating gene expression in the post-transcription step by binding to 3′-UTR-RNA regions, and inhibiting mRNA translation or degradation [2, 4, 13–15]. Being sustainable and measurable, circulating miRNAs in body fluids, especially blood, have been extensively used as a biomarker for the diagnosis of disorders [16–19]. In addition, several studies have suggested that circulating miRNA concentrations are associated with various metabolic diseases such as obesity, hyperlipidemia, and type 2 diabetes [20–22]. For example, a higher level of miR-122 is coincidental with the increased risk of obesity and insulin resistance in young adults [23, 24]. It correlates with an increased risk of metabolic syndrome, type 2 diabetes, and liver damage [25]. Previous studies have reported that miR-193 blood level was higher in obese and pre-diabetic adults [26]. In addition, miR-15 is reduced in type 2 diabetes, metabolic syndrome, and obesity, and is used as a biomarker to diagnose obese children at risk of type 2 diabetes henceforward [2]. MiRNAs are involved in adipogenesis, obesity, lipid and fatty acid metabolism, insulin resistance, adipocyte differentiation, appetite regulation, inflammation, oxidative stress, and cytokines expression [27, 28]. For example, miR-146 plays a crucial role in adipose tissue inflammation and is one of the most essential and effective mediators contributing to obesity and subsequent problems [29]. MiR-155 is involved in the metabolism of cholesterol and fatty acids in the liver and regulates many of the genes related to fat metabolism [30, 31]. It speculates that expression patterns of miRNAs in obese children could be used as a predictor of metabolic diseases in adulthood. miRNA expression profile of obese children, relative to others, has not been extensively investigated. The current study specified to compare the expression levels of miR-193a, miR-122, miR-155, miR-15a, and miR-146a in peripheral blood mononuclear cells (PBMCs) of 10 to 18 years-old obese children and adolescents with matched normal-weight subjects. Also, the association of these levels with metabolic conditions and inflammatory parameters is investigated. ## Study subjects First of all, the methodology and objectives of the study were explained to all of the participants and their parents; then they signed the informed consent form completely optional. 45 obese and 45 normal-weight 10 to 18 years old participants were involved in the present cross-sectional study. The criteria for matching the obese and normal-weight groups were the age and sex of the participants. Based on the 2–20 years old growth charts of CDC, children, and adolescents with a 5 ≤ BMI < 85 percentile and with a BMI ≥ 95th percentile were categorized as normal-weight and obese groups, respectively [32]. The exclusion criteria for the current study were unwillingness to enroll in the study for any reason, history of using medications including steroids, antiepileptic, and anti-psychotic medications, and history of chronic systemic diseases, including gastrointestinal, respiratory, cardiovascular, or neurologic disease, and being involved with obesity-related syndromes including Prader-Willi. ## Sample size For sample size calculation, we used Mahdavi et al. study’s data [33]. Based on the reported correlation between mir-155 and BMI (r = -0.31). The final calculated sample size was 90 persons (refusal rate = $15\%$) α (two-tailed) = 0.05 β = 0.2 r = -0.31 $$n = 79$.$ ## Physical and anthropometric variables The puberty status of children and adolescents was evaluated using a provided form of the Tanner stages (consisting of the pictures of five stages). The research team clarified the form to the participants; then they were asked to choose the proper stage that portrayed their physique in the most proper way. Weight, height, hip circumference (HC), waist circumference (WC), BMI (Body Mass Index), and blood pressure (systolic and diastolic) were measured after fasting for 12 h overnight. Waist to hip ratio (WHR) was calculated. Weight (Kg) was divided by height (m) square for calculating BMI. Body composition was analyzed using Tanita MC-780 S MA (Amsterdam, the Netherlands). Physical activity levels of children and adolescents were assessed by MAQ (Modifiable Activity Questionnaire). A previous study on an adolescent population has reported moderate validity and high reliability for the Persian translation of the MAQ in this age range [34]. ## Dietary assessment To evaluate participants’ dietary intakes, trained interviewers used the 168 Quantitative Food Frequency Questionnaire. The validity and reliability of the questionnaire have been previously studied [35, 36]. Energy intake per day for every participant was calculated using Nutritionist IV software (version 3.5.2). ## Resting energy expenditure (REE) estimation Indirect Calorimetry using Fitmate Pro., Rome, Italy was done to measure REE (VO2 and VCO2) between 06:00 to 08:00 AM. For more accuracy, doing exercise during the day before and eating any food from 21:00 the night before the visit was forbidden for all of the children and adolescents. ## Biochemical measurements After overnight fasting of 12 hours, blood samples were taken. Levels of fasting insulin, FBG (Fasting Blood Glucose), lipid profile, and hs-CRP (high-sensitivity C - reactive protein) were measured for every participant. Enzymatic methods and colorimetry techniques were utilized for measuring serum HDL-C (High-Density Lipoprotein Cholesterol), TG (triglyceride), and FBG. The commercial kits for the calorimetry technique were provided by Pars-Azmoon Co., Tehran, Iran. ELISA method (Monobind, Lake Forest, CA, USA) was used to estimate insulin levels. Hyperinsulinemia was distinguished as following cut-off levels: higher than 30 μU/mL in the pubertal period, 20 μU/mL in the post-pubertal period, and 15 μU/mL in the pre-pubertal period [37, 38]. The equation below was used to calculate HOMA-IR [35]. ( HOMA-IR= [fasting glucose (mg/dl) × fasting insulin (µIU/ml)]/405). Following cut-offs were considered for HOMA-IR to discern insulin resistance: ≥2.67 (specificity $65.5\%$, sensitivity $88.2\%$) and ≥ 2.22 (specificity $42.3\%$T sensitivity $100\%$) in boys and girls in the pre-pubertal stage respectively; ≥ 5.22 (specificity $93.3\%$, sensitivity $56\%$) and ≥ 3.82 (specificity $71.4\%$, sensitivity $77.1\%$) in boys and girls in the pubertal stage respectively [34]. Friedewald equation was employed for calculating LDL-C (Low-Density lipoprotein Cholesterol). Assessing hs-CRP levels in serum was done using the turbidimetric method. Based on gender and age, hs-CRP Reference values were spotted; Normal values in serum (mg/liter) including: (Boys: < 1.45 in 5–13 years old; < 2.13 in 14–18 years old) and (Girls: < 1.90 in 5–13 years old; < 3.33 in 14–18 years old). The subjects with MetS were diagnosed based on Cook et al. [ 39]. criteria, as follows: 1: HDL-C < 40 mg/dL. 2: TGs ≥ 110 mg/dL. 3: FBG ≥ 110 mg/dL. 4: WC ≥ 90th percentile for age and sex. 5: SBP and DBP ≥ 90th percentile for sex, age, and height. The subjects were categorized in the MetS group if they had at least three of the five mentioned criteria. ## miRNA extraction from PBMCs A panel of 5 miRNAs that have previously been shown to correlate with obesity and markers of metabolic disease mainly in adults or experimental models was selected [2, 24, 26, 29, 31]. Using density gradient centrifugation (Ficoll-Paque PLUS, Amersham Pharmacia Biotech, Sweden), PBMCs were isolated from ~ 4 ml peripheral blood. Phosphate-buffered saline (PBS) was used for washing isolated PBMCs. Total RNA including small RNAs was extracted by Mini Kit of miRNeasy (Qiagen, Germany). The standard stem-loop miRNA RT-qPCR method [35] was employed to synthesize the first-strand cDNA of the studied miRNAs (hsa-miR-155-5p, hsa-miR-193a-3p, hsa-miR-15a-5p, hsa-miR-122-5p, hsa-miR-146a-5p). Reverse transcriptions were performed in 10 μl reactions and the following program: 30 min at 16 °C, 30 min at 42 °C, and 10 min at 75 °C Table 1. presents the sequence of primers that were exploited in reverse transcription and qPCR reactions. Table 1Sequence of primers that were utilized in reverse transcription and qPCR reactionsgenePrimer typePrimer sequencemiR-155FCGGCGCTTAATGCTAATCGTGATAGRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACCCCTRGTGCAGGGTCCGAGGTmiR-146aFGTGTGGGTGAGAACTGAATTCCATGRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAACCCARGTGCAGGGTCCGAGGTmiR-122FGGCTGGAGTGTGACAATGGTRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCAAACARGTGCAGGGTCCGAGGTmiR-193aFAGCGAACTGGCCTACAAAGTCRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACTGGGRGTGCAGGGTCCGAGGTmiR-15aFGGCGTAGCAGCACATAATGGTRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCACAAARGTGCAGGGTCCGAGGTD24FCGCTATCTGAGAGATGGTGATGACATTRTGTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACTGCATCAGRGTGCAGGGTCCGAGGT ## Quantitative real-time PCR Light cycler 96 (Roche Diagnostics, Mannheim, Germany) was used to perform quantitative Real-Time PCR (qPCR) reactions using a 2 × qPCR master mix (RealQ Plus 2 × Master Mix Green without ROX, Ampliqon). Each 10 μl qPCR reaction contained the following components: 2 μl of forwarding primer (5 pmol/μl concentration), 0.5 μl of reverse primer, and, 5 μl master mix (RealQ Plus 2 × Master Mix Green without ROX, Ampliqon), 1 μl cDNA, 2 μl nuclease-free water and 1.5 μl dNTP. The qPCR reactions were performed in the following program: 15 min at 95 °C, followed by 40 cycles of 10 s at 95 °C and 30 s at 60 °C. At the end of qPCR cycles, the specificity of the reactions was evaluated by interpreting the melting curve step. To evaluate relative quantifications, the framework of Hellemans et al. Was used [40]. For a gene of interest (GOI) in each sample, ΔCq was equal to (mean Cq (GOI) in [all samples]) minus (Cq(GOI) in [that sample]). Then, relative quantities (RQs) were computed for each sample-GOI as RQ equals E(ΔCq), in which “E” was the mean efficiency of qPCR reactions for each gene. Normalized relative quantity (NRQ) for each sample-GOI was then calculated as NRQ (GOI) equals RQ (GOI) divided by the geometric mean of RQ values of the reference genes. NRQ values were converted to the Cq′ values by log2 transformation for the statistical analysis [41]. For normalizing miRNA quantities, the RQ value RNU24 was used. RNU24 is one of the stable small RNAs in PBMC which has previously been evaluated and used for normalizing miRNA quantities in PBMC in other studies [42, 43]. ## Statistical analysis SPSS Version 23.0 statistical software (SPSS Inc., Chicago, IL, USA) was used for statistical analysis. $P \leq 0.05$ was considered statistically significant. Quantitative variables are reported as mean (SD) or median (interquartile range [IQR]), as appropriate. Categorical variables are presented as numbers and percentages. Continuous variables that failed the normality test were logarithmically transformed before analysis. Non-parametric tests were used for the data that still did not have a normal distribution after this transformation. To compare differences between groups, independent samples T-test, Mann–Whitney U, or Kruskal–Wallis test were used. Spearman rank correlation coefficient (R) was used to test the association between selected miRNAs expression levels with some metabolic and inflammatory biomarkers. ## Results Anthropometric, demographic, and laboratory characteristics of obese and normal-weight children were compared and the results were shown in Table2. There was no significant difference between the two groups in age, gender, and pubertal stage variables ($P \leq 0.05$); but the mean birth weight of obese children was significantly lower than normal-weight children ($$P \leq 0.03$$). As expected, the mean values of weight, BMI, waist circumference, hip circumference, and WHR of obese cases were higher than the control group ($p \leq 0.001$).Table 2Baseline demographic, anthropometric and laboratory characteristics of participantsCharacteristicsNormal weight children($$n = 45$$)Mean (SD)Children with obesity($$n = 45$$)Mean (SD)@P-valueAge (years)13.73(3.02)13.92(2.98)0.83Gender $Boy22[49]22[49]1.00Girl23[51]23[51]Pubertal stage $Pre pubertal7(15.5)5(11.11)0.84pubertal26(57.8)24(53.33)Post pubertal12(26.7)16(35.56)Birth weight (gr)3402.71(497.29)3122.03(715.12)0.03Weight (kg)48.57(16.81)75.90(21.65) < 0.001Height (cm)158.47(18.09)160.00(12.92) < 0.001BMI (kg/m2)18.57(3.02)29.38(5.28) < 0.001Hip circumference (cm)85.90(10.87)101.56(21.95) < 0.001Waist circumference (cm)71.40(12.68)96.43(13.96) < 0.001WHR 0.83(0.10)0.91(0.11) < 0.001Total Energy Intake1970.06(734.03)2039.33(781.08)0.42Physical Activity (MET-h/week)14.15(4.31)15.03(5.25)0.37WBC (109 /L) 6.38(1.38)6.97(1.11)0.01RBC (16/mcL) 5.01(0.44)5.07(0.43)0.58HB (gr/dL)13.29(1.48)13.79(1.20)0.13HTC (%)40.68(3.38)41.54(3.07)0.27MCV (fl)81.46(4.82)82.02(3.87)0.54MCHC(g/dL)26.93(2.20)27.17(1.71)0.62PLAT (103/μl) 247.86(57.33)272.3(45.79)0.06@Independent Sample Test or Mann–Whitney u Test for Quantitative variables and Chi-square test for categorical variables. * Median (IQR). $N (%) Obese children had significantly higher SBP, DBP, TC ($P \leq 0.001$), LDL-C, TG (P Z 0.001), Insulin, HOMA-IR, hs-CRP, and significantly Lower HDL-C ($P \leq 0.001$). The results related to these parameters are given in detail in our previous article [44]. Among laboratory indices, WBC count was higher in obese children in comparison to normal-weight children ($$P \leq 0.01$$). Considering the effect of WBC count on the miRNA’s expression levels, we divided obtained miRNAs expression levels by the average WBC count in each group (obese and normal children). So, selected miRNAs expression levels per unit of WBC count among children with and without obesity are estimated and presented in Table 3. Among the five studied miRNAs, the miR-155 expression level was significantly lower in obese children compared to the normal-weight group (P-value = 0.01). After disaggregating the data by gender, the miR-155 expression level difference was significant in girls (P-value = 0.03), but not in boys. Although the expression levels of miR-122 and miR-193a were higher in obese children than in normal-weight children, the differences were not statistically significant (P-values = 0.07). Considering the gender distribution, only the difference between obese and normal girls became significant for miR-122 (P-value = 0.04). The expressions of the other two miRNAs did not differ between the two groups Table 4. presents the association of the five studied miRNAs expression levels with serum levels of insulin, HOMA-IR, fasting blood glucose, hs-CRP, IL-1ß, IL-10, and lipid profile. There was a significant positive association between the expression level of miR-122 and fasting insulin, HOMA-IR, hs-CRP, and TG ($P \leq 0.05$). The expression level of miR-193a was associated with fasting insulin, HOMA-IR, and TG level ($P \leq 0.05$) significantly and directly. MiR-155 showed a significant negative association with fasting insulin, HOMA-IR, and TG level ($P \leq 0.05$). MiR15a expression had a significant association with fasting insulin and HOMA-IR, ($P \leq 0.05$) Table 5, 6, and 7. present the comparison of miR-193a, mir122, and mir155 expression levels among children with and without some metabolic status (including obesity, metabolic syndrome indices, and different levels of REE, different body composition, hyper-insulinemia, insulin resistance, and high hs-CRP respectively. The median (IQR) expression level of miR-193a was significantly higher in children with metabolic syndrome, high TG, larger WC, insulin resistance, and high-fat mass (for a review of P-values between two groups, cf Table 5). There was also a significant difference in the expression level of miR-193a among children with slow REE and fast REE ($$P \leq 0.008$$). We couldn’t find any significant difference in miR-193a expression level of children with increased FBG, hyperinsulinemia, higher SBP/DBP, lower HDL, and higher hs-CRP with children who were in the normal spectrum for these biomarkers. Also, the expression level of miR-193a was not significantly different among children with different muscle mass. Table 3Selected miRNAs expression levels among children with and without obesityGeneNRQchildren with obesityNormal weight children†P- valuemiR-193aTotal1.01 (0.81)0.64 (0.68)0.07Boy1.11(0.79)0.49(0.31)0.06Girl1.00(0.98)0.69(0.82)0.40miR-122Total0.21 (3.11)0.06 (0.28)0.07Boy0.146(1.92)0.06(0.16)0.65Girl0.86(1.53)0.07(1.08)0.04miR-155Total0.17 (0.43)0.37 (0.61)0.01Boy0.29(0.51)0.47(0.66)0.21Girl0.06(0.15)0.34(0.59)0.03miR-15aTotal0.10 (0.29)0.08 (0.29)0.17Boy0.12(0.55)0.07(0.14)0.14Girl0.08(0.21)0.09(0.52)0.74miR-146aTotal0.09 (0.22)0.10 (0.14)0.66Boy0.08(0.16)0.08(0.12)0.95Girl0.12(0.51)0.11(0.22)0.70†Mann–Whitney u Test,NRQ Normalized relative quantity, IQR Inter Quartile RangeTable 4The associations between expression levels of selected miRNAs and fasting insulin, HOMA-IR, fasting blood glucose, lipid profile and some inflammatory biomarkers in childrenParametersmiR-122r(P-value)miR-193ar(P-value)miR-155r(P-value)miR-15ar(P-value)miR-146ar(P-value)Insulin0.38(0.005)0.40(0.003)-0.37(0.001)0.22(0.04)0.15(0.19)HOMA-IR0.38(0.004)0.40(0.004)-0.36(0.001)0.23(0.04)0.15(0.20)FBS0.09(0.44)-0.14(0.22)0.08(0.48)0.05(0.65)-0.08(0.49)TG0.28(0.03)0.38(0.04)0.15(0.18)0.09(0.42)0.07(0.56)TC0.05(0.67)-0.02(0.80)0.18(0.10)-0.02(0.82)-0.11(0.34)HDL-c-0.02(0.87)-0.07(0.52)-0.01(0.92)-0.03(0.75)-0.01(0.93)LDL-c-005(0.66)-0.10(0.37)0.16(0.16)-0.02(0.80)-0.10(0.43)IL-10-0.10(0.38)-0.12(0.28)-0.03(0.78)0.02(0.84)0.6(0.59)IL-1ß-0.06(0.61)-0.14(0.22)-0.11(0.31)0.01(0.91)-0.06(0.60)hs-CRP0.28(0.03)0.22(0.20)-0.25(0.02)0.20(0.08)0.04(0.69)Spearman rank correlation coefficientFBS Fasting Blood Glucose, TG Triglyceride, TC Total Cholesterol, HDL-C High Density Lipoprotein Cholesterol, LDL-C Low Density Lipoprotein Cholesterol, IL-10 Interlukin-10, IL-1ß Interlukin-1ß, hs-CRP high-sensitivity C Reactive ProteinTable 5The comparison of miR-193a expression level among children with and without some metabolic statusGroupsNMedian (IQR)@P-valueGroupsNMedian (IQR)@P-valueObesity0.07Muscle Mass Range0.09No450.64(0.68)Low410.56 (0.71)Yes451.01(0.81)Normal240.70 (0.59)Metabolic syndrome0.003High251.13 (0.82)No760.62(0.64)Fat Mass0.05Yes141.31(0.42)Normal550.67(0.64)&High TG0.05high351.11(0.85)No500.64(0.65)$$REE0.008Yes401.06(0.76)Slow10^1.11(0.49)&Low HDL0.07Normal590.59(0.65)No670.68 (0.64)Fast21^1.61(0.49)Yes231.21 (0.80)High hs-CRP0.23&Large WC0.01No600.70 (0.88)No670.64(0.64)yes290.85 (0.73)Yes231.31(0.75)%Hyper-insulinemia0.38&High SBP0.32No800.70(0.84)No800.70 (0.68)Yes100.80(0.87)Yes101.08 (1.12)#Insuline Resistance0.04&High DBP0.37No760.70(0.72)No390.70 (0.84)Yes141.13(1.10)Yes511 (0.83)@Mann–Whitney u Test or Kruskal–Wallis test /^P-value = 0.05&Normal Range according to metabolic syndrome definition in childrenDesired range reported in each person's body analyzer result sheet, was considered for classification of muscle mass (low-normal-high) and fat mass (low-normal-high)$$ Classification for REE (slow-normal-fast) according to desirable ranges reported for each child in Indirect Calorimeter result sheeths-CRP Reference Value according to sex and age. Normal value in serum (mg/liter): Boys: 5–13 year < 1.45 and 14–18 year < 2.13/Girls: 5–13 year < 1.90 and 14–18 year < $3.33\%$Fasting insulin levels above 15 μU/mL in the pre-pubertal period, 30 μU/mL in the pubertal period and 20 μU/mL in the post-pubertal period are cut-off levels for hyper insulinemia# HOMA-IR cut-off values for insulin resistance were calculated to be 2.67 (sensitivity $88.2\%$, specificity $65.5\%$) in boys and 2.22 (sensitivity $100\%$, specificity $42.3\%$) in girls in the prepubertal period, and 5.22 (sensitivity $56\%$, specificity $93.3\%$) in boys and 3.82 (sensitivity $77.1\%$, specificity $71.4\%$) in girls in the pubertal periodTable 6The comparison of miR-122 expression level among children with and without some metabolic statusGroupsNMedian (IQR)@P-valueGroupsNMedian (IQR)@P-valueObesity0.08Muscle Mass Range0.13No450.06 (0.28)Low410.05(0.33)Yes450.21(3.11)Normal240.16(2.00)Metabolic syndrome0.008High250.21(3.95)No760.06(0.33)Fat Mass0.04Yes141.07(13.17)Normal550.06(0.25)&High TG0.17high350.27(2.20)No500.06 (0.88)$$REE0.09Yes400.22 (3.98)~Slow100.57(2.75)&Low HDL0.27~Normal590.07(0.31)No670.08 (0.88)~Fast210.09(18.32)Yes230.22 (3.98)High hs-CRP0.04&Large WC0.03No600.06(0.70)No670.21 (1.02)Yes290.31(3.07)Yes230.96 (1.56)%Hyper-insulinemia0.30&High SBP0.55No800.09 (1.39)No800.08(1.25)Yes100.21 (4.86)Yes100.24(14.6)#Insulin Resistance0.06&High DBP0.62No760.08(1.22)No390.091(1.53)Yes140.27(18.29)Yes510.02(1.35)@Mann–Whitney u Test or Kruskal–Wallis test& Normal Range according to metabolic syndrome definition in childrenDesired range reported in each person's body analyzer result sheet, was considered for classification of muscle mass (low-normal-high) and fat mass (low-normal-high)$$ Classification for REE (slow-normal-fast) according to desirable ranges reported for each child in Indirect Calorimeter result sheeths-CRP Reference Value according to sex and age. Normal value in serum (mg/liter): Boys: 5–13 year < 1.45 and 14–18 year < 2.13/Girls: 5–13 year < 1.90 and 14–18 year < 3.33# HOMA-IR cut-off values for insulin resistance were calculated to be 2.67 (sensitivity $88.2\%$, specificity $65.5\%$) in boys and 2.22 (sensitivity $100\%$, specificity $42.3\%$) in girls in the prepubertal period, and 5.22 (sensitivity $56\%$, specificity $93.3\%$) in boys and 3.82 (sensitivity $77.1\%$, specificity $71.4\%$) in girls in the pubertal periodTable 7The comparison of miR-155 expression level among children with and without some metabolic statusGroupsNMedian (IQR)@P-valueGroupsNMedian (IQR)@P-valueObesity0.01Muscle Mass Range0.13No450.37(0.61)Low410.4(0.54)Yes450.17 (0.43)Normal240.30(0.58)Metabolic syndrome0.10High250.18(0.39)No760.33(0.52)Fat Mass0.03Yes140.08(0.32)Normal550.36(0.60)&High TG0.26high350.11(0.44)No500.35(0.53)$$REE0.59Yes400.18(0.54)~Slow100.17(0.42)&Low HDL0.51~Normal590.33(0.52)No670.34(0.52)~Fast210.33(0.53)Yes231.00(0.54)High hs-CRP0.02&Large WC0.29No600.34(0.54)No670.36(0.54)Yes290.10(0.33)Yes230.17(0.39)%Hyper-insulinemia0.18&High SBP0.55No800.33(0.52)No800.34(0.52)Yes100.09(0.16)Yes100.19(0.30)#Insulin Resistance0.06&High DBP0.62No760.35(0.52)No390.32(0.52)Yes140.06(0.16)Yes510.12(0.60)@Mann–Whitney u Test or Kruskal–Wallis test& Normal Range according to metabolic syndrome definition in childrenDesired range reported in each person's body analyzer result sheet, was considered for classification of muscle mass (low-normal-high) and fat mass (low-normal-high)$$ Classification for REE (slow-normal-fast) according to desirable ranges reported for each child in Indirect Calorimeter result sheeths-CRP Reference Value according to sex and age. Normal value in serum (mg/liter): Boys: 5–13 year < 1.45 and 14–18 year < 2.13/Girls: 5–13 year < 1.90 and 14–18 year < 3.33# HOMA-IR cut-off values for insulin resistance were calculated to be 2.67 (sensitivity $88.2\%$, specificity $65.5\%$) in boys and 2.22 (sensitivity $100\%$, specificity $42.3\%$) in girls in the prepubertal period, and 5.22 (sensitivity $56\%$, specificity $93.3\%$) in boys and 3.82 (sensitivity $77.1\%$, specificity $71.4\%$) in girls in the pubertal period The median (IQR) expression level of miR-122 was higher in children with metabolic syndrome, high-fat mass, high hs-CRP, larger WC, and insulin resistance significantly (for a review of P-values between two groups, cf Table 6). We found no significant difference in the miR-122 expression level of children with increased FBG, higher TG, higher SBP-DBP, lower HDL, hyperinsulinemia, and children with different REE with children who were in the normal range for these factors. Also, the expression level of miR-122 was not significantly different among children with different muscle mass. The median (IQR) expression level of miR-155 was significantly lower in children with high hs-CRP and high-fat mass compared to the normal group (for a review of P-values between two groups, cf Table 7). Similar analyses were done for miR-15a and miR-146 expression levels; we find no significant association between miR-15a and miR-146 with metabolic status, so the data were not reported. ## Discussion Circulating miRNA concentrations have been suggested to be associated with a variety of metabolic diseases such as obesity and type 2 diabetes [20, 21, 45, 46]. Obesity-related miRNAs have been called potential biomarkers for the prevention and diagnosis of obesity and obesity-related metabolic disorders [46]. Obtained evidence also indicated that obesity-related miRNAs are promising new therapeutic tools for curing obesity and related diseases [46]. The current cross-sectional study was performed to compare the expression levels of miR-193a, miR-122, miR-155, miR-15a, and miR-146a, in peripheral blood mononuclear cells of obese children and adolescents with normal weight groups. The relation of expressed miRNAs with fasting insulin, HOMA-IR, fasting blood glucose, lipid profile and some inflammatory biomarkers were also assessed. Furthermore, the relation of differentially expressed miRNAs with some metabolic status (including metabolic syndrome, different body composition, hyper-insulinemia, insulin resistance, REE and high hs-CRP was evaluated. Among the five studied miRNAs, obese children significantly had lower miR-155 than normal-weight children. After disaggregating the data by gender, the observed relationship was in place only for girls. MiR-155 had a negative association with fasting insulin, HOMA-IR, and hs-CRP. Mahdavi et al. showed that obese non-diabetic subjects had lower serum levels of miR-155 than normal-weight non-diabetic individuals [33]. Mazloom et al. reported that the expression of miR-155 in PBMC cells of diabetic patients was reduced compared to the control group [47]. They also reported a negative association between miR-155 and BMI, serum cholesterol, fasting insulin, HOMA-IR, and WC in the diabetic group [47]. According to previous studies, miR-155 plays a crucial role in the metabolism of cholesterol, and fatty acids in the liver through a direct effect on the regulatory factor X receptor alpha (LXRa). LXRa is involved in the regulation of many genes that contribute to fat metabolism [30, 31]. In animal models, reduction of miR-155 levels increases the risk of NAFLD in diabetic patients. In return, the miR-155 increase has a protective role in slowing NAFLD progression [31]. In line with these observations, in our study, the miR-155 was significantly lower in children with high fat-mass and insulin resistance compared to normal children. In the present study, although the expression levels of miR-193a and miR-122 were higher in obese children, these observations were statistically significant only for Mir122 in girls. We found no previous study that compared the miR-193a expression level of obese children and adolescents with normal-weight controls, but similar studies had been done on adult participants. Some studies reported blood levels of miR-193a and miR-122 are elevated in participants with obesity, pre-diabetes, diabetes, insulin resistance, liver damage, and cardiovascular disorders significantly [48–54]. The possible reasons for this inconsistency could be due to different participant ages, the low sample size, and the metabolic phenotype of the studied children than the other studies. In this study, $55\%$ of obese children did not have any obesity associated metabolic disorders like as metabolic syndrome and insulin resistance, While this percentage has been reported as 20-$30\%$ in other previous studies [7, 9]. Hess et al. [ 2020] assessed the serum levels of some miRNAs before and after a weight loss diet [55]. They showed that serum levels of miR-122 and miR-193a were reduced after an average weight loss of 5.7 Kg. They also demonstrated the levels of these miRNAs were positively correlated with metabolic syndrome, serum insulin, HOMA-IR, BMI, waist circumference, lean body mass, and visceral adipose tissue at the baseline. MiR-122 and miR-193a also had a positive correlation with TG level and fat mass, respectively [55]. In a recent study, serum levels of miR-29a and miR-122 were compared between obese with T2DM, obese without T2DM, and normal-weight children [56]. They also reported the positive relationship between the assessed miRNAs and waist circumference, BMI, TG, insulin, HOMA-IR, IL-6, hs-CRP, and TNF-a [56]. Wang et al. showed serum levels of miR-122 had a significant direct correlation with BMI, triglyceride, and HOMA-IR index, and a significant inverse correlation with HDL cholesterol levels [24]. Also, in this study, miR-122 was associated with an increased odds of insulin resistance in humans. Therefore, they suggested circulating miR-122 levels may act as a marker of obesity and insulin resistance [24]. Lischka et al. [ 2021] reported that miR-122 is positively correlated with HOMA-IR, TG, cholesterol and ALT as metabolic biomarkers and TNF-α, IL-1Ra, and pro-calcitonin as inflammatory biomarkers in pediatric patients. They also reported positive correlation of miR-193b with cholesterol, ALT, and MRI PDFF (magnetic resonance imaging-proton density fat fraction). On the other hand, miR-122 and miR-193b had no significant association with CRP and IL-6. Moreover, Lischka et al. revealed miR-122 expression level of patients with pre diabetes, impaired glucose tolerance or metabolic syndrome was significantly different from those obese children without these conditions [57]. In line with the mentioned studies, in the current study, miR-193a and miR-122 had a significant positive association with fasting insulin level, HOMA-IR, and TG. There was also a positive significant association between miR-122 and hs-CRP. The following comparative analysis conducted between these two miRNAs and some metabolic status showed expression levels of both were higher in children with metabolic syndrome, insulin resistance, and high-fat mass. Furthermore, expression levels of miR-193a and miR-122 were also higher in children with fast REE (than slow REE) and children with high hs-CRP, respectively. Also, the present study could not find a significant difference in the expression of miR-15 between the two groups. However, previous studies have shown that miR-15 expression is reduced in type 2 diabetes, metabolic syndrome, and obesity, thereby miR-15 is known as a diagnostic biomarker for obese children at risk of type 2 diabetes mellitus [2]. Lischka et al., in 2021 reported miR-15a expression level was significantly lower in IGT pediatric patients [57]. In the current study, miR-15a had a significant association with serum insulin and HOMA-IR. this result was in line with previous observations regarding miR-15a importance as a predictor of diabetes in obese people regarding down-regulation of miR-15a in hyperglycemic conditions [58]. In the present study, none of the children had FBS higher than the normal range (neither obese children nor normal weight children), which may be the reason for observing no significant difference in the miR-15a level between the two study groups. In the present study, there was no significant difference between the two groups for the expression of miR-146a. Chartoumpekis et al. compared the expression levels of 530 miRNAs in adipose tissue of mice feeding a standard diet and mice feeding a high-fat diet. For the first time, they showed that miR-146a, miR-146b, and miR-379 were up-regulated during the development of obesity [59]. Ortego et al. reported that obese participants after Bariatric Surgery lost about 30 percent of their original weight and miR-146a was up-regulated by about $146\%$ [60]. Sharma et al. designed a case-control study to investigate the relation of adipose tissue miRNAs with insulin sensitivity. They couldn’t find a significant difference in miR-146 expression level between the participants with and without insulin resistance, but the miR-146a and miR-146b had a significant correlation with body fat percentage, BMI, and insulin sensitivity index [29]. A few shortcomings of the study that can be improved in the future is to have a group of overweight children included to compare with obese and normal-weight children. Secondly, due to the cross-sectional design of the study cause and effect relationship could not be marked. Therefore, planning further studies to assess cause-effect correlations seems essential. For future studies, it is recommended that participants be categorized into two groups, metabolically healthy obese and metabolically unhealthy obese. Besides, considering groups of obese children with diabetes and metabolic syndrome will help grasp the affiliation of miRNAs with these diseases. ## Conclusions and future directions In conclusion, among the five studied miRNAs (miR-193a, miR-122, miR-155a, miR-15a, and miR-146), only the expression level of miR-155 was significantly different in PBMC of obese children compared to the normal group (lower in obese children). Moreover, the expression level of miR-155 was negatively associated with fasting insulin and HOMA and also significantly lower in children with high hs-CRP and high-fat mass compared to the normal group. MiR-193a and miR-122 were positively correlated with insulin, HOMA, and TG. MiR-122 also was correlated with hs-CRP. After disaggregating the data by gender, the expression levels of miR-122 were higher in obese girls than normal-weight girls. The expression levels of both miR-122 and miR-193a were higher in children with metabolic syndrome, insulin resistance, and high fat mass. Furthermore, the expression levels of miR-193a and miR-122 were also higher in children with fast REE (than slow REE) and high hs-CRP, respectively. Despite the limitations of our study, expression levels of miR-193a, miR-122, and miR-155a seem to be associated with some metabolic statuses, including obesity, metabolic syndrome, hyperinsulinemia, insulin resistance, and hyperlipidemia in obese children. Thereby, using the expression pattern of miRNAs in obese children and adolescents as a predictive marker of adulthood chronic disease appears to be very helpful. Further studies are suggested to evaluate the relationships and underlying mechanisms to achieve therapeutic interventions. ## References 1. Youssef EM. Expression profiling and analysis of some miRNAs in subcutaneous white adipose tissue during development of obesity. Genes Nutr (2020.0) 15 8. DOI: 10.1186/s12263-020-00666-0 2. Cui X. Change in circulating microRNA profile of obese children indicates future risk of adult diabetes. Metabolism (2018.0) 78 95-105. DOI: 10.1016/j.metabol.2017.09.006 3. Hatipoglu N. Waist circumference percentiles for 7- to 17-year-old Turkish children and adolescents. Eur J Pediatr. (2008.0) 167 383-9. DOI: 10.1007/s00431-007-0502-3. 4. Fau Kissebah Ah, Vydelingum N. Relation of body fat distribution to metabolic complications of obesity. J Clin Endocrinol Metab (1982.0) 54 254-60. DOI: 10.1210/jcem-54-2-254. 5. Akinci G. Evaluation of markers of inflammation, insulin resistance and endothelial dysfunction in children at risk for overweight. Hormones (2008.0) 7 156-62. DOI: 10.1007/BF03401507 6. Park JH. Murine hepatic miRNAs expression and regulation of gene expression in diet-induced obese mice. Mol Cells (2010.0) 31 33-8. DOI: 10.1007/s10059-011-0009-7 7. Blüher S, Schwarz P. Metabolically healthy obesity from childhood to adulthood - Does weight status alone matter?. Metabolism (2014.0) 63 1084-92. DOI: 10.1016/j.metabol.2014.06.009 8. Landgraf K. Evidence of early alterations in adipose tissue biology and function and its association with obesity-related inflammation and insulin resistance in children. Diabetes (2015.0) 64 1249-61. DOI: 10.2337/db14-0744 9. Prince RL. Predictors of metabolically healthy obesity in children. Diabetes Care (2014.0) 37 1462-1468. DOI: 10.2337/dc13-1697 10. 10.Blüher M. The distinction of metabolically “healthy” from “unhealthy” obese individuals. Curr Opin Lipidol. 2010;21(1):38–43. 10.1097/MOL.0b013e3283346ccc. 11. 11.Vukovic R, et al. Children With Metabolically Healthy Obesity: A Review. Front Endocrinol. 2019:10(865):865–75. 10.3389/fendo.2019.00865. 12. 12.Dexheimer PJ, Cochella L. MicroRNAs: From Mechanism to Organism. Front Dev Biol. 2020;8:409–26. 10.3389/fcell.2020.00409. 13. Foltz JL. Population-level intervention strategies and examples for obesity prevention in children. Annu Rev Nutr (2012.0) 32 391-415. DOI: 10.1146/annurev-nutr-071811-150646 14. Kumar S. Decreased Circulating Levels of Spexin in Obese Children. J Clin Endocrinol Metab (2016.0) 101 2931-2936. DOI: 10.1210/jc.2016-1177 15. 15.Pandita A, et al. Childhood obesity: prevention is better than cure. Diabetes, metabolic syndrome and obesity : targets and therapy. 2016;9:83–9. 16. 16.Dumortier, O., E. Hinault C Fau - Van Obberghen, and E. Van Obberghen, MicroRNAs and metabolism crosstalk in energy homeostasis, in Cell Metab 312–24. doi: 10.1016/j.cmet.2013.06.004. Epub 2013 Jul 11. 17. Fritz JV. Sources and Functions of Extracellular Small RNAs in Human Circulation. Annu Rev Nutr (2016.0) 17 301-336. DOI: 10.1146/annurev-nutr-071715-050711 18. Ross SA, Davis CD. The emerging role of microRNAs and nutrition in modulating health and disease. Annu Rev Nutr (2014.0) 34 305-336. DOI: 10.1146/annurev-nutr-071813-105729 19. Sun W. microRNA: a master regulator of cellular processes for bioengineering systems. Annu Rev Biomed Eng (2010.0) 12 1-27. DOI: 10.1146/annurev-bioeng-070909-105314 20. Fischer-Posovszky P. Functional Significance and Predictive Value of MicroRNAs in Pediatric Obesity: Tiny Molecules with Huge Impact?. Horm Res Paediatr (2016.0) 86 3-10. DOI: 10.1159/000444677 21. Nesca V. Identification of particular groups of microRNAs that positively or negatively impact on beta cell function in obese models of type 2 diabetes. Diabetologia (2013.0) 56 2203-2212. DOI: 10.1007/s00125-013-2993-y 22. Xu J. Several circulating miRNAs related to hyperlipidemia and atherosclerotic cardiovascular diseases. Lipids Health Dis (2019.0) 18 104. DOI: 10.1186/s12944-019-1046-z 23. Shah R. Extracellular RNAs are associated with insulin resistance and metabolic phenotypes. Diabetes Care (2017.0) 40 546-553. DOI: 10.2337/dc16-1354 24. Wang R. Elevated circulating microRNA-122 is associated with obesity and insulin resistance in young adults. Eur J Endocrinol (2015.0) 172 291-300. DOI: 10.1530/EJE-14-0867 25. Chakraborty C. Influence of miRNA in insulin signaling pathway and insulin resistance: micro-molecules with a major role in type-2 diabetes. Wiley Interdiscip Rev RNA (2014.0) 5 697-712. DOI: 10.1002/wrna.1240 26. Párrizas M. Circulating miR-192 and miR-193b are markers of prediabetes and are modulated by an exercise intervention. J Clin Endocrinol Metab (2015.0) 100 E407-E415. DOI: 10.1210/jc.2014-2574 27. 27.Alexander R, L. Lodish H Fau - Sun, and L. Sun,. MicroRNAs in adipogenesis and as therapeutic targets for obesity. Expert Opin Ther Targets. 2011;15(5):623–36. 10.1517/14728222.2011.561317. (Epub 28 Feb 2011). 28. 28.Williams, M.D. and G.M. Mitchell, MicroRNAs in insulin resistance and obesity, in Exp Diabetes Res. 2012;2012:484696. doi: 10.1155/2012/484696. Epub 2012 Jul 18. 29. Sharma NK. Obesity associated modulation of miRNA and co-regulated target transcripts in human adipose tissue of non-diabetic subjects. Microrna (2015.0) 4 194-204. DOI: 10.2174/2211536604666151103121817 30. 30.Miller, A.M., et al., MiR-155 has a protective role in the development of non-alcoholic hepatosteatosis in mice, in PLoS One. p. e72324. doi: 10.1371/journal.pone.0072324. eCollection 2013. 31. 31.Wang, B., et al., Role of microRNA-155 at early stages of hepatocarcinogenesis induced by choline-deficient and amino acid-defined diet in C57BL/6 mice, in Hepatology. p. 1152–61. doi: 10.1002/hep.23100. 32. Kuczmarski RJ. 2000 CDC Growth Charts for the United States: methods and development. Vital Health Stat (2002.0) 11 1-190 33. Mahdavi R. Decreased serum level of miR-155 is associated with obesity and its related metabolic traits. Clinical Laboratory (2018.0) 64 77-84. PMID: 29479888 34. Delshad M. Reliability and validity of the modifiable activity questionnaire for an Iranian urban adolescent population. Int J Prev Med (2015.0) 6 3. DOI: 10.4103/2008-7802.151433 35. Esfahani FH. Reproducibility and relative validity of food group intake in a food frequency questionnaire developed for the Tehran Lipid and Glucose Study. J Epidemiol (2010.0) 20 150-158. DOI: 10.2188/jea.JE20090083 36. Asghari G. Reliability, comparative validity and stability of dietary patterns derived from an FFQ in the Tehran Lipid and Glucose Study. Br J Nutr (2012.0) 108 1109-1117. DOI: 10.1017/S0007114511006313 37. Ten S, Maclaren N. Insulin resistance syndrome in children. J Clin Endocrinol Metab (2004.0) 89 2526-2539. DOI: 10.1210/jc.2004-0276 38. Kurtoglu S. Insulin resistance in obese children and adolescents: HOMA-IR cut-off levels in the prepubertal and pubertal periods. J Clin Res Pediatr Endocrinol (2010.0) 2 100-106. DOI: 10.4274/jcrpe.v2i3.100 39. Cook S. Prevalence of a metabolic syndrome phenotype in adolescents: findings from the third National Health and Nutrition Examination Survey, 1988–1994. Arch Pediatr Adolesc Med (2003.0) 157 821-827. DOI: 10.1001/archpedi.157.8.821 40. Hellemans J. qBase relative quantification framework and software for management and automated analysis of real-time quantitative PCR data. Genome Biol (2007.0) 8 R19. DOI: 10.1186/gb-2007-8-2-r19 41. Rieu I, Powers SJ. Real-Time Quantitative RT-PCR: Design, Calculations, and Statistics. Plant Cell (2009.0) 21 1031-1033. DOI: 10.1105/tpc.109.066001 42. Bastami M. Perturbation of miR-146b and relevant inflammatory elements in esophageal carcinoma patients supports an immune downregulatory mechanism. Pathol Res Pract (2021.0) 225 20 43. 43.Saadatian, Z., et al.,Dysregulated expression of STAT1, miR-150, and miR-223 in peripheral blood mononuclear cells of coronary artery disease patients with significant or insignificant stenosis, in J Cell Biochem. 2019. p. 19810–19824. doi: 10.1002/jcb.29286. Epub 2019 Jul 18. 44. Behrooz M, Vaghef-Mehrabany E, Ostadrahimi A. Different spexin level in obese vs normal weight children and its relationship with obesity related risk factors. Nutr Metab Cardiovasc Dis (2020.0) 30 674-682. DOI: 10.1016/j.numecd.2019.11.008 45. Peng M, Yang X. Controlling diabetes by chromium complexes: The role of the ligands. J Inorg Biochem (2015.0) 146 97-103. DOI: 10.1016/j.jinorgbio.2015.01.002 46. Ji CA-O, Guo XA-O. The clinical potential of circulating microRNAs in obesity. Nat Rev Endocrinol (2019.0) 15 731-743. DOI: 10.1038/s41574-019-0260-0 47. Mazloom H. Downregulated microRNA-155 expression in peripheral blood mononuclear cells of type 2 diabetic patients is not correlated with increased inflammatory cytokine production. Cytokine (2015.0) 76 403-408. DOI: 10.1016/j.cyto.2015.07.007 48. 48.Chang K-P, et al. MicroRNA signatures in ischemia-reperfusion injury. Annals of plastic surgery. 2012;69(6):668–71. 49. Fang L. Circulating microRNAs as biomarkers for diffuse myocardial fibrosis in patients with hypertrophic cardiomyopathy. Journal of translational medicine (2015.0) 13 314. DOI: 10.1186/s12967-015-0672-0 50. Holvoet P. The metabolic syndrome, circulating oxidized LDL, and risk of myocardial infarction in well-functioning elderly people in the health, aging, and body composition cohort. Diabetes (2004.0) 53 1068-73. DOI: 10.2337/diabetes.53.4.1068 51. Luo X. Regulation of human cardiac ion channel genes by microRNAs: theoretical perspective and pathophysiological implications. Cellular Physiology and Biochemistry (2010.0) 25 571-586. DOI: 10.1159/000315076 52. Yamada H. Associations between circulating microRNAs (miR-21, miR-34a, miR-122 and miR-451) and non-alcoholic fatty liver. Clin Chim Acta (2013.0) 424 99-103. DOI: 10.1016/j.cca.2013.05.021 53. Yi F. MicroRNA-193-5p modulates angiogenesis through IGF2 in type 2 diabetic cardiomyopathy. Biochemical and biophysical research communications (2017.0) 491 876-882. DOI: 10.1016/j.bbrc.2017.07.108 54. Oses MA-O. Circulating miRNAs as Biomarkers of Obesity and Obesity-Associated Comorbidities in Children and Adolescents: A Systematic Review.. Nutrients (2019.0) 11 2890. DOI: 10.3390/nu11122890 55. Hess AL. Levels of Circulating miR-122 are Associated with Weight Loss and Metabolic Syndrome. Obesity (2020.0) 28 493-501. DOI: 10.1002/oby.22704 56. Mohany KM. Investigation of the levels of circulating miR-29a, miR-122, sestrin 2 and inflammatory markers in obese children with/without type 2 diabetes: a case control study. BMC Endocrine Disorders (2021.0) 21 152. DOI: 10.1186/s12902-021-00829-z 57. Lischka JA-O. Circulating microRNAs 34a, 122, and 192 are linked to obesity-associated inflammation and metabolic disease in pediatric patients. Int J Obes (Lond). (2021.0) 45 1763-1772. DOI: 10.1038/s41366-021-00842-1 58. Ye E-A, Steinle JJ. miR-15b/16 protects primary human retinal microvascular endothelial cells against hyperglycemia-induced increases in tumor necrosis factor alpha and suppressor of cytokine signaling 3. Journal of neuroinflammation (2015.0) 12 1-8. DOI: 10.1186/s12974-015-0265-0 59. Chartoumpekis DV. Differential expression of microRNAs in adipose tissue after long-term high-fat diet-induced obesity in mice. PLOS ONE (2012.0) 7 e34872. DOI: 10.1371/journal.pone.0034872 60. Ortega FJ. Targeting the circulating microRNA signature of obesity. Clin Chem (2013.0) 59 781-792. DOI: 10.1373/clinchem.2012.195776
--- title: Berberine ameliorates depression-like behaviors in mice via inhibiting NLRP3 inflammasome-mediated neuroinflammation and preventing neuroplasticity disruption authors: - Zongshi Qin - Dong-Dong Shi - Wenqi Li - Dan Cheng - Ying-Dan Zhang - Sen Zhang - Bun Tsoi - Jia Zhao - Zhen Wang - Zhang-Jin Zhang journal: Journal of Neuroinflammation year: 2023 pmcid: PMC9976521 doi: 10.1186/s12974-023-02744-7 license: CC BY 4.0 --- # Berberine ameliorates depression-like behaviors in mice via inhibiting NLRP3 inflammasome-mediated neuroinflammation and preventing neuroplasticity disruption ## Abstract ### Objectives Neuroinflammation has been suggested that affects the processing of depression. There is renewed interest in berberine owing to its anti-inflammatory effects. Herein, we investigated whether berberine attenuate depressive-like behaviors via inhibiting NLRP3 inflammasome activation in mice model of depression. ### Methods Adult male C57BL/6N mice were administrated corticosterone (CORT, 20 mg/kg/day) for 35 days. Two doses (100 mg/kg/day and 200 mg/kg/day) of berberine were orally administrated from day 7 until day 35. Behavioral tests were performed to measure the depression-like behaviors alterations. Differentially expressed gene analysis was performed for RNA-sequencing data in the prefrontal cortex. NLRP3 inflammasome was measured by quantitative reverse transcription polymerase chain reaction, western blotting, and immunofluorescence labeling. The neuroplasticity and synaptic function were measured by immunofluorescence labeling, Golgi–Cox staining, transmission electron microscope, and whole-cell patch-clamp recordings. ### Results The results of behavioral tests demonstrated that berberine attenuated the depression-like behaviors induced by CORT. RNA-sequencing identified that NLRP3 was markedly upregulated after long-term CORT exposure. Berberine reversed the concentrations of peripheral and brain cytokines, NLRP3 inflammasome elicited by CORT in the prefrontal cortex and hippocampus were decreased by berberine. In addition, the lower frequency of neuronal excitation as well as the dendritic spine reduction were reversed by berberine treatment. Together, berberine increases hippocampal adult neurogenesis and synaptic plasticity induced by CORT. ### Conclusion The anti-depressants effects of berberine were accompanied by reduced the neuroinflammatory response via inhibiting the activation of NLRP3 inflammasome and rescued the neuronal deterioration via suppression of impairments in synaptic plasticity and neurogenesis. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12974-023-02744-7. ## Introduction Depression has been identified as a high incidence and severe psychiatric disease [1]. The health burden of depression by using disability-adjusted life-years (DALYs) estimation accounted for $1.85\%$ of all DALYs worldwide, which increased $61.1\%$ from 1990 to 2019 [2]. It is known to significantly increase the risk of suicide for all ages, especially in adolescents. Although the understanding of the pathology of depression has developed considerably. Currently, no single mechanism can satisfactorily explain the pathophysiology of depression [3, 4]. Studies have focused on many components of brain including prefrontal cortex (PFC), hippocampus, amygdala, ventral tegmental area (VTA), and nucleus accumbens (NAc), leading to the theories of depression as well as antidepressant response that have been involved in the molecular and cellular signaling mechanisms that mediate synaptic plasticity, contributing to a broader neuroplasticity hypothesis of depression [5, 6]. To date, increasing evidence indicated that overexpressed peripheral inflammatory responses could injure the integrity of the blood–brain-barrier (BBB) and result in neuroinflammation in the brain [7]. Consequently, the neuroinflammation-mediated neuroplasticity and neurogenesis defects might be a vital process under the mechanism in neuropsychiatric conditions, including depression [8]. It has been observed that an excess in peripheral acute phase proteins and proinflammatory cytokines production in depression patients, which have been identified to be linked with emotional alterations and severity of psychiatric symptoms [9, 10]. Besides, the remission of patients is often occurring after normalization of the inflammatory response, whereas a failure to remission is accompanied by the persistently elevated inflammatory response. This information promotes the hypothesis that the emotional alterations in depression patients might be attributed by an anomalous link between the central nervous system (CNS) and the innate immune response. Moreover, the pooled data from meta-analysis also supported that several anti-inflammatories have significant antidepressant effects [11]. The interactions between the immune system and the CNS are not only involved in shaping behavior, but also in responding to therapeutics [12]. NLRP3 (NLR family, pyrin domain containing 3) inflammasome complex is an intracellular multiprotein complex responsible for several innate immune processes associated with infection, inflammation, and autoimmunity [13]. As a component of the innate immune system that functions as a pattern recognition receptor that recognizes pathogen-associated molecular patterns (PAMPs) and danger-associated molecular patterns (DAMPs), NLRP3 appears to bridge the gap between immune activation and metabolic danger signals or stress exposure, which might be key factors in the pathogenesis of depression [14, 15]. Stimulated by NLRP3, the over-released pro-inflammatory cytokine IL-1β can cross the BBB and alter synaptic plasticity by directly acting on neurons or stimulating the microglia activation [16–18]. In addition, the hypothalamus–pituitary–adrenal (HPA) axis could be stimulated by cytokines and result in glucocorticoids overproduction, exacerbating the stress response [19, 20]. On the other hand, the neurotoxic effects of neuroinflammation consequently contribute to the synaptic remodeling, suggesting that neural plasticity also plays a vital role in the pathophysiology of depression and antidepressant function [21]. High levels of inflammatory molecules have been reported to decrease a wide range of neural plasticity markers such as synaptic transmission, membrane excitability, plasticity in pyramidal neurons, as well as neurogenesis. NLRP3 matured IL-1β plays functional roles in the mechanisms of synaptic plasticity and cognitive functions. In the depression mice, the spine density and critical morphologies were significantly decreased, especially in the specific brain regions related to depression, such as the prefrontal cortex and hippocampus. Berberine is a natural isoquinoline alkaloid and there is renewed interest in berberine of its potential role in neurodegenerative and neuropsychiatric disorders because of its effect on neuroinflammation, hormonal regulation, and neurotransmitters [22–24]. In this study, we investigated the differentially expressed genes in corticosterone induced depression mice model (CORT) using a high-throughput microarray. The theory that disruption of neurotrophic factors and synaptic connectivity in the PFC and hippocampus is related to neuroplasticity mechanisms is one of the leading neuroplasticity hypotheses of depression [25]. We found that NLRP3 showed significantly differential expression within the PFC of CORT-induced mice model versus wildtype mice controls and berberine-treated mice. Complementing these findings, the CORT mice result in neuroplasticity deficits and neurogenesis injury and induced depression-like behaviors. Accordingly, these results provide insights into mechanisms involving the functional regulation of corticosterone in depression and specifically, identify berberine as a potential therapy for depression. ## Animal models and drug treatment Adult male C57BL/6N mice (age 8–10 weeks) were obtained from the Centre for Comparative Medicine Research (CCMR), the University of Hong Kong. All mice were raised in the experimental holding areas (12 h light/dark cycle at 18–22 °C, with lights on at 8:00 A.M., ad libitum access to dry food pellets and water) following the ethics roles of the HKU (CULATR No. 5582-20) in the CCMR. Body weight was daily recorded during the entire experimental period. The mice were divided into 4 groups ($$n = 18$$ for each group), including saline-treated (control), corticosterone (20 mg/kg/day) treated (CORT), CORT + berberine (100 mg/kg/day) treated (BBR100), and CORT + berberine (200 mg/kg/day) treated (BBR200). CORT was consecutively administered from day1 to day35. From day7, 100 mg/kg/day or 200 mg/kg/day of berberine were given to mice in BBR100 or BBR200 groups via daily intragastrical administration, respectively. Mice in control and CORT groups received solution without berberine. Mice were given BrdU as daily intraperitoneal injections from day 8 to day 12. From day 28 to day 31, a series of behavioral tests were conducted for mice, CORT and berberine administration were sustained during behavioral tests until to the end of the experiment (day35). The dose of CORT was selected in accordance with a previous study that successfully induced depression in mice from the same laboratory [26]. CORT and berberine were dissolved in a $0.5\%$ aqueous solution of sodium carboxymethyl cellulose and administered via oral gavage. The solutions were freshly prepared before daily use. Figure 1a illustrates the experimental design timeline with the time of assays and manipulations. Fig. 1Berberine prevents CORT-induced behavioral changes. a Schematic of experiment paradigm. CORT was consecutively administered from day1 to day35. From day7, 100 mg/kg/day or 200 mg/kg/day of berberine were given to mice in BBR100 or BBR200 groups via daily intragastrical administration, respectively. Mice in control and CORT groups received solution without berberine. Mice were given BrdU as daily intraperitoneal injections from day8 to day12. From day28 to day31, a series of behavioral tests were conducted for mice, CORT and berberine administration were sustained during behavioral tests until to the end of the experiment (day35). b Heatmap of mice exploration during the open field test. c Duration in central zone in the open field test. CORT mice spent less time in the central zone than the control mice (One-way ANOVA, F [3, 32] = 5.448, $$P \leq 0.0038$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0012$$, $$n = 9$$ in each group). d Total distance in the open field test. CORT mice moved less in the field than the control and high-dose berberine mice (One-way ANOVA, F [3, 32] = 4.203, $$P \leq 0.0129$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0077$$, CORT vs. BBR200, $$P \leq 0.0212$$, $$n = 9$$ in each group). e Zone transition number in the open field test. CORT administration decreased the number of zone transitions in mice in the open field test compared to the control and high-dose berberine mice (One-way ANOVA, F [3, 32] = 5.711, $$P \leq 0.0030$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0008$$, CORT vs. BBR200, $$P \leq 0.0429$$, $$n = 9$$ in each group). f Immobile duration in the tail suspension test. CORT mice spent more time in immobile duration (One-way ANOVA, F [3, 28] = 6.435, $$P \leq 0.0019$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0005$$, CORT vs. BBR100, $$P \leq 0.0425$$, CORT vs. BBR200, $$P \leq 0.0345$$, $$n = 8$$ in each group). g Immobile duration in the forced swim test. CORT mice spent more time in immobile duration (One-way ANOVA, F [3, 32] = 6.316, $$P \leq 0.0017$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0007$$, CORT vs. BBR100, $$P \leq 0.0145$$, CORT vs. BBR200, $$P \leq 0.0141$$, $$n = 9$$ in each group). h Sucrose consumption in the sucrose preference test. The consumption of sucrose in CORT mice was significantly lower than in the mice in control and high-dose berberine groups (One-way ANOVA, F [3, 32] = 4.638, $$P \leq 0.0084$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0070$$, CORT vs. BBR200, $$P \leq 0.0090$$, $$n = 16$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, *$P \leq 0.001.$ CORT corticosterone, BBR berbberine, ANOVA analysis of variance ## Open field test (OFT) For assessment of locomotion and anxiety-like behavior in mice. The OFT was performed in a white plastic apparatus (50 height × 50 widths × 40 cm depth) with a 30 × 30 cm central zone. The total distance and velocity moved and the frequency of transfer between the central and surrounding zones was recorded over a 10-min test period. The $95\%$ ethanol was sprayed to clean the apparatus between each test to avoid odor and waste left by the last mouse. We performed the test for all mice on the first day before the experiment. SMART video tracking software (V.3.0, Panlab, USA) was used to record and analyzed data. ## Tail suspension test (TST) The TST was performed in a white plastic chamber (55 height × 10 widths × 10 cm depth). Each mouse was suspended from its tail tip with adhesive tape in a head-down position and lasted for 6 min. Immobility time is defined as the cessation of any movements of limbs and the trunk. To avoid the bias affected by the stress response, each mouse was adapted for 2 min after being suspended, only the remaining 4 min was recorded and analyzed. SMART video tracking software (V.3.0, Panlab, USA) was used to record and analyzed data. ## Forced swimming test (FST) To evaluate a depressive-like behavioral state, the FST was performed in a clear polycarbonate cylinder (30 height × 20 cm diameter). Each mouse was forced to swim in each cylinder filled with water (23 to 25 °C) for 6 min and videotaped. Immobility time is defined as the absence of all movements except the motions required to keep the mice's heads above the surface of the water. To avoid the bias affected by the stress response, each mouse was adapted for 2 min in the water, only the remaining 4 min was recorded and analyzed. ## Sucrose preference test (SPT) SPT was performed to assess the anhedonia, a core symptom of depression. First, mice were adaptively exposed to two bottles containing $1\%$ sucrose solution (w/v) with ad libitum access for 24 h in groups of 5 per cage. Then, one bottle of $1\%$ sucrose solution and another bottle of tap water were administrated for 24 h. On the last day, the position of the two bottles was switched for 24 h to avoid side preference. At the end of the adaptation period, all mice were deprived of food and water for 12 h before the test. After that, SPT was conducted in an individual mouse housed in a cage with free access to two respective bottles containing $1\%$ sucrose solution and tap water for 2 h. To prevent side preferences in drinking behavior, the position of the two bottles was switched in the middle of the testing. Water and sucrose consumption was measured as changes in the weight of fluid consumed. The sucrose preference was calculated from the following formula was the sucrose preference (%) = the sucrose consumption (g)/[the sucrose consumption(g) + the water consumption (g)] × $100\%$. ## RNA isolation, qPCR and RNA-sequencing Total RNA was isolated from dissected prefrontal cortex using QIAzol lysis reagent and purified using a miRNAeasy mini kit (Qiagen). cDNA was acquired from total RNA using a high-capacity cDNA Reverse Transcription Kit (Life Technologies). qPCR was performed using the Fast Start Universal SYBR Green Master kit (Takara, Japan) using CFX96 Real-Time System (Bio-Rad, USA). Each reaction was performed in triplicates. The relative quantification was determined by the ΔΔCT method. The values were normalized to those of β-actin mRNA in the same cDNA samples. Additional file 1: Table S1 summarizes the information on primers for qPCR. The details for library preparation and transcriptome sequencing are shown in the supplementary information. ## Brain slice preparation and electrophysiology recording After the end of behavioral tests, mice were anesthetized with a combined anesthetic (ketamine 100 mg/kg and xylazine 10 mg/kg i.p.) then perfused with chilled dissection buffer (110 mM choline chloride, 25 mM NaHCO3, 1.25 mM NaH2PO4, 2.5 mM KCl, 0.5 mM CaCl2, 7 mM MgCl2, 11.6 mM ascorbic acid, 3.1 mM pyruvic acid, and 25 mM d-glucose). The brains were immediately transferred into ice-cold oxygenated ($95\%$ O2 and $5\%$ CO2) dissection buffer and coronal mPFC slices (300 µm) were cut using a vibratome (VT1000s, Leica, Germany). Then, slices were incubated in oxygenated ($95\%$ O2/$5\%$ CO2) artificial cerebrospinal fluid solution (aCSF; 118 mM NaCl, 2.5 mM KCl, 26.5 mM NaHCO3, 1 mM NaH2PO4, 1 mM MgCl2, 2 mM CaCl2, and 20 mM D-glucose) at room temperature for 60 min to recover from the mechanical shock of slicing. The brain slices were soaked in the running artificial cerebrospinal fluid solution at 5 mL/min flow rate in the holding chamber of the electrophysiology recording platform. The glass recording pipettes were made by a pipette puller (P-97, Sutter Instrument, USA). The cells were located by a fine control micromanipulator under an optical microscope (Zeiss, Germany). The resistance ranged between 2 and 5 MΩ following fire polishing to enhance seal quality. For current-clamp recordings, the intracellular solution contained (in mM) 130 K gluconates, 5 KCL, 10 HEPES, 2.5 MgCl2, 4 Na2ATP, 0.4 Na3GTP, 10 Na phosphocreatine, 0.6 EGTA. For voltage-clamp recordings, the intracellular solution contained (in mM) 115 CsMeCO3, 20 CsCl, 10 HEPES, 2.5 MgCl2, 4 Na2ATP, 0.4 NaGTP, 10 Na phosphocreatine, and 0.6 EGTA. Current clamp recordings were filtered at 2.5 kHz and sampled at 5 kHz. Voltage clamp recordings were filtered at 2.5 kHz and sampled at 10 kHz. The sequential currents from − 50 to 400 pA in a 50 pA step for 500 ms were injected. The currents were injected every 60 s in the current clamp. The electrophysiology recording was performed using a patch-clamp amplifier (EPC10USB, HEKA, Germany), and the spontaneous or miniature excitatory events were analyzed with Mini Analysis Program (v.6.0.3, Synaptosoft Inc., USA). ## Enzyme-linked immunosorbent assay (ELISA) Serum concentrations of cytokines including IL-1β (KE10003, Proteintech), IL-6 (KE00007, Proteintech), IL-10 (KE00170, Proteintech), and TNF-α (KE10002, Proteintech) were measured using the quantitative ELISA kits. All measurements were performed in duplicate. The absorbance of each well for reactions was detected at 450 nm using the CLARIOstar Plus Microplate Reader (BMG LABTECH, Germany). The cytokine concentrations were determined by the standard curve of each cytokine. ## Western blotting To determine the expression of the protein, the hippocampus and prefrontal cortex tissue was extracted by RIPA buffer (Sigma-Aldrich, USA) containing $1\%$ protease inhibitor cocktail (MCE, USA). The protein concentrations were detected by Bradford protein assay using Coomassie brilliant blue G-250 (Bio-red Laboratories Inv., USA). Proteins were separated via $10\%$ or $12\%$ SDS–PAGE gels according to the different protein sizes and transferred to PVDF membranes (0.45 μm, Bio-red Laboratories Inv., USA). The blots were subsequently blocked for 1 h with $5\%$ BSA at room temperature. After being incubated with primary antibodies at 4 ℃ overnight, the membranes were incubated with fluorescence-conjugated secondary antibodies for 1 h at room temperature. Proteins were detected using an ECL kit (GE Healthcare, UK) and quantified using the Image Lab software (v.5.2.1, Bio-Rad, USA). The details of antibody information are shown in Additional file 1: Table S2. ## Immunofluorescence labeling The BrdU was injected for 5 days continuously from day 8 to day 12, mice were sacrificed by cardiac perfusion with $4\%$ paraformaldehyde. Brains were collected postfixed in $4\%$ paraformaldehyde at 4 ℃ overnight and then dehydrated in $30\%$ sucrose PBS solution for 2 days at 4 ℃ cold room. Coronal brain sections (25 μm) were prepared with a freezing microtome (Leica Inc., Germany). For immunofluorescence study, slides were processed antigen retrieved with citrate acid buffer with microwave oven for 30 min. Next, slides were blocked by BSA for 60 min at room temperature and then incubated at 4 ℃ overnight in the primary antibodies. After washing by PBST, the slides were incubated with fluorescent-dye-conjugated secondary antibodies (DyLight 594-conjugate donkey anti-goat, 1:200, Abcam; DyLight 488-conjugate donkey anti-rabbit, 1:200, Abcam; DyLight 594-conjugate donkey anti-rabbit, 1:200, Abcam) in dark for 1 h at room temperature. Before mounting, the slides were stained with DAPI to stain the nucleus for 10 min. Images were captured using a confocal microscope (LSM880, Zeiss, Germany). Analyses of pictures were performed using the ImageJ software (v.1.53c, NIH, USA). The details of antibody information are shown in Additional file 1: Table S3. ## Golgi-Cox staining Golgi-Cox staining was performed by using the commercial staining kit (Hito Golgi-Cox OptimStain Kit, Hitobiotec Corp, USA) to characterize potential changes in the density and feature of neuronal dendritic spines. The whole brain was soaked in the staining solution at room temperature for 14 days to avoid light according to the instruction manual. The tissues were prepared as paraffin sections into 60 μm. Images were captured using a confocal microscope (LSM880, Zeiss, Germany). Three-D reconstruction was performed by using Imaris software (v.9.0.1, Bitplane AG, Switzerland) to detect the categories of dendrite spine, the cells with straight terminal branches that had a clear resolution of spines and were longer than 10 μm were selected for dendritic spines counting and analysis. ## Transmission electron microscope (TEM) After perfusion with $4\%$ paraformaldehyde, the hippocampus of mice was manually cut into 1 mm3 tissue and immediately fixed in $2.5\%$ glutaraldehyde in 0.1 M phosphate buffer overnight at 4 °C. The tissues were transferred into $4\%$ osmium tetroxide with $3\%$ potassium ferrocyanide in 0.1 M cacodylate buffer for 1 h at 4 °C and avoid light then embedded in Epon 812 after dehydration. Samples were sectioned into fine sections (0.4 μm) using an ultramicrotome (Ultracut UCT, Leica, USA) and moved to copper grids. After staining with $2\%$ aqueous uranyl acetate and followed by Reynold’s lead citrate, the images were captured by using a transmission electron microscopy (CM100, Philips, Germany) at 100 kV attached with a charge-coupled device camera (SIS, Olympus, Japan). ## Statistical analysis Data are expressed as mean ± standard error of the mean with two-sided $p \leq 0.05$ considered significant. PROC POWER of SAS (v.9.4, SAS Institute Inc, USA) was used to determine the proper sample size for animal experiments. One-way or two-way analysis of variance (ANOVA) was appropriately used for comparisons among multiple groups in terms of behavioral tests, qPCR, western blot, ELISA, and immunofluorescence, followed by Dunnett’s or Tukey’s post hoc multiple comparisons adjustment. Data analyses were performed using GraphPad Prism software (v.9.0.0, GraphPad Software, USA). To detect correlation between behavioral parameters as dependent variables and NLRP3 inflammasome as independent variables, the Spearman’s correlation coefficient was calculated and the correlation matrix showing correlation coefficients between variables were summarized. The quantification analysis of immunofluorescence staining was calculated according to cell density analysis. One out of four 25 μm thick sections were analyzed thus representing the entire rostrocaudal extent of different brain regions at × 20 magnification. For analysis of NLRP3 + and Iba-1 + cells in PFC and hippocampus, the positive cells were counted and compared between groups. For analysis of PSD95 +, SYN +, BrdU + and DCX + cells in hippocampus, the positive cells were separately counted. Cells were counted using the plugin “Manual counting” (https://icy.bioimageanalysis.org/plugin/manual-counting/). The details for bioinformatic analysis for RNA-seq data are shown in the supplementary information. ## Berberine attenuates the emotional dysfunctions in CORT-induced depression mice Two doses of berberine (100 mg/kg/day and 200 mg/kg/day) were orally administrated daily from the middle of the CORT modeling until the end of the behavioral tests (Fig. 1a). Long-term high-dose CORT administration significantly decreased the body weight of mice, and a high dose of berberine reversed the negative effects of CORT on the body weight (Additional file 1: Fig. S1). CORT markable decreases the center zone time duration, total distance as well as zone transition number in OFT in mice (Fig. 1b–e). The high dosage (200 mg/kg/day) of berberine significantly reversed CORT-induced behavioral alterations (Fig. 1c–e). In the TST, CORT significantly increased immobile duration compared to other groups (Fig. 1f) and the same pattern was observed in the FST (Fig. 1g). In the SPT, the results showed a significant effect of CORT on the consumption of sucrose solution, the high dosage (200 mg/kg/day) of berberine reversed CORT-induced behavioral alterations (Fig. 1h). Collectively, these results suggest that berberine treatment especially in high doses (200 mg/kg) could attenuate the behavioral alterations associated with depression induced by CORT. ## Berberine reverses the peripheral and brain cytokine levels The cytokines of TNF-α, IL-1β, IL-6, and IL-10 were evaluated in the blood, prefrontal cortex (PFC), and hippocampus using ELISA and western blotting assay (Fig. 2a). One-way ANOVA analysis showed a significant effect of CORT on both peripheral, PFC, as well as hippocampus. Berberine effectively prevented the CORT-induced increase in pro-inflammatory cytokines, while increasing the anti-inflammatory cytokine IL-10 reduced by CORT (Fig. 2b–e). The results of western blotting in the PFC and hippocampus showed a similar pattern as ELISA that CORT treatment significantly enhanced proinflammatory cytokines levels including TNF-α, IL-1β, and IL-6. Two doses of berberine markedly reversed the levels of cytokines in the PFC and hippocampus (Fig. 2f, g).Fig. 2Effects of berberine on serum cytokines and neuroinflammatory responses in the prefrontal cortex and hippocampus. a Timing of CORT administration, behavioral tests, and western blotting and ELISA assay. b Effects of berberine on serum cytokine level of IL-6. The CORT significantly increased the IL-6 concentration compared to the mice in control and high-dose berberine groups (One-way ANOVA, F [3, 8] = 13.80, $$P \leq 0.0016$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0020$$, CORT vs. BBR200, $$P \leq 0.0021$$, $$n = 3$$ in each group). c Effects of BBR on serum cytokine level of IL-1β. The CORT significantly increased the IL-1β concentration (One-way ANOVA, F [3, 8] = 147.1, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). d Effects of berberine on serum cytokine level of TNF-α. The CORT significantly increased the TNF-α concentration (One-way ANOVA, F [3, 8] = 46.96, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). e Effects of berberine on serum cytokine level of IL-10. The CORT significantly decreased the IL-10 concentration (One-way ANOVA, F [3, 8] = 253.1, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). f The effects of berberine on cytokine level in prefrontal cortex measured by western blot. The CORT mice showed the significant different in four cytokines (Two-way ANOVA, row factor F [3, 41] = 26.78, $P \leq 0.0001$, column factor F [3, 41] = 13.61, $P \leq 0.0001$; Dunnett’s multiple comparisons test, CORT vs. Control, $P \leq 0.0001$, CORT vs. BBR100, $P \leq 0.0001$, CORT vs. BBR200, $$P \leq 0.0003$$, $$n = 3$$ in each group). g The effects of berberine on cytokine level in hippocampus measured by western blot. The CORT mice showed the significant different in four cytokines (Two-way ANOVA, row factor F [3, 41] = 18.07, $P \leq 0.0001$, column factor F [3, 41] = 8.34, $$P \leq 0.0002$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0007$$, CORT vs. BBR100, $$P \leq 0.0010$$, CORT vs. BBR200, $$P \leq 0.0003$$, $$n = 3$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001.$ CORT corticosterone, BBR berberine, ANOVA analysis of variance, ELISA enzyme-linked immunosorbent assay ## Bioinformatic analysis of RNA-seq and qPCR analysis identified NLRP3 as a hub gene in CORT mice To explore the transcriptome-wide alterations after long-term CORT exposure in the mice brain, differentially expressed gene (DEG) of RNA sequencing data from the prefrontal cortex was analyzed to identify a hub gene. Additional file 1: Table S4 shows the quality control of mapped reads for each RNA-seq sample. Long-term administration of high concentration CORT significantly altered the expression of 310 genes compared with the control group, and berberine (200 mg/kg/day) altered the expression of 128 genes compared with the CORT group (Fig. 3a). Forty-four genes were altered at criteria between the control group and CORT group, and the heatmap of DEG lists revealed that genes were altered in the opposite direction. DEG analysis identified NLRP3, as one of the hub genes among those altered in both comparisons of control vs. CORT and CORT vs. berberine. Long-term CORT-administration potentially increased the expression of NLRP3 inflammasome in the PFC compared to normal raised mice and the berberine treated mice (Fig. 3b, c). The mRNA expression alteration of NLRP3, caspase-1 and IL-1β in the PFC was confirmed by qPCR assay (Fig. 3d–f). DEGs were enriched for several relevant gene ontology (GO) terms. GO cellular component analysis indicated that CORT exhibited significant differences in genes involved in the pathways of response to the synaptic membrane, neuron to neuron synapse, and postsynaptic density, which has been associated with depression and neural plasticity, compared to the control group BBR exhibited significant differences in genes involved in the pathways of the neuron to neuron synapse, postsynaptic specialization, synaptic membrane, and postsynaptic density compared to CORT (Fig. 3g, h). KEGG pathway enrichment analysis of the DEGs between control and CORT group found that MAPK signaling pathway, axon guidance, HIF-1 signaling pathway, and steroid biosynthesis. Both MAPK and HIF-1 are upstream signals for NLRP3 inflammasome via mediated activation of NF-κB (Fig. 3i, j). Together, these results indicate that CORT may induce neuroinflammation as well as neuroplasticity deficit in the PFC. Therefore, we hypothesized that neuronal anomalies resulting from long-term CORT-induced chronic stress exposure are associated with NLRP3 inflammasome and synaptic dysfunction in the brain networks and result in emotional episodes alteration. Fig. 3Analysis of RNA-seq. a Representation of the differentially expressed genes represented by the Venn diagram. b Representation of the heatmap represented top differentially expressed genes between CON and CORT groups, $$n = 3$$ in each group. c Representation of the heatmap represented top differentially expressed genes between CORT and BBR groups, $$n = 3$$ in each group. d Results of qPCR showed that NLRP3 expression in PFC in CORT was higher than in CON and BBR mice (One-way ANOVA, F [2, 6] = 44.11, $$P \leq 0.0003$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0002$$, CORT vs. BBR, $$P \leq 0.0012$$, $$n = 3$$ in each group). e Results of qPCR showed that caspase-1 expression in PFC in CORT was higher than in CON and BBR mice (One-way ANOVA, F [2, 6] = 28.61, $$P \leq 0.0009$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0006$$, CORT vs. BBR, $$P \leq 0.0049$$, $$n = 3$$ in each group). f Results of qPCR showed that IL-1beta expression in PFC in CORT was higher than in CON and BBR mice (One-way ANOVA, F [2, 6] = 17.86, $$P \leq 0.0030$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0018$$, CORT vs. BBR, $$P \leq 0.0397$$, $$n = 3$$ in each group). g Representation of the top associated gene ontology terms between CON and CORT groups. h Representation of the top associated gene ontology terms between CORT and BBR group. i Representation of the top KEGG enrichment pathways between CON and CORT groups. j Representation of the top KEGG enrichment pathways between CORT and BBR group. CON control, CORT corticosterone, BBR berberine, ANOVA analysis of variance, NLRP3 NOD-like receptor thermal protein domain associated protein 3, qPCR quantitative reverse transcription polymerase chain reaction, PFC prefrontal cortex ## Berberine reverses neuroinflammation in the prefrontal cortex and hippocampus by inhibiting the NLRP3 signaling pathway Long-term CORT administration activates the inflammasome via NLRP3 and its linked signaling molecules, including caspase-1 and ASC regulation, which subsequently play a significant role in neuroinflammation and neurotoxicity. We examined the NLRP3 expression in PFC and hippocampus (Fig. 4a). Enhanced NLRP3 expression was detected in the hippocampal DG and CA1 subregions as well as the prefrontal cortex in CORT-induced depression mice (Fig. 4b–d). Both low and high doses of berberine treatment significantly reduced inflammasome activation, as demonstrated by decreased expression of NLRP3 (Fig. 4e–g). The western blotting results were consistent with the immunofluorescence labeling, indicating that long-term CORT administration remarkably increased the expression levels of NLRP3, caspase-1, and ASC in the hippocampus and prefrontal cortex in mice brain. Berberine treatment reversed the activation of the NLRP3 inflammasome, as demonstrated by decreased expression levels of NLRP3, caspase-1, and ASC (Fig. 4h, i). We further investigated whether the level of NLRP3 inflammasome could be correlated with key behavioral features of mice including time in center zone of OFC, number of zone transition of OFT, immobility duration of FST, immobility duration of TST, and sucrose preference of SPT. The NLRP3 inflammasome data obtained from 28 mice ($$n = 7$$ in each group) were pooled for correlation analysis. The components of NLRP3 inflammasome showed highly correlation between each other. A cluster of prefrontal NLRP3 inflammasome level inversely correlated with time in central zone, zone transition number, and sucrose preference. However, the NLRP3 inflammasome level in the PFC showed a positive correlation with immobile duration (Additional file 1: Fig. S3). These findings are consistent with the overall conclusions from the results presented in mice, indicating that NLRP3 expression is elevated in depressed mice and is decreased in the normal. Together, these results identify that CORT induced depression of mice via an elevated NLRP3 signaling pathway, while berberine exhibited antidepressant effects via a suppressed NLRP3 signaling pathway. Fig. 4Berberine reverses neuroinflammation in the prefrontal cortex and hippocampus by inhibiting the NLRP3 inflammasome. a Representation of the timing of BrdU injection and illustration of an image of the target of PFC and hippocampus. b Immunofluorescence of NLRP3 inflammasome in the dentate gyrus subregion of the hippocampus. Scar bar, 100 µm. c Immunofluorescence of NLRP3 inflammasome in the CA1 subregion of the hippocampus. Scar bar, 100 µm. d Immunofluorescence of NLRP3 inflammasome in the PFC. Scar bar, 100 µm. e NLRP3 + cells (green) and Iba1 + cells (red) in DG of hippocampus, CORT increased NLRP3 + /Iba1 + cells in DG (One-way ANOVA, F [3, 8] = 18.63, $$P \leq 0.0006$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0002$$, CORT vs. BBR100, $$P \leq 0.0112$$, CORT vs. BBR200, $$P \leq 0.0047$$, $$n = 3$$ in each group). f NLRP3 + cells (green) and Iba1 + cells (red) in CA1 of hippocampus, CORT increased NLRP3 + /Iba1 + cells in CA1 (One-way ANOVA, F [3, 8] = 12.28, $$P \leq 0.0023$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0021$$, CORT vs. BBR100, $$P \leq 0.7604$$, CORT vs. BBR200, $$P \leq 0.0115$$, $$n = 3$$ in each group). g NLRP3 + cells (green) and Iba1 + cells (red) in PFC, CORT increased the NLRP3 + /Iba1 + cells in PFC (One-way ANOVA, F [3, 8] = 141.0, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). h The western blotting of NLRP3/Caspase-1/ASC in the PFC, the CORT mice showed a significant difference in the NLRP3 signaling pathway (Two-way ANOVA, row factor F [2, 36] = 10.30, $$P \leq 0.0003$$, column factor F [3, 36] = 24.85, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). i The western blotting of NLRP3/Caspase-1/ASC in the hippocampus, the CORT mice showed a significant difference in the NLRP3 signaling pathway (Two-way ANOVA, row factor F [2, 41] = 25.79, $P \leq 0.0001$, column factor F [3, 41] = 37.04, $P \leq 0.0001$; Dunnett’s multiple comparisons tests, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001.$ CON control, CORT corticosterone, BBR berberine, ANOVA analysis of variance, NLRP3 NOD-like receptor thermal protein domain associated protein 3, qPCR quantitative reverse transcription polymerase chain reaction, PFC prefrontal cortex ## Effects of berberine on neural plasticity and hippocampal adult neurogenesis Combined with the alterations of behaviors and cytokine levels, we hypothesize that long-term administration of CORT induced neurotoxicity via neuroinflammation that is associated with NLRP3 inflammasome, resulting in cognitive deficits, which disrupts synaptic plasticity and contributes to dysregulated synaptogenesis. To further explore whether the effects of berberine could regulate neuroplasticity and neurogenesis, we detect the spine density in synaptic plasticity and neurogenesis in the hippocampus of mice via western blotting, immunofluorescence labeling, Golgi-Cox staining, and TEM (Fig. 5a). In the hippocampal DG and CA1 subregions of CORT mice, the expression levels of PSD95 and SYN were significantly decreased (Fig. 5b, c, f, g). Both low and high doses of berberine treatment significantly attenuated CORT-induced changes, indicating that berberine reduced the synaptic defects generated under CORT-induced stress results (Fig. 5b, c, f, g). The immunofluorescence labeling illustrated that the long-term CORT administration resulted in a significant reduction density of newly generated immature neuron labels with BrdU and DCX in the hippocampus in mice (Fig. 5d, e, h, i). Berberine treatment also increased both BrdU and DCX positive cells (Fig. 5d, e, h, i).Fig. 5Berberine increases hippocampal DG and CA1 subregional synaptic plasticity and adult neurogenesis induced by CORT. a Representation of the timing of experiments and illustration of the image of the target of PFC and hippocampus. b Hippocampal DG images show the decreasing signal intensity of PSD95 and SYN in the CORT group, and both BBR100 and BBR200 increased the PSD95 + and SYN + cell number (One-way ANOVA, F [3, 8] = 10.12, $$P \leq 0.0043$$; Dunnett’s multiple comparisons test, CORT vs. Control, $$P \leq 0.0021$$, CORT vs. BBR100, $$P \leq 0.0251$$, CORT vs. BBR200, $$P \leq 0.0081$$, $$n = 3$$ in each group). Scar bar, 100 µm. c Representation of the immunofluorescence labeling of hippocampal CA1, CORT significantly decreased the PSD95 + and SYN + cell number compared to BBR100 and BBR200 (One-way ANOVA, F [3, 8] = 84.63, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Scar bar, 100 µm. d NeuN (red) and BrdU (green) in hippocampal DG images (One-way ANOVA, F [3, 8] = 155.0, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Scar bar, 50 µm. e DAPI (blue) and DCX (red) in hippocampal DG images (One-way ANOVA, F [3, 8] = 220.0, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Scar bar, 100 µm. f percentage of PSD95 + /SYN + area in hippocampal DG. g percentage of PSD95 + /SYN + area in hippocampal CA1. h BrdU + cell density in hippocampal DG. i DCX + cell density in hippocampal DG. j Western blotting assay for PSD95 and SYN, the CORT mice showed a significant difference in the expression of PSD95 and SYN (Two-way ANOVA, row factor F [1, 19] = 11.90, $$P \leq 0.0027$$, column factor F [3, 19] = 30.06, $P \leq 0.0001$; Dunnett’s multiple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, and $P \leq 0.001.$ CORT corticosterone, BBR berberine, ANOVA analysis of variance, DG dentate gyrus, DCX Doublecortin, BrdU Bromodeoxyuridine, PSD95 postsynaptic density protein 95, SYN synapsin To confirm the function of dendritic function and morphology, the elimination of postsynaptic dendritic spines on the hippocampus was observed by Golgi-Cox staining. The results showed a significant loss of synaptic density in the mice after long-term CORT exposure (Fig. 6a, b). To explore how CORT exposure is related to dendritic spines, we classified dendritic spines into two categories based on the maximal diameter of the spine head (both thin and mushroom spines maximal diameter < 0.6 µm) and the length of spines (thin spines maximal length > 0.9 µm vs. mushroom spines maximal length < 0.9 µm). Long-term CORT administration markedly decreased the number of both mushroom and thin dendrite spines (Fig. 6a, c, d). Both low dosage (100 mg/kg/day) and high dosage (200 mg/kg/day) of berberine treatment attenuated the decrease of several spines in CORT-treated mice (Fig. 6a, c, d). The results from TEM also suggested that CORT-induced stress condition significantly decreases both mean depth and length of postsynaptic densities in CORT-induced mice than in control and berberine treatment groups (Fig. 6e–g).Fig. 6Berberine increases hippocampal adult neurogenesis induced by CORT. a Representation of the Golgi-Cox staining of hippocampal adult neuron. b Spine density results from Golgi staining (One-way ANOVA, F [3, 8] = 17.10, $$P \leq 0.0008$$; Dunnett’s multple comparisons test, CORT vs. Control, $$P \leq 0.0005$$, CORT vs. BBR100, $$P \leq 0.0112$$, CORT vs. BBR200, $$P \leq 0.0010$$, $$n = 3$$ in each group). c Proportion of mushroom spines (One-way ANOVA, F [3, 8] = 6.068, $P \leq 0.0001$; Dunnett’s multple comparisons test, CORT vs. Control, $$P \leq 0.0117$$, CORT vs. BBR200, $$P \leq 0.0432$$, $$n = 3$$ in each group). d Proportion of thin spines (One-way ANOVA, F [3, 8] = 52.74, $P \leq 0.0001$; Dunnett’s multple comparisons test, CORT vs. Control, $$P \leq 0.0001$$, CORT vs. BBR100, $$P \leq 0.0054$$, CORT vs. BBR200, $P \leq 0.0001$, $$n = 3$$ in each group). e Representative electron micrographs showing the synaptic structure and postsynaptic densities on neurons. Scar bar, 200 µm. f Depth of post synaptic dendrites (One-way ANOVA, F [3, 8] = 38.93, $P \leq 0.0001$; Dunnett’s multple comparisons test, CORT vs. Control, $P \leq 0.0001$, CORT vs. BBR100, $$P \leq 0.0033$$, CORT vs. BBR200, $$P \leq 0.0001$$, $$n = 3$$ in each group). g Length of post synaptic dendrites (One-way ANOVA, F [3, 8] = 220.0, $P \leq 0.0001$; Dunnett’s multple comparisons test, all comparisons $P \leq 0.0001$, $$n = 3$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, and $P \leq 0.001.$ CORT corticosterone, BBR berberine, ANOVA analysis of variance ## The effects of berberine on long-term CORT administration impaired the synaptic function of pyramidal neurons in the mPFC NLRP3 inflammasome activation induced neuroinflammatory response may also negatively affect the neuroplasticity, and the GO and KEGG analysis from RNA-seq results suggested that long-term CORT exposure involved the function of the synapse. To detect whether long-term CORT exposure alters the excitability of mPFC, the frequency of action potentials in response to depolarizing current steps was measured from pyramidal neurons in the mPFC. The number of action potentials elicited (induced spikes) throughout 500 ms was measured, as the current was varied in steps of 50 pA from − 50 to 400 pA (Fig. 7a, b). The pyramidal neurons were characterized as the cells that displayed spike-frequency adaption and broad action potentials and lacked spontaneous discharge at resting membrane potential. Depolarizing currents evoked higher firing in the mice of the control group and berberine-treated (BBR200) mice than in the CORT group. Similarly, less current was required to drive the cell to fire spikes at a given frequency (Fig. 7a, b). To examine the functional consequences of long-term CORT exposure on the synaptic transmission in the mPFC, a whole-cell patch-clamp recording of pyramidal neurons was performed. Spontaneous excitatory postsynaptic currents (sEPSCs) and mini excitatory postsynaptic currents (mEPSCs) were recorded in the same cells by alternate clamping at the reversal potential of glutamate receptor-mediated and α-amino-3-hydroxy-5-453 methylisoxazole-4-propionic acid receptor (AMPAR)-mediated currents, respectively (Fig. 7c, f). The sEPSCs and mEPSCs amplitude were comparable among groups (Fig. 7d, g), while for frequency of sEPSCs and mEPSCs, whereas CORT administration significantly inhibited both the frequency of sEPSCs and mEPSCs compared with mice in control and berberine group (200 mg/kg) (Fig. 7e, h), suggesting a deficit in CORT inhibits synaptic transmission and abnormal discharge in mPFC pyramidal neurons, which may then contribute to the depression-like behaviors observed in CORT administration mice. Fig. 7Berberine prevents CORT-induced patch-clamp alteration. a Representation of the data from the excitatory neuron of contrl (black), CORT (red), and BBR 200 mg/kg/day (green), which were stimulated by electricity from 50 to 350 pA, b representation of the data from the excitatory neuron, which were stimulated by electricity from 50 to 400 pA and records of the number of spikes (Two-way ANOVA, row factor F [7, 360] = 22.10, $P \leq 0.0001$, column factor F [2, 360] = 86.48, $P \leq 0.0001$; Tukey’s multple comparisons test, for 150 pA, CORT vs. control, $$P \leq 0.0008$$, CORT vs. BBR, $$P \leq 0.0042$$; for 250 pA, CORT vs. control, $P \leq 0.0001$, CORT vs. BBR, $$P \leq 0.0021$$; for 350 pA, CORT vs. control, $P \leq 0.0001$, CORT vs. BBR, $$P \leq 0.0023$$, $$n = 3$$ in each group). c Representation of the mEPSC of contrl (black), CORT (red), and BBR 200 mg/kg/day (green) d Representation of the amplitude of mEPSC (One-way ANOVA, F [2, 31] = 1.038, $$P \leq 0.3662$$). e Frequency of mEPSC was significantly lower in CORT compared to CON and BBR 200 mg/kg/day (One-way ANOVA, F [2, 31] = 6.972, $$P \leq 0.0032$$; Tukey’s multple comparisons test, CORT vs. control, $$P \leq 0.0046$$, CORT vs. BBR, $$P \leq 0.0388$$, $$n = 3$$ in each group). f Representation of the sEPSC of contrl (black), CORT (red), and BBR 200 mg/kg/day (green) g Representation of the amplitude of sEPSC (One-way ANOVA, F [2, 30] = 0.7402, $$P \leq 0.4855$$, $$n = 3$$ in each group). h Representation of the the frequency of sEPSC was significantly lower in CORT compared to CON and BBR 200 mg/kg/day (One-way ANOVA, F [2, 32] = 20.70, $P \leq 0.0001$; Tukey’s multple comparisons test, CORT vs. control, $$P \leq 0.0040$$, CORT vs. BBR, $P \leq 0.0001$, $$n = 3$$ in each group). Bar graphs show the mean ± SEM; $P \leq 0.05$, $P \leq 0.01$, and $P \leq 0.001.$ CORT, corticosterone; BBR, berberine; ANOVA, analysis of variance; mEPSC, miniature excitatory postsynaptic current; sEPSC, spontaneous excitatory postsynaptic current ## Discussion Despite the growing number of evidence suggesting a strong link between neuropsychiatric disorders and immune responses. To date, pharmacological therapies with satisfactory effectiveness are still elusive [19]. The current findings indicate that after long-term CORT exposure, the induced depression-like behaviors were associated with remarkably neuroinflammatory responses regarding NLRP3 inflammasome and cytokines as well as electrophysiological function in PFC and structural neuronal in the hippocampus of mice. Notably, berberine attenuated the behavioral alterations and regulated the NLRP3 signaling pathway with overexpressed cytokines, consequently, reversed the deterioration of neural plasticity induced by CORT. In this study, DEG analysis were initially identified in PFC via assessing the functions of genes by using high-throughput sequencing. NLRP3 was found to be overexpressed as a hub gene in the PFC of CORT-induced depression mice, suggesting that NLRP3 might play a key role and take part in the development and pathogenesis of depression. Findings from previous preclinical and clinical studies have indicated that NLRP3 inflammasome-driven pathways might be involved in numerous neuropsychiatric disorders including neuroinflammation-induced depression [14, 27–30]. As revealed by western blotting and immunofluorescence labeling, the current data suggest that the NLRP3 signaling pathway was activated in CORT-induced depression mice, while berberine downregulated the NLRP3 signaling pathway in the hippocampus and PFC. IL-1β is one of the main mediators of the crosstalk between the immune system and the CNS. The NLRP3 inflammasome processes pro-IL-1β into mature interleukins and act as a pro-inflammatory mediator, it has been confirmed that IL-1β was elevated in the serum of depression patients and associated with depression inventory scores [31, 32]. Together, these results confirmed that NLRP3 was involved in the process of neuroinflammation and served as a modulator in the development of depression. Meanwhile, several studies have demonstrated that depression is accompanied by dendritic remodeling in neurons of the PFC and hippocampal [33–35]. The brain-derived RNA could regulate dendritic spine development and may thus be involved in regulating neural plasticity and behaviors [36]. The structural and morphologic characteristics of neurons from Golgi-staining and TEM showed that CORT induced specific alterations in the synapse. In addition, CORT administration decreased excitatory synaptic transmission function as revealed from electrophysiological recordings of mEPSCs and sEPSCs in pyramidal neurons in the mPFC. These changes might be attributed to impaired neurobiological functions as associated with the NLRP3 inflammasome activation. It is well-documented that glucocorticoids are efficacy anti-inflammatories and have been frequently applied for inflammatory conditions including autoimmune diseases. However, hypersecretion glucocorticoids lay a foundation for the neuroinflammation hypothesis regarding the pathogenesis of depression [37, 38]. Immunosuppressive activities of glucocorticoids negatively regulate pro-inflammatory related pathways including glucocorticoid-mediated activation [39]. Specifically, Toll-like receptors (TLRs) could be activated by glucocorticoids, increasing the expression of several members of the TLRs family which is critical for inflammatory response [40]. As the agonist of the glucocorticoid receptor, CORT activated glucocorticoid receptor increased the expression of purinergic receptor and result in the over-secretion of IL-6 [41, 42]. Meanwhile, it can stimulate the activation of the NLRP3 inflammasome, by facilitating NLRP3 induction and inflammasome formation, glucocorticoids promote activation of neuroinflammation and the release of pro-inflammatory cytokine IL-1β. TLRs in circulating monocytes could be activated by DAMPs or PAMPs, the NLRP3 inflammasome transcription and combine ASC and pro-caspase-1, then processes pro-IL-1β to IL-1β, leading to release of mature cytokines in the extracellular milieu and inflammatory response [15]. The overproduction of cytokines such as IL-1β, TNF-α, and IFN-γ further result in the dysfunction of monoamine metabolism, including overexpression of indoleamine 2,3-dioxygenase (IDO), leading to the production of kynurenine metabolites from tryptophan [43]. As a natural IDO inhibitor, berberine decreased the production of kynurenine, which is subsequently converted into metabolites having modulatory effects on glutamatergic neurotransmission [44, 45]. These findings illustrate that the potential anti-depressant effects of berberine are attributed to its anti-inflammatory especially for inhibiting NLRP3 inflammasome activation. It should be noted that our results were limited to PFC and hippocampus, some other brain regions related to depression such as hypothalamus, amygdala, and locus coeruleus were not involved in the current study. ## Conclusion In conclusion, the findings of this study illustrated that berberine, via suppression of the NLRP3/caspase-1/ASC/IL-1β signaling pathway, play a key role in the alleviation of neuronal anomalies accompanied by emotional alteration induced by long-term CORT administration. Berberine attenuates the impairment of neuroplasticity and neurogenesis by inhibiting the neuroinflammatory response in PFC and hippocampus. ## Supplementary Information Additional file 1. Library preparation, transcriptome sequencing, and bioinformatic analysis for RNA-seq data. Fig. S1. Body weight of mice during experiments. Fig. S2. Uncropped western blot imaging. Fig. S3. Scatter plots of linear regression demonstrating the association between behavioral parameters and NLRP3 inflammasome. Table S1. The primers used in qPCR. Table S2. Antibody information for Western blotting. Table S3. Antibody information of immunofluorescence. Table S4. Sequencing quality control measures. ## References 1. Diseases Injuries Collaborators: global, regional, and national burden of 12 mental disorders in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet Psychiatry* (2019) 2022 137-150 2. Diseases Injuries Collaborators: global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet (2019) 2020 1204-1222 3. Malhi GS, Mann JJ. Depression. Lancet (2018) 392 2299-2312. DOI: 10.1016/S0140-6736(18)31948-2 4. Belmaker RH, Agam G. Major depressive disorder. N Engl J Med (2008) 358 55-68. DOI: 10.1056/NEJMra073096 5. Abdallah CG, Sanacora G, Duman RS, Krystal JH. Ketamine and rapid-acting antidepressants: a window into a new neurobiology for mood disorder therapeutics. Annu Rev Med (2015) 66 509-523. DOI: 10.1146/annurev-med-053013-062946 6. Duman RS, Aghajanian GK. Synaptic dysfunction in depression: potential therapeutic targets. Science (2012) 338 68-72. DOI: 10.1126/science.1222939 7. Miller AH, Maletic V, Raison CL. Inflammation and its discontents: the role of cytokines in the pathophysiology of major depression. Biol Psychiatry (2009) 65 732-741. DOI: 10.1016/j.biopsych.2008.11.029 8. Wu Y, Dissing-Olesen L, MacVicar BA, Stevens B. Microglia: dynamic mediators of synapse development and plasticity. Trends Immunol (2015) 36 605-613. DOI: 10.1016/j.it.2015.08.008 9. Ellul P, Mariotti-Ferrandiz E, Leboyer M, Klatzmann D. Regulatory T cells as supporters of psychoimmune resilience: toward immunotherapy of major depressive disorder. Front Neurol (2018) 9 167. DOI: 10.3389/fneur.2018.00167 10. Wang H, He Y, Sun Z, Ren S, Liu M, Wang G, Yang J. Microglia in depression: an overview of microglia in the pathogenesis and treatment of depression. J Neuroinflamm (2022) 19 132. DOI: 10.1186/s12974-022-02492-0 11. Bai S, Guo W, Feng Y, Deng H, Li G, Nie H, Guo G, Yu H, Ma Y, Wang J. Efficacy and safety of anti-inflammatory agents for the treatment of major depressive disorder: a systematic review and meta-analysis of randomised controlled trials. J Neurol Neurosurg Psychiatry (2020) 91 21-32. DOI: 10.1136/jnnp-2019-320912 12. Adzic M, Brkic Z, Mitic M, Francija E, Jovicic MJ, Radulovic J, Maric NP. Therapeutic strategies for treatment of inflammation-related depression. Curr Neuropharmacol (2018) 16 176-209. DOI: 10.2174/1570159X15666170828163048 13. Mangan MSJ, Olhava EJ, Roush WR, Seidel HM, Glick GD, Latz E. Targeting the NLRP3 inflammasome in inflammatory diseases. Nat Rev Drug Discov (2018) 17 688. DOI: 10.1038/nrd.2018.149 14. Kaufmann FN, Costa AP, Ghisleni G, Diaz AP, Rodrigues ALS, Peluffo H, Kaster MP. NLRP3 inflammasome-driven pathways in depression: clinical and preclinical findings. Brain Behav Immun (2017) 64 367-383. DOI: 10.1016/j.bbi.2017.03.002 15. Gong T, Liu L, Jiang W, Zhou R. DAMP-sensing receptors in sterile inflammation and inflammatory diseases. Nat Rev Immunol (2020) 20 95-112. DOI: 10.1038/s41577-019-0215-7 16. Albert PR. Adult neuroplasticity: a new "cure" for major depression?. J Psychiatry Neurosci (2019) 44 147-150. DOI: 10.1503/jpn.190072 17. Huang YJ, Lane HY, Lin CH. New treatment strategies of depression: based on mechanisms related to neuroplasticity. Neural Plast (2017) 2017 4605971. DOI: 10.1155/2017/4605971 18. Wang J, Hodes GE, Zhang H, Zhang S, Zhao W, Golden SA, Bi W, Menard C, Kana V, Leboeuf M. Epigenetic modulation of inflammation and synaptic plasticity promotes resilience against stress in mice. Nat Commun (2018) 9 477. DOI: 10.1038/s41467-017-02794-5 19. Troubat R, Barone P, Leman S, Desmidt T, Cressant A, Atanasova B, Brizard B, El Hage W, Surget A, Belzung C, Camus V. Neuroinflammation and depression: a review. Eur J Neurosci (2021) 53 151-171. DOI: 10.1111/ejn.14720 20. Worthen RJ, Garzon Zighelboim SS, Torres Jaramillo CS, Beurel E. Anti-inflammatory IL-10 administration rescues depression-associated learning and memory deficits in mice. J Neuroinflamm (2020) 17 246. DOI: 10.1186/s12974-020-01922-1 21. Li D, Chen M, Meng T, Fei J. Hippocampal microglial activation triggers a neurotoxic-specific astrocyte response and mediates etomidate-induced long-term synaptic inhibition. J Neuroinflamm (2020) 17 109. DOI: 10.1186/s12974-020-01799-0 22. Fan D, Liu L, Wu Z, Cao M. Combating neurodegenerative diseases with the plant alkaloid berberine: molecular mechanisms and therapeutic potential. Curr Neuropharmacol (2019) 17 563-579. DOI: 10.2174/1570159X16666180419141613 23. Fan J, Zhang K, Jin Y, Li B, Gao S, Zhu J, Cui R. Pharmacological effects of berberine on mood disorders. J Cell Mol Med (2019) 23 21-28. DOI: 10.1111/jcmm.13930 24. Liu YM, Niu L, Wang LL, Bai L, Fang XY, Li YC, Yi LT. Berberine attenuates depressive-like behaviors by suppressing neuro-inflammation in stressed mice. Brain Res Bull (2017) 134 220-227. DOI: 10.1016/j.brainresbull.2017.08.008 25. Duman RS, Aghajanian GK, Sanacora G, Krystal JH. Synaptic plasticity and depression: new insights from stress and rapid-acting antidepressants. Nat Med (2016) 22 238-249. DOI: 10.1038/nm.4050 26. Gao C, Du Q, Li W, Deng R, Wang Q, Xu A, Shen J. Baicalin modulates APPL2/glucocorticoid receptor signaling cascade, promotes neurogenesis, and attenuates emotional and olfactory dysfunctions in chronic corticosterone-induced depression. Mol Neurobiol (2018) 55 9334-9348. DOI: 10.1007/s12035-018-1042-8 27. Yang XJ, Zhao BC, Li J, Shi C, Song YQ, Gao XZ, Jiang HL, Yu QY, Liang XC, Feng SX. Serum NLRP3 inflammasome and BDNF: potential biomarkers differentiating reactive and endogenous depression. Front Psychiatry (2022) 13 814828. DOI: 10.3389/fpsyt.2022.814828 28. Wang D, Wang H, Gao H, Zhang H, Zhang H, Wang Q, Sun Z. P2X7 receptor mediates NLRP3 inflammasome activation in depression and diabetes. Cell Biosci (2020) 10 28. DOI: 10.1186/s13578-020-00388-1 29. Komleva YK, Lopatina OL, Gorina IV, Shuvaev AN, Chernykh A, Potapenko IV, Salmina AB. NLRP3 deficiency-induced hippocampal dysfunction and anxiety-like behavior in mice. Brain Res (2021) 1752 147220. DOI: 10.1016/j.brainres.2020.147220 30. Li S, Sun Y, Song M, Song Y, Fang Y, Zhang Q, Li X, Song N, Ding J, Lu M, Hu G. NLRP3/caspase-1/GSDMD-mediated pyroptosis exerts a crucial role in astrocyte pathological injury in mouse model of depression. JCI Insight (2021) 6 e146852. DOI: 10.1172/jci.insight.146852 31. Alcocer-Gomez E, Cordero MD. NLRP3 inflammasome: a new target in major depressive disorder. CNS Neurosci Ther (2014) 20 294-295. DOI: 10.1111/cns.12230 32. Alcocer-Gomez E, de Miguel M, Casas-Barquero N, Nunez-Vasco J, Sanchez-Alcazar JA, Fernandez-Rodriguez A, Cordero MD. NLRP3 inflammasome is activated in mononuclear blood cells from patients with major depressive disorder. Brain Behav Immun (2014) 36 111-117. DOI: 10.1016/j.bbi.2013.10.017 33. Wang L, Chang X, She L, Xu D, Huang W, Poo MM. Autocrine action of BDNF on dendrite development of adult-born hippocampal neurons. J Neurosci (2015) 35 8384-8393. DOI: 10.1523/JNEUROSCI.4682-14.2015 34. Pavlov D, Bettendorff L, Gorlova A, Olkhovik A, Kalueff AV, Ponomarev ED, Inozemtsev A, Chekhonin V, Lessmall es CKP, Anthony DC, Strekalova T. Neuroinflammation and aberrant hippocampal plasticity in a mouse model of emotional stress evoked by exposure to ultrasound of alternating frequencies. Prog Neuropsychopharmacol Biol Psychiatry (2019) 90 104-116. DOI: 10.1016/j.pnpbp.2018.11.014 35. Rogers JT, Morganti JM, Bachstetter AD, Hudson CE, Peters MM, Grimmig BA, Weeber EJ, Bickford PC, Gemma C. CX3CR1 deficiency leads to impairment of hippocampal cognitive function and synaptic plasticity. J Neurosci (2011) 31 16241-16250. DOI: 10.1523/JNEUROSCI.3667-11.2011 36. Di Filippo M, Sarchielli P, Picconi B, Calabresi P. Neuroinflammation and synaptic plasticity: theoretical basis for a novel, immune-centred, therapeutic approach to neurological disorders. Trends Pharmacol Sci (2008) 29 402-412. DOI: 10.1016/j.tips.2008.06.005 37. Pariante CM. Why are depressed patients inflamed? A reflection on 20 years of research on depression, glucocorticoid resistance and inflammation. Eur Neuropsychopharmacol (2017) 27 554-559. DOI: 10.1016/j.euroneuro.2017.04.001 38. Sorrells SF, Sapolsky RM. An inflammatory review of glucocorticoid actions in the CNS. Brain Behav Immun (2007) 21 259-272. DOI: 10.1016/j.bbi.2006.11.006 39. Revest JM, Di Blasi F, Kitchener P, Rouge-Pont F, Desmedt A, Turiault M, Tronche F, Piazza PV. The MAPK pathway and Egr-1 mediate stress-related behavioral effects of glucocorticoids. Nat Neurosci (2005) 8 664-672. DOI: 10.1038/nn1441 40. Fu S, Wang J, Hao C, Dang H, Jiang S. Tetramethylpyrazine ameliorates depression by inhibiting TLR4-NLRP3 inflammasome signal pathway in mice. Psychopharmacology (2019) 236 2173-2185. DOI: 10.1007/s00213-019-05210-6 41. Yue N, Huang H, Zhu X, Han Q, Wang Y, Li B, Liu Q, Wu G, Zhang Y, Yu J. Activation of P2X7 receptor and NLRP3 inflammasome assembly in hippocampal glial cells mediates chronic stress-induced depressive-like behaviors. J Neuroinflamm (2017) 14 102. DOI: 10.1186/s12974-017-0865-y 42. Zhang B, Zhang Y, Wu W, Xu T, Yin Y, Zhang J, Huang D, Li W. Chronic glucocorticoid exposure activates BK-NLRP1 signal involving in hippocampal neuron damage. J Neuroinflamm (2017) 14 139. DOI: 10.1186/s12974-017-0911-9 43. Huang YS, Ogbechi J, Clanchy FI, Williams RO, Stone TW. IDO and kynurenine metabolites in peripheral and CNS disorders. Front Immunol (2020) 11 388. DOI: 10.3389/fimmu.2020.00388 44. Ge PY, Qu SY, Ni SJ, Yao ZY, Qi YY, Zhao X, Guo R, Yang NY, Zhang QC, Zhu HX. Berberine ameliorates depression-like behavior in CUMS mice by activating TPH1 and inhibiting IDO1-associated with tryptophan metabolism. Phytother Res (2023) 37 342-357. DOI: 10.1002/ptr.7616 45. Yu D, Tao BB, Yang YY, Du LS, Yang SS, He XJ, Zhu YW, Yan JK, Yang Q. The IDO inhibitor coptisine ameliorates cognitive impairment in a mouse model of Alzheimer's disease. J Alzheimers Dis (2015) 43 291-302. DOI: 10.3233/JAD-140414
--- title: Research on the wound healing effect of Shengji Huayu Formula ethanol extract-derived fractions in streptozotocin-induced diabetic ulcer rats authors: - Jing-Ting Zhang - Min-Feng Wu - Ming-Hua Ma - Liang Zhao - Jian-Yong Zhu - Hua Nian - Fu-Lun Li journal: BMC Complementary Medicine and Therapies year: 2023 pmcid: PMC9976525 doi: 10.1186/s12906-023-03894-0 license: CC BY 4.0 --- # Research on the wound healing effect of Shengji Huayu Formula ethanol extract-derived fractions in streptozotocin-induced diabetic ulcer rats ## Abstract ### Background Diabetic ulcer is a common complication of diabetes. It is characterized by a long-term disease course and high recurrence rate. Shengji Huayu Formula (SHF) is an effective formula for treating diabetic ulcers. However, the specific effective parts of SHF remain unclear. Clarifying the active polar site of SHF would be helpful to refine research on the components in SHF that promote wound healing. This research aims to focus on evaluating the activity of polar fractions. ### Methods A diabetic rat model was established by intraperitoneally injecting streptozotocin (STZ) and was adopted to confirm the therapeutic effect of SHF. Four different polarity parts were extracted from SHF and prepared into a cream to evaluate the activity. High-performance liquid chromatography (HPLC) was used to detect chemical constituents in chloroform extracts. ### Results It was discovered that dracorhodin, aloe-emodin, rhein, imperatorin, emodin, isoimperatorin, chrysophanol, physcion, and tanshinone IIA were the main components of the chloroform extract from SHF. The results revealed that chloroform extract could effectively accelerate diabetic wound healing by promoting collagen regeneration and epidermal repair. Chloroform extract of SHF could stimulate the generation of vascular endothelial growth factor (VEGF). The results are also indicated that the effective active fraction was the chloroform part, and the method of detecting the main chemical constituents in the active part was successfully established. ### Conclusion SHF could improve diabetic ulcers by promoting granulation tissue synthesis. In this study, four polar parts (petroleum ether, chloroform, ethylacetate, n-butanol) were extracted from a $95\%$ ethanol extract. In contrast, chloroform polar parts showed a higher wound closure rate, stimulated more collagen regeneration and promoted more production of vascular endothelial cells. In conclusion, the chloroform extract of SHF was the effective polar part in ameliorating diabetic wound healing. ## Background Diabetes mellitus (DM) is a type of endocrine disease that is estimated to affect 284.6 million people worldwide [1]. The International Diabetes Federation (IDF) reported that 451 million people suffered from diabetes in 2017, and this number is expected to increase to 693 million by 2045 [2]. Diabetic ulcer, one of the most common complications of DM, is a kind of chronic and refractory cutaneous ulcer with several characteristics, including long-term refractoriness, easy recurrence and high incidence. In China, the morbidity of diabetic ulcers has reached $8.1\%$ [3]. It is estimated that chronic trauma causes losses of over $25 billion a year in the U.S., leading to an increase in medical costs [4, 5]. Diabetic ulcers cause great pain and a heavy economic burden to patients. Therefore, studying effective drugs and treatment methods to improve diabetic ulcers is an urgent task. At present, anti-infection, hyperbaric oxygen therapy and surgery are common treatments in the clinic. However, these therapies have several limitations, including a slow effect and difficulties in scab formation. Traditional Chinese medicine (TCM) has advantages in treating diabetic ulcers [6, 7]. External treatment has been a characteristic TCM therapy since ancient times. The effective substances and chemical constituents of TCM play a vital role in the healing of diabetic ulcers. Shengji Huayu Formula (SHF) has been applied in the clinic for decades and is a safe and effective therapy for treating chronic cutaneous ulcers, especially diabetic wounds. SHF could reduce wound closure time with subtle pain and low treatment expenditure [8–11]. SHF can improve local blood circulation and accelerate wound granulation, epithelium regeneration and healing processes [8–11]. However, the effect of the active polar component of this formula remains unknown. We previously used several animal models to assess the active effects of SHF, including the high-fat diet-induced diabetic ulcer model to assess the $95\%$ ethanolic extract [11], female clean-grade diabetic mouse models to evaluate the $70\%$ ethanolic extract [12], and STZ-induced animal models to assess the $95\%$ ethanolic extract [13]. All animal experiment results showed that SHF ethanol extract could accelerate re-epithelialization and reduce diabetic mouse wound healing time. In in vitro studies, human dermal microvascular endothelial cells (HDMECs) and shRNA interference were used to explore the effects of SHF ethanol extract on cell migration, PGT, PGE2, and the angiogenesis factor VEGF. Our in vitro studies confirmed that SHF ethanol extract could accelerate re-epithelialization and reduce inflammation by regulating the Activin/Follistatin imbalance [11, 12]. In addition, the molecular mechanisms of SHF in treating diabetic ulcers were revealed by transcriptional profiling and network analysis in recent years [9, 10]. All of these studies explain the SHF mechanism from numerous perspectives, but the active polar parts of the SHF ethanol extract need to be investigated further. Thus, this research was designed to verify the active polar parts in SHF with the function of promoting wound healing. The most active part, the chloroform fraction, was established as the HPLC method to detect the main chemical constituents. Comprehensive and objective evaluation of the chemical constituents in SHF contributes to providing scientific evidence for diabetic ulcer treatment. ## External TCM ointment preparation SHF contained eight Chinese herbs, as shown in Table 1. The dosage used in the present study was determined according to the Chinese Pharmacopoeia (2015 edition). The air-dried powder of herbs in SHF was extracted three times with $95\%$ ethanol at room temperature to produce a crude extract upon removal of the solvent. The extract was suspended in water and partitioned successively with petroleum ether, chloroform, ethyl acetate, and n-butanol at a 1:1 ratio 4 times to afford four corresponding portions. The four concentrated solutions were evaporated in vacuo to produce a semisolid residue, which was mixed with swollen carbomer and triethanolamine to adjust the pH value to 6–8, labeled and stored in a refrigerator at 4 °C. Therefore, four polar fractions were prepared as shown (Fig. 1). As a negative control, the blank gel matrix, 0.60 g carbomer, and 0.50 g glycerin were dissolved in 18 mL water and were swelled overnight. The positive control was recombinant bovine basic fibroblast growth factor (rb-bFGF).Table 1The composition of SHFLatin scientific nameChinese namePlant partWeight (g)%Astragalus membranceus (Fisch.) BgeHuangqiRadix30.015.8Salvia miltiorrhiza BgeDanshenRhizoma15.07.9Angelica dahurica (Fisch ex Hoffm.) Benth. et Hook. fBaizhiRadix30.015.8Rheum palmatum LDahuangRhizoma15.07.9Daemonorops draco BlXuejieResin10.05.3Arnebia euchroma (Royle) JohnstZicaoRadix30.015.8Pteria martensii (Dunker)Zhenzhufen/30.015.8CalamineLuganshi/30.015.8Fig. 1Experimental drugs used in negative control group, positive control drug group and four polar parts group. Take carbomer as substrate. Left (up): negative control (carbomer) and positive control (rb-bFGF). Right: four polarity components a Petroleum ether-extract; b Chloroform-extract; c Ethyl acetate-extract; d n-butanol-extract ## Animal Sprague Dawley (SD) rats (8 weeks old, 150 ± 5 g) were obtained from Shanghai SLAC Laboratory Animal Co., Ltd. (SLAC Shanghai 2012–0002) and kept under standard temperature (25 °C) in the laboratory of Shanghai University of Traditional Chinese Medicine. In total, 96 male SD rats were randomly divided into six groups (excluding rats that died in the process of modeling and whose blood glucose did not meet the standard), 16 in the negative control group, 16 in the positive control (rb-bFGF), 10 in the petroleum ether group, 14 in the chloroform group, 10 in the ethyl acetate group, and 14 in the n-butanol group. In this study, rats were anesthetized with isoflurane and a small animal anesthesia machine. After the experiments were finished, all rats were placed in a closable box and euthanized by CO2 inhalation to be suffocated. After CO2 inhalation, rats were subjected to cervical dislocation with eyeball whitening, heart failure and respiratory arrest, which confirmed death. The study on rats was approved by the ethical committee of Shanghai University of Traditional Chinese Medicine (No. 16661, 16,702). ## Diabetic wound model Diabetic rat models were established according to the classic modeling method [14]. After 3 days of acclimatization, the 8‑week‑old mice were fed a high-fat and high-sugar diet consisting of $54.6\%$ basic mouse feed, $16.9\%$ lard, $14\%$ sugar, $10.2\%$ casein, $2.1\%$ premix, and $2.2\%$ maltodextrin for two weeks. Then, the diabetic model was induced by intraperitoneal injection of $1\%$ STZ solution of 50 mg/kg and intragastric administration of $10\%$ glucose solution 2 h later to balance blood glucose. The weight of the rats was measured on the day of modeling, and tail tip blood was collected to measure blood glucose. One week after the last STZ injection, rats with blood glucose levels over 16.7 mmol/L, polyuria, polydipsia and intense hunger symptoms were considered to have successfully developed diabetes mellitus. On the day of modeling, rats were weighed and anesthetized with $2\%$-$3\%$ isoflurane. After anesthesia, the model area (both sides of the spine) was depilated with a depilatory knife. Under aseptic conditions, skin wounds with a diameter of 0.6 cm were made into diabetic rats with an area of 0.28 cm2. Each rat had 6 holes on the back. The depth of the wound reached the level of the subfascial dressing, and every rat was fed continuously in an individual cage separately. Intergroup markers should be made for high-fat and high-sugar diets. ## Drug delivery and specimen collection The rats in the petroleum ether group, chloroform group, ethyl acetate group, n-butanol group, positive control group, and negative control group were treated by intergroup comparison. After the model was established, the rats in the treatment group were treated with four polar parts of SHF, while the rats in the positive control group were treated with rb-bFGF, and the rats in the negative control group were treated with carbomer. The ointment was applied at a dose of 0.5 g/cm2/day immediately after the punch, and wound dressing was performed once a day. The wound was left uncovered. On Days 3, 7, and 9 after model establishment, the wound area of the rats was measured. On Day 9, half of the rats in each group were euthanized by CO2 inhalation, and the remaining rats were sacrificed after healing. A piece of basal muscle tissue of the wound was cut quickly. Two wounds in each group were fixed with $4\%$ paraformaldehyde, and $75\%$ alcohol was exchanged after 24 h. The remaining wounds were placed in the refrigerator at − 80 ℃. ## Wound healing process evaluation The wound area of the rats was recorded on Days 3, 7, 9, and 11. The measurement of wound area was conducted by visual inspection, digital camera and ImageJ software. First, a circular piece of paper with a diameter as a standard reference was placed over the wound. A digital camera was used to record the wound shape and outline from a fixed distance. Then, ImageJ 1.49v was adopted to analyze and evaluate the size of the wound according to wound photos at Day 9. ImageJ can calculate the irregular wound area by extracting the background color, enhancing the color difference and accumulating the sum of pixels. The area data were obtained, and the percentage of wound area was calculated as WC %. Wound closure (%) = (1 − WC)/WO × $100\%$. WC: wound area at the current observation time point. WO: original wound area. ## Histological examination by HE staining The collected samples were dehydrated with different concentrations of ethanol, embedded and sliced at a thickness of 5 µm. The samples were baked for 1 h in a 60 ℃ thermostat, paraffin was removed three times with xylene, 10 min each time, washed twice, 5 min each time, and treated with Harris hematoxylin staining for 5 min. Then, the samples were washed with water for 5 min, differentiated by $1\%$ hydrochloric acid alcohol solution for 5 s, washed with tap water for 15 min and stained with $0.5\%$ eosin (water solubility) for 1 min, $80\%$ ethanol for 2 min, $95\%$ ethanol 2 times for 5 min each time, and $100\%$ ethanol 2 times for 5 min each time. Finally, xylene was transparently treated twice, 5 min each time, 1–2 drops of gum were added, and glasses were added to seal it. After the samples were collected, they were placed into $4\%$ formalin solution. The samples were then paraffin-embedded, sectioned, and stained with hematoxylin and eosin (HE). Histopathological changes were observed under a light microscope. ## Immunohistochemical staining and evaluation The wounds were resected immediately after the rats were killed and fixed in $4\%$ neutral buffered paraformaldehyde at 4 °C for 24 h. Selected samples were embedded in paraffin, sectioned 5 µm thick, deparaffinized, and rehydrated with PBS (pH 7.4), and the antigen was retrieved with high temperature and pressure for 5 min, incubated with aqueous $3\%$ H2O2-methanol for 10 min, washed with PBS 3 times × 5 min, and stained serially with anti-PCNA and anti-VEGF at 4 °C overnight. The slices were incubated with a secondary antibody of IgG-HRP at 37 °C for 60 min, washed with PBS 3 times × 5 min, and incubated with DAB for 5 min. The reaction was terminated with water for 15 min and counterstained with hematoxylin. Sections were mounted with 1–2 drops of gum after transparency with xylene. For the nuclear staining protein PCNA and cytoplasmic staining protein VEGF, semiquantitative analysis was conducted using ImageJ software. Percentage-positive cells were calculated as the number of positively stained cells × 100/total number of cells in photomicrographs of tissue. The percentage of positive cells was calculated in a high-power field (HPF) (magnification 400 ×) and repeated for 10 HPFs. The arithmetic mean ± standard error deviation of counts was used for statistical analysis. ## Preparation of standard and sample solutions of chloroform for HPLC To certify the HPLC method, a standard stock solution was prepared and treated with a gradient mixed reference solution in methanol to the spiked concentration (2.58–258 µg/mL for dracorhodin perchlorate, 1.86–185.71 µg/mL for aloe emodin, 2.26–226 µg/mL for rhein, 1.31–131 µg/mL for imperatorin, 3.06–306.43 µg/mL for emodin, 4.42–442 µg/mL for isoimperatorin, 4.14–414 µg/mL for chrysophanol, 1.59–159.29 µg/mL for physcion, 1.63–162.86 µg/mL for tanshinone IIA). For the determination of the chloroform part of the SHF, the concentrated solution was extracted with chloroform at a 1:1 ratio 4 times, and the extractions were mixed together, concentrated to 0.72 g/mL by rotary evaporation, and then diluted to a proper concentration. Three replicates were used for each sample. All standards and sample preparations were filtered through a 0.45 µm membrane filter before injection into the HPLC system for analysis. ## Equipment and chromatographic conditions Chromatographic column: Agilent 1100 series HPLC, Agilent Zorbax Eclipse XDB-C18 (4.6 mm × 250 mm, 5 µm), column number: 990967–902, mobile phase: A phase is acetonitrile, B phase is $0.2\%$ formic acid aqueous solution (0 min-10 min: $30\%$ A, 50 min: $45\%$ A, 60 min: $50\%$ A, 75–90 min: $65\%$ A, 95–110 min: $95\%$ A), flow rate: 1 mL/min. The detection wavelength was 254 nm. The column temperature was 30 ℃. The injection volume was 10 µL, and the gradient of mobile phase was used as the initial condition to balance 20 min before injection. ## Statistical analysis SPSS 21.0 software was used for statistics, and data are expressed as the mean ± SEM. Statistical analysis for differences among groups was tested by one-way ANOVA with Dunnett’s or Tukey’s multiple comparisons test. Statistically significant results were expressed as $p \leq 0.05.$ ## General observation and random blood glucose Body weight was reduced, and intake, water intake and excretion were significantly increased, which was in accordance with the classic characteristics of diabetes mellitus. The random blood glucose of the tail vein of rats was stable after 3, 7, 9, and 11 days of wound modeling (Fig. 2).Fig. 2The blood glucose values of rat tail vein were stable after 3, 7, 9, and 11 days of wound modeling. Normal: Before modeling; Negative control:carbomer; Positive control:rb-bFGF; PE: Petroleum ether-extract; CHCl3: Chloroform-extract; EA: Ethyl acetate- extract; n-butanol: n-butanol-extract. * $p \leq 0.05$ vs normal ## Analysis of the wound closure area and the active polar parts of the SHF On Days 3 and 7 of administration, no significant difference was observed in the wound area (Fig. 3) between the four polar parts (petroleum ether, chloroform, ethylacetate, and n-butanol) group and the negative control group ($p \leq 0.05$) or among the four polar parts ($p \leq 0.05$). On Days 9 and 11, the healing degree of wounds in the positive control group and chloroform group was significantly different from that in the negative control group ($p \leq 0.05$). Significant differences were discovered between the chloroform group and four other groups (negative control, petroleum ether, ethyl acetate, and n-butanol groups) ($p \leq 0.05$).Fig. 3Photographic representation of wound closure on different post wounding days. A The morphological changes of polarity components wound at different time points. Wounds were marked with a ruler and photographed by a camera on day 3, 7, 9, and 11 after rat were modeled. B Effect of polarity components wound area ratio of each group at day 3, 7, 9, and 11 (Mean ± SEM). Negative control:carbomer; Positive control:rb-bFGF; PE: Petroleum ether-extract; CHCl3: Chloroform-extract; EA: Ethyl acetate-extract; n-butanol: n-butanol-extract. * $p \leq 0.05$ vs negative control group ## HE staining of active polar parts of SHF The results of HE staining on Day 9 showed that the wound epidermis of the negative control group was thin, and collagen in the dermis was loose with blurred cell layers and irregular layers. In contrast, the epidermis of the positive control group was obviously much thicker, and the repair condition was more complete with no necrotic tissues. In addition, the wound epidermis of the chloroform group had the best repair condition with the thickest epidermis layer and most abundant collagen regeneration. However, collagen tissues of the n-butanol group were incomplete with little collagen regeneration and epidermis layer (Fig. 4A). As a result, there was a significant difference in wound healing width and epidermal thickness between the four groups ($p \leq 0.05$, Fig. 4B and 4C).Fig. 4A Effect of HE staining of wound granulation tissue at day 9. B and C The epidermis thickness and healing width at day 9. Negative control:carbomer; Positive control:rb-bFGF; CHCl3: Chloroform-extract; n-butanol: n-butanol-extract. * $p \leq 0.05$ vs negative control ## PCNA staining of active polar parts of SHF On Day 9, proliferating cell nuclear antigen (PCNA) staining results showed that cell layers in the negative control group were blurred and irregular with few PCNA-positive cells. In the positive control group, dermal cells repaired well with regularly arranged positive cells. In the chloroform group, the best condition for epidermal repair was observed, and the epidermal layer was obviously much thicker. Positive cells were regularly arranged, and collagen regeneration was obvious. In the n-butanol group, collagen tissues exhibited a poor repair condition of few positive cells, and the arrangement was comparatively regular (Fig. 5A). As a result, the chloroform group and positive drug group contained significantly more PCNA-positive cells than that of the negative control group ($p \leq 0.05$, Fig. 5B).Fig. 5A Effect of PCNA staining of wound tissue at day 9. Red dotted lines describe cell proliferation and positive cells marked in black arrows in the pictures. B PCNA positive expression of wound tissue at day 9. Negative control:carbomer; Positive control:rb-bFGF; CHCl3: Chloroform-extract; n-butanol: n-butanol-extract. * $p \leq 0.05$ vs negative control ## VEGF staining of active polar parts of SHF On Day 9, the staining results showed that the vascular endothelial growth factor (VEGF)-positive cells became brown. The negative control group showed little positive expression of VEGF with irregular stratification and few vascular endothelial cells. In the positive control group, positive expression of VEGF was observed with abundant vascular endothelial cells. In the chloroform group, VEGF-positive cells were accompanied by regularly arranged vascular endothelial cells, complete repair of endothelial cells and much collagen regeneration. In the n-butanol group, the collagen tissue was comparatively repaired, and compared to the chloroform group, VEGF expression and positive cell arrangement were poorer (Fig. 6A). As a result, the number of VEGF-positive cells indicated the number of vascular endothelial cells within different groups. There were significantly more vascular endothelial cells in the chloroform group and the positive drug group than in the negative control group ($p \leq 0.05$) (Fig. 6B).Fig. 6A Effect of VEGF immunohistochemistry at day 9. Positive cells marked in black arrows in the pictures. B VEGF positive expression of wound tissues at day 9. Negative control:carbomer; Positive control:rb-bFGF; CHCl3: Chloroform-extract; n-butanol: n-butanol-extract. * $p \leq 0.05$ vs negative control ## HPLC analysis of the chloroform extract of SHF Dracorhodin is the main compound in Daemonorops draco Bl., and the main compounds in Rheum palmatum L. are aloe-emodin, rhein, emodin, chrysophanol, and physcion. Imperatorin and isoimperatorin are the main compounds in Angelica dahurica (Fisch ex Hoffm.) Benth. et Hook. f., and tanshinone IIA is the main compound in Salvia miltiorrhiza Bge. The nine compounds are also used by the Chinese Pharmacopoeia to assess the quality of Daemonorops draco Bl., Rheum palmatum L., Angelica dahurica (Fisch ex Hoffm.) Benth. et Hook. f., and Salvia miltiorrhiza Bge., respectively (Chinese Pharmacopoeia Commission 2015). Under the conditions of the “equipment and chromatographic conditions” experiment, the chromatogram was obtained by injecting the reference solution (Fig. 7). System suitability was determined by injecting a sample of the chloroform extract of SHF, including theoretical plates, and the resolution and tailing factor were calculated (Table 2). The calibration curves of 9 analytes were fitted with coefficients of determination greater than 0.999. The linear ranges were set as 12.90 to 258.00 μg/mL for dracorhodin perchlorate, 9.29 to 185.71 µg/mL for aloe emodin, 11.30 to 226.00 µg/mL for rhein, 6.55 to 131.00 µg/mL for imperatorin, 15.32 to 306.43 µg/mL for emodin, 4.42 to 221.00 µg/mL for isoimperatorin, 20.70 to 414.00 µg/mL for chrysophanol, 7.96 to 159.29 µg/mL for physcion, and 8.14 to 162.86 µg/mL for tanshinone IIA, according to the approximate concentrations of the sample. The relative standard deviations (RSD) of the precision, stability and repeatability tests were all less than $5\%$. The accuracy of the system was observed by recovery. The average of 9 analytes of the chloroform extract of SHF recoveries ($$n = 6$$) were dracorhodin perchlorate: $103.62\%$ (RSD = 0.62), emodin: $101.87\%$ (RSD = 2.42), rhein: $102.09\%$ (RSD = 1.64), imperatorin: $102.55\%$ (RSD = 1.72), emodin: $103.01\%$ (RSD = 1.36), isoimperatorin: $102.80\%$ (RSD = 2.11), chrysophanol: $102.06\%$ (RSD = 2.16), physcion: $95.84\%$ (RSD = 0.62), and tanshinone IIA: $99.33\%$ (RSD = 2.67). The limits of detection (LODs) and limits of quantification (LOQs) were determined by using signal-to-noise ratios of 3:1 and 10:1. LOD and LOQ results of 9 analytes (Table 2). Three samples of the same batch of original medicinal materials were prepared by the “Preparation of standard and sample solutions of chloroform part for HPLC” method, and contents of 9 analytes (Table 2) were calculated by the regression equation. Fig. 7HPLC chromatogram of chloroform part of SHF. A The standard control. B The chloroform extract of SHF. 1. Dracorhodin perchlorate 2. Aloe-emodin 3. Rhein 4. Imperatorin 5. Emodin 6. Isoimperatorin 7. Chrysophanol 8. Physcion 9. Tanshinone IIATable 2System suitability, LOD, LOQ and the concentration for 9 analytesAnalytesTheoretical plateResolutionTailing factorLOD (µg/mL)LOQ (µg/mL)sample 1 (mg/g)sample 2 (mg/g)sample 3 (mg/g)Mean ($$n = 3$$)RSD (%)Dracorhodin perchlorate92331.741.4160.371.293.423.543.413.462.18Aloe-emodin44,1451.61.0330.250.930.810.850.820.822.85Rhein50,1033.351.0820.271.131.041.071.031.051.74Imperatorin103,35221.0280.441.640.890.90.860.882.17Emodin98,8851.481.0450.691.391.952.031.961.982.12Isoimperatorin187,2722.191.0330.130.440.580.580.580.580.41Chrysophanol271,2692.21.0410.220.734.454.654.474.522.48Physcion270,8131.491.0690.361.321.241.291.251.262.27Tanshinone IIA225,4964.051.0460.140.450.570.60.570.582.87 ## Discussion The pathogenesis of diabetic ulcers remains unclear, and existing therapies include anti-infection, hyperbaric oxygen therapy, and surgery [15]. Wound healing is a complex process that includes fibroblast proliferation, angiogenesis and granulation tissue formation. Reducing the secretion of inflammatory factors, enriching collagen and promoting epidermal cells play an important role in different stages of diabetic ulcers [16–20]. External treatment with TCM has advantages in curing cutaneous ulcers and has been widely applied in China. Herbs regulate the immune [21] and skin microenvironments [22]. TCM has gradually been used to treat chronic ulcers and has achieved good effects [22–24]. SHF has been used to treat diabetic ulcers for over 30 years. Clinical studies have indicated that SHF can significantly improve diabetic ulcer wound healing by removing necrotic tissues and stimulating granulation tissues [25]. In our previous study, it was confirmed that SHF could reduce activin/follistatin protein levels, thus accelerating re-epithelialization during wound healing [11]. In addition, SHF could also reduce local inflammation by downregulating the protein expression of TNF-α, IL-1β, and IL-6 [9]. However, the effect of the exact active polar parts of this formula remains unclear. Few studies have focused on the active polar parts of SHF in treating diabetic wound healing. Due to the complicated pharmaceutical composition of SHF, finding the effective polar parts of SHF that treat diabetic ulcers is difficult. In this research, we further explored and compared different extractions of SHF and thus determined the potential active polar parts of this formula that treat STZ-induced diabetic ulcer rats. The activity of active parts in SHF is closely related to basement membrane reconstruction and epithelial regeneration. At present, skin inflammation, epidermal proliferation, scar formation and tissue remodeling are thoroughly understood [26–28]. Astragaloside IV from Astragalus membranceus (Fisch.) Bge can promote wound repair in diabetic mice. The mechanism is mainly related to enhancing collagen deposition and extracellular matrix (ECM)-related gene expression, promoting angiogenesis and improving the expression of vascular endothelial growth factors [29]. Deoxyshikonin from Angelica dahurica enhances the migration of vascular endothelial cells, stimulates the phosphorylation of p38 and extracellular signal-regulated kinase, and thus accelerates wound healing [30, 31]. The chloroform extract of SHF was confirmed to be effective in treating diabetic ulcers. There were nine main components that synergistically improved wound healing in the chloroform extract. Among these components, dracorhodin and rhein were characteristic. Dracorhodin could promote the proliferation of fibroblasts [32] and keratinocytes [33] during wound healing. In addition, dracorhodin can inhibit the secretion of IL-1α and TNF-α, alleviate inflammation, stimulate the expression of vascular endothelial growth factors and TGF, and promote fibroblast proliferation and collagen deposition [34, 35]. Rhein could reduce inflammation, expediting angiogenesis, and promoting wound healing [36]. Therefore, further research on the main components in SHF might help to explore the most effective extract in the formula during diabetic wound healing. In this study, STZ was injected into rats to establish a diabetic model. High-fat and high-sugar diets were fed to ensure the stability of the animal model. The positive control group used Bei Fuxin (rb-bFGF), which can promote vascular regeneration, improve local blood circulation, and accelerate wound healing and is used to treat burn wounds, chronic wounds and fresh wounds [37]. The measurement of wound area is the main parameter used to evaluate ulcer repair. Based on the results, no significant difference was observed between the treatment group and negative control group in the initial stage of wound treatment. During the inflammation stage, large amounts of inflammatory factors were discovered within the wound, which was not conducive to wound healing. However, from Day 9 of treatment, the effect of the positive group and chloroform group was significantly better than that of the negative control group. During this stage, abundant granulation regeneration could be found within the wound. The wound showed a faster closure rate. It has been reported [38] that the indices are width of wound healing and thickness of epidermis repair, which are used to judge the degree of wound healing. Complete repair of the epidermis, close connection of granulation tissues and vigorous angiogenesis are the primary conditions for wound healing. The results indicated that the wound healing degree of each group showed differences from the 9th day. Histology and HE staining showed that the wound healing degree of the positive control group and chloroform group was better than that of the negative control group, indicating that the positive control group and chloroform could promote wound healing. No significant difference was observed between the two groups. Therefore, chloroform extract could improve wound healing by accelerating granulation regeneration. PCNA is an indication of the degree of cell proliferation and could help epidermal regeneration and wound repair. PCNA staining results showed that PCNA-positive cells were distributed in wound tissues of the negative control group, positive control group, chloroform group and n-butanol group. However, the expression of PCNA-positive cells in the positive drug group and chloroform group was higher than that in the negative control group. VEGF can enhance vascular permeability, regulate endothelial cell growth and promote cell migration [39, 40]. VEGF staining results showed that the VEGF-positive cells were accompanied by regularly arranged vascular endothelial cells in the chloroform group, which was more than that in the negative control group. Therefore, high expression of VEGF promoted granulation tissue formation and new angiogenesis. In cases of hypoxia, hypoxia-inducible factor-1α (HIF-1α) is known to be involved in mediating protein expression [41]. Additionally, keratinocytes are important cell types during wound repair [42]. It was reported that HIF-1α could regulate VEGF expression in human keratinocytes treated with chloroform [43]. Thus, we speculated that the chloroform extract of SHF might promote the generation of VEGF although mediating HIF-1α; however, more in-depth molecular studies are needed for confirmation. Our HPLC results indicated that dracorhodin, aloe-emodin, rhein, imperatorin, emodin, isoimperatorin, chrysophanol, physcion, and tanshinone IIA were the main components of the chloroform extract from SHF. It was reported that dracorhodin could accelerate wound healing by facilitating the expression of VEGF and supporting collagen deposition [44]. Aloe-emodin was found to promote wound healing by regulating exosome release [45]. Rhein improved wound healing by decreasing inflammation and stimulating collagen deposition [46]. Similarly, imperatorin was confirmed to improve wound healing by increasing the secretion of VEGF, EGF and TGF-β1, thereby facilitating re-epithelization [47]. Tanshinone IIA was discovered to activate the PI3K/Akt/eNOS pathway, thus ameliorating wound healing [48]. As the abovementioned monomers were the main components of the chloroform extract from SHF and many of these monomers exhibited a capacity to promote wound healing, we therefore speculated that the chloroform extract might accelerate wound repair. ## Conclusion In this study, the effective active parts of SHF were concentrated in chloroform by exploring the pharmacodynamic material basis of TCM. The nine main chemical components in the chloroform extract were determined by HPLC. By investigating chromatographic conditions, such as column, mobile phase, and detection wavelength, a method of evaluating the main chemical constituents in chloroform extract of SHF was established to provide a scientific basis to comprehensively and accurately evaluate the quality of medicinal materials. It was discovered that the chloroform extract of SHF could stimulate granulation tissues and improve the generation of PCNA and VEGF compared with petroleum ether, ethylacetate and n-butanol extracts of SHF. The chloroform extract was clarified to be the effective polar part. However, the molecular mechanisms of SHF in treating diabetic wounds remain unclear. The monomer compound of SHF ethanol extract-derived fractions on wound healing and potential specific markers for the recovery and outcome of diabetic wounds require more convincing evidence and research. ## References 1. Zhang P, Lu J, Jing Y, Tang S, Zhu D, Bi Y. Global epidemiology of diabetic foot ulceration: a systematic review and meta-analysis. Ann Med (2017.0) 49 106-116. DOI: 10.1080/07853890.2016.1231932 2. Cho NH, Shaw JE, Karuranga S, Huang Y, da Rocha Fernandes JD, Ohlrogge AW, Malanda B. IDF diabetes atlas: global estimates of diabetes prevalence for 2017 and projections for 2045. Diabetes Res Clin Pract (2018.0) 138 271-281. DOI: 10.1016/j.diabres.2018.02.023 3. Jiang Y, Wang X, Xia L, Fu X, Xu Z, Ran X, Yan L, Li Q, Mo Z, Yan Z. A cohort study of diabetic patients and diabetic foot ulceration patients in China. Wound Repair Regen (2015.0) 23 222-230. DOI: 10.1111/wrr.12263 4. Shan GQ, Zhang YN, Ma J, Li YH, Zuo DM, Qiu JL, Cheng B, Chen ZL. Evaluation of the effects of homologous platelet gel on healing lower extremity wounds in patients with diabetes. Int J Low Extrem Wounds (2013.0) 12 22-29. DOI: 10.1177/1534734613477113 5. Couture M. A single-center, retrospective study of cryopreserved umbilical cord for wound healing in patients suffering from chronic wounds of the foot and ankle. Wounds (2016.0) 28 217-225. PMID: 27428716 6. Zhang J, Zhou R, Xiang C, Jia Q, Wu H, Yang H. Huangbai liniment accelerated wound healing by activating Nrf2 signaling in diabetes. Oxid Med Cell Longev (2020.0) 2020 4951820. PMID: 32566084 7. Wu M, Huang J, Shi J, Shi L, Zeng Q, Wang H. Ruyi Jinhuang powder accelerated diabetic ulcer wound healing by regulating Wnt/beta-catenin signaling pathway of fibroblasts in vivo and in vitro. J Ethnopharmacol (2022.0) 293 115321. DOI: 10.1016/j.jep.2022.115321 8. Jiang JS, Zhang Y, Luo Y, Ru Y, Luo Y, Fei XY, Song JK, Ding XJ, Zhang Z, Yang D. The identification of the biomarkers of Sheng-Ji Hua-Yu formula treated diabetic wound healing using modular pharmacology. Front Pharmacol (2021.0) 12 726158. DOI: 10.3389/fphar.2021.726158 9. Xiang Y, Kuai L, Ru Y, Jiang J, Li X, Li F, Chen Q, Li B. Transcriptional profiling and circRNA-miRNA-mRNA network analysis identify the biomarkers in Sheng-ji Hua-yu formula treated diabetic wound healing. J Ethnopharmacol (2021.0) 268 113643. DOI: 10.1016/j.jep.2020.113643 10. Ru Y, Zhang Y, Xiang YW, Luo Y, Luo Y, Jiang JS, Song JK, Fei XY, Yang D, Zhang Z. Gene set enrichment analysis and ingenuity pathway analysis to identify biomarkers in Sheng-ji Hua-yu formula treated diabetic ulcers. J Ethnopharmacol (2022.0) 285 114845. DOI: 10.1016/j.jep.2021.114845 11. Kuai L, Zhang JT, Deng Y, Xu S, Xu XZ, Wu MF, Guo DJ, Chen Y, Wu RJ, Zhao XQ. Sheng-ji Hua-yu formula promotes diabetic wound healing of re-epithelization via Activin/Follistatin regulation. BMC Complement Altern Med (2018.0) 18 32. DOI: 10.1186/s12906-017-2074-8 12. Lu Y, Ding X, Qi F, Ru Y, Kuai L, Chen S, Yang Y, Li X, Li F, Li B. Quyu Shengji formula facilitates diabetic wound healing via inhibiting the expression of prostaglandin transporter. Evid Based Complement Alternat Med (2021.0) 2021 8849935. DOI: 10.1155/2021/8849935 13. Zhang J, Minghua MA, Minfeng WU, Wang X, Fulun LI, Hua N. Therapeutic effect of 95% ethanol extract of Shengji Huayu formula on diabetic wound. Chin J Modern Appl Pharm (2017.0) 34 1369-1372 14. Ming Z, Lv XY, Jing L, Xu ZG, Chen L. The characterization of high-fat diet and multiple low-dose streptozotocin induced type 2 diabetes rat model. Exp Diabetes Res (2008.0) 2008 704045. PMID: 19132099 15. Isei T, Abe M, Nakanishi T, Matsuo K, Yamasaki O, Asano Y, Ishii T, Ito T, Inoue Y, Imafuku S. The wound/burn guidelines - 3: guidelines for the diagnosis and treatment for diabetic ulcer/gangrene. J Dermatol (2016.0) 43 591-619. DOI: 10.1111/1346-8138.13285 16. Piotto C, Julier Z, Martino MM. Immune regulation of tissue repair and regeneration via miRNAs-New therapeutic target. Front Bioeng Biotechnol (2018.0) 6 98. DOI: 10.3389/fbioe.2018.00098 17. Wichaiyo S, Lax S, Montague SJ, Li Z, Grygielska B, Pike JA, Haining EJ, Brill A, Watson SP, Rayes J. Platelet glycoprotein VI and C-type lectin-like receptor 2 deficiency accelerates wound healing by impairing vascular integrity in mice. Haematologica (2019.0) 104 1648-1660. DOI: 10.3324/haematol.2018.208363 18. Du Y, Ren P, Wang Q, Jiang SK, Zhang M, Li JY, Wang LL, Guan DW. Cannabinoid 2 receptor attenuates inflammation during skin wound healing by inhibiting M1 macrophages rather than activating M2 macrophages. J Inflamm (Lond) (2018.0) 15 25. DOI: 10.1186/s12950-018-0201-z 19. Zhang J, Wu M, Ma M, Zhu Q, Li G, Nian H. Study advances in traditional chinese medicines used in diabetic refractory wound. Chin Pharm (2015.0) 18 2138-2140 20. Kreimendahl F, Marquardt Y, Apel C, Bartneck M, Zwadlo-Klarwasser G, Hepp J, Jockenhoevel S, Baron JM. Macrophages significantly enhance wound healing in a vascularized skin model. J Biomed Mater Res A (2019.0) 107 1340-1350. DOI: 10.1002/jbm.a.36648 21. Wu M, Deng Y, Li S, Chen Y, Guo D, Jin X, Xu Q, Li B, Li F. The immunoregulatory effects of traditional Chinese medicine on psoriasis via its action on interleukin: advances and considerations. Am J Chin Med (2018.0) 46 739-750. DOI: 10.1142/S0192415X18500386 22. Wu M, Li Y, Guo D, Kui G, Li B, Deng Y, Li F. Microbial diversity of chronic wound and successful management of traditional Chinese medicine. Evid Based Complement Alternat Med (2018.0) 2018 9463295. DOI: 10.1155/2018/9463295 23. Wang G, Wang Y, Wang F, Ding Y, Kang Y, Dong Y, Lv P, Xu X. Lower limb nonhealing ulcers associated with chronic diabetes: a case series. Wounds (2019.0) 31 7-14. PMID: 30372417 24. Li FL, Deng H, Wang HW, Xu R, Chen J, Wang YF, Li X, Fan B, Li B. Effects of external application of Chinese medicine on diabetic ulcers and the expressions of beta-catenin, c-myc and K6. Chin J Integr Med (2011.0) 17 261-266. DOI: 10.1007/s11655-011-0703-5 25. Li FL, Wang YF, Li X, Li F, Xu R, Chen J, Geng L, Li B. Characteristics and clinical managements of chronic skin ulcers based on traditional chinese medicine. Evid Based Complement Alternat Med (2012.0) 2012 930192. DOI: 10.1155/2012/930192 26. Kulkarni M, O'Loughlin A, Vazquez R, Mashayekhi K, Rooney P, Greiser U, O'Toole E, O'Brien T, Malagon MM, Pandit A. Use of a fibrin-based system for enhancing angiogenesis and modulating inflammation in the treatment of hyperglycemic wounds. Biomaterials (2014.0) 35 2001-2010. DOI: 10.1016/j.biomaterials.2013.11.003 27. Hinz B. The role of myofibroblasts in wound healing. Curr Res Transl Med (2016.0) 64 171-177. DOI: 10.1016/j.retram.2016.09.003 28. Zhou Q, Jiang S, Ma K, Li C. Expression of a novel recombinant fusion protein BVN-Tbeta4 and its effects on diabetic wound healing. J Biosci Bioeng (2014.0) 118 341-343. DOI: 10.1016/j.jbiosc.2014.02.022 29. Luo X, Huang P, Yuan B, Liu T, Lan F, Lu X, Dai L, Liu Y, Yin H. Astragaloside IV enhances diabetic wound healing involving upregulation of alternatively activated macrophages. Int Immunopharmacol (2016.0) 35 22-28. DOI: 10.1016/j.intimp.2016.03.020 30. Park JY, Kwak JH, Kang KS, Jung EB, Lee DS, Lee S, Jung Y, Kim KH, Hwang GS, Lee HL. Wound healing effects of deoxyshikonin isolated from Jawoongo: in vitro and in vivo studies. J Ethnopharmacol (2017.0) 199 128-137. DOI: 10.1016/j.jep.2016.10.031 31. Park JY, Shin MS, Hwang GS, Yamabe N, Yoo JE, Kang KS, Kim JC, Lee JG, Ham J, Lee HL. Beneficial effects of deoxyshikonin on delayed wound healing in diabetic mice. Int J Mol Sci (2018.0) 19 3660. DOI: 10.3390/ijms19113660 32. Xiong Z, Huo M, Jia Y, Zhou C, Ma X, Yin H, Jiang X, Yu W. Dracorhodin perchlorate regulates the expression of inflammatory cytokines through the TLR4 pathway and improves skin wound healing in diabetic rats. Evid Based Complement Alternat Med (2022.0) 2022 9050686. DOI: 10.1155/2022/9050686 33. Lu CC, Yang JS, Chiu YJ, Tsai FJ, Hsu YM, Yin MC, Juan YN, Ho TJ, Chen HP. Dracorhodin perchlorate enhances wound healing via beta-catenin, ERK/p38, and AKT signaling in human HaCaT keratinocytes. Exp Ther Med (2021.0) 22 822. DOI: 10.3892/etm.2021.10254 34. Jiang XW, Qiao L, Liu L, Zhang BQ, Wang XW, Han YW, Yu WH. Dracorhodin perchlorate accelerates cutaneous wound healing in wistar rats. Evid Based Complement Alternat Med (2017.0) 2017 8950516. DOI: 10.1155/2017/8950516 35. Jiang X, Liu L, Qiao L, Zhang B, Wang X, Han Y, Yu W. Dracorhodin perchlorate regulates fibroblast proliferation to promote rat's wound healing. J Pharmacol Sci (2018.0) 136 66-72. DOI: 10.1016/j.jphs.2017.12.003 36. Yin C, Han X, Lu Q, Qi X, Guo C, Wu X. Rhein incorporated silk fibroin hydrogels with antibacterial and anti-inflammatory efficacy to promote healing of bacteria-infected burn wounds. Int J Biol Macromol (2022.0) 201 14-19. DOI: 10.1016/j.ijbiomac.2021.12.156 37. Li L, Lin Y, Bin G, Shu Q, Chen D, Shan Y, Wang S, Huang B, Wang J, Tang Q. MEBT/MEBO promoted chronic non-healing cutaneous wound by modulating the P13K-Akt-mTOR signaling pathway. Chin J Integr Tradit Western Med (2019.0) 39 583-590 38. Takagi N, Kawakami K, Kanno E, Tanno H, Takeda A, Ishii K, Imai Y, Iwakura Y, Tachi M. IL-17A promotes neutrophilic inflammation and disturbs acute wound healing in skin. Exp Dermatol (2017.0) 26 137-144. DOI: 10.1111/exd.13115 39. Seo E, Lim JS, Jun JB, Choi W, Hong IS, Jun HS. Exendin-4 in combination with adipose-derived stem cells promotes angiogenesis and improves diabetic wound healing. J Transl Med (2017.0) 15 35. DOI: 10.1186/s12967-017-1145-4 40. Zhou J, Ni M, Liu X, Ren Z, Zheng Z. Curcumol promotes Vascular Endothelial Growth Factor (VEGF)-mediated diabetic wound healing in streptozotocin-induced hyperglycemic rats. Med Sci Monit (2017.0) 23 555-562. DOI: 10.12659/MSM.902859 41. Guodong L, Chung-Nga K, Dan L, Chao Y, Wanhe W, Guan-Jun Y, Carmelo D, Vincent K, Yaozu X, Ligen Lin, Dik-Lung M, Chung-Hang L. A small molecule HIF-1α stabilizer that accelerates diabetic wound healing. Nat Commun (2021.0) 12 3363. DOI: 10.1038/s41467-021-23448-7 42. Ratklao S, Anh Q, Makio H, Kentaro A, Satoko S, Akira W, Shuh N, Dean T. Single-cell RNA sequencing identifies a migratory keratinocyte subpopulation expressing THBS1 in epidermal wound healing. iScience (2022.0) 25 104130. DOI: 10.1016/j.isci.2022.104130 43. Nam J, Lee H, Bae HC, Kim J, Jeong S, Ryu W, Son S. Chloroform induces HIF-1α-dependent VEGF expression in human keratinocytes. Mol Cell Toxicol (2013.0) 9 335-340. DOI: 10.1007/s13273-013-0042-z 44. Jiang X, Qiao L, Liu L, Zhang B, Wang X, Han Y, Yu W. Dracorhodin perchlorate accelerates cutaneous wound healing in wistar rats. Evid-based Compl Alt (2017.0) 2017 8950516. DOI: 10.1155/2017/8950516 45. Koken G, Abamor E, Allahverdiyev A, Karaoz E. Wharton jelly derived mesenchymal stem cell's exosomes demonstrate significant antileishmanial and wound healing effects in combination with aloe-emodin: an in vitro study. J Pharm Sci (2022.0) 111 3232-3242. DOI: 10.1016/j.xphs.2022.08.016 46. Li J, Wang C, Han X, Liu S, Gao X, Guo C, Wu X. Aramid nanofibers-reinforced rhein fibrous hydrogels as antibacterial and anti-inflammatory burn wound dressings. ACS Appl Mater Interfaces (2022.0) 14 45167-45177. DOI: 10.1021/acsami.2c12869 47. 47.Gao T, Zhao Y, Zhao Y, He Y, Huang Q, Yang J, Zhang L, Chen J. Curative effect and mechanisms of radix arnebiae oil on burn wound healing in rats. Planta Med. 2022. 10.1055/a-1997-5566. 48. Qu K, Cha H, Ru Y, Que H, Xing M. Buxuhuayu decoction accelerates angiogenesis by activating the PI3K-Akt-eNOS signalling pathway in a streptozotocin-induced diabetic ulcer rat model. J Ethnopharmacol (2021.0) 273 113824. DOI: 10.1016/j.jep.2021.113824
--- title: Influence of discontinuation of prophylactic antimicrobial agent for trabeculectomy authors: - Yuuka Ushio - Hiroshi Yoshikawa - Tetsuya Murase - Tatsuo Kataoka - Shohei Miyamoto - Kazunari Maruko - Shoko Okamoto - Yuuka Shibata - Ryotaro Toda - Yoshiaki Kiuchi - Hiroaki Matsuo journal: Journal of Pharmaceutical Health Care and Sciences year: 2023 pmcid: PMC9976534 doi: 10.1186/s40780-023-00276-z license: CC BY 4.0 --- # Influence of discontinuation of prophylactic antimicrobial agent for trabeculectomy ## Abstract ### Background There is no unified view of the necessity of prophylactic antimicrobial agents in trabeculectomy. Preoperative prophylactic antimicrobial agent injection and cefazolin sodium (CEZ) for trabeculectomy were discontinued at the Hiroshima University Hospital. In this study, we evaluated whether discontinuation of preoperative administration of CEZ in ophthalmology affects the incidence of postoperative infections. ### Methods We retrospectively investigated patient background, concomitant medications, subconjunctival dexamethasone sodium phosphate (DEX) injection at the end of the surgery, and the incidence of infective endophthalmitis within 6 weeks after surgery in the CEZ and non-CEZ groups. We also performed propensity score matching for background matching. Statistical analysis was performed using the Mann-Whitney U-test and Fisher’s exact test. ### Results The incidence of postoperative endophthalmitis was not significantly different between 629 and 751 patients in the CEZ and no-CEZ groups, respectively (0 in the CEZ group and 2 in the no-CEZ group, $$P \leq 0.504$$). More patients in the CEZ group were taking diabetes drugs preoperatively ($$P \leq 0.028$$) and fewer patients were receiving subconjunctival DEX at the end of surgery ($P \leq 0.001$) than those in the non-CEZ group. Propensity scores were calculated using the risk factors for postoperative infection as covariates, and matching (580 patients in the CEZ group and 580 patients in the non-CEZ group) showed no significant difference in the incidence of postoperative endophthalmitis ($$P \leq 0.500$$). ### Conclusions There was no significant difference in the incidence of endophthalmitis after trabeculectomy between the CEZ and non-CEZ groups, suggesting a decreased need for CEZ injections before trabeculectomy. ## Background Endophthalmitis is a common surgical site infection after trabeculectomy, with acute cases occurring within 6 weeks after surgery. Although the incidence is low, it is a serious complication that can lead to severe vision loss and blindness [1]. Since postoperative endophthalmitis is caused by bacteria indigenous to the conjunctiva and eyelid, antimicrobial agents and povidone-iodine disinfection are generally used to prevent endophthalmitis [2]. In the United States, preoperative administration of antimicrobial eye drops and subconjunctival administration of antimicrobial agents at the end of surgery is recommended to prevent postoperative endophthalmitis [2]. However, most of the literature on which these guidelines are based refers to cataract surgery, and there is no unified view of prophylactic antimicrobial agents in trabeculectomy. This is because the number of glaucoma surgeries is much lower than the number of cataract surgeries. Similarly, Japanese prophylactic antimicrobial guidelines recommend the preoperative use of antimicrobial eye drops for cataract surgery, while intravenous cefazolin sodium (CEZ) is recommended for trabeculectomy with a low level of evidence [3]. Since there have been few studies on prophylactic antimicrobial agents in trabeculectomy [4], no clear evidence of its usefulness has been obtained. Hiroshima University Hospital performs approximately 600 trabeculectomies per year, ranked second in Japan in terms of the number of surgeries in the Diagnosis Procedure Combination database in the fiscal year 2018 [5]. In addition, CEZ was administered prophylactically before trabeculectomy as a clinical pathway. However, there is no clear evidence that CEZ injections prevent postoperative endophthalmitis. There is also the risk of administering unnecessary antimicrobials, which may cause minor side effects, such as diarrhea, and serious side effects, such as anaphylactic shock with dyspnea. In addition, the “Action Plan for Antimicrobial Resistance Control [6]“has been formulated as a national policy, and it is important to discontinue unnecessary antimicrobial agents based on the viewpoint of controlling drug-resistant bacteria. Therefore, in April 2018, our hospital reviewed the clinical pathway for trabeculectomy and discontinued CEZ injection. Perioperative infection prophylaxis consisted only of povidone-iodine disinfection and antimicrobial eye drops. In this study, we compared the incidence of postoperative endophthalmitis before and after clinical pathway change to determine the impact of discontinuation of prophylactic CEZ injection for trabeculectomy on the incidence of surgical site infection. ## Patients A summary of our trabeculectomy clinical pathway and antimicrobials used is presented in Table 1. We removed the CEZ injection from our clinical pathway starting on April 12, 2018. Patients who underwent trabeculectomy at our hospital on admission were included in this study. The patients were classified into two groups: the CEZ group before the clinical pathway change (February 2, 2016, to April 11, 2018) and the non-CEZ group after the clinical pathway change (April 12, 2018, to December 31, 2020). We excluded patients who were taking oral antimicrobials at admission, who had undergone concomitant procedures other than trabeculectomy, and who were difficult to follow up for 6 weeks postoperatively after trabeculectomy. Table 1Summary of the trabeculectomy clinical pathway and antimicrobials usedThree days before surgeryLevofloxacin $1.5\%$ eye drops Three times a dayBefore surgeryCEZ injection 1 gDisinfection with iodine and polyvinyl alcohol 6x diluted solution($0.033\%$ effective iodine)During surgeryApplication of mitomycin C to the surgical wound a)(Subconjunctival administration of DEX injection a))After surgeryOfloxacin $0.3\%$ eye ointmentDay after surgery~ 1–3 monthsLevofloxacin $1.5\%$ eye dropsFluorometholone $0.1\%$ eye dropsNepafenac $0.1\%$ suspension eye drops Three times a daya) Not covered by insurance ## Patient background and primary end point Patient characteristics included age, sex, obesity (body mass index [BMI] ≥25 kg/m2), and concomitant medications. To adjust for factors that might affect the incidence of postoperative endophthalmitis, a propensity score was calculated using patient background parameters as covariates for matching (caliper coefficient:0.2). As it has been reported that age > 85 years is a risk factor for postoperative endophthalmitis in cataract surgery [7], we checked the number of patients in each age group. Concomitant medications known to affect surgical site infection, such as diabetes drugs, immunosuppressants, and corticosteroids (eye drops, eye ointment, oral administration, and subconjunctival dexamethasone sodium phosphate (DEX) injection at the end of surgery), were collected from medical records. Subconjunctival DEX was administered at the discretion of the primary surgeon for anti-inflammatory purposes. Immunosuppressants and corticosteroids were defined based on the efficacy classification of the Japanese Standard Commodity Classification [8]. Patients with postoperative endophthalmitis were defined as those diagnosed with postoperative endophthalmitis by an ophthalmologist within 6 weeks after surgery. ## Statistical analysis The Mann-Whitney U-test was used for age, and the chi-square test was used for sex, obesity, the presence of subconjunctival DEX at the end of the surgery, and concomitant medications other than immunosuppressants. The Fisher’s exact test was used to assess immunosuppressant usage and presence of postoperative endophthalmitis. Using propensity score matching methods, we calculated the propensity scores for CEZ injection using a logistic regression model that included the following variables: age, patients aged 85 years or older, obesity, concomitant medications, and subconjunctival DEX injection. One-to-one nearest-neighbor matching without replacement was performed for the estimated propensity scores in the patients using a caliper width set at $20\%$ of the pooled standard deviation of the logit of the propensity score. A significance level of less than $5\%$ was considered statistically significant. EZR Ver. 1.55 [9] was used for the statistical analysis. ## Patient background Of the 889 patients in the CEZ group, 15 taking oral antimicrobial agents on admission and 245 who had undergone procedures other than trabeculectomy were excluded, and 629 patients were included in the analysis. Of the 1063 patients in the non-CEZ group, 15 who were taking oral antimicrobial agents on admission and 297 who had undergone procedures other than trabeculectomy were excluded, and 751 patients were included in the analysis. In both groups, there were no patients with difficulty following up for 6 weeks postoperatively. The patient backgrounds of the two groups are shown in Table 2. There were no significant differences in sex, median age, or obesity (BMI ≥25) between the two groups. There were statistically significant differences between the two groups in terms of subconjunctival DEX at the end of surgery and diabetes drugs. Subconjunctival DEX was administered more frequently at the end of surgery in the non-CEZ group (161 patients [$21.4\%$]) than in the CEZ group (11 patients [$1.7\%$]). The use of diabetes drugs was less common in the non-CEZ group (134 patients ($17.8\%$)) than in the CEZ group (143 patients ($22.7\%$)). There were no significant differences in other concomitant medications between the two groups. Table 2Patient backgroundCEZ ($$n = 629$$)Non-CEZ ($$n = 751$$)P-valueSex (Male / Female)340 / 289426 / 3250.347a)Age (years)Median72700.106b)Quartile range64–7862–78Patients 85 years and older65($10.3\%$)66($8.8\%$)0.377a)Obesity (BMI ≥ 25)194($30.8\%$)213($28.4\%$)0.344a)Concomitant medications Diabetes drugs143($22.7\%$)134($17.8\%$)0.028a) Immunosuppressant3($0.5\%$)9($1.2\%$)0.244c) Oral corticosteroids31($4.9\%$)24($3.2\%$)0.133a) Corticosteroid eye drops and ointment58($9.2\%$)71($9.5\%$)0.956a)Subconjunctival DEX injection11($1.7\%$)161($21.4\%$) < 0.001a)a) Chi-square testb) Mann-Whitney U testc) Fisher’s exact test, $P \leq 0.05$ Since the rate of subconjunctival DEX injection at the end of surgery and diabetes drugs differed between the two groups and their possible influence on postoperative infection could be ruled out, propensity score matching was performed to reduce bias. After propensity score matching, 580 patients in the CEZ group and 580 patients in the non-CEZ group were included in the study. The patient backgrounds of the two groups are shown in Table 3. There were no significant differences between the two groups in any of the items. Table 3Patient background after matchingCEZ ($$n = 580$$)Non-CEZ ($$n = 580$$)P-valueSex (Male / Female)328 / 252330 / 2500.906a)Age (years)Median72700.319b)Quartile range(63–78)(63–78)Patients 85 years and older58($10.0\%$)53($9.1\%$)0.690a)Obesity (BMI ≥ 25)163($28.1\%$)167($28.8\%$)0.845a)Concomitant medications Diabetes drugs120($20.7\%$)118($20.3\%$)0.942a) Immunosuppressant3($0.5\%$)5($0.9\%$)0.726c) Oral corticosteroids23($4.0\%$)21($3.6\%$)0.878a) Corticosteroid eye drops and ointment53($9.1\%$)55($9.5\%$)0.920a)Subconjunctival DEX injection11($1.9\%$)9($1.6\%$)0.822a)a)Chi-square testb) Mann-Whitney U test, andc) Fisher’s exact test ## Incidence of postoperative Endophthalmitis The number of patients diagnosed with postoperative endophthalmitis was 0 ($0.0\%$) in the CEZ group and 2 ($0.3\%$) in the non-CEZ group, with no significant difference ($$P \leq 0.504$$) (Table 4). The number of patients diagnosed with postoperative endophthalmitis after matching was 0 ($0.0\%$) in the CEZ group and 2 ($0.3\%$) in the non-CEZ group, with no significant difference ($$P \leq 0.500$$) (Table 4).Table 4Incidence of postoperative endophthalmitisAll casesAfter matchingCEZNon-CEZP-valueCEZNon-CEZP-valuePostoperative endophthalmitis0 / 629 ($0.0\%$)2 / 751 ($0.3\%$)0.5040 / 580 ($0.0\%$)2 / 580 ($0.3\%$)0.500Fisher’s exact test The background and symptom course of the patients diagnosed with postoperative endophthalmitis are shown in Table 5. None of the patients was aged ≥85 years. In one patient, Enterococcus faecalis* was detected in the anterior chamber aqueous humor, while in the other patient, no bacteria were detected in either the anterior chamber aqueous humor or vitreous humor. All patients showed improvement in visual acuity after the vitrectomy. Table 5Postoperative endophthalmitis casesCase 1Female in her 80s, Obesity 1st degree, Diabetes drug user[Treatment] Day after surgery: Vitrectomy and intravitreal administration of vancomycin and ceftazidime[Bacteria detection] Anterior chamber aqueous humor: Enterococcus faecalis, Vitreous humor (−)[Visual acuity] Before surgery 50 cm/m.m., Day after surgery sl (+), 3 months after 30 cm/m.m. Case 2Male in his 70s, Obesity 1st degree[Treatment] 16 days after surgery: Vitrectomy and intravitreal administration of vancomycin[Bacteria detection] Anterior chamber aqueous humor (−), Vitreous humor (−)[Visual acuity] Before surgery 0.3 (0.5 × S), 16 days after surgery 0.01 (0.3 × S), 2 months after 0.3 (0.5 × S) ## Discussion This report investigated the necessity of administering injectable antimicrobials for the prophylaxis of postoperative endophthalmitis following trabeculectomy. Although there were differences in the use of subconjunctival DEX and diabetes drugs between the CEZ and non-CEZ groups, there were no significant differences in the number of cases of postoperative endophthalmitis between these groups after adjusting for patient background using the propensity score matching method. This suggests that discontinuation of the CEZ injection prior to trabeculectomy does not increase the incidence of postoperative endophthalmitis. There was no change in the clinical pathway other than administering antimicrobials. In addition, the 10-fold increase in the number of subconjunctival DEX injection in the non-CEZ group remains unclear. The reason for the lack of an increase in the incidence of endophthalmitis after discontinuation of CEZ injection may be that the risk of infection was controlled by povidone-iodine disinfection and antimicrobial eye drops. In addition, the Enterococcus species detected in the patient diagnosed with postoperative endophthalmitis in this study were resistant to cephem antimicrobials, suggesting that the use of preoperative CEZ injection was not feasible for prevention in this case. However, in our hospital, quinolone antibacterial eye drops have been used continuously for 1–3 months postoperatively, but reports of cataract surgery at other hospitals indicate that the administration of quinolone antibacterial eye drops is terminated 3–7 days after surgery [2]. There are no reports investigating the necessity and appropriate duration of eye drop administration, and further studies are required. This study has several limitations. First, although the risk of surgical site infection associated with oral corticosteroids is related to the dosage and duration of administration [10], it is difficult to evaluate the history of corticosteroid administration in a retrospective study. Second, although poor postoperative glycemic control is considered a risk factor for surgical site infection in patients with or without diabetes mellitus [11], perioperative blood glucose monitoring was not performed in many patients, and as an alternative, the use of diabetes drugs was compared according to the patient background. Finally, the incidence of postoperative endophthalmitis before and after discontinuation of CEZ injections should be compared in a non-inferiority study. However, because the incidence of endophthalmitis after trabeculectomy is very low, a large number of cases are required, which is difficult with the present number of cases. Therefore, we compared all the patients before and after discontinuation using the Fisher’s exact test, which can verify the null hypothesis of independence even when the expected value is small. The number of cases and data analyses at a single institution are limited, and future multicenter data collection and analysis are desirable. ## Conclusion These results suggest that preoperative CEZ injection is not necessary after trabeculectomy, and that prevention of postoperative endophthalmitis can be achieved with iodine disinfection and antimicrobial eye drops. Further large-scale studies are needed to examine the effects of corticosteroid administration and blood glucose control. ## References 1. Wallin Ö, Al-ahramy AM, Lundström M, Montan P. Endophthalmitis and severe blebitis following trabeculectomy. Epidemiology and risk factors; a single-Centre retrospective study. Acta Ophthalmol (2014.0) 92 426-431. DOI: 10.1111/aos.12257 2. Bratzler DW, Dellinger EP, Olsen KM, Perl TM, Auwaerter PG, Bolon MK. Clinical practice guidelines for antimicrobial prophylaxis in surgery. Am J Health Syst Pharm (2013.0) 70 195-283. DOI: 10.2146/ajhp120568 3. 3.Japanese Society of Chemotherapy; 2020. https://www.chemotherapy.or.jp/uploads/files/guideline/jyutsugo_shiyou_tuiho.pdf. Accessed 25 Sep 2022. 4. Aaberg TM, Flynn HW, Schiffman J, Newton J. Nosocomial acute-onset postoperative endophthalmitis survey. A 10-year review of incidence and outcomes. Ophthalmology (1998.0) 105 1004-1010. DOI: 10.1016/S0161-6420(98)96000-6 5. 5.The Government of Japan; 2021. https://www.mhlw.go.jp/stf/shingi2/0000196043_00004.html. Accessed 25 Sep 2022. 6. 6.The Government of Japan; 2016. https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/0000120172.html. Accessed 25 Sep 2022. 7. Lundström M, Wejde G, Stenevi U, Thorburn W, Montan P. Endophthalmitis after cataract surgery: a nationwide prospective study evaluating incidence in relation to incision type and location. Ophthalmology (2007.0) 114 866-870. DOI: 10.1016/j.ophtha.2006.11.025 8. 8.The government of JapanMinistry of internal affairs and communications1990. Ministry of internal affairs and communications (1990.0) 9. Kanda Y. Investigation of the freely available easy-to-use software ‘EZR’ for medical statistics. Bone Marrow Transplant (2013.0) 48 452-458. DOI: 10.1038/bmt.2012.244 10. Dixon WG, Abrahamowicz M, Beauchamp ME, Ray DW, Bernatsky S, Suissa S. Immediate and delayed impact of oral glucocorticoid therapy on risk of serious infection in older patients with rheumatoid arthritis: a nested case-control analysis. Ann Rheum Dis (2012.0) 71 1128-1133. DOI: 10.1136/annrheumdis-2011-200702 11. Kwon S, Thompson R, Dellinger P, Yanez D, Farrohki E, Flum D. Importance of perioperative glycemic control in general surgery: a report from the surgical care and outcomes assessment program. Ann Surg (2013.0) 257 8-14. DOI: 10.1097/SLA.0b013e31827b6bbc
--- title: 'Distinct DNA methylation signatures associated with blood lipids as exposures or outcomes among survivors of childhood cancer: a report from the St. Jude lifetime cohort' authors: - Qian Dong - Cheng Chen - Nan Song - Na Qin - Noel-Marie Plonski - Emily R. Finch - Kyla Shelton - John Easton - Heather Mulder - Emily Plyer - Geoffrey Neale - Emily Walker - Qian Li - I-Chan Huang - Jinghui Zhang - Hui Wang - Melissa M. Hudson - Leslie L. Robison - Kirsten K. Ness - Zhaoming Wang journal: Clinical Epigenetics year: 2023 pmcid: PMC9976538 doi: 10.1186/s13148-023-01447-3 license: CC BY 4.0 --- # Distinct DNA methylation signatures associated with blood lipids as exposures or outcomes among survivors of childhood cancer: a report from the St. Jude lifetime cohort ## Abstract ### Background DNA methylation (DNAm) plays an important role in lipid metabolism, however, no epigenome-wide association study (EWAS) of lipid levels has been conducted among childhood cancer survivors. Here, we performed EWAS analysis with longitudinally collected blood lipid data from survivors in the St. Jude lifetime cohort study. ### Methods Among 2052 childhood cancer survivors of European ancestry (EA) and 370 survivors of African ancestry (AA), four types of blood lipids, including high-density lipoprotein (HDL), low-density lipoprotein (LDL), total cholesterol (TC), and triglycerides (TG), were measured during follow-up beyond 5-years from childhood cancer diagnosis. For the exposure EWAS (i.e., lipids measured before blood draw for DNAm), the DNAm level was an outcome variable and each of the blood lipid level was an exposure variable; vice versa for the outcome EWAS (i.e., lipids measured after blood draw for DNAm). ### Results Among EA survivors, we identified 43 lipid-associated CpGs in the HDL ($$n = 7$$), TC ($$n = 3$$), and TG ($$n = 33$$) exposure EWAS, and 106 lipid-associated CpGs in the HDL ($$n = 5$$), LDL ($$n = 3$$), TC ($$n = 4$$), and TG ($$n = 94$$) outcome EWAS. Among AA survivors, we identified 15 lipid-associated CpGs in TG exposure ($$n = 6$$), HDL ($$n = 1$$), LDL ($$n = 1$$), TG ($$n = 5$$) and TC ($$n = 2$$) outcome EWAS with epigenome-wide significance ($P \leq 9$ × 10−8). There were no overlapping lipids-associated CpGs between exposure and outcome EWAS among EA and AA survivors, suggesting that the DNAm changes of different CpGs could be the cause or consequence of blood lipid levels. In the meta-EWAS, 12 additional CpGs reached epigenome-wide significance. Notably, 32 out of 74 lipid-associated CpGs showed substantial heterogeneity (Phet < 0.1 or I2 > $70\%$) between EA and AA survivors, highlighting differences in DNAm markers of blood lipids between populations with diverse genetic ancestry. Ten lipid-associated CpGs were cis-expression quantitative trait methylation with their DNAm levels associated with the expression of corresponding genes, out of which seven were negatively associated. ### Conclusions We identified distinct signatures of DNAm for blood lipids as exposures or outcomes and between EA and AA survivors, revealing additional genes involved in lipid metabolism and potential novel targets for controlling blood lipids in childhood cancer survivors. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13148-023-01447-3. ## Background Mounting evidence suggests that epigenetics, specifically DNA methylation (DNAm), plays an important role in lipid metabolism, and epigenome-wide association studies (EWAS) of blood lipid levels have identified robust 5′-cytosine-phosphate-guanine-3′ (CpG) sites and plausible underlying genes associated with lipid metabolism and related diseases [1]. However, an EWAS analysis of lipid levels has not been conducted among survivors of childhood cancer who experience early onset and a substantially higher burden of chronic health conditions (CHCs), compared to community controls without a history of childhood cancer [2, 3]. These health disparities are mostly attributable to genotoxic cancer treatment exposures at a young age with the most notable link being between cardiovascular diseases and exposures to anthracyclines and/or chest-directed radiation therapy (RT) [4]. Recognizing the high burden of CHCs among childhood cancer survivors [2, 3, 5], we have comprehensively analyzed DNAm variations among long-term survivors and conducted systematic investigations of potential casual pathways for treatment-associated CHCs [6]. Our previous findings provide compelling evidence of mediation effect of DNAm between abdominal-RT and dyslipidemia (triglycerides > 150 mg/dL or total cholesterol > 200 mg/dL) [6]. Dyslipidemia is highly prevalent within the broad spectrum of morbidities of childhood and adolescent cancer survivors [7], and a major risk factor for cardiac events, which are the leading cause of noncancer-related premature mortality and account for approximately $26\%$ of deaths among survivors within 45 years of diagnosis [8]. In the general population, African American adults have higher prevalence of high low-density lipoprotein (LDL) and low high-density lipoprotein (HDL) levels but lower prevalence of high triglycerides (TG) than European American adults in both men and women [9, 10]. A study considering racial/ethnic differences among childhood cancer survivors in the St. Jude lifetime cohort study (SJLIFE) reported that childhood cancer survivors of African ancestry (AA) had higher risk of cardiovascular diseases overall including specific conditions such as stroke, heart attack, and heart failure than survivors of European ancestry (EA), potentially explained by the higher prevalence of obesity, diabetes, hypertension, and dyslipidemia among AA survivors [11]. Studies have reported notable population-specific DNAm differences in multiple physical functions (e.g., immunity and kidney development) [12, 13], suggesting that EWAS across populations is critical to the interpretation of health disparities [14]. However, there is a lack of diversity in currently available EWAS data, with most studies conducted in individuals of EA. To further our understanding of the underlying biological mechanisms of different blood lipid levels among childhood cancer survivors and the differences between EA and AA populations as determined by their genetic ancestry, we employed a comprehensive and agnostic EWAS approach across these two populations. Taking advantage of longitudinal clinical assessments of SJLIFE survivors, we analyzed association of DNAm with blood lipids as exposures (i.e., blood lipids were measured before DNAm) and outcomes (i.e., blood lipids were measured after DNAm). Findings were compared between these two scenarios as well as between the two ancestral groups (i.e., EA and AA). The potential function of significant CpG sites were further demonstrated by their correlations with gene expression levels measured by RNA sequencing. We compared our findings among childhood cancer survivors with the known blood lipid-associated CpGs previously reported in non-cancer general populations. Clinically, the set of lipid-associated CpG sites (i.e., signatures) would facilitate the identification of survivors who have already experienced abnormal lipid levels or at higher risk of abnormal lipid levels in the future. ## Characteristics of the study population and EWAS analysis design Adult survivors of childhood cancer from the SJLIFE study [15, 16] were included in this analysis (Table 1). Among 2052 EA survivors (median age at blood draw for DNAm = 32.3 years, interquartile range [IQR] = 26.5–40.1 years; $47.2\%$ female), body mass index (BMI) was 9.9–67.7 kg/m2 (Table 1). Among 370 AA survivors (median age at blood draw for DNAm = 29.6 years, IQR = 23.8–37.0 years; $53.2\%$ female), BMI was 12.6–58.9 kg/m2 (Table 1). The summary statistics of the weighted average levels of HDL, LDL, TG, and TC (including the number of survivors with multiple lipid measurements) before or after DNA sampling are shown in Table 1. Compared with survivors of EA, those of AA had lower mean of weighted average of TG as an exposure (81.2 vs. 125.3 mg/dL, $P \leq 0.0001$) and outcome (83.4 vs. 130.6, $P \leq 0.0001$), TC as an exposure (171.3 vs 179.2 mg/dL, $$P \leq 0.02$$) and an outcome (163.8 vs. 182.3 mg/dL, $P \leq 0.0001$), and LDL as an outcome (95.0 vs. 106.1 mg/dL, $P \leq 0.0001$). AA survivors also had higher mean of weighted average of HDL as an exposure (57.2 vs. 51.1 mg/dL, $P \leq 0.0001$). The correlations of weighted average levels of lipids before and after DNA sampling were shown in Additional file 1: Table S1. The percentage of survivors taking any lipid control medications before DNAm sampling was $8.04\%$ in EA and $4.86\%$ in AA. The median time and range between the DNAm and pre-lipid profiles are 1.6, 0.0–5.3, years for EA and 1.6, 0.5–5.1, years for AA, and the median time and range between DNAm and post-lipid profiles are 2.2, 0.0–5.5, years for EA, and 2.3, 0.1–16.6, years (Table 1).Table 1Characteristics of the SJLIFE study populationCharacteristicSurvivors of European ancestrySurvivors of African ancestryPan(%)n(%)Total2052(100.0)370(100.0)Sex0.03 Male1084(52.8)173(46.8) Female968(47.2)197(53.2)DiagnosisLeukemia699(34.1)77(20.8)< 0.0001 Acute lymphoblastic leukemia644(31.4)67(18.1) Acute myeloid leukemia53(2.6)9(2.4) Other leukemia2(0.1)1(0.3) Lymphoma448(21.8)69(18.6)0.17 Hodgkin lymphoma288(14.0)45(12.2) Non-Hodgkin lymphoma160(7.8)24(6.5) Sarcoma274(13.4)56(15.1)0.36 Ewing sarcoma74(3.6)2(0.5) Osteosarcoma74(3.6)18(4.9) Rhabdomyosarcoma71(3.5)18(4.9) Soft tissue sarcoma55(2.7)18(4.9) CNS tumors231(11.3)45(12.2)0.61 Astrocytoma or glioma93(4.5)18(4.9) Medulloblastoma or PNET56(2.7)10(2.7) Ependymoma26(1.3)5(1.4) Other CNS tumors56(2.7)12(3.2) Embryonal276(13.5)81(21.9)< 0.0001 Wilms tumor134(6.5)44(11.9) Neuroblastoma107(5.2)14(3.8) Germ cell tumor35(1.7)23(6.2) Other124(6.0)42(11.4)0.0002 Retinoblastoma45(2.2)21(5.7) Hepatoblastoma13(0.6)2(0.5) Melanoma12(0.6)2(0.5) Carcinomas24(1.2)14(3.8) Others30(1.5)3(0.8)Chemotherapy Alkylating agent, classical1194(58.2)202(54.6)0.20 Alkylating agent, heavy metal239(11.7)63(17.0) 0.004 Alkylating agent, nonclassical67(3.3)12(3.2) 0.98 Anthracyclines1190(58.0)180(48.6)0.0008 Antimetabolites1024(49.9)133(35.9)< 0.0001 Asparaginase enzymes631(30.8)76(20.5)< 0.0001 Epipodophyllotoxins709(34.6)108(29.2)0.04 Corticosteroids965(47.0)122(33.0)< 0.0001 Vinca alkaloids1482(72.2)236(63.8)0.0009Radiation therapy, region exposed Brain629(30.7)98(26.5)0.11 Chest577(28.1)102(27.6)0.83 Abdominal412(20.1)84(22.7)0.25 Pelvic352(17.2)80(21.6)0.04Tobacco smoking status0.11 Never smoking956(46.6)183(49.5) Ever smoking349(17.0)51(13.8) Unknown747(36.4)136(36.8)Lipid control medication before DNA sampling0.03 Never used1887(92.0)352(95.1) Ever used165(8.0)18(4.9)Lipid control medication after DNA sampling0.06 Never used1958(95.4)361(97.6) Ever used94(4.6)9(2.4)Chronic health condition (Weighted average, mg/dL)nMean ± SDnMean ± SDPa Triglycerides, exposure734125.3 ± 92.911781.2 ± 46.6< 0.0001 Triglycerides, outcome1126130.6 ± 98.019783.4 ± 47.8< 0.0001 Total cholesterol, exposure734179.2 ± 34.9117171.3 ± 32.30.02 Total cholesterol, outcome1124182.3 ± 39.2197163.8 ± 48.0< 0.0001 High-density lipoprotein, Exposure73451.1 ± 14.711757.2 ± 14.2< 0.0001 High-density lipoprotein, Outcome112450.5 ± 16.319752.3 ± 20.00.21 Low-density lipoprotein, Exposure717103.6 ± 29.611698.3 ± 28.40.08 Low-density lipoprotein, Outcome1092106.1 ± 31.519695.0 ± 35.3< 0.0001Body mass index, kg/m2205128.5 ± 7.436829.2 ± 8.00.13MedianRangeMedianRangePaAge at DNA sampling, years32.318.0, 66.429.618.4, 65.1< 0.0001Median age of multiple lipid measurements, yearsb Exposure30.014.6, 65.126.615.2, 56.60.03 Outcome35.319.6, 67.834.020.0, 67.50.01Time duration between DNA sampling and median age of multiple lipid measurements, years Exposure1.60.0, 5.31.60.5, 5.10.41 Outcome2.20.0, 5.52.30.1, 16.60.44CNS central nervous system, IQR interquartile range, PNET primitive neuroectodermal tumoraChi-square test for categorical variables, Student’s t-test for continuous variables. Distribution differences for both age and time duration were assessed by Wilcoxon rank sum testbThe median age of multiple lipid measurements for each survivor was calculated, and the median across all survivors was subsequently derived After quality control of DNAm data, a total of 689,414 CpGs were further advanced for EWAS. The associations between the DNAm level of each CpG and specific blood lipid level (HDL, LDL, TG, or TC) as an exposure or outcome were analyzed separately (Fig. 1). Quantile–quantile plots of each EWAS among survivors of EA and AA were shown in Additional file 1: Fig. S1 and Additional file 1: Fig. S2, respectively. EWAS of blood lipids among survivors of EA showed moderately low genomic inflation factors between 0.92 and 1.13 (Additional file 1: Fig. S1). EWAS of blood lipids among survivors of AA showed moderately low to high genomic inflation factors between 0.92 and 2.32 (Additional file 1: Fig. S2).Fig. 1Schematic framework of study design. EWAS epigenome-wide association study, DNAm DNA methylation, SJLIFE St. Jude lifetime cohort study. Median (range) of time, the median/range of time between DNAm sampling age and median age of multiple lipid measurements ## CpG sites associated with blood lipids among survivors of European ancestry The landscapes of the overall association results among survivors of EA were shown in Fig. 2. Seven, three, and 33 epigenome-wide significant blood lipid-associated CpGs were identified for HDL, TC, and TG, respectively, in the exposure EWAS ($P \leq 9$ × 10−8, Fig. 2A–C). No significant CpG achieved epigenome-wide significance in the LDL exposure EWAS among survivors of EA ($P \leq 9$ × 10−8, Fig. 2D). Detailed estimates for the association between each CpG and specific blood lipid level as an exposure were provided in Additional file 1: Table S2. Notably, a cluster of three CpGs (cg00574958, cg05325763, and cg17058475), mapped to the 5′UTR of the CPT1A gene, were common in the TG and TC exposure EWAS (Table 2 and Additional file 1: Fig. S3). Five, three, four, and 94 CpGs were significantly associated with HDL, LDL, TC, and TG, respectively, in the outcome EWAS ($P \leq 9$ × 10−8, Fig. 2E–H and Additional file 1: Table S3). Three CpGs were common across HDL, LDL, and TC outcome EWAS, including ch.1.829344F mapped to the 5′UTR region of the SRPM1 gene, cg20935223 mapped to the 3′UTR region of the CYTH3 gene, and cg21750129 mapped to the 3′UTR region of the TRPM3 gene (Table 2 and Additional file 1: Fig. S3). No significant CpGs were common between blood lipid exposure and outcome EWAS ($P \leq 9$ × 10−8).Fig. 2Manhattan plots of exposure and outcome EWAS of blood lipids among survivors of European ancestry in SJLIFE cohort. EWAS epigenome-wide association study, SJLIFE St. Jude lifetime cohort study, HDL high-density lipoprotein, LDL low-density lipoprotein, TG triglycerides, TC total cholesterolTable 2Overlapping significant CpGs in exposure and outcome EWAS of blood lipid among survivors of European ancestry in SJLIFE cohortCpGchr_hg38Start_hg38Nearby gene gengebnegeneEWASLipidEffectSEPcg005749581168840153CPT1AExposureTC− 2.25E−033.30E−042.15E−11TG− 1.28E−031.20E−041.50E−24cg053257631168840250CPT1ATC− 1.84E−033.20E−041.31E−08TG− 1.13E−031.17E−041.48E−20cg170584751168840268CPT1ATC− 2.05E−033.52E−048.81E−09TG− 1.11E−031.31E−041.27E−16ch.1.829344F124657005SRRM1OutcomeHDL6.390.891.03E−12LDL172.192.18E−14TC25.512.941.36E−17cg2093522376168606CYTH3HDL8.301.503.96E−08LDL22.273.753.99E−09TC30.225.022.40E−09cg21750129971249778TRPM3HDL7.031.053.97E−11LDL15.862.642.63E−09TC24.533.525.73E−12EWAS epigenome-wide association study, chr_hg38 chromosome in GRCh38/hg38, Start_hg38 CpG start position in GRCh38/hg38, SE standard error, HDL high-density lipoprotein, LDL low-density lipoprotein, TG triglycerides, TC total cholesterol ## CpG sites associated with blood lipids among survivors of African ancestry The overall landscape of CpG associations with blood lipid EWAS among survivors of AA were shown in the Additional file 1: Fig. S4. Six TG-associated CpGs were identified in the exposure EWAS (Additional file 1: Fig. S4A and Table S4), and five TG-associated CpGs, two TC-associated CpGs, one HDL-associated CpG, and one LDL-associated CpG were found in the outcome EWAS (Additional file 1: Fig. S4E–H and Table S4) ($P \leq 9$ × 10−8). No significant CpG was found in HDL, LDL, and TC exposure EWAS. Similarly, there was no significant CpGs common in exposure and outcome EWAS among survivors of AA ($P \leq 9$ × 10−8). In TG exposure EWAS, there were five significant CpGs mapping to nearby genes including cg26675329 and cg04747445 within 1500 bp upstream of the transcription start site of the IL18RAP gene and the BBX gene, respectively; cg05416955 and cg21376908 in the gene body of the CARD9 gene and the MSI2 gene, respectively; and cg16197879 in the 5’UTR region of the CLDN14 gene ($P \leq 9$ × 10−8, Additional file 1: Table S4). In TC outcome EWAS, cg16411101 in the 5’UTR of the SSCP3 gene was significant in both LDL and TC outcome EWAS, and the other significant CpG was cg23724016 in the 3’UTR of the CDHR5 gene ($P \leq 9$ × 10−8). In HDL outcome, one significant CpG cg14558275 is mapped to the PIK3CG gene. In TG outcome, cg05416345 and cg01111718 are in the gene body of the IFFO2 gene and the GRIA4 gene, respectively; cg04348872 and cg12686539 are in the first and second exon of the ADCY3 gene and the ZNF891 gene, respectively. We did not identify any overlapping significant CpG between survivors of EA and AA in SJLIFE cohort in any of the exposure or outcome EWAS of blood lipids ($P \leq 9$ × 10−8). ## Trans-ethnic meta-analysis In the meta-analysis of blood lipid EWAS among EA and AA survivors, we identified 74 significant lipid-CpG associations (70 unique CpGs, $P \leq 9$ × 10−8). Specifically, four, one, and 33 significant CpGs were associated with HDL, TC, and TG exposures, respectively; and two, one, two, and 31 were associated with HDL, LDL, TC, and TG outcomes, respectively ($P \leq 9$ × 10−8, Additional file 1: Table S5). Among these significant lipid-CpG associations, twelve did not reach epigenome-wide significance level in either EWAS among survivors of EA or AA alone, including three for HDL exposure, three for TG exposure, and six for TG outcome (Table 3). All 12 had homogeneous effects with the same direction of association in survivors of EA and AA (Phet > 0.1, Table 3). Among the remaining 62 lipid-CpG associations that were significant in EWAS among survivors of EA or AA alone ($P \leq 9$ × 10−8), twenty-two had homogeneous effects with the same direction between survivors of EA and AA (Phet > 0.1), twenty-four had opposite directions of association with significant heterogeneity between survivors of AA and EA (Phet < 0.1), and the remaining 16 significant lipid-CpG associations either had the same direction of association but with significant heterogeneity between survivors of EA and AA (Phet < 0.1) or had the opposite direction of association but with homogenous effect (Phet > 0.1) (Additional file 1: Table S5).Table 3Additional significant CpGs identified in meta-analysis of blood lipids EWAS among survivors of European and African ancestry ($P \leq 9$ × 10−8)EWASLipidCpGchr_hg38Start_hg38Nearby genesPopulationEffectSEPDirections (EA/AA)I2PhetExposureHDLcg27324117chr4153793134NAEA− 0.00250.00052.11E−07AA− 0.00160.00120.20Meta− 0.00240.00048.71E−08−/−00.4765cg02443467chr1047191407NAEA− 0.00220.00041.17E−07AA− 0.00160.00100.09Meta− 0.00210.00042.30E−08−/−00.5916cg07373304chr16998982RP11-161M6.3EA− 0.00210.00042.26E−07AA− 0.00440.00170.01Meta− 0.00230.00041.41E−08−/−360.2115TGcg01082498chr1168840756CPT1AEA− 0.00060.00011.24E−07AA− 0.00050.00070.48Meta− 0.00060.00017.23E−08−/−00.8322cg06004232chr1611293335RMI2EA0.00036.09E−051.20E−07AA0.00070.00030.04Meta0.00030.00011.79E−08+/+5.70.3032cg11686214chr1958573616CENPBD1P1;MZF1;MZF1-AS1EA0.00060.00011.45E−07AA0.00060.00060.24Meta0.00060.00015.20E−08+/+00.929OutcomeTGcg23550429chr230797935CAPN13EA− 40.364020.27150.05AA− 14.99632.83603.37E−07Meta− 15.48332.80873.54E−08−/−34.90.2152cg03260858chr349103337QARSEA20.115813.60510.14AA38.55966.93929.14E−08Meta34.75216.18161.89E−08+/+31.40.2272cg20601481chr1167288314ANKRD13DEA− 27.967310.88750.01AA− 37.67087.69832.10E−06Meta− 34.43646.28574.29E−08−/−00.4668cg19590858chr1434540645EAPP;RP11-671J11.5EA− 31.97529.18365.17E−04AA− 27.29946.63245.73E−05Meta− 28.90225.37687.64E−08−/−00.6798cg20433103chr1763941062SCN4AEA43.845113.16758.98E−04AA36.70398.07319.67E−06Meta38.65496.88251.95E−08+/+00.6438cg07439166chr1935902502AD000864.6;HCST;NFKBIDEA− 51.556415.23187.37E−04AA− 34.03107.41378.01E−06Meta− 37.38766.6662.04E−08−/−6.60.3009chr_hg38 chromosome in GRCh38/hg38, Start_hg38 CpG start position in GRCh38/hg38, HDL high-density lipoprotein, LDL low-density lipoprotein, TG triglycerides, TC total cholesterol, EA survivors of European ancestry, AA survivors of African ancestry ## Association between DNAm levels of lipid–associated CpGs and gene expression For each of the blood-lipid associated CpGs, we estimated the linear association between DNAm levels and gene expression levels (adjusting for DNA/RNA sampling age and sex). Among the nearby genes of the significant lipid-associated CpGs among EA, there were no count data (number of reads from RNA sequencing) for 52 genes. Of the remaining 56 CpG-gene pairs, there were ten CpG-gene pairs with significant (FDR < 0.05) associations between the DNAm level of lipid-associated CpGs and gene expression of their nearby genes, including HDAC7 (cg01620154), AXIN2 (cg23475474), ECE1 (cg01758046), TRERF1 (cg07507418), TNRC6B (cg00543524), MICU1 (cg08641767), NUDCD3 (cg01507280), NLN (cg01710244), AKAP1 (cg18807499), and LRP5 (cg24040155) among EA (Table 4). Most of the estimated effects of these CpGs were negative (i.e., increased methylation associated with decreased gene expression), except for ECE1 (cg01758046), NUDCD3 (cg01507280), and NLN (cg01710244). Among the 12 annotated genes of significant lipid-associated CpGs among AA survivors (Additional file 1: Table S4), nine gene had no count data (i.e., number of reads from RNA sequencing). Of the remaining three CpG-gene pairs (PIK3CG-cg14558275, BBX-cg04747445, and IFFO2-cg05416345), there was no significant association between DNAm levels of lipid-associated CpGs and the gene expression among AA.Table 4Significant associations between significant CpGs identified in blood lipid (exposure/outcome) EWAS among survivors of European ancestry and the expression of their nearby genes (FDR < 0.05)CpGGene Ensembl IDNearby geneEffectSEPFDRcg01620154ENSG00000061273HDAC7− 3.680.634.43E−081.24E−06cg23475474ENSG00000168646AXIN2− 4.300.722.54E−081.24E−06cg01758046ENSG00000117298ECE12.280.568.54E−051.34E−03cg07507418ENSG00000124496TRERF1− 3.490.879.55E−051.34E−03cg00543524ENSG00000100354TNRC6B− 1.490.391.69E−041.89E−03cg08641767ENSG00000107745MICU1− 1.740.472.75E−042.57E−03cg01507280ENSG00000015676NUDCD33.130.928.48E−046.78E−03cg01710244ENSG00000123213NLN5.501.691.45E−031.01E−02cg18807499ENSG00000121057AKAP1− 3.881.458.27E−034.63E−02cg24040155ENSG00000162337LRP5− 6.122.288.09E−034.63E−02SE standard error, FDR false discovery rate ## Cross-reference with the EWAS Catalog By comparing our findings with previously reported blood lipid-associated CpGs in the general population from the EWAS Catalog, only four overlapping CpGs were identified among EA survivors, including cg00574958 (associated with TG and TC exposure), cg09737197 (associated with TG exposure), and cg17058475 (associated with TG and TC exposure) in CPT1A, and cg03725309 (associated with TG exposure) in SARS (Additional file 1: Tables S6 and S7), and none among AA survivors. Among the remaining 136 novel lipid-associated CpGs in childhood cancer survivors of EA, 26 were mapped to 23 genes that have been previously reported as lipid-associated (Additional file 1: Table S8). Among 12 additional blood lipid-associated CpGs identified in the meta-EWAS, five CpGs were reported to be associated with other traits (e.g., sex, age, Schizophrenia, and ADHD (attention-deficit and hyperactivity disorder)) but only cg01082498 (in the 5’UTR region of the CPT1A gene) was associated with blood lipid level in the EWAS Catalog (Additional file 1: Table S9). ## Discussion Genetic and epigenetic (specifically, DNAm) studies have identified numerous genetic variants or CpG sites that are associated with blood lipids in the general population, hence at least 2572 genes have been implicated in lipid metabolism (Additional file 1: Fig. S5) [17, 18]. We conducted the first EWAS of blood lipids among childhood cancer survivors, including EA and AA survivors from the SJLIFE cohort. Among EA survivors, we identified 149 (140 unique CpGs) significant associations with blood lipid levels; 136 of these were novel findings. Among AA survivors, we found 14 novel significant blood lipid-associated CpGs. There was no overlapping CpGs between EA and AA survivors. A majority of these findings are unique to the survivor population, which may be attributable to childhood cancer diagnoses and/or treatments. For example, two TG exposure associated CpGs, cg24327132 and cg19120513, were associated with chest-RT and abdominal-RT [6]. In the meta-EWAS, twenty-four CpGs had opposite direction of association with significant heterogeneity between EA and AA survivors (Phet < 0.1), suggesting substantial disparity in lipid-associated CpGs between the two ancestral groups. Meta-EWAS yielded eight additional epigenome-wide significant CpGs with heterogeneity in effect size between EA and AA survivors (Phet < 0.1). However, future replication in EA or AA alone with independent data set is warranted to validate such findings. TG outcome EWAS yielded the greatest number of significant CpGs among EA, with multiple novel lipid-associated CpGs mapped to the same nearby genes, including CDK5RAP3, FCGR2B, HSPA6, and HSPA7. CDK5RAP3, known to play important roles in liver development and hepatic function. Previous research showed that hepatocyte-specific Cdk5rap3 knockout mice suffered post-weaning lethality because of impaired lipid metabolism and serious hypoglycemia [19]. FCGR2B gene encodes FcγRIIb, with a novel role in CD11c+ cells in modulating serum cholesterol and triglyceride levels and maintaining liver cholesterol homeostasis [20]. HSPA6 and HSPA7 are family members of the HSP70 proteins, which are abundantly present in cancer and play crucial roles in cancer development, progression, and metastasis, clinically resulting in diverse outcomes for patient survival [21]. Moreover, cg20935223 was significantly associated with multiple lipid traits in the outcome EWAS and mapped to CYTH3 gene. CYTH3 gene encodes Cytohesin-3, which is essential for insulin receptor signaling and body fat regulation via lipid excretion [22]. Among novel genes (i.e., not genes with nearby lipid-associated CpGs in EWAS Catalog), four of them were reported as high-confidence genes that play a role in lipid levels, including LPIN2 (near cg07616376 associated with TG exposure among EA), SCARB1 (near cg08458758 associated with TG outcome among EA), MSI2 (near cg21376908 associated with TG exposure among AA), and SSBP3 (near cg16411101 associated with LDL and TC outcome among AA) [23]. We integrated gene expression levels from RNA sequencing to further characterize the associations between DNAm and blood lipid levels, which strengthened this study. For example, we demonstrated that ten blood lipid-associated CpGs were associated with levels of expression of the annotated genes, in which seven were inversely associated. However, it is important to note that there are several limitations in this study. First, although we innovatively designed both exposure and outcome EWAS based on our longitudinal follow-up study, the cross-sectional nature of the data prevented us from disentangling the complex interplay between DNAm and blood lipid levels. Nevertheless, we demonstrated potential regulation of gene expression as plausible mechanisms for DNAm alterations by performing RNA-sequencing analysis. Second, the sample size of survivors of AA was limited, that led to the limited power and the exploratory nature of the AA EWAS (i.e., some findings might be identified by chance). However, differences between EA and AA populations as determined by their genetic ancestry were observed with no overlapping blood lipid levels associated CpG between survivors of EA and those of AA. To further validate the findings, a larger sample size of AA survivors is warranted in the future. Previous methodological work suggested that more than 1,000 subjects are required to achieve $80\%$ power for detection of differential DNAm at nominal genome-wide significance with an odds ratio of 1.15 [24]. Third, we obtained DNAm data at only one time point. In the outcome EWAS, all the blood lipid levels were measured after blood draw for DNAm, so the DNAm may be predictive of blood lipid levels. However, to better assess and interpret the changes of blood lipids level in exposure EWAS, longitudinal DNAm measurement (ideally, after the first blood lipid level measurement) is required to correlate changes of DNAm between two time points with changes in blood lipid levels. Fourth, the follow-up of our cohort is limited and still-ongoing, so there was large proportion of missing data in the analytic setting of bi-directional association between DNAm and lipid levels which requires multiple clinical assessments of lipid levels. Lastly, we did not consider cell type-specific DNAm in the current work. Recent research identified that DNAm variation in diseases, such as type 1 diabetes, can be cell type-specific [25]. Therefore, in the future, we may deconvolute bulk DNAm measured in blood leukocytes into cell type–specific quantities and analyze the DNAm associations of each specific cell type. ## Conclusions Our findings demonstrated distinct DNAm signatures associated with blood lipid levels in EA and AA survivors, and that an additional set of genes may be implicated in lipid metabolism in the survivor population compared to the general population. Further longitudinal studies are warranted to replicate and validate DNAm biomarkers for blood lipid levels and other CHCs to facilitate the clinical translation for improved survivorship care. ## Study population SJLIFE is a retrospectively-constructed cohort with periodic evaluations of survivors beyond 5-years from childhood cancer diagnosis who were treated at St. Jude Children’s Research Hospital. The details of SJLIFE cohort study have been previously described [15, 16, 26]. Participants complete questionnaires assessing demographic and clinical factors, and receive comprehensive medical and laboratory assessments at each visit to determine health conditions. In this study, a total of 2,052 survivors of EA and 370 survivors of AA, with genome-wide DNAm profiling data, were included [27]. The ancestry for each survivor was determined using genotypes derived from whole-genome sequencing and population admixture analysis as previously described [28]. Primary childhood cancer diagnoses, exposure to chemotherapeutic agents and region-specific radiation dosimetry was obtained from medical records. All SJLIFE survivors completed at least one comprehensive clinical assessment that included a battery of laboratory tests including blood lipid measurement (HDL, LDL, TC, and TG) [26]. The blood lipid levels measured before blood sampling for DNAm were used for exposure EWAS and the blood lipid levels measured after were used in outcome EWAS (Fig. 1). Weighted average was calculated if there were multiple measurements, and time intervals between two consecutive measurements were used as weights. We excluded lipid measurements without fasting. Samples with only one lipid measurements (coinciding with the time point for the blood draw for DNAm) were excluded to ensure that our exposure and outcome EWAS examined the temporal association between DNAm and blood lipid levels. All participants provided written informed consent, with institutional review board approval at St. Jude Children’s Research Hospital. ## DNAm profiling and data processing Illumina Infinium® MethylationEPIC BeadChip array including 850K CpG sites was used to generate genome-wide DNAm profiling on DNA derived from peripheral blood mononuclear cells (PBMC) collected at each follow-up visit for SJLIFE survivors. Details about laboratory experimental processes, array scanning, and DNAm bioinformatics data analysis were previously described by Song et al. [ 6]. ## Genotyping based on whole-genome sequencing (WGS) Genotyping was based on whole-genome sequencing data of blood derived DNA from 4402 SJLIFE survivors as previously described [29, 30]. Details about data processing, genotyping calling as well as additional genotype quality control criteria and procedures were previously described in Dong et al. [ 31]. ## Epigenome-wide association analysis Bidirectional EWAS was conducted using a multivariable linear regression to test the association of DNAm levels at each CpG (M-value, continuous variable) with blood lipid levels (continuous variable). We performed principal components analysis of methylation levels of all CpG sites to quantify potential batch effects in the DNAm data. The top four principal components were determined by the change rate of eigenvalues [6] and were included as covariates in the regression model. We also performed principal components analysis of genotypes derived from WGS to quantify the population substructure in EA and AA survivors. The top four principal components were determined by the change rate of eigenvalues and were included as covariates in the regression model. In the exposure EWAS, a multivariable linear regression model was used with lipid level (weighted average was calculated if there were multiple measurements, and time intervals between two consecutive measurements were used as weights) prior to DNA sampling as an independent variable and DNAm as a dependent variable, adjusting for sex, age at DNA sampling, leukocyte subtype proportions, top four significant genetic principal components, top four methylation principal components, cancer treatments, median age of lipid measurement, BMI, cigarette smoking, and lipid lowering medicine use. All these covariates were potential confounding factors for DNAm level of each CpG, and hence were considered in the exposure EWAS. Cancer treatments included chemotherapy and radiation therapy within 5 years from primary childhood cancer diagnosis. The chemotherapy agents included classical alkylating agent, anthracyclines, corticosteroids, vinca alkaloids, asparaginase enzymes, antimetabolites, and epipodophyllotoxins. The region-specific RT included brain-RT, chest-RT, abdomen-RT, and pelvis-RT. For smoking status as a categorical variable, we included three levels (“never”, “ever”, and “unknown”) in the model. BMI was measured at the same time as DNAm sampling. CpGassoc R package was used for the exposure EWAS analyses [32]. In the outcome EWAS, a multivariable linear regression model was used for DNAm (age-, sex-, cell-type-, genotype principal components-, and methylation principal components- adjusted) as an independent variable and lipid level (weighted average was calculated if there were multiple measurements, and time intervals between two consecutive measurements were used as weights) after DNA sampling as a dependent variable. For EA survivors, a base model without DNAm level but including the complete set of covariate (i.e., sex, cancer treatments, median age of lipid measurement, BMI, smoking, lipid lowering medicine use, lipid level measured at DNA sampling, age at DNA sampling, and polygenic risk score for specific lipid level (in EA only) was fitted. Cancer treatment exposures that were not statistically significant ($P \leq 0.05$) in the base model were subsequently excluded. In the final model, DNAm level of each CpG was added for the EWAS analysis. For AA survivors, considering the smaller sample size and potential overfitting, a similar but slightly different variable selection approach was taken by additionally excluding BMI, smoking status, and lipid lowering medicine use if any of these was not statistically significant ($P \leq 0.05$) in the base model. Polygenic risk score for specific lipid level was constructed by following the same approach described previously [28] for EA survivors. Custom R code was used for the outcome EWAS analyses. A P value less than 9 × 10−8 was deemed as epigenome-wide significance level corresponding to $5\%$ family-wise error [33]. ## RNA-sequence profiling and data processing RNA was extracted from the same PBMC used for DNA methylation profiling. Details of library construction, sequencing, and data processing were described previously [27]. Briefly, paired-end 100 cycle sequencing was performed on a NovaSeq 6000 (Illumina). After quality control procedures, raw reads from the fastq files were aligned to the GRCh38.p13 version (v31) of the reference human genome from GENCODE through the automated internal pipeline [34]. The generated bam files were sorted and used to build an index using Samtools (version 1.9) [35] then used as inputs for counting reads using htseq-count [36] with GENCODE v31 gene annotation gtf file. A total of 165 samples of RNA-seq data (135 EA survivors and 30 AA survivors) were available for further analysis. After removing transcripts with mean read counts across all 165 samples less than 10, a total of 12,882 genes were determined to be expressed in PBMC. Transcripts per million (TPM) [37] were calculated and transformed in the form of log2(TPM + 0.01). The function normalizeQuantiles in the limma package [38] in R (version 3.6.1) was used for quantile normalization [39] of the log-transformed values before further downstream analyses. ## Expression quantitative trait methylation We used the Infinium® MethylationEPIC BeadChip array annotations (v1.0 B5) provided by Illumina (https://webdata.illumina.com/downloads/productfiles/methylationEPIC/infinium-methylationepic-v-1-0-b5-manifest-file-csv.zip) to map CpGs to their annotated genes. For 135 RNA-seq samples from EA survivors (out of 165 in total), the normalized expression values of the nearest genes of each lipid-associated CpGs were extracted to fit a linear regression against the DNAm levels of each CpG. ## Additional statistical and bioinformatic analyses Quantile–Quantile plots were generated from P-values in each EWAS using the R lattice package. GenABEL R package [40] was used to estimate genomic inflation factor (i.e., lambda). We searched lipid-associated CpGs and nearby genes identified in our study and compared with those previously reported from the EWAS Catalog [17]. ## Supplementary Information Additional file 1. Supplementary Information. ## References 1. Mittelstraß K, Waldenberger M. DNA methylation in human lipid metabolism and related diseases. Curr Opin Lipidol (2018) 29 116-124. DOI: 10.1097/MOL.0000000000000491 2. Robison LL, Hudson MM. Survivors of childhood and adolescent cancer: life-long risks and responsibilities. Nat Rev Cancer (2014) 14 61-70. DOI: 10.1038/nrc3634 3. Bhakta N, Liu Q, Ness KK, Baassiri M, Eissa H, Yeo F. The cumulative burden of surviving childhood cancer: an initial report from the St Jude lifetime cohort study (SJLIFE). Lancet (2017) 390 2569-2582. DOI: 10.1016/S0140-6736(17)31610-0 4. Armenian SH, Armstrong GT, Aune G, Chow EJ, Ehrhardt MJ, Ky B. Cardiovascular disease in survivors of childhood cancer: insights into epidemiology, pathophysiology, and prevention. J Clin Oncol (2018) 36 2135-2144. DOI: 10.1200/JCO.2017.76.3920 5. Khanna A, Pequeno P, Gupta S, Thavendiranathan P, Lee DS, Abdel-Qadir H. Increased risk of all cardiovascular disease subtypes among childhood cancer survivors: population-based matched cohort study. Circulation (2019) 140 1041-1043. DOI: 10.1161/CIRCULATIONAHA.119.041403 6. Song N, Hsu CW, Pan H, Zheng Y, Hou L, Sim JA. Persistent variations of blood DNA methylation associated with treatment exposures and risk for cardiometabolic outcomes in long-term survivors of childhood cancer in the St. Jude lifetime cohort. Genome Med (2021) 13 53. DOI: 10.1186/s13073-021-00875-1 7. Hudson MM, Ness KK, Gurney JG, Mulrooney DA, Chemaitilly W, Krull KR. Clinical ascertainment of health outcomes among adults treated for childhood cancer. JAMA (2013) 309 2371-2381. DOI: 10.1001/jama.2013.6296 8. Reulen RC, Winter DL, Frobisher C, Lancashire ER, Stiller CA, Jenney ME. Long-term cause-specific mortality among survivors of childhood cancer. JAMA (2010) 304 172-179. DOI: 10.1001/jama.2010.923 9. Frank AT, Zhao B, Jose PO, Azar KM, Fortmann SP, Palaniappan LP. Racial/ethnic differences in dyslipidemia patterns. Circulation (2014) 129 570-579. DOI: 10.1161/CIRCULATIONAHA.113.005757 10. Sheet SF. Older Americans and cardiovascular diseases (2013) 11. Liu Q, Leisenring WM, Ness KK, Robison LL, Armstrong GT, Yasui Y. Racial/ethnic differences in adverse outcomes among childhood cancer survivors: the childhood cancer survivor study. J Clin Oncol (2016) 34 1634-1643. DOI: 10.1200/JCO.2015.66.3567 12. Husquin LT, Rotival M, Fagny M, Quach H, Zidane N, McEwen LM. Exploring the genetic basis of human population differences in DNA methylation and their causal impact on immune gene regulation. Genome Biol (2018) 19 222. DOI: 10.1186/s13059-018-1601-3 13. Breeze CE, Batorsky A, Lee MK, Szeto MD, Xu X, McCartney DL. Epigenome-wide association study of kidney function identifies trans-ethnic and ethnic-specific loci. Genome Med (2021) 13 74. DOI: 10.1186/s13073-021-00877-z 14. Breeze CE, Wong JYY, Beck S, Berndt SI, Franceschini N. Diversity in EWAS: current state, challenges, and solutions. Genome Med (2022) 14 71. DOI: 10.1186/s13073-022-01065-3 15. Hudson MM, Ness KK, Nolan VG, Armstrong GT, Green DM, Morris EB. Prospective medical assessment of adults surviving childhood cancer: study design, cohort characteristics, and feasibility of the St. Jude lifetime cohort study. Pediatr Blood Cancer (2011) 56 825-836. DOI: 10.1002/pbc.22875 16. Hudson MM, Ehrhardt MJ, Bhakta N, Baassiri M, Eissa H, Chemaitilly W. Approach for classification and severity grading of long-term and late-onset health events among childhood cancer survivors in the St. Jude lifetime cohort. Cancer Epidemiol Biomark Prev (2017) 26 666-674. DOI: 10.1158/1055-9965.EPI-16-0812 17. Battram T, Yousefi P, Crawford G, Prince C, Sheikhali Babaei M, Sharp G. The EWAS catalog: a database of epigenome-wide association studies. Wellcome Open Res (2022) 7 41. DOI: 10.12688/wellcomeopenres.17598.2 18. Buniello A, MacArthur JAL, Cerezo M, Harris LW, Hayhurst J, Malangone C. The NHGRI-EBI GWAS catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res (2019) 47 D1005-d1012. DOI: 10.1093/nar/gky1120 19. Yang R, Wang H, Kang B, Chen B, Shi Y, Yang S. CDK5RAP3, a UFL1 substrate adaptor, is crucial for liver development. Development (2019) 146 1-13 20. Marvin J, Rhoads JP, Major AS. FcγRIIb on CD11c(+) cells modulates serum cholesterol and triglyceride levels and differentially affects atherosclerosis in male and female Ldlr(-/-) mice. Atherosclerosis (2019) 285 108-119. DOI: 10.1016/j.atherosclerosis.2019.04.221 21. Albakova Z, Armeev GA, Kanevskiy LM, Kovalenko EI, Sapozhnikov AM. HSP70 multi-functionality in cancer. Cells (2020) 9 587. DOI: 10.3390/cells9030587 22. Jux B, Gosejacob D, Tolksdorf F, Mandel C, Rieck M, Namislo A. Cytohesin-3 is required for full insulin receptor signaling and controls body weight via lipid excretion. Sci Rep (2019) 9 3442. DOI: 10.1038/s41598-019-40231-3 23. Kanoni S, Graham SE, Wang Y, Surakka I, Ramdas S, Zhu X. Implicating genes, pleiotropy, and sexual dimorphism at blood lipid loci through multi-ancestry meta-analysis. Genome Biol (2022) 23 1-42. DOI: 10.1186/s13059-022-02837-1 24. Tsai PC, Bell JT. Power and sample size estimation for epigenome-wide association scans to detect differential DNA methylation. Int J Epidemiol (2015) 44 1429-1441. DOI: 10.1093/ije/dyv041 25. Paul DS, Teschendorff AE, Dang MA, Lowe R, Hawa MI, Ecker S. Increased DNA methylation variability in type 1 diabetes across three immune effector cell types. Nat Commun (2016) 7 13555. DOI: 10.1038/ncomms13555 26. Howell CR, Bjornard KL, Ness KK, Alberts N, Armstrong GT, Bhakta N. Cohort profile: the St. Jude lifetime cohort study (SJLIFE) for paediatric cancer survivors. Int J Epidemiol (2021) 50 39-49. DOI: 10.1093/ije/dyaa203 27. Song N, Sim JA, Dong Q, Zheng Y, Hou L, Li Z. Blood DNA methylation signatures are associated with social determinants of health among survivors of childhood cancer. Epigenetics (2022) 17 1-15. DOI: 10.1080/15592294.2022.2030883 28. Wang Z, Liu Q, Wilson CL, Easton J, Mulder H, Chang T-C. Polygenic determinants for subsequent breast cancer risk in survivors of childhood cancer: the St Jude lifetime cohort study (SJLIFE) polygenic determinants for subsequent breast cancer risk. Clin Cancer Res (2018) 24 6230-6235. DOI: 10.1158/1078-0432.CCR-18-1775 29. Qin N, Wang Z, Liu Q, Song N, Wilson CL, Ehrhardt MJ. Pathogenic germline mutations in DNA repair genes in combination with cancer treatment exposures and risk of subsequent neoplasms among long-term survivors of childhood cancer. J Clin Oncol (2020) 38 2728. DOI: 10.1200/JCO.19.02760 30. Wang Z, Wilson CL, Easton J, Thrasher A, Mulder H, Liu Q. Genetic risk for subsequent neoplasms among long-term survivors of childhood cancer. J Clin Oncol (2018) 36 2078. DOI: 10.1200/JCO.2018.77.8589 31. Dong Q, Song N, Qin N, Chen C, Li Z, Sun X. Genome-wide association studies identify novel genetic loci for epigenetic age acceleration among survivors of childhood cancer. Genome Med (2022) 14 1-12. DOI: 10.1186/s13073-022-01038-6 32. Barfield RT, Kilaru V, Smith AK, Conneely KN. CpGassoc: an R function for analysis of DNA methylation microarray data. Bioinformatics (2012) 28 1280-1281. DOI: 10.1093/bioinformatics/bts124 33. Mansell G, Gorrie-Stone TJ, Bao Y, Kumari M, Schalkwyk LS, Mill J. Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array. BMC Genom (2019) 20 366. DOI: 10.1186/s12864-019-5761-7 34. Parker M, Mohankumar KM, Punchihewa C, Weinlich R, Dalton JD, Li Y. C11orf95-RELA fusions drive oncogenic NF-κB signalling in ependymoma. Nature (2014) 506 451-455. DOI: 10.1038/nature13109 35. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N. The sequence alignment/map format and SAMtools. Bioinformatics (2009) 25 2078-2079. DOI: 10.1093/bioinformatics/btp352 36. Anders S, Pyl PT, Huber W. HTSeq–a Python framework to work with high-throughput sequencing data. Bioinformatics (2015) 31 166-169. DOI: 10.1093/bioinformatics/btu638 37. Wagner GP, Kin K, Lynch VJ. Measurement of mRNA abundance using RNA-seq data: RPKM measure is inconsistent among samples. Theory Biosci (2012) 131 281-285. DOI: 10.1007/s12064-012-0162-3 38. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res (2015) 43 e47. DOI: 10.1093/nar/gkv007 39. Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics (2003) 19 185-193. DOI: 10.1093/bioinformatics/19.2.185 40. Aulchenko YS, Ripke S, Isaacs A, van Duijn CM. GenABEL: an R library for genome-wide association analysis. Bioinformatics (2007) 23 1294-1296. DOI: 10.1093/bioinformatics/btm108
--- title: Comparative study of extracellular vesicles derived from mesenchymal stem cells and brain endothelial cells attenuating blood–brain barrier permeability via regulating Caveolin-1-dependent ZO-1 and Claudin-5 endocytosis in acute ischemic stroke authors: - Yiyang Li - Bowen Liu - Tingting Zhao - Xingping Quan - Yan Han - Yaxin Cheng - Yanling Chen - Xu Shen - Ying Zheng - Yonghua Zhao journal: Journal of Nanobiotechnology year: 2023 pmcid: PMC9976550 doi: 10.1186/s12951-023-01828-z license: CC BY 4.0 --- # Comparative study of extracellular vesicles derived from mesenchymal stem cells and brain endothelial cells attenuating blood–brain barrier permeability via regulating Caveolin-1-dependent ZO-1 and Claudin-5 endocytosis in acute ischemic stroke ## Abstract ### Background Blood–brain barrier (BBB) disruption is a major adverse event after ischemic stroke (IS). Caveolin-1 (Cav-1), a scaffolding protein, played multiple roles in BBB permeability after IS, while the pros and cons of Cav-1 on BBB permeability remain controversial. Numerous studies revealed that extracellular vesicles (EVs), especially stem cells derived EVs, exerted therapeutic efficacy on IS; however, the mechanisms of BBB permeability needed to be clearly illustrated. Herein, we compared the protective efficacy on BBB integrity between bone marrow mesenchymal stem cells derived extracellular vesicles (BMSC-EVs) and EVs from brain endothelial cells (BEC-EVs) after acute IS and investigated whether the mechanism was associated with EVs antagonizing Cav-1-dependent tight junction proteins endocytosis. ### Methods BMSC-EVs and BEC-EVs were isolated and characterized by nanoparticle tracking analysis, western blotting, and transmission electron microscope. Oxygen and glucose deprivation (OGD) treated b. End3 cells were utilized to evaluate brain endothelial cell leakage. CCK-8 and TRITC-dextran leakage assays were used to measure cell viability and transwell monolayer permeability. Permanent middle cerebral artery occlusion (pMCAo) model was established, and EVs were intravenously administered in rats. Animal neurological function tests were applied, and microvessels were isolated from the ischemic cortex. BBB leakage and tight junction proteins were analyzed by Evans Blue (EB) staining and western blotting, respectively. Co-IP assay and Cav-1 siRNA/pcDNA 3.1 vector transfection were employed to verify the endocytosis efficacy of Cav-1 on tight junction proteins. ### Results Both kinds of EVs exerted similar efficacies in reducing the cerebral infarction volume and BBB leakage and enhancing the expressions of ZO-1 and Claudin-5 after 24 h pMCAo in rats. At the same time, BMSC-EVs were outstanding in ameliorating neurological function. Simultaneously, both EVs treatments suppressed the highly expressed Cav-1 in OGD-exposed b. End3 cells and ischemic cerebral microvessels, and this efficacy was more prominent after BMSC-EVs administration. Cav-1 knockdown reduced OGD-treated b. End3 cells monolayer permeability and recovered ZO-1 and Claudin-5 expressions, whereas Cav-1 overexpression aggravated permeability and enhanced the colocalization of Cav-1 with ZO-1 and Claudin-5. Furthermore, Cav-1 overexpression partly reversed the lower cell leakage by BMSC-EVs and BEC-EVs administrations in OGD-treated b. End3 cells. ### Conclusions Our results demonstrated that Cav-1 aggravated BBB permeability in acute ischemic stroke, and BMSC-EVs exerted similar antagonistic efficacy to BEC-EVs on Cav-1-dependent ZO-1 and Claudin-5 endocytosis. BMSC-EVs treatment was superior in Cav-1 suppression and neurological function amelioration. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12951-023-01828-z. ## Introduction As a severe cerebral vascular disease that accounts for nearly $10\%$ of mortality and $5\%$ of disability globally, stroke has brought miserable medical and economic burdens to patients and society [1]. Annually, about 795,000 people suffered from stroke based on the data from the National Heart, Lung, and Blood Institute (NHLBI) of the United States, which led to $52.8 billion in disease-related costs between 2017 and 2018. Among all types of stroke occurrences, about $87\%$ are categorized as ischemic stroke (IS) [2]. Effective intervention measures are still limited for IS treatment, and one of the therapeutics is intra-arterial or intravenous thrombolysis by administration of tissue plasminogen activator (t-PA). However, due to the narrow therapeutic time window and individual contraindications, only < $6.5\%$ of patients benefited from the therapy in the USA, and hemorrhagic transformation is difficult to avoid [3–5]. Another therapeutic measure is mechanical thrombectomy. As an invasive endovascular approach, it is suitable for patients with large artery occlusion, but it also produces high risks, e.g., vascular injury, intracerebral hemorrhage, and new territory occlusion [6]. Moreover, although $95\%$ of the published preclinical studies showed neuroprotective drugs had positive efficacies in animal IS models from 1990–2018, few achieved satisfactory outcomes in clinical Phase III trials [7]. Therefore, exploiting novel therapeutic approaches is essential. The blood–brain barrier (BBB) acts as a dynamic “doorkeeper” to mediate the liquid and substance exchange for maintaining the homeostasis between brain parenchyma and peripheral circulation [8]. However, once the function and structure of the neurovascular unit (NVU) are destroyed, it renders plenty of peripheral substances and immune cells to enter the brain parenchyma and eventually aggravates neurological functional deficit [9]. During pathological progression, transcytosis and paracellular barrier opening mediated by translocation and degradation of tight junction (TJ) proteins account for two significant modes of BBB permeability [9]. Caveolin-1 (Cav-1), as a caveolae scaffolding protein, predominated in the transcytosis process instead of the paracellular barrier in the early stage of ischemia/reperfusion (IR) [10]. However, the relationship between Cav-1 and TJ proteins is still controversial [11]. Studies indicated focal cerebral IR enhanced the production of nitric oxide (NO), resulting in the loss of Cav-1 and eventually contributed to matrix metalloproteinases (MMPs) activation for the degradation of TJ proteins [12]. Besides, the redistribution of Claudin-5 in cerebral microvessels aggravated BBB permeability, and the mechanism was related to the high expression of Cav-1 in ischemic brain endothelial cells (BECs) after two-hour middle cerebral artery occlusion (MCAo) in rats [13]. Although the paradoxical phenomenon was attributed to different experimental models and the discrepancy of observation time points after IS [14], the efficacy of Cav-1 on TJ protein regulation still awaits to be illustrated in acute IS. Extracellular vesicles (EVs) are natural phospholipid bilayer tiny vesicles secreted by almost all kinds of cells. Their size and characteristics enable them to penetrate BBB easily [15, 16]. In IS treatment, EVs have presented the potential to facilitate neuro-angiogenesis, regulate immune response and inflammation, and ameliorate neurological function [17]. Since cell transplantation therapy became popular in recent studies, stem cell therapy has been widely proven to bring magnificent benefits in IS recovery and has been conducted in clinical trials [18], however, potential side effects, including uncertainty of biodistribution and uncontrollable cell differentiation, limit cell transplantation therapy development. Compared with bone marrow mesenchymal stem cells (BMSC) transplantation for IS treatment, stem cells originated EVs exert similar efficacy and partly avoid adverse reactions such as undesirable differentiation, vascular embolism, and seizure [19]. Transplanting EVs from optimal cell sources should be a promising approach for IS treatment. NVU cells-derived BEC-EVs exerted satisfied neurological recovery efficacy as previously reported after IS [20–22]. Similarly, BMSCs derived EVs treatment more widely mediated multiple pathways for IS recovery [23–25]. Nevertheless, few studies report their efficacies on BBB disruption in acute ischemic stroke, and virtually seldom research demonstrates the outcome discrepancy among EV treatments from different cell origins. In this study, we isolated EVs from cultured BMSCs and BECs and illustrated their efficacies on BBB integrity and neurological functional improvement in the rat permanent MCAO model. Simultaneously, we firstly defined the effectiveness of BMSC-EVs against Cav-1 expression was superior to that of BEC-EVs and described the colocalization of Cav-1 with TJ proteins (ZO-1 and Claudin-5) in BECs under hypoxic conditions. Also, our results suggested that the therapeutic mechanism of EVs from two sources of cells on BBB permeability was characterized by the regulation of Cav-1- dependent ZO-1 and Claudin-5 endocytosis. ## Isolation of EVs EVs derived from BMSCs and BECs were isolated by ultrafiltration and Exo-Prep kit (HansaBioMed Life Sciences) according to the manufacturer’s protocol. Briefly, BMSCs and BECs were seeded into 75 cm3 flasks and cultured to 50–$60\%$ confluence. Then, EVs-deprived fetal bovine serum (EVs-free FBS) was prepared by ultracentrifuge at 100,000 g for 18 h under 4 ℃ (XPN-100, Beckman Coulter) according to a previous report [26] and then was added into culture medium when cells grew to 70–$80\%$ confluence. Eventually, cells were allowed to grow for another 24 h. After that, the supernatants were collected, centrifuged under 1000 g for 5 min, filtered by 0.22 μm membrane filters (Millipore), and then concentrated by ultrafiltration spin columns (28932358, Cytiva). Exo-prep reagent was added to the concentrated supernatant and left to stand for 1 h on ice until centrifuged at 10,000 g for another 1 h under 4 ℃. The EVs-enriched precipitation was washed with cold PBS 3 times and centrifuged again at 10,000 g under 4 ℃ for 5 min to fully abandon the supernatant and reagent. The final precipitation, including BEC-EVs or BMSC-EVs, was resuspended in 200 μl PBS and stored at − 80 ℃ refrigerator till further experiments. All samples were filtered again using 0.22 μm membrane filters to ensure sample sterility before administration. ## EVs size distribution, morphology identification, and phagocytosis experiment The size distribution of BEC-EVs and BMSC-EVs was determined by a nanoparticle tracking analysis (NTA) device (Nanosight-NS500, Malvern Panalytical Ltd), and the morphological images were taken by transmission electron microscope (TEM) under 40 k × magnification (Service provided by Servicebio Inc.). Western blotting was used to identify the surface markers (TSG 101, HSP 70, Alix, CD 9, and CD 63). To verify EVs from two sources of cells were phagocytized by BECs, they were labeled by 5 μM DiI (V22885, Invitrogen) and then washed by PBS 3 times to remove the excess dye, and DiI labeled BEC/BMSC-EVs were then co-cultured with b. End3 cells for 10–120 min under 37 ℃ or 4 ℃. Subsequently, cells were fixed by $4\%$ paraformaldehyde (PFA) and permeabilized by $0.5\%$ Triton X-100. Nucleus was stained by DAPI (C1005, Beyotime) before observation. Images were taken by a confocal microscope (SP8, Leica) under 63 × objective. For negative control, DMEM containing EVs-free FBS was subjected to EVs isolation and DiI dye labeling, the same as EVs samples. The negative control samples were also applied to NTA, TEM, and phagocytosis experiments. ## Cell cultures and OGD treatment BECs (b. End3 mouse brain endothelial cell line) and rat BMSCs (Purchased from Shanghai Zhong Qiao Xin Zhou Biotechnology Co., Ltd) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (11965092, Gibco) with $1\%$ penicillin/streptomycin (15070063, Gibco) and $10\%$ fetal bovine serum (FBS) (26140079, Gibco) under 37℃ with $95\%$ O2 and $5\%$ CO2. Cells within 10 passages were employed in the study. BECs were subjected to 4 or 6 h-oxygen and glucose deprivation (OGD) by a hypoxic chamber to mimic the hypoxic and glucose shortage status. Briefly, the medium of BECs was replaced with glucose-deprived DMEM (11966025, Gibco), and cells were transferred to a hypoxic chamber (MIC-101, Billups-Rothenberg). The oxygen percentage in the chamber was determined by Nuvair O2 QuickStick (Nuvair). The ventilation rate of N2 was controlled between 10–20 L/min, and the chamber was sealed until the O$2\%$ ≤ $0.5\%$, and then BECs were cultured at 37 ℃ for 4 or 6 h. ## Determination of optimal OGD duration and EVs administered dosage To determine optimal OGD duration and dosage of EVs administration, b. End3 cells at the density of 1 × 104 cells/well were seeded into 96 well plates and cultured overnight at 37 ℃ with $5\%$ CO2, and then 4 or 6-h-OGD was employed. After that, the medium containing $10\%$ CCK-8 solution (C0037, Beyotime) was replaced in each well. BECs were cultured for another 2 h until the absorbance of each well was read by a microplate reader (M5, SpectraMax) with the excitation wavelength at 450 nm. OD value of each well was read, normalized, and presented as ratio vs. Control group (set as 1). For suitable OGD duration assessment, cell viability in groups of Control, OGD 4 h, and OGD 6 h was evaluated. And for optimal EVs dosage appraisal, the particle number of isolated EVs was determined by NTA. BECs were divided into 8 groups as follows: Control, OGD, OGD + low dosage of BEC-EVs (~ 5 × 109 BEC-EVs), OGD + middle dosage of BEC-EVs (~ 1 × 1010 BEC-EVs), OGD + high dosage of BEC-EVs (~ 2 × 1010 BEC-EVs), and OGD + low dosage of BMSC-EVs (~ 5 × 109 BMSC-EVs), OGD + middle dosage of BMSC-EVs (~ 1 × 1010 BMSC-EVs), OGD + high dosage of BMSC-EVs (~ 2 × 1010 BMSC-EVs). ## Permanent MCAo model establishment, EVs administration, and animal grouping The animal experiment protocol was approved by the Ethics Committee of the University of Macau (Ethics number: UMARE-035–2020), and animal benefits were ensured according to the Guide for the Care and Use of Laboratory Animals (8th edition, Washington, DC: The National Academies Press, 2011). Healthy male Sprague–Dawley (SD) rats (weighing 250-280 g, ~ 8 weeks) were anesthetized by $1.5\%$ (w/v) sodium pentobarbital intraperitoneal injection (30 mg/kg) before they suffered from surgical procedure. A longitudinal midline incision in the neck was made at 2–3 cm below the incisor tooth, and common carotid, internal carotid, and external carotid arteries separation was performed. Then, a micro-cut was made at the common carotid artery, and 0.25 mm-diameter monofilament thread with a silicone top was inserted via the cut and advanced 1.6–1.8 cm into the internal carotid artery to reach the bifurcation of middle cerebral artery. Finally, the incision was closed and disinfected. Surgery that only made the cut and artery separation without thread insertion was done in SHAM group rats. Rats were grouped according to random digits table generated by computer software. 4 groups were randomly generated according to the table, including SHAM, pMCAo, pMCAo + BEC-EVs (1 × 1010 BEC-EVs), and pMCAo + BMSC-EVs (1 × 1010 BMSCs-EVs). Rats with the modified neurological severity score (mNSS) above 6 and Bederson score above 2 post surgery were considered to be successful modeled, and those without successful established pMCAo or dead were excluded in this study. BEC-EVs and BMSC-EVs at the amount of ~ 1 × 1010 particles were intravenously administered once via tail vein immediately after surgery whereas PBS without EVs was employed in pMCAo group with the same routine. Total 65 rats were used in this study, and 9 rats were dead before 24 h after surgery, which were excluded from the study. For TTC staining and Evans blue (EB) leakage experiments, 8 rats per groups were applied, and 3 rats per group for brain microvessels isolation, western blotting, and immunofluorescence. 10 rats per group for neurological function evaluation. During the pMCAo surgery, all procedures were conducted gently by skilled investigators, and animal body temperature were kept at 37 ℃ by warm pad to avoid body temperature loss. All rats were kept and treated evenly in an animal room with 12 h of light/dark cycle lighting environment under 20–25 °C room temperature and 50–$60\%$ humidity for 24 h. Drug administration order for different rats were randomized in every individual experiment to minimize confounders. Foods and water were freely available in cages after the surgery and EVs treatment, and the rats were euthanized by CO2 inhalation at the end of the study. ## Brain cortex microvessels isolation Rat brain microvessels (BMV) were isolated based on previously reported methods [27], and all procedures were conducted on ice. Briefly, brain samples were collected and shortly preserved in MCDB 131 medium (10372019, Gibco). After white-matter was roughly removed, cortex samples were homogenized and centrifuged (2000 g, 5 min at 4 °C). The supernatant was discarded, and the pellet was resuspended in $15\%$ dextran (MW ~ 70 kDa; 31,390, Sigma-Aldrich). After centrifuged at 10,000 g for 15 min under 4 °C, the pellets with enriched BMV were achieved for further analysis. The isolated BMV was identified by immunofluorescence staining and western blotting. ## The evaluation of neurological function, infarct volume and EB leakage Neurological function in rats was measured by Bederson score (5-points scoring scale) and mNSS (18-points scoring scale) as described before [28, 29]. Scores in each group were recorded immediately at 6 h, 12 h and 24 h after pMCAo by trained investigators who were blinded to the experiment design. The detailed score scales were provided in (Additional file 1). For brain infarct volume evaluation, $2\%$ 2,3,5-Triphenyl tetrazolium chloride (TTC, T819366, Macklin) were used to stain 2 mm coronal slices under 37 ℃ for 20 min avoiding light. ImageJ (Version 1.53f51, NIH) software was used to calculate the ratio of infarct area to the whole brain area. For EB leakage analysis, rats were intravenously administered $2\%$ EB (4 ml/kg) and allowed to circulate for 1 h. Rats were sacrificed by intraventricularly perfused with 50 ml cold PBS under anesthesia. Whole brains were collected and imaged by IVIS® Spectrum small animal image system (PerkinElmer) at excitation wavelength: 620 nm and emission wavelength: 710 nm. ## BECs permeability assay 5 × 104/well b. End3 cells were seeded into upper chamber with 0.4 μm pore sized 24-well transwell inserts (11820050, Costar) and cultured for 72 h to reach confluence. Before OGD stimulation, DMEM and glucose deprived DMEM (11966–025, Gibco) containing 2 mg/ml TRITC-Dextran (4.4 kDa; T1037, Sigma-Aldrich) were added into upper inserts in Control and OGD groups, respectively, and glucose deprived DMEM medium was added into every lower chamber. Subsequently, EVs from two sources of cells were diluted in glucose deprived culture medium and added to upper chambers in treatment groups, while equal volume of culture medium was added to normalize the chamber volume in other groups. After OGD 4 h, 50 μl medium of each upper and lower chambers was collected, and TRITC-dextran fluorescence intensity was read by microplate reader (SpectraMax M5) at wavelength of excitation: 550 nm and emission: 572 nm. The permeability coefficient was calculated by the following method as previously reported [30]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{P}}_{{{\text{dextran}}}} \, = \,\left({{\text{RFU}}_{{\text{lower chamber}}} /{\text{ RFU}}_{{\text{upper insert}}} } \right) \, \left({{1 }/{\text{ S}}} \right) \, \left({\text{V}} \right) \, ({1 }/{\text{ t}}).$$\end{document}Pdextran=RFUlower chamber/RFUupper insert1/SV(1/t). “RFU” was the fluorescent intensity of upper insert and lower chamber, and “S” indicated the surface area of cell monolayer, while “V” was the volume of lower chamber and “t” represented the time TRITC-dextran spread. ## Western blotting Total protein extracts of cell samples, BEC-EVs and BMSC-EVs, and isolated brain microvessels were collected, and ProteoExtract Subcellular Proteome Extraction Kit (539790, Calbiochem) was used to extract subcellular protein components in cytosolic fraction (CF), membrane fraction (MF), and actin cytoskeletal fraction (ACF) followed manufacture’s protocol. Samples with equal protein concentration were boiled, electrophoresis separated by $10\%$ sodium dodecyl sulfate (SDS) polyacrylamide gels and transferred into 0.22 μm pore sized polyvinylidene difluoride (PVDF) membranes (1620177, Bio-Rad). After blocking in $5\%$ skimmed milk, membranes were incubated with primary and secondary antibodies, and then washed by TBS-T (Tris-buffered saline with $0.1\%$ Tween 20). ChemiDoc MP Imaging System (Bio-Rad) were applied for imaging. Bands grey value was calculated by ImageJ software and relative expression level of proteins was presented as ratio of β-actin or Fractions REF (Calpain I, Calnexin, Vimentin). Antibodies used in this study were as follows: ZO-1 (1:1000, 61–7300, Invitrogen); Claudin-5 (1:1000, 34–1600, Invitrogen); CD 31 (1:1000, ab281583, Abcam); NeuN (1:5000, ab104225, Abcam); β-actin (1:5000, ab8227,Abcam); Caveolin-1 (1:1000, ab2910, Abcam); TSG 101 (1:1000, ab125011, Abcam); HSP 70 (1:1000, ab137680, Abcam); Calpain I (1:1000, ab108400, Abcam); Calnexin (1: 1000, ab22595, Abcam); Vimentin (1:1000, ab92547, Abcam); Goat Anti-Rabbit IgG H&L (HRP) (1:3000, ab6721, Abcam); Goat Anti-Mouse IgG H&L (HRP) (1:3000, ab67879, Abcam); Normal rabbit IgG (1 μg/ml, 2729S, Cell Signaling Technology). ## Co-immunoprecipitation (Co-IP) assay Cell samples were lysed by lysis buffer (P0013, Beyotime) with proteinase inhibitor cocktail, and then total protein was extracted, and concentration was evaluated as the protocol of western blotting. Protein G-Magnetic Beads (HY-K0204, MCE) were conjugated with Anti-Cav-1 antibodies or Anti-Normal rabbit IgG for 2 h under 4 ℃. After magnetic separation and washed by PBS-T ($0.5\%$ Tween-20 in PBS, pH 7.4) for 4 times, and 300 μg cell total proteins were added to beads and incubated overnight at 4 ℃. Excessive antigens were washed by PBS-T, and conjugated antigens were eluted by heating under 95 ℃ for 5 min with loading buffer and separated by electrophoresis as western blotting. ## Immunofluorescence staining For immunofluorescence samples preparation, rats were sacrificed, and then, 50 ml cold PBS and 50 ml cold $4\%$ paraformaldehyde (PFA) were intraventricularly perfused slowly. Whole brains were collected, fixed by $4\%$ PFA and dehydrated in $15\%$ and $30\%$ sucrose solution. 10 μm brain cryosections were cut by microtome (CryoStar NX70, Thermo Fisher Scientific). Brain microvessels were resuspended in PBS and dropped at glass slides after isolation, and samples were prepared after air dry. Cell samples were prepared in confocal dish after treatment. All samples were washed by PBS, fixed by $4\%$ PFA and permeabilized by $0.1\%$ Triton X-100. Primary and secondary antibodies (Alexa Fluor® 488 or 594 of rabbit or mouse, 1:500, ab150113; ab150080; ab150077, Abcam) were incubated with samples, and then DAPI was stained for nucleus before sections were sealed to be imaged by confocal (Lecia, SP8) or fluorescent microscope (Lecia, DMi8) under 63 ×, 10 ×, and 40 × objectives for cells, brain slices and microvessels, respectively. Images were processed by the microscopy software LAS X (Lecia), and fluorescence intensity was measured by ImageJ software. The following primary antibodies were used: ZO-1 (1:200, 61–7300, Invitrogen); Claudin-5 (1:150, 34–1600, Invitrogen), Caveolin-1 (1:250, ab2910, Abcam); CD31 (1:200, ab64543, Abcam). ## Cav-1 siRNA and pcDNA 3.1 vector transfection Cav-1 siRNA/pcDNA 3.1 and FAM-negative control siRNA/pcDNA 3.1-GFP (GenePhrama) were transfected by utilizing lipofectamine 3000 reagent (L3000015, Invitrogen) according to the manufacturer’s protocol. Cav-1 knockdown and overexpression were verified by western blotting and GFP/FAM labels imaging under 10 × objective by fluorescent microscope (Lecia, DMi8). Detailed pcDNA 3.1 vector gene map has been provided in (Additional file 1).siRNA sequences applied: Cav-1 (sense 5'-3'): CUGCGAUCCACUCUUUGAATT. Cav-1 (antisense 5'-3'): UUCAAAGAGUGGAUCGCAGTT. Negative Control siRNA oligo (sense 5'-3'): UUCUCCGAACGUGUCACGUTT. Negative Control siRNA oligo (antisense 5'-3'): ACGUGACACGUUCGGAGAATT. Cav-1 gene information applied in pcDNA 3.1 vector: NCBI Gene ID:12389. ## Transmission electron microscopy TEM was employed to observe the morphology of the EVs and the rat brain microvessel microstructure. Briefly, for EVs samples, 20 μl of EVs suspension was dropped onto the 150 meshes carbon filmed copper grid for 5 min. Thereafter, $2\%$ phosphotungstic acid was dropped on the copper grid to stain for 2 min. The samples were observed under TEM (HT7800, Hitachi) at the 40 k × magnification. For rat microvessel, 1 mm3 rat cortex brain tissues were harvested and fixed in $2.5\%$ glutaraldehyde solution and $1\%$ OsO4. Afterward, samples were dehydrated by using $30\%$-$95\%$ ethanol, and then subjected to resin penetration and embedding. The embedded samples were cut into 60 nm section by using the ultra-microtome (Leica UC7, Leica). Tissue samples were fished out onto the 150 meshes formvar filmed cuprum grids and stained by $2\%$ uranium acetate saturated alcohol solution for 8 min. After rinsed in $70\%$ ethanol, and ultra-pure water for 3 times, respectively, samples were then stained by $2.6\%$ lead citrate for 8 min, followed by rinsed in ultra-pure water 3 times. Sections were dried overnight and observed under TEM with the magnification of 2 k × or 20 k ×. ImageJ software was used to quantify the caveolae structure density. ## Statistical analysis All data of this study were collected and analyzed by GraphPad Prism 8 software and presented as mean ± standard deviation (SD). Statistical significance was considered when p value < 0.05. Additionally, two-way ANOVA was performed in experiments which involve two variations. All data were normalized and calculated as fold of Control group. ## Characteristics of BMSC-EVs and BEC-EVs To identify the isolated EVs, the size distribution, morphology, and the phagocytized ability were characterized. The average sizes of BMSC-EVs and BEC-EVs were 145 nm and 142 nm as determined by NTA (Fig. 1A), whereas negative control sample showed no typical EVs distribution, and had extremely low nanoparticle tracks (less than 200 tracks) (Additional file 2: Fig. S1A) and EVs surface markers TSG 101, HSP 70, Alix, CD 9 and CD 63 were exclusively expressed in EVs from two sources of cells whereas were not detected in parent cells and supernatant after EVs isolation (Fig. 1B). TEM images indicated EVs had an intact sphere structure (Fig. 1C). In contrast, negative control samples had no obvious nanoparticle feedback (Additional file 2: Fig. S1B). Notably, DiI labeled BMSC-EVs and BEC-EVs could be phagocytized by b. End3 cells (Fig. 1D), and DiI labeled negative control nearly failed to be observed in b. End3 cells, which indicated no residue of excessive DiI dye (Additional file 2: Fig. S1C). Additionally, DiI labeled BMSC-EVs and BEC-EVs were internalized by b. End3 cell over time under 37 ℃, however, the cells failed to phagocytize DiI labeled EVs at any time point within 120 min under 4 ℃, which indicated the EVs intracellular internalization was affected by non-physiological ambient temperature (Additional file 2: Fig. S1D). Collectively, the results indicated the isolated EVs preserved ideal size, morphology, and the ability of intracellular internalization. Fig. 1Characterization of BEC-EVs and BMSC-EVs. A NTA of BEC-EVs and BMSC-EVs showed the size of EVs. B Representative western blotting of EVs marker HSP 70, TSG 101, ALIX, CD9, and CD63. Cells and supernatant without EVs were regarded as negative control, respectively, and calnexin was defined as endoplasmic reticulum marker which presented exclusively in cells. C Representative TEM images of BEC-EVs and BMSC-EVs (Black arrowheads). Scale Bar: 200 nm. Magnification: 40 k ×. All images were taken in the same scale. D Representative confocal microscope images of b. End3 cells phagocytizing DiI labeled BEC-EVs and BMSC-EVs. Scale Bar: 50 μm. Magnification: 63 ×. All images were taken in the same scale ## BMSC-EVs and BEC-EVs treatments attenuated BECs hyperpermeability after OGD insult To evaluate the in vitro efficacy of BBB integrity exerted by two kinds of EVs, OGD stimulated b. End3 cells were treated with BMSC-EVs and BEC-EVs. The morphology of b. End3 cells gradually presented injury with soma shrinkage and rupture, after OGD stimulation for 4 or 6 h (Fig. 2A). Consistent with the alteration of cell morphology, cell viability was decreased by about $20\%$ to $60\%$ after 4 to 6 h OGD stimulation (Fig. 2B). As cells morphology was extremely upset under 6 h OGD insult, we chose 4 h OGD for the following studies. Then, to determine the optimal dosages of EVs from two sources of cells, low, medium, and high doses of BEC-EVs and BMSC-EVs were applied to evaluate cell viability. The results suggested three dosages of BEC-EVs presented cell viability reservation by about $30\%$ whereas three dosages of BMSC-EVs recovered cell viability by about $20\%$ against hypoxic insult (Fig. 2C), and no significant difference existed between three doses of BMSC-EVs and BEC-EVs groups (low dosage: ~ 5 × 109 EVs; middle dosage: ~ 1 × 1010 BEC-EVs; high dosage: ~ 2 × 1010 EVs). Medium dosage of EVs was determined for the following experiments to achieve stable therapeutic outcomes. In addition, the leakage of monolayer was dramatically increased after OGD insult by transwell assay (Fig. 2D, E), and the hyperpermeability of OGD insulted cell monolayer was attenuated about $30\%$ by BMSC-EVs and BEC-EVs administrations (Fig. 2E).Fig. 2BEC-EVs and BMSC-EVs treatments reduced the leakage and enhanced the expressions of ZO-1 and Claudin-5 in OGD insulted b. End3 cells. A Representative images of b. End3 cell morphology after 4 and 6 h OGD insult. Scale Bar: 200 μm. Magnification: 20 ×. All images were taken in the same scale. B The viability analysis of OGD insulted b. End3 cells after BEC-EVs and BMSC-EVs treatments ($$n = 6$$). C Viability analysis of b. End3 subjected to OGD after different dosages of BEC-EVs and BMSC-EVs administrations ($$n = 6$$). D Schematic of transwell insert for the evaluation of TRITC-Dextran leakage after the treatments of EVs from two sources. E Relative permeability coefficient of OGD insulted b. End3 cells after BEC-EVs and BMSC-EVs treatments ($$n = 4$$). F Representative immunofluorescent staining images of ZO-1 and Claudin-5 in OGD insulted b. End3 cells after BEC-EVs and BMSC-EVs treatments, and quantification of mean fluorescent intensity. Scale Bar: 50 μm. Magnification: 63 ×. All images were taken in the same scale. * $P \leq 0.001$ vs. Control group; ###$P \leq 0.001$, ##$P \leq 0.01$, #$P \leq 0.05$ vs. OGD or OGD 4 h group ## BMSC-EVs and BEC-EVs treatments enhanced ZO-1 and Claudin-5 expressions and inhibited their redistribution in OGD insulted BECs We next assessed the EVs’ therapeutic effects on the expression and dislocation of tight junction proteins in b. End3 cells after OGD insult. Immunofluorescence staining of ZO-1 and Claudin-5 showed that BMSC-EVs and BEC-EVs treatments had similar efficacies to increase their fluorescence intensity in OGD insulted b. End3 cells (Fig. 2F). Western blotting results also suggested the similar trend by the treatments of EVs from two sources, especially by BMSC-EVs treatment (Fig. 3A). Additionally, subcellular membrane, cytoplasm, and actin cytoskeleton fractions proteins from b. End3 cells were isolated, and the subcellular fraction markers Calpain I, Calnexin, and Vimentin were dominantly expressed in corresponded fractions (Fig. 3B), suggesting the extraction purity was reliable. The expressions of ZO-1 and Claudin-5 in membrane fraction were decreased after OGD insult, which were significantly reversed by BMSC-EVs and BEC-EVs treatments. Simultaneously, the expressions of ZO-1 and Claudin-5 were dramatically increased in cytoplasm whereas the treatments of EVs from two sources, especially BMSC-EVs treatment reversed this trend (Fig. 3C–E). In actin cytoskeleton, only Claudin-5 expression were downregulated by BMSC-EVs treatment (Fig. 3E). Altogether, the results suggested BMSC-EVs treatment more efficiently antagonized the redistribution of ZO-1 and Claudin-5 from cellular membrane to cytoplasm and actin cytoskeleton after OGD stimulation. Fig. 3BEC-EVs and BMSC-EVs treatments recovered ZO-1 and Claudin-5 expressions and inhibited their intracellular translocation in OGD insulted b. End3 cells. A Representative western blotting of ZO-1 and Claudin-5 after the treatments of EVs from two sources and quantification of ZO-1 and Claudin-5 expressions ($$n = 3$$). B The purity determination in subcellular fractions by western blotting. C Representative western blotting of ZO-1 and Claudin-5 in subcellular fractions. MF Membrane fraction; CF Cytoplasm fraction; ACF Actin cytoskeleton fraction; Fractions REF. ( Calpain I for AF, Calnexin for MF and Vimentin for ACF, respectively). D, E Quantification of ZO-1 and Claudin-5 expressions in subcellular fractions ($$n = 3$$). * $P \leq 0.001$, *$P \leq 0.01$, $P \leq 0.05$ vs. Control group; ###$P \leq 0.001$, ##$P \leq 0.01$, #$P \leq 0.05$ vs. OGD group; †††$P \leq 0.001$, ††$P \leq 0.01$, †$P \leq 0.05$ vs. OGD + BEC-EVs group ## BMSC-EVs and BEC-EVs treatments increased ZO-1 and Claudin-5 expressions in isolated cerebral microvessels Furthermore, brain microvessels (BMV) from rat ischemic cortex were isolated. BMV from normal cortex was identified by western blotting and immunofluorescence staining, as the results presented, NeuN (a neuron marker) was highly expressed in brain total extract (BTE), while it was almost absent in BMV (Additional file 3: Fig. S2A). Moreover, the vascular marker CD 31 mainly expressed in BMV by western blotting and immunofluorescence staining, suggesting the successful microvessels isolation (Additional file 3: Fig. S2A-B). After 24 h pMCAo, the fluorescence intensity of Claudin-5 and ZO-1 presented weak and unclear in BMV from ischemic cortex post-stroke, while partly salvaged by BMSC-EVs and BEC-EVs treatments (Fig. 4A). To quantify such differences, western blotting of BMV was performed, and Claudin-5 and ZO-1 expressions were extremely lowered after pMCAo, however, the treatments of EVs from two sources of cells reversed this trend. Moreover, the expression of total protein of Claudin-5 in BMV in BEC-EVs group was more significant when compared with that in BMSC-EVs group (Fig. 4B).Fig. 4BEC-EVs and BMSC-EVs treatments enhanced ZO-1 and Claudin-5 expressions in BMV from ischemic cortex. A Representative immunofluorescent staining of Claudin-5, ZO-1 and CD31 in BMV. Scale Bar: 50 μm. Magnification: 40 ×. All images were taken in the same scale. B Representative western blotting of Claudin-5 and ZO-1 in BMV and quantification of ZO-1 and Claudin-5 expressions ($$n = 3$$). * $P \leq 0.001$ vs. SHAM group; ###$P \leq 0.001$, ##$P \leq 0.01$, #$P \leq 0.05$ vs. pMCAo group; ††$P \leq 0.01$ vs. pMCAo + BMSC-EVs group ## BMSC-EVs and BEC-EVs treatments rescued pMCAo rats against ischemic injury and BBB leakage Next, we further evaluated the therapeutic efficacy of EVs in vivo, and administered two kinds of EVs to pMCAo rats. After the administrations of BMSC-EVs and BEC-EVs for 24 h, respectively, cerebral infarct volume, neurological function and BBB permeability were evaluated. It displayed similar effects on the reduction of infarct volume and EB leakage in ischemic ipsilateral hemisphere (Fig. 5A, B). Deficient neurological function evaluated by mNSS and Bederson tests was partly ameliorated by the administrations of EVs from two sources, and particularly, the efficacy of BMSC-EVs treatment was superior to that of BEC-EVs treatment (Fig. 5C). In addition, the fluorescence intensity of ZO-1 and Claudin-5 in infarct border zone was observed according to previous report [31]. Briefly, we made a vertical cut at about 2 mm from the midline in ischemic hemisphere, and then made a cut at 60° from midline to separate ischemic core from infarct border zone (Fig. 5D). The intensity of ZO-1 and Claudin-5 was dramatically decreased after pMCAo, which was restored by the treatments of EVs from two sources (Fig. 5E), and the improved trend was similar to that in BMV (Fig. 4A).Fig. 5BEC-EVs and BMSC-EVs treatments improved neurological injury, BBB leakage and ZO-1 and Claudin-5 expressions after acute IS. A Representative TTC staining images of total brain slices from pMCAo rats and relative infarction volume quantification ($$n = 8$$). B Representative images of EB leakage determined by IVIS and quantification of fluorescent efficacy ($$n = 8$$). C The evaluation of neurological function by mNSS and Bederson tests ($$n = 10$$). D Schematic of cortex infarct border area observed in immunofluorescent staining. E Representative immunofluorescent staining and expression quantification of ZO-1 and Claudin-5 in cortex infarct border area ($$n = 3$$). Scale Bar: 200 μm. Magnification: 10 ×. All images were taken in the same scale. * $P \leq 0.001$ vs. SHAM group; ###$P \leq 0.001$, ##$P \leq 0.01$, #$P \leq 0.05$ vs. pMCAo group; †††$P \leq 0.001$, ††$P \leq 0.01$, vs. pMCAo + BEC-EVs ## BMSC-EVs and BEC-EVs treatments antagonized Cav-1-dependent ZO-1 and Claudin-5 endocytosis To investigate the pros and cons of Cav-1 on BBB permeability, we focused on the relationship between Cav-1 and TJ proteins. Results showed the expression of Cav-1 in OGD insulted b. End3 cells and BMV from ischemic cortex was obviously upregulated, while this trend was reversed by BEC-EVs and BMSC-EVs treatments (Fig. 6A), and TEM image revealed that the density of caveolae-like vesicles in pMCAo rats’ brain microvascular endothelial cells were notably accumulated but dwindled by two kinds of EVs therapies (Fig. 6B). Consistent with immunofluorescence staining result in BMV, western blotting of the increased Cav-1 in BMV of pMCAo model was downregulated in EVs administration groups (Fig. 7A). Interestingly, the decreased expressions of ZO-1 and Claudin-5 in OGD insulted b. End3 cells were recovered by Cav-1 siRNA transfection (Fig. 7B). Additionally, it also indicated Cav-1 knockdown, and overexpression were successfully performed in b. End3 cells (Additional file 4: Fig. S3A-B), and attenuating Cav-1 expression by siRNA contributed to the decrease of BEC monolayer permeability, whereas Cav-1 overexpression by Cav-1 pcDNA 3.1 increased the monolayer leakage (Fig. 7C). Furthermore, Cav-1 expression was abundant in MF and ACF, and particularly upregulated in CF after OGD insult, but in a significant manner, it was downregulated by the treatments of EVs from two cell sources, especially by BMSC-EVs treatment (Fig. 7D). Co-IP experiment revealed that the colocalization of Cav-1 with ZO-1 and Claudin-5 was enhanced in b. End3 cells by Cav-1 overexpression (Fig. 7E), and the results showed that Cav-1 interacted closely with ZO-1 and Claudin-5. Finally, to prove the relationship of Cav-1 with EVs therapeutic efficacy, we administered EVs from two sources into Cav-1-overexpressed b. End3 cells insulted by OGD, and as expected, the decreased cell monolayer leakage by BMSC-EVs and BEC-EVs treatments was partly muted in Cav-1 overexpression groups (Fig. 7F). In summary, the results indicated Cav-1 mediated ZO-1 and Claudin-5 endocytosis after OGD insult, which contributed to endothelial hyperpermeability. Fig. 6Distribution and expression of Caveolin-1 and Caveolae in brain microvessels. A Representative immunofluorescence staining images and quantification of Caveolin-1 and CD 31 in BMV ($$n = 3$$). Scale Bar: 50 μm. Magnification: 40 ×. All images were taken in the same scale. B TEM image of caveolae-like vesicles in brain microvascular endothelial cells and quantification of the density. Scale Bar: 2 μm and 200 nm. Magnification: 2 k × and 20 k ×. All images were taken in the same scale. * $P \leq 0.001$ vs. SHAM group; ###$P \leq 0.001$, ##$P \leq 0.05$, #$P \leq 0.01$ vs. pMCAo groupFig. 7BEC-EVs and BMSC-EVs treatments antagonized Cav-1-dependent ZO-1 and Claudin-5 endocytosis to decrease BBB permeability. A Representative western blotting of Cav-1 in OGD insulted b. End3 cells and BMV from ischemic cortex; and quantification of Cav-1 expression ($$n = 3$$). B Representative western blotting of ZO-1, Claudin-5 and Cav-1 after Cav-1 siRNA transfection in OGD insulted b. End3 cells and quantification of ZO-1, Claudin-5 and Cav-1 expressions ($$n = 3$$). C Relative permeability coefficient of OGD insulted b. End3 cells after siRNA/pcDNA 3.1 transfection ($$n = 4$$). D Representative western blotting of Cav-1 in subcellular fractions. MF Membrane fraction; CF Cytoplasm fraction; ACF Actin cytoskeleton fraction; Fractions REF. ( Calpain I for AF, Calnexin for MF and Vimentin for ACF, respectively) and quantification of Cav-1 expression in subcellular fractions ($$n = 3$$). E Representative co-ip results of Cav-1 with ZO-1 and Claudin-5 after Cav-1 pcDNA 3.1 transfection in b. End3 cells, and IgG was served as negative control. IP: Immunoprecipitation; IB: Immunoblotting. F Relative permeability coefficient of OGD insulted b. End3 cells with pcDNA 3.1 transfection after the treatments of EVs from two sources ($$n = 4$$). * $P \leq 0.001$, $P \leq 0.01$ vs. Control/SHAM group; ###$P \leq 0.001$, #$P \leq 0.05$ vs. pMCAo/OGD group; †††$P \leq 0.0001$, vs. pMCAo + BEC-EVs. §§§ $P \leq 0.001$, §§$P \leq 0.01$ vs. Ctrl-siRNA/pcDNA 3.1 group. ‡‡‡$P \leq 0.001$ vs. OGD + pcDNA 3.1 + BEC-EVs group; ¶¶¶$P \leq 0.001$ vs. OGD + pcDNA 3.1 + BMSC-EVs group ## Discussion In this study, our results demonstrated the adverse role of Cav-1 in BBB permeability in acute IS, which was associated with Cav-1-dependent TJ proteins endocytosis from endothelial cell membrane to cytoplasm. Both BMSC-EVs and BEC-EVs treatments decreased BBB leakage and infarction volume, as well as improved neurological function. By comprehensive comparative analysis, the efficacies of BMSC-EVs treatment on neurological functional amelioration and antagonizing Cav-1-denpendent ZO-1 and Claudin-5 endocytosis are superior to BEC-EVs treatment. Recently, increasing attentions were attracted on the therapeutic paracrine factors of stem cells rather than themselves [32, 33]. EVs can be released by nearly all kind of cells, and based on Minimal Information for Studies of Extracellular Vesicles 2018 (MISEV2018) [34], our isolated EVs can be considered as small EVs (sEVs) which has a smaller size less than 200 nm. As one of the recognized cells paracrine factors, with protein, RNA and DNA cargos packed by their parent cells, EVs were reported to mediate multiple biological functions and penetrate BBB to improve neurodegeneration/cardiovascular recovery [35]. Therefore, EVs have ridden the new wave of cell-free therapeutic with similar efficacy and tend to be safer than their parent cells [32, 36]. The treatment of EVs derived from BMSCs has been extensively reported to achieve good effects on IS via multiple pathways [24, 37–40], and the treatment of EVs derived from BECs also exhibited satisfying efficacy on experimental IS [22, 41, 42]. However, few studies reported the discrepancy of therapeutic efficacies on BBB integrity between the two cells sources' EVs post-stroke. In our study, we compared the therapeutic efficacies of BMSC-EVs and BEC-EVs on acute IS in vitro and in vivo, and found that both kinds of EVs had similar abilities to attenuate hyperpermeability of b. End3 cell insulted by OGD, followed by restoring ZO-1 and Claudin-5 expressions and antagonizing their endocytosis abnormality. Moreover, BMSC-EVs treatment more definitely intervened with the redistribution of ZO-1 and Claudin-5 in subcellular fractions than BEC-EVs treatment did. In pMCAo rats, the efficacies of EVs from two sources on BBB leakage were the same as those in vitro. Notably, the therapeutic effect of BMSC-EVs on neurological function was superior to that of BEC-EVs. The potential reason may be attributed to the pluripotent characteristic of BMSCs, which enabled them to produce EVs to exert broad protective functions on diverse NVU cells as previously concluded [40], whereas BEC-EVs derived from BECs which compose an important structure of BBB, may specifically present vascular protective effects. It can be employed to explain why the expressions of ZO-1 and Claudin-5 in BEC-EVs group were higher than those in BMSC-EVs group in BMV. Cav-1 is a critical scaffolding/regulatory protein of Caveolae in lipid raft and mediates pathophysiological processes including Cav-1-dependent endocytosis [43]. However, its role in regulating of BBB permeability after IS always under controversy in recent years [11, 14]. In rat cortical cold injury model and cerebral I/R model, Cav-1 upregulation occurred before TJs breakdown, and in early stage of ischemia, BBB leakage resulted mainly from Cav-1-dependent transcytosis [10, 44], suggesting Cav-1 is a potential therapeutic target for BBB integrity. Additionally, Cav-1 was also reported to alter the subcellular distribution of TJ proteins. Liu and colleagues found that Cav-1 mediated the redistribution of Claudin-5, which contributed to BBB permeability in early IS phase [13]. Furthermore, their results demonstrated that Claudin-5 was finally degraded by autophagy via the interaction of NO with Cav-1 in OGD-insulted ECs [45]. Similarly, Cav-1, MMP-$\frac{2}{9}$ and autophagy-lysosome involved ZO-1 intracellular translocation and degradation resulted in BBB hyperpermeability in rats with I/R [46]. Although above results suggested Cav-1 plays an important role in ZO-1 and Claudin-5 redistribution and degradation post-stroke, whether Cav-1 directly or indirectly affects ZO-1 and Claudin-5 is still ambiguous, especially in permanent stroke model. Currently, our results demonstrated Cav-1 expression was significantly upregulated in OGD-insulted b. End3 cells and BMV from ischemic cortex, and it accumulated in cellular cytoplasm fraction after OGD stimulation. Moreover, by Cav-1 knockdown/overexpression and Co-IP assays, we defined a Cav-1-dependent endocytic pathway for ZO-1 and Claudin-5. Combing with the significant inhibition of Cav-1 in BMV from ischemic cortex in BMSC-EVs group, it suggested that BMSC-EVs treatment had more powerful ability to antagonize Cav-1-dependent ZO-1 and Claudin-5 endocytosis compared with BEC-EVs treatment. However, our results showed that there were no statistical significances in the attenuation of BBB leakage in vivo between BMSC-EVs and BEC-EVs treatments. We hypothesize BEC-EVs treatment regulated other Cav-1 independent approaches to supplement TJ proteins on cellular membrane (e.g., the suppression of MMP$\frac{2}{9}$, NO generation inhibition and endothelial protection and so on). In present study, we failed to further address “the fate” of the endocytosed ZO-1 and Claudin-5, and based on previous report, autophagy can be the following pathway that is responsible for the degradation of TJ proteins [45, 46]. In future works, the mechanism of how EVs adjust autophagy-lysosome dependent ZO-1 and Claudin-5 degradation should be further explored. Additionally, we also failed to explore what components of EVs could intervene with Cav-1 mediated ZO-1 and Claudin-5 translocation or disruption, and we hypothesize that the potential effects were related to microRNAs (miRNAs) loaded in EVs. miRNAs are small stem-ring structured no-coding RNA, which can be packed in EVs by their parent cells to mediate cell-to-cell communication and multiple biological functions [47]. Therefore, studies had verified some exosomal miRNAs were the key to IS recovery such as miR-126 in endothelial-EVs and miR-17–92 in BMSC-EVs [23]. Further studies should apply EVs RNA sequencing to identify potential miRNA within EVs to illustrate the related mechanisms. There are additional some limitations in our current study: Firstly, our EVs isolation protocol can only preserve small EVs (< 220 nm) by using 0.22 μm filters, however, those larger EVs may also exert unknown effects on BBB and IS recovery. Further studies are required to analyze the biological function of those large EVs in detail. In addition, our isolation procedure by Exo-Prep kit may influence the purity of collected EVs. Although at present there is still no perfect method for achieving EVs with high recovery and specificity, the second-step purification methods (e.g. ultracentrifugation and gradient centrifugation) may be optimal to achieve EVs with high purity and specificity in future studies. ## Conclusions In summary, we firstly defined high expression of Cav-1 aggravate BBB permeability in early rat pMCAo model, which is related to Cav-1-dependent ZO-1 and Claudin-5 endocytosis. By comparative analysis, our results demonstrated both of BMSC-EVs and BEC-EVs treatments antagonized Cav-1 endocytic pathway for the maintenance of BBB integrity, and the overall therapeutic efficacy of BMSC-EVs was superior to BEC-EVs in acute IS treatment. ## Supplementary Information Additional file 1: Supplemental methods and information of neurological function tests, pcDNA 3.1 vector gene map. Additional file 2: Fig. S1. Characterization of negative control of EVs isolation. ( A) Size distribution of negative control sample. ( B) Representative low magnificent TEM images of negative control sample. Scale Bar: 1 μm. Magnification: 4k ×. (C) Representative confocal images of b. End3 cells uptake DiI labeled negative control sample. Scale Bar: 20 μm. Magnification: 63 × with 2.5 × zoom in. ( D) Representative confocal images and fluorescence quantification of DiI labeled BEC and BMSC-EVs phagocytized by b. End3 cells in different ambient temperatures ($$n = 3$$). $P \leq 0.0001$, $P \leq 0.001$ vs. 4 ℃ group. Scale Bar: 20 μm. Magnification: 63 × with 2.5 × zoom in. Additional file 3: Fig. S2. Characterization of isolated BMV. ( A) Representative western blotting of NeuN, CD 31 in BMV. ( B) Representative immunofluorescence staining images of CD 31 in BMV. Scale Bar: 200 μm. Magnification: 10 ×.Additional file 4: Fig. S3. Verification for successful transfection of siRNA and pcDNA 3.1. ( A) Representative western blotting and quantification of Cav-1 in b. End3 cells transfected with Ctrl/Cav-1 siRNA and Cav-1 DNA/ pcDNA 3.1 ($$n = 3$$). $P \leq 0.001$, $P \leq 0.01$ vs. Ctrl-siRNA/pcDNA 3.1 group. ( B) Representative fluorescent microscope images of b. End3 transfected with FAM-negative control siRNA and pcDNA 3.1-GFP. Scale Bar: 200 μm. Magnification: 10 ×. ## References 1. Feigin VL, Nguyen G, Cercy K, Johnson CO, Alam T, Parmar PG. Global, regional, and country-specific lifetime risks of stroke, 1990 and 2016. N Engl J Med (2018) 379 2429-2437. DOI: 10.1056/NEJMoa1804492 2. Tsao CW, Aday AW, Almarzooq ZI, Alonso A, Beaton AZ, Bittencourt MS. Heart disease and stroke statistics-2022 update: a report from the american heart association. Circulation (2022) 145 e153-e639. DOI: 10.1161/CIR.0000000000001052 3. Emberson J, Lees KR, Lyden P, Blackwell L, Albers G, Bluhmki E. Effect of treatment delay, age, and stroke severity on the effects of intravenous thrombolysis with alteplase for acute ischaemic stroke: a meta-analysis of individual patient data from randomised trials. Lancet (2014) 384 1929-1935. DOI: 10.1016/s0140-6736(14)60584-5 4. Joo H, Wang G, George MG. Use of intravenous tissue plasminogen activator and hospital costs for patients with acute ischaemic stroke aged 18–64 years in the USA. Stroke Vasc Neurol (2016) 1 8-15. DOI: 10.1136/svn-2015-000002 5. Khandelwal P, Yavagal DR, Sacco RL. Acute ischemic stroke intervention. J Am Coll Cardiol (2016) 67 2631-2644. DOI: 10.1016/j.jacc.2016.03.555 6. Ganesh A, Goyal M. Thrombectomy for acute ischemic stroke: recent insights and future directions. Curr Neurol Neurosci Rep (2018) 18 59. DOI: 10.1007/s11910-018-0869-8 7. Shi L, Rocha M, Leak RK, Zhao J, Bhatia TN, Mu H. A new era for stroke therapy: integrating neurovascular protection with optimal reperfusion. J Cereb Blood Flow Metab (2018) 38 2073-2091. DOI: 10.1177/0271678x18798162 8. Keaney J, Campbell M. The dynamic blood-brain barrier. FEBS J (2015) 282 4067-4079. DOI: 10.1111/febs.13412 9. Jiang X, Andjelkovic AV, Zhu L, Yang T, Bennett MVL, Chen J. Blood-brain barrier dysfunction and recovery after ischemic stroke. Prog Neurobiol (2018) 163–164 144-171. DOI: 10.1016/j.pneurobio.2017.10.001 10. Knowland D, Arac A, Sekiguchi KJ, Hsu M, Lutz SE, Perrino J. Stepwise recruitment of transcellular and paracellular pathways underlies blood-brain barrier breakdown in stroke. Neuron (2014) 82 603-617. DOI: 10.1016/j.neuron.2014.03.003 11. Xu L, Guo R, Xie Y, Ma M, Ye R, Liu X. Caveolae: molecular insights and therapeutic targets for stroke. Expert Opin Ther Targets (2015) 19 633-650. DOI: 10.1517/14728222.2015.1009446 12. Gu Y, Zheng G, Xu M, Li Y, Chen X, Zhu W. Caveolin-1 regulates nitric oxide-mediated matrix metalloproteinases activity and blood-brain barrier permeability in focal cerebral ischemia and reperfusion injury. J Neurochem (2012) 120 147-156. DOI: 10.1111/j.1471-4159.2011.07542.x 13. Liu J, Jin X, Liu KJ, Liu W. Matrix metalloproteinase-2-mediated occludin degradation and caveolin-1-mediated claudin-5 redistribution contribute to blood-brain barrier damage in early ischemic stroke stage. J Neurosci (2012) 32 3044-3057. DOI: 10.1523/jneurosci.6409-11.2012 14. Huang Q, Zhong W, Hu Z, Tang X. A review of the role of cav-1 in neuropathology and neural recovery after ischemic stroke. J Neuroinflammation (2018) 15 348. DOI: 10.1186/s12974-018-1387-y 15. Rajendran L, Bali J, Barr MM, Court FA, Krämer-Albers EM, Picou F. Emerging roles of extracellular vesicles in the nervous system. J Neurosci (2014) 34 15482-15489. DOI: 10.1523/jneurosci.3258-14.2014 16. Patel MM, Patel BM. Crossing the blood-brain barrier: recent advances in drug delivery to the brain. CNS Drugs (2017) 31 109-133. DOI: 10.1007/s40263-016-0405-9 17. Li Y, Liu B, Chen Y, Quan X, Han Y, Zheng Y. Extracellular vesicle application as a novel therapeutic strategy for ischemic stroke. Transl Stroke Res (2021). DOI: 10.1007/s12975-021-00915-3 18. Trounson A, McDonald C. Stem cell therapies in clinical trials: progress and challenges. Cell Stem Cell (2015) 17 11-22. DOI: 10.1016/j.stem.2015.06.007 19. Boltze J, Arnold A, Walczak P, Jolkkonen J, Cui L, Wagner DC. The dark side of the force—constraints and complications of cell therapies for stroke. Front Neurol (2015) 6 155. DOI: 10.3389/fneur.2015.00155 20. Yue KY, Zhang PR, Zheng MH, Cao XL, Cao Y, Zhang YZ. Neurons can upregulate Cav-1 to increase intake of endothelial cells-derived extracellular vesicles that attenuate apoptosis via miR-1290. Cell Death Dis (2019) 10 869. DOI: 10.1038/s41419-019-2100-5 21. Xiao B, Chai Y, Lv S, Ye M, Wu M, Xie L. Endothelial cell-derived exosomes protect SH-SY5Y nerve cells against ischemia/reperfusion injury. Int J Mol Med (2017) 40 1201-1209. DOI: 10.3892/ijmm.2017.3106 22. Venkat P, Cui C, Chopp M, Zacharek A, Wang F, Landschoot-Ward J. MiR-126 mediates brain endothelial cell exosome treatment-induced neurorestorative effects after stroke in type 2 diabetes mellitus mice. Stroke (2019) 50 2865-2874. DOI: 10.1161/STROKEAHA.119.025371 23. Xin H, Liu Z, Buller B, Li Y, Golembieski W, Gan X. MiR-17-92 enriched exosomes derived from multipotent mesenchymal stromal cells enhance axon-myelin remodeling and motor electrophysiological recovery after stroke. J Cereb Blood Flow Metab (2021) 41 1131-1144. DOI: 10.1177/0271678x20950489 24. Zeng Q, Zhou Y, Liang D, He H, Liu X, Zhu R. Exosomes secreted from bone marrow mesenchymal stem cells attenuate oxygen-glucose deprivation/reoxygenation-induced pyroptosis in PC12 cells by promoting AMPK-dependent autophagic flux. Front Cell Neurosci (2020) 14 182. DOI: 10.3389/fncel.2020.00182 25. Xiao Y, Geng F, Wang G, Li X, Zhu J, Zhu W. Bone marrow-derived mesenchymal stem cells-derived exosomes prevent oligodendrocyte apoptosis through exosomal miR-134 by targeting caspase-8. J Cell Biochem (2018). DOI: 10.1002/jcb.27519 26. Tian Y, Gong M, Hu Y, Liu H, Zhang W, Zhang M. Quality and efficiency assessment of six extracellular vesicle isolation methods by nano-flow cytometry. J Extracell Vesicles (2020) 9 1697028. DOI: 10.1080/20013078.2019.1697028 27. Lee YK, Uchida H, Smith H, Ito A, Sanchez T. The isolation and molecular characterization of cerebral microvessels. Nat Protoc (2019) 14 3059-3081. DOI: 10.1038/s41596-019-0212-0 28. Bieber M, Gronewold J, Scharf AC, Schuhmann MK, Langhauser F, Hopp S. Validity and reliability of neurological scores in mice exposed to middle cerebral artery occlusion. Stroke (2019) 50 2875-2882. DOI: 10.1161/STROKEAHA.119.026652 29. Chen J, Li Y, Wang L, Zhang Z, Lu D, Lu M. Therapeutic benefit of intravenous administration of bone marrow stromal cells after cerebral ischemia in rats. Stroke (2001) 32 1005-1011. DOI: 10.1161/01.str.32.4.1005 30. McMillin MA, Frampton GA, Seiwell AP, Patel NS, Jacobs AN, DeMorrow S. TGFβ1 exacerbates blood-brain barrier permeability in a mouse model of hepatic encephalopathy via upregulation of MMP9 and downregulation of claudin-5. Lab Invest (2015) 95 903-913. DOI: 10.1038/labinvest.2015.70 31. Ashwal S, Tone B, Tian HR, Cole DJ, Pearce WJ. Core and penumbral nitric oxide synthase activity during cerebral ischemia and reperfusion. Stroke (1998) 29 1037-46. DOI: 10.1161/01.STR.29.5.1037 32. Eleuteri S, Fierabracci A. Insights into the secretome of mesenchymal stem cells and its potential applications. Int J Mol Sci. (2019). DOI: 10.3390/ijms20184597 33. Cai Y, Liu W, Lian L, Xu Y, Bai X, Xu S. Stroke treatment: is exosome therapy superior to stem cell therapy?. Biochimie (2020). DOI: 10.1016/j.biochi.2020.09.025 34. Théry C, Witwer KW, Aikawa E, Alcaraz MJ, Anderson JD, Andriantsitohaina R. Minimal information for studies of extracellular vesicles 2018 (MISEV2018): a position statement of the international society for extracellular vesicles and update of the MISEV2014 guidelines. J Extracell Vesicles (2018) 7 1535750. DOI: 10.1080/20013078.2018.1535750 35. Kalluri R, LeBleu VS. The biology, function, and biomedical applications of exosomes. Science (2020). DOI: 10.1126/science.aau6977 36. Baglio SR, Pegtel DM, Baldini N. Mesenchymal stem cell secreted vesicles provide novel opportunities in (stem) cell-free therapy. Front Physiol (2012) 3 359. DOI: 10.3389/fphys.2012.00359 37. Li Y, Cheng Q, Hu G, Deng T, Wang Q, Zhou J. Extracellular vesicles in mesenchymal stromal cells: a novel therapeutic strategy for stroke. Exp Ther Med (2018) 15 4067-4079. DOI: 10.3892/etm.2018.5993 38. Wang C, Borger V, Sardari M, Murke F, Skuljec J, Pul R. Mesenchymal stromal cell-derived small extracellular vesicles induce ischemic neuroprotection by modulating leukocytes and specifically neutrophils. Stroke (2020). DOI: 10.1161/STROKEAHA.119.028012 39. Xin H, Li Y, Cui Y, Yang JJ, Zhang ZG, Chopp M. Systemic administration of exosomes released from mesenchymal stromal cells promote functional recovery and neurovascular plasticity after stroke in rats. J Cereb Blood Flow Metab (2013) 33 1711-1715. DOI: 10.1038/jcbfm.2013.152 40. Rahmani A, Saleki K, Javanmehr N, Khodaparast J, Saadat P, Nouri HR. Mesenchymal stem cell-derived extracellular vesicle-based therapies protect against coupled degeneration of the central nervous and vascular systems in stroke. Ageing Res Rev (2020) 62 101106. DOI: 10.1016/j.arr.2020.101106 41. Chen J, Cui C, Yang X, Xu J, Venkat P, Zacharek A. MiR-126 affects brain-heart interaction after cerebral ischemic stroke. Transl Stroke Res (2017) 8 374-385. DOI: 10.1007/s12975-017-0520-z 42. Wang J, Chen S, Zhang W, Chen Y, Bihl JC. Exosomes from miRNA-126-modified endothelial progenitor cells alleviate brain injury and promote functional recovery after stroke. CNS Neurosci Ther (2020) 26 1255-1265. DOI: 10.1111/cns.13455 43. Parton RG. Caveolae: structure, function, and relationship to disease. Annu Rev Cell Dev Biol (2018) 34 111-136. DOI: 10.1146/annurev-cellbio-100617-062737 44. Nag S, Venugopalan R, Stewart DJ. Increased caveolin-1 expression precedes decreased expression of occludin and claudin-5 during blood-brain barrier breakdown. Acta Neuropathol (2007) 114 459-469. DOI: 10.1007/s00401-007-0274-x 45. Liu J, Weaver J, Jin X, Zhang Y, Xu J, Liu KJ. Nitric oxide interacts with caveolin-1 to facilitate autophagy-lysosome-mediated claudin-5 degradation in oxygen-glucose deprivation-treated endothelial cells. Mol Neurobiol (2016) 53 5935-5947. DOI: 10.1007/s12035-015-9504-8 46. Zhang S, An Q, Wang T, Gao S, Zhou G. Autophagy- and MMP-2/9-mediated reduction and redistribution of ZO-1 contribute to hyperglycemia-increased blood-brain barrier permeability during early reperfusion in stroke. Neuroscience (2018) 377 126-137. DOI: 10.1016/j.neuroscience.2018.02.035 47. Zhang J, Li S, Li L, Li M, Guo C, Yao J. Exosome and exosomal microRNA: trafficking, sorting, and function. Genom Proteom Bioinform (2015) 13 17-24. DOI: 10.1016/j.gpb.2015.02.001
--- title: 'Role of immune cell infiltration and small molecule drugs in adhesive capsulitis: Novel exploration based on bioinformatics analyses' authors: - Hailong Liu - Baoxi Yu - Zengfa Deng - Hang Zhao - Anyu Zeng - Ruiyun Li - Ming Fu journal: Frontiers in Immunology year: 2023 pmcid: PMC9976580 doi: 10.3389/fimmu.2023.1075395 license: CC BY 4.0 --- # Role of immune cell infiltration and small molecule drugs in adhesive capsulitis: Novel exploration based on bioinformatics analyses ## Abstract ### Background Adhesive capsulitis (AC) is a type of arthritis that causes shoulder joint pain, stiffness, and limited mobility. The pathogenesis of AC is still controversial. This study aims to explore the role of immune related factors in the occurrence and development of AC. ### Methods The AC dataset was downloaded from Gene Expression Omnibus (GEO) data repository. Differentially expressed immune-related genes (DEIRGs) were obtained based on R package “DESeq2” and Immport database. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) were performed to explore the functional correlation of DEIRGs. MCC method and Least Absolute Shrinkage and Selection Operator (LASSO) regression were conducted to identify the hub genes. The immune cell infiltration in shoulder joint capsule between AC and control was evaluated by CIBERSORTx, and the relationship between hub genes and infiltrating immune cells was analyzed by Spearman’s rank correlation. Finally, potential small molecule drugs for AC were screened by the Connectivity Map database (CMap) and further verified by molecular docking. ### Results A total of 137 DEIRGs and eight significantly different types of infiltrating immune cells (M0 macrophages, M1 macrophages, regulatory T cells, Tfh cells, monocytes, activated NK cells, memory resting CD4+T cells and resting dendritic cells) were screened between AC and control tissues. MMP9, FOS, SOCS3, and EGF were identified as potential targets for AC. MMP9 was negatively correlated with memory resting CD4+T cells and activated NK cells, but positively correlated with M0 macrophages. SOCS3 was positively correlated with M1 macrophages. FOS was positively correlated with M1 macrophages. EGF was positively correlated with monocytes. Additionally, dactolisib (ranked first) was identified as a potential small-molecule drug for the targeted therapy of AC. ### Conclusions This is the first study on immune cell infiltration analysis in AC, and these findings may provide a new idea for the diagnosis and treatment of AC. ## Introduction Adhesive capsulitis (AC), also known as frozen shoulder, is a common shoulder disease. The main clinical manifestations of AC are pain, stiffness, and gradual loss of both active and passive activities of the affected shoulder resulting from progressive fibrosis and contracture of the glenohumeral capsule [1]. The prevalence of AC in the general population is $2\%$ to $5\%$, and a majority of patients are middle-aged women aged 40-70 years (2–4). Traditionally, AC is considered a self-limited disease, which is usually relieved completely within 1-2 years. However, several studies have reported that $20\%$ to $50\%$ of patients may suffer from long-term symptoms (2, 5–7). The causes of the disease are still unclear. Some mainstream academic views have linked its occurrence and development to the following factors: inflammatory reaction (8–11), fibrous tissue hyperplasia (12–15), immune factors [14, 16, 17], endocrine factors (18–20), and vascular factors (16, 21–23). Therefore, identifying the biomarkers and revealing the potential mechanism of AC is the key to the early treatment of AC. Immune cells play an important role in the occurrence and development of many diseases. AC also contains some immune components, such as macrophages, B lymphocytes, and mast cells [24]. Previous studies have shown that the pathological process of AC begins with an immune response, which then progressively worsens inflammation and eventually leads to fibrosis of the shoulder capsule [16, 17]. Moeed et al. [ 25]confirmed the role of IL-17A driven AC pathogenesis and revealed the immune landscape of AC from the immune system dominated by macrophages to the immune system rich in T cells. However, the immune mechanisms of AC in the shoulder capsule have not been investigated thoroughly. Therefore, a systematic and effective method is urgently needed to identify immune-related genes and assess the contribution of immune cells in AC. With the rapid development of RNA sequencing technology, bioinformatics analysis can be applied to identify key genes and biomarkers for many diseases, as well as to differentiate immune cell types [26]. CIBERSORTx is a popular analytical tool to estimate the abundances of 22 immune cell types in a mixed cell population, using gene expression data [27]. Hence, it can help us analyze the composition of immune cells in AC. In this study, we used CIBERSORTx for the first time to evaluate the immune cell infiltration in AC. R package “DESeq2” was conducted to screen differential expressed genes (DEGs). PPI and LASSO logistic regression were used to identify hub genes. Wilcoxon test was conducted to identify the significant differences of immune cells in AC and control. Moreover, we explored the association between the hub genes and the infiltrating immune cells, providing the cornerstone for future study in this area. More importantly, the potential small molecule drugs targeted AC were carried out by using the CMap database and the verification of the potential mechanism was conducted by molecular docking. This study not only systematically analyzed the infiltration of immune cells in the shoulder capsule of AC, but also identified immune-related key genes and possible small molecule drugs for AC. The findings of this study will provide new perspectives on the early diagnosis and treatment of AC. The flow chart of this study is showed in Figure 1. Figure 1: Flow chart of this study. p<0.01,p<0.001 and *p<0.0001. ## Data download and preprocessing Gene expression profiling of AC was downloaded from Gene Expression Omnibus (GEO) data repository (https://www.ncbi.nlm.nih.gov/geo/). The search strategy [adhesive capsulitis (All Fields) OR frozen shoulder (All Fields)] and [“Homo sapiens” (Organism)] was adopted. GSE140731 was used in our study, which contained a total of 48 samples [28]. Gene annotation file (GENCODE - Human Release 40) was downloaded from GENCODE (https://www.gencodegenes.org) and used to convert “probe id” to “symbol” in the expression matrix. For multiple probes corresponding to the same gene symbol, we calculated the maximum value as its expression level. All analyses in this study were performed using the R software version (4.1.2). ## Immune-related genes collection A total of 1,793 non-duplicated immune-related genes (IRGs) were obtained from the ImmPort database [29], which is one of the largest subject level open repositories of human immunology data and funded by the National Institutes of Health (NIH), National Institute of Allergy and Infectious Diseases (NIAID), Division of Allergy, Immunology, and Transplantation (DAIT). ## Identification of differentially expressed genes (DEGs) and differential expressed immune-related genes (DEIRGS) R package “DEseq2” was used to screen for differential expressed genes (DEGs). The select criteria of DEGs were set as adjusted p-value < 0.05 and \|log2 fold change\| > 1. To obtain differentially expressed immune-related genes (DEIRGS), the DEGs were overlapped with the above 1,793 IRGs. Heat map, volcano plots and venn diagram created by “ggplot2” package were used for visualization. ## GO, KEGG, PPI network analysis and hub genes screening Gene Ontology (GO), which involved biological processes (BP), cellular components (CC) and molecular functions (MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis were conducted by using “clusterProfiler” package, and the threshold values of p and q were set as 0.05. The results were visualized by R packages “ggpubr”, “ggplot2”, and “Goplot”. STRING (version11.5, https://cn.string-db.org/) was used to construct the protein-protein interaction (PPI) network with a confidence score >0.4. Cytoscape (version3.9.1) was used for visualization and the plug-in cytoHubba was used to calculate the ranking of DEIRGs. We selected the top 10 genes of the MCC method. Then, based on the 10 genes, we used the Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression analysis by using the R package”glmnet” to identify the best hub genes. ## Assessment of immune cell infiltration CIBERSORTx (https://cibersortx.stanford.edu/) is an analytical tool to provide an estimation of the abundances of infiltrating immune cell types in a mixed cell population, using a normalized gene expression matrix. The raw counts of the gene expression matrix were converted into TPM (transcripts per million) in R software before being submitted. The immune infiltration analyses were performed with 1000 permutations and the LM22 was adopted as a reference gene expression signature. The LM22 dataset contains 547 signature genes that distinguish between 22 human immune cell phenotypes [27]. Only samples with a CIBERSORTx p value < 0.05 was filtered and selected for the subsequent analysis. R packages “ggplot2”, “ggraph”, and “corrplot” were used for visualization. ## Correlation analysis between hub genes and infiltrating immune cells The relationship of the 4 hub genes with the levels of infiltrating immune cells was explored using Spearman’s rank correlation analysis in R software. The results were visualized using the R package “ggplot2”. ## Identification of candidate small molecule compounds The Connectivity Map [30], or CMap, is a resource that uses cellular responses to perturbation to find relationships between diseases, genes, and therapeutics. It could be used to predict potential small molecules that affect that phenotype caused by specific gene expression. To explore the potential small molecule drugs that may treat AC, the 137 DEIRGs mentioned previously were divided into two groups (up-regulation down-regulation) and then submitted to the CMap database. A negative enrichment score indicates that small molecules may reverse the expression of the genes and have potential therapeutic value. ## Molecular docking verification Molecular docking between hub genes and small molecule compounds was carried out to predict the accuracy of the pivotal components and prediction targets using AutoDock Vina (v1.5.7). PubChem database (https://pubchem.ncbi.nlm.nih.gov/), RCSB protein data (http://www.rcsb.org/), and PDBe-KB database (https://www.ebi.ac.uk/pdbe/pdbe-kb/) were selected to download the MOL2 format of ligands and PDB format of proteins. Crystal of proteins was introduced to Pymol software (https://pymol.org/2/; version 2.4.1) to conduct dehydration and separation of ligands. Subsequently, the crystal conducted was introduced to AutoDockTools to build a docking grid box of targets. Molecular dockings were achieved via AutoDock Vina. The lower affinity scores, one of the results of molecular docking, represent a more stable binding affinity of protein and ligand. Eventually, the complexes of protein and compound were visualized by Pymol software. ## Identification of DEGs and DEIRGs The GSE140731 dataset was platform-based on GPL24676 and contained a total of 48 samples, including 22 AC samples and 26 control samples. We used the R package “DESeq2” to identify 1,012 DEGs (698 up-regulated and 314 down-regulated) from the dataset, as shown in the heatmap and volcano map (Figures 2A, B). Next, to obtain DEIRGs, the intersection of DEGs with immune-related genes from the Immport database was visualized by venn diagram (Figure 2C). Finally, 137 genes were screened, including 99 up-regulated and 38 down-regulated (Figure 2D). Figure 2: Identification of DEGs and DEIRGs. DEGs were visualized by heatmap (A) and volcano map (B). Venn diagram was used to visualize the acquisition of 137 DEIRGs (C). DEIRGs contain 99 up-regulated genes and 38 down-regulated genes (D). ## GO and KEGG enrichment analysis of DEIRGs To explore the potential biological functions and signaling pathways of DEIRGs, we performed GO and KEGG enrichment analysis with R package “clusterProfiler”. The GO analysis results revealed that the DEIRGs were mostly enriched in positive regulation of response to external stimulus, cytokine-mediated signaling pathway, cell chemotaxis, regulation of chemotaxis and leukocyte migration for biology process (BP); external side of plasma membrane, collagen-containing extracellular matrix, endocytic vesicle, secretory granule lumen and cytoplasmic vesicle lumen for cellular component (CC); receptor ligand activity, signaling receptor activator activity, cytokine activity, cytokine receptor binding and growth factor activity for molecular function (MF) (Figures 3A, B). Otherwise, KEGG pathway enrichment analysis showed that pathways were mainly associated with cytokine-cytokine receptor interaction, viral protein interaction with cytokine and cytokine receptor, chemokine signaling pathway, PI3K-Akt signaling pathway, and JAK-STAT signaling pathway (Figures 3C, D). Figure 3: Functional enrichment and PPI of DEIRGs. (A, B) The results of GO analysis were displayed in lollipop and circle charts. (C, D) The results of KEGG analysis were displayed in treemap and circle charts. ## PPI, LASSO analysis and hub genes identification Two different algorithms, namely, PPI and LASSO regression, were used to identify the hub genes of DEIRGs. We obtained the PPI results from the STRING database and then used the CytoHubba plugin MCC to calculate the score of each node gene. The top 10 genes (CCL4, CCL3, MMP9, CXCL2, PTGS2, SOCS3, JUN, CXCL3, EGF, FOS) of the MCC method are shown in Figure 4A. Finally, 4 hub genes (MMP9, SOCS3, EGF, FOS) were identified using the least absolute shrinkage and selection operator (LASSO) regression algorithm (Figures 4B–D). Figure 4: PPI network establishment and hub genes identification. (A) Top 10 genes based on MCC method, and yellow to red represent progressively higher scores. (B, C) LASSO logistic regression was performed to further identify hub genes. The red dotted line on the left represents lambda.min, and the red dotted line on the right represents lambda.1se. (D) The final 4 hub genes were shown in the violin diagram. ∗∗ p < 0.01,∗∗∗p < 0.001 and ∗∗∗∗ p < 0.0001. ## Immune cell infiltration in AC tissue The results of immune cell infiltration were downloaded from the CIBERSORTx website and visualized by R software. First, a bar chart (Figure 5A) and a heat map (Figure 5B) were used to show the composition of 22 kinds of immune cells in each sample. In the bar chart, the color represents the estimated proportion of different immune cells in each sample, and the sum of the total proportion is 1. The heat map represents the difference of immune cell abundance between AC samples and control samples. The results showed that M2 macrophages, activated NK cells, resting mast cells, Monocytes, and memory resting CD4+T cells were the main infiltration immune cells. After removing the immune cells with an expression abundance of 0 in all samples, we evaluated the correlation of the remaining 20 kinds of infiltration immune cells in AC shoulder capsule tissues. The correlation map (Figure S1, A) showed that M2 macrophages were negatively correlated with M0 macrophages (r = -0.43) and regulatory T cells (r = -0.68), positively correlated with resting mast cells ($r = 0.58$). Activated NK cells were strongly positively associated with memory resting CD4+T cells ($r = 0.87$), eosinophils ($r = 0.59$), plasma cells ($r = 0.57$) and resting mast cells ($r = 0.45$), but substantially negatively associated with resting NK cells (r = -0.49) and M0 macrophages (r = -0.59). Resting mast cells were strongly positively connected with M2 macrophages ($r = 0.58$), activated NK cells ($r = 0.45$), memory resting CD4+T cells ($r = 0.43$), but substantially negatively correlated with M0 macrophages (r = -0.8), resting NK cells (r = -0.46), Gamma delta T cells (r = -0.43), Tfh cells (r = -0.43), memory activated CD4+T cells (r= -0.44) and active mast cells (r = -0.48). Monocytes were strongly positively connected with naïve B cells ($r = 0.64$), but negatively correlated with M0 macrophages (r = -0.53). Memory resting CD4+T cells were strongly positively associated with plasma cells ($r = 0.46$), activated NK cells ($r = 0.87$), resting mast cells ($r = 0.43$) and eosinophils ($r = 0.59$), negatively correlated with M0 macrophages (r = -0.73). The network diagram (Figure 5C) showed that M0 macrophages, plasma cells, and resting mast cells were closely related to other infiltrating immune cells, but naïve B cells and resting dendritic cells were weakly related to other infiltrating immune cells. Otherwise, based on the composition of infiltrating immune cells in shoulder capsule tissues, we could completely distinguish AC from normal tissues by PCA analysis (Figure S1, B). As shown in the violin diagram (Figure 5D), the degree of M0 macrophages, M1 macrophages, regulatory T cells and Tfh cells infiltration in AC tissues were significantly higher than in normal tissues ($p \leq 0.05$), but the degree of monocytes, activated NK cells, memory resting CD4+T cells and resting dendritic cells infiltration were significantly lower than in normal tissues ($p \leq 0.05$). Figure 5: The results of immune cell infiltration analysis. Composition of 22 types of infiltrating immune cells in each sample was shown in a bar chart (A) and a heat map(B). Network diagram of 22 types of infiltrating immune cells. Larger circles represent stronger interactions with other immune cells (C). Wilcoxon test was used to identify significantly different infiltrating immune cells in AC and normal tissues (D). ## Correlation between hub genes and differential infiltrating immune cells in AC We analyzed the correlation of 4 hub genes (MMP9, SOCS3, EGF, FOS) with 8 significantly differential infiltrating immune cells in AC. The results are presented in Figure 6A. Significantly correlated hub genes and immune cells were screened by adjusted p-value < 0.05. As shown in Figure 6B, MMP9 was negatively correlated with memory resting CD4+T cells (r = -0.57, $$p \leq 2.8$$e-05) and activated NK cells (r = -0.6, $$p \leq 7.9$$e-06), but positively correlated with M0 macrophages ($r = 0.46$, $$p \leq 0.0011$$). SOCS3 was positively correlated with M1 macrophages ($r = 0.5$, $$p \leq 0.00035$$). FOS was positively correlated with M1 macrophages ($r = 0.46$, $$p \leq 0.001$$). EGF was positively correlated with monocytes ($r = 0.46$, $$p \leq 0.001$$). Figure 6: Visualization of correlation between hub genes and infiltrating immune cells. (A) Correlation between 4 hub genes and 8 significantly different immune cells. Red represents positive correlation, blue represents negative correlation. (B) Significantly correlated hub genes and immune cells were screened by adjusted p-value < 0.05. ## Small molecule drugs screening and molecular docking As shown in Table 1, the top 10 small molecule compounds (dactolisib, indinavir, NVP-AUY922, WYE-354, fostamatinib, selumetinib, loteprednol, velnacrine, tizanidine, tivozanib) with highest negative score were screened as potential drugs for AC. The 2D chemical structures downloaded from PubChem are presented in Figures 7A–J. Then, these 10 small molecule compounds were docked with screened 4 core targets (MMP9, SOCS3, EGF, FOS) by using AutoDock Vina software. The binding energy for molecular docking is presented in Table 2. We selected the lowest binding energy between each small molecule compound and the core target for visualization (Figures 8A–J). The yellow dotted lines in the figure represent hydrogen bonds. For instance, dactolisib may play its biological role by binding to MMP9 and forming a hydrogen bond on the amino acid GLY-428 near the active site. ## Discussion AC is a chronic shoulder disease characterized by pain, stiffness, and dysfunction. Although it is traditionally believed that most people’s symptoms can be completely relieved within 1-2 years, more and more clinical studies challenge this theory [2, 4, 6]. At present, there are few studies on the molecular mechanism of AC, and there are still controversies about its pathogenesis. Therefore, the most effective treatment for AC is still uncertain. Immune factors play an important role in the diagnosis and treatment of many diseases, but little is known about their role in AC. In this study, we used a systematic and comprehensive bioinformatics method to explore the immune-related hub genes of AC, analyze the role of immune cell infiltration in the shoulder capsule, and predict potential small molecule drugs for AC. We performed a comprehensive analysis of the GSE140731 dataset and identified a total of 137 DEIRGs, including 99 up-regulated and 38 down-regulated. GO enrichment analysis showed that the DEIRGs were associated with cytokine activity, chemokine activity, and cytokine-mediated signaling pathway. KEGG analysis indicated that these DEIRGs were primarily enriched in cytokine-cytokine receptor interaction, chemokine signaling pathway and natural killer cell mediated cytotoxicity. These results confirm previous findings that inflammation plays an important role in the development of AC and suggest that immune response may be involved, which is consistent with our goal. To improve the reliability of the results, we integrated PPI and LASSO to screen hub genes. Four genes were finally screened, namely MMP9, EGF, FOS, and SOCS3. Among them, only MMP9 has been reported in AC. MMP9 is a member of the matrix metalloproteinase (MMP) family and its main function is to maintain the dynamic balance of extracellular matrix. As early as 2005, Blaine et al. [ 31] and Voloshin et al. [ 32] successively confirmed that patients with rotator cuff tears were prone to bursitis, and the expression of MMP9 in subacromial bursitis was significantly higher than that in the control group. Yi Wang et al. [ 33]found that targeted knockout of TNF-α can downregulate the expression of MMP9, thereby reducing the inflammatory response of bursitis. In 2019, a Brazilian study [34] pointed out that women carrying the allele of MMP9 would increase the risk of frozen shoulder. In the same year, a Korean study [35] confirmed that MMP9 was significantly overexpressed in frozen shoulder patients. These studies are consistent with our bioinformatics analysis results and indicate that MMP9 has potential value in the diagnosis and treatment of AC. Although there have been no studies on MMP9 regulating immune responses in AC, MMP9 has been shown to exert immune function in many other diseases (36–40). SOCS3 is a member of suppressor of cytokine signaling (SOCS) family. It is an important regulator of cytokine signal transduction and immune response. SOCS3-mediated m6A mRNA methylation can regulate T cell homeostasis [41]. SOCS3 acts as a regulator of macrophage polarization, and its deficiency can skew macrophages toward an M1 phenotype [42]. In IBD-related diseases, SOCS3 can regulate the expression and differentiation of T cells and B cells [43]. FOS is one of the four members of the FOS gene family (FOS, FOSB, FOSL1, and FOSL2), which can form the transcription factor complex AP-1 and is considered a regulator of cell proliferation. Non-coding RNA can regulate immune response by targeting FOS [44, 45]. And FOS can transcribe and activate the target gene NFATc1 and participate in the active immune response [46]. Ryoko Yoshida et al. [ 47]confirmed that FOS can inhibit some innate and adaptive immune responses in dendritic cells. Epidermal growth factor (EGF) is a multifunctional growth factor. By combining with its receptor (EGFR), EGF can induce the growth and migration of tissue cells, promote the expression of differentiation genes, and maintain the normal metabolism of epithelial cells [48]. In early life, EGF can promote the maturation of the immune system [49]. Christina Groepper et al. [ 50]found that EGF signaling can be modified by HCV to exert antiviral immunity by upregulating CXCR2 expression. These studies provide some theoretical support for further exploring the immune function of these genes in AC. To our knowledge, this is the first study about immune cell infiltration in AC tissue. Compared with control group, M0 macrophages, M1 macrophages, regulatory T cells, and Tfh cells were significantly higher in AC shoulder capsule tissues, while monocytes, activated NK cells, memory resting CD4+T cells, and resting dendritic cells were significantly lower in AC. Interestingly, although the infiltration proportions of M2 macrophages and resting mast cells were relatively high, the difference between AC and control groups was not statistically significant. Our results showed that M0 macrophages were significantly increased and mainly polarized into M1 macrophages in AC tissues, which may be an important reason for aggravating AC. Although the role of immune cells in AC has not been elucidated, their relationship with inflammation has been reported. During inflammation or tissue injury, pro-inflammatory mediators attract migrating monocytes to sites of inflammation and promote their differentiation toward macrophages to activate them [51]. Regulatory T cells and Tfh cells may mediate the immune inflammatory response by promoting fibrogenesis and cytokine production [52, 53]. Imbalance of dendritic cells leads to disturbance of immune homeostasis, which in turn causes abnormal inflammatory activation [54, 55]. NK cells are a double-edged sword in the process of inflammation, which may be related to the activation of T cells and their recruitment by DCs [56]. To further explore the important role of immune infiltrating cells and hub genes in AC, we calculated the correlation among them. The most intriguing result was that SOCS3 was positively correlated with M1 macrophages ($r = 0.5$, $$p \leq 0.00035$$). Yao Chun Wang et al. [ 57] found that SOCS3 could be activated by a notch signal, thus promoting the polarization of M1 macrophages. However, another study [42] suggested that the lack of SOCS3 could promote the polarization of M1 macrophages. Therefore, further experiments are needed to verify the relationship between immune cells and hub genes. CMap is a database often used to explore potential therapeutic drugs for diseases [58]. We uploaded 99 up-regulated DEIRGs and 38 down-regulated DEIRGs to the database and successfully screened the top ten compounds with negative scores, of which dactolisib was the first. Dactolisib is a dual ATP competitive PI3K and mTOR inhibitor, which has been proven to have certain effects on tumors [59, 60], inflammatory diseases [61], polycystic kidney disease [62], and Alzheimer’s disease [63]. Meanwhile, we performed molecular docking to validate the binding of hub genes and small molecules. This provides the basis for future basic pharmacological experiments in AC. This study has some limitations. First, only one dataset was used to screen DEGs, which was considered a possible constraint. Second, it lacks useful clinical information, including the duration of the disease, etc. Last, we only used bioinformatics methods, and more in vitro and in vivo experiments are needed in the future. In conclusion, we not only screened four hub DEIRGs, but also analyzed the immune cell infiltration of AC for the first time. Meanwhile, the potential small molecule drugs were predicted. These findings provide a new idea to the study the pathogenesis of AC and we will further validate the above results through in vivo and in vitro experiments in future studies. ## Data availability statement Publicly available datasets were analyzed in this study. This data can be found here: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE140731. ## Author contributions MF and RL designed the study. HL, BY, and ZD drafted the manuscript. HZ and AZ made a significant contribution to the acquisition and integration of the data. MF and RL reviewed and revised the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1075395/full#supplementary-material ## References 1. Neviaser AS, Neviaser RJ. Adhesive capsulitis of the shoulder. J Am Acad Orthop Surg (2011) 19. DOI: 10.5435/00124635-201109000-00004 2. Binder AI, Bulgen DY, Hazleman BL, Roberts S. Frozen shoulder: a long-term prospective study. Ann Rheum Dis (1984) 43. DOI: 10.1136/ard.43.3.361 3. Lloyd-Roberts GC, French PR. Periarthritis of the shoulder: a study of the disease and its treatment. Br Med J (1959) 1. DOI: 10.1136/bmj.1.5137.1569 4. Neviaser AS, Hannafin JA. Adhesive capsulitis: a review of current treatment. Am J Sports Med (2010) 38. DOI: 10.1177/0363546509348048 5. Manske RC, Prohaska D. Diagnosis and management of adhesive capsulitis. Curr Rev Musculoskelet Med (2008) 1. DOI: 10.1007/s12178-008-9031-6 6. Shaffer B, Tibone JE, Kerlan RK. Frozen shoulder. a long-term follow-up. J Bone Joint Surg Am (1992) 74. DOI: 10.2106/00004623-199274050-00013 7. Hand C, Clipsham K, Rees JL, Carr AJ. Long-term outcome of frozen shoulder. J Shoulder Elbow Surg (2008) 17. DOI: 10.1016/j.jse.2007.05.009 8. Mullett H, Byrne D, Colville J. Adhesive capsulitis: human fibroblast response to shoulder joint aspirate from patients with stage II disease. J Shoulder Elbow Surg (2007) 16. DOI: 10.1016/j.jse.2006.08.001 9. Kabbabe B, Ramkumar S, Richardson M. Cytogenetic analysis of the pathology of frozen shoulder. Int J Shoulder Surg (2010) 4. DOI: 10.4103/0973-6042.76966 10. Bunker TD, Reilly J, Baird KS, Hamblen DL. Expression of growth factors, cytokines and matrix metalloproteinases in frozen shoulder. J Bone Joint Surg Br (2000) 82. DOI: 10.1302/0301-620x.82b5.9888 11. Rodeo SA, Hannafin JA, Tom J, Warren RF, Wickiewicz TL. Immunolocalization of cytokines and their receptors in adhesive capsulitis of the shoulder. J Orthop Res (1997) 15. DOI: 10.1002/jor.1100150316 12. Bunker TD. Frozen shoulder: unravelling the enigma. Ann R Coll Surg Engl (1997) 79 13. Lubis AMT, Lubis VK. Matrix metalloproteinase, tissue inhibitor of metalloproteinase and transforming growth factor-beta 1 in frozen shoulder, and their changes as response to intensive stretching and supervised neglect exercise. J Orthop Sci (2013) 18. DOI: 10.1007/s00776-013-0387-0 14. Bunker TD, Anthony PP. The pathology of frozen shoulder. A Dupuytren-like disease J Bone Joint Surg Br (1995) 77. DOI: 10.1302/0301-620x.77b5.7559688 15. Cohen C, Leal MF, Belangero PS, Figueiredo EA, Smith MC, Andreoli CV. The roles of tenascin c and fibronectin 1 in adhesive capsulitis: a pilot gene expression study. Clinics (Sao Paulo) (2016) 71. DOI: 10.6061/clinics/2016(06)07 16. Kanbe K, Inoue K, Inoue Y, Chen Q. Inducement of mitogen-activated protein kinases in frozen shoulders. J Orthop Sci (2009) 14 56-61. DOI: 10.1007/s00776-008-1295-6 17. Hand GCR, Athanasou NA, Matthews T, Carr AJ. The pathology of frozen shoulder. J Bone Joint Surg Br (2007) 89. DOI: 10.1302/0301-620X.89B7.19097 18. Kim Y-S, Kim J-M, Lee Y-G, Hong O-K, Kwon H-S, Ji J-H. Intercellular adhesion molecule-1 (ICAM-1, CD54) is increased in adhesive capsulitis. J Bone Joint Surg Am (2013) 95. DOI: 10.2106/JBJS.K.00525 19. Gundtoft PH, Kristensen AK, Attrup M, Vobbe JW, Luxhøi T, Rix FG. Prevalence and impact of diabetes mellitus on the frozen shoulder. South Med J (2018) 111. DOI: 10.14423/SMJ.0000000000000886 20. Arkkila PE, Kantola IM, Viikari JS, Rönnemaa T. Shoulder capsulitis in type I and II diabetic patients: association with diabetic complications and related diseases. Ann Rheum Dis (1996) 55. DOI: 10.1136/ard.55.12.907 21. Hwang KR, Murrell GAC, Millar NL, Bonar F, Lam P, Walton JR. Advanced glycation end products in idiopathic frozen shoulders. J Shoulder Elbow Surg (2016) 25. DOI: 10.1016/j.jse.2015.10.015 22. Ryu J-D, Kirpalani PA, Kim J-M, Nam K-H, Han C-W, Han S-H. Expression of vascular endothelial growth factor and angiogenesis in the diabetic frozen shoulder. J Shoulder Elbow Surg (2006) 15. DOI: 10.1016/j.jse.2006.01.002 23. Karvinen H, Pasanen E, Rissanen TT, Korpisalo P, Vähäkangas E, Jazwa A. Long-term VEGF-a expression promotes aberrant angiogenesis and fibrosis in skeletal muscle. Gene Ther (2011) 18. DOI: 10.1038/gt.2011.66 24. Cho C-H, Song K-S, Kim B-S, Kim DH, Lho Y-M. Biological aspect of pathophysiology for frozen shoulder. BioMed Res Int (2018) 2018. DOI: 10.1155/2018/7274517 25. Akbar M, Crowe LAN, McLean M, Garcia-Melchor E, MacDonald L, Carter K. Translational targeting of inflammation and fibrosis in frozen shoulder: Molecular dissection of the T cell/IL-17A axis. Proc Natl Acad Sci U.S.A. (2021) 118. DOI: 10.1073/pnas.2102715118 26. Shen-Orr SS, Gaujoux R. Computational deconvolution: extracting cell type-specific information from heterogeneous samples. Curr Opin Immunol (2013) 25. DOI: 10.1016/j.coi.2013.09.015 27. Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods (2015) 12. DOI: 10.1038/nmeth.3337 28. Kamal N, McGee SL, Eng K, Brown G, Beattie S, Collier F. Transcriptomic analysis of adhesive capsulitis of the shoulder. J Orthop Res (2020) 38. DOI: 10.1002/jor.24686 29. Bhattacharya S, Dunn P, Thomas CG, Smith B, Schaefer H, Chen J. ImmPort, toward repurposing of open access immunological assay data for translational and clinical research. Sci Data (2018) 5 180015. DOI: 10.1038/sdata.2018.15 30. Lamb J. The connectivity map: a new tool for biomedical research. Nat Rev Cancer (2007) 7 54-60. DOI: 10.1038/nrc2044 31. Blaine TA, Kim Y-S, Voloshin I, Chen D, Murakami K, Chang S-S. The molecular pathophysiology of subacromial bursitis in rotator cuff disease. J Shoulder Elbow Surg (2005) 14. DOI: 10.1016/j.jse.2004.09.022 32. Voloshin I, Gelinas J, Maloney MD, O’Keefe RJ, Bigliani LU, Blaine TA. Proinflammatory cytokines and metalloproteases are expressed in the subacromial bursa in patients with rotator cuff disease. Arthroscopy (2005) 21 1076.e1-1076.e9. DOI: 10.1016/j.arthro.2005.05.017 33. Wang Y, Li Q, Wei X, Xu J, Chen Q, Song S. Targeted knockout of TNF-α by injection of lentivirus-mediated siRNA into the subacromial bursa for the treatment of subacromial bursitis in rats. Mol Med Rep (2015) 12. DOI: 10.3892/mmr.2015.3985 34. Cohen C, Leal MF, Loyola LC, Santos SEB, Ribeiro-Dos-Santos ÂKC, Belangero PS. Genetic variants involved in extracellular matrix homeostasis play a role in the susceptibility to frozen shoulder: A case-control study. J Orthop Res (2019) 37. DOI: 10.1002/jor.24228 35. Cho C-H, Lho Y-M, Hwang I, Kim DH. Role of matrix metalloproteinases 2 and 9 in the development of frozen shoulder: human data and experimental analysis in a rat contracture model. J Shoulder Elbow Surg (2019) 28. DOI: 10.1016/j.jse.2018.11.072 36. Buttacavoli M, Di Cara G, Roz E, Pucci-Minafra I, Feo S, Cancemi P. Integrated multi-omics investigations of metalloproteinases in colon cancer: Focus on MMP2 and MMP9. Int J Mol Sci (2021) 22. DOI: 10.3390/ijms222212389 37. Xu T, Gao S, Liu J, Huang Y, Chen K, Zhang X. MMP9 and IGFBP1 regulate tumor immune and drive tumor progression in clear cell renal cell carcinoma. J Cancer (2021) 12. DOI: 10.7150/jca.48664 38. Lu H, Yang H-L, Zhou W-J, Lai Z-Z, Qiu X-M, Fu Q. Rapamycin prevents spontaneous abortion by triggering decidual stromal cell autophagy-mediated NK cell residence. Autophagy (2021) 17. DOI: 10.1080/15548627.2020.1833515 39. Moussa C, Hebron M, Huang X, Ahn J, Rissman RA, Aisen PS. Resveratrol regulates neuro-inflammation and induces adaptive immunity in alzheimer’s disease. J Neuroinflamm (2017) 14. DOI: 10.1186/s12974-016-0779-0 40. Ye Y, Kuang X, Xie Z, Liang L, Zhang Z, Zhang Y. Small-molecule MMP2/MMP9 inhibitor SB-3CT modulates tumor immune surveillance by regulating PD-L1. Genome Med (2020) 12 83. DOI: 10.1186/s13073-020-00780-z 41. Li H-B, Tong J, Zhu S, Batista PJ, Duffy EE, Zhao J. m6A mRNA methylation controls T cell homeostasis by targeting the IL-7/STAT5/SOCS pathways. Nature (2017) 548. DOI: 10.1038/nature23450 42. Qin H, Holdbrooks AT, Liu Y, Reynolds SL, Yanagisawa LL, Benveniste EN. SOCS3 deficiency promotes M1 macrophage polarization and inflammation. J Immunol (2012) 189. DOI: 10.4049/jimmunol.1201168 43. Li Y, de Haar C, Peppelenbosch MP, van der Woude CJ. SOCS3 in immune regulation of inflammatory bowel disease and inflammatory bowel disease-related cancer. Cytokine Growth Factor Rev (2012) 23. DOI: 10.1016/j.cytogfr.2012.04.005 44. Fan H, Lv Z, Gan L, Ning C, Li Z, Yang M. A novel lncRNA regulates the toll-like receptor signaling pathway and related immune function by stabilizing FOS mRNA as a competitive endogenous RNA. Front Immunol (2019) 10. DOI: 10.3389/fimmu.2019.00838 45. Zhu J, Yao K, Guo J, Shi H, Ma L, Wang Q. miR-181a and miR-150 regulate dendritic cell immune inflammatory responses and cardiomyocyte apoptosis. J Cell Mol Med (2017) 21. DOI: 10.1111/jcmm.13201 46. Wagner EF, Eferl R. Fos/AP-1 proteins in bone and the immune system. Immunol Rev (2005) 208. DOI: 10.1111/j.0105-2896.2005.00332.x 47. Yoshida R, Suzuki M, Sakaguchi R, Hasegawa E, Kimura A, Shichita T. Forced expression of stabilized c-fos in dendritic cells reduces cytokine production and immune responses. Biochem Biophys Res Commun (2012) 423. DOI: 10.1016/j.bbrc.2012.05.097 48. Carpenter G, Cohen S. Epidermal growth factor. J Biol Chem (1990) 265. DOI: 10.1016/S0021-9258(19)38983-5 49. Grases-Pintó B, Torres-Castro P, Marín-Morote L, Abril-Gil M, Castell M, Rodríguez-Lagunas MJ. Leptin and EGF supplementation enhance the immune system maturation in preterm suckling rats. Nutrients (2019) 11. DOI: 10.3390/nu11102380 50. Groepper C, Rufinatscha K, Schröder N, Stindt S, Ehlting C, Albrecht U. HCV modifies EGF signalling and upregulates production of CXCR2 ligands: Role in inflammation and antiviral immune response. J Hepatol (2018) 69 594-602. DOI: 10.1016/j.jhep.2018.04.005 51. Lin Y, Huang M, Wang S, You X, Zhang L, Chen Y. PAQR11 modulates monocyte-to-macrophage differentiation and pathogenesis of rheumatoid arthritis. Immunology (2021) 163 60-73. DOI: 10.1111/imm.13303 52. Zhang M, Zhang S. T Cells in fibrosis and fibrotic diseases. Front Immunol (2020) 11. DOI: 10.3389/fimmu.2020.01142 53. Wu R, Li N, Zhao X, Ding T, Xue H, Gao C. Low-dose interleukin-2: Biology and therapeutic prospects in rheumatoid arthritis. Autoimmun Rev (2020) 19. DOI: 10.1016/j.autrev.2020.102645 54. Liu J, Zhang X, Cheng Y, Cao X. Dendritic cell migration in inflammation and immunity. Cell Mol Immunol (2021) 18. DOI: 10.1038/s41423-021-00726-4 55. Joffre O, Nolte MA, Spörri R, Sousa C. Inflammatory signals in dendritic cell activation and the induction of adaptive immunity. Immunol Rev (2009) 227. DOI: 10.1111/j.1600-065X.2008.00718.x 56. Zitti B, Bryceson YT. Natural killer cells in inflammation and autoimmunity. Cytokine Growth Factor Rev (2018) 42 37-46. DOI: 10.1016/j.cytogfr.2018.08.001 57. Wang Y-C, He F, Feng F, Liu X-W, Dong G-Y, Qin H-Y. Notch signaling determines the M1 versus M2 polarization of macrophages in antitumor immune responses. Cancer Res (2010) 70. DOI: 10.1158/0008-5472.CAN-10-0269 58. Subramanian A, Narayan R, Corsello SM, Peck DD, Natoli TE, Lu X. A next generation connectivity map: L1000 platform and the first 1,000,000 profiles. Cell (2017) 171 1437-1452.e17. DOI: 10.1016/j.cell.2017.10.049 59. Massard C, Chi KN, Castellano D, de Bono J, Gravis G, Dirix L. Phase ib dose-finding study of abiraterone acetate plus buparlisib (BKM120) or dactolisib (BEZ235) in patients with castration-resistant prostate cancer. Eur J Cancer (2017) 76 36-44. DOI: 10.1016/j.ejca.2017.01.024 60. Liu G, Jin Z, Lu X. Differential targeting of gr-MDSCs, T cells and prostate cancer cells by dactolisib and dasatinib. Int J Mol Sci (2020) 21. DOI: 10.3390/ijms21072337 61. Gholizadeh S, Kamps JAAM, Hennink WE, Kok RJ. PLGA-PEG nanoparticles for targeted delivery of the mTOR/PI3kinase inhibitor dactolisib to inflamed endothelium. Int J Pharm (2018) 548. DOI: 10.1016/j.ijpharm.2017.10.032 62. Shi H, Leonhard WN, Sijbrandi NJ, van Steenbergen MJ, Fens MHAM, van de Dikkenberg JB. Folate-dactolisib conjugates for targeting tubular cells in polycystic kidneys. J Control Release (2019) 293. DOI: 10.1016/j.jconrel.2018.11.019 63. Bellozi PMQ, Gomes GF, de Oliveira LR, Olmo IG, Vieira ÉLM, Ribeiro FM. NVP-BEZ235 (Dactolisib) has protective effects in a transgenic mouse model of alzheimer’s disease. Front Pharmacol (2019) 10. DOI: 10.3389/fphar.2019.01345
--- title: 'Prolonged dual antiplatelet therapy for Chinese ACS patients undergoing emergency PCI with drug-eluting stents: Benefits and risks' authors: - Yong Zhang - Chao Chu - Zhong Zhong - Yong-bai Luo - Fei-Fei Ning - Ning Guo journal: Frontiers in Cardiovascular Medicine year: 2023 pmcid: PMC9976624 doi: 10.3389/fcvm.2023.1080673 license: CC BY 4.0 --- # Prolonged dual antiplatelet therapy for Chinese ACS patients undergoing emergency PCI with drug-eluting stents: Benefits and risks ## Abstract ### Background In patients with acute coronary syndrome (ACS), prolonged dual antiplatelet therapy (DAPT) may reduce ischemic events and increase the risks of bleeding events differently in different ethnic groups. However, whether prolonged DAPT in Chinese patients with ACS following emergency percutaneous coronary intervention (PCI) with drug-eluting stents (DES) will be beneficial or dangerous remains unclear. This study aimed to examine the potential benefits and risks of prolonged DAPT in Chinese patients with ACS who have undergone emergency PCI with DES. ### Methods This study included 2,249 patients with ACS who underwent emergency PCI. If DAPT was continued for 12 or 12–24 months, it was classified as the standard ($$n = 1$$,011) or prolonged ($$n = 1$$,238) DAPT group, respectively. The incidence of the following endpoint events was determined and compared between the two groups: composite bleeding event (BARC 1 or 2 types of bleeding and BARC 3 or 5 types of bleeding) and major adverse cardiovascular and cerebrovascular events (MACCEs) [ischemia-driven revascularization, non-fatal ischemia stroke, non-fatal myocardial infarction (MI), cardiac death, and all-cause death]. ### Results After a median period of 47 months of follow-up [47 [40, 54]], the rate of composite bleeding events was $13.2\%$ ($$n = 163$$) in the prolonged DAPT group and $7.9\%$ ($$n = 80$$) in the standard DAPT group [odds ratio (OR) 1.765, $95\%$ confidence interval (CI) 1.332–2.338, $p \leq 0.001$]. The rate of MACCEs was $11.1\%$ ($$n = 138$$) in the prolonged DAPT group and $13.2\%$ ($$n = 133$$) in the standard DAPT group (OR 0.828, $95\%$ CI 0.642–1.068, $$p \leq 0.146$$). The DAPT duration was further shown to be insignificantly correlated with MACCEs as per the multivariable Cox regression model (HR, 0.813; $95\%$ CI, 0.638–1.036; $$p \leq 0.094$$). No statistically significant difference was observed between the two groups. However, the DAPT duration was a separate predictor of composite bleeding events according to the multivariable Cox regression model (HR 1.704, $95\%$ CI 1.302–2.232, $p \leq 0.001$). Compared with the standard DAPT group, the prolonged DAPT group had substantially more BARC 3 or 5 types of bleeding events (3.0 vs. $0.9\%$ in those with standard DAPT, OR 3.430, $95\%$ CI 1.648–7.141, $p \leq 0.001$) and BARC 1 or 2 types of bleeding events (10.2 vs. $7.0\%$ in those with standard DAPT, OR 1.500, $95\%$ CI 1.107–2.032, $$p \leq 0.008$$). ### Conclusion The prolonged DAPT group had a considerably greater incidence of composite bleeding events than the standard DAPT group. No statistically significant difference was observed in the incidence of MACCEs between the two groups. ## 1. Introduction The most severe form of atherosclerotic cardiovascular disease—acute coronary syndrome (ACS)—is responsible for the majority of cardiovascular disease-related morbidity and mortality worldwide [1, 2]. Patients who have experienced an ACS event in the past are at a higher risk for readmission and further severe adverse cardiac events [3]. Dual antiplatelet therapy (DAPT) and percutaneous coronary intervention (PCI) have been proven to be effective clinical treatments for patients with ACS [4, 5]. To lower the risk of ischemic events, such as stent thrombosis (ST) and recurrent myocardial infarction (MI), recent guidelines in Europe and the United States recommend DAPT with aspirin and a P2Y12 inhibitor (clopidogrel, prasugrel, and ticagrelor) for up to 12 months (6–8). After surviving an ACS event, patients still face a high risk of recurrent ischemic events. Studies of patients with ACS from the UK and Belgium reported that $20\%$ of patients died within five years post-ACS, with $13\%$ of those deaths attributed to cardiovascular causes. These findings underscore the need for additional secondary prevention measures beyond the first year of treatment [9]. DAPT may be a viable option for lowering the risk level in patients with ACS after one year. The risk of long-term ST and cardiovascular events can theoretically be decreased with prolonged DAPT; however, it will always result in higher risks of bleeding events [10]. Whether the administration of DAPT for 12 months allowed patients in certain patient categories to lower their risk of ST or atherothrombotic consequences associated with sites outside the stented segment remains controversial (10–14). While some trials have confirmed its benefit [10, 11], others have not (12–14). According to a previous study, East Asians may have a similar or even reduced risk of developing post-PCI ischemic attacks than Westerners [15]. A total of 15,603 patients with atherothrombosis, including 775 Asians, were enrolled in the CHARISMA research (a median follow-up period of 28 months). Asians are more likely to experience moderate Global Utilization of Streptokinase and Tissue-Plasminogen Activator for Occluded Coronary Arteries (GUSTO) bleeding than other races. They also have a lower rate of the composite of cardiovascular death, MI, and stroke during antiplatelet therapy [16]. Prolonged DAPT in patients with ACS may reduce ischemic events and increase the risk of bleeding events differently in different ethnic groups. However, the effectiveness and safety of prolonged DAPT in Chinese patients with ACS after emergency PCI with drug-eluting stents (DES) are unknown. In this study, we explored the benefits and risks of prolonged DAPT in Chinese patients with ACS after emergency PCI with DES. ## 2.1. Study population The current analysis is an observational, retrospective cohort study conducted at a single location between October 2013 and February 2017 on patients with ACS who underwent emergency PCI with DES at the First Affiliated Hospital of Xi'an Jiaotong University, Yanta. The inclusion criteria of the current analysis are as follows: [1] patients between the age of 18 and 80 years; [2] patients with ACS who received DAPT for 12–24 months and who had no clinical ischemic or bleeding events during the first 12 months; and [3] patients who successfully underwent emergency PCI with DES. A total of 3,236 patients with ACS were investigated. The exclusion criteria for this study are as follows: [1] a history of coronary artery bypass grafting, cardiogenic shock, malignant tumor, significant infection, or autoimmune disease; [2] a renal disorder with an estimated glomerular filtration rate (eGFR) of < 30 mL/min/1.73 m2) or accepted renal replacement treatment; [3] hepatic dysfunction with aspartate transaminase or alanine transaminase levels greater than five upper limits of normal; [4] non-obstructive coronary disease, primary cardiomyopathy, and valvular heart disease; [5] heart failure with left ventricular ejection fraction (LVEF) <$30\%$; [6] oral anticoagulants during follow-up; [7] anemia with hemoglobin (Hb) <60 g/L; [8] a history of gastrointestinal bleeding and hemorrhagic stroke; and [9] missing clinical data. A total of 987 patients were excluded following the exclusion criteria. Finally, 2,249 patients were included in the group. If DAPT was continued for 12 or 12–24 months, it was classified as the standard ($$n = 1$$,011) or prolonged ($$n = 1$$,238) DAPT group. The duration of the prolonged DAPT group is 22 [20, 24] months (Figure 1). Figure 1: The flowchart of study subject enrollment. ACS, acute coronary syndrome; PCI, percutaneous coronary intervention; DES, drug-eluting stent; CABG, coronary artery bypass grafting; DAPT, dual antiplatelet therapy. ## 2.2. Data collection and follow-up Trained physicians gathered clinical data from electronic medical records. The records contain information on the population, anthropometry, laboratory results, medical diagnoses, and procedures. After an overnight fast, venous blood samples were collected in the morning and examined the same day at the central laboratory using the standard procedures. After admission, all patients were routinely followed up by trained clinicians for major adverse cardiovascular and cerebrovascular events (MACCEs) and composite bleeding events at 3, 6, and 12 months and, then, at every 6 months; the longest individual follow-up period was 66 months. Follow-up data were obtained from hospital records or through telephone or in-person interviews with patients and their families. The first observational endpoints of this study were MACCEs and composite bleeding events during the follow-up period of 47 [40, 54] months. We also analyzed the observational endpoints at the 24-month follow-up period after discharge (during the prolonged DAPT duration). MACCEs are defined as the composite of ischemic-driven revascularization, non-fatal ischemic stroke, non-fatal myocardial infarction (MI), cardiac death, and all-cause death. Bleeding Academic Research Composite bleeding events were created by combining BARC 3 or 5 types of bleeding events with BARC 1 or 2 types of bleeding events [17]. Only the most serious event (all-cause death > non-fatal ischemic stroke > non-fatal MI > ischemia-driven revascularization) was chosen to perform our analysis for patients with multiple MACCEs occurring virtually and simultaneously throughout the follow-up. Similarly, BARC 3 or 5 types of bleeding events were selected for our analysis for patients who experienced BARC 3 or 5 types of bleeding events and BARC 1 or 2 types of bleeding events throughout the follow-up. Only the initial occurrence of the same event was intended to be used for our analysis of patients when it occurred more than once. ## 2.3. Definitions According to the relevant guidelines, the diagnostic criteria for ACS included ST-segment elevation MI (STEMI) and non-ST-segment elevation ACS (NSTE-ACS) [non-ST-segment elevation MI (NSTEMI) or unstable angina (UA)] [6, 7]. Patients were considered to have hypertension if they received treatment with a conclusive diagnosis or if their systolic blood pressure (SBP) was ≥140 mmHg or if their diastolic blood pressure (DBP) ≥90 mmHg was higher than two times on different days during the baseline hospitalization. According to the practical guidelines, patients with type 2 diabetes mellitus either had a prior, conclusive diagnosis or had the condition recently verified [18]. Patients were considered to have hyperlipidemia if they received treatment with lipid-lowering medications or had fasting total cholesterol >6.22 mmol/L or low-density lipoprotein cholesterol (LDL-C) >4.14 mmol/L. Patients suffering from an ischemic stroke had a cerebral infarction or a transient ischemic attack. Patients with peripheral artery disease (PAD) had previously been diagnosed with artery disease as well as in the coronary and aortic arteries. They had $50\%$ stenosis and/or signs of ischemia. Patients with the eGFR levels between 30 and 60 mL/min/1.73 m2 were considered to have renal dysfunction. Emergency PCI was defined as PCI performed within 24 h of hospital admission for patients with NSTE-ACS or within 12 h of symptom onset for patients with STEMI. Elective PCI was performed 24 h after hospital admission for—patients with NSTE-ACS or 12 h after symptom onset for patients with STEMI. Weight (kg)/[height (m)]2 was the formula used to calculate body mass index (BMI). The formula for calculating the eGFR was 186 × serum creatinine (mg/dL)−1.154 × age−0.203 (× 0.742 if the patient was a woman) [19]. A number of the main coronary arteries, including the left anterior descending artery, the left circumflex artery, and the right coronary artery, must have a stenosis of ≥$50\%$ to be considered to have a several-vessel disease. Chronic total occlusion (CTO) lesions were defined as total obstruction persisting for more than 3 months, as determined by the coronary angiography findings or prior medical history. An individual stenotic lesion of over 20 mm is considered a diffuse lesion. The term “in-stent restenosis” (ISR) was used to describe a stenosis of ≥$50\%$ in a segment that was inside the stent or 5 mm away from it [20]. ## 2.4. Statistical analysis Continuous variables were presented as the mean and standard deviation or the median (IQR). A Mann–Whitney U-test or an independent-sample t-test was used to compare the two groups. Counts (percentages) were used to characterize categorical variables, which were then compared using either the Fisher's exact test or the Pearson chi-square test (Pearson χ2 test). Univariate and multivariable Cox proportional hazards analyses evaluated the predictive value of the variables for MACCEs and composite bleeding events. Several risk factors were present in the multivariate model, including clinically significant variables ($p \leq 0.2$) from the univariate model. The Kaplan–Meier survival curves estimated the cumulative incidence of MACCEs and composite bleeding events. Further stratified analysis was performed to determine the prognostic impact of standard DAPT and prolonged DAPT for MACCEs and composite bleeding events. The propensity score for matching (PSM) was calculated using a binary logistic regression model, which took into account the use of statins, ACEI/ARB, β-blockers, P2Y12 inhibitors, and aspirin at the time of discharge. Finally, 986 standard patients with DAPT were individually matched at a ratio of 1:1 to patients with prolonged DAPT. IBM SPSS Statistics (version 24.0) was used for data analysis. A statistically significant correlation was defined as a two-tailed p-value of <0.05. ## 3.1. Basic characteristics of the standard and prolonged DAPT groups The baseline characteristics of the standard and prolonged DAPT groups are displayed in Table 1. A total of 2,249 patients (60.98 ± 9.95 years; $23.5\%$ women) were enrolled in the present study, with 1,011 ($45.0\%$) in the standard DAPT group and 1,238 ($55.0\%$) in the prolonged DAPT group. The baseline demographic characteristics, medical history, laboratory data, and angiographic information were similar between the two groups. Patients in the prolonged DAPT group had a higher proportion of β-blocker and angiotensin-converting enzyme inhibitor/angiotensin receptor blocker treatment at the time of discharge. There was no statistical difference in basic characteristics between the two groups after PSM (Supplementary Table 1). Table 1 \| Characteristics \| Total population \| Standard DAPT group \| Prolonged DAPT group \| p -value \| \| --- \| --- \| --- \| --- \| --- \| \| \| (n = 2,249) \| (n = 1,124) \| (n = 1,125) \| \| \| Age, years \| 60.98 ± 9.95 \| 60.83 ± 9.90 \| 61.11 ± 9.98 \| 0.502 \| \| Gender, men, n (%) \| 1,720 (76.5%) \| 768 (76.0%) \| 952 (76.9%) \| 0.603 \| \| BMI, kg/m2 \| 23.28 ± 2.44 \| 23.28 ± 2.46 \| 23.28 ± 2.42 \| 0.943 \| \| SBP, mmHg \| 128.28 ± 21.33 \| 127.39 ± 21.04 \| 129.00 ± 21.54 \| 0.074 \| \| DBP, mmHg \| 77.66 ± 11.91 \| 77.47 ± 11.74 \| 77.81 ± 12.05 \| 0.500 \| \| Heart rate, bpm \| 74.39 ± 10.03 \| 73.94 ± 11.56 \| 74.36 ± 12.40 \| 0.103 \| \| Smoking history, n (%) \| 1,237 (55.0%) \| 560 (55.4%) \| 677 (54.7%) \| 0.738 \| \| Drinking history, n (%) \| 617 (27.4%) \| 295 (29.2%) \| 322 (26.0%) \| 0.094 \| \| Family history of CAD, n (%) \| 191 (8.5%) \| 89 (8.8%) \| 102 (8.2%) \| 0.633 \| \| Initial diagnosis, n (%) \| \| \| \| 0.218 \| \| UA \| 1,364 (60.6%) \| 633 (62.6%) \| 731 (59.0%) \| \| \| NSTEMI \| 163 (7.2%) \| 68 (6.7%) \| 95 (7.7%) \| \| \| STEMI \| 722 (32.1%) \| 310 (30.7%) \| 412 (33.3%) \| \| \| Medical history, n (%) \| Medical history, n (%) \| Medical history, n (%) \| Medical history, n (%) \| Medical history, n (%) \| \| Hypertension \| 1,297 (57.7%) \| 563 (55.7%) \| 734 (59.3%) \| 0.085 \| \| AF \| 61 (2.7%) \| 27 (2.7%) \| 34 (2.7%) \| 0.912 \| \| CHA2DS2-VASc score \| 2.00 (1.00, 3.00) \| 2.00 (1.00, 3.00) \| 2.00 (1.00, 3.00) \| 0.671 \| \| HAS-BLED score \| 1.00 (1.00, 2.50) \| 1.00 (1.00, 3.00) \| 1.00 (1.00, 2.25) \| 0.878 \| \| DM \| 599 (26.6%) \| 325 (32.1%) \| 359 (29.0%) \| 0.106 \| \| Dyslipidemia \| 251 (11.2%) \| 105 (10.4%) \| 146 (11.8%) \| 0.295 \| \| Renal dysfunction \| 45 (2.0%) \| 17 (1.7%) \| 28 (2.3%) \| 0.328 \| \| Previous MI \| 197 (8.8%) \| 79 (7.8%) \| 118 (9.5%) \| 0.152 \| \| Previous PCI \| 245 (10.9%) \| 112 (11.1%) \| 133 (10.8%) \| 0.799 \| \| Previous stroke \| 482 (21.4%) \| 200 (19.8%) \| 282 (22.8%) \| 0.085 \| \| Previous cerebral infarction \| 184 (8.2%) \| 80 (7.9%) \| 104 (8.4%) \| 0.675 \| \| Previous PAD \| 360 (16.0%) \| 156 (15.4%) \| 204 (16.5%) \| 0.500 \| \| Laboratory results \| Laboratory results \| Laboratory results \| Laboratory results \| Laboratory results \| \| WBC (× 109/L) \| 7.30 ± 2.55 \| 7.33 ± 2.56 \| 7.28 ± 2.54 \| 0.580 \| \| PLT (× 109/L) \| 157.57 ± 55.69 \| 155.30 ± 55.49 \| 159.42 ± 55.80 \| 0.081 \| \| Hb (g/L) \| 137.44 ± 18.49 \| 137.82 ± 17.88 \| 137.14 ± 18.97 \| 0.386 \| \| BUN (mmol/L) \| 5.42 ± 1.95 \| 5.36 ± 1.91 \| 5.46 ± 1.98 \| 0.259 \| \| Cr (umol/L) \| 68.50 ± 18.98 \| 67.85 ± 18.28 \| 68.63 ± 19.48 \| 0.326 \| \| eGFR (mL/min/1.73 m2) \| 97.26 ± 28.63 \| 98.47 ± 29.32 \| 96.27 ± 28.03 \| 0.070 \| \| FBG (mmol/L) \| 6.79 ± 2.92 \| 6.86 ± 2.99 \| 6.74 ± 2.86 \| 0.338 \| \| HbA1c (%) \| 6.14 ± 1.28 \| 6.20 ± 1.34 \| 6.10 ± 1.22 \| 0.068 \| \| HDL-C (mmol/L) \| 0.97 ± 0.22 \| 0.98 ± 0.23 \| 0.97 ± 0.22 \| 0.108 \| \| TC (mmol/L) \| 3.72 ± 1.18 \| 3.76 ± 1.18 \| 3.69 ± 1.19 \| 0.115 \| \| TG (mmol/L) \| 1.51 ± 0.88 \| 1.50 ± 0.88 \| 1.52 ± 0.88 \| 0.588 \| \| LDL-C (mmol/L) \| 1.83 ± 0.86 \| 1.87 ± 0.85 \| 1.81 ± 0.87 \| 0.089 \| \| NT-proBNP (pg/mL) \| 710.69 ± 1244.39 \| 695.67 ± 1177.56 \| 722.96 ± 1296.76 \| 0.605 \| \| LVEF (%) \| 59.77 ± 11.35 \| 59.84 ± 11.21 \| 59.71 ± 11.46 \| 0.779 \| \| Angiographic data \| Angiographic data \| Angiographic data \| Angiographic data \| Angiographic data \| \| LM disease, n (%) \| 264 (11.7%) \| 106 (10.5%) \| 158 (12.8%) \| 0.095 \| \| CTO, n (%) \| 661 (29.4%) \| 291 (28.8%) \| 370 (29.9%) \| 0.568 \| \| Number-vessel disease. n (%) \| \| \| \| 0.996 \| \| Single-vessel disease \| 561 (24.9%) \| 253 (25.0%) \| 308 (24.9%) \| \| \| Two-vessel disease \| 667 (29.7%) \| 300 (29.7%) \| 367 (29.6%) \| \| \| Three-vessel disease \| 1021 (45.4%) \| 458 (45.3%) \| 563 (45.5%) \| \| \| Diffuse lesion, n (%) \| 1390 (61.8%) \| 620 (61.3%) \| 770 (62.2%) \| 0.672 \| \| In-stent restenosis, n (%) \| 78 (3.5%) \| 35 (3.5%) \| 43 (3.5%) \| 0.988 \| \| Calcification lesion, n (%) \| 62 (2.8%) \| 30 (3.0%) \| 32 (2.6%) \| 0.582 \| \| Number of stents \| 1.76 ± 1.18 \| 1.78 ± 1.19 \| 1.74 ± 1.18 \| 0.392 \| \| Medication at the time of discharge, n (%) \| Medication at the time of discharge, n (%) \| Medication at the time of discharge, n (%) \| Medication at the time of discharge, n (%) \| Medication at the time of discharge, n (%) \| \| ACEI/ARB \| 1,869 (83.1%) \| 812 (80.3%) \| 1,057 (85.4%) \| 0.001 \| \| β-blocker \| 1,839 (81.8%) \| 800 (79.1%) \| 1,039 (83.9%) \| 0.003 \| \| Statins \| 2,241 (99.6%) \| 1,006 (99.5%) \| 1,235 (99.8%) \| 0.480 \| \| P2Y12 inhibitor \| \| \| \| 0.120 \| \| Clopidogrel \| 1,954 (86.9%) \| 866 (85.7%) \| 1,088 (87.9%) \| \| \| Ticagrelor \| 295 (13.1%) \| 145 (14.3%) \| 150 (12.1%) \| \| \| Aspirin \| 2,249 (100.0%) \| 1,011 (100%) \| 1,238 (100%) \| - \| \| CRUSADE score \| 23.76 ± 10.90 \| 23.39 ± 10.72 \| 24.06 ± 11.05 \| 0.147 \| ## 3.2. Incidence of MACCEs in the standard and prolonged DAPT groups A total of 271 ($12.0\%$) MACCEs, including 112 ($5.0\%$) all-cause deaths, 74 ($3.3\%$) cardiac deaths, 14 ($0.6\%$) non-fatal MIs, 33 ($1.5\%$) non-fatal ischemic strokes, and 112 ($5.0\%$) ischemia-driven revascularizations, were recorded at a median of 47 months of follow-up [47 [40, 54]]. No statistically significant difference in the prevalence of MACCEs ($11.1\%$ vs. $13.2\%$ in those with standard DAPT, OR 0.828, $95\%$ CI 0.642–1.068, $$p \leq 0.146$$), all-cause death (4.4 vs. $5.6\%$ in those with standard DAPT, OR 0.778, $95\%$ CI 0.532–1.138, $$p \leq 0.195$$), cardiac death (2.7 vs. $4.1\%$ in those with standard DAPT, OR 0.648, $95\%$ CI 0.407–1.033, $$p \leq 0.066$$), non-fatal MI (0.6 vs. $0.6\%$ in those with standard DAPT, OR 1.089, $95\%$ CI 0.377–3.150, $$p \leq 0.874$$), non-fatal ischemic stroke (1.4 vs. $1.6\%$ in those with standard DAPT, OR 0.866, $95\%$ CI 0.435–1.722, $$p \leq 0.681$$), and ischemia-driven revascularization (4.7 vs. $5.3\%$ in those with standard DAPT, OR 0.871, $95\%$ CI 0.596–1.274, $$p \leq 0.477$$) was observed between the two groups. A total of 133 ($5.9\%$) MACCEs, including 70 ($3.1\%$) all-cause deaths, 48 ($2.1\%$) cardiac deaths, 6 ($0.3\%$) non-fatal MIs, 11 ($0.5\%$) non-fatal ischemic strokes, and 46 ($2.0\%$) ischemia-driven revascularizations, were recorded at 24 months after discharge. No statistically significant difference in the prevalence of MACCEs (5.2 vs. $6.8\%$ in those with standard DAPT, OR 0.744, $95\%$ CI 0.524–1.057, $$p \leq 0.098$$), all-cause death (2.7 vs. $3.6\%$ in those with standard DAPT, OR 0.764, $95\%$ CI 0.475–1.232, $$p \leq 0.269$$), cardiac death (1.7 vs. $2.7\%$ in those with standard DAPT, OR 0.629, $95\%$ CI 0.353–1.119, $$p \leq 0.112$$), non-fatal MI (0.2 vs. $0.4\%$ in those with standard DAPT, OR 0.407, $95\%$ CI 0.074–2.227, $$p \leq 0.418$$), non-fatal ischemic stroke (0.4 vs. $0.6\%$ in those with standard DAPT, OR 0.679, $95\%$ CI 0.207–2.232, $$p \leq 0.736$$), and ischemia-driven revascularization (1.9 vs. $2.3\%$ in those with standard DAPT, OR 0.813, $95\%$ CI 0.454–1.458, $$p \leq 0.487$$) was found between the two groups (Table 2). There was no statistical difference in the incidence of MACCE and its components between the two groups after PSM (Supplementary Table 2). Table 2 \| Endpoint event \| Total population \| Standard DAPT group \| Prolonged DAPT group \| OR (95%CI) \| p -value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| (n = 2,249) \| (n = 1,011) \| (n = 1,238) \| \| \| \| MACCE, n (%) \| MACCE, n (%) \| MACCE, n (%) \| MACCE, n (%) \| MACCE, n (%) \| MACCE, n (%) \| \| 24 months \| 133 (5.9%) \| 69 (6.8%) \| 64 (5.2%) \| 0.744 (0.524, 1.057) \| 0.098 \| \| 47 months* \| 271 (12.0%) \| 133 (13.2%) \| 138 (11.1%) \| 0.828 (0.642, 1.068) \| 0.146 \| \| All-cause death, n (%) \| All-cause death, n (%) \| All-cause death, n (%) \| All-cause death, n (%) \| All-cause death, n (%) \| All-cause death, n (%) \| \| 24 months \| 70 (3.1%) \| 36 (3.6%) \| 34 (2.7%) \| 0.764 (0.475, 1.232) \| 0.269 \| \| 47 months* \| 112 (5.0%) \| 57 (5.6%) \| 55 (4.4%) \| 0.778 (0.532, 1.138) \| 0.195 \| \| Cardiac death, n (%) \| Cardiac death, n (%) \| Cardiac death, n (%) \| Cardiac death, n (%) \| Cardiac death, n (%) \| Cardiac death, n (%) \| \| 24 months \| 48 (2.1%) \| 27 (2.7%) \| 21 (1.7%) \| 0.629 (0.353, 1.119) \| 0.112 \| \| 47 months* \| 74 (3.3%) \| 41 (4.1%) \| 33 (2.7%) \| 0.648 (0.407, 1.033) \| 0.066 \| \| Non-fatal MI, n (%) \| Non-fatal MI, n (%) \| Non-fatal MI, n (%) \| Non-fatal MI, n (%) \| Non-fatal MI, n (%) \| Non-fatal MI, n (%) \| \| 24 months \| 6 (0.3%) \| 4 (0.4%) \| 2 (0.2%) \| 0.407 (0.074, 2.227) \| 0.418 \| \| 47 months* \| 14 (0.6%) \| 6 (0.6%) \| 8 (0.6%) \| 1.089 (0.377, 3.150) \| 0.874 \| \| Non-fatal ischemic stroke, n (%) \| Non-fatal ischemic stroke, n (%) \| Non-fatal ischemic stroke, n (%) \| Non-fatal ischemic stroke, n (%) \| Non-fatal ischemic stroke, n (%) \| Non-fatal ischemic stroke, n (%) \| \| 24 months \| 11 (0.5%) \| 6 (0.6%) \| 5 (0.4%) \| 0.679 (0.207, 2.232) \| 0.736 \| \| 47 months* \| 33 (1.5%) \| 16 (1.6%) \| 17 (1.4%) \| 0.866 (0.435, 1.722) \| 0.681 \| \| Ischemia-driven revascularization, n (%) \| Ischemia-driven revascularization, n (%) \| Ischemia-driven revascularization, n (%) \| Ischemia-driven revascularization, n (%) \| Ischemia-driven revascularization, n (%) \| Ischemia-driven revascularization, n (%) \| \| 24 months \| 46 (2.0%) \| 23 (2.3%) \| 23 (1.9%) \| 0.813 (0.454, 1.458) \| 0.487 \| \| 47 months* \| 112 (5.0%) \| 54 (5.3%) \| 58 (4.7%) \| 0.871 (0.596, 1.274) \| 0.477 \| ## 3.3. Cox proportional hazard analysis to assess the impact of MACCEs on prognosis The relationship between DAPT duration and MACCEs was investigated using the Cox proportional hazard model. The DAPT duration was insignificantly related to MACCEs according to a univariate model (HR 0.848, $95\%$ CI 0.668–1.076, $$p \leq 0.177$$) and a multivariate model (HR 0.813, $95\%$ CI 0.638–1.036, $$p \leq 0.094$$) (Table 3). Meanwhile, the multivariate and univariate analyses revealed the absence of a meaningful relationship between the DAPT duration and cardiac death (HR 0.643, $95\%$ CI 0.361–1.142, $$p \leq 0.132$$). Table 3 \| Characteristics \| Univariate analysis \| Univariate analysis.1 \| Multivariate analysis \| Multivariate analysis.1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| HR (95%CI) \| p -value \| HR (95%CI) \| p -value \| \| Age \| 1.025 (1.013, 1.038) \| < 0.001 \| 1.018 (1.003, 1.033) \| 0.017 \| \| Previous MI \| 1.563 (1.091, 2.240) \| 0.015 \| \| \| \| Previous PCI \| 1.640 (1.184, 2.272) \| 0.003 \| \| \| \| Previous stroke \| 1.030 (0.773, 1.372) \| 0.840 \| 1.045 (0.777, 1.407) \| 0.771 \| \| AF \| 1.870 (1.071, 3.265) \| 0.028 \| 1.656 (0.942, 2.909) \| 0.080 \| \| DM \| 1.246 (0.970, 1.600) \| 0.085 \| \| \| \| Drinking history \| 0.809 (0.611, 1.071) \| 0.139 \| \| \| \| DBP \| 0.987 (0.977, 0.997) \| 0.011 \| \| \| \| HR \| 1.013 (1.004, 1.023) \| 0.007 \| \| \| \| Number-vessel disease \| 1.329 (1.138, 1.554) \| < 0.001 \| \| \| \| LM disease \| 1.749 (1.284, 2.383) \| < 0.001 \| 1.469 (1.067, 2.022) \| 0.018 \| \| CTO \| 1.211 (0.940, 1.561) \| 0.139 \| \| \| \| ISR \| 1.957 (1.181, 3.244) \| 0.009 \| \| \| \| Number of stents \| 1.194 (1.089, 1.309) \| < 0.001 \| 1.114 (1.009, 1.229) \| 0.032 \| \| WBC \| 1.048 (1.003, 1.094) \| 0.036 \| \| \| \| Hb \| 0.991 (0.984, 0.997) \| 0.003 \| 0.993 (0.986, 1.000) \| 0.047 \| \| LVEF \| 0.977 (0.967, 0.986) \| < 0.001 \| 0.987 (0.976, 0.998) \| 0.019 \| \| BUN \| 1.094 (1.035, 1.156) \| 0.001 \| \| \| \| Cr \| 1.007 (1.002, 1.013) \| 0.005 \| \| \| \| eGFR \| 0.997 (0.993, 1.001) \| 0.167 \| \| \| \| FBG \| 1.045 (1.007, 1.083) \| 0.019 \| \| \| \| HbA1c \| 1.187 (1.099, 1.282) \| < 0.001 \| 1.200 (1.068, 1.348) \| 0.002 \| \| HDL-C \| 0.618 (0.356, 1.073) \| 0.087 \| \| \| \| LDL-C \| 1.125 (0.992, 1.276) \| 0.066 \| 1.161 (1.018, 1.324) \| 0.026 \| \| DAPT duration \| 0.848 (0.668, 1.076) \| 0.177 \| 0.813 (0.638, 1.036) \| 0.094 \| \| ACEI/ARB \| 0.768 (0.543, 1.087) \| 0.136 \| \| \| \| β-blocker \| 0.584 (0.404, 0.844) \| 0.003 \| \| \| ## 3.4. Sensitivity analysis We further analyzed different subgroups to evaluate the independent association of DAPT duration with MACCEs. According to Figure 2, being men (HR 0.538, $95\%$ CI 0.307–0.941, $$p \leq 0.028$$), having a history of PAD (HR 0.730, $95\%$ CI 0.555–0.960, $$p \leq 0.024$$), having three-vessel disease (HR 0.672, $95\%$ CI 0.475–0.949, $$p \leq 0.023$$), and having >2 stents implanted (HR 0.604, $95\%$ CI 0.384–0.950, $$p \leq 0.028$$) primarily reflected the significant predictive effect of DAPT duration on MACCEs. It is worth noting that patients with a history of PAD appeared to have a higher predictive value for DAPT duration [HR ($95\%$CI) with previous PAD 0.730 (0.555–0.960) vs. without previous PAD 0.904 (0.793–1.031), p for interaction = 0.016]. Figure 2: Forest plot investigating the association between the DAPT duration and MACCEs in different subgroups. DAPT, dual antiplatelet therapy; UA, unstable angina; NSTEMI, non ST-segment elevation myocardial infarction; STEMI, ST-segment elevation myocardial infarction; MI, myocardial infarction; PCI, percutaneous coronary intervention; PAD, peripheral artery disease; DM, diabetes mellitus; CAD, coronary artery disease; LM, left main; CTO, chronic total occlusion; SBP, systolic blood pressure; Hb, hemoglobin; LVEF, left ventricular ejection fraction; eGFR, estimated glomerular filtration rate; BMI, body mass index; HbA1c, glycosylated hemoglobin A1c; LDL-C, low-density lipoprotein cholesterol; CJ, confidence interval. ## 3.5. Incidence of composite bleeding events in the standard and prolonged DAPT groups A total of 243 ($10.8\%$) composite bleeding events, including 46 ($2.0\%$) BARC 3 or 5 types of bleeding events and 197 ($8.8\%$) BARC 1 or 2 types of bleeding events, were recorded at a median of 47 months of follow-up [47 [40, 54]]. The incidence of composite bleeding events (13.2 vs. $7.9\%$ in those with standard DAPT, OR 1.765, $95\%$ CI 1.332–2.338, $p \leq 0.001$), BARC 3 or 5 types of bleeding events (3.0 vs. $0.9\%$ in those with standard DAPT, OR 3.430, $95\%$ CI 1.648–7.141, $p \leq 0.001$), and BARC 1 or 2 types of bleeding events (10.2 vs. $7.0\%$ in those with standard DAPT, OR 1.500, $95\%$ CI 1.107–2.032, $$p \leq 0.008$$) increased significantly in the prolonged DAPT group. A total of 143 ($6.4\%$) composite bleeding events, including 24 ($1.1\%$) BARC 3 or 5 types of bleeding events and 119 ($5.3\%$) BARC 1 or 2 types of bleeding events, were recorded at 24 months after discharge, and the incidence of composite bleeding events (9.3 vs. $2.8\%$ in those with standard DAPT, OR 3.597, $95\%$ CI 2.358–5.495, $p \leq 0.001$), BARC 3 or 5 types of bleeding events (1.3 vs. $0.6\%$ in those with standard DAPT, OR 2.469, $95\%$ CI 1.001–6.250, $$p \leq 0.048$$), and BARC 1 or 2 types of bleeding events (7.8 vs. $2.2\%$ in those with standard DAPT, OR 3.817, $95\%$ CI 2.387–6.135, $p \leq 0.001$) also increased significantly in the prolonged DAPT group (Table 4). The incidence of composite bleeding events and its components was statistically different between the two groups after PSM (Supplementary Table 3). Table 4 \| Endpoint event \| Total population (n = 2,249) \| Standard DAPT group (n = 1,011) \| Prolonged DAPT group (n = 1,238) \| OR (95%CI) \| p -value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Composite bleeding events, n (%) \| Composite bleeding events, n (%) \| Composite bleeding events, n (%) \| Composite bleeding events, n (%) \| Composite bleeding events, n (%) \| Composite bleeding events, n (%) \| \| 24 months \| 143 (6.4%) \| 28 (2.8%) \| 115 (9.3%) \| 3.597 (2.358, 5.495) \| <0.001 \| \| 47 months* \| 243 (10.8%) \| 80 (7.9%) \| 163 (13.2%) \| 1.765 (1.332, 2.338) \| <0.001 \| \| BARC 3 or 5 bleeding events, n (%) \| BARC 3 or 5 bleeding events, n (%) \| BARC 3 or 5 bleeding events, n (%) \| BARC 3 or 5 bleeding events, n (%) \| BARC 3 or 5 bleeding events, n (%) \| BARC 3 or 5 bleeding events, n (%) \| \| 24 months \| 24 (1.1%) \| 6 (0.6%) \| 18 (1.3%) \| 2.469 (1.001, 6.250) \| 0.048 \| \| 47 months* \| 46 (2.0%) \| 9 (0.9%) \| 37 (3.0%) \| 3.430 (1.648, 7.141) \| <0.001 \| \| BARC 1 or 2 bleeding events, n (%) \| BARC 1 or 2 bleeding events, n (%) \| BARC 1 or 2 bleeding events, n (%) \| BARC 1 or 2 bleeding events, n (%) \| BARC 1 or 2 bleeding events, n (%) \| BARC 1 or 2 bleeding events, n (%) \| \| 24 months \| 119 (5.3%) \| 22 (2.2%) \| 97 (7.8%) \| 3.817 (2.387, 6.135) \| <0.001 \| \| 47 months* \| 197 (8.8%) \| 71 (7.0%) \| 126 (10.2%) \| 1.500 (1.107, 2.032) \| 0.008 \| ## 3.6. Cox proportional hazard analysis to assess the impact of composite bleeding events on prognosis The relationship between DAPT duration and composite bleeding events was investigated using the Cox proportional hazard model. The DAPT duration was substantially related to composite bleeding events according to a univariate model (HR 1.724, $95\%$ CI 1.319–2.252, $p \leq 0.001$). The other significant risk factors included sex, WBC count, Hb level, LVEF, HbA1c level, previous history of MI, and use of ticagrelor. The multivariate model for analysis included various risk factors, such as significant variables ($p \leq 0.2$) from the univariate model, and the DAPT duration remained an independent predictor of composite bleeding events (HR 1.704, $95\%$ CI 1.302–2.232, $p \leq 0.001$). The other independent predictors included age, Hb, and HbA1c, a previous history of diabetes mellitus (DM), and the use of ticagrelor (Table 5). Meanwhile, the multivariate and univariate analyses indicated that a meaningful relationship between DAPT duration and BARC 3 or 5-type bleeding events exists (Supplementary Table 4). Table 5 \| Characteristics \| Univariate analysis \| Univariate analysis.1 \| Multivariate analysis \| Multivariate analysis.1 \| \| --- \| --- \| --- \| --- \| --- \| \| Characteristics \| HR (95%CI) \| p -value \| HR (95%CI) \| p -value \| \| Gender \| 1.447 (1.101, 1.902) \| 0.008 \| \| \| \| Age \| 1.008 (0.996, 1.020) \| 0.107 \| 1.017 (1.003, 1.032) \| 0.019 \| \| Previous MI \| 2.032 (1.110, 3.717) \| 0.022 \| \| \| \| Hypertension \| 1.207 (0.931, 1.564) \| 0.155 \| \| \| \| DM \| 1.260 (0.968, 1.639) \| 0.085 \| 1.749 (1.256, 2.437) \| 0.001 \| \| Smoking \| 1.245 (0.969, 1.603) \| 0.087 \| \| \| \| DBP \| 0.993 (0.982, 1.003) \| 0.169 \| \| \| \| SBP \| 0.995 (0.990, 1.001) \| 0.124 \| \| \| \| WBC \| 1.063 (1.007, 1.122) \| 0.028 \| \| \| \| PLT \| 0.998 (0.995, 1.001) \| 0.122 \| \| \| \| Hb \| 0.987 (0.981, 0.994) \| <0.001 \| 0.987 (0.979, 0.995) \| 0.002 \| \| LVEF \| 0.986 (0.975, 0.998) \| 0.020 \| \| \| \| FBG \| 1.049 (0.999, 1.101) \| 0.055 \| \| \| \| HbA1c \| 1.125 (1.004, 1.259) \| 0.042 \| 1.229 (1.043, 1.447) \| 0.014 \| \| P2Y12 inhibitor \| 1.808 (1.325, 2.468) \| <0.001 \| 1.843 (1.342, 2.533) \| <0.001 \| \| DAPT duration \| 1.724 (1.319, 2.252) \| <0.001 \| 1.704 (1.302, 2.232) \| <0.001 \| ## 3.7. Sensitivity analysis We further analyzed different subgroups to evaluate the independent association of DAPT duration with composite bleeding events. According to Figure 3, the significant predictive effect of DAPT duration on composite bleeding events was primarily reflected in the subgroups of patients aged < 70 years and patients aged ≥70 years, patients who were men, patients with a history of hypertension, patients with and without a history of DM, patients who experienced ischemic stroke, and patients with a history of PAD, Hb ≥120 g/L and <120 g/L, LVEF <$50\%$, eGFR <60 ml/min/1.73 m2 and ≥60 ml/min/1.73 m2, BMI <25 kg/m2, HbA1c <$6.5\%$ and ≥$6.5\%$, SBP ≥120 mmHg and <120 mmHg, and heart rate ≥80 bpm and <80 bpm. Female patients appeared to have a higher predictive value for DAPT duration [HR ($95\%$CI) for women vs. men = 2.329 (1.636–3.315) vs. 1.000 (0.611–1.636), p for interaction =0.025]. Figure 3: Forest plot investigating th e association between the DA PT duration and composite bleeding events in different subgroups. DAPT, dual antiplatelet therapy; DM, diabetes mellitus; PAD, peripheral artery disease; Hb, hemoglobin; LVEF, left ventricular ejection fraction; eGFR, estimated glomerular filtration rate; BMI, body mass index; HbA1c, glycosylated hemoglobin A1c; SEP, systolic blood pressure; Cl, confidence interval. ## 4. Discussion In this cohort of Chinese patients with ACS who were treated with emergency PCI with drug-eluting stents, the prolonged DAPT group had a significantly higher risk of composite bleeding events than the standard DAPT group. DAPT duration was an independent predictor of composite bleeding events. However, we did not find a statistically significant difference in the prevalence of MACCEs between the two groups, and DAPT duration was not an independent predictor of MACCEs. To the best of our knowledge, this study is the first to examine the benefits and risks of prolonged DAPT in Chinese patients with ACS who underwent emergency PCI with DES. STEMI, NSTEMI, and UA are clinical diagnoses caused by acute myocardial ischemia, which is referred to as ACS. Several studies confirmed that the pathophysiology of ACS includes coronary vulnerable plaque rupture, vasospasm, and vascular endothelial dysfunction caused by oxidative damage and inflammation, which result in platelet activation, adhesion, aggregation, and secondary thrombosis [21]. Patients who have experienced an ACS are at a higher risk of having recurrent ischemic events. The EPICOR Asia study enrolled 12,922 patients with ACS [mostly from China ($63.6\%$)]: 6,616 ($51.2\%$) patients with STEMI, 2,570 ($19.9\%$) patients with NSTEMI, and 3,736 ($28.9\%$) patients with UA. The study showed that all-cause mortality during the 2-year follow-up period was $5.2\%$, and the composite endpoint of death, MI, and stroke during the 2-year follow-up period was $8.4\%$ [22]. In the present study, all-cause mortality during the 24-month follow-up period was $3.1\%$, and the composite endpoint of death, non-fatal MI, non-fatal ischemic stroke, and ischemia-driven revascularization during the 24-month follow-up period was $5.9\%$. The lower risk of an adverse event in the present study may be due to the higher proportion of UA. Therefore, enhanced antiplatelet therapy is important for preventing and treating thrombosis. In recent years, new P2Y12 inhibitors (prasugrel and ticagrelor) have been affirmed by many large-scale clinical studies and have been recommended by guidelines due to their more powerful antiplatelet effect. However, new P2Y12 inhibitors (prasugrel and ticagrelor) may increase the incidence of bleeding events in East Asians as opposed to clopidogrel, according to various studies. A Korean study enrolled 4,421 patients (637 patients prescribed prasugrel and 3,784 patients prescribed clopidogrel) with acute MI who underwent successful revascularization. No statistically significant difference was detected between prasugrel and clopidogrel in the composite ischemic events of cardiac death, MI, stroke, or target vessel revascularization at 6 months (2.4 vs. $2.9\%$, $$p \leq 0.593$$). Compared with clopidogrel, prasugrel increased the presence of nosocomial thrombolysis in myocardial infarction (TIMI) major or minor bleeding (5.3 vs. $2.7\%$, $$p \leq 0.015$$) [23]. Meanwhile, a Korean study enrolled 800 patients with ACS accepted for PCI management. No statistically significant difference between ticagrelor and clopidogrel was found in the composite ischemic events of cardiac death, MI, or stroke at 12 months (9.2 vs. $5.8\%$; HR, 1.62; $$p \leq 0.593$$). Compared with clopidogrel, ticagrelor increased the prevalence rates of clinically significant bleeding (11.7 vs. $5.3\%$; HR, 2.26; $$p \leq 0.002$$), major bleeding (7.5 vs. $4.1\%$, $$p \leq 0.04$$), and fatal bleeding (1.0 vs. $0.0\%$, $$p \leq 0.04$$) [24]. As in the current research, the proportion of clopidogrel treatment was up to $86.9\%$, and the proportion of ticagrelor was only $13.1\%$ during the DAPT period. Furthermore, the proportion of ticagrelor in the two groups was similar (14.3 vs. $12.1\%$, $$p \leq 0.120$$). The multivariate analysis based on the Cox proportional hazard model showed that the use of ticagrelor was insignificantly associated with MACCEs, cardiac death, or BARC 3 or 5 bleeding events. However, the use of ticagrelor was an independent predictor of composite bleeding events. Recent guidelines in Europe and the United States advocate DAPT combined with aspirin and a P2Y12 inhibitor (clopidogrel, prasugrel, and ticagrelor) for up to 12 months after ACS to lower the risk of ischemic events, such as recurrent MI and ST (6–8). However, the risk of target lesion failure is still 2–$4\%$ annually after 1 year of DAPT [25]. The guidelines in Europe recommend DAPT at >12 months in patients at a high risk of ischemic events and without an increased risk of major bleeding (Class IIa indication) [6]. The EPICOR Asia trial showed that $78.8\%$ of patients with NSTEMI continued DAPT for over 12 months [26]. In the present study, $55.0\%$ of patients with ACS continued DAPT at 12–24 months. A study enrolled 9,961 patients (5,020 patients accepted prolonged DAPT and 4,941 patients accepted standard DAPT) after they underwent successful revascularization with DES, and the study showed that the incidence of MACCEs (4.3 vs. $5.9\%$; OR, 0.71; $p \leq 0.001$), MI (2.1 vs. $4.1\%$; OR, 0.47; $p \leq 0.001$), and ST (0.4 vs. $1.4\%$; OR, 0.29; $p \leq 0.001$) was significantly lower in the prolonged DAPT group compared with the standard group. However, the incidence of severe or moderate bleeding events was elevated with prolonged DAPT (2.5 vs. $1.6\%$; $$p \leq 0.001$$) [10]. The PEGASUS-TIMI 54 trial demonstrated that TIMI severe bleeding events were much more prevalent in the prolonged group (2.5 vs. $1.1\%$; OR, 2.36; $p \leq 0.001$); however, the composite ischemic events of cardiovascular death, MI, or stroke were less common in the prolonged DAPT group (7.9 vs. $9.6\%$; OR, 0.80; $$p \leq 0.001$$) [27]. EPICOR Asia showed that the composite endpoint occurred less frequently in the prolonged DAPT group (3.1 vs. $10.6\%$), and only four patients had severe bleeding events in the prolonged DAPT group [26]. ARCTIC interruption showed that the incidence of endpoints had no statistically significant difference between the standard DAPT and prolonged DAPT groups (4.0 vs. $4.0\%$; HR, 1.17; $$p \leq 0.58$$), and either minor or severe bleeding was much more common in the prolonged DAPT group than in the standard DAPT group (2.0 vs. $1.0\%$; OR, 0.26; $$p \leq 0.04$$) [12]. To evaluate the benefit and risk of prolonged DAPT for predicting MACCEs and composite bleeding events in patients with ACS who underwent emergency PCI with DES, we analyzed a cohort of 2,249 Chinese patients with ACS and found no statistically significant difference between the standard DAPT and prolonged DAPT groups regarding the incidence of MACCEs. However, the prolonged DAPT group experienced significantly more composite bleeding events than the standard DAPT group, and DAPT duration was an independent predictor of composite bleeding events. Compared with the EPICOR Asia study, prolonged DAPT duration did not reduce the incidence of MACCEs in the present study due to enrolling a higher proportion of UA. The risk of ischemia and bleeding needs to be evaluated when prolonging DAPT. A previous study showed that the risk factors for ischemic events included older age, ACS, previous MI, complex coronary artery disease (≥3 stents implanted, ≥3 lesions treated, LM, bifurcation, CTO, and previous ST on antiplatelet treatment), DM, PAD, and chronic kidney disease (CKD) [3, 7]. The risk factors for bleeding events included a previous history of intracerebral hemorrhage or gastrointestinal bleeding, a previous history of moderate or severe ischemic stroke, a history of consuming oral anticoagulants, being women, being of older age, patients with low weight, patients with CKD, patients with liver failure, patients with anemia, and patients with long-term treatment with steroids or non-steroidal anti-inflammatory drugs (NSAIDs) [3, 7]. As part of the current research, we analyzed a cohort of 2,249 Chinese patients with ACS. The independent predictors of MACCEs included age, Hb, LVEF, HbA1c, LDL-C, LM disease, and the number of stents implanted. Meanwhile, the independent predictors of composite bleeding events included age, DM, Hb, HbA1c, and use of ticagrelor. In the subgroup analysis, we found that prolonged DAPT could significantly reduce the risk of MACCEs in the three-vessel disease and being men compared with standard DAPT. Meanwhile, we discovered that prolonged DAPT had no effect on the risk of composite bleeding events in women without a prior history of hypertension, patients with LVEF ≥$50\%$, and patients with BMI ≥25 kg/m2 compared with standard DAPT. This study has several limitations that should be acknowledged. First, this study is an observational, retrospective, single-center study. Therefore, this trial could not determine the benefits or risks of prolonged DAPT in Chinese patients with ACS after emergency PCI with DES. Second, the sample size and the follow-up period might be insufficient. Third, our study's exclusion of patients with ACS who underwent elective PCI or were treated with a drug-coated balloon (DCB) may have limited the generalizability of our findings to patients with ACS who underwent primary PCI with DES and may have resulted in selection bias. ## 5. Conclusion Compared with the standard DAPT group, the prolonged DAPT group had a statistically significant higher prevalence of composite bleeding events. However, the incidence of MACCEs showed no statistically significant difference between the two groups. In the subgroup analysis, prolonged DAPT significantly reduced the risk of MACCEs in the three-vessel disease and male subgroups. In contrast, prolonged DAPT did not increase the risk of composite bleeding events in men with no prior history of hypertension or DM, LVEF ≥$50\%$, and BMI ≥25 kg/m2 subgroups. Additional large-scale, prospective cohort, multicenter studies with a sizable sample and a prolonged follow-up period will be needed to support our findings. ## Data availability statement The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author. ## Ethics statement The studies involving human participants were reviewed and approved by the Academic Committee of the First Affiliated Hospital of Xi'an Jiaotong University. The patients/participants provided their written informed consent to participate in this study. ## Author contributions YZ and NG designed and drafted the manuscript. Y-bL, YZ, and ZZ carried out the cohort's follow-up and gathered the data. CC and F-FN evaluated the data and edited the text. YZ, ZZ, and F-FN did the planning and coordination of the research. The final manuscript was reviewed and approved by all authors. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcvm.2023.1080673/full#supplementary-material ## References 1. Eisen A, Giugliano RP, Braunwald E. Updates on acute coronary syndrome: a review. JAMA Cardiol (2016) 1 718-30. DOI: 10.1001/jamacardio.2016.2049 2. Benjamin EJ, Virani SS, Callaway CW, Chamberlain AM, Chang AR, Cheng S. Heart disease and stroke statistics-2018 update: a report from the american heart association. Circulation (2018) 137 e67-492. DOI: 10.1161/CIR.0000000000000573 3. Hammer Y, Iakobishvili Z, Hasdai D, Goldenberg I, Shlomo N, Einhorn M. Guideline-recommended therapies and clinical outcomes according to the risk for recurrent cardiovascular events after an acute coronary syndrome. J Am Heart Assoc (2018) 7 e009885. DOI: 10.1161/JAHA.118.009885 4. Chan Pin Yin D, Ten Berg JM. Long-term residual cardiovascular risk after acute coronary syndrome: antithrombotic treatment options. Neth Heart J (2022) 30 38-46. DOI: 10.1007/s12471-021-01604-4 5. Byrne RA, Joner M, Kastrati A. Stent thrombosis and restenosis: what have we learned and where are we going? The Andreas Grüntzig Lecture ESC 2014. Eur Heart J (2015) 36 3320-31. DOI: 10.1093/eurheartj/ehv511 6. Collet JP, Thiele H, Barbato E, Barthélémy O, Bauersachs J, Bhatt DL. 2020 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J (2021) 42 1289-367. DOI: 10.1093/eurheartj/ehaa575 7. Ibanez B, James S, Agewall S, Antunes MJ, Bucciarelli-Ducci C, Bueno H. 2017 ESC Guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation: the task force for the management of acute myocardial infarction in patients presenting with ST-segment elevation of the European Society of Cardiology (ESC). Eur Heart J (2018) 39 119-77. DOI: 10.1093/eurheartj/ehx393 8. Levine GN, Bates ER, Bittl JA, Brindis RG, Fihn SD, Fleisher LA. 2016 ACC/AHA guideline focused update on duration of dual antiplatelet therapy in patients with coronary artery disease: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. J Am Coll Cardiol (2016) 68 1082-115. DOI: 10.1016/j.jacc.2016.03.513 9. Fox KA, Carruthers KF, Dunbar DR, Graham C, Manning JR, De Raedt H. Underestimated and under-recognized: the late consequences of acute coronary syndrome (GRACE UK-Belgian Study). Eur Heart J (2010) 31 2755-64. DOI: 10.1093/eurheartj/ehq326 10. Mauri L, Kereiakes DJ, Yeh RW, Driscoll-Shempp P, Cutlip DE, Steg PG. Twelve or 30 months of dual antiplatelet therapy after drug-eluting stents. N Engl J Med (2014) 371 2155-66. DOI: 10.1056/NEJMoa1409312 11. Bonaca MP, Bhatt DL, Cohen M, Steg PG, Storey RF, Jensen EC. Long-term use of ticagrelor in patients with prior myocardial infarction. N Engl J Med. (2015) 372 1791-800. DOI: 10.1056/NEJMoa1500857 12. Collet JP, Silvain J, Barthélémy O, Rangé G, Cayla G, Van Belle E. Dual-antiplatelet treatment beyond 1 year after drug-eluting stent implantation (ARCTIC-Interruption): a randomized trial. Lancet (2014) 384 1577-85. DOI: 10.1016/S0140-6736(14)60612-7 13. Lee CW, Ahn JM, Park DW, Kang SJ, Lee SW, Kim YH. Optimal duration of dual antiplatelet therapy after drug-eluting stent implantation: a randomized, controlled trial. Circulation (2014) 129 304-12. DOI: 10.1161/CIRCULATIONAHA.113.003303 14. Helft G, Steg PG, Le Feuvre C, Georges JL, Carrie D, Dreyfus X. Stopping or continuing clopidogrel 12 months after drug-eluting stent placement: the OPTIDUAL randomized trial. Eur Heart J. (2016) 37 365-74. DOI: 10.1093/eurheartj/ehv481 15. Goto S, Toda E. Antiplatelet therapy after coronary intervention in Asia and Japan: the Asian perspective of antiplatelet intervention. Hamostaseologie (2009) 29 321-5. DOI: 10.1055/s-0037-1617131 16. Mak KH, Bhatt DL, Shao M, Hankey GJ, Easton JD, Fox KA. Ethnic variation in adverse cardiovascular outcomes and bleeding complications in the Clopidogrel for High Atherothrombotic Risk and Ischemic Stabilization, Management, and Avoidance (CHARISMA) study. Am Heart J (2009) 157 658-65. DOI: 10.1016/j.ahj.2008.08.031 17. Mehran R, Rao SV, Bhatt DL, Gibson CM, Caixeta A, Eikelboom J. Standardized bleeding definitions for cardiovascular clinical trials: a consensus report from the Bleeding Academic Research Consortium. Circulation (2011) 123 2736-47. DOI: 10.1161/CIRCULATIONAHA.110.009449 18. Diagnosis and classification of diabetes mellitus. Diabetes Care (2011) 34 S62-9. DOI: 10.2337/dc11-S062 19. Levey AS, Coresh J, Greene T, Stevens LA, Zhang YL, Hendriksen S. Using standardized serum creatinine values in the modification of diet in renal disease study equation for estimating glomerular filtration rate. Ann Intern Med. (2006) 145 247-54. DOI: 10.7326/0003-4819-145-4-200608150-00004 20. Alfonso F, Byrne RA, Rivero F, Kastrati A. Current treatment of in-stent restenosis. J Am Coll Cardiol (2014) 63 2659-73. DOI: 10.1016/j.jacc.2014.02.545 21. Patel VB, Topol EJ. The pathogenesis and spectrum of acute coronary syndromes: from plaque formation to thrombosis. Cleve Clin J Med (1999) 66 561-71. DOI: 10.3949/ccjm.66.9.561 22. Huo Y, Lee SW, Sawhney JPS, Kim HS, Krittayaphong R, Pocock SJ. Two-year outcomes post-discharge in Asian patients with acute coronary syndrome: findings from the EPICOR Asia study. Int J Cardiol (2020) 315 1-8. DOI: 10.1016/j.ijcard.2020.05.022 23. Park KH, Jeong MH, Kim HK, Ahn TH, Seung KB, Oh DJ. Comparison of prasugrel versus clopidogrel in Korean patients with acute myocardial infarction undergoing successful revascularization. J Cardiol (2018) 71 36-43. DOI: 10.1016/j.jjcc.2017.05.003 24. Park DW, Kwon O, Jang JS, Yun SC, Park H, Kang DY. Clinically significant bleeding with ticagrelor versus clopidogrel in korean patients with acute coronary syndromes intended for invasive management: a randomized clinical trial. Circulation (2019) 140 1865-77. DOI: 10.1161/CIRCULATIONAHA.119.041766 25. Iqbal J, Serruys PW, Silber S, Kelbaek H, Richardt G, Morel MA. Comparison of zotarolimus- and everolimus-eluting coronary stents: final 5-year report of the RESOLUTE all-comers trial. Circ Cardiovasc Interv (2015) 8 e002230. DOI: 10.1161/CIRCINTERVENTIONS.114.002230 26. Zou Y, Yang S, Wang S, Lv B, Xiu L, Li L. Prolonged dual antiplatelet therapy in patients with non-ST-segment elevation myocardial infarction: 2-year findings from EPICOR Asia. Clin Cardiol (2020) 43 346-54. DOI: 10.1002/clc.23322 27. Dellborg M, Bonaca MP, Storey RF, Steg PG, Im KA, Cohen M. Efficacy and safety with ticagrelor in patients with prior myocardial infarction in the approved European label: insights from PEGASUS-TIMI 54. Eur Heart J Cardiovasc Pharmacother (2019) 5 200-6. DOI: 10.1093/ehjcvp/pvz020
--- title: 'Risk factor-based screening compared to universal screening for gestational diabetes mellitus in marginalized Burman and Karen populations on the Thailand-Myanmar border: An observational cohort' authors: - Janna T. Prüst - Tobias Brummaier - Mu Wah - Htay Htay Yee - Nyo Nyo Win - Mupawjay Pimanpanarak - Aung Myat Min - Mary Ellen Gilder - Nay Win Tun - Onaedo Ilozumba - Basirudeen Syed Ahamed Kabeer - Annalisa Terranegra - Francois Nosten - Sue J. Lee - Rose McGready journal: Wellcome Open Research year: 2023 pmcid: PMC9976631 doi: 10.12688/wellcomeopenres.17743.2 license: CC BY 4.0 --- # Risk factor-based screening compared to universal screening for gestational diabetes mellitus in marginalized Burman and Karen populations on the Thailand-Myanmar border: An observational cohort ## Abstract Background: Gestational diabetes mellitus (GDM) contributes to maternal and neonatal morbidity. As data from marginalized populations remains scarce, this study compares risk-factor-based to universal GDM screening in a low resource setting. Methods: This is a secondary analysis of data from a prospective preterm birth cohort. Pregnant women were enrolled in the first trimester and completed a 75g oral glucose tolerance test (OGTT) at 24-32 weeks' gestation. To define GDM cases, Hyperglycaemia and Adverse Pregnancy Outcomes (HAPO trial) criteria were used. All GDM positive cases were treated. Sensitivity and specificity of risk-factor-based selection for screening (criteria: age ≥30y, obesity (Body mass index (BMI) ≥27.5kg/m 2), previous GDM, 1 st degree relative with diabetes, previous macrosomia (≥4kg), previous stillbirth, or symphysis-fundal height ≥90th percentile) was compared to universal screening using the OGTT as the gold standard. Adverse maternal and neonatal outcomes were compared by GDM status. Results: GDM prevalence was $13.4\%$ ($\frac{50}{374}$) ($95\%$ CI: 10.3-17.2). Three quarters of women had at least one risk factor ($$n = 271$$ women), with $\frac{37}{50}$ OGTT positive cases correctly identified: sensitivity $74.0\%$ (59.7-85.4) and specificity $27.8\%$ (3.0-33.0). Burman women (self-identified) accounted for $29.1\%$ of the cohort population, but $38.0\%$ of GDM cases. Percentiles for birthweight ($$p \leq 0.004$$), head circumference ($$p \leq 0.002$$), and weight-length ratio ($$p \leq 0.030$$) were higher in newborns of GDM positive compared with non-GDM mothers. $21.7\%$ ($\frac{75}{346}$) of newborns in the cohort were small-for-gestational age (≤10 th percentile). In Burman women, overweight/obese BMI was associated with a significantly increased adjusted odds ratio 5.03 ($95\%$ CI: 1.43-17.64) for GDM compared with normal weight, whereas in Karen women, the trend in association was similar but not significant (OR 2.36; $95\%$ CI 0.95-5.89). Conclusions: Risk-factor-based screening missed one in four GDM positive women. Considering the benefits of early detection of GDM and the limited additional cost of universal screening, a two-step screening program was implemented. ## Amendments from Version 1 Version 2 of the manuscript was submitted in response to the reviewer comments on version 1. Comments and suggestions by the reviewers were considered to revise the manuscript or rebuked were appropriate (see response to reviewer comments). Most notably all references to environmental factors (i.e., seasonality of GDM diagnosis) has been removed from the manuscript. This also resulted in the removal of Figure 3a and Figure 3b from version 2 of the manuscript. The abstract was amended to include all risk factors that were screened for the risk-factor based screening procedure. Sample size justification and a power estimation was added. Some items (e.g., Asian BMI definitions, exact nature of the risk-factor based screening or rational to include symphysis-fundal height (SFH) in the analysis) were clarified. Table 1 was complemented with the addition of number of pregnant women presenting with a SFH ≥90 th centile and figures describing gestational weight gain; how these compare between non-GDM and GDM women was also added. Centiles for head circumference, length, and weight for length ratio, as published by the Intergrowth 21 st consortium, were added to Table 2, together with the number of neonates admitted to the special care baby unit (SCBU) and number of neonates diagnosed with hypoglycaemia. As suggested by the reviewer, a paragraph presenting results of oral glucose tolerance test (OGTT) results was added. A sentence on historical context of GDM screening was added to the discussion and the fact the there is no international consensus on the best screening approach exists, together with the ongoing debate whether screening criteria derived from high-resource settings are applicable to low-resource settings is now discussed in more detail. Strengths and limitations were expanded to address comments from the reviewers. Lastly, the conclusion was amended to highlight the translational impact of this analysis. ## Introduction Gestational diabetes mellitus (GDM) is rising in tandem with obesity globally, including in South- and South-East Asia 1. The prevalence of GDM in Thailand is estimated between $6.1\%$ and $29.2\%$ 1, 2. In Myanmar, there is insufficient data to provide reliable estimations of the GDM prevalence 1. Detection of GDM is important as it is associated with neonatal macrosomia, neonatal hypoglycaemia and an increased risk for birth complications, such as shoulder dystocia and the need for caesarean section 3– 5. Furthermore, GDM is associated with an increased risk of preeclampsia, and entails a tenfold risk of developing type II diabetes and doubles the risk of cardiovascular events later in life 6, 7. In absolute numbers more women are diagnosed with GDM in low- and middle-income countries (LMIC although relative estimates are similar between LMIC and high-income countries (HIC): $13.5\%$ and $13.4\%$, respectively) 8. Within HIC, migrant women have a higher risk for GDM and associated adverse birth outcomes, but this is poorly evidenced for migrants in LMIC 9. In South-East Asia domestic as well as international migration is a dominant feature and access to health care for migrants is problematic 10, 11. While most women receive some form of antenatal care (ANC), screening for GDM is often not available 12, 13. In addition, awareness of GDM is limited, as are adequate protocols and tools to monitor blood glucose, which hinders best-practice management 13, 14. Officially Thailand has approximately 2 million migrant workers predominantly from Myanmar, as well as an unknown number of undocumented migrants. Shoklo Malaria Research Unit (SMRU) has provided health care to both the refugee and migrant populations residing along the Thailand-Myanmar border. In the pregnant migrant population attending SMRU antenatal (ANC) clinics, the nutrition transition has been marked by a two-fold increase in first trimester overweight in just over a decade, aggravated by limited awareness of healthy diets and lifestyle 15, 16. These trends in marginalized populations are worrying given the greater risk of cardiometabolic effects occurring at lower BMI in Asians than in white Europeans 17. In a meta-analysis, Lee et al. described a GDM prevalence of $11.5\%$ in Asian women and identified the following risk factors: multiparity, previous GDM, or pregnancy-induced hypertension (PIH), a family history of GDM and an increased maternal body mass index (BMI ≥25kg/m 2) 18. An obstetric history of preterm birth, macrosomia, stillbirth, or an infant with congenital anomalies are also recognised GDM risk factors 18. GDM diagnosis and management improves maternal and perinatal outcomes, although this is largely evidenced from HIC 13, 19. Both universal and risk-factor-based screening are common practices, with no international consensus about best practice 2, 20, 21. In 2011–2012, one of the first surveys conducted in a refugee camp reported a GDM prevalence of $10.1\%$ ($95\%$ CI 6.2-$14.0\%$) on the Thailand-Myanmar border with GDM being significantly associated with increased maternal age and parity, and low literacy 20. Although the proportion of caesarean section and obesity (BMI ≥27.5kg/m 2) were higher among women with GDM, this difference was not significant 20. In the low-resource setting of the refugee camp, the decision at that time was to commence efforts to screen for GDM based on risk factors using the Hyperglycaemia and Adverse Pregnancy Outcomes (HAPO) criteria 22. SMRU implemented this approach in all its antenatal care clinics on the border in 2018. The study presented here aimed to evaluate the performance of two screening methods for GDM detection: risk-factor-based identification of pregnant women who were then screened by an OGTT, which was routinely used in antenatal care clinics for migrant women, to universal screening by OGTT. Within this cohort, risk factors for GDM were examined and adverse maternal and neonatal outcomes were evaluated in women with and without GDM. ## Ethical approval The study was approved by the ethics committee of the Faculty of Tropical Medicine, Mahidol University, Bangkok, Thailand (Ethics Reference: TMEC 15–062, initial approval 1 December 2015), the Oxford Tropical Research Ethics Committee (Ethics Reference: OxTREC: 33–15, initial approval 16 December 2015) and reviewed by the local Tak Province Community Ethics Advisory Board. The study was conducted in full conformity with the Declaration of Helsinki and followed regulations of the ICH Guidelines for Good Clinical Practice. ## Study design This is a secondary analysis of data from an observational preterm birth cohort study with data collected prospectively between September 2016 and February 2019 in women enrolled in their first trimester of pregnancy (ClinicalTrials.gov Identifier: NCT02797327) with GDM screening occurring from December 2016 to November 2018. ## Study setting SMRU was established more than three decades ago and combines research and humanitarian work that serves the migrant population alongside the Thailand-Myanmar border. To be accessible within these communities, which largely depend on below minimum wage jobs, SMRU operates free-of-charge walk-in clinics offering universal antenatal care, as well as 24-hour delivery services, led by trained personnel originating from the local population. At the same clinics, women may be invited to participate in research. The study was explained to all pregnant women attending SMRU ANC clinics in the first trimester and they were invited to participate if they met the study inclusion criteria and enrolled if consent was forthcoming. Informed consent was obtained in the form of a signature or in the event of an illiterate participant by thumbprint coupled with a confirmatory signature by an impartial literate witness. ## Sample size A detailed description of the study protocol and SMRU routine ANC procedures are available elsewhere 23. Briefly, women were followed fortnightly throughout pregnancy, at delivery, and in the postpartum period. The planned sample size of 400 in the original cohort study was based on estimated preterm birth rates (of approximately $8\%$) and on the following inclusion criteria: a viable, singleton first trimester pregnancy and an unremarkable medical and obstetric history e.g., no history of caesarean section. For this secondary analysis of the original cohort to determine appropriateness of GDM risk-factor-based screening, additional exclusion criteria were miscarriage prior to GDM screening, maternal death, lost to follow-up, withdrawal of consent (primary cohort), and if OGTT was performed late (gestational age (GA) ≥33 weeks) or not done at all. Women who did not complete follow-up to delivery were replaced as permitted in the original protocol. At an expected GDM rate of $10\%$, a sample size of 400 is expected to be sufficient to determine population prevalence 20. ## Study variables Baseline characteristics, regular prenatal symphysis-fundal height (SFH) measurements, blood pressure, weight, and assessment of gestation by ultrasound, as well as birth outcomes, were collected by trained ANC staff and midwives in accordance with the study protocol. GA was estimated by crown rump length measured by first trimester ultrasound 24. Body-mass index (BMI) definitions followed recommendations for Asian BMI groups: underweight <18.5 kg/m 2; normal weight 18.5 to <23 kg/ m 2; overweight 23 to <27.5 kg/m 2; obese ≥27.5 kg/m 2 17. While the study protocol specified GDM screening with OGTT at 24–26 weeks of gestation, the HAPO study target time for testing was at 28 weeks (24–32 weeks) 22. Therefore, OGTTs to 32 weeks of gestation were included in this analysis. In women with a history of GDM, an OGTT was performed as early as possible in pregnancy and repeated at 24–26 weeks if previously negative. GDM diagnosis was based on HAPO trial cut-offs: a fasting capillary blood glucose measurement of ≥92mg/dL, ≥180mg/dL one hour or ≥153mg/dL two hours after ingestion of 75g glucose were considered positive 22. ## Risk-factor based screening In 2018, risk-factor-based screening for GDM commenced at SMRU clinics. The risk factors were based on a survey in Karen and Burmese women in a SMRU refugee clinic screened at 24-28 weeks with a 75-gram OGTT using the HAPO trial cut-offs, where prevalence was $10.1\%$ ($95\%$ CI 6.1-14.0). Risk factors in positive cases and review of recommendations from UK and Australia, both of which have populations of South-East Asian women, and Thailand resulted in the final list 20. The risk factors for GDM screening required at least one positive finding among the following 10 criteria: (i) age ≥30 years, (ii) obesity (BMI ≥27.5kg/m 2, the WHO definition for Asian populations) 17, (iii) GDM in a previous pregnancy, (iv) family history (1 st degree relative) of diabetes mellitus (although this is of reduced sensitivity in LMIC as access to diabetes screening is limited), (v) previous macrosomia (≥4kg), (vi) previous stillbirth, (vii) SFH ≥90th percentile, (viii) previous caesarean section regardless of birth-weight, (ix) 2+/3+ glucose on a urine dipstick test, or (x) polycystic ovarian syndrome (PCOS). The following criteria were not included in the analysis: women with a previous caesarean section, as they were excluded from the original study protocol, PCOS, as it was not encountered, and glucosuria, as there was no routine screening, leaving seven criteria. ## Maternal and Neonatal Outcomes In resource-limited settings, assessment of the uterus size by SFH measurement as a proxy for fetal size has been suggested as a first level screening tool for fetal growth assessment. SFH measurement is a straightforward and inexpensive method, but its precision is controversial 3. A previously published bespoke SFH growth curve has been in use for more than 10 years in the pregnant population along the Thailand-Myanmar border 25; however, whether increased SFH using this local growth curve is a useful addition to the identification of GDM (macrosomia is a common adverse effect of GDM) has not been assessed. Serial SFH measurements were included from 16 weeks of gestation on a two-weekly basis and data was examined using both, local population and international centiles 25, 26. Gestational weight gain was defined as the final maternal weight measured not more than four weeks prior to birth, minus the weight measured at the first antenatal visit. For women with a normal BMI at enrolment (between 18.50 and 24.99kg/m 2), Intergrowth-21 st standard percentiles for each weight measurement from ≥26 weeks and ≤40 weeks of gestation were calculated 27. Neonatal anthropometry (i.e., birthweight, head circumference, and length) were only considered if measured within 72 hours of birth. If women gave birth at SMRU, the neonate was weighed on a digital SECA 354 scale (precision 5g) with weekly calibration. Percentiles and z-scores for neonatal anthropometry were calculated using standards as published by the Intergrowth-21 st Project 28. Born too small or large for GA (SGA, LGA) were defined as ≤10 th and ≥90 th percentile, respectively. Standard management of infants admitted to the special care baby unit included measurement of blood glucose and treatment for neonates with blood glucose below 45 mg/dL ## GDM management If GDM was diagnosed, all women were counselled about lifestyle modification (e.g., diet and exercise) and, due to the unavailability of glucose self-monitoring in the population, the status of GDM control was monitored weekly or every two weeks at the clinic. Monitoring was as follows: women with GDM were asked to attend fasting and blood glucose was checked on arrival; then women ate a typical meal and were retested after one hour (post-prandial) with the desired value of <90 mg/dL (fasting) and <140 mg/dL (after one hour) for satisfactory control. Treatment was provided either directly or if non-pharmacologic interventions led to insufficient glucose control, with metformin as the first choice and glibenclamide as an additional oral agent. Due to the lack of home-based glucose monitoring options and the absence of adequate storage facilities, insulin is rarely prescribed in this population. ## Statistical analysis Data were analysed using Stata, version 17.0 (TX, USA) (Stata, RRID:SCR_012763, https://www.stata.com/). Normally distributed continuous data were presented as means with standard deviation (SD) and non-normally distributed data as medians with interquartile range (IQR). Baseline characteristics as well as birth outcomes were compared between women with and without GDM. For continuous variables, the Student’s t-test or Mann-Whitney U test were used, and categorical variables were compared using the Fisher’s exact or Chi-square test. Univariate associations were quantified using logistic regression. To evaluate the predictive ability of the risk factors used in the current screening approach to identify women with GDM, all risk factors were combined into one logistic regression model, using GDM as the outcome. The sensitivity and specificity of risk-factor-based screening criteria was calculated using OGTT as the gold standard. An any positive test principle (i.e., if any of the GDM risk factors stated above were positive, an OGTT was performed) was the basis for this assessment. For further in-depth analysis and to identify risks and potential risk groups for GDM in this population, age (30 or older, vs. all others), smoking (yes/no), ethnicity (Karen and Burman), and BMI groups underweight, normal weight (reference group) and overweight/obese were explored using interaction terms and logistic regression modelling. ## Results Following exclusions, $87.4\%$ ($\frac{374}{428}$) of pregnant women from the original cohort were available for analysis (Figure 1). Of these, $13.4\%$ ($\frac{50}{374}$, $95\%$ CI 10.3-17.2), were diagnosed with GDM by OGTT. The median number of antenatal care visits was 16 (IQR 15-17). Baseline maternal characteristics of women with and without GDM were compared (Table 1). Women with GDM were significantly more likely to have had previous GDM ($4.0\%$ vs. 0, $p \leq 0.001$) and postpartum hypertension ($4.0\%$ vs. $0.3\%$, $$p \leq 0.006$$) and less likely to have had previous preterm labour ($0\%$ vs. $7.41\%$, $$p \leq 0.047$$). A family history of diabetes was rarely reported ($$n = 6$$) by women irrespective of GDM status. Figure 1.: Flow diagram of participant selection.Abbreviations: GDM gestational diabetes mellitus, OGTT oral glucose tolerance test. Sudden death due to mixed mitral valve disease at seven months gestation.* TABLE_PLACEHOLDER:Table 1. Overall, 23 women ($6.1\%$) were obese (BMI ≥27.5kg/m 2). In the group of women who self-identified as being of Burman descent the GDM prevalence was $17.4\%$ ($\frac{19}{109}$) compared to $11.7\%$ ($\frac{29}{247}$) in women of Karen descent and $11.1\%$ ($\frac{2}{18}$) in women of other ethnicities. Burman women accounted for $29.1\%$ of the cohort population, but $38.0\%$ of GDM cases (Table 1). There were more women with GDM with an SFH ≥90 th centile during pregnancy with gestational week ≥24, $68.0\%$ vs. $52.8\%$, $$p \leq 0.044$$ (Table 2). In particular, from about 224 days (32 weeks) onwards, women with GDM appeared to have larger SFH when compared with women without GDM (Figure 2). ## Birth outcomes Newborns from mothers with GDM were heavier (mean birthweight (SD): 3096g [408] vs. 2952g [398], $$p \leq 0.019$$), and nearly five times more likely to be born large for gestational age ($6.1\%$ ($\frac{3}{49}$) vs. $1.3\%$ ($\frac{4}{297}$), OR 4.78, $95\%$ CI 1.04-22.1) (Table 2). They were also more likely to be in a higher percentile for birthweight and head circumference, adjusted for GA and sex: median [IQR]: 40.5 [16.3, 61.0] vs. 23.2 [11.2, 43.9], $$p \leq 0.004$$), and 30.6 [12.8, 60.5] vs. 19.3 [6.69, 37.6],, $$p \leq 0.002$$ respectively. Infants born to mothers with GDM had a higher weight-length ratio (mean (SD): $6.4\%$ WLR (0.7) vs. $6.1\%$ w/l (0.7), $$p \leq 0.010$$), Table 2. Overall, the proportion of SGA was relatively high ($21.7\%$, $\frac{75}{346}$) with a lower proportion of SGA in the GDM positive group which was not statistically significant ($14.3\%$ ($\frac{7}{49}$) vs. $22.9\%$ ($\frac{68}{297}$, $$p \leq 0.175$$). Other adverse birth complications such as stillbirth ($0\%$, $\frac{0}{50}$ of GDM positive; $1.2\%$, $\frac{4}{324}$ of GDM negative), and preterm birth ($2.0\%$, $\frac{1}{50}$ in GDM positive; $5.2\%$, $\frac{17}{324}$ in GDM negative) were low. ## OGTT test results As expected, the absolute blood sugar levels (BSL) levels were higher in the GDM positive group (Table 3). Of the women with GDM, $88.0\%$ ($\frac{44}{50}$) had only one of the three glucose measurements above the cut-off, $10\%$ ($\frac{5}{50}$) had two of three glucose measurements above the defined threshold and in only one study participant ($\frac{1}{50}$, $2.0\%$) all three measurements were above the defined limits. Screening with fasting and two-hour results, as performed in some institutions to reduce costs, would result in only $66\%$ ($\frac{33}{50}$) of the GDM cases being detected in this study population (Table 3). Table 3. \| OGTT test results and GDM treatment \| Total \| Without GDM \| With GDM \| p-value \| \| --- \| --- \| --- \| --- \| --- \| \| N \| 374 \| 324 \| 50 \| \| \| GA (weeks) at OGTT, median [IQR] \| 26.6 [25.7, 27.6] \| 26.6 [25.7, 27.6] \| 26.6 [25.9, 27.4] \| 0.949 \| \| OGTT * results (mg/dL), median [IQR] \| \| \| \| \| \| BSL fasting \| 79 [74, 84] \| 78 [73, 83] \| 86 [81, 96] \| <0.001 \| \| BSL one hour \| 132 [114, 154] \| 129 [112, 147] \| 173 [142, 191] \| <0.001 \| \| BSL two hours \| 111 [97, 127] \| 110 [96, 123] \| 129 [113, 157] \| <0.001 \| \| Proportion of positivity at each OGTT timepoint \| \| \| \| \| \| Fasting only \| \| \| 17 (34%) \| \| \| One hour only \| \| \| 17 (34%) \| \| \| Two hours only \| \| \| 10 (20%) \| \| \| Fasting and one hour \| \| \| 2 (4%) \| \| \| Fasting and two hours \| \| \| 0 (0%) \| \| \| One hour and two hours \| \| \| 3 (6%) \| \| \| All three \| \| \| 1 (2%) \| \| \| GDM treatment, n (%) \| \| \| \| \| \| Diet and exercise only \| \| \| 18 (36%) \| \| \| Diet & metformin \| \| \| 27 (54%) \| \| \| Metformin and glibenclamide \| \| \| 4 (8%) \| \| \| Metformin and insulin \| \| \| 1 (2%) \| \| ## Risk-factor-based screening for GDM There were 37 women in the GDM positive group and 234 women in the GDM negative group who had at least one risk factor, translating into an overall proportion of $72.5\%$ ($\frac{271}{374}$) (Table 1). Of the 50 OGTT positive cases, 37 were correctly identified by risk factors alone, resulting in a sensitivity of $74.0\%$ ($59.7\%$-$85.4\%$). Specificity was low, with 90 of 324 being correctly identified as negative for GDM using risk-factor-based screening: $27.8\%$ ($23.0\%$-$33.0\%$). The positive and negative predictive values were $13.7\%$ ($9.8\%$-$18.3\%$) and $87.4\%$ ($79.4\%$-$93.1\%$), respectively. Of the seven risk-factor-based screening items included in this analysis, a history of GDM and previous stillbirth could not be included in a multivariable model due to zero counts. None of the risk-factor-based screening criteria significantly predicted GDM status in this migrant population. History of macrosomia had a positive (wide confidence interval) and non-significant association due to the small number of cases (6.59, $95\%$ CI 0.41-107.1, $$p \leq 0.185$$). All other risk factors were not significant at $p \leq 0.20.$ ## GDM management and treatment Approximately two out of three women, $64\%$ ($\frac{32}{50}$), were medicated for their GDM (Table 3). Most received metformin only ($54\%$ ($\frac{27}{50}$)), with a smaller proportion receiving metformin plus glibenclamide ($8.0\%$ ($\frac{4}{50}$)), and only one patient ($2.0\%$) received insulin due to metformin failure at 27+3 weeks of gestation. This case required referral to the government hospital. ## GDM risk in Burman and Karen ethnic groups Risk factors for GDM were examined separately for the two main ethnic groups in the population by multivariate analysis (Table 4). After adjustment, overweight or obese Burman women were at a five-fold higher risk of GDM. A different relationship between BMI and GDM was apparent for Karen women where the risks were similarly elevated (non-significant) for both underweight and overweight or obese women (Table 4). Table 4. \| Risk factors \| Karen n=247 \| Karen n=247.1 \| Karen n=247.2 \| Karen n=247.3 \| Burman n=109 \| Burman n=109.1 \| Burman n=109.2 \| Burman n=109.3 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Risk factors \| No GDM, n=218 \| GDM, n=29 \| Adjusted Odd Ratio (95% CI) \| p-value \| No GDM, n=90 \| GDM, n=19 \| Adjusted Odd Ratio (95% CI) \| P-value \| \| Age 30 and older, n (%) \| 56 (25.7) \| 6 (20.7) \| 0.52 (0.18-1.52) \| 0.231 \| 24 (26.7) \| 5 (26.3) \| 0.54 (0.15-1.92) \| 0.343 \| \| Smoker, n (%) \| 19 (8.72) \| 5 (17.2) \| 3.09 (0.92-10.39) \| 0.069 \| 2 (2.22) \| 1 (5.26) \| 5.27 (0.39-71.88) \| 0.213 \| \| BMI, kg/m 2 * \| \| \| \| \| \| \| \| \| \| Normal (18.50-22.99) \| 126 (57.8) \| 11 (38.0) \| reference \| \| 46 (51.1) \| 6 (31.6) \| reference \| \| \| Underweight (≤18.5) \| 31 (14.2) \| 7 (24.1) \| 2.41 (0.85-6.79) \| 0.097 \| 26 (28.9) \| 4 (21.1) \| 1.20 (0.30-4.73) \| 0.704 \| \| Overweight / obese (≥23) \| 61 (28.0) \| 11 (37.9) \| 2.36 (0.95-5.89) \| 0.064 \| 18 (20.0) \| 9 (47.4) \| 5.03 (1.43-17.64) \| 0.012 \| ## Discussion The most consequential early GDM definition was published by O´Sullivan and Mahan in 1964 29. Their criteria were then tried and adapted over decades with the culmination in the HAPO trial 30, 31. Currently there is no consensus on the optimal screening approach with Europe leaning more to risk-factor based screening and USA towards glucose challenge tests; and it is not entirely clear whether criteria derived from high-resource settings are adequate for institutions in low-resource settings 30, 32. Hence, as the main objective of this manuscript was to assess the performance of the risk-factor-based screening used in routine clinical practice and draw conclusions of its fitness, the presented cohort was explored by an any positive approach. This was possible because all women had data on the relevant risk-factors collected, and as they were part of a preterm birth study cohort, all women had an OGTT done. The analysis identified the shortcomings of current clinical practice as almost one in four women with GDM would have been missed based on risk-factor-based selection for screening when compared with universal screening by 75g OGTT. While the risk-factor-based screening had a sensitivity of $74.0\%$ ($95\%$ CI 59.7-85.4), it lacked specificity $27.8\%$ ($95\%$ CI 23.0-33.0) and resulted in an inadequate positive predictive value of $13.7\%$ ($95\%$ CI 9.8-18.3). Reasons for this underperformance could be related to the limited size of the cohort; due to exclusion of women with a previous caesarean section (potentially due to undiagnosed GDM) from the original cohort; or that risk-factor-based screening is inherently weak for GDM diagnosis in South-East Asian women. The low incidence of reported prior history of GDM or family history of diabetes, most likely results from the limited extent of testing in this population that has limited access to health care 33. At least one in seven ‘healthy’ migrant women presenting to antenatal care in this study cohort had GDM based on the 75g OGTT and thus identifying GDM as a significant health problem in Burman and Karen migrants on the Thailand-Myanmar border. These findings are similar to other migrant populations globally who have to make food choices based on limited expenditure 34. The BMI-related differences in risk factors observed on regression analysis for GDM in Karen and Burman women may relate to different diets and smoking habits between these ethnic groups. A more detailed dietary analysis based on quantitative 24-hour food recall is currently under evaluation. The similar odds for GDM in underweight and overweight/obese Karen women may be related to the thin-type II diabetic phenotype where individuals are at increased risk at a lower BMI 35. Gujral et al. and Rajakramikan et al. have proposed pathogenic mechanisms including impaired insulin secretion, in utero undernutrition, or epigenetic alterations, to explain thin-type II diabetes 36, 37. Of greatest concern is the propensity for this group of patients with undernutrition to have worse diabetes. Ethnohistorical Burman and Karen are distinct populations with their own pheno- and genotypic peculiarities 38. As the slightly different GDM risk-profile is based on a small sample size, these findings must be confirmed in larger cohorts. In this analysis, there was a positive association between GDM and higher percentiles for infant birthweight, larger head circumference and weight-length ratio composition but no difference was seen in mode of delivery, postpartum haemorrhage, perineal damage or Apgar score by GDM status 39– 41. Given that pregnant women with an unremarkable medical and obstetric history were prioritized in the cohort and women with GDM received treatment following the abnormal OGTT result, the low rate of adverse birth outcomes is not unexpected. The high rate of small for gestational age (one in five) newborns has been reported previously and highlights the double burden of nutrition in this population but may also signal a risk for thin-type II diabetes 15, 35. Data from other South-East Asian populations suggest that obese women with GDM have a higher risk of adverse outcome when compared to normal weight pregnant women with GDM 42. However, considering a significant increase in perinatal morbidity in women with uncontrolled GDM compared to women with adequately treated GDM, different strategies of GDM management for obese and non-obese pregnant women does not seem appropriate at this point 43. Early detection of GDM may prevent the need for caesarean section, which limits total expenditure per pregnancy. While the cost for an individual OGTT is small (i.e., approximately 18 THB (0.54 USD) for one glucose test strip, 7.5 THB (0.22 USD) for 75g glucose powder), costs add up if thousands of pregnant women are universally screened each year. Considering the average cost for caesarean section in 2020 for migrant women was 27,695 THB (approximately 824 USD) when referred to the public hospital system, one averted caesarean section would be equivalent to 1,539 glucose test strips – enough for OGTTs in 500 women. Mo et al. concluded that cost effectiveness of universal GDM screening is likely favourable over screening of targeted high-risk populations in a meta-analysis in mostly HIC, while others suggest that universal screening is not useful 44, 45. Since access to adequate diabetes monitoring and pharmacological intervention is severely limited outside of pregnancy in resource-limited settings, there may be added benefit to universal screening in LMIC. The counselling women receive during pregnancy about their GDM may be the first and only information provided on lifestyle modification to prevent the development of type II diabetes later in life 46. Reducing from three (fasting, one hour, two hours) to two (fasting, two hours) tests to bring down costs is not a useful alternative in this population as nearly nine in 10 were positive at a single timepoint distributed across all three time points. As the majority ($68.7\%$) of GDM positive women in this study used oral hypoglycaemic agents, there is a need for a better understanding of effective lifestyle interventions in this marginalized group 2, 16, 47. The findings on the usefulness of SFH contributes to the ongoing debate on the use of international vs. local centiles. The proportion of pregnant women presenting with a SFH ≥90 th centile using local centiles differs markedly compared to the proportion when using international centiles. Using international standards for SFH, most GDM positive women would not be signalled as women with a problem in this population 26. This most likely arises from maternal anthropometric differences (e.g., the greater than 10cm difference in maternal height) between the populations participating to the cohorts for centile curve calculation. From 24 weeks EGA there was a significantly higher proportion of women in the GDM positive group with a SFH above the 90 th centile compared to women without GDM. While this suggests that SFH may have a role, the fact that more than half (52,$8\%$) the women with no GDM had at least one SFH measurement ≥90 th centile renders SFH for GDM as rather unspecific. In addition, the timeframe of detection of increased SFH (32 weeks) is later than when an OGTT identifies GDM. ## Strengths of this study The strengths of this study include first trimester enrolment and ultrasound dating allowing accurate assessment of neonatal anthropometry based on gestation. The risk of information bias is reduced by the prospective cohort design with minimal missing data. There was also close monitoring throughout pregnancy with a high number of antenatal care visits (median 16, IQR 15-17). Furthermore, weight and SFH were measured with calibrated instruments and by well-trained personnel. In addition, this analysis has had a direct local impact resulting in the implementation of universal GDM screening for all women with a two-step approach; with the first step being a glucose challenge test (i.e., 50 g non-fasting oral glucose load, followed by a 1-hour glucose measurement) with 1-hour levels of ≥200 mg/dL being diagnostic of GDM and values 140–199 mg/dL requiring a 2 nd step, namely a complete OGTT. This pragmatic choice to increase the number of women screened and minimize the burden of a full OGTT in all women follows the recommendation of the American College of Obstetricians and Gynecologists. ## Potential study limitations Women with a complicated obstetric or medical history were excluded from the original study. As SMRU does not perform caesarean sections in their clinics, women thought to be at risk of this pregnancy complication were excluded from the original study as they were predicted to not be able to provide a complete set of samples. This was a selection bias for healthier pregnant women, potentially leading to an underestimate of the GDM prevalence in this border population, i.e., the study likely presents the minimum GDM rate in the community of pregnant women. With the selection bias and treatment for all GDM positive women there was a low number of complications; among 37 risk-factor positive cases (vs 17 risk-factor negative cases), there were seven complications overall (preterm ($$n = 1$$), stillbirths ($$n = 0$$), caesarean section ($$n = 2$$), postpartum haemorrhage ($$n = 1$$), and LGA ($$n = 3$$)). The study design did not allow exploration of whether those identified in the high-risk group were also the same women who are likely to have complications from GDM. Due to the relatively small sample, the suggestion of differences in the risk of GDM between the two major ethnic groups requires further verification. ## Conclusions These findings imply that GDM is a problem at the Thailand-Myanmar border with Burman women who are overweight/obese being at the highest risk. GDM determined by risk-factor-based screening performed sub-optimally in this rural, resource-constrained pregnant population. Access to universal screening for GDM can potentially reduce negative impacts for an individual pregnancy but also provide an opportunity to sensitize people in marginalized populations of their potential increased risk for type II diabetes later in life. Considering that additional costs for universal screening appear limited, this is the preferred policy in this population. ## Underlying data Oxford University Research Archives: MSP COHORT GDM SCREEN. Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0). ## Reporting guidelines Figshare: STARD checklist for ‘ Risk factor-based screening compared to universal screening for gestational diabetes mellitus in marginalized Burman and Karen populations on the Thailand-Myanmar border: an observational cohort’. https://doi.org/10.6084/m9.figshare.19382624 48. ## References 1. Nguyen CL, Pham NM, Binns CW. Prevalence of Gestational Diabetes Mellitus in Eastern and Southeastern Asia: A Systematic Review and Meta-Analysis.. (2018.0) 2018 6536974. DOI: 10.1155/2018/6536974 2. Kunasegaran T, Balasubramaniam VRMT, Arasoo VJT. Gestational Diabetes Mellitus in Southeast Asia: A Scoping Review.. (2021.0) 18 1272. DOI: 10.3390/ijerph18031272 3. Goto E. Symphysis-fundal height to identify large-for-gestational-age and macrosomia: a meta-analysis.. (2020.0) 40 929-935. DOI: 10.1080/01443615.2019.1673713 4. Plows JF, Stanley JL, Baker PN. The Pathophysiology of Gestational Diabetes Mellitus.. (2018.0) 19 3342. DOI: 10.3390/ijms19113342 5. Tieu J, McPhee AJ, Crowther CA. Screening for gestational diabetes mellitus based on different risk profiles and settings for improving maternal and infant health.. (2017.0) 8 CD007222. DOI: 10.1002/14651858.CD007222.pub4 6. Vounzoulaki E, Khunti K, Abner SC. Progression to type 2 diabetes in women with a known history of gestational diabetes: systematic review and meta-analysis.. (2020.0) 369 m1361. DOI: 10.1136/bmj.m1361 7. Paschou SA, Sydney GI, Ioakim KJ. Comment on the systematic review and meta-analysis titled "Gestational diabetes and the risk of cardiovascular disease in women".. (2020.0) 19 447-448. DOI: 10.1007/s42000-019-00158-w 8. Guariguata L, Linnenkamp U, Beagley J. Global estimates of the prevalence of hyperglycaemia in pregnancy.. (2014.0) 103 176-185. DOI: 10.1016/j.diabres.2013.11.003 9. Seghieri G, Di Cianni G, Seghieri M. Risk and adverse outcomes of gestational diabetes in migrants: A population cohort study.. (2020.0) 163 108128. DOI: 10.1016/j.diabres.2020.108128 10. World Migration Report 2020. 11. Thailand (April - June 2019).. (2019.0) 12. Myanmar Demographic and Health Survey 2015-16. (2017.0) 13. Nielsen KK, de Courten M, Kapur A. Health system and societal barriers for gestational diabetes mellitus (GDM) services - lessons from World Diabetes Foundation supported GDM projects.. (2012.0) 12 33. DOI: 10.1186/1472-698X-12-33 14. Lorga T, Srithong K, Manokulanan P. Public knowledge of diabetes in Karen Ethnic rural residents: a community-based questionnaires study in the far north-west of Thailand.. (2012.0) 5 799-804. DOI: 10.2147/IJGM.S33177 15. Hashmi AH, Solomon N, Lee SJ. Nutrition in transition: historical cohort analysis summarising trends in under- and over-nutrition among pregnant women in a marginalised population along the Thailand-Myanmar border from 1986 to 2016.. (2019.0) 121 1413-1423. DOI: 10.1017/S0007114519000758 16. Hashmi AH, Paw MK, Nosten S. 'Because the baby asks for it': a mixed-methods study on local perceptions toward nutrition during pregnancy among marginalised migrant women along the Myanmar-Thailand border.. (2018.0) 11 1473104. DOI: 10.1080/16549716.2018.1473104 17. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies.. (2004.0) 363 157-163. DOI: 10.1016/S0140-6736(03)15268-3 18. Lee KW, Ching SM, Ramachandran V. Prevalence and risk factors of gestational diabetes mellitus in Asia: a systematic review and meta-analysis.. (2018.0) 18 494. DOI: 10.1186/s12884-018-2131-4 19. Farrar D, Simmonds M, Bryant M. Treatments for gestational diabetes: a systematic review and meta-analysis.. (2017.0) 7 e015557. DOI: 10.1136/bmjopen-2016-015557 20. Gilder ME, Zin TW, Wai NS. Gestational diabetes mellitus prevalence in Maela refugee camp on the Thai-Myanmar border: a clinical report.. (2014.0) 7 23887. DOI: 10.3402/gha.v7.23887 21. Rani PR, Begum J. Screening and Diagnosis of Gestational Diabetes Mellitus, Where Do We Stand.. (2016.0) 10 QE01-04. DOI: 10.7860/JCDR/2016/17588.7689 22. Metzger BE, Lowe LP. Hyperglycemia and adverse pregnancy outcomes.. (2008.0) 358 1991-2002. DOI: 10.1056/NEJMoa0707943 23. Brummaier T, Syed Ahamed Kabeer B, Lindow S. A prospective cohort for the investigation of alteration in temporal transcriptional and microbiome trajectories preceding preterm birth: a study protocol.. (2019.0) 9 e023417. DOI: 10.1136/bmjopen-2018-023417 24. Rijken MJ, Mulder EJ, Papageorghiou AT. Quality of ultrasound biometry obtained by local health workers in a refugee camp on the Thai-Burmese border.. (2012.0) 40 151-157. DOI: 10.1002/uog.11091 25. White LJ, Lee SJ, Stepniewska K. Estimation of gestational age from fundal height: a solution for resource-poor settings.. (2012.0) 9 503-510. DOI: 10.1098/rsif.2011.0376 26. Papageorghiou AT, Ohuma EO, Gravett MG. International standards for symphysis-fundal height based on serial measurements from the Fetal Growth Longitudinal Study of the INTERGROWTH-21. (2016.0) 355 i5662. DOI: 10.1136/bmj.i5662 27. Cheikh Ismail L, Bishop DC, Pang R. Gestational weight gain standards based on women enrolled in the Fetal Growth Longitudinal Study of the INTERGROWTH-21. (2016.0) 352 i555. DOI: 10.1136/bmj.i555 28. Villar J, Cheikh Ismail L, Victora CG. International standards for newborn weight, length, and head circumference by gestational age and sex: the Newborn Cross-Sectional Study of the INTERGROWTH-21. (2014.0) 384 857-868. DOI: 10.1016/S0140-6736(14)60932-6 29. O’Sullivan JB, Mahan CM. Criteria for the oral glucose tolerance test in pregnancy.. (1964.0) 13 278-285. PMID: 14166677 30. McIntyre HD, Catalano P, Zhang C. Gestational diabetes mellitus.. (2019.0) 5 47. DOI: 10.1038/s41572-019-0098-8 31. The Hyperglycemia and Adverse Pregnancy Outcome (HAPO) Study.. (2002.0) 78 69-77. DOI: 10.1016/s0020-7292(02)00092-9 32. Mishra S, Rao CR, Shetty A. Trends in the Diagnosis of Gestational Diabetes Mellitus.. (2016.0) 2016 5489015. DOI: 10.1155/2016/5489015 33. Saito M, Keereevijit A, San TD. Challenges to primary healthcare services in the management of non-communicable diseases in marginalised populations on the Thailand-Myanmar border: a pilot survey.. (2018.0) 48 273-277. DOI: 10.1177/0049475518786853 34. Pottie K, Hadi A, Chen J. Realist review to understand the efficacy of culturally appropriate diabetes education programmes.. (2013.0) 30 1017-1025. DOI: 10.1111/dme.12188 35. Gujral UP, Weber MB, Staimez LR. Diabetes Among Non-Overweight Individuals: an Emerging Public Health Challenge.. (2018.0) 18 60. DOI: 10.1007/s11892-018-1017-1 36. Gujral UP, Pradeepa R, Weber MB. Type 2 diabetes in South Asians: similarities and differences with white Caucasian and other populations.. (2013.0) 1281 51-63. DOI: 10.1111/j.1749-6632.2012.06838.x 37. Rajamanickam A, Munisankar S, Dolla CK. Impact of malnutrition on systemic immune and metabolic profiles in type 2 diabetes.. (2020.0) 20 168. DOI: 10.1186/s12902-020-00649-7 38. Bancone G, Gilder ME, Chowwiwat N. Prevalences of inherited red blood cell disorders in pregnant women of different ethnicities living along the Thailand-Myanmar border [version 2; peer review: 2 approved].. (2017.0) 2 72. DOI: 10.12688/wellcomeopenres.12338.2 39. Logan KM, Gale C, Hyde MJ. Diabetes in pregnancy and infant adiposity: systematic review and meta-analysis.. (2017.0) 102 F65-F72. DOI: 10.1136/archdischild-2015-309750 40. Villar J, Puglia FA, Fenton TR. Body composition at birth and its relationship with neonatal anthropometric ratios: the newborn body composition study of the INTERGROWTH-21. (2017.0) 82 305-316. DOI: 10.1038/pr.2017.52 41. Yau KI, Chang MH. Weight to length ratio--a good parameter for determining nutritional status in preterm and full-term newborns.. (1993.0) 82 427-429. DOI: 10.1111/j.1651-2227.1993.tb12715.x 42. Yue S, Thi VTK, Dung LP. Clinical consequences of gestational diabetes mellitus and maternal obesity as defined by asian BMI thresholds in Viet Nam: a prospective, hospital-based, cohort study.. (2022.0) 22 195. DOI: 10.1186/s12884-022-04533-1 43. Langer O, Yogev Y, Most O. Gestational diabetes: the consequences of not treating.. (2005.0) 192 989-997. DOI: 10.1016/j.ajog.2004.11.039 44. Mo X, Gai Tobe R, Takahashi Y. Economic Evaluations of Gestational Diabetes Mellitus Screening: A Systematic Review.. (2021.0) 31 220-230. DOI: 10.2188/jea.JE20190338 45. Fitria N, van Asselt ADI, Postma MJ. Cost-effectiveness of controlling gestational diabetes mellitus: a systematic review.. (2019.0) 20 407-417. DOI: 10.1007/s10198-018-1006-y 46. Hashmi AH, Nyein PB, Pilaseng K. Feeding practices and risk factors for chronic infant undernutrition among refugees and migrants along the Thailand-Myanmar border: a mixed-methods study.. (2019.0) 19 1586. DOI: 10.1186/s12889-019-7825-7 47. Martis R, Crowther CA, Shepherd E. Treatments for women with gestational diabetes mellitus: an overview of Cochrane systematic reviews.. (2018.0) 8 CD012327. DOI: 10.1002/14651858.CD012327.pub2 48. Brummaier T. STARD checklist for Risk factor-based screening compared to universal screening for gestational diabetes mellitus in marginalized Burman and Karen populations on the Thailand-Myanmar border: an observational cohort.. figshare. (2022.0). DOI: 10.6084/m9.figshare.19382624.v1 49. Clinical consequences of gestational diabetes mellitus and maternal obesity as defined by asian BMI thresholds in Viet Nam: a prospective, hospital-based, cohort study.. (2022.0) 22 195. DOI: 10.1186/s12884-022-04533-1
--- title: 'Equity within AI systems: What can health leaders expect?' authors: - Emma Gurevich - Basheer El Hassan - Christo El Morr journal: Healthcare Management Forum year: 2022 pmcid: PMC9976641 doi: 10.1177/08404704221125368 license: CC BY 4.0 --- # Equity within AI systems: What can health leaders expect? ## Abstract Artificial Intelligence (AI) for health has a great potential; it has already proven to be successful in enhancing patient outcomes, facilitating professional work and benefiting administration. However, AI presents challenges related to health equity defined as an opportunity for people to reach their fullest health potential. This article discusses the opportunities and challenges that AI presents in health and examines ways in which inequities related to AI can be mitigated. ## AI and ML in health Artificial Intelligence (AI) aims to imitate human intelligence and can be used to enable better decision-making processes in many areas including health. Machine Learning (ML) is a field of AI that aims to develop models for prediction and clustering. A ML algorithm uses a dataset to learn how to predict or cluster; this dataset is called the learning dataset. When a ML model predicts a class to which a data instance belongs, the model is called a classifier; on the other hand, when the model predicts a number (e.g. age and number of months) it is called a regressor. Both classification and regression are part of a larger category called supervised learning. In supervised learning, the learning dataset contains the target or outcome (i.e. the dependent variable) of each instance in the dataset. In the case of clustering, a ML algorithm aims to build a model that groups data into clusters based on a certain similarity measure among the data instances. It will then indicate the cluster to which each data instance belongs. The outcome is known in the learning dataset. When faced with new data, the clustering model chooses the cluster to which the new data belongs. Since the outcome of the new data instance is not known in the learning dataset, clustering is said to belong to unsupervised learning. The application of AI and ML in healthcare is expanding. AI has been proven to be successful in early diagnosis, early detection, prediction, and choosing between treatment alternatives,1 in medical imaging interpretation and processing, in pathology, gastroenterology, and ophthalmology, to name a few domains.2 ## Health equity Equity is defined as fairness and justice for all. Health equity is a principle dedicated to maximizing people’s health potential and reducing health disparities. Hence, health equity considers people’s social factors, also known as the Social Determinants of Health (SDoH), as determinants of their ability to equitably access health. SDoH are known to affect individual and population health with ample evidence indicating that poor health is directly related to social factors.3 In Canada, SDoH include Aboriginal status, race, disability, early life development, education, sexual orientation, social exclusion, social safety net, unemployment and job security, employment and working conditions, food insecurity, health services, gender and gender identity, housing, income, and income distribution.4 Canada is a multicultural society, and racialized populations include South Asians, Chinese, and Black communities. Racialized Canadians’ physical, mental, and social health are due to experience of lower rates of income, higher rates of unemployment, and lower occupational status.4 While the implementation of AI in health has potential benefits, AI can also undermine health equity.2 The objective of this paper is to assess the interplay between equity and AI. ## Equity in health: AI potential benefits AI has a high potential in transforming decision-making and medical treatment, specifically, in primary care.5,6 Many vulnerable populations access healthcare services through primary care; AI systems in these settings can have a positive impact on vulnerable populations.7 AI solutions have proven to be beneficial for patients in areas of clinical oncology, dermatology, the prediction of postpartum depression, the diagnosis of diabetic retinopathy in youth, and in the management and nutrition counselling for patients with diabetes and other chronic diseases.6,8-14 AI has also been emerging in preventative care15,16 and the medical robot sector.17-19 Furthermore, AI-assisted medical services can benefit underserved rural areas.20 Currently, initiatives have been implemented to properly manage health systems, track interactions, improve cost-efficiency, and to effectively increase well-being.21 In addition, patient-centred care is expected to be positively impacted by AI applications, specifically in communication with patients. Many patients in healthcare settings have limited English proficiency and, as a result, may suffer from a larger number of medical complications.22 AI can play a role in overcoming language barriers. Indeed, AI-based applications have been developed for patients to choose their preferred language through standardized instructions.23 The list of areas in healthcare that can benefit from AI, including individual and public health, is endless.24-37 Not only can AI potentially enhance health equity by improving healthcare provision, but it also has potential to help overcome human decision-making, which is often clouded by biases (including cognitive bias); for instance, AI-based systems have helped in reducing the number of incorrectly denied refugee claims.38 ## Equity in health: Potential AI concerns While AI has great potential in enhancing health equity, there are concerns related to its use in healthcare. It is imperative that AI initiatives do not continue perpetuating the same inequities already faced by vulnerable individuals.6,7,9,11-13,37,39,40 For instance, an AI application that aimed to predict how likely an individual is to recommit a crime was proven to be substantially biased against Black people as it consistently predicted that they were at a high risk of recommitting a crime in comparison to White individuals. However, statistics show that they were only half as likely to recommit a crime as their White counterparts.7 This software reflects inherent and explicit social biases surrounding race. The same risk applies to Canada, while race correction is used in kidney and lung function measurements, for example, variation exists within the healthcare professionals’ body.41 As LLana James, AI, Medicine and Data Justice Post-Doctoral Fellow at Queen’s University puts it: “Race-medicine is not solely about Black people, it is also about how White people have used themselves as a primary reference in clinical assessment, and have in so doing, not necessarily tended to the precision of the science. ”41 AI models trained on past data will reflect the data biases. Other instances of unfairness towards vulnerable groups have been reported across algorithms used for medical management, public health, and federal compensation programs.12 Health data used to train algorithms is often collected from a mostly White population, and/or excludes ethno-racial information altogether; the resulting models may be biased against Black, Indigenous, and People of Colour (BIPOC). On the other hand, historically, when ethno-racial data has been included, it has been incorporated inappropriately. For instance, pulmonary function and pain scores that are adjusted for race continue to be used throughout the healthcare systems contributing to poor health outcomes for People of Colour.12 These are a few examples of SDoH impact on AI algorithms, the main lesson is that those with privilege (i.e. White people, men, higher socioeconomic status, and English speaking) tend to have better outcomes with the use of these algorithms as opposed to those with less privilege (i.e. women, non-binary folks, BIPOC, and English as a second language); hence, the need to mitigate algorithmic biases. Building ML models based on biased tools only exacerbates bias. Biases in such algorithms reflect historical influence that encapsulates systemic racism, sexism, and other types of socioeconomic biases. This often occurs due to over-/under-representation of specific populations in training data sets, or due to the implicit biases of those creating the algorithms. This is later reflected in the predictive power of algorithms. Undesirable biases further perpetuate existing health inequities, putting vulnerable populations at a greater risk of experiencing poor health outcomes.7 There is a need to train AI and ML algorithms to be inclusive so that biases are addressed.11 ## Lesson for health leaders: Mitigating AI inequities AI solutions can only be as successful as their benefits; it is imperative that the disadvantages of such technologies and their potential pitfalls are mitigated. Despite this, it should be noted that inequalities exist in access to AI technology, as well as unfairness in who it may provide an advantage and disadvantage to.39 ## Equity assessment The largest concern surrounding AI solutions is the potential for systems to continue perpetuating inequities.6-8,11,12,39,40 Thus, AI initiatives should have two main goals: [1] they should be designed and utilized in a manner that does not create or maintain health disparities currently experienced by vulnerable groups, and [2] they should address and remove existing health disparities.6,39 To ensure that all healthcare-based AI embodies these two goals, it is important to create system level changes such as a federal and/or provincial regulatory framework that oversees the equity dimensions in the implementation of AI solutions.12,39 The Federal Drug Agency (FDA) in the United States has introduced a regulation for AI applications that are designed for use in clinical decision-making or for inpatient health data analysis or medical imaging. This step forward is still limited as it leaves a myriad of applications designed for other purposes (e.g. resource allocation and access to public health) and affecting patients and healthcare delivery from regulations.39 It is our view that such applications should also be regulated. Currently, the government of Canada is tabling Bill C-27 that will enact the Artificial Intelligence and Data Act (AIDA) “to regulate international and interprovincial trade and commerce in artificial intelligence systems by requiring that certain persons adopt measures to mitigate risks of harm and biased output related to high-impact artificial intelligence systems. ”42 While it is not enacted yet, it addresses assessment, mitigation, and monitoring obligations; it has a provision to establish measure “to identify, assess, and mitigate the risks of harm or biased output that could result from the use of the system. ”42 The definition of a “high-impact system” is not clear yet and is left to be established in AIDA section 5[1]. It is yet to be discovered how the law will impact the Canadian innovation and application landscape. A regulatory body for AI applications will probably take shape on the provincial and territorial levels; however, some levels of coordination and collaboration among national, provincial, and territorial entities would be expected. For AI applications intended for health, a regulatory body would collect evidence from available research, and might recommend or require [1] AI-reporting based on current recommendations in the field,43,44 especially those related to AI-equity and AI-interpretability,45,46 as well as [2] submission of specific evidence (i.e. randomized control trial).47 ## Equity at the core of AI projects It is important to incorporate an equity dimension in the different stages of AI creation, from assessing the representativeness of data, to continuous surveillance of systems after deployment.12 In the development stage, for example, it is imperative that data used in the training of predictive algorithms includes ethno-racial, sex, and gender characteristics as there are apparent differences in the risk factors for certain diseases and health outcomes based on these factors.9,11 This in turn will reduce the chance of a distributional shift, a phenomenon where the training data is not representative of the population.11 Likewise, it is important to disclose the distribution of factors that are not routinely reported as these may increase desirable bias while exposing undesirable biases.9,11 Moreover, one should report limitations related to the training data set (e.g. ethno-racial).11 Furthermore, there is a need to validate models using data samples other than retrospective data as these may not fully capture biases.12 When implementing the model, it is good practice to make certain that systems undergo continuous evaluation,12 to ensure that models can perform as designed and work to remove existing systemic inequalities within the healthcare system.6 ## Involving stakeholders The implementation of AI in projects must be a collaborative effort. It should include physicians, patients, and communities from diverse backgrounds of social, cultural, and economic contexts. One way to involve recipients of care is by using Patient-Reported Outcome Measures (PROMs) to understand health-related outcome measures from a patient's perspective. Furthermore, engaging patients and their communities with Information Technology (IT) teams that produce the algorithms can help assess and address AI bias. In this context, training and education on health equity is important for IT teams to understand the potential effects of AI initiatives on health equity.48 ## Algorithmovigilance Due to the number of systemic inequities and health disparities, developing and testing algorithms that allow systematic surveillance and vigilance in the development of AI models in healthcare becomes important. Algorithmovigilance involves algorithms’ evaluation and monitoring to prevent AI bias and, thus, must be part of AI projects. Debiasing steps can be taken within a project as well. For instance, debiasing can be the result of retraining models without race variables (fairness through unawareness) or measuring the differences in outcomes between privileged and unprivileged groups.49 ## Need to address SDoH The use of AI is emerging in public health; however, it faces multiple challenges from a social justice perspective. Challenges include focusing on data while drawing the attention away from the causes of health inequities such as the SDoH. AI intended for social good that neglects this aspect may create new vulnerabilities and fail to attain the projects’ aim. Employing SDoH lens in AI initiatives will benefit the public and help create a digital world oriented toward social justice and health equity.50 ## Need for data regarding social context AI technologies and advanced analytics are being integrated into healthcare to make key clinical decisions. Thus, AI technologies must be provided with data related to social contexts, otherwise the work produced will be short of considering health equity, especially in primary care. In one example, AI models were used in a primary care setting, $20\%$ of patients preferring to use Spanish were misclassified as preferring English due to imbalanced training.48 Integrating lived experiences of diverse communities is key to increasing equity of AI models. ## Challenges specific to the Canadian context One limitation is the cost of implementing AI in northern and remote communities as the number of people using the AI-based application is significantly lower. However, AI-based may produce cost savings, if the AI-related cost could be balanced by cost savings is something to be studied. There is also a challenge regarding Indigenous health practices and values, as these are different from medicine as practiced in the healthcare system. AI models based on Indigenous practices and values would be needed for an Indigenous population practicing medicine. Moreover, given the multicultural environment in the Canadian society, culturally inclusive AI models that respect the variety of culture could be needed and would be need to be designed.51 A large portion of AI research and development in Canada is a part of the Pan-Canadian AI Strategy which is directed by the Canadian Institute for Advanced Research (CIFAR). CIFAR partners with the following institutions: Alberta Machine Intelligence Institute (AMII), Montreal Institute for Learning Algorithms (MILA), and Vector Institute to bring together researchers from across the country.52 While CIFAR involves AI in multiple areas, health-related projects include understanding how gene interactions impact health and development as well as the effect of human microbiomes on health, development, and behaviour. Ethical implications of AI in health would be important to research with these institutions. ## Important considerations It is critical for policy-makers to understand that bias mitigation should not end with AI model development but, rather, extend across the product lifecycle. We believe in line with Thomasian et al.53 that the following considerations are key for future development of equitable AI for health:1) Bias alleviation during model development• Study how the quality and availability of equity related data (e.g. immigration, race, and gender) can impact model performance.• Assemble and organize open databases with non-identifiable patient information to overcome imbalance in equity related data.• Make use of collaborative model training (e.g. federated learning and cyclic weight transfer) as they can increase data size without transferring patient data between health organizations.2) Bias mitigation of the machine learning model• Consider non-routinely reported factors, such as socioeconomic status and race, when developing models, especially when the models are intended to serve in areas where inequity is well documented.• Use appropriate bias metrics selected based on the algorithm's objectives.3) Post deployment validation• Validate the model prospectively and not only based on retrospective datasets. This is important as models trained on retrospective data alone might behave differently (e.g. cause harm) when new instances emerge in real-life.4) Auditing for interpretability and bias• Audit for equity/biases continuously throughout implementation.• Audit for interpretability of the models to avoid unintended consequences of technology and mitigate human factors/errors post implementation. ## Conclusion While AI can handle the complex and multidimensional fabric of the Canadian population and deal with big data, it cannot do so unless trained to do it. Hence, mitigation of AI potential biases is needed; particularly, processes and frameworks to follow during the design, and quality monitoring processes are important to implement. It is important to note that AI has proven to be cost effective in many cases. For example, autonomous AI was as effective for and less costly (up to $34 for compared with telemedicine and $64 and $91 compared with ophthalmoscopy) for retinopathy of prematurity screening.54 Also, research shows that AI-based tools produce cost savings if used as a strategy in screening colonoscopy,55 and for breast cancer screening.56 While this is encouraging, it cannot be generalized and need to be studied on a case by case basis.57 Equity as an aim in healthcare delivery is an important and often overlooked factor in health informatics. AI can provide potential benefits and risks to patients as it can enhance or diminish equity. While steps to mitigate equity concerns in AI projects are needed and available, a systematic equitable AI approach is yet to be developed. While it is likely that the proper inclusion of SDoH will require more work on the side of the creators of algorithms (and will be more resource intensive), the cost implications of disregarding the SDoH within the current healthcare system are also high and defeat the very purpose of the healthcare system. The Canadian healthcare system would benefit from implementing SDoH informed AI solutions in order to prevent health incidents and provide an equitable access to health. It is our view that health leaders need to support the inclusion of SDoH within Canadian healthcare in general and particularly the expected upcoming wave of AI-based systems. Simultaneously leaders should advocate for the inclusion of equity in AI projects and support the inclusion of anti-racism and anti-oppressive practices in the healthcare industry. ## ORCID iD Christo El Morr https://orcid.org/0000-0001-6287-3438 ## References 1. Hulsen T, Petkovic M, Varga OE, Jamuar SS. Editorial: AI in healthcare: From data to intelligence. Editorial. Frontiers in Artificial Intelligence (2022) 5. DOI: 10.3389/frai.2022.909391 2. Rajpurkar P, Chen E, Banerjee O, Topol EJ. AI in health and medicine. Nature Medicine (2022) 28 31-38. DOI: 10.1038/s41591-021-01614-0 3. Castillo EG, Harris C. Directing research toward health equity: a health equity research impact assessment. Journal of General Internal Medicine (2021) 36 2803-2808. DOI: 10.1007/s11606-021-06789-3 4. Raphael D, Bryant T, Mikkonen J, Raphael A. Social Determinants of Health: the Canadian Facts (2020) 5. Vogel L. Record gift to University of Toronto faculty of medicine supports AI, equity. Cmaj (2020) 192 E1223-e1224. DOI: 10.1503/cmaj.1095900 6. Zhongming Z, Linong L, Xiaona Y, Wangqiang Z, Wei L. Eliminating bias from healthcare AI critical to improve health equity. (2021) 7. Wang JX, Somani S, Chen JH, Murray S, Sarkar U. Health equity in artificial intelligence and primary care research: Protocol for a scoping review. JMIR research protocols (2021) 10 e27799. PMID: 34533458 8. Johnson-Mann CN, Loftus TJ, Bihorac A. Equity and artificial intelligence in surgical care. JAMA Surg (2021) 156 509-510. DOI: 10.1001/jamasurg.2020.7208 9. Lee MS, Guo LN, Nambudiri VE. Towards gender equity in artificial intelligence and machine learning applications in dermatology. J Am Med Inform Assoc (2022) 29 400-403. DOI: 10.1093/jamia/ocab113 10. Monlezun DJ. Artificial Intelligence-Driven Population Health Management Improving Healthcare Value & Equity: Culinary Medicine & Its Multi-Site Cohort Study with Nested Bayesian Adaptive Randomized Trial of 3,785 Medical Trainees/Professionals & Patients (2017) 11. Pham Q, Gamble A, Hearn J, Cafazzo JA. The Need for Ethnoracial Equity in Artificial Intelligence for Diabetes Management: Review and Recommendations. J Med Internet Res (2021) 23 e22320. DOI: 10.2196/22320 12. Thomasian NM, Eickhoff C, Adashi EY. Advancing health equity with artificial intelligence. Journal of public health policy (2021) 42 1-10. PMID: 33674687 13. Urman A, Wang C-K, Dankwa-Mullan I, Scheinberg E, Young MJ. Harnessing AI for health equity in oncology research and practice (2018) 14. Wolf RM, Liu TYA, Thomas C. The SEE study: Safety, efficacy, and equity of implementing autonomous artificial intelligence for diagnosing diabetic retinopathy in youth. Diabetes Care (2021) 44 781-787. DOI: 10.2337/dc20-1671 15. Noorbakhsh-Sabet N, Zand R, Zhang Y, Abedi V. Artificial intelligence transforms the future of health care. Am J Med (2019) 132 795-801. DOI: 10.1016/j.amjmed.2019.01.017 16. Rowe JP, Lester JC. Artificial intelligence for personalized preventive adolescent healthcare. J Adolesc Health (2020) 67 S52-s58. DOI: 10.1016/j.jadohealth.2020.02.021 17. Fiske A, Henningsen P, Buyx A. Your robot therapist will see you now: Ethical implications of embodied artificial intelligence in psychiatry, psychology, and psychotherapy. J Med Internet Res (2019) 21 e13216. DOI: 10.2196/13216 18. Gumbs AA, Frigerio I, Spolverato G. Artificial intelligence surgery: How do we get to autonomous actions in surgery?. Sensors (Basel) (2021) 21. DOI: 10.3390/s21165526 19. Stai B, Heller N, McSweeney S. Public perceptions of artificial intelligence and robotics in medicine. J Endourol (2020) 34 1041-1048. DOI: 10.1089/end.2020.0137 20. Guo J, Li B. The application of medical artificial intelligence technology in rural areas of developing countries. Health Equity (2018) 2 174-181. DOI: 10.1089/heq.2018.0037 21. van Leeuwen KG, de Rooij M, Schalekamp S, van Ginneken B, Rutten M. How does artificial intelligence in radiology improve efficiency and health outcomes?. Pediatr Radiol (2021). DOI: 10.1007/s00247-021-05114-8 22. Wilson E, Chen AHM, Grumbach K, Wang F, Fernandez A. Effects of limited English proficiency and physician language on health care comprehension. Journal of general internal medicine (2005) 20 800-806. DOI: 10.1111/j.1525-1497.2005.0174.x 23. Chonde DB, Pourvaziri A, Williams J. RadTranslate: An artificial intelligence-powered intervention for urgent imaging to enhance care equity for patients with limited English proficiency during the COVID-19 pandemic. J Am Coll Radiol (2021) 18 1000-1008. DOI: 10.1016/j.jacr.2021.01.013 24. Banerjee M, Chiew D, Patel KT. The impact of artificial intelligence on clinical education: Perceptions of postgraduate trainee doctors in London (UK) and recommendations for trainers. BMC Med Educ (2021) 21 429. DOI: 10.1186/s12909-021-02870-x 25. Benke K, Benke G. Artificial intelligence and big data in public health. Int J Environ Res Public Health (2018) 15 2796. DOI: 10.3390/ijerph15122796 26. Greatbatch O, Garrett A, Snape K. The impact of artificial intelligence on the current and future practice of clinical cancer genomics. Genet Res (Camb) (2019) 101 e9. DOI: 10.1017/s0016672319000089 27. Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts H. Artificial intelligence in radiology. Nat Rev Cancer (2018) 18 500-510. DOI: 10.1038/s41568-018-0016-5 28. Keskinbora KH. Medical ethics considerations on artificial intelligence. J Clin Neurosci (2019) 64 277-282. DOI: 10.1016/j.jocn.2019.03.001 29. Lee J, Wallace MB. State of the art: The impact of artificial intelligence in endoscopy 2020. Curr Gastroenterol Rep (2021) 23 7. DOI: 10.1007/s11894-021-00810-9 30. Maron JL. Impact of artificial intelligence on clinical decision-making in health care. Clin Ther (2022) 44 825-826. DOI: 10.1016/j.clinthera.2022.05.005 31. Miller DD, Brown EW. Artificial intelligence in medical practice: The question to the answer?. Am J Med (2018) 131 129-133. DOI: 10.1016/j.amjmed.2017.10.035 32. Wolff J, Pauling J, Keck A, Baumbach J. The economic impact of artificial intelligence in health care: Systematic review. J Med Internet Res (2020) 22 e16866. DOI: 10.2196/16866 33. Wuni AR, Botwe BO, Akudjedu TN. Impact of artificial intelligence on clinical radiography practice: Futuristic prospects in a low resource setting. Radiography (Lond) (2021) 27 S69-s73. DOI: 10.1016/j.radi.2021.07.021 34. Moghisi R, El Morr C, Pace KT, Hajiha M, Huang J. A machine learning approach to predict the outcome of urinary Calculi treatment using shock wave lithotripsy: Model development and validation study. Interact J Med Res (2022) 11 e33357. DOI: 10.2196/33357 35. Kundi B, Dharmalingam D, Tadesse R. Disability advocacy using a smart virtual community. (2022) 316-319. DOI: 10.5220/0010748600003123 36. El Morr C, Maret P, Muhlenbach F. A virtual community for disability advocacy: Development of a searchable artificial intelligence–Supported platform. JMIR Form Res (2021) 5 e33335. DOI: 10.2196/33335 37. Kundi B, El Morr C. Need of equity in virtual mental health in Canada in the times of COVID-19. Stud Health Technol Inform (2022) 295 358-359. DOI: 10.3233/shti220737 38. Cameron HE, Goldfarb A, Morris L. Artificial intelligence for a reduction of false Denials in refugee claims. Journal of Refugee Studies (2021) 35 493-510. DOI: 10.1093/jrs/feab054 39. Smith MJ, Axler R, Bean S, Rudzicz F, Shaw J. Four equity considerations for the use of artificial intelligence in public health. Bulletin of the World Health Organization (2020) 98 290-292. PMID: 32284656 40. Zhang J, Whebell S, Gallifant J. An interactive dashboard to track themes, development maturity, and global equity in clinical artificial intelligence research. Lancet Digit Health (2022) 4 e212-e213. DOI: 10.1016/s2589-7500(22)00032-2 41. Alberga H. How can we keep algorithmic racism out of Canadian health care’s AI toolkit?. (2022) 42. BILL C-27: An act to enact the consumer privacy protection act, the personal information and data protection tribunal act and the artificial intelligence and data act and to make consequential and related amendments to other Acts. Minister of Innovation, Science and Industry. (2022) 43. Liu X, Cruz Rivera S, Moher D. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Nature Medicine (2020) 26 1364-1374. DOI: 10.1038/s41591-020-1034-x 44. Mongan J, Moy L, Kahn CE,. Checklist for artificial intelligence in medical imaging (CLAIM): A guide for authors and reviewers. Radiol Artif Intell (2020) 2 e200029. DOI: 10.1148/ryai.2020200029 45. Reyes M, Meier R, Pereira S. On the interpretability of artificial intelligence in radiology: Challenges and opportunities. Radiol Artif Intell (2020) 2 e190043. DOI: 10.1148/ryai.2020190043 46. Stiglic G, Kocbek P, Fijacko N, Zitnik M, Verbert K, Cilar L. Interpretability of machine learning-based prediction models in healthcare. WIREs Data Mining and Knowledge Discovery (2020) 10 e1379. DOI: 10.1002/widm.1379 47. Zhou Q, Chen Z-h, Cao Y-h, Peng S. Clinical impact and quality of randomized controlled trials involving interventions evaluating artificial intelligence prediction tools: a systematic review. npj Digital Medicine (2021) 4 154. DOI: 10.1038/s41746-021-00524-2 48. Clark CR, Wilkins CH, Rodriguez JA. Health care equity in the use of advanced analytics and artificial intelligence technologies in primary care. Journal of General Internal Medicine (2021) 36 3188-3193. DOI: 10.1007/s11606-021-06846-x 49. Embi PJ. Algorithmovigilance-advancing methods to analyze and monitor artificial intelligence-driven health care for effectiveness and equity. JAMA Netw Open (2021) 4 e214622. DOI: 10.1001/jamanetworkopen.2021.4622 50. Holzmeyer C. Beyond ‘AI for social good’ (AI4SG): Social transformations—not tech-fixes—for health equity. Interdisciplinary Science Reviews (2021) 46 94-125. DOI: 10.1080/03080188.2020.1840221 51. Collaborative culture-sensitive AI. The detector. (2022) 52. 52Canadian Institute for Advanced Research. Pan-Canadian AI Strategy. Canadian Institute for Advanced Research (CIFAR). 2022. https://cifar.ca/ai/. Pan-Canadian AI Strategy (2022) 53. Thomasian NM, Eickhoff C, Adashi EY. Advancing health equity with artificial intelligence. J Public Health Policy (2021) 42 602-611. DOI: 10.1057/s41271-021-00319-5 54. Morrison SL, Dukhovny D, Chan RVP, Chiang MF, Campbell JP. Cost-effectiveness of artificial intelligence-based retinopathy of prematurity screening. JAMA Ophthalmol (2022) 140 401-409. DOI: 10.1001/jamaophthalmol.2022.0223 55. Areia M, Mori Y, Correale L. Cost-effectiveness of artificial intelligence for screening colonoscopy: A modelling study. The Lancet Digital Health (2022) 4 e436-e444. DOI: 10.1016/S2589-7500(22)00042-5 56. Mital S, Nguyen HV. Cost-effectiveness of using artificial intelligence versus polygenic risk score to guide breast cancer screening. BMC Cancer (2022) 22 501. DOI: 10.1186/s12885-022-09613-1 57. Gomez Rossi J, Rojas-Perilla N, Krois J, Schwendicke F. Cost-effectiveness of artificial intelligence as a decision-support system applied to the detection and grading of Melanoma, dental caries, and diabetic retinopathy. JAMA Network Open (2022) 5 e220269-e220269. DOI: 10.1001/jamanetworkopen.2022.0269
--- title: 'Meteorological extremes and their impact on tinnitus-related emergency room visits: a time-series analysis' authors: - Markus Haas - Mateo Lucic - Franziska Pichler - Alexander Lein - Faris F. Brkic - Dominik Riss - David T. Liu journal: European Archives of Oto-Rhino-Laryngology year: 2023 pmcid: PMC9976663 doi: 10.1007/s00405-023-07894-1 license: CC BY 4.0 --- # Meteorological extremes and their impact on tinnitus-related emergency room visits: a time-series analysis ## Abstract ### Purpose Extreme weather events are rising due to the accelerating pace of climate change. These events impact human health and increase emergency room visits (EV) for many morbidities. Tinnitus is a common cause of EVs within otolaryngology in Germany and Austria. The effect of extreme weather conditions on tinnitus-related EVs is unknown. ### Methods A total of 526 tinnitus-related EVs at a tertiary care hospital in Vienna were identified. A distributed lag non-linear model with a maximum lag period of 14 days was fitted to investigate the immediate and delayed effect of single-day and prolonged (three-day) extreme atmospheric pressure, relative humidity, mean temperature, precipitation and mean wind speed on EV rates. Extreme conditions were defined as the 1st, 5th, 95th, and 99th percentile of the meteorological variables. Relative risk (RR) is defined as risk for tinnitus-related EVs at an extreme condition compared to the risk at the median weather condition. Cumulative RR (cRR) is the total cumulated EV risk for a given time period. ### Results High relative humidity increased same-day RR for tinnitus-related EVs to 1.75. Both low and high atmospheric pressure raised cRR as early as three days after an event to a maximum of 3.24. Low temperatures mitigated cRR within 4 days, while high temperatures tended to increase risk. Prolonged precipitation reduced cRR within one day. ### Conclusion Extreme meteorological conditions are associated with tinnitus-related EV rates. Further investigation into potential causative links and underlying pathophysiological mechanisms is warranted. ### Supplementary Information The online version contains supplementary material available at 10.1007/s00405-023-07894-1. ## Introduction The latest Intergovernmental Panel on Climate Change report paints a devastating picture of the current trajectory of the global climate, describing a sharp rise in extreme weather events [1]. Heat waves and increased climate variability negatively affect human health and lead to higher mortality and morbidity [2]. Emergency departments constitute the first point of contact in managing diseases affected or exacerbated by extreme weather events, resulting in a considerable increase in emergency room visits (EV) during and after such conditions [3]. Tinnitus is the perception of sound without an external stimulus. Although frequently idiopathic, it may also be the first manifestation of severe health issues [4, 5]. Recent studies estimate that, worldwide, 740 million adults are affected by tinnitus, with an annual incidence of approximately $1\%$, making its global burden comparable to that of pain and migraine [6]. About one in 50 patients suffering from tinnitus experience severe symptoms that require urgent medical attention [6]. From an otorhinolaryngological perspective, tinnitus is a common reason for emergency department visits and ranked third at $4.9\%$ of all ear, nose and throat (ENT)-related EVs a in recent analysis of ENT emergencies in Germany [7]. Hearing loss, chronic noise exposure, stress and psychiatric comorbidities, particularly depression and anxiety, are regarded as important risk factors for tinnitus [5, 8, 9]. Patients with normal hearing and tinnitus may still display subclinical anomalies during audiological testing [10]. The pathophysiology of tinnitus is complex and involves several levels of the auditory pathway. According to the current European tinnitus guidelines, potential peripheral mechanisms include activation of cochlear N-methyl-d-aspartate receptors, outer hair cell alterations and shifts in endo-cochlear potentials [11–13]. Proposed central mechanisms range from central hyperactivity following hearing loss through neural homeostatic plasticity to cortical tonotopic remapping and hyperpolarization of thalamic neurons due to reduced sensory inputs [13, 14]. Recent studies even reported an association between tinnitus and COVID-19 [15, 16]. Given the plethora of potential underlying mechanisms and risk factors, tinnitus remains difficult to treat. Cognitive behavioral therapy is currently the only recommended treatment by European and American guidelines [13, 17]. EV rates for psychiatric (anxiety, mood disorders) and somatic risk factors (hypertension, diabetes) for tinnitus are associated with extreme weather events [18–20]. However, the current literature on weather-associated tinnitus occurrence is limited to Ménière's disease (MD), a distinct inner ear disorder characterized by a symptom triad of tinnitus, hearing loss and vertigo. Low atmospheric pressure and high humidity have been reported to increase the risk of a Ménière’s attack. In contrast, atmospheric pressure and mean temperature have been correlated with tinnitus severity [21]. However, the impact of extreme weather events on tinnitus as a primary symptom, independent of MD, remains to be investigated. Furthermore, we currently lack an understanding of how delayed effects of extreme meteorological events impact tinnitus beyond the same day. Identifying environmental factors to predict tinnitus-related EV rates has potential implications for improving the resilience of healthcare systems towards a changing climate and optimizing health resource management. In addition, the possible correlation with weather conditions could provide further insight into tinnitus pathophysiology. In the current study, we analyzed tinnitus-related EVs at a tertiary care hospital in Austria employing a distributed lag non-linear model (DLNM) to investigate the immediate and delayed effects of extreme weather events. ## Study population and meteorological data A total of 526 tinnitus-related EVs were identified by screening the electronic patient records of the Vienna General Hospital’s emergency room (ENT division) from January 1st, 2015, to December 31st, 2018, for EVs mentioning tinnitus as a current symptom. Only patients with tinnitus as a primary symptom for their EV, absence of suspicion for or former diagnosis of MD, lack of signs for ear infections or labyrinthitis and lack of reported acute hearing loss as a primary symptom were included. Basic patient information (date of visit, age, sex, referral status, admission status) and relevant clinical information regarding diagnostic workup (laterality, results of Weber’s and Rinne’s test, suspected etiology and blood pressure at presentation) were extracted for every tinnitus-related EV. Meteorological data for Vienna was provided by the national meteorological and geophysical service of Austria (Central Institution for Meteorology and Geodynamics—“Zentralanstalt für Meteorologie und Geodynamik”) from January 1st, 2015, to December 31st, 2018, including the daily atmospheric pressure, relative humidity, mean temperature, precipitation and mean wind speed. All measurements were taken within three kilometers of the Vienna General Hospital (elevation: 177 m above sea level, latitude: 48.198°; longitude: 16.3669°). ## Statistical analysis The immediate and delayed effects of extreme weather conditions on tinnitus-related EV rates were analyzed using a DLNM [22], an established model commonly used to quantify the effects of meteorological conditions and air pollution on mortality and morbidity [23, 24]. The meteorological conditions were used as independent variables. The total number of daily tinnitus-related EVs was chosen as the response variable. We describe the relationship between meteorological conditions and lags (lag days, i.e., the days following the first exposure to a certain weather condition) using natural cubic splines with five degrees of freedom (df) at equally spaced quantiles for daily meteorological conditions and equal intervals on the logarithmic scale of lags [25]. To account for potential harvesting effects, we chose a maximum lag period of 14 days [26]. Hereby, we control for a potential decrease in EV rates after extreme weather events trigger tinnitus episodes in vulnerable patients, who are less likely to seek emergency care soon after an initial visit, where they are referred to follow-up diagnostics and treatment. Natural cubic splines of time with seven df per year were employed to control for seasonality and long-term trends. We included a weekday indicator variable in the model to account for differing demand throughout the week (e.g., due to the closing of doctor’s offices, and external providers on weekends). We used a dummy variable to control for public holidays due to the expected increase in demand during those days. Akin to weekends, general practitioner’s offices, ENT doctor’s offices and outpatient ENT departments of hospitals are closed during Austrian public holidays. Thus, emergency rooms remain the sole providers during those days, which is expected to increase visitation rates. As public holidays and weekend days sometimes coincide, which may reduce the effect of public holidays on EV rates, an interaction between the weekday indicator and the public holiday dummy was included. According to the coefficient estimates, EVs were constant during the week, with a statistically significant increase in EVs on public holidays, Fridays, Saturdays and Sundays. The interaction coefficients between weekend days and public holidays were negative for Saturdays and Sundays and statistically significant for Saturdays. Hence, tinnitus-related EVs are not expected to increase further when public holidays and Saturdays coincide. Furthermore, we controlled for a procedural change in emergency room admissions using a dummy variable after the introduction of a pre-screening outpatient department in December 2016. The estimated coefficient did not show statistical significance. Finally, we calculated the Pearson correlation coefficient for every combination of the five independent weather variables. Temperature and relative humidity showed a strong negative correlation, while all other combinations showed weak to no correlation. Extreme weather events were defined as the 1st, 5th, 95th, and 99th percentile of atmospheric pressure, relative humidity, mean temperature and mean wind speed and the 95th and 99th percentile for precipitation. The median of each weather condition was used as a reference for calculating relative risk (RR) for each lag day. RR describes the risk for an increase or decrease in total daily emergency room cases with tinnitus as their chief complaint on a given lag day at extreme conditions compared to median conditions. The cumulative relative risk (cRR) was calculated by the cumulation of RR from lag0 up to lag14. cRR describes the cumulated risk for tinnitus-related EV after extreme weather events compared to the EV risk at the median weather condition within a stated period. Next to single-day events, we investigated risk after prolonged three-day long extreme weather events. The effects of prolonged events were calculated using a rolling window. The number of tinnitus-related EVs over the previous three days acted as the response variable. Lag0 refers to the last day of the three-day window. The 1st, 5th, 95th, and 99th percentile of the three-day mean for atmospheric pressure, relative humidity, mean temperature and mean wind speed and the 95th and 99th percentile of the sum of precipitation over three days were defined as prolonged extreme conditions. Numeric values for RR (single-day: Supplementary Table 1; prolonged: Supplementary Table 3) and cRR (single-day: Supplementary Table 2; prolonged: Supplementary Table 4) were extracted at day 0, day 1, day 3, day 7 and day 14. RR and cRR values are reported with their $95\%$ confidence interval (CI) and p-value throughout the manuscript. R software (version 4.1.3) was used to perform statistical testing [27]. The DLNM was fitted using the R package “dlnm” [28]. ## Study population and weather From January 1st, 2015, to December 31st, 2018, 526 tinnitus-related EVs occurred at the Vienna General Hospital in Vienna, Austria. Patient characteristics for each EV are shown in Table 1. The study population had a higher proportion of male patients ($58.4\%$) with a median age of 37 (range: 11–89). Most patients suffered from one-sided tinnitus ($75.5\%$) and showed no indication of hearing loss, as determined by Weber’s and Rinne’s test ($80.4\%$). In those with pathologic results, $15.7\%$, $1.3\%$, and $2.6\%$ of patients showed signs of sensorineural loss, conductive loss and combined loss, respectively. The etiology was largely idiopathic ($95\%$). However, some cases were attributable to recent acoustic trauma ($2.5\%$) and cerumen ($2.5\%$). From the small number of cases with available blood pressure measurements ($$n = 88$$), $55.7\%$ of patients were hypertensive. On average, 2.5 tinnitus-related EVs occurred per week. Daily presentation rates throughout 2015–2018 are shown in Fig. 1. While the highest numbers of tinnitus-related EVs occurred during January ($$n = 55$$) and October ($$n = 54$$), no clear seasonal trends were observed. All meteorological variables in Vienna during the observational period are shown in Supplementary Fig. 1. Relative humidity and mean temperature showed clear differences between the seasons. Atmospheric pressure, precipitation and mean wind speed were less affected by seasonality. Table 1Patient characteristics of the study cohortPatient characteristics for tinnitus-related EVn (% missing)Count (%)Age526 ($0\%$)–0–18–19 ($3.6\%$)19–29–165 ($31.4\%$)30–44–155 ($29.5\%$)45–64–141 ($26.8\%$)65+–46 ($8.7\%$)Sex526 ($0\%$)–Male–307 ($58.4\%$)Female–219 ($41.6\%$)Referral by an external provider526 ($0\%$)–Yes–83 ($15.8\%$)No–443 ($84.2\%$)Admission required526 ($0\%$)–Yes–3 ($0.1\%$)No–523 ($99.9\%$)Laterality of tinnitus458 ($12.9\%$)–One-sided–346 ($75.5\%$)Two-sided–112 ($24.5\%$)Weber’s and Rinne’s test312 ($40.7\%$)–No indication of hearing loss–251 ($80.4\%$)Indicating sensorineural loss–49 ($15.7\%$)Indicating conductive loss–4 ($1.3\%$)Indicating combined loss–8 ($2.6\%$)Etiology526 ($0\%$)–Idiopathic–496 ($95.0\%$)Acoustic trauma–13 ($2.5\%$)Cerumen–13 ($2.5\%$)Arterial hypertension at presentationa88 ($83.3\%$)–Yes–49 ($55.7\%$)No–39 ($44.3\%$)aSystolic pressure > 140 mmHgFig. 1Weekly tinnitus-related EVs are shown as the percentage of total EVs from any cause from 2015 to 2018 using smoothing splines (lines) and daily values (background) ## Atmospheric pressure Based on previous reports on the link between atmospheric pressure and tinnitus symptoms in patients suffering from MD [21, 29], we first aimed to investigate the effect of extreme atmospheric pressure events on tinnitus symptoms in patients without MD. On the same day, extreme atmospheric pressure events did not significantly affect tinnitus-related EV. Over the subsequent lag period of single-day events, the earliest impact of extremely low atmospheric pressure occurred after 3 days at 983 hPa (P5) with an RR of 1.40 [1.04–1.89; $$p \leq 0.028$$]. For the same condition, cRR was elevated within 4 days to 1.88 [1.03–3.44; $$p \leq 0.038$$] and was significantly elevated for the remaining lag period with the highest cRR of 3.24 [1.04–10.15; $$p \leq 0.044$$] at day 14 (Fig. 2A). Similarly, extremely high atmospheric pressure at 1009 hPa (P95) increased RR to 1.12 [1.01–1.25; $$p \leq 0.028$$] starting at day 8 and elevated cRR to 2.31 [1.04–5.10; $$p \leq 0.038$$] on day 11.Fig. 2Line-plots of cRR for tinnitus-related EV are shown for single-day extreme weather events from lag0 to lag14 defined as the 1th, 5th, 95th, and 99th percentile of atmospheric pressure in hPa (A), relative humidity in % (B), mean temperature in °C (C), precipitation in mm [P95 & P99 only] (D) and mean wind speed in m/s (E). Confidence intervals ($95\%$) are shown in grey. Significant decreases (green) and increases (red) in cRR (p ≤ 0.05) are highlighted on the lag-axis Prolonged extremely low atmospheric pressure showed comparable results with the earliest elevation of cRR at 980 hPa (P1) to 1.7 [1.09–2.66; $$p \leq 0.020$$] within 4 days and at 985 hPa (P5) to 1.42 [1.09–1.85; $$p \leq 0.010$$] within 3 days, respectively (Fig. 3A). Following the single-day results, extremely high atmospheric pressure over 3 days at an average 1008 hPa (P95) raised the cRR to 1.36 [1.04–1.77; $$p \leq 0.026$$] within 3 days and led to a sustained cRR elevation over the remaining 14-day observational period with its maximum on day 13 at 2.23 [1.43–3.47; $p \leq 0.001$].Fig. 3Line-plots of cRR for tinnitus-related EV are shown for prolonged extreme weather events over three days from lag0 to lag14 defined as the 1th, 5th, 95th, and 99th percentile of atmospheric pressure in hPa (A), relative humidity in % (B), mean temperature in °C (C), precipitation in mm [P95 & P99 only] (D) and mean wind speed in m/s (E). Confidence intervals ($95\%$) are shown in grey. Significant decreases (green) and increases (red) in cRR (p ≤ 0.05) are highlighted on the lag-axis Taken together, atmospheric pressure increased tinnitus-related EV risk at low and high extremes as early as 3 days after the weather event. ## Relative humidity After we found extreme atmospheric pressure conditions to be a risk-increasing factor for tinnitus-related EV, we then analyzed the impact of relative humidity due to its correlation with Ménière’s attacks [21]. On the same day, extremely high relative humidity at $92\%$ (P99) increased the risk for tinnitus-related EVs to 1.75 [1.01–3.03; $$p \leq 0.046$$]. Over the subsequent lag period for single-day events, extremely low relative humidity at $34\%$ (P1) significantly decreased RR between 8 and 11 days after the event. cRR at $34\%$ (P1) was reduced within 1 day to 0.40 [0.16–1.00; $$p \leq 0.050$$] and remained significantly decreased over the entire 14-day observational period with a cRR as low as 0.09 [0.02–0.39; $$p \leq 0.002$$] by day 12 (Fig. 2B). Extremely high relative humidity had no significant delayed effect. Prolonged extremely low relative humidity over 3 days impacted the risk for tinnitus-related EV in a comparable way, with a significantly reduced RR between day 8 and day 12. cRR was decreased within 2 days to 0.44 [0.26–0.74, $$p \leq 0.002$$] at − 4 °C (P1) and within 3 days to 0.59 [0.43–0.79; $p \leq 0.001$] for 0 °C (P5). ( Fig. 3B) An extremely high three-day average relative humidity of $89\%$ (P99) decreased RR to 0.77 [0.62–0.95; $$p \leq 0.016$$] at day 4. However, cRR was significantly increased to 1.78 [1.11–2.85; $$p \leq 0.018$$] within 2 days after a prolonged extremely high humidity of $89\%$ (P99). In summary, extremely low relative humidity showed a pronounced reductive effect on tinnitus-related EV as early as one day after the event. Conversely, extremely high relative humidity significantly increased same-day risk and showed a bidirectional impact over the subsequent lag period. ## Mean temperature Next, we investigated mean daily temperature as a possible factor in tinnitus-related EV rates based on its described role in tinnitus severity in MD [21]. On the same day, extreme temperature events did not significantly affect EV risk. Over the subsequent lag period of single-day events, extremely low temperatures of 0 °C (P5) led to a decreased RR of 0.66 [0.47–0.92; $$p \leq 0.014$$] on day 4 and a decreased cRR from day 5 to day 7 with a low-point of 0.41 [0.19–0.91; $$p \leq 0.028$$] at day 6 (Fig. 2C). Extremely high temperatures of 27 °C (P95) significantly affected RR in both directions across different lag days, with a reduction to 0.33 [0.12–0.93; $$p \leq 0.036$$] on day 2 and an increase to 1.39 [1.01–1.91; $$p \leq 0.044$$] on day 4. cRR was not significantly impacted by extremely high temperatures. Prolonged extreme conditions over three days in the form of cold waves exerted a risk-mitigating effect on tinnitus-related EVs (Fig. 3C). The highest reduction of RR was observed on day 4 to 0.68 [0.46–1.00, $$p \leq 0.048$$] after cold waves averaging 0 °C (P5). cRR was reduced after cold waves averaging both − 4 °C (P1) and 0 °C (P5) within 4 to 10 days. The lowest risk was observed at − 4 °C (P1) within 6 days at a cRR of 0.34 [0.18–0.64, $p \leq 0.001$]. Heat waves affected RR for tinnitus-related EV in both directions, although the risk-increasing effect was more pronounced. At day 3, RR was increased to 1.83 [1.02–3.30; $$p \leq 0.044$$] for 26 °C (P95) and to 2.61 [1.15–5.89; $$p \leq 0.022$$] for 30 °C (P99). cRR was not significantly affected by heatwaves. In brief, extremely low temperatures reduced the risk for tinnitus-related EV as early as 4 days after the weather event. Extremely high temperatures showed a non-linear effect on RR starting 2 days after an event with a tendency towards increased risk. ## Precipitation Since the effect of precipitation on tinnitus has not yet been described, we chose to analyze it as an additional potential meteorological factor in tinnitus-related EV rates. For single-day events, extremely high precipitation had no significant effect on same-day risk, nor did it show delayed effects in the subsequent 14-day observational period. ( Fig. 2D) On the other hand, prolonged extremely high precipitation over 3 days had a risk-mitigating effect on tinnitus-related EV. At its earliest, RR was decreased to 0.62 [0.41–0.94; $$p \leq 0.026$$] 1 day after prolonged precipitation totaling 24 mm (P95). cRR for the same condition was decreased within 1 day to 0.53 [0.36–0.78, $$p \leq 0.002$$] and remained significantly reduced over the entire lag period (Fig. 3D). Late effects of prolonged precipitation totaling 40 mm (P99) moderately increased RR between day 8 and day 11 ($p \leq 0.050$). However, these effects did not remain significant after cumulation. Therefore, the data indicates a risk-reducing effect of prolonged, but not single-day, extreme precipitation for tinnitus-related EV as early as 1 day after the weather event. ## Mean wind speed Finally, we investigated mean wind speed, which does not appear to affect tinnitus in MD [21], to dissect potential differences in weather impact in tinnitus unrelated to MD. On the same day, extreme wind speeds did not significantly impact risk for tinnitus-related EV. Over the subsequent lag period of single-day events, the earliest effect of extremely high wind speeds was observed at 6 m/s (P95) on day 3 with a RR of 1.36 [1.01–1.84, $$p \leq 0.046$$]. However, these effects did not remain significant after risk cumulation. ( Fig. 2E) Extremely low wind speeds showed no significant impact on RR or cRR. Prolonged extremely low wind speeds at 2 m/s led to a decreased RR between day 8 and day 14, with a low-point of 0.88 [0.81–0.95, $p \leq 0.001$] by day 14. cRR was significantly reduced by day 12 to 0.67 [0.46–0.98, $$p \leq 0.040$$] (Fig. 3E). Extremely high wind speeds over 3 days only showed late, marginal effects on RR after day 9, with no significant results after cumulation. To summarize, extremely high single-day wind speeds had minor, early effects on tinnitus-related EV risk by day 2. In contrast, extremely low prolonged wind speeds led to a moderate, late reduction of EV risk starting at day 8. ## Discussion In this study, we investigated the effect of extreme meteorological conditions on tinnitus-related EVs. Previous reports on meteorological effects on tinnitus are limited to study populations suffering from MD. Here, we report both parallels and differences in the impact of weather on EV risk for tinnitus without signs of MD. Atmospheric pressure has been the best-described meteorological factor influencing MD attacks and tinnitus severity. Higher atmospheric pressure was positively correlated with the diagnosis of MD in a large-scale study analyzing over 7000 cases in South Korea [30]. Another prospective study identified increases in atmospheric pressure as risk factors for an MD episode on the following day [29]. More specifically for tinnitus, a study utilizing a mobile app for self-reporting symptoms for MD patients showed that low atmospheric pressure was associated with increased tinnitus severity [21]. In our study, extremely low and high atmospheric pressure led to delayed increases in tinnitus-related EV risk as early as two days after an event. The possible underlying mechanisms at work are likely multifactorial. For one, associated risk factors for tinnitus, such as headaches [31] and depressive symptoms [32], have been correlated with atmospheric pressure. They may, therefore, indirectly promote tinnitus symptoms. On the other hand, environmental air pressure may disturb the tympanic pressure equilibrium, affecting tinnitus cases with underlying middle and inner ear etiologies. An animal model investigating middle-ear and intracranial pressure and its effect on electrocochleographic response has shown that static middle-ear pressure and an unimpaired eustachian tube are vital in maintaining normal cochlear function [33]. A recent study demonstrated that pressure changes in the external auditory canal can result in middle ear muscle contractions and eustachian tube dysfunction, which, in turn, can lead to tinnitus [34]. With regards to direct cochlear effects, atmospheric pressure changes have been shown to cause alterations in the cochlea of rats including partial loss of stereocilia [35]. Loss of stereocilia may impair the function of outer hair cells (OHC) and lead to aberrant spontaneous activity of the cochlea, which is one of the current hypotheses for the pathophysiology of cochlear tinnitus [13]. Spontaneous activity and prolonged depolarization of inner hair cells is thought to arise from an excitatory shift following stereocilia damage or degeneration of OHC, detachment of OHC from the tectorial membrane or from pressure increases in the scala media [12, 36]. Although no conclusions regarding tinnitus pathophysiology can be drawn from our study, the reported associations between atmospheric pressure extremes and EV rates lend further support to the previously reported links between atmospheric pressure and tinnitus. Nevertheless, since the exact mechanisms still need to be fully understood, further investigation is needed to better explain the association between atmospheric pressure and tinnitus both in MD- and non-MD-related cases. Another link has been drawn between high relative humidity and increased frequency of Ménière’s attacks and tinnitus [21, 30]. Our study found comparable results for tinnitus-related EVs independent of MD with extremely low relative humidity reducing EV risk within one day and a same-day increase in risk under extremely humid conditions. Currently, the scientific literature lacks any leads to how relative humidity could affect tinnitus directly. An infodemiological study reported a winter peak in web-based inquiries for tinnitus [37]. Relative humidity is generally higher in winter as colder air has a decreased capability of holding water vapor and saturates at lower levels than hot air. However, our study did not show seasonal variations in tinnitus-related EVs, which occur mostly due to acute, severe, or debilitating cases of tinnitus, making an effect of relative humidity due to seasons unlikely. Another risk factor for tinnitus is SSNHL. A study on audiogram configurations in patients with SSNHL has suggested a correlation between ascending audiogram patterns and high relative humidity on the day of onset [38]. The authors discuss a potential pathophysiological similarity between MD and SSNHL with ascending patterns due to endolymphatic hydrops based on the overlapping findings on high relative humidity as a risk factor. However, in our study, over $80\%$ of patients with available Weber’s and Rinne’s test assessments did not show indication of hearing loss, albeit not confirmed by pure-tone audiometry due to the emergency setting. As a result, we cannot attribute the association between extreme relative humidity and tinnitus to hearing loss for most of the study population. In contrast with a reported negative correlation between temperature and tinnitus severity in MD [21], our study found extremely low temperatures to be a mitigating risk factor for tinnitus-related EVs within 5 and 4 days after the single-day events and cold waves, respectively. These findings are further contrasted by a Japanese report of higher rates of MD attacks after the passing of a cold front [39]. Our results, therefore, suggest a differing temperature response in non-MD-related tinnitus cases. Precipitation and its potential effect on tinnitus has so far not been investigated. Our findings suggest that prolonged extremely high precipitation exerts a mitigating effect on tinnitus-related EVs. Although speculative, the lack of impact of single-day events compared to prolonged precipitation could potentially suggest that the auditory stimulus of prolonged precipitation in the form of rainfall during warmer months has a soothing effect on tinnitus. Sound therapy is one treatment option for patients with chronic tinnitus. Although acoustic stimulation may bring some relief to patients by masking tinnitus symptoms, the treatment approach is not aimed at the underlying cause of tinnitus and high-level evidence is limited. While European guidelines currently do not recommend sound therapy, American guidelines mention sound therapy as an option for patients with bothersome, persistent tinnitus [13, 17]. A recent study employed a smartphone-based noise generator to investigate the benefit of enriching the sound environment of chronic tinnitus patients [40]. Most participants opted to listen to environmental sounds and rainfall. At the 3-month follow-up, participants reported a reduced overall tinnitus severity. Extreme precipitation in the form of prolonged rainfall may, therefore, provide an auditory stimulus leading to transient symptom relief in patients suffering from tinnitus and resulting in a temporary reduction in EV rates. Lastly, extreme mean wind speeds showed minor late effects on cumulative risk within 12 days, which are unlikely to be clinically relevant. In accordance, wind speed did not correlate with tinnitus severity or attack onset in MD patients [21]. Our study had several limitations. First, the retrospective study design carries inherent bias concerning data availability and homogeneity concerning diagnostic follow-up and whereabouts of patients during extreme weather events. Although EVs of patients who reported acute hearing loss as their primary complaint were excluded, we were unable to distinguish between patients suffering from tinnitus with or without hearing loss due to the lack of pure tone audiometry data. Additionally, seasons and occupations of patients can have a major impact on the degree to which patients are directly exposed to weather conditions. Nevertheless, our study provides new avenues for future studies that are appropriately designed to receive real-time data of patients’ tinnitus symptoms and whereabouts during such events (e.g., smart phone app-based studies). Second, our model considered a lag period of 14 days. While we controlled for demand fluctuations related to weekdays and public holidays, we were unable to control for other potential confounders, such as degree of weather exposure, that may have impacted EV rates during the two weeks after extreme conditions. Next, we could not distinguish between rainfall and snowfall, which is common during winter months in Vienna, based on the available meteorological data. Regarding generalizability, the tinnitus-related EVs took place at a single tertiary care hospital resulting in a study population with acute and more severe cases. Our findings are, therefore, not applicable to all tinnitus cases. Equivalently, this study focused on extreme weather events. Therefore, the reported associations between extreme weather and tinnitus-related EVs are not applicable to moderate weather conditions. Finally, while our study generates potential hypotheses for underlying mechanisms of how extreme meteorological conditions impact EVs for tinnitus, our study was not designed to prove any causative link between the reported associations. ## Conclusion Extreme weather conditions are associated with changes in tinnitus-related EV rates. Extremely high relative humidity significantly increased same-day RR. Extreme atmospheric pressure and temperature, as well as prolonged extreme precipitation, showed delayed effects on cRR. These findings warrant further clinical and experimental research on how meteorological variables affect tinnitus. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 449 KB) ## References 1. Tollefson J. IPCC climate report: earth is warmer than it's been in 125,000 years. Nature (2021.0) 596 171-172. DOI: 10.1038/d41586-021-02179-1 2. Ebi KL. Extreme weather and climate change: population health and health system implications. Annu Rev Public Health (2021.0) 42 293-315. DOI: 10.1146/annurev-publhealth-012420-105026 3. Hughes HE. Using an emergency department syndromic surveillance system to investigate the impact of extreme cold weather events. Public Health (2014.0) 128 628-635. DOI: 10.1016/j.puhe.2014.05.007 4. Zenner HP. A multidisciplinary systematic review of the treatment for chronic idiopathic tinnitus. Eur Arch Otorhinolaryngol (2017.0) 274 2079-2091. DOI: 10.1007/s00405-016-4401-y 5. Baguley D, McFerran D, Hall D. Tinnitus. Lancet (2013.0) 382 1600-1607. DOI: 10.1016/s0140-6736(13)60142-7 6. Jarach CM. Global prevalence and incidence of tinnitus: a systematic review and meta-analysis. JAMA Neurol (2022.0) 79 888-900. DOI: 10.1001/jamaneurol.2022.2189 7. Kuhr E, Werner JA, Dietz A, Wiegand S. Analysis of the emergency patients of a university ENT hospital. Laryngorhinootologie (2019.0) 98 625-630. DOI: 10.1055/a-0916-8916 8. Jafari Z. Tinnitus, sound intolerance, and mental health: the role of long-term occupational noise exposure. Eur Arch Otorhinolaryngol (2022.0) 279 5161-5170. DOI: 10.1007/s00405-022-07362-2 9. Elarbed A, Fackrell K, Baguley DM, Hoare DJ. Tinnitus and stress in adults: a scoping review. Int J Audiol (2021.0) 60 171-182. DOI: 10.1080/14992027.2020.1827306 10. Park Y, Shin SH, Byun SW, Lee ZY, Lee HY. Audiological and psychological assessment of tinnitus patients with normal hearing. Front Neurol (2022.0) 13 1102294. DOI: 10.3389/fneur.2022.1102294 11. Guitton MJ. Salicylate induces tinnitus through activation of cochlear NMDA receptors. J Neurosci (2003.0) 23 3944-3952. DOI: 10.1523/jneurosci.23-09-03944.2003 12. Noreña AJ. Revisiting the cochlear and central mechanisms of tinnitus and therapeutic approaches. Audiol Neurootol (2015.0) 20 53-59. DOI: 10.1159/000380749 13. Cima RFF. A multidisciplinary European guideline for tinnitus: diagnostics, assessment, and treatment. HNO (2019.0) 67 10-42. DOI: 10.1007/s00106-019-0633-7 14. Eggermont JJ, Roberts LE. The neuroscience of tinnitus. Trends Neurosci (2004.0) 27 676-682. DOI: 10.1016/j.tins.2004.08.010 15. Kartal A, Kılıç M. Tinnitus in patients recovering after COVID-19: observational and cross-sectional study. Eur Arch Otorhinolaryngol (2023.0) 280 573-580. DOI: 10.1007/s00405-022-07501-9 16. Nocini R, Lippi G, Mattiuzzi C. Impact of COVID-19 pandemic on the worldwide burden of tinnitus. Eur Arch Otorhinolaryngol (2023.0) 280 945-946. DOI: 10.1007/s00405-022-07790-0 17. Tunkel DE. Clinical practice guideline: tinnitus. Otolaryngol Head Neck Surg (2014.0) 151 S1-s40. DOI: 10.1177/0194599814545325 18. Davis RE, Driskill EK, Novicoff WM. The association between weather and emergency department visitation for diabetes in Roanoke, Virginia. Int J Biometeorol (2022.0) 66 1589-1597. DOI: 10.1007/s00484-022-02303-4 19. Lin YK, Zafirah Y, Ke MT, Andhikaputra G, Wang YC. The effects of extreme temperatures on emergency room visits-a population-based analysis by age, sex, and comorbidity. Int J Biometeorol (2021.0) 65 2087-2098. DOI: 10.1007/s00484-021-02166-1 20. Yoo EH, Eum Y, Roberts JE, Gao Q, Chen K. Association between extreme temperatures and emergency room visits related to mental disorders: a multi-region time-series study in New York, USA. Sci Total Environ (2021.0) 792 148246. DOI: 10.1016/j.scitotenv.2021.148246 21. Schmidt W. The weather and Ménière's disease: a longitudinal analysis in the UK. Otol Neurotol (2017.0) 38 225-233. DOI: 10.1097/mao.0000000000001270 22. Gasparrini A, Armstrong B, Kenward MG. Distributed lag non-linear models. Stat Med (2010.0) 29 2224-2234. DOI: 10.1002/sim.3940 23. Wang YC, Lin YK. Association between temperature and emergency room visits for cardiorespiratory diseases, metabolic syndrome-related diseases, and accidents in metropolitan Taipei. PLoS ONE (2014.0) 9 e99599. DOI: 10.1371/journal.pone.0099599 24. Goldberg MS. Associations between ambient air pollution and daily mortality among elderly persons in Montreal, Quebec. Sci Total Environ (2013.0) 463–464 931-942. DOI: 10.1016/j.scitotenv.2013.06.095 25. Bai L. Temperature, hospital admissions and emergency room visits in Lhasa, Tibet: a time-series analysis. Sci Total Environ (2014.0) 490 838-848. DOI: 10.1016/j.scitotenv.2014.05.024 26. Zhou H, Geng H, Dong C, Bai T. The short-term harvesting effects of ambient particulate matter on mortality in Taiyuan elderly residents: a time-series analysis with a generalized additive distributed lag model. Ecotoxicol Environ Saf (2021.0) 207 111235. DOI: 10.1016/j.ecoenv.2020.111235 27. 27.R Core Team (2022) R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/ 28. Gasparrini A. Distributed lag linear and non-linear models in R: the package dlnm. J Stat Softw (2011.0) 43 1-20. DOI: 10.18637/jss.v043.i08 29. Gürkov R. Atmospheric pressure and onset of episodes of menière's disease—a repeated measures study. PLoS ONE (2016.0) 11 e0152714. DOI: 10.1371/journal.pone.0152714 30. Choi HG. Effects of short- and long-term exposure to air pollution and meteorological factors on Meniere's disease. Sci Rep (2021.0) 11 16063. DOI: 10.1038/s41598-021-95491-9 31. Mukamal KJ, Wellenius GA, Suh HH, Mittleman MA. Weather and air pollution as triggers of severe headaches. Neurology (2009.0) 72 922-927. DOI: 10.1212/01.wnl.0000344152.56020.94 32. Brazienė A. Relationship between depressive symptoms and weather conditions. Int J Environ Res Public Health (2022.0). DOI: 10.3390/ijerph19095069 33. Franz B, Anderson C. Effect of static middle-ear and intracranial pressure changes on differential electrocochleographic response. Int Tinnitus J (2008.0) 14 101-107. PMID: 19205159 34. Fournier P. Exploring the middle ear function in patients with a cluster of symptoms including tinnitus, hyperacusis, ear fullness and/or pain. Hear Res (2022.0) 422 108519. DOI: 10.1016/j.heares.2022.108519 35. Eroglu S. Repeated atmospheric pressure alteration effect on the cochlea in rats: experimental animal study. Aerosp Med Hum Perform (2021.0) 92 550-555. DOI: 10.3357/amhp.5732.2021 36. Liberman MC, Dodds LW. Acute ultrastructural changes in acoustic trauma: serial-section reconstruction of stereocilia and cuticular plates. Hear Res (1987.0) 26 45-64. DOI: 10.1016/0378-5955(87)90035-9 37. Plante DT, Ingram DG. Seasonal trends in tinnitus symptomatology: evidence from Internet search engine query data. Eur Arch Otorhinolaryngol (2015.0) 272 2807-2813. DOI: 10.1007/s00405-014-3287-9 38. Zhang J, Ji S, Ma X, Yu L, Jing Y. Association between meteorological factors and audiogram configurations in patients with sudden sensorineural hearing loss: a cross-sectional study. BMJ Open (2021.0) 11 e045768. DOI: 10.1136/bmjopen-2020-045768 39. Mizukoshi K. Influence of a cold front upon the onset of Meniére's disease in Toyama. Japan Acta Otolaryngol Suppl (1995.0) 520 412-414. DOI: 10.3109/00016489509125285 40. Kutyba JJ, Jędrzejczak WW, Gos E, Raj-Koziak D, Skarzynski PH. Chronic tinnitus and the positive effects of sound treatment via a smartphone app: mixed-design study. JMIR Mhealth Uhealth (2022.0) 10 e33543. DOI: 10.2196/33543
--- title: 'Association of scrub typhus with incidence of dementia: a nationwide population-based cohort study in Korea' authors: - Jooyun Kim - Hyeri Seok - Ji Hoon Jeon - Won Suk Choi - Gi Hyeon Seo - Dae Won Park journal: BMC Infectious Diseases year: 2023 pmcid: PMC9976677 doi: 10.1186/s12879-023-08107-0 license: CC BY 4.0 --- # Association of scrub typhus with incidence of dementia: a nationwide population-based cohort study in Korea ## Abstract ### Background Scrub typhus is a mite-borne infectious rickettsial disease that can occur in rural and urban areas, with an especially high prevalence in older populations. This disease causes systemic vasculitis that can invade the central nervous system. Considering these characteristics, here we examined whether scrub typhus was associated with the occurrence of dementia, using large population-based cohort data. ### Method This population-based cohort study enrolled patients aged 60–89 years using data from the Health Insurance Review and Assessment database of South Korea between 2009 and 2018. We defined scrub typhus and dementia using International Classification of Diseases, Tenth Edition diagnostic codes. The control group was stratified according to age and sex at a ratio of 1:5 to the case group in the study population. The index date was set after 90 days beyond the date of the scrub typhus diagnosis, while the observation period was from the time of the index appointment to December 31, 2020. The primary outcome was newly diagnosed dementia. The secondary outcome was dementia classification, such as Alzheimer’s disease, vascular dementia, and other. All analyses were conducted by matching age, gender, and comorbidity. ### Results During the observation period, 10,460 of 71,047 ($14.7\%$) people who had a history of scrub typhus versus 42,965 of 355,235 ($12.1\%$) people in the control group, that is, with no history of scrub typhus, were diagnosed with dementia (adjusted hazard ratio, 1.12; $95\%$ confidence interval, 1.10–1.15, $p \leq 0.001$). The Kaplan–Meier curves for time to cumulative incidence of dementia showed that the dementia incidence in both groups increased over time, while individuals with a past history of scrub typhus had a higher incidence of dementia than the control group. Second, the risk of Alzheimer’s disease was significantly higher among patients with a history of scrub typhus (adjusted hazard ratio, 1.15; $95\%$ confidence interval 1.13–1.18, $p \leq 0.001$). ### Conclusion In conclusion, a history of scrub typhus infection in old age is significantly associated with an increase in dementia, especially Alzheimer’s disease. Our results suggest that prevention and appropriate treatment of scrub typhus should be emphasized as a dementia prevention measure. ## Background Scrub typhus is an infectious disease that is transmitted by human bites from mites infected with Orienta tsutsugamushi, a gram-negative coccobacillus in the family Rickettsiaceae. This bacterium infects host vascular endothelial cells and is released from infected cells, causing systemic vasculitis such as Scrub typhus [1]. Most cases are found in Southeast Asia, China, Japan, India, Indonesia and northern Australia, especially in rural areas and both residents and visitors are susceptible to infection [2]. In 2016, there were nearly 11,000 new cases in South Korea. Since then, more than 4,000 new cases have been reported annually [3]. In terms of age distribution, the proportion of affected older individuals aged 60 years and older is high [4, 5]. Unlike other infectious diseases, scrub typhus is characterized by systemic vasculitis. In the early stages of the infection, proper antibiotic treatment results in good responses. However, delayed treatment can lead to severe systemic progression and complications ranging from pneumonia and acute renal injury to invasion of the central nervous system (CNS) resulting in meningitis or encephalitis [6]. The role of scrub typhus in the long-term development of CNS complications remains unknown. Dementia is a degenerative disorder that affects the brain, with more than $90\%$ of cases diagnosed after 65 years old as the age is the leading risk factor. As a result, the increase in the mean population age worldwide is associated with a steady increase in the incidence and prevalence of dementia [7]. In 2015, according to the World Alzheimer Report, there are nearly 46.8 million dementia cases worldwide, which is expected to become 131.5 million by 2050 [8]. Although the pathogenesis of dementia remains unclear, it may be associated with comorbidities affecting the blood vessels, such as hypertension and diabetes [9, 10]. Other infectious diseases such as Borrelia burgdorferi, Helicobacter pylori, Chlamydia pneumoniae and herpes simplex virus type 1 (HSV-1) may also be associated with dementia [11–16]. Considering the characteristics of scrub typhus infection (occurring frequently in old age and involving the inflammation of systemic blood vessels that can invade the CNS), scrub typhus may be associated with the occurrence of dementia. This study aimed to investigate this hypothesis using large population-based cohort data from the Health Insurance Review and Assessment Service (HIRA) in South Korea. ## Data source HIRA is the data source responsible for claim reviews and quality assessments of the National Health Insurance (NHI) and National Medical Aid (NMA) programs, of which all medical institutions and local pharmacies are compulsory participants. All medical expenditures within the NHI and NMA programs are monitored by HIRA for redemption purposes. The HIRA claims database contains basic information on patients’ sociodemographic characteristics and prescriptions, diagnosis and diagnostic procedures [17]. This dataset contains coded information for all of the diagnoses, tests, and treatments prescribed in South Korea since January 2007. ## Study population We set the washout period for all factors that may affect the occurrence of dementia from January 2007 to December 2008. From January 2009 to December 2018, among people aged 60 to 89, those diagnosed with scrub typhus were classified into the scrub typhus group. The index date was set after 90 days of scrub typhus diagnosis date to exclude the screening effect on dementia diagnosis. The operational definition of scrub typhus was as follows: [1] International Classification of Diseases, Tenth Edition (ICD-10) diagnostic code corresponding to scrub typhus (A75.3) during the registration period; and [2] prescription of doxycycline, the treatment of scrub typhus, for at least three days a week before or after the day when the diagnostic code was added. The operational definition of dementia was as follows: [1] code for dementia (V-code V800, V801, V810, V811) in expending benefit coverage, reduces medical costs for patients with fatal diseases, during the follow-up period; or [2] at least one diagnostic code for dementia (F00.x, F01.x, F02.x, F03.x, G30.x), examination code (Mini-Mental State Examination [MMSE F6216] and one Global Deterioration Scale score [GDS, F6221] and Clinical Dementia Rating score [CDR, F6222]), or drug code (donepezil [1486, 6434], rivastigmine [2245], and/or galantamine [3852]) within the adjacent 4 weeks. In South Korea’s Health Insurance System, hospitals can receive medical expenses after the evaluation and approval of the Review and Assessment Service; it was considered that the possibility of falsely coding as dementia would be low. We also classified patients by dementia subgroup into Alzheimer’s disease (AD), vascular dementia, and other, setting the operational definition of the subtype of dementia by using diagnostic codes. Among people diagnosed with dementia, those diagnosed with F00.x or G30.x code corresponding to Alzheimer’s were set as Alzheimer’s group, and those who received F01.x code corresponding to vascular dementia were set as vascular dementia group. As a control group, among the data sets of HIRA, those aged 60–89 who were not infected with scrub typhus from January 2007 to December 2020 and applied to the exclusion criteria were randomly selected, stratified by age and sex at a ratio of 1:5. We excluded those who were diagnosed with scrub typhus or dementia before the end of the washout period, for whom covariate data were missing, or who died during the observation period. Institutional Review Board of Korea University Ansan Hospital (2022AS0001) and the HIRA [2022-001] approved the study. They also waived the informed consent to the study. We report the results in compliance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines [18]. ## Outcomes The observation period was set from the index date to December 31, 2020. The primary outcome of this study was the event and timing of newly diagnosed dementia during the observation period. The secondary outcome was set as a subgroup of dementia that occurred during the observation period. ## Covariates All variables that were used to treat the disease at least once between January 2007 and the index date were included as covariates. Age and sex were set at the time of the index date, and age was classified as 60–69 years or 70–89 years. The other covariates were defined as those with an ICD-10 diagnostic code. The analysis was adjusted for several known risk factors, such as age, sex, central degenerative disease (G20.x and G31.x), stroke (I63.x, I64.x, and I69.x), and hypertension (I10.x–I15.x), diabetes mellitus (E11.x-E14. x), and major depressive disorder (F32.x, F33.x) [14]. ## Statistical analysis We represent the Categorical variables using numbers (%) and mean ± SD for continuous variables. Fisher’s exact test, Student’s t-test and the chi-squared test were conducted for intergroup comparisons. Cox proportional hazards analysis for the incidence of dementia was performed to estimate the hazards of dementia incidence with the adjustment for age, sex, and comorbidities, and hazard ratios (HRs) were calculated with $95\%$ confidence intervals (CI). The cumulative incidence curves of dementia were constructed and compared between groups using Kaplan–Meier curves. The follow-up period started on the index date and was censored on the date of diagnosis of dementia or the end of the study. All analyses were conducted by matching age, gender, and comorbidity. Statistical significance was considered when the two-tailed P value was < 0.05. On the other hand, R version 4.0.2 was the implemented statistical problem. ## Results At the baseline, 76,073 individuals aged 60–89 years old were confirmed to have scrub typhus between January 2009 and December 2018. After the exclusion of participants who were diagnosed with scrub typhus during the washout period ($$n = 233$$), those diagnosed with dementia before the index date ($$n = 3991$$), and those who died before the index date ($$n = 802$$), a total of 71,047 subjects were included in the scrub typhus group. A control group was randomly selected at a ratio of 1:5 with sex and age matching; thus, 355,235 subjects were included (Fig. 1). The mean follow-up period was 5.6 ± 2.9 years. The case group was followed for 5.7 ± 2.8 years versus the control group for 5.6 ± 2.9 years. Fig. 1Flow chart of the study population All subjects of characteristics are presented in Table 1. Since the control group was selected by matching sex and age with the case group, the ratio of males to females was 37.3–$62.7\%$. Likewise, for age, the ratios were the same between groups: $47.1\%$ were 60–69 years old, while $52.9\%$ were 70–89 years old. Table 1Participants’ characteristics values are shown as n (%) or mean ± standard deviationTotal ($$n = 426$$,282)Past history of scrub typhus($$n = 71$$,047)Non-history of scrub typhus($$n = 355$$,235)P valueAge Mean age (years ± SD)70.6 ± 7.070.6 ± 7.070.6 ± 7.0> 0.999 60–69 (n, %)200,580 (47.1)33,430 (47.1)167,150 (47.1)> 0.999 70–89 (n, %)225,702 (52.9)37,617 (52.9)188,085 (52.9)Sex Male (n, %)159,126 (37.3)26,521 (37.3)132,605 (37.3)> 0.999 Female (n, %)267,156 (62.7)44,526 (62.7)222,630 (62.7)Comorbidity (n, %) Stroke78,923 (18.5)13,804 (19.4)65,119 (18.3)< 0.001 Central degenerative disease33,325 (7.8)6876 (9.7)26,449 (7.4)< 0.001 Diabetes mellitus219,506 (51.5)40,010 (56.3)179,496 (50.5)< 0.001 Hypertension286,724 (67.3)47,067 (66.2)239,657 (67.5)< 0.001 Major depressive disorder108,146 (25.4)20,291 (28.6)87,855 (24.7)< 0.001Dementia subtype All (n, %)53,425 (12.5)10,460 (14.7)42,965 (12.1)< 0.001 Alzheimer (n, %)38,828 (9.1)7770 (10.9)31,058 (8.7) Vascular (n, %)947 (0.2)163 (0.2)784 (0.2) Others (n, %)13,650 ($3.2\%$)2527 ($3.6\%$)11,123 ($3.1\%$) Follow-up duration (years ± SD)5.6 ± 2.95.7 ± 2.85.6 ± 2.9< 0.001 Regarding comorbidities, stroke ($19.4\%$ vs. $18.3\%$, $p \leq 0.001$), central degenerative disease ($9.7\%$ vs. $7.4\%$, $p \leq 0.001$), diabetes mellitus ($56.3\%$ vs. $50.5\%$, $p \leq 0.001$), and major depressive disorder ($28.6\%$ vs. $24.7\%$, $p \leq 0.001$) were significantly more prevalent in the case group. However, hypertension was more common in the control group ($66.2\%$ vs. $67.5\%$, $p \leq 0.001$). Of the 426,282 participants, 53,425 ($12.5\%$) were diagnosed with dementia during the observation period between the index date and December 31, 2020. In the scrub typhus group, 10,460 ($14.7\%$) people were diagnosed with dementia, a significantly higher incidence comparing to the control group ($$n = 42$$,965 [$12.1\%$]; $p \leq 0.001$) (Table 1). Cox proportional hazards analysis revealed that dementia incidence was higher among individuals with a past history of scrub typhus than in the controls. Crude HR was 1.19 ($95\%$ CI 1.16–1.21, $p \leq 0.001$) and the adjusted HR (aHR) was 1.12 ($95\%$ CI 1.10–1.15, $p \leq 0.001$) (Table 2). When classified into subgroups, the risk of AD was higher in the scrub typhus group (aHR, 1.15; $95\%$ CI 1.13–1.18, $p \leq 0.001$). However, there was no statistically significant differences regarding the vascular dementia risk (aHR, 0.98; $95\%$ CI 0.83–1.17, $$p \leq 0.853$$) and other types of dementia (aHR, 1.05; $95\%$ CI 1.00–1.09, $$p \leq 0.033$$). Table 2Multivariate Cox proportional hazards analysis of dementia by subtypepatients (n, %)Crude HRP valueAdjusted HRP valueScrub typhus groupControl groupDementia subtype All10,460 (14.7)42,965 (12.1)1.19 (1.16–1.21)< 0.0011.12 (1.10–1.15)< 0.001 Alzheimer’s disease7770 (10.9)31,058 (8.7)1.22 (1.19–1.25)< 0.0011.15 (1.13–1.18)< 0.001 Vascular163 (0.2)784 (0.2)1.02 (0.86–1.20)0.8370.98 (0.83–1.17)0.853 Others2527 (3.6)11,123 (3.1)1.11 (1.06–1.16)< 0.0011.05 (1.00-1.09)0.033Model adjusted for age, sex, and comorbidities (Stroke, Central degenerative disease, Diabetes mellitus, Hypertension, Major depressive disorder) The Kaplan–Meier curves for the cumulative incidence of dementia are shown in Fig. 2. It shows dementia incidence in both groups increased over time, and the incidence of dementia was higher in the scrub typhus group than in the non-history of scrub typhus group. As time goes, the gap in the incidence rate also increased. Fig. 2Cumulative incidence of dementia during the follow-up period according to the past history of scrub typhus ## Discussion This nationwide retrospective population-based cohort study found that the history of scrub typhus is related to an increased risk of dementia, especially AD. The prevalence of all comorbidities except hypertension was significantly higher in the scrub typhus group than in the control group. However, the association between a high prevalence of dementia and a history of scrub typhus persisted after the adjustment for age, sex, and comorbidities. This trend became more pronounced over time, that can be attributed to the continued effects of the infection in scrub typhus over time. Dementia, a degenerative brain disorder, is a major cause of disability. AD is the most common subtype of dementia in the world. Aging is considered an important risk factor as it increases chronic low-grade inflammation, which may be involved in the pathological progress of AD [19–21]. One of neuropathological features of AD is that extracellular neuronal plate containing amyloid beta (Aβ) and neural fiber tangles consisting of intracellular hyperphosphorylation tau filaments, which can be accompanied by deposition of vascular Aβ and inflammation of vessel [22–24]. In AD, several mechanisms contribute to the pathogenesis of neurodegeneration. Among them, neuroinflammation plays an important role in pathogenesis due to mechanisms that exacerbate Aβ and tau pathologies [25–27]. Increased levels of pro-inflammatory cytokines in postmortem brain and the serum of AD patients suggest that neuroinflammation plays an important role in AD pathogenesis [28]. Many studies have suggested that infectious diseases such as bacteria and viruses may be associated with AD development by this neuroinflammatory pathogenesis [29]. This eventually leads to accumulation of Aβ and hyperphosphorylation of tau by inducing excessive neuroinflammation. For example, HSV-1 is the most common cause of viral encephalitis and previous studies have shown that HSV-1 was detected in 70–$100\%$ of AD patients over the age of 65. When comparing AD patients with controls, more HSV-1 DNA was identified in AD, which was associated with Aβ plaque [30–33]. This suggests that the recent infection or reactivation of HSV-1 is associated with the pathogenesis of AD. As another example, spirochetes are the bacteria that can cause chronic inflammation in CNS, among which Borrelia burgdorperi and Treponema pallidum are widely studied [34, 35]. B. burgdorferi antigens were observed in neuron and Aβ plaques and neurofibrillary tangles in patients with both AD and neuroborreliosis [34, 36]. According to the study, spirochetal-specific antigens and T. pallidum DNA were detected on Aβ plaques, and the risk of AD increased in patients with T. pallidum infection [37, 38]. Rickettsial infections can affect CNS, and the most common neurological manifestations are meningitis, encephalitis, and acute disseminated encephalomyelitis [6]. In some studies, rickettsia lasted for a long period in the CNS of immunodeficiency mice and reappeared, causing severe neuroinflammation [39, 40]. This shows that Rickets disease is a potential risk factor for AD. The most representative involvement of CNS in rickettsia is Rocky Mountain spotted fever and epidemic typhus, followed by scrub typhus [41]. Pathological studies of brain autopsy specimens in patients with fatal scrub typhus showed the presence of mononuclear cell infiltration of the leptomeninges and typhoid nodules, microglial clusters, and hemorrhage [42]. These pathological findings are related to the significant upregulation of the pro-inflammatory cytokine gene in scrub typhus infection in murine models. Therefore, this neurotrophic O. tsutsugamushi seems to be able to induce neuroinflammation by directly inducing AD through CNS infection, or indirectly inducing AD through various systemic inflammatory effects in the brain. Although scrub typhus is a disease that causes systemic vasculitis, the relationship between its infection and an increased incidence of vascular dementia remains unclear. Vascular dementia can only be diagnosed if there is sufficient evidence for cerebrovascular disease (CVD) on brain imaging, such as multiple infarctions, subcortical ischemic changes, or hemorrhage [43]. However, systemic vasculitis caused by scrub typhus is usually microscopic and pathologic findings and might not progress to ischemic change to brain parenchyma. Our findings are supported by a previous study that Scrub typhus showed no difference in CVD incidence rate compared to the control group, even though it showed a greater incidence rate of cardiovascular disease when analyzing the National Health Insurance Service-National Sample Cohort data in South Korea [44]. To our knowledge, this is the first to show the long-term progression of dementia-related to a history of scrub typhus infection. Its main strength is its use of medical codes that can minimize interference, such as recall bias and selection bias, and the results could be derived on a large scale. It was also designed to reduce bias through washout time and correct the risk factors for dementia. Because only computerized anonymous information was used, it may be free of ethical problems and personal information leakage. This study has several limitations. First, since each person’s inclusion or exclusion criteria and diagnosis were made indirectly, such as via operational definitions by code, it is possible that the data may have been over-or underestimated. Second, although some potential risk factors were adjusted for, important risk factors, such as socioeconomic status and family history, could not be considered but may affect the results. Third, after January 20, 2020, there is a possibility of a potential bias due to changes in medical behavior related to the COVID-19 epidemic in Korea. ## Conclusion In conclusion, our findings suggest that a history of scrub typhus infection in old age is significantly associated with increased dementia risk, especially AD. Additionally, prevention and appropriate treatment of scrub typhus should be emphasized as a preventive measure for dementia. ## References 1. Fisher J, Card G, Soong L. Neuroinflammation associated with scrub typhus and spotted fever group rickettsioses. PLoS Negl Trop Dis (2020.0) 14 e0008675. DOI: 10.1371/journal.pntd.0008675 2. Kelly DJ, Fuerst PA, Ching WM, Richards AL. Scrub typhus: the geographic distribution of phenotypic and genotypic variants of Orientia tsutsugamushi. Clin Infect Dis (2009.0) 48 203-30. DOI: 10.1086/596576 3. Bonell A, Lubell Y, Newton PN, Crump JA, Paris DH. Estimating the burden of scrub typhus: a systematic review. PLoS Negl Trop Dis (2017.0) 11 e0005838. DOI: 10.1371/journal.pntd.0005838 4. 4.Infectious diseases statistics sysmtem. https://www.index.go.kr/potal/main/EachDtlPageDetail.do?idx_cd=1442. 5. Chung MH, Kang JS. History of Tsutsugamushi disease in Korea. Infect Chemother (2019.0) 51 196-209. DOI: 10.3947/ic.2019.51.2.196 6. Sekeyova Z, Danchenko M, Filipcik P, Fournier PE. Rickettsial infections of the central nervous system. PLoS Negl Trop Dis (2019.0) 13 e0007469. DOI: 10.1371/journal.pntd.0007469 7. Elahi FM, Miller BL. A clinicopathological approach to the diagnosis of dementia. Nat Rev Neurol (2017.0) 13 457-76. DOI: 10.1038/nrneurol.2017.96 8. 8.International AsD: World Alzheimer Report 2015. In.; 2015. 9. Posner HB, Tang MX, Luchsinger J, Lantigua R, Stern Y, Mayeux R. The relationship of hypertension in the elderly to AD, vascular dementia, and cognitive function. Neurology (2002.0) 58 1175-81. DOI: 10.1212/WNL.58.8.1175 10. Schmidt R, Launer LJ, Nilsson LG, Pajak A, Sans S, Berger K, Breteler MM, de Ridder M, Dufouil C, Fuhrer R. Magnetic resonance imaging of the brain in diabetes: the Cardiovascular Determinants of Dementia (CASCADE) Study. Diabetes (2004.0) 53 687-92. DOI: 10.2337/diabetes.53.3.687 11. Blanc F, Philippi N, Cretin B, Kleitz C, Berly L, Jung B, Kremer S, Namer IJ, Sellal F, Jaulhac B. Lyme neuroborreliosis and dementia. J Alzheimers Dis (2014.0) 41 1087-93. DOI: 10.3233/JAD-130446 12. Jamieson GA, Maitland NJ, Wilcock GK, Yates CM, Itzhaki RF. Herpes simplex virus type 1 DNA is present in specific regions of brain from aged people with and without senile dementia of the Alzheimer type. J Pathol (1992.0) 167 365-8. DOI: 10.1002/path.1711670403 13. Kountouras J, Boziki M, Gavalas E, Zavos C, Deretzi G, Grigoriadis N, Tsolaki M, Chatzopoulos D, Katsinelos P, Tzilves D. Increased cerebrospinal fluid. Int J Neurosci (2009.0) 119 765-77. DOI: 10.1080/00207450902782083 14. Mawanda F, Wallace R. Can infections cause Alzheimer’s disease?. Epidemiol Rev (2013.0) 35 161-80. DOI: 10.1093/epirev/mxs007 15. Miklossy J. Alzheimer’s disease—a spirochetosis?. Neuroreport (1993.0) 4 841-848. DOI: 10.1097/00001756-199307000-00002 16. Paradowski B, Jaremko M, Dobosz T, Leszek J, Noga L. Evaluation of CSF-. J Neurol (2007.0) 254 154-9. DOI: 10.1007/s00415-006-0298-5 17. Kim JA, Yoon S, Kim LY, Kim DS. Towards actualizing the value potential of Korea Health Insurance Review and Assessment (HIRA) data as a resource for Health Research: strengths, Limitations, applications, and strategies for optimal use of HIRA data. J Korean Med Sci (2017.0) 32 718-28. DOI: 10.3346/jkms.2017.32.5.718 18. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP, Initiative S. The strengthening the reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet (2007.0) 370 1453-7. DOI: 10.1016/S0140-6736(07)61602-X 19. Chen WW, Zhang X, Huang WJ. Role of neuroinflammation in neurodegenerative diseases (review). Mol Med Rep (2016.0) 13 3391-6. DOI: 10.3892/mmr.2016.4948 20. Gross AL, Walker KA, Moghekar AR, Pettigrew C, Soldan A, Albert MS, Walston JD. Plasma markers of inflammation linked to clinical progression and decline during preclinical AD. Front Aging Neurosci (2019.0) 11 229. DOI: 10.3389/fnagi.2019.00229 21. Kennedy BK, Berger SL, Brunet A, Campisi J, Cuervo AM, Epel ES, Franceschi C, Lithgow GJ, Morimoto RI, Pessin JE. Geroscience: linking aging to chronic disease. Cell (2014.0) 159 709-13. DOI: 10.1016/j.cell.2014.10.039 22. DeTure MA, Dickson DW. The neuropathological diagnosis of Alzheimer’s disease. Mol Neurodegener (2019.0) 14 32. DOI: 10.1186/s13024-019-0333-5 23. Goedert M, Spillantini MG. A century of Alzheimer’s disease. Science (2006.0) 314 777-81. DOI: 10.1126/science.1132814 24. Masters CL, Bateman R, Blennow K, Rowe CC, Sperling RA, Cummings JL. Alzheimer’s disease. Nat Rev Dis Primers (2015.0) 1 15056. DOI: 10.1038/nrdp.2015.56 25. Heppner FL, Ransohoff RM, Becher B. Immune attack: the role of inflammation in Alzheimer disease. Nat Rev Neurosci (2015.0) 16 358-72. DOI: 10.1038/nrn3880 26. Ising C, Venegas C, Zhang S, Scheiblich H, Schmidt SV, Vieira-Saecker A, Schwartz S, Albasset S, McManus RM, Tejera D. NLRP3 inflammasome activation drives tau pathology. Nature (2019.0) 575 669-73. DOI: 10.1038/s41586-019-1769-z 27. Sala Frigerio C, Wolfs L, Fattorelli N, Thrupp N, Voytyuk I, Schmidt I, Mancuso R, Chen WT, Woodbury ME, Srivastava G. The major risk factors for Alzheimer’s Disease: age, sex, and genes modulate the Microglia response to Abeta plaques. Cell Rep (2019.0) 27 1293-1306. DOI: 10.1016/j.celrep.2019.03.099 28. Wang WY, Tan MS, Yu JT, Tan L. Role of pro-inflammatory cytokines released from microglia in Alzheimer’s disease. Ann Transl Med (2015.0) 3 136. PMID: 26207229 29. Danics K, Forrest SL, Kapas I, Erber I, Schmid S, Toro K, Majtenyi K, Kovacs GG. Neurodegenerative proteinopathies associated with neuroinfections. J Neural Transm (Vienna) (2021.0) 128 1551-66. DOI: 10.1007/s00702-021-02371-7 30. Harris SA, Harris EA. Herpes simplex virus type 1 and other pathogens are key causative factors in sporadic Alzheimer’s Disease. J Alzheimers Dis (2015.0) 48 319-53. DOI: 10.3233/JAD-142853 31. Jamieson GA, Maitland NJ, Wilcock GK, Craske J, Itzhaki RF. Latent herpes simplex virus type 1 in normal and Alzheimer’s disease brains. J Med Virol (1991.0) 33 224-7. DOI: 10.1002/jmv.1890330403 32. Laval K, Enquist LW. The potential role of herpes simplex virus type 1 and neuroinflammation in the pathogenesis of Alzheimer’s Disease. Front Neurol (2021.0) 12 658695. DOI: 10.3389/fneur.2021.658695 33. Looker KJ, Magaret AS, May MT, Turner KM, Vickerman P, Gottlieb SL, Newman LM. Global and regional estimates of prevalent and incident herpes simplex virus type 1 infections in 2012. PLoS ONE (2015.0) 10 e0140765. DOI: 10.1371/journal.pone.0140765 34. Miklossy J. Alzheimer’s disease—a neurospirochetosis. Analysis of the evidence following Koch’s and Hill’s criteria. J Neuroinflamm (2011.0) 8 90. DOI: 10.1186/1742-2094-8-90 35. Ramesh G, Alvarez AL, Roberts ED, Dennis VA, Lasater BL, Alvarez X, Philipp MT. Pathogenesis of Lyme neuroborreliosis:. Eur J Immunol (2003.0) 33 2539-50. DOI: 10.1002/eji.200323872 36. MacDonald AB. Alzheimer’s neuroborreliosis with trans-synaptic spread of infection and neurofibrillary tangles derived from intraneuronal spirochetes. Med Hypotheses (2007.0) 68 822-5. DOI: 10.1016/j.mehy.2006.08.043 37. Maheshwari P, Eslick GD. Bacterial infection and Alzheimer’s disease: a meta-analysis. J Alzheimers Dis (2015.0) 43 957-66. DOI: 10.3233/JAD-140621 38. Miklossy J. Historic evidence to support a causal relationship between spirochetal infections and Alzheimer’s disease. Front Aging Neurosci (2015.0) 7 46. DOI: 10.3389/fnagi.2015.00046 39. Bohacsova M, Filipcik P, Opattova A, Valarikova J, Quevedo Diaz M, Skultety L, Novak M, Sekeyova Z. Survival of rat cerebrocortical neurons after rickettsial infection. Microbes Infect (2015.0) 17 845-9. DOI: 10.1016/j.micinf.2015.09.024 40. Osterloh A, Papp S, Moderzynski K, Kuehl S, Richardt U, Fleischer B. Persisting Rickettsia typhi causes fatal central nervous system inflammation. Infect Immun (2016.0) 84 1615-32. DOI: 10.1128/IAI.00034-16 41. Mahajan SK, Rolain JM, Kanga A, Raoult D. Scrub typhus involving central nervous system, India, 2004–2006. Emerg Infect Dis (2010.0) 16 1641-3. DOI: 10.3201/eid1610.100456 42. Allen AC, Spitz S. A comparative study of the pathology of Scrub Typhus (Tsutsugamushi Disease) and other Rickettsial Diseases. Am J Pathol (1945.0) 21 603-81. PMID: 19970829 43. O’Brien JT, Thomas A. Vascular dementia. Lancet (2015.0) 386 1698-706. DOI: 10.1016/S0140-6736(15)00463-8 44. Jung LY, Lee J, Jeon M, Hwang JH, Hong JY, Lee CS. Risk of cerebro- and cardiovascular disease in patients with scrub typhus. Eur J Clin Microbiol Infect Dis (2020.0) 39 451-4. DOI: 10.1007/s10096-019-03743-4
--- title: The effects of exposure to solar radiation on human health authors: - R. E. Neale - R. M. Lucas - S. N. Byrne - L. Hollestein - L. E. Rhodes - S. Yazar - A. R. Young - M. Berwick - R. A. Ireland - C. M. Olsen journal: Photochemical & Photobiological Sciences year: 2023 pmcid: PMC9976694 doi: 10.1007/s43630-023-00375-8 license: CC BY 4.0 --- # The effects of exposure to solar radiation on human health ## Abstract This assessment by the Environmental Effects Assessment Panel (EEAP) of the Montreal Protocol under the United Nations Environment Programme (UNEP) evaluates the effects of ultraviolet (UV) radiation on human health within the context of the Montreal Protocol and its Amendments. We assess work published since our last comprehensive assessment in 2018. Over the last four years gains have been made in knowledge of the links between sun exposure and health outcomes, mechanisms, and estimates of disease burden, including economic impacts. Of particular note, there is new information about the way in which exposure to UV radiation modulates the immune system, causing both harms and benefits for health. The burden of skin cancer remains high, with many lives lost to melanoma and many more people treated for keratinocyte cancer, but it has been estimated that the Montreal Protocol will prevent 11 million cases of melanoma and 432 million cases of keratinocyte cancer that would otherwise have occurred in the United States in people born between 1890 and 2100. While the incidence of skin cancer continues to rise, rates have stabilised in younger populations in some countries. Mortality has also plateaued, partly due to the use of systemic therapies for advanced disease. However, these therapies are very expensive, contributing to the extremely high economic burden of skin cancer, and emphasising the importance and comparative cost-effectiveness of prevention. Photodermatoses, inflammatory skin conditions induced by exposure to UV radiation, can have a marked detrimental impact on the quality of life of sufferers. More information is emerging about their potential link with commonly used drugs, particularly anti-hypertensives. The eyes are also harmed by over-exposure to UV radiation. The incidence of cataract and pterygium is continuing to rise, and there is now evidence of a link between intraocular melanoma and sun exposure. It has been estimated that the Montreal Protocol will prevent 63 million cases of cataract that would otherwise have occurred in the United States in people born between 1890 and 2100. Despite the clearly established harms, exposure to UV radiation also has benefits for human health. While the best recognised benefit is production of vitamin D, beneficial effects mediated by factors other than vitamin D are emerging. For both sun exposure and vitamin D, there is increasingly convincing evidence of a positive role in diseases related to immune function, including both autoimmune diseases and infection. With its influence on the intensity of UV radiation and global warming, the Montreal Protocol has, and will have, both direct and indirect effects on human health, potentially changing the balance of the risks and benefits of spending time outdoors. ### Supplementary Information The online version contains supplementary material available at 10.1007/s43630-023-00375-8. ## Introduction The Montreal Protocol on Substances that Deplete the Ozone Layer and its Amendments (most recently Kigali in 2016) have prevented substantial depletion of stratospheric ozone and facilitated its recovery, with a marked effect on ultraviolet (UV) radiation and reduction in global warming. In the absence of the Montreal Protocol, the erythemally weighted UV irradiance, indicated by the UV Index, would have increased by up to $20\%$ between 1996 and 2020 in the region where most of the world’s population lives (between 50°N and 50°S of the equator) [1]. With the Montreal Protocol, it is projected that UV radiation will decline at mid-latitudes over the remainder of the twenty-first century, although in urban areas where air quality is improving, UV radiation at the Earth’s surface is likely to increase. The Montreal Protocol has contributed to a reduction in global warming, as the ozone-depleting chemicals controlled under the Protocol are also potent greenhouse gases. The changes brought about by the Montreal Protocol have important effects on human well-being, both directly and indirectly. In this assessment, we focus largely on direct effects due to human exposure to UV radiation, but human health is also influenced by air quality [2] and impacts of UV radiation on terrestrial [3] and aquatic [4] ecosystems, and materials [5]. Direct effects occur due to ozone-driven changes in the intensity of UV radiation, influencing the time outdoors before damage to the skin and eyes occurs. These changes in UV irradiance, along with climate change, influence sun exposure and sun protection behaviour. However, changes in health outcomes linked to UV radiation also need be considered within the context of broader societal influences and changes in health service use. For example, over the past several decades day-to-day occupational and recreational activities have moved predominantly indoors, but in many countries with temperate climates, annual holidays in regions with high ambient UV radiation have become common and use of sunbeds has increased. Alongside this, the sun protection factor of sunscreens has increased and the public has been educated about how to protect the skin from the sun. In developed countries, changing practices in screening and diagnosis, particularly for skin cancer, make a considerable contribution to the observed trends. It is, thus, challenging to attribute trends in human health solely to changes in ambient UV radiation. However, with the increases in UV radiation that would have occurred in the absence of the Montreal Protocol, balancing the risks and benefits of sun exposure would have presented a far greater challenge. We present an assessment of findings regarding the effect of UV radiation on health published since our previous Quadrennial Assessment [6]. This assessment is not a systematic literature review. Rather, we conducted a broad critical assessment of the literature to identify publications containing information that may be of interest to policy makers whose remit is to make decisions about controls of ozone-depleting substances. This Perspective is part of the topical collection: Environmental effects of stratospheric ozone depletion, UV radiation, and interactions with climate change: UNEP Environmental Effects Assessment Panel, 2022 Quadrennial Assessment (10.1007/s43630-023-00374-9). ## Genes and skin cancer Skin cancer arises primarily as a consequence of UV-induced DNA damage that remains unrepaired, combined with immune suppression (Fig. 1). The past decade has seen an in-depth discovery of the genetic basis of skin cancers. Cutaneous melanomas carry distinct UV radiation mutational signatures (C > T substitutions at TpC dinucleotides (mutated base underlined), C > T substitutions at CpC and CpC dinucleotides, and high levels of T > C and T > A mutations (see Online Resource Fig. 1); the latter mutations may be caused by indirect DNA damage following exposure to UV radiation [7]. Melanocytes, from which melanomas arise, contain over 2000 genomic sites that are up to 170-fold more susceptible to UV radiation-induced damage than the average site in the genome [8]. These may serve as genetic dosimeters (i.e. indicators of UV radiation dose), which could be developed as a tool to determine risk of melanoma and, thus, the need for surveillance. Fig. 1Skin cancer arises primarily as a consequence of direct and indirect (via reactive oxygen species) DNA damage and immune suppression. ( Figure created by Rachael Ireland) Until recently, it was believed that cyclobutane pyrimidine dimers (CPDs) could only be formed during exposure to UV radiation. New studies have shown that CPDs can be formed after UV radiation exposure has ended, with maximal expression 2–3 h post-irradiation, including in human skin in vivo [9]. These “dark CPDs” are formed by chemiexcitation, in which energy from UV radiation photons is transferred to chemical intermediates, including melanin intermediates, which then transfer energy to DNA, resulting in CPD formation. The biological significance of dark CPDs is unknown. Many genetic loci associated with the risk of melanoma have been discovered. Additional variants have been identified through the use of multi-trait analysis of genome-wide association studies. Of note, new variants include those related to autoimmune traits; further functional analyses of these may identify new targets for chemoprevention of melanoma [10]. This method has also been used to identify new loci underpinning risk of keratinocyte cancer. Most variants affect both basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) (collectively called keratinocyte cancer (KC)), demonstrating their shared susceptibility [11]. Loci in pigmentation, DNA repair and cell-cycle control, telomere length and immune response pathways have been identified. ## The role of UV radiation-induced immune modulation in the harms and benefits of sun exposure Many of the harmful and beneficial effects of exposure to UV radiation are mediated through UV-induced effects on the immune system, both locally and systemically. Our immune system is responsible for protecting us from pathogens and destroying aberrant (potentially malignant) cells. At the same time, it must self-regulate to avoid over-reactions to pathogens, and to tolerate ‘self’ by not attacking self-antigens that could lead to autoimmune diseases. In most people, exposing the skin to UV radiation suppresses local (skin) immune processes, enabling malignant cells to escape immune control, but it also upregulates anti-microbial processes in the skin. It also suppresses aberrant immune responses systemically; i.e. in other, non-sun exposed, parts of the body. Exposure to UV radiation is, thus, ‘immune modulatory’ rather than solely ‘immune suppressive’. ## Mechanisms and consequences of UV radiation-induced modulation of immunity Modulation of the immune system occurs through the direct or indirect activation of cells that reside within the epidermis and dermis, including epidermal keratinocytes, dendritic cells such as Langerhans’s cells, dermal lymphocytes, nerves, and mast cells [12]. Indirect pathways include UV radiation-induced changes in the action of cytokines and other mediators of the immune response, such as nitric oxide, cis-urocanic acid, ligands of the aryl hydrocarbon receptor, platelet-activating factor (PAF), prostaglandin E2, anti-microbial peptides, and vitamin D [13]. Some of these mediators lead to the recruitment of circulating immune cells from the blood. For example, following a sunburn (see Sect. 3.2), the skin is rapidly infiltrated by neutrophils, the most abundant leucocyte (white blood cell) in the circulation. Neutrophil infiltration peaks at ~ 24 h after exposure to an inflammatory (3 minimal erythema dose (MED)) dose of broadband UV-B radiation, returning to baseline 7–14 days later [14]. Neutrophils perform important anti-bacterial functions which, together with the induction of anti-microbial peptides, partly explains why skin infections are uncommon following exposure of the skin to UV radiation. UV-recruited neutrophils also produce anti-inflammatory cytokines such as IL-4 which leads to local immune suppression. Dendritic cells in the skin capture, process and present antigens to other immune cells, initiating an immune response. They are versatile and ‘plastic’ in their ability to take up, process and present foreign and tumour antigens to T cells. It is this property that makes dendritic cells the ‘conductors’ of the adaptive immune response. In response to UV radiation, dendritic cells and mast cells migrate from the site of exposure to the lymph nodes that drain the skin. There, they regulate T-cell-dependent responses (reviewed in [15]) and activate immune regulatory B cells (BRegs—Fig. 2) [16]. Importantly, in mouse models and using solar-simulated UV radiation, blocking this UV radiation-induced migration of mast cells [17] and/or the activity of UV-activated B cells [18] prevents carcinogenesis induced by UV radiation. Other regulatory immune cells are also activated and may migrate back to UV-irradiated skin [14]. There they suppress the skin and anti-tumour immune responses, modulate inflammation, potentially enhancing wound healing [19], and/or proliferate and migrate into the circulation (reviewed in [12]). Together, these events explain why UV radiation is considered a complete carcinogen; it is able to both mutate DNA and suppress the anti-tumour immune response. Fig. 2Ultraviolet radiation is immunomodulatory. The absorption of UV radiation by chromophores in the skin directly and indirectly activates cells in the epidermis and dermis, including keratinocytes, Langerhans cells (LCs), mast cells and dermal lymphocytes. Exposing the skin to UV radiation stimulates keratinocytes and mast cells to release microvesicle particles, cytokines and immunomodulatory lipids such as platelet-activating factor (PAF), which induce neutrophil and monocyte infiltration into the skin and can affect distant, non-skin cells. Skin mast cells and dendritic cells migrate into the skin-draining lymph nodes where they activate regulatory phenotypes (e.g. Breg). Elevated sphingosine-1-phosphate (S1P) lipid levels in the draining lymph nodes after exposure of the skin to UV radiation also contribute to systemic immune suppression by preventing lymphocyte circulation. UCA urocanic acid, 5-HT 5-hydroxytryptamine, PG prostaglandin (Figure created by Rachael Ireland) Research published since our last assessment [6] has highlighted new mechanisms by which exposing the skin to UV radiation influences immunity, including upregulation of lipids, changes in white blood cells, and alterations in the skin microbiome and transcriptome. Exposing the skin to solar-simulated UV radiation causes an increase in the production of immunomodulatory lipids such as platelet-activating factor (PAF) and PAF-like species [20]. These bioactive lipids, and changes in lipid metabolism, directly affect immune-cell phenotype and function, including increasing the production of cytokines that suppress the immune system (Fig. 2). In addition, activation of the PAF receptor in human skin induces the release of large numbers of microvesicle particles [21]. These may transport PAF and other bioactive chemicals from epidermal keratinocytes to distant immune cells and organs, thus effecting UV-B-mediated systemic immune modulation [21]. This discovery provides crucial insight into the mechanism by which exposure to UV-B radiation alters the immune system at sites that are not directly exposed to the radiation. The effects of exposure to UV radiation on white blood cell (leukocyte) subsets in blood have been recently reviewed [13]. Exposure of mice to a single 8 kJ m−2 dose of solar-simulated UV radiation induces changes in the number, phenotype and function of these cells in both the innate and adaptive immune systems that typically lead to reduced activity and capacity to recirculate [22], consistent with benefits for immune-mediated disorders such as multiple sclerosis (MS) and potentially COVID-19 [23]. Several studies have identified seasonal changes in the number of leukocytes and have found the overall inflammatory milieu to be more pro-inflammatory in winter and anti-inflammatory in summer. While vitamin D is known to have effects on immune function, the effects on leukocytes were independent of vitamin D status (reviewed in [13]). In support of this work, a randomised controlled trial (RCT) of low-dose (400 IU/day) vitamin D3 supplementation (compared to placebo) in vitamin D-deficient (mean 25-hydroxy vitamin D [25(OH)D1] blood concentration = 36.1 nmol L−1) but otherwise healthy participants in Aberdeen, Scotland, found seasonal variation in natural T-regulatory cell populations and functions that was independent of blood 25(OH)D concentration [24]. UV irradiation of the skin causes changes in the skin microbiome [25] and transcriptome (the set of coding and non-coding RNA in cells) [26]. In people with atopic dermatitis (the most common type of eczema), 12–25 treatments over 6–8 weeks with narrowband UV-B radiation caused a shift to greater microbial diversity accompanied by reduced skin inflammation [25]. Irradiation of the skin of seven healthy male volunteers (skin type II) using solar-simulated UV radiation and doses equivalent to 0, 3 and 6 standard erythemal doses (SED) led to altered expression, mainly upregulation, of multiple genes (primarily related to DNA repair and apoptosis, immunity and inflammation, pigmentation, and vitamin D synthesis) [26]. The number of genes affected increased with increasing dose of UV radiation. UV-B (280–320 nm) and UV-A1 (340–400 nm) had similar effects on gene expression. An abnormal cutaneous response to exposure to UV radiation may result in overactive immune responses to substances in the skin, resulting in UV-induced allergic skin conditions [27]. Evidence is also accruing to suggest that dysfunction of the skin’s innate immune system contributes to some photodermatoses, including conditions aggravated by sun exposure such as systemic lupus erythematosus (SLE) [28] and rosacea [29] (Sect. 4.3). Abnormalities of innate immunity can explain the enhanced UV-B-induced keratinocyte damage observed in cutaneous manifestations of SLE [28], and the inflammatory response to UV-B-induced keratinocyte damage in rosacea [29]. Recent studies show that irradiating the skin of mice with UV-B radiation can lead to changes in distant organs. One study demonstrated changes in gene expression in the kidney, upregulating inflammatory responses [30]. This may be one mechanism by which sun exposure in people with SLE causes acute exacerbation of nephritis (inflammation of the kidney). In another study in mice, chronic exposure of the skin to broadband UV-B radiation (100–300 m J cm−2 for 3 days per week for 10 weeks) significantly reduced levels of dopamine and related enzymes (tyrosine hydroxylase and dopamine beta-hydroxylase) in the blood and adrenal glands and induced marked damage in the adrenal medulla [31]. These studies add to our emerging understanding of wide-ranging systemic effects of exposing the skin to UV radiation, noting that studies in mice do not always translate to humans but that similar studies in humans may not be feasible. ## Harms of exposure to UV radiation Human exposure to UV radiation causes harms to the skin and eyes. For the skin in particular, the risks vary according to skin pigmentation. People with deeply pigmented skin are at particularly low risk of UV-induced skin cancer, due to the type of melanin and the degree of pigmentation. In contrast, people with lightly pigmented skin are at markedly increased risk of skin cancer, particularly if they reside in areas with high ambient UV radiation. Low-dose repeated exposures to UV radiation can increase pigmentation and skin thickness, offering some protection against skin damage during subsequent exposures, a concept called habituation. However, the protection afforded is modest, with photoprotection factors (interpreted similarly to the sun protection factor (SPF) used for sunscreens) of 2–3 for people with darker skin at high northern latitudes and 10–12 for people with lighter skin types at lower European latitudes (e.g. 35°North) [32]. ## The association between exposure to UV radiation and skin cancer Exposing the skin to UV radiation is the primary modifiable cause of melanoma and KC. The main mechanisms underlying UV-induced tumourigenesis are DNA mutation, suppression of anti-tumour immune responses, and promotion of cutaneous inflammation. However, the patterns of exposure that give rise to these tumours, and the proportion estimated to be attributable to exposure to UV radiation, differ by geographic location, skin type, and tumour type. The association between sun exposure and melanoma is complex, and appears to differ according to the site of the tumour. A recent study supports the dual pathway hypothesis, where melanoma on sites that are less frequently exposed to the sun occurs in people with many naevi (moles), whereas melanomas on the head and neck are associated with cumulative sun exposure [33, 34]. Despite their complex association with pattern and dose of sun exposure, $75\%$ of melanomas globally are estimated to be attributable to population exposure to excess UV radiation compared with a reference population [35]. This figure is higher in countries with higher ambient UV radiation, particularly Australia and New Zealand ($96\%$) [35], than in those where the intensity of UV radiation is lower, such as Canada ($62\%$) [36] and France ($83\%$) [37]. In people with skin of colour, melanomas tend to occur on the palms of the hands, soles of the feet, and mucosal surfaces, and UV radiation is not a risk factor for these lesions [38]. With respect to KCs, SCCs have a straightforward association with cumulative exposure to UV radiation. The pattern of exposure that gives rise to BCC is less well established, but intermittent exposure in both childhood and adulthood appears to play an important role. This notion is supported by a recent meta-analysis that found stronger associations between sunburns and sunbathing in adulthood and BCC than was apparent for SCC. Sunburn in adulthood was associated with a 1.85-fold increased risk of BCC ($95\%$ CI 1.15–3.00) and a 1.41-fold increased risk of SCC ($95\%$ CI 0.91–2.18). Similar findings were reported for sunbathing in adulthood [39]. Nevertheless, one study did find that cumulative sun exposure was associated with BCC but the association with exposure before the age of 25 years of age was stronger than the association with exposure in adulthood [40]. There is little information about the link between exposure to UV radiation and risk of KC in people with skin of colour. Studies in east Asia suggest associations with measures of sun exposure, such as UV Index, outdoors occupational exposure, and lifetime exposure, but the quality of the studies is low to moderate. There are no studies in people with black skin [41]. The strong association between exposure to UV radiation and KC, combined with high prevalence of exposure, translates into a very high proportion of KCs being attributable to this exposure factor. In Canada, estimates suggested that $81\%$ of BCCs and $83\%$ of SCCs diagnosed in 2015 were attributable to exposure to UV radiation [39]. Easily modifiable risk factors were responsible for BCC in particular; $19\%$ of BCCs were attributable to sunburn in adulthood and $28\%$ to adult sunbathing (the equivalent values for SCC were $10\%$ and $17\%$). Outdoor workers are at particular risk of developing KC [42]. In a systematic review, 18 of the 19 included studies suggested an increased risk of KC among outdoors workers, although estimates were imprecise in many studies [43]. In Canada $6\%$ of KCs in 2011 were attributed to occupational exposure to UV radiation [44]. This is similar to previous studies, where in women $1\%$ of skin cancer (i.e. KCs and rare skin cancers) cases and $4\%$ of skin cancer deaths were attributable to exposure to UV radiation in an occupational setting. The equivalent numbers for men were $7\%$ of cases and $13\%$ of deaths [45, 46]. A possible synergistic effect of simultaneous exposure to UV radiation and excessive alcohol consumption on sunburn and skin damage has previously been raised in epidemiological studies (reviewed in [47]). New work in mouse models and using human skin explants suggests that this is not due to alcohol-induced risky sun exposure behaviour, but rather that synergistic metabolic pathways induce more DNA mutations and immune dysfunction [47]. ## Skin cancers avoided by the Montreal Protocol Estimates from the United States Environmental Protection Agency indicate that the Montreal Protocol will prevent 11 million cases of melanoma and 432 million cases of KC that would have occurred in the United States in people born between 1890 and 2100 [48]. The model estimated that cohorts born in 2040 or later will not experience any excess incidence of skin cancer caused by the effects of ozone depletion, assuming continued compliance with the Montreal Protocol. While this highlights the critical importance of the Montreal Protocol, an important limitation is that these estimates assume no changes in sun exposure behaviour and skin cancer surveillance, and no changes in population structure, such as in the distribution of skin types. Other limitations include uncertainty regarding stratospheric ozone trends, the impacts of climate change, and the action spectrum for skin cancer development. ## Geographic variability in the incidence of melanoma Worldwide in 2020 an estimated 325,000 new cases of invasive melanoma were diagnosed and 57,000 people died from melanoma [49]. The estimated age-standardised (World Standard) incidence per 100,000 people per year of invasive cutaneous melanoma was 3.8 for men and 3.0 for women. Incidence was highest in Oceania (30.1) and lowest in Africa (0.9) and Asia (0.42). Australia and New Zealand continue to report the highest incidence of all countries (Fig. 3), and the highest burden in terms of disability-adjusted life years (DALYs) lost, followed by North America and Europe [50, 51].Fig. 3Estimated age-standardised incidence rate (world-standard population) of invasive cutaneous melanoma in the year 2020, by world region: A men; and B women (Data from the Global Cancer Observatory Database) In 2018, it was estimated that melanoma accounted for $1.6\%$ of all new cancer cases and was responsible for $0.6\%$ of all cancer deaths worldwide [52]. In comparison, the most common cancer at that time (lung), excluding keratinocyte cancer, was responsible for $11.6\%$ of cases and $18.4\%$ of deaths. The cumulative risk of developing melanoma (birth to age 74 years, globally) was estimated to be $0.39\%$ in men and $0.31\%$ in women (noting that this is an average of the markedly different risks in people with light and dark skin); estimates of the cumulative risk of death from melanoma were $0.08\%$ for men and $0.05\%$ for women [52]. Melanoma constituted $11\%$ of all cancer cases in Australia in 2019, and was responsible for $2.7\%$ of deaths from cancer [53]. In Europe in 2018, melanoma accounted for $3.7\%$ of all cancer cases (men: $3.5\%$; women: $3.9\%$), and was responsible for $2.5\%$ of deaths from cancer (men: $3.2\%$; women: $1.9\%$) [54]. By 2040, the number of new melanoma cases globally is predicted to increase to 510,000 per year and deaths to 96,000, assuming changes in population size and age structure but no change in the incidence rates [49]. ## Trends in the incidence of melanoma, based on published reports Trends in melanoma incidence need to be interpreted in light of changing surveillance practices. In the United States [55, 56], Australia [57], and Europe [58, 59] there has been a much greater increase in the incidence of in situ (confined to the epidermis) and thin melanomas compared with thick melanomas. The increase in melanoma incidence has also greatly outstripped increases in the mortality rate. These patterns are thought to reflect the detection of lesions that are unlikely to cause significant morbidity or mortality within a person’s lifetime, a phenomenon known as over-diagnosis, which is occurring due to the combined effect of an increase in skin examinations, lower clinical thresholds for taking a biopsy of pigmented lesions, and lower pathological thresholds for diagnosing melanomas [60, 61]. Over-diagnosis of melanomas could lead to an under-estimate of the impact of the Montreal Protocol. Recent trends in incidence of melanoma vary across populations. Incidence increased in the United Kingdom, Norway, Sweden and Canada (1982–2015) [62], particularly the Eastern Newfoundland and Labrador provinces (2007–2015) [63], and in France (1990–2018) [64]. Of recent reports from Eastern Europe, those from Lithuania (1991–2015) [65], Ukraine (2002–2013) [66], and the Czech Republic (1977–2018) [67] described increases for all age groups and in both men and women, while a study from Hungary found increases between 2011 and 2015 followed by a significant decrease between 2015 and 2019 [68]. For Australia, New Zealand and Denmark (1982–2015) there is a recent trend of stabilising or even declining incidence, likely due to concerted efforts in primary prevention over the past 2–4 decades [62]. While incidence is very low in China and South Korea, small increases in incidence were noted (from $\frac{0.4}{100}$,000 in 1990 to $\frac{0.9}{100}$,000 in 2019) in China [69] and in South Korea (from $\frac{2.6}{100}$,000 in 2004 to $\frac{3.0}{100}$,000 in 2017) [70]. In China in 2017, the highest incidence rates were recorded for the eastern and northeast provinces compared with the western provinces, a trend which may be due to heightened awareness and greater access to medical services in these regions [71]. A study from Singapore reported very low incidence among Chinese, Malay and Indian Singaporeans [72]. A study of trends in melanoma incidence using data from the Surveillance, Epidemiology, and End Results (SEER) programme in the United States showed that across all ethnicities incidence stabilised between 2010 and 2018 (average annual percent change [AAPC], $0.39\%$; $95\%$ CI − 0.40 to $1.18\%$), following five decades of continuous increases [73]. However, the incidence of the thickest melanomas (T4, > 4.0 mm) continued to rise (AAPC $3.32\%$; $95\%$ CI 2.06–$4.60\%$). Populations with lower socioeconomic status or from minority groups were more likely to have thicker melanomas over the time period examined, likely due to poorer access to screening and early detection activities. While the incidence of melanoma in children is very low, between 2000 and 2015 in the United States declines in incidence were reported for children aged 10–19 years, while incidence in younger children remained stable [74]. Several studies have reported different trends according to age. Studies from Canada [75], Italy [76], and England [77] report increases in incidence in older age groups, possibly at least partly due to longer life-span, but a stabilisation or decline in younger age groups. In contrast, a Finnish study of melanoma incidence in children and adolescents reported a fourfold increase between 1990 and 2014, most notable among adolescents [78]. It is unclear whether this represents a true increase or is due to changes in diagnostic criteria and/or cancer registry coverage. ## Trends in incidence of melanoma according to age: analysis of Global Cancer Observatory data It is difficult to compare trends in incidence of melanoma based on reports from the published literature due to the use of different populations for standardising age, as has been noted in the Panel’s annual assessments 2019–2021 [79–81]. We, therefore, extracted population-based registry statistics for six high-risk populations with data available for the period 1982–2016 (namely Australia, United States Whites, Norway, Sweden, Denmark and the United Kingdom) from the Global Cancer Observatory (age standardised to the World Standard Population) [82]. While incidence began to stabilise in Australia after 2005, it continues to increase in the other countries for both men (Fig. 4A) and women (Fig. 4B). However, there is marked variation with age, with modest increases among people aged less than 50 years (Fig. 4C, D) and much more notable increases among older age groups (50 years and over) (Fig. 4E, F). For Australia only, there has been a decline in incidence among younger age groups that began around 2007. In the most recent 10-year period, the estimated average annual percent change in incidence was highest for Norway ($4.0\%$ for men and $4.2\%$ for women) and Sweden ($3.8\%$ for men and $4.0\%$ for women). These trends are attributable to population-specific changes in time outdoors and implementation of sun-protection programmes; these will influence trends into the future as younger cohorts, who have been exposed to these behavioural changes from a younger age, enter middle and older age. Fig. 4Age-standardised incidence rate (ASIR, World) of invasive cutaneous melanoma 1982–2016 in 6 populations [Australia, United States Whites, Norway, Sweden, Denmark and United Kingdom (England and Wales)] from 1982 to 2016. Trends presented separately for men and women, and for all ages and separately for those < 50 years and ≥ 50 years ## Trends in melanoma mortality Trends in mortality are underpinned by changes in incidence and case-fatality rates. The latter has been decreasing markedly in some countries in recent years due to the introduction of new and highly effective systemic therapies for advanced melanoma [83], and this will continue to affect mortality rates with increasing use for earlier stage disease. A study using data from the WHO Mortality Database covering 31 countries over the time period 1985–2015 reported an overall increase in melanoma mortality for men in all countries, in contrast with stable or declining rates in women [84]. For the most recent time period (2013–2015) the median mortality rate was 2.6 deaths per 100,000 for males and 1.6 per 100,000 for females; the highest mortality rates were recorded for Australia and Norway for men, and Norway and Slovenia for women (noting that New Zealand, which has the highest mortality globally, was not included in the report). The increase in most countries reflected increasing mortality rates in people aged 50 years or older; mortality rates were generally stable or declining in younger age groups. The latter trend likely reflects lower incidence among younger birth cohorts exposed to lower cumulative exposure to damaging UV radiation. A separate report for Spain over the period 1982–2016 showed a similar trend, with mortality rates stabilising in men and women younger than 64 years from the mid-90 s, while rates continued to rise in older age groups [85]. Recent declines in melanoma mortality have been reported for New Zealand (2015–2018) [86] and China (1990–2019) [69], but increases were reported for the Netherlands (1950–2018) [87] and Brazil (1996–2016) [88], while mortality was stable in France (1990–2018) [64] and South Korea (2014–2017) [70]. These disparate trends are difficult to interpret given heterogeneity in the introduction (and timing thereof) of new systemic treatments (particularly immunotherapy about 10 years ago) across jurisdictions. ## Trends in the incidence of Merkel cell carcinoma Merkel cell carcinoma (MCC) is a rare skin cancer that may be associated with exposure to UV radiation. An increase in the incidence of MCC between 1997 and 2016 has been reported for the United States, Norway, Scotland, New Zealand, and Queensland, Australia at a rate of 2–$4\%$ per year [89]. Increases have been greater in Brazil, with average annual percent change from 2000 to 2017 of $9.4\%$ for men and $3.1\%$ for women [90]. These findings are consistent with an earlier report covering 20 countries for the period 1990–2007 [91]. The increase in the United States has been attributed to three factors: increased detection, an ageing population, and higher exposure to UV radiation in more recent birth cohorts [92]. The cause of MCC is not well understood; the Merkel cell polyomavirus (MCPyV) is clonally integrated in up to $80\%$ of tumours [93]. While several studies have reported more mutations in MCPyV-negative tumours (dominated by UV signature mutations) [93, 94], a new study based on 9 tumours reported more mutations in MCPyV-positive compared to MCPyV-negative tumours [95]. Because MCC is such a rare tumour, all existing studies are based on limited tumour series, and further studies using larger sample sizes are needed to understand the role of exposure to UV radiation in the aetiology of these cancers. Survival from MCC is much lower than for melanoma ($50\%$ at 5-years for local and < $14\%$ for metastatic disease [96]), although immunotherapy trials are reporting improved outcomes [97–99]; the costs of treatment are likely to increase if these therapies are widely adopted. ## Trends in incidence of keratinocyte cancer Accurately reporting the burden, incidence, and trends in KC remains a challenge. KCs are not routinely reported in most cancer registries. Further, people frequently experience more than one lesion, but this multiplicity is often not considered, with only the first lesion in a person being reported. Accounting for multiple KCs per person results in an approximately $50\%$ increase in incidence rates [100, 101]. An analysis of Global Burden of Disease data found that in 2019 KC was the most common cancer globally, affecting almost 3 times as many people as the next most common cancer (lung—2.2 million people) [102, 103]; there were ~ 6.4 million new patients with KC. Death due to BCC is very rare, but ~ 56,000 people died due to SCC. The burden of disease, as measured by DALYs, increased by almost $25\%$ between 2010 and 2019. Age-standardised incidence rates of KC are highest and increasing in Australia and New Zealand [104–107], with age-standardised rates as high as $\frac{1907}{100}$,000 (standardised to the 2001 Australian population). In Europe, increasing incidence of KC has been reported. For example, in Iceland there was a two–fourfold increase in the incidence of BCC [108] and a 16-fold increase in incidence of SCC between 1981 and 2017 [109], attributed to increased holidays to destinations with high ambient UV radiation and use of sunbeds. In Serbia between 1999 and 2015, there was an annual increase in KCs of $2.3\%$ [110]. In the United Kingdom, SCC incidence increased by $31\%$ and BCC by $21\%$ between 2004 and 2014 [111]. In the United States, the incidence of KC increased from 1990 to 2004, but then remained fairly stable from 2005 to 2019 [112]. Among populations with predominantly light skin, the lifetime risk of KC is much higher in areas with high ambient UV radiation. In the United Kingdom, where ambient UV radiation is comparatively low, lifetime risk is estimated to be $20\%$ [113]. In contrast, lifetime risk in Australia, where the ambient UV radiation is high, is estimated at $69\%$ ($73\%$ for men and $65\%$ for women) [114]. Benign and premalignant keratinocyte lesions caused by sun exposure add an additional burden to the already high cost of skin cancer for healthcare systems and individuals. The prevalence of actinic keratosis (benign lesions) is high and estimated to be between $25\%$ (in a general practice population in Switzerland) and $29\%$ (in patients attending dermatology outpatient clinics in Spain) in European populations [115, 116]. The incidence of in situ skin cancers (premalignant lesions) is also increasing, and in some countries the incidence of these lesions is increasing more rapidly than that of invasive cancers. For example, in the Netherlands the incidence of SCC increased by 6–$8\%$ per year between 2002 and 2017, compared with a 12–$14\%$ annual increase since 2010 for SCC in situ [101, 117]. ## Risks of skin cancer in people who are immunosuppressed Immunosuppression is a risk factor for melanoma, BCC and SCC. Populations with compromised immunity at increased risk include organ transplant recipients [118], those diagnosed with HIV/AIDS (fourfold increased risk of melanoma) [119, 120], and those treated for rheumatoid arthritis (~ 1.3-fold increased risk of KC and melanoma) [121], inflammatory bowel disease (~ 1.5-fold increased risk of KC), and some lymphoproliferative disorders including non-Hodgkin lymphoma and chronic lymphocytic leukaemia (~ twofold increased risk of melanoma) [122]. In solid-organ transplant recipients the magnitude of the increased risk differs between skin cancer types: the increased risk in a high ambient UV radiation environment is two–threefold for melanoma, six–tenfold for BCC, and as high as 100-fold for SCC [123]. ## Costs associated with skin cancer management The average paid and unpaid productivity loss per premature death from melanoma in Europe is estimated to be €450,694 [124], the second highest loss of all cancer types after Hodgkin’s lymphoma, likely due to the relatively earlier age of onset (and, thus, greater paid productivity losses). The introduction of new systemic treatments for advanced melanoma and their increasing use as an adjuvant treatment for non-metastatic disease is causing a rise in the overall cost per capita associated with melanoma treatment globally. In the United States between 1997 and 2015, total expenditure for treatment of melanoma increased at a faster rate than for other cancers [125]. In the Netherlands malignant skin tumours were the 4th most costly cancer in 2017; drug costs increased from €0.7 million to €121 million from 2007 to 2017 [126]. The largest cost drivers in France, Germany and the United Kingdom are medications and hospitalisation and/or emergency department treatment [127]. Adverse events from the use of new treatments are also responsible for a sizable cost burden [128, 129]. A modelling study on the cost of melanoma in Europe estimated national costs ranging between €1.1 million in Iceland and €543.8 million in Germany (€2.7 billion for all European Union states) [130]. A recent study estimated the national costs of treating newly diagnosed melanoma in Australia and New Zealand for the year 2021, and reported total costs of AUD 481.6 million (€310 million2), and NZD 74.5 million (€43 million), respectively [131]. In Australia, the mean cost per patient was AUD 14,268 (€9,198), ranging from AUD 644 (€415) for in situ melanoma to AUD 100,725 (€64,930) for stage III/IV (advanced) disease. These costs will increase as expensive immunotherapy becomes a therapy of choice for earlier stage melanoma, either alone or in combination with targeted therapies [132]. Examining the skin to identify melanoma can lead to the detection of benign lesions, often resulting in additional treatments that may or may not be needed. A study in the United States reported on the costs of diagnosis and treatment of actinic keratoses and other benign lesions associated with screening for melanoma via total body skin examination [133]. In an analysis of 36,647 total body skin examinations in 20,270 adults, the estimated cost of treatment (including consultation, biopsy and pathology charges) for each melanoma detected was USD 32,594 (€33,614), with an additional cost of USD 7840 (€8085) to treat actinic keratoses and other benign lesions. Given the very high and escalating costs of treatment, public health agencies have strengthened their focus on primary prevention, for which there is evidence of cost-effectiveness. A modelling study to evaluate the cost-effectiveness of prevention compared with early detection for melanoma control [134] used data from two randomised controlled trials (RCTs) conducted in Australia [135, 136]. Compared with annual clinical skin examinations (early detection), and no intervention, advice to use sunscreen daily (prevention) resulted in lower numbers of melanoma and KC cases, and significantly lower costs associated with diagnosis and treatment [134]. However, these findings may not be applicable to locations with lower ambient UV radiation, and potential costs of over-diagnosis have not yet been considered. In Canada, the costs of KC and other rare skin cancers due to occupational exposure to UV radiation were estimated to be CAD [2011] 29 million (€22 million) in direct and indirect costs, and CAD [2011] 6 million (€4.5 million) in intangible (due to effects on quality of life) costs in 2011 [137]. These costs can be mitigated; estimates suggest that for every dollar invested in personal protective equipment and shade structures, CAD 0.49 and CAD 0.35 will be returned, respectively [138]. Another modelling study of cost-effectiveness showed that primary prevention by systematic use of sunscreen at a population level will prevent substantial numbers of new skin tumours ($26\%$ less excised KC), and save healthcare costs [134]. Among people at high risk of KC, costs for treatment of KC and actinic keratoses were reduced 1 year after treatment with topical 5-fluorouracil, showing that chemoprevention may be an option to reduce the incidence of skin cancer in this subgroup [139]. The very high and increasing costs of managing skin cancer underscore the need to protect the stratospheric ozone layer; in the absence of control of ozone-depleting substances, the intensity of UV radiation in some regions would increase to the point where many more people would be exposed to sufficient UV radiation to initiate skin cancers. ## Sunburn Sunburn is an acute inflammatory skin reaction caused by over-exposing the skin to UV radiation, primarily the UV-B wavelengths; it is clinically manifested as erythema (redness) in people with Fitzpatrick skin types I–IV3 (modified from [140]), and may cause pain and blistering. Despite the definition of sunburn varying between studies, it is a well-established risk factor for the development of cutaneous melanoma and KC [141, 142], and number of severe sunburns may be associated with increased risk of herpes zoster (i.e. shingles) [143]. Moreover, inflammation from sunburn is a health burden, independently of its association with other conditions. In the United States National Emergency Department Sample, including information about presentations to 950 hospital emergency departments from 2013 to 2015, there were 82,048 visits for sunburn, with $21\%$ classified as severe sunburn (second or third degree burns and/or requiring inpatient admission) [144]. The average cost of an emergency department visit for sunburn was USD 1132. Presentation for all sunburns and for severe sunburns showed highest frequency in lower-income young men, and the incidence was higher in the sunnier states. ## Trends in rates of sunburn Data from the United States National Health Interview Surveys reveal that $34\%$ of community-dwelling adults reported one or more sunburns in the prior 12 months in both 2005 and 2015 (sample sizes 29,250 and 31,399, respectively) [145]. The percentage of adolescents reporting sunburn was considerably higher. Between 2015 and 2017, $57\%$ of 21,894 people aged 14–18 years reported being sunburnt at least once in the previous 12 months [146]. Sunburn was also more common in adolescents than in adults in Spain; $75\%$ of 776 adolescents reported being sunburnt in the previous year compared with ~ $54\%$ of 632 adults and $44\%$ of 324 children [147]. In Germany, $22\%$ of children aged 1–10 years surveyed in 2020 had been sunburnt in the previous year, and there was a positive association with age [148]. In some countries, there has been a reduction in the prevalence of sunburn, coinciding with increased use of sun protection behaviours. In Australia, a comprehensive skin cancer prevention campaign, SunSmart, began in 1988. Surveys conducted in the state of Victoria over the subsequent three decades, in which participants were asked about their sun protection behaviour on the weekend prior to the interview, showed a marked increase in the percentage of people using at least one sun protection behaviour (seeking shade, or using hat or sunscreen) in the first decade after SunSmart began (from 29 to $65\%$) and more modest increases thereafter [149]. Sunscreen use increased from $11\%$ pre SunSmart to $68\%$ in the 2010s. In the state of New South Wales, the percentage of people reporting often or always using sunscreen increased from ~ $30\%$ in 2003 to ~ $40\%$ in 2016, but there was no increase in use of hats [150]. The increase in sun protection is evident in sunburn trends. In Australian adults ($$n = 3614$$), the percentage reporting sunburn during the previous weekend in summer decreased from 14 to $11\%$ between $\frac{2003}{2004}$ and $\frac{2016}{2017}$ [151], accompanied by an increase in the percentage of people using two more sun protection behaviours (from 41 to $45\%$). Sunburn occurred more frequently in Australian adolescents than in adults, but there was a decline from 20 to $15\%$ across this period. In adults in Denmark ($$n = 33$$,315) a $1\%$ annual decrease in sunburn in the previous 12 months was seen across 2007–2015, coinciding with a national sun safety campaign [152]. A birth cohort analysis of melanoma-prone families demonstrates changes in sun protection behaviour and sunburns over time. People from 17 centres in Europe, North and South America, Australia and the Middle East ($$n = 2407$$) were questioned about sun exposure and sunburns at various anchor points across their lives [153]. These behaviours were analysed according to birth cohort (in decades from those born in the 1910s and 1920s through to those born in the 1980s). There was a clear secular trend in the reported frequency of sunscreen use; people born more recently were more likely to use sunscreen at a younger age than those born earlier. Time outdoors on weekends at less than 20 years of age was lower in more recent birth cohorts. Within each birth cohort sunburn occurred more frequently in early vs later life, but more recent cohorts were less likely to experience early life sunburns. Changes in sun exposure and prevention behaviour in some countries have been marked, which may be underpinned, at least in part, by sun protection campaigns. In the absence of the Montreal Protocol, however, it is likely that the benefits of these changes would have been less evident, as the time to sunburn would have been markedly shorter. ## Sunburn prevalence in people of darker skin type Skin melanisation provides some protection against sunburn and UV radiation-induced skin cancer. Traditionally, people with dark skin (skin types V–VI) have been thought to be at very low risk of sunburn [140]. However, the difficulties of detecting sunburn erythema in people with dark skin can contribute to an over-estimation of the amount of UV radiation required to cause sunburn, and an under-estimation of sunburn prevalence [154]. Known differences in sun protection behaviours between ethnically diverse populations could also influence sunburn prevalence [155]. In a survey of people of Black African or Black Caribbean heritage living in the United Kingdom ($$n = 222$$ respondents), over $50\%$ reported a lifetime history of sunburn [156], with frequencies of $47\%$, $54\%$ and $71\%$ in those self-classifying as dark, medium and light skin tone. In the United States, nearly $10\%$ of 4157 non-Hispanic Black participants in the National Health Interview Survey 2015 reported being sunburnt in the previous year, compared with nearly $25\%$ of Hispanic people ($$n = 5208$$) and $42\%$ of non-Hispanic Whites ($$n = 19$$,784) [145]. These surveys suggest that sunburn occurs more frequently in people with darker skin types than traditionally appreciated, but in light of the lower severity of sunburn compared with that in people with light skin, and the extremely low risk of UV-induced skin cancer in these populations, the significance of this is unclear. ## Photodermatoses Photodermatoses are inflammatory skin disorders that are induced or exacerbated by exposure to UV radiation and, in certain conditions, visible light [27]. Both UV-B and UV-A radiation can contribute to the development of photodermatoses. Photodermatoses fall into aetiological groups: dysregulated immune responses to UV radiation; disorders of DNA repair; intrinsic biochemical defects; photosensitivity reactions to drugs and/or exogenous chemicals; and photoaggravated disorders. ## The burden of photodermatoses and their impact on health and psychological well-being The lack of registry data and of consistent case definition for the most common dermatoses make it very challenging to estimate the population prevalence of photodermatoses. However, some photodermatoses, such as the immune-mediated condition, polymorphic light eruption (PLE), have been reported commonly from dermatology clinics in light-skinned populations in temperate regions, particularly during spring [157]. Comprehensive reviews of data from photodiagnostic units in dermatology departments indicate that the photodermatoses most commonly seen are PLE, photoaggravated atopic dermatitis, actinic prurigo, chronic actinic dermatitis, solar urticaria and drug-induced photosensitivity. Photodermatoses occur in dark-skin populations, although with differing frequencies and characteristics from light-skin populations [158]. In a systematic review of population-based and dermatology outpatient studies of rosacea, a photoaggravated chronic inflammatory facial condition, a global prevalence of up to $5\%$ was estimated; however, studies in which rosacea was self-reported yielded higher prevalence than in those where the condition was determined by examination [159]. Studies of the prevalence of most photodermatoses are scarce; for example, there are no reported population-based studies in solar urticaria. Photodermatoses involve a wide range of clinical features, which vary according to the individual condition; these include pain in the skin within a few minutes of sun exposure, severe itching, erythema, blistering, and scarring. The adverse impact on sufferers occurs both directly due to symptoms, and indirectly through restrictions imposed by sun avoidance. In a systematic review of 20 studies (2487 adult and 119 child participants), in which an assessment of quality of life or psychological well-being was performed, one-third of adults and children with photodermatoses were found to experience a very large negative impact on quality of life (Dermatology Life Quality index > 10), and anxiety and depression occurred twice as frequently as in the unaffected population [160]. ## The association between commonly used photosensitising drugs and photodermatoses and skin cancer The pathologic mechanisms underlying drug photosensitivity are broadly classified as phototoxic or photoallergic. Oral medication-induced photosensitivity commonly involves phototoxicity, which can theoretically occur in anyone upon exposure to sufficient dose of a drug and UV radiation. Clinically, drug phototoxicity most often manifests as skin redness, swelling and burning, and can be misdiagnosed as severe sunburn. An analysis of more than 745 million drugs dispensed in Germany and Austria between 2010 and 2017 indicated that nearly $50\%$ had photosensitising potential, with diuretics and anti-inflammatory drugs being primarily responsible [161]. However, the global incidence of drug photosensitivity is uncertain. Analysis of the Japanese Adverse Drug Event Report database (2004–2016) found less than $0.1\%$ of 430,587 reports concerned photosensitivity reactions [162]. A systematic review identified 1134 reported cases of suspected drug phototoxicity associated with 129 oral drugs [163]. However, the quality of the evidence for an association with drugs is low; fewer than $25\%$ of studies performed phototesting, and only $10\%$ confirmed the diagnosis with drug challenge–rechallenge testing. In a report of 2243 patients with photodermatosis evaluated at a photodiagnostic unit, $5\%$ were diagnosed with photodermatosis induced by oral medication. All underwent broadband UV radiation testing and monochromatic testing to wavelengths from 300 to 600 nm (i.e. in the UV-B, UV-A, and visible spectra). UV-A was the main provoking waveband with UV-B contributing in $15\%$ of cases [164]. It is possible that commonly prescribed photosensitising drugs may induce skin cancer. Some mechanisms by which drugs induce acute photosensitivity are also relevant for skin cancer induction, such as promotion of UV-induced DNA damage. In a nested case–control study, using data from the Danish Cancer Registry, of people with their first diagnosis of BCC ($$n = 71$$,533) or SCC ($$n = 8629$$) and population controls ($$n = 1$$,430,883), there was an increased risk of KC with long-term use of hydrochlorothiazide (a diuretic medication commonly used for treatment of high blood pressure); adjusted odds ratios (ORs) for high use vs. never use were 1.29 ($95\%$ CI 1.23–1.35) for BCC and 3.98 ($95\%$ CI 3.68–4.31) for SCC [165]. This led to the European Medicines Agency recommending that advice on increased risk of KC should be included in hydrochlorothiazide product information [166]. Further studies, based in different geographic locations and demographic groups, reveal heterogeneous and conflicting results for an increased risk of KC and melanoma with hydrochlorothiazide use [167–170] [171]. Given its potential public health significance, this issue needs to be resolved. ## Eye diseases associated with exposure to UV radiation Exposure to UV radiation, either directly or through intermediate factors, is associated with increased risk of cataract of the lens, pterygium, squamous cell carcinoma of the cornea and/or conjunctiva, photokeratitis (affecting the cornea) and photoconjunctivitis, pinguecula, and possibly intraocular melanomas, macular degeneration and glaucoma. This section assesses evidence available since our last assessment [6] on conditions that are directly related to exposure to UV radiation. The superficial layers of the eye are exposed to UV radiation and incur damage through the same pathways of DNA damage and production of reactive oxygen species as is seen in the skin. When the individual is in an upright position and the sun is overhead, there is some inherent protection from exposure to UV radiation provided by the protrusion of the brow, the eyebrows, and the eyelids. These provide less protection at other body positions (e.g. lying down), or when the sun is at a lower angle [172, 173]. Wearing a hat and using shade can also reduce exposure, while high surface albedo can increase exposure; large and wraparound sunglasses that block both UV-A and UV-B radiation provide good sun protection [174–176]. UV wavelengths also penetrate to the deeper structures of the eye (reviewed in a previous assessment [177]). The cornea absorbs wavelengths below 295 nm, but allows longer wavelengths to reach the iris and lens. In adults, the lens of the eye absorbs all wavelengths below 370 nm, and greater than $98\%$ of wavelengths between 370 and 400 nm, with higher absorbance in the posterior part of the lens [178]. Over time, the chemical changes induced by that absorption—direct UV-B induced damage and (indirect) UV-A induced photo-oxidation of soluble lens proteins—cause clouding of the lens; i.e. cataract [178]. In young children, the lens may transmit a greater proportion of shorter UV wavelengths, allowing these to reach, and potentially damage, the retina. ## Trends in the prevalence and incidence of cataract Cataract is the major eye condition associated with long-term exposure to UV radiation. The main types of cataracts, as defined by their location in the lens, are nuclear, cortical, or posterior subcapsular. In many cases, there is a mixed phenotype and, within any individual, the two eyes may contain cataracts with a different predominant phenotype. The two subtypes most clearly associated with exposure to UV radiation are nuclear and cortical cataracts. According to the latest reports of the Vision Loss Expert Group of the Global Burden of Disease Study, cataract was the leading cause of blindness between 1990 and 2015 around the world, accounting for $35\%$ ($95\%$ CI 26–44) of the total blindness in 2015 [179–183]. Projections to 2020 from several countries/regions indicate that cataract would remain the main cause of blindness in 2020 [179–184]. Compared with global figures, the proportion of moderate to severe vision impairment caused by cataract was estimated to be higher in East Asia [179], South-east Asia [181], Oceania [181] and Sub-Saharan Africa [183] where exposure to sunlight may be higher and access to suitable medical care may be limited. The disability from cataract (measured in DALYs) increased from 3.5 million in 1990 to 6.7 million in 2019—an increase of $191\%$ [185]. New studies further demonstrate the high prevalence of cataract. In the cross-sectional Ural Eye and Medical Study set in a rural area of Russia, the prevalence of cataract was $45\%$ in people aged ≥ 40 years (of 5899 participants, $81\%$ of eligible residents). Nuclear and cortical cataracts affected $38\%$ and $15\%$ of participants, respectively [186]. A population-based study conducted in Finland found that the prevalence of cataracts increased from $8.1\%$ ($95\%$ CI 7.8–8.5) to $11.4\%$ ($95\%$ CI 10.9–11.9) among individuals ≥ 30 years between 2000 and 2011 [187]. The annual average incidence over the 11-year period was estimated to be 109 cases per year per 10,000 people ($95\%$ CI 104–114) [187]. The cumulative incidence over a similar time period (baseline 2004–2006; follow-up 2011–2013) was greater in Singapore; in the Malay Eye Study the age-standardised cumulative incidence of nuclear and cortical cataract over this time period was estimated to be $13.6\%$ and $14.1\%$ (equating to an annual average crude incidence of 227 and 189 cases per year per 10,000 individuals), respectively [188]. Greater exposure to UV radiation has been clearly linked to an increased risk of cataract (reviewed in [178]). A recent study provides additional supporting evidence. In a population-based cross-sectional study in three different rural areas of India ($$n = 12$$,021), $33\%$ of participants aged 40 years and older had a cataract in at least one eye [189]. Compared with the lowest quintile of a lifetime effective sun exposure score (calculated taking into account the years of exposure, hours of sun exposure accounting for type of headgear used (none, caps, hats, umbrellas, veils, sunglasses)), the prevalence of cataract was significantly higher in the 3rd, 4th and 5th quintiles of exposure. Those in the fifth quintile were 9 times more likely to have cataracts than those in the first quintile (adjusted OR 9.4; $95\%$ CI 7.9–11.2), rising to nearly 26 times more likely in analyses confined to the highest altitude region (Guwahati/Hills region). Differences in exposure to UV radiation, solar angle and sun protection behaviours each had an additional influence on prevalence of cataracts. Nevertheless, in data from the 2008–12 Korea National Health and Nutritional Examination Survey of economically active people, there was no significant association between higher sunlight exposure (≥ 5 h vs. < 5 h/day in the sun without sunglasses or hat) and medically diagnosed cataract (adjusted OR 0.88, $95\%$ CI 0.77–1.00) [190]. Globally, and across diverse individual regions for which there are recent data, the incidence of cataract continues to increase, at least partly due to ageing populations. Where there is good access to high-quality medical care, including cataract surgery, this may not contribute greatly to the burden of disability. However, in many regions, cataract remains a leading cause of blindness, resulting in considerable morbidity due to vision loss and its sequelae (e.g. falls) [191]. A recent study has estimated the effectiveness of the Montreal Protocol in preventing eye diseases, with a focus on cataract. It was estimated that the implementation of the Montreal Protocol with all of its Amendments and adjustments compared to a scenario of no control of ozone-depleting substances, will prevent 63 million cataract cases in people born in the United States between 1890 and 2100. When the comparison scenario is the original Montreal Protocol, this figure is 33 million fewer cases of cataract, demonstrating the importance of the ongoing strengthening of the Protocol [48]. ## Prevalence of pterygium Pterygium is a non-cancerous, self-limiting pink, fleshy tissue growth on the conjunctiva, that is initially induced by exposure to both UV-B and UV-A radiation. The mechanisms of how and why pterygium is self-limiting have been clarified by recent studies [192]. As this condition commonly occurs in surfers who are exposed to significant amounts of sunlight, it is often referred to as ‘surfer’s eye’. The impact of pterygium on vision is minimal unless it reaches the cornea, but it is painful to remove and often recurs after surgical removal. Evidence suggests that the prevalence of pterygium has slightly increased in recent years. In a recent meta-analysis of 55 studies (including data from > 400,000 people in 24 countries), the overall prevalence of pterygium was estimated at $12\%$ ($95\%$ CI 11–14) [193]. However, studies included a diverse range of age groups and not all were population based. The reported prevalence was higher than that from a 2013 meta-analysis based on 20 articles from 12 countries ($10.2\%$; $95\%$ CI 6.3–$16.1\%$) [194]. Studies from Brazil demonstrate the high variability in prevalence estimates according to location and study methods. In a population-based study in the Brazilian Amazon, including 2041 people ($86\%$ of those eligible to participate) aged 45 years and over, the prevalence was $58\%$ [195]. The recent meta-analysis estimated prevalence in Brazil to be $52.0\%$ (in an ophthalmic clinic-based study in Manaus, age range 21–61 years); $21.2\%$ in the Amazon rainforest (population-based study of people 11 years and older); $18.4\%$ in the Brazilian rainforest (population-based, no age data); and $8.1\%$ in São Paulo (population-based, median age 49.6 years) [193]. Such variability challenges the simple combining of estimates across studies. However, the prevalence of pterygium seems to be modest and consistent across various regions of China, estimated to be approximately $6\%$ [196, 197]. A population-based cohort study, the Gutenberg Health Study, including the German city of Mainz and the surrounding regions (latitude 50°N), found a very low prevalence of pterygium with an estimate for the weighted prevalence of $0.9\%$ ($95\%$ CI 0.8–1.2) in people aged 40–80 years [198]. The presence of pterygium was associated with male sex, higher age, and migration from Arabic-Islam countries, the former Soviet Union, and former Yugoslavia. In a slightly higher latitude region, including both an urban and rural multi-ethnic population in Ufa city and surrounds in Russia, the prevalence of pterygium was $2.3\%$ ($95\%$ CI 2.0–2.7) among people over 40 years [199]. Risk factors for pterygium were rural residence, higher age, and lower level of education. The incidence of pterygium has been reported in two longitudinal studies. In Southern India, which lies within the ‘pterygium belt’ (37° north and south of the equator where pterygia are most common [200]), the age- and sex-adjusted incidence was 25.4 per 100 person years ($95\%$ CI 24.8, 25.7) over a 15-year period in residents ($$n = 2290$$) of rural areas aged 30 years and older at baseline [201]. The overall incidence rate in 6122 adults aged 40 years and over was considerably lower (age-adjusted 6-year incidence = $1.2\%$; $95\%$ CI 1.0–$1.6\%$) over 6 years of follow-up in the Singapore Epidemiology of Eye Diseases Study [202]. In a meta-analysis of risk factors for pterygium several factors associated with solar exposure of the eyes increased the risk of pterygium, including spending more vs. less than 5 h outdoors per day (OR 1.24; $95\%$ CI 1.11–1.36), or having outdoor vs. indoor occupations (OR 1.46; $95\%$ CI 1.36, 1.55). Furthermore, in a recent study reporting on findings from the Korean National Health and Nutritional Examination Survey, an average of ≥ 5 h/day in the sun without sunglasses or hat, compared to < 5 h, was associated with an increased risk of pterygium in women (OR 1.47, $95\%$ CI 1.16–1.73) but not men (OR 0.88, $95\%$ CI 0.70, 1.10) [190]. There was a dose response apparent, with greater time outdoors associated with higher risk. Importantly, in the meta-analysis, wearing sunglasses reduced the odds of pterygium by approximately $50\%$ (OR 0.47; $95\%$ CI 0.19–0.74) [193]. In support of this finding, in a longitudinal study of young adults in Australia, wearing sunglasses for at least half of the time outdoors resulted in a significantly greater decline in the area of conjunctival UV fluorescence, a biomarker of sun exposure, over 8 years compared to never or seldom use [203]. ## The link between exposure to UV radiation and intraocular melanoma Intraocular melanoma is the most common type of cancer that develops within the eyeball, but it is rare compared to cutaneous melanoma. Intraocular melanomas predominantly occur on the uvea and conjunctiva, but uveal are considerably more common than conjunctival melanomas. Exposure to sunlight, light pigmentation of the eye and skin, and living at high latitudes are often reported as risk factors for both types of intraocular melanoma, akin to melanoma of the skin. We have previously assessed the epidemiological and genetic evidence regarding the role of UV radiation in the aetiology of intraocular melanoma, with more convincing evidence for conjunctival vs. uveal melanoma [177]. Recent genetic studies provide further evidence of similarity of intraocular to cutaneous melanoma and, thus, a possible causal role of exposure to UV radiation. A study comparing the genetic changes in uveal melanomas with those in cutaneous melanomas has shown many shared mutations, including UV signature mutations, suggesting that some uveal melanomas may be UV dependent [204]. In a similar study, tissue samples from conjunctival melanomas displayed evidence of genetic changes consistent with UV-related damage similar to those found in melanoma of the skin [205]. Published recent incidence data for intraocular melanoma are available from four developed countries that have well-established cancer registries: United States, Canada, Australia and Ireland (Table 1). The age-standardised incidence rate ranged from 3.3 per million in Canada [206] to 9.5 per million in Ireland [207], but rates are not directly comparable due to the use of different populations for age standardisation and different periods of observation. On average the age-adjusted incidence increased by $0.5\%$ per year in the United States between 1973 and 2013 ($p \leq 0.05$) [208]. In Canada, there was minimal change from 1992 to 2010 [206]. In Australia, there was an increase of $2.5\%$ per year from 1982 to 1993, followed by a decrease of $1.2\%$ per year from 1993 to 2014 [209]. Thus, in these countries, the incidence of intraocular melanomas has remained relatively constant over time, in contrast to that of cutaneous melanomas. Table 1Age-standardised incidence rates of intraocular melanomas across developed countries and regionsStudyCountry/RegionPeriodAge-standardised incidence rate per million person years ($95\%$ CI)Uveal melanoma Aronow et al. [ 208]United States1973–20135.2 (5.0, 5.4)a Baily et al. [ 207]Ireland2010–20159.5 (8.4, 10.7)b Ghazawi et al. [ 206]Canada1992–20103.3 (3.2, 3.5)c Beasley et al. [ 209]Australia1982–20147.6 (7.3, 7.9)dConjunctival melanoma Ghawazi et al. [ 210]Canada1992–20100.32 (0.28, 0.37)c Virgili et al. [ 211]Europe1995–2007overall 0.42e Virgili et al. [ 211]Northern Europe1995–20070.81 (0.59, 1.09)e Virgili et al. [ 211]UK and Ireland1995–20070.40 (0.36, 0.45)e Virgili et al. [ 211]Central Europe1995–20070.59 (0.51, 0.68)e Virgili et al. [ 211]Southern Europe1995–20070.35 (0.26, 0.47)e Virgili et al. [ 211]Eastern Europe1995–20070.27 (0.22, 0.33)eaAge-adjusted to the US population 2000bAge-standardised using the 1976 European standard populationcAge-standardised using the World Standard PopulationdAge-standardised using the 2001 Australian standard populationeAge-standardised using the European standard population The incidence of conjunctival melanoma was substantially lower compared to uveal melanoma in incidence studies from Canada and Europe, replicating previous findings. The age-standardised incidence rate of conjunctival melanoma was 0.32 cases per million people per year (age standardised to the World Standard Population) between 1992 and 2010 in Canada [210], while it was 0.42 cases per million people per year (age standardised to the European Standard Population) in Europe [211]. ## Damage to the eye from drug-induced phototoxicity A number of drugs absorb in the UV range and have phototoxic side effects affecting various structures in the eye [212]. For example, fluoroquinolone antibiotics such as ciprofloxacin and norfloxacin (used to treat ocular infections), in the presence of UV-A radiation, caused damage to epithelial cells (in cell culture) and proteins of the lens. Exposure of the eye to UV-A radiation while using these compounds could accelerate the development of cataract [213]. Use of ophthalmic formulations containing ketoconazole, diclofenac, or sulphacetamide were found to be toxic or irritating in the presence of UV-A radiation [214]. While there is growing awareness of cutaneous photosensitivity in relation to systemic drugs, the focus for eyes appears to have been on exposure to UV-A radiation in conjunction with topical medications. It will be important to better understand the potential photosensitisation resulting from both topical and systemic drugs for the eye, given its vulnerability to damage from exposure to UV radiation and the clear protection that sunglasses provide. ## Increased risk of systemic infections and reduced vaccine effectiveness Hart and Norval [12] hypothesised that vaccination through acutely or chronically sun-exposed skin (e.g. the upper arm, a common site for intramuscular vaccination) may result in a less effective immune response compared to unexposed skin (e.g. buttock). However, there remains little confirmatory evidence for this at present. In a cluster randomised trial in children in rural South Africa, an intervention to protect vaccinees from solar UV radiation did not result in higher antibody levels following a measles booster [215]. However, in a small clinical trial testing the immune response to a novel antigen (keyhole limpet haemocyanin)—although higher natural exposure to UV radiation was not associated with a change in antigen-specific antibodies—there was a reduced T-cell response [216]. Any effect of exposure to UV radiation may be more important for vaccines that rely on a cell-mediated, rather than humoral (antibody), response to vaccination; e.g. Bacille Calmette Guerin (BCG) for tuberculosis, particularly in low latitude (higher UV radiation) locations. ## UV radiation and reactivation of viruses The association of intense exposure to UV radiation with subsequent reactivation of Herpes simplex virus 1 (HSV), causing cold sores of the lip, is well described (reviewed in [177]). The presence of IgM class antibodies to HSV reflects recent viral activity, either primary or recurrent infection [217]. In a recent study from Sweden, the odds for anti-HSV IgM positivity were nearly twofold higher (odds ratio = 1.99 per mean MED difference) in summer than in winter (mean MED difference was 9.967 equivalent to 2093.1 J m−2), consistent with UV-induced reactivation of HSV, with or without the manifestation of cold sores [217]. There is considerable current interest in another herpes virus, Epstein–Barr virus (EBV), in relation to risk of multiple sclerosis (MS), nasopharyngeal carcinoma and other diseases. Results from a recent study from Hong Kong [218] suggest that higher personal sun exposure is associated with reactivation of EBV. The measures of personal exposure included ambient UV radiation at the date of blood collection, serum 25(OH)D concentration, and self-reported duration of sunlight exposure (hours/day) over four life periods (6–12 years, 13–18 years, 19–30 years, and 10 years prior to recruitment). EBV reactivation was measured as seropositivity to EBV viral capsid antigen (VCA) IgA. Only duration of sunlight exposure at 19–30 years and 10 years prior to recruitment (for ≥ 8 h compared to < 2 h, OR 2.44, $95\%$ CI 1.04–5.73, OR 3.59, $95\%$ CI 1.46–8.77, respectively) were associated with increased odds of VCA-IgA seropositivity (inferred as evidence of reactivation). Reactivation of EBV may be a trigger of relapses in MS [219]. Thus, higher levels of sun exposure, leading to EBV reactivation, might be expected to also be associated with relapse. However, previous research suggests that higher sun exposure (over the life-course prior to MS onset) is associated with fewer relapses in people with MS [220]. Nevertheless, the time course of sun exposure may be of importance; higher sun exposure earlier in life may be protective for the development of MS through immune mechanisms, but after EBV infection higher sun exposure may be associated with increased risk of relapse through reactivation of EBV. Datasets are available that could test this hypothesis. Varicella zoster virus is a herpesvirus that causes chicken pox during primary infection and shingles on reactivation. Using data from Thailand, a recent study examined seasonal variation in case reports of chickenpox and shingles [221]. Both chickenpox and shingles showed strong seasonality. Chickenpox was characterised by outbreaks beginning during November and December, with seasonal peaks in February and March (with deep troughs from June to October). The amplitude of the seasonal effect decreased closer to the equator. Shingles showed a peak in May–June, with a shallow trough in February–March and a deep trough in October–December. Again, higher latitudes had more pronounced seasonal cycles. Changes in ambient UV radiation were the main driver of the seasonal cycle for shingles reactivation, but not chickenpox, consistent with an effect of UV radiation on reactivation but not primary infection with Varicella zoster virus. ## Benefits of exposure to UV radiation Sun exposure has numerous benefits, many of which are mediated by exposure to UV radiation, and some by exposure to other wavelengths. People need to be able to safely spend time outdoors to gain these benefits. The Montreal Protocol has likely enabled the benefits to be gained, by preventing the intensity of ambient UV radiation from increasing to an extent where it would have been very difficult for light-skinned people in particular to spend time outdoors without markedly increasing their risk of UV-induced skin and eye diseases. ## Health benefits of greater time outdoors and sun exposure We have previously reported on the evidence of health benefits of exposure to sunlight for autoimmune and cardiovascular diseases, as well as myopia and some cancers [6]. It is challenging to generate high-quality evidence from human studies of benefits of exposure to UV radiation, primarily because it is difficult to capture accurate exposure data over a relevant time period. In addition, it is challenging to determine which wavelengths of sunlight are most important and further, how much of any effect of exposure to the sun is through vitamin D vs. non-vitamin D pathways. Determining whether associations are causal, and thus whether the balance of risks and benefits of sun exposure needs to be reconsidered, will require accumulation of evidence across epidemiological and mechanistic studies [222]. Recent studies have been largely cross-sectional, and/or used population-level exposures such as sunshine duration [223], ambient UV radiation or location [224], or remote sensing of green space coverage [225]. From these studies, benefits of higher green space coverage, longer duration of sunshine, or higher individual levels of sun exposure have included lower blood pressure in adults [226] and children [225], and reduced prevalence of obesity [223] and depression [227]. In a large study ($$n = 342$$,457) of patients undergoing dialysis in 2189 facilities across the United States, monthly average ambient UV irradiation at the clinic location had a linear inverse association with monthly average pre-dialysis systolic blood pressure, including after adjustment for ambient temperature [224]. The effect size was greater in Whites than in Blacks. These data are consistent with new analyses of the Melanoma in Southern Sweden study in which women with low or moderate past sun exposure (assessed by questionnaire including items on deliberate sun bathing, use of a sun bed, and travel for sunny holidays) had a greater risk of being prescribed anti-hypertensive medication by their physician than those with higher sun exposure; the association persisted after adjustment for being a smoker, exercise category, BMI, and education (adjusted OR 1.41, $95\%$ CI 1.3–1.6; adjusted OR 1.15, $95\%$ CI 1.1–1.2, respectively, for low and moderate sun exposure) [228]. In a recent study from South Korea, there were fewer cardiovascular (adjusted hazard ratio (HR) 0.68, $95\%$ CI 0.49–0.94) and cerebrovascular (adjusted HR 0.60, $95\%$ CI 0.47–0.77) events over 11 years in patients with vitiligo who had received long-term narrowband UV-B phototherapy (≥ 100 sessions) compared with those who had received < 3 phototherapy sessions [229]. There is accumulating evidence, including from small clinical trials, that UV-A (and possibly UV-B) irradiation influences blood pressure (and cardiovascular disease risk) through release of nitric oxide from stores in skin [230–232]. There is compelling evidence from multiple studies supporting reduced risk of myopia with greater exposure to UV radiation and/or high intensity visible light. Longitudinal cohort studies from China [233], the Netherlands [234], and Australia [235] show that more time spent outdoors during childhood, measured using a variety of metrics, was associated with reduced risk of developing myopia in childhood and young adulthood. Furthermore, two studies found that greater outdoor activity in childhood could reduce the adverse effect of higher levels of screen time [233, 234]. Another study showed that lower area of conjunctival autofluorescence was associated with a greater risk of developing myopia between ages 20 and 28 years [236], suggesting that the protective effects of sun exposure may continue into young adulthood. In a cross-sectional analysis within the Singapore birth cohort study, more time outdoors, but not light levels or the timing and frequency of light exposure, was associated with lower odds of myopia [237]. Another study found that greater green space coverage was associated with lower prevalence of myopia [238]. While more time outdoors seems well-established as protective for the development of myopia, details of optimal exposure to minimise myopia are not fully elucidated. Of note, the effect of greater exposure to UV radiation appears to be distinct from any effect of varying focal length during time outdoors [239]. There is now considerable evidence that there may be benefits of spending more time outdoors/sun exposure for the onset and progression of MS in addition to those ascribed to vitamin D (see below). A recent multi-ethnic case–control study confirmed a protective effect of higher sun exposure on risk of developing MS in white populations, and extended this to show the benefits were also apparent for blacks and Hispanics [240]. In contrast, benefits of higher 25(OH)D concentration were apparent only in United States Whites, possibly because 25(OH)D concentration is a better indicator of recent sun exposure in people with lighter skin. A case–control study in Canada, Italy, and Norway demonstrated that an accumulation model for sun exposure to age 15 years, rather than a critical periods model, provided the best fit for the protective effects of higher sun exposure on risk of MS in adulthood [241]. Importantly, among those who spent a lot of time outdoors in summer, use of sun protection did not alter MS risk. These findings highlight the need to provide balanced sun exposure messages that take account of geographical differences in weather patterns, skin pigmentation, and cultural practices [241]. Another case–control study showed a strong protective effect of greater time outdoors in the summer prior to diagnosis or during the first year of life, as well as higher ambient UV radiation, on the risk of developing paediatric MS [242]. The focus in relation to MS has been largely on the risk of developing the disease. There is new evidence that higher sun exposure prior to developing MS, and increasing sun exposure post-diagnosis, are associated with a more favourable post-diagnostic disease course [220, 243] [244], although there is some evidence that sun exposure may be detrimental for people with MS who have a sun-sensitive genotype [243]. A trial of narrowband UV-B radiation in people with clinically isolated syndrome to prevent the development of MS [245] found a lower risk of progression to MS in people receiving phototherapy than in the control group, although in this small study this was not statistically significant. Analyses of data from this trial have since revealed some novel potential pathways activated by narrowband (311 nm) UV-B, including transient changes in both the number of circulating leukocytes [246] and the production of pro-inflammatory cytokines [247]. Given the non-solar spectrum used, this is likely to be most relevant to a treatment setting, but it does indicate the possible importance of UV-B radiation for this condition. There is also recent evidence that exposure to higher intensity of UV radiation during early life may protect from the development of type 1 diabetes—an autoimmune disease of the pancreas. In a data-linkage-based cohort study of 29,078 children in Western Australia (~ $6\%$ of whom were diagnosed with type 1 diabetes by age 16), higher ambient (erythemally weighted) UV radiation was associated with reduced risk of developing type 1 diabetes, but only in males and only for UV radiation during the 3rd trimester and 1st year of life [248]. The authors concluded, assuming a causal association, that for every 100 kJ m−2 increase in total lifetime dose of ambient UV radiation dose, the relative risk of developing type 1 diabetes in males decreased by $29\%$. Emerging evidence suggests that higher antenatal sun exposure may reduce the risk of pre-term birth [249] and learning disabilities [250]. However, higher pre-delivery ambient temperatures have been linked to increased risk of pre-term birth [251, 252], complicating analyses where data on personal exposures and potential confounders are not available, and multiple environmental exposures acting at different time-points during pregnancy need to be considered. Additional research will be required to clarify the role of personal sun exposure during pregnancy on the many facets of the health of the offspring. Exposing the skin to UV radiation enhances feelings of well-being, possibly through the release of beta-endorphins following UV-B-induced DNA damage in keratinocytes [253]. This could provide a biological underpinning to an ‘addiction’ to tanning. In addition, serotonin is produced in the brain in response to bright sunlight [254], with this pathway potentially important for seasonal variability in mood and seasonal affective disorder. Observational studies show links between higher exposure to sunlight and reduced risk of depressive disorders, but confounding and reverse causation are possible explanations for these findings. However, artificial light therapy is established as a treatment for disorders such as seasonal affective disorder [255], and a recent experimental study confirms the benefit of sunlight. The single-blind clinical trial tested the effect of sunlight therapy (exposure of sun-protected forearms or calves to sunlight on sunny days (at least 10,000 lx) in Taiwan for an accumulated minimum of 30 min/day for a total of 14 days in 4 weeks) on depression in participants who were at least 1 month post-stroke. Testing at 1 month after the completion of the intervention showed a significant reduction in the depression score in the group receiving the sunlight therapy compared to a control (usual treatment) group [256]. There is growing interest in better understanding the potential benefits of sun exposure and the pathways and wavelengths involved. This information is critical to providing appropriate messaging to different populations on safe sun exposure to balance harms and benefits. ## Vitamin D Perhaps the best known benefit of sun exposure to the skin, driven by UV-B radiation, is the synthesis in the skin of vitamin D. Most populations derive very little of their vitamin D needs from diet, thus relying primarily on this UV-B-induced synthesis. ## The role of vitamin D in health outcomes Vitamin D is best known for its role in musculoskeletal health. Vitamin D status is defined according to the blood concentration of 25(OH)D, with a concentration of < 50 nmol L−1 commonly considered vitamin D deficient (including here unless specifically stated otherwise). Vitamin D deficiency as defined at this concentration is associated with increased risk of hip fractures in people aged 60 years and over [257]. It has been estimated that, assuming this association is causal, approximately $8\%$ of hip fractures occurring in adults aged ≥ 65 years in Australia are attributable to vitamin D deficiency (25(OH)D < 50 nmol L−1) [258]. Falls in older adults have also been linked to 25(OH)D concentration < 50 nmol L−1 [259]. Despite the established link between vitamin D and musculoskeletal health, the optimal 25(OH)D concentration to minimise fractures and falls is uncertain. Meta-analyses of randomised controlled trials (RCTs) show that vitamin D supplementation alone is only of benefit in people who are vitamin D deficient (< 50 nmol L−1) [260] or that it has no effect [261]. The Vitamin D and Omega-3 Trial in the United States did not find any benefit of supplementing older adults with vitamin D for 5 years on fractures, including in people whose baseline 25(OH)D concentration was < 50 nmol L−1 [262], but there was insufficient power to assess the effect in people with more severe vitamin D deficiency. These findings collectively suggest that the risk of falls and fractures may not increase until 25(OH)D concentration drops to the range currently considered to be severely deficient (< 25 nmol L−1). The importance of vitamin D for other health outcomes remains unclear. Observational studies are prone to confounding or reverse causality. This can be overcome by Mendelian randomisaton (MR) studies (which examine the association between genetically determined, rather than measured 25(OH)D concentration, and health outcomes), although most MR studies have not allowed for non-linear associations between genetically predicted 25(OH)D concentration and disease, and are mostly silent on the consequences of severe vitamin D deficiency. RCTs provide additional information regarding the causality of associations. However, RCTs test the effect of a particular supplement dose and dosing regimen in a specific population for a set length of time during one life period. The absence of effect in an RCT cannot, therefore, be used as proof of lack of a causal association. With these cautions in mind, we present below a summary of recent evidence for some common disease conditions. Low 25(OH)D concentration has been consistently linked with increased risk of depression in observational studies [263]. MR studies suggest that this association may not be causal [264, 265], and it is likely that adequate vitamin D status is a good marker of exposure to other beneficial wavelengths in sunlight that have an important effect on mood. Data from RCTs are somewhat inconsistent. A meta-analysis revealed an effect of vitamin D on negative emotion, but with very high heterogeneity, and the effect was predominantly seen in people who were vitamin D deficient or who were depressed at study baseline [266]. In support of this, a very large trial in the United States among adults without depression at baseline did not find any benefit of 5 years of vitamin D supplementation [267]. There are similarly inconsistent findings for type 2 diabetes mellitus (T2DM). A meta-analysis of observational studies found that each 1 standard deviation (SD) higher 25(OH)D concentration was associated with a $20\%$ lower risk of T2DM ($p \leq 0.001$), but a genetically predicted 1 SD increase was not significantly associated with T2DM [268]. An MR study in a Chinese population also found no association between genetically predicted 25(OH)D and T2DM [269]. An RCT in which 2423 people with prediabetes were supplemented with 4000 IU of vitamin D per day for ~ 2.5 years did not find a statistically significant reduction in the incidence of T2DM, although it is important to note that the mean 25(OH)D concentration at baseline was in the sufficient range (70 nmol L−1) and only $22\%$ of participants were vitamin D deficient (< 50 nmol L−1) [270, 271]. In a meta-analysis of RCTs in people without T2DM, vitamin D supplementation significantly reduced fasting glucose and fasting insulin but had no effect on incident T2DM overall or in progression from prediabetes to T2DM [272]. It is plausible that the findings in the observational studies reflect a non-vitamin D pathway of sun exposure, whereby higher 25(OH)D concentration is an indicator of having received sufficient sun exposure to gain the other benefits. This hypothesis is supported by mouse studies that suggest UV radiation-induced release of nitric oxide from the skin can suppress the development of glucose intolerance and hepatic lipid accumulation [273]. Observational studies consistently demonstrate inverse associations between 25(OH)D concentration and cancer incidence [274], but confounding and reverse causality are possible explanations for this finding, and this is not supported by MR studies [275] or RCTs [274]. Evidence is emerging, however, for a possible beneficial effect of vitamin D supplementation on cancer mortality [274], [276]. Case–control and cohort studies support an increased risk of MS with low 25(OH)D concentration [277], and this is supported by MR studies [278]. The association is less clear for other autoimmune diseases such as type 1 diabetes mellitus [279] and inflammatory bowel disease [278]; although there are suggestive protective effects, confidence intervals are wide and small effects cannot be ruled out. In terms of infectious diseases, observational studies [280], RCTs [281], and MR studies [282] indicate that low 25(OH)D increases risk and severity [283] of respiratory tract infection. An analysis including over 500,000 participants found strong evidence for a non-linear association between serum 25(OH)D concentration and coronary heart disease, stroke, and all-cause mortality. This was supported by an MR analysis, which suggested the risk associated with low genetically predicted 25(OH)D was only evident in those with measured 25(OH)D below 40 nmol L−1 [284]. However, a recent re-analysis (published after the reference list for this paper was finalised), using different model assumptions, found no significant association with genetically predicted 25(OH)D and mortality outcomes, irrespective of 25(OH)D concentration, suggesting that the earlier analysis may have generated incorrect findings (https://www.pubmed.ncbi.nlm.nih.gov/36528346/). Collectively, these findings suggest that vitamin D plays a causal role in some health outcomes, in addition to falls and fractures. However, there is no strong evidence to support increasing the recommended 25(OH)D target concentration to greater than 50 nmol L−1 [284], which is the concentration recommended by many organisations internationally. ## Revised action spectrum for vitamin D Action spectra are biological weighting functions that are used to assess the risks and benefits of exposure to different wavelengths of UV radiation. An action spectrum for the production of pre-vitamin D in the skin was produced by the Commission Internationale l’Éclairage (CIE) in 1982, showing a maximum effect at 297 nm, with essentially no production above 315 nm. However, the validity of this has been questioned because it is based on the use of human skin ex vivo. A recently published study calculated the action spectrum for serum 25(OH)D, the accepted molecule to determine vitamin D status, using an in vivo experiment [285]. The action spectrum was shifted 5 nm towards shorter wavelengths, suggesting that the CIE action spectrum may need to be revised. However, the effect of the shift is likely to be less relevant for natural sunlight than for artificial light sources [1]. Thus while further research is needed to elucidate the implications of a revised action spectrum for calculating the ratio of harms vs benefits of exposure to sunlight, the CIE action spectrum is likely to be adequate for risk benefit calculations. ## Effect of clothing, sunscreen, and skin pigmentation on vitamin D production Clothing provides good protection against erythema but also has a strong inhibitory effect on vitamin D synthesis. Full body clothing cover, especially in females, may contribute to the high prevalence of vitamin D deficiency in many countries with high insolation. Recent studies confirm the influence of clothing on 25(OH)D concentration [286, 287]. Sunscreen reduces the risk of skin cancer and premalignant lesions and is a mainstay of sun protection globally, but concerns have been raised that regular application of sunscreen may increase the risk of vitamin D deficiency. Two reviews suggest this is not the case [288, 289], although there are no RCTs of the effect of routine application of high SPF sunscreen on 25(OH)D concentration. Studies conducted since these reviews continue to suggest that sunscreen users have higher 25(OH)D concentration than those who do not use sunscreen [290, 291]. This is most likely because sunscreen users spend more time outdoors, but these studies suggest that using sunscreen does not obviate the benefits for vitamin D of spending more time outdoors. Dark-skinned immigrants to northern European countries tend to have lower 25(OH)D concentration than those with lighter skin. This is likely due to a combination of reduced vitamin D production in darker compared with lighter skins, and to behavioural differences. For example, an observational study comparing Danes with dark and light skin found that those with dark skin received a lower UV radiation dose and exposed less body surface area than those with lighter skin. There was only minimal difference in the increase in 25(OH)D concentration per joule of UV radiation exposure (light = 0.63 nmol L−1 J−1; dark = 0.53 nmol L−1 J−1) [292], although the analysis assumed a proportional response in 25(OH)D concentration with increasing body surface area exposed, which may not be the case. Experimental studies have generated discrepant estimates of the inhibitory effect of melanin. One study examined the effect on 25(OH)D concentration of exposing people with different skin types to five serial whole-body sub-erythemal exposures of solar-simulated UV radiation [293]. Comparing people with very light and very dark skin, the melanin inhibitory factor was estimated at ~ 1.3. In contrast, in a study in which the dose of solar-simulated radiation was given as a function of minimum erythemal dose (i.e. people with darker skins received a higher dose), and UV radiation was delivered to commonly exposed skin sites only, the melanin inhibitory factor was estimated to be ~ 8 [294]. This issue needs to be resolved as it has implications for public health advice for people with darker skin. ## Prevalence of vitamin D deficiency Vitamin D deficiency is prevalent across many parts of the world. However, obtaining accurate estimates is hampered by unreliable laboratory assays used in many studies. The prevalence of deficiency also depends on the 25(OH)D concentration used to define deficiency and on the time of year when samples were collected; these factors must be considered when interpreting these data. Figure 5 (with detailed data in Online Resource Table 1) shows the prevalence of vitamin D deficiency (25(OH)D concentration < 50 nmol L−1) derived from national studies (albeit using a range of assays, not all standardised to an international standard reference method) as well as some recent population studies. The results of these prevalence studies emphasise the apparent high prevalence of vitamin D deficiency in many parts of the world. With recovery of stratospheric ozone under the Montreal Protocol, projections are for lower UV-B radiation at high-latitude locations [1], which could increase the prevalence of vitamin D deficiency. This effect may be ameliorated by warming temperatures due to climate change, resulting in greater time outdoors, as demonstrated by a study from Germany, which found significantly higher 25(OH)D concentration in two extreme summers (2018 and 2019) compared with the preceding 4 summers [295]. However, in lower-latitude locations where the temperature is already high, warming temperatures due to climate change may reduce time outdoors and exacerbate the problem of vitamin D deficiency, particularly in urban populations. Fig. 5Prevalence of vitamin D deficiency (25(OH)D < 50 nmol/L). Figures for south-Asian countries (Sri Lanka, Nepal, Bangladesh, Pakistan, and India) are derived from a meta-analysis of studies (see Online Resource Table 2) that included a range of different populations and 25(OH)D assays. Similarly, figures for African countries are derived from a meta-analysis of studies (see Online Resource Table 3) that included a range of different populations and 25(OH)D assays. All other figures are based on population surveys. Data for Chile and Fiji are restricted to women. Data for Mongolia are restricted to men. Data for Denmark, Norway, Greece, Mexico, Ireland, and Iran are restricted to children and/or adolescents. For details of the adult age ranges for other countries see Online Resource Table 1 ## Climate change, depletion of stratospheric ozone, and human health In a previous assessment [79] we comprehensively reviewed the links between climate change, stratospheric ozone depletion and recovery, and human health. Little has been published on this topic in the last 4 years. The Sixth Assessment of the IPCC Working Group II on the effects of climate change on human health does not mention skin cancer, or other UV-induced health outcomes. Projections for ambient UV radiation in the coming years [1] suggest that with recovery of stratospheric ozone there will be a reduction in the UVI of 2–$5\%$ in northern mid-latitudes, a reduction of 4–$6\%$ in southern mid-latitudes, and no change in the tropics. Although large reductions in the UV Index in southern high latitudes (> 60°S) as stratospheric ozone recovers are projected, this is in a region with no resident populations (Ushuaia in southern Argentina has latitude of 55°S). However, with a growing number of tourists to Antarctica each summer season (approximately 170,000/season in recent years [296]) as well as base staff and researchers, the high variability in the UV Index, including maxima reaching UV Index of 14 [1] may pose risks to health. In addition to the effects of recovery of stratospheric ozone, reductions in cloud cover are projected to increase DNA-weighted UV radiation levels by $1.3\%$ per decade from 2050, based on data from 1998 to 2016 at four mid-latitude sites (Lauder, New Zealand; Table Mountain, South Africa; Haute Provence, France; Hohenpeissenberg, Germany) and one tropical high altitude site (Mauna Loa, Hawaii) [297]. In an extension of a previous analysis [298], Piacentini and colleagues applied a temperature modification to the carcinogenicity of UV radiation (‘effective carcinogenicity’) to estimate the incidence of KC in current and next centuries as a result of rising ambient temperature under different climate change scenarios (RCP2.6, RCP4.5, RCP6.0, RCP 8.5) [299]. The model projects increases in incidence for SCC for 2100 of $5.8\%$, $10.4\%$, $13.8\%$ and $21.4\%$ (for respective RCPs), and for BCC, $2.1\%$, $4.9\%$, $6.5\%$ and $9.9\%$. The model does not take account of changing UV radiation (as a result of changes in stratospheric ozone and/or cloud cover), or changes in sun exposure behaviour in relation to temperature. ## Gaps in knowledge During our assessment of the literature, we identified the following gaps in knowledge:Dynamic modelling is needed to better quantify the benefits of the Montreal Protocol: Trends in skin cancer in different countries are likely to be due to a combination of: [1] immigration patterns, leading to changed distribution of skin types; [2] changing recreational and occupational exposures; [3] concerted efforts to encourage populations to adopt sun-protective behaviours; and [4] changing surveillance habits, potentially resulting in over-diagnosis. Importantly, any predictions will need to take account of the influence of climate change on human behaviour, which will be an increasingly important driver of exposure to UV radiation. Better methods are needed to estimate prevalence of health conditions, including keratinocyte cancer, related to UV radiation: Other than internal cancers and melanoma, the lack of population-based registries makes it extremely challenging to estimate the population incidence or prevalence of conditions, including keratinocyte cancers, related to exposure to sunlight. Harnessing the power of data linkage may be one way of resolving this problem, recognising that this may under-estimate the burden of conditions that present less frequently in the health system. Photobiological studies to define the dose and pattern of UV radiation that confers minimal harm are required: There is currently no known UV radiation dose or exposure pattern that confers minimal harm to the skin and eyes. Related to this, over-exposure is poorly defined and, in some settings, sunburn is considered the only relevant indicator of over-exposure. A greater understanding of this issue would enable messages to be developed that balance the benefits and harms of exposure to sunlight. The extent of the problem relating to the use of photosensitising medications needs to be elucidated: Photosensitising medications can result in damage to skin and eyes. However, the extent of the problem, particularly for the eyes, is unclear. Studies are needed to better understand beneficial effects of exposure to UV radiation: Exposing the skin and eyes to the sun is likely to have benefits beyond those mediated by production of vitamin D. However, while evidence of benefit and mechanisms is maturing, it is still in its infancy. Clearly defined non-vitamin D biomarkers of benefit are needed so that studies can be conducted to identify the mechanisms, along with the dose and pattern of exposure needed to confer benefits. Public health messaging to guide personal sun exposure to minimise harms and maximise any benefits requires more detail on the relative effective doses of UV radiation: In particular, we need to quantify the effect on the balance of risks and harms of smaller doses of UV radiation to a greater body surface area. For example, a comparison of the effect of exposure to UV radiation with $5\%$ and $85\%$ of the body surface area exposed suggests that there may not be a linear increase in 25(OH)D concentration, but there is little information about percentages between these extremes. Current public health messages focus on lightly pigmented populations, with doses of UV radiation for harms and benefits uncertain for deeply pigmented skin: Skin cancers are rare, but vitamin D deficiency common, in those with deeply pigmented skin. Skin melanin protects the skin from UV-B-induced harms (e.g. DNA damage) and reduces vitamin D production, but the UV radiation dose at which these events occur needs to be quantified. Further, the action spectrum for vitamin D production may vary with skin type, but this is not currently known. Resolving these questions is important, enabling the development of evidence-based messages that recognise the increasing diversity of populations within countries. ## Conclusions Exposure to UV radiation has multiple harms and benefits. By preventing large increases in UV-B radiation, the Montreal Protocol has avoided many adverse health outcomes, consistent with Sustainable Development Goal (SDG) 3 (Ensure healthy lives and promote well-being for all at all ages). Further, the costs of the adverse effects of exposure to UV radiation are high and increasing, and occupational exposures represent a considerable economic burden. The Montreal Protocol plays a role in protecting outdoor workers, consistent with SDG 8 (Promote sustained, inclusive and sustainable economic growth, full and productive employment and decent work for all). In addition to avoiding large increases in UV radiation, the Montreal Protocol has stimulated research into the harms and benefits of sunlight exposure. The resulting knowledge has enabled harms to the skin and eyes to be ameliorated through the use of sun protection strategies. For the skin, this has been particularly important for people with lightly pigmented skin, and is evident in the plateauing trends in skin cancer seen in younger age groups in some countries. For the eyes, blindness caused by cataracts disproportionately affects people in developing countries due to lack of access to lens replacement surgery. These diverse effects are consistent with SDG 10 (Reduce inequality within and among countries). Alongside the harms, increasing recognition of the benefits is informing public health and clinical practice. For people with lightly pigmented skin, this underpins strategies to balance the risks and benefits of sun exposure. For those with deeply pigmented skin, knowledge of the importance of sun exposure may be particularly relevant for those living in areas with low ambient UV radiation for whom the benefits of sun exposure for most people (with the exception of those at risk of inflammatory skin disorders) are likely to outweigh the harms. In conclusion, sun exposure is critical for human life on Earth. The Montreal Protocol and its Amendments have prevented large increases in ambient UV-B radiation. This has both mitigated the adverse effects and enabled access to the beneficial effects of sun exposure, thus playing a vital role globally in health and economies. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 347 KB) ## References 1. 1.Bernhard, G. H., Bais, A. F., Aucamp, P. J., Klekociuk, A. R., Liley, J. B., & McKenzie, R. L. (2023). Stratospheric ozone, UV radiation, and climate interactions. Photochemical & Photobiological Sciences, This issue. 10.1007/s43630-023-00371-y 2. 2.Madronich, S., Sulzberger, B., Longstreth, J., Schikowski, T., Sulbaek Andersen, M. P. S., Solomon, K. R., & Wilson, S. R. (2023). Changes in tropospheric air quality related to the protection of stratospheric ozone and a changing climate. Photochemical & Photobiological Sciences, This issue. 10.1007/s43630-023-00369-6 3. 3.Barnes, P. W., Robson, T. M., Zepp, R. G., Bornman, J. F., Jansen, M. A. K., Ossola, R., Wang, Q.-W., Robinson, S. A., Foereid, B., Klekociuk, A. R., Martinez-Abaigar, J., Hou, W. C., & Paul, N. D. (2023). Interactive effects of changes in UV radiation and climate on terrestrial ecosystems, biogeochemical cycles, and feedbacks to the climate system. Photochemical & Photobiological Sciences, This issue. 10.1007/s43630-023-00376-7 4. 4.Neale, P. J., Williamson, C. E., Banaszak, A. T., Häder, D. P., Hylander, S., Ossola, R., Rose, K. A., Wängberg, S.-Å., & Zepp, R. G. (2023). The response of aquatic ecosystems to the interactive effects of stratospheric ozone depletion, UV radiation, and climate change. Photochemical & Photobiological Sciences, This issue. 10.1007/s43630-023-00370-z 5. 5.Andrady, A. L., Heikkila, A. M., Pandey, K. K., Bruckman, L. S., White, C. C., Zhu, M., & Zhu, L. (2023). Effects of UV radiation on natural and synthetic materials. Photochemical & Photobiological Sciences, This issue. 10.1007/s43630-023-00377-6 6. Lucas RM, Yazar S, Young AR, Norval M, de Gruijl FR, Takizawa Y, Rhodes LE, Sinclair CA, Neale RE. Human health in relation to exposure to solar ultraviolet radiation under changing stratospheric ozone and climate. Photochemical & Photobiological Sciences (2019.0) 18 641-680. DOI: 10.1039/c8pp90060d 7. Hayward NK, Wilmott JS, Waddell N, Johansson PA, Field MA, Nones K, Patch AM, Kakavand H, Alexandrov LB, Burke H, Jakrot V, Kazakoff S, Holmes O, Leonard C, Sabarinathan R, Mularoni L, Wood S, Xu Q, Waddell N, Tembe V. Whole-genome landscapes of major melanoma subtypes. Nature (2017.0) 545 175-180. DOI: 10.1038/nature22071 8. Premi S, Han L, Mehta S, Knight J, Zhao D, Palmatier MA, Kornacker K, Brash DE. Genomic sites hypersensitive to ultraviolet radiation. Proceedings of the National Academy of Sciences of the United States of America (2019.0) 116 24196-24205. DOI: 10.1073/pnas.1907860116 9. Lawrence KP, Delinasios GJ, Premi S, Young AR, Cooke MS. Perspectives on cyclobutane pyrimidine dimers—rise of the dark dimers. Photochemistry and Photobiology (2022.0) 98 609-616. DOI: 10.1111/php.13551 10. Liyanage UE, MacGregor S, Bishop DT, Shi J, An J, Ong JS, Han X, Scolyer RA, Martin NG, Medland SE, Byrne EM, Green AC, Saw RPM, Thompson JF, Stretch J, Spillane A, Jiang Y, Tian C, Agee M, Aslibekyan S. Multi-trait genetic analysis identifies autoimmune loci associated with cutaneous melanoma. Journal of Investigative Dermatology (2022.0) 142 1607-1616. DOI: 10.1016/j.jid.2021.08.449 11. Liyanage UE, Law MH, Han X, An J, Ong J-S, Gharahkhani P, Gordon S, Neale RE, Olsen CM, MacGregor S, Whiteman DC. Combined analysis of keratinocyte cancers identifies novel genome-wide loci. Human Molecular Genetics (2019.0) 28 3148-3160. DOI: 10.1093/hmg/ddz121 12. Hart PH, Norval M. Are there differences in immune responses following delivery of vaccines through acutely or chronically sun-exposed compared with sun-unexposed skin?. Immunology (2020.0) 159 133-141. DOI: 10.1111/imm.13128 13. Hart PH, Norval M. More than effects in skin: Ultraviolet radiation-induced changes in immune cells in human blood. Frontiers in Immunology (2021.0) 12 694086. DOI: 10.3389/fimmu.2021.694086 14. Hawkshaw NJ, Pilkington SM, Murphy SA, Al-Gazaq N, Farrar MD, Watson REB, Nicolaou A, Rhodes LE. UV radiation recruits CD4+GATA3+ and CD8+GATA3+ T cells while altering the lipid microenvironment following inflammatory resolution in human skin in vivo. Clinical & Translational Immunology (2020.0) 9 e01104. DOI: 10.1002/cti2.1104 15. Bernard JJ, Gallo RL, Krutmann J. Photoimmunology: how ultraviolet radiation affects the immune system. Nature Reviews Immunology (2019.0) 19 688-701. DOI: 10.1038/s41577-019-0185-9 16. Byrne SN, Limon-Flores AY, Ullrich SE. Mast cell migration from the skin to the draining lymph nodes upon ultraviolet irradiation represents a key step in the induction of immune suppression. Journal of Immunology (2008.0) 180 4648-4655. DOI: 10.4049/jimmunol.180.7.4648 17. Sarchio SNE, Scolyer RA, Beaugie C, McDonald D, Marsh-Wakefield F, Halliday GM, Byrne SN. Pharmacologically antagonizing the CXCR4-CXCL12 chemokine pathway with AMD3100 inhibits sunlight-induced skin cancer. Journal of Investigative Dermatology (2014.0) 134 1091-1100. DOI: 10.1038/jid.2013.424 18. Kok L-F, Ferguson AL, Marshall JE, Tse BCY, Halliday GM, Byrne SN. B cell-targeted immunotherapy limits tumor growth, enhances survival, and prevents lymph node metastasis of UV-induced keratinocyte cancers in mice. Journal of Investigative Dermatology (2020.0) 140 1459-1463. DOI: 10.1016/j.jid.2019.12.018 19. Shime H, Odanaka M, Tsuiji M, Matoba T, Imai M, Yasumizu Y, Uraki R, Minohara K, Watanabe M, Bonito AJ, Fukuyama H, Ohkura N, Sakaguchi S, Morita A, Yamazaki S. Proenkephalin(+) regulatory T cells expanded by ultraviolet B exposure maintain skin homeostasis with a healing function. Proceedings of the National Academy of Sciences of the United States of America (2020.0) 117 20696-20705. DOI: 10.1073/pnas.2000372117 20. Tse BCY, Byrne SN. Lipids in ultraviolet radiation-induced immune modulation. Photochemical & Photobiological Sciences (2020.0) 19 870-878. DOI: 10.1039/d0pp00146e 21. Liu L, Awoyemi AA, Fahy KE, Thapa P, Borchers C, Wu BY, McGlone CL, Schmeusser B, Sattouf Z, Rohan CA, Williams AR, Cates EE, Knisely C, Kelly LE, Bihl JC, Cool DR, Sahu RP, Wang J, Chen Y, Rapp CM. Keratinocyte-derived microvesicle particles mediate ultraviolet B radiation-induced systemic immunosuppression. Journal of Clinical Investigation (2021.0) 131 e144963. DOI: 10.1172/JCI144963 22. Tse BCY, Ireland RA, Lee JY, Marsh-Wakefield F, Kok LF, Don AS, Byrne SN. Exposure to systemic immunosuppressive ultraviolet radiation alters T cell recirculation through sphingosine-1-phosphate. Journal of Immunology (2021.0) 207 2278-2287. DOI: 10.4049/jimmunol.2001261 23. Lau FH, Powell CE, Adonecchi G, Danos DM, DiNardo AR, Chugden RJ, Wolf P, Castilla CF. Pilot phase results of a prospective, randomized controlled trial of narrowband ultraviolet B phototherapy in hospitalized COVID-19 patients. Experimental Dermatology (2022.0) 31 1109-1115. DOI: 10.1111/exd.14617 24. Maboshe W, Macdonald HM, Wassall H, Fraser WD, Tang JCY, Fielding S, Barker RN, Vickers MA, Ormerod A, Thies F. Low-dose vitamin D3 supplementation does not affect natural regulatory T cell population but attenuates seasonal changes in T cell-produced IFN-gamma: results from the D-SIRe2 randomized controlled trial. Frontiers in Immunology (2021.0) 12 623087. DOI: 10.3389/fimmu.2021.623087 25. Lossius AH, Sundnes O, Ingham AC, Edslev SM, Bjornholt JV, Lilje B, Bradley M, Asad S, Haraldsen G, Skytt-Andersen P, Holm JO, Berents TL. Shifts in the skin microbiota after UVB treatment in adult atopic dermatitis. Dermatology (2022.0) 238 109-120. DOI: 10.1159/000515236 26. Bustamante M, Hernandez-Ferrer C, Tewari A, Sarria Y, Harrison GI, Puigdecanet E, Nonell L, Kang W, Friedlander MR, Estivill X, Gonzalez JR, Nieuwenhuijsen M, Young AR. Dose and time effects of solar-simulated ultraviolet radiation on the in vivo human skin transcriptome. British Journal of Dermatology (2020.0) 182 1458-1468. DOI: 10.1111/bjd.18527 27. Hart PH, Norval M, Byrne SN, Rhodes LE. Exposure to ultraviolet radiation in the modulation of human diseases. Annual Review of Pathology (2019.0) 14 55-81. DOI: 10.1146/annurev-pathmechdis-012418-012809 28. Sarkar MK, Hile GA, Tsoi LC, Xing X, Liu J, Liang Y, Berthier CC, Swindell WR, Patrick MT, Shao S, Tsou P-S, Uppala R, Beamer MA, Srivastava A, Bielas SL, Harms PW, Getsios S, Elder JT, Voorhees JJ, Gudjonsson JE. Photosensitivity and type I IFN responses in cutaneous lupus are driven by epidermal-derived interferon kappa. Annals of Rheumatic Diseases (2018.0) 77 1653-1664. DOI: 10.1136/annrheumdis-2018-213197 29. Kulkarni NN, Takahashi T, Sanford JA, Tong Y, Gombart AF, Hinds B, Cheng JY, Gallo RL. Innate immune dysfunction in rosacea promotes photosensitivity and vascular adhesion molecule expression. Journal of Investigative Dermatology (2020.0) 140 645-655.e646. DOI: 10.1016/j.jid.2019.08.436 30. Skopelja-Gardner S, Tai J, Sun X, Tanaka L, Kuchenbecker JA, Snyder JM, Kubes P, Mustelin T, Elkon KB. Acute skin exposure to ultraviolet light triggers neutrophil-mediated kidney inflammation. Proceedings of the National Academy of Sciences of the United States of America (2021.0) 118 e2019097118. DOI: 10.1073/pnas.2019097118 31. Lim H-S, Yoon K-N, Chung JH, Lee Y-S, Lee DH, Park G. Chronic ultraviolet irradiation to the skin dysregulates adrenal medulla and dopamine metabolism in vivo. Antioxidants (2021.0) 10 920. DOI: 10.3390/antiox10060920 32. Diffey B. Erythema and acclimatization following repeated sun exposure: a modeling study. Photochemistry and Photobiology (2021.0) 97 1558-1567. DOI: 10.1111/php.13466 33. Laskar R, Ferreiro-Iglesias A, Bishop DT, Iles MM, Kanetsky PA, Armstrong BK, Law MH, Goldstein AM, Aitken JF, Giles GG, Robbins HA, Cust AE. Risk factors for melanoma by anatomical site: an evaluation of aetiological heterogeneity. British Journal of Dermatology (2021.0) 184 1085-1093. DOI: 10.1111/bjd.19705 34. Ghiasvand R, Robsahm TE, Green AC, Rueegg CS, Weiderpass E, Lund E, Veierød MB. Association of phenotypic characteristics and UV radiation exposure with risk of melanoma on different body sites. JAMA Dermatology (2018.0) 155 39-49. DOI: 10.1001/jamadermatol.2018.3964 35. Arnold M, de Vries E, Whiteman DC, Jemal A, Bray F, Parkin DM, Soerjomataram I. Global burden of cutaneous melanoma attributable to ultraviolet radiation in 2012. International Journal of Cancer (2018.0) 143 1305-1314. DOI: 10.1002/ijc.31527 36. O'Sullivan DE, Brenner DR, Villeneuve PJ, Walter SD, Demers PA, Friedenreich CM, King WD, Com PST. Estimates of the current and future burden of melanoma attributable to ultraviolet radiation in Canada. Preventive Medicine (2019.0) 122 81-90. DOI: 10.1016/j.ypmed.2019.03.012 37. Arnold M, Kvaskoff M, Thuret A, Guenel P, Bray F, Soerjomataram I. Cutaneous melanoma in France in 2015 attributable to solar ultraviolet radiation and the use of sunbeds. Journal of the European Academy of Dermatology and Venereology (2018.0) 32 1681-1686. DOI: 10.1111/jdv.15022 38. Lopes FCPS, Sleiman MG, Sebastian K, Bogucka R, Jacobs EA, Adamson AS. UV exposure and the risk of cutaneous melanoma in skin of color: a systematic review. JAMA Dermatology (2021.0) 157 213-219. DOI: 10.1001/jamadermatol.2020.4616 39. O'Sullivan DE, Brenner DR, Villeneuve PJ, Walter SD, Demers PA, Friedenreich CM, King WD, Com PST. The current burden of non-melanoma skin cancer attributable to ultraviolet radiation and related risk behaviours in Canada. Cancer Causes & Control (2021.0) 32 279-290. DOI: 10.1007/s10552-020-01382-1 40. Little MP, Linet MS, Kimlin MG, Lee T, Tatalovich Z, Sigurdson AJ, Cahoon EK. Cumulative solar ultraviolet radiation exposure and basal cell carcinoma of the skin in a nationwide US cohort using satellite and ground-based measures. [Article]. Environmental Health (2019.0) 18 114. DOI: 10.1186/s12940-019-0536-9 41. Kolitz E, Lopes FCPS, Arffa M, Pineider J, Bogucka R, Adamson AS. UV exposure and the risk of keratinocyte carcinoma in skin of color: a systematic review. JAMA Dermatology (2022.0) 158 542-546. DOI: 10.1001/jamadermatol.2022.0263 42. 42.World Health Organisation (2021) The effect of occupational exposure to solar ultraviolet radiation on malignant skin melanoma and nonmelanoma skin cancer: a systematic review and meta-analysis from the WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury. Geneva: World Health Organization; 2021. Licence: CC BY-NC-SA 3.0 IGO 43. Loney T, Paulo MS, Modenese A, Gobba F, Tenkate T, Whiteman DC, Green AC, John SM. Global evidence on occupational sun exposure and keratinocyte cancers: a systematic review. British Journal of Dermatology (2021.0) 184 208-218. DOI: 10.1111/bjd.19152 44. Peters CE, Kim J, Song C, Heer E, Arrandale VH, Pahwa M, Labreche F, McLeod CB, Davies HW, Ge CB, Demers PA. Burden of non-melanoma skin cancer attributable to occupational sun exposure in Canada. International Archives of Occupational and Environmental Health (2019.0) 92 1151-1157. DOI: 10.1007/s00420-019-01454-z 45. Nurminen M, Karjalainen A. Epidemiologic estimate of the proportion of fatalities related to occupational factors in Finland. Scandinavian Journal of Work, Environment & Health (2001.0) 27 161-213. DOI: 10.5271/sjweh.605 46. Young C, Rushton L. Occupational cancer in Britain. Skin cancer. British Journal of Cancer (2012.0) 107 S71-75. DOI: 10.1038/bjc.2012.120 47. Brand RM, Stottlemyer JM, Paglia MC, Carey CD, Falo LD. Ethanol consumption synergistically increases ultraviolet radiation induced skin damage and immune dysfunction. Journal of Dermatological Science (2021.0) 101 40-48. DOI: 10.1016/j.jdermsci.2020.11.001 48. Madronich S, Lee-Taylor JM, Wagner M, Kyle J, Hu Z, Landolfi R. Estimation of skin and ocular damage avoided in the United States through implementation of the montreal protocol on substances that deplete the Ozone layer. ACS Earth and Space Chemistry (2021.0) 5 1876-1888. DOI: 10.1021/acsearthspacechem.1c00183 49. Arnold M, Singh D, Laversanne M, Vignat J, Vaccarella S, Meheus F, Cust AE, de Vries E, Whiteman DC, Bray F. Global burden of cutaneous melanoma in 2020 and projections to 2040. JAMA Dermatology (2022.0) 158 495-503. DOI: 10.1001/jamadermatol.2022.0160 50. Urban K, Mehrmal S, Uppal P, Giesey RL, Delost GR. The global burden of skin cancer: a longitudinal analysis from the Global Burden of Disease Study, 1990–2017. JAAD International (2021.0) 2 98-108. DOI: 10.1016/j.jdin.2020.10.013 51. Li Z, Fang Y, Chen H, Zhang T, Yin X, Man J, Yang X, Lu M. Spatiotemporal trends of the global burden of melanoma in 204 countries and territories from 1990 to 2019: results from the 2019 global burden of disease study. Neoplasia (2022.0) 24 12-21. DOI: 10.1016/j.neo.2021.11.013 52. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA A Cancer Journal for Clinicians (2018.0) 68 394-424. DOI: 10.3322/caac.21492 53. 53.Australian Institute of Health and Welfare (AIHW). (2021). Cancer in Australia 2021. Australian Government, Canberra. www.aihw.gov.au/reports/cancdr/cancer-in-austrlia-2021/data. Accessed 01 Apr 2022. Canberra, Australia 54. Ferlay J, Colombet M, Soerjomataram I, Dyba T, Randi G, Bettio M, Gavin A, Visser O, Bray F. Cancer incidence and mortality patterns in Europe: Estimates for 40 countries and 25 major cancers in 2018. European Journal of Cancer (2018.0) 103 356-387. DOI: 10.1016/j.ejca.2018.07.005 55. Lashway SG, Harris RB, Farland LV, O'Rourke MK, Dennis LK. Age and cohort trends of malignant melanoma in the United States. Cancers (Basel) (2021.0) 13 3866. DOI: 10.3390/cancers13153866 56. Kurtansky NR, Dusza SW, Halpern AC, Hartman RI, Geller AC, Marghoob AA, Rotemberg VM, Marchetti MA. An epidemiologic analysis of melanoma overdiagnosis in the United States, 1975–2017. Journal of Investigative Dermatology (2022.0) 142 1804-1811. DOI: 10.1016/j.jid.2021.12.003 57. Aitken JF, Youlden DR, Baade PD, Soyer HP, Green AC, Smithers BM. Generational shift in melanoma incidence and mortality in Queensland, Australia, 1995–2014. International Journal of Cancer (2018.0) 142 1528-1535. DOI: 10.1002/ijc.31141 58. Eriksson H, Nielsen K, Vassilaki I, Lapins J, Mikiver R, Lyth J, Isaksson K. Trend shifts in age-specific incidence for in situ and invasive cutaneous melanoma in Sweden. Cancers (Basel) (2021.0) 13 2838. DOI: 10.3390/cancers13112838 59. Sacchetto L, Zanetti R, Comber H, Bouchardy C, Brewster DH, Broganelli P, Chirlaque MD, Coza D, Galceran J, Gavin A, Hackl M, Katalinic A, Laronningen S, Louwman MWJ, Morgan E, Robsahm TE, Sanchez MJ, Tryggvadottir L, Tumino R, Van Eycken E. Trends in incidence of thick, thin and in situ melanoma in Europe. European Journal of Cancer (2018.0) 92 108-118. DOI: 10.1016/j.ejca.2017.12.024 60. Welch HG, Mazer BL, Adamson AS. The rapid rise in cutaneous melanoma diagnoses. The New England Journal of Medicine (2021.0) 384 72-79. DOI: 10.1056/NEJMsb2019760 61. Matsumoto M, Wack S, Weinstock MA, Geller A, Wang H, Solano FX, Kirkwood JM, Ferris LK. Five-year outcomes of a melanoma screening initiative in a large health care system. JAMA Dermatology (2022.0) 158 504-512. DOI: 10.1001/jamadermatol.2022.0253 62. Olsen CM, Thompson JF, Pandeya N, Whiteman DC. Evaluation of sex-specific incidence of melanoma. JAMA Dermatology (2020.0) 156 553-560. DOI: 10.1001/jamadermatol.2020.0470 63. Gulliver W, Gulliver S, Power RJ, Penney M, Lane D. The incidence of cutaneous malignant melanoma in Eastern Newfoundland and Labrador, Canada, from 2007 to 2015. Dermatology (2021.0) 238 527-533. DOI: 10.1159/000519193 64. Defossez G, Uhry Z, Delafosse P, Dantony E, d'Almeida T, Plouvier S, Bossard N, Bouvier AM, Molinié F, Woronoff AS, Colonna M, Grosclaude P, Remontet L, Monnereau A. Cancer incidence and mortality trends in France over 1990–2018 for solid tumors: the sex gap is narrowing. BMC Cancer (2021.0) 21 726. DOI: 10.1186/s12885-021-08261-1 65. Dulskas A, Cerkauskaite D, Vincerževskiene I, Urbonas V. Trends in incidence and mortality of skin melanoma in Lithuania 1991–2015. International Journal of Environmental Research and Public Health (2021.0) 18 4165. DOI: 10.3390/ijerph18084165 66. Korovin S, Fedorenko Z, Michailovich Y, Kukushkina M, Sekerija M, Ryzhov A. Burden of malignant melanoma in Ukraine in 2002–2013: incidence, mortality and survival. Experimental Oncology (2020.0) 42 324-329. DOI: 10.32471/exp-oncology.2312-8852.vol-42-no-4.15334 67. Pehalova L, Krejci D, Snajdrova L, Dusek L. Cancer incidence trends in the Czech Republic. Cancer Epidemiology (2021.0) 74 101975. DOI: 10.1016/j.canep.2021.101975 68. Liszkay G, Kiss Z, Gyulai R, Oláh J, Holló P, Emri G, Csejtei A, Kenessey I, Benedek A, Polányi Z, Nagy-Erdei Z, Daniel A, Knollmajer K, Várnai M, Vokó Z, Nagy B, Rokszin G, Fábián I, Barcza Z, Polgár C. Changing trends in melanoma incidence and decreasing melanoma mortality in Hungary between 2011 and 2019: a nationwide epidemiological study. Frontiers in Oncology (2020.0) 10 612459. DOI: 10.3389/fonc.2020.612459 69. Bai R, Huang H, Li M, Chu M. Temporal trends in the incidence and mortality of skin malignant melanoma in China from 1990 to 2019. Journal of Oncology (2021.0) 2021 9989824. DOI: 10.1155/2021/9989824 70. Kim T, Yoon S, Shin DE, Lee SC, Oh J, Lee SY, Kim DK, Kim S, Jung B, Kim M, Lee S. Incidence and survival rates of cutaneous melanoma in South Korea using nationwide health insurance claims data. Cancer Research and Treatment (2021.0) 54 937-949. DOI: 10.4143/crt.2021.871 71. Wu Y, Wang Y, Wang L, Yin P, Lin Y, Zhou M. Burden of melanoma in China, 1990–2017: findings from the 2017 global burden of disease study. International Journal of Cancer (2020.0) 147 692-701. DOI: 10.1002/ijc.32764 72. Oh CC, Jin A, Koh WP. Trends of cutaneous basal cell carcinoma, squamous cell carcinoma, and melanoma among the Chinese, Malays, and Indians in Singapore from 1968–2016. JAAD International (2021.0) 4 39-45. DOI: 10.1016/j.jdin.2021.05.006 73. Chen ML, de Vere Hunt IJ, John EM, Weinstock MA, Swetter SM, Linos E. Differences in thickness-specific incidence and factors associated with cutaneous melanoma in the US from 2010 to 2018. JAMA Oncology (2022.0) 8 755-759. DOI: 10.1001/jamaoncol.2022.0134 74. Kelm RC, Ali Y, Orrell K, Rangel SM, Kruse L, Wagner AM, Gerami P, West DP, Nardone B. Age and sex differences for malignant melanoma in the pediatric population-childhood versus adolescence: analysis of current nationwide data from the national cancer institute surveillance, epidemiology, and end results (SEER) program. JAMA Dermatology (2021.0) 84 862-864. DOI: 10.1016/j.jaad.2020.10.050 75. Heer EV, Harper AS, Sung H, Jemal A, Fidler-Benaoudia MM. Emerging cancer incidence trends in Canada: the growing burden of young adult cancers. Cancer (2020.0) 126 4553-4562. DOI: 10.1002/cncr.33050 76. Bucchi L, Mancini S, Crocetti E, Dal Maso L, Baldacchini F, Vattiato R, Giuliani O, Ravaioli A, Caldarella A, Carrozzi G, Ferretti S, Filiberti RA, Fusco M, Gatti L, Gili A, Magoni M, Mangone L, Mazzoleni G, Michiara M, Panato C. Mid-term trends and recent birth-cohort-dependent changes in incidence rates of cutaneous malignant melanoma in Italy. International Journal of Cancer (2021.0) 148 835-844. DOI: 10.1002/ijc.33259 77. Memon A, Banniser P, Rogers I, Sundin J, Al-Ayadhy B, James PW, McNally RJQ. Changing epidemiology and age-specific incidence of cutaneous malignant melanoma in England: an analysis of the national cancer registration data by age, gender and anatomical site, 1981–2018. The Lancet Regional Health Europe (2021.0) 2 100024. DOI: 10.1016/j.lanepe.2021.100024 78. Rousi EK, Kallionpää RA, Kallionpää RE, Juteau SM, Talve LAI, Hernberg MM, Vihinen PP, Kähäri VM, Koskivuo IO. Increased incidence of melanoma in children and adolescents in Finland in 1990–2014: nationwide re-evaluation of histopathological characteristics. Annals of Medicine (2022.0) 54 244-252. DOI: 10.1080/07853890.2022.2026001 79. Bernhard GH, Neale RE, Barnes PW, Neale PJ, Zepp RG, Wilson SR, Andrady AL, Bais AF, McKenzie RL, Aucamp PJ, Young PJ, Liley JB, Lucas RM, Yazar S, Rhodes LE, Byrne SN, Hollestein LM, Olsen CM, Young AR, Robson TM. Environmental effects of stratospheric ozone depletion, UV radiation and interactions with climate change: UNEP environmental effects assessment panel, update 2019. Photochemical & Photobiological Sciences (2020.0) 19 542-584. DOI: 10.1039/d0pp90011g 80. Neale RE, Barnes PW, Robson TM, Neale PJ, Williamson CE, Zepp RG, Wilson SR, Madronich S, Andrady AL, Heikkilä AM, Bernhard GH, Bais AF, Aucamp PJ, Banaszak AT, Bornman JF, Bruckman LS, Byrne SN, Foereid B, Häder DP, Hollestein LM. Environmental effects of stratospheric ozone depletion, UV radiation, and interactions with climate change: UNEP environmental effects assessment panel, update 2020. Photochemical & Photobiological Sciences (2021.0) 20 1-67. DOI: 10.1007/s43630-020-00001-x 81. Barnes PW, Robson TM, Neale PJ, Williamson CE, Zepp RG, Madronich S, Wilson SR, Andrady AL, Heikkilä AM, Bernhard GH, Bais AF, Neale RE, Bornman JF, Jansen MAK, Klekociuk AR, Martinez-Abaigar J, Robinson SA, Wang QW, Banaszak AT, Häder DP. Environmental effects of stratospheric ozone depletion, UV radiation, and interactions with climate change: UNEP environmental effects assessment panel, update 2021. Photochemical & Photobiological Sciences (2022.0) 21 275-301. DOI: 10.1007/s43630-022-00176-5 82. Ferlay J, Colombet M, Soerjomataram I, Parkin DM, Piñeros M, Znaor A, Bray F. Cancer statistics for the year 2020: an overview. International Journal of Cancer (2021.0) 149 778-789. DOI: 10.1002/ijc.33588 83. Berk-Krauss J, Stein JA, Weber J, Polsky D, Geller AC. New systematic therapies and trends in cutaneous melanoma deaths among US Whites, 1986–2016. American Journal of Public Health (2020.0) 110 731-733. DOI: 10.2105/AJPH.2020.305567 84. Yang DD, Salciccioli JD, Marshall DC, Sheri A, Shalhoub J. Trends in malignant melanoma mortality in 31 countries from 1985 to 2015. British Journal of Dermatology (2020.0) 183 1056-1064. DOI: 10.1111/bjd.19010 85. Gutierrez-Gonzalez E, Lopez-Abente G, Aragones N, Pollan M, Pastor-Barriuso R, Sanchez MJ, Perez-Gomez B. Trends in mortality from cutaneous malignant melanoma in Spain (1982–2016): sex-specific age-cohort-period effects. Journal of the European Academy of Dermatology and Venereology (2019.0) 33 1522-1528. DOI: 10.1111/jdv.15565 86. 86.Environmental Health Intelligence New Zealand (ehinz) (2020). Fact sheet: melanoma mortality. https://www.ehinz.ac.nz/indicators/uv-exposure/melanoma/. Accessed 14 Jan 2021 87. van Niekerk CC, Groenewoud HMM, Verbeek ALM. Trends and projections in cutaneous melanoma death in the Netherlands from 1950 to 2045. Medicine Open (2021.0) 100 e27784 88. Nader Marta G, Munhoz RR, Teixeira MP, Waldvogel BC, Pires de Camargo V, Feher O, Sanches JA. Trends in melanoma mortality in Brazil: a registry-based study. JCO Global Oncology (2020.0) 6 1766-1771. DOI: 10.1200/go.20.00426 89. Olsen CM, Pandeya N, Whiteman DC. International increases in Merkel cell carcinoma incidence rates between 1997 and 2016. Journal of Investigative Dermatology (2021.0) 141 2596-2601. DOI: 10.1016/j.jid.2021.04.007 90. de Melo AC, Santos Thuler LC. Trends in the incidence and morbidity of Merkel cell carcinoma in Brazil. Future Oncology (2021.0) 17 2857-2865. DOI: 10.2217/fon-2020-1313 91. Stang A, Becker JC, Nghiem P, Ferlay J. The association between geographic location and incidence of Merkel cell carcinoma in comparison to melanoma: an international assessment. European Journal of Cancer (2018.0) 94 47-60. DOI: 10.1016/j.ejca.2018.02.003 92. Jacobs D, Huang H, Olino K, Weiss S, Kluger H, Judson BL, Zhang Y. Assessment of age, period, and birth cohort effects and trends in Merkel cell carcinoma incidence in the United States. JAMA Dermatology (2021.0) 157 59-65. DOI: 10.1001/jamadermatol.2020.4102 93. Feng H, Shuda M, Chang Y, Moore PS. Clonal integration of a polyomavirus in human Merkel cell carcinoma. Science (2008.0) 319 1096-1100. DOI: 10.1126/science.1152586 94. Goh G, Walradt T, Markarov V, Blom A, Riaz N, Doumani R, Stafstrom K, Moshiri A, Yelistratova L, Levinsohn J, Chan TA, Nghiem P, Lifton RP, Choi J. Mutational landscape of MCPyV-positive and MCPyV-negative Merkel cell carcinomas with implications for immunotherapy. Oncotarget (2016.0) 7 3403-3415. DOI: 10.18632/oncotarget.6494 95. Mokánszki A, Méhes G, Csoma SL, Kollár S, Chang Chien YC. Molecular profiling of Merkel cell polyomavirus-associated Merkel cell carcinoma and cutaneous melanoma. Diagnostics (Basel) (2021.0) 11 212. DOI: 10.3390/diagnostics11020212 96. Harms KL, Healy MA, Nghiem P, Sober AJ, Johnson TM, Bichakjian CK, Wong SL. Analysis of prognostic factors from 9387 Merkel cell carcinoma cases forms the basis for the new 8th edition AJCC staging system. Annals of Surgical Oncology (2016.0) 23 3564-3571. DOI: 10.1245/s10434-016-5266-4 97. D'Angelo SP, Bhatia S, Brohl AS, Hamid O, Mehnert JM, Terheyden P, Shih KC, Brownell I, Lebbé C, Lewis KD, Linette GP, Milella M, Georges S, Shah P, Ellers-Lenz B, Bajars M, Güzel G, Nghiem PT. Avelumab in patients with previously treated metastatic Merkel cell carcinoma: long-term data and biomarker analyses from the single-arm phase 2 JAVELIN Merkel 200 trial. Journal for Immunotherapy of Cancer (2020.0) 8 e000674. DOI: 10.1136/jitc-2020-000674 98. Chang WC, Lin AY, Hsu JC, Wu CE, Goh C, Chou P, Kuo K, Chang A, Palencia R. A cost-utility analysis of avelumab for metastatic Merkel cell carcinoma in Taiwan. Cancer Reports (2021.0) 4 e1399. DOI: 10.1002/cnr2.1399 99. Garza-Davila VF, Valdespino-Valdes J, Barrera FJ, Ocampo-Candiani J, Garza-Rodríguez V. Clinical impact of immunotherapy in Merkel cell carcinoma patients: a systematic review and meta-analysis. JAMA Dermatology (2022.0) 87 121-130. DOI: 10.1016/j.jaad.2021.04.024 100. Venables ZC, Nijsten T, Wong KF, Autier P, Broggio J, Deas A, Harwood CA, Hollestein LM, Langan SM, Morgan E, Proby CM, Rashbass J, Leigh IM. Epidemiology of basal and cutaneous squamous cell carcinoma in the UK 2013–15: a cohort study. [Article in Press]. British Journal of Dermatology (2019.0) 181 474-482. DOI: 10.1111/bjd.17873 101. Tokez S, Hollestein L, Louwman M, Nijsten T, Wakkee M. Incidence of multiple vs first cutaneous squamous cell carcinoma on a nationwide scale and estimation of future incidences of cutaneous squamous cell carcinoma. JAMA Dermatology (2020.0) 156 1300-1306. DOI: 10.1001/jamadermatol.2020.3677 102. Cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life years for 29 cancer groups from 2010 to 2019: a systematic analysis for the global burden of disease study 2019. JAMA Oncology (2021.0) 8 420. DOI: 10.1001/jamaoncol.2021.6987 103. Zhang W, Zeng W, Jiang A, He Z, Shen X, Dong X, Feng J, Lu H. Global, regional and national incidence, mortality and disability-adjusted life-years of skin cancers and trend analysis from 1990 to 2019: an analysis of the Global Burden of Disease Study 2019. Cancer Medicine (2021.0) 10 4905-4922. DOI: 10.1002/cam4.4046 104. Pondicherry A, Martin R, Meredith I, Rolfe J, Emanuel P, Elwood M. The burden of non-melanoma skin cancers in Auckland, New Zealand. [Article]. Australasian Journal of Dermatology (2018.0) 59 210-213. DOI: 10.1111/ajd.12751 105. Pandeya N, Olsen CM, Whiteman DC. The incidence and multiplicity rates of keratinocyte cancers in Australia. Medical Journal of Australia (2017.0) 207 339-343. DOI: 10.5694/mja17.00284 106. Staples MP, Elwood M, Burton RC, Williams JL, Marks R, Giles GG. Non-melanoma skin cancer in Australia: the 2002 national survey and trends since 1985. Medical Journal of Australia (2006.0) 184 6-10. DOI: 10.5694/j.1326-5377.2006.tb00086.x 107. Adelson P, Sharplin GR, Roder DM, Eckert M. Keratinocyte cancers in South Australia: incidence, geographical variability and service trends. [Article]. Australian and New Zealand Journal of Public Health (2018.0) 42 329-333. DOI: 10.1111/1753-6405.12806 108. Adalsteinsson JA, Ratner D, Olafsdottir E, Grant-Kels J, Ungar J, Silverberg JI, Kristjansson AK, Jonasson JG, Tryggvadottir L. Basal cell carcinoma: an emerging epidemic in women in Iceland. British Journal of Dermatology (2020.0) 183 847-856. DOI: 10.1111/bjd.18937 109. Adalsteinsson JA, Olafsdottir E, Ratner D, Waldman R, Feng H, Ungar J, Silverberg JI, Kristjansson AK, Jonasson JG, Tryggvadottir L. Invasive and in situ squamous cell carcinoma of the skin: a nationwide study in Iceland. British Journal of Dermatology (2021.0) 185 537-547. DOI: 10.1111/bjd.19879 110. Ilić D, Videnović G, Kozomara R, Radaković SS, Vlahović Z, Matvijenko V, Živković S. Non-melanoma skin cancers in Serbia (1999–2015)—the need for national prevention and control strategy. [Article]. Vojnosanitetski Pregled (2020.0) 77 1154-1160. DOI: 10.2298/vsp181112201i 111. Mirza FN, Yumeen S, Walter FM. The epidemiology of malignant melanoma, squamous cell carcinoma and basal cell carcinoma in the UK from 2004 to 2014: a population-based cohort analysis using the Clinical Practice Research Datalink. [Article]. British Journal of Dermatology (2021.0) 184 365-367. DOI: 10.1111/bjd.19542 112. Aggarwal P, Knabel P, Fleischer AB. United States burden of melanoma and non-melanoma skin cancer from 1990 to 2019. JAMA Dermatology (2021.0) 85 388-395. DOI: 10.1016/j.jaad.2021.03.109 113. Kwiatkowska M, Ahmed S, Ardern-Jones MR, Bhatti LA, Bleiker TO, Gavin A, Hussain S, Huws DW, Irvine L, Langan SM, Millington GWM, Mitchell H, Murphy R, Paley L, Proby CM, Thomson CS, Thomas R, Turner C, Vernon S, Venables ZC. A summary of the updated report on the incidence and epidemiological trends of keratinocyte cancers in the UK 2013–2018. [Letter]. British Journal of Dermatology (2022.0) 186 367-369. DOI: 10.1111/bjd.20764 114. Olsen CM, Pandeya N, Green AC, Ragaini BS, Venn AJ, Whiteman DC. Keratinocyte cancer incidence in Australia: a review of population-based incidence trends and estimates of lifetime risk. Public Health Research and Practice (2022.0) 32 e3212203. DOI: 10.17061/phrp3212203 115. Dziunycz PJ, Schuller E, Hofbauer GFL. Prevalence of actinic keratosis in patients attending general practitioners in Switzerland. [Article]. Dermatology (2018.0) 234 186-191. DOI: 10.1159/000491820 116. Ferrándiz-Pulido C, Lera-Imbuluzqueta M, Ferrándiz C, Plazas-Fernandez MJ. Prevalence of actinic keratosis in different regions of Spain: the EPIQA Study. [Article]. Actas Dermo-Sifiliográficas (2018.0) 109 83-86. DOI: 10.1016/j.adengl.2017.05.023 117. Tokez S, Wakkee M, Louwman M, Noels E, Nijsten T, Hollestein L. Assessment of cutaneous squamous cell carcinoma (cSCC) in situ incidence and the risk of developing invasive cSCC in patients with prior cSCC in situ vs the general population in the Netherlands, 1989–2017. JAMA Dermatology (2020.0) 156 973-981. DOI: 10.1001/jamadermatol.2020.1988 118. Friman TK, Jäämaa-Holmberg S, Åberg F, Helanterä I, Halme M, Pentikäinen MO, Nordin A, Lemström KB, Jahnukainen T, Räty R, Salmela B. Cancer risk and mortality after solid organ transplantation: a population-based 30-year cohort study in Finland. International Journal of Cancer (2022.0) 150 1779-1791. DOI: 10.1002/ijc.33934 119. Poizot-Martin I, Lions C, Delpierre C, Makinson A, Allavena C, Fresard A, Brégigeon S, Rojas TR, Delobel P. Prevalence and spectrum of second primary malignancies among people living with HIV in the French Dat'AIDS Cohort. Cancers (Basel) (2022.0) 14 401. DOI: 10.3390/cancers14020401 120. Yuan T, Hu Y, Zhou X, Yang L, Wang H, Li L, Wang J, Qian H-Z, Clifford GM, Zou H. Incidence and mortality of non-AIDS-defining cancers among people living with HIV: A systematic review and meta-analysis. eClinicalMedicine (2022.0) 52 101613. DOI: 10.1016/j.eclinm.2022.101613 121. D'Arcy ME, Beachler DC, Pfeiffer RM, Curtis JR, Mariette X, Seror R, Mahale P, Rivera DR, Yanik EL, Engels EA. Tumor necrosis factor inhibitors and the risk of cancer among older Americans with rheumatoid arthritis. Cancer Epidemiology, Biomarkers & Prevention (2021.0) 30 2059-2067. DOI: 10.1158/1055-9965.EPI-21-0125 122. Turk T, Saad AM, Al-Husseini MJ, Gad MM. The risk of melanoma in patients with chronic lymphocytic leukemia: a population-based study. Current Problems in Cancer (2020.0) 44 100511-100511. DOI: 10.1016/j.currproblcancer.2019.100511 123. Plasmeijer E, Jiyad Z, Way M, Marquart L, Miura K, Campbell S, Isbel N, Fawcett J, Ferguson LE, Davis M, Whiteman DC, Soyer HP, O'Rourke P, Green AC. Extreme incidence of skin cancer in kidney and liver yransplant recipients living with high sun exposure. Acta Dermato-Venereologica (2019.0) 99 929-930. DOI: 10.2340/00015555-3234 124. Ortega-Ortega M, Hanly P, Pearce A, Soerjomataram I, Sharp L. Paid and unpaid productivity losses due to premature mortality from cancer in Europe in 2018. International Journal of Cancer (2021.0) 150 580-593. DOI: 10.1002/ijc.33826 125. Yang JJ, Maloney NJ, Cheng K, Bach DQ. Financial burden in US patients with melanoma from 1997 to 2015: racial disparities, trends, and predictors of high expenditures. JAMA Dermatology (2021.0) 84 819-821. DOI: 10.1016/j.jaad.2020.07.051 126. Noels E, Hollestein L, Luijkx K, Louwman M, de Uyl-de Groot C, van den Bos R, van der Veldt A, Grunhagen D, Wakkee M. Increasing costs of skin cancer due to increasing incidence and introduction of pharmaceuticals, 2007–2017. Acta Dermato-Venereologica (2020.0) 100 adv00147. DOI: 10.2340/00015555-3463 127. Grange F, Mohr P, Harries M, Ehness R, Benjamin L, Siakpere O, Barth J, Stapelkamp C, Pfersch S, McLeod LD, Kaye JA, Wolowacz S, Kontoudis I. Economic burden of advanced melanoma in France, Germany and the UK: a retrospective observational study (Melanoma Burden-of-Illness Study). Melanoma Research (2017.0) 27 607-618. DOI: 10.1097/cmr.0000000000000372 128. Fernandes J, Bregman B, Combemale P, Amaz C, de Leotoing L, Vainchtock A, Gaudin AF. Hospitalisation costs of metastatic melanoma in France; the MELISSA study (MELanoma In hoSpital coSts Assessment). BMC Health Services Research (2017.0) 17 542. DOI: 10.1186/s12913-017-2472-0 129. Copley-Merriman C, Stevinson K, Liu FX, Wang J, Mauskopf J, Zimovetz EA, Chmielowski B. Direct costs associated with adverse events of systemic therapies for advanced melanoma: systematic literature review. Medicine (Baltimore) (2018.0) 97 e11736. DOI: 10.1097/md.0000000000011736 130. Krensel M, Schafer I, Augustin M. Cost-of-illness of melanoma in Europe—a modelling approach. Journal of the European Academy of Dermatology and Venereology (2019.0) 33 34-45. DOI: 10.1111/jdv.15308 131. Gordon LG, Leung W, Johns R, McNoe B, Lindsay D, Merollini KMD, Elliott TM, Neale RE, Olsen CM, Pandeya N, Whiteman DC. Estimated healthcare costs of melanoma and keratinocyte skin cancers in Australia and Aotearoa New Zealand in 2021. International Journal of Environmental Research and Public Health (2022.0) 19 3178. DOI: 10.3390/ijerph19063178 132. Luke JJ, Ascierto PA, Carlino MS, Gershenwald JE, Grob J-J, Hauschild A, Kirkwood JM, Long G, Mohr P, Robert C, Ross M, Scolyer RA, Yoon CH, Poklepovic A, Rutkowski P, Anderson JR, Ahsan S, Ibrahim N, Eggermont AMM. KEYNOTE-716: Phase III study of adjuvant pembrolizumab versus placebo in resected high-risk stage II melanoma. Future Oncology (2020.0) 16 4429-4438. DOI: 10.2217/fon-2019-0666 133. Anderson A, Matsumoto M, Secrest A, Saul MI, Ho J, Ferris LK. Cost of treatment of benign and premalignant lesions during skin cancer screening. JAMA Dermatology (2021.0) 157 876-879. DOI: 10.1001/jamadermatol.2021.1953 134. Gordon L, Olsen C, Whiteman DC, Elliott TM, Janda M, Green A. Prevention versus early detection for long-term control of melanoma and keratinocyte carcinomas: a cost-effectiveness modelling study. BMJ Open (2020.0) 10 e034388. DOI: 10.1136/bmjopen-2019-034388 135. van der Pols JC, Williams GM, Neale RE, Clavarino A, Green AC. Long-term increase in sunscreen use in an Australian community after a skin cancer prevention trial. Preventive Medicine (2006.0) 42 171-176. DOI: 10.1016/j.ypmed.2005.10.007 136. Janda M, Youl P, Neale R, Aitken J, Whiteman D, Gordon L, Baade P. Clinical skin examination outcomes after a video-based behavioral intervention: analysis from a randomized clinical trial. JAMA Dermatology (2014.0) 150 372-379. DOI: 10.1001/jamadermatol.2013.9313 137. Mofidi A, Tompa E, Spencer J, Kalcevich C, Peters CE, Kim J, Song C, Mortazavi SB, Demers PA. The economic burden of occupational non-melanoma skin cancer due to solar radiation. Journal of Occupational and Environmental Hygiene (2018.0) 15 481-491. DOI: 10.1080/15459624.2018.1447118 138. Mofidi A, Tompa E, Song C, Tenkate T, Arrandale V, Jardine JK, Davies H, Demers PA. Economic evaluation of interventions to reduce solar ultraviolet radiation (UVR) exposure among construction workers. Journal of Occupational and Environmental Hygiene (2021.0) 18 250-264. DOI: 10.1080/15459624.2021.1910278 139. Yoon J, Phibbs CS, Chow A, Weinstock MA. Impact of topical fluorouracil cream on costs of treating keratinocyte carcinoma (nonmelanoma skin cancer) and actinic keratosis. [Article]. JAMA Dermatology (2018.0) 79 501-507.e502. DOI: 10.1016/j.jaad.2018.02.058 140. Fitzpatrick TB. The validity and practicality of sun-reactive skin types I through VI. Archives of Dermatology (1988.0) 124 869-871. DOI: 10.1001/archderm.124.6.869 141. Olsen CM, Pandeya N, Law MH, MacGregor S, Iles MM, Thompson BS, Green AC, Neale RE, Whiteman DC. Does polygenic risk influence associations between sun exposure and melanoma? A prospective cohort analysis. British Journal of Dermatology (2020.0) 183 303-310. DOI: 10.1111/bjd.18703 142. Savoye I, Olsen CM, Whiteman DC, Bijon A, Wald L, Dartois L, Clavel-Chapelon F, Boutron-Ruault M-C, Kvaskoff M. Patterns of ultraviolet radiation exposure and skin cancer risk: the E3N-SunExp Study. Journal of Epidemiology (2018.0) 28 27-33. DOI: 10.2188/jea.JE20160166 143. Kawai K, VoPham T, Drucker A, Curhan SG, Curhan GC. Ultraviolet radiation exposure and the risk of Herpes zoster in three prospective cohort studies. Mayo Clinic Proceedings (2020.0) 95 283-292. DOI: 10.1016/j.mayocp.2019.08.022 144. Tripathi R, Mazmudar RS, Knusel KD, Ezaldein HH, Bordeaux JS, Scott JF. Trends in emergency department visits due to sunburn and factors associated with severe sunburns in the United States. Archives of Dermatological Research (2021.0) 313 79-88. DOI: 10.1007/s00403-020-02073-2 145. Holman DM, Ding H, Berkowitz Z, Hartman AM, Perna FM. Sunburn prevalence among US adults, National Health Interview Survey 2005, 2010, and 2015. JAMA Dermatology (2019.0) 80 817-820. DOI: 10.1016/j.jaad.2018.10.044 146. Wei G, Farooq J, Castelo-Soccio L, Mhaskar R. Correlates between physical activity and sunburn prevalence among a nationally representative sample of US high school students, 2015–2017. Journal of Physical Activity and Health (2021.0) 18 1113-1119. DOI: 10.1123/jpah.2020-0711 147. Blazquez-Sanchez N, Rivas-Ruiz F, Bueno-Fernandez S, Fernandez-Morano MT, Arias-Santiago S, Rodriguez-Martinez A, DeCastro-Maqueda G, DeTroya-Martin M. Photoprotection habits, attitudes and knowledge among school communities in the Costa del sol (Spain). European Journal of Public Health (2021.0) 31 508-514. DOI: 10.1093/eurpub/ckab010 148. Goerig T, Soedel C, Pfahlberg AB, Gefeller O, Breitbart EW, Diehl K. Sun protection and sunburn in children aged 1–10 years in Germany: prevalence and determinants. Children (Basel) (2021.0) 8 668. DOI: 10.3390/children8080668 149. Tabbakh T, Volkov A, Wakefield M, Dobbinson S. Implementation of the SunSmart program and population sun protection behaviour in Melbourne, Australia: results from cross-sectional summer surveys from 1987 to 2017. PLoS Medicine (2019.0) 16 e1002932-e1002932. DOI: 10.1371/journal.pmed.1002932 150. Liew AYS, Cust AE. Changes in sun protection behaviours, sun exposure and shade availability among adults, children and adolescents in New South Wales, 2003–2016. Australian and New Zealand Journal of Public Health (2021.0) 45 462-468. DOI: 10.1111/1753-6405.13112 151. 151.Cancer Australia (2019). National cancer control indicators: sunburn and sun protection. https://www.ncci.canceraustralia.gov.au/prevention/sun-exposure/sunburn-and-sun-protection. Accessed 30 Mar 2022 152. Køster B, Meyer M, Andersson T, Engholm G, Dalum P. Development in sunburn 2007–2015 and skin cancer projections 2007–2040 of campaign results in the Danish population. Medicine (Baltimore) (2018.0) 97 e12738-e12738. DOI: 10.1097/MD.0000000000012738 153. Lacson JCA, Zamani SA, Froes LAR, Mitra N, Qian L, Doyle SH, Azizi E, Balestrini C, Bishop DT, Bruno W, Carlos-Ortega B, Cuellar F, Cust AE, Elder DE, Gerdes A-M, Ghiorzo P, Grazziotin TC, Gruis NA, Hansson J, Hocevar M. Birth cohort-specific trends of sun-related behaviors among individuals from an international consortium of melanoma-prone families. BMC Public Health (2021.0) 21 692-692. DOI: 10.1186/s12889-021-10424-5 154. Shih BB, Allan D, de Gruijl FR, Rhodes LE. Robust detection of minimal sunburn in pigmented skin by 785 nm laser speckle contrast imaging of blood flux. Journal of Investigative Dermatology (2015.0) 135 1197-1199. DOI: 10.1038/jid.2014.507 155. Calderón TA, Bleakley A, Jordan AB, Lazovich D, Glanz K. Correlates of sun protection behaviors in racially and ethnically diverse US adults. Preventive Medicine Reports (2019.0) 13 346-353. DOI: 10.1016/j.pmedr.2018.12.006 156. Bello O, Sudhoff H, Goon P. Sunburn prevalence is underestimated in UK-based people of African ancestry. Clinical, Cosmetic and Investigational Dermatology (2021.0) 14 1791-1797. DOI: 10.2147/CCID.S334574 157. Rhodes LE, Bock M, Soe Janssens A, Ling TC, Anastasopoulou L, Antoniou C, Aubin F, Bruckner T, Faivre B, Gibbs NK, Jansen C, Pavel S, Stratigos AJ, de Gruijl FR, Diepgen TL. Polymorphic light eruption occurs in 18% of Europeans and does not show higher prevalence with increasing latitude: Multicenter survey of 6,895 individuals residing from the Mediterranean to Scandinavia. Journal of Investigative Dermatology (2010.0) 130 626-628. DOI: 10.1038/jid.2009.250 158. Gutierrez D, Gaulding JV, Motta Beltran AF, Lim HW, Pritchett EN. Photodermatoses in skin of colour. Journal of the European Academy of Dermatology and Venereology (2018.0) 32 1879-1886. DOI: 10.1111/jdv.15115 159. Gether L, Overgaard LK, Egeberg A, Thyssen JP. Incidence and prevalence of rosacea: a systematic review and meta-analysis. British Journal of Dermatology (2018.0) 179 282-289. DOI: 10.1111/bjd.16481 160. Rutter KJ, Ashraf I, Cordingley L, Rhodes LE. Quality of life and psychological impact in the photodermatoses: a systematic review. British Journal of Dermatology (2020.0) 182 1092-1102. DOI: 10.1111/bjd.18326 161. Hofmann GA, Gradl G, Schulz M, Haidinger G, Tanew A, Weber B. The frequency of photosensitizing drug dispensings in Austria and Germany: a correlation with their photosensitizing potential based on published literature. Journal of the European Academy of Dermatology and Venereology (2020.0) 34 589-600. DOI: 10.1111/jdv.15952 162. Nakao S, Hatahira H, Sasaoka S, Hasegawa S, Motooka Y, Ueda N, Abe J, Fukuda A, Naganuma M, Kanoh H, Seishima M, Ishiguro M, Kinosada Y, Nakamura M. Evaluation of drug-induced photosensitivity using the Japanese adverse drug event report (JADER) Database. Biological and Pharmaceutical Bulletin (2017.0) 40 2158-2165. DOI: 10.1248/bpb.b17-00561 163. Kim WB, Shelley AJ, Novice K, Joo J, Lim HW, Glassman SJ. Drug-induced phototoxicity: a systematic review. JAMA Dermatology (2018.0) 79 1069-1075. DOI: 10.1016/j.jaad.2018.06.061 164. Alrashidi A, Rhodes LE, Sharif JCH, Kreeshan FC, Farrar MD, Ahad T. Systemic drug photosensitivity—culprits, impact and investigation in 122 patients. Photodermatology, Photoimmunology & Photomedicine (2020.0) 36 441-451. DOI: 10.1111/phpp.12583 165. Pedersen SA, Gaist D, Schmidt SAJ, Holmich LR, Friis S, Pottegard A. Hydrochlorothiazide use and risk of nonmelanoma skin cancer: a nationwide case-control study from Denmark. JAMA Dermatology (2018.0) 78 673-681 e679. DOI: 10.1016/j.jaad.2017.11.042 166. 166.Pharmacovigilance Risk Assessment Committee PRAC recommendations on signals, European Medicines Agency 2018. https://www.ema.europa.eu/en/documents/prac-recommendation/prac-recommendations-signals-adopted-3-6-september-2018-prac-meeting_en-0.pdf. Accessed 30 Mar 2022 167. Pottegård A, Pedersen SA, Schmidt SAJ, Lee C-N, Hsu C-K, Liao T-C, Shao S-C, Lai EC-C. Use of hydrochlorothiazide and risk of skin cancer: a nationwide Taiwanese case-control study. British Journal of Cancer (2019.0) 121 973-978. DOI: 10.1038/s41416-019-0613-4 168. Gallelli L, Cione E, Siniscalchi A, Vasta G, Guerra A, Scaramuzzino A, Longo L, Muraca L, De Sarro G, Leuzzi G, Gerace A, Scuteri A, Vasapollo P, Natale V, Zampogna S, Luciani F. Is there a link between non-melanoma skin cancer and hydrochlorothiazide?. Current Drug Safety (2021.0) 17 211-216. DOI: 10.2174/1574886316666211103164412 169. Daniels B, Pearson SA, Vajdic CM, Pottegård A, Buckley NA, Zoega H. Risk of squamous cell carcinoma of the lip and cutaneous melanoma in older Australians using hydrochlorothiazide: a population-based case-control study. Basic and Clinical Pharmacology and Toxicology (2020.0) 127 320-328. DOI: 10.1111/bcpt.13463 170. Bendinelli B, Masala G, Garamella G, Palli D, Caini S. Do thiazide diuretics increase the risk of skin cancer? A critical review of the scientific evidence and updated meta-analysis. Current Cardiology Reports (2019.0) 21 92. DOI: 10.1007/s11886-019-1183-z 171. Williams NM, Vincent LT, Rodriguez GA, Nouri K. Antihypertensives and melanoma: an updated review. Pigment Cell & Melanoma Research (2020.0) 33 806-813. DOI: 10.1111/pcmr.12918 172. Hatsusaka N, Seki Y, Mita N, Ukai Y, Miyashita H, Kubo E, Sliney D, Sasaki H. UV Index does not predict ocular ultraviolet exposure. Translational Vision Science & Technology (2021.0) 10 1. DOI: 10.1167/tvst.10.7.1 173. Marro M, Moccozet L, Vernez D. Assessing human eye exposure to UV light: a narrative review. Frontiers in Public Health (2022.0) 10 900979-900979. DOI: 10.3389/fpubh.2022.900979 174. Lanca C, Teo A, Vivagandan A, Htoon HM, Najjar RP, Spiegel DP, Pu S-H, Saw S-M. The effects of different outdoor environments, sunglasses and hats on light levels: implications for myopia prevention. Translational Vision Science & Technology (2019.0) 8 7-7. DOI: 10.1167/tvst.8.4.7 175. Backes C, Religi A, Moccozet L, Behar-Cohen F, Vuilleumier L, Bulliard JL, Vernez D. Sun exposure to the eyes: predicted UV protection effectiveness of various sunglasses. Journal of Exposure Science & Environmental Epidemiology (2019.0) 29 753-764. DOI: 10.1038/s41370-018-0087-0 176. Rabbetts R, Sliney D. Technical report: solar ultraviolet protection from sunglasses. Optometry and Vision Science (2019.0) 96 523-530. DOI: 10.1097/OPX.0000000000001397 177. Lucas RM, Norval M, Neale RE, Young AR, de Gruijl FR, Takizawa Y, van der Leun JC. The consequences for human health of stratospheric ozone depletion in association with other environmental factors. Photochemical & Photobiological Sciences (2014.0) 14 53-87. DOI: 10.1039/c4pp90033b 178. Lofgren S. Solar ultraviolet radiation cataract. Experimental Eye Research (2017.0) 156 112-116. DOI: 10.1016/j.exer.2016.05.026 179. Cheng C-Y, Wang N, Wong TY, Congdon N, He M, Wang YX, Braithwaite T, Casson RJ, Cicinelli MV, Das A, Flaxman SR, Jonas JB, Keeffe JE, Kempen JH, Leasher J, Limburg H, Naidoo K, Pesudovs K, Resnikoff S, Silvester AJ. Prevalence and causes of vision loss in East Asia in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology (2020.0) 104 616-622. DOI: 10.1136/bjophthalmol-2018-313308 180. Kahloun R, Khairallah M, Resnikoff S, Cicinelli MV, Flaxman SR, Das A, Jonas JB, Keeffe JE, Kempen JH, Leasher J, Limburg H, Naidoo K, Pesudovs K, Silvester AJ, Tahhan N, Taylor HR, Wong TY, Bourne RRA. Prevalence and causes of vision loss in North Africa and Middle East in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology (2019.0) 103 863-870. DOI: 10.1136/bjophthalmol-2018-312068 181. Keeffe JE, Casson RJ, Pesudovs K, Taylor HR, Cicinelli MV, Das A, Flaxman SR, Jonas JB, Kempen JH, Leasher J, Limburg H, Naidoo K, Silvester AJ, Stevens GA, Tahhan N, Wong TY, Resnikoff S, Bourne RRA. Prevalence and causes of vision loss in South-east Asia and Oceania in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology (2019.0) 103 878-884. DOI: 10.1136/bjophthalmol-2018-311946 182. Leasher JL, Braithwaite T, Furtado JM, Flaxman SR, Lansingh VC, Silva JC, Resnikoff S, Taylor HR, Bourne RRA. Prevalence and causes of vision loss in Latin America and the Caribbean in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology (2019.0) 103 885-893. DOI: 10.1136/bjophthalmol-2017-311746 183. Naidoo K, Kempen JH, Gichuhi S, Braithwaite T, Casson RJ, Cicinelli MV, Das A, Flaxman SR, Jonas JB, Keeffe JE, Leasher J, Limburg H, Pesudovs K, Resnikoff S, Silvester AJ, Tahhan N, Taylor HR, Wong TY, Bourne RRA. Prevalence and causes of vision loss in sub-Saharan Africa in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology (2020.0) 104 1658-1668. DOI: 10.1136/bjophthalmol-2019-315217 184. Foreman J, Keel S, McGuiness M, Liew D, Wijngaarden P, Taylor HR, Dirani M. Future burden of vision loss in Australia: projections from the national eye health survey. Clinical & Experimental Ophthalmology (2020.0) 48 730-738. DOI: 10.1111/ceo.13776 185. 185.GBD results tool Available from: https://ghdx.healthdata.org/gbd-results-tool. Accessed 11 Apr 22 186. Bikbov MM, Kazakbaeva GM, Gilmanshin TR, Zainullin RM, Nuriev IF, Zaynetdinov AF, Israfilova GZ, Panda-Jonas S, Arslangareeva II, Rakhimova EM, Rusakova IA, Jonas JB. Prevalence and associated factors of cataract and cataract-related blindness in the Russian Ural Eye and Medical Study. Scientific Reports (2020.0) 10 18157. DOI: 10.1038/s41598-020-75313-0 187. Purola PKM, Nättinen JE, Ojamo MUI, Koskinen SVP, Rissanen HA, Sainio PRJ, Uusitalo HMT. Prevalence and 11-year incidence of common eye diseases and their relation to health-related quality of life, mental health, and visual impairment. Quality of Life Research (2021.0) 30 2311-2327. DOI: 10.1007/s11136-021-02817-1 188. Tan AG, Tham YC, Chee ML, Mitchell P, Cumming RG, Sabanayagam C, Wong TY, Wang JJ, Cheng CY. Incidence, progression and risk factors of age-related cataract in Malays: the Singapore Malay Eye Study. Clinical & Experimental Ophthalmology (2020.0) 48 580-592. DOI: 10.1111/ceo.13757 189. Vashist P, Tandon R, Murthy GVS, Barua CK, Deka D, Singh S, Gupta V, Gupta N, Wadhwani M, Singh R, Vishwanath K. Association of cataract and sun exposure in geographically diverse populations of India: The CASE study. First Report of the ICMR-EYE SEE Study Group. PLoS One (2020.0) 15 e0227868. DOI: 10.1371/journal.pone.0227868 190. Lee J, Kim UJ, Lee Y, Han E, Ham S, Lee W, Choi WJ, Kang SK. Sunlight exposure and eye disorders in an economically active population: data from the KNHANES 2008–2012. Annals of Occupational and Environmental Medicine (2021.0) 33 e24. DOI: 10.35371/aoem.2021.33.e24 191. Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: the Right to Sight: an analysis for the Global Burden of Disease Study. The Lancet Global Health (2021.0) 9 e144-e160. DOI: 10.1016/S2214-109X(20)30489-7 192. Van Acker SI, van den Bogerd B, Haagdorens M, Siozopoulou V, Ni Dhubhghaill S, Pintelon I, Koppen C. Pterygium—the good, the bad, and the ugly. Cells (Basel) (2021.0) 10 1567. DOI: 10.3390/cells10071567 193. Rezvan F, Khabazkhoob M, Hooshmand E, Yekta A, Saatchi M, Hashemi H. Prevalence and risk factors of pterygium: a systematic review and meta-analysis. Survey of Ophthalmology (2018.0) 63 719-735. DOI: 10.1016/j.survophthal.2018.03.001 194. Liu L, Wu J, Geng J, Yuan Z, Huang D. Geographical prevalence and risk factors for pterygium: a systematic review and meta-analysis. BMJ Open (2013.0) 3 e003787. DOI: 10.1136/bmjopen-2013-003787 195. Fernandes AG, Salomão SR, Ferraz NN, Mitsuhiro MH, Furtado JM, Muñoz S, Cypel MC, Cunha CC, Vasconcelos GC, Sacai PY, Morales PH, Cohen MJ, Cohen JM, Watanabe SS, Campos M, Belfort Junior R, Berezovsky A. Pterygium in adults from the Brazilian Amazon Region: prevalence, visual status and refractive errors. British Journal of Ophthalmology (2020.0) 104 757-763. DOI: 10.1136/bjophthalmol-2019-314131 196. Pan Z, Cui J, Shan G, Chou Y, Pan L, Sun Z, Cui Z, Sun J, Cao Y, Zhao J, Ma X, Ma J, He H, Ma J, Zhong Y. Prevalence and risk factors for pterygium: a cross-sectional study in Han and Manchu ethnic populations in Hebei China. BMJ Open (2019.0) 9 e025725. DOI: 10.1136/bmjopen-2018-025725 197. Wang Y, Shan G, Gan L, Qian Y, Chen T, Wang H, Pan X, Wang W, Pan L, Zhang X, Wang M, Ma J, Zhong Y. Prevalence and associated factors for pterygium in Han and Mongolian adults: a cross-sectional study in inner Mongolian, China. BMC Ophthalmology (2020.0) 20 45. DOI: 10.1186/s12886-020-1324-6 198. Hampel U, Wasielica-Poslednik J, Ries L, Faysal R, Schulz A, Nickels S, Wild PS, Schmidtmann I, Münzel T, Beutel ME, Lackner KJ, Pfeiffer N, Schuster AK. Prevalence of pterygium and identification of associated factors in a German population—results from the Gutenberg Health Study. Acta Ophthalmologica (2021.0) 99 e130-e131. DOI: 10.1111/aos.14505 199. Bikbov MM, Zainullin RM, Kazakbaeva GM, Gilmanshin TR, Salavatova VF, Arslangareeva II, Nikitin NA, Panda-Jonas S, Zaynetdinov AF, Kazakbaev RA, Nuriev IF, Khikmatullin RI, Uzianbaeva YV, Yakupova DF, Aminev SK, Jonas JB. Pterygium prevalence and Its associations in a Russian population: the ural eye and medical study. American Journal of Ophthalmology (2019.0) 205 27-34. DOI: 10.1016/j.ajo.2019.02.031 200. Cameron ME. Pterygium Throughout the World (1965.0) 201. Khanna RC, Marmamula S, Cicinelli MV, Mettla AL, Giridhar P, Banerjee S, Shekhar K, Chakrabarti S, Murthy GVS, Gilbert CE, Rao GN. Fifteen-year incidence rate and risk factors of pterygium in the Southern Indian state of Andhra Pradesh. British Journal of Ophthalmology (2021.0) 105 619-624. DOI: 10.1136/bjophthalmol-2020-316359 202. Fang XL, Chong CCY, Thakur S, Da Soh Z, Teo ZL, Majithia S, Lim ZW, Rim TH, Sabanayagam C, Wong TY, Cheng C-Y, Tham Y-C. Ethnic differences in the incidence of pterygium in a multi-ethnic Asian population: the Singapore Epidemiology of Eye Diseases Study. Scientific Reports (2021.0) 11 501. DOI: 10.1038/s41598-020-79920-9 203. Lingham G, Kugelman J, Charng J, Lee SS, Yazar S, McKnight CM, Coroneo MT, Lucas RM, Brown H, Stevenson LJ, Mackey DA, Alonso-Caneiro D. Conjunctival ultraviolet autofluorescence area decreases with age and sunglasses use. British Journal of Ophthalmology (2021.0). DOI: 10.1136/bjophthalmol-2021-320284 204. Goh AY, Ramlogan-Steel CA, Jenkins KS, Steel JC, Layton CJ. Presence and prevalence of UV related genetic mutations in uveal melanoma: similarities with cutaneous melanoma. Neoplasma (2020.0) 67 958-971. DOI: 10.4149/neo_2020_190815N768 205. Mundra PA, Dhomen N, Rodrigues M, Mikkelsen LH, Cassoux N, Brooks K, Valpione S, Reis-Filho JS, Heegaard S, Stern M-H, Roman-Roman S, Marais R. Ultraviolet radiation drives mutations in a subset of mucosal melanomas. Nature Communications (2021.0) 12 259. DOI: 10.1038/s41467-020-20432-5 206. Ghazawi FM, Darwich R, Le M, Rahme E, Zubarev A, Moreau L, Burnier JV, Sasseville D, Burnier MN, Litvinov IV. Uveal melanoma incidence trends in Canada: a national comprehensive population-based study. British Journal of Ophthalmology (2019.0). DOI: 10.1136/bjophthalmol-2018-312966 207. Baily C, O'Neill V, Dunne M, Cunningham M, Gullo G, Kennedy S, Walsh PM, Deady S, Horgan N. Uveal melanoma in Ireland. Ocular Oncology and Pathology (2019.0) 5 195-204. DOI: 10.1159/000492391 208. Aronow ME, Topham AK, Singh AD. Uveal melanoma: 5-Year update on incidence, treatment, and survival (SEER 1973–2013). Ocular Oncology and Pathology (2018.0) 4 145-151. DOI: 10.1159/000480640 209. Beasley AB, Preen DB, McLenachan S, Gray ES, Chen FK. Incidence and mortality of uveal melanoma in Australia (1982–2014). British Journal of Ophthalmology (2021.0). DOI: 10.1136/bjophthalmol-2021-319700 210. Ghazawi FM, Darwich R, Le M, Jfri A, Rahme E, Burnier JV, Sasseville D, Burnier MN, Litvinov IV. Incidence trends of conjunctival malignant melanoma in Canada. British Journal of Ophthalmology (2020.0) 104 23-25. DOI: 10.1136/bjophthalmol-2019-313977 211. Virgili G, Parravano M, Gatta G, Capocaccia R, Mazzini C, Mallone S, Botta L. Incidence and survival of patients with conjunctival melanoma in Europe. JAMA Ophthalmology (2020.0) 138 601-608. DOI: 10.1001/jamaophthalmol.2020.0531 212. Roberts JE. Screening for ocular phototoxicity. International Journal of Toxicology (2002.0) 21 491-500. DOI: 10.1080/10915810290169918 213. Zhao B, Chignell CF, Rammal M, Smith F, Hamilton MG, Andley UP, Roberts JE. Detection and prevention of ocular phototoxicity of ciprofloxacin and other fluoroquinolone antibiotics. Photochemistry and Photobiology (2010.0) 86 798-805. DOI: 10.1111/j.1751-1097.2010.00755.x 214. Sahu RK, Singh B, Saraf SA, Kaithwas G, Kishor K. Photochemical toxicity of drugs intended for ocular use. Archives of Industrial Hygiene and Toxicology (2014.0) 65 157-167. DOI: 10.2478/10004-1254-65-2014-2461 215. Wright CY, Lucas RM, D'Este C, Kapwata T, Kunene Z, Swaminathan A, Mathee A, Albers PN. Effect of a sun protection intervention on the immune response to measles booster vaccination in infants in rural South Africa. Photochemistry and Photobiology (2019.0) 95 446-452. DOI: 10.1111/php.13004 216. Swaminathan A, Harrison SL, Ketheesan N, van den Boogaard CHA, Dear K, Allen M, Hart PH, Cook M, Lucas RM. Exposure to solar UVR suppresses cell-mediated immunization responses in humans: the Australian Ultraviolet Radiation and Immunity Study. Journal of Investigative Dermatology (2019.0) 139 1545-1553.e1546. DOI: 10.1016/j.jid.2018.12.025 217. LopatkoLindman K, Lockman-Lundgren J, Weidung B, Olsson J, Elgh F, Lövheim H. Long-term time trends in reactivated herpes simplex infections and treatment in Sweden. BMC Infectious Diseases (2022.0) 22 547-547. DOI: 10.1186/s12879-022-07525-w 218. Mai Z-M, Lin J-H, Ngan RK-C, Kwong DL-W, Ng W-T, Ng AW-Y, Ip K-M, Chan Y-H, Lee AW-M, Ho S-Y, Lung ML, Lam T-H. Solar ultraviolet radiation and vitamin D deficiency on Epstein-Barr Virus reactivation: Observational and genetic evidence from a nasopharyngeal carcinoma-endemic population. Open Forum Infectious Diseases (2020.0) 7 ofaa426. DOI: 10.1093/ofid/ofaa426 219. Bar-Or A, Pender MP, Khanna R, Steinman L, Hartung HP, Maniar T, Croze E, Aftab BT, Giovannoni G, Joshi MA. Epstein-Barr virus in multiple sclerosis: theory and emerging immunotherapies. Trends in Molecular Medicine (2020.0) 26 296-310. DOI: 10.1016/j.molmed.2019.11.003 220. Simpson S, van der Mei I, Lucas RM, Ponsonby A-L, Broadley S, Blizzard L, Taylor B. Sun exposure across the life course significantly modulates early multiple sclerosis clinical course. Frontiers in Neurology (2018.0) 9 16. DOI: 10.3389/fneur.2018.00016 221. Bakker KM, Eisenberg MC, Woods R, Martinez ME. Exploring the seasonal drivers of Varicella zoster virus transmission and reactivation. American Journal of Epidemiology (2021.0) 190 1814-1820. DOI: 10.1093/aje/kwab073 222. Lucas RM, Rodney Harris R. On the nature of evidence and ‘proving’ causality: smoking and lung cancer vs. sun exposure, vitamin D and multiple sclerosis. International Journal of Environmental Research and Public Health (2018.0) 15 1726. DOI: 10.3390/ijerph15081726 223. Chen R, Yang C, Li P, Wang J, Liang Z, Wang W, Wang Y, Liang C, Meng R, Wang HY, Peng S, Sun X, Su Z, Kong G, Wang Y, Zhang L. Long-term exposure to ambient PM2.5, sunlight, and obesity: a nationwide study in China. Frontiers in Endocrinology (2021.0) 12 790294. DOI: 10.3389/fendo.2021.790294 224. Weller RB, Wang Y, He J, Maddux FW, Usvyat L, Zhang H, Feelisch M, Kotanko P. Does incident solar ultraviolet radiation lower blood pressure?. Journal of the American Heart Association (2020.0) 9 e013837. DOI: 10.1161/JAHA.119.013837 225. Luo YN, Yang BY, Zou Z, Markevych I, Browning M, Heinrich J, Bao WW, Guo Y, Hu LW, Chen G, Ma J, Ma Y, Chen YJ, Dong GH. Associations of greenness surrounding schools with blood pressure and hypertension: a nationwide cross-sectional study of 61,229 children and adolescents in China. Environmental Research (2022.0) 204 112004. DOI: 10.1016/j.envres.2021.112004 226. Twohig-Bennett C, Jones A. The health benefits of the great outdoors: a systematic review and meta-analysis of greenspace exposure and health outcomes. Environmental Research (2018.0) 166 628-637. DOI: 10.1016/j.envres.2018.06.030 227. Cui Y, Gong Q, Huang C, Guo F, Li W, Wang Y, Cheng X. The relationship between sunlight exposure duration and depressive symptoms: a cross-sectional study on elderly Chinese women. PLoS One (2021.0) 16 e0254856. DOI: 10.1371/journal.pone.0254856 228. Lindqvist PG, Landin-Olsson M, Olsson H. Low sun exposure habits is associated with a dose-dependent increased risk of hypertension: a report from the large MISS cohort. Photochemical & Photobiological Sciences (2021.0) 20 285-292. DOI: 10.1007/s43630-021-00017-x 229. Bae JM, Kim Y-S, Choo EH, Kim M-Y, Lee JY, Kim H-O, Park YM. Both cardiovascular and cerebrovascular events are decreased following long-term narrowband ultraviolet B phototherapy in patients with vitiligo: a propensity score matching analysis. Journal of the European Academy of Dermatology and Venereology (2021.0) 35 11-12. DOI: 10.1111/jdv.17073 230. Mackay DF, Clemens TL, Hastie CE, Cherrie MPC, Dibben C, Pell JP. UVA and seasonal patterning of 56 370 myocardial infarctions across Scotland, 2000–2011. Journal of the American Heart Association (2019.0) 8 e012551. DOI: 10.1161/JAHA.119.012551 231. Liddle L, Monaghan C, Burleigh MC, Baczynska KA, Muggeridge DJ, Easton C. Reduced nitric oxide synthesis in winter: a potential contributing factor to increased cardiovascular risk. Nitric Oxide (2022.0) 127 1-9. DOI: 10.1016/j.niox.2022.06.007 232. Weller RB, Macintyre IM, Melville V, Farrugia M, Feelisch M, Webb DJ. The effect of daily UVA phototherapy for 2 weeks on clinic and 24-h blood pressure in individuals with mild hypertension. Journal of Human Hypertension (2022.0). DOI: 10.1038/s41371-022-00729-2 233. Huang L, Schmid KL, Yin XN, Zhang J, Wu J, Yang G, Ruan ZL, Jiang XQ, Wu CA, Chen WQ. Combination effect of outdoor activity and screen exposure on risk of preschool myopia: findings from Longhua Child Cohort Study. Frontiers in Public Health (2021.0) 9 607911. DOI: 10.3389/fpubh.2021.607911 234. Enthoven CA, Tideman JWL, Polling JR, Yang-Huang J, Raat H, Klaver CCW. The impact of computer use on myopia development in childhood: the Generation R study. Preventive Medicine (2020.0) 132 105988. DOI: 10.1016/j.ypmed.2020.105988 235. Lingham G, Yazar S, Lucas RM, Milne E, Hewitt AW, Hammond CJ, MacGregor S, Rose KA, Chen FK, He M, Guggenheim JA, Clarke MW, Saw SM, Williams C, Coroneo MT, Straker L, Mackey DA. Time spent outdoors in childhood is associated with reduced risk of myopia as an adult. Scientific Reports (2021.0) 11 6337. DOI: 10.1038/s41598-021-85825-y 236. Lee SS-Y, Lingham G, Sanfilippo PG, Hammond CJ, Saw S-M, Guggenheim JA, Yazar S, Mackey DA. Incidence and progression of myopia in early adulthood. JAMA Ophthalmology (2022.0) 140 162-169. DOI: 10.1001/jamaophthalmol.2021.5067 237. Li M, Lanca C, Tan CS, Foo LL, Sun CH, Yap F, Najjar RP, Sabanayagam C, Saw SM. Association of time outdoors and patterns of light exposure with myopia in children. British Journal of Ophthalmology (2021.0). DOI: 10.1136/bjophthalmol-2021-318918 238. Yang X, Yang Y, Wang Y, Wei Q, Ding H, Zhong X. Protective effects of sunlight exposure against PRK-induced myopia in infant rhesus monkeys. Ophthalmic and Physiological Optics (2021.0) 41 911-921. DOI: 10.1111/opo.12826 239. Lingham G, Mackey DA, Lucas R, Yazar S. How does spending time outdoors protect against myopia? A review. British Journal of Ophthalmology (2020.0) 104 593-599. DOI: 10.1136/bjophthalmol-2019-314675 240. Langer-Gould A, Lucas R, Xiang AH, Chen LH, Wu J, Gonzalez E, Haraszti S, Smith JB, Quach H, Barcellos LF. MS Sunshine Study: Sun exposure but not vitamin D is associated with multiple sclerosis risk in Blacks and Hispanics. Nutrients (2018.0) 10 268. DOI: 10.3390/nu10030268 241. Magalhaes S, Pugliatti M, Riise T, Myhr KM, Ciampi A, Bjornevik K, Wolfson C. Shedding light on the link between early life sun exposure and risk of multiple sclerosis: results from the EnvIMS Study. International Journal of Epidemiology (2019.0) 48 1073-1082. DOI: 10.1093/ije/dyy269 242. Sebastian P, Cherbuin N, Barcellos LF, Roalstad S, Casper C, Hart J, Aaen GS, Krupp L, Benson L, Gorman M, Candee M, Chitnis T, Goyal M, Greenberg B, Mar S, Rodriguez M, Rubin J, Schreiner T, Waldman A, Weinstock-Guttman B. Association between time spent outdoors and risk of multiple sclerosis. Neurology (2022.0) 98 e267-e278. DOI: 10.1212/WNL.0000000000013045 243. Ostkamp P, Salmen A, Pignolet B, Goerlich D, Andlauer TFM, Schulte-Mecklenbeck A, Gonzalez-Escamilla G, Bucciarelli F, Gennero I, Breuer J, Antony G, Schneider-Hohendorf T, Mykicki N, Bayas A, Bergh FT, Bittner S, Hartung H-P, Friese MA, Linker RA, Luessi F. Sunlight exposure exerts immunomodulatory effects to reduce multiple sclerosis severity. Proceedings of the National Academy of Sciences of the United States of America (2021.0) 118 e2018457118. DOI: 10.1073/pnas.2018457118 244. Chapman C, Lucas RM, Ponsonby AL, Taylor B, Ausimmune Investigator G. Predictors of progression from a first demyelinating event to clinically definite multiple sclerosis. Brain Communications (2022.0) 4 fcac181. DOI: 10.1093/braincomms/fcac181 245. Hart PH, Jones AP, Trend S, Cha L, Fabis-Pedrini MJ, Cooper MN, d’Este C, Geldenhuys S, Carroll WM, Byrne SN, Booth DR, Cole JM, Lucas RM, Kermode AG. A randomised, controlled clinical trial of narrowband UVB phototherapy for clinically isolated syndrome: The PhoCIS study. Multiple Sclerosis Journal Experimental Translational and Clinical (2018.0) 4 2055217318773112. DOI: 10.1177/2055217318773112 246. Trend S, Jones AP, Cha L, Cooper MN, Geldenhuys S, Fabis-Pedrini MJ, Carroll WM, Cole JM, Booth DR, Lucas RM, French MA, Byrne SN, Kermode AG, Hart PH. Short-term changes in frequencies of circulating leukocytes associated with narrowband UVB phototherapy in people with clinically isolated syndrome. Scientific Reports (2019.0) 9 7980. DOI: 10.1038/s41598-019-44488-6 247. Trend S, Leffler J, Cooper MN, Byrne SN, Kermode AG, French MA, Hart PH. Narrowband UVB phototherapy reduces TNF production by B-cell subsets stimulated via TLR7 from individuals with early multiple sclerosis. Clinical & Translational Immunology (2020.0) 9 e1197. DOI: 10.1002/cti2.1197 248. Miller KM, Hart PH, Lucas RM, Davis EA, de Klerk NH. Higher ultraviolet radiation during early life is associated with lower risk of childhood type 1 diabetes among boys. Scientific Reports (2021.0) 11 18597. DOI: 10.1038/s41598-021-97469-z 249. Megaw L, Clemens T, Daras K, Weller RB, Dibben C, Stock SJ. Higher sun exposure in the first trimester Is associated with reduced preterm birth A Scottish population cohort study using linked maternity and meteorological records. Frontiers in Reproductive Health (2021.0). DOI: 10.3389/frph.2021.674245 250. Hastie CE, Mackay DF, Clemens TL, Cherrie MPC, King A, Dibben C, Pell JP. Antenatal exposure to solar radiation and learning disabilities: Population cohort study of 422,512 children. Scientific Reports (2019.0) 9 9356-9356. DOI: 10.1038/s41598-019-45562-9 251. Chersich MF, Pham MD, Areal A, Haghighi MM, Manyuchi A, Swift CP, Wernecke B, Robinson M, Hetem R, Boeckmann M, Hajat S. Associations between high temperatures in pregnancy and risk of preterm birth, low birth weight, and stillbirths: systematic review and meta-analysis. BMJ British Medical Journal (2020.0) 371 m3811. DOI: 10.1136/bmj.m3811 252. Sun S, Weinberger KR, Spangler KR, Eliot MN, Braun JM, Wellenius GA. Ambient temperature and preterm birth: a retrospective study of 32 million US singleton births. Environment International (2019.0) 126 7-13. DOI: 10.1016/j.envint.2019.02.023 253. Yardman-Frank JM, Fisher DE. Skin pigmentation and its control: from ultraviolet radiation to stem cells. Experimental Dermatology (2021.0) 30 560-571. DOI: 10.1111/exd.14260 254. Lambert GW, Reid C, Kaye DM, Jennings GL, Esler MD. Effect of sunlight and season on serotonin turnover in the brain. Lancet (2002.0) 360 1840-1842. DOI: 10.1016/s0140-6736(02)11737-5 255. Wirz-Justice A, Skene DJ, Münch M. The relevance of daylight for humans. Biochemical Pharmacology (2021.0) 191 114304. DOI: 10.1016/j.bcp.2020.114304 256. Wang S-J, Chen M-Y. The effects of sunlight exposure therapy on the improvement of depression and quality of life in post-stroke patients: a RCT study. Heliyon (2020.0) 6 e04379-e04379. DOI: 10.1016/j.heliyon.2020.e04379 257. Wang N, Chen Y, Ji J, Chang J, Yu S, Yu B. The relationship between serum vitamin D and fracture risk in the elderly: a meta-analysis. Journal of Orthopaedic Surgery and Research (2020.0) 15 81-81. DOI: 10.1186/s13018-020-01603-y 258. Neale RE, Wilson LF, Black LJ, Waterhouse M, Lucas RM, Gordon LG. Hospitalisations for falls and hip fractures attributable to vitamin D deficiency in older Australians. British Journal of Nutrition (2021.0) 126 1682-1686. DOI: 10.1017/S0007114521000416 259. Annweiler C, Beauchet O. Questioning vitamin D status of elderly fallers and nonfallers: a meta-analysis to address a ‘forgotten step’. Journal of Internal Medicine (2015.0) 277 16-44. DOI: 10.1111/joim.12250 260. Ling Y, Xu F, Xia X, Dai D, Xiong A, Sun R, Qiu L, Xie Z. Vitamin D supplementation reduces the risk of fall in the vitamin D deficient elderly: an updated meta-analysis. Clinical Nutrition (2021.0) 40 5531-5537. DOI: 10.1016/j.clnu.2021.09.031 261. Chakhtoura M, Bacha DS, Gharios C, Ajjour S, Assaad M, Jabbour Y, Kahale F, Bassatne A, Antoun S, Akl EA, Bouillon R, Lips P, Ebeling PR, el-Hajj Fuleihan G. Vitamin D supplementation and fractures in adults: a systematic umbrella review of meta-analyses of controlled trials. Journal of Clinical Endocrinology & Metabolism (2022.0) 107 882-898. DOI: 10.1210/clinem/dgab742 262. LeBoff MS, Chou SH, Ratliff KA, Cook NR, Khurana B, Kim E, Cawthon PM, Bauer DC, Black D, Gallagher JC, Lee IM, Buring JE, Manson JE. Supplemental vitamin D and incident fractures in midlife and older adults. The New England Journal of Medicine (2022.0) 387 299-309. DOI: 10.1056/NEJMoa2202106 263. Li H, Sun D, Wang A, Pan H, Feng W, Ng CH, Ungvari GS, Tao L, Li X, Wang W, Xiang Y-T, Guo X. Serum 25-hydroxyvitamin D levels and depression in older adults: a dose–response meta-analysis of prospective cohort studies. American Journal of Geriatric Psychiatry (2019.0) 27 1192-1202. DOI: 10.1016/j.jagp.2019.05.022 264. Libuda L, Laabs B-H, Ludwig C, Bühlmeier J, Antel J, Hinney A, Naaresh R, Föcker M, Hebebrand J, König IR, Peters T. Vitamin D and the risk of depression: a causal relationship? Findings from a Mendelian randomization study. Nutrients (2019.0) 11 1085. DOI: 10.3390/nu11051085 265. Arathimos R, Ronaldson A, Howe LJ, Fabbri C, Hagenaars S, Hotopf M, Gaughran F, Lewis CM, Dregan A. Vitamin D and the risk of treatment-resistant and atypical depression: a Mendelian randomization study. Translational Psychiatry (2021.0) 11 561-561. DOI: 10.1038/s41398-021-01674-3 266. Cheng YC, Huang YC, Huang WL. The effect of vitamin D supplement on negative emotions: a systematic review and meta-analysis. Depression and Anxiety (2020.0) 37 549-564. DOI: 10.1002/da.23025 267. Okereke OI, Reynolds CF, Mischoulon D, Chang G, Vyas CM, Cook NR, Weinberg A, Bubes V, Copeland T, Friedenberg G, Lee IM, Buring JE, Manson JE. Effect of long-term vitamin D3 supplementation vs placebo on risk of depression or clinically relevant depressive symptoms and on change in mood scores: a randomized clinical trial. JAMA Journal of the American Medical Association (2020.0) 324 471-480. DOI: 10.1001/jama.2020.10224 268. Zheng J-S, Luan Ja, Sofianopoulou E, Sharp SJ, Day FR, Imamura F, Gundersen TE, Lotta LA, Sluijs I, Stewart ID, Shah RL, van der Schouw YT, Wheeler E, Ardanaz E, Boeing H, Dorronsoro M, Dahm CC, Dimou N, El-Fatouhi D, Franks PW. The association between circulating 25-hydroxyvitamin D metabolites and type 2 diabetes in European populations: a meta-analysis and Mendelian randomisation analysis. PLoS Medicine (2020.0) 17 e1003394-e1003394. DOI: 10.1371/journal.pmed.1003394 269. Wang N, Wang C, Chen X, Wan H, Chen Y, Chen C, Han B, Lu Y. Vitamin D, prediabetes and type 2 diabetes: bidirectional Mendelian randomization analysis. European Journal of Nutrition (2019.0) 59 1379-1388. DOI: 10.1007/s00394-019-01990-x 270. Pittas AG, Dawson-Hughes B, Sheehan P, Ware JH, Knowler WC, Aroda VR, Brodsky I, Ceglia L, Chadha C, Chatterjee R, Desouza C, Dolor R, Foreyt J, Fuss P, Ghazi A, Hsia DS, Johnson KC, Kashyap SR, Kim S, LeBlanc ES. Vitamin D supplementation and prevention of type 2 diabetes. The New England Journal of Medicine (2019.0) 381 520-530. DOI: 10.1056/NEJMoa1900906 271. Pittas A, Dawson-Hughes B, Staten M. Vitamin D supplementation and prevention of type 2 diabetes. Reply. The New England Journal of Medicine (2019.0) 381 1785-1786. DOI: 10.1056/NEJMc1912185 272. Tang H, Li D, Li Y, Zhang X, Song Y, Li X. Effects of vitamin D supplementation on glucose and insulin homeostasis and incident diabetes among nondiabetic adults: a meta-analysis of randomized controlled trials. International Journal of Endocrinology (2018.0) 2018 7908764-7908769. DOI: 10.1155/2018/7908764 273. Dhamrait GK, Panchal K, Fleury NJ, Abel TN, Ancliffe MK, Crew RC, Croft K, Fernandez BO, Minnion M, Hart PH, Lucas RM, Mark PJ, Feelisch M, Weller RB, Matthews V, Gorman S. Characterising nitric oxide-mediated metabolic benefits of low-dose ultraviolet radiation in the mouse: a focus on brown adipose tissue. Diabetologia (2019.0) 63 179-193. DOI: 10.1007/s00125-019-05022-5 274. Sluyter JD, Manson JE, Scragg R. Vitamin D and clinical cancer outcomes: a review of meta-analyses. JBMR Plus (2021.0) 5 e10420. DOI: 10.1002/jbm4.10420 275. Ong J-S, Dixon-Suen SC, Han X, An J, Liyanage U, Dusingize J-C, Schumacher J, Gockel I, Boehmer A, Jankowski J, Palles C, O'Mara T, Spurdle A, Law MH, Iles MM, Pharoah P, Berchuck A, Zheng W, Thrift AP, Olsen C. A comprehensive re-assessment of the association between vitamin D and cancer susceptibility using Mendelian randomization. Nature Communications (2021.0) 12 246-246. DOI: 10.1038/s41467-020-20368-w 276. Keum N, Chen QY, Lee DH, Manson JE, Giovannucci E. Vitamin D supplementation and total cancer incidence and mortality by daily vs infrequent large-bolus dosing strategies: a meta-analysis of randomised controlled trials. British Journal of Cancer (2022.0) 127 872-878. DOI: 10.1038/s41416-022-01850-2 277. Sintzel MB, Rametta M, Reder AT. Vitamin D and multiple sclerosis: a comprehensive review. Neurology and Therapy (2018.0) 7 59-85. DOI: 10.1007/s40120-017-0086-4 278. Jiang X, Ge T, Chen C-Y. The causal role of circulating vitamin D concentrations in human complex traits and diseases: a large-scale Mendelian randomization study. Scientific Reports (2021.0) 11 184-184. DOI: 10.1038/s41598-020-80655-w 279. Manousaki D, Harroud A, Mitchell RE, Ross S, Forgetta V, Timpson NJ, Smith GD, Polychronakos C, Richards JB. Vitamin D levels and risk of type 1 diabetes: a Mendelian randomization study. PLoS Medicine (2021.0) 18 e1003536-e1003536. DOI: 10.1371/journal.pmed.1003536 280. Pham H, Rahman A, Majidi A, Waterhouse M, Neale RE. Acute respiratory tract infection and 25-hydroxyvitamin D concentration: a systematic review and meta-analysis. International Journal of Environmental Research and Public Health (2019.0) 16 3020. DOI: 10.3390/ijerph16173020 281. Jolliffe DA, Camargo CA, Sluyter JD, Aglipay M, Aloia JF, Ganmaa D, Bergman P, Bischoff-Ferrari HA, Borzutzky A, Damsgaard CT, Dubnov-Raz G, Esposito S, Gilham C, Ginde AA, Golan-Tripto I, Goodall EC, Grant CC, Griffiths CJ, Hibbs AM, Janssens W. Vitamin D supplementation to prevent acute respiratory infections: a systematic review and meta-analysis of aggregate data from randomised controlled trials. The Lancet Diabetes & Endocrinology (2021.0) 9 276-292. DOI: 10.1016/S2213-8587(21)00051-6 282. Çolak Y, Nordestgaard BG, Afzal S. Low vitamin D and risk of bacterial pneumonias: mendelian randomisation studies in two population-based cohorts. Thorax (2021.0) 76 468-478. DOI: 10.1136/thoraxjnl-2020-215288 283. Pham H, Waterhouse M, Baxter C, Duarte Romero B, McLeod DSA, Armstrong BK, Ebeling PR, English DR, Hartel G, Kimlin MG, Martineau AR, O'Connell R, van der Pols JC, Venn AJ, Webb PM, Whiteman DC, Neale RE. The effect of vitamin D supplementation on acute respiratory tract infection in older Australian adults: an analysis of data from the D-Health trial. The Lancet Diabetes & Endocrinology (2021.0) 9 69-81. DOI: 10.1016/S2213-8587(20)30380-6 284. Estimating dose-response relationships for vitamin D with coronary heart disease, stroke, and all-cause mortality: observational and Mendelian randomisation analyses. The Lancet Diabetes & Endocrinology (2021.0) 9 837-846. DOI: 10.1016/S2213-8587(21)00263-1 285. Young AR, Morgan KA, Harrison G, Lawrence KP, Petersen B, Wulf HC, Philipsen PA. A revised action spectrum for vitamin D synthesis by suberythemal UV radiation exposure in humans in vivo. Proceedings of the National Academy of Sciences of the United States of America (2021.0) 118 e2015867118. DOI: 10.1073/pnas.2015867118 286. Al Zarooni AAR, Nagelkerke N, Al Marzouqi FI, Al Darmaki SH. Risk factors for vitamin D deficiency in Abu Dhabi Emirati population. PLoS ONE (2022.0) 17 e0264064-e0264064. DOI: 10.1371/journal.pone.0264064 287. Al-Yatama FI, AlOtaibi F, Al-Bader MD, Al-Shoumer KA. The effect of clothing on vitamin D status, bone turnover markers, and bone mineral density in young Kuwaiti females. International Journal of Endocrinology (2019.0) 2019 6794837-6794810. DOI: 10.1155/2019/6794837 288. Passeron T, Bouillon R, Callender V, Cestari T, Diepgen TL, Green AC, van der Pols JC, Bernard BA, Ly F, Bernerd F, Marrot L, Nielsen M, Verschoore M, Jablonski NG, Young AR. Sunscreen photoprotection and vitamin D status. British Journal of Dermatology (2019.0) 181 916-931. DOI: 10.1111/bjd.17992 289. Neale RE, Khan SR, Lucas RM, Waterhouse M, Whiteman DC, Olsen CM. The effect of sunscreen on vitamin D: a review. British Journal of Dermatology (2019.0) 181 907-915. DOI: 10.1111/bjd.17980 290. Fayet-Moore F, Brock KE, Wright J, Ridges L, Small P, Seibel MJ, Conigrave AD, Mason RS. Determinants of vitamin D status of healthy office workers in Sydney, Australia. Journal of Steroid Biochemistry and Molecular Biology (2019.0) 189 127-134. DOI: 10.1016/j.jsbmb.2019.02.017 291. Feketea GM, Bocsan IC, Tsiros G, Voila P, Stanciu LA, Zdrenghea M. Vitamin D status in children in Greece and its relationship with sunscreen application. Children (Basel) (2021.0) 8 111. DOI: 10.3390/children8020111 292. Datta P, Philipsen PA, Idorn LW, Wulf HC. Low vitamin D in dark-skinned immigrants is mainly due to clothing habits and low UVR exposure: a Danish observational study. Photochemical & Photobiological Sciences (2021.0) 20 1573-1584. DOI: 10.1007/s43630-021-00115-w 293. Young AR, Morgan KA, Ho T-W, Ojimba N, Harrison GI, Lawrence KP, Jakharia-Shah N, Wulf HC, Cruickshank JK, Philipsen PA. Melanin has a small inhibitory effect on cutaneous vitamin D synthesis: A comparison of extreme phenotypes. Journal of Investigative Dermatology (2020.0) 140 1418-1426.e1411. DOI: 10.1016/j.jid.2019.11.019 294. Neville JJ, Palmieri T, Young AR. Physical determinants of vitamin D photosynthesis: a review. JBMR Plus (2021.0) 5 e10460. DOI: 10.1002/jbm4.10460 295. Kraus FB, Medenwald D, Ludwig-Kraus B. Do extreme summers increase blood vitamin D (25-hydroxyvitamin D) levels?. PLoS One (2020.0) 15 e0242230. DOI: 10.1371/journal.pone.0242230 296. 296.Boyall, L., & Samingpai, B. (2021). Tourism in Antarctica. Accessed 25/9/2022 297. Eleftheratos K, Kapsomenakis J, Zerefos C, Bais A, Fountoulakis I, Dameris M, Jöckel P, Haslerud A, Godin-Beekmann S, Steinbrecht W, Petropavlovskikh I, Brogniez C, Leblanc T, Liley JB, Querel R, Swart D. Possible effects of greenhouse gases to ozone profiles and DNA active UV-B irradiance at ground level. Atmosphere (2020.0) 11 228. DOI: 10.3390/atmos11030228 298. van der Leun JC, Piacentini RD, de Gruijl FR. Climate change and human skin cancer. Photochemical & Photobiological Sciences (2008.0) 7 730-733. DOI: 10.1039/b719302e 299. Piacentini RD, Della Ceca LS, Ipiña A. Climate change and its relationship with non-melanoma skin cancers. Photochemical & Photobiological Sciences (2018.0) 17 1913-1917. DOI: 10.1039/c7pp00405b
--- title: Mousepost 2.0, a major expansion of the resource authors: - Steven Timmermans - Jolien Vandewalle - Claude Libert journal: Nucleic Acids Research year: 2023 pmcid: PMC9976886 doi: 10.1093/nar/gkad064 license: CC BY 4.0 --- # Mousepost 2.0, a major expansion of the resource ## Abstract The Mousepost 1.0 online search tool, launched in 2017, allowed to search for variations in all protein-coding gene sequences of 36 sequenced mouse inbred strains, compared to the reference strain C57BL/6J, which could be linked to strain-specific phenotypes and modifier effects. Because recently these genome sequences have been significantly updated and sequences of 16 extra strains added by the Mouse Genomes Project, a profound update, correction and expansion of the Mousepost 1.0 database has been performed and is reported here. Moreover, we have added a new class of protein disturbing sequence polymorphisms (besides stop codon losses, stop codon gains, small insertions and deletions, and missense mutations), namely start codon mutations. The current version, Mousepost 2.0 (https://mousepost.be), therefore is a significantly updated and invaluable tool available to the community and is described here and foreseen by multiple examples. ## INTRODUCTION The mouse is the most important and most widely used mammalian model organism species. It takes a prominent place in biomedical and genetic research, both for fundamental research and for disease treatment/drug development. It has a rich history in research, dating back more than a century, but the sequencing and publication of the mouse reference genome has marked the real start of the modern genetic and genomic sequence-based research [1]. The year 2022 marks the twentieth anniversary of the publication of the first draft of the mouse reference genome, derived from the C57BL/6J strain [2]. Mouse research is usually performed using inbred mice, which are, by definition, homozygous over their entire diploid genome and belong to a certain inbred strain. Several hundreds of such inbred strains have been described, the most popular ones being commercially available at several sources. Some of these inbred strains exhibit strain-specific phenotypes, such as resistance to lipopolysaccharides in the mice belonging to the C3H/HeJ strain [3]. Also, many examples have been described, illustrating that mutations introduced in mice via embryonic stem cells or CRISPR/Cas9 exhibit very significantly different phenotypes, depending on the mouse inbred strain in which the mutation was introduced [4]. Such data suggest the existence of strain-specific genome variations that should be mapped to the nucleotide level, to understand the role, function and mechanisms of these elements in the aforementioned observations [5,6]. During the last two decades, significant advancements were made in the quality of the mouse C57BL/6J reference genome sequence [7] and projects towards obtaining sequence data from other inbred lines were also started. The Mouse Genomes Project (MGP), initiated and maintained by the Wellcome Sanger Institute, is the primary effort towards whole genome shotgun sequencing and variant detection in mouse inbred lines [8,9]. The initial release of the MGP data in 2011 contained 36 popular and frequently used inbred strains (selected in collaboration with the mouse research community) [9]. These data were available as complete genome assemblies for a subset of the strains, and ‘variant call format’ (VCF) files containing information of single-nucleotide polymorphisms (SNPs) as well as small insertions and deletions (indels) for all strains. Our research group has used the nucleotide level data released by the MGP to build the ‘Mousepost’ resource (an acronym of mouse polymorphic sequence tags; https://mousepost.be). Initially started from a project to provide an overview of the mutations in the wild-derived inbred strain SPRET/EiJ [10], this endeavour was expanded to all protein-coding transcripts in all MGP sequenced strains and we released the first version of the data and tool in 2017 [11]. This web-based tool provided strain-specific protein sequence information for the 36 strains in the MGP. We processed the data from the VCF files and only retained variants causing non-synonymous amino acid substitutions on a per codon basis. The first release of Mousepost (Mousepost 1.0) included a classification of the protein sequences [stop-gain (SG), stop-loss (SL) or others] and data export and search functions, making it an excellent complementary resource of the MGP [7]. We expanded Mousepost several times with novel functionalities, such as variants in the reference strain compared to the others and direct pairwise comparisons between strains [12,13]. Furthermore, we also made a derivate of Mousepost, Ratpost, providing the same type of information for rat inbred strains [14,15]. Based on the popular use of Mousepost 1.0 and recent significant changes in (i) the underlying sequence data in the reference genome, (ii) the MGP sequences and (iii) the addition of new sequenced strains, the construction of a new version as Mousepost 2.0 was mandatory. This new Mousepost 2.0 version adds data from 16 new strains, bringing the total number of strains present to 52. The refinements to the MGP data cause the addition of many new events and removal of others compared to the previous version. In this paper, we disclose and discuss the main additions and removals to the Mousepost database and describe changes to the processing pipeline. Moreover, we have now also added a new class of sequence polymorphisms, which cause start codon (SC) mutations, as a new class and have also processed genome sequence patches on the reference genome meaning that we involved the incremental sequence changes of the last 5 years that were not part of the previous version. The multitude of changes in the Mousepost 1.0 database are illustrated by several examples that illustrate the power of the new Mousepost 2.0 release, available at https://mousepost.be, and the tool in general. ## MATERIALS AND METHODS Variant data were obtained from the ftp site of the MGP [8] as a VCF file containing SNPs and indels from all included strains, as release REL-2005. We used the Picard tools (https://broadinstitute.github.io/Picard/) ‘LiftoverVcf’ command with the appropriate chain file from the University of California, Santa Cruz (UCSC) to reprocess the mm10 annotated VCF file to the new mm39 coordinates and structural annotation. This VCF was processed to remove heterozygous and low-quality calls and split per strain. All files were processed into a MySQL relational database, which holds the DNA sequence for every variant position in all strains, including C57BL/6J. Structural reference annotation. We used the GRCm39 version of the mouse genome reference strain (C57BL/6J), which we obtained from the Genome Reference Consortium (GRC) ftp site. The structural annotation used was the complete annotation, version M28 from the gencode resource [16], as in ‘gene transfer format’ (GTF) format. Mousepost data processing was done with the Mousepost pipeline as described in [12]. The only difference to the pipeline itself was that the conserved domain database for the rps-blast was updated to the release of 22 February 2021. PROVEAN (Protein Variation Effect Analyzer) [17] score calculation was performed on a compute cluster using saved supporting gene sets whenever possible and the same blast database. This allows us to keep −2.5 as a score threshold below which the mutation is predicted to be deleterious to protein function with a balanced accuracy of $80\%$ as described by the PROVEAN authors. Gene Ontology. Information for mouse was downloaded from the Gene Ontology (GO) Consortium website (www.geneontology.org/). We processed this file to update the GO terms for all genes in our dataset. Sequence alignments were made with the needle algorithm from the EMBOSS tools [18] in case of pairwise alignments or with the MUSCLE aligner [19] for multiple sequence alignments. Multiple sequence alignments were used to obtain majority vote consensus sequences to compare the reference C57BL/6J genome too. ## The Mousepost 2.0 pipeline data All source datasets used to create the Mousepost 2.0 resource were updated: the SNP/indel data, the C57BL/6J reference genome sequences and the reference genome annotation. A complete comparison of the data and workflow used in the original Mousepost and Mousepost 2.0 can be found in Figure 1. Briefly, the following updates were performed: (i) In order to obtain the most relevant dataset, the genome annotation was changed to the most recent release of the mouse genome, which is mm39 at the time of writing. As the variants were called using the mm10 genome annotation, we used Picard tools (https://broadinstitute.github.io/Picard/), specifically the command LiftoverVcf, with the ‘mm10tomm39’ chain file obtained from the UCSC ftp servers to lift the mm10 VCF file to the mm39 mouse genome assembly [20]. We then used the gencode v28 release [21] as the structural annotation from which transcripts were obtained, changed from Ensemble release 85. ( ii) The SNP/indel data were updated from REL-1505 to REL-2005, which is the most recent release available on the ftp site [4]. This REL-2005 release contained data of a total of 52 mouse inbred strains, which is 16 more than the REL-1505 version that was used in the initial Mousepost version [7]. The newly added strains are B10/RIII, BALB/cByJ, C57BL/10SnJ, CE/J, CZECHII/EiJ, JF1/MsJ, LG/J, MA/MyJ, NON/LtJ, PL/J, QSi3, QSi5, RIIIS/J, SJL/J, SM/J and SWR/J. Figure 1.: Schematic representation of the differences between Mousepost 1.0 and Mousepost 2.0, the updated version described in this work. ## Processing and annotation of transcripts The variant VCF file was first converted to the mm39 version of the mouse reference genome. This was done using the LiftoverVcf command from the Picard tools software and the mm10 to mm39 UCSC chain file obtained from https://hgdownload.soe.ucsc.edu/goldenPath/mm10/liftOver/. The transcript GTF files and lifted VCF file were converted to ‘browser extensible data’ file format with BEDTools [22], in order to process the files with BEDOPS [23]. To minimize processing time, only relevant transcripts and events were selected: (i) we made the intersect of the variants with the transcripts to retain only those variants that overlapped a transcript location; and (ii) to select relevant transcripts, the opposite overlap was made, with only transcripts overlapping one or more variants being kept. Furthermore, all heterozygous events, where one of the alleles matches the reference (C57BL/6J), low-quality events (filtered on the FI tag, with FI = 1 events retained) and non-protein-coding transcripts or pseudogene transcripts were filtered out of the dataset. This provided a total of 71 035 transcripts across all strains to be processed by the Mousepost 2.0 pipeline, as described in [11,13], an increase of 12 988 compared to the original Mousepost release. Briefly, we assembled the untranslated sequences (5′UTR and 3′UTR) and coding sequence (CDS) from the structural annotation and sequence data. We also included the sequence data from the so-called reference mouse (C57BL/6J) ‘genome patches’. These are 87 sequence scaffolds with sequence data that would lead to coordinate changes if integrated in the mm10 reference genome. We mapped the coordinates to the standard chromosomes and used the patch sequence as a replacement where applicable. The sequences from the patches were given precedence in the construction of the reference version of the UTR, CDS and eventual translated sequences. Strain-specific sequences were translated and compared for classification. The number of classes was expanded from the three original ones [SG, SL and non-synonymous variants (MUT)] with a fourth specific class, namely SC variants. Indeed, the initial release of Mousepost did not include these variants fully appropriately in any of the other three classes, and although only a small group, such SC losses can be very deleterious and thus have a high impact on protein function. Novel MUT class variants were also subjected to a prediction of their potential effect on protein function with the PROVEAN tool [17]. All PROVEAN predictions were made using the same database used in the initial version of Mousepost. As always, a PROVEAN score of −2.5 or lower suggests a significant loss of function of the protein involving a sequence variation. ## Updated statistics First, the REL-2005 SNP/indel data from the MGP (from the mouse strains that are compared to the reference genome) contain significantly more data than the previously used REL-1505 (45 164 532 versus 38 700 845 in REL-1505). This leads to (i) a strong increase in number of variant transcripts compared to the reference genome. But in addition, (ii) several previously defined events are no longer present in the dataset. The main reason for such event losses is the inclusion of more accurate sequencing data, which improves SNP/indel calling and thus removes events that were previously false positive calls. Second, use of the new version of the C57BL/6J reference genome also leads to removal of some events, but this is an almost negligible amount compared to the total number of events removed for the previous reason. On the other hand, one strain, closely related to the C57BL/6J reference strain, namely C57BL/6NJ, in the new version of the Mousepost database has no longer SL variant transcripts, while it had 46 of them in the initial version. Third, the proportion of different classes differs markedly between both Mousepost versions, even when the new SC class is removed from the comparison (Table 1). However, the class proportions are largely the same when compared between the different strains, with strong deviations only seen in strains that are closely related to C57BL/6J and have relatively few mutated transcripts. Taking into consideration all 52 included strains (Figure 2), the largest group consists of the MUT variants, which substitute single amino acids or constitute small indels ($46.20\%$). The second largest group of transcripts contains only synonymous events ($44.42\%$). These are either fully silent nucleotide changes (e.g. most wobble position variants) or double mutations that cancel each other at the codon level (e.g. codon TCC→AGC, both encoding serine). Stop codon gains (nonsense variants) make up $5.14\%$ of all variant transcripts. Furthermore $3.46\%$ of the transcripts have an SC variation and only $0.78\%$ belong to the SL class (Tables 2 and 3). If the synonymous class is not considered and only transcripts with mutations leading to amino acid changes are taken into account, then $1.40\%$ of those have an SL mutation, $6.23\%$ have a mutation in the canonical SC, $9.24\%$ have gained a premature stop codon (SG) and $83.13\%$ have a non-synonymous mutation that is in no way related to start or stop codons. Next to the correction and expansion of the data for the 36 previously studied strains, the REL-2005 also has SNP and indel data for 16 extra strains not involved in the previous release (Figure 1). These new strains were processed in the pipeline and added to the database (Table 2), expanding the number of mouse strains to 52 that are compared to C57BL/6J. Of the added strains, two are closely related to C57BL/6J and have only a few mutated transcripts, while two other strains (CZECHII/EiJ and JF1/MsJ) are (evolutionary) far removed from C57BL/6J and are in the same order as other ‘wild-derived strains’ (e.g. SPRET/EiJ). The addition of these strains also resulted in a complete rebuild of the ‘C57BL/6J’ section of Mousepost 2.0. In this section, we provide an overview of those transcripts that can be considered as mutated in the reference strain, because they differ in C57BL/6J compared to synthetic reference transcripts that were constructed from all strain-specific transcripts by a per position majority vote mechanism. For Mousepost 2.0, the synthetic reference transcripts were rebuilt completely using the REL-2005 SNP/indel data from the MGP, as well as the data from the new strains, and any newly added or removed variant information in the already included strains. The impact of this change is large: (i) There are no longer C57BL/6J-specific SG transcripts as C57BL/6NJ does not have any SL variants compared to the reference (Table 1) and, overall, the number of SG C57BL/6J variants drops to 27 (from 38). ( ii) The number of SL variants increases from 121 to 277 and the number of MUT class transcripts with a PROVEAN score of ≤−2.5 is 2684 in Mousepost 2.0 versus 1892 in Mousepost 1.0. All transcripts where the protein sequence was different in C57BL/6J from the synthetic consensus reference were included as missense transcripts in the database. ## Mousepost use and applications In this section, we will provide direct examples of Mousepost use at https://mousepost.be. We will highlight the changes to the previous version by the new data and provide a step-by-step overview as to how the tool can be used to obtain the data of interest. ## A general overview of the number of transcripts per class and per strain On the welcome page of the tool, a short description is provided as well as four submenus. The ‘data overview’ item gives an overview of how many distinct transcripts and genes can be found in each of the variant classes of all included strains, compared to the reference C57BL/6J. This table can be updated with the application of filters for SG, SL and missense (MUT) classes. For SG and SL, this filter is the maximal and the minimal length ratio, respectively, of the strain-specific transcript compared to the reference (C57BL/6J) one. For MUT class genes, this is the maximal value of the PROVEAN score below which at least one scored variant must fall; i.e. if there are three variants in the transcript, and the filter is set to ‘−5’, at least one of those must have a score of −5 or lower to be included in the overview. The numbers that are shown in the table also act as internal links: for each strain, the numbers of genes and transcripts displayed are direct links to the strain-specific lists of variants immediately filtered to return the actual transcripts that are behind the number shown in the main table. Compared to the previous version of Mousepost, the main change is the inclusion of a new class, SC variants. The other two submenus (‘Compare to Reference’ and ‘Deviations in C57BL/6J’) are a set of guided links to the main functions, all of which can be accessed from the top menu bar. The submenus will guide the user to the correct page for performing the desired query, through a set of descriptions. The main functions, which will be discussed in detail, are as follows: (i) per strain lists (‘Lists’) of each variant type, optionally filtered by chromosomal region, which is also what the number of transcripts/genes from table in the previous paragraph links to; (ii) gene, gene location and gene function (ontology) searches (‘Search’), to search for information for specific genes, genomic regions or gene sets related to specific functions in multiple strains at the same time; and (iii) the ability to perform direct pairwise comparisons of any two strains (‘Pairwise’). Next to these comparisons, there is also the option (iv) to obtain gene lists per mutation type of those genes that are considered to be mutated in the reference genome, obtained by comparison with a synthetic reference (‘C57BL/6J’). ## The new SC variant class One of the novelties of Mousepost 2.0 is the addition of the SC variant class. All strains have several SC variants assigned to them. We assume that these will mostly result in significant loss of function of the protein that should normally be produced, which is why we have chosen to include this set as a separate class. We can use gene set enrichment analysis to find phenotypes and pathways that are related to these genes. Due to the low number of such genes, no statistically significant enrichments could be found; however, there was a trend to lipid metabolism-related functions and cardiac development and/or function. The A/J strain has both these trends and, according to the mouse phenome database, also displays phenotypical characteristics in accordance with the predicted functions: a relatively low level of free fatty acids circulating in the blood compared to other tested inbred strains (Supplementary Figure S1) and a very low heart size compared to body weight (the lowest measured) [24] (Figure 3). In order to perform such analyses, the list of SC transcripts for the A/J strain can be easily obtained from the Mousepost tool. Here, we can use the ‘Lists’ function of the menu bar and fill in the strain of interest (SOI), A/J, the variant type of interest and the SC variants. This gives a complete list of all SC variants in A/J and can be easily exported for further processing, such as gene set enrichment analyses. This list may also be obtained through the main overview table on the tool homepage. As previously described, the numbers of transcripts per strain and per class act as internal hyperlinks, which lead to the strain-specific transcript page. As the data are retrieved from the database with HTTP_GET requests, direct passing of the search/filter parameters is possible. Thus, the links directly give the results for the correct strain with application of the selected filters for length ratio or PROVEAN score; other filters such as lost protein domains or location restrictions must be set on the ‘Lists’ page. Figure 3.: Distribution of ventricle weight (median with complete measured range) (left and right) of the mouse heart in those mouse strains in which this has been measured, corrected for the total body weight of the mice. The A/J strain has the overall lowest relative ventricle weight, with especially male A/J mice being on the extreme low end of the distribution. Image adapted from the mouse phenome database (https://phenome.jax.org/). When studying the SC mutants for the A/J stain, we find a potential link to cardiac development in these mice to the Mybpc3 gene [25,26]. This gene has an SC loss that affects all transcripts (the methionine codon is mutated to an isoleucine codon, M1I). The protein will likely be truncated from the N-terminal side, as there is a second, in-frame, SC just 8 AAs downstream of the normal one, which would result in the partial removal of an ‘immunoglobulin-like domain’ from the N-terminal part of the protein (normally from AA position 2→103). The question remains how severely this would affect protein function and whether the downstream ATG can be used as a substitute SC. ## Stop codon and non-synonymous variants Obtaining lists of transcripts for the other classes works for any SOI in almost the same manner. For stop codon variants, the strain and length ratio (>1 for SL and <1 for SG) must be selected. In the case of SG, it is also possible to check a ‘domains’ option, which will also give an overview of which domains are part of the truncated region in the result set. The results are returned as a table with, for each transcript, the length in the C57BL/6J and the SOI version, and a visual ratio as a red/green bar where the red part is the part that is truncated (SG) or added (SL) to the reference sequence. This update increases the number of variants in each class and caused the reclassification of several genes and transcripts. The latter is mostly due to inclusion of new SG mutations, and an example is seen when retrieving a listing for the AKR/J SG variants since one of the new genes recovered in this release is Nlrp1b. This gene is related to the sensitivity to anthrax lethal toxin, to which the AKR/J strain has previously been shown to be resistant [27]. This resistance has been linked to Nlrp1b [28], the gene coding for a part of the Nlrp1 inflammasome. Nlrp1b loss of function in other strains also confers this resistance to the toxin (27–29). This gene has multiple mutations, several of which were included in the previous Mousepost database and have PROVEAN scores (−7 and −9) predicting loss of function. However, multiple new events in this gene were included from REL-2005, one of which causes an SG mutation in all coding transcripts of this gene. This SG results in the loss of 287 AAs ($24\%$) truncating the protein from $\frac{1174}{1177}$ to $\frac{887}{890}$ AAs depending on the transcript. This will very likely result in a partial loss of the ‘function to find’ (FIIND) domain and total loss of the ‘caspase activation and recruitment domain’ (CARD) domain (Figure 4 and Supplementary Figure S2). There is one caveat the user must take into account when querying SG or SL variants. In case of SG variants, depending on where the nonsense mutation has occurred, there is a (high) possibility of ‘nonsense-mediated decay’ (NMD) [30]. In vivo or in vitro, when a nonsense mutation occurs ‘early’ in a transcript, the mRNA may be recognized as aberrant and be degraded immediately and so no meaningful amount of truncated peptide will remain. As it is not possible to determine which proteins may (not) undergo NMD, we have chosen to report the length of the encoded protein that is affected by the SG variants. These reported data do not depend on NMD, but users working with mice should be careful when working in vivo or in vitro with SG variants as the mRNAs can be degraded and the truncated protein absent. In addition, a mechanism (called ‘non-stop decay’) exists to degrade SL mRNA [31,32], but is mostly applicable to those transcripts lacking any stop codon at all, which does not occur that often. Figure 4.: Representation of the new SG variant in Nlrp1b in AKR/J mice and its effects on the protein, which is in C57BL/6J 1177 AAs in size. Domains are added based on UniProt/InterPro. FIIND and CARD domains are, respectively, partially or completely affected by the truncation. NACHT, nucleoside triphosphatase (NTPase) domain; LRR, leucine-rich repeat; FIIND, function to find domain; CARD, caspase activation and recruitment domain. Non-synonymous variants can be queried per strain by selecting ‘MUT’ and specifying a maximal PROVEAN score. The result table obtained here lists all transcripts that have at least one variant with PROVEAN score lower than the selected cut-off. Next to the gene and transcript, the total number of variants in the protein sequence is displayed as well as the score of the lowest scoring variant. In all cases, the result table contains several links to more information: the Ensemble IDs are internal links to a ‘details’ page providing in detail information for the transcript and strain and the final column contains external links to UniProt, Ensembl, Mouse Genome Informatics (MGI) and UCSC Genome Browser. ## Searching the database Next to per strain and per variant type listings, it is also possible to perform an extensive search function consisting of three main options: a gene of interest (GOI), with optional filters, a region of interest and a GO term. In all cases, results are returned as a table, with results grouped by variant type and displaying the strain, transcript ratio and length ratio, for SG, SL and SC, or lowest PROVEAN score, for MUT transcripts. For the GOI search, one or more gene names can be entered, and the official gene name or an Ensembl gene ID can be used. The main application is to study whether one or more GOI(s) have variants (of any type) in any strain (by default all, but any can be selected). As an illustration, we will search for a previously described mutation in the Tlr3 gene [25] in one of the strains that was added in this update: the Tlr3 P369L variant found in CZECHII/EiJ. Using the search function with the query ‘Tlr3’ as input, and filtering on strains and mutation types, we are able to easily recover and confirm the Tlr3 mutation, a P369L change [33], which ruins the function of TLR3, similar to the deleterious Tlr4 P712H change in the genome of strain C3H/HeJ [3]. CZECHII/EiJ mice are resistant to diet-induced obesity since they remain lean when they are fed a high-fat diet (HFD) for 8 weeks [mouse phenome database, Paigen1 dataset [34]; Supplementary Figure S3]. Searching the Mousepost 2.0 dataset using the GO search function from the website as well as data from the literature for genes related to diet-induced obesity resulted in transcripts from two genes being affected. One is the Akap1 gene, which has 5 AA substitutions, and is known in relation to this phenotype [35]. However, none of the AA changes individually have a PROVEAN score of −2.5 (lowest is −1.4) or lower, and are thus not predicted to be deleterious, but all of them combined may still impact function. The use of the GO search function using the GO term ‘regulation of appetite’ yields a large result table, which was narrowed down by entering CZECHII in the table search/filter field. One of the genes found here, as well as in the literature, is the cholecystokinin-coding gene (Cck). Mice that have a loss of function of this gene were shown to be resistant to HFD-induced obesity [36]. In CZECH/EiJ mice, Cck shows an SG mutation leading to the production of a truncated peptide of the canonical transcript. The resulting deletion is small, only 10 AAs preceded by a 4 AA frameshift of a normally 115 AA protein, but this affects the active peptide region, which indeed is known to involve the very C-terminal end of the protein. The final option is the possibility to search a genomic region. This is an extension on the location filter that can be optionally enabled when obtaining strain-specific gene lists. The main difference between the location filter in the strain-specific lists and the location search is that the search is performed across all strains and variant types and without any cut-offs, e.g. for the PROVEAN score. When this functionality is used with a certain region of interest, such as can be obtained from quantitative trait loci (QTL) studies, it is possible to obtain candidate genes for the trait investigated that also have at least one variant in the strain used. In the case of CZECHII/EiJ, an in silico QTL study from 2012 for nurturing ability [37], in which the authors showed a very low ‘average daily weight gain’ of pups for this strain, nine genes were identified as candidates in the QTL analysis. Based on Mousepost 2.0, we can associate two of these (Fn1 and Rbm44) with protein-coding variants in the CZECHII/EiJ inbred line when using the regions provided to search our tool database and comparing the results. ## The updated reference and C57BL/6J variants The change from mm10 to mm39 version of the reference genome has directly influenced the variants included in the database: several previously included variants have been removed as they had become irrelevant, indicating that these were in fact false positive calls. Furthermore, a large number of new variants were included. The largest impact of these facts can be found in the number of variants per strain being higher in this release as reported previously and in the ‘C57BL/6J’ part of Mousepost. In this part of the tool, we present those transcripts, divided per class, that were found to be deviant in the reference genome (C57BL/6J) compared to the synthetic consensus reference. To determine this, we compared this genome against a synthetic reference as described previously [12]. When selecting one of the four mutation classes, a table of C57BL/6J variants is returned (in the same form as for the strain-specific lists) with one additional column: the agreement score. This score consists of two elements and shows how many other strains have the same variant as C57BL/6J or have a different variant at the same position. A gene that was removed in the update due to corrections to the reference genome is Nadk2, which was previously annotated as an SG variant exclusively in C57BL/6J. However, after Mousepost 1.0 was made available, the GRC published sequence patches, changing the reference sequence in these regions. In the new version of the mouse reference genome, mm39, the sequence in the Nadk2 region differs from the mm10 version that was used in Mousepost 1.0. Hence, the Nadk2 variants were present only due to (sequencing) errors and have been removed in Mousepost 2.0. ## Pairwise comparisons The final functionality related to protein-coding variants in the Mousepost 2.0 tool is the pairwise comparison function (‘Pairwise’). Here, the user can directly compare two strains, without involvement of any reference sequence. Both strains to be compared need to be selected as ‘strain A’ and ‘strain B’. To speed up the query or if one has a specific region of interest, such as from a QTL study, a genomic location filter can be added. The result table returned provides an overview of each transcript and how many AA differences there are between the selected strains. By clicking on the numbers provided in that table, the user can directly access more information: selecting the number of differences will add a window with a complete overview of non-identical positions in the form of ‘position StrainA_sequence StrainB_sequence’. The table also shows how many strains in the entire dataset have the sequence of strain A and how many have the sequence of strain B. It is important to note that these may be ‘0’ as the selected strains are not themselves counted. Clicking on these numbers will give a listing of these strains in alphabetical order, and here the chosen SOI will be included, in contrast to the number shown in the table. ## Detailed strain-specific and gene-specific data Each of the result tables from comparisons to the reference strain, be it from the listings per strain or search page, provides an overview summary only and contains several useful links, with the link in the final column leading to gene and genomic location information from UCSC, Ensembl and MGI. The gene names and gene/transcript IDs are internal links that lead to the ‘gene details’ page. On the ‘gene details’ page, all transcripts of the gene are listed as expandable sections giving a complete overview of all information in the database on the transcript. For SG, SL and SC classed transcripts, the information included complete AA sequence of the protein in the reference strain and selected SOI. SC transcripts also have a list of all annotated protein domains that were affected by the truncation. In case of MUT transcripts, there is a table of all variants and their PROVEAN scores. ## A complementary resource to the MGP The Mousepost 2.0 tool also serves as an interesting complementary tool to the data directly provided by the MGP. The MGP data can be accessed through their own variant querying site. This resource provides an extensive search function to query the database based on gene name/ID and genomic coordinates. The main difference between the MGP variant site and Mousepost 2.0 is that the MGP variant site provides a nucleotide focused result set, i.e. an overview of all variants at the level of the genome with links to the variant database from the MGI, where more information can be obtained, such as the effects on the protein, if any. Mousepost 2.0, in contrast, directly allows queries and provides information at the protein level. Mousepost 2.0 also integrates the effects of all variants, because, on several occasions, there are multiple variants affecting a single codon. This is a major difference with the information available in the MGP directly and provides significant added value. The effect is the largest for variants that are annotated as ‘stop gain’ (or nonsense) in the VCF file, as these often have a neighbouring variant (always co-occurring) resulting in a missense variant instead of a nonsense variant. A practical example is the Cracdl gene (ENSMUSG00000026090; reverse strand) in the A/J strain, which is annotated as having an SG variant (MGP VCF file) at genomic position chr1:37664116–37664117, with the actual mutation resulting in a GAA (E) to TAA (STOP). However, this gene also has a variant at the neighbouring position in the same strain (at chr1:37664115–37664116). By itself this variant is annotated as a missense (GAA to GTA: E→V), but both variations are always present together; thus, neither of the annotations in the VCF gives the correct real situation in A/J, which is in fact GAA→TTA resulting in E→L at the AA level, which is the single variant that Mousepost 2.0 will show, from the integration of all nucleotide variants on a per codon basis. MGP reports on a per nucleotide basis instead. We also directly incorporate exon skipping or intron retention results into the sequences for transcripts where splice sites are (predicted to) be affected. This allows us to provide comprehensive protein comparisons between the strains. However, non-coding sequence-related variants (e.g. upstream of genes) cannot be found in Mousepost 2.0, as these are only present at the nucleotide level, but they are already comprehensively annotated in the MGP. Finally, we also integrate data from all available strains into a synthetic reference to interrogate the C57BL/6J strain for mutations, which is not possible with the MGP tools. In contrast, the MGP offers complete reference genome, with all variants included, for several popular strains. ## A complementary resource to the ‘wildmouse’ dataset Mousepost 2.0 may also be used as a resource for users who are more interested in wild mice and wild-derived strains. It serves as a complementary resource to the ‘wildmouse’ dataset, which was published in 2016 by Harr et al. [ 38] and includes information about genomic variation in mice living in the wild, specifically individuals from the species Mus musculus domesticus, Mus musculus helgolandicus, Mus musculus musculus and Mus spretus. These data are available through their own ftp site or in the USCS Genome Browser as a public session (https://genome.ucsc.edu/cgi-bin/hgPublicSessions; search ‘wildmouse’). Using the genomic location of the variants in Mousepost, it is possible to link back to the data in the wildmouse project. The occurrence of variants from inbred strains can thus be checked in wild mice, which can be used to predict loss-of-function variants in wild mouse strains. ## CONCLUSION This work presents a major update of the underlying data on the Mousepost 1.0 resource, based on significantly updated sequencing data of the reference strain as well as the 36 previously studied inbred strains and addition of 16 new mouse inbred strains. The expansion significantly increases the usefulness of Mousepost to the community. ## DATA AVAILABILITY All data in this study are available from public repositories, which are listed and linked in the manuscript. All processed data are made available on the mousepost.be website. ## SUPPLEMENTARY DATA Supplementary Data are available at NAR Online. ## FUNDING Universiteit Gent [BOF.MET.2021.0001.0]. Funding for open access charge: Universiteit Gent [BOF.MET.2021.0001.0]. Conflict of interest statement. None declared. ## References 1. Battey J., Jordan E., Cox D., Dove W.. An action plan for mouse genomics. Nat. Genet. (1999) 21 73-75. PMID: 9916794 2. Waterston R.H., Lindblad-Toh K., Birney E., Rogers J., Abril J.F., Agarwal P., Agarwala R., Ainscough R., Alexandersson M., An P.. Initial sequencing and comparative analysis of the mouse genome. Nature (2002) 420 520-562. PMID: 12466850 3. Poltorak A., He X., Smirnova I., Liu M.Y., Van Huffel C., Du X., Birdwell D., Alejos E., Silva M., Galanos C.. Defective LPS signaling in C3H/HeJ and C57BL/10ScCr mice: mutations in Tlr4 gene. Science (1998) 282 2085-2088. PMID: 9851930 4. Chandler C.H., Chari S., Tack D., Dworkin I.. Causes and consequences of genetic background effects illuminated by integrative genomic analysis. Genetics (2014) 196 1321-1336. PMID: 24504186 5. Dejager L., Libert C., Montagutelli X.. Thirty years of. Trends Genet. (2009) 25 234-241. PMID: 19361882 6. Mahieu T., Park J.M., Revets H., Pasche B., Lengeling A., Staelens J., Wullaert A., Vanlaere I., Hochepied T., van Roy F.. The wild-derived inbred mouse strain SPRET/Ei is resistant to LPS and defective in IFN-beta production. Proc. Natl Acad. Sci. U.S.A. (2006) 103 2292-2297. PMID: 16455798 7. Church D.M., Goodstadt L., Hillier L.W., Zody M.C., Goldstein S., She X.W., Bult C.J., Agarwala R., Cherry J.L., DiCuccio M.. Lineage-specific biology revealed by a finished genome assembly of the mouse. PLoS Biol. (2009) 7 e1000112. PMID: 19468303 8. Adams D.J., Doran A.G., Lilue J., Keane T.M.. The Mouse Genomes Project: a repository of inbred laboratory mouse strain genomes. Mamm. Genome (2015) 26 403-412. PMID: 26123534 9. Keane T.M., Goodstadt L., Danecek P., White M.A., Wong K., Yalcin B., Heger A., Agam A., Slater G., Goodson M.. Mouse genomic variation and its effect on phenotypes and gene regulation. Nature (2011) 477 289-294. PMID: 21921910 10. Steeland S., Timmermans S., Van Ryckeghem S., Hulpiau P., Saeys Y., Van Montagu M., Vandenbroucke R.E., Libert C.. Efficient analysis of mouse genome sequences reveal many nonsense variants. Proc. Natl Acad. Sci. U.S.A. (2016) 113 5670-5675. PMID: 27147605 11. Timmermans S., Van Montagu M., Libert C.. Complete overview of protein-inactivating sequence variations in 36 sequenced mouse inbred strains. Proc. Natl Acad. Sci. U.S.A. (2017) 114 9158-9163. PMID: 28784771 12. Timmermans S., Libert C.. Easy access to and applications of the sequences of all protein-coding genes of all sequenced mouse strains. Trends Genet. (2018) 34 899-902. PMID: 30243593 13. Timmermans S., Libert C.. Overview of inactivating mutations in the protein-coding genome of the mouse reference strain C57BL/6J. JCI Insight (2018) 3 e121758. PMID: 29997285 14. Timmermans S., Libert C.. Coding variants in mouse and rat model organisms: Mousepost and Ratpost. Mamm. Genome (2022) 33 81-87. PMID: 34313794 15. Timmermans S., Libert C.. Ratpost: a searchable database of protein-inactivating sequence variations in 40 sequenced rat-inbred strains. Mamm. Genome (2021) 32 1-11. PMID: 33481094 16. Frankish A., Diekhans M., Jungreis I., Lagarde J., Loveland J.E., Mudge J.M., Sisu C., Wright J.C., Armstrong J., Barnes I.. Gencode 2021. Nucleic Acids Res. (2021) 49 D916-D923. PMID: 33270111 17. Choi Y., Sims G.E., Murphy S., Miller J.R., Chan A.P.. Predicting the functional effect of amino acid substitutions and indels. PLoS One (2012) 7 e46688. PMID: 23056405 18. Rice P., Longden I., Bleasby A.. EMBOSS: the European molecular biology open software suite. Trends Genet. (2000) 16 276-277. PMID: 10827456 19. Edgar R.C.. MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics (2004) 5 113. PMID: 15318951 20. Navarro Gonzalez J., Zweig A.S., Speir M.L., Schmelter D., Rosenbloom K.R., Raney B.J., Powell C.C., Nassar L.R., Maulding N.D., Lee C.M.. The UCSC Genome Browser database: 2021 update. Nucleic Acids Res. (2021) 49 D1046-D1057. PMID: 33221922 21. Frankish A., Diekhans M., Jungreis I., Lagarde J., Loveland J.E., Mudge J.M., Sisu C., Wright J.C., Armstrong J., Barnes I.. Gencode 2021. Nucleic Acids Res. (2021) 49 D916-D923. PMID: 33270111 22. Quinlan A.R., Hall I.M.. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics (2010) 26 841-842. PMID: 20110278 23. Neph S., Kuehn M.S., Reynolds A.P., Haugen E., Thurman R.E., Johnson A.K., Rynes E., Maurano M.T., Vierstra J., Thomas S.. BEDOPS: high-performance genomic feature operations. Bioinformatics (2012) 28 1919-1920. PMID: 22576172 24. Deschepper C.F., Olson J.L., Otis M., Gallo-Payet N.. Characterization of blood pressure and morphological traits in cardiovascular-related organs in 13 different inbred mouse strains. J. Appl. Physiol. (2004) 97 369-376. PMID: 15047670 25. Carrier L., Mearini G., Stathopoulou K., Cuello F.. Cardiac myosin-binding protein C (MYBPC3) in cardiac pathophysiology. Gene (2015) 573 188-197. PMID: 26358504 26. Barefield D., Kumar M., Gorham J., Seidman J.G., Seidman C.E., de Tombe P.P., Sadayappan S.. Haploinsufficiency of MYBPC3 exacerbates the development of hypertrophic cardiomyopathy in heterozygous mice. J. Mol. Cell. Cardiol. (2015) 79 234-243. PMID: 25463273 27. Moayeri M., Martinez N.W., Wiggins J., Young H.A., Leppla S.H.. Mouse susceptibility to anthrax lethal toxin is influenced by genetic factors in addition to those controlling macrophage sensitivity. Infect. Immun. (2004) 72 4439-4447. PMID: 15271901 28. Chavarria-Smith J., Vance R.E.. Direct proteolytic cleavage of NLRP1B is necessary and sufficient for inflammasome activation by anthrax lethal factor. PLoS Pathog. (2013) 9 e1003452. PMID: 23818853 29. Levinsohn J.L., Newman Z.L., Hellmich K.A., Fattah R., Getz M.A., Liu S., Sastalla I., Leppla S.H., Moayeri M.. Anthrax lethal factor cleavage of Nlrp1 is required for activation of the inflammasome. PLoS Pathog. (2012) 8 e1002638. PMID: 22479187 30. Kurosaki T., Popp M.W., Maquat L.E.. Quality and quantity control of gene expression by nonsense-mediated mRNA decay. Nat. Rev. Mol. Cell Biol. (2019) 20 406-420. PMID: 30992545 31. Vasudevan S., Peltz S.W., Wilusz C.J.. Non-stop decay—a new mRNA surveillance pathway. Bioessays (2002) 24 785-788. PMID: 12210514 32. Ge Z., Mehta P., Richards J., Karzai A.W.. Non-stop mRNA decay initiates at the ribosome. Mol. Microbiol. (2010) 78 1159-1170. PMID: 21091502 33. Stephan K., Smirnova I., Jacque B., Poltorak A.. Genetic analysis of the innate immune responses in wild-derived inbred strains of mice. Eur. J. Immunol. (2007) 37 212-223. PMID: 17163447 34. Paigen B., Bouchard G., Carey M.C.. Diet effects on gallstone formation and the assessment of liver morphology, plasma lipids, and atherosclerosis in 44 inbred strains of mice on high-fat atherogenic diet (not under pathogen-free conditions). (2000) 35. Ji L.L., Zhao Y., He L.J., Zhao J., Gao T., Liu F.Z., Qi B.C., Kang F., Wang G., Zhao Y.L.. AKAP1 deficiency attenuates diet-induced obesity and insulin resistance by promoting fatty acid oxidation and thermogenesis in brown adipocytes. Adv. Sci. (2021) 8 2002794 36. Lo C.M., King A., Samuelson L.C., Kindel T.L., Rider T., Jandacek R.J., Raybould H.E., Woods S.C., Tso P.. Cholecystokinin knockout mice are resistant to high-fat diet-induced obesity. Gastroenterology (2010) 138 1997-2005. PMID: 20117110 37. Hadsell D.L., Wei J., Olea W., Hadsell L.A., Renwick A., Thomson P.C., Shariflou M., Williamson P.. Physiol. Genomics (2012) 44 787-798. PMID: 22759921 38. Harr B., Karakoc E., Neme R., Teschke M., Pfeifle C., Pezer Z., Babiker H., Linnenbrink M., Montero I., Scavetta R.. Genomic resources for wild populations of the house mouse,. Sci. Data (2016) 3 160075. PMID: 27622383
--- title: Simultaneous inhibition of endocytic recycling and lysosomal fusion sensitizes cells and tissues to oligonucleotide therapeutics authors: - Brendan T Finicle - Kazumi H Eckenstein - Alexey S Revenko - Brooke A Anderson - W Brad Wan - Alison N McCracken - Daniel Gil - David A Fruman - Stephen Hanessian - Punit P Seth - Aimee L Edinger journal: Nucleic Acids Research year: 2023 pmcid: PMC9976930 doi: 10.1093/nar/gkad023 license: CC BY 4.0 --- # Simultaneous inhibition of endocytic recycling and lysosomal fusion sensitizes cells and tissues to oligonucleotide therapeutics ## Abstract Inefficient endosomal escape remains the primary barrier to the broad application of oligonucleotide therapeutics. Liver uptake after systemic administration is sufficiently robust that a therapeutic effect can be achieved but targeting extrahepatic tissues remains challenging. Prior attempts to improve oligonucleotide activity using small molecules that increase the leakiness of endosomes have failed due to unacceptable toxicity. Here, we show that the well-tolerated and orally bioavailable synthetic sphingolipid analog, SH-BC-893, increases the activity of antisense oligonucleotides (ASOs) and small interfering RNAs (siRNAs) up to 200-fold in vitro without permeabilizing endosomes. SH-BC-893 treatment trapped endocytosed oligonucleotides within extra-lysosomal compartments thought to be more permeable due to frequent membrane fission and fusion events. Simultaneous disruption of ARF6-dependent endocytic recycling and PIKfyve-dependent lysosomal fusion was necessary and sufficient for SH-BC-893 to increase non-lysosomal oligonucleotide levels and enhance their activity. In mice, oral administration of SH-BC-893 increased ASO potency in the liver by 15-fold without toxicity. More importantly, SH-BC-893 enabled target RNA knockdown in the CNS and lungs of mice treated subcutaneously with cholesterol-functionalized duplexed oligonucleotides or unmodified ASOs, respectively. Together, these results establish the feasibility of using a small molecule that disrupts endolysosomal trafficking to improve the activity of oligonucleotides in extrahepatic tissues. ## Graphical Abstract Graphical AbstractWe show that the small molecule SH-BC-893 increases intracellular oligonucleotide levels boosting activity >100-fold in vitro and >10-fold in vivo by simultaneously limiting endosomal recycling and lysosomal accumulation of oligonucleotides. ## INTRODUCTION Oligonucleotide therapeutics have the potential to revolutionize medicine by making almost any target accessible. Oligonucleotides targeting RNA include single-stranded antisense oligonucleotides (ASOs) that base pair with a target RNA to elicit RNaseH-dependent degradation, inhibition of translation, or changes to splicing [1,2]. Double-stranded small interfering RNAs (siRNAs) degrade target RNAs after being loaded into the RNA-induced silencing complex (RISC). Medicinal chemistry optimization of the drug-like properties of ASOs and siRNAs solved historical problems with stability and rapid clearance. At least 13 therapeutic oligonucleotides have been FDA-approved and hundreds are in preclinical development (3–8). Despite these successes, inefficient delivery to targets in the cytosol and nucleus of cells remains a major barrier to the broad application of oligonucleotide therapeutics (2,9–11). Only $1\%$ of the oligonucleotide that is delivered to patients engages its target. The high concentrations required in target tissues are readily achieved in the liver after subcutaneous administration. Extrahepatic tissues generally require local (e.g. intrathecal or aerosol) delivery or frequent administration of high doses to achieve significant target engagement [9]. Conjugation to ligands that bind to cell surface receptors can increase oligonucleotide activity by increasing cellular uptake [9,12,13]. Thus far, the only liganded oligonucleotides that are FDA-approved are N-acetylgalactosamine (GalNAc) conjugates that target hepatocytes. Other ligand-target pairs will require optimization for extrahepatic delivery. Approaches that address post-endocytic blocks to delivery offer an alternate strategy to extend the range of tissues and cells that are accessible to oligonucleotide therapeutics. As large (4–14 kDa) polar molecules, ASOs and siRNAs do not readily diffuse across lipid bilayers; both receptor-targeted and unconjugated oligonucleotides enter cells via endocytosis [9,10]. Oligonucleotides in endocytic vesicles are either recycled back to the extracellular space through exocytosis or progress to lysosomes, the degradative compartment of the cell [9,14,15]. Chemical modifications render therapeutic oligonucleotides resistant to lysosomal nucleases [2,9,10]. Therefore, the majority of endocytosed oligonucleotides accumulate within lysosomes where they are stable but unable to reach their cytosolic targets, although recent evidence suggests slow leakage from lysosomes supports long-term oligonucleotide activity in the liver [16]. Attempts to improve the escape of oligonucleotides from endosomes and lysosomes into the cytosol have met little success. Genetic knockdown screens have failed to identify targetable proteins that significantly enhance oligonucleotide activity [17,18]. Molecules that increase the leakiness of endosomes produce large increases in oligonucleotide activity but have a narrow therapeutic index because permeabilizing endosomes and lysosomes is toxic (19–25). Therefore, novel approaches that improve oligonucleotide escape into the cytosol without lysing endocytic compartments are required to solve the delivery problem that limits the therapeutic use of oligonucleotides. In the absence of permeabilizing agents, oligonucleotides likely escape from endocytic compartments at sites of membrane fission and fusion (18,26–28). During these dynamic membrane remodeling events, the lipid bilayer is deformed to create non-bilayer regions that have increased permeability (29–33). Consistent with this model, pre-lysosomal compartments that undergo high rates of vesicle budding and fusion have been identified as sites of oligonucleotide escape (17,34–36). Escape from lysosomes is much less efficient because the limiting membrane is heavily decorated with glycoproteins (e.g. LAMP1 and LAMP2) and glycolipids that reduce permeability relative to other endocytic structures [37,38]. Approaches that increase oligonucleotide uptake and/or residency time in pre-lysosomal compartments where oligonucleotide escape is most efficient could offer significant gains in potency that would make extrahepatic tissues therapeutically accessible. We have published that the synthetic sphingosine analog SH-BC-893 disrupts endocytic recycling by inactivating the small GTPase ARF6 and blocks lysosomal fusion reactions that depend on the lipid kinase PIKfyve (39–42). We hypothesized that simultaneous disruption of these endolysosomal trafficking pathways would synergistically increase oligonucleotide activity by causing accumulation within pre-lysosomal compartments where endosomal release is most efficient (Figure 1A). Importantly, these changes in trafficking are well tolerated as SH-BC-893 is non-toxic at the effective dose even with chronic administration (39–41). Here, we demonstrate that the parallel actions of SH-BC-893 on endocytic recycling and lysosomal fusion are necessary and sufficient to increase intracellular oligonucleotide levels in extra-lysosomal compartments and significantly enhance oligonucleotide activity both in vitro and in vivo with no toxicities detected. Figure 1.: SH-BC-893 increases oligonucleotide accumulation in non-lysosomal compartments. (A) Model showing how SH-BC-893 (893) alters intracellular trafficking. (B) Phase contrast images of HeLa cells treated with SH-BC-893 (5 μM) for 3 h. Scale bar = 20 μm. (C) Viability measured by vital dye (DAPI) exclusion through flow cytometry in HeLa cells treated with indicated concentrations of SH-BC-893 for 24 h. Arrow indicates concentration used in all in vitro oligonucleotide assays. Mean ± SD shown, n = 3. (D) HeLa cells treated with a 3–10–3 cEt ASO targeting MALAT1 (2 μM) ± SH-BC-893 (5 μM) for 6 h and stained with antibodies to endogenous LAMP2 or PS-ASOs. Scale bar = 20 μm. For inset, scale bar = 10 μm. (E) Quantification of the raw intensity values for ASO from images in (D) within LAMP2-positive, LAMP2-negative, and total cellular areas. At least 100 cells were quantified from each of 2 independent experiments. Using a Mann–Whitney t test to correct for data that is not normally distributed, *P < 0.001. ## Cell lines and cell culture MDA-MB-468, MDA-MB-231, SW620, NCI-H358, A549, BxPC3 and PANC1 were obtained from the ATCC. HeLa cells were obtained from Steve Caplan (University of Nebraska Medical Center, Omaha, NE, USA). p53–/– MEFs were generated in-house from embryos from C57BL/6 mice (stock no. 008462, The Jackson Laboratory) using standard techniques. All cells were maintained at 37°C in $5\%$ CO2. HeLa, MEFs, A549, BxPC3 and PANC1 cells were cultured in DMEM media supplemented with $10\%$ fetal bovine serum (FBS) without antibiotics. MDA-MB-468, MDA-MB-231 and SW620 were cultured in DMEM media supplemented with $10\%$ FBS and $1\%$ sodium pyruvate without antibiotics. All cells were maintained in culture for no more than 3 weeks before low-passage vials were thawed. Mycoplasma testing was performed monthly using the published PCR protocol from Uphoff and Drexeler [43]. ## Chemicals and reagents SH-BC-893 was synthesized by IntelliSyn RD (Montreal, Quebec, Canada) following the methods in [44] and with advice from S. Hanessian. The following chemicals were purchased: YM201636 (Cayman Chemicals, cat# 13576), apilimod (SelleckChem, cat# S6414), NAV2729 (R&D systems, cat# 5986), SecinH3 (Cayman Chemicals, cat# 10009570), perphenazine (PPZ, Sigma, cat# P6402-1G), UNC10217938A (Medchemexpress, cat# HY-136151), 6BIO (Cayman Chemicals, cat# 13123), AZD8055 (Cayman Chemicals, cat# 16978) and retro-2 (Sigma, cat# SML1085-5MG). Stock solutions were prepared as follows, aliquoted, and stored at –20°C: SH-BC-893 (5 mM in H2O), YM201636 (1.6 mM in DMSO), apilimod (100 μM in DMSO), NAV2729 (12.5 mM in DMSO), SecinH3 (30 mM in DMSO), perphenazine (50 mM in DMSO), UNC10217938A (10 mM in DMSO), 6BIO (15 mM in DMSO), AZD8055 (1 mM in DMSO), and retro-2 (100 mM in DMSO). All chemical structures of compounds are shown in Supplementary Figure S1. Oligonucleotides used are shown in Supplementary Tables 1, 3 and 4 and were obtained from Ionis Pharmaceuticals. ## Fluorescence microscopy sample preparation Cells were seeded into glass bottom 8-chamber slides at 12 000 cells per chamber (Cellvis, cat# C8-1.5H-N). 16–24 h after seeding, cells were incubated with ASOs (2 μM) for indicated time points. Post-incubation, cells were washed three times with PBS, fixed in $4\%$ paraformaldehyde (VWR, cat# 100503–917) for 15 min, and then washed again with PBS. For samples where no other proteins were immunostained, cells were imaged following DAPI staining (1 mg/ml in PBS, VWR, cat# 422801-BL) for 5 min. For samples where other proteins or molecules were immunostained (e.g. LAMP1, LAMP2, PS-ASOs or myc-tag), samples were permeabilized and incubated in blocking solution for 30 min at RT ($10\%$ FCS, $0.3\%$ saponin, $0.05\%$ azide in PBS). Samples were incubated with primary antibody diluted in block solution for 1 h at RT or overnight at 4°C, washed three times with PBS, and then stained in Alexa Fluor-conjugated secondary antibody solution at RT for 1 h with rocking. Samples were washed again three times, stained with DAPI, washed, and then imaged in PBS by confocal microscopy. Antibodies: anti-LAMP1 (Cell Signaling Technologies, cat# 9091S, 1:400 dilution), anti-LAMP2 (Developmental Studies Hybridoma Bank, cat# H4B4, 1:200 dilution), anti-PS-ASOs (provided by Ionis Pharmaceuticals, 1:200 dilution), anti-myc-tag (Cell Signaling Technologies, cat# 2278S, 1:200 dilution), Alexa Fluor 594 goat anti-mouse (Fisher Scientific, cat# A11032, 1:200 dilution), and Alexa Fluor 594 goat anti-rabbit (Fisher Scientific, cat# A-11012, 1:200 dilution). ## Microscopy, image analysis, and quantification All microscopy images were collected with ZEN digital imaging software on a Zeiss LSM780 confocal microscope with a Plan-Apochromat 63×/1.40 Oil DIC objective or a Zeiss LSM900 with Airyscan 2 with a Plan-Apochromat 63×/1.40 Oil DIC objective. All quantification of microscopy data was performed using ImageJ. In brief, regions of interest (ROIs) enclosing individual cells in each field of view were drawn using cell autofluorescence to define cell boundaries. LAMP$\frac{1}{2}$-positive area was defined by turning LAMP1 or LAMP2 fluorescent images into a binary image and utilizing ImageJ to automatically draw a ROI around each LAMP1- or LAMP2-positive lysosome. LAMP$\frac{1}{2}$-negative area was defined by subtracting the LAMP$\frac{1}{2}$-positive ROI from the ROI enclosing the total cell. ASO fluorescence that did or did not colocalize with LAMP$\frac{1}{2}$-positive pixels was measured as Integrated Density per cell. Total intracellular ASO fluorescence was measured by adding the fluorescence of ASOs within LAMP$\frac{1}{2}$-positive or LAMP$\frac{1}{2}$-negative ROIs. For cytoplasmic ASO quantification, endosomal ASO signal was eliminated by generating regions of interest (ROIs) on thresholded images. With the diffuse cytoplasmic signal remaining, the Integrated Density per cell was measured. Background subtraction was performed by quantifying fluorescent signal in cells that were not exposed to ASOs but stained with both primary and secondary antibodies. At least 100 cells per experiment from a total of 2–3 independent experiments were analyzed for each experiment. ## RNA isolation and RT-qPCR To monitor ASO or siRNA activity in vitro, 3000 cells were plated in duplicate or triplicate wells of a 96-well flat bottom plate. After 16–24 h, cells were treated with a serial dilution of ASOs starting at 20 μM and including 3-fold serial dilutions. Cells were lysed 24 h after ASO addition in GTC lysis buffer (4 M guanidine isothiocyanate, 50 mM Tris–HCl pH 7.5, 25 mM EDTA). RNA was immediately purified or lysates were stored in this buffer at –20°C for later purification. Samples were mixed with an equal amount of $70\%$ EtOH and pipetted into a Pall 96-well glass fiber filter plate (VWR, cat# 97052-104) fitted onto a vacuum manifold. Samples were then washed twice with RNA wash buffer (60 mM potassium acetate, 10 mM Tris–HCl pH 7.5, $60\%$ EtOH) before digesting genomic DNA using DNaseI (Fisher, cat# 18047019) in DNaseI buffer (400 mM Tris–HCl pH 7.5, 100 mM NaCl, 100 mM CaCl2, 100 mM MgCl2) for 15 min at RT. Samples were then washed twice with GTC wash buffer (1 M guanidine isothiocyanate, 12.5 mM Tris–HCl (pH 7.5), 6.25 mM EDTA) followed by three washes in RNA wash buffer. After the final wash, residual buffers were removed from the glass fiber plate by spinning in a tabletop centrifuge for 5 min at 3696 × g. RNA was eluted in nuclease-free water into a 96-well round bottom plate. Following RNA elution, RT-qPCR reaction plates were set up by loading 5 μl of RNA into each well and using the Taqman AgPath-ID™ One-Step RT-PCR Reagents (Fisher, cat# 4387391) according to supplier's recommendations. Taqman primers and probes are described in Supplementary Table 2. RNA levels were quantified using a StepOnePlus Real-Time PCR system (Applied Biosystems) and normalized to total RNA loaded using Quant-iT™ RiboGreen™ RNA Assay Kit (Invitrogen, cat# R11490) according to supplier's recommendations. RNA levels are expressed relative to control or drug-treated samples that received no ASOs. Two to three technical replicates each were performed for each ASO activity assay. At least three independent experiments were performed to produce biological replicates. To monitor ASO activity in tissues in vivo, ∼10–100 mg of flash-frozen tissue were lysed in 1 ml of Trizol using a Bullet Blender Storm Pro Tissue Homogenizer (Next Advance, cat# BT24M). After complete tissue homogenization, samples were transferred to a new Eppendorf tube and incubated at RT for 5 min to permit complete dissociation of nucleoprotein complexes. Following addition of 0.2 ml of chloroform to the Trizol-homogenate, the tube was inverted vigorously 3–4 times and then by vortexing for 20 s. Samples were then centrifuged at 13 000 rpm in a microcentrifuge at 4°C for at least 15 min. Following centrifugation, the upper aqueous layer containing RNA was transferred to a new tube. RNA was precipitated by adding 0.5 ml of ice-cold isopropanol followed by centrifugation at 13 000 rpm in a microcentrifuge at 4°C for 30–60 min. The resulting RNA pellet was washed twice with 1 ml of $70\%$ EtOH (centrifuging for 5 min at 13 000 rpm to clear washes). The washed RNA pellet was dissolved in 40–100 μl of nuclease-free water, and then quantified using a NanoDrop™ 2000 spectrophotometer. To achieve pure RNA, approximately 10 μg of RNA was treated with DNaseI (Zymo Research Corporation, cat# E1010) and then cleaned up on spin columns (New England Biolabs, cat# T2030L) according to the supplier's recommendations. qRT-PCR reaction plates were set up by loading 50 ng of RNA into each well and using the Taqman AgPath-ID™ One-Step RT-PCR Reagents (Fisher, cat# 4387391) according to the supplier's recommendations. Taqman primers and probes are described in Supplementary Table 2. RNA levels were quantified using a StepOnePlus Real-Time PCR system (Applied Biosystems) and expressed relative to housekeeping gene Ppia using the 2−ΔΔCt method prior to normalization as described in the figure legends. ## Viability measurement by flow cytometry Viability was measured by vital dye exclusion using DAPI. In brief, adherent cells were plated in 24-well plates at 30 000 cells per well (matching the confluency of the cells in 96-well plates in the RT-qPCR experiments). After 16–24 h, cells were treated with indicated compounds at indicated concentrations for 24 h. After 24 h, the following was collected in FACS tubes: media containing floating dead cells, a PBS wash, trypsinized cells, and a final media wash to collect all remaining cells and debris. This live and dead cell containing solution was pelleted by centrifugation and the supernatant discarded. The pellet was placed back in single-cell suspension with 0.6 ml of media containing 1 μg/ml of DAPI. Samples were evaluated on a BD Fortessa X20. ## Plasmids and stable cell line generation pQCXIP-mCherry and pQCXIP-mCherry-VAC14 were generously provided by Thomas Weide (University Hospital of Muenster, Germany). The pQCXIB Firefly Luciferase plasmid was a gift from Reuben Shaw (Addgene plasmid # 74445; http://n2t.net/addgene:74445; RRID:Addgene_74445). myc-GRP1DD was produced by PCR amplification of the insert in pEF6-myc-GRPS155D/T240D (a generous gift from Victor Hsu, Harvard Medical School, Boston, MA, USA) and subsequent Gateway cloning into pQCXIB CMV/TO DEST (39–41), a gift from Eric Campeau & Paul Kaufman (Addgene plasmid # 17400; http://n2t.net/addgene:17400; RRID:Addgene_17400). Stable cell lines were generated by transducing target cells with retrovirus and drug selection. ## General animal procedures All experiments measuring ASO activity in animals were approved by the Institutional Animal Care and Use Committee of University of California, Irvine. Six- to eight-week old, male Balbc/J mice were purchased from the Jackson Laboratory and acclimated to the university vivarium for at least 7 days prior to experimentation. Mice were housed under a 12 h light/dark cycle at 20–22°C in groups of 2–4. Cages contained $\frac{1}{8}$′ corncob bedding (7092A, Envigo, Huntingdon, UK) enriched with ∼6 g of cotton fiber nestlets (Ancare, Corp., Bellmore, NY). Mice were fed the vivarium stock diet (chow, 2020x, Envigo). Access to food and water was ad libitum. For oral administration of SH-BC-893, polypropylene feeding tubes (20 g × 38 mm; Instech Laboratories Inc., Plymouth, PA) were used to dose 120 mg/kg of SH-BC-893 dissolved in H2O (stock = 24 mg/ml). To aid gavage by inducing salivation, feeding tubes were dipped into a 1 g/ml sucrose solution immediately prior to treatment. For subcutaneous administration, ASOs were dissolved in PBS at a concentration such that 10 ml/kg was administered using a 27 G needle. For blood chemistry analysis, blood was collected by decapitation from nine-week old male Balbc/J mice. Serum was separated from whole blood in a SST-MINI tube with clot activator gel (Greiner, cat# 450571VET). Serum samples were sent to IDEXX Bioanalytics for a comprehensive chemistry panel (cat# 6006). ## Tissue distribution and pharmacokinetics of SH-BC-893 Tissue PK studies were performed at Pharmaron (Beijing, China) on a fee-for-service basis and were approved by their Institutional Animal Care and Use Committee. Six- to eight-week old male or female CD1 mice were treated daily for 5 days with 120 mg/kg SH-BC-893 dissolved in H2O (P.O., stock = 12 mg/ml) and then sacrificed 0.5, 1, 2, 4, 8 or 24 h after the last dose. Tissues were perfused with 10 mL saline and prior to collection and snap frozen in liquid nitrogen. Frozen tissue was homogenized in PBS (W/V 1:4). 10 μl of tissue or plasma homogenate was mixed with 10 μl blank solution and added to 200 μl acetonitrile containing FTY720 as an internal standard. Samples were vortexed and then spun for 30 min at 4700 rpm at 4°C to precipitate protein. The resultant supernatant was diluted 2-fold with water and 10 μl of diluted supernatant was injected into the LC–MS/MS for quantitative analysis. SH-BC-893 was quantified by LC–MS/MS using a HALO 90A AQ-C18, 2.7 μm 2.1 × 50 mm column and an AB Sciex Triple Quad 5500 LC–MS/MS instrument (serial no. BB214861610) and corrected for extraction efficiency using the FTY720 internal standard. ## Statistical analysis For microscopy experiments, box and whisker plots showing median and quartiles are presented because data was not normally distributed. In bar graphs depicting ASO or siRNA IC50s or target RNA levels in tissues, mean ± SD is presented. All experimental data are from ≥ 3 independent biological replicates except where otherwise indicated. Statistical analysis was performed using GraphPad Prism software. Corrections for multiple comparisons were made as indicated in the legends and adjusted P-values reported: n.s., not significant, P ≥ 0.05; * $P \leq 0.05$; $P \leq 0.01$; * $P \leq 0.001$; key comparisons are shown in the figures. ## SH-BC-893 increases ASO delivery and activity in vitro SH-BC-893 (Supplementary Figure S1) disrupts endocytic recycling by inactivating the small GTPase ARF6 and blocks lysosomal fusion reactions that depend on the lipid kinase PIKfyve (39–42), activities that we predicted would synergize to promote oligonucleotide delivery (Figure 1A). HeLa cervical carcinoma cells were used initially to test this hypothesis because they are well suited to confocal microscopy and widely used to study both intracellular trafficking and oligonucleotide delivery. As expected, SH-BC-893 disrupted endolysosomal trafficking in HeLa cells at non-toxic concentrations (Figure 1B, C). While toxic in vitro at high concentrations, SH-BC-893 is well tolerated in mice at the effective dose (120 mg/kg) even with chronic administration [39,41]. Initial studies used a phosphorothioate (PS) gapmer complementary to the long non-coding RNA (lncRNA) MALAT1 with 10 deoxynucleotides in the center flanked on each side by 3 nucleotides containing riboses with a 2′,4′ constrained ethyl (cEt) group (Supplementary Table S1); this modification is common among preclinical ASOs and is currently being evaluated in patients [9]. ASO localization was tracked using a polyclonal antibody specific for PS ASOs [17]. ASO localization was evaluated by confocal fluorescence microscopy after gymnotic delivery (‘free uptake’). In control cells, the majority of internalized ASOs colocalized with the lysosomal marker LAMP2 after 6 h (Figure 1D, E and Supplementary Figure S2A). Co-treatment with SH-BC-893 (5 μM) reduced the percent of ASOs in LAMP2-positive lysosomes from $70\%$ to $9\%$. Moreover, SH-BC-893 treatment increased the total amount of intracellular ASOs by 4-fold (Figure 1D, E). Reducing the delivery of ASOs to LAMP2-positive lysosomes and concomitantly increasing intracellular ASO levels translated to a >250-fold increase in the amount of ASOs in extra-lysosomal, LAMP2-negative compartments. Similar results were obtained with an untagged 2′-O-methoxyethyl (2′MOE) PS-gapmer, an ASO tagged on the 5’ end with a 6-carboxyfluorescein (FAM) fluorophore, and an antibody against another lysosomal marker, LAMP1 (Supplementary Table 1 and Supplementary Figure S2B–G). In sum, co-incubation with SH-BC-893 increased the total amount of intracellular ASOs and dramatically reduced colocalization with lysosomal markers. ASOs must escape from endosomal structures to produce knockdown. It is estimated that 1–$4\%$ of the oligonucleotides that are endocytosed escape to the cytoplasm [10,45,46]. Such a small quantity is hard to observe in fluorescence microscopy images that contain very bright endosomal signals. To better illustrate changes in the low, cytoplasmic levels of ASOs, the endosomal ASO signal was removed and the cytoplasmic signal quantified (Supplementary Data Figure S3A). Background subtraction was performed using cells that were stained with both primary and secondary antibodies but not exposed to ASOs. In control cells, ∼$3\%$ of the ASOs were cytoplasmic which is consistent with prior estimates (Supplementary Figure S3B, C and [10,45,46]). SH-BC-893 treatment increased the cytosolic fraction to $10\%$ which, given the increase in intracellular ASOs, translated to a 9-fold increase in the absolute amount of cytoplasmic ASO (Supplementary Figure S3B–D). In summary, SH-BC-893 significantly increased the amount of ASO that reaches the cytoplasm. Whether this increase in cytoplasmic ASOs translated into increased target RNA degradation was assessed by RT-qPCR under the same experimental conditions. Co-treatment with SH-BC-893 increased MALAT1 knockdown in cells treated with a cEt gapmer ASO (2 μM) from $25\%$ to $85\%$ (Supplementary Figure S4A); similar potentiation was observed with a 2’MOE gapmer ASO (Supplementary Figure S4B). To quantify the degree of potentiation more rigorously [47], IC50s were determined for both cEt and 2’MOE ASOs in the presence or absence of SH-BC-893. With gymnotic delivery in HeLa cells, cEt and 2’MOE gapmer IC50s were 19 μM and 13 μM, respectively (Figure 2A–D). Co-incubation with SH-BC-893 shifted the dose response curves to the left by 2 logs, reducing the ASO IC50s to 171 nM (cEt) or 61 nM (2’MOE), a 111- or 215-fold increase in activity. ASO potentiation by SH-BC-893 was equally robust when normalized to total RNA or to the housekeeping gene ACTB (Figure 2E). Additionally, ASO potentiation by SH-BC-893 was dose responsive, ranging from an 11-fold increase in ASO activity at 2.5 μM to a ∼100-fold increase at 5 and 10 μM (Figure 2F). Notably, SH-BC-893-mediated potentiation plateaued at 5 μM, a non-toxic concentration (Figure 1C). Figure 2.: SH-BC-893 enhances oligonucleotide activity. (A) MALAT1 levels in HeLa cells treated with the indicated concentrations of 3–10–3 cEt gapmer targeting MALAT1 ± SH-BC-893 (5 μM) for 24 h. Mean ± SD shown, n = 6. (B) IC50s from each biological replicate in (A); mean ± SD shown. Using a Welch's t test to correct for unequal SD, P < 0.001. (C, D) As in (A, B), except using a 5–10–5 2’MOE gapmer; mean ± SD shown, n = 3. Using a Welch's t test to correct for unequal SD’s, P < 0.01. (E) As in (A), except data expressed relative to total RNA or to housekeeping gene ACTB, n = 1. (F) As in (A), except with indicated concentrations of 893, n = 1. (G) As in (A), except using a 3–10–3 cEt gapmer targeting the ACTN1 mRNA and n = 3. IC50 values for control could not be calculated due to the low basal activity of this ASO. (H) The indicated cell lines were treated with the cEt MALAT1 ASO ± SH-BC-893 (5 μM) for 24 h and an IC50 (nM) calculated. Dose response curves shown in Supplementary Figure S5. (I) HPRT1 mRNA levels in HeLa cells treated with the indicated doses of a palmitate-conjugated siRNA targeting HPRT1 ± SH-BC-893 (5 μM) for 24 h. Mean ± SD shown, n = 3. (J) IC50s from each biological replicate in (I); mean ± SD shown. Using a Welch's t test to correct for unequal SD, **P < 0.001. In our experience, the MALAT1 ASOs widely used in proof-of-concept studies are 2–3-fold more potent than most ASOs targeting other sequences. Importantly, SH-BC-893 increased the activity of oligonucleotides of other sequences in different cell types. SH-BC-893 enhanced the activity of a cEt gapmer targeting α-actinin-1 (ACTN1) >63-fold (Figure 2G). In addition to HeLa cells, potentiation was observed in SH-BC-893-treated mouse embryonic fibroblasts and in cancer cell lines derived from tumors of the breast, colon, lung, or pancreas (Figure 2H and Supplementary Figure S5). The same barriers limit ASO and siRNA delivery to their targets in the cytosol and nucleus [14,48]. SH-BC-893 also improved the activity of a palmitate-conjugated, nuclease-resistant siRNA targeting HPRT1 by 20-fold, shifting the IC50 from 1.9 μM to 97 nM (Figure 2I, J). These results suggest that SH-BC-893 could improve oligonucleotide activity across multiple platforms and in extrahepatic tissues. ## SH-BC-893 is less toxic or more effective than previously identified oligonucleotide-potentiating small molecules Small molecules that enhance endosomal release of oligonucleotides through lysis have been described. For example, the small molecule UNC10217938A dramatically enhances the activity of ASOs, siRNAs, and splice-switching oligonucleotides by permeabilizing endosomes and lysosomes [19]. However, the therapeutic use of endolytic agents is limited by their toxicity. Under our standard assay conditions where cells are continuously exposed to ASOs and compound for 24 h, the effective dose of UNC10217938A [19] was cytotoxic, killing $90\%$ of the cells (Supplementary Figure S6A, B). In [19], toxicity was avoided by pre-loading cells with ASOs for 16 h and then pulsing cells with UNC10217938A for only 2 h. Under these conditions, UNC10217938A produced profound ASO potentiation without cell death (Supplementary Figure S6C, D). In contrast, SH-BC-893 did not increase ASO activity when added after ASO had reached lysosomes. SH-BC-893’s lack of toxicity at the efficacious dose (Figure 1C, Figure 2F, and Supplementary Figure S6B) and the observation that SH-BC-893 is not effective once ASO are sequestered in lysosomes (Supplementary Fig. S6D) suggests that SH-BC-893 does not function as an endolytic agent like UNC10217938A. To directly test whether SH-BC-893 permeabilizes endosomal membranes, cells pre-loaded with ASO and 10 kDa dextran were evaluated by microscopy for release into the cytosol and nucleus (Supplementary Figure S6E,F). At their effective doses, UNC10217938A, but not SH-BC-893, released both ASOs and dextran from endosomes. Together, this data shows that SH-BC-893 does not enhance ASO activity by permeabilizing lysosomes and is consistent with our proposed mechanism of action: trapping ASOs in a pre-lysosomal compartment from which oligonucleotide escape is naturally more efficient (Figure 1A). Several other small molecules have been reported to enhance ASO and/or siRNA activity without permeabilizing endosomes or lysosomes. For example, the GSK3 inhibitor 6BIO enhances ASO and siRNA activity through an unknown mechanism [23], the mTOR kinase inhibitor AZD8055 increases ASO activity by stimulating autophagy [49], and the retrograde trafficking inhibitor retro-1 increases the activity of ASOs [21]. Because the assay conditions and cell lines used in these publications vary, we directly compared the ASO potentiating ability of these compounds and SH-BC-893 in HeLa cells treated with an RNaseH-dependent ASO. Notably, many prior studies utilized splice-switching ASOs that trigger expression of luciferase or GFP, reporter assays that give a larger fold-change in activity than would an equivalent effect size in an RNaseH-dependent assay measuring RNA knockdown. Using the concentrations employed in the prior publications, 6BIO, AZD8055 and retro-2 enhanced ASO activity 3-, 4- or 5-fold, respectively, while SH-BC-893 increased ASO activity 130-fold under the same conditions (Supplementary Figure S6G-H); the structurally-related molecule retro-2 was utilized because retro-1 is not commercially available [50]. None of these molecules were cytotoxic under these assay conditions (Supplementary Figure S6I). In summary, SH-BC-893 is less toxic than endosome-permeabilizing agents and more effective than previously identified small molecule ASO potentiators that do not lyse endocytic structures. ## Simultaneous PIKfyve and ARF6 inhibition is both necessary and sufficient for ASO potentiation by SH-BC-893 The mechanism by which SH-BC-893 promotes antisense activity was next evaluated. We hypothesized that the previously reported effects of SH-BC-893 on endocytic trafficking, disrupting ARF6-dependent endocytic recycling and PIKfyve-dependent lysosomal fusion reactions [39,40], were responsible for the observed oligonucleotide potentiation (Figure 1A). To test this model, the impact of selective PIKfyve (YM201636 and apilimod) or ARF6 (SecinH3 and NAV2729) inhibitors on ASO uptake and localization was compared to SH-BC-893 (structures provided in Supplementary Figure S1). These structurally distinct inhibitors are unlikely to have the same off-target effects as each other or as SH-BC-893, thereby increasing confidence that any shared effects on oligonucleotide trafficking and activity are due to ARF6 or PIKfyve inhibition. At concentrations previously established to fully inhibit their targets [39,40], the two ARF6 inhibitors increased intracellular ASO levels to a similar extent as SH-BC-893 (Figure 3A, B). In contrast, PIKfyve inhibition with YM201636 or apilimod slightly decreased ASO accumulation within cells. ARF6 promotes endosomal recycling [15,40]. To determine whether ARF6 inhibition increased intracellular ASO levels by reducing their recycling out of the cell (Figure 1A), a pulse-chase protocol was designed. Cells were pulsed with ASOs for 1 h in the absence of inhibitors, washed, and then maintained in medium lacking ASOs but containing vehicle, SH-BC-893, an ARF6 inhibitor, or a PIKfyve inhibitor for an additional 2 h (Figure 3C, D). Control cells treated with vehicle lost >$80\%$ of the ASO signal during the 2 h chase indicating that much of the ASO that enters cells is recycled. The loss of signal during the chase was not due to quenching of the 6-FAM fluorophore as similar results were obtained with untagged ASOs detected with an antibody recognizing PS ASOs (Supplementary Figure S7A, B). In contrast to the >$80\%$ loss of signal in control cells, only 25–$30\%$ of the ASO signal was lost when cells were chased in medium containing SH-BC-893 or the ARF6 inhibitors SecinH3 and NAV2729 (Figure 3C, D). Thus, ARF6 inhibition is sufficient to account for SH-BC-893’s ability to increase intracellular ASO levels and reduce ASO recycling. Figure 3.: Simultaneous PIKfyve and ARF6 inhibition is both necessary and sufficient to recapitulate the effects of SH-BC-893 on ASO uptake and localization. (A) FAM-tagged cEt 3–10–3 ASO and LAMP1 localization in HeLa cells treated with SH-BC-893 (5 μM), NAV2729 (12.5 μM), SecinH3 (30 μM), YM201636 (800 nM), apilimod (100 nM), or both NAV2729 and YM20636 for 6 h. (B) Quantification of the total intracellular ASO fluorescence intensity from the images in (A). At least 100 cells were quantified from each of 3–4 independent experiments. Because the data is not normally distributed, a Kruskal–Wallis ANOVA was used with Dunn's test to correct for multiple comparisons. P < 0.001. (C) HeLa cells were pulsed with FAM-tagged 3–10–3 cEt ASO (2 μM) for 1 h, washed, and then chased in media containing vehicle (DMSO), SH-BC-893 (5 μM), NAV2729 (12.5 μM), SecinH3 (30 μM), or YM201636 (800 nM) for 2 h prior to imaging. (D) Quantification of the intracellular ASO fluorescence of cells in (C). At least 100 cells were quantified from each of two independent experiments. Because data is not normally distributed, a Kruskal–Wallis ANOVA was used with Dunn's test to correct for multiple comparisons. P < 0.001. (E) HeLa cells expressing luciferase or GRP1DD were subjected to an ASO pulse-chase as in (C). (F) Quantification of the intracellular ASO fluorescence intensity in (E) performed as in (D). Scale bars, 20 μm (A and E) or 10 μm (inset in A and in C). (G) As in (B), except ASO fluorescence intensity within LAMP1-positive pixels is measured. P < 0.01; *P < 0.001. Results using ARF6 inhibitors were validated with genetic approaches. The classic knockdown/knockout target validation experiments could not be performed because knockdown of ARF6 compromised cellular health to an extent that would confound the interpretation of results. We have previously shown that inhibition of ARF6-mediated endocytic recycling by SH-BC-893 occurs downstream of activation of the serine and threonine protein phosphatase 2A (PP2A) [40]. ARF6 is activated by its guanine nucleotide exchange factor GRP1 which is in turn inactivated by PP2A-dependent dephosphorylation [40,51]. Replacing serine 255 and threonine 280 with phosphomimetic aspartic acid residues (GRP1DD) renders GRP1 resistant to inactivation by PP2A and restores recycling in SH-BC-893-treated cells. Consistent with a model where SH-BC-893 increases intracellular ASO levels by inactivating ARF6, SH-BC-893 failed to block ASO recycling in GRP1DD expressing cells (Figure 3E, F). Thus, chemical and genetic approaches indicate that ARF6 inactivation is both necessary and sufficient to explain the ability of SH-BC-893 to boost intracellular ASO levels. Although the ARF6 inhibitors SecinH3 and NAV2729 increased intracellular ASO levels, they did not block delivery to lysosomes like SH-BC-893 (Figure 3A, G). In contrast, the PIKfyve inhibitors YM201636 and apilimod reduced the amount of ASOs within lysosomes by 80–$90\%$ similar to SH-BC-893. However, these PIKfyve inhibitors did not increase intracellular ASO levels and had only a minimal effect on recycling (Figure 3A–D). In keeping with the model in Figure 1A, combining ARF6 and PIKfyve inhibition was necessary and sufficient to recapitulate the full effects of SH-BC-893 on ASO localization. Cells treated with both the ARF6 inhibitor NAV2729 and the PIKfyve inhibitor YM201636 had more intracellular ASOs outside of lysosomes matching the effects of SH-BC-893. Taken together, these experiments indicate that inhibiting ARF6 and PIKfyve-dependent trafficking in parallel accounts for effects of SH-BC-893 on intracellular ASO accumulation and trafficking as proposed in Figure 1A. To determine the extent to which increased intracellular ASO levels and extra-lysosomal localization contribute to the improved ASO activity in SH-BC-893-treated cells (Figure 2 and Supplementary Figures S4 and S5), the effects of ARF6 inhibitors and PIKfyve inhibitors alone and in combination on ASO activity were assessed by RT-qPCR. Despite increasing the amount of intracellular ASOs to a similar extent as SH-BC-893 (Figure 3A, B), the ARF6 inhibitors NAV2729 and SecinH3 failed to increase ASO activity (Figure 4A,B and Supplementary Figure S8A,B). Lack of potentiation most likely reflects ASO accumulation in lysosomes when only ARF6 is inhibited (Figure 3A, G) as escape from this site is expected to be inefficient. Although they reduced lysosomal co-localization to a similar degree as SH-BC-893 (Figure 3A,G), the PIKfyve inhibitors YM201636 or apilimod increased ASO activity by only 3–4-fold (Figure 4A, B and Supplementary Figure S8A, B). The failure of PIKfyve inhibitors to increase ASO activity to a similar extent as SH-BC-893 likely reflects their inability to increase intracellular ASO accumulation (Figure 3A-B). Combining the ARF6 inhibitor NAV2729 with the PIKfyve inhibitor YM201636 produced marked synergy, improving ASO activity to the same extent as SH-BC-893 (Figure 4A, B). Importantly, SH-BC-893, the ARF6 inhibitors, the PIKfyve inhibitors, and the combination were not toxic to cells at the concentrations that disrupt endolysosomal trafficking and modulate ASO activity (Supplementary Figure S8C). As for SH-BC-893 (Figure 2H–J), the synergy between ARF6 inhibitors and PIKfyve inhibitors was not cell type- or platform-specific. PIKfyve/ARF6 inhibitor combinations also improved the activity of a Malat1-targeting cEt gapmer in MEFs and a HPRT1-targeting siRNA in HeLa cells (Supplementary Figure S8D, E and Figure 4C, D). In sum, structurally distinct chemical inhibitors and genetic studies indicate that simultaneous ARF6 and PIKfyve inhibition synergistically improve ASO activity. Figure 4.: Simultaneous PIKfyve and ARF6 inhibition is both necessary and sufficient to account for the increase in ASO and siRNA activity in SH-BC-893-treated cells. (A) MALAT1 levels in HeLa cells treated with the indicated concentrations of cEt gapmer ASO targeting MALAT1 ± SH-BC-893 (5 μM), NAV2729 (12.5 μM), YM201636 (800 nM), or both for 24 h. Mean ± SD shown, n = 3. (B) IC50s from each biological replicate in (A); mean ± SD shown. Due to unequal SD, a Brown–Forsythe and Welch ANOVA test was used with Dunnett's T3 test to correct for multiple comparisons; P < 0.01. (C) HPRT1 mRNA levels in HeLa cells treated with the indicated concentrations of a palmitate-conjugated siRNA targeting HPRT1 ± SH-BC-893 (5 μM), NAV2729 (12.5 μM), apilimod (100 nM) or NAV2729 and apilimod for 24 h. Mean ± SD shown, n = 3. (D) IC50s from each biological replicate in (C); mean ± SD shown. Using an ordinary one-way ANOVA with Sidak's multiple comparison test, P < 0.01; P < 0.001. (E) MALAT1 levels in HeLa cells stably expressing luciferase or GRP1DD treated with the indicated concentrations of cEt ASO targeting MALAT1 ± SH-BC-893 (5 μM) for 24 h. Mean ± SD shown, n = 3. (F) IC50s from each biological replicate in (E); mean ± SD shown. (G) MALAT1 levels in HeLa cells treated with the cEt ASO targeting MALAT1 ± SH-BC-893 (5 μM) or PPZ (15 μM) for 24 h. Mean ± SD shown, n = 3. (H) IC50s from each biological replicate in (G); mean ± SD shown. Because SD are not equal, Brown-Forsythe and Welch ANOVA test was used with Dunnett's T3 test to correct for multiple comparisons; P < 0.01.* Genetic experiments confirmed that simultaneous ARF6 and PIKfyve inhibition was required for ASO potentiation. Similar to ARF6 knockdown, PIKfyve knockdown severely limited cell viability and growth. As small molecule inhibition of ARF6 or PIKfyve is well tolerated (Supplementary Figure S8C), these proteins likely have essential non-enzymatic functions in cells. As was done for ARF6 (Figure 3E, F), PIKfyve activity was modulated indirectly. PIKfyve functions as part of a heterotrimeric complex, and can be inhibited by over-expressing its scaffold VAC14 [52,53]. VAC14 over-expression produced robust cytoplasmic vacuolation (Supplementary Figure S8F). Consistent with published reports that PIKfyve-dependent vacuolation is not cytotoxic [54,55], VAC14 over-expressing cells were fully viable and proliferated normally for several weeks of continuous culture despite their highly vacuolated state. Recapitulating the effects of chemical PIKfyve inhibitors (Figure 4A–B and Supplementary Figure S8A–B, D–E), VAC14 over-expression increased ASO activity by ∼2-fold (Supplementary Figure S8G, H). This result confirms that PIKfyve inhibition alone is not sufficient to promote ASO activity. Consistent with the model that dual ARF6 and PIKfyve inactivation is required to phenocopy the effects of SH-BC-893, VAC14 over-expression synergized with the otherwise ineffective ARF6 inhibitor NAV2729 to potentiate ASO activity (Supplementary Figure S8G-H). Conversely, inhibition of ARF6-mediated endocytic recycling was necessary for SH-BC-893-mediated ASO potentiation. GRP1DD expressing cells that are resistant to endocytic recycling inhibition by SH-BC-893 ([40] and Figure 3E, F) were also significantly less sensitive to the ASO-potentiating effects of SH-BC-893 (Figure 4E, F). In summary, studies with chemical inhibitors and genetic approaches both support the model that simultaneous ARF6 and PIKfyve inhibition is necessary and sufficient to account for the effects of SH-BC-893 on ASO uptake, distribution, and activity. SH-BC-893 disrupts both endocytic recycling and lysosomal fusion by activating PP2A (Figure 1A and [39,40,42]). The dopaminergic antagonist perphenazine (PPZ) is structurally distinct from SH-BC-893 (Supplementary Figure S1) but also activates PP2A [56,57]. Like SH-BC-893, PPZ inhibits ARF6-dependent endocytic recycling [40] and vacuolates cells similar to PIKfyve inhibitors (Supplementary Figure S9A). Consistent with the model that SH-BC-893’s effects on ASO trafficking and activity lie downstream of PP2A-dependent ARF6 and PIKfyve inactivation (Figure 1A), PPZ enhanced the intracellular accumulation of ASOs, blocked endocytic recycling, and increased the extralysosomal fraction of ASOs phenocopying the effects of SH-BC-893 (Supplementary Figure S9B–E). Like SH-BC-893, PPZ is also not cytotoxic under the conditions where it disrupts endolysosomal trafficking (Supplementary Figure S9F). Consistent with its effects on intracellular ASO levels and localization (Supplementary Figure S9B–E), PPZ increased ASO activity to a similar extent as SH-BC-893 (Figure 4G, H). Notably, SH-BC-893 is 3 times more potent than PPZ. As with SH-BC-893, potentiation by the PP2A activator PPZ was independent of the RNA target or cell type (Supplementary Figure S9G-I). Together, these results with molecules that are structurally unrelated to SH-BC-893, the ARF6 inhibitors NAV2729 and SecinH3, the PIKfyve inhibitors YM201636 and apilimod, and the PP2A activator PPZ, support the model that SH-BC-893 improves oligonucleotide activity via PP2A-dependent changes in ARF6- and PIKfyve-dependent endolysosomal trafficking (Figure 1A and [39,40]). ## Oral administration of SH-BC-893 safely potentiates systemically delivered ASOs in both the liver and extra-hepatic tissues Inefficient oligonucleotide uptake and endosomal escape limits target engagement in all tissues, including the liver [9,10,46]. Prior work established that SH-BC-893 reaches active concentrations in the liver 4 h after oral administration of 120 mg/kg [41]. To determine whether SH-BC-893 potentiated ASO activity in the liver, the cEt gapmer targeting Malat1 was administered to mice subcutaneously. Malat1 is often targeted in proof-of-concept studies as this lncRNA is ubiquitously expressed and not essential for cellular homeostasis [58]. To control for sequence-independent effects of oligonucleotides on Malat1 expression [47], a non-targeting, control ASO was also utilized (Supplementary Table 1). As expected, 50 mg/kg of targeted but not control ASO administered subcutaneously reduced Malat1 RNA levels in the liver by $80\%$ (Figure 5A). At this high ASO dose, knockdown efficiency was not significantly improved by SH-BC-893. At lower ASO doses, the potentiating effect of SH-BC-893 became apparent (Figure 5B, C). With SH-BC-893 co-administration, 5 and 0.5 mg/kg Malat1-targeted ASO produced similar Malat1 knockdown as the 50 and 5 mg/kg doses in the absence of SH-BC-893 (Figure 5A–C and Supplementary Figure S10A). Fitting a curve to the dose response shown in Figure 5C, the ED50 for knockdown was 15 mg/kg in control mice and 1 mg/kg in SH-BC-893-treated mice. Targeted delivery via GalNAc conjugation increases ASO uptake and activity in hepatocytes [59,60]. As expected, the GalNAc3-conjugated form of the MALAT1 ASO (GN3-ASO) was already extremely potent in the liver (Supplementary Figure S10B). SH-BC-893 only modestly improved its activity. Given the different route of entry and pharmacology, a kinetic study combined with a full dose response curve would be required to fully appreciate the extent to which SH-BC-893 would further improve ASO potency for GalNAc-conjugates. In sum, SH-BC-893 increases the activity of systemically delivered ASOs in the liver. Figure 5.: SH-BC-893 enhances activity of systemically administered ASOs in the liver. (A) Malat1 knockdown in the livers of male Balbc/J mice treated with SH-BC-893 (120 mg/kg P.O.) 2 h before ASO (50 mg/kg S.C.) and sacrificed 24 h after a single dose. Non-targeting (control) or Malat1-targeting cEt gapmer ASO were used. Mean ± SD shown, n = 8. Using an ordinary one-way ANOVA with Tukey's correction for multiple comparisons, *P < 0.001. (B) As in (A), except mice were given 5 mg/kg ASO and n = 4. (C) Malat1 knockdown in mice treated as in (A) with 120 mg/kg SH-BC-893 and the indicated dose of cEt Malat1 ASO. Mean ± SD shown, n = 4 except 50 mg/kg group where n = 8. Using an unpaired t-test to compare results ± SH-BC-893, P < 0.05 and *P < 0.01. RNA levels are expressed relative to the housekeeping gene Ppia using the 2−ΔΔCt method. In (A, B), knockdown is calculated relative to the mean from the mice receiving the non-targeting ASO and water vehicle. In (C), knockdown is expressed relative to the mean from the non-targeting ASO group for either the vehicle- or SH-BC-893-treated mice. Orally administered SH-BC-893 also reaches active concentrations in the brain [41]. Many CNS diseases, including lethal neurodegenerative diseases where there is a high unmet need, could be treated with oligonucleotide therapeutics. Because unconjugated oligonucleotides do not cross the blood brain barrier (BBB), intrathecal administration is required to access CNS targets [9,10]. However, intrathecal or intracerebroventricular administration is not patient-friendly. If oligonucleotides could engage CNS targets after systemic delivery, it could allow home administration and improve provider and patient uptake. It was recently discovered that systemically-delivered oligonucleotides can cross the BBB if duplexed with a cholesterol-conjugated sense strand [61]. However, multiple high doses of these duplexes are required to achieve significant knockdown in the CNS. We therefore evaluated whether SH-BC-893 could improve target RNA knockdown in the CNS of mice treated with the Malat1 targeting cEt ASO duplexed with a cholesterol-functionalized DNA sense strand (Supplementary Table 4). Consistent with prior reports [61], two subcutaneous 50 mg/kg doses of duplexed ASOs were insufficient to produce robust Malat1 knockdown in CNS tissues (Figure 6A–E). However, co-administration with SH-BC-893 enabled statistically significant knockdown of ∼$30\%$ in the brain stem and spinal cord with a promising trend in the hippocampus and cerebellum. No knockdown was observed in the cortex in control or 893-treated mice. The lack of activity in the cortex could be due to low penetration of ASOs in this region as SH-BC-893 has been shown to reach active concentrations in the cortex [41]. These results clearly demonstrate that SH-BC-893 can improve oligonucleotide activity in at least a subset of CNS tissues. Figure 6.: SH-BC-893 increases the activity of systemically-delivered, cholesterol-functionalized DNA/DNA duplexed oligonucleotides in the CNS. (A–E) Malat1 knockdown in the brain stem (A), spinal cord (B), hippocampus (C), cerebellum (D) or cortex (E) of male Balbc/J mice treated with SH-BC-893 (120 mg/kg P.O.) 2 h before subcutaneous dosing with cholesterol-functionalized duplexed cEt gapmer targeting Malat1 (100 mg/kg of duplex, 50 mg/kg of ASO strand S.C.). Mice received two doses 2 days apart and were sacrificed 5 days after the last dose. Mean ± SEM shown, n = 5. Using an ordinary one-way ANOVA with Tukey's correction for multiple comparisons, *P < 0.001. RNA levels were expressed relative to the housekeeping gene Ppia using the 2−ΔΔCt method and knockdown expressed relative to the mean of the group receiving both vehicles. To predict which additional extrahepatic tissues might be susceptible to SH-BC-893-mediated ASO potentiation, SH-BC-893 levels were determined in a variety of organs after oral administration (Figure 7A). SH-BC-893 levels were highest in the lung (Figure 7A, B), a tissue that is basally resistant to systemically administered ASOs due to limited ASO accumulation at this site [62,63]. As expected, a single 50 or 5 mg/kg subcutaneous dose of ASO did not produce knockdown in the lung (Figure 7C, D). In contrast, oral administration of SH-BC-893 enabled ASO-dependent reductions in lung Malat1 RNA levels of $54\%$ or $26\%$ 24 h after a single 50 or 5 mg/kg subcutaneous ASO dose, respectively. Potentiation by SH-BC-893 was similarly robust when measured 3 d after dosing (Supplementary Figure S10C). Although ASOs and SH-BC-893 are both present at reasonably high levels in the kidney and spleen ([64] and Figure 7A and Supplementary Figure S10D), no potentiation was observed in these tissues (Supplementary Figure S10E–I). The lack of potentiation could reflect different limitations on ASO delivery in different tissues, accumulation of ASO and SH-BC-893 in different cell types, and/or the fact that tissue-level LC-MS/MS measurements fail to discriminate between intracellular and extracellular ASOs and/or SH-BC-893. Statistically significant ASO potentiation was not observed in tissues with low SH-BC-893 levels such as skeletal and heart muscle, although a promising trend was observed in the quadriceps that might be improved with repeated dosing (Figure 7A and Supplementary Figure S10J, K). In summary, co-administration of SH-BC-893 rendered systemically administered ASO active in the lung. Figure 7.: SH-BC-893 sensitizes the lung to systemically administered ASOs. (A) Tissue SH-BC-893 levels in male (n = 3) or female (n = 3) CD1 mice treated with 120 mg/kg P.O. Q.D. for 5 days and sacrificed 8 h after the last dose. Mean ± SD shown, n = 6. (B) As in (A) but in mice sacrificed at the indicated time points after the last dose. (C) Malat1 knockdown in the lungs of male Balbc/J mice treated with SH-BC-893 (120 mg/kg P.O.) 2 h before ASO (50 mg/kg S.C.) and sacrificed 24 h after a single dose. Non-targeting (control) or Malat1-targeting cEt gapmer ASO were used. Mean ± SD shown, n = 8. Using an ordinary one-way ANOVA with Tukey's correction for multiple comparisons, *P < 0.001. (D) As in (C), except mice were given 5 mg/kg ASO and n = 4. (E) Malat1 knockdown in mice treated as in (C) with 120 mg/kg SH-BC-893 and the indicated dose of cEt Malat1 ASO. Mean ± SD shown, n = 4 except 50 mg/kg group where n = 8. Using an unpaired t-test to compare results ± SH-BC-893, P < 0.05 and P < 0.01. (F) Scnn1a (aka ENaCα) knockdown in the lungs of male Balbc/J mice treated with SH-BC-893 (120 mg/kg P.O.) 2 h before ASO (50 mg/kg S.C.) and sacrificed 72 h after a single dose. Non-targeting (control) or Scnn1a-targeting cEt gapmers was used. Mean ± SD shown, n = 4. Using a one-way ANOVA with Tukey's correction for multiple comparisons, P < 0.001. (G) As in (F), except three doses of 5 mg/kg ASO were given at 7 days intervals and mice sacrificed 7 days after the last dose. (H) As in (G), except expressed as a function of number of doses received and normalized to the non-targeting ASO control. Using an unpaired t-test to compare results ± SH-BC-893, P < 0.01. RNA levels are expressed relative to the housekeeping gene Ppia using the 2−ΔΔCt method. In (C, D, F, G), knockdown is calculated relative to the mean from the mice receiving the non-targeting ASO and water vehicle. In (E, H), knockdown is expressed relative to the mean from the non-targeting ASO group for either the vehicle- or SH-BC-893-treated mice. Potentiating effects in the lung were intriguing as oligonucleotide therapeutics are being developed to treat lung diseases [62,65,66]. An ASO targeting the Scnn1a mRNA that encodes a subunit of the ENaC sodium channel has been tested in cystic fibrosis patients and may also have benefits in other lung diseases characterized by mucus dehydration such as chronic obstructive pulmonary disease (COPD) (65–67). SH-BC-893-dependent potentiation may vary with different ASO sequences and targets. Because the MALAT1 ASO widely used in proof-of-concept studies is 2–3-fold more potent than most other cEt gapmers, other oligonucleotides may exhibit less potentiation with SH-BC-893. Additionally, individual cell types present within a tissue may be differentially sensitive to SH-BC-893-dependent potentiation (62–64). For these reasons, SH-BC-893 was also tested with a cEt gapmer ASO targeting the Scnn1a mRNA expressed in lung epithelial cells. At the 50 mg/kg ASO dose, SH-BC-893 improved Scnn1a knockdown from $17\%$ to $59\%$ (Figure 7F). While ENaC is also expressed in the kidney, statistically significant knockdown was not observed with subcutaneous administration of the Scnn1a ASO even in mice treated with SH-BC-893 (Supplementary Figure S10L). Although knockdown did not achieve statistical significance after a single 5 mg/kg dose of Scnn1a ASO and SH-BC-893, $30\%$ knockdown was achieved after repeat dosing with the combination while Scnn1a ASO alone was ineffective (Figure 7G, H and Supplementary Figure S10M). Thus, SH-BC-893 improves ASO activity in lung epithelial cells that are targeted by oligonucleotide therapeutics designed to treat cystic fibrosis. In sum, the results presented here establish the feasibility of improving therapeutic oligonucleotide delivery to both hepatic and extrahepatic tissues by co-administering small molecules like SH-BC-893 that block endocytic recycling and lysosomal fusion. In contrast to the historical toxicity problems associated with endolytic agents, even chronic daily administration of SH-BC-893 is well tolerated [39]. After 11 weeks of daily oral administration of the effective dose of SH-BC-893 (120 mg/kg), blood chemistry revealed no signs of organ toxicity. Rapidly proliferating cells in the bone marrow and intestinal crypts were not compromised based on normal complete blood counts and grossly normal histopathology. In fact, oral administration of 120 mg/kg SH-BC-893 every other day promotes metabolic homeostasis in mice maintained on a high fat diet [41]. In this study, mice provided with a running wheel engaged in the same amount of voluntary exercise whether they were treated with vehicle or with 120 mg/kg SH-BC-893 PO over the 4-week study. Voluntary wheel running is a holistic and sensitive measure of animal health [68]. Together, these studies clearly establish the safety of chronic administration of the effective dose of SH-BC-893. To confirm that SH-BC-893 did not induce acute toxicity that might have been missed in the published long-term studies, we evaluated blood chemistry in mice 24 h after treatment with vehicle, 120 or 240 mg/kg SH-BC-893 (Supplementary Figure S11C). Neither the effective dose (120 mg/kg) nor twice the effective dose (240 mg/kg) disrupted blood chemistry values, indicating that SH-BC-893 is not acutely toxic to the liver, kidney, or muscle. Consistent with the bloodwork, no significant changes in body weight were noted in mice that were maintained for more than 24 h after administration of SH-BC-893 and oligonucleotide (Supplementary Figure S11D-F). When used as an oligonucleotide potentiator, the safety margin would be further enhanced compared to published studies with repeat dosing because SH-BC-893 would only be administered as frequently as the oligonucleotide (weekly, monthly or even every 6 months) [16]. Taken together, the data presented here and in published studies clearly establish that SH-BC-893 is not toxic to mice at the effective dose. ## DISCUSSION Here, we demonstrate that the small molecule SH-BC-893 increases the activity of ASOs and siRNAs by up to 100-fold in multiple cell types without lysing endosomes. Rather than disrupting membranes, SH-BC-893 traps oligonucleotides in a pre-lysosomal compartment where they likely escape when fission and fusion reactions deform the lipid bilayer increasing permeability (27,29–33). Consistent with this mechanism of action and the well-established low permeability of the lysosomal membrane, SH-BC-893 cannot increase oligonucleotide activity if it is added after ASOs have reached lysosomes (Supplementary Figure S6C, D). Genetic approaches and multiple, structurally distinct chemical inhibitors confirmed that simultaneous inhibition of ARF6-dependent endocytic recycling and PIKfyve-dependent lysosomal fusion was necessary and sufficient for SH-BC-893 to boost intracellular ASO levels, increase accumulation of ASOs within extra-lysosomal compartments, and enhance oligonucleotide activity (Figures 3 and 4 and Supplementary Figure S8). As single agents, ARF6 inhibitors increase intracellular ASO levels but fail to potentiate ASO activity because ASOs still end up in lysosomes (Figure 3A–B, G and Figure 4B and Supplementary Figure S8A–B, D–E). Similarly, inactivating the PIKfyve-dependent lysosomal fusion pathway in isolation fails to block recycling and produces only a 2–4-fold increase in oligonucleotide activity similar to what can be achieved with previously reported small molecule potentiators (Figures 3C–D and 4A–D and Supplementary Figures S6G–H and S8A–B, D–H). In contrast, the synergistic effects of blocking recycling out of the cell and preventing ASO transit to lysosomes allows SH-BC-893 to dramatically increase oligonucleotide activity (Figures 3 and 4 and Supplementary Figure S8). SH-BC-893 is distinct from previously identified small molecule potentiators in that it robustly increases oligonucleotide activity without lysing endosomes, avoiding the toxic consequences that have prevented endolytic agents from advancing to the clinic. Agents like SH-BC-893 may have some utility for liver-targeted ASOs. The 15-fold increase in ASO activity observed in the livers of mice treated with SH-BC-893 (Figure 5C) is on par with the potency increase reported with GalNAc conjugation, a clinically validated means to improve oligonucleotide delivery to hepatocytes [9,59]. Thus, for liver targets, SH-BC-893 might offer greater access to cell types that do not express the ASGPR or offer an alternative approach to GalNAc conjugation. SH-BC-893 slightly enhanced the activity of the already quite potent GalNAc3-MALAT1 ASO (Supplementary Figure S10B). Additional studies with GalNAc-conjugates and other targeted oligonucleotides would be required to fully assess the potential of SH-BC-893 to improve ligand-conjugated oligonucleotide uptake in the liver and extrahepatic tissues [13,69,70]. Agents that act like SH-BC-893 are poised to render new targets in extrahepatic tissues accessible even without ligand-receptor targeting strategies. In the CNS, SH-BC-893 increased the activity of cholesterol-functionalized ASO duplexes in the brainstem and spinal cord (Figure 6A, B) with positive trends noted in the hippocampus and cerebellum (Figure 6C, D). The most sensitive extrahepatic tissue was the lung (Figure 7C–E) likely due to the accumulation of SH-BC-893 in this tissue (Figure 7A, B). Interestingly, although both ASOs and SH-BC-893 accumulate in the kidney and spleen (Figure 7A, Supplementary Figure S10D, and [64]), potentiation was limited or absent in these tissues (Supplementary Figure S10E–I, L). These results could be explained by lower basal levels of ARF6-dependent ASO recycling, reduced PIKfyve-dependent lysosomal fusion, and/or differences in the endocytic trafficking pathways used for oligonucleotide entry in these tissues. Alternatively, SH-BC-893 and ASOs may not accumulate in the same cell types in these organs; ASOs accumulate primarily in proximal tubular epithelial cells in the kidney and endothelial cells in the spleen. In heart and skeletal muscle, limited potentiation is likely explained by the low concentrations of both SH-BC-893 and ASOs in these tissues ([63,64] and Supplementary Figure S10D, J–K). Given the ubiquitous expression of PP2A, ARF6 and PIKfyve, additional tissues with SH-BC-893 levels higher than the brain (Figure 7A) may also be sensitive to potentiation. The results reported here offer an important proof of concept, but additional studies will be required to produce a comprehensive inventory of the tissues and cell types that are responsive to SH-BC-893-mediated potentiation with single or repeat oligonucleotide dosing. While this study utilized systemic ASO delivery, SH-BC-893 should also improve the efficacy of locally administered oligonucleotide therapeutics. Administration of ASOs directly at the site of action (e.g. intravitreal, intrathecal, or aerosol delivery) increases uptake by elevating local concentrations but does not address the negative effects of ARF6-dependent recycling or PIKfyve-dependent lysosomal fusion on delivery (Figure 1A). Local administration of SH-BC-893 along with the oligonucleotide might additionally overcome any limitations imposed by the tissue pharmacokinetics of SH-BC-893. In some cases, local delivery of oligonucleotide is preferred to limit target engagement to specific tissues. For example, in cystic fibrosis therapy it is desirable to reduce ENaC levels in the lung while leaving ENaC expression in the kidney unaffected [71]. Notably, SH-BC-893 increased ENaC knockdown in the lung (Figure 7F-H) but not in the kidney (Supplementary Figure S10L). Formulation work would be required to evaluate whether SH-BC-893 could be administered intrathecally or via inhalation along with the oligonucleotide, but local delivery of oligonucleotide and/or SH-BC-893 might further extend the clinical reach of small molecule-mediated potentiation. It is also worth noting that SH-BC-893 is unlikely to diminish any long-term activity that might be sustained by slow leak of oligonucleotides from lysosomes [16]. A single dose of SH-BC-893 administered with oligonucleotides would be cleared from the body in 24–48 h releasing the trafficking block and allowing trapped oligonucleotides to continue their progress through the endocytic pathway. Long-term activity likely tracks with the level of initial oligonucleotide uptake by target cells, and thus SH-BC-893 may improve knockdown duration by increasing the intracellular amount of oligonucleotide (Figures 1D–E and 3A–B and Supplementary Figure S2B, D, E, G). In summary, SH-BC-893 is likely to be compatible with other approaches that improve extrahepatic oligonucleotide delivery. The in vivo results with SH-BC-893 are noteworthy both due to the significant gain in oligonucleotide activity and the surprising tolerability of this compound despite its profound effects on endolysosomal trafficking [39,41]. SH-BC-893 does not harm even rapidly proliferating normal tissues after chronic administration of the effective dose; bone marrow suppression was not observed and rapidly dividing intestinal crypts were not compromised [39]. In fact, SH-BC-893 promotes metabolic homeostasis in mice, and treated animals engage in normal levels of voluntary exercise, a holistic measure of animal health [41,68]. When used as an oligonucleotide delivery potentiator, the safety margin would be further increased by infrequent dosing. Consistent with these earlier studies, acute administration of even twice the effective dose of SH-BC-893 produced no detectable changes in liver, kidney, or muscle function by blood chemistry analysis (Supplementary Figure S11C). The tolerability of SH-BC-893 likely stems from its origin as a synthetic analog of the natural sphingolipid, phytosphingosine, that promotes survival in yeast under stress [72]. This sphingolipid-responsive signaling pathway is conserved in mammalian cells suggesting that it was fine-tuned by evolution to modulate intracellular trafficking in a manner that preserves cell viability and tissue homeostasis (39–42,73). Consistent with this hypothesis, SH-BC-893 selectively kills cancer cells while sparing non-transformed cells [39]. Oncogenic mutations make cancer cells constitutively anabolic and therefore hypersensitive to nutrient limitation. Indeed, this liability provides the therapeutic index for many standard-of-care cancer therapies [74]. While cancer cells starve to death secondary to SH-BC-893-induced changes in endolysosomal trafficking, normal cells adapt to the moderate nutrient restriction by reducing their energy demands [39]. In summary, the low toxicity of the anti-neoplastic agent SH-BC-893 in non-transformed cells and tissues likely augurs well for its use in combination with oligonucleotide therapeutics. Going forward, oligonucleotide therapeutics will need to compete with next-generation small molecules that are also capable of hitting what were previously ‘undruggable’ targets [75]. A potentiator like SH-BC-893 that permits at-home oligonucleotide administration could improve the ability of oligonucleotide therapeutics to compete with new orally administered small molecule alternatives [76]. By lowering the required dose, a small molecule potentiator like SH-BC-893 could make expensive oligonucleotide therapeutics accessible to more patients. Given the established pre-clinical activities of SH-BC-893 as a single agent in cancer and obesity models [39,41], oligonucleotide therapeutics targeting these diseases might be prioritized for assessing the therapeutic value of combination therapy. For example, several oligonucleotides that entered cancer trials failed to reach efficacy benchmarks. SH-BC-893 slows autochthonous prostate tumor growth [39] and could be even more effective in combination with relevant oncology ASOs [7,77]. Given its accumulation in the lung (Figure 7A), primary or metastatic lung tumors might also be responsive to an SH-BC-893/ASO combination targeting KRAS [6]. The oligonucleotide-independent actions of SH-BC-893 on mitochondrial dynamics [41] may complement the activities of oligonucleotides designed to treat cancer [78], non-alcoholic steatohepatitis (NASH) [79], and/or neurodegenerative diseases [9,41,80]. In conclusion, the proof-of-concept studies presented here provide a strong rationale for future work exploring the therapeutic value and safety of SH-BC-893 and/or related small molecules as oligonucleotide potentiating agents. ## DATA AVAILABILITY All data generated and analyzed in this study are included in the manuscript and supplementary data. ## SUPPLEMENTARY DATA Supplementary Data are available at NAR Online. ## FUNDING Ono Pharma Foundation Breakthrough Science Initiative [G-1902-0039]; unrestricted gift from Ionis Pharmaceuticals (to A.L.E.); NIH [1R44CA257568 to D.G.]; B.T.F. was supported by NCI [T32 CA009054, F31 CA261085, F99 CA264430]; Chao Family Comprehensive Cancer Center Optical Biology Center shared resource, supported by the National Cancer Institute of the National Institutes of Health [P30 CA062203]. Funding for open access charge: Ono Pharma Foundation Breakthrough Science Initiative grant [G-1902-0039]. Conflict of interest statement. A.L.E., S.H., P.P.S. and B.T.F. are inventors on patents relevant to the use of SH-BC-893 and agents with similar activities for oligonucleotide potentiation. A.N.M., D.G., S.H. and A.L.E. own shares in Siege Pharmaceuticals, which has licensed these patents and is developing SH-BC-893 for clinical use. A.S.R., B.A.A., W.B.W. and P.P.S. were employees of Ionis Pharmaceuticals during the project. P.P.S. is currently an employee of Alnylam Pharmaceuticals. A.L.E. has consulted for Alnylam Pharmaceuticals. Other authors declare no conflicts of interest. ## References 1. Shen X., Corey D.R.. Chemistry, mechanism and clinical status of antisense oligonucleotides and duplex rnas. Nucleic Acids Res. (2018) 46 1584-1600. PMID: 29240946 2. Roberts T.C., Langer R., Wood M.J.A.. Advances in oligonucleotide drug delivery. Nat. Rev. Drug. Discov. (2020) 19 673-694. PMID: 32782413 3. Hong D., Kurzrock R., Kim Y., Woessner R., Younes A., Nemunaitis J., Fowler N., Zhou T., Schmidt J., Jo M.. AZD9150, a next-generation antisense oligonucleotide inhibitor of STAT3 with early evidence of clinical activity in lymphoma and lung cancer. Sci. Transl. Med. (2015) 7 314ra185 4. Odate S., Veschi V., Yan S., Lam N., Woessner R., Thiele C.J.. Inhibition of STAT3 with the generation 2.5 antisense oligonucleotide, AZD9150, decreases neuroblastoma tumorigenicity and increases chemosensitivity. Clin. Cancer Res. (2017) 23 1771-1784. PMID: 27797972 5. Raal F.J., Santos R.D., Blom D.J., Marais A.D., Charng M.-J., Cromwell W.C., Lachmann R.H., Gaudet D., Tan J.L., Chasan-Taber S.. Mipomersen, an apolipoprotein B synthesis inhibitor, for lowering of LDL cholesterol concentrations in patients with homozygous familial hypercholesterolaemia: a randomised, double-blind, placebo-controlled trial. Lancet (2010) 375 998-1006. PMID: 20227758 6. Ross S.J., Revenko A.S., Hanson L.L., Ellston R., Whalley N., Pandey S.K., Revill M., Rooney C., Buckett K., Klein S.K.. Targeting KRAS dependent tumors with AZD4785, a high-affinity therapeutic antisense oligonucleotide inhibitor of KRAS. Sci. Trans. Med. (2017) 9 eaal5253 7. De Velasco M.A., Kura Y., Sakai K., Hatanaka Y., Davies B.R., Campbell H., Klein S., Kim Y., MacLeod A.R., Sugimoto K.. Targeting castration-resistant prostate cancer with androgen receptor antisense oligonucleotide therapy. JCI Insight (2019) 4 e122688. PMID: 31484823 8. Yamamoto Y., Loriot Y., Beraldi E., Zhang F., Wyatt A.W., Al Nakouzi N., Mo F., Zhou T., Kim Y., Monia B.P.. Generation 2.5 antisense oligonucleotides targeting the androgen receptor and its splice variants suppress enzalutamide-resistant prostate cancer cell growth. Clin. Cancer Res. (2015) 21 1675-1687. PMID: 25634993 9. Gökirmak T., Nikan M., Wiechmann S., Prakash T.P., Tanowitz M., Seth P.P.. Overcoming the challenges of tissue delivery for oligonucleotide therapeutics. Trends Pharmacol. Sci. (2021) 42 588-604. PMID: 34020790 10. Dowdy S.F.. Overcoming cellular barriers for RNA therapeutics. Nat. Biotechnol. (2017) 35 222-229. PMID: 28244992 11. Corey D.R., Damha M.J., Manoharan M.. Challenges and opportunities for nucleic acid therapeutics. Nucleic Acid Ther. (2022) 32 8-13. PMID: 34931905 12. Tanowitz M., Hettrick L., Revenko A., Kinberger G.A., Prakash T.P., Seth P.P.. Asialoglycoprotein receptor 1 mediates productive uptake of N-acetylgalactosamine-conjugated and unconjugated phosphorothioate antisense oligonucleotides into liver hepatocytes. Nucleic Acids Res. (2017) 45 12388-12400. PMID: 29069408 13. Seth P.P., Tanowitz M., Bennett C.F.. Selective tissue targeting of synthetic nucleic acid drugs. J. Clin. Invest. (2019) 129 915-925. PMID: 30688661 14. Crooke S.T., Wang S., Vickers T.A., Shen W., Liang X.-H.. Cellular uptake and trafficking of antisense oligonucleotides. Nat. Biotechnol. (2017) 35 230-237. PMID: 28244996 15. Grant B.D., Donaldson J.G.. Pathways and mechanisms of endocytic recycling. Nat. Rev. Mol. Cell Biol. (2009) 10 597-608. PMID: 19696797 16. Brown C.R., Gupta S., Qin J., Racie T., He G., Lentini S., Malone R., Yu M., Matsuda S., Shulga-Morskaya S.. Investigating the pharmacodynamic durability of GalNAc-siRNA conjugates. Nucleic Acids Res. (2020) 48 11827-11844. PMID: 32808038 17. Linnane E., Davey P., Zhang P., Puri S., Edbrooke M., Chiarparin E., Revenko A.S., Macleod A.R., Norman J.C., Ross S.J.. Differential uptake, kinetics and mechanisms of intracellular trafficking of next-generation antisense oligonucleotides across human cancer cell lines. Nucleic Acids Res. (2019) 47 4375-4392. PMID: 30927008 18. Wagenaar T.R., Tolstykh T., Shi C., Jiang L., Zhang J., Li Z., Yu Q., Qu H., Sun F., Cao H.. Identification of the endosomal sorting complex required for transport-I (ESCRT-I) as an important modulator of anti-miR uptake by cancer cells. Nucleic Acids Res. (2015) 43 1204-1215. PMID: 25550434 19. Yang B., Ming X., Cao C., Laing B., Yuan A., Porter M.A., Hull-Ryde E.A., Maddry J., Suto M., Janzen W.P.. High-throughput screening identifies small molecules that enhance the pharmacological effects of oligonucleotides. Nucleic Acids Res. (2015) 43 1987-1996. PMID: 25662226 20. Gilleron J., Paramasivam P., Zeigerer A., Querbes W., Marsico G., Andree C., Seifert S., Amaya P., Stöter M., Koteliansky V.. Identification of siRNA delivery enhancers by a chemical library screen. Nucleic Acids Res. (2015) 43 7984-8001. PMID: 26220182 21. Ming X., Carver K., Fisher M., Noel R., Cintrat J.-C., Gillet D., Barbier J., Cao C., Bauman J., Juliano R.L.. The small molecule Retro-1 enhances the pharmacological actions of antisense and splice switching oligonucleotides. Nucleic Acids Res. (2013) 41 3673-3687. PMID: 23396438 22. Osborn M.F., Alterman J.F., Nikan M., Cao H., Didiot M.C., Hassler M.R., Coles A.H., Khvorova A.. Guanabenz (Wytensin. Nucleic Acids Res. (2015) 43 8664-8672. PMID: 26400165 23. Zhang X., Castanotto D., Nam S., Horne D., Stein C.. 6BIO Enhances oligonucleotide activity in cells: a potential combinatorial anti-androgen receptor therapy in prostate cancer cells. Mol. Ther. (2017) 25 79-91. PMID: 28129131 24. Kendall G.C., Mokhonova E.I., Moran M., Sejbuk N.E., Wang D.W., Silva O., Wang R.T., Martinez L., Lu Q.L., Damoiseaux R.. Dantrolene enhances antisense-mediated exon skipping in human and mouse models of Duchenne muscular dystrophy. Sci. Transl. Med. (2012) 4 164ra160 25. Juliano R.L.. Chemical manipulation of the endosome trafficking machinery: implications for oligonucleotide delivery. Biomedicines (2021) 9 512-523. PMID: 34063104 26. Wittrup A., Ai A., Liu X., Hamar P., Trifonova R., Charisse K., Manoharan M., Kirchhausen T., Lieberman J.. Visualizing lipid-formulated siRNA release from endosomes and target gene knockdown. Nat. Biotechnol. (2015) 33 870-876. PMID: 26192320 27. Juliano R.L.. Intracellular trafficking and endosomal release of oligonucleotides: what we know and what we don’t. Nucleic Acid Ther. (2018) 28 166-177. PMID: 29708838 28. Wang S., Sun H., Tanowitz M., Liang X.-H., Crooke S.T.. Intra-endosomal trafficking mediated by lysobisphosphatidic acid contributes to intracellular release of phosphorothioate-modified antisense oligonucleotides. Nucleic Acids Res. (2017) 45 5309-5322. PMID: 28379543 29. Wang T., Smith E.A., Chapman E.R., Weisshaar J.C.. Lipid mixing and content release in single-vesicle, SNARE-driven fusion assay with 1-5 ms resolution. Biophys. J. (2009) 96 4122-4131. PMID: 19450483 30. Renard H.-F., Johannes L., Morsomme P.. Increasing diversity of biological membrane fission mechanisms. Trends Cell Biol. (2018) 28 274-286. PMID: 29307448 31. Wickner W., Rizo J.. A cascade of multiple proteins and lipids catalyzes membrane fusion. Mol. Biol. Cell (2017) 28 707-711. PMID: 28292915 32. Bennett W.F.D., Tieleman D.P.. The importance of membrane defects-lessons from simulations. Acc. Chem. Res. (2014) 47 2244-2251. PMID: 24892900 33. Cullis P.R., Hope M.J., Tilcock C.P.. Lipid polymorphism and the roles of lipids in membranes. Chem. Phys. Lipids. (1986) 40 127-144. PMID: 3742670 34. Wang S., Sun H., Tanowitz M., Liang X.-H., Crooke S.T.. Annexin A2 facilitates endocytic trafficking of antisense oligonucleotides. Nucleic Acids Res. (2016) 44 7314-7330. PMID: 27378781 35. Liang X.-H., Sun H., Hsu C.-W., Nichols J.G., Vickers T.A., De Hoyos C.L., Crooke S.T.. Golgi-endosome transport mediated by M6PR facilitates release of antisense oligonucleotides from endosomes. Nucleic Acids Res. (2020) 48 1372-1391. PMID: 31840180 36. Paramasivam P., Franke C., Stöter M., Höijer A., Bartesaghi S., Sabirsh A., Lindfors L., Arteta M.Y., Dahlén A., Bak A.. Endosomal escape of delivered mRNA from endosomal recycling tubules visualized at the nanoscale. J. Cell Biol. (2022) 221 e202110137. PMID: 34882187 37. Wilke S., Krausze J., Büssow K.. Crystal structure of the conserved domain of the DC lysosomal associated membrane protein: implications for the lysosomal glycocalyx. BMC Biol. (2012) 10 62. PMID: 22809326 38. Rudnik S., Damme M.. The lysosomal membrane-export of metabolites and beyond. FEBS J. (2021) 288 4168-4182. PMID: 33067905 39. Kim S.M., Roy S.G., Chen B., Nguyen T.M., McMonigle R.J., McCracken A.N., Zhang Y., Kofuji S., Hou J., Selwan E.. Targeting cancer metabolism by simultaneously disrupting parallel nutrient access pathways. J. Clin. Invest. (2016) 126 4088-4102. PMID: 27669461 40. Finicle B.T., Ramirez M.U., Liu G., Selwan E.M., McCracken A.N., Yu J., Joo Y., Nguyen J., Ou K., Roy S.G.. Sphingolipids inhibit endosomal recycling of nutrient transporters by inactivating ARF6. J. Cell Sci. (2018) 131 jcs213314. PMID: 29848659 41. Jayashankar V., Selwan E., Hancock S.E., Verlande A., Goodson M.O., Eckenstein K.H., Milinkeviciute G., Hoover B.M., Chen B., Fleischman A.G.. Drug-like sphingolipid SH-BC-893 opposes ceramide-induced mitochondrial fission and corrects diet-induced obesity. EMBO Mol. Med. (2021) 13 e13086. PMID: 34231322 42. Kubiniok P., Finicle B.T., Piffaretti F., McCracken A.N., Perryman M., Hanessian S., Edinger A.L., Thibault P.. Dynamic phosphoproteomics uncovers signaling pathways modulated by anti-oncogenic sphingolipid analogs. Mol. Cell. Proteomics (2019) 18 408-422. PMID: 30482847 43. Uphoff C.C., Drexler H.G.. Detection of Mycoplasma contamination in cell cultures. Curr. Protoc. Mol. Biol. (2014) 106 28.4.1-28.4.14 44. Chen B., Roy S.G., McMonigle R.J., Keebaugh A., McCracken A.N., Selwan E., Fransson R., Fallegger D., Huwiler A., Kleinman M.T.. Azacyclic FTY720 analogues that limit nutrient transporter expression but lack S1P receptor activity and negative chronotropic effects offer a novel and effective strategy to kill cancer cells in vivo. ACS Chem. Biol. (2016) 11 409-414. PMID: 26653336 45. He C., Migawa M.T., Chen K., Weston T.A., Tanowitz M., Song W., Guagliardo P., Iyer K.S., Bennett C.F., Fong L.G.. High-resolution visualization and quantification of nucleic acid-based therapeutics in cells and tissues using Nanoscale secondary ion mass spectrometry (NanoSIMS). Nucleic Acids Res. (2021) 49 1-14. PMID: 33275144 46. Dowdy S.F., Setten R.L., Cui X.-S., Jadhav S.G.. Delivery of RNA therapeutics: the great endosomal escape!. Nucleic Acid Ther. (2022) 32 361-368. PMID: 35612432 47. Gagnon K.T., Corey D.R.. Guidelines for experiments using antisense oligonucleotides and double-stranded rnas. Nucleic Acid Ther. (2019) 29 116-122. PMID: 30907681 48. Dominska M., Dykxhoorn D.M.. Breaking down the barriers: siRNA delivery and endosome escape. J. Cell Sci. (2010) 123 1183-1189. PMID: 20356929 49. Ochaba J., Powers A.F., Tremble K.A., Greenlee S., Post N.M., Matson J.E., MacLeod A.R., Guo S., Aghajan M.. A novel and translational role for autophagy in antisense oligonucleotide trafficking and activity. Nucleic Acids Res. (2019) 47 11284-11303. PMID: 31612951 50. Stechmann B., Bai S.-K., Gobbo E., Lopez R., Merer G., Pinchard S., Panigai L., Tenza D., Raposo G., Beaumelle B.. Inhibition of retrograde transport protects mice from lethal ricin challenge. Cell (2010) 141 231-242. PMID: 20403321 51. Li J., Malaby A.W., Famulok M., Sabe H., Lambright D.G., Hsu V.W.. Grp1 plays a key role in linking insulin signaling to glut4 recycling. Dev. Cell (2012) 22 1286-1298. PMID: 22609160 52. Schulze U., Vollenbröker B., Braun D.A., Van Le T., Granado D., Kremerskothen J., Fränzel B., Klosowski R., Barth J., Fufezan C.. The Vac14-interaction network is linked to regulators of the endolysosomal and autophagic pathway. Mol. Cell. Proteomics (2014) 13 1397-1411. PMID: 24578385 53. Burack W.R., Shaw A.S.. Signal transduction: hanging on a scaffold. Curr. Opin. Cell Biol. (2000) 12 211-216. PMID: 10712921 54. Giridharan S.S.P., Luo G., Rivero-Rios P., Steinfeld N., Tronchere H., Singla A., Burstein E., Billadeau D.D., Sutton M.A., Weisman L.S.. Lipid kinases VPS34 and PIKfyve coordinate a phosphoinositide cascade to regulate retriever-mediated recycling on endosomes. Elife (2022) 11 e69709. PMID: 35040777 55. Zhang Y., McCartney A.J., Zolov S.N., Ferguson C.J., Meisler M.H., Sutton M.A., Weisman L.S.. Modulation of synaptic function by VAC14, a protein that regulates the phosphoinositides PI(3,5)P. EMBO J. (2012) 31 3442-3456. PMID: 22842785 56. Morita K., He S., Nowak R.P., Wang J., Zimmerman M.W., Fu C., Durbin A.D., Martel M.W., Prutsch N., Gray N.S.. Allosteric activators of protein phosphatase 2A display broad antitumor activity mediated by dephosphorylation of MYBL2. Cell (2020) 181 702-715. PMID: 32315619 57. Gutierrez A., Pan L., Groen R.W.J., Baleydier F., Kentsis A., Marineau J., Grebliunaite R., Kozakewich E., Reed C., Pflumio F.. Phenothiazines induce PP2A-mediated apoptosis in T cell acute lymphoblastic leukemia. J. Clin. Invest. (2014) 124 644-655. PMID: 24401270 58. Arun G., Aggarwal D., Spector D.L.. MALAT1 Long non-coding RNA: functional implications. Noncoding RNAs (2020) 6 22-38 59. Prakash T.P., Graham M.J., Yu J., Carty R., Low A., Chappell A., Schmidt K., Zhao C., Aghajan M., Murray H.F.. Targeted delivery of antisense oligonucleotides to hepatocytes using triantennary N-acetyl galactosamine improves potency 10-fold in mice. Nucleic Acids Res. (2014) 42 8796-8807. PMID: 24992960 60. Crooke S.T., Baker B.F., Xia S., Yu R.Z., Viney N.J., Wang Y., Tsimikas S., Geary R.S.. Integrated assessment of the clinical performance of GalNAc3-conjugated 2′-O-methoxyethyl chimeric antisense oligonucleotides: I. Human volunteer experience. Nucleic Acid Ther. (2019) 29 16-32. PMID: 30570431 61. Nagata T., Dwyer C.A., Yoshida-Tanaka K., Ihara K., Ohyagi M., Kaburagi H., Miyata H., Ebihara S., Yoshioka K., Ishii T.. Cholesterol-functionalized DNA/RNA heteroduplexes cross the blood-brain barrier and knock down genes in the rodent CNS. Nat. Biotechnol. (2021) 39 1529-1536. PMID: 34385691 62. Kumar M., Moschos S.A.. Oligonucleotide therapies for the lung: ready to return to the clinic?. Mol. Ther. (2017) 25 2604-2606. PMID: 29174842 63. Geary R.S., Norris D., Yu R., Bennett C.F.. Pharmacokinetics, biodistribution and cell uptake of antisense oligonucleotides. Adv. Drug. Deliv. Rev. (2015) 87 46-51. PMID: 25666165 64. Geary R.S.. Antisense oligonucleotide pharmacokinetics and metabolism. Expert Opin. Drug Metab. Toxicol. (2009) 5 381-391. PMID: 19379126 65. Crosby J.R., Zhao C., Jiang C., Bai D., Katz M., Greenlee S., Kawabe H., McCaleb M., Rotin D., Guo S.. Inhaled ENaC antisense oligonucleotide ameliorates cystic fibrosis-like lung disease in mice. J. Cyst. Fibros. (2017) 16 671-680. PMID: 28539224 66. Zhao C., Crosby J., Lv T., Bai D., Monia B.P., Guo S.. Antisense oligonucleotide targeting of mRNAs encoding ENaC subunits α, β, and γ improves cystic fibrosis-like disease in mice. J. Cyst. Fibros. (2019) 18 334-341. PMID: 30100257 67. Ghosh A., Boucher R.C., Tarran R.. Airway hydration and COPD. Cell. Mol. Life Sci. (2015) 72 3637-3652. PMID: 26068443 68. Häger C., Keubler L.M., Talbot S.R., Biernot S., Weegh N., Buchheister S., Buettner M., Glage S., Bleich A.. Running in the wheel: defining individual severity levels in mice. PLoS Biol. (2018) 16 e2006159. PMID: 30335759 69. Ämmälä C., Drury W.J., Knerr L., Ahlstedt I., Stillemark-Billton P., Wennberg-Huldt C., Andersson E.M., Valeur E., Jansson-Löfmark R., Janzén D.. Targeted delivery of antisense oligonucleotides to pancreatic β-cells. Sci. Adv. (2018) 4 eaat3386. PMID: 30345352 70. Sugo T., Terada M., Oikawa T., Miyata K., Nishimura S., Kenjo E., Ogasawara-Shimizu M., Makita Y., Imaichi S., Murata S.. Development of antibody-siRNA conjugate targeted to cardiac and skeletal muscles. J. Control Release (2016) 237 1-13. PMID: 27369865 71. Christopher Boyd A., Guo S., Huang L., Kerem B., Oren Y.S., Walker A.J., Hart S.L.. New approaches to genetic therapies for cystic fibrosis. J. Cyst. Fibros. (2020) 19 S54-S59. PMID: 31948871 72. Dickson R.C., Nagiec E.E., Skrzypek M., Tillman P., Wells G.B., Lester R.L.. Sphingolipids are potential heat stress signals in Saccharomyces. J. Biol. Chem. (1997) 272 30196-30200. PMID: 9374502 73. Perryman M.S., Tessier J., Wiher T., O’Donoghue H., McCracken A.N., Kim S.M., Nguyen D.G., Simitian G.S., Viana M., Rafelski S.. Effects of stereochemistry, saturation, and hydrocarbon chain length on the ability of synthetic constrained azacyclic sphingolipids to trigger nutrient transporter down-regulation, vacuolation, and cell death. Bioorg. Med. Chem. (2016) 24 4390-4397. PMID: 27475534 74. Luengo A., Gui D.Y., Vander Heiden M.G.. Targeting metabolism for cancer therapy. Cell Chem. Biol. (2017) 24 1161-1180. PMID: 28938091 75. Makurvet F.D.. Biologics vs. Small molecules: drug costs and patient access. Med. Drug Discov. (2021) 9 100075 76. Paik J.. Risdiplam: a review in spinal muscular atrophy. CNS Drugs (2022) 36 401-410. PMID: 35284988 77. Xiao L., Tien J.C., Vo J., Tan M., Parolia A., Zhang Y., Wang L., Qiao Y., Shukla S., Wang X.. Epigenetic reprogramming with antisense oligonucleotides enhances the effectiveness of androgen receptor inhibition in castration-resistant prostate cancer. Cancer Res. (2018) 78 5731-5740. PMID: 30135193 78. Chen H., Chan D.C.. Mitochondrial dynamics in regulating the unique phenotypes of cancer and stem cells. Cell Metab. (2017) 26 39-48. PMID: 28648983 79. Friedman S.L., Neuschwander-Tetri B.A., Rinella M., Sanyal A.J.. Mechanisms of NAFLD development and therapeutic strategies. Nat. Med. (2018) 24 908-922. PMID: 29967350 80. Burté F., Carelli V., Chinnery P.F., Yu-Wai-Man P.. Disturbed mitochondrial dynamics and neurodegenerative disorders. Nat. Rev. Neurol. (2015) 11 11-24. PMID: 25486875
--- title: Is Triglyceride-Glucose Index a Valuable Parameter in Peripheral Artery Disease? journal: Cureus year: 2023 pmcid: PMC9976948 doi: 10.7759/cureus.35532 license: CC BY 3.0 --- # Is Triglyceride-Glucose Index a Valuable Parameter in Peripheral Artery Disease? ## Abstract Background The aim of this study was to investigate the relationship between the triglyceride-glucose (TyG) index and peripheral artery disease. Methodology This was a single-center, observational, retrospective study that included patients evaluated with color Doppler ultrasonography. A total of 440 individuals, 211 peripheral artery patients and 229 healthy controls, were included in the study. Results The TyG index levels were significantly higher in the peripheral artery disease group than in the control group (9.19 ± 0.57 vs. 8.80 ± 0.59; $p \leq 0.001$). The multivariate regression analysis conducted to determine the independent predictors of peripheral artery disease revealed that age (odds ratio (OR) = 1.111, $95\%$ confidence interval (CI) = 1.083-1.139; $p \leq 0.001$), male gender (OR = 0.441, $95\%$ CI = 0.249-0.782; $$p \leq 0.005$$), diabetes mellitus (OR = 1.925, $95\%$ CI = 1.018-3.641; $$p \leq 0.044$$), hypertension (OR = 0.036, $95\%$ CI = 0.285- 0.959; $$p \leq 0.036$$), coronary artery disease (OR = 2.540, $95\%$ CI = 1.376-4.690; $$p \leq 0.003$$), white blood cell count (OR = 1.263, $95\%$ CI = 1.029-1.550; $$p \leq 0.026$$), creatinine (OR = 0.975, $95\%$ CI = 0.952-0.999; $$p \leq 0.041$$), and TyG index (OR = 1.111, $95\%$ CI = 1.083-1.139; $p \leq 0.001$) were independent predictors of peripheral artery disease. The cut-off value of the TyG index in predicting peripheral artery disease was determined to be 9.06 with a sensitivity of $57.8\%$ and a specificity of $70\%$ (area under the curve = 0.689; $95\%$ CI = 0.640-0.738; $p \leq 0.001$). Conclusions High TyG index values can be used as an independent predictor of peripheral artery disease. ## Introduction Peripheral artery disease is a commonly encountered disease in the population, usually with atherosclerosis as its etiology, with high morbidity and mortality [1]. According to previous studies, its prevalence is estimated to range between $3\%$ and $13\%$ [2]. Because it mostly develops in the background of atherosclerosis, its concomitance with other vascular pathologies such as coronary artery disease and cerebrovascular diseases is common. Its incidence has been increasing with the aging population worldwide. It is a high-cost disease for health systems as it causes the loss of workforce and low quality of life, as well as morbidity and mortality [3]. The risk factors of atherosclerosis such as advanced age, hypertension, hyperlipidemia, diabetes, smoking, and chronic kidney failure are important in the development and course of peripheral artery disease [4]. Particularly, in peripheral artery disease accompanied by diabetes, arterial bed involvement is more commonly seen. In addition, severe pathologies such as extremity amputations are more frequently noted in peripheral artery patients with the diagnosis of diabetes mellitus [5,6]. As a result of insulin resistance, caused by hyperinsulinemia, the release of nitric oxide, a vasodilator, from the endothelium decreases, and the level of endothelin-1, which has vasoconstrictor properties, increases, which accelerates the development of atherosclerosis by contributing to endothelial dysfunction. Moreover, an increase in levels of fibrinogen and thromboxane as well as platelet aggregation is observed in connection with the increased insulin level [7,8]. Many methods are used to measure insulin resistance. Although the gold standard method for the detection of insulin resistance is the hyperinsulinemic-euglycemic clamp technique, it is utilized in clinical studies rather than being used routinely because it is time-consuming and costly to be performed in clinical practice [9]. Due to the difficulties experienced, some inexpensive and practical mathematical formulas have been developed to measure insulin resistance. The most commonly used method is the Homeostasis Model Assessment of Insulin Resistance (HOMA-IR) method which is calculated using fasting serum glucose and fasting serum insulin values. However, the fact that insulin is not studied in primary healthcare centers constitutes the biggest limitation of this method [10]. Thus, a simple and low-cost method that can be used in primary healthcare centers is needed. Hence, the triglyceride-glucose (TyG) index that is measured with the use of triglyceride and fasting blood sugar values has been developed [11]. Studies have demonstrated that insulin resistance measured by the TyG index highly correlates with insulin resistance detected by the HOMA-IR method [12]. Doppler ultrasonography is a rapid, non-invasive test that can be applied bedside in the anatomical evaluation of peripheral arteries and in the determination of the morphology of artery stenosis. It provides anatomical and functional information in peripheral artery disease and has a sensitivity of $92\%$ and a specificity of $97\%$ in the prediction of artery stenosis [13]. The most common symptom seen in peripheral artery disease is intermittent claudication [14]. However, about half of the patients present with asymptomatic or atypical pain [15]. Although it affects the quality of life, causes serious complications, and is associated with mortality, asymptomatic and symptomatic patients are diagnosed late and cannot receive adequate treatment. This situation negatively affects the prognosis of the disease [16]. Previous studies have shown the relationship between the TyG index and atherosclerosis and cardiovascular mortality. In this study, we investigated the role of the TyG index, a practical and inexpensive method, in the early diagnosis and prediction of peripheral artery disease. ## Materials and methods Study design and participants This was a single-center, observational, retrospective study of patients with peripheral artery disease. We included 211 consecutive patients with a diagnosis of peripheral artery disease. A total of 229 subjects without peripheral artery disease served as the control group. All included patients visited either the Cardiology or Cardiovascular Surgery department at Bahçelievler State Hospital, Istanbul between October 01, 2017, and October 01, 2022. At the time of diagnosis, patients under the age of 18, those with malignancy, those with chronic kidney failure and chronic liver disease, pregnant women, those receiving antidiabetic and antihyperlipidemic therapy, and those with active infection or chronic inflammatory disease were not included in the study. The study conformed to the principles outlined in the Declaration of Helsinki and was approved by the local ethics committee of Istanbul Bakirkoy Dr. Sadi Konuk Training and Research Hospital (05.20.2019, 2019-10-10). Data collection Hypertension was defined as systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg in clinical measurements or the use of antihypertensive drugs [17]. Diabetes mellitus was defined as a fasting glucose level ≥126 mg/dL (≥7.0 mmol/L) or the use of any antidiabetic drugs [18]. Hyperlipidemia was considered as the patient’s total cholesterol level above 200 mg/dL [19]. History of coronary artery disease was defined by a patient’s self-reported history of myocardial infarction or prior coronary revascularization, and history of stroke was defined as a patient’s self-reported history of stroke. Those who currently smoke and those who quit smoking one month before the study were considered active smokers. A Toshiba Applio500 (TUS A500) ultrasonography device and an 11 MHz linear probe were used in the arterial Doppler assessment of the lower extremity of the patients. The patients were evaluated from the aorta distal to the ankle after resting for 15 minutes. The proximal assessment of the popliteal artery and trifurcation arteries was performed in the prone position, while other arteries were evaluated in the supine position. Maximum velocities, flow form, and spectral changes were analyzed in the Doppler ultrasonography assessment. The peak systolic velocity ratio was calculated by dividing the maximum velocity at the narrowest part of any stenotic segment by the maximal velocity at 1.5-2 cm proximal to the stenosis. A ratio of 2 or greater was regarded as severe stenosis (over $50\%$). The diagnosis of occlusion was made in the case of no flow in the artery [20]. Blood samples were taken from the antecubital vein after 12 hours of fasting. The hematological parameters were determined using an XT-4000i Hematology Analyzer (Sysmex, Kobe, Japan). The biochemical analyses to detect alanine aminotransferase, aspartate aminotransferase, creatinine, and lipid profile (total cholesterol, high-density lipoprotein-cholesterol, low-density lipoprotein-cholesterol, and triglyceride levels) were performed with the AU5800 Clinical Chemistry System (Beckman Coulter, Inc., California, USA). The following formula was used to calculate the TyG index: TyG index ln[TG (mg/dL) fasting blood glucose (mg/dL)/2] [21]. Body mass index was calculated using the following formula: weight (kg)/height (m)2. Statistical analysis All statistical analyses were conducted using IBM SPSS Statistics version 22.0 (IBM Corp., Armonk, NY, USA) and MedCalc version 12.3.0.0 software packages. Normally distributed numerical variables are presented as mean ± standard deviation, whereas those that were not normally distributed are presented as median (minimum-maximum). To compare two independent samples, the independent-sample t-test was used when the normality assumption was met, whereas the Mann-Whitney U test was performed when not. The categorical variables are presented as number and percentage values, and the Pearson chi-square test was employed to compare the groups in terms of categorical variables. Correlations between parameters were determined by the Pearson correlation test for the normally distributed variables and by the Spearman correlation test for the non-normally distributed variables. Univariate and multivariate regression analyses were performed to identify risk factors for peripheral artery disease. Receiver operating characteristic (ROC) analysis was conducted to determine the cut-off positivity value of the TyG index. A p-value of less than 0.05 was considered statistically significant. ## Results Demographic characteristics and laboratory results of the patients included in our study are presented in Table 1. A total of 440 individuals, 211 patients with peripheral artery disease and 229 control individuals without peripheral artery disease, were included in the study. According to the gender of the patients, 159 ($75.4\%$) patients in the peripheral artery disease group and 114 ($49.2\%$) patients in the control group were male ($p \leq 0.001$). The mean age was 69.41 ± 10.61 years for patients with peripheral artery disease and 51.79 ± 15.15 years for the control group ($p \leq 0.001$). Table 1 \| Unnamed: 0 \| PAD group (n = 211) \| Control group (n = 229) \| P-value \| \| --- \| --- \| --- \| --- \| \| Age (years) \| 69.41 ± 10.61 \| 51.79 ± 15.15 \| <0.001 \| \| Gender, male, n (%) \| 159 (75.4%) \| 114 (49.8%) \| <0.001 \| \| Diabetes mellitus, n (%) \| 84 (39.8%) \| 33 (14.4%) \| <0.001 \| \| Hypertension, n (%) \| 120 (56.9%) \| 77 (33.6%) \| <0.001 \| \| Hyperlipidemia, n (%) \| 67 (31.8%) \| 48 (21.0%) \| 0.01 \| \| CAD, n (%) \| 92 (43.6%) \| 30 (13.1%) \| <0.001 \| \| CVD, n (%) \| 24 (11.4%) \| 12 (5.2%) \| 0.019 \| \| Glucose \| 121 (59–245) \| 96 (71–332) \| <0.001 \| \| Creatinine (mg/dL) \| 0.9 (0.42–3.32) \| 0.75 (0.27–1.70) \| <0.001 \| \| TC (mg/dL) \| 207.12 ± 46.54 \| 207.03 ± 39.74 \| 0.893 \| \| HDLc (mg/dL) \| 45.92 ± 35.34 \| 49.47 ± 12.91 \| 0.157 \| \| TG (mg/dL) \| 175 (58–882) \| 122(30–445) \| <0.001 \| \| LDLc (mg/dL) \| 121.62 ± 38.77 \| 124.11 ± 35.02 \| 0.479 \| \| WBC (103/mm3) \| 8.16 ± 2.09 \| 7.64 ± 1.84 \| <0.001 \| \| Hemoglobin (g/dL) \| 13.38 ± 1.99 \| 13.66 ± 1.53 \| 0.107 \| \| Platelets (103/mm3) \| 250.34 ± 71.01 \| 250.28 ± 60.31 \| 0.993 \| \| CRP (mg/L) \| 3.20 ± 1.19 \| 2.55 ± 1.42 \| <0.001 \| \| TyG index \| 9.19 ± 0.57 \| 8.80 ± 0.59 \| <0.001 \| The presence of coronary artery disease ($p \leq 0.001$), cerebrovascular disease ($$p \leq 0.019$$), diabetes mellitus ($p \leq 0.001$), hypertension ($p \leq 0.001$), and hyperlipidemia ($$p \leq 0.01$$) was higher in the peripheral artery disease group than in the control group (Table 1). Serum glucose (121 mg/dL, min-max = 59-245 vs. 96 mg/dL, min-max = 71-332; $p \leq 0.001$), creatinine (0.9 mg/dL, min-max = 0.42-3.32 vs. 0.75 mg/dL, min-max = 0.27-1.7; $p \leq 0.001$), triglyceride (175 mg/dL, min-max = 58-882 vs. 122 mg/dL, min-max = 30-445; $p \leq 0.001$), C-reactive protein (CRP) (3.20 ± 1.19 mg/L vs. 2.55 ± 1.42 mg/L; $p \leq 0.001$), white blood cell (WBC) count (8.16 ± 2.09 103/mm3 vs. 7.64 ± 1.84 103/mm3; $p \leq 0.001$), and TyG index (9.19 ± 0.57 vs. 8.80 ± 0.59; $p \leq 0.001$) levels were significantly higher in the peripheral artery disease group than in the control group. There were no significant differences between the groups regarding other parameters (Table 1). The results of the multivariate analysis conducted with the significant parameters found in the univariate analysis that was performed to identify the independent predictors of peripheral artery disease showed that age (odds ratio (OR) = 1.111, $95\%$ confidence interval (CI) = 1.083-1.139; $p \leq 0.001$), male gender (OR = 0.441, $95\%$ CI = 0.249-0.782; $$p \leq 0.005$$), diabetes mellitus (OR = 1.925, $95\%$ CI = 1.018-3.641; $$p \leq 0.044$$), hypertension (OR = 0.036, $95\%$ CI = 0.285-0.959; $$p \leq 0.036$$), coronary artery disease (OR = 2.540, $95\%$ CI = 1.376-4.690; $$p \leq 0.003$$), WBC (OR = 1.263, $95\%$ CI = 1.029-1.550; $$p \leq 0.026$$), creatinine (OR = 0.975, $95\%$ CI = 0.952-0.999; $$p \leq 0.041$$), and TyG index (OR = 1.111, $95\%$ CI = 1.083-1.139; $p \leq 0.001$) were the independent predictors of peripheral artery disease (Tables 2, 3). ROC curve analysis was performed to determine the best cut-off value of the TyG index in predicting peripheral artery disease. The cut-off value of the TyG index in the prediction of peripheral artery disease was found to be 9.06 with a sensitivity of $57.8\%$ and a specificity of 70 (area under the curve (AUC) = 0.689; $95\%$ CI = 0.640-0.738; $p \leq 0.001$) (Figure 1). Figure 1: Receiver operator characteristic curve (ROC) of the triglyceride-glucose (TyG) index in predicting peripheral artery disease. The correlation analysis to evaluate the relationship of the TyG index with clinical and laboratory variables showed that the TyG index was correlated positively with creatinine ($r = 0.128$, $$p \leq 0.007$$), total cholesterol ($r = 0.352$, $p \leq 0.001$), low-density lipoprotein-cholesterol ($r = 0.179$, $p \leq 0.001$), WBC ($r = 0.232$, $p \leq 0.001$), CRP ($r = 0.107$, $$p \leq 0.024$$), and age ($r = 0.243$, $p \leq 0.001$), and negatively with high-density lipoprotein (r = -0.411, $p \leq 0.001$) (Table 4). Table 4 \| Variables \| r-value \| P-value \| \| --- \| --- \| --- \| \| Creatinine \| 0.128 \| 0.007 \| \| TC \| 0.352 \| <0.001 \| \| HDLc \| -0.411 \| <0.001 \| \| LDLc \| 0.179 \| <0.001 \| \| WBC \| 0.232 \| <0.001 \| \| Hemoglobin \| 0.076 \| 0.110 \| \| Platelets \| 0.073 \| 0.126 \| \| CRP \| 0.107 \| 0.024 \| \| Age \| 0.243 \| <0.001 \| ## Discussion The main finding of our study is that the TyG index was higher in patients with peripheral artery disease than in patients in the control group. The relationship between the TyG index and cardiovascular diseases has been previously reported in various studies. In this study, with the use of multivariate analysis, we noted that a high TyG index value was an effective parameter in predicting peripheral artery disease. Atherosclerosis and the cardiovascular diseases it leads to are the most common causes of death worldwide [22]. Chronic and slowly progressing peripheral artery disease leads to a high cost for health systems with its high prevalence and the consequent loss of workforce [3]. Although different symptoms are observed according to the arterial area involved and the severity of the stenosis, the most frequent symptom is intermittent claudication, which occurs as a consequence of the inability to meet the increased blood demand of the lower extremity muscles during exercise [23]. However, in approximately half of the patients, the disease can manifest asymptomatically or present with atypical complaints [24]. It is crucial to make the diagnosis and begin the treatment at an early stage in patients with atypical complaints or asymptomatic patients to prevent complications and mortality the disease causes. Many parameters have been tested so far for this purpose. Hyperinsulinemia and increased insulin resistance are known to play a role in the pathophysiology of atherosclerosis through more than one mechanism. These can be listed as the disruption of the balance between nitric oxide and endothelin-1, which are vasodilator and vasoconstrictor substances that play a role in the maintenance of vascular tone in endothelial cells, against nitric oxide; increased chronic inflammation; vascular smooth muscle cell proliferation; and increased connective tissue in the vessel wall [25,26]. Plasma insulin level with its role in the atherosclerotic process constitutes an independent risk factor for cardiovascular diseases [27], which increases the importance of early detection of insulin resistance before cardiovascular diseases are clinically present, and of intervening. The TyG index, which can be easily calculated from fasting glucose and triglyceride levels, is an effective parameter demonstrating insulin resistance. After being associated with insulin resistance, it has been shown that the TyG index also positively correlates with the development of type 2 diabetes [28]. It is more practical than the HOMA-IR and the hyperinsulinemic-euglycemic clamp test, the gold standard test for determining insulin resistance, because it does not require any additional tests such as insulin level, and because it can be easily calculated using fasting glucose and triglyceride levels, which are routine tests in daily practice. Studies have shown that results of the TyG index and HOMA-IR correlate with the hyperinsulinemic-euglycemic clamp test result [29,30]. The relationship of the TyG index with atherosclerotic cardiovascular diseases has been demonstrated in many studies. In a study investigating the SYNTAX score and major adverse cardiovascular events of 438 patients with non-ST-segment elevation acute coronary syndrome, the TyG index was found to be an independent predictor of cardiovascular outcomes [31]. In another study in which 5,014 patients were followed for 10 years, the TyG index was shown to be related to newly emerging cardiovascular diseases that are independent of other cardiovascular risk factors such as age, gender, smoking, diabetes, and hypertension [32]. A study conducted in South Korea retrospectively evaluating 12,326 asymptomatic adult patients revealed that increased TyG index is an independent predictor of coronary artery calcification progression [33]. In a meta-analysis of eight cohort studies that included 5,731,234 participants with no known prior atherosclerotic cardiovascular disease, high TyG index values were associated with increased stroke, atherosclerotic cardiovascular disease, and coronary artery disease, independent of age, gender, and diabetes status [34]. In this study, we also detected that increased TyG index value is an independent predictor of peripheral artery disease, in line with the literature. Smoking is one of the most important preventable risk factors that play a role in the development of atherosclerosis. Endothelial damage and inflammation increase as a result of elevated oxidative stress due to smoking [35]. With endothelial damage and the direct effect of oxidative substances, platelet activation and predisposition to thrombosis develop [36]. Smoking also leads to the development of lipid abnormalities and glucose intolerance [37]. Many studies have reported that there is a strong association between smoking and peripheral artery disease depending on the consumed dose, and that continuation of smoking increases peripheral artery disease by two to sixfold [38]. It has been shown that in the follow-up of patients who quit smoking, when compared to patients who continue to smoke, they benefit more from revascularization treatments and there is a significant decrease in mortality rates [39]. Dyslipidemia and hypertension are two important risk factors in the development of atherosclerosis and play an important role in cardiovascular diseases. Clinical studies have revealed that high serum cholesterol levels and hypertension are among the leading risk factors for peripheral artery disease, as in coronary artery disease [40]. Every 40 mg/dL increase in total cholesterol level has been associated with a 1.2-fold increase in the risk of claudication. Similarly, it has been observed that the risk of claudication increases three to fourfold when blood pressure values are above $\frac{160}{95}$ mmHg [41]. As hypertension and dyslipidemia frequently coexist, it is recommended to treat the two conditions simultaneously [42]. CRP is an acute-phase protein synthesized in the liver during infection and inflammation, and its role in the course of atherosclerosis, which is an inflammatory process, has been shown in numerous studies [43]. In a study, it was determined that CRP level is a more powerful risk indicator than low-density lipoprotein-cholesterol in terms of cardiovascular events (myocardial infarction, cerebrovascular incidents, coronary revascularization, and death due to cardiovascular causes) [44]. Another study conducted among 144 healthy men followed for 60 months reported that CRP is a predictive indicator for the development of peripheral artery disease [45]. The fact that, in our study, CRP levels were statistically significantly higher in the peripheral artery disease group than in the control group suggests that high CRP levels can be an independent predictor in peripheral artery disease patients, in accordance with other studies in the literature. Limitations The biggest limitation of our study is that it was a single-center and retrospective study. Because insulin level was not routinely studied in patients, HOMA-IR values were not calculated, and their comparison with the TyG index results was not performed. Furthermore, grading peripheral artery disease in a larger sample group and revealing its variation with the TyG index could have strengthened our study. ## Conclusions It is important to screen and detect at-risk individuals before atherosclerosis-related diseases and related complications occur for preventive cardiology. In our study, the TyG index, which is a predictor of atherosclerosis, was found to be significantly higher in patients with peripheral artery disease than those in the control group. These results suggest that patients with a high TyG index value may be at risk for peripheral artery disease. Using the TyG index, patients at risk of peripheral artery disease can be early detected, and a decrease in mortality and morbidity can be achieved by preventing related complications. ## References 1. Jimbo M. Diagnosis and treatment of peripheral arterial disease. JAMA (2002) 287 314-315 2. Olin JW, Sealove BA. Peripheral artery disease: current insight into the disease and its diagnosis and management. Mayo Clin Proc (2010) 85 678-692. PMID: 20592174 3. Hirsch AT, Hartman L, Town RJ, Virnig BA. National health care costs of peripheral arterial disease in the Medicare population. Vasc Med (2008) 13 209-215. PMID: 18687757 4. Shammas NW. Epidemiology, classification, and modifiable risk factors of peripheral arterial disease. Vasc Health Risk Manag (2007) 3 229-234. PMID: 17580733 5. Beckman JA, Creager MA, Libby P. Diabetes and atherosclerosis: epidemiology, pathophysiology, and management. JAMA (2002) 287 2570-2581. PMID: 12020339 6. Pande RL, Perlstein TS, Beckman JA, Creager MA. Secondary prevention and mortality in peripheral artery disease: National Health and Nutrition Examination Study, 1999 to 2004. Circulation (2011) 124 17-23. PMID: 21690489 7. De Vriese AS, Verbeuren TJ, Van de Voorde J, Lameire NH, Vanhoutte PM. Endothelial dysfunction in diabetes. Br J Pharmacol (2000) 130 963-974. PMID: 10882379 8. Kakouros N, Rade JJ, Kourliouros A, Resar JR. Platelet function in patients with diabetes mellitus: from a theoretical to a practical perspective. Int J Endocrinol (2011) 2011 742719. PMID: 21869886 9. Bonora E, Targher G, Alberiche M. Homeostasis model assessment closely mirrors the glucose clamp technique in the assessment of insulin sensitivity: studies in subjects with various degrees of glucose tolerance and insulin sensitivity. Diabetes Care (2000) 23 57-63. PMID: 10857969 10. Zhang M, Wang B, Liu Y. Cumulative increased risk of incident type 2 diabetes mellitus with increasing triglyceride glucose index in normal-weight people: the Rural Chinese Cohort Study. Cardiovasc Diabetol (2017) 16 30. PMID: 28249577 11. Maratou E, Hadjidakis DJ, Kollias A. Studies of insulin resistance in patients with clinical and subclinical hypothyroidism. Eur J Endocrinol (2009) 160 785-790. PMID: 19141606 12. Du T, Yuan G, Zhang M, Zhou X, Sun X, Yu X. Clinical usefulness of lipid ratios, visceral adiposity indicators, and the triglycerides and glucose index as risk markers of insulin resistance. Cardiovasc Diabetol (2014) 13 146. PMID: 25326814 13. Norgren L, Hiatt WR, Dormandy JA, Nehler MR, Harris KA, Fowkes FG. Inter-society consensus for the management of peripheral arterial disease (TASC II). J Vasc Surg (2007) 45 Suppl S 0-67 14. Murphy TP, Cutlip DE, Regensteiner JG. Supervised exercise versus primary stenting for claudication resulting from aortoiliac peripheral artery disease: six-month outcomes from the claudication: exercise versus endoluminal revascularization (CLEVER) study. Circulation (2012) 125 130-139. PMID: 22090168 15. Benjamin EJ, Muntner P, Alonso A. Heart Disease and Stroke Statistics-2019 update: a report from the American Heart Association. Circulation (2019) 139 0 16. 8. Cardiovascular disease and risk management. Diabetes Care (2016) 39 Suppl 1 0-71 17. Williams B, Mancia G, Spiering W. 2018 ESC/ESH Guidelines for the management of arterial hypertension. Eur Heart J (2018) 39 3021-3104. PMID: 30165516 18. Standards of medical care in diabetes--2009. Diabetes Care (2009) 32 Suppl 1 0-61 19. Third report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) final report. Circulation (2002) 106 3143-3421. PMID: 12485966 20. Ranke C, Creutzig A, Alexander K. Duplex scanning of the peripheral arteries: correlation of the peak velocity ratio with angiographic diameter reduction. Ultrasound Med Biol (1992) 18 433-440. PMID: 1509618 21. Simental-Mendía LE, Rodríguez-Morán M, Guerrero-Romero F. The product of fasting glucose and triglycerides as surrogate for identifying insulin resistance in apparently healthy subjects. Metab Syndr Relat Disord (2008) 6 299-304. PMID: 19067533 22. Frostegård J. Immunity, atherosclerosis and cardiovascular disease. BMC Med (2013) 11 117. PMID: 23635324 23. Aboyans V, Ricco JB, Bartelink ME. 2017 ESC Guidelines on the Diagnosis and Treatment of Peripheral Arterial Diseases, in collaboration with the European Society for Vascular Surgery (ESVS): Document covering atherosclerotic disease of extracranial carotid and vertebral, mesenteric, renal, upper and lower extremity arteriesEndorsed by: the European Stroke Organization (ESO)The Task Force for the Diagnosis and Treatment of Peripheral Arterial Diseases of the European Society of Cardiology (ESC) and of the European Society for Vascular Surgery (ESVS). Eur Heart J (2018) 39 763-816. PMID: 28886620 24. Hirsch AT, Criqui MH, Treat-Jacobson D. Peripheral arterial disease detection, awareness, and treatment in primary care. JAMA (2001) 286 1317-1324. PMID: 11560536 25. Sandoo A, van Zanten JJ, Metsios GS, Carroll D, Kitas GD. The endothelium and its role in regulating vascular tone. Open Cardiovasc Med J (2010) 4 302-312. PMID: 21339899 26. Festa A, Hanley AJ, Tracy RP, D'Agostino R Jr, Haffner SM. Inflammation in the prediabetic state is related to increased insulin resistance rather than decreased insulin secretion. Circulation (2003) 108 1822-1830. PMID: 14517163 27. Reaven GM. Banting lecture 1988. Role of insulin resistance in human disease. Diabetes (1988) 37 1595-1607. PMID: 3056758 28. da Silva A, Caldas AP, Rocha DM, Bressan J. Triglyceride-glucose index predicts independently type 2 diabetes mellitus risk: a systematic review and meta-analysis of cohort studies. Prim Care Diabetes (2020) 14 584-593. PMID: 32928692 29. Hameed EK. TyG index a promising biomarker for glycemic control in type 2 diabetes mellitus. Diabetes Metab Syndr (2019) 13 560-563. PMID: 30641766 30. Selvi NM, Nandhini S, Sakthivadivel V, Lokesh S, Srinivasan AR, Sumathi S. Association of triglyceride-glucose index (TyG index) with HbA1c and insulin resistance in type 2 diabetes mellitus. Maedica (Bucur) (2021) 16 375-381. PMID: 34925590 31. Mao Q, Zhou D, Li Y, Wang Y, Xu SC, Zhao XH. The triglyceride-glucose index predicts coronary artery disease severity and cardiovascular outcomes in patients with non-ST-segment elevation acute coronary syndrome. Dis Markers (2019) 2019 6891537. PMID: 31281548 32. Sánchez-Íñigo L, Navarro-González D, Fernández-Montero A, Pastrana-Delgado J, Martínez JA. The TyG index may predict the development of cardiovascular events. Eur J Clin Invest (2016) 46 189-197. PMID: 26683265 33. Won KB, Park EJ, Han D. Triglyceride glucose index is an independent predictor for the progression of coronary artery calcification in the absence of heavy coronary artery calcification at baseline. Cardiovasc Diabetol (2020) 19 34. PMID: 32178666 34. Ding X, Wang X, Wu J, Zhang M, Cui M. Triglyceride-glucose index and the incidence of atherosclerotic cardiovascular diseases: a meta-analysis of cohort studies. Cardiovasc Diabetol (2021) 20 76. PMID: 33812373 35. Bermudez EA, Rifai N, Buring JE, Manson JE, Ridker PM. Relation between markers of systemic vascular inflammation and smoking in women. Am J Cardiol (2002) 89 1117-1119. PMID: 11988205 36. Matetzky S, Tani S, Kangavari S. Smoking increases tissue factor expression in atherosclerotic plaques: implications for plaque thrombogenicity. Circulation (2000) 102 602-604. PMID: 10931797 37. Benowitz NL, Kuyt F, Jacob P 3rd. Influence of nicotine on cardiovascular and hormonal effects of cigarette smoking. Clin Pharmacol Ther (1984) 36 74-81. PMID: 6734053 38. Tendera M, Aboyans V, Bartelink ML. ESC Guidelines on the diagnosis and treatment of peripheral artery diseases: Document covering atherosclerotic disease of extracranial carotid and vertebral, mesenteric, renal, upper and lower extremity arteries: the Task Force on the Diagnosis and Treatment of Peripheral Artery Diseases of the European Society of Cardiology (ESC). Eur Heart J (2011) 32 2851-2906. PMID: 21873417 39. 2011 ACCF/AHA focused update of the guideline for the management of patients with peripheral artery disease (updating the 2005 guideline): a report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines. Circulation (2011) 124 2020-2045. PMID: 21959305 40. Haider AW, Larson MG, Franklin SS, Levy D. Systolic blood pressure, diastolic blood pressure, and pulse pressure as predictors of risk for congestive heart failure in the Framingham Heart Study. Ann Intern Med (2003) 138 10-16. PMID: 12513039 41. Murabito JM, D'Agostino RB, Silbershatz H, Wilson WF. Intermittent claudication. A risk profile from the Framingham Heart Study. Circulation (1997) 96 44-49. PMID: 9236415 42. Meigs JB, D'Agostino RB Sr, Wilson PW, Cupples LA, Nathan DM, Singer DE. Risk variable clustering in the insulin resistance syndrome. The Framingham Offspring Study. Diabetes (1997) 46 1594-1600. PMID: 9313755 43. Ross R. The pathogenesis of atherosclerosis: a perspective for the 1990s. Nature (1993) 362 801-809. PMID: 8479518 44. Ridker PM. Inflammatory biomarkers and risks of myocardial infarction, stroke, diabetes, and total mortality: implications for longevity. Nutr Rev (2007) 65 0-9 45. Ridker PM, Cushman M, Stampfer MJ, Tracy RP, Hennekens CH. Plasma concentration of C-reactive protein and risk of developing peripheral vascular disease. Circulation (1998) 97 425-428. PMID: 9490235
--- title: recountmethylation enables flexible analysis of public blood DNA methylation array data authors: - Sean K Maden - Brian Walsh - Kyle Ellrott - Kasper D Hansen - Reid F Thompson - Abhinav Nellore journal: Bioinformatics Advances year: 2023 pmcid: PMC9976962 doi: 10.1093/bioadv/vbad020 license: CC BY 4.0 --- # recountmethylation enables flexible analysis of public blood DNA methylation array data ## Abstract ### Summary Thousands of DNA methylation (DNAm) array samples from human blood are publicly available on the Gene Expression Omnibus (GEO), but they remain underutilized for experiment planning, replication and cross-study and cross-platform analyses. To facilitate these tasks, we augmented our recountmethylation R/Bioconductor package with 12 537 uniformly processed EPIC and HM450K blood samples on GEO as well as several new features. We subsequently used our updated package in several illustrative analyses, finding (i) study ID bias adjustment increased variation explained by biological and demographic variables, (ii) most variation in autosomal DNAm was explained by genetic ancestry and CD4+ T-cell fractions and (iii) the dependence of power to detect differential methylation on sample size was similar for each of peripheral blood mononuclear cells (PBMC), whole blood and umbilical cord blood. Finally, we used PBMC and whole blood to perform independent validations, and we recovered 38–$46\%$ of differentially methylated probes between sexes from two previously published epigenome-wide association studies. ### Availability and implementation Source code to reproduce the main results are available on GitHub (repo: recountmethylation_flexible-blood-analysis_manuscript; url: https://github.com/metamaden/recountmethylation_flexible-blood-analysis_manuscript). All data was publicly available and downloaded from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). Compilations of the analyzed public data can be accessed from the website recount.bio/data (preprocessed HM450K array data: https://recount.bio/data/remethdb_h5se-gm_epic_0-0-2_1589820348/; preprocessed EPIC array data: https://recount.bio/data/remethdb_h5se-gm_epic_0-0-2_1589820348/). ### Supplementary information Supplementary data are available at Bioinformatics Advances online. ## 1 Introduction DNA methylation (DNAm) is the most commonly studied epigenetic mark, and most public DNAm array samples are generated from blood (Maden et al., 2021c). In prior work (Maden et al., 2021c), we conducted comprehensive cross-study analyses of human DNAm array studies with raw data deposited on the Gene Expression Omnibus (GEO) (Barrett et al., 2012; Edgar et al., 2002), the largest archive of publicly available array data. We confined attention to the HumanMethylation450K (HM450K) platform introduced by Illumina in 2012. HM450K arrays profile 485 577 CpG loci concentrated in protein-coding genes and CpG island regions (Bibikova et al., 2011; Sandoval et al., 2011). We found that: (i) a subset of Illumina’s prescribed BeadArray quality metrics explained most quality variances; (ii) samples clustered by tissue and cancer status in a principal component analysis (PCA) of autosomal DNAm; and (iii) subsets of CpG probes showed high tissue-specific DNAm variation among seven normal tissues. We further released the recountmethylation Bioconductor package (Huber et al., 2015) along with uniformly processed data compilations pairing DNAm with harmonized metadata labels for age, sex, tissue and disease state. The initial recountmethylation release left open two important issues. First, the prevalence of raw data from the newer EPIC platform (Pidsley et al., 2016) is rapidly increasing while our initial data compilation included only samples run on the older HM450K platform. Second, several practical research concerns were not accommodated in the initial package release, including how to leverage public array data compilations to determine the required number of samples to test a new hypothesis, how to account for confounding factors in cross-study analyses and how to leverage public data to independently validate previously published differentially methylated probes (DMPs) and identify subsets of high-confidence biomarker candidates. We address these outstanding issues in this article using novel cross-study and cross-platform analyses, confining attention to normal human blood samples. Blood DNAm is often probed in epigenome-wide association studies (EWAS) to discover, test and validate biomarkers (Li et al., 2012; Locke et al., 2019; Mikeska and Craig, 2014) for diseases, such as type II diabetes (Bacos et al., 2016; Dayeh et al., 2016; Willmer et al., 2018), obesity (Samblas et al., 2019), non-alcoholic fatty liver disease (Hyun and Jung, 2020), asthma (Thibeault and Laprise, 2019) and dementia (Fransquet et al., 2020), as well as colorectal (Alizadeh-Sedigh et al., 2021; Dong and Ren, 2018; Jensen et al., 2019; Lin et al., 2021), esophageal (Yu et al., 2018), breast (Guan et al., 2019), pancreatic (Henriksen and Thorlacius-Ussing, 2021) and head-and-neck (Danstrup et al., 2020) cancers. It is widely used to study biological aging (Haftorn et al., 2021; Hannum et al., 2013; Horvath, 2013; Merid et al., 2020) and normal tissue epigenetics (Åsenius et al., 2020; Huang et al., 2016), including development and function of the immune system (Parveen and Dhawan, 2021). Recent work studied how gestational age-related differential DNAm relates to fetal health and disease risk (Mayne et al., 2017; Merid et al., 2020). Further, cord blood DNAm is increasingly used to precisely quantify fetal gestational age (Bohlin et al., 2016; Haftorn et al., 2021; Knight et al., 2016; Lee et al., 2019), which may lead to improvement in the efficacy of prenatal screening (Fung et al., 2020). In addition, many software tools were trained and designed for use with blood DNAm data; these included methods for cell-type deconvolution (Houseman et al., 2012; Koestler et al., 2016; Salas et al., 2022), inference of population genetic structure and shared genetic ancestry (Rahmani et al., 2017) and power analyses (Graw et al., 2019). DNAm differences between sexes have been observed in mouse (Masser et al., 2017) and multiple human tissues including brain (Masser et al., 2017), pancreas (Hall et al., 2014), nasal epithelium (Nino et al., 2018), cord blood (Maschietto et al., 2017) and whole blood (Grant et al., 2021; Inoshita et al., 2015). These DNAm differences can impact insulin secretion (Hall et al., 2014), risk of disease (Maschietto et al., 2017; Nino et al., 2018) and biological age (Masser et al., 2017). Using cross-study and cross-platform compilations of whole blood and peripheral blood mononuclear cells (PBMC), we performed novel independent validation of previously published sets of probes with differential methylation between the sexes (a.k.a. ‘ sex DMPs’) from two previous studies in whole blood (Grant et al., 2021; Inoshita et al., 2015). ## 2.1 12 537 normal blood samples spanning 3 sample types were incorporated into recountmethylation We uniformly processed raw intensity data generated on the HM450K or EPIC platforms for 68 758 samples available on GEO before March 31, 2021 (Fig. 1, Section 4) (Aryee et al., 2014; Triche et al., 2013). We narrowed focus to 12 537 normal human blood samples from 63 studies, each of which had ≥10 samples after quality control. After harmonizing metadata across studies, we found these samples were predominantly of three types (Fig. 2a): whole blood, umbilical cord blood (a.k.a. ‘ cord blood’) and ‘PBMC’. Each blood sample type included ≥245 samples from ≥2 studies per respective platform (Fig. 2b and Supplementary Table S1). Fig. 1.: Workflow to obtain public DNAm array data from GEO. Collection, preparation and processing of array samples (top left) as well as publication of GEO datasets were performed by other investigators (top right). We downloaded raw intensity data (IDATs) and metadata (SOFTs; top right), processed GEO metadata (middle) and DNAm signals (bottom right) into HDF5-based data formats (bottom middle), and finally updated our server and the recountmethylation Bioconductor package (bottom left). Color outlines indicate data access and processing using tools we developed [green=recountmethylation_server (Maden et al., 2021a), blue=recountmethylation_pipeline (Maden et al., 2021b), green=recountmethylation_instance (Maden et al., 2022)]. Diagrams were created with BioRender.com Fig. 2.: Blood specimen collection and DNAm array data availability by sample type. (a) Blood sample collection and handling prior to upload to GEO. (b) Barplot summaries of available samples (left) and studies (right), showing counts (top) and percentages (bottom) of blood sample types (black=other/not otherwise specified, gray=whole blood and red=PBMCs and yellow=cord blood). Bar heights indicate the aggregate sample group ‘all’. Diagrams were created with BioRender.com Whole blood was distinguishable from PBMC from erythrocyte and granulocyte DNA, as these cell types are removed during PBMC preparation (Murray and Rajeevan, 2013) (Fig. 2a). However, PBMC granulocyte proportions showed strong study-specific trends (Supplementary Fig. S1). We further observed slight-to-moderate but highly significant correlations between estimated granulocyte proportions and quality metrics (Supplementary Table S2). We subsequently updated our Bioconductor package recountmethylation (Maden et al., 2021c) to facilitate cross-study and cross-platform analyses of the blood samples. The package’s new features permit search for samples with DNAm profiles similar to a query sample (Malkov and Yashunin, 2018), inference of shared genetic ancestry (Rahmani et al., 2017) and novel power analyses (Graw et al., 2019). These features are explained in package vignettes. Further, a new recountmethylation_instance Snakemake workflow available on GitHub (Maden et al., 2022) allows users to create their own compilations of public DNAm array data on GEO (Mölder et al., 2021), with the functionality to customize output data types and attributes predicted from GEO metadata. As shown below, our resources enable identification of biomarker candidates, independent validation and replication of previous research, experiment planning and more. Since the analyses contained in this article were performed, we have released updated recountmethylation compilations. The resource now spans all 93 306 HM450K and 38 122 EPIC samples with IDATs available on GEO before October 16, 2022. ## 2.2 Study ID adjustment increased variation explained by biological and demographic variables We conducted simulations investigating the impact of bias correction by study ID, a surrogate for technical confounders (Maden et al., 2021c). Three DNAm values were modeled in multiple regressions: (i) unadjusted DNAm, (ii) uniform adjustment on five randomly selected studies (a.k.a. ‘ adjustment 1’) and (iii) exact adjustment on 2–4 randomly selected studies (a.k.a. ‘ adjustment 2’). Regression models 2 and 3 were compared to test whether two distinct study ID bias adjustment strategies had comparable outcomes. We determined the fraction of explained variance (FEV) for each of 13 variables from ANOVA, yielding three results per variable per repetition of the simulation (Section 4). Total non-residual variances almost invariably decreased after applying either of the two study ID adjustment strategies (Supplementary Fig. S3a, median fractions of non-residual variances, adjusted over unadjusted, adj. 1 = 6.88e−1, adj. 2 = 6.84e−1). Variance reduction magnitudes were identical across adjustment strategies, with the exception of a few outlying models from Adjustment 1 simulations (Supplementary Fig. S3b). We categorized variables as biological (e.g. six predicted blood cell-type fractions), demographic (e.g. predicted sex, age and genetic ancestry), or technical (e.g. platform, where applicable). Across all three variable categories, FEV increased relative to unadjusted models after either adjustment strategy, and FEV distributions were far more similar among adjusted models than between adjusted and unadjusted models (Fig. 3a). The largest median FEV differences were observed for demographic variables, while the smallest were observed for technical variables. Among individual variables, median FEV was <0.1 across most models and variables, where study ID showed the maximum median FEV of 0.47 for unadjusted DNAm. After either adjustment, study median FEV decreased drastically to ≤2e–3, while median FEV for all remaining variables increased (Supplementary Table S3). Fig. 3.: Variance analyses of study bias adjustments and principal components. (a) Distributions of FEV. Violin plots show results grouped by three variable categories (plot titles, one of biological on left, demographic in middle, or technical on right), and color fills show model type (pink=Adjustment 1 or adjustment on five studies, green=Adjustment 2 or adjustment on 2–4 studies and blue=unadjusted). (b) Autosomal DNAm PCA results across normal blood samples. Stacked barplot y axes show the eigenvalue magnitudes at left and percentages at right for the top 10 components on the x axes. Fill colors indicate magnitudes of component sum of squared variances explained by variable categories (red=biological, green=demographic, blue=technical and purple=residuals). Variables were grouped into three categories: (i) biological (blood sample type and cell type); (ii) demographic (age, sex and two genetic ancestry components); and (iii) technical (platform and study ID) Because performing compilation-wide corrections on study ID substantially increased variation explained by biological and demographic variables, we included Beta-values under our adjusted models in recountmethylation for reuse in cross-study analyses. ## 2.3 Most DNAm variation was explained by predicted genetic ancestry and predicted cell composition To better understand key sources of variation in compiled blood data, we performed PCA on normalized (Triche et al., 2013), study ID-corrected autosomal DNAm, followed by ANOVA on regressions with 13 variables categorized as biological, demographic, or technical (Section 4). While most variation was residual across most components, explained variation was mainly from demographic variables at Components 1 and 3, biological variables at Components 4, 5, 6 and 10, and from technical variables at Component 8, and split between demographic and biological variables at Component 2 (Fig. 3b). Most explained variation from demographic variables was from genetic ancestry in the first component, while CD4+ T-cell fraction explained substantial biological variation across remaining top components (Supplementary Fig. S4). The top two principal components showed samples clustered largely independent from sample type and platform labels, but showed distinct gradient patterns for genetic ancestry, CD8+ T-cells, CD4+ T-cells and B-cells (Supplementary Fig. S5). ## 2.4 Dependence of statistical power on sample size was similar across blood sample types We conducted power analyses on the blood samples included in recountmethylation by applying the simulation-based pwrEWAS approach (Graw et al., 2019) (Section 4). To attain ≥$80\%$ power to detect DMPs between two groups of roughly equal size, the N estimated total samples required were similar across sample types, where N≈300 samples at mean Beta-value difference between groups δ=0.05, N≈150 samples at δ=0.1 and N≈80 samples at δ=0.2. We assumed an FDR threshold of $5\%$. Outcomes were similar within each of the whole blood, cord blood and PBMC groups, but they were worse when including all blood samples, likely due to greater sample heterogeneity (Fig. 4). Fig. 4.: Results of power analyses with the simulation-based pwrEWAS method (Graw et al., 2019). Curves indicate tradeoffs between power on the y axes and total samples N on the x axes, across three delta values (0.05 at left, 0.1 at middle and 0.2 at right). Curve colors show results for each sample type (line colors, blue=all, yellow=cord blood, red=PBMCs and gray=whole blood). Dotted horizontal lines show the 0.8 (80%) target power. Grayed regions and insets show magnifications of regions with low N Our results suggest fewer samples are necessary than the results of Graw et al. [ 2019], where adult PBMCs showed ≥$80\%$ power with $$n = 220$$ samples at δ=0.1. Further, an independent power analysis using whole-blood EPIC arrays (Mansell et al., 2019) found $85\%$ of probes had >$80\%$ power with $$n = 200$$ and δ=0.1, although their FDR cutoff value of $15\%$ was less stringent than our cutoff value of $5\%$. ## 2.5 Sex-specific differences in estimated blood cell fractions were consistent across sample types We investigated differences in blood cell proportions between sexes after correcting for sources of confounding (see above, Section 4). When we compared males with females using all four combined blood sample types, we found slight (<$3\%$) but significant (P-adjusted ≤1.7e−7) differences in five of six cell types—or every blood cell type except monocytes. Mean proportions of CD4+ T-cells, natural killer cells and B-cells were higher in males, while mean proportions of CD8+ T-cells and granulocytes were higher in females. Comparisons by blood sample type were similar in terms of the directionality and magnitude of mean cell proportion differences by sex. PBMC and whole blood showed the most significant differences (T-test P-adjusted, Benjamini–Hochberg method, <1e−3), where females had greater mean proportions of granulocytes and monocytes (≥$3.1\%$), and males had greater mean proportions of (≥$1.7\%$) CD8+ T-cells, CD4+ T-cells and B-cells (Supplementary Fig. S6 and Supplementary Table S4). ## 2.6 38% of sex DMPs from a previously published EWAS study were replicated in either whole blood or PBMC We queried a search index of blood autosomal CpG DNAm, which is included in the updated recountmethylation resource, for each of the 113 whole-blood samples from Inoshita et al. [ 2015]. In the process, we quantified the similarity of queried sample methylation profiles to other samples by analyzing the k nearest neighbors returned (Section 4). Among the 1000 nearest neighbors returned per queried sample, the whole-blood label was common while the PBMC label was rare (Fig. 5a), in agreement with Methods in Inoshita et al. [ 2015] describing the queried samples as ‘peripheral whole blood’. This greater similarity to compiled whole blood may reflect greater similarity in subject ages, cell composition (Murray and Rajeevan, 2013) and/or genetic ancestry (Supplementary Fig. S7), and we corrected for these potential confounders in regressions for identifying sex DMPs from either compilation (Section 4). Fig. 5.: Replication of sex DMPs from Inoshita et al. (2015) (a.k.a ‘Inoshita et al 2015’) using cross-study compilations of whole blood and PBMC. (a) Sample label distributions among the 1000 nearest neighbors from querying Inoshita et al. (2015) samples, where density and box plots show returned frequencies of three sample labels (red=PBMCs, black=other/not otherwise specified and gray=whole blood). (b) Concordance of Inoshita et al. (2015) DMPs on the y-axis among the top significant compilation DMPs on the x-axis, ranked on P-values, for whole blood in gray and PBMCs in red. The zoom shows the top 1000 DMPs, and colored dotted lines and colored numbers indicate total DMPs from Inoshita et al. (2015) (Section 4). (c and d) Mean beta-value differences (male–female) at Inoshita et al. (2015) DMPs on y axes and in (c) whole blood and (d) PBMC compilations. Region colors show direction agreement in gold and disagreement in blue, and insets show DMP counts by region We next considered the threshold of the top 1000 most significant DMPs from whole blood and PBMC. We set this threshold because these DMPs captured the long tail of high between-sex DNAm differences for each tissue (Supplementary Fig. S8a–d), and because of less replication divergence between tissues compared to less stringent thresholds observed from our concordance at the top analysis (Fig. 5b). ( $\frac{112}{292}$ =) $38\%$ of sex DMPs from Inoshita et al. [ 2015] were replicated in either whole blood or PBMC (Fig. 5b). Of these, 42 were only replicated in whole blood, 17 were only replicated in PBMC and 53 were replicated in both tissues (Supplementary Fig. S9a). Further, ($\frac{250}{544}$ =) $46\%$ of whole-blood sex DMPs independently reported in Grant et al. [ 2021] overlapped DMPs in whole blood or PBMC. However, just 26 sex DMPs appeared in all of PBMC, whole blood, Inoshita et al. [ 2015] and Grant et al. [ 2021]. Mean (normalized) Beta-value was typically higher in females than males in both whole blood (81−$93\%$ of DMPs) and PBMC ($68\%$ of DMPs). There was $100\%$ agreement in mean Beta-value differences between males and females among the subset of replicated DMPs in each compilation (Fig. 5c and d). Cytosine- and guanine-rich regions are known as CpG islands (Bird, 2002; Deaton and Bird, 2011; Gardiner-Garden and Frommer, 1987; Illumina, 2016; Takai and Jones, 2002), and a substantial number of replicated sex DMPs mapped at or proximal to a CpG island ($\frac{84}{112}$ = $75\%$). The most significant of these DMPs (P-adjusted <5.1e−47, Bonferroni method) mapped to a variety of functional gene regions, including two body-mapping DMPs (cg26355737 and cg04946709) at TFDP1 and LOC644649, and one promoter-mapping DMP (cg09066361) at GRM8. ## 3 Discussion We analyzed DNAm array data from the three most prevalent blood sample types in the GEO database and updated the recountmethylation Bioconductor package to make reproducible (Beaulieu-Jones and Greene, 2017; Heil et al., 2021) cross-study and cross-platform analyses of these data easier. Since HM450K and EPIC data continue to accumulate rapidly on GEO, we further developed the recountmethylation_instance Snakemake workflow to enable semi-automated compilation of the DNAm array data on GEO (Maden et al., 2022). We replicated $38\%$ of sex DMPs from Inoshita et al. [ 2015] and $46\%$ of sex DMPs from Grant et al. [ 2021] using independent whole blood and PBMC compilations. These rates were similar to prior studies of sex DNAm differences, including a $38\%$ validation rate of cord blood sex DMPs between two independent cohorts (Maschietto et al., 2017), and $44\%$ validation rate of genes in nasal epithelium with DNAm differences by sex (Nino et al., 2018). These results could represent a baseline expectation for replication or independent validation rate of DMPs for sex, and potentially other variables, across independent EWAS. Our work has several limitations. First, we excluded blood spots from our analyses due to insufficient raw DNAm array data available from GEO, although this blood sample type accounts for a substantial fraction of publicly available data from younger subjects. Another limitation related to data availability is that far fewer blood samples were available for the EPIC platform compared to HM450K as of March 31, 2021. The larger EPIC platform could help expand analyses to new genome regions and clarify regional DNAm signals at CpG islands and genes. The pwrEWAS method assumes a technical detection threshold of Beta-value = 0.01 by default, and using this threshold ensures our findings are relevant for both single-study and cross-study analyses. However, this technical threshold likely should be lowered if the study being planned involves cross-study analyses using study ID bias correction, because we found this correction reduced explained variances (section Study ID adjustment increased variation explained by biological and demographic variables) and resulted in lower between-group differences in our sex DMP cross-study analysis compared to the single-study discovery EWAS (Section $38\%$ of sex DMPs from a previously published EWAS study were replicated in either whole blood or PBMC). A further shortcoming of our analyses is that we did not investigate the influence of SNPs on DNAm, both proximally and distally. We note, however, that recountmethylation includes a feature to filter out probes overlapping known SNPs, as described in a package vignette. We also did not quantify nucleated red blood cells, a cell type with a highly distinctive DNAm profile characteristic of cord blood samples (Gervin et al., 2019). We intend to perform this quantification in a future version of recountmethylation. Finally, we did not conduct orthogonal or wet-lab validation of replicated sex DMPs. Such steps would be essential to narrow biomarker candidates and elucidate biological mechanisms explaining differential DNAm. Our work invites further exploration in several directions. First, future studies could apply our cross-study and cross-platform compilation of cord blood samples to validate other independent EWAS, such as Solomon et al. [ 2022], a recent meta-analysis of cord blood autosomal DNAm. Second, our linear correction for study-specific bias could be compared to alternative approaches, such as embedding alignment methods, which have been used to harmonize transcriptomics data across disparate platforms and data sources (Butler et al., 2018). Third, the population diversity encompassed by our cross-study compilations may permit more nuanced array-based studies of relationships between genetic variants and DNAm than previously possible. Fourth, our cross-study and cross-platform approach could be applied to other tissues with high prevalence among public datasets, such as brain (Maden et al., 2021c). Finally, our DNAm array data compilations could be expanded to include public bisulfite sequencing samples from the Sequence Read Archive (Leinonen et al., 2011), and these new samples can help clarify the genome region specificity and phased DNAm patterns proximal to significant DMPs and biomarker candidates (Noble et al., 2021; Pidsley et al., 2016; Wang et al., 2015). A future update of recountmethylation may include such samples. ## 4.1 Compiling recent public DNAm array data across platforms DNAm array data were identified, downloaded and processed using the recountmethylation_instance v0.0.1 (Maden et al., 2022) Snakemake workflow. It comprises the recountmethylation_server v1.0.0 (Maden et al., 2021a) and recountmethylation.pipeline v1.0.0 (Maden et al., 2021b) tools, which we used previously to compile HM450K data (Maden et al., 2021c). We uniformly processed samples into cross-study and cross-platform data compilations. Data were from samples run on the HM450K (Bibikova et al., 2011) and EPIC (Pidsley et al., 2016) DNAm array platforms and available on GEO by March 31, 2021. Sample DNAm was normalized using out of band signal (a.k.a. ‘ noob-normalization’) (Triche et al., 2013). Compilations paired normalized DNAm fractions, or Beta-values, with harmonized sample metadata for 68 758 cumulative samples for which raw image data files, or IDATs, were available as gzip-compressed Supplementary files. Compilations were stored as Hierarchical Data Format 5 (HDF5)-based SummarizedExperiment files generated using the HDF5Array v1.18.0 and rhdf5 v2.34.0 R/Bioconductor packages (Fischer et al., 2020; Pagès, 2021). These formats used DelayedArray v0.18.0 (Pagès et al., 2021) to support rapid access, summaries and filters. For most analyses, DNAm data were merged across platforms for the 453 093 CpG probes (Pidsley et al., 2016) they shared. We made compiled data available online at https://recount.bio/data/gr-gseadj_h5se_hm450k-epic-merge_0-0-$\frac{3}{.}$ ## 4.2 Prediction and harmonization of sample metadata We generated harmonized sample metadata from heterogeneous metadata mined from SOFT files accompanying GEO studies. We wrote regex terms to detect keywords in metadata-containing files, and we mapped detected terms to controlled vocabularies under ‘tissue’, ‘disease’ and other categories, as described in Maden et al. ( 2021c) and Lowe and Rakyan [2013]. The suitability of regex patterns for capturing informative metadata terms was spot-checked across studies and iteratively updated to more precisely map terms and avoid erroneous mappings. We further predicted sample types from mined metadata using the method from Bernstein et al. [ 2017]. To add sex annotations, we applied the getSex() function from minfi v1.37.1 R package with argument defaults. To add six blood type cell fractions, we applied the estimateCellCounts() function from minfi with argument defaults (Aryee et al., 2014; Fortin et al., 2017). This applies the deconvolution method from Houseman et al. [ 2012] on the blood reference dataset originally published in Reinius et al. [ 2012] and distributed in the FlowSorted. Blood.450k v1.34.0 R package. We added age annotations using the pan-tissue epigenetic clock model from Horvath [2013], which was implemented in the wateRmelon v2.2.0 package. Finally, we calculated the top components of genetic ancestry using the EPISTRUCTURE method (Rahmani et al., 2017). We noted limited overlap among CpG probes used in the above models to predict or impute missing sample metadata fields. Genetic ancestry was predicted from 4913 probes previously found to have strong association with genetic ancestry-defining SNPs after correction for bias from factors including predicted cell-type heterogeneity (Rahmani et al., 2017). Ages were obtained from 353 CpG probes previously found to robustly predict age across tissues and platforms (Horvath, 2013), of which two (cg03760483 and cg04431054) overlapped probes for genetic ancestry predictions. Blood cell-type fractions were predicted from deconvolution of cell type-specific array-wide DNAm signals. The original method on which this approach is based was previously validated using probes from the older HM27K platform (Houseman et al., 2012), none of which overlapped probes used to predict either genetic ancestry or age. ## 4.3 Sample QC filters We used metadata filters to find the three most prevalent blood sample types (whole blood, cord blood and PBMCs), and to define the aggregate type ‘all’, which includes the above types and blood samples whose specific type could not be determined from their metadata. We then performed QC with reference to prior findings from Maden et al. ( 2021c). We removed samples for which either: (i) log2 median M and U signals were both <10; or (ii) the sample failed ≥$\frac{2}{5}$ most informative BeadArray quality metrics. These metrics, described by Illumina in Illumina [2010, 2015], pertain to CpG probe chemistry and performance [also see Maden et al. ( 2021c) for details]. These filtering criteria removed 245 samples, and all but one was run on the HM450K platform. We additionally filtered PBMCs with high estimated granulocyte proportions (≥0.25), and this threshold was set to remove the long tail (93rd quantile) of the granulocyte proportion distribution across studies (Supplementary Fig. S1 and Supplementary Table S2). This removed 47 PBMC samples and left 580 remaining. ## 4.4 Simulation of study bias adjustments We used simulations to show the impact of study ID adjustment on explained variance. As detailed in Supplementary Figure S2, simulations consisted of four steps: (i) calculate sample DNAm M-values from 500 CpG probes and 5 studies, selected randomly; (ii) adjust study ID across all five selected studies (i.e. ‘adjustment 1’) or subsets of 2–4 studies (i.e. ‘adjustment 2’); (iii) perform ANOVA for three models; (iv) get FEV for each variable across three models. In total, simulations used 29 028 unique CpG probes and 62 unique studies. Multiple regression models accounted for sample type, platform, study ID, DNAm-based predictions for age, sex and six cell-type fractions and two genetic ancestry components, which were determined as described above. Variables were grouped into three categories: (i) biological (blood sample type and cell type); (ii) demographic (age, sex and two genetic ancestry components); and (iii) technical (platform and study ID). Study bias adjustments were performed using the removeBatchEffect() function from the limma v3.46.0 (Ritchie et al., 2015) R package. Parallel sessions were deployed using the parallel v4.1.1 R package. ## 4.5 PCA of autosomal DNAm We performed autosomal DNAm PCA on compiled blood samples using a reduced 1000D representation of the normalized and bias-corrected Beta-values (Kane and Nelson, 2014; Williams, 2005) obtained via feature hashing [see Maden et al. ( 2021c) for details on this approach]. For the top 10 components, we calculated FEV from ANOVA using multiple regression models containing the 13 variables from the 3 categories described above. ## 4.6 Blood autosomal DNAm search index construction We used the hnswlib v0.5.2 Python library to make a DNAm-based search index from which one can rapidly identify the nearest samples which neighbor one or more queried DNAm profiles (Malkov and Yashunin 2018). We used the Hierarchical and Navigable Small Worlds graph algorithm implemented in hnswlib, as this was among the top performing algorithms from a recent comprehensive benchmark of search algorithms (Aumüller et al., 2018). With the mmh3 v3.0.0 and numpy v1.20.1 (Harris et al., 2020) Python libraries, we applied feature hashing to generate a reduced 1000D representation of each sample (Kane and Nelson, 2014; Weinberger et al., 2010) of each blood sample’s noob-normalized Beta-values. The search index files are available online at https://recount.bio/data/sindex-hnsw_bval-gseadj-fh10k_all-blood-2-platforms.pickle and https://recount.bio/data/sidict-hnsw__bval-gseadj-fh10k__all-blood-2-platforms.pickle. ## 4.7 Power analyses using pwrEWAS We used the method provided in the pwrEWAS v1.4.0 R/Bioconductor library to perform power analyses across DNAm array platforms (Graw et al., 2019). Parameters for these analyses included 100 total simulations varying the total samples N from 50 to 850. We targeted 500 DMPs and assessed test group Beta-value differences δ of 0.05, 0.1 and 0.2. ## 4.8 Replication of whole-blood sex DMPs We replicated sex DMPs from Inoshita et al. [ 2015], a study of whole blood from Japanese individuals, using independent compilations of whole blood and PBMC samples in recountmethylation. After filtering out sex chromosome and cross-reactive probes (Chen et al., 2013), there were 375 244 CpG probes in whole blood and 375 244 CpG probes in PBMCs. After filtering for sample quality, we used data from 5980 whole-blood samples (3942 females and 2924 males) and 580 PBMC samples (357 females and 223 males). Ages tended toward young adult and middle-aged for whole blood (age, mean±SD, 39±21 years) and samples from Inoshita et al. [ 2015] (46±12 years), but were more frequently from adolescents and young adults among PBMC (25±19 years). We preprocessed DNAm M-values using surrogate variables analysis with the sva v3.4.0 R package (Leek et al., 2021). We determined sex DMPs using coefficient P-values for the sex variable in multiple regressions, where regression models corrected for bias from biological (six predicted blood cell-type fractions), demographic (predicted age) and technical variables (platform and study ID). ## 4.9 Statistical analyses and visualizations Data processing and analyses were performed using the R v4.1.0 and Python v3.7.1 programming languages (Python Core Team, 2019; R Core Team, 2021). Statistical summaries and tests were performed using base R libraries. DNAm array processing, normalization, analysis and prediction of sex and six blood cell-type fractions were performed using the minfi, minfiData and minfiDataEPIC R packages. Workflow diagrams were created using BioRender.com. Visualizations in Section 2 made use of the ggplot2 v3.3.2, grid v4.1.3, gridExtra v2.3, UpSetR v1.4.0, ggpubr v0.4.0, ggforce v0.3.3 and png v0.1-7 R packages (Gehlenborg, 2019; Wickham, 2016). P-value adjustments used either the Bonferroni method or the Benjamini–Hochberg method (Benjamini and Hochberg, 1995). Enrichment tests used the binom.test() base R function with the background of 453 093 total probes overlapping both array platforms (R Core Team, 2021). Supplementary scripts and functions recreating our results are available online at https://www.github.com/metamaden/recountmethylation_flexible-blood-analysis_manuscript. ## 4.10 Supplementary data, files and code The following resources have been provided to reproduce results, figures and tables in this article: ## Funding This work was supported by the National Institutes of Health [5R01GM121459-02 to K.D.H.]. Conflict of Interest: none declared. ## Data availability Data used in this study were publicly available and downloaded from the Gene Expression Omnibus (GEO) repository at National Center for Biotechnology Information (NCBI) website (https://www.ncbi.nlm.nih.gov/geo). Compilations of the analyzed public data can be accessed from the website recount.bio/data (preprocessed HM450K array data: https://recount.bio/data/remethdb_h5se-gm_epic_0-0-2_1589820348/; preprocessed EPIC array data: https://recount.bio/data/remethdb_h5se-gm_epic_0-0-2_1589820348/). ## Disclaimer The contents do not represent the views of the U.S. Department of Veterans Affairs or the United States Government. ## References 1. Alizadeh-Sedigh M.. Methylation of FBN1, SPG20, ITF2, RUNX3, SNCA, MLH1, and SEPT9 genes in circulating cell-free DNA as biomarkers of colorectal cancer. Cancer Biomark (2021) 34 221-250 2. Aryee M.J.. Minfi: a flexible and comprehensive Bioconductor package for the analysis of Infinium DNA methylation microarrays. Bioinformatics (2014) 30 1363-1369. PMID: 24478339 3. Åsenius F.. The DNA methylome of human sperm is distinct from blood with little evidence for tissue-consistent obesity associations. PLoS Genet (2020) 16 e1009035. PMID: 33048947 4. Aumüller M.. (2018) 5. Bacos K.. Blood-based biomarkers of age-associated epigenetic changes in human islets associate with insulin secretion and diabetes. Nat. Commun (2016) 7 11089. PMID: 27029739 6. Barrett T.. NCBI GEO: archive for functional genomics data sets—update. Nucleic Acids Res (2012) 41 D991-D995. PMID: 23193258 7. Beaulieu-Jones B.K., Greene C.S.. Reproducibility of computational workflows is automated using continuous analysis. Nat. Biotechnol (2017) 35 342-346. PMID: 28288103 8. Benjamini Y., Hochberg Y.. Controlling the false discovery rate: a practical and powerful approach to multiple testing. JSRRB (1995) 57 289-300 9. Bernstein M.N.. MetaSRA: normalized human sample-specific metadata for the sequence read archive. Bioinformatics (2017) 33 2914-2923. PMID: 28535296 10. Bibikova M.. High density DNA methylation array with single CpG site resolution. Genomics (2011) 98 288-295. PMID: 21839163 11. Bird A.. DNA methylation patterns and epigenetic memory. Genes Dev (2002) 16 6-21. PMID: 11782440 12. Bohlin J.. Prediction of gestational age based on genome-wide differentially methylated regions. Genome Biol (2016) 17 207. PMID: 27717397 13. Butler A.. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol (2018) 36 411-420. PMID: 29608179 14. Chen Y-a.. Discovery of cross-reactive probes and polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray. Epigenetics (2013) 8 203-209. PMID: 23314698 15. Danstrup C.S.. DNA methylation biomarkers in peripheral blood of patients with head and neck squamous cell carcinomas. A systematic review. PLoS One (2020) 15 e0244101. PMID: 33332423 16. Dayeh T.. DNA methylation of loci within ABCG1 and PHOSPHO1 in blood DNA is associated with future type 2 diabetes risk. Epigenetics (2016) 11 482-488. PMID: 27148772 17. Deaton A.M., Bird A.. CpG islands and the regulation of transcription. Genes Dev (2011) 25 1010-1022. PMID: 21576262 18. Dong L., Ren H.. Blood-based DNA methylation biomarkers for early detection of colorectal cancer. J. Proteomics Bioinform (2018) 11 120-126. PMID: 30034186 19. Edgar R.. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res (2002) 30 207-210. PMID: 11752295 20. Fischer B.. (2020) 21. Fortin J.-P.. Preprocessing, normalization and integration of the Illumina HumanMethylationEPIC array with minfi. Bioinformatics (2017) 33 558-560. PMID: 28035024 22. Fransquet P.D.. Blood DNA methylation signatures to detect dementia prior to overt clinical symptoms. Alzheimers Dement. (Amst.) (2020) 12 e12056. PMID: 32671182 23. Fung R.. Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study. Lancet Digit. Health (2020) 2 e368-e375. PMID: 32617525 24. Gardiner-Garden M., Frommer M.. CpG islands in vertebrate genomes. J. Mol. Biol (1987) 196 261-282. PMID: 3656447 25. Gehlenborg N.. (2019) 26. Gervin K.. Systematic evaluation and validation of reference and library selection methods for deconvolution of cord blood DNA methylation data. Clin. Epigenetics (2019) 11 125. PMID: 31455416 27. Grant O.A.. (2021). DOI: 10.1101/2021.09.02.458717 28. Graw S.. pwrEWAS: a user-friendly tool for comprehensive power estimation for epigenome wide association studies (EWAS). BMC Bioinformatics (2019) 20 218. PMID: 31035919 29. Guan Z.. Whole-blood DNA methylation markers in early detection of breast cancer: a systematic literature review. Cancer Epidemiol. Biomarkers Prev (2019) 28 496-505. PMID: 30487132 30. Haftorn K.L.. An EPIC predictor of gestational age and its application to newborns conceived by assisted reproductive technologies. Clin. Epigenetics (2021) 13 82. PMID: 33875015 31. Hall E.. Sex differences in the genome-wide DNA methylation pattern and impact on gene expression, microRNA levels and insulin secretion in human pancreatic islets. Genome Biol (2014) 15 522. PMID: 25517766 32. Hannum G.. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol. Cell (2013) 49 359-367. PMID: 23177740 33. Harris C.R.. Array programming with NumPy. Nature (2020) 585 357-362. PMID: 32939066 34. Heil B.J.. Reproducibility standards for machine learning in the life sciences. Nat. Methods (2021) 18 1132-1135. PMID: 34462593 35. Henriksen S.D., Thorlacius-Ussing O.. Cell-free DNA methylation as blood-based biomarkers for pancreatic adenocarcinoma—a literature update. Epigenomes (2021) 5 8. PMID: 34968295 36. Horvath S.. DNA methylation age of human tissues and cell types. Genome Biol (2013) 14 R115. PMID: 24138928 37. Houseman E.A.. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics (2012) 13 86. PMID: 22568884 38. Huang Y.-T.. Epigenome-wide profiling of DNA methylation in paired samples of adipose tissue and blood. Epigenetics (2016) 11 227-236. PMID: 26891033 39. Huber W.. Orchestrating high-throughput genomic analysis with bioconductor. Nat. Methods (2015) 12 115-121. PMID: 25633503 40. Hyun J., Jung Y.. DNA methylation in nonalcoholic fatty liver disease. Int. J. Mol. Sci (2020) 21 8138. PMID: 33143364 41. Illumina. (2010) Illumina Genome Studio Methylation Module v1.8. https://support.illumina.com/content/dam/illumina-support/documents/documentation/software_documentation/genomestudio/genomestudio-2011-1/genomestudio-methylation-v1-8-user-guide-11319130-b.pdf (1 December 2021, date last accessed).. (2010) 42. Illumina. (2015) BeadArray Controls Reporter Software Guide. https://support.illumina.com/content/dam/illumina-support/documents/documentation/chemistry_documentation/infinium_assays/infinium_hd_methylation/beadarray-controls-reporter-user-guide-1000000004009-00.pdf (11 December 2021, date last accessed).. (2015) 43. Illumina. (2016) Field Guide to Methylation Methods. https://www.illumina.com/content/dam/illumina-marketing/documents/products/other/field_guide_methylation.pdf (23 November 2021, date last accessed).. Field Guide to Methylation Methods (2016) 44. Inoshita M.. Sex differences of leukocytes DNA methylation adjusted for estimated cellular proportions. Biol. Sex Differ (2015) 6 11. PMID: 26113971 45. Jensen S.Ø.. Novel DNA methylation biomarkers show high sensitivity and specificity for blood-based detection of colorectal cancer—a clinical biomarker discovery and validation study. Clin. Epigenetics (2019) 11 158. PMID: 31727158 46. Kane D.M., Nelson J.. Sparser Johnson-Lindenstrauss transforms. J. ACM (2014) 61 1-23 47. Knight A.K.. An epigenetic clock for gestational age at birth based on blood methylation data. Genome Biol (2016) 17 206. PMID: 27717399 48. Koestler D.C.. Improving cell mixture deconvolution by identifying optimal DNA methylation libraries (IDOL). BMC Bioinformatics (2016) 17 120. PMID: 26956433 49. Lee Y.. Placental epigenetic clocks: estimating gestational age using placental DNA methylation levels. Aging (Albany NY) (2019) 11 4238-4253. PMID: 31235674 50. Leek J.T.. (2021) 51. Leinonen R.. The sequence read archive. Nucleic Acids Res (2011) 39 D19-D21. PMID: 21062823 52. Li L.. DNA methylation in peripheral blood: a potential biomarker for cancer molecular epidemiology. J. Epidemiol (2012) 22 384-394. PMID: 22863985 53. Lin W.-H.. Circulating tumor DNA methylation marker MYO1-G for diagnosis and monitoring of colorectal cancer. Clin. Epigenetics (2021) 13 232. PMID: 34961566 54. Locke W.J.. DNA methylation cancer biomarkers: translation to the clinic. Front. Genet (2019) 10 1150. PMID: 31803237 55. Lowe R., Rakyan V.K.. Marmal-aid - a database for Infinium HumanMethylation450. BMC Bioinformatics (2013) 14 359. PMID: 24330312 56. Maden S.. (2021a) 57. Maden S.. (2021b) 58. Maden S.. (2022) 59. Maden S.K.. Human methylome variation across Infinium 450K data on the Gene Expression Omnibus. NAR Genom. Bioinform (2021c) 3 lqab025. PMID: 33937763 60. Malkov Y.A., Yashunin D.A.. (2018) 61. Mansell G.. Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array. BMC Genomics (2019) 20 366. PMID: 31088362 62. Maschietto M.. Sex differences in DNA methylation of the cord blood are related to sex-bias psychiatric diseases. Sci. Rep (2017) 7 44547. PMID: 28303968 63. Masser D.R.. Sexually divergent DNA methylation patterns with hippocampal aging. Aging Cell (2017) 16 1342-1352. PMID: 28948711 64. Mayne B.T.. Accelerated placental aging in early onset preeclampsia pregnancies identified by DNA methylation. Epigenomics (2017) 9 279-289. PMID: 27894195 65. Merid S.K.. Epigenome-wide meta-analysis of blood DNA methylation in newborns and children identifies numerous loci related to gestational age. Genome Medicine (2020) 12 25. PMID: 32114984 66. Mikeska T., Craig J.M.. DNA methylation biomarkers: cancer and beyond. Genes (2014) 5 821-864. PMID: 25229548 67. Mölder F.. Sustainable data analysis with Snakemake. F1000Res (2021) 10 33. PMID: 34035898 68. Murray J.R., Rajeevan M.S.. Evaluation of DNA extraction from granulocytes discarded in the separation medium after isolation of peripheral blood mononuclear cells and plasma from whole blood. BMC Res. Notes (2013) 6 440. PMID: 24176175 69. Nino C.L.. Characterization of Sex-Based DNA methylation signatures in the airways during early life. Sci. Rep (2018) 8 5526. PMID: 29615635 70. Noble A.J.. A validation of Illumina EPIC array system with bisulfite-based amplicon sequencing. PeerJ (2021) 9 e10762. PMID: 33614276 71. Pagès H.. (2021) 72. Pagès H.. (2021) 73. Parveen N., Dhawan S.. DNA methylation patterning and the regulation of beta cell homeostasis. Front. Endocrinol. (Lausanne) (2021) 12 651258. PMID: 34025578 74. Pidsley R.. Critical evaluation of the Illumina MethylationEPIC BeadChip microarray for whole-genome DNA methylation profiling. Genome Biol (2016) 17 208. PMID: 27717381 75. Python Core Team. (2019) Python: A Dynamic, Open Source Programming Language. Python version 3.7. Python Software Foundation, Fredericksburg, Virginia. https://www.python.org/.. Python: A Dynamic, Open Source Programming Language (2019) 76. R Core Team. (2021) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.. R: A Language and Environment for Statistical Computing (2021) 77. Rahmani E.. Genome-wide methylation data mirror ancestry information. Epigenetics Chromatin (2017) 10 1. PMID: 28149326 78. Reinius L.E.. Differential DNA methylation in purified human blood cells: implications for cell lineage and studies on disease susceptibility. PLoS One (2012) 7 e41361. PMID: 22848472 79. Ritchie M.E.. Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res (2015) 43 e47. PMID: 25605792 80. Salas L.A.. Enhanced cell deconvolution of peripheral blood using DNA methylation for high-resolution immune profiling. Nat. Commun (2022) 13 761. PMID: 35140201 81. Samblas M.. DNA methylation markers in obesity, metabolic syndrome, and weight loss. Epigenetics (2019) 14 421-444. PMID: 30915894 82. Sandoval J.. Validation of a DNA methylation microarray for 450,000 CpG sites in the human genome. Epigenetics (2011) 6 692-702. PMID: 21593595 83. Solomon O.. Meta-analysis of epigenome-wide association studies in newborns and children show widespread sex differences in blood DNA methylation. Mutat. Res. Rev. Mutat. Res (2022) 789 108415. PMID: 35690418 84. Takai D., Jones P.A.. Comprehensive analysis of CpG islands in human chromosomes 21 and 22. Proc. Natl. Acad. Sci. USA (2002) 99 3740-3745. PMID: 11891299 85. Thibeault A.-A.H., Laprise C.. Cell-specific DNA methylation signatures in asthma. Genes (2019) 10 932. PMID: 31731604 86. Triche T.J.. Low-level processing of Illumina Infinium DNA Methylation BeadArrays. Nucleic Acids Res (2013) 41 e90. PMID: 23476028 87. Wang T.. A systematic study of normalization methods for Infinium 450K methylation data using whole-genome bisulfite sequencing data. Epigenetics (2015) 10 662-669. PMID: 26036609 88. Weinberger K.. (2010) 89. Wickham H.. ggplot2: Elegant Graphics for Data Analysis (2016) 90. Williams R.. A new algorithm for optimal 2-constraint satisfaction and its implications. Theor. Comput. Sci (2005) 348 357-365 91. Willmer T.. Blood-based DNA methylation biomarkers for type 2 diabetes: potential for clinical applications. Front. Endocrinol. (Lausanne) (2018) 9 744. PMID: 30564199 92. Yu M.. Subtypes of Barrett’s oesophagus and oesophageal adenocarcinoma based on genome-wide methylation analysis. Gut (2018) 68 389-399. PMID: 29884612
--- title: 'Facilitators and Barriers of Tai Chi Practice in Community-Dwelling Older Adults: Qualitative Study' journal: Asian/Pacific Island Nursing Journal year: 2023 pmcid: PMC9976991 doi: 10.2196/42195 license: CC BY 4.0 --- # Facilitators and Barriers of Tai Chi Practice in Community-Dwelling Older Adults: Qualitative Study ## Abstract ### Background Numerous studies have documented the beneficial effects of Tai Chi on a variety of health outcomes, especially in older adults. However, only few studies have examined how to improve the practice and adherence of this Asian-originated exercise among older adults in Western countries. ### Objective This study aimed to identify facilitators and barriers to Tai Chi practice and adherence in community-dwelling older adults. ### Methods This study analyzed the qualitative data collected from 13 participants (mean age 62.0, SD 10.3) at the end of a 15-week randomized controlled trial conducted at a day activity senior center. Semistructured interviews were conducted, recorded, and transcribed; and the data were analyzed using inductive thematic analysis. ### Results Four themes emerged: perceived benefit, threats, facilitators, and barriers. Perceived threats (eg, aging and side effects of medications) and perceived benefits of Tai Chi (eg, balance) inspired participants’ engagement in Tai Chi exercise. On the other hand, barriers to Tai Chi practice and adherence included instructor’s teaching style, the complexity of Tai Chi postures and movements, and existing health conditions (eg, hip problems). In essence, factors like Tai Chi class availability, family and peer support, as well as practicing Tai Chi with music may facilitate Tai Chi exercise adherence. ### Conclusions The study findings could provide valuable information to health professionals, such as nurses and physical therapists, in developing and implementing effective Tai Chi programs in care plans. Considering health conditions, tailoring Tai Chi exercise instruction styles, encouraging social and peer support, and incorporating music may promote Tai Chi practice and adherence. ## Background The population worldwide is rapidly aging, and the global percentage of adults aged 65 years and older is projected to double by the year 2050 [1]. Aging is frequently accompanied with increased chronic health conditions, including but not limited to osteoporosis, sarcopenia, cancer, heart disease, stroke, diabetes, and Alzheimer disease [2-4]. It is widely evidenced that chronic conditions significantly increase the risk of falls and physical disability, resulting in poor quality of life and premature death [5-8]. Tai Chi, a body-mind practice originating in China, has generated increasing attention from health professionals, including nurses, due to evidence that suggests Tai Chi’s ability to enhance health and well-being indices. A growing body of studies have documented the beneficial effects of Tai Chi on a variety of health outcomes, especially in the older population. The number of Tai Chi studies that are indexed in MEDLINE or PubMed increased from 9 before 1990 to 105 between 1990 and 2003, then rising to 234 between 2004 and 2008, and even higher between 2009 and 2013 to 362 [9]; this number increased to 2336 between 2014 and 2021. The health benefits of Tai Chi practice include but are not limited to physical function [10-13], cardiovascular diseases [14], mental health [15,16], the musculoskeletal system [17,18], balance and fall prevention [13,19], and cognitive function [20,21]. One of the implications of practicing Tai Chi consistently is relative to improved health benefits. For example, a systematic review reports that the frequency of Tai Chi practice is important for fall prevention in older adults [19]. However, like most types of exercise programs, barriers exist that limits adherence to Tai Chi exercise. Understanding these barriers and facilitators becomes essential for health professionals to develop effective Tai Chi interventions that promotes mind-body exercise for optimal health benefits. ## Objective In spite of the fact that there have been many studies demonstrating numerous health benefits associated with Tai Chi practice, only a small number of studies have looked at the barriers and facilitators involved in the practice. Gryffin et al [22] suggest that inadequate information and teaching style may serve as an obstacle for Tai Chi practice. However, this study did not address the facilitators of engaging in and adhering to Tai Chi practice. Another study found that encouragement from social supports is a factor that motivates older people to start practicing Tai Chi, and subsequent positive health outcomes from the exercise program can help motivate people to continue practicing Tai Chi [23]. This study was conducted in Taiwan [23]; therefore, the results may not be generalized to individuals living in the Western countries. In addition, even though Tai Chi has been proven to provide health benefit to certain patient populations, we are not aware of any examples of facilitators and barriers in the African American community as it relates to Tai Chi. The objective of this study was to explore the facilitators and barriers of Tai Chi practice and adherence in both White and African American older adults. ## Study Participants Our study reports the findings of the qualitative data collected at the end of a 15-week randomized controlled trial, which assessed the effects of practicing Tai Chi with music on fall-related factors. The trial was conducted in the fall of 2014 at a day center in the Southern United States, offering a variety of creative arts and activity programs for adults aged 50 and older. A total of 13 women were enrolled, and block randomly assigned into a Tai Chi practice with music group or Tai Chi practice without music group. A detailed study design of the randomized controlled trial was documented in early reports [24]. ## Ethics Approval This study was approved by the Tulane University Institute Review Board (#630231). Written consents were obtained from all participants. ## Data Collection At the end of the 15-week Tai Chi exercise intervention, a semistructured interview was conducted with each of the 13 study participants. An interview guide was developed based on the Health Belief Model, which consists of the following 6 concepts: perceived susceptibility, perceived severity, perceived benefits, perceived barriers, cues to action, and self-efficacy (Multimedia Appendix 1) [25]. Health Belief Model is a theory designed to predict health behaviors to promote good health outcomes [26]. This has been used frequently in nursing to identify factors relative to positive behavior changes [27]. This model suggests that an individual’s perception about health problems, perceived benefits of intervention and barriers to intervention, self-efficacy, and cues to action explains engagement (or lack of engagement) in health-promoting behaviors [25]. The semistructured interview questions were organized in the following segments: [1] perceived susceptibility and severity of health issues occurring with aging, [2] motivations of Tai Chi practice, [3] perceived benefits of Tai Chi practice, and [4] perceived facilitators and barriers for Tai Chi practice. Open ended questions such as “What made you sign up for Tai Chi class?” and “Was there anything that stopped you from practicing Tai Chi?” were included in the interview. Probe questions were further asked when appropriate or deemed necessary to explore participants’ experience and perceptions of Tai Chi practice. Two trained graduate students conducted the semistructured interviews, and each interview lasted about 15-30 minutes. All interviews were audio-recorded and transcribed. ## Data Analysis Data were analyzed using NVivo (version 8.0; QSR International). Although the Health Belief Model was used to develop the interview guide, the data were analyzed using inductive thematic analysis—a method of identifying, analyzing, and reporting patterns (ie, themes, topics, and ideas) within data without predetermined themes to guide coding processes [28,29]. First, 2 researchers immersed themselves in the qualitative data to become acquainted with the content; throughout, they made notes, comments, and ideas of coding the data. Second, the two researchers independently coded the 13 interviews using open coding, and then the researchers gathered together to reconcile code differences in their respective analyses. Coding discrepancies were discussed between the two researchers until a consensus was reached. Third, one researcher grouped the codes into themes based on the similarities and differences of the codes and cited relevant quotes for each theme, while the other researcher reviewed the themes created, and then both discussed to reach an agreement as different opinions arose. ## Trustworthiness The trustworthiness related to credibility, transferability, dependability, and confirmability was enhanced through various approaches starting at the study design stage [30,31]. For instance, prolonged engagement at the study site, member checking, and team meetings were used to improve credibility. Even though the study was conducted at a single senior site, it included both White and African American participants, which might improve the transferability, given that little is known about African Americans regarding this topic. Audio recordings of all conducted interviews were adopted to increase dependability. Confirmability was enhanced through approaches such as coding the data independently with 2 researchers. ## Sample Characteristics Table 1 shows characteristics of the study participants. Participants in the study were female, aged 50 to 84 years, with an average age of 69.2 (SD 8.5) years. Half of the participants were African American. This was a group of relatively high educated adults, with $82\%$ ($\frac{11}{13}$) having higher than high school education. Only $23\%$ ($\frac{3}{13}$) still worked full- or part-time. Around $46\%$ ($\frac{6}{13}$) of the participants either were married or lived with a partner. The average reported exercise hours per week was 4.5 (SD 2.1) hours, including walking, yoga, ballet, and strength training (not shown in Table 1). In total, $61\%$ ($\frac{9}{13}$) of the participants had practiced Tai Chi before this study, mostly for one semester and at the same facility as in this study, with a different volunteer instructor. The average class attendance rate for the clinical trial was $71\%$. Four major themes related to this study topic emerged from the qualitative interviews, as follows: perceived threats, perceived benefits, perceived facilitators, and perceived barriers for Tai Chi practice and adherence. The corresponding codes for each theme are displayed in Figure 1. ## Perceived Threats The perceived threats due to aging were typical reasons that inspired participants to engage in exercise, including Tai Chi. One of the oldest participants in the study stated the following: When talking about what drives Tai Chi practice, some participants responded with a combination of reasons related to aging and chronic health conditions, like one participant said: In particular, cancer was frequently repeated as a health threat in this study population. One cancer survivor said the following: Relatedly, the side effect of drugs was also motivation for these reasons. For example, another cancer survivor responded: ## Perceived Benefits Perceived health benefits consist mainly of two aspects: the benefits they learned from scientific report or other sources; and the benefits they experienced themselves. Perceived benefits from other sources were usually stated as the motivation to start Tai Chi practice. Participants described appreciation of Tai Chi in a variety of ways. Most participants stated that they had prior first- or second-hand knowledge of Tai Chi. Many were aware of the reported health benefits of Tai Chi, particularly balance improvement. For example, one participant in response to being asked about reasons for wanting to take part in Tai Chi, said the following: Additionally, as they started to practice Tai Chi exercises, experienced benefits were stated as factors that facilitate the continuity to practice and adhere to Tai Chi. In addition to frequently perceived balance improvement, participants also mentioned the psychological aspect they gained from the movements, such as the following statement by a participant: Likewise, characteristics of Tai Chi movements were perceived positively by participants. The gracefulness of the movements was especially attractive; In that regard, a participant said the following: The slowness of the movements was also acknowledged; one participant, after stating that she had some health issues and considered her age, mentioned, ## Perceived Barriers Similar to participating in other exercises, self-discipline and time management are among the most common barriers for engaging in Tai Chi practice, with several participants mentioning self-discipline as the most challenging barrier that kept them from adhering to Tai Chi class schedules. For instance, one participant said the following: Time was another factor several individuals talked about, as one participant said: The complexity of Tai Chi postures and movements were likewise identified as barriers by the participants. Several participants stated it was difficult for them to master movements. For example, one participant said the following: Furthermore, finding the right instructor with an appropriate teaching style for the older population was also a barrier. One participant stated the following: Some subjects perceived multiple barriers at the same time. A participant stated: Lastly, some existing health conditions also restricted participants from practicing, such as hip and knee issues. One participant said the following in this regard: ## Perceived Facilitators Facilitators included class availability, music with Tai Chi practice, and support from family, friends, and peers. Availability of classes can also promote Tai Chi practice or adherence. One person stated, “I would commit to the class if I had a class (available to me).” In this study, music was added as another component to increase motivation with Tai Chi practice, and participants in the Tai Chi group with music component indicated that music helped them focus and enjoy more of the practice: Similarly, all participants enrolled in the Tai Chi without music group wanted to add music in the future when practicing Tai Chi. In addition, peers, friends, and family members were among the most common facilitators for adhering to Tai Chi exercise, with one participant commenting the following: One member of the Tai Chi classes reported: ## Perceived Facilitators and Barriers This qualitative study identified several facilitators (eg, practicing with music and class availability) and barriers (eg, lack of quality instructors and complexity of Tai Chi movements) of Tai Chi practice and adherence perceived by community-dwelling older adults. These findings are important for nurses and other health care professionals to develop and recommend effective Tai Chi programs and interventions for older adults’optimal health benefits. Perceived health conditions and aging were the two major motivators for Tai Chi practice, and perceived health benefit of Tai Chi was another motivator for Tai Chi practice. This is consistent with the literature suggesting that perceived threats of health could be a motivator for engaging in healthy lifestyle behaviors [32] and Tai Chi practice [33]. Previous studies have widely documented the health benefits of Tai Chi practice in the older population [11,19,34,35], especially because of its gentle and slow movements [22]. Meanwhile, certain health problems, such as major physical disability, might be an obstacle for engaging in Tai Chi exercise. Fortunately, Tai Chi exercise can be modified and tailored to individuals with physical limitations. For example, the wheelchair Tai Chi includes modified exercises for participants in wheelchairs and has been proven effective for people with disabilities [36]. Therefore, providing tailored instructions and recommendations to target populations is warranted to improve Tai Chi practice and adherence. In addition to similar barriers to participating in other exercises, such as time restriction and self-discipline [37], participants perceived some unique challenges when practicing this Eastern exercise. First, unlike other exercise, Tai Chi classes are not always available in the community. Even though learning from videos is possible, it is very different from learning on site, especially considering that Tai Chi movements have a substation stretching and turning of the body, which may be difficult to perceive over video and easier to learn with the assistance of an instructor. This could be one of the reasons for retaining participants in Tai Chi practice and Tai Chi studies [38]. Second, Tai Chi was considered both physically and cognitively challenging by the participants. Tai Chi’s origin is from martial arts, and it is an intricate combination of individual head, hand, arm, leg, ankle, upper body, and lower body movements. Tai Chi involves continuous, slow, and rhythmic dynamic loading and unloading with the ability to gradually modify the difficulty of the task, all of which is needed for joint health. Current Tai Chi research can be divided into those analyzing the practice and those that introduce Tai Chi movements, further analyzing their therapeutic effects on particular maladies. Tai Chi, whether performed as an exercise or woven into daily life for fall prevention, is beneficial to the body without causing secondary problems, especially to the joints. Tai Chi has several different styles, including but not limited to the Chen, Yang, Sun, and Wu styles [39]; some of these styles are more physically and cognitively challenging than others. If instructors do not consider older adults’ physical and cognitive changes, Tai Chi exercise may be unnecessarily taxing for this population and deter them from practicing it. In addition, the selection of Tai Chi forms is critical to the success of Tai Chi as a therapeutic intervention; thus, it is crucial that a more precise estimate of joint movement within Tai Chi forms be incorporated into future studies to understand how Tai Chi can optimize joint kinematics and kinetics, then identify the biomechanical mechanisms and their association with different Tai Chi forms. Therefore, instructors who teach Tai Chi to older adults could select the most optimal forms and movements to maximize Tai Chi’s benefits and minimize its harms. It would also be beneficial to standardize the training process for Tai Chi instructors in both future research studies and general practice in the community. In addition to instructor’s teaching style, music may also play a critical role in influencing Tai Chi learning and adherence. Studies have documented improved learning occurring when music is paired with movements in the music therapy technique of entrainment. Entrainment occurs when music is paired with an activity, further described as “a temporal locking process in which one system’s motion or signal frequency entrains the frequency of another system” [40]. Via this principle, linking movement to rhythm may establish a kinetic pattern that is easier and faster, increases confidence, and therefore, promptly leads to increased compliance in attendance. Although teaching long and complicated movement patterns is traditionally taught with “chunking” (ie, grouping together chunks of information and focusing on one chunk at a time), compound cues may actually improve acquisition [41]; and the addition of music, therefore, encodes basic movements and facilitate progression to difficult patterns. Lastly, this is one of the first studies to include African American participants in the study of Tai Chi practice. Perceived facilitators and barriers of Tai Chi practice among African American participants were similar to those of their counterparts. Even though the longevity of African Americans is increasing, they generally undergo more chronic conditions and have a higher risk of disability [42,43], which may be improved with Tai Chi. Literature supports the health benefits of Tai Chi practice [9]; therefore, it is important to conduct further studies with larger samples and thoughtful research designs to examine Tai Chi pratice in African American pupulation and other minority groups. ## Limitations There were a few limitations to this study. One of the limitations was that all study participants were from a single senior center and were previously enrolled in a Tai Chi class. In addition, all our participants reported having at least high school education or a higher level of education. Therefore, the generalizability of the study findings to other populations is limited, and studies that include a diverse population are still needed. Data from this study were collected in 2014, and despite there being a few other studies examining similar topics since 2014 [22,44], our study population included White and African American participants, which resulted in some unique findings. For example, we found that using music may promote Tai Chi learning experience and Tai Chi practice adherence among this racial diversity, which would be very helpful in implementing Tai Chi in the community settings, particularly; in the face of increasing evidence that reveals the health benefits of Tai Chi exercise in older adults, little is known regarding how to disseminate Tai Chi to diverse older populations. Thus, these results are worthy of being reported and publicized, as this would help guide the development of Tai Chi programs, and it will benefit the aging community. Nurses play an essential role in health promotion, educating the public and patients on the prevention and management of health conditions, providing evidence-based care and support, advocating for health-related programs and policies, as well as advancing nursing care through research. Tai Chi, as a mind-body exercise, can be practiced in various community settings, including but not limited to hospitals, senior communities, clinics, and nursing homes. The study findings provide valuable information for nurses to develop or identify effective Tai Chi programs to improve health outcomes in older adults. In addition, research exploring strategies to tailor Tai Chi programs to promote Tai Chi practice in populations with different health conditions and background is needed. ## Conclusions This study found that perceived aging, health issues, and health benefits were common reasons for choosing to practice Tai Chi. Importantly, the barriers to its practice and adherence (eg, lack of quality instructors) need to be addressed; and facilitators, such as practicing with music and class availability, need to be promoted. Although studies have been trending upward about the health benefits of Tai Chi constantly, most of them are very limited in terms of translational forethought. Therefore, research in exploring the dissemination and promotion of Tai Chi exercise is warranted. For instance, strategies must be explored to address the shortage of qualified instructors and train them to meet specific health needs, especially for older adults. Additionally, incorporating music into Tai Chi may reduce anxiety and promote adherence to Tai Chi practices. ## References 1. He W, Goodkind D, Kowal P. An aging world: 2015 International population reports. United States Census Bureau (2016) 2. Jaul E, Barron J. Age-related diseases and clinical and public health implications for the 85 years old and over population. Front Public Health (2017) 5 335. DOI: 10.3389/fpubh.2017.00335 3. Dao T, Green AE, Kim YA, Bae S, Ha K, Gariani K, Lee M, Menzies KJ, Ryu D. Sarcopenia and muscle aging: a brief overview. Endocrinol Metab (2020) 35 716-732. DOI: 10.3803/enm.2020.405 4. Johnston CB, Dagar M. Osteoporosis in older adults. Med Clin North Am (2020) 104 873-884. DOI: 10.1016/j.mcna.2020.06.004 5. Ning H, Du Y, Zhao Y, Liu Q, Li X, Zhang H, Jiang D, Feng H. Longitudinal impact of metabolic syndrome and depressive symptoms on subsequent functional disability among middle-aged and older adults in China. J Affect Disord (2022) 296 216-223. DOI: 10.1016/j.jad.2021.09.061 6. Gandham A, Mesinovic J, Jansons P, Zengin A, Bonham MP, Ebeling PR, Scott D. Falls, fractures, and areal bone mineral density in older adults with sarcopenic obesity: a systematic review and meta-analysis. Obes Rev (2021) 22 e13187. DOI: 10.1111/obr.13187 7. Vaishya R, Vaish A. Falls in older adults are serious. Indian J Orthop (2020) 54 69-74. DOI: 10.1007/s43465-019-00037-x 8. Zhang X, Huang P, Dou Q, Wang C, Zhang W, Yang Y, Wang J, Xie X, Zhou J, Zeng Y. Falls among older adults with sarcopenia dwelling in nursing home or community: a meta-analysis. Clin Nutr (2020) 39 33-39. DOI: 10.1016/j.clnu.2019.01.002 9. Harmer PA. So much research, so little application: barriers to dissemination and practical implementation of Tai Ji Quan. J Sport Health Sci (2014) 3 16-20. DOI: 10.1016/j.jshs.2013.10.005 10. Tai H, Chou Y, Tzeng I, Wei C, Su C, Liu W, Kung W. Effect of Tai Chi synergy T1 exercise on autonomic function, metabolism, and physical fitness of healthy individuals. Evid Based Complement Alternat Med (2018) 2018 6351938-7. DOI: 10.1155/2018/6351938 11. Choo YT, Jiang Y, Hong J, Wang W. Effectiveness of Tai Chi on quality of life, depressive symptoms and physical function among community-dwelling older adults with chronic disease: A systematic review and meta-analysis. Int J Nurs Stud (2020) 111 103737. DOI: 10.1016/j.ijnurstu.2020.103737 12. Gu R, Gao Y, Zhang C, Liu X, Sun Z. Effect of Tai Chi on cognitive function among older adults with cognitive impairment: a systematic review and meta-analysis. Evid Based Complement Alternat Med (2021) 2021 6679153-9. DOI: 10.1155/2021/6679153 13. Nyman S. Tai Chi for the prevention of falls among older adults: a critical analysis of the evidence. J Aging Phys Act (2021) 29 343-352. DOI: 10.1123/japa.2020-0155 14. Liu T, Chan AW, Liu YH, Taylor-Piliae RE. Effects of Tai Chi-based cardiac rehabilitation on aerobic endurance, psychosocial well-being, and cardiovascular risk reduction among patients with coronary heart disease: a systematic review and meta-analysis. Eur J Cardiovasc Nurs (2018) 17 368-383. DOI: 10.1177/1474515117749592 15. Hu L, Wang Y, Liu X, Ji X, Ma Y, Man S, Hu Z, Cheng J, Huang F. Tai Chi exercise can ameliorate physical and mental health of patients with knee osteoarthritis: systematic review and meta-analysis. Clin Rehabil (2021) 35 64-79. DOI: 10.1177/0269215520954343 16. Wang F, Lee EO, Wu T, Benson H, Fricchione G, Wang W, Yeung AS. The effects of tai chi on depression, anxiety, and psychological well-being: a systematic review and meta-analysis. Int J Behav Med (2014) 21 605-17. DOI: 10.1007/s12529-013-9351-9 17. Woo J, Hong A, Lau E, Lynn H. A randomised controlled trial of Tai Chi and resistance exercise on bone health, muscle strength and balance in community-living elderly people. Age Ageing (2007) 36 262-8. DOI: 10.1093/ageing/afm005 18. Chang T, Ting YT, Sheu SL, Chang HY. [Effects of tai chi in postmenopausal women with osteoporosis: a systematic review]. Hu Li Za Zhi (2014) 61 75-84. DOI: 10.6224/JN.61.5.75 19. Huang Z, Feng Y, Li Y, Lv C. Systematic review and meta-analysis: Tai Chi for preventing falls in older adults. BMJ Open (2017) 7 e013661. DOI: 10.1136/bmjopen-2016-013661 20. Wayne PM, Walsh JN, Taylor-Piliae RE, Wells RE, Papp KV, Donovan NJ, Yeh GY. Effect of tai chi on cognitive performance in older adults: systematic review and meta-analysis. J Am Geriatr Soc (2014) 62 25-39. DOI: 10.1111/jgs.12611 21. Liu F, Chen X, Nie P, Lin S, Guo J, Chen J, Yu L. Can Tai Chi improve cognitive function? A systematic review and meta-analysis of randomized controlled trials. J Altern Complement Med (2021) 27 1070-1083. DOI: 10.1089/acm.2021.0084 22. Gryffin PA, Chen WC, Chaney BH, Dodd VJ, Roberts B. Facilitators and barriers to tai chi in an older adult community: a theory-driven approach. Am J Health Educ (2015) 46 109-118. DOI: 10.1080/19325037.2014.999964 23. Chen K, Snyder M, Krichbaum K. Facilitators and barriers to elders' practice of t'ai chi. A mind-body, low-intensity exercise. J Holist Nurs (2001) 19 238-55. DOI: 10.1177/089801010101900305 24. Du Y, Roberts P, Xu Q. The effects of Tai Chi practice with asynchronous music on compliance and fall-related risk factors in middle-aged and older women: a pilot study. J Holist Nurs (2017) 35 142-150. DOI: 10.1177/0898010116636972 25. Jones CL, Jensen JD, Scherr CL, Brown NR, Christy K, Weaver J. The Health Belief Model as an explanatory framework in communication research: exploring parallel, serial, and moderated mediation. Health Commun (2015) 30 566-76. DOI: 10.1080/10410236.2013.873363 26. Rosenstock IM, Strecher VJ, Becker MH. Social learning theory and the Health Belief Model. Health Educ Q (1988) 15 175-83. DOI: 10.1177/109019818801500203 27. Cengiz B, Bahar Z, Canda AE. The effects of patient care results of applied nursing intervention to individuals with stoma according to the health belief model. Cancer Nurs (2018) 43 E87-E96. DOI: 10.1097/ncc.0000000000000678 28. Castleberry A, Nolen A. Thematic analysis of qualitative research data: Is it as easy as it sounds?. Curr Pharm Teach Learn (2018) 10 807-815. DOI: 10.1016/j.cptl.2018.03.019 29. Chapman AL, Hadfield M, Chapman CJ. Qualitative research in healthcare: an introduction to grounded theory using thematic analysis. J R Coll Physicians Edinb (2015) 45 201-5. DOI: 10.4997/JRCPE.2015.305 30. Guba EG. Criteria for assessing the trustworthiness of naturalistic inquiries. ECTJ (1981) 29 75-91. DOI: 10.1007/bf02766777 31. Roberts P, Du Y, Xu Q. The effects of T'ai Chi with asynchronous music on the health of older women: a pilot study. Altern Complement Ther (2017) 23 14-19. DOI: 10.1089/act.2016.29093.per 32. Villar OAE, Montañez-Alvarado P, Gutiérrez-Vega M, Carrillo-Saucedo IC, Gurrola-Peña GM, Ruvalcaba-Romero NA, García-Sánchez MD, Ochoa-Alcaraz SG. Factor structure and internal reliability of an exercise health belief model scale in a Mexican population. BMC Public Health (2017) 17 229. DOI: 10.1186/s12889-017-4150-x 33. Yang G, Li X, Peel N, Klupp N, Liu J, Bensoussan A, Kiat H, Chang D. Perceptions of participants on trial participation and adherence to Tai Chi: a qualitative study. PPA (2022) Volume 16 2695-2707. DOI: 10.2147/ppa.s380780 34. Huston P, McFarlane B. Health benefits of tai chi: what is the evidence?. Can Fam Physician (2016) 62 881-890. PMID: 28661865 35. Wayne PM, Walsh JN, Taylor-Piliae RE, Wells RE, Papp KV, Donovan NJ, Yeh GY. Effect of tai chi on cognitive performance in older adults: systematic review and meta-analysis. J Am Geriatr Soc (2014) 62 25-39. DOI: 10.1111/jgs.12611 36. Wang YT, Li Z, Yang Y, Zhong Y, Lee S, Chen S, Chen Y. Effects of wheelchair Tai Chi on physical and mental health among elderly with disability. Res Sports Med (2016) 24 157-70. DOI: 10.1080/15438627.2016.1191487 37. Picorelli AMA, Pereira LSM, Pereira DS, Felício Diogo, Sherrington C. Adherence to exercise programs for older people is influenced by program characteristics and personal factors: a systematic review. J Physiother (2014) 60 151-6. DOI: 10.1016/j.jphys.2014.06.012 38. Wayne PM, Kaptchuk TJ. Challenges inherent to t'ai chi research: part I--t'ai chi as a complex multicomponent intervention. J Altern Complement Med (2008) 14 95-102. DOI: 10.1089/acm.2007.7170a 39. Wolf SL, Coogler C, Xu T. Exploring the basis for Tai Chi Chuan as a therapeutic exercise approach. Arch Phys Med Rehabil (1997) 78 886-892. DOI: 10.1016/s0003-9993(97)90206-9 40. Thaut MH, McIntosh GC, Hoemberg V. Neurobiological foundations of neurologic music therapy: rhythmic entrainment and the motor system. Front Psychol (2014) 5 1185. DOI: 10.3389/fpsyg.2014.01185 41. Cohen NR, Sekuler R. Chunking and compound cueing of movement sequences: learning, retention, and transfer. Percept Mot Skills (2010) 110 736-750. DOI: 10.2466/pms.110.3.736-750 42. Ellis KR, Hecht HK, Young Tiffany L, Oh Seyoung, Thomas Shikira, Hoggard Lori S, Ali Zaire, Olawale Ronke, Carthron Dana, Corbie-Smith Giselle, Eng Eugenia. Chronic disease among African American fFamilies: a systematic scoping review. Prev Chronic Dis (2020) 17 E167-8. DOI: 10.5888/pcd17.190431 43. Thorpe RJ, Wynn AJ, Walker Janiece L, Smolen Jenny R, Cary Michael P, Szanton Sarah L, Whitfield Keith E. Relationship between chronic conditions and disability in African American men and women. J Natl Med Assoc (2016) 108 90-8. DOI: 10.1016/j.jnma.2015.12.012 44. Jones DL, Starcher RW, Eicher JL, Wilcox S. Adoption of a Tai Chi intervention, Tai Ji Quan: moving for better balance, for fall prevention by rural faith-based organizations, 2013-2014. Prev Chronic Dis (2016) 13 E92. DOI: 10.5888/pcd13.160083
--- title: 'RNA sequencing-based transcriptome analysis of granulosa cells from follicular fluid: Genes involved in embryo quality during in vitro fertilization and embryo transfer' authors: - Eun Jeong Yu - Won Yun Choi - Mi Seon Park - Jin Hee Eum - Dong Ryul Lee - Woo Sik Lee - Sang Woo Lyu - Sook Young Yoon journal: PLOS ONE year: 2023 pmcid: PMC9977003 doi: 10.1371/journal.pone.0280495 license: CC BY 4.0 --- # RNA sequencing-based transcriptome analysis of granulosa cells from follicular fluid: Genes involved in embryo quality during in vitro fertilization and embryo transfer ## Abstract ### Background Granulosa cells play an important role in folliculogenesis, however, the role of RNA transcripts of granulosa cells in assessing embryo quality remains unclear. Therefore, we aims to investigate that RNA transcripts of granulosa cells be used to assess the probability of the embryonic developmental capacity. ### Methods This prospective cohort study was attempted to figure out the probability of the embryonic developmental capacity using RNA sequencing of granulosa cells. Granulosa cells were collected from 48 samples in good-quality embryo group and 79 in only poor- quality embryo group from women undergoing in vitro fertilization and embryo transfer treatment. Three samples from each group were used for RNA sequencing. ### Results 226 differentially expressed genes (DEGs) were related to high developmental competence of embryos. Gene Ontology enrichment analysis indicated that these DEGs were primarily involved in biological processes, molecular functions, and cellular components. Additionally, pathway analysis revealed that these DEGs were enriched in 13 Kyoto Encyclopedia of Genes and Genomes pathways. Reverse transcription quantitative polymerase chain reaction verified the differential expression of the 13 selected DEGs. Among them,10 genes were differently expressed in the poor-quality embryo group compared to good-quality embryo group, including CSF1R, CTSH, SERPINA1, CYP27A1, ITGB2, IL1β, TNF, TAB1, BCL2A1, and CCL4. ### Conclusions RNA sequencing data provide the support or confute granulosa expressed genes as non-invasive biomarkers for identifying the embryonic developmental capacity. ## Introduction The ability to identify oocyte quality remains one of the most significant challenges in assisted reproductive technology. The quality of oocyte is an important predictor of implantation and live birth [1, 2]. The presence of a mature and high-quality oocyte plays an essential role in the development of a high-quality embryo [3]. This means that the selection of high-quality embryos begins at the time of oocyte selection. Embryo quality is a strong indicator for the success rate of in vitro fertilization (IVF) program, as the live birth rate increases when good quality embryos are transferred [4]. The incidence of only poor-quality embryos is usual in IVF cycles of patients with advanced maternal age and in low responding patients. According to Semondade et al. [ 5], the incidence rate of only poor embryos is approximately $10\%$ at the first IVF cycle, and the recurrence rate is $3\%$. Morphological features and development rates are key indicators for oocyte and embryo selection during IVF. These methods are familiar for both clinician and embryologist and are used as the standard method in embryos selection step. However, the usefulness of their assessment is being questioned because of personal bias from the embryologist [6]. In recent years, techniques for embryo selection that provide chromosomal analysis to improve clinical pregnancy have recently been developed, such as preimplantation genetic testing (PGT) [7]. Although PGT is a strong predictor for implantation, it is expensive and requires invasive embryo biopsy, which involves technical expertise [8]. Therefore, there is a high demand for a non-invasive and easy-to-perform screening tools to improve the selection of the most pregnancy-competent embryo. Good-quality and mature oocyte in IVF are important for fertilization and embryo development [9]. Oocyte maturation occurred during folliculogenesis through orchestrated cross-talk between the oocyte and granulosa cells (GCs). Therefore, it is advocated that the functions of GCs indirectly reflect oocyte developmental competence. GCs can be easily recovered in large quantities during oocyte collection. Thus, the gene expression analysis in GCs could provide a non-invasive assessment for identifying the most competent oocytes and embryos. However, the transcriptomic analysis of GC used to identify embryo quality is still controversial. While some studies have identified candidate genes expressed in GC that could be expected as biomarkers of oocyte and embryo quality [10, 11] and successful clinical outcomes [12–14], others have reported that there are no significant differences in gene expression between embryos that did or did not successfully implant [15, 16]. Therefore, the present study aims to investigate and compare the GC transcriptomic obtained from subjects producing good-quality embryos and those producing only poor quality embryos in human IVF. We carried out RNA sequencing (RNA-seq) of GCs isolated from follicular fluid to identify novel gene transcription factors correlated with embryo quality among the embryos with good quality. ## Study population This prospective cohort study was conducted between January 2019 and February 2021 at the CHA Fertility Center Gangnam. The study was approved by the CHA University Gangnam CHA hospital institutional review board (GCI-19-10, May 15, 2019), Republic of Korea. All women gave written informed consent to provide material for this study. All procedures followed the rules for studies with human-origin materials established by the IRB. Of 141 embryos transfer (ETs), 48 were good quality (GQ) and 79 were poor quality (PQ) ETs; 14 cases of GQ and PQ ETs cases with endometriosis, endometrial pathologies, uterine fibroids or hydrosalpinx were excluded. Couples with severe male factor infertility as defined based on severe oligozoospermia (<5 million sperm/mL) or a history of testicular biopsy were excluded. There were 141 women undergoing IVF and embryo transfers (ETs), which included 48 women who had good-quality (GQ) ETs, 79 with only poor-quality (PQ) ETs, and 14 with either GQ or PQ that were excluded from this study (Fig 1). Most patients ($$n = 83$$) underwent double ETs, 59 samples came from a single ET, and 3 from triple ETs. ETs for 91 samples occurred on day 3 or 4, and for 50 on day 5, no significant difference between the GQ and PQ groups according to the day of ET. Fig 1: Study flowchart. Table 1 presented the characteristics of patients between the GQ and the PQ groups. There were no significant differences in age, body mass index, infertility duration, and anti-Mullerian hormone level. However, there were more ETs in the PQ group than in the GQ group. The clinical pregnancy rate was higher in the GQ ($52.0\%$) than in the PQ embryo transfer group ($40.5\%$); however, the difference was statistically insignificant ($$p \leq 0.09$$), and the ongoing pregnancy rate of the GQ group was significantly higher than that of the PQ group ($45.8\%$ vs. $24\%$, respectively, $$p \leq 0.02$$). Table 1 \| Unnamed: 0 \| Good-quality(GQ) embryo (N = 48) \| Poor-quality(PQ) embryo (N = 79) \| P-value \| \| --- \| --- \| --- \| --- \| \| Age (yr) \| 35.3 ± 2.9 \| 36.3 ± 3.7 \| 0.11 \| \| BMI (kg/m2) \| 20.3 ± 4.2 \| 21.5 ± 3.2 \| 0.48 \| \| Infertility duration (yr) \| 3.6 ± 2.3 \| 3.4 ± 2.5 \| 0.54 \| \| Previous IVF attempts (n) \| 1.4 ± 0.8 \| 2.2 ± 1.8 \| 0.001 \| \| AMH (ng/mL) \| 3.2 ± 2.7 \| 2.7 ± 2.4 \| 0.26 \| \| Basal E2 (pg/mL) \| 40.2 ± 17.8 \| 54.3 ± 41.1 \| 0.02 \| \| Basal FSH (mIU/mL) \| 7.4 ± 2.7 \| 7.9 ± 2.6 \| 0.17 \| \| Days of stimulation \| 12.7 ± 1.3 \| 12.8 ± 1.3 \| 0.51 \| \| E2 on hCG day (pg/mL) \| 2376.3 ± 1301.1 \| 1872.2 ± 926.0 \| 0.01 \| \| LH on hCG day (mIU/mL) \| 4.4 ± 3.4 \| 4.1 ± 2.6 \| 0.57 \| \| P4 on hCG day (ng/mL) \| 0.6 ± 0.4 \| 0.5 ± 0.4 \| 0.24 \| \| Number of retrieved oocytes \| 13.0 ± 6.9 \| 10.7 ± 4.9 \| 0.04 \| \| Number of transferred embryos \| 1.5 ± 0.7 \| 1.9 ± 0.4 \| 0.01 \| \| Clinical pregnancy rate \| 54.1% (26/48) \| 31.6% (25/79) \| 0.02 \| \| Ongoing pregnancy rate \| 50% (24/48) \| 26.5% (21/79) \| 0.04 \| ## Ovarian stimulation and embryo transfer procedures Patients underwent controlled ovarian stimulation using either the midluteal long gonadotropin-releasing hormone (GnRH) agonist protocol or GnRH antagonist protocol. Gonadotropin doses were individualized according to patients’ age, anti-Mullerian hormone level, antral follicle count, and previous response to stimulation. Cycle monitoring with transvaginal ultrasonography and serum estradiol measurement, was continued until hCG administration. When at least two follicles reached 18 mm in diameter, 250–500 μg of recombinant hCG was administered for final oocyte maturation. Transvaginal ultrasound-guided oocyte-retrieval under conscious sedation was carried out 34–36 hours after hCG administration. The mature oocytes were inseminated using intracytoplasmic sperm insemination (ICSI). One or two embryos were transferred to each patient, and ETs occurred on day 3, 4, or 5. All patients received luteal phase support with progesterone after ET until 8–10 weeks after pregnancy. ## Isolation of GCs and assessment of embryo quality Follicular fluids were aspirated and pooled for each patient during oocyte-retrieval. Follicle aspirates, which were not clear and were contaminated with endometriosis cysts, were discarded. Granulosa cells recovered from one woman were used as one sample. The 127 samples were divided into two groups according to the patient’s embryo quality; 48 patients with good quality blastocysts having a grade of at least 3BB and 97 patients with only poor quality blastocysts. The follicular fluid was centrifuged at 1000 x g for 20 min to separate erythrocytes, leukocytes and GCs. The cell pellets were washed in RBC lysis buffer (Roche Diagnostics, Basel, Switzerland) for RBC removal and centrifuged at 500g for 10 min and separated enzymatically with instigation at 37°C for 30 min in the enzyme solution [Hanks’ balanced salt solution (HBSS, Gibco, Grand Island, NY) containing 0.5 mg/ml collagenase (type IV; Gibco) and 0.25mg/ml dispase II (neutral protease, grade III, Roche)]. The suspension of GCs was washed and filtered through a 40μm mesh (BD, Franklin Lakes, NJ). GCs using TRIzol Reagent (Life Technologies, San Diego, CA, USA) were stored -75°C until analysis. Embryo grading was evaluated by two embryologists and was assessed by another senior embryologist with over at least ten years of work experience before embryo transfer. For standardization, all embryologists at our center were trained in embryo grading by the same laboratory director. Embryo quality assessment was carried out according to previously described protocol [17]. Briefly, good quality embryos were identified using the following characteristics. Cleavage embryos were classified as GQ according to Cummins criteria [18, 19]. Blastocyst quality was assessed regarding to the degree of blastocoel expansion, inner cell mass (ICM) and trophectoderm (TE) morphology on day 5 or 6 [20]. ## RNA isolation and NGS library preparation Total RNA was isolated from cells using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. RNA integrity was checked by using an Agilent 2100 BioAnalyzer (Agilent, CA, USA) with an RNA integrity number value greater than 6. Only qualified samples were used for RNA library constitution. The libraries were arranged for 151 bp paired-end sequencing by the TruSeq stranded mRNA sample preparation kit (Illumina, CA, USA). After the sequential process of end repair, A-tailing, and adapter ligation, cDNA libraries were amplified by Polymerase Chain Reaction (PCR). They were quantified using the KAPA library quantification kit (Kapa Biosystems, MA, USA) following the manufacturer’s quantification protocol. After cluster amplification of denatured templates, sequencing was carried out as paired-end (2×151bp) using IlluminaNovaSeq6000 (Illumina, CA, USA). ## DEG analysis Based on the estimated read counts in the previous step, DEGs were screened using the R package TCCv.1.26.0 [21]. The TCC package uses well-set normalization strategies to compare the tagcount data. Normalization factors were estimated by the iterative DESeq2 [22] /edgeR method [23]. The Q-value was assessed based on the p-value using the p.adjust function of the R package by default parameter programs. The DEGs were evaluated with a Q- value threshold of < 0.05 for fixing errors due to multiple testing. ## GO analysis The GO database indicates a set of hierarchically controlled vocabulary classified into three categories: biological process, cellular component, and molecular function. For functional characterization of the DEGs, GO-based trend test was assessed by the R package called GOseq [24] through the Wallenius non-central hypergeometric distribution. Selected genes with P-values < 0.05 following the test were regarded as statistically significant. ## RT-qPCR Approximately 1 μg RNA from each sample was reverse transcribed to cDNA using a SensiFAST™ cDNA Synthesis Kit (Meridian Bioscience, Tennessee, USA) following the manufacturer’s instructions. Thirteen genes were selected for validation from the list of DEGs. Primers were designed using the Primer3 primer-design program (version 0.4.0; http://bioinfo.ut.ee/primer3-0.4.0/) (S1 Table). The qPCR reactions were performed on the CFX Connect Real-Time PCR detection System (Bio-Rad Laboratories, Inc., USA) using iQ™ SYBR® Green Supermix (Bio-Rad Laboratories, Inc., USA). Relative quantification analyses were carried out using the comparative CT method, and relative gene expression levels were calculated using the 2−ΔΔCT method. ## Clinical outcome and statistics The primary outcomes of the study were clinical pregnancy outcomes, including clinical pregnancy rate, and ongoing pregnancy rate. Clinical pregnancy was defined as the presence of at least one gestational sac with a fetal heartbeat on ultrasonography. Ongoing pregnancy was defined when a positive heartbeat was at 12 weeks or more of gestation on ultrasonography. All statistical analyses were conducted using SPSS (version 25.0; Chicago, IL, USA) software. Categorical variables were performed using the chi‐square and Fisher’s exact tests. Continuous variables were assessed using Student’s t test. A probability (p) value <0.05 was considered to determine statistical significance. A p value below 0.05 was considered statistically significant. ## Identification of differentially expressed genes (DEGs): samples clustered according to embryo quality Three GCs samples from each group were selected and used for RNA-seq analysis. Six hundred and forty genes were differentially expressed in our cohort. We analyzed DEGs between the two groups. Compared with that in the GQ embryo group, 18 genes were upregulated, whereas 208 genes were downregulated in the PQ embryo group ($p \leq 0.05$; FC of log2-transformed fragments per kilobase of transcript per million (FPKM) > 1) (Fig 2A). As shown in Fig 2B, we developed unsupervised hierarchical clustering of DEGs. A detailed list of 226 DEGs that were upregulated and downregulated in each group is presented in S2 Table. Fig 2: Hierarchical clustering and Differential Expression (DE) analysis of granulosa cells from good-quality (GQ) embryo transfer group compared with granulosa cells from poor-quality (PQ) embryo transfer group.(A) A Venn diagram depicting the distribution of significant differentially expressed genes (DEGs) (fold change > 2.5, q <0.05) in our study. Numbers represent the number of different genes showing upregulated or downregulated expression in PQ embryo samples compared with GQ embryo samples with red representing upregulated expression and blue indicating downregulated expression. Numbers in parenthesis represent the number of DEGs in only PQ or GQ group. (B) Heatmap of the DEGs between the GQ and PQ embryo groups. The color key from blue to red represents the relative gene expression level from low to high, respectively. (C) Volcano plot showing DEG analysis between GQ embryo and PQ embryo groups using DESeq2; 640 genes were differentially expressed (210 downregulated (FC < -2 and FDR < 0.05) and 25 upregulated (FC > 2 and FDR < 0.05). The upregulated genes are represented using red dots; downregulated genes are denoted using blue dots; and the black dots indicate genes with no significant changes. ## Gene ontology (GO) enrichment analysis: Differential expression revealed significant differences in gene expression regarding to the degree of embryo quality To further extend the molecular properties of the 226 DEGs, we performed a GO analysis of the up- or down-regulated groups (Fig 3). The DEGs were divided into three categories: biological processes, molecular functions, and cellular components. In the GO category biological process, DEGs were enriched in the response to stimulus, metabolic process, biological regulation, immune system process, binding and metabolic process, cellular process, single-organism process, cell and cell part, developmental process, cellular component organization or biogenesis, and the reproductive process. The top 20 genes displayed in the different embryo qualities are presented in Tables 2 and 3. KEGG pathway enrichment analysis showed that the DEGs in our study participated in 13 pathways (Fig 4). Several of these are related to reproduction, such as the chemokine signaling pathway, phagosome, cytokine-cytokine receptor interaction, cell adhesion molecules, and NF-kB signaling pathway. Fig 3: Histogram of the Gene Ontology(GO) analysis of the differentially expressed genes (DEGs).Genes were classified into three GO domains: biological process, cellular component, and molecular function. The left y-axis shows the gene count in each category. The solid columns indicate DEGs. Fig 4: Validation of RNAseq results using RT-qPCR of 13 targets normalized to GAPDH in duplicate.Fold change was calculated using the 2ΔΔCt method between the good-quality (GQ) and poor-quality (PQ) embryos. $p \leq 0.05.$ (A) Six genes associated with embryo quality both in our and previous studies. (B) Seven genes randomly selected from the NF-kB pathway.* TABLE_PLACEHOLDER:Table 2 TABLE_PLACEHOLDER:Table 3 ## Our findings enhanced the available literature exploring processes associated with embryo quality When comparing our DEGs to those previously reported in the literature, Table 4 summarizes the findings of the detailed list of 68 genes that show genes positively associated with embryo quality. The GC gene biomarker studies have varied depending on the endpoints chosen, including ovarian development, follicular development, folliculogenesis, oocyte and embryo quality. Table 4 \| Subject outcomes \| Known Function \| Sample \| Method of detection \| Gene \| Previous study \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Ovarian development \| Signal transduction \| Pooled GC \| Microarray \| ACP5, ADCY7, AKAP9, ALPL, ARL5B, ATP2A1, COL16A1, DUSP2, GPR65, GPRC5C, HK3, ITGAL, KRT5, MAPK15, PIK3CG, PPF1A4, PRKCB, PTPN6, PTPRC, RAB7B, RGS19, TNFAIP2, TNFRSF10C, TYROBP \| [25–28] \| \| Oocyte and embryo competence \| Extra-cellular matrix proteases \| Pooled GC \| RNAseq, RT-qPCR \| ADAM8, ARHGDIB, ARHGEF33, CCL22, CD14, CD44, CYBA, CYP1B1,DHRS3, DOCK7, GPR137C, HIST1H4I, HLA-DPA1, IFI30, IL1B, LAPTM5, MMP9, PLA2G2D, PLD4, PLEK, PPBP, S100A8, TLR8, TMEM41B, TREM1, TRIM29, UCP2, \| [29–31] \| \| Oocyte quality \| Cytokine-cytokine receptor interaction, Signal transduction, Cytoskeleton organization, Chemokine signaling pathway \| Pooled GC \| Microarray, RT-qPCR \| AQP8, CXCL3, CXCR2, CYP27A1, ITBG2, OSM, PDE10A, TNFSF13, VAV1, \| [26, 32] \| \| Embryo development \| Apoptosis regulator \| Embryo \| RNAseq \| BCL2A1, SERPINA1, \| [33] \| \| Folliculogenesis \| Cytokine-cytokine interaction \| Pooled GC \| RNAseq \| CCL4, TNF, \| [34] \| \| Follicle maturation \| Extra-cellular matrix molecules \| Pooled GC \| RT-qPCR \| CDH13, DHCR7 \| [35, 36] \| \| Follicular survival \| Cellular protease \| Pooled GC \| RT-qPCR \| CTSH \| [37] \| \| Follicular development \| Growth factor \| Individual GC \| RT-qPCR \| CSF1R \| [38] \| ## Validation of DEGs using RT-qPCR analysis To examine the reliability of the RNA-seq data, we selected 13 genes to verify their expression in GC using RT-qPCR (Fig 4). When comparing our DEGs to those reported in previous studies, we found six genes that were related to embryo quality both in our study and previous published literature, including colony stimulating factor receptor 1 (CSFR1) [38], cathepsin H (CTSH) [37], cadherin 13 (CDH13) [35], serpin family A member 1 (SERPINA1) [39], cytochrome P450 family 27 subfamily A member 1 (CYP27A1) [40], and integrin subunit beta 2 (ITGB2) [16]. Among the significantly enriched pathways, we randomly selected seven genes from the NF-kB pathway: four genes that have refuting findings between different studies (interleukin 1 beta (IL1β), tumor necrosis factor (TNF), BCL2 related protein A1 (BCL2A1), C-C motif chemokine ligand 4 (CCL4)), and three genes that did not support an association with pregnancy or embryo quality (MYD88 innate immune signal transduction adaptor(MYD88), mitogen-activated protein kinase kinase kinase 7(MAP3K7), and TGF-beta activated kinase 1 binding protein 1(TAB1)). Thirteen candidate genes, comprising eleven downregulated genes in GCs from only PQ embryo groups, CSF1R, CTSH, IL1β, TNF, BCL2A1, CCL4, SERPINA1, CYP27A1, ITGB2, MYD88, MAP3K7, and two upregulated genes in GC from PQ embryo groups, TAB1, and CDH13, were selected and analyzed using RT-qPCR. The RT-qPCR results were consistent with the RNA-seq data, which means that the RNA-seq results were dependable and that it could be used to perform accurate differential expression analysis of mRNA. ## Discussion In the present study, we analyzed GC expression of CSF1R, CTSH, IL1β, TNF, BCL2A1, CCL4, SERPINA1, CYP27A1, ITGB2 from GCs collected during IVF from oocytes that developed into GQ embryos. Selected genes were shown to be well differentiated between immature MI and mature MII oocytes. In recent years, many studies have been performed to analyze GC expression in association with various endpoints: oocyte quality, embryo development and pregnancy (Table 4). Since GC is an easily accessible material that is normally discarded during the IVF cycle, it represents a good biological material for research and diagnostic purposes. Several signaling pathways are important in mediating or modifying the proliferative response of GCs to FSH stimulation, the primary mediator of GC proliferation during folliculogenesis [41]. Our RNA-Seq results showed that some differential pathways might verify oocyte development, such as the chemokine signaling pathway, cytokine-cytokine receptor interaction, phagosome, cell adhesion molecules, and NF-kB signaling pathway. Thus, we could infer that the PQ embryos were mainly related to the expression levels of cytokine reaction, apoptosis, and the adherent junction pathway. In our study, the differential expression of transcripts in the NF-kB signaling pathway was related to embryo quality (Fig 4B). This pathway participate in apoptosis and cell growth, as well as in immune, inflammatory and acute phase responses [42]. The downregulation of immune and inflammation-related genes in the PQ embryo group was in agreement with the hypothesis that ovulatory process is an acute inflammatory reaction [43]. Consistent with previous studies [44, 45], various ovulation-associated factors such as PDE2A (3′5’-cyclic nucleotide phosphodiesterase 2A), RGS1 and RGS16 (a regulator of G-protein signaling 1 and 16), ADAMTS1 (a disintegrin-like and metalloprotease with thrombospondin type 1 motif, 1), and PTGS1 (prostaglandin-endoperoxide synthetase 1), were all upregulated in GCs from the PQ embryo group. CSF-1R is activated in a similar way by homodimeric growth factors colony-stimulating factor-1 (CSF-1) and interleukin (IL)-34 [46]. CSF-1 may be essential in regulating the response of GCs to gonadotropin and may promote in the early embryonic development [38]. The biological activity of CSF-1 depends on its binding to CSF-1R in target cells [47]. CTSH expression is increased in human GCs during follicular maturation in vivo [37]. Our study demonstrated that the expression of CTSH was higher in GCs from GQ embryos than that from PQ embryos with statistical significance, suggesting embryo development failure was caused by poor follicular maturation and GC dysfunction. SERPINA1 is the gene for an acute-phase protein that increases in response to inflammation [48]. This gene, implicated in the inflammatory response, reduces the levels of proinflammatory mediators, and increases anti-inflammatory cytokines [49]. In our study, increased SERPINA1 expression levels were observed in GQ embryo. Overexpression of SERPINA1 in GCs from the GQ embryo group indicates decreased proinflammatory cytokine levels, which leads to increased developmental competence. This finding suggests that an impaired follicular fluid microenvironment characterized by elevated pro-inflammatory cytokines may cause poor quality of oocyte. Changes in the cellular structure of the corpus luteum (CL) on natural luteolysis may contribute to elevated concentrations of CYP27A1 mRNA. For example, macrophages invade the CL during luteolysis and express CYP27A1 [50]. In addition, increased CYP27A1 protein concentrations in mitochondria cause elevated conversion of cholesterol into 27OH, which decreases progesterone secretion in human luteinized GC [40]. ITGB2 is an integrin subunit that is a pivotal mediator of uterine receptivity and embryonic development [51]. Overexpression of this gene is associated with GCs of healthy follicles and decreased with atresia [52]. Reorganization of cellular composition for ovulation includes follicular wall degradation, and oocyte expulsion during the advanced stages of follicular development [53]. IL1β, TNF-α, and CCL4 are pro-inflammatory cytokines that act locally on ovarian follicular cells and are involved in the ovulation process [54]. During the periovulatory phase, ovarian macrophages collect in the ovary and secret these pro-inflammatory cytokines [55–57]. They regulate the secretion of steroid hormones for follicle growth and are important for ovulation process and the development and regression of the CL [58, 59]. For instance, IL-1β has been shown to induce ovulation by promoting follicular rupture [60]. In the present study, IL1β, TNF-α, and CCL4 were significantly upregulated in GCs from oocytes that yielded GQ embryos. This finding suggests that these cytokines may modulate oocyte quality and embryo development. TAB1 may represents the initiation of inflammatory functions and may serve as a connector of the IL-1 induced signaling pathway leading to activation of c-Jun N-terminal kinase (JNK) and NF-κB signaling pathways [61, 62]. B-cell lymphoma-2 (BCL2) family proteins are pivotal regulators of apoptosis [63]. The ovaries of Bcl2 knockout mice were revealed to have fewer primordial follicles [64]. However, this loss may be due to GC apoptosis, which indirectly affects follicle development. BCL2 family members, such as BCL2A1, are expressed during embryonic genome activation and maintained until the blastocyst stage [33], suggesting that these are required throughout early embryonic development. The strength of our study is that it was prospective and sampled a large group of patients who were followed up clinically. We also performed sorting and enrichment analysis of RNA-sequencing data, the most unbiased approach that can be used to explore transcriptomic signatures [65]. Furthermore, by incorporating information from DEGs, pathway analysis, and related studies, the present study provides a list of candidate genes with functions and expression levels closely related to embryo quality in IVF. In this study, RNA-sequencing data of whole GCs collection revealed a difference between the group that produced good-quality embryos and the group that produced only poor-quality embryos. This difference could be used for therapeutic purposes to improve embryo quality in the future. The present study, however, has some limitations. First, GCs obtained from the follicular fluid were GCs in all follicles and cannot accurately reflect the quality of single oocyte. In clinical practice, it may be more difficult to collect GCs from a single follicle than cumulus cells. Second, we reported expression levels of the genes, but did not analyze expression ot corresponding proteins and the role of post-translational modifications. ## Conclusions We found that the expression profiles of specific genes in GCs was associated with embryo quality during IVF and suggested the need to develop treatment strategies that can compensate for the poor quality due to the deficiency of specific genes during follicular development. Furthermore, we reported the published data that support or confute granulosa expressed genes as biomarkers for identifying oocyte or embryo quality. This study improves our understanding of reproductive function of GCs, which could be helpful for more targeted studies aiming to improve oocyte and embryo competence in the future. ## References 1. Ahlstrom A, Westin C, Reismer E, Wikland M, Hardarson T. Trophectoderm morphology: an important parameter for predicting live birth after single blastocyst transfer. Hum Reprod (2011) 26 3289-96. DOI: 10.1093/humrep/der325 2. della Ragione T, Verheyen G, Papanikolaou EG, Van Landuyt L, Devroey P, Van Steirteghem A. Developmental stage on day-5 and fragmentation rate on day-3 can influence the implantation potential of top-quality blastocysts in IVF cycles with single embryo transfer. Reprod Biol Endocrinol (2007) 5. DOI: 10.1186/1477-7827-5-2 3. Sirard MA, Richard F, Blondin P, Robert C. Contribution of the oocyte to embryo quality. Theriogenology (2006) 65 126-36. DOI: 10.1016/j.theriogenology.2005.09.020 4. Ziebe S, Petersen K, Lindenberg S, Andersen AG, Gabrielsen A, Andersen AN. Embryo morphology or cleavage stage: how to select the best embryos for transfer after in-vitro fertilization. Hum Reprod (1997) 12 1545-9. DOI: 10.1093/humrep/12.7.1545 5. Sermondade N, Delarouziere V, Ravel C, Berthaut I, Verstraete L, Mathieu E. Characterization of a recurrent poor-quality embryo morphology phenotype and zygote transfer as a rescue strategy. Reprod Biomed Online (2012) 24 403-9. DOI: 10.1016/j.rbmo.2012.01.004 6. Alfarawati S, Fragouli E, Colls P, Stevens J, Gutierrez-Mateo C, Schoolcraft WB. The relationship between blastocyst morphology, chromosomal abnormality, and embryo gender. Fertil Steril (2011) 95 520-4. DOI: 10.1016/j.fertnstert.2010.04.003 7. Sigalos G, Triantafyllidou O, Vlahos NF. Novel embryo selection techniques to increase embryo implantation in IVF attempts. Arch Gynecol Obstet (2016) 294 1117-24. DOI: 10.1007/s00404-016-4196-5 8. Cimadomo D, Capalbo A, Ubaldi FM, Scarica C, Palagiano A, Canipari R. The Impact of Biopsy on Human Embryo Developmental Potential during Preimplantation Genetic Diagnosis. Biomed Res Int (2016) 2016. DOI: 10.1155/2016/7193075 9. Gilchrist RB, Lane M, Thompson JG. Oocyte-secreted factors: regulators of cumulus cell function and oocyte quality. Hum Reprod Update (2008) 14 159-77. DOI: 10.1093/humupd/dmm040 10. McKenzie LJ, Pangas SA, Carson SA, Kovanci E, Cisneros P, Buster JE. Human cumulus granulosa cell gene expression: a predictor of fertilization and embryo selection in women undergoing IVF. Hum Reprod (2004) 19 2869-74. DOI: 10.1093/humrep/deh535 11. van Montfoort AP, Geraedts JP, Dumoulin JC, Stassen AP, Evers JL, Ayoubi TA. Differential gene expression in cumulus cells as a prognostic indicator of embryo viability: a microarray analysis. Mol Hum Reprod (2008) 14 157-68. DOI: 10.1093/molehr/gam088 12. Hamel M, Dufort I, Robert C, Gravel C, Leveille MC, Leader A. Identification of differentially expressed markers in human follicular cells associated with competent oocytes. Hum Reprod (2008) 23 1118-27. DOI: 10.1093/humrep/den048 13. Hamel M, Dufort I, Robert C, Leveille MC, Leader A, Sirard MA. Genomic assessment of follicular marker genes as pregnancy predictors for human IVF. Mol Hum Reprod (2010) 16 87-96. DOI: 10.1093/molehr/gap079 14. Hamel M, Dufort I, Robert C, Leveille MC, Leader A, Sirard MA. Identification of follicular marker genes as pregnancy predictors for human IVF: new evidence for the involvement of luteinization process. Mol Hum Reprod (2010) 16 548-56. DOI: 10.1093/molehr/gaq051 15. Burnik Papler T, Vrtacnik Bokal E, Lovrecic L, Kopitar AN, Maver A. No specific gene expression signature in human granulosa and cumulus cells for prediction of oocyte fertilisation and embryo implantation. PLoS One (2015) 10 e0115865. DOI: 10.1371/journal.pone.0115865 16. Devjak R, Fon Tacer K, Juvan P, Virant Klun I, Rozman D, Vrtacnik Bokal E. Cumulus cells gene expression profiling in terms of oocyte maturity in controlled ovarian hyperstimulation using GnRH agonist or GnRH antagonist. PLoS One (2012) 7 e47106. DOI: 10.1371/journal.pone.0047106 17. Kim MK, Park JK, Jeon Y, Choe SA, Lee HJ, Kim J. Correlation between Morphologic Grading and Euploidy Rates of Blastocysts, and Clinical Outcomes in In Vitro Fertilization Preimplantation Genetic Screening.. J Korean Med Sci (2019) 34 e27. DOI: 10.3346/jkms.2019.34.e27 18. Cummins JM, Breen TM, Harrison KL, Shaw JM, Wilson LM, Hennessey JF. A formula for scoring human embryo growth rates in in vitro fertilization: its value in predicting pregnancy and in comparison with visual estimates of embryo quality.. J In Vitro Fert Embryo Transf (1986) 3 284-95. DOI: 10.1007/BF01133388 19. Reinblatt SL, Ishai L, Shehata F, Son WY, Tulandi T, Almog B. Effects of ovarian endometrioma on embryo quality. Fertil Steril (2011) 95 2700-2. DOI: 10.1016/j.fertnstert.2011.03.002 20. Gardner DK, Surrey E, Minjarez D, Leitz A, Stevens J, Schoolcraft WB. Single blastocyst transfer: a prospective randomized trial. Fertil Steril (2004) 81 551-5. DOI: 10.1016/j.fertnstert.2003.07.023 21. Sun J, Nishiyama T, Shimizu K, Kadota K. TCC: an R package for comparing tag count data with robust normalization strategies.. BMC Bioinformatics (2013) 14. DOI: 10.1186/1471-2105-14-219 22. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol (2014) 15 550. DOI: 10.1186/s13059-014-0550-8 23. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics (2010) 26 139-40. DOI: 10.1093/bioinformatics/btp616 24. Young MD, Wakefield MJ, Smyth GK, Oshlack A. Gene ontology analysis for RNA-seq: accounting for selection bias. Genome Biol (2010) 11 R14. DOI: 10.1186/gb-2010-11-2-r14 25. Sasson R, Rimon E, Dantes A, Cohen T, Shinder V, Land-Bracha A. Gonadotrophin-induced gene regulation in human granulosa cells obtained from IVF patients. Modulation of steroidogenic genes, cytoskeletal genes and genes coding for apoptotic signalling and protein kinases. Mol Hum Reprod (2004) 10 299-311. DOI: 10.1093/molehr/gah041 26. Liu C, Su K, Chen L, Zhao Z, Wang X, Yuan C. Prediction of oocyte quality using mRNA transcripts screened by RNA sequencing of human granulosa cells. Reprod Biomed Online (2021) 43 413-20. DOI: 10.1016/j.rbmo.2021.05.018 27. Conti M.. Specificity of the cyclic adenosine 3’,5’-monophosphate signal in granulosa cell function. Biol Reprod (2002) 67 1653-61. DOI: 10.1095/biolreprod.102.004952 28. Seger R, Hanoch T, Rosenberg R, Dantes A, Merz WE, Strauss JF. The ERK signaling cascade inhibits gonadotropin-stimulated steroidogenesis. J Biol Chem (2001) 276 13957-64. DOI: 10.1074/jbc.M006852200 29. Koks S, Velthut A, Sarapik A, Altmae S, Reinmaa E, Schalkwyk LC. The differential transcriptome and ontology profiles of floating and cumulus granulosa cells in stimulated human antral follicles. Mol Hum Reprod (2010) 16 229-40. DOI: 10.1093/molehr/gap103 30. Duffy DM, Seachord CL, Dozier BL. An ovulatory gonadotropin stimulus increases cytosolic phospholipase A2 expression and activity in granulosa cells of primate periovulatory follicles. J Clin Endocrinol Metab (2005) 90 5858-65. DOI: 10.1210/jc.2005-0980 31. Kawano Y, Fukuda J, Nasu K, Nishida M, Narahara H, Miyakawa I. Production of macrophage inflammatory protein-3alpha in human follicular fluid and cultured granulosa cells. Fertil Steril (2004) 82 1206-11. DOI: 10.1016/j.fertnstert.2004.05.070 32. Lu CL, Yan ZQ, Song XL, Xu YY, Zheng XY, Li R. Effect of exogenous gonadotropin on the transcriptome of human granulosa cells and follicular fluid hormone profiles. Reprod Biol Endocrinol (2019) 17 49. DOI: 10.1186/s12958-019-0489-4 33. Metcalfe AD, Hunter HR, Bloor DJ, Lieberman BA, Picton HM, Leese HJ. Expression of 11 members of the BCL-2 family of apoptosis regulatory molecules during human preimplantation embryo development and fragmentation. Mol Reprod Dev (2004) 68 35-50. DOI: 10.1002/mrd.20055 34. Yamamoto Y, Kuwahara A, Taniguchi Y, Yamasaki M, Tanaka Y, Mukai Y. Tumor necrosis factor alpha inhibits ovulation and induces granulosa cell death in rat ovaries. Reprod Med Biol (2015) 14 107-15. DOI: 10.1007/s12522-014-0201-5 35. Makrigiannakis A, Coukos G, Christofidou-Solomidou M, Gour BJ, Radice GL, Blaschuk O. N-cadherin-mediated human granulosa cell adhesion prevents apoptosis: a role in follicular atresia and luteolysis?. Am J Pathol (1999) 154 1391-406. DOI: 10.1016/S0002-9440(10)65393-X 36. Yerushalmi GM, Salmon-Divon M, Yung Y, Maman E, Kedem A, Ophir L. Characterization of the human cumulus cell transcriptome during final follicular maturation and ovulation. Mol Hum Reprod (2014) 20 719-35. DOI: 10.1093/molehr/gau031 37. Oksjoki S, Soderstrom M, Vuorio E, Anttila L. Differential expression patterns of cathepsins B, H, K, L and S in the mouse ovary. Mol Hum Reprod (2001) 7 27-34. DOI: 10.1093/molehr/7.1.27 38. Wei Z, Song X, Zhifen Z. Molecular mechanism and functional role of macrophage colonystimulating factor in follicular granulosa cells. Mol Med Rep (2017) 16 2875-80. DOI: 10.3892/mmr.2017.6857 39. Burnik Papler T, Vrtacnik Bokal E, Maver A, Kopitar AN, Lovrecic L. Transcriptomic Analysis and Meta-Analysis of Human Granulosa and Cumulus Cells.. PLoS One (2015) 10 e0136473. DOI: 10.1371/journal.pone.0136473 40. Xu Y, Hutchison SM, Hernandez-Ledezma JJ, Bogan RL. Increased 27-hydroxycholesterol production during luteolysis may mediate the progressive decline in progesterone secretion. Mol Hum Reprod (2018) 24 2-13. DOI: 10.1093/molehr/gax061 41. Richards JS. Hormonal control of gene expression in the ovary. Endocr Rev (1994) 15 725-51. DOI: 10.1210/edrv-15-6-725 42. Kucharczak J, Simmons MJ, Fan Y, Gelinas C. To be, or not to be: NF-kappaB is the answer—role of Rel/NF-kappaB in the regulation of apoptosis. Oncogene (2003) 22 8961-82. DOI: 10.1038/sj.onc.1207230 43. Duffy DM, Ko C, Jo M, Brannstrom M, Curry TE. Ovulation: Parallels With Inflammatory Processes.. Endocr Rev (2019) 40 369-416. DOI: 10.1210/er.2018-00075 44. Wyse BA, Fuchs Weizman N, Kadish S, Balakier H, Sangaralingam M, Librach CL. Transcriptomics of cumulus cells—a window into oocyte maturation in humans.. J Ovarian Res (2020) 13 93. DOI: 10.1186/s13048-020-00696-7 45. Wissing ML, Kristensen SG, Andersen CY, Mikkelsen AL, Host T, Borup R. Identification of new ovulation-related genes in humans by comparing the transcriptome of granulosa cells before and after ovulation triggering in the same controlled ovarian stimulation cycle. Hum Reprod (2014) 29 997-1010. DOI: 10.1093/humrep/deu008 46. Stanley ER, Chitu V. CSF-1 receptor signaling in myeloid cells. Cold Spring Harb Perspect Biol (2014) 6. DOI: 10.1101/cshperspect.a021857 47. Chitu V, Stanley ER. Regulation of Embryonic and Postnatal Development by the CSF-1 Receptor. Curr Top Dev Biol (2017) 123. DOI: 10.1016/bs.ctdb.2016.10.004 48. Badola S, Spurling H, Robison K, Fedyk ER, Silverman GA, Strayle J. Correlation of serpin-protease expression by comparative analysis of real-time PCR profiling data. Genomics (2006) 88 173-84. DOI: 10.1016/j.ygeno.2006.03.017 49. Boots CE, Jungheim ES. Inflammation and Human Ovarian Follicular Dynamics.. Semin Reprod Med (2015) 33 270-5. DOI: 10.1055/s-0035-1554928 50. Fu X, Menke JG, Chen Y, Zhou G, MacNaul KL, Wright SD. 27-hydroxycholesterol is an endogenous ligand for liver X receptor in cholesterol-loaded cells. J Biol Chem (2001) 276 38378-87. DOI: 10.1074/jbc.M105805200 51. Monniaux D, Huet-Calderwood C, Le Bellego F, Fabre S, Monget P, Calderwood DA. Integrins in the ovary.. Semin Reprod Med (2006) 24 251-61. DOI: 10.1055/s-2006-948554 52. Le Bellego F, Pisselet C, Huet C, Monget P, Monniaux D. Laminin-alpha6beta1 integrin interaction enhances survival and proliferation and modulates steroidogenesis of ovine granulosa cells. J Endocrinol (2002) 172 45-59. DOI: 10.1677/joe.0.1720045 53. Abedel-Majed MA, Romereim SM, Davis JS, Cupp AS. Perturbations in Lineage Specification of Granulosa and Theca Cells May Alter Corpus Luteum Formation and Function.. Front Endocrinol (Lausanne) (2019) 10. DOI: 10.3389/fendo.2019.00832 54. Sarapik A, Velthut A, Haller-Kikkatalo K, Faure GC, Bene MC, de Carvalho Bittencourt M. Follicular proinflammatory cytokines and chemokines as markers of IVF success. Clin Dev Immunol (2012) 2012. DOI: 10.1155/2012/606459 55. Cohen PE, Nishimura K, Zhu L, Pollard JW. Macrophages: important accessory cells for reproductive function. J Leukoc Biol (1999) 66 765-72. DOI: 10.1002/jlb.66.5.765 56. Wu R, Van der Hoek KH, Ryan NK, Norman RJ, Robker RL. Macrophage contributions to ovarian function. Hum Reprod Update (2004) 10 119-33. DOI: 10.1093/humupd/dmh011 57. Turner EC, Hughes J, Wilson H, Clay M, Mylonas KJ, Kipari T. Conditional ablation of macrophages disrupts ovarian vasculature. Reproduction (2011) 141 821-31. DOI: 10.1530/REP-10-0327 58. Jabbour HN, Sales KJ, Catalano RD, Norman JE. Inflammatory pathways in female reproductive health and disease. Reproduction (2009) 138 903-19. DOI: 10.1530/REP-09-0247 59. Greenfeld CR, Roby KF, Pepling ME, Babus JK, Terranova PF, Flaws JA. Tumor necrosis factor (TNF) receptor type 2 is an important mediator of TNF alpha function in the mouse ovary.. Biol Reprod (2007) 76 224-31. DOI: 10.1095/biolreprod.106.055509 60. Vassiliadis S, Relakis K, Papageorgiou A, Athanassakis I. Endometriosis and infertility: a multi-cytokine imbalance versus ovulation, fertilization and early embryo development. Clin Dev Immunol (2005) 12 125-9,. DOI: 10.1080/17402520500125484 61. Takaesu G, Kishida S, Hiyama A, Yamaguchi K, Shibuya H, Irie K. TAB2, a novel adaptor protein, mediates activation of TAK1 MAPKKK by linking TAK1 to TRAF6 in the IL-1 signal transduction pathway. Mol Cell (2000) 5 649-58. DOI: 10.1016/s1097-2765(00)80244-0 62. Owerbach D, Pina L, Gabbay KH. A 212-kb region on chromosome 6q25 containing the TAB2 gene is associated with susceptibility to type 1 diabetes. Diabetes (2004) 53 1890-3. DOI: 10.2337/diabetes.53.7.1890 63. Aouacheria A, Brunet F, Gouy M. Phylogenomics of life-or-death switches in multicellular animals: Bcl-2, BH3-Only, and BNip families of apoptotic regulators. Mol Biol Evol (2005) 22 2395-416. DOI: 10.1093/molbev/msi234 64. Ratts VS, Flaws JA, Kolp R, Sorenson CM, Tilly JL. Ablation of bcl-2 gene expression decreases the numbers of oocytes and primordial follicles established in the post-natal female mouse gonad. Endocrinology (1995) 136 3665-8. DOI: 10.1210/endo.136.8.7628407 65. Wang Z, Gerstein M, Snyder M. RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet (2009) 10 57-63. DOI: 10.1038/nrg2484
--- title: 'Predictors of death among severe COVID-19 patients admitted in Hawassa City, Sidama, Southern Ethiopia: Unmatched case-control study' authors: - Samuel Misganaw - Betelhem Eshetu - Adugnaw Adane - Tarekegn Solomon journal: PLOS ONE year: 2023 pmcid: PMC9977030 doi: 10.1371/journal.pone.0282478 license: CC BY 4.0 --- # Predictors of death among severe COVID-19 patients admitted in Hawassa City, Sidama, Southern Ethiopia: Unmatched case-control study ## Abstract ### Introduction Since COVID-19 was announced as a worldwide pandemic, the world has been struggling with this disease. In Ethiopia, there is some information on the epidemiological characteristics of the disease and treatment outcomes of COVID-19 patients. But, there is limited evidence related to predictors of death in COVID-19 patients. ### Objective To assess the predictor of death among severely ill COVID-19 patients admitted in Hawassa city COVID-19 treatment centers. ### Methods An institution-based unmatched case-control study was conducted at Hawassa city COVID-19 treatment centers from May 2021 to June 2021. All severe COVID-19-related deaths from May 2020 to May 2021 were included in the case group whereas randomly selected discharged severe COVID-19 patients were included in the control group. Extracted information was entered into Epi-data 4.6 and exported to SPSS 25 for analysis. Multivariable binary logistic regression was run to assess predictors. The result was presented as an adjusted odds ratio with a $95\%$ confidence interval. Variables with a $95\%$ confidence interval which not included one were considered statistically significant. ### Result A total of 372 (124 cases and 248 controls) patients were included in the study. Multivariable analysis revealed age ≥ 65 years (AOR = 2.62, $95\%$ CI = 1.33–5.14), having shortness of breath (AOR = 1.87, $95\%$ CI = 1.02–3.44), fatigue (AOR 1.78, $95\%$ CI = 1.09–2.90), altered consciousness (AOR 3.02, $95\%$ CI = 1.40, 6.49), diabetic Mellitus (AOR = 2.79, $95\%$ CI = 1.16–6.73), chronic cerebrovascular disease (AOR = 2.1, $95\%$ CI = 1.23, 3.88) were found to be predictors of death. ### Conclusion Older age, shortness of breath, fatigue, altered consciousness, and comorbidity were predictors of death in Severe COVID-19 patients. ## Introduction Coronavirus disease (COVID-19) is an infectious disease caused by a single-stranded RNA virus called severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1]. In December 2019, a cluster of patients with pneumonia of unknown cause emerged in Wuhan, China [2]. On January 2020, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was identified as a causative agent for that observed pneumonia cases [3]. World health organization named the disease caused by SARS-CoV-2 “COVID-19” (coronavirus disease 2019). On 11 March 2020, COVID-19 spreads to 144 countries and more than 118,000 cases and 4,000 deaths were reported; and WHO announced it as a pandemic [4]. In Africa, Egypt reported the first COVID-19 case on 14 February followed by Algeria on 25 February 2020 [5, 6], and Ethiopia reported the first COVID-19 case on 13 March 2020 [7]. Even if precautionary measures were practiced to prevent the rapid spread of the disease, the virus was quickly expanding all over the world and causing millions of death. As of the WHO report on 28 September 2021, there have been over 231 million cases and more than 4.7 million deaths worldwide. The United States of America reported the highest number of cases and deaths followed by India [8]. The case fatality rate of the disease caused by SARS-CoV-2 is 3.26–$4.16\%$ in Latin America; $5.8\%$ in the United States [9]. In Africa, a total of 5,998,863 cases and 144,957 deaths were reported [8]. As of the EPHI report on 19 September 2021, there were over 332 thousand confirmed COVID-19 cases and more than 5 thousand deaths in Ethiopia [10]. Since COVID-19 was announced as a worldwide pandemic, the world has been struggling with this disease [4], but the pandemic is still ongoing and the number of confirmed cases and mortality rates are changing every day. Virus mutations and appearing of new variants are also challenging to public health as they are more contagious and cause more severe illnesses [11, 12]. Similar to other RNA viruses, SARS-CoV-2 also continually mutates and new variants are appearing [11]. A variant of interest is defined as an isolate of SARS-CoV-2 that has genotypic and/or phenotypic changes compared to the reference genome. A variant of concern is defined as a variant of interest that has evidence of one or more increases in transmissibility or detrimental change in COVID-19 epidemiology, increase in virulence or change in clinical disease presentation; decrease in the effectiveness of available diagnostics, vaccines, therapeutics or public health and social measures [12, 13]. Globally, cases of the Alpha variant have been reported in 193 countries, while 142 countries have reported cases of the Beta variant; 96 countries have reported cases of the Gamma variant and the Delta variant has been reported in 187 countries [8]. In Ethiopia Alpha, Beta, and Delta SARS-COV-2 variants were detected [10]. Based on the WHO report on 28 September 2021, over 3.3 million new cases and over 55,000 new deaths were reported globally during the week of 20–26 September 2021. The African region reported over 87,000 new cases and over 2,500 new deaths in the reported week. The number of deaths in Africa shows a $5\%$ increase as compared to the previous week [8]. In Ethiopia, currently, people neglect the benefit of social distancing, hand washing, staying at home, using a mask, and suffering from the critical phase of COVID-19 [14]. As of the EPHI report on 19 September 2021, Ethiopia was ranked in the fourth position by the number of confirmed cases and in the sixth position by the number of deaths due to COVID-19 in Africa. A total of 9,857 new cases and 201 new deaths were reported during the week of 13–19 September 2021. The weekly number of cases and death increased by $21\%$ compared with the past week. The rate of positivity was higher in Sidama region compared with other regions which was $36\%$ [10]. The severity of the disease and treatment outcome of the patients varies from person to person. It may range from asymptomatic infection to severe disease with complications and lastly even death [15, 16]. It displayed a wide spectrum of clinical signs and symptoms, which included: fever, cough, sore throat, nasal congestion, sputum, headache, diarrhea, fatigue, dyspnea, chest tightness, myalgia, nausea, rhinorrhea, dizziness or confusion, hemoptysis, anorexia, vomiting, chest and abdominal pain [17]. After the first case of COVID-19 disease were reported in China the virus spread around the world and cause millions of deaths. Following that, international studies identified potential predictors of mortality among COVID-19 patients [18–20]. Regarding disease severity and predictor of mortality, studies identified different potential risk factors for disease severity and mortality among COVID-19 patients; including older age, male sex, pre-existing comorbidities, patient’s vital sign, and clinical symptoms during admission, radiographic and laboratory findings [19, 21–25]. Advanced age, male sex, cardiovascular comorbidities, acute cardiac or kidney injury, and lymphocytopenia, hypertension, diabetes, COPD, and history of CVD), acute organ injury, conferred an increased risk of death [26–28]. Age ⩾65 years was identified as a strong predictor of death for COVID-19 patients [20, 21, 29, 30]. Males were more likely to develop severe disease with complications and had much higher mortality than females [19, 22, 31–33]. Related to comorbidity status, hypertension, diabetes [20], chronic kidney disease, stroke [34], cardiovascular or cerebrovascular disease [21], and liver disease [18] were identified as predictors of death among severe COVID-19 patients. Patient’s vital signs and presenting symptoms on admission such as SpO2 less than $90\%$, RR greater than 20 breath /minute, heart rate greater than 100peats/ minute systolic blood pressure less than 90 mmHg, dyspnea, cough, breathing difficulty, vomiting and consciousness disorder were identified as a predictor of death [19, 24, 35, 36]. Identification of predictors of death and understanding the characteristic of severe COVID-19 patients is very important to provide efficient, equitable, and appropriate management for COVID-19 patients. In Africa, there are limited data regarding regional predictors of mortality among COVID -19 patients [33]. In Ethiopia, there is some information on epidemiological characteristics of the disease, Knowledge, attitudes, and practices of the population related to the COVID-19 pandemic, and treatment outcome of COVID-19 patients. But there is limited evidence related to predictors of death in COVID-19 in Ethiopia including Hawassa. To fill the gap it is important to conduct more research on the area of identifying risk factors for disease severity and predictor of death among COVID-19 patients in our local setting. Therefore, the purpose of this study was to assess the predictors of death in severe COVID-19 patients in Hawassa, southern Ethiopia in 2021. ## Study setting and period This study was conducted at Hawassa city COVID-19 treatment centers. Severe COVID-19 patients admitted in the centers from May 2020 to May 2021 were included in the study and data was extracted from 30 May 2021 to 30 June 2021. Hawassa city is located 273 km south of Addis Ababa, Ethiopia. During the pandemic, there was one quarantine center (at Hawassa University), two isolation centers (at pyramid hotel and Dukale wakayo hotel), and two treatment centers (Mehal sub-city and Hawassa University Comprehensive Specialized Hospital). The COVID-19 treatment center located in Mehal sub-city has 100 beds including 8 critical beds and two mechanical ventilators. Hawassa University Compressive Specialized Hospital COVID-19 treatment center has 100 beds including six intensive care unit beds of which four of them are equipped with mechanical ventilation. ## Study design An institution-based unmatched case-control study. ## Source population All patients with a confirmed diagnosis of severe COVID-19 using PT-PCR and admitted to COVID-19 treatment centers in Hawassa city. ## Study population All patients admitted to Hawassa city COVID-19 treatment centers with a confirmed diagnosis of severe COVID-19 from May 2020 to May 2021. Case. Cases were patients admitted to Hawassa city COVID-19 treatment centers from May 2020 to May 2021 with a confirmed diagnosis of severe COVID-19 and whose treatment outcome was death. Control. Controls were patients admitted to Hawassa city COVID-19 treatment centers from May 2020 to May 2021 with a confirmed diagnosis of severe COVID-19 and who recover from the disease and were discharged. ## Inclusion criteria All severe COVID-19 patients who were on treatment and follow-up at the Centers during the study period and whose treatment outcome status is known as dead or discharged were included in the study. ## Exclusion criteria Severe COVID-19 patients whose chart is incomplete for age, sex, comorbidity, vital sign, and clinical symptom were excluded. Severe COVID-19 patients whose outcome status is unknown due to transfer to other hospitals or any other reason that resulted in the discharge of the patient before the observation of the outcome were excluded from the study. ## Sample size determination The sample size was calculated via online Open Epi-info from the findings of previous similar studies. The calculation was done for four potential determinants (age ≥65, cough, male sex, and oxygen saturation (spo2) ≤$89\%$) which were consistently significant in many studies. To determine the appropriate sample size; two-sided confidence level, the power of the study, the ratio of controls to cases, the proportion of controls with exposure, relatively least extreme odds ratio to be detected, and lastly, $10\%$ for the incomplete chart was added. Based on this, the maximum calculated sample size was 366 (122 cases and 244 controls) and it was considered the minimum sample size of the study [36] (Table 1). Table 1 \| Variables \| CI \| Power \| AOR \| The ratio of control to case \| The proportion of control exposed \| case \| control \| 10% \| Total \| Ref. \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age ≥65 \| 95% \| 80% \| 3.765 \| 2 \| 30.4 \| 33 \| 66 \| 10 \| 109 \| [21] \| \| Cough \| 95% \| 80% \| 2.06 \| 2 \| 25.7 \| 111 \| 222 \| 33 \| 366 \| [36] \| \| SPO2 ≤89% \| 95% \| 80% \| 2.959 \| 2 \| 25.7 \| 49 \| 98 \| 15 \| 162 \| [19] \| \| Male sex \| 95% \| 80% \| 2.876 \| 2 \| 51.4 \| 54 \| 108 \| 16 \| 178 \| [19] \| Sample of cases. All confirmed severe COVID-19-related death in the study setting from May 2020 to May 2021 and whose chart is completed related to age, sex, comorbidity, and presenting signs and symptoms. Sample of controls. From all severe COVID-19 patients who were discharged with recovery from the COVID-19 treatment centers from May 2020 to May 2021, the sample of the control group was selected using a 1:2 case-to-control ratio. ## Sampling technique and procedure From May 2020 to May 2021, a total of 1,032 confirmed COVID-19 patients were admitted to COVID-19 treatment centers in the study area. Of the total admitted patients, 673 patients were severe cases. These severe COVID-19 patients are divided into the case (dead) and control (discharged) groups based on their outcome status. Of the total 673 severe COVID-19 patients, 128 were dead and 545 were discharged. Among the dead, 4 incomplete charts were excluded and 124 patients were included in the study. From the discharge, the sample of control was selected using a 1:2 case-to-control ratio. Finally, 372 (124 cases and 248 controls) were included in the study. To select the sample of control, a sampling frame was developed using discharged patient’s medical registration number. Then simple random sampling technique was used to select the required number of controls from each COVID-19 treatment center. The number of controls selected from each setting was proportional to the number of selected cases (Fig 1). Fig 1: Sampling technique for assessment of predictor of death among severely ill COVID-19 patients admitted to Hawassa city COVID-19 treatment centers, Hawassa, Sidama, Ethiopia, 2021. ## Dependent variables Treatment outcome of COVID-19 patients (Death versus Discharge). ## Independent variables Demographic variables; Age, sex, place of residence Signs and symptoms during admission; Temperature, heart rate, respiratory rate, blood pressure, fever, cough, sore trout, shortness of breath, chest pain, headache, loss of appetite, fatigue, altered consciousness, joint pain, and vomiting. Laboratory findings; White blood cell count, hemoglobin, hematocrit, platelets count, creatinine, urea, sodium, potassium, aspartate transaminase, alkaline phosphatase. Comorbidity; Hypertension, diabetes, cardiovascular disease, cerebrovascular disease, asthma, chronic obstructed pulmonary disease, chronic kidney disease, chronic liver disease, neurological disease, cancer, hematologic disease, malnutrition, HIV/AIDS, and tuberculosis. ## Data collection technique Data were extracted from the patient’s card using a standardized data collection tool, which was a modified version of the WHO/ International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) case record form for severe acute respiratory infection clinical characterization [37]. The extracted information about the patients were including demographic data, pre-existing comorbidities, vital signs, clinical symptoms, signs during admission, and the patient’s outcome status. ## Data quality and management Amendment to the tool was made after a pre-test on the data extraction tool conducted on 19 randomly selected charts at shashamene treatment center. Training on the data collection tool was given to four BSC nurse data collectors and one supervisor. Data consistency and completeness were checked before data entry and analysis. ## Data management and statistical analysis The data were coded, entered, cleaned, and stored in epi-data version 4.6 and exported to SPSS version 25 software for analysis. Descriptive statistics were presented by frequency and percentage for categorical variables and median and interquartile range for continuous variables. The statistically significant difference between died and discharged groups in terms of independent variables was assessed by Pearson’s chi-square test for categorical variables and Continuous variables were assessed by the Mann-Whitney U test due to their non-normally distribution. Those variables with p-value < 0.05 was defined as having statistical significance difference in terms of COVID-19 outcome. Multivariable binary logistic regression was run to assess the association between the dependent variable and independent variables. Initially, bivariate analysis was conducted and variables having p-value < 0.25 were further included in multivariate analysis to determine independent risk factors associated with COVID-19 mortality. For the final model, model assumption such as multicollinearity and Hosmer and Lemeshow goodness of fit test was checked and the result of the final model was presented as an adjusted odds ratio with a $95\%$ confidence interval. Variables having a $95\%$ confidence interval that does not include one in the final model were considered independent predictors of COVID-19 death. ## Operational definitions Sever COVID-19 disease. Adolescent or adult with clinical signs of pneumonia (fever, cough, dyspnea, fast breathing) plus one of the following: Respiratory rate ≥ 30 breaths/min; Peripheral oxygen saturation (SpO2) < $90\%$ on room air, arterial partial pressure of oxygen (PaO2)/fraction of inspired oxygen (FiO2) ≤300mmHg (1mmHg = 0.133 kPa). A child with clinical signs of pneumonia (cough or difficulty in breathing) plus at least one of the following: Central cyanosis or SpO2 < $90\%$, Severe respiratory distress (e.g. fast breathing, grunting, very severe chest indrawing), general danger sign (inability to breastfeed or drink, lethargy or unconsciousness, or convulsions), fast breathing (in breaths/min): < 2 months: ≥ 60; 2–11 months: ≥ 50; 1–5 years: ≥ 40 [38]. ## COVID-19 death A death due to COVID-19 is defined for surveillance purposes as a death resulting from a clinically compatible illness, in a probable or confirmed COVID-19 case, unless there is a clear alternative cause of death that cannot be related to COVID disease (e.g. trauma). There should be no period of complete recovery from COVID-19 between illness and death [39]. ## Ethical considerations Ethical approval was obtained from the Hawassa University, College of Medicine and Health Sciences Institutional Review Board. Besides, permission letters were given to respective institutions in the study area, (Hawassa University comprehensive specialized Hospital, Mehal sub city COVID-19 treatment center). Moreover, permission was obtained from the responsible person of each study institute after explaining the purpose and significance of the study. Face masks, sanitizer and gloves were provided for data collectors. The patient data was taken anonymously by using the patient’s medical registration number and the personal information of the patients stayed confidential. ## Socio-demographic characteristics and clinical presentation A total of 372 severely ill COVID-19 patients were included in the study (124 cases and 248 controls). The age of the patients ranged from 1 year to 90 years with a median of 46 (IQR: 25) years. The majority 233 ($62.6\%$) of the patients were male. Compared between the two groups, the patients in the death event group had a higher male proportion ($70.2\%$ vs. $58.9\%$), and older age (median 58 vs. 42) than patients who are discharged. In addition, the case group was more likely to present with fever ($49.6\%$ vs. $41.1\%$), shortness of breath ($80.6\%$ vs. $68.1\%$), fatigue ($51.6\%$ vs. $37\%$), and altered consciousness ($21\%$ vs. $6.5\%$) compared to control group (Table 2). Table 2 \| Variables \| Case (n = 124) \| Control (n = 248) \| p-value \| \| --- \| --- \| --- \| --- \| \| Median (IQR) age (Years) \| 58(28) \| 42(22) \| 0.001 \| \| Age groups (Years) \| \| \| \| \| <65 \| 91(73.4%) \| 225(90.7%) \| 0.001 \| \| ≥65 \| 33(26.6%) \| 23(9.3%) \| \| \| Sex \| \| \| \| \| Male \| 87(70.2%) \| 146(58.9%) \| 0.034 \| \| Female \| 37(29.8%) \| 102(41.1%) \| \| \| Residence \| \| \| \| \| Rural \| 35 (28.2%) \| 69 (27.8%) \| 0.935 \| \| Urban \| 89 (71.8%) \| 179 (72.2%) \| \| \| Fever \| \| \| \| \| Yes \| 51 (41.1%) \| 123 (49.6%) \| 0.123 \| \| No \| 73 (58.9%) \| 125 (50.4%) \| \| \| Cough \| \| \| \| \| Yes \| 111 (89.5%) \| 222 (89.5%) \| 1.0 \| \| No \| 13 (10.5%) \| 26 (10.5%) \| \| \| Sore throat \| \| \| \| \| Yes \| 8 (6.5%) \| 10 (4.0%) \| 0.305 \| \| No \| 116 (93.5%) \| 238 (96.0%) \| \| \| Shortness of breath \| \| \| \| \| Yes \| 100 (80.6%) \| 169 (68.1%) \| 0.011 \| \| No \| 24 (19.4%) \| 79 (31.9%) \| \| \| Chest Pain \| \| \| \| \| Yes \| 36 (29.0%) \| 86 (34.7%) \| 0.274 \| \| No \| 88 (71.0%) \| 162 (65.3%) \| \| \| Headache \| \| \| \| \| Yes \| 38 (30.6%) \| 96 (38.7%) \| 0.127 \| \| No \| 86 (69.4%) \| 152 (61.3%) \| \| \| Loss of appetite \| \| \| \| \| Yes \| 37 (29.8%) \| 68 (27.4%) \| 0.625 \| \| No \| 87 (70.2%) \| 180 (72.6%) \| \| \| Fatigue \| \| \| \| \| Yes \| 64 (51.6%) \| 92 (37.1%) \| 0.007 \| \| No \| 60 (48.4%) \| 156 (62.9%) \| \| \| Altered consciousness \| \| \| \| \| Yes \| 26 (21.0%) \| 16 (6.5%) \| 0.001 \| \| No \| 98 (79.0%) \| 232 (93.5%) \| \| \| Joint pain \| \| \| \| \| Yes \| 12 (9.7%) \| 35 (14.1%) \| 0.225 \| \| No \| 112 (90.3%) \| 213 (85.9%) \| \| \| Vomiting \| \| \| \| \| Yes \| 21(16.9) \| 31(12.5) \| 0.245 \| \| No \| 103(83.1) \| 217(87.5) \| \| ## Baseline vital signs and laboratory findings On admission, the median respiratory rate was 32 breaths per minute (IQR: 8) and the median heart rate was 120 beats per minute (IQR: 24 beats/min). The patients in the case group had higher respiratory rate (median 35.8 vs. 30 beats/min) and heart rate (median 105 vs. 98 beats/min). On the laboratory findings, patients in the case group had a lower value of Hematocrit (median 39.8 vs. 41.15), platelets count (median 234 vs. 225), and higher white blood cell count (median 13 vs. 8.4), and urea (median 46.5 vs. 29.6) than the control group (Table 3). Table 3 \| Variables \| Case \| Control \| p-value \| \| --- \| --- \| --- \| --- \| \| Presenting vital sign \| \| \| \| \| Median (IQR) body temperature in °C \| 37.2 (1.2) \| 37.1 (1.2) \| 0.412 \| \| Median (IQR) heart rate per minute \| 105 (86) \| 98 (22) \| 0.003 \| \| Median (IQR) respiratory rate per minute \| 35.81(16) \| 30 (6) \| 0.001 \| \| Median (IQR) systolic blood pressure \| 122.5(25) \| 120 (22) \| 0.407 \| \| Median (IQR) diastolic blood pressure \| 73.5(18) \| 73(14) \| 0.947 \| \| Complete cell count \| \| \| \| \| Median (IQR) White blood cell count (x 103/L) \| 13 (10) \| 8.4 (6.6) \| 0.001 \| \| Median (IQR) Hemoglobin (g/dL) \| 12.65 (4.2) \| 13.2(2.9) \| 0.052 \| \| Median (IQR) Hematocrit (%) \| 39.8(12.9) \| 41.15(8.8) \| 0.179 \| \| Median (IQR) platelets count (x 103/L) \| 234(147) \| 225(133.5) \| 0.85 \| \| Renal function test \| \| \| \| \| Median (IQR) creatinine (mg/dl) \| 1.08(0.8) \| 0.91(0.37) \| 0.001 \| \| Median (IQR) urea (mg/dl) \| 46.5(30.4) \| 29.6(20) \| 0.001 \| \| Electrolyte \| \| \| \| \| Median (IQR) sodium (mEq/L) \| 135 (10) \| 134(7.7) \| 0.17 \| \| Median (IQR) potassium (mEq/L) \| 4.3(1.6) \| 3.9(0.87) \| 0.039 \| \| Liver function test \| \| \| \| \| Median (IQR) Aspartate transaminase (IU/L) \| 47.65(34.65) \| 36(31) \| 0.051 \| \| Median (IQR) Alkaline phosphatase (IU/L) \| 139 (141) \| 106 (103) \| 0.154 \| ## Comorbidity status of the patients Of the total participants, 223 ($59.9\%$) patients presented with one or more comorbidity. The proportion of patients presenting with comorbidity was higher in the case group ($79\%$ vs. $50\%$) than in the controls. Diabetes mellitus ($37.1\%$ vs. $17.7\%$), hypertension ($35.5\%$ vs. $20.2\%$), and cerebrovascular disease ($17.7\%$ vs. $4.4\%$) were noted with a higher proportion in the case group than the control group (Table 4). Table 4 \| Variable \| Case (n = 124) # (%) \| Control (n = 248) # (%) \| p-value \| \| --- \| --- \| --- \| --- \| \| At least one comorbidity \| \| \| \| \| Yes \| 98 (79.0) \| 125 (50.4) \| 0.001 \| \| No \| 26 (21.0) \| 123 (49.6) \| \| \| Hypertension \| \| \| \| \| Yes \| 44 (35.5) \| 50 (20.2) \| 0.001 \| \| No \| 80 (64.5) \| 198 (79.8) \| \| \| Cardiovascular disease/non-hypertension/ \| \| \| \| \| Yes \| 15 (12.1) \| 24 (9.7) \| 0.473 \| \| No \| 109 (87.9) \| 224 (90.3) \| \| \| Diabetes Mellitus \| \| \| \| \| Yes \| 46 (37.1) \| 44 (17.7) \| 0.001 \| \| No \| 78 (62.9) \| 204 (82.3) \| \| \| Cerebrovascular disease \| \| \| \| \| Yes \| 22 (17.7) \| 11(4.4) \| 0.001 \| \| No \| 102 (82.3) \| 237 (95.6) \| \| \| Asthma \| \| \| \| \| Yes \| 7 (5.6) \| 15 (6.0) \| 0.876 \| \| No \| 117 (94.4) \| 233 (94) \| \| \| Obstructive pulmonary disease/non-Asthma/ \| \| \| \| \| Yes \| 1 (3.2) \| 4 (0.4) \| 0.026 \| \| No \| 120 (96.8) \| 247 (99.6) \| \| \| Kidney disease \| \| \| \| \| Yes \| 3 (2.4) \| 7 (2.8) \| 0.821 \| \| No \| 121 (97.6) \| 241 (97.2) \| \| \| Liver disease \| \| \| \| \| Yes \| 2 (1.6) \| 3 (1.2) \| 0.75 \| \| No \| 122 (98.4) \| 245 (98.8) \| \| \| Neurological disease \| \| \| \| \| Yes \| 3 (2.4) \| 1 (0.4) \| 0.076 \| \| No \| 121 (97.6) \| 247 (99.6) \| \| \| Cancer \| \| \| \| \| Yes \| 3 (2.4) \| 1 (0.4) \| 0.076 \| \| No \| 121 (97.6) \| 247 (99.6) \| \| \| Hematologic disease/non-cancer/ \| \| \| \| \| Yes \| 6 (4.8) \| 3 (1.2) \| 0.032 \| \| No \| 118 (95.2) \| 245 (98.8) \| \| \| Malnutrition \| \| \| \| \| Yes \| 3 (2.4) \| 6 (2.4) \| 1.0 \| \| No \| 121 (97.6) \| 242 (97.6) \| \| \| HIV/ADIS \| \| \| \| \| Yes \| 7 (5.6) \| 5 (2.0) \| 0.062 \| \| No \| 117 (94.4) \| 243 (98.0) \| \| \| Tuberculosis \| \| \| \| \| Yes \| 8 (6.5) \| 7 (2.8) \| 0.093 \| \| No \| 116 (93.5) \| 241 (97.2) \| \| ## Predictors of death among patients with severe COVID-19 The bivariate logistic regression analysis was initially performed on socio-demographic variables, presenting symptom and comorbidity parameters of the patients. As a result, variables including male sex, age ≥ 65 years, shortness of breath, fatigue, altered consciousness, Diabetic Mellitus, hypertension, and cerebrovascular disease showed significant predictors of death events. To determine independent risk factors associated with COVID-19 mortality, a multivariable analysis was performed. Variables that have a p-value of < 0.25 in bivariate analysis were selected for multivariate analysis. Multicollinearity effect was not observed (variance inflation factor (VIF) value lies between 1 and 2 in each involved variable). The final model was checked for goodness of fit using Hosmer and Lemeshow test and the value was 0.422 which means the actual model and the expected one has no significant difference. Based on multivariate analysis results age ≥ 65 years, shortness of breath, fatigue, altered consciousness, diabetic Mellitus, and chronic cerebrovascular disease were the independent predictors of COVID-19 death. The risk of mortality among severe COVID-19 patients with age ≥ 65 years was 2.62 times higher as compared to the patients with age < 65 years (AOR = 2.62, $95\%$CI = 1.33–5.14). Patients who presented with shortness of breath were $87\%$ more likely to have died than a patient without shortness of breath (AOR = 1.87, $95\%$ CI = 1.02–3.44). Mortality among COVID-19 patients who presented with a symptom of fatigue was $78\%$ higher compared to patients without fatigue (AOR = 1.78, $95\%$ CI = 1.09–2.90). Altered consciousness also increased the risk of COVID-19 death by 3 times (AOR = 3.02, $95\%$ CI = 1.40–6.49). The risk of mortality among COVID-19 patients with cerebrovascular disease was 2.79 times higher than patients without cerebrovascular disease (AOR = 2.79, $95\%$CI = 1.16–6.73). Patients with diabetics were two times as high as the patients without diabetics to die due to COVID-19 disease (AOR = 2.18, $95\%$ CI = 1.23–3.88) (Table 5). Table 5 \| Variable \| Case (n = 124) # (%) \| Control (n = 248) # (%) \| COR(95%CI) \| AOR(95%CI) \| \| --- \| --- \| --- \| --- \| --- \| \| Age \| \| \| \| \| \| <65 \| 91(73.4) \| 225(90.7) \| 1 \| 1 \| \| ≥65 \| 33(26.6) \| 23(9.3) \| 3.55 (1.98–6.37) \| 2.62 (1.33–5.14) * \| \| Sex \| \| \| \| \| \| Male \| 87(70.2) \| 146(58.9) \| 1.64 (1.04–2.6) \| 1.19 (0.71–1.99) \| \| Female \| 37(29.8) \| 102(41.1) \| 1 \| 1 \| \| Fever \| \| \| \| \| \| Yes \| 51 (41.1) \| 123 (49.6) \| 0.71 (0.46–1.1) \| 0.77 (0.45–1.28) \| \| No \| 73 (58.9) \| 125 (50.4) \| 1 \| 1 \| \| Shortness of breath \| \| \| \| \| \| Yes \| 100 (80.6) \| 169 (68.1) \| 1.95 (1.16–3.27) \| 1.87 (1.02–3.44) * \| \| No \| 24 (19.4) \| 79 (31.9) \| 1 \| 1 \| \| Headache \| \| \| \| \| \| Yes \| 38 (30.6) \| 96 (38.7) \| 0.7 (0.44–1.11) \| 0.75 (0.45–1.25) \| \| No \| 86 (69.4) \| 152 (61.3) \| 1 \| 1 \| \| Fatigue \| \| \| \| \| \| Yes \| 64 (51.6) \| 92 (37.1) \| 1.81 (1.17–2.8) \| 1.78 (1.09–2.9) * \| \| No \| 60 (48.4) \| 156 (62.9) \| 1 \| 1 \| \| Altered consciousness \| \| \| \| \| \| Yes \| 26 (21) \| 16 (6.5) \| 3.85 (1.98–7.49) \| 3.02 (1.4–6.49) * \| \| No \| 98 (79.0) \| 232 (93.5) \| 1 \| 1 \| \| Joint pain \| \| \| \| \| \| Yes \| 12 (9.7) \| 35 (14.1) \| 0.65 (0.33–1.31) \| 0.59 (0.27–1.29) \| \| No \| 112 (90.3) \| 213 (85.9) \| 1 \| 1 \| \| Vomiting \| \| \| \| \| \| Yes \| 21(16.9) \| 31(12.5) \| 1.43 (0.79–2.6) \| 1.47 (0.75–2.91) \| \| No \| 103(83.1) \| 217(87.5) \| 1 \| 1 \| \| Hypertension \| \| \| \| \| \| Yes \| 44 (35.5) \| 50 (20.2) \| 2.18(1.35–3.52) \| 1.29 (0.71–2.34) \| \| No \| 80 (64.5) \| 198 (79.8) \| 1 \| 1 \| \| Cerebrovascular disease \| \| \| \| \| \| Yes \| 22 (17.7) \| 11(4.4) \| 4.65 (2.17–9.94) \| 2.79 (1.16–6.73) * \| \| No \| 102 (82.3) \| 237 (95.6) \| 1 \| 1 \| \| Diabetic Mellitus \| \| \| \| \| \| Yes \| 46 (37.1) \| 44 (17.7) \| 2.73 (1.68–4.46) \| 2.18 (1.23–3.88) * \| \| No \| 78 (62.9) \| 204 (82.3) \| 1 \| 1 \| \| Tuberculosis \| \| \| \| \| \| Yes \| 8 (6.5) \| 7 (2.8) \| 2.37 (0.84–6.71) \| 2.9 (0.89–9.38) \| \| No \| 116 (93.5) \| 241 (97.2) \| 1 \| 1 \| ## Discussion This study assessed the predictors of death in severe COVID-19 patients who were admitted at Hawassa city COVID-19 treatment centers. Age ≥ 65 years, shortness of breath, fatigue, altered consciousness, diabetic Mellitus, and chronic cerebrovascular diseases were the independent predictors of COVID-19 death. The finding of this study revealed that older age was associated with a higher risk of death among severe COVID-19 patients. Patients with age ≥ 65 years had higher odds of death as compared to patients with age < 65 years. Similar to this finding, studies in the USA reported older age was found as a potential risk factor for death among COVID-19 patients [18, 26, 27]. A cohort study in china also reveals that age ⩾ 65 years was a strong predictor of death for COVID-19 patients [21]. A study conducted in Bangladesh also explained that mortality among hospitalized COVID-19 patients with age> 65 years was 3.59 times more likely higher as compared to the patients with age < 65 years [20]. This is due to there is correlation between age and natural immunity; as natural immunity declines gradually at older ages. As age increases, decreases the production of T cells and immunological responses to pathogens. It leads to less body fitness to fight infection and increased vulnerability and susceptibility to adverse health outcomes or death when exposed to infection [40]. Older people are also vulnerable to adverse drug reactions which may either reduce organ function at an older age or taking multiple drugs due to comorbidities [41, 42]. The result of this study also found that having shortness of breath at admission is a significant factor that predicts a death outcome. Patients who had a history of SOB at presentation had a higher risk of death outcome than those who had no history of SOB. Similarly, studies conducted in Ethiopia [43], Nigeria, and Congo reported shortness of breath as a significant risk factor for death among COVID-19 patients [24, 25, 36]. This is due to that severe pneumonia can cause significant gas exchange disturbances and lead to hypoxemia. Hypoxia reduces the energy production required for cell metabolism and increases the body’s anaerobic digestion. Acidosis and oxygen free radicals accumulated in the cell destroy the phospholipid layer of the cell membrane. As hypoxia continues, the intracellular calcium ion concentration increases significantly, leading to a series of cell damage processes [44]. As a result, shortness of breath is a manifestation of decreased lung function and is considered a sign of a life-threatening condition. In addition, fatigue was identified as a significant risk factor for death among severe Covid -19 patients in this study. In the same way, a study conducted in china explained patients presented with fatigue were at a $20\%$ higher risk of death than those without fatigue. However, no significant relationships were found between mortality and fever, cough, diarrhea, headache, abdominal pain, dizziness, nausea, and chest pain in this study as well as in the previous study [23]. This study also identified that altered consciousness was a risk factor for mortality among COVID-19 patients. The patient presented with altered consciousness were three times high risk to die than those without altered consciousness. Similarly, a study conducted in china suggested that there was a direct link between consciousness impairment and death in COVID- 19 patients. Patients whose Glasgow coma scale core was less than 9 were at high risk of death than patients with a GCS score >14 [45]. This may be because altered consciousness is a manifestation of organ failure due to continued hypoxemia. In the present analysis, diabetes was found to be an important predictor of death among severe COVID-19 patients. Patients with diabetics had higher odds of death than patients without diabetics. This result is consistence with other previous studies [20, 22]. A study conducted in Bangladesh explained patients with diabetics were 1.87 times higher than patients without diabetes to die [20]. A study conducted in Addis Ababa, Ethiopia also showed death in severe COVID-19 patients is found to be associated with being diabetic [43]. This could be because diabetes mellitus, especially if poorly controlled, lead to compromised immunity that decreases the body’s ability to fight off any infection [46]. They are more susceptible to being infected by viruses, bacteria, and fungi than individuals without diabetes. Also, the chances of having or developing another chronic illness are higher than in non-diabetic individuals [46, 47]. As a result, these patients might be at an increased risk of SARS-COV-2 infection which could result in a worse disease prognosis. This study identified risk of death among COVID-19 patients with cerebrovascular disease was higher than without the disease. In a previous study, a history of cerebrovascular disease is associated with a 2.78-fold increased risk of mortality compared to patients without underlying cerebrovascular disease [48]. A cohort study in china also consistently explained the risk of death among COVID-19 patient with cerebrovascular disease were 2.4 times higher than patients without the disease [21]. ## Conclusion This study tried to assess the predictors of death among severe COVID-19 patients admitted to Hawassa treatment centers. The study used a relatively larger sample size compared with a study done in Addis Ababa. It also addresses many variables and it is the study in the southern Ethiopian district which may include sociocultural differences from other study areas. Majorities of studies on predictors of death have been conducted in the context of Asian and American countries. In Africa, there are limited data regarding regional predictors of mortality among COVID -19 patients including Ethiopians and this finding may be one input. In conclusion, this study found that age was an important demographic variable that predict death outcomes among severely ill COVID-19 patients. Mortality among severe COVID-19 patients with age ≥65 years was higher as compared to the patients with age < 65 years. Clinical signs and symptoms such as shortness of breath, fatigue, and altered consciousness were the most significant predictor of death in severe COVID-19 patients. Regarding pre-existing comorbidities, having diabetes and cerebrovascular disease at admission were significant predictors to have death outcomes among severely ill COVID-19 patients. ## Recommendations Policymakers and those responsible to develop COVID-19 triage protocols shall give more focus on those risk groups and should be done with a more sensitive triaging method to pick them. If severe COVID-19 patients present with older age, difficulty in breathing, fatigue, altered consciousness, and comorbidities like diabetes and cerebrovascular disease, it is necessary to be alert for further deterioration of the patient’s condition and high risk for death. Therefore, Health care providers should be used these patients’ characteristics as a warning sign in a patient follow-up to provide early detection and intervention for a favorable outcome. Future researchers shall conduct a prospective study to get a chance to include all important variables like laboratory and radiologic findings. ## Limitation Due to the retrospective study design, radiographic and laboratory tests-related variables have high missing values and are not included in logistic regression analysis. The effect of the emergency of a new variant virus on the disease outcome is not covered in this study. ## References 1. Adam S, Emma B. Genetic Variants of SARS-CoV-2—What Do They Mean?. JAMA (2021) 325 529-531. DOI: 10.1001/jama.2020.27124 2. 2World Health Organization, Novel Coronavirus (‎ 2019-nCoV)‎: situation report, 1. 2020.. Novel Coronavirus (‎ 2019-nCoV)‎: situation report, 1 (2020) 3. Lu H., Stratton C.W., Tang Y.W.. Outbreak of pneumonia of unknown etiology in Wuhan, China: the mystery and the miracle. Journal of medical virology (2020) 92 401-402. PMID: 31950516 4. 4World Health Organization. WHO Director-General’s opening remarks at the media briefing on COVID-19-11 March 2020. 2020; Available from: https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19—11-march-2020.. WHO Director-General’s opening remarks at the media briefing on COVID-19-11 (2020) 2020 5. 5World Health Organization. COVID-19 cases top 10 000 in Africa. 2020; Available from: https://www.afro.who.int/news/covid-19-cases-top-10-000-africa.. COVID-19 cases top 10 000 in Africa (2020) 6. 6World Health Organization. A second COVID-19 case is confirmed in Africa. 25 Febuary 2020; Available from: https://www.afro.who.int/pt/node/12312.. A second COVID-19 case is confirmed in Africa (2020) 7. 7world Health Organization. First case of COVID-19 confirmed in Ethiopia. 13 March 2020; Available from: https://www.afro.who.int/news/first-case-covid-19-confirmed-ethiopia.. First case of COVID-19 confirmed in Ethiopia (2020) 8. 8World Health Organization, Weekly COVID-19 Epidemiological Update. 28 September 2021.. Weekly COVID-19 Epidemiological Update (2021) 9. Undurraga E.A., Chowell G., Mizumoto K.. COVID-19 case fatality risk by age and gender in a high testing setting in Latin America: Chile, March-August 2020. Infect Dis Poverty (2021) 10 11. DOI: 10.1186/s40249-020-00785-1 10. COVID-19 pandemic preparedness and response in Ethiopia, in Weekly bulletin. Bulletin No:73 (2021) 11. Detection of new SARS-CoV-2 variants related to mink. ECDC: Stockholm (2020) 12. Sars-cov-2 variant classifications and definitions.. (2021) 13. Tracking SARS-CoV-2 variants. (2021) 14. Yoseph A., Tamiso A., Ejeso A.. Knowledge, attitudes, and practices related to COVID-19 pandemic among adult population in Sidama Regional State, Southern Ethiopia: A community-based cross-sectional study. PloS one (2021) 16 e0246283. DOI: 10.1371/journal.pone.0246283 15. Huang C.. Clinical features of patients infected with 2019 novel coronavirus in Wuhan. China. The lancet (2020) 395 497-506 16. The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19)—China, 2020. China CDC weekly (2020) 2 113-122. DOI: 10.1016/j.ijid.2020.01.009 17. Sharma J.. Clinical Predictors of COVID-19 Severity and Mortality: A Perspective. Front Cell Infect Microbiol (2021) 11 674277. DOI: 10.3389/fcimb.2021.674277 18. Gupta S.. Factors Associated With Death in Critically Ill Patients With Coronavirus Disease 2019 in the US.. JAMA Intern Med (2020) 180 1-12. DOI: 10.1001/jamainternmed.2020.3596 19. Pan F.. Factors associated with death outcome in patients with severe coronavirus disease-19 (COVID-19): a case-control study. International journal of medical sciences (2020). DOI: 10.7150/ijms.46614 20. Noor F.M., Islam M.M.. Prevalence and associated risk factors of mortality among COVID-19 patients: A meta-analysis. Journal of community health (2020) 45 1270-1282. DOI: 10.1007/s10900-020-00920-x 21. Du R.-H.. Predictors of mortality for patients with COVID-19 pneumonia caused by SARS-CoV-2: a prospective cohort study. European Respiratory Journal (2020) 55 22. Chidambaram V.. Factors associated with disease severity and mortality among patients with COVID-19: A systematic review and meta-analysis.. PLoS One (2020) 15 e0241541. DOI: 10.1371/journal.pone.0241541 23. Yang L.. Risk factors for predicting mortality of COVID-19 patients: A systematic review and meta-analysis.. PLoS One (2020) 15 e0243124. DOI: 10.1371/journal.pone.0243124 24. Bepouka B.I.. Predictors of mortality in COVID-19 patients at Kinshasa University Hospital, Democratic Republic of the Congo, from March to June 2020. The Pan African Medical Journal (2020) 25. Abayomi A.. Presenting symptoms and predictors of poor outcomes among 2,184 patients with COVID-19 in Lagos State, Nigeria. International Journal of Infectious Diseases (2021) 102 226-232. DOI: 10.1016/j.ijid.2020.10.024 26. Mikami T.. Risk factors for mortality in patients with COVID-19 in New York City.. Journal of general internal medicine (2021) 36 17-26. DOI: 10.1007/s11606-020-05983-z 27. Chilimuri S.. Predictors of mortality in adults admitted with COVID-19: Retrospective cohort study from New York City. Western Journal of Emergency Medicine (2020) 21 779. PMID: 32726241 28. Figliozzi S.. Predictors of adverse prognosis in COVID-19: A systematic review and meta-analysis. European Journal of Clinical Investigation (2020) 50 e13362. PMID: 32726868 29. Biswas M.. Association of sex, age, and comorbidities with mortality in COVID-19 patients: a systematic review and meta-analysis. Intervirology (2020) 1-12 30. Chang M.C.. Risk factors for disease progression in COVID-19 patients.. BMC infectious diseases (2020) 20 1-6. DOI: 10.1186/s12879-020-05144-x 31. Su W.. Sex differences in clinical characteristics and risk factors for mortality among severe patients with COVID-19: a retrospective study. Aging (Albany NY) (2020) 12 18833. DOI: 10.18632/aging.103793 32. Li X.. Risk factors for severity and mortality in adult COVID-19 inpatients in Wuhan. Journal of Allergy and Clinical Immunology (2020) 146 110-118. PMID: 32294485 33. Mohammed M.. Risk Factors Associated with Mortality Among Patients with Novel Coronavirus Disease (COVID-19) in Africa.. Journal of racial and ethnic health disparities (2020) 1-6. DOI: 10.1007/s40615-020-00888-3 34. Tehrani S.. Risk factors for death in adult COVID-19 patients: Frailty predicts fatal outcome in older patients. International Journal of Infectious Diseases (2021) 102 415-421. PMID: 33130213 35. Sun H.. Risk factors for mortality in 244 older adults with COVID‐19 in Wuhan, China: a retrospective study. Journal of the American Geriatrics Society (2020) 68 E19-E23. DOI: 10.1111/jgs.16533 36. Elimian K.O.. Patient characteristics associated with COVID-19 positivity and fatality in Nigeria: retrospective cohort study. BMJ open (2020) 10 e044079 37. 37WHO. COVID-19 core case report form acute respiratory infection clinical characterisation data tool. 2020 April 2020 [cited 2021 02/02/2021]; Available from: https://www.elso.org/Portals/0/Files/ECMOCard/Update1/ISARIC%20WHO%20nCoV%20CORE%20CRF%2023.4.20.. COVID-19 core case report form acute respiratory infection clinical characterisation data tool (2020) 2020 38. COVID-19 clinical management: living guidance. (2021) 39. International guidelines for certification and classification (coding) of COVID-19 as cause of death. (2020) 40. Trzonkowski P.. Immunosenescence increases the rate of acceptance of kidney allotransplants in elderly recipients through exhaustion of CD4+ T-cells. Mech Ageing Dev (2010) 131 96-104. PMID: 20060852 41. Leng J., Daniel R. Impact of aging on viral infections. Microbes and infection (2010) 12 1120-1124. PMID: 20849976 42. Lavan A.H., Gallagher P.. Predicting risk of adverse drug reactions in older adults. Therapeutic advances in drug safety (2016) 7 11-22. DOI: 10.1177/2042098615615472 43. Leulseged T.W.. Predictors of death in severe COVID-19 patients at millennium COVID-19 care center in Ethiopia: a case-control study. Pan Afr Med J (2021) 38 351. DOI: 10.11604/pamj.2021.38.351.28831 44. Lenihan Colin R., Taylor Cormac T.. The impact of hypoxia on cell death pathways. Biochemical Society Transactions (2013) 41 657-663. PMID: 23514172 45. Xiong W.. Association of consciousness impairment and mortality in people with COVID‐19.. Acta Neurologica Scandinavica (2021). DOI: 10.1111/ane.13471 46. Hodgson K.. Immunological mechanisms contributing to the double burden of diabetes and intracellular bacterial infections. Immunology (2015) 144 171-185. PMID: 25262977 47. Xia C., Rao X., Zhong J.. Role of T lymphocytes in type 2 diabetes and diabetes-associated inflammation. Journal of diabetes research (2017) 2017 48. Florez-Perdomo W.A.. Relationship between the history of cerebrovascular disease and mortality in COVID-19 patients: a systematic review and meta-analysis. Clinical Neurology and Neurosurgery (2020) 197 106183. DOI: 10.1016/j.clineuro.2020.106183
--- title: 'Interleukin6 prediction of mortality in critically ill COVID19 patients: A prospective observational cohort study' authors: - Amira Jamoussi - Lynda Messaoud - Fatma Jarraya - Emna Rachdi - Nacef Ben Mrad - Sadok Yaalaoui - Mohamed Besbes - Samia Ayed - Jalila Ben Khelil journal: PLOS ONE year: 2023 pmcid: PMC9977034 doi: 10.1371/journal.pone.0279935 license: CC BY 4.0 --- # Interleukin6 prediction of mortality in critically ill COVID19 patients: A prospective observational cohort study ## Abstract ### Objective The aim of this study is to explore the role of IL6 in predicting outcome in critically ill COVID-19 patients. Design Prospective observational cohort study. Setting 20-bed respiratory medical intensive care unit of Abderrahmen Mami Teaching Hospital between September and December 2020. ### Methods We included all critically ill patients diagnosed with COVID-19 managed in ICU. IL6 was measured during the first 24 hours of hospitalization. ### Results 71 patients were included with mean age of 64 ± 12 years, gender ratio of 22. Most patients had comorbidities, including hypertension ($$n = 32$$, $45\%$), obesity ($$n = 32$$, $45\%$) and diabetes ($$n = 29$$, $41\%$). Dexamethasone 6 mg twice a day was initiated as treatment for all patients. Thirty patients ($42\%$) needed high flow oxygenation; 59 ($83\%$) underwent non-invasive ventilation for a median duration 2 [1–5] days. Invasive mechanical ventilation was required in 44 ($62\%$) patients with a median initiation delay of 1 [0–4] days. Median ICU length of stay was 11 [7–17] days and overall mortality was $61\%$. During the first 24 hours, median IL6 was 34.4 [12.5–106] pg/ml. Multivariate analysis shows that IL-6 ≥ 20 pg/ml, CPK < 107 UI/L, AST < 30 UI/L and invasive ventilation requirement are independent risk factors for mortality. ### Conclusions IL-6 is a strong mortality predictor among critically ill COVID19 patients. Since IL-6 antagonist agents are costly, this finding may help physicians to consider patients who should benefit from that treatment. ## Introduction Coronavirus disease 2019 (COVID-19) causes an inflammatory response and the degree of inflammatory cytokine storm is linked to COVID-19 related severity [1]. Cytokine storm is a maladaptive cytokine release in response to infection and other stimuli with a complex pathogenesis [2]. Studies describing the immunological profile of critically ill COVID-19 patients suggest hyperactivation of the humoral immune pathway Interleukin6 (IL6). Specifically, IL6 was highlighted to predict occurrence of respiratory failure, shock and multiorgan dysfunction [3]. It is now recognized that serum levels of IL-6 are significantly elevated in severe COVID-19 disease. Indeed, patients with complicated forms of COVID-19 had nearly threefold higher serum IL-6 levels than those with noncomplicated forms of the disease [3]. In the ICU, we need to rely on reliable outcome indicators as points of assessment. In critically ill COVID-19 patients, the ability of biomarkers to predict poor outcomes such as death is still under review. The aim of the present study is to explore whether IL6 levels can predict outcomes. Secondarily, we aimed to search for correlations between initial IL6 levels and others clinico-biological parameters. ## Study design and patients’ selection A single centre prospective cohort study was conducted in a dedicated COVID-19 ICU of Abderrahmen Mami Teaching Hospital between September and December 2020. Participants were recruited for this study from the inpatient ICU population with full verbal consent acquired from each patient prior to inclusion in our study. Inclusion criteria included all critically ill patients diagnosed with COVID-19 by RT-PCR and managed in the ICU between September and December 2020. ## Study protocol For all patients admitted to the ICU, IL-6 levels were measured during the first 24 hours. Quantitative determination of serum IL-6 levels was performed with a solid phase, enzyme labelled, chemiluminescent immunometric assay on an Immulite 1000 analyser (Siemens™). Normal range of serum IL-6 quantification was under 2 pg/mL. ## Clinical data We collected information on demographic characteristics, underlying diseases and chest CT imaging. Severity of illness was evaluated according to the Simplified Acute Physiology Score II (SAPS II) and the Acute Physiology and Chronic Health Evaluation (APACHE) II during the first 24 hours in the ICU. Management features such as steroids, the need for invasive mechanical ventilation and/or non-invasive mechanical ventilation, were also recorded. Laboratory results including arterial blood gases (ABG), lactates, blood cell counts, Zinc, C-reactive protein (CRP), fibrinogen, Ddimer, urea, creatinine, creatine-phospho-kinase (CPK), lactico-deshydrogenase (LDH), Aspartate aminotransferase (AST), Alanine aminotransferase (ALT), N-terminal pro-B type natriuretic peptide (NT pro BNP) and Troponin were reported for each patient. Lastly, outcome data such as length of stay and ICU mortality were recorded. ## Objectives Primary outcome was to explore the role of IL6 in the prediction of COVID-19 related ICU mortality and invasive mechanical ventilation requirement. Secondary outcomes were to investigate IL-6’s correlations with biological data also measured during the first 24 hours of hospitalization; and to search for all independent ICU mortality risk factors. ## Statistical analyses SPSS 23.0. was used for statistical analyses. Descriptive statistics of the patients were calculated and reported in terms of absolute frequencies and percentages for the qualitative variables. Quantitative variables in our cohort according to the Kolmogorov-Smirnov test were predominantly non-parametric. Quantitative variables were expressed as either medians and IQR 25th and 75th percentiles or in terms of means ± standard deviation (SD). Analysis of the differences in clinical characteristics, biological data and management requirements between survivors and non-survivors was performed. The differences between independent groups regarding continuous variables were evaluated by Student’s t-test. Nominal data were analysed by Pearson’s Chi-square test or Fisher’s Exact test, when appropriate. Medians of quantitative variables between groups were compared using the Mann-Whitney nonparametric test. Optimal cut off values were also determined using receiver operating characteristic curve (ROC) analysis. Variables which showed a significant p value in the univariate analysis were entered into the model. A logistic regression was performed to obtain an adjusted estimate of the odds ratios (ORs) and to identify which factors were independently associated with ICU mortality. We tested the role of IL-6 as risk factor for negative outcome. The correlation between IL-6 and biological data was studied according to Spearman’s coefficient and curves built if the correlation was significant. Data were considered to be statistically significant, if the p values were less than 0.05. ## Ethical considerations All data used in the analysis were collected in the routine surveillance activities suggested during COVID-19 pandemic, so did not require informed consent. All data were fully anonymized before we accessed them. Study ethical approval was given by institutional review board (Abderrahmen Mami hospital’s local committee) and waived the requirement for informed consent. Patients agreed to study results dissemination. ## Baseline characteristics A total of seventy-one patients who were diagnosed with coronavirus and admitted to the ICU were included in this study, with a mean age 64 ± 12 years and gender ratio of 2,2. The reason for admission to the ICU was acute respiratory failure in all patients with a median PaO2/FiO2 ratio of 120 IQR [85–156] mmHg. Chest CT was performed in 45 ($63\%$) patients showing ground glass opacities and/or consolidations in all cases. Pulmonary parenchyma damage extent was estimated to be >$75\%$ ($$n = 13$$), 50–$75\%$ ($$n = 10$$), 25–$50\%$ ($$n = 17$$) and < $25\%$ ($$n = 5$$). Baseline characteristics and onset severity of all participants are further detailed in Table 1. Table 1 \| Unnamed: 0 \| n = 71 \| \| --- \| --- \| \| Age, mean ± SD, years \| 64±12 \| \| Gender, n (%) \| \| \| Male \| 49 (69) \| \| Female \| 22 (31) \| \| SAPSII, mean ± SD \| 30 ± 10 \| \| APACHEII, mean ± SD \| 10 ± 6 \| \| Comorbidities, n (%) \| \| \| Hypertension \| 32 (45) \| \| Diabetes \| 29 (41) \| \| Obesity \| 32 (45) \| \| Symptoms delay, med [IQR] \| 10 [7–13] \| \| Acute hypoxemic respiratory failure, n (%) \| 71 (100) \| ## Laboratory findings At ICU admission, lymphopenia (< 1.500 103/mm3) was noticed in 54 cases ($76\%$) of which 30 ($42\%$) was very low (< 1.000 103/mm3). Zinc serum deficiency (<70 μg/L) was recorded in 47 patients (%). Rhabdomyolysis (CPK > 1000 UI/l) was noticed in 5 patients ($7\%$) and marked transaminases elevation in 9 patients ($13\%$). Main laboratory findings of all participants are detailed in Table 2. Table 2 \| Unnamed: 0 \| Median \| IQR \| \| --- \| --- \| --- \| \| White blood cells, 103/mm3 \| 9.5 \| 6.800–13.400 \| \| Lymphocytes, 103/mm3 \| 1.02 \| 0.760–1.420 \| \| Hemoglobin, g/dL \| 13.0 \| 11.6–14 \| \| Platelets, 103/mm3 \| 251.0 \| 201–293 \| \| CPK, UI/L \| 114.0 \| 56–233 \| \| LDH, UI/L \| 425.0 \| 341–603 \| \| AST, UI/L \| 31.5 \| 22–45.2 \| \| ALT, UI/L \| 27.0 \| 20–42.5 \| \| IL-6, pg/ml \| 34.4 \| 12.5–106 \| \| Zinc, μg/L \| 51.5 \| 36.2–69 \| \| CRP, mg/L \| 119.5 \| 54.8–187.7 \| \| Troponin, ng/ml \| 0.01 \| 0–0.03 \| \| NT-proBNP, pg/ml \| 31.8 \| 14–72 \| \| Fibrinogen, g/L \| 6.36 \| 4.68–6.89 \| \| D-Dimer, μg/L \| 670.0 \| 420–2950 \| \| pH \| 7.47 \| 7.43–7.51 \| \| PaO2/FiO2, mmHg \| 120.0 \| 85–156 \| \| PaCO2, mmHg \| 36.0 \| 32–42 \| ## IL-6 description Serum IL-6 concentration was high in all patients, median and IQR were 34.4 [12.5–106] pg/ml. IL-6 showed a significant correlation with age ($r = 0.301$, $$p \leq 0.011$$), CRP ($r = 0.287$, $$p \leq 0.021$$), zinc (r = -0.326, $$p \leq 0.011$$), and D-dimers ($r = 0.281$, $$p \leq 0.033$$) (Fig 1). We focused on IL-6 level association with outcome parameters (Table 3) which were significant for ICU mortality. Under ROC curve analysis (Fig 2), IL-6 ≥ 20 pg/ml was found to be significantly associated with ICU mortality (AUC = 0.805, $p \leq 10$−3, sensitivity = 0.837 and specificity = 0.679). Fig 1: Serum IL-6 concentration correlation with age, C-reactive protein, zinc, and D-dimers (r = Spearman’s coefficient). Fig 2: Receiver operating characteristic (ROC) curve showing the predictive power of IL-6 for predicting ICU mortality among critically ill COVID-19 patients. TABLE_PLACEHOLDER:Table 3 ## Management and outcome Dexamethasone was the main treatment provided at a dose of 6 mg twice a day. All patients required respiratory assistance. Thirty patients ($42\%$) needed high flow oxygenation; 59 ($83\%$) underwent non-invasive ventilation for a median duration 2 [1–5] days. Invasive mechanical ventilation was required in 44 ($62\%$) patients with a median initiation delay of 1 [0–4] days. Median ICU length of stay was 11 [7–17] days and overall mortality was of $61\%$. ## Mortality analysis A comparison of the characteristics of survivors and non survivors is detailed in Table 4. Table 4 \| Unnamed: 0 \| Non-Survivors (n = 43) \| Survivors (n = 28) \| p \| \| --- \| --- \| --- \| --- \| \| Age, mean ± SD, years \| 65.9 ± 7.4 \| 60.5 ± 15.6 \| 0.093 \| \| Male gender, n (%) \| 34 (70%) \| 15 (30%) \| 0.023 \| \| SAPS II, med [IQR] \| 32 [27–37] \| 24 [21–33.7] \| 0.004 \| \| APACHE II, med [IQR] \| 10 [7–14] \| 8 [6–11.7] \| 0.038 \| \| IL-6, med [IQR], pg/ml \| 70.8 [27.7–143] \| 14.7 [3.4–33.3] \| < 10−3 \| \| Zinc, med [IQR], μg/L \| 51.5 [36.7–70.5] \| 51.5 [32–69] \| 0.0662 \| \| WBC, med [IQR], 103/mm3 \| 10.000 [7.100–13.600] \| 8.650 [6.425–11.925] \| 0.158 \| \| Lymphocytes, mean ± SD, 103/mm3 \| 1117 ± 517 \| 1498 ± 1353 \| 0.098 \| \| Haemoglobin, mean ± SD, g/dL \| 12.7 ± 2.3 \| 12.5 ± 2.4 \| 0.316 \| \| Platelets, med [IQR], 103/mm3 \| 266.000 [203.000–315.000] \| 235.500 [193.750–288.000] \| 0.129 \| \| CPK, med [IQR], UI/L \| 84 [43–183] \| 171 [73–325] \| 0.050 \| \| LDH, med [IQR], UI/L \| 450 [357–603] \| 408 [308–613] \| 0.488 \| \| AST, med [IQR], UI/L \| 26.5 [21–41.7] \| 39.5 [27–48.7] \| 0.048 \| \| ALT, med [IQR], UI/L \| 27.5 [19.7–40.5] \| 27 [20–46] \| 0.986 \| \| CRP, med [IQR], mg/L \| 115 [54–190] \| 125 [56–187] \| 0.725 \| \| Troponin, med [IQR] \| 0.01 [0–0.28] \| 0.015 [0–0.08] \| 0.332 \| \| NT-proBNP, med [IQR], pg/ml \| 36 [16.2–78] \| 28 [11.4–51.2] \| 0.105 \| \| Fibrinogen, med [IQR], g/L \| 6.36 [4.6–6.9] \| 5.24 [4.6–6.9] \| 0.974 \| \| Lactates, med [IQR], meq/L \| 1.6 [1–2] \| 1.4 [1–1.9] \| 0.776 \| \| PaO2/FiO2, med [IQR], mmHg \| 97 [83–152] \| 134.5 [102–188] \| 0.067 \| \| DDimer, med [IQR], μg/L \| 1040 [410–4340] \| 630 [420–980] \| 0.259 \| \| Invasive MV, n (%) \| 33 (76.7%) \| 11 (39%) \| 0.001 \| Multivariate logistic regression, detailed in Table 5, indicates that IL-6 ≥ 20 pg/ml, CPK < 107 UI/L, AST < 30 UI/L and invasive ventilation are independent risk factors for mortality. Table 5 \| Unnamed: 0 \| p \| OR \| CI 95% \| \| --- \| --- \| --- \| --- \| \| Male gender \| 0.213 \| 2.9 \| 0.542–15.584 \| \| SAPS II ≥ 28 \| 0.160 \| 4.02 \| 0.578–27.989 \| \| APACHE II ≥ 9 \| 0.483 \| 2.007 \| 0.287–14.055 \| \| IL-6 ≥ 20 pg/ml \| < 10−3 \| 55.3 \| 5.910–517.77 \| \| CPK < 107 UI/L \| 0.021 \| 10.263 \| 1.413–74.521 \| \| AST < 30 UI/L \| 0.022 \| 9.37 \| 1.372–64.069 \| \| Invasive MV requirement \| 0.042 \| 6.444 \| 1.066–38.972 \| ## Discussion In this prospective cohort study of critically ill COVID19 patients, we report 3 major findings. First, IL-6 is a strong mortality predictor, more accurately a cut-off concentration of IL-6 ≥ 20 pg/mL at initial assessment in ICU. Second, IL-6 has significant correlations with age, CRP, zinc and D-dimer. Third, independent risk factors for ICU mortality, in addition to IL-6, were invasive mechanical ventilation requirement, AST < 30 UI/L and CPK < 107 UI/L. Relevance of IL-6 in mortality prediction was already studied for several clinical conditions, other than COVID19. This biomarker has been shown to predict mortality in heart failure [4], haemodialysis patients [5], end-stage liver disease [6] or hospitalized patients with cancer [7]. There is a substantial body of evidence linking the IL6 concentration to the severity of disease and unfavourable outcome of Covid-19 [8–10]. In our study, the optimal cut-off value of IL-6 to predict mortality was 20 pg/mL. Grifonin E. et al [11] reported IL-6 of > 25 pg/ml to be a sufficient predictor for severe COVID19 and/or in hospital mortality. It should be noted that the authors had excluded patients requiring immediate ICU admission (indicating patients with a lesser degree of disease severity). Targeting the cytokine storm induced by SARS-CoV-2 by using anti-IL-6 drugs is considered as a therapeutic option in several countries, with documented evidence for its efficiency [12, 13]. In Tunisia, medications that target the cytokine storm caused by COVID-19 (tocilizumab and sarilumab) are costly and not always available. In fact, they cannot be provided where indicated according to scientific evidence due to these factors. Determining an IL-6 cut-off score to predict fatal outcome could help physicians to decide who should benefit from these treatments, where available. Significant correlations highlighted in our study are partially in accordance with Zhou J. and colleagues who reported similar correlations between serum IL-6 concentrations and age, urea, creatinine, NT-proBNP, cTroponin I, C-reactive protein and procalcitonin [14]. Zinc is an anti-inflammatory and antioxidant micronutrient available in food with a well-demonstrated relationship to immunity [15]. Zinc deficiency causes immunodeficiency with severe lymphopenia that is characterized by a considerable decrease in developing B cell compartments in the bone marrow [16]. A prospective study investigated the role of zinc deficiency in COVID-19 outcomes and found an OR of 5.54 for developing complications in zinc deficient COVID-19 patients ($95\%$ CI 1.56–19.6, $$p \leq 0.008$$) and 5.48 ($95\%$ CI 0.61–49.35, $$p \leq 0.129$$) for mortality [17]. We highlight in this study significant zinc serum deficiency among participants as well as a negative correlation with IL-6. Nevertheless, this correlation is statistically significant but fairly weak. Since the severity of COVID-19 is related, in a large measure, to the extent of pulmonary involvement and consistent hypoxemia, the requirement for invasive mechanical ventilation was expected to be a fatal outcome predictor. Indeed, it simply means escalation after failure of non-invasive respiratory assistance among most severe patients. Data on CPK, which is a marker of muscular damage, is only briefly mentioned in most papers on COVID-19. Muscle pain and fatigue are often reported during SARS-COV2 infections, independent of severity. Elevated CPK may occur in COVID-19, but it remains unclear whether it is due to a virus-triggered inflammatory response or direct muscle toxicity [18]. Besides, in critically ill patients with acute respiratory failure, CPK elevation may result from additional respiratory muscle effort observable during respiratory distress. In a retrospective cohort study in Italy which included 331 COVID-19 patients, authors reported that increased CPK > 200 UI/L may predict a worse COVID-19 outcome [19]. Our data suggest that CPK < 107 UI/L is an independent marker of mortality risk. It appears that elevated CPK is a healthier host response against virus invasion. All studied patients had acute respiratory failure documented with hypoxemia in blood gases. Nevertheless, many of them were ‘happy hypoxemic’ and did not show any polypnoea or respiratory distress signs. Those patients often delay consulting, are tardily managed and had worse outcomes. So as a possible explanation, we have hypothesized that a lack of elevated CPK may be observed among these patients, but this remains to be confirmed. ## Strengths and limitations Key strengths were prospective study design and quality correlations between IL-6 and other biological data measured in the same 24-hour period. To our knowledge, this is the first study providing IL-6 characteristics in COVID-19 critically ill patients managed in Tunisia and determining fatal cut-off, which may be helpful for determining treatment regimens and management in severe cases. Limitations are mainly represented by single-center design. ## Conclusions The COVID-19 pandemic continues to threaten patients, societies, economies and healthcare systems around the world. During the first 24 hours of ICU admission, IL-6 ≥ 20 pg/mL as a cut off value is a predictor of fatal outcome. In low-income countries where IL-6 antagonists are not constantly available, dosing IL-6 at ICU admission may help physicians to decide who should mandatory benefit from these treatments. This can take part of a cost saving approach. In severe COVID-19 patients in ICU, we recommend serum level determination at admission to predict outcome and establish a therapeutic plan. Nevertheless, we need multicenter studies to improve our knowledge of the sensitivity and specificity of serum IL-6 cut-off values associated with fatal outcomes. ## References 1. Wan S, Yi Q, Fan S, Lv J, Zhang X, Guo L. Characteristics of lymphocyte subsets and cytokines in peripheral blood of 123 hospitalized patients with 2019 novel coronavirus pneumonia (NCP). MedRxiv (2020) 2. Ye Q, Wang B, Mao J. The pathogenesis and treatment of theCytokine Storm’in COVID-19. Journal of infection (2020) 80 607-13. DOI: 10.1016/j.jinf.2020.03.037 3. Coomes EA, Haghbayan H. Interleukin-6 in COVID-19: a systematic review and meta-analysis. Reviews in medical virology (2020) 30 1-9. DOI: 10.1002/rmv.2141 4. Markousis-Mavrogenis G, Tromp J, Ouwerkerk W, Devalaraja M, Anker SD, Cleland JG. The clinical significance of interleukin-6 in heart failure: results from the BIOSTAT-CHF study. European journal of heart failure (2019) 21 965-73. DOI: 10.1002/ejhf.1482 5. Loc ND, Kien NT, Dung NH, Quyen DBQ, Tuan NM, Kien TQ. Interleukin 6 is a better predictor of 5-year cardiovascular mortality than high-sensitivity C-reactive protein in hemodialysis patients using reused low-flux dialyzers. International urology and nephrology (2020) 52 1135-42. DOI: 10.1007/s11255-020-02461-7 6. Ponzetto A, Figura N. Interleukin 6 Predicts Mortality in Patients With End-Stage Liver Disease. Clinical Gastroenterology and Hepatology (2018) 16 783. DOI: 10.1016/j.cgh.2017.10.012 7. Stoll JR, Vaidya TS, Mori S, Dusza SW, Lacouture ME, Markova A. Association of interleukin-6 and tumor necrosis factor-α with mortality in hospitalized patients with cancer. Journal of the American Academy of Dermatology (2021) 84 273-82. PMID: 32171811 8. Mandel M, Harari G, Gurevich M, Achiron A. Cytokine prediction of mortality in COVID19 patients. Cytokine (2020) 134 155190. DOI: 10.1016/j.cyto.2020.155190 9. Zhu Z, Cai T, Fan L, Lou K, Hua X, Huang Z. Clinical value of immune-inflammatory parameters to assess the severity of coronavirus disease 2019. International Journal of Infectious Diseases (2020) 95 332-9. DOI: 10.1016/j.ijid.2020.04.041 10. Liu F, Li L, Xu M, Wu J, Luo D, Zhu Y. Prognostic value of interleukin-6, C-reactive protein, and procalcitonin in patients with COVID-19. Journal of clinical virology (2020) 127 104370. DOI: 10.1016/j.jcv.2020.104370 11. Grifoni E, Valoriani A, Cei F, Lamanna R, Gelli AMG, Ciambotti B. Interleukin-6 as prognosticator in patients with COVID-19. Journal of Infection (2020) 81 452-82. DOI: 10.1016/j.jinf.2020.06.008 12. Domingo P, Mur I, Mateo GM, del Mar Gutierrez M, Pomar V, de Benito N. Association between administration of IL-6 antagonists and mortality among patients hospitalized for COVID-19: a meta-analysis. Jama (2021) 326 499-518. DOI: 10.1001/jama.2021.11330 13. Zeraatkar D, Cusano E, Martinez JPD, Qasim A, Mangala SO, Kum E. Tocilizumab and sarilumab alone or in combination with corticosteroids for COVID-19: A systematic review and network meta-analysis. medRxiv (2021) 14. Zhou J, He W, Liang J, Wang L, Yu X, Bao M. Association of interleukin-6 level with morbidity and mortality in patients with coronavirus disease 2019 (COVID-19). Jpn J Infect Dis (2020) 15. Pal A, Squitti R, Picozza M, Pawar A, Rongioletti M, Dutta AK. Zinc and COVID-19: basis of current clinical trials. Biological Trace Element Research (2020) 1-11. DOI: 10.1007/s12011-020-02437-9 16. Bonaventura P, Benedetti G, Albarède F, Miossec P. Zinc and its role in immunity and inflammation. Autoimmunity reviews (2015) 14 277-85. DOI: 10.1016/j.autrev.2014.11.008 17. Jothimani D, Kailasam E, Danielraj S, Nallathambi B, Ramachandran H, Sekar P. COVID-19: Poor outcomes in patients with zinc deficiency. International Journal of Infectious Diseases (2020) 100 343-9. DOI: 10.1016/j.ijid.2020.09.014 18. Pitscheider L, Karolyi M, Burkert FR, Helbok R, Wanschitz JV, Horlings C. Muscle involvement in SARS-CoV-2 infection. European journal of neurology (2020). DOI: 10.1111/ene.14564 19. Orsucci D, Trezzi M, Anichini R, Blanc P, Barontini L, Biagini C. Increased Creatine Kinase May Predict A Worse COVID-19 Outcome. Journal of Clinical Medicine (2021) 10 1734. DOI: 10.3390/jcm10081734
--- title: Looking beyond the individual–The importance of accessing health and cultural services for Indigenous women in Thunder Bay, Ontario authors: - Jonathan C. Lin - Elaine Toombs - Chris Sanders - Candida Sinoway - Marni Amirault - Christopher J. Mushquash - Linda Barkman - Melissa Deschamps - Meghan Young - Holly Gauvin - Anita C. Benoit journal: PLOS ONE year: 2023 pmcid: PMC9977040 doi: 10.1371/journal.pone.0282484 license: CC BY 4.0 --- # Looking beyond the individual–The importance of accessing health and cultural services for Indigenous women in Thunder Bay, Ontario ## Abstract Access to cultural activities and culturally relevant healthcare has always been significant for achieving holistic Indigenous health and continues to be a key factor in shaping the health journey of Indigenous individuals and communities. Previous research has indicated the importance of cultural practices and services in sustaining cultural identity for Indigenous peoples, which is a major influence on their wellbeing. This study marks the first phase in a project aimed at establishing an Indigenous healing program and uses a qualitative research approach to understand the health and cultural services that Indigenous women want and require in Thunder Bay, Ontario. During interviews, participants ($$n = 22$$) answered questions around their understandings of health and wellbeing, and how they are able to incorporate cultural practices into their circle of care. Thematic analysis was performed on interview transcripts, and 4 key themes were identified: ‘independence and self-care’, ‘external barriers to accessing services’, ‘finding comfort in the familiar’ and ‘sense of community’. Together these themes illustrate how Indigenous women feel a strong sense of personal responsibility for maintaining their health despite the multiple environmental factors that may act as barriers or supports. Furthermore, the necessity of embedding cultural practices into Indigenous women’s circle of care is highlighted by the participants as they describe the mental, spiritual, social, and emotional health benefits of engaging in cultural activities within their community. The findings demonstrate the need for current modes of care to look beyond the individual and consider the impacts that socio-environmental factors have on Indigenous women. To accomplish this, we hope to increase access to health and cultural services through the creation of an Indigenous healing program that can be adequately incorporated into Indigenous women’s circle of care if they wish to do so. ## Introduction Cultural continuity is the level of social and cultural cohesion within a community, particularly involving intergenerational connectedness, which is maintained through engagement with family members or other community members that transmit knowledge and pass on traditions to subsequent generations [1, 2]. Cultural continuity has been identified as a determinant of health for Indigenous Peoples, and is associated with positive health outcomes that are achieved through ceremonies, intergenerational transmission of cultural knowledge, and creating a sense of belonging and cultural identity [1, 3]. Cultural identity acts as a major influence on improving Indigenous girls’ and women’s confidence and self-esteem by dispelling common negative stereotypes or assumptions attached to them by society [4]. Having this strong sense of cultural identity and self-esteem has been linked with positive outcomes for wellbeing by playing a protective role for mental and emotional health in Indigenous populations, particularly Indigenous youth [3, 5–8]. Indigenous peoples have long engaged in their cultural practices as part of healing and medicine, sharing teachings, and maintaining reciprocal relationships with the environment [9–12]. Ensuring culture-based treatment options for Indigenous service users has the potential to influence broader wellbeing for Indigenous peoples, as self-care is largely embedded within cultural practices such as smudging, ceremony, and the gathering and use of medicinal plants [13, 14]. There is a diverse range of culturally-based practices and treatments for Indigenous peoples, which derive from a variety of Indigenous traditions that exist among different Indigenous groups [14]. It is necessary to support continued access to these cultural practices and healing approaches for Indigenous women in their circle of care, as maintaining these practices also help to improve autonomy and access to diverse health services [15]. Indigenous peoples continue to face various barriers in obtaining adequate and equitable health care in Canada [16]. This has been partially attributed to the emphasis on predominately non-Indigenous health services of biomedical paradigms that are unable to encapsulate the unique needs and experiences of Indigenous Peoples fully [17–19]. Previous studies have shown that health care provisioning can be improved when the model of care is tailored to the needs of, or owned and managed by Indigenous communities themselves [20, 21]. Having the autonomy to control and decide what services they want is likely to lead to a model of health care provisioning that minimizes experiences of discrimination for Indigenous Peoples while maximizing cultural relevance [3, 22]. Beyond Indigenous healing approaches, Indigenous health and wellbeing also signifies the importance of relationships and interdependence among individuals and their families, communities, nature, and spirit [23, 24]. Having support from relationships and the wider community have been associated with thriving health for Indigenous individuals, and it has been suggested that health programs may have greater health effects if they simultaneously build on supporting positive social interactions across the community level [25, 26]. Both service users and service providers must understand that many socio-environmental factors play a role in holistically shaping an individual’s health, and self-care practices are only part of the greater whole. Maintaining relationships with land, nature, and the social and cultural environment is a critical component in achieving holistic health for Indigenous Peoples, and acknowledging this in mainstream services will be key in expanding the relevance and acceptability of services for Indigenous women while minimizing instances of self-blame for those experiencing poor health outcomes [25, 27–29]. This paper will describe four key themes identified from participant interviews that illustrate the importance of considering the multiple actors involved in influencing an individual’s ability to access health and cultural services for their wellbeing. We then conclude by discussing how these themes may be useful when informing or conceptualizing the creation of an Indigenous healing program. This study describes the first phase in a project aimed at establishing an Indigenous healing program that is sustainable and accessible for Indigenous women living in Thunder Bay, Ontario. The study objectives were to determine Indigenous women’s understanding of health and wellbeing, and to identify Indigenous healing approaches that they use or desire. Findings will inform the development of the Indigenous healing program. ## Research team Our community partners were Elevate NWO and the Ontario Aboriginal HIV/AIDS Strategy (Oahas). Elevate NWO is taking action to address the Truth and Reconciliation Committee of Canada’s (TRCC) [30] Call to Action #22: “We call upon those who can effect change within the Canadian health-care system to recognize the value of Aboriginal healing practices and use them in the treatment of Aboriginal patients in collaboration with Aboriginal healers and Elders where requested by Aboriginal patients.” Elevate now is a support service in Northwestern Ontario that provides counselling and referrals regarding issues related to HIV/AIDS, hepatitis C, and harm reduction information or supplies. They are focused on prevention, community outreach, advocacy, and case management, and are committed to the TRC Calls to Action across all their work. The Indigenous healing program resulting from this study will be housed at Elevate NWO and be made available to its clients. In addition, it will also be available to clients accessing services from the Oahas, which is based within Elevate NWO. Oahas delivers culturally grounded programs and services including harm reduction to prevent the transmission of HIV and other sexually transmitted blood borne infections. They closely collaborate with Elevate NWO in research and the delivery of services. Members of both organizations are represented on our research team (CAS, LB, MY, MD, HG). Moreover, our research team included First Nations individuals as service providers, academic and community researchers. Five of the 11 co-authors are First Nations (CAS, CJM, LB, MY, MD) with one author being First Nations and French Acadian (ACB). Several of the authors (ET, CAS, CJM, LB, MY, MD, HG, MA) work in and/or lead organizations which provides health, social and cultural services to Indigenous people. Five authors are academic researchers (JCL, ET, CS, CM, ACB) and 6 are community researchers (CAS, MA, LB, MY, MD, HG). Three authors (JCL, CAS, LB) were hired to work on the research project. All authors are experienced in community-based research (CBR). Seven authors (ET, CHS, CAS, CM, LB, MD, HG) call Thunder Bay home and one co-author (MY) is the executive director of an organization that has several regional sites including Thunder Bay. ## Two-eyed seeing guiding principles and community-based research The research process for this project is guided by Etuaptmumk, or Two-Eyed Seeing, which means “learning to see from one eye with the strengths of Indigenous knowledge and ways of knowing, and from the other eye with the strengths of western knowledges and ways of knowing, and to use both these eyes together for the benefit of all” [31]. Beyond acknowledging different types of perspectives, deeper premises of Etuaptmumk involve inclusion of spiritual knowledge in human understandings of the world, co-learning processes, and considerations on the value and impact of our current actions for seven generations ahead [31–33]. Using the Two-Eyed Seeing guiding principle enables the strengths of research team members as well as study participants to be harnessed, and further reinforces the project to be shaped by diverse perspectives that include the knowledges of both Indigenous and allied stakeholders. Two-Eyed Seeing was implemented during data collection to allow a safe space for participants to speak about their lived experiences openly, without fear of providing “wrong answers” or being subject to judgement from the interviewers. To be able to see multiple perspectives, it is important to be reflexive of the diverse knowledges held in the research setting. Active reflexivity allows researchers to continually practice critical self-awareness throughout the research process, helping to enhance both the integrity of research as well as the quality of knowledge produced [34]. Further, a CBR approach was followed, which inherently values multiple ways of knowing, is collaborative, and equitably involves all research team members who choose how they would like to participate [35]. The conceptualization of the study had originated from our partner organization Elevate NWO to address the TRCC’s calls to action #22 and support the agenda of their long-time collaborator Oahas, another partner, as well as respond to the needs of their Indigenous clients. Also, various community member and knowledge user perspectives who were part of the research team contributed to conceptualizing the study. Using the CRediT methodology, we have outlined the research team members contributions as authors. Prior to implementation of any study activities, the research team consulted with relevant stakeholders including community leaders and individuals with lived experience to participate in the development of the socio-demographic questionnaire and interview guide. We hired local Indigenous persons as research staff (i.e., Elder, research assistant) who became part of the research team. Our research staff recruited study participants, led the data collection process (e.g., informed consent, administering questionnaire, interviews) and participated in data analysis to prepare this manuscript. Our student research assistant took the lead on developing the codebook from the interview transcripts. The student held research team analysis meetings to review the codebook and analyze the transcripts. They prepared the draft manuscript, circulated it for feedback and incorporated their feedback to finalize the submitted manuscript. ## Participant sampling and recruitment The study used convenience and snowball sampling to recruit potential participants. Recruitment flyers were posted, and cards were distributed by the community partners as part of the recruitment process in Thunder Bay, Ontario. Eligible participants were those who spoke and read in English, self-identified as First Nations, Métis, and/or Inuk (i.e., Indigenous), self-identified as a woman (including transgender women) and were aged 16 years or over at the time of data collection. Participants were deemed eligible if they initially endorsed that they needed or wanted to access cultural services (e.g., ceremonies, meeting with Elders) and emergency services (e.g., food banks, crisis response services) during the recruitment phase. The eligibility criteria were chosen to reflect the demographic who can theoretically access the Indigenous healing program because they qualify for services from our community partner, and as such, included those living with HIV, hepatitis C, or in need of emergency services. Thus, we selected study participants who could qualify for these services to best inform the Indigenous healing program. ## Study consent Each eligible participant provided both verbal and written informed consent prior to beginning research activities. Verbal consent was obtained using the online platform Zoom or over the phone and written consent was obtained by delivering the consent form through mail, email or during the in-person meeting. Ongoing consent is required by the research ethics board and was obtained verbally throughout the research. Our study received ethics clearance from Women’s College Hospital Research Ethics Board at Women’s College Hospital and the University of Toronto Research Ethics Board in Toronto, Ontario. ## Data collection Data collection occurred during the COVID-19 pandemic. Canadian public health guidelines were followed as well as research ethics requirements related to return to in-person research. Screening for eligible participants was done over Zoom or the phone. Following screening, participants could complete the questionnaire and interview over Zoom, the phone or in-person. In-person data collection activities took place in a well-ventilated room at the community partner site and personal protective equipment (e.g., surgical masks, face shields, and hand sanitizer) was provided. Plexiglass separated the participants from the interviewer who were sitting face-to-face. Participants were provided with an individually wrapped refreshment and a drink during the interview. ## Sociodemographic questionnaire Participants completed a 10-minute sociodemographic questionnaire, which also asked individuals to describe their housing and sleeping situations, caregiving responsibilities, ongoing access to a health service provider or site, and where they go to access different types of health services. The interviewer used a tablet pre-loaded with the questionnaire in Qualtrics™ (Provo, UT). ## Semi-structured interviews Thirty-minute interviews took place between September 2020 –June 2021. A semi-structured interview guide was used to facilitate participant interviews. The focus of these interviews was primarily how participants understood their own individual health and wellbeing, how participants stay healthy, including how they apply Indigenous cultural practices for their health, and how participants would measure if something was working for their health. Some examples of interview questions included: Participants were encouraged to share as much detail as they were willing to when discussing these themes. Discussions were audio-recorded and transcribed. ## Data analysis Data were analyzed using thematic analysis, which consisted of coding the transcripts line by line to formulate descriptive themes [36]. Each transcript was read through once and initial codes of interest were highlighted by several research team members. Once recurring themes and patterns began to emerge, a second reading of each transcript was done by the research assistant to map out how coded lines with commonalities could be grouped under the same theme. Generated codes under the same theme were then added to a codebook and defined, and direct quotes were pulled from transcripts during a third reading to be included as examples of each code. The coding process occurred through a collaborative approach, with discussions between research team members on potential themes and key points of interest. ## Inclusivity in global research Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in S1 Checklist. ## Participant characteristics Thirty-four participants responded to the demographic questionnaire. The median age of participants at interview was 39.5 years [IQR: 31–44] (S2 Table). Participant characteristics are described in Table 1. Table 1 \| Variable \| Unnamed: 1 \| N \| \| --- \| --- \| --- \| \| Age (years) \| \| \| \| \| <30 \| 5.0 \| \| \| 30–49 \| 24.0 \| \| \| ≥50 \| 5.0 \| \| Living situation \| \| \| \| \| Sleep and spend majority of time in same spaces \| 29.0 \| \| \| Sleep and spend majority of time in different spaces \| 5.0 \| \| Caregiver responsibilities1 \| \| \| \| \| Yes \| 9.0 \| \| \| No \| 24.0 \| \| \| Prefer not to answer \| 1.0 \| \| Ongoing access to a healthcare provider \| \| \| \| \| Yes2 \| 29.0 \| \| \| Doctor \| 23.0 \| \| \| Nurse \| 14.0 \| \| \| Social worker or counselor \| 15.0 \| \| \| Other3 \| 8.0 \| \| \| No \| 5.0 \| \| Ongoing access to healthcare sites \| \| \| \| \| Yes4 \| 26.0 \| \| \| Community health centre \| 17.0 \| \| \| Emergency room \| 8.0 \| \| \| Other5 \| 16.0 \| \| \| No \| 8.0 \| \| Chronic health conditions6 \| \| \| \| \| Hepatitis C \| 13.0 \| \| \| Depression \| 23.0 \| \| \| Anxiety \| 23.0 \| \| \| Other7 \| 20.0 \| ## Interviews Of the 34 respondents who were screened and completed the questionnaire, 22 participants were interviewed to reach data saturation. Eighteen participated in one-on-one interviews resulting in 271 pages of transcripts. Four participants participated in groups of two with one interviewer, resulting in 38 pages of transcripts. A total of 309 pages of transcribed audio-recordings were read through as part of the thematic analysis process. After thematic analysis was completed, 10 total subthemes were identified, and were grouped into 4 separate composite themes. These composite themes were: [1] independence and self-care; [2] external barriers to accessing services; [3] sense of community; and [4] finding comfort in the familiar. ‘ Independence and self-care’ were found to be a stand-alone theme with no subthemes grouped within it. Subthemes grouped under ‘external barriers to accessing services’ include (i) distance and transportation; and (ii) COVID-19. Subthemes grouped under ‘finding comfort in the familiar’ include (i) personal relationships; (ii) connection to land and nature; and (iii) demonstrated interest in culturally relevant activities. Finally, subthemes grouped under ‘sense of community’ include (i) knowledge and learning; and (ii) gathering. The first theme illustrates the Indigenous women’s understanding that their health status is indicative of their (in)ability to take care of themselves and make independent, healthy choices. The second theme contrasts with the first by spotlighting external factors that often act as barriers outside of Indigenous women’s control. The third and fourth theme together are sources of strength and familiarity that allow women to overcome socio-environmental barriers in accessing services by providing a sense of comfort and hopefulness. The total number of counted quotations representing each theme or subtheme are summarized in S1 Table. Theme 1: Independence and self-care. Self-reliance and independence were associated with health status by participants. Several of the Indigenous women described that they needed to hold themselves responsible for their own health, and that the onus was on themselves when it comes to maintaining and protecting their wellbeing. This belief resulted in participants vocalizing that poor health outcomes were a result of an individual’s incapacity to manage their own health. While feelings of self-blame and discouragement resulted from this belief that each individual holds responsibility over their health, participants also suggest feeling optimistic when they can listen to their minds and bodies to make health-conscious decisions. Being able to tune in to what your body and mind are telling you was described by a participant as helpful towards understanding when or how to practice self-care. The participant perspectives highlight the importance of taking an active role in one’s wellbeing. These values and beliefs prove significant in the following sections when the women discuss both the external factors and structural forces that limit their ability to freely exercise decision-making power for their wellbeing. ## Theme 2: External barriers to accessing services Although the women believed they are responsible for their own health and must act independently in achieving good health, there are several barriers that may prevent them from doing so. The basis of this theme is on external factors identified by participants that prevent or deter them from being able to readily access health or cultural services. In this context, external barriers to accessing services are socio-economic or environmental factors that are often beyond individual’s control (e.g., availability of transportation, the ongoing COVID-19 pandemic). Being able to physically attend health and cultural services in-person was considered important by participants to their overall wellbeing, particularly when these services are convenient to access. These external barriers identified by the women demonstrate that despite their independence and responsibility in managing their own health, there are other challenges present that make it more difficult to maintain good health status. Distance and transportation. The ability to both physically and financially access appropriate health services is widely acknowledged as a determinant of health [37, 38]. Areas where health services are located compared to participants’ households were identified as a barrier to physically accessing services. In conjunction, lacking methods of convenient transportation exacerbated the problem of geographically distant health services. For the women, this included lacking a car, or the added costs of purchasing bus passes every month. One of the women mention that knowing what services to access aren’t the problem; rather, being able to physically reach them is the challenging part. She mentions how walking is not a suitable option for her personally, as she has limited capacity for physical movement due to other pre-existing health issues. COVID-19. At the time the interviews were conducted, lockdowns due to the coronavirus (COVID-19) pandemic were occurring across Canada. Public health regulations such as physical and social distancing, as well as bans on gatherings, were cited by research participants as obstacles to fully engaging in social and cultural activities to stay healthy. Public health messaging at the time was also strongly recommending Indigenous individuals to halt participation in ceremonies, due to the potential for COVID-19 infection [39]. These regulations were a main source of dissension and frustration among the Indigenous women, as they described the many ways COVID-19 was impacting their ability to connect with family, friends, and other community members. Because of COVID-19 restrictions on gatherings, participants vocalized feeling a decline in their overall wellbeing not due to contracting the disease itself, but because of their inability to partake in activities for their health. These activities included going to the gym, sage picking, and moose hunting–all of which participants demonstrated having a strong desire to engage in. Similarly, transitioning to virtual methods of health care delivery (e.g., telemedicine, phone calls) was a challenge for some participants, with one particular participant mentioning that she discontinued sessions with her therapist as it felt uncomfortable and less personal to describe her problems over the phone. The impacts to social wellbeing are highlighted by a participant who describes how COVID-19 has “isolated a lot of people” and because of this physical and social isolation, “people are not getting the necessary help that they need.” ## Theme 3: Finding comfort in the familiar This theme focuses on participants describing how being in environments of familiarity or around familiar individuals elicit feelings of comfort and closeness. These settings allowed individuals to be at ease and express themselves openly, which was mentioned to be ideal for health and wellbeing. Participants drew on familiarity in their physical and social environments, using these facets of their life as sources of strength to overcome burdens on their health. Personal relationships. A commonly identified factor that acts as a facilitator for good health was the connections that each Indigenous woman had with those closest to them–friends, family, colleagues, Elders, and other confidants. Many participants described their relationships with close friends or family members as being key components to feeling or staying well, sometimes relating these emotional bonds as ways to uplift their mood or provide motivation to get through the day. Relationships were acknowledged by participants as supporting networks that can “help each other,” and are meant to benefit both parties involved by looking out for one another where necessary. One participant mentions the potential for relationships to be harmful on the emotional and mental state, saying that they only “stay around positive people if [I] can. I try to stay away from the negative people that bring me down about their problems”. The majority of participants spoke positively about the relationships that they had constructed, and described how they continue to foster and maintain these relationships within their social circle. However, it is also important to note that in the case for one participant, relationships were described as a burden to her health. The participant describes feeling uncomfortable at home due to her controlling boyfriend and having a lack of autonomy to do or say what she wants around him: Connection to land and nature. Maintaining a closeness to nature and the land was described by participants as a comforting feeling and a major contributor to overall wellbeing. In combination, engaging in recreational activities outdoors were strongly appreciated and allowed participants to feel connected to the natural world. There was an association between participants’ connection to land and nature, and their interest for culturally relevant activities. Many of the women described being able to feel closer to their culture by being outside. Land and nature can act as outlets for the women to draw strength from while partaking in cultural activities, and some of the women also felt that being in nature improved their spiritual wellbeing. One woman described being spiritually healthy means being in the bush and having the “freedom of being in the wilderness.” Another woman echoed these thoughts more broadly, saying that to be spiritually healthy, she would have to feel “connected with the world around [her].” Demonstrated interest for culturally relevant activities. Strong desires to participate in culturally appropriate and relevant activities were a focal point during the interviews. Participants mentioned various types of cultural activities and ceremonies that they were either interested in wanting to attempt for the first time or partaking in on a consistent basis. These included pow-wows, pipe ceremonies, cedar baths, ceremonial dancing, drum-making, drumming, beading, learning their native language, and singing. The women vocalized the importance of connecting to culture by engaging in these activities as a positive contribution to their emotional state and wellbeing. One participant described that when they’re feeling down or depressed, they like to smudge to help clear the emotions going through their mind. ## Theme 4: Sense of community Indigenous women vocalized the importance of feeling a sense of community through social activities or ceremonies as another source of strength. This feeling of togetherness does not necessarily have to stem from traditional ceremonies like the sweat lodge but can be from informal and casual gatherings. The sentiment of being around other relatable individuals were highlighted by several participants as comforting and facilitates a safe space to share knowledge and stories. Knowledge and learning. Sharing knowledge and lived experiences with others was one of the ways that Indigenous women described building a sense of community and the intergenerational transmission of knowledge. Passing on traditional teachings or stories was described by participants as a way of feeling healthier by connecting with others in the community. As well, the women themselves also expressed wanting to learn more about Indigenous practices and teachings from fellow community members. One woman described the importance of maintaining a cycle in knowledge sharing, mentioning that it is important to speak to Elders regularly to seek guidance and then to pass on that knowledge to others you know. When asked about what she does to feel healthy, another woman says: Gathering. The physical act of gathering around family, friends, or other familiar members in the community as a form of social connectedness was associated with good health by participants. This subtheme contrasts with the subtheme ‘personal relationships’ because in this theme, participants emphasized the importance of being in a social setting with multiple people at one time. The Indigenous women described that while they may not have close personal relationships with each individual in settings such as get-togethers or sharing groups, they felt that the aspect of togetherness was beneficial to their mental and social health. More specifically, getting involved in social gatherings allowed women to feel included and a sense of belonging to social circles that they can relate to. ## Discussion The study findings reveal that despite health and wellness being attributed to independence and self-reliance of participants, there are external factors beyond the individual that can impact their health condition–for better or worse. Thus, it is increasingly relevant for health service models to look beyond the individual, and to consider the contexts of the environment surrounding them; particularly for Indigenous Peoples, where cultural-relevance, social relationships, and connection to land are held so closely in line with wellbeing [1, 40]. The theme ‘independence and self-care’ exemplifies how Indigenous women associate being healthy with self-reliance. When asked about how they interpret meanings of health and wellbeing, participants set expectations for themselves in achieving good health and believe they must hold themselves accountable in reaching those expectations. Self-blame and having a sense of responsibility for one’s own wellbeing has been a pattern noticed in other areas of health, including chronic disease management [41–43]. This unyielding belief of independence in managing one’s own health status can be related to individualist ideologies in mainstream health services that are used to explicitly or implicitly place blame on Indigenous Peoples when poor health outcomes are observed [44–47]. Within the field of health promotion, there have been criticisms of focusing too much attention on individualistic and lifestyle factors, while not fully taking into consideration the environmental and contextual forces that may similarly influence an individual’s health journey [46–48]. Thus, the socio-ecological model has been used as an approach towards better encapsulating the interpersonal, community, and societal levels of influences on individual health [49]. For example, at the time this study was conducted, the two most prominent external factors identified as barriers included the ongoing COVID-19 pandemic and inconvenient distances or the lack of transportation to physically access services. Not only did the pandemic prevent Indigenous gatherings for social and cultural practices, it also exacerbated problems with healthcare access, particularly in contexts regarding testing and the shift to virtual modes of healthcare delivery [39]. This creates potential challenges for Indigenous individuals who may not have internet or cellular connections strong enough to support the demands of video or telephone exchanges with health service providers [39]. Although this problem specifically was not described by participants in the study, one participant mentions feeling that virtual methods of health care delivery felt less personal than in-person sessions and discontinued talking with her therapist as a result. Maintenance of therapeutic relationships and limited engagement between patient and provider have previously been cited as challenges in the shift towards virtual health care delivery methods [50–52]. Participants also cited that knowing where to access health services was not an issue, but being able to physically access them posed more of a challenge. Physical access to health services often is related to the geographical distance to services, or an individual’s ability to overcome the time and cost associated with transportation to the service [1, 37]. While physical distances and lack of transportation options were described by participants as barriers towards attending services, it is equally critical to consider the relevance of social distances between patient and provider. Reducing social distances in health care means establishing spaces that are safe, culturally-appropriate, and socially inviting [37, 53, 54]. Adopting this culturally-safe approach to providing care may encourage greater trust and willingness for Indigenous women to seek out health care workers for their health concerns, rather than avoiding health care workers under the fear of being discriminated against [39, 37, 53–55]. These barriers signify the relevance of considering external or environmental factors beyond the individual level as potential influences on poor health outcomes. Despite the strengths of the socio-ecological model in acknowledging the interactions across individual, interpersonal, and societal levels, it does not fully apply to Indigenous communities due to the lack of acknowledging connections to culture, spirit, and land. As such, we propose referencing the First Nation Health Authority’s (FNHA) visual depiction of First Nations Perspective on Health and Wellness [40] as a more appropriate and relevant model in discussions on Indigenous wellbeing. The First Nations Perspective on Health and Wellness model includes considerations of environmental, cultural, economic, and social aspects of an individual in the most outer ring [40]. Other important determinants of health including knowledge, family, and land are included in the model as well [40], and are able to clearly represent the range of factors that influence health status for Indigenous Peoples compared to the standard socio-ecological model. This is reflected in this study’s findings, as the women described the importance of having supports from the wider community and natural world for achieving and sustaining good health. In these instances, participants perceive health holistically and beyond the individual level–bringing up the benefits of attending cultural ceremonies, maintaining strong relationships with loved ones, and having a connection to the land. Holding a close connection to land and the natural environment has been recognized as a significant determinant of Indigenous peoples’ wellbeing, given how important land is in supporting physical, mental, emotional, and spiritual health [27–29, 56, 57]. Communities, as well, have long been understood as a part of Indigenous identities and can play an integral role in healing processes and overall wellbeing [58]. Within our study, participants described how they look out for others and establish security by socially confiding with each other. Social support gained from close relationships or community bonds have been understood to be as important as other more established protective factors for health [26, 59, 60]. Not only are communities able to act as networks for social supports and relevance through shared experiences, they are also sources of strength for Indigenous women to draw on [58]. Individuals can draw strength from their communities through feeling like they belong to a wider collective, but also because engaging within communities is critical to expressing Indigenous identity [58]. In connection to social gatherings, participants described a longing for culture and the importance of cultural activities in promoting mental, emotional, and spiritual health. Cultural activities appeared to act as a facilitator for social gatherings to occur. Social support and social capital have been interlinked with reinforcing cultural identity for Indigenous peoples [25, 61] and participants vocalized this by recounting the opportunities to build their social networks during or after cultural ceremonies and activities. This reciprocity in creating community bonds allows for cyclical and sustained knowledge sharing, which serves to further compound the long-term benefits of both Indigenous individuals and communities’ health [62, 63]. Feeling a sense of community was mentioned as important to their overall wellbeing, and the participants described instances when they would act as both contributors and beneficiaries to creating a sense of community. For example, participants would act as contributors by passing on knowledge to younger generations within the community, and as beneficiaries by attending cultural ceremonies guided by Elders. Such is true that the transmission of traditional knowledge amongst Indigenous communities can occur in both formal and informal contexts, including social encounters, ceremonial practices or other activities within the community [64]. These descriptions by participants might demonstrate that while societal understandings of health are from an individual standpoint, foundational underpinnings of Indigenous wellbeing are constructed through social bonds and cultural continuity, aspects which are not typically considered in current models of healthy provisioning that are accessed by Indigenous women [3, 26, 65]. ## Implications for future research Referencing the FNHA Health and Wellness model while planning culturally relevant health services or an Indigenous healing program is beneficial for prospective service users as it describes determinants of health that exist across different levels of society for every individual. The ability for Indigenous healing programs to take into account the racist, sexist, and colonialist mindsets of society will be crucial towards program success, particularly in the case of Indigenous women, where intersectionality may result in disproportionate burdens of poor health being experienced [66]. This, in turn, may also help to deepen understandings of the socio-environmental complexities impacting health beyond the individual’s control that equally should be addressed in Indigenous women’s circle of care. Continuing to shift the paradigm from individualistic ideologies of healthcare to a more holistic mindset (as presented by the FNHA model) can be significant in changing the ways which Indigenous women perceive self-reliance as the key to health and wellbeing. Further research should investigate how an Indigenous healing program with health and cultural services can benefit the holistic wellbeing of Indigenous women, as well as how it can encourage more women to seek out supports within the community rather than relying on themselves to care for their body, mind, and soul. ## Limitations Our study employed convenience sampling as the main recruitment strategy, which limits the population reached, as all recruited participants had already been accessing services offered at Elevate NWO and harm reduction sites. Thus, suggestions made by participants on health and cultural services that they would want or need within the Indigenous healing program are not necessarily indicative of the preferences of those who did not previously access the same harm reduction sites or services. This makes it challenging to create a comprehensive healing program offering services that would be culturally relevant and appropriate for a diverse group of Indigenous individuals. ## Conclusion Considering the unique needs that Indigenous women require in contrast to standard, westernized health services will be critical in ensuring successful healing programs are developed and implemented. Despite the importance of accessing Indigenous healing approaches for Indigenous health being deeply understood and substantially cited by the literature over the past decades [15, 67], Statistics Canada reported that only roughly one third of Indigenous Peoples living in urban areas have access to traditional medicines and healing practices [68]. Aspects of cultural continuity, self-determination, and knowledge transmission have been identified as critical health determinants, yet they remain excluded from major health care reform initiatives in Canada [69]. Allowing Indigenous control and autonomy over the Healing Program will help to establish cultural safety in Indigenous women’s circle of care. Successful Indigenous health strategies must be grounded in factors such as self-determination in order to ensure responsiveness to the unique needs of every community and embed cultural continuity in models of Indigenous health care provisioning [69]. Furthermore, we are hopeful that this Indigenous healing program will support women towards perceiving the importance of social and cultural determinants of health, and that independence and self-care are only one piece towards achieving holistic wellbeing. As such, the Healing Program resulting from this study will be able to act as a facilitator for Indigenous women in accessing culturally relevant health and social services within their community. ## References 1. Reading C, Wien F. Health inequalities and social determinants of Aboriginal Peoples’ Health. 1-36 2. Chandler MJ, Lalonde C. Cultural Continuity as a Hedge against Suicide in Canada’s First Nations.. Transcult Psychiatry. (1998.0) 35 191-219. DOI: 10.1177/136346159803500202 3. Auger M.. Cultural Continuity as a Determinant of Indigenous Peoples’ Health: A Metasynthesis of Qualitative Research in Canada and the United States.. International Indigenous Policy Journal (2016.0) 7. DOI: 10.18584/iipj.2016.7.4.3 4. Krieg B.. Understanding the Role of Cultural Continuity in Reclaiming the Identity of Young Indigenous Women.. Girlhood Studies. (2016.0) 9. DOI: 10.3167/ghs.2016.090204 5. Brunanski DM. Counselling with Aboriginal street youth: an aboriginalizing narrative inquiry.. University of British Columbia. (2009.0). DOI: 10.14288/1.0053867 6. Lucero NM. Making Meaning of Urban American Indian Identity: A Multistage Integrative Process.. Social Work. (2010.0) 55 327-336. DOI: 10.1093/sw/55.4.327 7. Smith D, Varcoe C, Edwards N. Turning around the intergenerational impact of residential schools on Aboriginal people: Implications for health policy and practice.. The Canadian journal of nursing research = Revue canadienne de recherche en sciences infirmières (2006.0) 37 38-60 8. Wexler L.. Looking across three generations of Alaska Natives to explore how culture fosters indigenous resilience.. Transcult Psychiatry (2014.0) 51 73-92. DOI: 10.1177/1363461513497417 9. Robbins JA, Dewar J. Traditional Indigenous Approaches to Healing and the modern welfare of Traditional Knowledge, Spirituality and Lands: A critical reflection on practices and policies taken from the Canadian Indigenous Example.. The International Indigenous Policy Journal (2011.0) 2. DOI: 10.18584/iipj.2011.2.4.2 10. Redvers N, Blondin B. Traditional Indigenous medicine in North America: A scoping review.. PLOS ONE. (2020.0) 15 e0237531. DOI: 10.1371/journal.pone.0237531 11. Waldram JB. Transformative and Restorative Processes: Revisiting the Question of Efficacy of Indigenous Healing.. Medical Anthropology. (2013.0) 32 191-207. DOI: 10.1080/01459740.2012.714822 12. Redvers J.. “The land is a healer”: Perspectives on land-based healing from Indigenous practitioners in northern Canada.. International Journal of Indigenous Health (2020.0) 15 90-107. DOI: 10.32799/ijih.v15i1.34046 13. Barwin L, Crighton E, Shawande M, Veronis L. Teachings around self-care and medicine gathering in Manitoulin Island, Ontario: Rebuilding capacity begins with youth.. Pimatisiwin: A Journal of Aboriginal and Indigenous Community Health. (2013.0) 11 323-344 14. Rowan M, Poole N, Shea B, Gone JP, Mykota D, Farag M. Cultural interventions to treat addictions in Indigenous populations: findings from a scoping study.. Subst Abuse Treat Prev Policy (2014.0) 9 34. DOI: 10.1186/1747-597X-9-34 15. Auger M, Howell T, Gomes T. Moving toward holistic wellness, empowerment and self-determination for Indigenous peoples in Canada: Can traditional Indigenous health care practices increase ownership over health and health care decisions?. Can J Public Health (2016.0) 107 e393-e398. DOI: 10.17269/cjph.107.5366 16. Aspin C, Brown N, Jowsey T, Yen L, Leeder S. Strategic approaches to enhanced health service delivery for Aboriginal and Torres Strait Islander people with chronic illness: a qualitative study.. BMC Health Serv Res (2012.0) 12 143. DOI: 10.1186/1472-6963-12-143 17. Anaya SJ. Peoples UHRCSR on R of I. Report of the Special Rapporteur on the rights of indigenous peoples. (2012.0) 18. Senese LC, Wilson K. Aboriginal urbanization and rights in Canada: examining implications for health.. Soc Sci Med (2013.0) 91 219-228. DOI: 10.1016/j.socscimed.2013.02.016 19. Haskell DL, Randall M. Disrupted Attachments: A Social Context Complex Trauma Framework and the Lives of Aboriginal Peoples in Canada.. International Journal of Indigenous Health (2009.0) 5 48-99 20. Reibel T, Morrison L, Griffin D, Chapman L, Woods H. Young Aboriginal women’s voices on pregnancy care: factors encouraging antenatal engagement.. Women Birth. (2015.0) 28 47-53. DOI: 10.1016/j.wombi.2014.10.003 21. Hayman N.. Strategies to improve indigenous access for urban and regional populations to health services. Heart Lung Circ (2010.0) 19 367-371. DOI: 10.1016/j.hlc.2010.02.014 22. Khoury P.. Beyond the Biomedical Paradigm: The Formation and Development of Indigenous Community-Controlled Health Organizations in Australia.. Int J Health Serv (2015.0) 45 471-494. DOI: 10.1177/0020731415584557 23. Walls M, Hautala D, Cole A, Kosobuski L, Weiss N, Hill K. Socio-cultural integration and holistic health among Indigenous young adults.. BMC Public Health (2022.0) 22 1002. DOI: 10.1186/s12889-022-13395-3 24. Ullrich J.. For the love of our children: an Indigenous connectedness framework.. AlterNative: An International Journal of Indigenous Peoples. (2019.0) 15 25. Yeung S, Rosenberg M, Banach D, Mayotte L, Anand SS. Bonding social capital and health within four First Nations communities in Canada: A cross-sectional study.. SSM—Population Health. (2021.0) 16 100962. DOI: 10.1016/j.ssmph.2021.100962 26. Richmond CAM, Ross NA, Egeland GM. Social Support and Thriving Health: A New Approach to Understanding the Health of Indigenous Canadians.. Am J Public Health. (2007.0) 97 1827-1833. DOI: 10.2105/AJPH.2006.096917 27. Lines L-A, Jardine CG. Connection to the land as a youth-identified social determinant of Indigenous Peoples’ health.. BMC Public Health. (2019.0) 19 176. DOI: 10.1186/s12889-018-6383-8 28. Hatala AR, Njeze C, Morton D, Pearl T, Bird-Naytowhow K. Land and nature as sources of health and resilience among Indigenous youth in an urban Canadian context: a photovoice exploration. BMC Public Health (2020.0) 20 538. DOI: 10.1186/s12889-020-08647-z 29. Richmond CAM, Ross NA. The determinants of First Nation and Inuit health: a critical population health approach.. Health Place (2009.0) 15 403-411. DOI: 10.1016/j.healthplace.2008.07.004 30. 30Truth and Reconciliation Commission of Canada. Honouring the truth, reconciling for the future: summary of the final report of the Truth and Reconciliation Commission of Canada. 2015. Available: https://irsi.ubc.ca/sites/default/files/inline-files/Executive_Summary_English_Web.pdf. Honouring the truth, reconciling for the future: summary of the final report of the Truth and Reconciliation Commission of Canada. (2015.0) 31. Bartlett C, Marshall M, Marshall A. Two-Eyed Seeing and other lessons learned within a co-learning journey of bringing together indigenous and mainstream knowledges and ways of knowing.. J Environ Stud Sci (2012.0) 2 331-340. DOI: 10.1007/s13412-012-0086-8 32. Roher SIG, Yu Z, Martin DH, Benoit AC. How is Etuaptmumk/Two-Eyed Seeing characterized in Indigenous health research? A scoping review.. PLOS ONE. (2021.0) 16 e0254612. DOI: 10.1371/journal.pone.0254612 33. Bartlett C, Marshall M, Marshall A. Integrative Science and Two-Eyed Seeing: Enriching the Discussion Framework for Healthy Communities.. Ecosystems, Society and Health: Pathways through Diversity, Convergence and Integration. (2015.0) 280-326 34. Probst B.. The Eye Regards Itself: Benefits and Challenges of Reflexivity in Qualitative Social Work Research.. Social Work Research. (2015.0) 39 37-48. DOI: 10.1093/swr/svu028 35. Israel BA, Schulz AJ, Parker EA, Becker AB. REVIEW OF COMMUNITY-BASED RESEARCH: Assessing Partnership Approaches to Improve Public Health.. Annual Review of Public Health. (1998.0) 19 173-202. DOI: 10.1146/annurev.publhealth.19.1.173 36. Braun V, Clarke V. Using thematic analysis in psychology.. Qualitative Research in Psychology (2006.0) 3 77-101. DOI: 10.1191/1478088706qp063oa 37. Horrill T, McMillan DE, Schultz ASH, Thompson G. Understanding access to healthcare among Indigenous peoples: A comparative analysis of biomedical and postcolonial perspectives.. Nurs Inq (2018.0) 25 e12237. DOI: 10.1111/nin.12237 38. 38Commission on Social Determinants of Health. Closing the gap in a generation: health equity through action on the social determinants of health—Final report of the commission on social determinants of health. 2008. Available: https://www.who.int/publications/i/item/WHO-IER-CSDH-08.1. Closing the gap in a generation: health equity through action on the social determinants of health—Final report of the commission on social determinants of health. (2008.0) 39. Mashford-Pringle A, Skura C, Stutz S, Yohathasan T. What we heard: Indigenous Peoples and COVID-19 (2021.0) 32 40. 40First Nations Health Authority. First Nations Perspective on Health and Wellness. [cited 5 Jul 2022]. Available: https://www.fnha.ca:443/wellness/wellness-for-first-nations/first-nations-perspective-on-health-and-wellness. First Nations Perspective on Health and Wellness. 41. Beverly EA, Ritholz MD, Brooks KM, Hultgren BA, Lee Y, Abrahamson MJ. A Qualitative Study of Perceived Responsibility and Self-Blame in Type 2 Diabetes: Reflections of Physicians and Patients.. J Gen Intern Med (2012.0) 27 1180-1187. DOI: 10.1007/s11606-012-2070-0 42. Eccleston C, De C, Williams AC, Rogers WS. Patients’ and professionals’ understandings of the causes of chronic pain: Blame, responsibility and identity protection.. Social Science & Medicine. (1997.0) 45 699-709. DOI: 10.1016/S0277-9536(96)00404-2 43. Tuncay T, Musabak I, Gok DE, Kutlu M. The relationship between anxiety, coping strategies and characteristics of patients with diabetes. Health and Quality of Life Outcomes (2008.0) 6 79. DOI: 10.1186/1477-7525-6-79 44. Browne AJ, Varcoe C, Lavoie J, Smye V, Wong ST, Krause M. Enhancing health care equity with Indigenous populations: evidence-based strategies from an ethnographic study.. BMC Health Services Research (2016.0) 16 544. DOI: 10.1186/s12913-016-1707-9 45. Wylie L, McConkey S. Insiders’ Insight: Discrimination against Indigenous Peoples through the Eyes of Health Care Professionals.. J Racial and Ethnic Health Disparities (2019.0) 6 37-45. DOI: 10.1007/s40615-018-0495-9 46. Hook CJ, Rose Markus H. Health in the United States: Are Appeals to Choice and Personal Responsibility Making Americans Sick?. Perspect Psychol Sci. (2020.0) 15 643-664. DOI: 10.1177/1745691619896252 47. McClean S.. ‘The illness is part of the person’: discourses of blame, individual responsibility and individuation at a centre for spiritual healing in the North of England.. Sociology of Health & Illness. (2005.0) 27 628-648. DOI: 10.1111/j.1467-9566.2005.00459.x 48. Golden SD, Earp JAL. Social Ecological Approaches to Individuals and Their Contexts: Twenty Years of Health Education & Behavior Health Promotion Interventions.. Health Educ Behav. (2012.0) 39 364-372. DOI: 10.1177/1090198111418634 49. Stokols D.. Translating Social Ecological Theory Into Guidelines for Community Health Promotion. American journal of health promotion. AJHP (1996.0) 10 282-98. DOI: 10.4278/0890-1171-10.4.282 50. Breton M, Sullivan EE, Deville-Stoetzel N, McKinstry D, DePuccio M, Sriharan A. Telehealth challenges during COVID-19 as reported by primary healthcare physicians in Quebec and Massachusetts.. BMC Family Practice (2021.0) 22 192. DOI: 10.1186/s12875-021-01543-4 51. Grady CB, Claus EB, Bunn DA, Pagliaro JA, Lichtman JH, Bhatt AB. Disparities in patient engagement with video telemedicine among high-video-use providers during the COVID-19 pandemic. European Heart Journal—Digital Health (2021.0) 2 691-694. DOI: 10.1093/ehjdh/ztab067 52. Wijesooriya NR, Mishra V, Brand PLP, Rubin BK. COVID-19 and telehealth, education, and research adaptations.. Paediatr Respir Rev (2020.0) 35 38-42. DOI: 10.1016/j.prrv.2020.06.009 53. Browne AJ, Smye VL, Rodney P, Tang SY, Mussell B, O’Neil J. Access to primary care from the perspective of Aboriginal patients at an urban emergency department.. Qual Health Res (2011.0) 21 333-348. DOI: 10.1177/1049732310385824 54. Cameron BL, Carmargo Plazas MDP, Salas AS, Bourque Bearskin RL, Hungler K. Understanding inequalities in access to health care services for aboriginal people: a call for nursing action.. ANS Adv Nurs Sci (2014.0) 37 E1-E16. DOI: 10.1097/ANS.0000000000000039 55. Churchill ME, Smylie JK, Wolfe SH, Bourgeois C, Moeller H, Firestone M. Conceptualising cultural safety at an Indigenous-focused midwifery practice in Toronto, Canada: qualitative interviews with Indigenous and non-Indigenous clients. BMJ Open (2020.0) 10 e038168. DOI: 10.1136/bmjopen-2020-038168 56. Durkalec A, Furgal C, Skinner MW, Sheldon T. Climate change influences on environment as a determinant of Indigenous health: Relationships to place, sea ice, and health in an Inuit community.. Soc Sci Med. (2015.0) 136-137. DOI: 10.1016/j.socscimed.2015.04.026 57. Toombs E, Lund J, Mushquash AR, Mushquash CJ. Predictors of Land-Based Activity Participation in a National Representative Sample of Indigenous Individuals Living Off-Reserve.. Int J Environ Res Public Health (2022.0) 19 8029. DOI: 10.3390/ijerph19138029 58. Maracle G. Connections and Processes: Indigenous Community and Identity’s place in the Healing Journey.. Turtle Island Journal of Indigenous Health (2021.0) 1. DOI: 10.33137/tijih.v1i2.36052 59. Harandi TF, Taghinasab MM, Nayeri TD. The correlation of social support with mental health: A meta-analysis.. Electron Physician. (2017.0) 9 5212-5222. DOI: 10.19082/5212 60. Ozbay F, Johnson DC, Dimoulas E, Morgan CA, Charney D, Southwick S. Social Support and Resilience to Stress.. Psychiatry (Edgmont). (2007.0) 4 35-40 61. Berry HL. Social capital and mental health among Aboriginal Australians, New Australians and Other Australians living in a coastal region.. Australian e-Journal for the Advancement of Mental Health. (2009.0) 8 142-154. DOI: 10.5172/jamh.8.2.142 62. Smith L.. Decolonizing methodologies: Research and Indigenous Peoples (1999.0) 63. Rowe G, Straka S, Hart M, Callahan A, Robinson D, Robson G. Prioritizing Indigenous Elders’ Knowledge for Intergenerational Well-being. Can J Aging (2020.0) 39 156-168. DOI: 10.1017/S0714980819000631 64. Bruchac M.. Indigenous Knowledge and Traditional Knowledge (2014.0) 65. Svenson K, Lafontaine C. The search for wellness. First Nations and Inuit Regional Health Survey, National Report. (1999.0) 181-216 66. Bourassa C, McKay-McNabb K, Hampton M. Racism, Sexism and Colonialism: The Impact on the Health of Aboriginal Women in Canada.. Canadian Woman Studies. (2005.0) 67. 67Environics Institute. The Urban Aboriginal Peoples Study. Toronto, ON; 2010. Available: https://www.uaps.ca/wp-content/uploads/2010/04/UAPS-FULL-REPORT.pdf 68. 68Statistics Canada. Aboriginal peoples survey 2001: initial findings: well being of the non-reserve Aboriginal population. Ottawa: Statistics Canada; 2003. Available: https://www150.statcan.gc.ca/n1/pub/89-589-x/89-589-x2003001-eng.pdf. Aboriginal peoples survey 2001: initial findings: well being of the non-reserve Aboriginal population. (2003.0) 69. Katz A, Enns J, Kinew KA. Canada needs a holistic First Nations health strategy.. CMAJ (2017.0) 189 E1006-E1007. DOI: 10.1503/cmaj.170261
--- title: 'Impact of digital meditation on work stress and health outcomes among adults with overweight: A randomized controlled trial' authors: - Rachel M. Radin - Elissa S. Epel - Ashley E. Mason - Julie Vaccaro - Elena Fromer - Joanna Guan - Aric A. Prather journal: PLOS ONE year: 2023 pmcid: PMC9977041 doi: 10.1371/journal.pone.0280808 license: CC BY 4.0 --- # Impact of digital meditation on work stress and health outcomes among adults with overweight: A randomized controlled trial ## Abstract Mindfulness meditation may improve well-being at work; however, effects on food cravings and metabolic health are not well known. We tested effects of digital meditation, alone or in combination with a healthy eating program, on perceived stress, cravings, and adiposity. We randomized 161 participants with overweight and moderate stress to digital meditation (‘MED,’ $$n = 38$$), digital meditation + healthy eating (‘MED+HE,’ $$n = 40$$), active control (‘HE,’ $$n = 41$$), or waitlist control (‘WL,’ $$n = 42$$) for 8 weeks. Participants ($$n = 145$$; M(SD) BMI: 30.8 (5.4) kg/m2) completed baseline and 8-week measures of stress (Perceived Stress Scale), cravings (Food Acceptance and Awareness Questionnaire) and adiposity (sagittal diameter and BMI). ANCOVAs revealed that those randomized to MED or MED+HE (vs. HE or WL) showed decreases in perceived stress ($F = 15.19$, $p \leq .001$, η2 =.10) and sagittal diameter ($F = 4.59$, $$p \leq .03$$, η2 =.04), with no differences in cravings or BMI. Those high in binge eating who received MED or MED+HE showed decreases in sagittal diameter ($$p \leq .03$$). Those with greater adherence to MED or MED+HE had greater reductions in stress, cravings, and adiposity (ps <.05). A brief digital mindfulness-based program is a low-cost method for reducing perceptions of stress and improving abdominal fat distribution patterns among adults with overweight and moderate stress. Future work should seek to clarify mechanisms by which such interventions contribute to improvements in health. Trial registration: Clinical trial registration http://www.ClinicalTrials.gov: identifier NCT03945214. ## Introduction Obesity remains a public health crisis [1, 2] and it is highly comorbid with work-related stress [3]. Work stress contributes to an estimated 5–$8\%$ of annual healthcare costs in the United States [4]. Epidemiological studies consistently demonstrate associations between high work stress and worse self-reported mental and physical health, including depression, anxiety, cardiovascular disease, and type 2 diabetes [5]. Mindfulness meditation may improve well-being in workplace settings [6]. Mindfulness, in general, aims to cultivate a non-judging awareness of experiences in the present moment and promote adaptive self-regulation [7]. Mindfulness-based psychological interventions decrease perceptions of stress in non-clinical populations [8], and improve psychosocial outcomes in clinical populations with anxiety and depression [9–11]. Recent data indicate that mindfulness-based trainings delivered in the workplace decrease global perceptions of psychological stress in healthy adults [12]. However, traditional in-person practice cannot be easily scaled and disseminated, making them less cost-effective than other approaches. In the current study, we used a commercially available digitally-delivered meditation platform. Overeating drive patterns, such as food cravings and binge eating, may explain the links between work stress and obesity. These eating patterns are strongly associated with obesity [13–16] and worsened metabolic health [17] affecting up to $30\%$ of those who seek weight-loss treatment [18–20]. Overeating drive may uniquely predict the development of cardiovascular and endocrine disorders, including heart disease and type 2 diabetes, even after accounting for obesity status [17, 21]. These data support the importance of overeating drive as a behavioral target. Mindfulness-based approaches may be a promising avenue for targeting reductions in overeating drive, including food cravings, and downstream metabolic outcomes. Mindfulness-based approaches are not diet-based, thus appealing to those with overeating drive patterns, who may have had many unsuccessful dieting attempts. There is little data assessing whether mindfulness approaches promote improvements in metabolic outcomes [22, 23]. It also remains unclear for whom mindfulness-based approaches are best suited. Our prior work on a weight loss intervention demonstrated that those with a tendency toward binge eating showed greater improvements in a range of weight-related factors following a mindfulness intervention compared to those without binge eating [24]. Thus, mindfulness-based approaches may be a better fit for adults with obesity and overeating drive, in comparison to standard behavioral weight loss interventions. Mindfulness training delivered via a self-guided smartphone app may offer a convenient alternative to in-person treatment, though research on their efficacy is limited [25, 26]. Three small studies using smartphone apps to deliver mindfulness interventions to healthy adults found benefits comparable to traditional delivery methods on subjective well-being, depressive symptoms, and compassion [27–29]. App-based interventions also offer the benefit of standardization of instruction across participants, as well as the ability for participants to control where and when they access the intervention, and objective measures of adherence, rather than self-report. Digital mindfulness interventions demonstrate significant reductions in perceived stress and increases in subjective mindfulness, compared to a wait list condition, among non-clinical populations [30]. A recent meta-analysis of digital occupational mental health interventions [31], which included mindfulness-based programs, found small, positive effects on psychological well-being and work effectiveness. Treatment adherence to digitally-based mindfulness interventions is an understudied, yet probable moderator of treatment effects. In a recent 8-week pilot, Carolan and colleagues [32] found greater treatment engagement in digital programs incorporating a discussion group. It is also unknown whether digitally-based mindfulness interventions improve overeating drive or metabolic health. A recent meta-analysis [33] of in-person work-based mindfulness meditation programs found them generally effective in lowering cortisol production, heart rate, and sympathetic activity. Previous work using in-person mindfulness has shown that mindful eating training reduces abdominal fat without reducing overall body mass index [34, 35]. The Current Study: To test the effects of digital meditation on stress, cravings, and abdominal adiposity, we tested 4 conditions including an active control group with information about healthy eating, and a no treatment wait list control. The healthy eating program, which we considered to be an “active control” condition, utilized mindfulness-based and motivational approaches to improve eating behaviors. We aimed to test whether digital meditation could out-perform an active control condition that was matched for time and attention and other non-specific intervention effects [36]. We were interested in whether the adjunctive treatment offered from the healthy eating program would further improve outcomes compared to digital mindfulness alone. We aimed to examine the effects of treatment randomization on global perceptions of psychological distress [37] and overeating drive [38]. Secondarily, we examined treatment effects on body mass index (BMI) and sagittal diameter. Finally, we examined the influence of treatment adherence (total minutes participants engaged in meditation on the app) on treatment outcomes. We hypothesized that mindfulness (in either form) would out-perform either control condition with respect to improvements in primary and secondary outcomes, and that treatment adherence would moderate these effects. We also anticipated that the combination of digital mindfulness + healthy eating (vs. digital mindfulness alone) would promote the greatest improvements. We also endeavored to examine the potential moderating role of binge eating presence. We hypothesized that those with binge eating would derive the greatest benefit from a mindfulness-based digital intervention, whereas those without binge eating would show no differences in outcomes across interventions. ## Study overview We aimed to test the effects of a digital meditation intervention vs. an active or wait list control, on subjective measures of perceived stress, food cravings, and adiposity in a sample of employees at a large university with overweight and obesity who reported mild to moderate stress (NCT03945214). We randomized participants to 8-weeks of a digital meditation intervention (using the commercially available application, Headspace), a healthy eating intervention (active control), a digital meditation + healthy eating intervention, or a waitlist control condition. We asked all participants to complete questionnaires and anthropomorphic measurements at an in-person clinic visit at baseline and week 8. Adherence to the digital meditation intervention was tracked remotely by Headspace. ## Participants Eligible participants were ≥18 years old, employed at a large academic medical center, had a BMI equal to or greater than 25 kg/m2, reported mild to moderate levels of stress in the previous month (as determined by a Perceived Stress Scale score of 15 or higher), and had daily access to a smartphone or computer. Exclusion criteria included being an experienced meditator (defined as 3 times per week for 10 minutes or more). We obtained written informed consent from all study participants. We aimed to enroll up to 150 participants. Our prior study [39] detected effects in a sample of <250 participants. We therefore expected that our sample size of 150 would be well-powered to detect improvements in our self-report measures in response to our treatment intervention. The university’s Institutional review board (IRB) approved all aspects of this study. Participants did not receive monetary compensation; however they received a one-year subscription to Headspace (value of $150), and were entered into a raffle drawing to win a 2-night expenses paid vacation in the local Bay Area. ## Study design Participants completed baseline assessment procedures, including measures of body composition and self-report assessments. Study personnel then randomly assigned participants to one of four possible conditions, using factorial assignment, on Qualtrics: [1] Meditation only, [2] Healthy Eating only, [3] Meditation + Healthy Eating, or [4] Waitlist control. The sequence of assignments was generated ahead of time with a computer script by a statistician who was not involved in running the study. Study personnel were not able to access the file containing the sequence of assignments or to see the next condition in the sequence until the moment they randomized the participant. Both participants and study staff were unblinded to the assignment after allocation. We re-assessed participants on all measures collected at baseline (body composition, self-report assessments) again at 8 weeks from randomization. ## Meditation group (‘MED’) We provided participants with access to digitally-based meditation program (Headspace app- Basics + ‘Letting go of stress’ packs) and asked them to engage with the app for at least 10 minutes a day for 8 weeks. We contacted participants who completed less than one meditation in the previous 10 to 17 days via phone, in order to re-engage them with the program. Participants were expected to meditate 5 days per week over the course of 8 weeks. ## Healthy eating group (‘HE’) Within week 1, we provided participants with an in-person 50-minute counseling session with a trained health counselor geared towards developing goals to improve eating behavior, along with three 10-minute booster phone calls at weeks 1, 4, and 8. The counseling session incorporated a motivational interviewing framework to assess areas of concern around eating behavior and to establish specific and achievable eating-related goals. We also asked participants to engage with a digitally-based mindful eating program once per week for 8 weeks, and sent text message reminders 3 times per week to increase accountability towards eating-related goals. The digitally-based mindful eating program was created specifically for this study by the research team, and was primarily a secured website that included information on mindful eating, and audio tools for mindful eating practice (~3–5 minute practices). This password-secured website contained up to six different brief audio exercises using mindful eating and urge-surfing strategies. The audio exercises were scripted and recorded by the study’s first author and adapted from a number of mindful eating resources including the mindfulness-based eating awareness training (MB-EAT) curriculum [45]. We instructed participants to access these audios during high vulnerability times for compulsive eating. For example, for those who identify cravings as a potent trigger for problematic consumption, participants could access a brief urge-surfing exercise to learn how to ‘ride out’ a craving. Participants had a total of approximately 1.5–2 hours of contact with a counselor, and were expected to engage with the online resources 1 day per week over the course of 8 weeks. The program is adapted from several sources, including motivational interviewing for binge eating, weight management, and sugar-sweetened beverage intake (from our recently completed trial), and mindfulness-based eating awareness training. ## Meditation + healthy eating group (‘MED+HE’) We provided participants with access to digitally-based meditation program (as described above under ‘MED’) in addition to the ‘HE’ program (as described above under ‘HE’). Participants were expected to meditate 5 days per week over the course of 8 weeks and they had a total of approximately 1.5–2 hours of contact with a counselor, and were expected to engage with the online resources 1 day per week over the course of 8 weeks. ## Waitlist control condition (‘WL’) We instructed participants to continue their normal activities and not add any meditation during the study period. We did not provide Headspace access codes to WL or HE participants until after they completed a 2-month follow-up questionnaire. Participants had no contact with a study counselor over the course of the 8 week intervention period. ## Primary outcome measures Perceived stress. The Perceived Stress Scale (PSS; [37] is a 10-item self-report questionnaire that measures a persons’ evaluation of the life stress they have experienced over the previous month, and has been extensively validated. The PSS has a total score scale range of 0 to 40, with higher values indicating more perceived stress. The PSS has demonstrated adequate reliability and validity among similar populations [37]. Among our study sample, scale reliability was high (α =.87) ## Tolerance for food cravings The Food Acceptance and Awareness Questionnaire (FAAQ) measures acceptance of urges and cravings to eat or the extent to which individuals might try to control or change these thoughts [38]. The FAAQ is made up of 10 items, each rated on a 6-point Likert scale (1 = very seldom true to 6 = always true). It has a total score scale range of 10 to 60, with higher scores indicating greater acceptance of motivations to eat and greater tolerance for food cravings. The FAAQ has demonstrated sound psychometric properties [38]. Among our study sample, scale reliability was high (α =.80). Among all 4 groups, we found no treatment effect (F[3,132] = 0.58, $$p \leq .63$$, η2 =.01). Comparing the estimated marginal means showed a pattern (while not significant) that those in HE showed the greatest increases in FAAQ (mean change: +1.80, SE = 1.37, $95\%$ CI: -0.91, 4.51), followed by those in WL (mean change: +0.81, SE = 1.33, $95\%$ CI: -1.83, 3.45), MED (mean change: +0.26, SE = 1.37, $95\%$ CI: -2.46, 2.97) and MED+HE (mean change: -0.83, SE = 1.51, $95\%$ CI: -3.81, 2.15; Fig 3). In sub-analyses, those randomized to ‘meditation’ vs. ‘no meditation’ did not differ; Both groups showed similar changes (meditation: $0.10\%$ reduction; vs no meditation: $4.3\%$ increase; F(1,134 = 1.21, $$p \leq .27$$, η2 =.01). Findings were identical when using non-parametric tests (Kruskal-Wallis), given the ordinal nature of the FAAQ scoring. Frequency of meditation did not moderate the effect of treatment on changes in FAAQ ($p \leq .10$). However, we observed a main effect of meditation frequency on FAAQ at 8-weeks ($F = 5.31$, $$p \leq .02$$), irrespective of treatment randomization. Treatment adherence was associated with higher FAAQ scores at 8-weeks ($r = .27$, $$p \leq .03$$), although it was not associated with changes in FAAQ score at 8 weeks ($r = .20$, $$p \leq .12$$). Fig 3: Effect of treatment randomization on Tolerance for Food Cravings (FAAQ) at 8 weeks, accounting for baseline values. ## Secondary outcome measures BMI. We calculated body mass index (BMI) as weight in kilograms divided by the square of height in meters (kg/m2).Weight was measured twice using a digital scale, and height was measured using a stadiometer. Sagittal diameter. We measured body fat distribution using an abdominal caliper placed just above the umbilicus, measuring the distance from the small of the back to the upper abdomen. Measurements were taken, using the two closest measurements that were within 0.5 cm, and recorded to the nearest 0.1 cm. Binge presence. We used the Questionnaire on Eating and Weight Patterns –5 (QEWP-5) to determine the presence of binge eating. The QEWP-5 [40] is a 24-item questionnaire that assesses frequency of reported binge eating and loss of control eating episodes, which has been shown to have reasonable agreement with interview-based measures such as the Eating Disorder Examination [40]. Binge presence was defined by the endorsement of the following: 1- During the last 3 months, did you ever eat, in a short period of time- for example, a two hour period- what most people would think was an unusually large amount of food?; 2- During the times when you ate an unusually large amount of food, did you often feel you could not stop eating or control what or how much you were eating? ## Treatment adherence Adherence to either meditation program (MED or MED+HE) was calculated by summing the total number of minutes spent meditating via Headspace over 8 weeks. The research team had access to individual user data via Headspace, in order to make these calculations. We also assessed meditation frequency with the following questions: “How often did you practice sitting meditation (for 10 min or more) in the past 8 weeks?” Participants selected of the following options: never, less than once a week, 1–3 times per month, 1–2 times per week, 3–4 times per week, or every day. We used this information to ensure that those in the control conditions (active and wait list control) abstained from meditation practice throughout the intervention period. Participants randomized to MED or MED+HE ($$n = 78$$) engaged with the Headspace app an average of 4.15 ± 4.22 minutes per day with no differences between meditation groups ($t = 1.50$, $$p \leq .14$$). Approximately $10\%$ ($$n = 8$$) were adherent to instructions to meditate ≥10 minutes per day over the course of the 8 week program. Participants randomized MED or MED+HE (vs. HE or WL) reported a greater frequency of meditation at 8 weeks, after accounting for baseline frequency ($F = 78.51$, $p \leq .001$). The majority of those in MED ($83\%$) or MED+HE ($72\%$) reported meditating up to two times per week at 8 weeks (compared to $9\%$ of those in HE and $3\%$ of those in WL), suggesting that both mindfulness groups were adherent to treatment (i.e., engaging in mindfulness). ## Data preparation We used SPSS (Version 27.0. Armonk, NY: IBM Corp.) for all variable preparation and statistical analysis. We computed summary statistics to evaluate the distributions of each study variable (i.e., PSS, FAAQ, BMI, sagittal diameter, binge presence, treatment adherence) and assess potential outliers. We did not find any outliers with regard to primary or secondary outcome variables (defined as > ± 3 standard deviations of the mean). ## Treatment effect on outcome variables In a series of Analysis of Covariance (ANCOVA) models, we compared treatment groups (IV: MED vs. MED+HE vs. HE vs. WL) on each 8-week outcome variable (DV: Treatment adherence, PSS, FAAQ, BMI, Sagittal Diameter), adjusting for baseline value of each corresponding measure (covariate). If the main ANCOVA model was significant, we used post-hoc (least square differences) tests to explore group differences. In sub-analyses, we ran an identical series of ANCOVA models, where we collapsed treatment groups (IV) into ‘meditation’ (MED or MED+HE) vs. ‘no meditation’ (HE or WL). ## Moderation analyses We ran a series of ANCOVA models adding an interaction term between treatment group and total meditation minutes (treatment adherence) and examined the simple slopes of the interaction term. We also ran a series of linear regressions to explore whether baseline binge presence (treated as a dichotomous variable of binge vs. no binge presence) moderated the effect of treatment group on primary and secondary outcome variables. We created an interaction term (between binge presence at baseline X intervention) as our independent variable. In all analyses, we considered p ≤.05 to be statistically significant (using two-tailed tests). ## Participant recruitment and retention We enrolled 161 participants, who we randomized to: MED ($$n = 38$$), MED+HE ($$n = 40$$), HE ($$n = 41$$), or WL ($$n = 42$$). At 8 weeks, 145 participants completed follow-up surveys and 128 participants completed an in-person follow-up visit (see Fig 1 for CONSORT diagram). Fig 1: CONSORT flow diagram. ## Participant characteristics Participants had a mean BMI of 30.78 kg/m2 ($40\%$ with obesity vs. $60\%$ with overweight). The majority ($40\%$) identified as White, and reported a four-year college or graduate degree ($85\%$). We classified the majority of participants as administrative staff ($30\%$), researchers ($19\%$) mid-level managers ($16\%$) or medical staff ($15\%$). By study design, participants endorsed a mean PSS score indicative of moderate stress [37] and the majority (>$95\%$) reported meditating less than once a week. Approximately $39\%$ endorsed binge eating presence (objectively large amount of food + loss of control; Tables 1 and 2 for demographic and health characteristics of the sample, respectively). ## Perceived stress Among all 4 groups, there was a treatment effect (F[3,139] = 5.91, $$p \leq .001$$, η2 =.11), such that those in MED (mean change: -5.97, SE = 0.94, $95\%$ CI: -7.84, -4.11) or MED+HE (mean change: -4.97, SE = 0.99, $95\%$ CI: -6.92, -3.02) showed the greatest decreases in PSS score (with no differences between MED vs. MED+HE, $$p \leq .30$$), compared to those in HE (mean change: -2.00, SE = 0.93, $95\%$ CI: -3.84, -0.16) or WL (mean change: -1.66, SE = 0.92, $95\%$ CI: -3.48, 0.16); with no differences between HE vs. WL, $$p \leq .80$$; Fig 2). In sub-analyses, those randomized to either ‘meditation’ (i.e., MED or MED+HE) group showed greater decreases in PSS score ($26\%$ reduction) vs. those in either ‘no meditation (i.e., HE or WL)’ group ($8\%$ reduction; F(1,142 = 15.19, $p \leq .001$, η2 =.10). Findings were identical when using non-parametric tests (Kruskal-Wallis), given the ordinal nature of the PSS scoring. Frequency of meditation moderated the effect of treatment on changes in PSS (interaction term, $F = 4.74$, $$p \leq .03$$), such that greater treatment adherence in meditation was associated with greater decreases in PSS score at 8 weeks (r = -.27, $$p \leq .03$$). Fig 2: Effect of treatment randomization on perceived stress (PSS) at 8 weeks, accounting for baseline values. ## Sagittal diameter Among all 4 treatment groups, we found no treatment effect (F[3,124] = 1.69, $$p \leq .18$$ = 7; η2 =.04). Comparing the estimated marginal means showed a pattern (while not significant) that those in MED+HE (mean change: -0.25, SE = 0.24, $95\%$ CI: -0.72, 0.22) and MED (mean change: -0.12; SE = 0.23, $95\%$ CI: -0.57, 0.33) showed decreases in sagittal diameter, whereas those in HE (mean change:+0.41, SE = 0.23, $95\%$ CI: -0.05, 0.86) and WL (mean change: +0.21, SE = 0.23, $95\%$ CI: -0.24, 0.66) showed slight increases. In sub-analyses, those randomized to either ‘meditation’ group showed greater decreases in sagittal diameter (-0.19 cm; $1\%$ reduction) vs. those in either ‘no meditation’ group (+0.31 cm; $1\%$ increase; F[1,126] = 4.59, $$p \leq .03$$; η2 =.04, Fig 4). Frequency of meditation did not moderate the effect of treatment randomization on changes in sagittal diameter. However, we observed a main effect of meditation frequency on sagittal diameter at 8-weeks ($F = 15.21$, $p \leq .001$), irrespective of treatment randomization. Treatment adherence was associated with greater decreases in sagittal diameter at 8 weeks (r = -.45, $p \leq .001$). Fig 4: Effect of treatment randomization on sagittal diameter at 8 weeks, accounting for baseline values. ## BMI Among all 4 treatment groups, we found no treatment effect (F[3,124] = 1.61, $$p \leq .19$$, η2 =.04) Comparing the estimated marginal means showed a pattern (while not significant) that those in MED+HE (mean change: -.66, SE = 0.29, $95\%$ CI: -1.25, -0.08) showed slight decreases in BMI, whereas those in HE (mean change: +0.04, SE = 0.28, $95\%$ CI: -0.52, 0.60), WL (mean change: +0.06, SE = 0.28, $95\%$ CI: -0.50, 0.61) and MED (mean change: +0.11, SE = 0.28, $95\%$ CI: -0.44, 0.67) showed slight increases. In sub-analyses, those randomized to ‘meditation’ vs. ‘no meditation’ did not differ; Both groups showed similar changes (meditation: -0.26 kg/m2; $1\%$ reduction; vs no meditation: +0.05; $1\%$ reduction; F[1,126] = 1.13, $$p \leq .29$$; η2 =.01). Frequency of meditation did not moderate the effect of treatment randomization on changes in BMI. Treatment adherence was not associated with changes in BMI at 8-weeks (r = -.03, $$p \leq .83$$). ## Moderation by baseline binge eating status We did not observe a main effect of treatment randomization on binge presence at 8 weeks, (chi2 = 0.78, $$p \leq .46$$). We did not find evidence for a moderating effect of baseline binge presence on our primary outcome variables (PSS, FAAQ, ps for interaction terms>.50). However, baseline binge presence moderated the effect of treatment randomization on changes in sagittal diameter at 8 weeks (F[1,123] = 4.95, $$p \leq .03$$, η2 =.04). Examining the simple slopes of this interaction term showed that the association between treatment randomization and sagittal diameter was stronger among those with binge presence but not among those without binge presence. Participants with baseline binge presence showed greater decreases in sagittal diameter if randomized to the meditation (vs. no meditation) group, whereas participants without binge presence did not differ in sagittal diameter changes based on treatment randomization (Fig 5). We observed a similar interaction effect on changes in BMI, although this effect approached statistical significance (F[1,123] = 3.09, $$p \leq .08$$, η2 =.03), such that those who reported binge presence tended to derive the greatest benefit when randomized to meditation vs. no meditation. Fig 5: Associations between treatment randomization and changes in sagittal diameter at 8 weeks among those with (n = 53) vs. without (n = 75) baseline binge presence. ## Discussion Participants with overweight and moderate stress, who received either one of the digitally-based mindfulness programs showed expected reductions in perceived stress, thus confirming prior findings [30, 31]. We also found a small, but significant treatment effect on reductions in sagittal diameter. Contrary to our hypothesis, there was no treatment effect on food cravings or BMI. In an exploratory analysis, we found that meditators who also reported binge eating significantly reduced sagittal diameter. We found preliminary evidence for a moderating effect of treatment adherence on reductions in perceived stress. Furthermore, meditation frequency was positively associated with greater tolerance for food cravings and decreases in sagittal diameter. It is plausible that treatment adherence, measured by meditation frequency using the Headspace app, accounts for treatment effects in a dose-like fashion, and suggests a mechanistic pathway, promoting reductions in stress, food cravings, and abdominal fat. Few digitally-based mindfulness interventions have examined treatment effects on weight and metabolic outcomes [33]. We found a small but significant treatment effect on reductions in sagittal diameter, despite no reductions in BMI. In-person mindfulness interventions have improved some physiological outcomes, including blood pressure, glucose, and abdominal fat [34, 35, 41, 42] despite no changes in BMI. To our knowledge, this is this first digitally-based mindfulness intervention to observe such an effect on abdominal fat distribution. Given the main effect of treatment adherence on reductions in sagittal diameter, it is plausible that this effect is mediated by stress-related pathways, including reductions in cortisol. We found that reduction in perceived stress were associated with reductions in sagittal diameter and increases in awareness of food cravings (ps≤.05). It is plausible that participants who received digital mindfulness may make healthier eating choices (e.g., increased mindfulness around satiety/hunger) which may contribute to downstream metabolic improvements. However, we were not adequately powered to test such a mechanistic pathway. This finding may also point to the importance of measuring abdominal fat distribution, in addition to BMI, in mindfulness-based digital trials. Contrary to our hypothesis, we did not find a treatment effect on food cravings. However, the mindfulness groups reduced in binge eating (although this finding approached statistical significance). Thus, we were unable to replicate known effects of in-person mindfulness-based and mindful eating-based interventions on reductions in dysregulated eating [43–46]. Further, the addition of a healthy eating program did not add a beneficial effect to our primary or secondary outcomes. Both digital mindfulness groups (alone or with healthy eating) performed equally well with regard to reductions in perceived stress and sagittal diameter. It should be noted that the healthy eating (active control) program was newly developed, and in need of further refining following feasibility and acceptability testing. Participants randomized to healthy eating showed good adherence (only $4\%$ declined participation following the initial counseling session). The majority ($73\%$) of participants rated the program as ‘good’ to ‘excellent’, and $91\%$ of completers would recommend the program. Thus, while the healthy eating program, as packaged, did not reduce our measures of food craving, the feasibility data provided the necessary preliminary evidence for future refinement and testing. Qualitative data point to the potential added value of face-to-face counseling to establish health-related eating goals. It is plausible that participants first need to learn general mindfulness skills before showing eating-related improvements. Finally, while exploratory in nature, we replicated our prior findings with regard to treatment matching [24], such that participants with baseline binge presence showed the greatest decreases in sagittal diameter in the meditation (vs. no-meditation) group. These findings suggest that mindfulness may be a better fit for adults with overweight and overeating drive, in comparison to treatment as usual. We did not actively recruit participants high in binge eating, although nearly $40\%$ endorsed engaging in some level of binge eating. Future RCTs should specifically seek to recruit adults with both overweight and binge eating, to fully examine whether mindfulness-based digital approaches contribute to greater improvements in psychological and metabolic health among this high-risk group. This study had several strengths. We were able to deliver a primarily self-guided, scalable treatment for meditation to adults who experienced both perceived stress and overweight. We observed generally good adherence to our digital intervention, with only $11\%$ being lost to follow-up. We had the added benefit of being able to compare our treatment (mindfulness) to what we considered to be an active control (healthy eating) matched for time and attention. However, our study was likely limited by a sample size that may have been too small to detect modest interaction effects. We were unable to truly ascertain whether participants in either control condition were accessing mindfulness programs or apps during the 8 week intervention period. Further, our measures of dysregulated eating may not fully reflect non-homeostatic eating behavior (vs. a semi-structured interview measure of eating pathology), and the scoring metrics for the FAAQ (a 6-point Likert scale ranging from 1 to 6) without the option for including negative (e.g., -1, -2) response may not yield particularly meaningful arithmetic means. Finally, our sample of participants were highly educated and primarily White. Thus, our findings may not fully generalize to the US population of adults with overweight. ## Conclusions A brief digital mindfulness-based intervention is a low-cost method to reduce perceived measures of stress and may have the potential to reduce abdominal fat distribution among adults with overweight and moderate stress. These findings add to the existing literature documenting salutary effects of mindfulness on reports of well-being and extends it to digital-based mindfulness interventions. Future work should seek to clarify mechanisms by which digitally-based mindfulness interventions may contribute to improvements in psychological and physiological health. ## References 1. Flegal KM, Carroll MD, Kit BK, Ogden CL. Prevalence of obesity and trends in the distribution of body mass index among US adults, 1999–2010. Jama (2012) 307 491-7. DOI: 10.1001/jama.2012.39 2. Ogden CL, Carroll MD, Fryar CD, Flegal KM. Prevalence of obesity among adults and youth: United States, 2011–2014. (2015). PMID: 26633046 3. Heraclides AM, Chandola T, Witte DR, Brunner EJ. Work Stress, Obesity and the Risk of Type 2 Diabetes: Gender-Specific Bidirectional Effect in the Whitehall II Study. Obesity (2012) 20 428-33. DOI: 10.1038/oby.2011.95 4. Goh J, Pfeffer J, Zenios SA. The relationship between workplace stressors and mortality and health costs in the United States. Management Science (2016) 62 608-28 5. Ganster DC, Rosen CC. Work stress and employee health: A multidisciplinary review. Journal of management (2013) 39 1085-122 6. Jamieson SD, Tuckey MR. Mindfulness interventions in the workplace: A critique of the current state of the literature. Journal of Occupational Health Psychology (2017) 22 180. DOI: 10.1037/ocp0000048 7. Brown KW, Ryan RM, Creswell JD. Mindfulness: Theoretical Foundations and Evidence for its Salutary Effects. Psychological Inquiry (2007) 18 211-37 8. Chiesa A, Serretti A. Mindfulness-based stress reduction for stress management in healthy people: a review and meta-analysis. The journal of alternative and complementary medicine (2009) 15 593-600. DOI: 10.1089/acm.2008.0495 9. Bohlmeijer E, Prenger R, Taal E, Cuijpers P. The effects of mindfulness-based stress reduction therapy on mental health of adults with a chronic medical disease: a meta-analysis. Journal of psychosomatic research (2010) 68 539-44. DOI: 10.1016/j.jpsychores.2009.10.005 10. Kuyken W, Warren FC, Taylor RS, Whalley B, Crane C, Bondolfi G. Efficacy of mindfulness-based cognitive therapy in prevention of depressive relapse: an individual patient data meta-analysis from randomized trials. JAMA psychiatry (2016) 73 565-74. DOI: 10.1001/jamapsychiatry.2016.0076 11. Hofmann SG, Heering S, Sawyer AT, Asnaani A. How to handle anxiety: The effects of reappraisal, acceptance, and suppression strategies on anxious arousal. Behaviour research and therapy (2009) 47 389-94. DOI: 10.1016/j.brat.2009.02.010 12. Virgili M.. Mindfulness-based interventions reduce psychological distress in working adults: a meta-analysis of intervention studies. Mindfulness (2015) 6 326-37 13. Bruce B, Agras WS. Binge eating in females: A population‐based investigation. International Journal of Eating Disorders (1992) 12 365-73 14. Striegel-Moore RH, Franko DL. Should binge eating disorder be included in the DSM-V? A critical review of the state of the evidence. Annu Rev Clin Psychol (2008) 4 305-24. DOI: 10.1146/annurev.clinpsy.4.022007.141149 15. Wonderlich SA, Gordon KH, Mitchell JE, Crosby RD, Engel SG. The validity and clinical utility of binge eating disorder. The International journal of eating disorders (2009) 42 687-705. DOI: 10.1002/eat.20719 16. Hudson J, Hiripi E, Pope HG, Kessler RC. The prevalence and correlates of eating disorders in the National Comorbidity Survey Replication. Biological psychiatry (2007) 61 348-58. DOI: 10.1016/j.biopsych.2006.03.040 17. Hudson J, Lalonde JK, Coit CE, Tsuang MT, McElroy SL, Crow SJ. Longitudinal study of the diagnosis of components of the metabolic syndrome in individuals with binge eating disorder. Am J Clin Nutr (2010). DOI: 10.3945/ajcn.2010.29203 18. Kessler RC, Berglund PA, Chiu WT, Deitz AC, Hudson JI, Shahly V. The prevalence and correlates of binge eating disorder in the World Health Organization World Mental Health Surveys. Biological Psychiatry (2013) 73 904-14. DOI: 10.1016/j.biopsych.2012.11.020 19. de França GVA, Gigante DP, Olinto MTA. Binge eating in adults: prevalence and association with obesity, poor self-rated health status and body dissatisfaction. Public health nutrition (2014) 17 932-8. DOI: 10.1017/S1368980013000591 20. Ivezaj V, White MA, Grilo CM. Examining binge‐eating disorder and food addiction in adults with overweight and obesity. Obesity (2016) 24 2064-9. DOI: 10.1002/oby.21607 21. Tanofsky-Kraff M, Shomaker LB, Stern EA, Miller R, Sebring N, Dellavalle D. Children’s binge eating and development of metabolic syndrome. Int J Obes (Lond) (2012) 36 956-62. DOI: 10.1038/ijo.2011.259 22. Goyal M, Singh S, Sibinga EM, Gould NF, Rowland-Seymour A, Sharma R. Meditation programs for psychological stress and well-being: a systematic review and meta-analysis. JAMA Internal Medicine (2014) 174 357-68. DOI: 10.1001/jamainternmed.2013.13018 23. Godfrey KM, Gallo LC, Afari N. Mindfulness-based interventions for binge eating: a systematic review and meta-analysis. J Behav Med (2015) 38 348-62. DOI: 10.1007/s10865-014-9610-5 24. Radin RM, Epel ES, Daubenmier J, Moran P, Schleicher S, Kristeller J. Do stress eating or compulsive eating influence metabolic health in a mindfulness-based weight loss intervention?. Health Psychol (2020) 39 147-58. DOI: 10.1037/hea0000807 25. Donker T, Petrie K, Proudfoot J, Clarke J, Birch MR, Christensen H. Smartphones for smarter delivery of mental health programs: a systematic review. J Med Internet Res (2013) 15 e247. DOI: 10.2196/jmir.2791 26. Fairburn CG, Patel V. The impact of digital technology on psychological treatments and their dissemination. Behav Res Ther (2017) 88 19-25. DOI: 10.1016/j.brat.2016.08.012 27. Howells A, Ivtzan I, Eiroa-Orosa FJ. Putting the ‘app’in happiness: a randomised controlled trial of a smartphone-based mindfulness intervention to enhance wellbeing. Journal of happiness studies (2016) 17 163-85 28. Lim D, Condon P, DeSteno D. Mindfulness and compassion: an examination of mechanism and scalability. PLoS One (2015) 10 e0118221. DOI: 10.1371/journal.pone.0118221 29. Ly KH, Truschel A, Jarl L, Magnusson S, Windahl T, Johansson R. Behavioural activation versus mindfulness-based guided self-help treatment administered through a smartphone application: a randomised controlled trial. BMJ Open (2014) 4 e003440. DOI: 10.1136/bmjopen-2013-003440 30. Jayewardene WP, Lohrmann DK, Erbe RG, Torabi MR. Effects of preventive online mindfulness interventions on stress and mindfulness: A meta-analysis of randomized controlled trials. Prev Med Rep (2017) 5 150-9. DOI: 10.1016/j.pmedr.2016.11.013 31. Carolan S, Harris PR, Cavanagh K. Improving Employee Well-Being and Effectiveness: Systematic Review and Meta-Analysis of Web-Based Psychological Interventions Delivered in the Workplace. J Med Internet Res (2017) 19 e271. DOI: 10.2196/jmir.7583 32. Carolan S, Harris PR, Greenwood K, Cavanagh K. Increasing engagement with an occupational digital stress management program through the use of an online facilitated discussion group: Results of a pilot randomised controlled trial. Internet Interv (2017) 10 1-11. DOI: 10.1016/j.invent.2017.08.001 33. Heckenberg RA, Eddy P, Kent S, Wright BJ. Do workplace-based mindfulness meditation programs improve physiological indices of stress? A systematic review and meta-analysis. J Psychosom Res (2018) 114 62-71. DOI: 10.1016/j.jpsychores.2018.09.010 34. Daubenmier J, Moran PJ, Kristeller J, Acree M, Bacchetti P, Kemeny ME. Effects of a mindfulness-based weight loss intervention in adults with obesity: A randomized clinical trial. Obesity (Silver Spring, Md) (2016) 24 794-804. DOI: 10.1002/oby.21396 35. Daubenmier J, Kristeller J, Hecht FM, Maninger N, Kuwata M, Jhaveri K. Mindfulness Intervention for Stress Eating to Reduce Cortisol and Abdominal Fat among Overweight and Obese Women: An Exploratory Randomized Controlled Study. J Obes (2011) 2011 651936. DOI: 10.1155/2011/651936 36. Kinser PA, Robins JL. Control group design: enhancing rigor in research of mind-body therapies for depression. Evid Based Complement Alternat Med (2013) 2013 140467. DOI: 10.1155/2013/140467 37. Cohen S, Kamarck T, Mermelstein R. A global measure of perceived stress. Journal of health and social behavior (1983) 385-96. PMID: 6668417 38. Juarascio A, Forman E, Timko CA, Butryn M, Goodwin C. The development and validation of the food craving acceptance and action questionnaire (FAAQ). Eat Behav (2011) 12 182-7. DOI: 10.1016/j.eatbeh.2011.04.008 39. Bostock S, Crosswell AD, Prather AA, Steptoe A. Mindfulness on-the-go: Effects of a mindfulness meditation app on work stress and well-being. Journal of occupational health psychology (2019) 24 127. DOI: 10.1037/ocp0000118 40. Yanovski SZ, Marcus MD, Wadden TA, Walsh BT. The Questionnaire on Eating and Weight Patterns-5: an updated screening instrument for binge eating disorder. The International journal of eating disorders (2015) 48 259-61. DOI: 10.1002/eat.22372 41. Carlson LE, Speca M, Faris P, Patel KD. One year pre–post intervention follow-up of psychological, immune, endocrine and blood pressure outcomes of mindfulness-based stress reduction (MBSR) in breast and prostate cancer outpatients. Brain, Behavior, & Immunity (2007) 21 1038-49. DOI: 10.1016/j.bbi.2007.04.002 42. Hoge EA, Bui E, Marques L, Metcalf CA, Morris LK, Robinaugh DJ. Randomized controlled trial of mindfulness meditation for generalized anxiety disorder: effects on anxiety and stress reactivity. J Clin Psychiatry (2013) 74 786-92. DOI: 10.4088/JCP.12m08083 43. Kristeller J, Hallett CB. An exploratory study of a meditation-based intervention for binge eating disorder. Journal of health psychology (1999) 4 357-63. DOI: 10.1177/135910539900400305 44. Kristeller J, Wolever RQ. Mindfulness-based eating awareness training for treating binge eating disorder: the conceptual foundation. Eating Disorders (2011) 19 49-61. DOI: 10.1080/10640266.2011.533605 45. Kristeller J, Wolever RQ, Sheets V. Mindfulness-based eating awareness training (MB-EAT) for binge eating: A randomized clinical trial. Journal of Mindfulness (2014) 5 282-97 46. Mason AE, Epel ES, Aschbacher K, Lustig RH, Acree M, Kristeller J. Reduced reward-driven eating accounts for the impact of a mindfulness-based diet and exercise intervention on weight loss: Data from the SHINE randomized controlled trial. Appetite (2016) 100 86-93. DOI: 10.1016/j.appet.2016.02.009
--- title: Validity and sensitivity of field tests’ heart-rate recovery assessment in recreational football players authors: - Susana Póvoas - Peter Krustrup - Carlo Castagna journal: PLOS ONE year: 2023 pmcid: PMC9977042 doi: 10.1371/journal.pone.0282058 license: CC BY 4.0 --- # Validity and sensitivity of field tests’ heart-rate recovery assessment in recreational football players ## Abstract We aimed at examining the criterion validity and sensitivity of heart-rate recovery (HRRec) in profiling cardiorespiratory fitness in male recreational football players in the untrained and trained status, using endurance field-tests. Thirty-two male untrained subjects (age 40 ± 6 years, VO2max 41.7 ± 5.7 ml·kg-1·min-1, body mass 82.7 ± 9.8 kg, stature 173.3 ± 7.4 cm) participated in a 12-week (2‒3 sessions per week) recreational football intervention and were tested pre- and post-intervention (i.e. untrained and trained status). The participants performed three intermittent field tests for aerobic performance assessment, namely Yo-Yo intermittent endurance level 1 (YYIE1) and level 2 (YYIE2) tests, and Yo-Yo intermittent recovery level 1 (YYIR1) test. VO2max was assessed by performing a progressive maximal treadmill test (TT) and maximal HR (HRmax) determined as the maximal value across the testing conditions (i.e., Yo-Yo intermittent tests or TT). HRRec was calculated as the difference between Yo-Yo tests’ HRpeak or HRmax and HR at 30 s (HR30), 60 s (HR60) and 120 s (HR120) and considered as beats·min-1 (absolute) and as % of tests’ HRpeak or HRmax values. Significant post-intervention improvements ($p \leq 0.0001$) were shown in VO2max ($8.6\%$) and Yo-Yo tests performance (23–$35\%$). Trivial to small ($p \leq 0.05$) associations were found between VO2max and HRRec (r = -0.05−0.27, $p \leq 0.05$) across the Yo-Yo tests, and training status either expressed as percentage of HRpeak or HRmax. The results of this study do not support the use of field-test derived HRRec to track cardiorespiratory fitness and training status in adult male recreational football players. ## Introduction Heart rate (HR) monitoring is a valid and popular method for controlling aerobic training aimed at enhancing cardiorespiratory health [1]. In individuals with different training status, health conditions, age and sex, maximal exercise and recovery HR are the variables usually considered to prescribe and monitor training, and to assess cardiorespiratory fitness [1, 2]. Heart rate recovery (HRRec) is most commonly measured as the rate at which heart rate decreases within the following seconds/minutes after the end of exercise [3, 4] and reflects the dynamic balance and coordinated interplay between parasympathetic reactivation and sympathetic withdrawal [4–6]. HRRec following exercise to exhaustion was deemed sensitive to the interplay between parasympathetic and sympathetic nervous activity, reflecting autonomic efficiency [2, 3, 7]. This, alongside with the high accessibility to HRRec measurements, promoted the development of normative values considered useful for detecting pernicious variations in post-maximal-exercise HR kinetics in daily practice [3, 8–10]. Indeed, faster HRRec was reported to be associated with a higher fitness level, and subjects with abnormal HRRec (i.e. a decrease of ≤ 12 beats·min-1 for HR at 60 s after the end of the test) [3, 8] were less likely to be engaged in regular and strenuous exercise [8]. Furthermore, HRRec revealed to be a prognostic indicator of adverse cardiometabolic outcomes and an independent factor for metabolic syndrome prediction [11, 12]. The published scientific evidence of a deleterious effect of attenuated HRRec on cardiovascular and metabolic health and all-cause mortality, promoted HRRec recording in clinical practice as “per se” routine for health risk assessment [12]. Training interventions using conventional aerobic exercise promoted positive changes in HRRec in cardiovascular patients and in athletes [13, 14]. However, most of the published research studies were carried out considering pre- to post-intervention HRRec at selected time points post-exercise, as training outcome variables, not considering changes in VO2max. In fact, the supposed low sensitivity of VO2max for tracking cardiorespiratory fitness changes, and its low absolute reliability, promoted the consideration of HRRec for performance monitoring [13]. Nevertheless, VO2max levels relate to cardiorespiratory health and to the likelihood of all-cause or cardiovascular mortality, suggesting consideration of VO2max tracking in cardiorespiratory fitness enhancing programmes [15]. The practical interest in evaluating cardiorespiratory fitness with an easily accessible variable like HRRec and with endurance field tests, warrants therefore experimental consideration in recreational sports [16, 17]. Recreational football research has provided compelling evidence of clinically sound training-induced improvements in cardiorespiratory fitness and aerobic performance [18, 19]. Regular weekly practice of recreational football in the form of small-sided games, has been proposed as an alternative exercise mode for improving cardiovascular health across age, sex and health status. The casually intermittent nature of recreational football and the associated variability of the individual responses to practice (i.e. small-sided games), suggests the periodical evaluation of aerobic fitness to assess the effectiveness of the training programmes [17, 20]. Furthermore, recreational football involving high-intensity bouts of exercise interspersed with activities performed at lower intensity for recovery, may constitute a viable training activity for improving HRRec [18, 21]. To the best of these study authors’ knowledge, no research has been published with the aim of evaluating the validity and sensitivity (i.e. external responsiveness) of HRRec in recreational football players. Information about the validity, sensitivity and applicability of HRRec monitoring in recreational football would be of great practical importance for the control, regulation and implementation of successful training programmes. The main aim of this study was therefore to examine the association between HRRec values obtained at arbitrarily chosen time points after intermittent maximal field tests with VO2max in adult recreational football players (convergent construct validity) in the untrained and trained states (i.e. longitudinal construct validity). Recovery HR was assessed using field tests deemed to induce exhaustion and popularly used in recreational football interventions (i.e. Yo-Yo intermittent tests) [17, 20]. An effect of individual and training-induced cardiorespiratory fitness improvements on HRRec was assumed as working hypothesis [13]. ## Participants In this study, thirty-two male adults (age 40 ± 6 years, VO2max 41.74 ± 5.72 ml·kg-1·min-1, body mass 82.7 ± 9.8 kg, stature 173.3 ± 7.4 cm, systolic and diastolic blood pressure 125 ± 11 and 74 ± 8 mmHg, respectively) volunteered to participate. The participants were tested at the untrained and trained states, i.e., before and after engaging in a 12-week recreational football training-based intervention. The untrained state (baseline conditions, i.e., pre-intervention) was defined as the participants having less than 20 min of exercise on 3 or more days a week [22]. All the participants were familiarised with the procedures used in the investigation during the two weeks before the commencement of the study by performing submaximal versions of the treadmill test and the Yo-Yo intermittent tests. The participants gave their written informed consent to participate in the study, which was conducted in accordance with the Declaration of Helsinki, and ethical approval was provided by the Ethics Committee of the Faculty of Sport, University of Porto (Porto, Portugal). All participants were informed of the risks and benefits of participating and made aware that they could withdraw from the study at any time without penalty. ## Design In this study, HRRec was determined as the difference between the Yo-Yo intermittent tests’ peak HR (HRpeak) or maximal HR (HRmax), depending on whether HRmax or only HRpeak was reached during the test conditions, and post-exhaustion HR at selected time points, i.e. 30s (HR30), 60s (HR60) and 120s (HR120) after the end of the tests [3, 9]. Specifically, HRRec (i.e. ΔHRRec = peak/maximal HR minus post-exhaustion HR value) was reported in absolute values (beats·min-1) and as a percentage of HRpeak or HRmax (%HRRec) reached during the tests [23]. With the aim of evaluating the proposed levels of validity, data normalisation was performed using HRpeak and HRmax in either the untrained or trained states. Maximal HR (HRmax) was assessed as the maximal value reached across the testing conditions (i.e. Yo-Yo intermittent tests or the treadmill test for VO2max assessment), using a multiple approach, as suggested by Póvoas et al. [ 16], in recreational football. HRpeak refers to the maximal value reached during a testing condition that requires maximal effort, but that is below the maximal reached by the participant in all testing conditions. The magnitude of HR60 was rated for clinical importance using the cut-off values suggested by Cole et al. [ 3, 24]. Given the relatively active recovery observed post-exhaustion during the field tests (deceleration and spontaneous ambulation), abnormality was considered when HR60 was ≤12 beats·min-1 [3, 8]. The intensity and duration of the exercise used to induce HRRec has been considered as a confounding variable [13]. With the aim of examining the interest in using intermittent endurance field tests in assessing HRRec, three intermittent versions of the Yo-Yo test were considered [17], namely levels 1 and 2 of the Yo-Yo intermittent endurance test (YYIE1 and YYIE2, respectively) and the Yo-Yo intermittent recovery test level 1 (YYIR1). The field test protocols were assumed to induce similar aerobic demands with different anaerobic involvement and time to exhaustion in order to stress different HRRec [13, 17]. After the baseline (i.e. untrained status) VO2max and field testing, the participants engaged in a recreational football training intervention (2‒3 60-min weekly sessions) and were retested after 12 weeks of training to access the responsiveness of the selected variables (i.e., pre- and post-intervention). The training intervention was carried out according to the guidelines suggested by Krustrup et al. [ 18, 19, 25] for recreational football interventions with male participants. ## Testing procedures The field tests (Yo-Yo intermittent tests) and the treadmill test for VO2max (TT) assessment were performed in random order with at least 4 days (i.e., 4–6 days) of recovery in between. Test standardisation was achieved by performing the Yo-Yo intermittent tests on the same artificial football pitch and at the same time of day for circadian performance consistency. Furthermore, a standardised warm-up consisting of 10 min of running at different intensities and with changes of direction preceded each Yo-Yo intermittent test. Two minutes of passive rest were considered for each of the participants, before the start of the field tests. On the day before testing, the players refrained from vigorous physical activity. The proposed Yo-Yo intermittent tests differ in their initial running speed and progression, and the between-bouts (40 m) recovery lasts 5–10 s, during which the participants are asked to cover 5–10 m. The Yo-Yo intermittent test protocols were implemented according to the procedures suggested by Krustrup et al. [ 26–28]. The TT (HP Cosmos Quasar, Nussdorf, Germany) consisted of 3 min of walking at 5 km·h-1 and 2 min of running at 8 km·h-1 with $0\%$ inclination, and then alternating between increases in speed (1 km·h-1) and inclination ($1\%$) every 30 s until voluntary exhaustion. Expired respiratory gas fractions were measured using an open-circuit breath-by-breath automated gas analysis system (Quark CPET, Cosmed, Rome, Italy). Attainment of VO2max was assumed when the participants achieved a plateau in VO2 despite an increase in exercise intensity and at least one of the following criteria: a respiratory exchange ratio (RER) greater than 1.10 and RPE equal to or higher than 7 [29, 30]. The highest 15-s VO2 during the final stages of the test was considered as proof of individual VO2max [16, 17]. Data analysis was performed with manual inspection of each TT data file using an Excel file (Microsoft, Redmont, USA). Attainment of individual maximal effort during field tests was considered when the participants, at their subjective exhaustion, reported a rating of perceived exertion (RPE) equal to or higher than 7, at a 0–10 scale or had a HRpeak equal to or higher than $90\%$ of their age-predicted HRmax. Visual inspection of HR profile was performed to assess possible artefacts and to evaluate possible HR plateau and peak. All exercise HRs were recorded at 1-s intervals using Polar Team System 2 HR monitors (Polar Electro Oy, Kempele, Finland). The players were allowed to drink water ad libitum in order to ensure proper hydration under all the exercise conditions considered in this study. After the completion of the field tests participants were instructed and guided to stay with minimal movement to standardise the recovery (2−3 min). No drinking was allowed during the recovery period. ## Training intervention After baseline testing (i.e. laboratory and field testing, $$n = 32$$), the participants engaged in a recreational football intervention comprising 2–3 60-min training sessions per week in the form of 45-min small-sided games played on an artificial pitch (7v7; 43 x 27 m pitch, 83 m2 per player) [31]. The training intervention was conducted over 12 weeks, and the intensity of the sessions was monitored using HR monitors and the subjective internal load estimated by the RPE method [32]. All participants repeated all the test procedures post-intervention in the week after the completion of the last recreational football training session. Participants were advised to follow the guidelines followed at baseline testing. ## Statistical analyses Results are expressed as means±standard deviations (±SD) and $95\%$ confidence intervals ($95\%$ CI). Normality assumption was verified using the Shapiro-Wilk W-test. A repeated-measurements analysis of variance (ANOVA) with post-hoc Bonferroni test was used to compare HRRec across the tests’ recovery time points (i.e. HR30, HR60, HR120). Practical differences were assessed as partial eta squared (η2p) and magnitudes rated as follows: η2p≥0.14 large effect, 0.14>η2p≥0.06 medium effect, 0.06>η2p≥0.01 small effect and η2p<0.01 trivial effect [33]. Pearson correlation (r) was used to assess the associations between variables. The magnitude of the reported effects was described using the Hopkins et al. [ 34] criteria. Within-test conditions variability was expressed as coefficient of variation (%CV). Relative reliability was assessed using the intraclass correlation coefficient (ICC3,1) with $95\%$ CI [35, 36]. According to Landis and Kock [37], ICC values of 0.00–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80, 0.81–1.00 were considered as slight, fair, moderate, substantial and almost perfect, respectively. The Cohen’s d was used to evaluate the effect size, with values above 0.8, between 0.8 and 0.5, between 0.5 and 0.2, and lower than 0.2 considered as large, moderate, small and trivial, respectively [24]. The smallest worthwhile change (SWC) in measurement was considered to test the practical difference between variables and calculated as 0.2 times the variable standard deviation [34]. Sample size estimation was performed for a sample power of $85\%$ with an effect size of 0.50 at a significance level of $5\%$, resulting in 29 participants to be recruited. Significance was set at $5\%$ ($P \leq 0.05$). ## Results The participants showed a relative mean attendance of 73 ± $15\%$ (26 ± 5 total training sessions out off a maximum of 36) with a weekly average of 2.2 ± 0.5 training sessions. The VO2max ($8.6\%$, ~1 MET) and Yo-Yo tests performances were significantly (large) improved (23, 37 and $35\%$ for YYIE1, YYIE2 and YYIR1, respectively) after the training intervention (Table 1). A large and significant decrement (-$2\%$) in HRmax was detected after 12 weeks of recreational football training (Table 1). Yo-Yo test HRpeak was significantly decreased (3, 1 and $1\%$ for YYIE1, YYIE2 and YYIR1, respectively) post-intervention (small to large). The relative reliability (pre- to post-intervention) of the above variables was substantial to almost perfect (ICC>0.71). Table 1 \| Variable \| Pre \| Post \| Diff \| 95%CI Diff \| P value \| d \| ICC \| TEM \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| VO2max (ml·kg-1·min-1) \| 41.7±5.7 \| 45.3±5.8 \| 3.6±3.8 \| (-4.9; -2.3) \| <0.0001 \| 1.0 \| 0.81 (0.64–0.90) \| 6.1 (4.9–8.2) \| \| HRmax (beats·min-1) \| 186±10 \| 182±10 \| 4.1±3.9 \| (2.7–5.5) \| <0.0001 \| 1.0 \| 0.93 (0.86–0.96) \| 1.5 (1.2–2.0) \| \| YYE1-HRpeak (beats·min-1) \| 184±10 \| 180±10 \| 4.7±5.1 \| (2.8–6.5) \| <0.0001 \| 0.8 \| 0.88 (0.77–0.94) \| 2.0 (1.6–2.7) \| \| YYE2-HRpeak (beats·min-1) \| 181±11 \| 179±10 \| 2.3±4.8 \| (0.6–4.0) \| 0.01 \| 0.4 \| 0.90 (0.80–0.95) \| 1.9 (1.5–2.5) \| \| YYR1-HRpeak (beats·min-1) \| 179±11 \| 176±12 \| 2.3±5.9 \| (0.2–4.5) \| 0.03 \| 0.6 \| 0.87 (0.74–0.93) \| 2.4 (1.9–3.2) \| \| YYE1-HR30 (beats·min-1) \| 168±13 \| 167±13 \| 1.4±9.1 \| (-1.9–4.6) \| 0.41 \| 0.1 \| 0.75 (0.55–0.87) \| 3.9 (3.1–5.2) \| \| YYE1-HR60 (beats·min-1) \| 150±16 \| 147±16 \| 3.2±11.6 \| (-1.0–7.4) \| 0.13 \| 0.4 \| 0.74 (0.54–0.87) \| 5.6 (4.5–7.5) \| \| YYE1-HR120 (beats·min-1) \| 132±17 \| 121±15 \| 10.5±13.3 \| (5.7–15.3) \| 0.0001 \| 0.8 \| 0.67 (0.42–0.82) \| 7.4 (5.9–10) \| \| YYE2-HR30 (beats·min-1) \| 170±13 \| 165±10 \| 4.3±7.9 \| (1.5–7.1) \| 0.004 \| 0.7 \| 0.77 (0.58–0.88) \| 3.3 (2.7–4.4) \| \| YYE2-HR60 (beats·min-1) \| 156±17 \| 149±14 \| 7.2±10.3 \| (3.5–10.9) \| 0.0004 \| 0.7 \| 0.79 (0.61–0.89) \| 5.2 (4.2–7.0) \| \| YYE2-HR120 (beats·min-1) \| 134±22 \| 126±16 \| 8.0±13.8 \| (3.0–13.0) \| 0.003 \| 1.1 \| 0.75 (0.55–0.87) \| 9.0 (7.2–12.2) \| \| YYR1-HR30 (beats·min-1) \| 162±13 \| 163±14 \| 0.6±9.3 \| (-2.7–4.0) \| 0.71 \| 0.1 \| 0.77 (0.58–0.88) \| 4.2 (3.3–5.6) \| \| YYR1-HR60 (beats·min-1) \| 147±16 \| 145±17 \| 1.4±11.2 \| (-2.6–5.4) \| 0.48 \| 0.2 \| 0.77 (0.58–0.88) \| 5.7 (4.6–7.7) \| \| YYR1-HR120 (beats·min-1) \| 125±16 \| 125±13 \| 0.2±11.4 \| (-3.9–4.3) \| 0.91 \| 0.0 \| 0.70 (0.47–0.84) \| 6.7 (5.4–9.1) \| \| ΔHRpeak (beats·min-1) \| \| \| \| \| \| \| \| \| \| YYE1-HR 30 \| 16±5 \| 13±5 \| 3.4±6 \| (1.1–5.6) \| 0.004 \| 0.5 \| 0.33 (-0.02–0.61) \| 42.4 (32.7–59.9) \| \| YYE1-HR 60 \| 34±10 \| 32±10 \| 1.5±9 \| (-1.7; 4.7) \| 0.34 \| 0.2 \| 0.61 (0.33–0.79) \| 1.1 (0.9–1.5) \| \| YYE1-HR 120 \| 49±14 \| 60±12 \| -11±13 \| (-15.0; -5.7) \| <0.0001 \| 0.8 \| 0.48 (0.17–0.71) \| 21.4 (16.8–29.3) \| \| YYE2-HR 30 \| 12±5 \| 14±4 \| -2±6 \| (-4.3–0.3) \| 0.08 \| 0.33 \| 0.08 (-0.27–0.42) \| 47.7 (36.7–68.0) \| \| YYE2-HR 60 \| 25±9 \| 30±8 \| -5±8 \| (-7.7; -2.1) \| 0.001 \| 0.65 \| 0.60 (0.32–0.78) \| 23.7 (18.6–32.7) \| \| YYE2-HR 120 \| 47±15 \| 55±11 \| -8±14 \| (-13.0; -3.0) \| 0.003 \| 0.6 \| 0.46 (0.14–0.69) \| 24.1 (18.9–33.3) \| \| YYR1-HR 30 \| 16±6 \| 13±5 \| 3±7 \| (0.5–5.5) \| 0.02 \| 0.42 \| 0.17 (-0.19–0.48) \| 45.7 (35.2–64.9) \| \| YYR1-HR 60 \| 32±9 \| 31±10 \| 1±9 \| (-2.2–3.9) \| 0.58 \| 0.12 \| 0.62 (0.35–0.79) \| 0.9 (0.7–1.2) \| \| YYR1-HR 120 \| 56±10 \| 57±11 \| -1±10 \| (-4.9–2.3) \| 0.46 \| 0.1 \| 0.57 (0.29–0.77) \| 14.8 (11.7–20.1) \| \| ΔHRmax(beats·min-1) \| \| \| \| \| \| \| \| \| \| YYE1-HR 30 \| 18±7 \| 15±7 \| 3±8 \| (-0.1–5.6) \| 0.05 \| 0.39 \| 0.42 (0.08–0.66) \| 45.4 (35.0–64.5) \| \| YYE1-HR 60 \| 36±11 \| 35±11 \| 1±11 \| (-2.9–4.8) \| 0.62 \| 0.09 \| 0.52 (0.21–0.73) \| 30.1 (23.5–41.9) \| \| YYE1-HR 120 \| 54±13 \| 61±12 \| -6±13 \| (-10.9; -1.8) \| 0.008 \| 0.55 \| 0.48 (0.17–0.71) \| 18.3 (14.4–25.0) \| \| YYE2-HR 30 \| 16±6 \| 17±5 \| -0.1±8 \| (-2.9–2.6) \| 0.92 \| 0.13 \| 0.05 (-0.30–0.39) \| 48.3 (37.2–68.9) \| \| YYE2-HR 60 \| 30±11 \| 33±8 \| -3±10 \| (-6.6–0.5) \| 0.09 \| 0.32 \| 0.49 (0.18–0.72) \| 27.0 (21.1–37.4) \| \| YYE2-HR 120 \| 52±16 \| 55±16 \| -3±14 \| (-8.0–2.0) \| 0.23 \| 0.22 \| 0.63 (0.37–0.80) \| 22.6 (17.8–31.2) \| \| YYR1-HR 30 \| 24±8 \| 19±6 \| 5±10 \| (1.2–8.3) \| 0.01 \| 0.51 \| 0.02 (-0.33–0.36) \| 43.1 (33.3–61.0) \| \| YYR1-HR 60 \| 39±10 \| 37±11 \| 3±12 \| (-1.4–6.9) \| 0.19 \| 0.17 \| 0.39 (0.05–0.65) \| 25.7 (20.1–35.5) \| \| YYR1-HR 120 \| 61±10 \| 57±11 \| 4±11 \| (-0.2–8.0) \| 0.06 \| 0.35 \| 0.43 (0.10–0.67) \| 16.6 (13.1–22.6) \| \| YYIE1 (m) \| 1600±621 \| 1969±757 \| 369±321 \| (-485; -253) \| <0.0001 \| 1.2 \| 0.90 (0.80–0.95) \| 14 (11.1–19.0) \| \| YYIE2 (m) \| 471±200 \| 648±230 \| 176±143 \| (-228; -125) \| <0.0001 \| 1.3 \| 0.79 (0.62–0.89) \| 17 (13.7–23.7) \| \| YYIR1 (m) \| 674±298 \| 909±374 \| 235±267 \| (-331; -139) \| <0.0001 \| 0.9 \| 0.71 (0.47–0.84) \| 21 (16.5–28.8) \| HR values during the considered recovery time points are reported in Table 2 as absolute (beats· min-1) and relative (% of tests’ HRpeak or HRmax values). Large and significant ($p \leq 0.0001$) differences were reported for the test conditions across the selected recovery time points and on the two testing occasions (i.e. pre- and post-training intervention). Participants achieved 87–94, 79–86 and 67–$74\%$ of their HRmax or HRpeak values at HR30, HR60 and HR120, respectively. The corresponding absolute HR difference ranges were 11–24, 25–39 and 51–61 beats·min-1 for HR30, HR60 and HR120, respectively. Table 2 \| Training status \| Yo-Yo test \| Variable \| HR (beats·min-1) \| %HRpeak \| %HRmax \| 95%CI Diff \| ΔHRpeak \| ΔHRmax \| η2p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Pre-intervention \| YYIE1 \| HRmax \| 186±10 \| \| \| (14–21)# \| \| \| 0.86 \| \| \| \| HRpeak \| 184±10 \| \| \| \| \| \| \| \| \| \| HR30 \| 168±13 \| 91±3 \| 90±4 \| (14–19) \| 16.0 \| 18.0 \| 0.91 \| \| \| \| HR60 \| 151±16 \| 80±10 \| 81±6 \| (15–21)* \| 33.0 \| 35.0 \| 0.88 \| \| \| \| HR120 \| 132±17 \| 71±7 \| 71±7 \| (15–22)* \| 52.0 \| 54.0 \| 0.87 \| \| \| YYIE2 \| HRmax \| 186±10 \| \| \| (13–20)# \| \| \| 0.88 \| \| \| \| HRpeak \| 181±11 \| \| \| \| \| \| \| \| \| \| HR30 \| 170±13 \| 94±3 \| 91±3 \| (15–21) \| 11.0 \| 16.0 \| 0.86 \| \| \| \| HR60 \| 156±17 \| 86±6 \| 84±6 \| (10–17)* \| 25.0 \| 30.0 \| 0.8 \| \| \| \| HR120 \| 134±22 \| 74±10 \| 72±8 \| (17–27)* \| 47.0 \| 52.0 \| 0.84 \| \| \| YYIR1 \| HRmax \| 186±10 \| \| \| (20–28)# \| \| \| 0.91 \| \| \| \| HRpeak \| 179±11 \| \| \| \| \| \| \| \| \| \| HR30 \| 162±13 \| 91±3 \| 87±4 \| (14–19) \| 17.0 \| 24.0 \| 0.9 \| \| \| \| HR60 \| 147±16 \| 82±5 \| 79±6 \| (12–19)* \| 32.0 \| 39.0 \| 0.86 \| \| \| \| HR120 \| 125±16 \| 70±6 \| 67±6 \| (19–25)* \| 54.0 \| 61.0 \| 0.93 \| \| Post-intervention \| YYIE1 \| HRmax \| 182±10 \| \| \| (11–19)# \| \| \| 0.82 \| \| \| \| HRpeak \| 180±10 \| \| \| \| \| \| \| \| \| \| HR30 \| 167±13 \| 93±3 \| 92±4 \| (10–16) \| 13.0 \| 15.0 \| 0.86 \| \| \| \| HR60 \| 147±16 \| 82±6 \| 81±6 \| (16–23)* \| 33.0 \| 35.0 \| 0.89 \| \| \| \| HR120 \| 121±15 \| 67±6 \| 67±7 \| (21–31)* \| 59.0 \| 61.0 \| 0.88 \| \| \| YYIE2 \| HRmax \| 182±10 \| \| \| (14–19)# \| \| \| 0.92 \| \| \| \| HRpeak \| 179±10 \| \| \| \| \| \| \| \| \| \| HR30 \| 165±10 \| 92±3 \| 91±3 \| (12–16) \| 14.0 \| 17.0 \| 0.91 \| \| \| \| HR60 \| 149±14 \| 83±5 \| 82±5 \| (13–20)* \| 30.0 \| 33.0 \| 0.87 \| \| \| \| HR120 \| 126±16 \| 71±7 \| 69±7 \| (19–26)* \| 53.0 \| 56.0 \| 0.91 \| \| \| YYIR1 \| HRmax \| 182±10 \| \| \| (16–22)# \| \| \| 0.9 \| \| \| \| HRpeak \| 176±12 \| \| \| \| \| \| \| \| \| \| HR30 \| 163±14 \| 92±3 \| 89±4 \| (11–16) \| 13.0 \| 19.0 \| 0.87 \| \| \| \| HR60 \| 145±17 \| 82±6 \| 80±7 \| (14–21)* \| 31.0 \| 37.0 \| 0.88 \| \| \| \| HR120 \| 125±13 \| 71±6 \| 69±6 \| (16–25)* \| 51.0 \| 57.0 \| 0.84 \| Significantly higher (small) relative (%) post-training HR30 values were found in YYIE1 when using HRpeak or HRmax for normalising HRRec (Table 3). Higher (small, $p \leq 0.04$) post-intervention %HR30 values in YYIR1 were reported for both HRpeak and HRmax. Lower and significant ($p \leq 0.04$) post-intervention %HR60 values were seen for HRpeak (moderate) and HRmax (small) in the YYIE2 test. When considering %HR120, a significant and moderate decrement was evident in YYIE1 for both HRpeak and HRmax. The YYIE$2\%$HR120 was small and significantly lower when considering HRpeak for normalisation. Table 3 \| Normalizing variable \| HRRec \| Pre (beats·min-1) \| Post (beats·min-1) \| 95%CI Diff \| P value \| d \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| HR peak \| YYIE1-HR 30 \| 91±3 \| 93±3 \| 0.44–2.82 \| 0.01 \| 0.49 \| \| \| YYIE2-HR 30 \| 94±3 \| 92±2 \| -2.40–0.15 \| 0.08 \| 0.51 \| \| \| YYIR1-HR 30 \| 91±3 \| 92±3 \| 0.17–3.07 \| 0.03 \| 0.4 \| \| HR max \| YYIE1-HR 30 \| 90±4 \| 92±4 \| -2.85–0.17 \| 0.08 \| 0.32 \| \| \| YYIE2-HR 30 \| 91±3 \| 91±3 \| -1.48–1.55 \| 0.97 \| 0.01 \| \| \| YYIR1-HR 30 \| 87±4 \| 89±4 \| 0.38–4.12 \| 0.02 \| 0.43 \| \| HR peak \| YYIE1-HR 60 \| 80±10 \| 82±6 \| -1.40–5.21 \| 0.25 \| 0.22 \| \| \| YYIE2-HR 60 \| 86±6 \| 83±5 \| 1.30–4.60 \| 0.001 \| 0.64 \| \| \| YYIR1-HR 60 \| 82±5 \| 82±6 \| -1.52–2.07 \| 0.75 \| 0.06 \| \| HR max \| YYIE1-HR 60 \| 81±6 \| 81±6 \| -2.03–2.09 \| 0.98 \| 0.0 \| \| \| YYIE2-HR 60 \| 84±6 \| 82±5 \| 0.20–4.05 \| 0.03 \| 0.41 \| \| \| YYIR1-HR 60 \| 79±6 \| 80±7 \| -1.38–3.13 \| 0.43 \| 0.14 \| \| HR peak \| YYIE1-HR 120 \| 71±7 \| 67±6 \| -6.13;-1.74 \| 0.01 \| 0.65 \| \| \| YYIE2-HR 120 \| 74±10 \| 71±7 \| -5.89;-0.69 \| 0.02 \| 0.49 \| \| \| YYIR1-HR 120 \| 70±6 \| 71±6 \| -1.23–3.16 \| 0.37 \| 0.16 \| \| HR max \| YYIE1-HR 120 \| 71±7 \| 67±7 \| -6.57;-1.67 \| 0.002 \| 0.61 \| \| \| YYIE2-HR 120 \| 72±10 \| 69±7 \| -0.01–5.32 \| 0.05 \| 0.39 \| \| \| YYIR1-HR 120 \| 67±6 \| 69±6 \| -0.78–3.84 \| 0.19 \| 0.24 \| At baseline, VO2max was not significantly associated (trivial to moderate) with HRRec at the selected time points, expressed as test HRpeak or HRmax, in any of the considered field-testing conditions (i.e. YYIE1, YYIE2 and YYIR1, Table 4). The lack of significant ($p \leq 0.05$) correlations persisted in the trained state. Table 4 \| Unnamed: 0 \| Unnamed: 1 \| HR recovery \| HR recovery.1 \| HR recovery.2 \| HR recovery.3 \| HR recovery.4 \| HR recovery.5 \| HR recovery.6 \| HR recovery.7 \| HR recovery.8 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Normalising Variables \| VO2max \| YYIE130 \| YYIE160 \| YYIE1120 \| YYIE230 \| YYIE260 \| YYIE2120 \| YYIR130 \| YYIR160 \| YYIR1120 \| \| HR peak \| \| \| \| \| \| \| \| \| \| \| \| \| Pre \| 0.05 \| -0.02 \| -0.04 \| 0.33 \| 0.30 \| 0.31 \| -0.05 \| 0.20 \| 0.05 \| \| \| 95%CI \| -0.30–0.39 \| -0.37–0.33 \| -0.38–0.31 \| -0.02–0.61 \| -0.05–0.59 \| -0.04–0.59 \| -0.39–0.31 \| -0.16–0.52 \| -0.30–0.39 \| \| \| P value \| 0.79 \| 0.91 \| 0.83 \| 0.07 \| 0.091 \| 0.084 \| 0.79 \| 0.28 \| 0.79 \| \| \| Post \| \| \| \| \| \| \| \| \| \| \| \| 95%CI \| -0.19–0.49 \| -0.13–0.53 \| -0.28–0.41 \| -0.49–0.18 \| -0.23–0.46 \| -0.12–0.54 \| -0.05–0.59 \| -0.16–0.52 \| -0.29–0.41 \| \| \| P value \| 0.36 \| 0.22 \| 0.69 \| 0.34 \| 0.46 \| 0.18 \| 0.10 \| 0.27 \| 0.71 \| \| HR max \| \| \| \| \| \| \| \| \| \| \| \| \| Pre \| 0.11 \| 0.05 \| -0.01 \| 0.16 \| 0.22 \| 0.27 \| 0.01 \| 0.20 \| 0.07 \| \| \| 95%CI \| -0.25–0.44 \| -0.30–0.39 \| -0.35–0.34 \| -0.20–0.49 \| -0.14–0.53 \| -0.09–0.57 \| -0.34–0.36 \| -0.16–0.52 \| -0.28–0.41 \| \| \| P value \| 0.55 \| 0.77 \| 0.97 \| 0.37 \| 0.22 \| 0.13 \| 0.95 \| 0.27 \| 0.69 \| \| \| Post \| 0.19 \| 0.24 \| 0.09 \| 0.14 \| -0.17 \| 0.24 \| 0.17 \| 0.15 \| 0.03 \| \| \| 95%CI \| -0.17–0.50 \| -0.12–0.54 \| -0.27–0.42 \| -0.22–0.46 \| -0.49–0.19 \| -0.12–0.54 \| -0.19–0.49 \| -0.21–0.48 \| -0.32–0.38 \| \| \| P value \| 0.31 \| 0.19 \| 0.63 \| 0.46 \| 0.34 \| 0.18 \| 0.35 \| 0.40 \| 0.85 \| Using the ≤12 beats·min-1 criterion to qualitatively evaluate HRRec, only 3‒$6\%$ and $3\%$ of the participants reported the supposed abnormalities in HR60 during the pre-intervention YYIE2 and post-intervention YYIE1, respectively (Table 5). Table 5 \| Unnamed: 0 \| Unnamed: 1 \| Pre \| Pre.1 \| Pre.2 \| Post \| Post.1 \| Post.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variable \| \| HR30 \| HR60 \| HR120 \| HR30 \| HR60 \| HR120 \| \| YYIE1 \| HRpeak \| 9 (28%) \| 0 \| 0 \| 19 (59%) \| 1 (3%) \| 0 \| \| \| HRmax \| 7 (22%) \| 0 \| 0 \| 17 (53%) \| 1 (3%) \| 0 \| \| YYIE2 \| HRpeak \| 22 (69%) \| 2 (6%) \| 0 \| 12 (38%) \| 0 \| 0 \| \| \| HRmax \| 9 (28%) \| 1 (3%) \| 0 \| 8 (25%) \| 0 \| 0 \| \| YYIR1 \| HRpeak \| 7 (22%) \| 0 \| 0 \| 15 (47%) \| 0 \| 0 \| \| \| HRmax \| 3 (9%) \| 0 \| 0 \| 5 (16%) \| 0 \| 0 \| ## Discussion This is the first study to examine the validity and sensitivity of using intermittent endurance field tests’ post-exhaustion HRRec values to characterise cardiorespiratory fitness in male participants that volunteered for a recreational football intervention, in the trained and untrained states [13, 17]. These tests were supposed to induce similar maximal aerobic demands and different anaerobic loads [38, 39]. The main finding was that no associations of significance or practical importance were found between HRRec and cardiorespiratory fitness in either the untrained or trained states. The reported significant, mainly moderate to large, changes in post-intervention HRpeak and HRmax affected the representation of HRRec values magnitude, suggesting a multifaceted interpretation of HRRec. The occurrence of HRRec abnormalities (i.e. ≤12 beats·min-1 for HR60) was dependent on the field test performed and training state (i.e. untrained or trained). This suggests that caution is advised when considering fixed recovery HR count as a reference for characterising poor HRRec in healthy male recreational football players in the age span here considered (30–50 years). In this study, the relative reliability of VO2max, HRmax and Yo-Yo HRpeak pre- to post-intervention was almost perfect (ICC, 0.81–0.93), supporting the consistency of participants’ training-induced changes ranking [37]. Interestingly, significant decrements in HRmax (large) and Yo-Yo HRpeak (small-to-large) were found at post-intervention (Table 1) [40]. These results provide evidence of the internal validity of this study design and confirm the applicability of the Yo-Yo intermittent tests in recreational football [16, 17, 20]. Bosquet et al. [ 23] assessed HRRec reliability by replicating a maximal treadmill test at least 72 hours apart in healthy subjects. A general low relative reliability of ΔHRRec variables was reported, except when considering absolute HRRec values (beats·min-1), with ICC ranging between 0.68 and 0.83 [23]. In line with the cited study, we found a substantial agreement between pre-to-post absolute HRRec values (0.67–0.79) across the post-exhaustion time points [23]. Similarly, mainly slight to moderate agreement (0.17–0.61 and 0.05–0.63 using HRpeak and HRmax, respectively) was reported for ΔHRRec across the Yo-Yo tests and recovery time points. These results (i.e. long-term relative reliability) add evidence of the poor reliability of ΔHRRec values when used to characterise HRRec after maximal testing [23]. This finding is of practical relevance as, in the clinical set-up, HRRec efficiency is mainly reported as ΔHRRec [3, 14, 41]. HRRec is deemed to be affected by the fitness level and training status of subjects of different age, gender and health condition [5, 7, 13, 42]. Cross-sectional studies found cardiorespiratory fitness as a possible cause for the faster HRRec in athletic populations compared to untrained or inactive subjects [13]. Darr et al. [ 41] reported an effect of VO2max level on HRRec with trained subjects (>60 ml·kg-1·min-1 mean VO2peak), reporting a faster decrement in HR than in untrained subjects. Interestingly, the untrained male subjects’ mean VO2max values (40 ml·kg-1·min-1) were similar to those of this study’s recreational football participants at pre-intervention evaluations (i.e. untrained status), suggesting an effect of training status on HRRec. This study’s findings did not provide empirical evidence of an association between cardiorespiratory fitness improvements and HRRec variations in recreational football players. This was supported by the analyses performed in both the untrained and trained status and occurs irrespective of consideration for physiological (i.e. VO2max) or performance (Yo-Yo tests) changes. Additionally, this study’s results are in line with those of Hautala et al. [ 43] in inactive subjects who trained for 2 weeks on an intensive endurance programme and reported no variations in HR60, despite significant positive VO2max changes (+$8\%$). Similarly, 3 weeks of intensive football training did not provide any changes in HRRec in competitive young football players [44]. Again, comparison with other studies addressing HRRec may be confounding, as different research designs and HRRec assessment tests were used [45]. The reported large and significant reduction in HRmax and HRpeak across the Yo-Yo tests further supports the previously reported findings of training-induced changes in heart physiology [40]. Although not based on robust mechanistic evidence, the reported significant and practical important decrement in HRmax could have been the result of variations in short-term neurological changes related to variation in parasympathetic and sympathetic nervous systems interplay [40]. Short-term effects of recreational football on players’ heart anatomy and physiology, may have also played a role [46]. However, this recreational football intervention revealed a limited effect on HRRec, with moderate to large changes in absolute HRRec values observed only for HR120 after YYIE2 and YYIE1 and for HR30 and HR60 after YYIE2 (Table 1), suggesting a test-dependent effect on HRRec. This was further supported by ΔHRRec analyses, revealing significantly higher post-intervention values only for a few test conditions, i.e. HR60 in YYIE2 and HR120 in both YYIE2 and YYIE1, when considering HRpeak as a reference for normalisation. Interestingly, a faster post-training intervention HRRec was only evident for the YYIE1 ΔHR120 when considering HRmax. The variations in HRRec across the tests and testing time points were also evident when considering %HRRec values with a general increase (trivial to moderate) in post-intervention values. Improvements in VO2max and aerobic performance as a consequence of endurance training and recreational football practice were shown to be associated with significant decrements in HRmax and changes in HRpeak in field tests [16, 20, 40]. In this study, large and significant decrements in HRmax were detected, providing practical relevance of changes in the range of 3‒6 beats·min-1 (SWC 2 beats·min-1). Given the period of the considered change (12 weeks), a reassessment of HRmax every 12 weeks may be advisable for controlling and regulating exercise intensity in recreational football interventions. The reported changes in HRmax were paralleled by moderate to large decrements ($p \leq 0.04$) in Yo-Yo HRpeak post-intervention, further suggesting caution when choosing ΔHRRec to profile HRRec. Indeed, variations in peak HR values as an effect of training, alongside with the reported fair-to-moderate HRRec relative reliability, discourage consideration of HRRec as raw data difference values [23]. Cole et al. [ 3, 8] provided longitudinal evidence of an association between decrements in HRRec and reductions in all-cause and cardiovascular mortality. Furthermore, the same authors demonstrated the predictive strength of absolute HR cut-off values when considering HR60 and HR120. Using the suggested cut-off values (i.e. ≤12 beats·min-1), we only found 3–$6\%$ of abnormalities in HR60 across the field tests. The training intervention and the associated increase in VO2max (~$9\%$) and corresponding decrement in HRmax (~$3\%$) produced a remarkable reduction in the initial HR60 abnormalities, when considering a highly demanding test like YYIE2 (Table 5). Indeed, no HR60 abnormalities were detected in the participants when examining pre-intervention HR60 after the YYIE1 and YYIR1 tests. Interestingly, only $3\%$ of participants reported HR60 abnormalities in YYIE1 HR60 post-intervention, suggesting a sort of independence between cardiorespiratory fitness improvement and HR60 abnormalities. The resulting occurrence of HR60 abnormalities may have been the direct consequence of the participants’ health-related inclusion criteria. However, the lower prevalence of abnormal HR60 values at post-intervention might be considered as evidence of a possible positive effect of recreational football practice on reducing potential health-related risk factors. The reported test-related prevalence of HR60 abnormalities has practical importance for preventive medicine that warrants future studies [3, 8, 10]. However, the normative value proposed by these authors was derived from a population of male inactive subjects who were approximately 20 years older than the recreational football participants considered in the present study, which promotes the interest in population-specific normative values for tracking abnormalities in HRRec [3, 10]. The intermittent nature of recreational football, involving high-intensity bouts of exercise interspersed with activities performed at lower intensity for recovery, or less demanding match-related actions, potentially would be effective for improving HRRec [18]. However, this study results did not provide evidence for enhanced HRRec as result of a typical recreational football intervention. This was probably the consequence of considering the training-induced variations in HRpeak and HRmax when profiling the HRRec variables at post-intervention [40, 47]. Further studies with larger samples and a mechanistic design, are warranted to understand the variation in heart physiology provided by a casually intermittent activity such as recreational football in different populations. Additionally, the use of a control group would be useful in future studies to fully understand the nature of HRRec and related variables. These findings question the use of HRRec as indicator of cardiorespiratory fitness level (see if you agree) or training-induced changes in healthy subjects participating in a recreational football intervention or in recreational players during the training process. The observed large changes (26–$43\%$) in field test performance suggest sensitivity in tracking the aimed enhancement of the individual aerobic performance in recreational football training interventions. Changes in YYIE1 and YYIR1 performance were moderately and significantly associated with changes in HR60 ($r = 0.36$, $$P \leq 0.045$$) and HR120 ($r = 0.38$, $$P \leq 0.034$$), respectively. These results support the general picture of a limited association between changes in cardiorespiratory fitness level in recreational football players and HRRec variables. In the published literature, HRRec assessment is reported as absolute values (beats·min-1) [3, 7, 13, 23]. Despite the practical interest of considering absolute values, differences in individual HRmax may provide biased data supposedly producing false positive results. Heart rate recovery normalisation using peak test HR or HRmax may potentially be the solution for avoiding biased data that may affect training prescription and clinical diagnosis. However, in this study the deliberate use of absolute or HRmax derived HRRec did not provide differences in the information supposed to have clinical importance. Inter-subject variability in test HRpeak (~$6\%$) may have been the cause of the reported limited effect of data normalisation on the considered variables. The practical interest of the effect of reporting HRRec data deserves further studies with populations of different age and cardiorespiratory fitness level. Attention should be paid when considering cut-off values (i.e. 12 or 43 beats·min-1) to qualitatively characterise the risk of developing cardiovascular diseases [3, 8–10]. A bias-limiting indicator of the cardiorespiratory risk could be developed using individual maximal values. Unfortunately, the reduced data variability of this study’s participants was not helpful in providing meaningful guidelines and further studies are warranted. However, the age range of this study’s participants (5.6 years, standard deviation range 22 years) may have affected the results, as in age-independent groups subjects with higher HRmax reported better absolute HRRec. In this study, a moderate association was reported between HRmax and HRRec. Future training studies should also investigate the possibility to find more meaningful HRRec reference time points as suggested in a cross-sectional study by Ostojic et al [48]. This would be of specific interest, as short-term HRRec (i.e., as short as 20s) has been reported to be faster in athletes of intermittent sports (i.e., basketball, soccer and team handball) with at least four years of participation in these sports [48]. Given that, knowledge about the applicability of the short-term HRRec concept in previously untrained subjects participating in a training intervention, using exclusively intermittent exercise like recreational football, would be of great practical interest. ## Conclusions It was not possible with this study design to support the validity of tracking post-exhaustion HRRec to estimate individual aerobic fitness (VO2max), either in the untrained or trained status (i.e. pre- and post-intervention, respectively). Indeed, no significant and practically important associations were found between HRRec variables and recreational football players’ VO2max. This study’s results are in line with those reported in the athletic populations, suggesting HRRec as a “per se” physiological adaptation that is independent of VO2max level and changes [13]. Given the interest of this issue for public health, further studies involving a larger number of participants followed for a longer time are warranted. From the practical point of view, HRRec is a reliable variable in the short-term (i.e. 12 weeks), nevertheless, it is not associated with the improvement in aerobic fitness in this population of recreational football players. Although this study’s results refer to recreational football players, the information obtained may be of interest for all professionals dealing with health-enhancing strategies evaluated under field conditions. ΔHRRec is considered as the variable for tracking the efficiency of the physiological processes that underpin HRRec. This study’s results suggest that caution is advised when considering ΔHRRec, as this variable may be affected by the concomitant reductions in HR at exhaustion values and, consequently, bias the reported differences. ## References 1. Garber CE, Blissmer B, Deschenes MR, Franklin BA, Lamonte MJ, Lee IM. American College of Sports Medicine position stand. Quantity and quality of exercise for developing and maintaining cardiorespiratory, musculoskeletal, and neuromotor fitness in apparently healthy adults: guidance for prescribing exercise. Med Sci Sports Exerc (2011) 43 1334-59. DOI: 10.1249/MSS.0b013e318213fefb 2. Best SA, Bivens TB, Dean Palmer M, Boyd KN, Melyn Galbreath M, Okada Y. Heart rate recovery after maximal exercise is blunted in hypertensive seniors. J Appl Physiol (1985) (2014) 117 1302-7. DOI: 10.1152/japplphysiol.00395.2014 3. Cole CR, Blackstone EH, Pashkow FJ, Snader CE, Lauer MS. Heart-rate recovery immediately after exercise as a predictor of mortality. N Engl J Med (1999) 341 1351-7. DOI: 10.1056/NEJM199910283411804 4. Pecanha T, Silva-Junior ND, Forjaz CL. Heart rate recovery: autonomic determinants, methods of assessment and association with mortality and cardiovascular diseases.. Clin Physiol Funct Imaging. (2014) 34 327-39. DOI: 10.1111/cpf.12102 5. Borresen J, Lambert MI. Autonomic control of heart rate during and after exercise: measurements and implications for monitoring training status. Sports Medicine (2008) 38 633-46. DOI: 10.2165/00007256-200838080-00002 6. Coote JH. Recovery of heart rate following intense dynamic exercise. Exp Physiol (2010) 95 431-40. DOI: 10.1113/expphysiol.2009.047548 7. Kim BJ, Jo EA, Im SI, Kim HS, Heo JH, Cho KI. Heart rate recovery and blood pressure response during exercise testing in patients with microvascular angina. Clin Hypertens (2019) 25 4. DOI: 10.1186/s40885-019-0108-x 8. Cole CR, Foody JM, Blackstone EH, Lauer MS. Heart rate recovery after submaximal exercise testing as a predictor of mortality in a cardiovascularly healthy cohort. Ann Intern Med (2000) 132 552-5. DOI: 10.7326/0003-4819-132-7-200004040-00007 9. Watanabe J, Thamilarasan M, Blackstone EH, Thomas JD, Lauer MS. Heart rate recovery immediately after treadmill exercise and left ventricular systolic dysfunction as predictors of mortality: the case of stress echocardiography. Circulation (2001) 104 1911-6. PMID: 11602493 10. Shetler K, Marcus R, Froelicher VF, Vora S, Kalisetti D, Prakash M. Heart rate recovery: validation and methodologic issues. J Am Coll Cardiol (2001) 38 1980-7. DOI: 10.1016/s0735-1097(01)01652-7 11. Jae SY, Bunsawat K, Kunutsor SK, Yoon ES, Kim HJ, Kang M. Relation of Exercise Heart Rate Recovery to Predict Cardiometabolic Syndrome in Men.. Am J Cardiol (2019) 123 582-7. DOI: 10.1016/j.amjcard.2018.11.017 12. Qiu S, Cai X, Sun Z, Li L, Zuegel M, Steinacker JM. Heart Rate Recovery and Risk of Cardiovascular Events and All-Cause Mortality: A Meta-Analysis of Prospective Cohort Studies.. J Am Heart Assoc (2017) 6. DOI: 10.1161/JAHA.117.005505 13. Daanen HA, Lamberts RP, Kallen VL, Jin A, Van Meeteren NL. A systematic review on heart-rate recovery to monitor changes in training status in athletes.. Articolo in rivista. (2012) 7 251-60. DOI: 10.1123/ijspp.7.3.251 14. Elshazly A, Khorshid H, Hanna H, Ali A. Effect of exercise training on heart rate recovery in patients post anterior myocardial infarction. Egypt Heart J (2018) 70 283-5. DOI: 10.1016/j.ehj.2018.04.007 15. Nes BM, Vatten LJ, Nauman J, Janszky I, Wisloff U. A simple nonexercise model of cardiorespiratory fitness predicts long-term mortality.. Med Sci Sports Exerc (2014) 46 1159-65. DOI: 10.1249/MSS.0000000000000219 16. Póvoas SC, Krustrup P, Pereira R, Vieira S, Carneiro I, Magalhaes J. Maximal heart rate assessment in recreational football players: A study involving a multiple testing approach. Scand J Med Sci Sports (2019) 29 1537-45. DOI: 10.1111/sms.13472 17. Castagna C, Krustrup P, Póvoas S. Yo-Yo intermittent tests are a valid tool for aerobic fitness assessment in recreational football. Eur J Appl Physiol (2019). DOI: 10.1007/s00421-019-04258-8 18. Krustrup P, Krustrup BR. Football is medicine: it is time for patients to play!. Br J Sports Med (2018) 52 1412-4. DOI: 10.1136/bjsports-2018-099377 19. Krustrup P, Williams CA, Mohr M, Hansen PR, Helge EW, Elbe AM. The "Football is Medicine" platform-scientific evidence, large-scale implementation of evidence-based concepts and future perspectives.. Scand J Med Sci Sports (2018) 28 3-7. DOI: 10.1111/sms.13220 20. Póvoas SC, Krustrup P, Castagna C. Submaximal field testing validity for aerobic fitness assessment in recreational football.. Scand J Med Sci Sports. (2019). DOI: 10.1111/sms.13606 21. Krustrup P, Bangsbo J. Recreational football is effective in the treatment of non-communicable diseases. British journal of sports medicine (2015). DOI: 10.1136/bjsports-2015-094955 22. Church TS, Blair SN, Cocreham S, Johannsen N, Johnson W, Kramer K. Effects of aerobic and resistance training on hemoglobin A1c levels in patients with type 2 diabetes: a randomized controlled trial. JAMA (2010) 304 2253-62. DOI: 10.1001/jama.2010.1710 23. Bosquet L, Gamelin FX, Berthoin S. Reliability of postexercise heart rate recovery. Int J Sports Med (2008) 29 238-43. DOI: 10.1055/s-2007-965162 24. Cohen J.. Statistical power analysis for the behavioral sciences. (1988) 25. Krustrup P, Aagaard P, Nybo L, Petersen J, Mohr M, Bangsbo J. Recreational football as a health promoting activity: a topical review.. Scand J Med Sci Sports (2010) 20 1-13. DOI: 10.1111/j.1600-0838.2010.01108.x 26. Krustrup P, Bradley PS, Christensen JF, Castagna C, Jackman S, Connolly L. The Yo-Yo IE2 test: physiological response for untrained men versus trained soccer players.. Med Sci Sports Exerc. (2015) 47 100-8. DOI: 10.1249/MSS.0000000000000377 27. Krustrup P, Mohr M, Amstrup T, Rysgaard T, Johansen J, Steensberg A. The Yo-Yo Intermittent Recovery Test: Physiological response, reliability, and validity.. Med Sci Sports Exer. (2003) 35 697-705. DOI: 10.1249/01.MSS.0000058441.94520.32 28. Krustrup P, Mohr M, Nybo L, Jensen JM, Nielsen JJ, Bangsbo J. The Yo-Yo IR2 test: physiological response, reliability, and application to elite soccer.. Med Sci Sports Exerc. (2006) 38 1666-73. DOI: 10.1249/01.mss.0000227538.20799.08 29. Midgley AW, Mc Naughton LR, Wilkinson M. Criteria and other methodological considerations in the evaluation of time at V.O2max.. J Sports Med Phys Fitness (2006) 46 183-8. PMID: 16823345 30. Midgley AW, McNaughton LR, Polman R, Marchant D. Criteria for determination of maximal oxygen uptake: a brief critique and recommendations for future research.. Sports Med. (2007) 37 1019-28. DOI: 10.2165/00007256-200737120-00002 31. Randers MB, Nybo L, Petersen J, Nielsen JJ, Christiansen L, Bendiksen M. Activity profile and physiological response to football training for untrained males and females, elderly and youngsters: influence of the number of players.. Scand J Med Sci Sports (2010) 20 14-23. DOI: 10.1111/j.1600-0838.2010.01069.x 32. Borg G, Hassmen P, Lagerstrom M. Perceived exertion related to heart rate and blood lactate during arm and leg exercise. Eur J Appl Physiol Occup Physiol (1987) 56 679-85. DOI: 10.1007/BF00424810 33. Bakeman R.. Recommended effect size statistics for repeated measures designs. Behav Res Methods (2005) 37 379-84. DOI: 10.3758/bf03192707 34. Hopkins WG, Marshall SW, Batterham AM, Hanin J. Progressive statistics for studies in sports medicine and exercise science. Med Sci Sports Exerc (2009) 41 3-13. DOI: 10.1249/MSS.0b013e31818cb278 35. Weir JP. Quantifying test-retest reliability using the intraclass correlation coefficient and the SEM.. J Strength Cond Res (2005) 19 231-40. DOI: 10.1519/15184.1 36. Hopkins WG. Measures of reliability in sports medicine and science. Sports Med (2000) 30 1-15. DOI: 10.2165/00007256-200030010-00001 37. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics (1977) 33 159-74. PMID: 843571 38. Bangsbo J, Iaia FM, Krustrup P. The Yo-Yo intermittent recovery test: a useful tool for evaluation of physical performance in intermittent sports.. Sports Med. (2008) 38 37-51. DOI: 10.2165/00007256-200838010-00004 39. Schmitz B, Pfeifer C, Kreitz K, Borowski M, Faldum A, Brand SM. The Yo-Yo Intermittent Tests: A Systematic Review and Structured Compendium of Test Results.. Front Physiol. (2018) 9 870. DOI: 10.3389/fphys.2018.00870 40. Zavorsky GS. Evidence and possible mechanisms of altered maximum heart rate with endurance training and tapering. Sports Med (2000) 29 13-26. DOI: 10.2165/00007256-200029010-00002 41. Darr KC, Bassett DR, Morgan BJ, Thomas DP. Effects of age and training status on heart rate recovery after peak exercise. Am J Physiol (1988) 254 H340. DOI: 10.1152/ajpheart.1988.254.2.H340 42. Dimkpa U.. Post-Exercise Heart Rate Recovery: An index of cardiovascular Fitness. Journal of Exercise Physiologyonline (2009) 12 19-22 43. Hautala AJ, Rankinen T, Kiviniemi AM, Makikallio TH, Huikuri HV, Bouchard C. Heart rate recovery after maximal exercise is associated with acetylcholine receptor M2 (CHRM2) gene polymorphism.. Am J Physiol Heart Circ Physiol (2006) 291 H459-66. DOI: 10.1152/ajpheart.01193.2005 44. Buchheit M, Mendez-Villanueva A, Quod MJ, Poulos N, Bourdon P. Determinants of the variability of heart rate measures during a competitive period in young soccer players. European journal of applied physiology (2010) 109 869-78. DOI: 10.1007/s00421-010-1422-x 45. Borresen J, Lambert MI. Changes in heart rate recovery in response to acute changes in training load. European journal of applied physiology (2007) 101 503-11. DOI: 10.1007/s00421-007-0516-6 46. Krustrup P, Hansen PR, Nielsen CM, Larsen MN, Randers MB, Manniche V. Structural and functional cardiac adaptations to a 10-week school-based football intervention for 9-10-year-old children.. Articolo in rivista. (2014) 24 4-9. DOI: 10.1111/sms.12277 47. Dobbin N, Highton J, Moss SL, Twist C. The Effects of In-Season, Low-Volume Sprint Interval Training With and Without Sport-Specific Actions on the Physical Characteristics of Elite Academy Rugby League Players.. Int J Sports Physiol Perform. (2020) 15 705-13. DOI: 10.1123/ijspp.2019-0165 48. Ostojic SM, Markovic G, Calleja-Gonzalez J, Jakovljevic DG, Vucetic V, Stojanovic MD. Ultra short-term heart rate recovery after maximal exercise in continuous versus intermittent endurance athletes. Eur J Appl Physiol (2010) 108 1055-9. DOI: 10.1007/s00421-009-1313-1
--- title: 'Prevalence of co-morbidity and history of recent infection in patients with neuromuscular disease: A cross-sectional analysis of United Kingdom primary care data' authors: - Iain M. Carey - Niranjanan Nirmalananthan - Tess Harris - Stephen DeWilde - Umar A. R. Chaudhry - Elizabeth Limb - Derek G. Cook journal: PLOS ONE year: 2023 pmcid: PMC9977045 doi: 10.1371/journal.pone.0282513 license: CC BY 4.0 --- # Prevalence of co-morbidity and history of recent infection in patients with neuromuscular disease: A cross-sectional analysis of United Kingdom primary care data ## Abstract ### Background People with neuromuscular disease (NMD) experience a broader range of chronic diseases and health symptoms compared to the general population. However, no comprehensive analysis has directly quantified this to our knowledge. ### Methods We used a large UK primary care database (Clinical Practice Research Datalink) to compare the prevalence of chronic diseases and other health conditions, including recent infections between 23,876 patients with NMD ever recorded by 2019 compared to 95,295 age-sex-practice matched patients without NMD. Modified Poisson regression estimated Prevalence Ratios (PR) to summarise the presence of the disease/condition ever (or for infections in 2018) in NMD patients versus non-NMD patients. ### Results Patients with NMD had significantly higher rates for 16 of the 18 conditions routinely recorded in the primary care Quality and Outcomes Framework (QOF). Approximately 1-in-10 adults with NMD had ≥4 conditions recorded (PR = 1.39, $95\%$CI 1.33–1.45). Disparities were more pronounced at younger ages (18–49). For other (non-QOF) health conditions, significantly higher recorded levels were observed for rarer events (pulmonary embolism PR = 1.96 $95\%$CI 1.76–2.18, hip fractures PR = 1.65 $95\%$CI 1.47–1.85) as well as for more common primary care conditions (constipation PR = 1.52 $95\%$CI 1.46–1.57, incontinence PR = 1.52 $95\%$CI 1.44–1.60). The greatest co-morbidity burden was in patients with a myotonic disorder. Approximately 1-in-6 ($17.1\%$) NMD patients had an infection recorded in the preceding year, with the risk of being hospitalised with an infection nearly double (PR = 1.92, $95\%$CI 1.79–2.07) compared to non-NMD patients. ### Conclusion The burden of chronic co-morbidity among patients with NMD is extremely high compared to the general population, and they are also more likely to present in primary and secondary care for acute events such as infections. ## Introduction People with neuromuscular disease (NMD) experience a broad range of health issues related to the progression of their disease, such as reduced mobility impacting quality of life [1], as well as pulmonary issues possibly leading to severe respiratory complications [2]. Additional health problems are also associated with specific types of NMD, such as cardiomyopathy in Duchenne or Becker muscular dystrophy [3], endocrine dysfunction in myotonic dystrophy [4] or dysphagia resulting from inflammatory myopathies [5]. Many other associations have been suggested but are less well established, such as a link between myasthenia gravis and diabetes, potentially related to the increased usage of corticosteroids in these patients [6]. Many observations have historically been based on NMD registries [3], and as a result few direct comparisons with the general population exist. Recent studies have indicated that the combined prevalence of all NMDs may now exceed 100 per 100,000 persons [7], and is rising over time [8]. Previously, we reported on trends in the incidence and prevalence of NMD recorded in UK primary care and showed an increasing burden among older patients [9]. Earlier studies of some NMDs, such as Duchene muscular dystrophy have been based almost exclusively primarily on younger patients, who may be less representative of the overall disease burden in the wider population as the associated life expectancy with the condition has increased over time [10]. As NMD patients are already at a greater risk of falls and fractures from a loss of muscle power over time [1], this risk may become more relevant to an ageing patient group. However, there is an absence of large-scale descriptive studies of older patients with a NMD. Better recognition of older patients with NMD is important, since they are likely to be frequently hospitalised, so better coordinated care might prevent some admissions such as fractures and infections [11]. In this study, we use a large UK primary care database to summarise the chronic diseases, health conditions and recent infections recorded in a group of patients with NMD and compare these directly to a comparator group without NMD, to quantity differences between them. Additionally, we wanted to explore differences by age and type of NMD. ## Data source The Clinical Practice Research Datalink (CPRD) is a primary care database in the UK jointly sponsored by the Medicines and Healthcare products Regulatory Agency and the National Institute for Health and Care Research [12]. For over 30 years researchers have used CPRD data to help inform clinical guidance and best practice. Diagnoses are recorded on CPRD using a hierarchical clinical classification system called Read codes [13], from clinical sources such as hospital discharge summaries or communication from specialists. The database recently expanded due to the inclusion of practices using EMIS software [14] and now includes 16 million currently registered patients. Our analysis includes a total of 1,418 practices actively providing data as of $\frac{1}{1}$/2019 [9]. Additionally, we also used some data from Hospital Episodes Statistics (HES), which has been linked to CPRD patient records [15]. HES is a data source recording every NHS hospital admission in England, including information on clinical diagnoses [16]. HES linkage was available for 1,088 practices in England in our dataset. ## Classifying patients with neuromuscular disease Previously, we classified NMD using a hierarchical classification based on the existence of Read codes anywhere previously in their primary care record [9]. As these represent diagnoses made outside of the primary care setting, generally from specialist settings, it is reasonable to assume most diagnoses are valid. Diagnoses were broadly classified into the following (S1 Table): motor neuron disorders, acquired myopathies (e.g. inflammatory myopathies), hereditary myopathies (including muscular dystrophies), mitochondrial disease, muscle channelopathies, hereditary neuropathies (e.g. Charcot-Marie Tooth disease), inflammatory & autoimmune neuropathies (e.g. Guillain-Barré Syndrome), neuromuscular junction disorders (e.g. myasthenia gravis), plus a non-specific category (“Muscular or neuromuscular disease unspecified”) as some Read codes would not allow clear classification into any other category. For this analysis, we wanted to describe the long-term health in NMD patients, so we excluded patients with motor neurone disease due to the shorter survival time from diagnosis, but still included other motor neuron disorders such as spinal muscular atrophy and post-polio syndrome. We present results for all NMD combined initially, but we also reported findings by the following 6 specific conditions: Charcot-Marie Tooth disease (CMT), Guillain-Barré syndrome (GBS), inflammatory myopathies (IIM), muscular dystrophy (MD), myotonic dystrophy type 1 (DM1) and myasthenia gravis (MG). ## Study cohort and matched non-NMD patients Patients were included in the study if they were actively registered on $\frac{1}{1}$/2019 with their GP and had been so for at least 90 days. We further restricted to NMD patients who had been originally diagnosed at least one year previously. Diagnoses made historically, either at a different practice or pre-computerisation, can be inferred from the record but these will become less reliable the further back in time one goes. Lastly, we only included patients who were at least age 2 on $\frac{1}{1}$/2019 as few outcomes in the study would be present below this age. A total of 23,876 patients with NMD were eligible for the analysis (S1 Fig). Four patients matched on age and sex from the same practice without any history of a NMD and registered for >90 days were selected to be the comparator group in the analysis. A total of 95,295 patients without NMD were randomly selected without replacement. Where the outcome required the patient to be registered with their general practice for 1 year (recording of infections), analyses were restricted to 22,946 NMD patients and 87,959 corresponding non-NMD patients who were registered in CPRD throughout 2018. Finally, analyses that relied on linked HES data (England only), were based on 19,012 NMD patients and 74,831 matched non-NMD patients. ## Defining co-morbidity and infections Our primary focus was describing co-morbidity in NMD patients using a list of conditions routinely collected as part of the Quality and Outcomes Framework (QOF), a UK wide system for performance management and payment of GPs in primary care [17]. Since its introduction in 2004, disease registers for approximately 20 different chronic diseases or conditions have been created and maintained. This has improved data quality and recoding, and we have previously shown how a score based on these conditions is highly predictive of mortality [18]. For the analysis here, we counted the presence of any Read codes for 18 of these conditions (S2 Table) in a patient’s record by 2019, using the published code definitions [17]. Additionally, we also wanted to describe a broader list of health conditions, including many that we would expect to find more commonly in patients with NMD. For this, we created a list of 30 further conditions (S2 Table). The majority of these were selected and adapted from a list of 308 physical and mental health conditions described by Kuan et al [19], who provided a comprehensive summary of recording of prevalence within a subset of CPRD data, including code list definitions. For some conditions we combined some classifications into a broader grouping (cardiomyopathy, uveitis). Finally, we also added constipation and dysphagia to this extended list, due to consensus from primary and secondary care clinician authors on their importance to quality of life in people living with NMD. For infections, we report only on events recorded in the prior year [2018], additionally utilising the linked HES data to distinguish more serious infections. Infections were grouped into 10 categories: cellulitis, eye, gastro-intestinal, genitourinary, lower respiratory tract, mycoses (candidiasis, other fungal), sepsis, skin and upper respiratory tract. For primary care analyses, presence of an infection for each category was indicated by the occurrence of a Read code in 2018, but we also created a summary group for any infection which also required the prescribing of an antibiotic, antifungal or antiviral drug in the 14 day period either side of the diagnosis, an approach we have used previously [20]. For hospitalisations, any new episode where the primary ICD-10 indicated an infection were counted. ## Statistical analysis Our summary measure for all analyses was the estimated Prevalence Ratio (PR) to summarise the presence of the condition ever (or for infections in 2018 only) in NMD patients versus non-NMD patients. We used modified Poisson regression, which fits a model with a robust error-variance correction and has been shown to provide reliable relative risk estimates [21]. To account for the matching, a Generalized Estimating Equation (GEE) approach was used which allows for the statistical dependence within the match-sets. All models were fitted using PROC GENMOD in SAS (Version 9.4). Terms for age and sex are included in the model even though they were matched on, but they have little impact on the prevalence ratio due to the balanced design. We also fitted models stratified by age group (e.g., 18–49, 50–64, 65+ for adult comparisons) as it was likely that the prevalence ratio was not constant by age, such that relative comparisons will become less extreme in older ages as the prevalence of disease rises in the general population. ## Ethics approval This study is based in part on data from the Clinical Practice Research Datalink obtained under licence from the UK Medicines and Healthcare products Regulatory Agency. The protocol (no. 19_211) was approved by the Independent Scientific Advisory Committee evaluation of joint protocols of research involving CPRD data in October 2019. The approval allows analysis of anonymous electronic patient data without the need for written or oral consent. ## Demographics of cohort The mean age of the 23,876 patients with NMD included in the study was 54.3 years, with $53.1\%$ of them recorded as being male (Table 1). Among specific neuromuscular disorders, Guillain-Barré syndrome was most common ever recorded ($20.1\%$), followed by Myasthenia Gravis ($16.2\%$) and Charcot-Marie Tooth disease ($14.7\%$). About $23\%$ were not classified any further (“Other”)–these were a combination of rarer conditions (e.g., neuralgic amyotrophy) or non-specific codes (“Myopathy or muscular dystrophy”). A small number of patients ($$n = 183$$) were classified into multiple NMD categories and appear in the analysis for each group. Patients with NMD were more likely to consult during 2018 with a GP (S3 Table) and were more likely to have had a GP referral for further care ($14.0\%$ vs. $5.8\%$) than non-NMD patients. Table 1 \| Unnamed: 0 \| N(%) \| Mean age (s.d.) \| Median age at diagnosis (IQR) \| Number of patients without NMD(%) ‡ \| \| --- \| --- \| --- \| --- \| --- \| \| All Patients with NMD * \| 23876 \| 54.3 (20.8) \| 39 (21–57) \| 95295 \| \| Age \| \| \| \| \| \| 2 to 17 \| 1,494 (6.3%) \| 11.1 (4.1) \| 4 (1–7) \| 5,967 (6.3%) \| \| 18 to 49 \| 7,412 (31.0%) \| 36.1 (9.0) \| 22 (12–32) \| 29,645 (31.1%) \| \| 50 to 64 \| 6,298 (26.4%) \| 57.0 (4.3) \| 42 (31–50) \| 25,189 (26.4%) \| \| 65+ \| 8,672 (36.3%) \| 75.3 (7.2) \| 61 (50–69) \| 34,494 (36.2%) \| \| Sex \| \| \| \| \| \| Female \| 11,206 (46.9%) \| 55.3 (20.3) \| 39 (22–56) \| 44,768 (47.0%) \| \| Male \| 12,670 (53.1%) \| 53.4 (21.2) \| 39 (19–57) \| 50,527 (53.0%) \| \| Neuromuscular disorder† \| \| \| \| \| \| Charcot-Marie Tooth \| 3,511 (14.6%) \| 51.2 (20.7) \| 35 (15–53) \| 14,029 (14.6%) \| \| Guillain-Barré syndrome \| 4,791 (19.9%) \| 57.9 (18.4) \| 40 (24–57) \| 19,126 (19.9%) \| \| Inflammatory myopathies \| 2,816 (11.7%) \| 57.9 (18.6) \| 45 (28–58) \| 11,248 (11.7%) \| \| Muscular dystrophy \| 2,711 (11.3%) \| 45.3 (22.3) \| 25 (7–45) \| 10.832 (11.3%) \| \| Myotonic dystrophy (Type 1) \| 851 (3.5%) \| 46.3 (16.6) \| 30 (18–44) \| 3,384 (3.5%) \| \| Myasthenia Gravis \| 3,866 (16.1%) \| 64.3 (17.2) \| 55 (35–67) \| 15,398 (16.0%) \| \| Other \| 5,519 (22.9%) \| 50.3 (22.0) \| 36 (17–53) \| 22,034 (22.9%) \| \| Country \| \| \| \| \| \| England \| 19,735 (82.7%) \| 54.2 (20.9) \| 39 (20–57) \| 78,786 (82.7%) \| \| Northern Ireland \| 391 (1.6%) \| 55.4 (19.7) \| 42 (23–60) \| 1,563 (1.6%) \| \| Scotland \| 2,176 (9.1%) \| 54.1 (20.1) \| 38 (20–56) \| 8,676 (9.1%) \| \| Wales \| 1,574 (6.6%) \| 55.3 (20.7) \| 41 (22–57) \| 6,270 (6.6%) \| ## Prevalence of chronic disease and health conditions in adults Among the 18 chronic conditions that were recorded in the QOF (Table 2), 16 of them were significantly higher among patients with NMD (e.g., lower $95\%$ CI for PR was >1). Only serious mental health disorders (e.g., psychosis, schizophrenia, bipolar disorder) and dementia did not show an increased prevalence. The largest relative associations among all NMD patients were seen for learning disability (PR = 2.82), rheumatoid arthritis (PR = 1.94) and osteoporosis (PR = 1.86). Osteoporosis was over 10 times more likely to have been recorded among 18–49-year-olds, as it was extremely rare among non-NMD patients at this age. Similarly in this age group, heart failure (PR = 11.65) and atrial fibrillation (PR = 3.77) produced large prevalence ratios compared to those seen in older age groups. Almost 1-in-4 of NMD patients ($24.4\%$) had ever received a diagnosis of depression, which was $24\%$ higher than the general population and remained constant across age groups. When we summarised multi-morbidity by adding up the total number of QOF conditions ever recorded, approximately 4-in-10 patients with NMD had 2 or more conditions ($25\%$ higher than general population), and 1-in-10 had 4 or more ($39\%$ higher). Table 2 \| Condition \| ALL Adults \| ALL Adults.1 \| Age 18–49 \| Age 18–49.1 \| Age 50–64 \| Age 50–64.1 \| Age 65- \| Age 65-.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| % \| PR (95% CI) \| % \| PR (95% CI) \| % \| PR (95% CI) \| % \| PR (95% CI) \| \| Atrial Fibrillation \| 6.2% \| 1.28 (1.21,1.36) \| 0.9% \| 3.77 (2.69,5.28) \| 3.1% \| 1.96 (1.66,2.32) \| 13.0% \| 1.17 (1.10,1.24) \| \| Asthma \| 16.7% \| 1.21 (1.17,1.25) \| 19.0% \| 1.17 (1.11,1.23) \| 16.0% \| 1.23 (1.15,1.31) \| 15.3% \| 1.25 (1.18,1.32) \| \| Cancer (excl. non-melanoma skin) \| 8.9% \| 1.21 (1.15,1.26) \| 1.9% \| 1.72 (1.41,2.09) \| 6.0% \| 1.22 (1.09,1.37) \| 17.0% \| 1.17 (1.11,1.24) \| \| Coronary Heart Disease \| 8.2% \| 1.21 (1.15,1.27) \| 0.6% \| 1.89 (1.32,2.71) \| 5.3% \| 1.42 (1.26,1.61) \| 16.9% \| 1.16 (1.10,1.22) \| \| Chronic Kidney Disease \| 8.1% \| 1.08 (1.04,1.14) \| 0.6% \| 2.39 (1.67,3.42) \| 3.2% \| 1.43 (1.22,1.67) \| 18.1% \| 1.04 (0.99,1.09) \| \| COPD \| 4.6% \| 1.11 (1.04,1.19) \| 0.5% \| 1.60 (1.08,2.37) \| 3.6% \| 1.21 (1.05,1.40) \| 8.8% \| 1.07 (1.00,1.16) \| \| Dementia \| 1.5% \| 0.86 (0.77,0.97) \| 0.0% \| 2.67 (0.45,15.96) \| 0.1% \| 0.85 (0.38,1.92) \| 3.8% \| 0.86 (0.77,0.97) \| \| Depression \| 24.4% \| 1.24 (1.20,1.27) \| 22.7% \| 1.25 (1.19,1.31) \| 29.4% \| 1.23 (1.18,1.29) \| 22.2% \| 1.23 (1.17,1.28) \| \| Diabetes \| 13.2% \| 1.29 (1.24,1.34) \| 4.1% \| 1.88 (1.65,2.15) \| 12.8% \| 1.37 (1.28,1.48) \| 21.4% \| 1.20 (1.14,1.25) \| \| Epilepsy \| 2.7% \| 1.72 (1.56,1.88) \| 3.5% \| 2.47 (2.12,2.88) \| 2.5% \| 1.49 (1.25,1.79) \| 2.2% \| 1.32 (1.13,1.56) \| \| Heart Failure \| 3.2% \| 1.70 (1.56,1.85) \| 1.3% \| 11.65 (7.91,17.14) \| 1.5% \| 2.34 (1.82,3.00) \| 5.9% \| 1.39 (1.26,1.53) \| \| Hypertension \| 29.8% \| 1.09 (1.07,1.12) \| 5.6% \| 1.49 (1.34,1.66) \| 25.4% \| 1.15 (1.10,1.21) \| 53.7% \| 1.05 (1.03,1.08) \| \| Learning Disability \| 1.3% \| 2.82 (2.42,3.28) \| 2.9% \| 5.36 (4.38,6.56) \| 0.8% \| 1.48 (1.08,2.03) \| 0.2% \| 0.59 (0.34,1.03) \| \| Mental Health (Psychosis, schizophrenia, bipolar) \| 1.3% \| 1.06 (0.93,1.21) \| 1.3% \| 1.18 (0.94,1.48) \| 1.5% \| 1.08 (0.86,1.36) \| 1.1% \| 0.95 (0.76,1.18) \| \| Osteoporosis \| 6.8% \| 1.86 (1.76,1.97) \| 1.7% \| 10.98 (7.85,15.35) \| 4.5% \| 2.86 (2.47,3.31) \| 12.8% \| 1.57 (1.48,1.67) \| \| Peripheral arterial disease \| 1.7% \| 1.23 (1.10,1.38) \| 0.1% \| 2.67 (1.09,6.52) \| 1.0% \| 1.74 (1.30,2.32) \| 3.6% \| 1.15 (1.02,1.30) \| \| Rheumatoid Arthritis \| 2.1% \| 1.94 (1.74,2.16) \| 0.6% \| 2.44 (1.70,3.50) \| 2.1% \| 2.29 (1.86,2.83) \| 3.4% \| 1.76 (1.54,2.01) \| \| Stroke (including TIA) \| 5.1% \| 1.31 (1.23,1.40) \| 0.6% \| 2.47 (1.72,3.57) \| 3.3% \| 1.74 (1.48,2.04) \| 10.3% \| 1.21 (1.13,1.30) \| \| 2 or more QOF conditions \| 38.7% \| 1.25 (1.23,1.27) \| 14.9% \| 1.75 (1.64,1.87) \| 33.3% \| 1.37 (1.32,1.43) \| 63.1% \| 1.14 (1.12,1.16) \| \| 4 or more QOF conditions \| 10.3% \| 1.39 (1.33,1.45) \| 0.7% \| 3.04 (2.14,4.32) \| 5.8% \| 2.05 (1.82,2.32) \| 21.8% \| 1.29 (1.24,1.35) \| Among the other non-QOF conditions we investigated (S4 Table), the recorded prevalence of rarer conditions such as cardiomyopathy (PR = 4.44), scoliosis (PR = 3.44) and aspiration pneumonitis (3.42) were higher in NMD patients compared to non-NMD patients, as expected. Venous thromboembolism (VTE), either with or without pulmonary embolism, was almost twice as likely to have been recorded, rising to three times higher in those under age 50. More common conditions (constipation, cataract, dysphagia, incontinence, post-viral fatigue syndrome) were all more than $50\%$ higher in NMD patients. ## Prevalence of childhood diseases and conditions We investigated 12 conditions that were recorded in the children with NMD in our study (S5 Table). Both visual impairment and a history of dysphagia were over 6 times more likely compared to the general population, while sleep apnoea was 4 times more likely. While asthma was higher among adults with NMD, no such association existed among children (PR = 1.00). ## Co-morbidity by type of neuromuscular disease We repeated the analysis for all different chronic diseases and conditions for the 6 different common NMD groups we investigated. These are summarised in Table 3 and listed according to their relative associations with the general population using the prevalence ratio. The complete set of results for the 18 QOF conditions (S6 Table) and 30 non-QOF (S7 Table) are available in the supplementary material. Table 3 \| Unnamed: 0 \| Charcot-Marie Tooth \| Guillain-Barré syndrome \| Inflammatory myopathies \| Muscular dystrophy \| Myotonic dystrophy (Type 1) \| Myasthenia Gravis \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| >5 times as likely \| Scoliosis \| Multiple sclerosis \| Aspiration pneumonitis \| CardiomyopathyScoliosisAspiration pneumonitis \| CardiomyopathyAspiration pneumonitisLearning DisabilityCataractSleep apnoeaHeart FailureAtrial FibrillationVisual impairmentDysphagiaAutism/Asperger’sPulmonary embolism \| \| \| >3 times as likely \| Diabetic NeuropathySleep apnoeaLearning Disability \| \| Rheumatoid Arthritis \| Heart FailureFracture of hipCollapsed vertebraSleep apnoeaLearning Disability \| Macular degenerationScoliosisSkin cancer‡Diabetic Neuropathy \| Aspiration pneumonitis \| \| >2 times as likely \| Multiple sclerosisCollapsed vertebraFracture of hipEpilepsy \| Diabetic NeuropathyPulmonary embolism \| CardiomyopathyDysphagiaSleep apnoeaPVFSPulmonary embolismVTE disease†OsteoporosisCollapsed vertebra \| DysphagiaOsteoporosisDiabetic NeuropathyVisual impairment \| Collapsed vertebraPADConstipationUrinary IncontinenceVTE disease†IBS \| DysphagiaSleep apnoeaMultiple sclerosisPulmonary embolism \| \| >50% higher & >10% prevalence \| ConstipationHearing LossErectile dysfunction \| \| HypothyroidismSpondylosisCataract \| Constipation \| Hearing Loss \| Hypothyroidism Urinary Incontinence \| \| >20% higher & >20% prevalence \| Osteoarthritis* DepressionAnxiety disorders \| \| Erectile dysfunctionOsteoarthritis \| \| \| Depression \| Patients with CMT not only had a high burden of recorded co-morbidity, but a significantly higher reporting of common conditions impacting quality of life (e.g., constipation, hearing loss, erectile dysfunction, urinary incontinence) than the general population. A history of depression and/or an anxiety disorder was also highest among patients with CMT, with the prevalence higher than the general population (PR = 1.34 depression, PR = 1.21 anxiety). Approximately 1-in-10 CMT patients have received a new depression diagnosis in the last 5 years ($$n = 355$$, $10.1\%$). Diabetic neuropathies were highest in patients with CMT ($1.5\%$, PR = 4.43). To reduce the possibility of misdiagnosis around the same time, we excluded cases with a code for diabetic neuropathy +/- 1 year of their initial CMT diagnosis, but the PR remained high (3.61). Compared to the other NMDs, patients who have had a prior history of GBS had lower overall co-morbidity for their age, and smaller relative associations for most conditions when compared to their matched non-NMD patients. The anomaly was Multiple Sclerosis (MS) where $1.8\%$ of GBS patients in our study also had a MS diagnosis (PR = 5.59). To try and discount misdiagnosis as explanation, we excluded all match-sets where the GBS case had a first MS diagnosis +/- 1 year of their initial GBS diagnosis, but the PR remained high (4.12). Additionally excluding all GBS cases who had MS at the time of diagnosis in their record still produced an elevated PR (3.16). Patients with a history of IIM were far more likely to have a range of conditions and complications recorded compared to non-NMD patients, particularly aspiration pneumonia (PR = 8.68) and rheumatoid arthritis (PR = 3.64). Also notable was a history of cancer (PR = 1.44 excluding non-melanoma skin, PR = 1.36 for non-melanoma skin), and diagnoses of coronary heart disease (PR = 1.59) and diabetes (PR = 1.44), which produced higher PRs than for other NMDs. Patients with myotonic dystrophy type 1 (DM1) had the greatest burden of co-morbidities with 21 different conditions being more than twice as likely to be recorded ever than their matched non-NMD patients. Cardiomyopathy, aspiration pneumonia, learning disability and cataract were all greater than 10 times more likely. An association that appeared specific to DM1 was with non-melanoma skin cancer (PR = 3.31). Diseases of the eye, such as cataracts (PR = 10.93), and circulatory system such atrial fibrillation (PR = 7.59) were particularly raised in DM1 patients compared to other NMDs. Almost 1-in-10 ($9.3\%$) had a co-occurring learning disability, far higher than for any other NMD. While other muscular dystrophies showed a similar pattern for many of these conditions, in general cardiovascular and eye diseases were lower, while musculoskeletal conditions such as hip fractures (PR = 3.92) were higher. Approximately half ($49.8\%$) of patients with a MG diagnosis had 2 or more QOF conditions. Almost 1-in-4 ($23.5\%$) patients had a diagnosis of diabetes, more the non-NMD group (PR = 1.47). A history of asthma was also noticeably higher in these patients (PR = 1.33). Among other conditions, other raised associations included aspiration pneumonia (PR = 3.45) and dysphagia (PR = 2.92). ## History of recent infection Table 4 summarises the recording of infections in primary care and hospital admissions for an infection during 2018 among NMD patients (now including children). Among all patients, those with a history of NMD were $43\%$ more likely (PR = 1.43, $95\%$ CI 1.38–1.48) to have had any infection recorded in primary care, affecting 1-in-6 NMD patients ($17.1\%$). A higher risk of infection was seen in all infection categories. When only hospitalisations were counted, the increased risk among NMD patients was now almost a doubling (PR = 1.92, $95\%$CI 1.79–2.07), with sepsis showing the largest association (PR = 2.37). In both healthcare settings, lower respiratory tract infections were raised among NMD patients, especially among children (PR = 3.63 primary care, PR = 15.1 hospital admissions). Table 4 \| Unnamed: 0 \| ALL Patients \| ALL Patients.1 \| Age 2–17 \| Age 2–17.1 \| Age 18–64 \| Age 18–64.1 \| Age 65- \| Age 65-.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| % \| PR (95% CI) \| % \| PR (95% CI) \| % \| PR (95% CI) \| % \| PR (95% CI) \| \| Recorded in primary care * \| \| \| \| \| \| \| \| \| \| • Any plus prescription \| 17.1% \| 1.43 (1.38,1.48) \| 17.1% \| 1.81 (1.57,2.08) \| 15.1% \| 1.49 (1.42,1.56) \| 20.1% \| 1.33 (1.27,1.40) \| \| • Cellulitis \| 2.3% \| 1.65 (1.49,1.82) \| 0.6% \| 2.18 (0.92,5.19) \| 1.4% \| 1.78 (1.49,2.12) \| 4.1% \| 1.57 (1.39,1.78) \| \| • Eye \| 1.2% \| 1.37 (1.20,1.57) \| 1.8% \| 1.62 (1.03,2.56) \| 0.9% \| 1.33 (1.09,1.63) \| 1.6% \| 1.36 (1.12,1.65) \| \| • Gastro-Intestinal Tract \| 0.8% \| 1.36 (1.16,1.61) \| 0.7% \| 1.09 (0.54,2.20) \| 0.8% \| 1.51 (1.22,1.88) \| 0.8% \| 1.21 (0.92,1.58) \| \| • Genito-Urinary \| 3.3% \| 1.46 (1.34,1.58) \| 0.6% \| 1.08 (0.52,2.22) \| 2.6% \| 1.51 (1.34,1.72) \| 4.8% \| 1.41 (1.26,1.58) \| \| • Lower Respiratory Tract \| 6.8% \| 1.61 (1.52,1.70) \| 5.5% \| 3.63 (2.70,4.87) \| 5.2% \| 1.89 (1.72,2.06) \| 9.4% \| 1.36 (1.26,1.47) \| \| • Mycoses—Candidiasis \| 1.2% \| 1.60 (1.39,1.83) \| 0.7% \| 3.17 (1.37,7.33) \| 1.2% \| 1.52 (1.27,1.83) \| 1.3% \| 1.62 (1.30,2.02) \| \| • Mycoses—Other Fungal \| 1.7% \| 1.31 (1.17,1.47) \| 1.3% \| 1.28 (0.75,2.18) \| 1.7% \| 1.38 (1.18,1.61) \| 1.8% \| 1.23 (1.03,1.47) \| \| • Skin (Other) \| 5.0% \| 1.37 (1.28,1.46) \| 5.8% \| 1.43 (1.12,1.83) \| 4.9% \| 1.48 (1.35,1.62) \| 4.9% \| 1.22 (1.09,1.35) \| \| • Upper Respiratory Tract (Other) \| 7.4% \| 1.23 (1.16,1.29) \| 15.3% \| 1.28 (1.11,1.47) \| 7.8% \| 1.24 (1.16,1.33) \| 5.5% \| 1.18 (1.06,1.30) \| \| Hospitalisations † \| \| \| \| \| \| \| \| \| \| • Any \| 5.3% \| 1.92 (1.79,2.07) \| 7.5% \| 5.82 (4.28,7.90) \| 3.9% \| 2.47 (2.19,2.78) \| 7.2% \| 1.46 (1.32,1.61) \| \| • Gastro-Intestinal Tract \| 0.9% \| 2.02 (1.69,2.42) \| 1.7% \| 11.90 (5.22,27.12) \| 0.8% \| 2.30 (1.77,2.98) \| 1.0% \| 1.41 (1.07,1.86) \| \| • Lower Respiratory Tract \| 2.2% \| 2.00 (1.78,2.25) \| 3.5% \| 15.14 (7.80,29.40) \| 1.3% \| 3.61 (2.87,4.53) \| 3.3% \| 1.39 (1.19,1.61) \| \| • Sepsis \| 0.7% \| 2.37 (1.91,2.94) \| 0.2% \| 7.95 (0.72,87.48) \| 0.5% \| 3.86 (2.61,5.70) \| 1.1% \| 1.86 (1.42,2.43) \| When the associations were explored by NMD (S8 Table), the most elevated estimates were seen for Myotonic Dystrophy with patients $80\%$ more likely to have had any infection in primary care (PR = 1.80, $95\%$ CI 1.52–2.13) and any hospitalisation (PR = 2.74, $95\%$CI 1.85–4.08). This was true across for most infection types except for upper respiratory tract which showed no association. Fungal infections tended to be more common among patients with a history of inflammatory myopathy or myasthenia gravis. ## Discussion We have used a large UK primary care database to quantify the higher overall burden of recorded chronic diseases, general health conditions and recent infections in NMD patients directly compared to an age-sex-practice matched sample from the general population. To our knowledge, this is a novel comparison and provides a broad overview by age and NMD as to the increased disease burden in these patients. The main strength of our study is the size, containing over 20,000 patients with recorded NMD from a nationwide sample of general practices in the UK. The CPRD database, at the time of our study date (January 1st 2019), contained approximately 12 million registered patients representing almost $20\%$ of the UK population [9]. So, the results are likely to be generalisable in terms of what is being recorded on primary care medical records across the UK. However, there are several limitations to our analyses. Firstly, we have not attempted to validate the diagnosis of NMD in our study as we are assuming that any Read codes used for these rare conditions represent diagnoses that have been made in a specialist setting outside of primary care and then transferred to the patients’ GP record [9]. So, while it is possible that some of the patients may have been mis-diagnosed or mis-classified, that would lead to our analysis underestimating the elevated associations we described. An exception here was for GBS and MS where it appears some diagnoses were made closely in time, and the validity of the GBS diagnosis could be queried, as well as acknowledging that historical dates of diagnosis will become less reliable the further back in time they were made. However, the higher finding of MS recorded in patients who have also had GBS still persisted even after excluding these patients. Secondly, it may be that some of the health conditions we included here are not consistently recorded on GP systems as they are diagnosed outside of primary care (e.g., eye diseases) or they are not always going to result in primary care contact (e.g., constipation). For example, while we found a high prevalence ratio between scoliosis and CMT, only $6.7\%$ of the CMT patients in our study had this recorded, less than the $15\%$ reported in a study of younger CMT patients [22]. A US study of constipation in DMD patients found higher rates than we did, but also reported that less than half were receiving treatment suggesting the condition could be underdiagnosed [23]. Consistency of coding and recording was why we primarily focused on the chronic diseases collated by the QOF. For conditions less consistently recorded, our analysis would be biased if patients with NMD were more likely to be seen and assessed in primary care, which we showed to be the case at all ages during 2018. Thirdly, an important limitation of our approach is that it is essentially cross-sectional in nature (using a census date of 1st January 2019) and is not exploring the implied future risk of any of these conditions. We have not attempted to disentangle the date ordering of diagnoses and events, and many of the diseases and conditions we reported on would have been recorded before the patient was diagnosed with a NMD, especially as some may be presenting symptoms prior to the initial diagnosis itself. Analyses using CPRD which have explored outcomes post-diagnosis in more common conditions such as chronic inflammatory disorders [24], have shown comparable increases in risk ($16\%$) for depression and anxiety events as we found using our approach here. Finally, patients with NMD in CPRD who died in 2018 from a complication of their disease are not included in our comparison. So, one might expect that any associations with a health condition associated with short-term mortality such as venous thromboembolism, stroke or sepsis, would be underestimated. For example, while we reported on a large relative association of aspiration pneumonitis ever being recorded in patients for some NMDs, it may still not represent the true risk for this patient group. Despite these limitations, we think our study provides a broad overview of the overall disease burden encountered by patients with NMD. Among the list of other chronic diseases routinely recorded in UK primary care, almost all were significantly higher with dementia and severe mental illness the only exceptions. Conditions such as depression and diabetes were both relatively common and significantly more likely to have been recorded in patients with NMD. Previous studies have linked CMT to depressive symptoms [25], and MG to diabetes [6]. Since we have previously reported a more than doubling in the prevalence of both recorded CMT and MG in the UK during 2000–19 [9], the number of potential patients with these conditions too will also be increasing. We observed that the recording of diabetic neuropathies was also much higher in patients with NMD, particularly CMT. An advantage of our analysis was that we were able to stratify comparisons by age and type of NMD, where more meaningful comparisons can be made. In younger adults, the differences are more marked between NMD patents and the general population, where many conditions are rare. Among the different types of NMD we investigated, it was clear that patients with type 1 myotonic dystrophy (DM1) had the greatest disease burden; previous population-based analysis specific to DM1 have demonstrated the level of co-morbidities [26]. In adults with DM1 the frequency of different symptoms has been reported to vary according to age of onset and clinical subtype [27]. Many of the conditions we found with elevated associations with DM1 have been documented in two recent reviews [26,27]. There were two other findings that appeared specific to DM1. We found a higher than expected recording of non-melanoma skin cancer, which mirrors a previous study of DM1 patients using CPRD, that found a higher risk of developing basal cell carcinoma over time [28]. Also of note was the significantly lower prevalence of hypertension compared to non-NMD patients and other NMD, which would back up a historical finding that hypotension was a clinical feature of myotonic dystrophy [29]. Some diseases have not been widely reported in patients with NMD previously. We found higher than expected rates of venous thromboembolism within each NMD, even among the patients with historical GBS diagnoses, many of whom may have recovered over time. The prevalence of DVT in patients with NMD could be presumed to be higher because patients typically have reduced physical activity and may adopt a more sedentary lifestyle [30]. While DVT has been reported as an important cause of mortality in patients with amyotrophic lateral sclerosis and Parkinson disease, there has been limited reporting with other NMD [31]. Although we queried the co-existence of the MS and GBS diagnoses, a case-control study has shown an association between GBS and prior infections such as Epstein-Barr virus [32], which is also thought to be a risk factor for MS [33]; so a more forensic analysis, ideally studied prospectively, is necessary here to understand our finding further. The associations we observed with osteoporosis and a recorded hip fracture are not surprising given that NMD patients often suffer from nutritional issues impacting bone health, in addition to low levels of physical activity [1]. Falls have also been reported in post-GBS patients, with over half the respondents in a recent survey reporting a fall in the last year [34]. The increase in risk we estimated was quite marked in younger adults compared to the generation population, suggesting that fall prevention methods when developing care plans should be assessed for all adults not just older patients [35]. It has also been advocated that clinicians should consider the administration of anti-osteoporotic medications such as bisphosphonates to prevent fragility fractures due to the prolonged use of glucocorticoids over time [1]. However it is worth noting for MG that neither a previous study using CPRD [36], nor large studies from Canada [37] and Denmark [38] found an increased risk of fracture among MG patients, even when they restricted to those who received high-dose oral glucocorticoids [36,38]. The most novel finding from our analysis, and potentially the most important, may be the consistently higher risk of recent infection in patients in NMD. Since we only included patients with NMD diagnosed prior to 2018, this analysis is based on recorded infections occurring post-diagnosis. So while gastrointestinal and respiratory tract infections have been shown to be associated with an increased risk of developing a IIM [39], our analysis suggests that infection risk may be present both pre- and post-diagnosis for some autoimmune disorders. The higher rates of infection in NMD patients were seen in primary care for common respiratory and skin infections as well as rarer fungal infections. We were able to utilise linked HES data to show that the increased risk for patients with NMD was almost a doubling for hospital admissions for infection, such as sepsis. Many of these admissions are where prevention or effective management in primary care could have decreased the risk of acute hospitalisation [40], so identifying ways to improve surveillance among this group of patients at higher risk could potentially reduce unplanned hospital admissions. Whilst individually rare, neuromuscular conditions are collectively relatively common with a population prevalence similar to that of Parkinson’s disease or multiple sclerosis [9]. The high levels of medical co-morbidity in patients with neuromuscular conditions highlight the important role of general practitioners in the care of this group of conditions, in terms of recognising and managing treatable co-morbidities and infections which may have a significant effect on quality of life. Case management approaches to linking primary care physicians and community services with specialist neuromuscular services may support care by raising awareness of the spectrum of associated comorbidities in this population and supporting them with early identification of disorder-specific comorbidities. ## Conclusion We have provided a broad overview of the level of co-morbidity of disease and health conditions experienced by patients with NMD, confirming most well observed associations but also highlighting some less well documented ones, particularly around recent infection. ## References 1. Iolascon G, Paoletta M, Liguori S, Curci C, Moretti A. Neuromuscular Diseases and Bone. (2019) 10. DOI: 10.3389/fendo.2019.00794 2. Benditt JO, Boitano LJ. Pulmonary Issues in Patients with Chronic Neuromuscular Disease. (2013) 187 1046-55. DOI: 10.1164/rccm.201210-1804CI 3. Connuck DM, Sleeper LA, Colan SD. Characteristics and outcomes of cardiomyopathy in children with Duchenne or Becker muscular dystrophy: A comparative study from the Pediatric Cardiomyopathy Registry. (2008) 155 998-1005. DOI: 10.1016/j.ahj.2008.01.018 4. Winters SJ. Endocrine Dysfunction in Patients With Myotonic Dystrophy.. (2021) 106 2819-27 5. Zeng R, Schmidt J. Impact and Management of Dysphagia in Inflammatory Myopathies. (2020) 22. DOI: 10.1007/s11926-020-00950-3 6. Yeh JH, Chen HJ, Lin CC. Risk of diabetes mellitus among patients with myasthenia gravis. (2015) 132 132-38. DOI: 10.1111/ane.12374 7. Deenen JCW, Horlings CGC, Verschuuren JJGM, Verbeek ALM, van Engelen BGM. The Epidemiology of Neuromuscular Disorders: A Comprehensive Overview of the Literature. (2015) 2 73-85. PMID: 28198707 8. Rose L, McKim D, Leasa D. Trends in incidence, prevalence, and mortality of neuromuscular disease in Ontario, Canada: A population-based retrospective cohort study (2003–2014). (2019) 14 9. Carey IM, Banchoff E, Nirmalananthan N. Prevalence and incidence of neuromuscular conditions in the UK between 2000 and 2019: A retrospective study using primary care data. (2021) 16. DOI: 10.1371/journal.pone.0261983 10. Ryder S, Leadley RM, Armstrong N. The burden, epidemiology, costs and treatment for Duchenne muscular dystrophy: an evidence review. (2017) 12. DOI: 10.1186/s13023-017-0631-3 11. Scalco RS, Quinlivan RM, Nastasi L, Jaffer F, Hanna MG. Improving specialised care for neuromuscular patients reduces the frequency of preventable emergency hospital admissions. (2020) 30 173-79. DOI: 10.1016/j.nmd.2019.11.013 12. Herrett E, Gallagher AM, Bhaskaran K. Data Resource Profile: Clinical Practice Research Datalink (CPRD). (2015) 44 827-36. DOI: 10.1093/ije/dyv098 13. Chisholm J.. The Read clinical classification. (1990) 300 1092. DOI: 10.1136/bmj.300.6732.1092 14. Wolf A, Dedman D, Campbell J. Data resource profile: Clinical Practice Research Datalink (CPRD) Aurum. (2019) 48 1740-40g. DOI: 10.1093/ije/dyz034 15. Padmanabhan S, Carty L, Cameron E. Approach to record linkage of primary care data from Clinical Practice Research Datalink to other health-related patient data: overview and implications. (2019) 34 91-99. DOI: 10.1007/s10654-018-0442-4 16. Herbert A, Wijlaars L, Zylbersztejn A, Cromwell D, Hardelid P. Data Resource Profile: Hospital Episode Statistics Admitted Patient Care (HES APC). (2017) 46 1093-+. DOI: 10.1093/ije/dyx015 17. Digital NHS. Quality and Outcomes Framework (2022) 18. Carey IM, Shah SM, Harris T, DeWilde S, Cook DG. A new simple primary care morbidity score predicted mortality and better explains between practice variations than the Charlson index. (2013) 66 436-44. DOI: 10.1016/j.jclinepi.2012.10.012 19. Kuan V, Denaxas S, Gonzalez-Izquierdo A. A chronological map of 308 physical and mental health conditions from 4 million individuals in the English National Health Service. (2019) 1 e63-e77. DOI: 10.1016/S2589-7500(19)30012-3 20. Carey IM, Critchley JA, DeWilde S. Risk of Infection in Type 1 and Type 2 Diabetes Compared With the General Population: A Matched Cohort Study. (2018) 41 513-21. DOI: 10.2337/dc17-2131 21. Zou GY. A modified Poisson regression approach to prospective studies with binary data. (2004) 159 702-06. DOI: 10.1093/aje/kwh090 22. Karol LA, Elerson E. Scoliosis in patients with Charcot-Marie-Tooth disease. (2007) 89 1504-10. DOI: 10.2106/JBJS.F.01161 23. Kraus D, Wong BL, Horn PS, Kaul A. Constipation in Duchenne Muscular Dystrophy: Prevalence, Diagnosis, and Treatment. (2016) 171 183-88. DOI: 10.1016/j.jpeds.2015.12.046 24. Dregan A, Matcham F, Harber-Aschan L. Common mental disorders within chronic inflammatory disorders: a primary care database prospective investigation. (2019) 78 688-95. DOI: 10.1136/annrheumdis-2018-214676 25. Cordeiro JLC, Marques W, Hallak JEC, Osorio FL. Charcot-Marie-Tooth disease, psychiatric indicators and quality of life: a systematic review. (2014) 6 8. DOI: 10.1042/AN20130048 26. Fisette-Paulhus I, Gagnon C, Girard-Cote L, Morin M. Genitourinary and lower gastrointestinal conditions in patients with myotonic dystrophy type 1: A systematic review of evidence and implications for clinical practice. (2022) 32 361-76. DOI: 10.1016/j.nmd.2022.01.008 27. De Antonio M, Dogan C, Hamroun D. Unravelling the myotonic dystrophy type 1 clinical spectrum: A systematic registry-based study with implications for disease classification. (2016) 172 572-80. DOI: 10.1016/j.neurol.2016.08.003 28. Wang YJ, Pfeiffer RM, Alsaggaf R. Risk of skin cancer among patients with myotonic dystrophy type 1 based on primary care physician data from the UK Clinical Practice Research Datalink. (2018) 142 1174-81. PMID: 29114849 29. O’Brien T, Harper PS, Newcombe RG. Blood pressure and myotonic dystrophy. (1983) 23 422-6. DOI: 10.1111/j.1399-0004.1983.tb01976.x 30. McDonald CM. Physical activity, health impairments, and disability in neuromuscular disease. (2002) 81 S108-S20. DOI: 10.1097/00002060-200211001-00012 31. Yamada H, Nishikawa T, Yamasaki M. Deep vein thrombosis in patients with neuromuscular disease who undergo tracheotomy with positive pressure ventilation. (2021) 9 63-67 32. Jacobs BC, Rothbarth PH, van der Meche FGA. The spectrum of antecedent infections in Guillain-Barre syndrome—A case-control study. (1998) 51 1110-15. PMID: 9781538 33. Ascherio A, Munger KL. Environmental risk factors for multiple sclerosis. Part I: The role of infection. (2007) 61 288-99. DOI: 10.1002/ana.21117 34. Davidson I, Parker ZJ. Falls in people post-Guillain-Barre syndrome in the United Kingdom: A national cross-sectional survey of community based adults. (2022) 35. Jimenez-Moreno AC, Raaphorst J, Babacic H. Falls and resulting fractures in Myotonic Dystrophy: Results from a multinational retrospective survey. (2018) 28 229-35. DOI: 10.1016/j.nmd.2017.12.010 36. Pouwels S, de Boer A, Javaid MK. Fracture rate in patients with myasthenia gravis: the general practice research database. (2013) 24 467-76. DOI: 10.1007/s00198-012-1970-5 37. Kassardjian C, Widdifield J, Paterson JM. Fracture Risk in Patients with Myasthenia Gravis: A Population-Based Cohort Study. (2021) 8 625-32. DOI: 10.3233/JND-200612 38. Safipour Z, van Der Zanden R, van den Bergh J. The use of oral glucocorticoids and the risk of major osteoporotic fracture in patients with myasthenia gravis. (2022) 33 649-58. DOI: 10.1007/s00198-021-06101-3 39. Svensson J, Holmqvist M, Lundberg IE, Arkema EV. Infections and respiratory tract disease as risk factors for idiopathic inflammatory myopathies: a population-based case-control study. (2017) 76 1803-08. DOI: 10.1136/annrheumdis-2017-211174 40. Bardsley M, Blunt I, Davies S, Dixon J. Is secondary preventive care improving? Observational study of 10-year trends in emergency admissions for conditions amenable to ambulatory care. (2013) 3. DOI: 10.1136/bmjopen-2012-002007
--- title: Predicting the development of T1D and identifying its Key Performance Indicators in children; a case-control study in Saudi Arabia authors: - Ahood Alazwari - Alice Johnstone - Laleh Tafakori - Mali Abdollahian - Ahmed M. AlEidan - Khalid Alfuhigi - Mazen M. Alghofialy - Abdulhameed A. Albunyan - Hawra Al Abbad - Maryam H. AlEssa - Abdulaziz K. H. Alareefy - Mohammad A. Alshamrani journal: PLOS ONE year: 2023 pmcid: PMC9977054 doi: 10.1371/journal.pone.0282426 license: CC BY 4.0 --- # Predicting the development of T1D and identifying its Key Performance Indicators in children; a case-control study in Saudi Arabia ## Abstract The increasing incidence of type 1 diabetes (T1D) in children is a growing global concern. It is known that genetic and environmental factors contribute to childhood T1D. An optimal model to predict the development of T1D in children using Key Performance Indicators (KPIs) would aid medical practitioners in developing intervention plans. This paper for the first time has built a model to predict the risk of developing T1D and identify its significant KPIs in children aged [0-14] in Saudi Arabia. Machine learning methods, namely Logistic Regression, Random Forest, Support Vector Machine, Naive Bayes, and Artificial Neural Network have been utilised and compared for their relative performance. Analyses were performed in a population-based case-control study from three Saudi Arabian regions. The dataset ($$n = 1$$,142) contained demographic and socioeconomic status, genetic and disease history, nutrition history, obstetric history, and maternal characteristics. The comparison between case and control groups showed that most children (cases = $68\%$ and controls = $88\%$) are from urban areas, $69\%$ (cases) and $66\%$ (control) were delivered after a full-term pregnancy and $31\%$ of cases group were delivered by caesarean, which was higher than the controls (χ2 = 4.12, P-value = 0.042). Models were built using all available environmental and family history factors. The efficacy of models was evaluated using Area Under the Curve, Sensitivity, F Score and Precision. Full logistic regression outperformed other models with Accuracy = 0.77, Sensitivity, F Score and Precision of 0.70, and AUC = 0.83. The most significant KPIs were early exposure to cow’s milk (OR = 2.92, $$P \leq 0.000$$), birth weight >4 Kg (OR = 3.11, $$P \leq 0.007$$), residency(rural) (OR = 3.74, $$P \leq 0.000$$), family history (first and second degree), and maternal age >25 years. The results presented here can assist healthcare providers in collecting and monitoring influential KPIs and developing intervention strategies to reduce the childhood T1D incidence rate in Saudi Arabia. ## Introduction Type 1 diabetes (T1D), commonly known as insulin-dependent diabetes, is a metabolic illness caused by an autoimmune reaction that attacks the pancreas’ insulin-producing beta cells, resulting in severe insulin deficiency and accompanying hyperglycemia. A deficiency of insulin production is characteristic of the disease, and insulin injections must be administered continuously for the rest of one’s life for survival [1]. Diabetes complications can result in a variety of long-term health issues, such as nerve and blood vessel damage, hospitalization, and death [2, 3]. Additionally, T1D can cause blindness, due to retinopathy [3], and kidney failure [4]. Chronic diseases like diabetes can disrupt physiology, affecting linear growth, pubertal development [5], and cause serious dermatological complications [6]. T1D is likely to be initiated by as yet unidentified environmental variables in genetically susceptible individuals, with the major genetic contribution identified within the HLA complex, specifically HLA class II [7]. Despite recent advances in the understanding of the disease’s genetics and immunology, its incidence continues to rise by 3–$4\%$ per year worldwide [8]. T1D is a key primary target for prevention due to its growing incidence, severe morbidity and mortality, and substantial health care costs [8]. According to the International Diabetes Federation (IDF), Atlas’ 10th edition [2022], there are 651,700 children under the age of 15 years globally with T1D [9]. Additionally, it is predicted that approximately 108,300 children under the age of 15 will be diagnosed with T1D each year, rising to 128,900 when the age range is extended to 20 years [9]. Saudi Arabia has a high incidence rate of new T1D cases in children (<15 year) each year (31.4 cases/100,000 children) and the number of new T1D cases among children and adolescents is estimated to be 3,800 annually [9]. Better knowledge of the development of T1D in Saudi Arabian children is urgently needed to inform monitoring practice in order to prevent complications and address the incidence of T1D, as early intervention may have a higher success rate in preventing the beginning of dysglycemia and slow the loss of functioning islet cells. ## Review of literature on risk factors of T1D in children The etiology of T1D is uncertain, environmental triggers and viruses are believed to activate the autoimmune process, resulting in pancreatic B-cell destruction and T1D [10]. Although the involvement of environmental and genetic influences on the development of T1D has been recognised for over 40 years [11], research into the potential risk factors for T1D is ongoing [12, 13]. Demographic and socioeconomic status, genetic, obstetric history, maternal characteristics and nutritional history have all been associated with T1D onset [12–14], although the relative contribution of each factor shows variation in the literature as outlined below. ## Demographic and socioeconomic status The research on the demographic profile has investigated and compared T1D incidence in urban and rural residents. An Australian study reported that the incidence of T1D in children in urban areas increased by an average of $2.5\%$ a year ($95\%$ CI [1.2, 3.8]) compared with $5.0\%$ a year ($95\%$ CI [2.4, 7.6]) in the non-urban area over the period from 1985 to 2002 [15]. This difference in incidence-rate trends was not statistically significant over the whole study period. However, the rate of increase in incidence in non-urban areas was significantly higher than that in urban areas when this study was limited to the last 10 years [15]. In addition, a study conducted in Taiwan found that there was an association between the incidence of T1D and living in an urban area [16]. However, studies conducted in Finland and Scotland [17, 18] indicated that urban areas were linked to lower incidence rates compared to rural areas. A recent study conducted in Germany also found that the incidence of T1D was higher in rural areas compared with urban areas [19]. The apparent difference in patterns of incidence between European countries and those of Asia and Australia could be due to the socioeconomic differences between these environments. A study conducted in Australia indicated that T1D incidence is more likely in families with a higher socioeconomic status [15]. This is supported by the findings of a study of European countries that showed T1D rates are most closely correlated with markers of country prosperity such as gross domestic product and low infant mortality [20]. T1D incidence was also shown to be the lowest in areas with the most material deprivation in the United Kingdom (Northern Ireland [21], Yorkshire [22], and Scotland [18]). In both case and control groups, a higher proportion of children were from Jeddah ($42\%$), with approximately equal proportions from Al-Ahsa and Riyadh. More females ($58\%$) were observed in both case and control groups than males ($42\%$). Further, we can see that urban residents were significantly higher in the control ($88\%$) than in case samples ($68\%$) (χ2 = 60.8, p-value = <0.001). There were more consanguineous parents in the control group ($46\%$) than in the case group ($44\%$) but the difference was not statistically significant ($$p \leq 0.289$$). ## Genetic (family history) It is well established that there is a genetic component that increases the risk of T1D in children. It is shown that children of mothers with T1D have a greater risk of developing T1D compared with offspring of non-diabetic mothers [23]. In [24], the authors showed that the risk of T1D for children whose fathers are affected by the disease is 11 times higher than in the control group [14]. While the risk for children whose brothers are affected by the disease is 20 times higher compared to the control group [14]. [ 7] presents a review and comparison of the genetic determinants of T1D in various populations. They have concluded that the worldwide variation in incidence is at least partially determined by differences in genetic risk factors. Authors in [24] have shown that children with the HLA-risk genotype along with a family history of T1D have more than a 1 in 5 risk for developing islet autoantibodies during childhood. The authors in [7] have found over 40 loci were associated with the risk of T1D. As expected from the prior literature, a higher percentage of children in the case group ($22.55\%$) had a first-degree family history of T1D when compared to the control group ($7.84\%$) (χ2 = 49.25, p-value = <0.001). The majority ($75\%$) of both groups had no second-degree of family history of T1D. ## Nutritional history When considering nutrition of the child, a recent review study reported that early exposure to cow’s milk could lead to the development of T1D in children [25]. Other studies reported that short breastfeeding periods (2–4 months) and early cow’s milk exposure (before 4 months) might increase susceptibility [26]. The absence of breastfeeding was found strongly associated with T1D [27]. Another study [28] has explored the associations between early solid food feeding and the risk of T1D. The authors found that early (<4 months) and late (>6 months) introduction of solid foods was associated with increased risk of T1D. Furthermore, studies from Saudi Arabia [29, 30] and other countries such as Italy [31] found that vitamin D deficiency may be linked to the occurrence of childhood T1D. Breast feeding alone was reported by $24\%$ of the control group and $17\%$ of the case group. However, a similar proportion of children in the control and case groups were fed with both breastfeeding and cow’s milk ($61\%$ and $63\%$) (χ2 = 7.69, p-value = <0.001), and the timing of solid food introduction was also consistent across both groups. The majority of children in this study were self-reported as middle-income ($81\%$). ## Obstetric history Several studies have examined associations between obstetric factors and T1D such as birth delivery method, gestational age, birth weight, and birth order. Birth delivery method was linked to the development of T1D in children [32–35]. It is well documented that children born by Caesarean section have a higher propensity to develop T1D compared to children born by normal delivery [32–35]. For gestational age, very preterm (<33 weeks) [35, 36] or post-term (>40 weeks) [35] were linked to a reduced risk of the onset of T1D in children. However, pre-term (33–36 weeks) and early term (37–38 weeks) were linked to an increased risk of T1D [35–37]. An association between a higher birth-weight (of the child) and the risk of developing T1D was found in [35, 38]. Children weighing either 3.5 kilograms (kg) to 4.0 kg or heavier than 4.0 kg at birth had an average increase of $6\%$ and $10\%$, respectively in their risk of diabetes [38]. Birth order may also contribute to the risk of developing T1D. A previous study indicated that the risk in firstborn children was highest and gradually declined with higher birth order after adjustment for the effects of maternal age at childbirth and child sex [39]. This is also supported by [40] where a higher birth order was associated with a substantial reduction in the risk of T1D comparing firstborn birth order with three or more. Full-term pregnancy (39–40 weeks) was the most common gestational age with ($69\%$) and ($66\%$) in the case and control groups respectively. There was a significant difference in the mode of birth delivery between the case and control group (χ2 = 4.12, p-value = 0.042). For the control group, $75\%$ were normal birth and only $25\%$ caesarean, whereas the caesarean rate increased to $31\%$ in the case group with $69\%$ by a normal birth. The child’s weight at birth showed similar percentages for all weight categories between case and control groups except for those with birth weight >4.0 kg. In the control group, only $2\%$ had a weight >4.0 kg compared with the case group which had $5\%$ with a birth weight >4.0 kg. The frequency of being born 2nd was significantly lower in case group ($19\%$) than control group ($29\%$) and a higher proportion in the case group of a birth order of 4th or higher (χ2 = 19.96, p-value = <0.001). ## Maternal characteristics Maternal health may play an important role in T1D. In a comparison of maternal age of greater than 35 with maternal age of less than 25, the occurrence of T1D in the older maternal age group increased substantially [40]. This was also confirmed by [39], who showed an increased chance of having T1D for children with a mother of 45 years or more. Previous studies reported that maternal diabetes (T1D, T2D, or gestational diabetes) is associated with an increased risk in children with T1D [23, 37]. Other maternal health concerns such as maternal asthma are associated with an increased risk of T1D in children [37]. Furthermore, pre-eclampsia was considered a significant risk factor for the development of T1D in children [41]. However, a meta-analysis study showed that children born to mothers with pre-eclampsia had on average $10\%$ increase in their risk of developing T1D but this association was not significant [42]. The maternal age group of 25–35 was the most common in both cases and control groups with ($61\%$) and ($58\%$) respectively. Mothers aged over 35 years were reported more in case samples ($18\%$) than in control samples ($11\%$) (χ2 = 15.45, p-value = <0.001). A higher proportion of cases ($9.8\%$) reported mothers that experienced gestational diabetes (χ2 = 6.81, P-value = 0.009). For pre-eclampsia and maternal history of asthma, there were no significant differences between case and control groups. ## The use of preventative studies to improve T1D clinical outcomes The possibility of preventing, delaying, or reducing complications of T1D diabetes in children is an important area of investigation [3, 43, 44]. In the preventative studies [45, 46] it was demonstrated that the elimination of cow’s milk proteins in infant formula (in the Finnish TRIGR pilot study [45]) and the elimination of bovine insulin in infant formula (in the FINDIA study [46]) both decreased the production of islet autoantibodies. Further studies have investigated methods for predicting T1D. Early exposure to respiratory infections has been linked to an increased probability of autoantibody seroconversion in children with a T1D family history [47]. This was observed by continuously monitoring islet autoantibodies over their first three years of life [47]. The risk of T1D has also been predicted using longitudinal autoantibody measures in families with a first-degree relative with T1D [48], in the general populations [49], or in individuals who have been identified as at risk [48, 50, 51]. In addition, genetic markers and genetic risk scores were used to detect islet autoantibodies in children with high-risk HLA genotypes [52, 53]. A composite risk score model that was constructed for high-risk children that included clinical, genetic, and immunological variables demonstrated a significantly improved T1D prediction compared to autoantibodies alone [54]. The benefits shown by the above research could be enhanced further by including more of the risk factors of T1D. ## Machine learning to predict T1D development in children Creating predictive model with the large number of risk factors requires a sophisticated approach. Machine learning is becoming a popular and important approach in the field of medical research due to its capability of modeling complex linear and nonlinear relationships. Researchers in T1D have started to use Machine learning methods [55–58]. In [55], the authors utilised Random Forest (RF), Support Vector Machines (SVM), and Generalized Linear Model (glmnet) with data from the intestinal microbiota of infants to search for species that influence the development of T1D. In addition, the authors in [56] used Naïve Bayes classifier to determine if a combination of genetic, immunologic, and metabolic features, measured at infancy, can be used to predict the likelihood that a child will develop T1D by age 6 years. In another study [57], Naïve Bayes, Decision Tree, Support Vector Machine SVM, and Random Forest were used to predict diabetes in children using glucose, blood pressure, skin thickness, insulin, BMI, age, and family history of diabetes. In [58], the authors used Random Forest and logistic regression to evaluate the association between the onset of T1D in a child and the mother’s vaginal bacteriome and mycobiome. However, beyond the above research, there is an opportunity for further application of machine learning methods to develop predictive models of T1D. ## Motivation and the objective of this study The existing research such as those conducted in Australia, Sweden, and Finland do not capture the diversity, ethnicity, and culture of the population in Saudi Arabia [59, 60]. It is reported [61] that there is a high rate of consanguinity in Arabic countries which could lead to increased homozygosity in the (HLA) haplotypes and non-HLA genes. This may alter the susceptibility to T1D. In addition, Saudi Arabia has an increasing incidence rate of T1D in children, ranking 7th in the world for the highest number of children with T1D and 5th in the world for the T1D incidence rate [9]. Despite the remarkably increased incidence of childhood T1D in Saudi Arabia, there is a lack of comprehensive research on T1D in Saudi children when compared to developed countries [62–64]. Existing T1D research in Saudi Arabia used small sample sizes, a single center, or a single city or region of the country, or limited number of factors [62, 64, 65]. Improvements are being made with a recent study conducted over three different regions in Saudi Arabia exploring age at onset of T1D in children [64], however this can be further enriched with the incorporation of control data as in this study. This study aims to fill the gap by creating predictive models of T1D and identifying the KPIs of T1D in children using data from Saudi Arabia. The proposed work differs from previous research as we have considered: Having a clinically relevant predictive model will contribute to better understanding and identifying the requirements to begin the development of a clinically relevant tool. This tool can then be used to develop intervention strategies to reduce the incidence rate of Childhood T1D in Saudi Arabia. The research presented here will improve public health in Saudi Arabia and contribute to filling the gap in T1D research for diverse populations. ## Population and sample size De-identified data for 1,142 individuals were collected based on a case-control study conducted in three cities (Al-ahsa, Jeddah, and Riyadh) located in the highest populated regions of Saudi Arabia; Eastern, Western, and Central to address the aim of this work. Ethical approval was granted by the RMIT University Human Research Ethics committee in Australia and the Research Ethics Committee of the Ministry of Health in Saudi Arabia. This was a retrospective study of medical records, with additional supporting demographic information collected through a survey of the parents of each child included in the cohort. The need for informed consent was waived by the ethics committee to collect existing data from medical records. Informed consent was obtained for the additional information collected via survey for their residency, income status, and nutritional history. All data were collected and reviewed by trained medical professionals and then were fully anonymised before analysis. Cases were children <15 years with a confirmed diagnosis of T1D between 2010 and 2020. Controls were children <15 years without any clinical indicators of T1D who attended the same hospital. All children with diabetes were matched with at least two control children with the same year of birth, same sex, and from the same city. In this study, the sample sizes were total: 1,142, cases: 377 and controls: 765. An a priori power calculation using the epiR package in R, indicated that a total sample size of 1,086 would be sufficient at $80\%$ power with a $95\%$ confidence level to detect an OR above 1.5. This was determined with a 1:2 case to control ratio and a prevalence of control exposure set to $24\%$, as reported in previous literature [66–68]. ## The Key Performance Indicators (KPIs) considered A structured questionnaire was used for collection of data from the medical records and parents of both case and control samples. De-identified data were collected on socio-demographic, potential genetic and environmental factors identified through the literature review. The collected demographic data included city where they live, gender, and birth year. Socioeconomic Status: residency and family’s income levels factors. Genetic: consanguinity marriage and having a family history of T1D (First degree and second degree). Environmental: nutrition history, and solid food, obstetric History: birth delivery mode, gestational age, weight at birth (Kg) and birth order. Maternal characterise at child birth: maternal age at child’s birth, gestational diabetes, maternal asthma and Pre-eclampsia. The pregnancy weeks used by the World Health Organization (WHO) and the American College of Obstetricians and Gynaecologists (ACOG) [69, 70] were used to categorise pregnancy length (in weeks) as: <33 (very pre-term), 33–36 (pre-term), 37–38 (early-term), 39–40 (full-term), 41 (late-term), and >42 (post-term). The child’s weight at birth was categorised into 4 categories using WHO and ACOG definitions; Birth weight (<2.5 kg), Birth weight (2.5–3.0 kg), Birth weight (3.5–4.0 kg), and Birth weight (>4 kg) [71, 72]. Also, maternal age at child birth were classified to 3 classes based on the previous studies [40]; maternal Age at child’s birth: (<25 years), maternal age (25–35 years), and maternal age (>35 years). Income was collected as a selection from the following 5 categories: High income: >12000 Saudi Riyals, Upper-middle income:(>9000–12000) Saudi Riyals, Middle income: (>6000–9000) Saudi Riyals, Lower-middle income: (>3000–6000) Saudi Riyals, and Low income: ≤ 3000 Saudi Riyals. ## Logistic regression and ML models Understanding the wide range of the potential KPIs of T1D could ease diagnosis, provide adequate classification and relative importance and allow for cost-effective disease management. Therefore, the first phase in effective intervention is to identify such KPIs associated with T1D. Majority of conducted T1D studies in children have used a single approach to identify KPIs whereas using different and comparable approaches can enhance the ability to find the best fit for the data and hence may more accurately provide the significant KPIs. Machine learning may help in understanding the intricacies of relationships between inputs and the main outcome. They have the flexibility and the advantages of a built-in feature selection method, are able to handle many input variables without the need to minimize dimensionality, and can control overfitting by using out-of-bag validation. This research combines five different approaches—Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Naive Bayes (NB) and Artificial Neural Network (ANN) to examine the most suitable model for predicting the risk of T1D and explore factors affecting this disease in Saudi Arabian children. All machine learning models have been utilised in a variety of studies to describe various systems such as environmental protection, agriculture as well as in the various areas of health [73–77], and hence they will be used and compared with the LR to identify the risk factors of T1D in children. The description of the individual ML models are summarised in the following sections. Additional references are provided for the detailed mathematical formula and theory behind the individual ML models. R statistical software was used to perform the analysis [78]. All code for data analysis associated with the current submission is available at https://github.com/Alazwari/R-code.git. ## Logistic Regression (LR) Logistic regression (LR) analysis is a commonly used statistical tool in medical research [79, 80]. It is the expansion of the linear regression model to use the probabilities of the two potential outcomes for classification problems. It is suitable for models involving disease state (diseased or healthy) and decision making (yes or no), and is thus widely applicable to health studies [81], including diabetes analysis of cases versus controls [82–84]. One of the important aspects of LR that makes it an effective and powerful tool in medical research, is the ability to determine the potential predictors [82, 85]. In addition, the advantages of LR are the simultaneous analysis of multiple explanatory variables, interactions, and reduction of the effect of confounding factors [82]. A LR model can be expressed in the form of: β0+β1x1+β2x2+…+βnxn=log[P(Y)/1-P(Y)], where log[P(Y)/1 − P(Y)] is the log (odds) of the outcomes, Y is the dependent variable, xi are the independent variables, β0, β1, ‥βn are the regression coefficients and β0 is the intercept. Odds ratios (with $95\%$ confidence intervals) will be used for assessing the KPIs of diabetes for significance. The LR assumptions have been checked and validated. ## Random Forest (RF) Decision trees are a commonly used machine learning tool because it is simple, easy to use, and interpretable for categorical data [86]. Several studies have been conducted to address the limitations of the traditional decision trees like their lack of robustness and suboptimal performance [87]. Based on these studies [86, 87], Random Forest (RF) was developed as an ensemble learning method in which the output of a number of weak learners, which could be a single decision tree, is improved by a voting scheme similar to other ensemble learning methods [88, 89]. RF can handle a large number of input variables without the need to reduce dimensionality because it has a built-in feature selection method [64, 75]. Also, in RF, out-of-bag validation can be used to control overfitting [75]. ## Support Vector Machine (SVM) In the field of Machine Learning, Support Vector Machine (SVM), developed by Vladimir Vapnik, is among the most well-known and widely used algorithms [90]. SVM is a very effective method for building a classifier. Its aim is to create a decision boundary between two classes that allows for the prediction of labels using one or more feature vectors [91–94]. This decision boundary, called the hyperplane, is oriented to be as far away from the closest data points from each class as possible. These closest points are referred to as support vectors [88, 91, 92, 94]. In addition, the kernel approach, which allows us to model higher-dimensional, and non-linear models, is another advantage of SVM [92]. The kernel function chosen can have a significant impact on the performance of a SVM model. However, there is no way to know which kernel is optimum for a particular pattern recognition problem, this must be determined by trial [92]. In order to improve the SVM model accuracy, there are several parameters that need to be tuned. The main hyperparameter of the SVM is the kernel (linear, polynomial, radial basis function (RBF), and sigmoid kernel, Gaussian, Exponential, Hyperbolic tangent, ANOVA radial basis kernel, and Laplacian kernel). Every one of these kernels requires optimisation of one or more parameters such as cost, degree and gamma. The parameter cost controls over-fitting of the model by specifying tolerance for misclassification, gamma controls the degree of non-linearity of the model, and in the polynomial kernel, degree is the degree of the polynomial kernel function that controls the flexibility of the decision boundary. Higher degree polynomial kernels allow a more flexible decision boundary. In this study, we will use linear, polynomial, radial basis function (RBF), and sigmoid kernel. For each kernel, cross-validation is used to select the value of the parameters that optimise the SVM model. ## Naive Bayes (NB) Naive Bayes is a classification strategy based on Bayes Theorem and the idea that the existence of a specific feature in a class is unrelated to the presence of other features [95, 96]. It has been widely used for classification and prediction in many domains because it can be constructed quickly and easily from data. Also, it reduces space complexity, allowing quick inference, and often outperforms more complex learning algorithms in practice [97, 98]. The Naive Bayes Classifier provides a highly efficient probability estimation based on a simple structure, requiring only a small amount of training data to predict the classification parameters. It is based on two main assumptions: feature independence and the absence of hidden or latent properties [99]. ## Artificial Neural Network (ANN) Artificial Neural Network (ANN) is a prediction approach used for finding a solution when other statistical methods are not suitable. The benefits of this method include the ability to learn from examples, fault tolerance (the property that allows an ANN to operate properly in the event one or more components are lost), and non-linear data forecasting, which all make it a commonly used statistical method [100]. The major benefit that ANN provides is the potential to distinguish hidden linear and nonlinear relationships, which often occur in high-dimensional and complex data sets [64, 101]. The ANN algorithms consist of input, hidden and output layer nodes where the nodes may also be referred to as “neurons” [102]. The number of input layer nodes represents the number of variables that describe the features evaluated, whereas the number of output layer neurons is equivalent to the number of levels of the outcome factor. The number of hidden layers and the number of neurons depend on the quantity of data and the complexity of the relationship. Every neuron in the hidden and output layer is linked by a corresponding numerical weight to all nodes in the proceeding layer [64, 103, 104]. It is these weights that determine the effect of neurons and impact the output. ## Models performance evaluation In order to assess the models’ performance prediction, data are randomly divided into two subsets of $70\%$ for training and $30\%$ for the testing set to build and evaluate the LR, RF, SVM, NB, and ANN models. In addition, there are five common measures used in ML models [105, 106]: Accuracy, Sensitivity (Recall rate (R)), Receiver Operating Characteristic (ROC) which is equal to the Area Under the Curve (AUC), Precision rate (P), and F Score. The diagnosis of T1D is a binary classification issue characterised by the ground truth that determines the performance of the prediction. The output of the prediction can be True Positive (TP), False Positive (FP), True Negative (TN), or False Negative (FN). The terms “positive case” and “negative case” were used to define the case group and the control group, respectively. To obtain the ROC curve, the true positive rate (TPR) was plotted against the false positive rate (FPR). In this case, better performance is indicated by higher values of the (AUC) corresponding to the ROC curve. The accuracy was also calculated and it is the proportion of the correctly predicted made for those with T1D and the controls divided by the total number of the dataset. In addition to the correct identification of having T1D, we have considered the number of true positives (children who had T1D) using the Precision rate P and the number of diabetic children predicted using Recall rate R. We also used the F Score to describe the efficiency of each model, with F Score close to 1 indicating better performance. ## Characteristics of the study sample For this cohort, the range of birth year is from 2003 to 2020 and the median birth year was 2010. In addition, the demographic characteristics and potential risk factors for each of the case and control groups are shown in Tables 1 and 2. The results of the Chi-square (χ2) test to determine if there was a difference in the distribution of categorical variables between cases and controls are also shown in Tables 1 and 2. ## Models performance Data was randomly divided into two subsets of $70\%$ for training and $30\%$ for the testing set to build and evaluate the LR, RF, SVM, NB, and ANN models. Also, we adopted a k-fold cross-validation approach to assess the models’ performance where k denotes the number of groups in which data is split and $k = 10.$ In addition, for ML models, hyperparameters were tuned in order to optimize the ML models (Fig 1). Table 3 shows a summary of the result of the models’ performance on the testing set and is described below. Fig 1: Flowchart of models performance. TABLE_PLACEHOLDER:Table 3 Table 3 shows the detailed performance results of the LR and ML classification methods. Logistic regression (LR) based on all variables (full model) achieved a prediction Accuracy (0.77, $95\%$ CI [0.7116, 0.8264]), Sensitivity (Recall), Precision, and F Score of (0.70), and AUC (0.83). The interactions between the independent variables were explored to further improve the LR model. However, due to the small number of observations in some levels of the independent variables, such as the maternal history of asthma ($$n = 14$$) (Table 2), the results did not show further improvement for this dataset. In addition, we built the reduced LR based on the most significant factors (based on their corresponding P-value) presented in (Table 4) and (S1 Table). The reduced LR achieved lower values of Accuracy (0.75, $95\%$ CI [0.6825, 0.8016]), Sensitivity (Recall) (0.59), Precision (0.61), F Score (0.60), and AUC (0.78) compared with the full LR model. Table 4 \| Variables \| OR \| CI \| P-value \| \| --- \| --- \| --- \| --- \| \| Intercept \| 0.014 \| (0.003, 0.058) \| 0.000* \| \| City (Jeddah) \| 1.618 \| (1.111, 2.370) \| 0.012* \| \| Residency (Rural) \| 3.745 \| (2.604, 5.415) \| 0.000* \| \| Nutrition history (Introduction to cow’s milk) \| 2.928 \| (1.847, 4.678) \| 0.000* \| \| Nutrition history (Both) \| 1.58 \| (1.095, 2.301) \| 0.016* \| \| First degree of T1D (Father) \| 6.181 \| (2.355, 17.189) \| 0.000* \| \| First degree of T1D (Siblings) \| 3.708 \| (2.282, 6.083) \| 0.000* \| \| First degree of T1D (Mother) \| 2.639 \| (1.228, 5.707) \| 0.012* \| \| Second degree of T1D (Yes) \| 1.598 \| (1.146, 2.225) \| 0.005* \| \| Birth weight (>4 kg) \| 3.114 \| (1.367, 7.356) \| 0.007* \| \| Maternal age at child’s birth (25–35 years) \| 1.691 \| (1.169, 2.463) \| 0.006* \| \| Maternal age at child’s birth (>35 years) \| 2.42 \| (1.393, 4.217) \| 0.002* \| \| Income status (Low income) \| 3.063 \| (0.828, 11.949) \| 0.098 \| \| Delivery birth mode (CS) \| 1.369 \| (0.997, 1.877) \| 0.051 \| \| Birth order (1st) \| 1.52 \| (0.987, 2.350) \| 0.057 \| \| Birth year \| 0.992 \| (0.952, 1.035) \| 0.739 \| \| City (Al-Ahsa) \| 1.323 \| (0.876, 2.002) \| 0.183 \| \| Gender (Female) \| 1.147 \| (0.863, 1.527) \| 0.346 \| \| Consanguineous parents (Yes) \| 0.964 \| (0.727, 1.276) \| 0.798 \| \| Solid food (less than 6 months) \| 0.81 \| (0.586, 1.114) \| 0.198 \| \| Income status (Lower-middle income) \| 2.275 \| (0.854, 6.566) \| 0.111 \| \| Income status (middle income) \| 1.778 \| (0.768, 4.583) \| 0.201 \| \| Income status (Upper-middle income) \| 1.17 \| (0.421, 3.464) \| 0.768 \| \| pregnancy length (full-term [39–40 weeks]) \| 1.385 \| (0.672, 2.998) \| 0.390 \| \| pregnancy length (early-term [37–38 weeks]) \| 1.156 \| (0.519, 2.681) \| 0.727 \| \| pregnancy length (pre-term [36–33 weeks]) \| 1.219 \| (0.513, 2.992) \| 0.658 \| \| pregnancy length (very pre-term [<33 weeks]) \| 0.427 \| (0.105, 1.550) \| 0.209 \| \| Gestational diabetes (Yes) \| 1.37 \| (0.801, 2.329) \| 0.245 \| \| Birth order (2nd) \| 0.737 \| (0.476, 1.140) \| 0.171 \| \| Birth order (3rd) \| 0.903 \| (0.590, 1.381) \| 0.640 \| \| Pre-eclampsia (Yes) \| 0.527 \| (0.178, 1.357) \| 0.209 \| \| Asthma (Yes) \| 0.824 \| (0.394, 1.636) \| 0.593 \| \| Birth weight (<2.5 kg) \| 1.318 \| (0.946, 1.834) \| 0.101 \| \| Birth weight (3.5–4.0 kg) \| 1.979 \| (1.059, 3.667) \| 0.347 \| Random Forest (RF) models have been built using all independent variables. In addition, the hyperparameter that controls the split variable randomization feature of RF is often referred to mtry. This is the number of variables randomly sampled as candidates at each split, and it helps to balance the trade-off between a low correlation reasonable strength. Through 10-fold cross-validation based on the training data, we observed that the best mtry for the full RF is 4 (Fig 2). Fig 2: Tuning hyperparameter for full RF model.The best mtry for the full RF model based on the 10-fold cross-validation. Table 3 shows the Full RF achieved Accuracy (0.75), Sensitivity (Recall) (0.61), Precision (0.64), F Score (0.62), and AUC (0.81). Significant variables from the full models (S1 Table) and (S1 Fig) are used to build the reduced RF. Similar to the full model, the hyperparameter was considered in reduced RF and we found that the best mtry for the reduced RF is 3 (Fig 3). Fig 3: Tuning hyperparameter for reduced RF model.The best mtry for the reduced RF model based on the 10-fold cross-validation. The results in Table 3 show that the reduced RF achieved Accuracy (0.73), Sensitivity (Recall) (0.58), Precision (0.60), F Score (0.56), and AUC (0.80). In Support Vector Machine (SVM), the main hyperparameter is the kernel, the mathematical function used to transform data that cannot initially be separated linearly, was considered to improve the SVM model accuracy. SVM can use different kernels and each of these kernels requires one or more parameters, including cost, degree, and gamma. Four different SVM kernels were explored linear, polynomial, radial, and sigmoid. The optimal model was found with the linear SVM with all variables, and it was based on a cost parameter of 10. However, the best polynomial SVM was based on the degree of 3 and the cost of 1. For the radial SVM, the cost of 2 and gamma of 0.1 was found to be optimal and the sigmoid SVM used a gamma of 0.1, and a cost of 1. The results of the evaluation of SVM models (Table 3) indicated that the linear and sigmoid kernel functions achieved a high AUC of (0.80 and 0.79), Sensitivity of 0.65, and F Score of (0.62 and 0.63) compared with other kernel functions. To further improve the best SVM model (linear SVM), the reduced model has been built based on the significant variables (S1 Table) and (S2 Fig). However, the reduced linear SVM model has the lowest values of Sensitivity and F Score (0.50 and 0.52) respectively compared with the full SVM models. In addition, we tuned the hyperparameter in Naive Bayes (NB) using the kernel density estimation. As a result, the NB model achieved an Accuracy (0.75), Sensitivity (0.63), F Score (0.61), and AUC (0.75). The reduced NB model also was considered and built using the significant variables identified (S1 Table) and (S3 Fig) for further improvement. The reduced NB model achieved Accuracy (0.72), Sensitivity (0.50), F Score (0.53), and AUC (0.73) (Table 3). Therefore, the full NB outperforms the reduced NB model. For ANN the number of neurons in the hidden layers is one of the hyperparameters to tune. There is no standardised approach for determining the number of neurons in a hidden layer [107]. The number of hidden layer neurons varies from problem to problem, and it depends on the number and quality of training patterns [64, 108]. We used 10-fold cross-validation to determine the optimal number of neurons in hidden layers. A similar accuracy and AUC were observed between ANN models with one and two hidden layers. ( Fig 4) shows that the highest accuracy is at 14 neurons when using one hidden layer and at 12 neurons when using two hidden layers. Fig 4: Tuning hyperparameter for ANN models.ANN with one hidden layer (a) and ANN with two hidden layers (b). Both ANN models with one and two hidden layers have the same Accuracy (0.70) and Precision (0.60) and similar F Score (0.56 and 0.57), and AUC (0.70 and 0.72) respectively. However, the ANN with two hidden layers achieved a high value of Sensitivity (0.58) compared with ANN with one hidden layer (0.52). So, the optimal ANN model was with two hidden layers and 12 neurons in each layer (Fig 5). Fig 5: An optimal ANN model with 2 hidden layers with all input variables.P.L: pregnancy length, F.H.2nd: Family history of T1D with a second-degree relative, I.S.:Income Status, C.P.: Consanguineous Parents, M.Age: Maternal Age, B.Weight: Birth weight, B.Order: Birth Order. Furthermore, the reduced ANN model with two hidden layers was also built using the significant variables from (S1 Table) and (S4 Fig). The reduced ANN model achieved lower performance with Accuracy (0.65), Precision (0.50), F score (0.51), and AUC (0.66) (Table 3). ## Models comparison As shown in Table 3, the full LR model has the highest value of Sensitivity, Precision, and F Score (0.70) compared to all other models. Whereas the reduced SVM, reduced NB, and ANN with one hidden layer and reduced ANN models showed weaker model performance with Sensitivity (0.50, 0.50, 0.52 and 0.54) and AUC (0.72,73, 0.70, and 0.66) respectively. The best performing models from every approach are shown in (Fig 6). LR has the highest value of AUC (0.83) followed by full RF which achieved AUC of 0.81. SVM linear showing a slightly lower AUC of 0.80. Therefore, LR (logistic regression) yields a better Accuracy, Sensitivity, Precision, F Score, and AUC while showing similar performance for Specificity as the machine learning models to predict childhood T1D in Saudi Arabia. An additional figure of the ROC curve performance for all other models is located in the Supporting information section (S5 Fig). Fig 6: ROC curve and AUC for the best-performing model of different approaches. ## Selection of the significant variables Logistic regression was subsequently used to estimate the importance of each predictor. The variables that were identified to be significantly associated with T1D included: rural residency, introduction to cow’s milk, first-degree T1D of father, mother, and siblings, presence of second-degree T1D, birth weight more than 4 kg, and maternal age of 25–35 and over 35 years at child’s birth (Table 4). In addition, the risk factors such as income status (low income), birth delivery(CS), and birth order(first) (Table 4), did not reach the level of statistical significance (P-value <0.05) but they show a trend towards a relationship. In addition, the significant variables identified by LR were also identified as significant variables by one or more of the ML models (S1 Table). For example: residency, and maternal age at child birth were identified as significant by all models. The first degree of T1D (siblings) was identified as a significant variable by LR and three ML models (SVM, NB, and ANN). S1–S4 Figs in (Supporting information section) show the variables importance based on each ML model. ## Discussion This study has several strengths. To the best of our knowledge, it was the largest study to investigate the association between the environmental and family history factors of T1D in children in Saudi Arabia. We have used local data to create a more robust and clinically relevant predictive model of T1D. Further different statistical and ML approaches were used to find the optimal model to predict T1D in children and identify its significant KPIs. To ensure a broad representation of the diverse population was covered, de-identified data of 1,142 children collected from three cities in Saudi Arabia were used for this case-control study. Having access to a statically sufficient sample size [1,142], we have compared the performance of LR and modern ML approaches to predict T1D. The efficacy of models was assessed using multiple criteria (Accuracy, Sensitivity, Precision, F Score, and AUC). The results presented in Table 3 showed that LR has the best performance with the highest values of Accuracy (0.77), the same value of Sensitivity, Precision, and F score (0.70), and AUC (0.83). This was followed by the performance of full RF and full SVM (linear) with AUC (0.81 and 0.80) and Accuracy (0.75 and 0.75) respectively. This was consistent with the previous study [109] where the authors concluded that LR “was as good as” ML in predicting major chronic diseases. The comparison results outlined above were obtained after tuning hyperparameters for each ML model and 10-fold cross-validation for LR and ML models. Subsequently, LR as the best performing model was used to identify the significant KPIs of childhood T1D. Our study has investigated both environmental and family history as risk factors for T1D. The results presented in Table 4 show that urban residence is linked to lower incident rates compared to rural areas. This result supports previous findings outlined in [17, 18] but it is in contrast with the findings of [15, 16, 110]. This contrast could be due to the differences in study design and the country of origin under investigation. For example, in [15] only the cases group was considered, and in [16] the population densities were categorised into four levels. Our result has identified early exposure to cow’s milk as one of the significant KPIs (OR = 2.92, $95\%$ CI [1.84, 4.67], $$P \leq 0.000$$). This is in agreement with the finding in the previous studies [25, 26]. However, other studies [111, 112] have reported that early exposure to cow’s milk had no association with T1D (OR = 1.06, $95\%$ CI [0.36, 3.09], $$p \leq 1.000$$) and (OR = 0.85, $95\%$ CI [0.61, 1.18], $$p \leq 0.332$$) respectively. Our study shows that the risk of T1D is significantly associated with a positive family history of T1D in agreement with the previous study [14, 113]. The results presented in Table 4 indicated that T1D in first and second-degree relatives increases the risk of T1D in children. Specifically, the chance of developing T1D for children whose father is affected by T1D is 6.18 times higher (OR = 6.18, $95\%$ CI [2.35, 17.18]). Also, the risk of T1D for children whose siblings are affected by T1D is 3.70 higher (OR = 3.70, $95\%$ CI [2.28, 6.08]). Moreover, our results indicated that child’s birth weight >4 kg was a risk factor for childhood T1D whereas a low birth weight had no statistically significant impact. This is in agreement with the previous studies [35, 38, 114]. However, the authors in [36] showed that there is no association between a child’s high birth weight and T1D (OR: 1.01, $95\%$CI: 0.96–1.05), and in [115] the authors reported a significantly lower risk of T1D in children with low birth weight (<2.5 kg) (OR: 0.82, $95\%$CI: 0.67–0.99). These results could be due to the difference in the definition of the birth weight reference levels. The reference levels of birth weight in [36] were (3.0—4.0 kg) and in [115] (3.0- 3.5 kg) which are different to (2.5—3.5 kg) considered in our study. The reference level (2.5—3.5 kg) considered in our study is the recommended normal weight for the children T1D [38, 71, 72]. It is worth mentioning that merging the categories of (3.5—4kg) and (>4 kg) provided a more reasonable sample size, consequently, a narrower confidence interval ($95\%$ CI [1.07, 3.24]) but the value of (OR = 1.86) confirms that birth weight is a significant risk factor. The child’s birth weight >4 kg is identified as high risk, therefore it was decided to include this category in the analysis (Table 4) to be consistent with the literature and recommendation of the American College of Obstetricians and Gynaecologists definitions. [ 38, 72]. Other obstetric factors such as gestational length (gestational age) were not associated with childhood T1D in our study in line with prior findings from study conducted in Israel [116] and conflicts with previous studies in [35–37, 114, 115] that demonstrated the preterm birth (34–36 weeks) has a negative impact on T1D. Our results showed that maternal T1D diabetes, and maternal age at child’s birth (greater than 25 years) were significant factors of childhood T1D. This is consistent with the results presented in [35, 37]. However, other maternal characteristics such as gestational diabetes, pre-eclampsia, and maternal history of asthma were not identified as significant factors of childhood T1D in our study but were shown as significant factors in [23, 37, 41, 42, 117]. This may reflect the small number of observations related to these characteristics in our study. Other risk factors such as the weight of the mother at childbirth [118] were not included in the medical records of Saudi Arabia children, which is a limitation of using secondary data in this study. This should be considered as a key factor in future data collection for research, particularly as female obesity has increased in Saudi Arabia over the last decade [119]. A further limitation of the current study is the use of controls from existing hospital records. Although all controls were identified as free from T1D, they had attended a clinic for medical treatment and hence may not fully represent the health population with no medical clinic attendance. The results presented here show the importance of collecting and monitoring the significant KPIs identified in this study to improve public health outcomes. The creation of a unified electronic health record linking all hospitals in the country would increase the efficacy of data collection (sample size, diversity and monitoring of pregnancy variables, birth characteristics, and child development over time) and enable further refinement of our predictive T1D model. ## Conclusion This study is the largest case-control study to investigate the association between environmental and family history factors of childhood T1D in Saudi Arabia. The country has an increasing incidence rate of T1D in children, currently the 5th highest rate in the World. With regards to the total number of children with T1D, Saudi Arabia ranks 7th in the World. Despite this remarkably high incidence, there is a lack of targeted, comprehensive T1D research considering the diversity of the Saudi Arabian population compared to developed countries. Existing T1D research in Saudi Arabia is limited to small sample sizes, a single center, a single city or region of the country, or a small number of associated factors. In this study, secondary data from a total of [1,142] individual medical records collected from three cities located in different regions of Saudi Arabia have been used in the analysis to represent the country’s diverse population. Both cases and matched controls data (birth year, gender, and location) are used to create a more robust and clinically relevant model by controlling confounders. Also, a wide range of potential KPIs suggested in the literature have been included in this study. We have utilised several approaches to find the optimal model, including modern Machine learning methods, as their application has increasingly gained interest in the healthcare domain. Specifically, we used logistic regression, Random Forest, Support Vector Machine, Naive Bayes, and Artificial Neural Network with local data to find the most clinically relevant optimal model to predict T1D in children. The optimal model was then used to identify the significant KPIs of T1D in children aged (0–14 years). The analysis of environmental and family history factors revealed significant differences across demographic and socioeconomic status, genetic and disease history, nutrition history, obstetric history, and maternal characteristics. The results show that most children (cases = $68\%$ and controls = $88\%$) are from urban areas of Saudi Arabia, $23\%$ (cases) and $8\%$ (control) have a family history of T1D (first-degree relative), and $31\%$ of the cases group were delivered through a caesarean section which was higher than the control group (χ2 = 4.12, $$P \leq 0.042$$). Also, the percentage of mothers older than 35 years was higher in case samples ($18\%$) than in control samples ($11\%$) (χ2 = 8.59, $$P \leq 0.003$$). Comparing the performance of different models based on Accuracy, Sensitivity, Precision, F Score, and AUC, it is shown that the logistic regression model outperforms the other models with Accuracy = 0.77, Sensitivity, F Score and Precision of 0.70, and AUC = 0.83. This was followed by the performance of the full Random Forest and the full Support Vector Machine (linear) models with AUC (0.81 and 0.80) and Accuracy (0.75 and 0.75) respectively. The comparison results outlined above were obtained after tuning hyperparameters for each ML model and 10-fold cross-validation for all LR and ML models. To investigate whether we could further enhance the prediction Accuracy and simplify the models, significant variables identified by the individual full model were used to build the reduced models. The performance of the reduced model in terms of efficacy criteria was less impressive compared with the full models. The full logistic regression was then employed to identify the most significant KPIs of childhood T1D in Saudi Arabia. The results show that the most significant KPIs are early exposure to cow’s milk (OR = 2.92, $$P \leq 0.000$$), birth weight >4 Kg (OR = 3.11, $$P \leq 0.007$$), residency(rural) (OR = 3.74, $$p \leq 0.000$$), family history of T1D in the first degree (father (OR = 6.18, $$P \leq 0.000$$), and siblings (OR = 3.07, $$P \leq 0.000$$)), and second degree (OR = 1.59, $$P \leq 0.005$$), and maternal age (25–35 years) and greater than 35 years with (OR = 1.69, $$P \leq 0.006$$) and (OR = 2.42, $$P \leq 0.002$$) respectively. This study makes a significant contribution to Saudi Arabian childhood T1D research by providing the most clinically relevant optimal model to predict T1D in children and identifying its associated KPIs using local data. The results presented in this paper will also assist healthcare providers in collecting and monitoring the influential KPIs data. This would enable the initiation of suitable intervention strategies to reduce the disease burden and potentially slow the incidence rate of childhood T1D in Saudi Arabia. Furthermore, the research demonstrates that access to a nationwide electronic health record database linking all hospitals in the country could further improve the efficacy of the predictive model with respect to the population diversity attributes. The results presented in this paper have also contributed to filling the gap in childhood T1D research of non-European populations. ## References 1. Xia Y, Xie Z, Huang G, Zhou Z. Incidence and trend of type 1 diabetes and the underlying environmental determinants. Diabetes Metab Res Rev (2019.0) 35 e3075. DOI: 10.1002/dmrr.3075 2. Laing SP, Swerdlow AJ, Slater SD, Botha JL, Burden AC, Waugh NR. The British Diabetic Association Cohort Study, II: cause-specific mortality in patients with insulin-treated diabetes mellitus. Diabetic Medicine (1999.0) 16 466-71. DOI: 10.1046/j.1464-5491.1999.00075.x 3. Dayan CM, Besser RE, Oram RA, Hagopian W, Vatish M, Bendor-Samuel O. Preventing type 1 diabetes in childhood. Science (2021.0) 373 506-10. DOI: 10.1126/science.abi4742 4. Carel JC, Levy-Marchal C. Renal complications of childhood type 1 diabetes. BMJ (2008.0) 336 677-8. DOI: 10.1136/bmj.39517.508241.80 5. Chowdhury S. Puberty and type 1 diabetes. Indian J Endocrinol Metab (2015.0) 19 S51-4. DOI: 10.4103/2230-8210.155402 6. Lombardo F, Passanisi S, Tinti D, Messina MF, Salzano G, Rabbone I. High Frequency of Dermatological Complications in Children and Adolescents with Type 1 Diabetes: A Web-Based Survey. J Diabetes Sci Technol (2021.0) 15 1377-81. DOI: 10.1177/1932296820947072 7. Barrett JC, Clayton DG, Concannon P, Akolkar B, Cooper JD, Erlich HA. Genome-wide association study and meta-analysis find that over 40 loci affect risk of type 1 diabetes. Nature genetics (2009.0) 41 703-7. DOI: 10.1038/ng.381 8. Tuomilehto J. The emerging global epidemic of type 1 diabetes. Curr Diab Rep (2013.0) 13 795-804. DOI: 10.1007/s11892-013-0433-5 9. 9 International diabetes federation. International diabetes federation. IDF Diabetes Atlas, 10th edn. Brussels, Belgium. 2021.. IDF Diabetes Atlas (2021.0) 10. Rewers M, Norris J, Dabelea D. Epidemiology of type 1 diabetes mellitus. Advances in experimental medicine and biology (2004.0) 552 219-46. PMID: 15622966 11. Wasserfall C, Nead K, Mathews C, Atkinson MA. The threshold hypothesis: solving the equation of nurture vs nature in type 1 diabetes. Diabetologia (2011.0) 54 2232-6. DOI: 10.1007/s00125-011-2244-z 12. Egro FM. Why is type 1 diabetes increasing?. J Mol Endocrinol (2013.0) 51 R1-13. DOI: 10.1530/JME-13-0067 13. Butalia S, Kaplan GG, Khokhar B, Rabi DM. Environmental Risk Factors and Type 1 Diabetes: Past, Present, and Future. Can J Diabetes (2016.0) 40 586-93. DOI: 10.1016/j.jcjd.2016.05.002 14. Altobelli E, Chiarelli F, Valenti M, Verrotti A, Blasetti A, Di Orio F. Family history and risk of insulin-dependent diabetes mellitus: a population-based case-control study. Acta diabetologica (1998.0) 35 57-60. DOI: 10.1007/s005920050102 15. Haynes A, Bulsara MK, Bower C, Codde JP, Jones TW, Davis EA. Independent effects of socioeconomic status and place of residence on the incidence of childhood type 1 diabetes in Western Australia. Pediatr Diabetes (2006.0) 7 94-100. DOI: 10.1111/j.1399-543X.2006.00153.x 16. Lee HY, Lu CL, Chen HF, Su HF, Li CY. Perinatal and childhood risk factors for early-onset type 1 diabetes: a population-based case-control study in Taiwan. Eur J Public Health (2015.0) 25 1024-9. DOI: 10.1093/eurpub/ckv059 17. Rytkönen M, Ranta J, Tuomilehto J, Karvonen M. Bayesian analysis of geographical variation in the incidence of type I diabetes in Finland. Diabetologia (2001.0) 44 B37-44. PMID: 11724415 18. Patterson CC, Waugh NR. Urban/rural and deprivational differences in incidence and clustering of childhood diabetes in Scotland. Int J Epidemiol (1992.0) 21 108-17. DOI: 10.1093/ije/21.1.108 19. Castillo-Reinado K, Maier W, Holle R, Stahl-Pehe A, Baechle C, Kuss O. Associations of area deprivation and urban/rural traits with the incidence of type 1 diabetes: analysis at the municipality level in North Rhine-Westphalia, Germany. Diabetic Medicine (2020.0) 37 2089-97. DOI: 10.1111/dme.14258 20. Patterson CC, Dahlquist G, Soltész G, Green A. Is childhood-onset type I diabetes a wealth-related disease? An ecological analysis of European incidence rates. Diabetologia (2001.0) 44 B9-16. DOI: 10.1007/PL00002961 21. Patterson CC, Carson DJ, Hadden DR. Epidemiology of childhood IDDM in Northern Ireland 1989–1994: low incidence in areas with highest population density and most household crowding. Diabetologia (1996.0) 39 1063-9. DOI: 10.1007/BF00400655 22. Staines A, Bodansky HJ, McKinney PA, Alexander FE, McNally RJ, Law GR. Small area variation in the incidence of childhood insulin-dependent diabetes mellitus in Yorkshire, UK: links with overcrowding and population density. Int J Epidemiol (1997.0) 26 1307-13. DOI: 10.1093/ije/26.6.1307 23. Hussen HI, Persson M, Moradi T. Maternal overweight and obesity are associated with increased risk of type 1 diabetes in offspring of parents without diabetes regardless of ethnicity. Diabetologia (2015.0) 58 1464-73. DOI: 10.1007/s00125-015-3580-1 24. Steck AK, Rewers MJ. Genetics of type 1 diabetes. Clinical chemistry (2011.0) 57 176-85. DOI: 10.1373/clinchem.2010.148221 25. Virtanen SM. Dietary factors in the development of type 1 diabetes. Pediatr Diabetes (2016.0) 17 49-55. DOI: 10.1111/pedi.12341 26. Giwa AM, Ahmed R, Omidian Z, Majety N, Karakus KE, Omer SM. Current understandings of the pathogenesis of type 1 diabetes: genetics to environment. World J Diabetes (2020.0) 11 13. DOI: 10.4239/wjd.v11.i1.13 27. Malcova H, Sumnik Z, Drevinek P, Venhacova J, Lebl J, Cinek O. Absence of breast-feeding is associated with the risk of type 1 diabetes: a case–control study in a population with rapidly increasing incidence. Eur J Pediatr (2006.0) 165 114-9. DOI: 10.1007/s00431-005-0008-9 28. Hakola L, Takkinen HM, Niinistö S, Ahonen S, Nevalainen J, Veijola R. Infant feeding in relation to the risk of advanced islet autoimmunity and type 1 diabetes in children with increased genetic susceptibility: a cohort study. Am. J. Epidemiol (2018.0) 187 34-44. DOI: 10.1093/aje/kwx191 29. Al-Ghamdi AH, Fureeh AA, Alghamdi JA, Alkuraimi WA, Alomar FF, Alzahrani FA. High prevalence of vitamin D deficiency among Saudi children and adolescents with type 1 diabetes in Albaha Region, Saudi Arabia. IOSR J. Pharm. Biol. Sci (2017.0) 12 5-10 30. Al-Daghri NM, Al-Attas OS, Alokail MS, Alkharfy KM, Yakout SM, Aljohani NJ. Lower vitamin D status is more common among Saudi adults with diabetes mellitus type 1 than in non-diabetics. BMC public health (2014.0) 14 153. DOI: 10.1186/1471-2458-14-153 31. Franchi B, Piazza M, Sandri M, Mazzei F, Maffeis C, Boner AL. Vitamin D at the onset of type 1 diabetes in Italian children. Eur J Pediatr (2014.0) 173 477-82. DOI: 10.1007/s00431-013-2204-3 32. Begum M, Pilkington R, Chittleborough C, Lynch J, Penno M, Smithers L. Caesarean section and risk of type 1 diabetes: whole-of-population study. Diabetic Medicine (2019.0) 36 1686-93. DOI: 10.1111/dme.14131 33. Cardwell CR, Stene LC, Joner G, Cinek O, Svensson J, Goldacre MJ. Caesarean section is associated with an increased risk of childhood-onset type 1 diabetes mellitus: a meta-analysis of observational studies. Diabetologia (2008.0) 51 726-35. DOI: 10.1007/s00125-008-0941-z 34. Tanaka M, Nakayama J. Development of the gut microbiota in infancy and its impact on health in later life. Allergol Int (2017.0) 66 515-22. DOI: 10.1016/j.alit.2017.07.010 35. Waernbaum I, Dahlquist G, Lind T. Perinatal risk factors for type 1 diabetes revisited: a population-based register study. Diabetologia (2019.0) 62 1173-1184. DOI: 10.1007/s00125-019-4874-5 36. Khashan AS, Kenny LC, Lundholm C, Kearney PM, Gong T, McNamee R. Gestational age and birth weight and the risk of childhood type 1 diabetes: a population-based cohort and sibling design study. Diabetes Care (2015.0) 38 2308-15. DOI: 10.2337/dc15-0897 37. Metsälä J, Hakola L, Lundqvist A, Virta LJ, Gissler M, Virtanen SM. Perinatal factors and the risk of type 1 diabetes in childhood and adolescence-A register-based case-cohort study in Finland, years 1987 to 2009. Pediatr Diabetes (2020.0) 21 586-96. DOI: 10.1111/pedi.12994 38. Cardwell CR, Stene LC, Joner G, Davis EA, Cinek O, Rosenbauer J. Birthweight and the risk of childhood-onset type 1 diabetes: a meta-analysis of observational studies using individual patient data. Diabetologia (2010.0) 53 641-51. DOI: 10.1007/s00125-009-1648-5 39. Bingley PJ, Douek IF, Rogers CA, Gale EA. Influence of maternal age at delivery and birth order on risk of type 1 diabetes in childhood: prospective population based family study. Bmj (2000.0) 321 420-4. DOI: 10.1136/bmj.321.7258.420 40. Cardwell CR, Carson DJ, Patterson CC. Parental age at delivery, birth order, birth weight and gestational age are associated with the risk of childhood Type 1 diabetes: a UK regional retrospective cohort study. Diabetic Medicine (2005.0) 22 200-6. DOI: 10.1111/j.1464-5491.2005.01369.x 41. Majeed AA, Mea KH. Risk factors for type 1 diabetes mellitus among children and adolescents in Basrah. Oman Med J (2011.0) 26 189. DOI: 10.5001/omj.2011.46 42. Henry EB, Patterson CC, Cardwell CR. A meta-analysis of the association between pre-eclampsia and childhood-onset Type 1 diabetes mellitus. Diabetic medicine (2011.0) 28 900-5. DOI: 10.1111/j.1464-5491.2011.03291.x 43. Eisenbarth GS. Banting Lecture 2009: An unfinished journey: molecular pathogenesis to prevention of type 1A diabetes. Diabetes (2010.0) 59 759-74. DOI: 10.2337/db09-1855 44. Bluestone JA, Herold K, Eisenbarth G. Genetics, pathogenesis and clinical interventions in type 1 diabetes. Nature (2010.0) 464 1293-1300. DOI: 10.1038/nature08933 45. Knip M, Virtanen SM, Seppä K, Ilonen J, Savilahti E, Vaarala O. Dietary intervention in infancy and later signs of beta-cell autoimmunity. N. Engl. J. Med (2010.0) 363 1900-8. DOI: 10.1056/NEJMoa1004809 46. Vaarala O, Ilonen J, Ruohtula T, Pesola J, Virtanen SM, Härkönen T. Removal of bovine insulin from cow’s milk formula and early initiation of beta-cell autoimmunity in the FINDIA pilot study. Arch. pediatr. adolesc. med (2012.0) 166 608-14. DOI: 10.1001/archpediatrics.2011.1559 47. Beyerlein A, Wehweck F, Ziegler AG, Pflueger M. Respiratory infections in early life and the development of islet autoimmunity in children at increased type 1 diabetes risk: evidence from the BABYDIET study. JAMA pediatrics (2013.0) 167 800-7. DOI: 10.1001/jamapediatrics.2013.158 48. Verge CF, Gianani R, Kawasaki E, Yu L, Pietropaolo M, Chase HP. Prediction of type I diabetes in first-degree relatives using a combination of insulin, GAD, and ICA512bdc/IA-2 autoantibodies. Diabetes (1996.0) 45 926-33. DOI: 10.2337/diabetes.45.7.926 49. LaGasse JM, Brantley MS, Leech NJ, Rowe RE, Monks S, Palmer JP. Successful prospective prediction of type 1 diabetes in schoolchildren through multiple defined autoantibodies: an 8-year follow-up of the Washington State Diabetes Prediction Study. Diabetes care (2002.0) 25 505-11. DOI: 10.2337/diacare.25.3.505 50. Ziegler AG, Danne T, Dunger DB, Berner R, Puff R, Kiess W. Primary prevention of beta-cell autoimmunity and type 1 diabetes–The Global Platform for the Prevention of Autoimmune Diabetes (GPPAD) perspectives. Molecular metabolism (2016.0) 5 255-62. DOI: 10.1016/j.molmet.2016.02.003 51. Ziegler AG, Rewers M, Simell O, Simell T, Lempainen J, Steck A. Seroconversion to multiple islet autoantibodies and risk of progression to diabetes in children. Jama (2013.0) 309 2473-9. DOI: 10.1001/jama.2013.6285 52. Redondo MJ, Geyer S, Steck AK, Sharp S, Wentworth JM, Weedon MN. A type 1 diabetes genetic risk score predicts progression of islet autoimmunity and development of type 1 diabetes in individuals at risk. Diabetes care (2018.0) 41 1887-94. DOI: 10.2337/dc18-0087 53. Beyerlein A, Bonifacio E, Vehik K, Hippich M, Winkler C, Frohnert BI. Progression from islet autoimmunity to clinical type 1 diabetes is influenced by genetic factors: results from the prospective TEDDY study. J. Med. Genet (2019.0) 56 602-5. DOI: 10.1136/jmedgenet-2018-105532 54. Bonifacio E, Beyerlein A, Hippich M, Winkler C, Vehik K, Weedon MN. Genetic scores to stratify risk of developing multiple islet autoantibodies and type 1 diabetes: a prospective study in children. PLoS medicine (2018.0) 15 e1002548. DOI: 10.1371/journal.pmed.1002548 55. Fernández-Edreira D, Liñares-Blanco J, Fernandez-Lozano C. Machine Learning analysis of the human infant gut microbiome identifies influential species in type 1 diabetes. Expert Syst. Appl (2021.0) 185 115648. DOI: 10.1016/j.eswa.2021.115648 56. Webb-Robertson BJ, Nakayasu ES, Frohnert BI, Bramer LM, Akers SM, Norris JM. Integration of Infant Metabolite, Genetic, and Islet Autoimmunity Signatures to Predict Type 1 Diabetes by Age 6 Years. J. Clin. Endocr (2022.0). DOI: 10.1210/clinem/dgac225 57. Altamimi AM, Hijjawi M. Performance Analysis of Supervised Classifying Algorithms to Predict Diabetes in Children. J. Xi’an Univ. Archit. amp; Technol (2020.0) 1006-7930 2010-17 58. Ruotsalainen AL, Tejesvi MV, Vänni P, Suokas M, Tossavainen P, Pirttilä AM. Child type 1 diabetes associated with mother vaginal bacteriome and mycobiome. Journal Med. Microbiol. Immunol (2022.0) 211 185-94. DOI: 10.1007/s00430-022-00741-w 59. Parviainen A, But A, Siljander H, Knip M. Decreased incidence of type 1 diabetes in young Finnish children. Diabetes care (2020.0) 43 2953-8. DOI: 10.2337/dc20-0604 60. Liu X, Vehik K, Huang Y, Elding Larsson H, Toppari J, Ziegler AG. Distinct growth phases in early life associated with the risk of type 1 diabetes: the TEDDY study. Diabetes care (2020.0) 43 556-62. DOI: 10.2337/dc19-1670 61. Zayed H. Genetic epidemiology of type 1 diabetes in the 22 Arab countries. Curr Diab Rep (2016.0) 16 1-11. DOI: 10.1007/s11892-016-0736-4 62. Robert AA, Al-Dawish A, Mujammami M, Dawish MA. Type 1 diabetes mellitus in Saudi Arabia: a soaring epidemic. Int J Pediatr (2018.0). DOI: 10.1155/2018/9408370 63. Alotaibi M, Alibrahim L, Alharbi N. Challenges associated with treating children with diabetes in saudi arabia. Diabetes Res Clin Pract (2016.0) 120 235-40. DOI: 10.1016/j.diabres.2016.08.015 64. Alazwari A, Abdollahian M, Tafakori L, Johnstone A, Alshumrani RA, Alhelal MT. Predicting age at onset of type 1 diabetes in children using regression, artificial neural network and Random Forest: A case study in Saudi Arabia. PloS one (2022.0) 17 e0264118. DOI: 10.1371/journal.pone.0264118 65. Albishi LA, AlAmri E, Mahmoud AA. Relationships among consanguinity, family history, and the onset of type 1 diabetes in children from Saudi Arabia. Prim Care Diabetes (2022.0) 16 102-6. DOI: 10.1016/j.pcd.2021.09.002 66. 66Stevenson M, Stevenson MM, BiasedUrn I. Package ‘epiR’. Tools for the analysis of epidemiological data R package version 0.9–62. 2015. 67. Wahabi H, Fayed A, Esmaeil S, Mamdouh H, Kotb R. Prevalence and complications of pregestational and gestational diabetes in Saudi women: analysis from Riyadh Mother and Baby cohort study (RAHMA). BioMed research international (2017.0). DOI: 10.1155/2017/6878263 68. Al-Rifai RH, Abdo NM, Paulo MS, Saha S, Ahmed LA. Prevalence of gestational diabetes mellitus in the middle east and North Africa, 2000–2019: A Systematic Review, Meta-Analysis, and Meta-Regression. Frontiers in endocrinology (2021.0) 12 668447. DOI: 10.3389/fendo.2021.668447 69. 69World Health Organization. WHO fact sheet: Preterm birth. World Health Organization. https://www.who.int/news-room/fact-sheets/detail/preterm-birth. Accessed 2020;26. 70. ACOG Committee Opinion No 579: definition of term pregnancy. Obstet Gynecol (2013.0) 122 1139-40. DOI: 10.1097/01.AOG.0000437385.88715.4a 71. World Health Organization: Care of the preterm and/or low-birth-weight newborn. World Health Organization (2014.0) 72. Macrosomia: Acog practice bulletin summary, number 216. Obstet Gynecol (2020.0) 135 246-248. DOI: 10.1097/AOG.0000000000003607 73. Parveen N, Zaidi S, Danish M. Development of SVR-based model and comparative analysis with MLR and ANN models for predicting the sorption capacity of Cr (VI). Process Saf Environ (2017.0) 107 428-37. DOI: 10.1016/j.psep.2017.03.007 74. Antonopoulos VZ, Papamichail DM, Aschonitis VG, Antonopoulos AV. Solar radiation estimation methods using ANN and empirical models. Comput. Electron. Agric (2019.0) 160 160-7. DOI: 10.1016/j.compag.2019.03.022 75. Shaikhina T, Lowe D, Daga S, Briggs D, Higgins R, Khovanova N. Decision tree and random forest models for outcome prediction in antibody incompatible kidney transplantation. Biomed Signal Process Control (2019.0) 52 456-62. DOI: 10.1016/j.bspc.2017.01.012 76. Berchialla P, Foltran F, Gregori D. Naïve Bayes classifiers with feature selection to predict hospitalization and complications due to objects swallowing and ingestion among European children. Saf. Sci (2013.0) 51 1-5. DOI: 10.1016/j.ssci.2012.05.021 77. Sueno HT, Gerardo BD, Medina RP. Multi-class document classification using support vector machine (SVM) based on improved Naïve bayes vectorization technique. International Journal of Advanced Trends in Computer Science and Engineering (2020.0) 9. DOI: 10.30534/ijatcse/2020/216932020 78. 78R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Vienna, Austria. 2021. https://www.R-project.org/ 79. Oommen T, Baise LG, Vogel RM. Sampling bias and class imbalance in maximum-likelihood logistic regression. Math. Geosci (2011.0) 43 99-120. DOI: 10.1007/s11004-010-9311-8 80. Reed P, Wu Y. Logistic regression for risk factor modelling in stuttering research. J. Fluen. Disord (2013.0) 38 88-101. DOI: 10.1016/j.jfludis.2012.09.003 81. Boateng EY, Abaye DA. A review of the logistic regression model with emphasis on medical research. Journal of data analysis and information processing (2019.0) 7 190-207. DOI: 10.4236/jdaip.2019.74012 82. Sperandei S. Understanding logistic regression analysis. Biochemia medica (2014.0) 24 12-8. DOI: 10.11613/BM.2014.003 83. Gholipour K, Asghari-Jafarabadi M, Iezadi S, Jannati A, Keshavarz S. Modelling the prevalence of diabetes mellitus risk factors based on artificial neural network and multiple regression. East. Mediterr. Health J (2018.0) 24. DOI: 10.26719/emhj.18.012 84. Jacobsen LM, Larsson HE, Tamura RN, Vehik K, Clasen J, Sosenko J. Predicting progression to type 1 diabetes from ages 3 to 6 in islet autoantibody positive TEDDY children. Pediatric diabetes (2019.0) 20 263-70. DOI: 10.1111/pedi.12812 85. Stoltzfus JC. Logistic regression: a brief primer. Acad Emerg Med (2011.0) 18 1099-104. DOI: 10.1111/j.1553-2712.2011.01185.x 86. Duda RO, Hart PE. Pattern classification and scene analysis. Neural Computing and Applications (1973.0) 87. Larivière B, Van den Poel D. Predicting customer retention and profitability by using random forests and regression forests techniques. Expert Syst. Appl (2005.0) 29 472-84. DOI: 10.1016/j.eswa.2005.04.043 88. Breiman L. Random forests. Mach Learn (2001.0) 45 5-32. DOI: 10.1023/A:1010933404324 89. Jiang R, Tang W, Wu X, Fu W. A random forest approach to the detection of epistatic interactions in case-control studies. BMC Bioinformatics (2009.0) 10 S65. DOI: 10.1186/1471-2105-10-S1-S65 90. Vapnik V. The nature of statistical learning theory. Springer science & business media (1999.0) 91. Noble WS. What is a support vector machine?. Nat. Biotechnol (2006.0) 24 1565-7. DOI: 10.1038/nbt1206-1565 92. Huang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W. Applications of support vector machine (SVM) learning in cancer genomics. Cancer genomics & proteomics (2018.0) 15 41-51. DOI: 10.21873/cgp.20063 93. Goh RY, Lee LS. Credit scoring: a review on support vector machines and metaheuristic approaches. Advances in Operations Research (2019.0). DOI: 10.1155/2019/1974794 94. 94Bennett KP, Blue JA. A support vector machine approach to decision trees. In1998 IEEE International Joint Conference on Neural Networks Proceedings. IEEE. 1998;3:2396–2401. 95. Wei W, Visweswaran S, Cooper GF. The application of naive Bayes model averaging to predict Alzheimer’s disease from genome-wide data. J Am Med Inform Assoc (2011.0) 18 370-5. DOI: 10.1136/amiajnl-2011-000101 96. Webb G.I. Naïve Bayes. Springer (2017.0) 895-96 97. Domingos P, Pazzani M. On the optimality of the simple Bayesian classifier under zero-one loss. Mach. Learn (1997.0) 29 103-30. DOI: 10.1023/A:1007413511361 98. 98Iyer A, Jeyalatha S, Sumbaly R. Diagnosis of diabetes using classification mining techniques. arXiv preprint arXiv:1502.03774. 2015. 99. Dong L, Li X, Xie G. Abstract and Applied Analysis (2014.0) 100. Carvalho NB, Minim VP, Silva RD, Della Lucia SM, Minim LA. Artificial neural networks (ANN): prediction of sensory measurements from instrumental data. Food Sci Technol Int (2013.0) 33 722-9. DOI: 10.1590/S0101-20612013000400018 101. Zhang X, Pan F, Wang W, Philip SY, Han J, Faloutsos C. Link Mining: Models, Algorithms, and Applications (2010.0) 505-534 102. 102Fojnica A, Osmanović A, Badnjević A. Dynamical model of tuberculosis-multiple strain prediction based on artificial neural network. In:2016 5th Mediterranean Conference on Embedded Computing (MECO);2016.290–293. 103. Kaushik AC, Sahi S. HOGPred: artificial neural network-based model for orphan GPCRs. Neural. Comput. Appl (2018.0) 29 985-92. DOI: 10.1007/s00521-016-2502-6 104. Dongare AD, Kharde RR, Kachare AD. Introduction to artificial neural network. Int. J. Eng. Technol. Innov (2012.0) 2 189-94 105. Chen C, Zhang J, Xie Y, Xiang Y, Zhou W, Hassan MM. A performance evaluation of machine learning-based streaming spam tweets detection. IEEE Trans Comput Soc Syst (2015.0) 2 65-76. DOI: 10.1109/TCSS.2016.2516039 106. Belavagi MC, Muniyal B. Performance evaluation of supervised machine learning algorithms for intrusion detection. Procedia Comput Sci (2016.0) 89 117-23. DOI: 10.1016/j.procs.2016.06.016 107. Ahmad MW, Mourshed M, Rezgui Y. Trees vs Neurons: Comparison between random forest and ANN for high-resolution prediction of building energy consumption. Energy Build (2017.0) 147 77-89. DOI: 10.1016/j.enbuild.2017.04.038 108. Li Q, Meng Q, Cai J, Yoshino H, Mochida A. Predicting hourly cooling load in the building: A comparison of support vector machine and different artificial neural networks. Energy Convers. Manag (2009.0) 50 90-6. DOI: 10.1016/j.enconman.2008.08.033 109. Nusinovici S, Tham YC, Yan MY, Ting DS, Li J, Sabanayagam C. Logistic regression was as good as machine learning for predicting major chronic diseases. J. Clin. Epidemiol (2020.0) 122 56-69. DOI: 10.1016/j.jclinepi.2020.03.002 110. Cherubini V, Carle F, Gesuita R, Iannilli A, Tuomilehto J, Prisco F. Large incidence variation of type I diabetes in central-southern Italy 1990–1995: lower risk in rural areas. Diabetologia (1999.0) 42 789-92. DOI: 10.1007/s001250051228 111. Rochmah N, Faizi M, Triastuti IW, Wicaksono G, Endaryanto A, Soetjipto S. Vitamin D level and early cow’s milk protein exposure in type 1 diabetes mellitus. Arch. Hell. Med (2022.0) 39 112. Rosenbauer J, Herzig P, Kaiser P, Giani G. Early nutrition and risk of type 1 diabetes mellitus-a nationwide case-control study in preschool children. Experimental and clinical endocrinology & diabetes (2007.0) 115 502-8. DOI: 10.1055/s-2007-973829 113. 113American Diabetes Association. Learn the genetics of diabetes. American Diabetes Association. https://www.diabetes.org/diabetes/genetics-diabetes. Accessed 2020;9. 114. Huang K, Si S, Chen R, Wang C, Chen S, Liang Y. Preterm birth and birth weight and the risk of type 1 diabetes in Chinese children. Frontiers in endocrinology (2021.0) 12 603277. DOI: 10.3389/fendo.2021.603277 115. Goldacre RR. Associations between birthweight, gestational age at birth and subsequent type 1 diabetes in children under 12: a retrospective cohort study in England, 1998–2012. Diabetologia (2018.0) 61 616-25. DOI: 10.1007/s00125-017-4493-y 116. Adar A, Shalitin S, Eyal O, Loewenthal N, Pinhas-Hamiel O, Levy M. Prevalence of early and late prematurity is similar among pediatric type 1 diabetes patients and the general population. Diabetes/Metab. Res. Rev (2018.0) 34 e2996. DOI: 10.1002/dmrr.2996 117. Tegethoff M, Olsen J, Schaffner E, Meinlschmidt G. Asthma during pregnancy and clinical outcomes in offspring: a national cohort study. Pediatrics (2013.0) 132 483-91. DOI: 10.1542/peds.2012-3686 118. Lindell N, Carlsson A, Josefsson A, Samuelsson U. Maternal obesity as a risk factor for early childhood type 1 diabetes: a nationwide, prospective, population-based case–control study. Diabetologia (2018.0) 61 130-7. DOI: 10.1007/s00125-017-4481-2 119. Fallatah AM, AlNoury A, Fallatah EM, Nassibi KM, Babatin H, Alghamdi OA. Obesity Among Pregnant Women in Saudi Arabia: A Retrospective Single-Center Medical Record Review. Cureus (2021.0) 13 e13454. DOI: 10.7759/cureus.13454
--- title: 'Psychiatric symptomatology in skin-restricted lupus patients without axis I psychiatric disorders: A post-hoc analysis' authors: - Fabien Rondepierre - Urbain Tauveron-Jalenques - Solène Valette - Aurélien Mulliez - Michel D’Incan - Sophie Lauron - Isabelle Jalenques journal: PLOS ONE year: 2023 pmcid: PMC9977055 doi: 10.1371/journal.pone.0282079 license: CC BY 4.0 --- # Psychiatric symptomatology in skin-restricted lupus patients without axis I psychiatric disorders: A post-hoc analysis ## Abstract ### Background Skin-restricted lupus is a chronic inflammatory disease associated with high rates of depression and anxiety disorders. Patients without psychiatric disorders can experience anxiety and depressive symptoms at a subclinical level, which could be risk factors for progression towards psychiatric disorders. It was decided, therefore, to investigate the presence of specific symptoms in skin-restricted lupus patients without axis I psychiatric disorders and their impact on the occurrence of axis I psychiatric disorders during the study follow-up. ### Methods Longitudinal data of 38 patients and 76 matched controls without active axis I psychiatric disorders from the LuPsy cohort were used. Depressive, neurovegetative, psychic and somatic anxiety symptom scores were established from the Montgomery-Asberg Depression Rating Scale (MADRS) and the Hamilton Anxiety Rating scale (HAMA). ### Results None of the participants had any current active axis I psychiatric disorders but the patients had personality disorders more frequently and had received more past psychotropic treatments than the controls. They also had higher MADRS and HAMA scores than the controls, in particular neurovegetative, psychic anxiety and somatic symptoms scores. No dermatological factor tested was associated with these scores, whereas being a lupus patient was associated with higher neurovegetative and somatic symptoms scores, having a current personality disorder with higher depressive and neurovegetative scores and receiving more past psychotropic treatments with psychic anxiety and somatic symptoms scores. The occurrence of psychiatric disorders during the study follow-up was associated with an elevated psychic anxiety score at baseline and past psychotropic treatment but not with history of psychiatric disorder. ### Limitations The LuPsy cohort included a large number of patients with axis I psychiatric disorders, the sample without axis I psychiatric disorders is therefore limited. ### Conclusions We observed numerous psychiatric symptoms among the skin-restricted lupus patients. They should therefore receive special attention in the management of their subclinical symptoms before they progress towards full psychiatric disorders. ## Introduction Skin-restricted lupus (SRL) is a chronic inflammatory disease characterized by many elevated cytokines [1]. SRL includes discoid lupus erythematosus (DLE), lupus tumidus (LT) and subacute cutaneous lupus erythematosus (SCLE). It is associated with high rates of depression and anxiety disorders [2, 3] as in other chronic inflammatory diseases such as psoriasis, alopecia areata, atopic dermatitis, systemic lupus erythematosus, inflammatory bowel disease, multiple sclerosis and rheumatoid arthritis [4–9]. All these diseases are characterized by longstanding systemic inflammation that, in vulnerable patients, can cause neuroinflammation and lead to depression and anxiety [10, 11]. Individuals can be considered vulnerable on the basis of genetic predisposition, psychiatric history or an exacerbated immune response [12]. In this inflammatory context, other patients can develop anxious and depressive symptoms that are not severe enough to meet DSM criteria for depression and anxiety disorders but which correspond to subclinical disorders [13, 14]. These subclinical disorders could be risk factors for progression towards genuine depression and anxiety disorders [13, 15]. It would therefore be useful to identify specific symptoms associated with chronic inflammatory diseases. Additionally, interventions in preventing progression may be of interest [16]. The first mention of the depressive and anxious effects of inflammation were made in a study of interferon therapy [17]. A later study described the chronology of symptom onset in interferon-treated patients, who first developed neurovegetative and somatic symptoms and then anxious and depressive symptoms [18]. Since then, many studies have looked for links between inflammation and depressive symptoms, including in chronic inflammatory diseases [4, 6, 8, 9, 19]. Additionally, inflammation and depressive symptoms were observed in association in obese men and in diabetic patients [20, 21]. In acute coronary heart disease patients with no history of depression, inflammation was associated with somatic symptoms [22]. In the same study, a longitudinal association was also found between inflammation and anxious symptoms [22]. In the present exploratory study, we used longitudinal data to investigate the presence of depressive, psychic anxiety, neurovegetative and somatic symptoms in SRL patients without current active axis I psychiatric disorders. First, we compared baseline symptom scores in patients and controls, looking for associated factors, and second, we attempted to identify risk factors in patients who developed psychiatric disorders during follow-up. ## Methods Data from the studies of the LuPsy cohort were used for this analysis. This cohort comprised 80 consecutive outpatients with chronic SRL (DLE and LT) or SCLE and 160 healthy control subjects without a history of lupus recruited among volunteers from a clinical research centre and workers in public hospitals, the national railway company and administrative personnel of the state education system. As reported elsewhere, the LuPsy studies were performed to compare the baseline prevalence of psychiatric and personality disorders in SRL patients and sex-, age-, and education level-matched controls [3, 23]. The cohort was approved by the local ethics committee (Comité de Protection des Personnes Sud-Est 6, reference 2008-A00343-52 / AU740, 18 June 2009). After presentation of the objectives and procedures of the study, all participants provided their written informed consent. ## Participants As the objective of this study was to investigate the presence of psychiatric symptoms in SRL patients without active axis I psychiatric disorders, all patients of the LuPsy cohort with current axis I psychiatric disorders according to the Mini International Neuropsychiatric Interview (MINI 5.0.0) [24, 25] at baseline were excluded from the study, leaving 38 patients. They were matched for gender, age (plus or minus 5 years) and level of education with 76 healthy controls without active axis I psychiatric disorders from the LuPsy cohort. The patients were followed for 2 years with a visit every 6 months (5 visits in total) including both a dermatological and a psychiatric evaluation. Controls had no follow-up. ## Dermatological evaluation The dermatologist determined the location and number of the lesions and the Cutaneous Lupus Erythematosus Disease Area and Severity Index score (CLASI) [26], whenever possible. The date of the first symptoms of the SRL, the consumption of tobacco and the past and current specific treatments were collected. ## Psychiatric evaluation The past and current axis I psychiatric disorders were explored by a psychiatrist using a structured interview, the MINI 5.0.0 [25]. The 99-item self-report Personality Diagnostic Questionnaire 4+ (PDQ-4+) was used to assess personality disorders. In its latest version, the clinical significance scale of the PDQ-4+ (CSS) which is an individual directive interview, allowed the psychiatrist to confirm or not the specific diagnosis of personality disorder suggested by a required number of positive criteria [27–29]. The psychiatric assessments could not be performed without perceiving the case-control status since most patients presented with skin lesions on a visible area. ## Symptom scoring The psychiatrist also recorded scores from the Montgomery-Asberg Depression Rating Scale (MADRS) [30] and the Hamilton Anxiety Rating scale (HAMA) [31]. Depressive, neurovegetative, psychic anxiety and somatic symptom scores were calculated by adding the scores of the following selected items: [1] apparent sadness, reported sadness, inability to feel, pessimistic thoughts and suicidal thoughts from the MADRS for depressive symptoms; [2] reduced sleep, reduced appetite and lassitude from the MADRS for neurovegetative symptoms; [3] anxious mood, tension, fears and behavior at interview from the HAMA scale for psychic anxiety symptoms and [4] muscular, sensory, cardiovascular, respiratory, gastrointestinal and genitourinary symptoms also from the HAMA scale for somatic symptoms. ## Data management Study data were collected and managed with Research Electronic Data Capture (REDCap) tools hosted at Clermont-Ferrand University Hospital [32]. The raw data supporting the conclusions of this manuscript are available at Mendeley [33]. ## Statistical analysis Statistics were computed with STATA V15 (Stata Corp, College Station, Texas, USA). The study sample was described by frequencies and percentages for categorical data and by means ± standard deviations (or median and interquartile range when data not normal) for continuous data. Normality was assessed graphically and using Shapiro Wilk’s test. Patients and controls were compared by Student’s t-test (or Mann & Whitney test when data not normal) for continuous data and by chi-squared test (or Fisher’s exact test when appropriate) for categorical data. Depressive, neurovegetative, psychic anxiety and somatic symptom scores were compared in all participants according to sociodemographic and psychiatric factors and in patients according to dermatological factors by Mann & Whitney test for categorical data and by Spearman correlation coefficient for continuous data, in order to determine the associated factors. Multivariate analyses of factors associated with the symptom scores were performed by multiple generalized linear regression, taking the group (patients vs controls) as main covariate and selecting other covariates, based on clinically relevant factors or statistically significant factors shown in univariate analysis. Results are shown as regression coefficient estimates and their $95\%$ Confidence Interval (CI). Those analyses were performed on the subgroup of patients using the same methods, to compare patients with and without occurrence of a psychiatric disorder during the follow-up. These analyses were completed by a factor analysis on mixed data using the dudi.mix function of the ade4 package in R (http://cran.r-project.org/web/packages/ade4/index.html) to visualize difference in patient’s baseline profile according to the occurrence or not of a psychiatric disorder during the follow-up. The parameters considered in this analysis were those that could impact the development of psychiatric disorder and were determined based on the univariate results, author’s experience and clinical relevance among sociodemographic (age, gender, smoking status), dermatological (number and size of lesion, type of lupus, medication) and psychiatric data (history, medication, consultation, symptom scores). Quantitative data were centered and scaled and qualitative data were converted into binary variables. All tests were two-sided and a p-value <0.05 was considered statistically significant. ## Participant characteristics The characteristics of the participants without current active axis I psychiatric disorders are similar to the whole LuPsy cohort and are given in Table 1. Briefly, patients were predominantly women ($82\%$) and smokers ($54\%$) with a mean age of 46.0 ± 15.4 years and median disease duration of 6.2 years [3.8–10.8]. Twenty eight ($74\%$) patients had chronic lupus and 12 ($26\%$) a subacute form. Twenty five patients ($68\%$) had lesions on a visible area of the body. Thirty three patients ($87\%$) were receiving treatment for lupus and 7 ($18\%$) a psychotropic medication. Table 1 \| Unnamed: 0 \| Patients \| Controls \| pValue \| \| --- \| --- \| --- \| --- \| \| \| N = 38 \| N = 76 \| pValue \| \| Sex (Female) \| 31 (82) \| 62 (82) \| 1.000 \| \| Age \| 46.0 ± 15.4 \| 46.5 ± 15.3 \| 0.853 \| \| Smokers \| 50 (54) \| 15 (19) \| < 0.001 \| \| Medical comorbidities† \| 17 (45) \| 35 (46) \| 0.894 \| \| Lupus type \| \| \| \| \| Chronic cutaneous \| 28 (74) \| \| \| \| Subacute \| 12 (26) \| \| \| \| Lupus duration, years \| 6.2 [3.8–10.8] \| \| \| \| Largest lesion size, cm2 \| 1.5 [0.2–5.5] \| \| \| \| Number of lupus lesions \| 2.0 [1.0–4.0] \| \| \| \| Number of affected areas \| 1.0 [1.0–2.5] \| \| \| \| Pruritus or burning sensations \| 17 (46) \| \| \| \| CLASI†† \| \| \| \| \| Activity \| 3.0 [0.5–4.0] \| \| \| \| Damage \| 1.0 [0.0–2.0] \| \| \| \| Lesions on visible areas††† \| 25 (68) \| \| \| \| Current lupus treatment \| 33 (87) \| \| \| \| Synthetic antimalarials \| 14 (37) \| \| \| \| Thalidomide \| 12 (32) \| \| \| \| Topical steroids \| 7 (18) \| \| \| \| Current psychotropic treatment \| 7 (18) \| \| \| \| Antidepressant \| 5 (13) \| \| \| \| Anxiolytic \| 4 (11) \| \| \| \| Hypnotic \| 1 (3) \| \| \| \| Current psychiatrist consultation \| 0 \| \| \| \| Current personality disorder †††† \| 11 (32) \| 11 (14) \| 0.030 \| \| Cluster A \| 4 (11) \| 5 (7) \| 0.472 \| \| Cluster B \| 0 \| 3 (4) \| 0.550 \| \| Cluster C \| 6 (16) \| 8 (11) \| 0.381 \| \| Depressive \| 2 (5) \| 3 (4) \| 0.746 \| \| Passive-aggressive (Negativistic) \| 0 \| 0 \| \| \| Past psychiatric disorder \| 19 (50) \| 29 (38) \| 0.227 \| \| Past psychotropic treatment \| 19 (50) \| 18 (24) \| 0.005 \| \| Antidepressant \| 9 (24) \| 10 (13) \| 0.155 \| \| Anxiolytic \| 14 (37) \| 11 (14) \| 0.007 \| \| Hypnotic \| 6 (16) \| 2 (3) \| 0.016 \| \| Past psychiatrist consultation \| 6 (16) \| 12 (16) \| 1.000 \| \| Baseline MADRS score \| 3 [1–8] \| 1 [0–3] \| 0.005 \| \| Baseline HAMA score \| 7 [2–12] \| 3 [1–5] \| 0.001 \| The patients included more commonly smokers ($54\%$ vs $19\%$, $p \leq 0.001$), had more frequently personality disorders ($32\%$ vs $14\%$, $$p \leq 0.030$$) and had received more past psychotropic treatment ($50\%$ vs $24\%$, $$p \leq 0.005$$) than controls. Unlike the controls, none of the patients, even among those who had a psychotropic treatment, had recently consulted a psychiatrist. Finally, patients had no more other medical comorbidities. ## MADRS, HAMA and symptom scores None of the participants had any current active axis I psychiatric disorders but the patients recorded higher MADRS and HAMA scores than controls, in particular for neurovegetative, psychic anxiety and somatic symptoms (Fig 1 and Table 1). In contrast, the difference in the depressive symptom scores between patients and controls was non-significant ($$p \leq 0.092$$, Fig 1). The highest total MADRS and HAMA scores for patients were 19 and 24, respectively. Fig 1: Comparison of MADRS, HAMA and symptom scores between patients and controls.Patients (in black) had significantly higher scores than controls (in white) for MADRS and HAMA total scores and for neurovegetative, psychic anxiety and somatic symptom scores. Scores are presented as Box plot with interquartile range and 95% confidence interval. Median and mean scores are presented as red line and red cross respectively. p < 0.05 and ** p < 0.01. NS, non-significant.* ## Factors associated with symptom scores in all participants Each symptom score was compared according to sociodemographic and psychiatric characteristics to identify associated factors. The main factors identified were current personality disorders and past psychotropic treatment: participants with current personality disorders had higher depressive and psychic anxiety symptom scores than participants without (2 [0–4] vs 0 [0–0], $p \leq 0.001$ and 2.5 [1–5] vs 1 [0–2], $$p \leq 0.009$$, respectively) and participants who had received psychotropic treatment had higher psychic anxiety and somatic symptom scores (2 [1–4] vs 1 [0–2], $$p \leq 0.001$$ and 2 [1–6] vs 1 [0–2], $$p \leq 0.004$$, respectively). Remarkably, no tested factor modified the neurovegetative symptom score (S1 Table). To understand why patients had higher scores than controls and to identify the factors responsible for the differences, we performed multiple generalized linear regression using the study population (patients vs controls) and the previously identified factors (current personality disorders and past psychotropic treatment) as covariates. The results are presented in Table 2. The depressive symptom score was related to the presence of a current personality disorder whereas the neurovegetative symptom score was affected only by patient status and thus no other factor affected the score. Several factors affected the psychic anxiety, including being a SRL patient, having received psychotropic treatment in the past and having a current personality disorder. The somatic symptoms score was only associated with having received past psychotropic treatment. Table 2 \| Unnamed: 0 \| Depressive symptoms \| Neurovegetative symptoms \| Psychic anxiety symptoms \| Somatic symptoms \| \| --- \| --- \| --- \| --- \| --- \| \| Patients \| 0.35 [-0.28; 0.97] \| 1.03 [0.24; 1.81]* \| 0.97 [0.12; 1.82]* \| 1.01 [-0.17; 2.18] \| \| Current personality disorder \| 1.46 [0.77; 2.14]*** \| 0.18 [-0.69; 1.04] \| 1.06 [0.13; 1.99]* \| 0.55 [-0.74; 1.85] \| \| Current psychotropic treatment \| -0.31 [-1.65; 1.04] \| -0.14 [-1.83; 1.55] \| 1.67 [-0.16; 3.50] \| 1.17 [-1.37; 3.71] \| \| Past psychotropic treatment \| 0.42 [-0.17; 1.02] \| 0.18 [-0.57; 0.93] \| 0.94 [0.12; 1.785]* \| 1.64 [0.51; 2.76] \| ## Factors associated with symptoms scores in patients Because being an SRL patient was related to neurovegetative, psychic anxiety and somatic symptoms scores, we investigated whether dermatological factors affected these scores. We found no such association (S2 Table). The same psychiatric factors, mentioned above, were related to patients’ scores (personality disorder, current antidepressant and past psychotropic treatments). Among the patients, women had higher scores for neurovegetative and somatic symptoms (S2 Table). These identified factors were included in a multiple generalized linear regression to assess their impact on symptoms scores. The results confirmed that the depressive symptoms score was only related to the presence of a current personality disorder (Coefficient estimates (CE) [$95\%$ CI], 2.05 [0.75; 3.35], $$p \leq 0.003$$). Interestingly, female patients had a higher neurovegetative symptom score (CE 1.99 [0.21; 3.77], $$p \leq 0.030$$), and patients with current antidepressant medication had a high psychic anxiety symptom score (CE 3.49 [0.43; 6.55], $$p \leq 0.027$$). In contrast, no factor in this model significantly modified the somatic score in SRL patients. ## Psychiatric evolution and associated baseline factors Longitudinal data were unavailable for 5 of the patients; the other 33 underwent 4.2 assessments on average, with an average follow-up of 21.9 ± 5.9 months. Twenty-five ($76\%$) patients completed the last assessment at 24 months. The distribution of patients’ MADRS and HAMA scores was stable during follow-up (S1 Fig). Nine patients ($27\%$) had a depressive or anxious disorder during follow-up: three depressive disorders (including one with anxiety disorder), two dysthymia, one suicide risk, two panic disorders and one social phobia. The psychiatric diagnosis was made at 12.8 ± 6.4 months of follow-up, at which time they were 47.0 ± 18.1 years old with lupus disease duration of 7.1 ± 4.8 years. The Fig 2A showed the correlation circle of the variables used in the factor analysis on mixed data. The first axis, which represented $24.7\%$ of the variance of the data considered in the analysis, was more linked to the sex, HAMA and MADRS scores, whereas the second axis, with $15.3\%$ of the variance, were more associated with the presence of past psychiatric disorder, of lupus or psychiatric medications. The Fig 2B revealed that the patients who developed a psychiatric disorder during the follow-up had a distinctive baseline profile than those who did not develop such disorders. These two profiles were mainly marked by different HAMA scores, with higher scores for the patients who developed a psychiatric disorder during the follow-up. Comparison of patients according to the occurrence or not of a psychiatric disorder confirmed that patients who developed a psychiatric disorder had higher psychic anxiety symptom scores at baseline (S3 Table). Although they had higher scores overall the only significant increases were in the psychic anxiety symptom scores (4 [3–6] vs 1.5 [0–3.5] $$p \leq 0.008$$) and the total HAMA scores (18 [6–21] vs 4.5 [1–10], $$p \leq 0.012$$) (Fig 3 and S3 Table). These patients had also received more past psychotropic treatment (hypnotic) (S3 Table). Four ($44\%$) of the 9 had a psychiatric history but did not differ from those who had not developed psychiatric disorders during follow-up (S3 Table). Fig 2:** Baseline profile of patients who developed a psychiatric disorder during the 2-years follow-up.A: Correlation circles obtained with factor analysis showed the correlation between the variables. The shorter vectors represented variables with weak correlations and less contribution to the axis (axis 1 = 24.7% of the variance and axis 2 = 15.3%). B: Patients representation (one point for each patient) showing 2 clusters: one with psychiatric disorder during follow-up (1, blue) and another without psychiatric disorder (0, red). The variables most involved in these two clusters were all HAMA scores. Fig 3: Comparison of baseline MADRS, HAMA and symptom scores in patients according to the occurrence or not of a psychiatric disorder.Patients who developed a psychiatric disorder are presented in blue (labelled psychiatric disorder), patients who never developed a psychiatric disorder during the follow-up are presented in red (labelled no psychiatric disorder). Baseline scores are presented as Box plot with interquartile range and 95% confidence interval. Median and mean scores are presented as red line and black cross respectively. Significant different: p < 0.05 and ** p < 0.01.* ## Discussion This cohort study indicated that SRL patients have a high prevalence of anxious and depressive symptoms. We show for the first time that even patients without current active axis I psychiatric disorders experience these symptoms and that over a 2-year follow-up period psychic anxiety symptoms at baseline were associated with the occurrence of psychiatric disorders. Although the SRL patients were not suffering from established axis I psychiatric disorders they had higher depression (MADRS) and anxiety (HAMA) scores than the controls. Some of the scores were very high, in certain cases close to those of patients with confirmed depressive or anxious disorders, which is evidence of subclinical psychiatric symptoms. The scores were significantly higher for somatic, neurovegetative and psychic anxiety symptoms but only tended towards significance for depressive symptoms. The depressive and psychic anxiety symptoms scores were associated with personality disorders, which was unsurprising given the strong comorbidity of the disorders [34, 35]. However, in the patient group, depressive symptoms were associated only with personality disorders. Thus, the non-significant increase in depressive symptoms scores of the patients can be explained by a greater occurrence of personality disorders. For the other symptoms, various factors could be involved in the difference in scores between the patients and controls. The psychic anxiety symptoms score was also related to past psychotropic treatment, as was the somatic symptoms score, but not to previous psychiatric disorders. Indeed, although the patients had a larger past consumption of psychotropic drugs they did not have a greater number of previous psychiatric disorders. In contrast, few patients, in comparison to controls, had consulted a psychiatrist. This could explain why patients who had no psychiatric disorders but only subclinical, mainly anxiety and somatic, symptoms, were prescribed psychotropic drugs. This result was confirmed in patients whose psychic anxiety symptoms score was associated with antidepressant treatment undergone during the study period. Interestingly, while most controls who were taking psychotropic drugs had consulted a psychiatrist no patient had. These findings are consistent with those for the general population in France, which show widespread psychotropic drug use in individuals with no established psychiatry disorders [36]. The search for past psychiatric disorders was carried out through a psychiatric interview using the MINI, a method which is considered the gold standard. However, we cannot exclude defects in the recall of the disorders, or the psychotropic treatments or the psychiatrist consultations. Being an SRL patient was related to high neurovegetative and psychic anxiety symptoms scores, irrespective of the other factors tested. Of these, the only significant factor was the neurovegetative score. These results suggest that SRL plays a role in the occurrence of these symptoms, possibly via chronic inflammation since no other dermatological factor had any effect. Several studies reported that all patients on interferon-α therapy had neurovegetative symptoms [37, 38] and others that these symptoms were correlated with the level of inflammation [39]. In the latter report, Jokela et al. point to the correlation between C-reactive protein (CRP) and sleep problems, tiredness and changes in appetite, the specific neurovegetative symptoms of depression. Other studies on inflammation evidenced an increase in anxious and somatic symptoms [40, 41]. Unfortunately, we are unable to confirm with certainty the precise effect of inflammation on these symptoms because biological data are not available. However, none of the dermatological factors tested were associated with the different symptoms and thus the hypothesis of inflammation could be envisaged. A link between inflammation and psychiatric symptoms has already been established in other chronic inflammatory diseases such as SLE [42], psoriasis [43], ulcerative colitis [44], diabetes [20] and obesity [21]. Our results showed a sex-related difference in the patient group in neurovegetative symptoms, which were more common among females. Still under the hypothesis of a role of inflammation, other studies have reported differences according to sex, notably in somatic symptoms in acute coronary heart disease [22]. In obese patients, inflammation was associated with depressive symptoms solely in the male participants [21] and in a model of inflammation induced by lipopolysaccharide, the women had more cytokines and more anxiety symptoms [41]. These findings suggest there are biological differences between women and men in the relationship between inflammation and psychiatric symptoms that could be explained by differences observed in inflammation [45, 46] and in depression and anxiety [47]. Our study also showed that the occurrence of psychiatric disorders was associated with high baseline symptom scores and with past psychotropic treatment. As mentioned above, the patients who were prescribed psychotropic drugs to alleviate their symptoms were nevertheless not suffering from more numerous psychiatric disorders nor had greater recourse to psychiatric consultations. A more detailed analysis shows that these patients had received more hypnotic drugs, in all likelihood for sleep disturbance, which is a neurovegetative symptom, and more antidepressants, which are effective against depressive, anxious and somatic symptoms [18]. Immunotherapy studies have shown that the development of depressive and anxious disorders are related to high depressive and anxious symptoms scores before the initiation of treatment [37, 38, 48]. These findings were confirmed in the present observational study of SRL patients. They also support the hypothesis of the “switch in vulnerable patients”, in which depressive, neurovegetative and anxious symptoms related at the outset to inflammation subsequently lead to psychiatric disorders [12]. In chronic diseases, depression is a factor associated with poor quality of life, exacerbation of symptoms and increased hospital admissions and mortality rates [11]. It is therefore necessary to identify patients at risk and to tailor their management so as to limit the occurrence of psychiatric disorders. Antidepressants, which are effective against depressive, anxious and somatic symptoms [18], could achieve this aim and at the same time, probably owing to their anti-inflammatory action, exert a beneficial effect on dermatological disorders [49, 50]. Another study also showed that the CRP level could be taken into consideration in the choice of antidepressant [51]. These findings show that psychiatric disorders are not always associated with an increase in inflammation but, if that is the case, particularly in certain chronic inflammatory diseases, their management and treatment need to be adapted. The results of the present study are in line with those from the works cited above: high psychic anxiety symptoms scores were associated with past psychotropic treatment (mainly anxiolytics and hypnotics that are mostly prescribed by general practitioners in France) and the occurrence of psychiatric disorders during the follow-up period. However, only two of the nine patients who developed a psychiatric disorder had been receiving antidepressant drug. Our post-hoc analysis hypothesized that the presence of psychiatric symptoms in SRL patients could have been due to the effects of inflammation but the initial cohort study had not included a biological assay thus it was not possible to study such link via a biological analysis. However, our study was performed in the context of chronic inflammatory disease with increased cytokine production [1] and no other dermatological factor tested was associated with the different symptoms. Many studies in the past used only overall scale scores and numerous others depression scales: publications on anxiety are rarer and more recent. In addition, many scales are designed to measure depressive and anxious symptoms but they do not all rate the same ones, and the symptoms are not always classified in the same manner. In our study, for example, we designated sleep problems, tiredness and changes in appetite as neurovegetative symptoms whereas Jokela et al. referred to them as “specific symptoms of depression” [39]. These discrepancies make it difficult to meaningfully compare studies. We used the MADRS for depressive and neurovegetative symptoms and the HAMA scale for psychic anxiety and somatic symptoms so as to cover all the symptoms as widely as possible using validated tools. Only the cognitive domain could not be studied, owing to the lack of data (there was just one cognitive item on each scale). Because of the limited number of patients enrolled in the study it was not possible to perform multivariate analysis of the factors associated with the occurrence of psychiatric disorders during the follow-up. Nevertheless, the results were sufficient to evidence a greater presence of symptoms in the SRL patients with no axis 1 psychiatric disorders than in the healthy matched controls. Besides the limited number of SRL patients enrolled in the study, we note differences at baseline between patients and controls concerning current personality disorders and past anxiolytic and hypnotic treatments. Moreover, the follow-up was not identical for all the patients since only $76\%$ of them attended the last visit, which could have led to an underestimation of the development of axis I psychiatric disorders; however, it is important to emphasize that there were no differences between them regarding the MADRS and HAMA scores at baseline. In conclusion, this study shows that SRL patients who were not suffering from current active axis I psychiatric disorders nevertheless experienced numerous neurovegetative, somatic and psychic anxiety symptoms. Those with marked symptoms of psychic anxiety were at risk of developing ulterior psychiatric disorders. This particular patient group should be closely monitored and their subclinical symptoms carefully managed before they progress towards a full psychiatric disorder. ## References 1. Robinson ES, Werth VP. The role of cytokines in the pathogenesis of cutaneous lupus erythematosus. Cytokine (2015) 73 326-34. DOI: 10.1016/j.cyto.2015.01.031 2. Achtman J, Kling MA, Feng R, Okawa J, Werth VP. A cross-sectional study of untreated depression and anxiety in cutaneous lupus erythematosus and dermatomyositis. J Am Acad Dermatol (2016) 74 377-9. DOI: 10.1016/j.jaad.2015.09.016 3. Jalenques I, Rondepierre F, Massoubre C, Haffen E, Grand JP, Labeille B. High prevalence of psychiatric disorders in patients with skin-restricted lupus: a case-control study. Br J Dermatol (2016) 174 1051-60. DOI: 10.1111/bjd.14392 4. Dowlatshahi EA, Wakkee M, Arends LR, Nijsten T. The prevalence and odds of depressive symptoms and clinical depression in psoriasis patients: a systematic review and meta-analysis. J Invest Dermatol (2014) 134 1542-51. DOI: 10.1038/jid.2013.508 5. Ghanizadeh A, Ayoobzadehshirazi A. A review of psychiatric disorders comorbidities in patients with alopecia areata. Int J Trichology (2014) 6 2-4. DOI: 10.4103/0974-7753.136746 6. Marrie RA, Walld R, Bolton JM, Sareen J, Walker JR, Patten SB. Increased incidence of psychiatric disorders in immune-mediated inflammatory disease.. J Psychosom Res (2017) 101 17-23. DOI: 10.1016/j.jpsychores.2017.07.015 7. Meszaros ZS, Perl A, Faraone SV. Psychiatric symptoms in systemic lupus erythematosus: a systematic review. J Clin Psychiatry (2012) 73 993-1001. DOI: 10.4088/JCP.11r07425 8. Rønnstad ATM, Halling-Overgaard AS, Hamann CR, Skov L, Egeberg A, Thyssen JP. Association of atopic dermatitis with depression, anxiety, and suicidal ideation in children and adults: A systematic review and meta-analysis. J Am Acad Dermatol (2018) 79 448-456.e30. DOI: 10.1016/j.jaad.2018.03.017 9. Zhang L, Fu T, Yin R, Zhang Q, Shen B. Prevalence of depression and anxiety in systemic lupus erythematosus: a systematic review and meta-analysis. BMC Psychiatry (2017) 17 70. DOI: 10.1186/s12888-017-1234-1 10. Dantzer R O. ’Connor JC, Freund GG, Johnson RW, Kelley KW. From inflammation to sickness and depression: when the immune system subjugates the brain. Nat Rev Neurosci (2008) 9 46-56. PMID: 18073775 11. Maes M, Kubera M, Obuchowiczwa E, Goehler L, Brzeszcz J. Depression’s multiple comorbidities explained by (neuro)inflammatory and oxidative & nitrosative stress pathways. Neuro Endocrinol Lett (2011) 32 7-24. PMID: 21407167 12. Cytokine Dantzer R.. sickness behavior, and depression. Immunol Allergy Clin North Am (2009) 29 247-64. PMID: 19389580 13. Cuijpers P, Smit F. Subthreshold depression as a risk indicator for major depressive disorder: a systematic review of prospective studies. Acta Psychiatr Scand (2004) 109 325-31. DOI: 10.1111/j.1600-0447.2004.00301.x 14. Karsten J, Nolen WA. Penninx BWJH, Hartman CA. Subthreshold anxiety better defined by symptom self-report than by diagnostic interview. J Affect Disord (2011) 129 236-43. PMID: 20956020 15. Korte KJ, Brown MAS, Schmidt NB. Anxiety sensitivity and impairment: evidence for a direct association and partial mediation by subclinical anxiety symptoms. J Affect Disord (2013) 151 875-81. DOI: 10.1016/j.jad.2013.07.026 16. Jorm AF, Griffiths KM. Population promotion of informal self-help strategies for early intervention against depression and anxiety. Psychol Med (2006) 36 3-6. DOI: 10.1017/S0033291705005659 17. McDonald EM, Mann AH, Thomas HC. Interferons as mediators of psychiatric morbidity. An investigation in a trial of recombinant alpha-interferon in hepatitis-B carriers. Lancet (1987) 2 1175-8. DOI: 10.1016/s0140-6736(87)91319-5 18. Capuron L, Gumnick JF, Musselman DL, Lawson DH, Reemsnyder A, Nemeroff CB. Neurobehavioral effects of interferon-alpha in cancer patients: phenomenology and paroxetine responsiveness of symptom dimensions. Neuropsychopharmacology (2002) 26 643-52. DOI: 10.1016/S0893-133X(01)00407-9 19. Jalenques I, Bourlot F, Martinez E, Pereira B, D’Incan M, Lauron S. Prevalence and Odds of Anxiety Disorders and Anxiety Symptoms in Children and Adults with Psoriasis: Systematic Review and Meta-analysis. Acta Derm Venereol (2022) 102 adv00769. DOI: 10.2340/actadv.v102.1386 20. Herder C, Fürstos JF, Nowotny B, Begun A, Strassburger K, Müssig K. Associations between inflammation-related biomarkers and depressive symptoms in individuals with recently diagnosed type 1 and type 2 diabetes. Brain Behav Immun (2017) 61 137-45. DOI: 10.1016/j.bbi.2016.12.025 21. Vetter ML, Wadden TA, Vinnard C, Moore RH, Khan Z, Volger S. Gender differences in the relationship between symptoms of depression and high-sensitivity CRP. Int J Obes (Lond). (2013) 37 S38-43. DOI: 10.1038/ijo.2013.95 22. Steptoe A, Wikman A, Molloy GJ, Messerli-Bürgy N, Kaski JC. Inflammation and symptoms of depression and anxiety in patients with acute coronary heart disease. Brain Behav Immun (2013) 31 183-8. DOI: 10.1016/j.bbi.2012.09.002 23. Jalenques I, Rondepierre F, Massoubre C, Bonnefond S, Schwan R, Labeille B. High Prevalence of Personality Disorders in Skin-restricted Lupus Patients.. Acta Derm Venereol (2017) 97 941-6. DOI: 10.2340/00015555-2691 24. Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E. The Mini-International Neuropsychiatric Interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry (1998) 59 22-33. PMID: 9881538 25. Lecrubier Y, Sheehan D, Weiller E, Amorim P, Bonora I, Harnett Sheehan K. The Mini International Neuropsychiatric Interview (MINI). A short diagnostic structured interview: reliability and validity according to the CIDI. European Psychiatry (1997) 12 224-31 26. Albrecht J, Taylor L, Berlin JA, Dulay S, Ang G, Fakharzadeh S. The CLASI (Cutaneous Lupus Erythematosus Disease Area and Severity Index): an outcome instrument for cutaneous lupus erythematosus. J Invest Dermatol (2005) 125 889-94. DOI: 10.1111/j.0022-202X.2005.23889.x 27. Hyler S.. (1994) 28. Bouvard M, Cosma P. An exploratory study of a personality disorders questionnaire. Encephale (2008) 34 517-25. PMID: 19068342 29. Laconi S, Cailhol L, Pourcel L, Thalamas C, Lapeyre-Mestre M, Chabrol H. Dépistage des troubles de la personnalité avec la version française du Personality Diagnostic Questionnaire-4+ dans une population psychiatrique: une étude préliminaire. Annales Médico-psychologiques, revue psychiatrique (2016) 174 39-44 30. Montgomery SA, Asberg M. A new depression scale designed to be sensitive to change. Br J Psychiatry (1979) 134 382-9. DOI: 10.1192/bjp.134.4.382 31. Hamilton M.. The assessment of anxiety states by rating. Br J Med Psychol (1959) 32 50-5. DOI: 10.1111/j.2044-8341.1959.tb00467.x 32. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)—a metadata-driven methodology and workflow process for providing translational research informatics support.. J Biomed Inform (2009) 42 377-81. DOI: 10.1016/j.jbi.2008.08.010 33. Rondepierre F, Tauveron-Jalenques U, Valette S, Mulliez A, D’Incan M, Lauron S. Psychiatric symptomatology in skin-restricted lupus patients without psychiatric disorders: a post-hoc analysis. Mendeley Data (2022) V1. DOI: 10.17632/bdxf94z5xw.1 34. Corruble E, Ginestet D, Guelfi JD. Comorbidity of personality disorders and unipolar major depression: a review. J Affect Disord (1996) 37 157-70. DOI: 10.1016/0165-0327(95)00091-7 35. Latas M, Milovanovic S. Personality disorders and anxiety disorders: what is the relationship?. Curr Opin Psychiatry (2014) 27 57-61. DOI: 10.1097/YCO.0000000000000025 36. Grolleau A, Cougnard A, Bégaud B, Verdoux H. Psychotropic drug use and correspondence with psychiatric diagnoses in the mental health in the general population survey. Encephale (2008) 34 352-9. PMID: 18922237 37. Capuron L, Ravaud A, Miller AH, Dantzer R. Baseline mood and psychosocial characteristics of patients developing depressive symptoms during interleukin-2 and/or interferon-alpha cancer therapy. Brain Behav Immun (2004) 18 205-13. DOI: 10.1016/j.bbi.2003.11.004 38. Constant A, Castera L, Dantzer R, Couzigou P, de Ledinghen V, Demotes-Mainard J. Mood alterations during interferon-alfa therapy in patients with chronic hepatitis C: evidence for an overlap between manic/hypomanic and depressive symptoms. J Clin Psychiatry (2005) 66 1050-7. DOI: 10.4088/jcp.v66n0814 39. Jokela M, Virtanen M, Batty GD, Kivimäki M. Inflammation and Specific Symptoms of Depression.. JAMA Psychiatry (2016) 73 87-8. DOI: 10.1001/jamapsychiatry.2015.1977 40. Hou R, Baldwin DS. A neuroimmunological perspective on anxiety disorders. Hum Psychopharmacol (2012) 27 6-14. DOI: 10.1002/hup.1259 41. Lasselin J, Elsenbruch S, Lekander M, Axelsson J, Karshikoff B, Grigoleit JS. Mood disturbance during experimental endotoxemia: Predictors of state anxiety as a psychological component of sickness behavior. Brain Behav Immun (2016) 57 30-7. DOI: 10.1016/j.bbi.2016.01.003 42. Postal M, Lapa AT, Sinicato NA, de Oliveira Peliçari K, Peres FA, Costallat LTL. Depressive symptoms are associated with tumor necrosis factor alpha in systemic lupus erythematosus. J Neuroinflammation (2016) 13 5. DOI: 10.1186/s12974-015-0471-9 43. Krishnadas R, Nicol A, Sassarini J, Puri N, Burden AD, Leman J. Circulating tumour necrosis factor is highly correlated with brainstem serotonin transporter availability in humans. Brain Behav Immun (2016) 51 29-38. DOI: 10.1016/j.bbi.2015.08.005 44. Häuser W, Schmidt C, Stallmach A. Depression and mucosal proinflammatory cytokines are associated in patients with ulcerative colitis and pouchitis—A pilot study.. J Crohns Colitis (2011) 5 350-3. DOI: 10.1016/j.crohns.2011.03.001 45. Gubbels Bupp MR. Sex, the aging immune system, and chronic disease. Cell Immunol (2015) 294 102-10. DOI: 10.1016/j.cellimm.2015.02.002 46. Nelson LH, Lenz KM. The immune system as a novel regulator of sex differences in brain and behavioral development. J Neurosci Res (2017) 95 447-61. DOI: 10.1002/jnr.23821 47. Altemus M, Sarvaiya N, Neill Epperson C. Sex differences in anxiety and depression clinical perspectives. Front Neuroendocrinol (2014) 35 320-30. DOI: 10.1016/j.yfrne.2014.05.004 48. Capuron L, Ravaud A. Prediction of the depressive effects of interferon alfa therapy by the patient’s initial affective state. N Engl J Med (1999) 340 1370. DOI: 10.1056/NEJM199904293401716 49. Eskeland S, Halvorsen JA, Tanum L. Antidepressants have Anti-inflammatory Effects that may be Relevant to Dermatology: A Systematic Review. Acta Derm Venereol. (2017) 97 897-905. DOI: 10.2340/00015555-2702 50. Jalenques I, Rondepierre F, Massoubre C, Labeille B, Perrot JL, Mulliez A. Treatment of Psychiatric Disorders and Skin-Restricted Lupus Remission: A Longitudinal Study.. JAMA Dermatol (2017) 153 1331-2. DOI: 10.1001/jamadermatol.2017.3590 51. Uher R, Tansey KE, Dew T, Maier W, Mors O, Hauser J. An inflammatory biomarker as a differential predictor of outcome of depression treatment with escitalopram and nortriptyline. Am J Psychiatry (2014) 171 1278-86. DOI: 10.1176/appi.ajp.2014.14010094
--- title: Contribution of labor related gene subtype classification on heterogeneity of polycystic ovary syndrome authors: - Jue Zhou - Zhou Jiang - Leyi Fu - Fan Qu - Minchen Dai - Ningning Xie - Songying Zhang - Fangfang Wang journal: PLOS ONE year: 2023 pmcid: PMC9977056 doi: 10.1371/journal.pone.0282292 license: CC BY 4.0 --- # Contribution of labor related gene subtype classification on heterogeneity of polycystic ovary syndrome ## Abstract ### Objective As one of the most common endocrine disorders in women of reproductive age, polycystic ovary syndrome (PCOS) is highly heterogeneous with varied clinical features and diverse gestational complications among individuals. The patients with PCOS have 2-fold higher risk of preterm labor which is associated with substantial infant morbidity and mortality and great socioeconomic cost. The study was designated to identify molecular subtypes and the related hub genes to facilitate the susceptibility assessment of preterm labor in women with PCOS. ### Methods Four mRNA datasets (GSE84958, GSE5090, GSE43264 and GSE98421) were obtained from Gene Expression Omnibus database. Twenty-eight candidate genes related to preterm labor or labor were yielded from the researches and our unpublished data. Then, we utilized unsupervised clustering to identify molecular subtypes in PCOS based on the expression of above candidate genes. Key modules were generated with weighted gene co-expression network analysis R package, and their hub genes were generated with CytoHubba. The probable biological function and mechanism were explored through Gene Ontology analysis and Kyoto Encyclopedia of Genes and Genomes pathway analysis. In addition, STRING and Cytoscape software were used to identify the protein-protein interaction (PPI) network, and the molecular complex detection (MCODE) was used to identify the hub genes. Then the overlapping hub genes were predicted. ### Results Two molecular subtypes were found in women with PCOS based on the expression similarity of preterm labor or labor-related genes, in which two modules were highlighted. The key modules and PPI network have five overlapping five hub genes, two of which, GTF2F2 and MYO6 gene, were further confirmed by the comparison between clustering subgroups according to the expression of hub genes. ### Conclusions Distinct PCOS molecular subtypes were identified with preterm labor or labor-related genes, which might uncover the potential mechanism underlying heterogeneity of clinical pregnancy complications in women with PCOS. ## Introduction Polycystic ovary syndrome (PCOS) is one of the most common endocrine disorders in women of reproductive age [1]. Its primary features are menstrual dysfunction, hyperandrogenism and polycystic ovary, but highly heterogeneous [2]. The current treatment strategies for the patients with PCOS are to reduce insulin resistance in order to reach a reduction of compensatory hyperinsulinemia, and to improve the metabolic and ovulatory features. For the overweight and obese PCOS patients, although physical activity and lifestyle change are the first steps to achieve weight loss, insulin-sensitizer drugs are the recommended first-line therapy, and many new insights have also been provided in the strategies for PCOS [3]. Myo-inositol and d-chiro-inositol have very specific physiological roles, however, they should be evaluated on the patients’ conditions before the treatment and the effects of inositol therapy on different PCOS phenotypes needs further investigation [4]. Moreover, as PCOS causes a rising risk of maternal, fetal, and neonatal complications, including pregnancy-induced hypertension, preeclampsia, gestational diabetes mellitus, spontaneous preterm birth, an increased necessity for a cesarean section, elevated neonatal morbidity, prematurity, fetal growth restriction, birth weight variations, and transfer to the Neonatal Intensive Care Unit, a closer follow-up should be offered to PCOS women during pregnancy [5]. Although the causes of PCOS remain obscure, it is underpinned by a complex genetic and epigenetic architecture [1, 6, 7]. PCOS and PCOS-related gestational complications influence the intrauterine environment, leading to adverse developmental programming of the offspring for long-term, chronic health conditions [8, 9]. As preterm birth affects 1 in 10 pregnancies worldwide [10], the women with PCOS seemed to have a 2-fold increased risk of preterm labor, including both spontaneous preterm labor and indicated preterm labor which attributes to certain medical scenarios [11, 12].The preterm labor was associated with the substantial infant morbidity and mortality, long-term consequences of offspring as well as a huge socioeconomic cost [13–15]. Although the etiology of spontaneous preterm birth and the mechanism of labor is complex and unclear, a series of candidate genes have been reported to be involved in the preterm labor and labor [16–25]. In the past decade, the wide application of microarray technology and accurate RNA-sequencing technology has made it more convenient to reveal the mechanism underlying complex diseases (such as PCOS), on the basis of which our recent work uncovered gene biomarkers and developed a novel diagnostic model of PCOS [26]. Here, to elaborate the heterogeneity of preterm labor risk in women with PCOS, we analyzed the expression of previously reported preterm labor or labor related genes in PCOS based on public database- Gene Expression Omnibus (GEO) database, and attempted to classified PCOS into molecular subtypes through bioinformatics analysis. ## Data sources The NCBI-GEO database was searched for screening expression datasets, including microarray and RNA-seq, in women with PCOS. To minimize the heterogeneity among various tissues, four independent expression datasets in adipose tissue of women with PCOS were finally selected. The expression profiling by high throughput sequencing GSE84958, GSE5090, GSE43264 and GSE98421 were based on GPL16791, GPL96, GPL15362 and GPL570 platforms, with sample size of 15, 9, 8 and 4, respectively (Table 1). Table 1 \| Dataset ID \| PCOS \| Data type \| Tissue type \| Country \| \| --- \| --- \| --- \| --- \| --- \| \| GSE84958 \| 15 \| RNA-seq \| adipose \| UK \| \| GSE5090 \| 9 \| microarray \| omental adipose \| Spain \| \| GSE43264 \| 8 \| microarray \| Subcutaneous adipose \| Ireland \| \| GSE98421 \| 4 \| microarray \| subcutaneous adipose \| USA \| ## Collecting preterm or labor related genes A literature review of English language studies was undertaken in the PubMed databases until November 23, 2020. Two independent review authors (JZ & ZJ) manually extracted the preterm or labor related genes from each eligible article, relevant review articles or book. Any disagreements were resolved by discussion with a third review author (FW). ## Data preprocessing All datasets were downloaded as txt files, and outputs from mRNA array and RNA-seq were normal-exponential background corrected and then between-arrays quantile normalized using limma R package. Unsupervised cluster analysis of preterm or labor related genes was performed using the Consensus Cluster Plus R package (1.46.0) to select the best cluster group. Differential expression analysis of subgroups was performed using the Limma R package (3.36.5). The differential expressed genes were determined by two criteria: 1) the threshold value was greater than 1.0, and 2) the p-value calculated from pooled t-test was less than 0.05 and the corresponding confidence intervals were $95\%$ [27]. ## Identification of molecular subtypes of PCOS The consensus k means clustering was utilized to perform consistent clustering and selecting of PCOS molecular subtypes based on preterm or labor related gene expression profiles. The optimal cluster number was determined by cumulative distribution function (CDF) curves of the consensus score, clear separation of the consensus matrix heatmaps, characteristics of the consensus cumulative distribution function plots, and adequate pair wise–consensus values between cluster members [26]. Principal Components Analysis (PCA) was used for confirmation of molecular clusters of PCOS samples with R package ggplot2. ## Functional annotation of the key module genes We used weighted gene co-expression network analysis (WGCNA) R package to determine the genes correlated to molecular subtypes within all expressed genes in four GEO datasets. Then, Gene Ontology (GO) and KEGG (Kyoto Encyclopedia of Genes and Genomes) analysis were performed to elaborate the functions and associated pathways of the key module genes in PCOS subtypes using ClusterProfiler R packages with $P \leq 0.05$ as the significance threshold. ## Construction of protein-protein interaction (PPI) networks In order to determine the molecular mechanisms of signaling pathways and cellular activities in PCOS, the PPI network of the key module genes was constructed and visualized using the STRING (https://string-db.org) database. ## Prediction of hub genes in PCOS Hub genes in the key modules were selected using CytoHubba through connection degree method. The Cytoscape software (http://www.cytoscape.org) was utilized to yeild the top 10 hub genes in PPI network using top degree method. Molecular Complex Detection (MCODE) was used to identify key clusters of genes within PPI network. Finally, we summarized the overlapping genes between results of MCODE and CytoHubba to create a consensus of predictions to identify more accurate hub genes [26]. ## Characteristics of datasets and patients After search strategy and selection, four mRNA datasets, which was from UK, Spain, Ireland and USA, were finally enrolled in current study with the total sample size of 36. One dataset GSE84958 was got from RNA-seq analysis, and all the rest datasets was got from Array analysis. Since the women in GSE5090 dataset were diagnosed as PCOS with the presence of oligoovulation, clinical and/or biochemical hyperandrogenism in 2006 [28], whereas the diagnostic criteria of the other studies in 2014, 2017 and 2018 were not available. The PCOS patients in GSE5090 underwent bariatric surgery because of morbid obesity, while those in GSE98421 were lean. ## The genes related to preterm labor or labor According to previously published literature [16–25] and unpublished data of our group, genes related to preterm labor or labor were yielded, and shown in Fig 1A. Four genes played roles in induction of uterine contraction, including hematopoietic prostaglandin D synthase (HPGDS), aldo-keto reductase family 1 member C3 (AKR1C3) and ATP binding cassette subfamily C member 4 (ABCC4) as well as corticotropin releasing hormone (CRH) and its receptor (CRHR1). Seven genes were associated with inflammation and immune response, including interleukin 6 (IL6), tumor necrosis factor (TNF), interleukin 1 beta (IL1B), complement C3 (C3), complement factor H (CFH), complement C1r (C1R), toll like receptor 8 (TLR8) and endoplasmic reticulum aminopeptidase 2 (ERAP2). Three genes were suggested as transcription regulators, including sirtuin 1 (SIRT1), tripartite motif containing 28 (TRIM28), nuclear factor kappa B subunit 1 (NFKB1). There existed seven genes which influenced cell proliferation, migration, adhesion and metabolism: ADAM metallopeptidase with thrombospondin type 1 motif 12 (ADAMTS12), ADAMTS16, insulin like growth factor binding protein 1 (IGFBP1), IGFBP2, IGFBP6, tenascin C (TNC), Fos proto-oncogene (FOS) and FosB proto-oncogene (FOSB). Proteins coded by hydroxysteroid 17-beta dehydrogenase 4 (HSD17B4), androgen receptor (AR), estrogen receptor 1 (ESR1), peroxisome proliferator activated receptor gamma (PPARG) genefunctioned in metabolism and function of steroids. Next, a correlation analysis was performed to explore the correlation among the genes of interest (Fig 1B). Fig 1: Identification of molecular subtypes in PCOS based on preterm labor or labor-related genes.(A) Chromosomal distribution diagram of preterm labor or labor-related genes involved in the present study. (B) The relationship among preterm labor or labor-related genes of interest were shown. (C) The cumulative distribution function (CDF) curves of consensus scores based on different subtype number (k = 2, 3, 4, 5, 6) and the corresponding color are represented. (D) The CDF Delta area curve of all samples when k = 2. (E) A relative stable partitioning of the samples at k = 2 in consensus heatmap. (F) In PCA analysis, the symbols represent the gene expressed differently in two clusters. (G) The expression heatmap of the preterm labor or labor-related genes among various molecular and clinical subtypes. ## The molecular subtyping in PCOS based on preterm labor or labor-related genes Based on the expression similarity of preterm labor or labor-related genes, women with PCOS were divided into two molecular subtypes with clustering stability $k = 2$ (Fig 1C–1E). The clustering classification of two subgroups in patients with PCOS was verified with principal component analysis (PCA) (Fig 1F). Fig 1G indicated a distinct expression pattern in the genes of interest profiles between the two molecular subtypes, and age group from 22–39 and subcutaneous distribution of adipose tissue was mainly included in cluster 2. ## Heterogeneity of biological process in key modules of PCOS WGCNA identified 4 modules in the PCOS population (Fig 2A–2C). Cluster 1 negatively correlated with 4 modules, whereas cluster 2 positively correlated with 4 modules. Number of the related genes in each module was as follow: 4, 36, 28 and 80 genes for brown, blue, grey and turquoise modules, respectively. Thus, we used blue and turquoise modules related genes to furthermore explore biological function. Fig 2: Construction of expression modules by WGCNA package.(A) Analysis of the scale-free fit index for various soft-thresholding powers and analysis of the mean connectivity for various soft-thresholding powers. (B) The cluster dendrogram of genes. Each branch represents one gene, and every color below represents one expression module. (C) Heatmap of the correlation between module genes and the clusters. (D-F) Top 10 terms from a GO analysis of molecular function, biological process and cellular component in the blue and turquoise modules. (G) Top 10 terms were clustered by KEGG pathway analysis in the blue and turquoise modules. GO enrichment and KEGG pathway analyses were conducted. With GO analyses, genes in the blue and turquoise modules were found to be primarily associated with phospholipid metabolism process, especially phosphotransferase activity, as well as DNA transcription (Fig 2D–2F). Additionally, KEGG pathway analysis lent support to the above result, and phospholipid metabolism and DNA transcription related pathways were enriched (Fig 2G). ## Hub genes for PCOS On one hand, to screen the upstream regulators with high connectivity, we identified the 81 hub genes for the key modules (blue and turquoise); On the other hand, following PPI networks construction, top 10 hub genes were identified with Cytoscape (Fig 3A). Finally, five hub genes were found overlapped between the above two analyses and they were considered as hub genes for PCOS, including MYO6, ACTL6A, NCBP2, GTF2F2 and MRPL13 (Fig 3B). Fig 3: Grouping based on hub genes of PCOS.(A) Top10 hub genes were identified and PPI networks were established with Cytoscape. (B) Venn diagram for the overlapping genes between the above 10 hub genes and hub genes for the key modules. (C) The expression comparison of the five overlapped hub genes between cluster1 and cluster2. (D) The relationship among these overlapped hub genes were shown. (E) A relative stable partitioning of the samples at k = 2 in consensus heatmap. (F) In PCA analysis, the symbols represent the gene expressed differently in hub_cluster1 and hub_cluster2. (G) The expression heatmap of these overlapped hub genes in hub_cluster1 and hub_cluster2. (H) The expression comparison of the five overlapped hub genes between hub_cluster1 and hub_cluster2. (I) Sankey diagram for the links among the preterm labor or labor related gene clusters, hub gene clusters, age subgroups and adipose tissue subtypes of PCOS. ## Differential expression of hub genes between subtypes in PCOS To confirm the roles of hub genes in PCOS subtypes, we compared the expression of the five hub genes mentioned above between various subtypes. First, we observed the significantly increased expression of all the five hub genes between cluster1 and cluster2 divided according to the expression of genes related to preterm labor or labor (Fig 3C). Then, we investigated the expression of these hub genes among adipose tissue subtypes (S1A Fig) and various age subgroups (S1B Fig), but did not find any difference. To understand the expression correlation among the five hub genes, we carried out a Pearson analysis, and found that they were positively related with each other (Fig 3D). ## Grouping by cluster analysis based on hub genes of PCOS Based on the five hub genes, cluster analysis revealed that the 36 women with PCOS could be classified into 2 subgroups: hub-cluster1 and hub-cluster2 (Fig 3E and 3F). The expression patterns of the five hub genes in both clinical subgroups were showed as heatmap (Fig 3G). To further confirm the impacts of hub genes, the expression comparisons were performed between the clinical subgroups, and then a significant change was found for GTF2F2 and MYO6 gene (Fig 3H). What’s more, we made a Sankey diagram to visualize the links among preterm labor or labor related gene clusters, hub gene clusters, age subgroups and adipose tissue subtypes (Fig 3I). ## Discussion The development of machine learning algorithms and the availability of gene expression data in the public databases provide approaches to infer biomarkers for disease diagnosis or prognosis in a wide range of fields [28–32]. The bioinformatic attempts for PCOS vary from susceptibility and pathogenesis, to precise diagnosis and tailed therapy [33–37]. According to the most widely used Rotterdam PCOS diagnostic criteria for adult, any 2 out of 3 following features should be met: androgen excess, ovulatory dysfunction, and polycystic ovaries, suggesting that the clinical manifestations and pregnancy complications of PCOS are highly heterogeneous [38–42]. Thus, it is important to find a way to differentiate the heterogeneity of preterm labor risk, and then guide the clinical intervention. In the clinical practice, a feasible genetic test can be expected to perform for the PCOS patients with higher risks, however the economic cost may be considered. Since there is conflicting evidence as to whether or not PCOS women predispose to preterm birth [43], it is reasonable for obstetricians to give the primary prevention strategy firstly, and then the secondary prevention strategies if necessary, to stratify subgroup of PCOS patients with genetic predisposition. The current research indicates that two molecular subtypes were identified in PCOS, by clustering based on the expression of candidate genes related preterm labor and labor. These two subtypes exhibited distinct biological processes and pathways. In addition, two hub genes were spotlighted to imply the key network nodes in the molecular subtypes of PCOS concerning to preterm labor. To our knowledge, this is the first study concerning the transcriptome-wide molecular subtyping of PCOS with preterm labor or labor associated genes. There exist some possible explanations for the PCOS-related preterm labor. On one hand, hyperandrogenism, one of main PCOS features, usually gets enhanced throughout the pregnancy period, which might increase the risk of pregnancy complication, such as preterm labor [11]. Androgens could induce indicated preterm labor in PCOS patients due to severe pregnancy complications, e.g. pre-eclampsia, possibly through changes of endovascular trophoblast invasion and placentation [44]. Androgens might increase the incidence of spontaneous preterm labor in women with PCOS by acting on cervical remodelling and myometrial function [45]. On the other hand, there may exist other molecular mechanisms underlying the preterm labor risk in non-hyperandrogenic PCOS patients. For instance, an abnormal pattern of low-grade chronic inflammation in combination with a subclinical impairment of vascular structure and function were found in both non-pregnant and pregnant women with PCOS, probably contributing to the subsequent reduced depth of endovascular trophoblast and abnormal placentation [46]. As one of the many pro-inflammatory cytokines involved in the induction of spontaneous preterm labor, IL-6 stands out for its pleiotropic effects in both acute and chronic inflammation [47]. Our previous study indeed observed elevated IL-6 levels in peripheral blood of non-hyperandrogenic pregnant women with PCOS [48], suggesting a potential link of preterm labor to non-hyperandrogenic PCOS. The results suggested that genetics would be used to stratify a proportion of women with PCOS into the subgroups with clinical significance. The primary prevention method for the PCOS patients with genetic predisposition is to control the risk factors (e.g. obesity) through lifestyle modification during pre-pregnancy and early pregnancy. And the secondary prevention strategy for this PCOS subgroup is to apply appropriate cervical length surveillance, and the precise vaginal administration of progesterone [43, 49]. Two hub genes of molecular subtyping in PCOS, GTF2F2 and MYO6, were highlighted in the current study. As GTF2F2 gene encodes general transcription factor IIF (TFIIF) subunit 2, the interaction between TFIIF and RPB5-mediating protein is critical to suppress the activated transcription [50], which might be involved in the biological events of energy metabolism, metabolic disorders and fertility [51–53]. MYO6 gene encodes a reverse-direction motor protein that moves toward the minus end of actin filaments and plays a role in intracellular vesicle and organelle transport [54, 55], and execute its functions at multiple steps in autophagy, microtubule polymerization, cell proliferation and metastasis, and hearing loss, spermatogenesis [54, 56–59]. There are several limitations in the present study. First, our sample size was relatively small, and the ethnicities were restricted to Europe and North America. Second, we were not able to access more clinical features (e.g. gestational outcomes) and demographic data of the available samples selected in the present study for more detailed investigation. In addition, our analysis strategy is promising to facilitate the risk assessment for the preterm labor in PCOS, however, the further confirmation in diverse tissue types through biomedical experiments should be conducted sequentially due to the limitations of bioinformatics. ## References 1. Wang F, Pan J, Liu Y, Meng Q, Lv P, Qu F. Alternative splicing of the androgen receptor in polycystic ovary syndrome. Proceedings of the National Academy of Sciences of the United States of America (2015) 112 4743-8. DOI: 10.1073/pnas.1418216112 2. Norman RJ, Dewailly D, Legro RS, Hickey TE. Polycystic ovary syndrome. Lancet (2007) 370 685-97. DOI: 10.1016/S0140-6736(07)61345-2 3. Laganà AS, Vitale SG, Noventa M, Vitagliano A. Current Management of Polycystic Ovary Syndrome: From Bench to Bedside.. Int J Endocrinol (2018) 2018 7234543. DOI: 10.1155/2018/7234543 4. Laganà AS, Forte G, Bizzarri M, Kamenov ZA, Bianco B, Kaya C. Inositols in the ovaries: activities and potential therapeutic applications.. Expert Opin Drug Metab Toxicol (2022) 18 123-33. DOI: 10.1080/17425255.2022.2071259 5. D’Alterio MN, Sigilli M, Succu AG, Ghisu V, Laganà AS, Sorrentino F. Pregnancy outcomes in women with polycystic ovarian syndrome.. Minerva Obstet Gynecol (2022) 74 45-59. DOI: 10.23736/s2724-606x.21.04758–4 6. Jones MR, Goodarzi MO. Genetic determinants of polycystic ovary syndrome: progress and future directions. Fertility and sterility (2016) 106 25-32. DOI: 10.1016/j.fertnstert.2016.04.040 7. Qu F, Wang FF, Yin R, Ding GL, El-Prince M, Gao Q. A molecular mechanism underlying ovarian dysfunction of polycystic ovary syndrome: hyperandrogenism induces epigenetic alterations in the granulosa cells. Journal of molecular medicine (2012) 90 911-23. DOI: 10.1007/s00109-012-0881-4 8. Boomsma CM, Eijkemans MJ, Hughes EG, Visser GH, Fauser BC, Macklon NS. A meta-analysis of pregnancy outcomes in women with polycystic ovary syndrome. Hum Reprod Update (2006) 12 673-83. DOI: 10.1093/humupd/dml036 9. Valent AM, Barbour LA. Management of Women with Polycystic Ovary Syndrome During Pregnancy. Endocrinology and metabolism clinics of North America (2021) 50 57-69. DOI: 10.1016/j.ecl.2020.10.005 10. Greer C, Troughton RW, Adamson PD, Harris SL. Preterm birth and cardiac function in adulthood. Heart (2021). DOI: 10.1136/heartjnl-2020-318241 11. Palomba S, de Wilde MA, Falbo A, Koster MP, La Sala GB, Fauser BC. Pregnancy complications in women with polycystic ovary syndrome. Human reproduction update (2015) 21 575-92. DOI: 10.1093/humupd/dmv029 12. Lovvik TS, Carlsen SM, Salvesen O, Steffensen B, Bixo M, Gomez-Real F. Use of metformin to treat pregnant women with polycystic ovary syndrome (PregMet2): a randomised, double-blind, placebo-controlled trial.. The lancet Diabetes & endocrinology (2019) 7 256-66. DOI: 10.1016/S2213-8587(19)30002-6 13. Bell EF. Red cell transfusion thresholds for preterm infants: finally some answers. Archives of disease in childhood Fetal and neonatal edition (2021). DOI: 10.1136/archdischild-2020-320495 14. Dean B, Ginnell L, Boardman JP, Fletcher-Watson S. Social cognition following preterm birth: A systematic review. Neuroscience and biobehavioral reviews (2021) 124 151-67. DOI: 10.1016/j.neubiorev.2021.01.006 15. Leung C.. Born too soon. Neuro endocrinology letters (2004) 25 133-6. PMID: 15735596 16. Oros D, Strunk M, Breton P, Paules C, Benito R, Moreno E. Altered gene expression in human placenta after suspected preterm labour. Placenta (2017) 55 21-8. DOI: 10.1016/j.placenta.2017.04.025 17. Mani S, Ghosh J, Lan Y, Senapati S, Ord T, Sapienza C. Epigenetic changes in preterm birth placenta suggest a role for ADAMTS genes in spontaneous preterm birth. Human molecular genetics (2019) 28 84-95. DOI: 10.1093/hmg/ddy325 18. Kim SH, Shim SH, Sung SR, Lee KA, Shim JY, Cha DH. Gene expression analysis of the microdissected trophoblast layer of human placenta after the spontaneous onset of labor. PloS one (2013) 8 e77648. DOI: 10.1371/journal.pone.0077648 19. Sitras V, Paulssen RH, Gronaas H, Vartun A, Acharya G. Gene expression profile in labouring and non-labouring human placenta near term. Molecular human reproduction (2008) 14 61-5. DOI: 10.1093/molehr/gam083 20. Cindrova-Davies T, Yung HW, Johns J, Spasic-Boskovic O, Korolchuk S, Jauniaux E. Oxidative stress, gene expression, and protein changes induced in the human placenta during labor.. The American journal of pathology. (2007) 171 1168-79. DOI: 10.2353/ajpath.2007.070528 21. Phillips RJ, Fortier MA, Lopez Bernal A. Prostaglandin pathway gene expression in human placenta, amnion and choriodecidua is differentially affected by preterm and term labour and by uterine inflammation.. BMC pregnancy and childbirth. (2014) 14 241. DOI: 10.1186/1471-2393-14-241 22. Szilagyi A, Gelencser Z, Romero R, Xu Y, Kiraly P, Demeter A. Placenta-Specific Genes, Their Regulation During Villous Trophoblast Differentiation and Dysregulation in Preterm Preeclampsia. International journal of molecular sciences (2020) 21. DOI: 10.3390/ijms21020628 23. Peng HH, Kao CC, Chang SD, Chao AS, Chang YL, Wang CN. The effects of labor on differential gene expression in parturient women, placentas, and fetuses at term pregnancy.. The Kaohsiung journal of medical sciences. (2011) 27 494-502. DOI: 10.1016/j.kjms.2011.06.012 24. Brockway HM, Kallapur SG, Buhimschi IA, Buhimschi CS, Ackerman WE, Muglia LJ. Unique transcriptomic landscapes identified in idiopathic spontaneous and infection related preterm births compared to normal term births.. PloS one. (2019) 14 e0225062. DOI: 10.1371/journal.pone.0225062 25. Mezouar SM J, Ahmed RG. L Gene Expression Profiling of Placenta from Normal to Pathological Pregnancies. (2018) 26. Xie NN, Wang FF, Zhou J, Liu C, Qu F. Establishment and Analysis of a Combined Diagnostic Model of Polycystic Ovary Syndrome with Random Forest and Artificial Neural Network.. BioMed research international (2020) 2020 2613091. DOI: 10.1155/2020/2613091 27. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic acids research (2015) 43 e47. DOI: 10.1093/nar/gkv007 28. Corton M, Botella-Carretero JI, Benguria A, Villuendas G, Zaballos A, San Millan JL. Differential gene expression profile in omental adipose tissue in women with polycystic ovary syndrome. The Journal of clinical endocrinology and metabolism (2007) 92 328-37. DOI: 10.1210/jc.2006-1665 29. Wang D, Li JR, Zhang YH, Chen L, Huang T, Cai YD. Identification of Differentially Expressed Genes between Original Breast Cancer and Xenograft Using Machine Learning Algorithms.. Genes (2018) 9. DOI: 10.3390/genes9030155 30. Wang C, Pu W, Zhao D, Zhou Y, Lu T, Chen S. Identification of Hyper-Methylated Tumor Suppressor Genes-Based Diagnostic Panel for Esophageal Squamous Cell Carcinoma (ESCC) in a Chinese Han Population.. Frontiers in genetics. (2018) 9 356. DOI: 10.3389/fgene.2018.00356 31. Zhang Y, Tseng JT, Lien IC, Li F, Wu W, Li H. mRNAsi Index: Machine Learning in Mining Lung Adenocarcinoma Stem Cell Biomarkers.. Genes (2020) 11. DOI: 10.3390/genes11030257 32. Tabl AA, Alkhateeb A, ElMaraghy W, Rueda L, Ngom A. A Machine Learning Approach for Identifying Gene Biomarkers Guiding the Treatment of Breast Cancer.. Frontiers in genetics (2019) 10 256. DOI: 10.3389/fgene.2019.00256 33. Dai X, Fu G, Reese R. Detecting PCOS susceptibility loci from genome-wide association studies via iterative trend correlation based feature screening. BMC bioinformatics (2020) 21 177. DOI: 10.1186/s12859-020-3492-z 34. Lu J, Xue Y, Wang Y, Ding Y, Zou Q, Pan M. CiRS-126 inhibits proliferation of ovarian granulosa cells through targeting the miR-21-PDCD4-ROS axis in a polycystic ovarian syndrome model. Cell and tissue research (2020) 381 189-201. DOI: 10.1007/s00441-020-03187-9 35. Qin L, Zhao S, Yang P, Cao Y, Zhang J, Chen ZJ. Variation analysis of anti-Mullerian hormone gene in Chinese women with polycystic ovary syndrome. Endocrine (2021) 72 287-93. DOI: 10.1007/s12020-020-02538-4 36. Zhou J, Huang X, Xue B, Wei Y, Hua F. Bioinformatics analysis of the molecular mechanism of obesity in polycystic ovary syndrome. Aging (2021) 13 12631-40. DOI: 10.18632/aging.202938 37. Devarbhavi P, Telang L, Vastrad B, Tengli A, Vastrad C, Kotturshetti I. Identification of key pathways and genes in polycystic ovary syndrome via integrated bioinformatics analysis and prediction of small therapeutic molecules.. Reproductive biology and endocrinology: RB&E (2021) 19 31. DOI: 10.1186/s12958-021-00706-3 38. Teede HJ, Misso ML, Costello MF, Dokras A, Laven J, Moran L. Recommendations from the international evidence-based guideline for the assessment and management of polycystic ovary syndrome. Human reproduction (2018) 33 1602-18. DOI: 10.1093/humrep/dey256 39. Teede HJ, Misso ML, Costello MF, Dokras A, Laven J, Moran L. Recommendations from the international evidence-based guideline for the assessment and management of polycystic ovary syndrome. Fertility and sterility (2018) 110 364-79. DOI: 10.1016/j.fertnstert.2018.05.004 40. Teede HJ, Misso ML, Costello MF, Dokras A, Laven J, Moran L. Recommendations from the international evidence-based guideline for the assessment and management of polycystic ovary syndrome. Clinical endocrinology (2018) 89 251-68. DOI: 10.1111/cen.13795 41. Revised 2003 consensus on diagnostic criteria and long-term health risks related to polycystic ovary syndrome. Fertility and sterility (2004) 81 19-25. DOI: 10.1016/j.fertnstert.2003.10.004 42. Revised 2003 consensus on diagnostic criteria and long-term health risks related to polycystic ovary syndrome (PCOS). Human reproduction (2004) 19 41-7. DOI: 10.1093/humrep/deh098 43. Singh N, Bonney E, McElrath T, Lamont RF. Prevention of preterm birth: Proactive and reactive clinical practice-are we on the right track. Placenta (2020) 98 6-12. DOI: 10.1016/j.placenta.2020.07.021 44. Naver KV, Grinsted J, Larsen SO, Hedley PL, Jorgensen FS, Christiansen M. Increased risk of preterm delivery and pre-eclampsia in women with polycystic ovary syndrome and hyperandrogenaemia. BJOG: an international journal of obstetrics and gynaecology (2014) 121 575-81. DOI: 10.1111/1471-0528.12558 45. Makieva S, Saunders PT, Norman JE. Androgens in pregnancy: roles in parturition. Human reproduction update (2014) 20 542-59. DOI: 10.1093/humupd/dmu008 46. Murray AJ. Oxygen delivery and fetal-placental growth: beyond a question of supply and demand?. Placenta (2012) 33 e16-22. DOI: 10.1016/j.placenta.2012.06.006 47. Prairie E, Cote F, Tsakpinoglou M, Mina M, Quiniou C, Leimert K. The determinant role of IL-6 in the establishment of inflammation leading to spontaneous preterm birth. Cytokine & growth factor reviews (2021) 59 118-30. DOI: 10.1016/j.cytogfr.2020.12.004 48. Wang F, Xie N, Zhou J, Dai M, Zhang Q, Hardiman PJ. Molecular mechanisms underlying altered neurobehavioural development of female offspring of mothers with polycystic ovary syndrome: FOS-mediated regulation of neurotrophins in placenta. EBioMedicine (2020) 60 102993. DOI: 10.1016/j.ebiom.2020.102993 49. Prediction and Prevention of Spontaneous Preterm Birth: ACOG Practice Bulletin, Number 234.. Obstetrics and gynecology. (2021) 138 e65-e90. DOI: 10.1097/aog.0000000000004479 50. Wei W, Gu JX, Zhu CQ, Sun FY, Dorjsuren D, Lin Y. Interaction with general transcription factor IIF (TFIIF) is required for the suppression of activated transcription by RPB5-mediating protein (RMP).. Cell research (2003) 13 111-20. DOI: 10.1038/sj.cr.7290155 51. Moreira GCM, Salvian M, Boschiero C, Cesar ASM, Reecy JM, Godoy TF. Genome-wide association scan for QTL and their positional candidate genes associated with internal organ traits in chickens. BMC genomics (2019) 20 669. DOI: 10.1186/s12864-019-6040-3 52. Lee T, Lee H. Shared Blood Transcriptomic Signatures between Alzheimer’s Disease and Diabetes Mellitus.. Biomedicines (2021) 9. DOI: 10.3390/biomedicines9010034 53. M KM, Kumaresan A, Yadav S, Mohanty TK, Datta TK. Comparative proteomic analysis of high- and low-fertile buffalo bull spermatozoa for identification of fertility-associated proteins.. Reproduction in domestic animals = Zuchthygiene. (2019) 54 786-94. DOI: 10.1111/rda.13426 54. Zakrzewski P, Lenartowska M, Buss F. Diverse functions of myosin VI in spermiogenesis. Histochemistry and cell biology (2021) 155 323-40. DOI: 10.1007/s00418-020-01954-x 55. Zakrzewski P, Redowicz MJ, Buss F, Lenartowska M. Loss of myosin VI expression affects acrosome/acroplaxome complex morphology during mouse spermiogenesisdagger. Biology of reproduction (2020) 103 521-33. DOI: 10.1093/biolre/ioaa071 56. Qiu Y, Wang J, Li H, Yang B, Wang J, He Q. Emerging views of OPTN (optineurin) function in the autophagic process associated with disease.. Autophagy (2021) 1-13. DOI: 10.1080/15548627.2021.1908722 57. Yoong LF, Lim HK, Tran H, Lackner S, Zheng Z, Hong P. Atypical Myosin Tunes Dendrite Arbor Subdivision. Neuron (2020) 106 452-67 e8. DOI: 10.1016/j.neuron.2020.02.002 58. Duan Q, Xu M, Wu M, Zhang X, Gan M, Jiang H. Long noncoding RNA UCA1 promotes cell growth, migration, and invasion by targeting miR-143-3p in oral squamous cell carcinoma.. Cancer medicine. (2020) 9 3115-29. DOI: 10.1002/cam4.2808 59. Boucher S, Tai FWJ, Delmaghani S, Lelli A, Singh-Estivalet A, Dupont T. Ultrarare heterozygous pathogenic variants of genes causing dominant forms of early-onset deafness underlie severe presbycusis. Proceedings of the National Academy of Sciences of the United States of America (2020) 117 31278-89. DOI: 10.1073/pnas.2010782117
--- title: The Association Between Lipedema and Attention-Deficit/Hyperactivity Disorder journal: Cureus year: 2023 pmcid: PMC9977104 doi: 10.7759/cureus.35570 license: CC BY 3.0 --- # The Association Between Lipedema and Attention-Deficit/Hyperactivity Disorder ## Abstract Introduction The current study aimed to investigate the overlap between symptoms of lipedema and attention-deficit/hyperactivity disorder (ADHD). Lipedema is a condition that causes abnormal fat accumulation and inflammation in the legs and buttocks, often accompanied by edema and pain. ADHD is a common condition characterized by difficulty paying attention and controlling behavior, affecting the social, academic, and occupational quality of life. The study’s primary objective was to assess the prevalence of ADHD symptoms in a population of women with lipedema symptoms and compare the clinical characteristics. Method The study used a lipedema screening questionnaire and the Adult Self-Report Scale (ASRS-18) to assess the prevalence of ADHD in a sample of 354 female volunteers with or without a prior lipedema diagnosis. Results Of the lipedema group, 100 ($77\%$) were ASRS positive, and 30 ($23\%$) were ASRS negative. In the group without lipedema, 121 ($54\%$) were ASRS positive, and 103 ($46\%$) were ASRS negative, with a relative risk of 1.424 ($p \leq 0.0001$). Conclusion Our results demonstrate a positive correlation between lipedema and ADHD and suggest that targeted strategies to improve clinic attendance for individuals with ADHD may improve lipedema treatment outcomes. Patients with lipedema symptoms are more likely to have ADHD symptoms. ## Introduction The current study examined the overlap between lipedema symptoms and attention-deficit/hyperactivity disorder (ADHD). Lipedema is characterized by the abnormal accumulation of fat in the lower limbs, which may be accompanied by complaints of pain and edema when standing up [1,2]. The cause of lipedema is not well understood, but it is known to be linked to inflammation. Lipedema is often mistaken for other conditions, such as obesity, gynoid lipodystrophy, and lymphedema, and is frequently not diagnosed during the first medical visit. Lipedema is more common in women, and imaging tests such as ultrasound [3], magnetic resonance imaging, and computed tomography can confirm the diagnosis. Recently, a self-administered questionnaire was developed with excellent screening accuracy for lipedema [1]. ADHD is a common condition affecting the quality of life for children and adults in social, academic, and occupational contexts. The underlying causes of ADHD are not fully understood. Still, proposed mechanisms include glial activation, neuronal damage and degeneration, increased oxidative stress, reduced neurotrophic support, altered neurotransmitter metabolism, and blood-brain barrier disruption. There is growing evidence that inflammation may also play a role in ADHD. However, this evidence is mainly from observational studies showing a strong co-occurrence of ADHD with inflammatory and autoimmune disorders, studies evaluating serum inflammatory markers, and genetic studies. Inflammation may be an essential factor in the development of ADHD, but further research is needed to confirm this [4]. ADHD is better understood as a multidimensional disorder that often persists into adulthood [5]. This study aims to evaluate the correlation between lipedema and attention-deficit/hyperactivity disorder (ADHD) in the Brazilian female population. The primary objective is to assess the prevalence of ADHD in women with lipedema characteristics. Understanding the prevalence rates of ADHD in this population has direct clinical implications since this subgroup has inflammatory triggers. Additionally, the study will describe the clinical characteristics of those presenting ADHD compared to those not meeting the criteria. ## Materials and methods We used the 10-question lipedema screening questionnaire previously published [6] and the standardized Adult Self-Report Scale (ASRS-18) questionnaire [7]. Additional data were collected regarding age, weight, and height, and body mass index (BMI) was calculated using self-reported weight and height. They were converted into an online digital version using secure software appropriate for developing and analyzing questionnaires (SurveyMonkey, CA, USA). The questionnaire was disclosed on a non-profit Brazilian lipedema website (www.lipedema.org.br), and it was administered to 354 volunteers with or without a prior lipedema diagnosis. The participants were women over 18 years old who registered on the online survey platform and volunteered. The sampling method used is convenience sampling. Participants who did not digitally sign the consent form were excluded. Questionnaires completed in less than 300 seconds were excluded. A mean time was previously calculated as three minutes [6] for the lipedema questionnaire and less than two minutes [8] for the ASRS. Screening probability of lipedema The measure was based on our group’s previously proposed lipedema screening questionnaire [6] and cutoff validation [1]. The lipedema screening questionnaire total score cutoff method was used, with an area under the receiver operating curve (ROC curve) of 0.8615, which can be considered an excellent level of accuracy [1]. We took a conservative approach, aiming to achieve specificity closer to 0.9, setting the cutoff at 12. At that point, the probability of a lipedema diagnosis is $77.8\%$ ($95\%$ confidence interval (CI): $64.2\%$-$87.3\%$), with a sensitivity of 0.46 and specificity of 0.88, as previously proposed [1]. The methodology is based on the sum of points from a self-administered questionnaire in a population survey, indicating the probability of a lipedema diagnosis. Screening probability of ADHD All eligible subjects completed the Adult Self-Report Scale (ASRS-18) in Portuguese [7], an 18-item questionnaire on current ADHD symptoms. Volunteers completed the ASRS online just after completing the lipedema screening. Since there is no psychometric data on the ASRS in Brazil, we used the 18-item version instead of the six-item screening version. Shaded boxes in the ASRS were ignored, and the checklist was interpreted in a binary fashion: symptoms reported as often or very often were considered positive. All other frequencies were considered “negative,” as previously suggested by others [5]. Volunteers reporting at least five positive symptoms in the inattention or hyperactivity-impulsivity domain were considered “ASRS positive.” This strategy rendered sensitivity and specificity rates of 0.97 and 0.40, respectively [5]. Statistical analysis A sample size of 152 questionnaires was calculated to achieve a $95\%$CI, considering a $5.5\%$ margin of error. Statistical analysis was conducted after the consistency of the data had been checked manually and automatically with software developed explicitly for this analysis using Xojo 4.1 (Xojo, Austin, TX, USA). Descriptive statistics and frequencies were calculated. Correlations between questionnaire variables were assessed using the chi-square test (z-score), Mann-Whitney, and Student’s t-test. We adopted a $p \leq 0.05$ level of statistical significance for the correlations. The software used for data analysis is MedCalc® Statistical Software version 20.211 (MedCalc Software Ltd., Ostend, Belgium) (https://www.medcalc.org [2023]) and Wizard 2.0.12 (Evan Miller, MA, USA). This study complies with the National Health Council standards set out in resolution $\frac{196}{96}$ regulating research involving human beings. It also adheres to the Helsinki Declaration. The institution’s ethical committee approved the study, and all subjects signed an informed consent form before participation. ## Results The total sample included 354 eligible volunteers who met the inclusion criteria. The mean (±standard deviation (SD)) age of the whole study population was 36.8 (±9.1 SD) years, while the mean (±SD) age of volunteers with the lipedema diagnostic criterion was 38.2 (±9.6 SD) years, which is equivalent. The mean (±SD) height of the whole study population was 163.4 (±6.3 SD) cm, while the mean height of volunteers with the diagnostic criterion was 163.3 (±6.6 SD) cm, which is equivalent. Within the lipedema group, 100 ($77\%$) were ASRS positive, and 30 ($23\%$) were ASRS negative. Of the subgroup of volunteers without diagnostic criteria for lipedema, 121 ($54\%$) were ASRS positive, and 103 ($46\%$) were ASRS negative (Table 1). Table 1 \| Unnamed: 0 \| Volunteers with diagnostic criteria for lipedema \| Volunteers without diagnostic criteria for lipedema \| Total \| Statistics (chi-square) \| \| --- \| --- \| --- \| --- \| --- \| \| Number \| 130 (36.7%±5) \| 224 (63.3%±5) \| 354 \| Unequal proportions p<0.001 \| \| Age (years) \| 38.2 (±9.6 SD) \| 36.1 (±8.8 SD) \| 36.8 (±9.1 SD) \| Independent p=0.251 \| \| BMI (kg/m2) \| 30.9 (±4.9 SD) \| 28.5 (±5.4 SD) \| 29.4 (±5.3 SD) \| Independent p<0.001 \| \| Weight (kg) \| 82.7 (±14.7 SD) \| 76.3 (±15.2 SD) \| 78.6 (±15.3 SD) \| Independent p<0.001 \| \| Height (cm) \| 163.3 (±6.6 SD) \| 163.4 (±6.1 SD) \| 163.4 (±6.3 SD) \| Independent p=0.317 \| \| ASRS + \| 100 (76.9%±7.15) \| 121 (54%±6.45) \| 221 (62.4%±5) \| z-score p<0.001 \| \| ASRS - \| 30 (23.1%±7.15) \| 103 (46%±6.45) \| 133 (37.6%±5) \| z-score p<0.001 \| There was a $35\%$ dropout rate (less than 300 seconds and fill-out mistakes) during the completion of the questionnaire ($$n = 190$$, total 547), and the mean time taken to respond was 482 seconds. It was observed that $36.7\%$±5 ($$n = 130$$, $95\%$CI, unequal proportions, z-score $p \leq 0.001$) of the study population met the lipedema diagnosis criteria, while $62.4\%$±5 ($$n = 221$$, $95\%$CI, unequal proportions, z-score $p \leq 0.001$) were ASRS positive, and $37.6\%$ ($$n = 133$$) were ASRS negative, with a relative risk of 1.424 ($95\%$CI: 1.2218-1.6598, $p \leq 0.0001$) (Table 2). Table 2 \| Unnamed: 0 \| ASRS positive \| ASRS negative \| Statistics (Mann-Whitney) \| \| --- \| --- \| --- \| --- \| \| Age (years) \| 36.2 (±8.8 SD) \| 37.9 (±9.6 SD) \| Equal medians p=0.1794 \| \| BMI (kg/m2) \| 29.3 (±4.99 SD) \| 29.5 (±5.9 SD) \| Equal medians p=0.7652 \| \| Weight (kg) \| 78.5 (±14.2 SD) \| 78.7 (±16.9 SD) \| Equal medians p=0.670 \| \| Height (cm) \| 163.5 (±6.0 SD) \| 163.2 (±6.7 SD) \| Equal medians p=0.6961 \| When evaluating the sum of points of the lipedema and ASRS questionnaire, a positive correlation was found and plotted in Figure 1 (Pearson correlation, $p \leq 0.001$). Figure 1: Scatterplot visualizing the relationship between lipedema sum points and ASRS sum points.Pearson correlation, p<0.001, positive correlationASRS: Adult Self-Report Scale ## Discussion ADHD is a psychiatric disorder that affects approximately $5\%$ of children and has a high genetic component and various potential causes [9]. About $6.76\%$ of adults have ADHD, although there is some variation in the studies on this topic. ADHD symptoms tend to lessen as adults age [10], but this trend was not statistically shown in our data, despite a slightly younger age in the ASRS positive group. Lipedema is characterized by abnormal fat accumulation in the subcutaneous tissue, leading to inflammation and fibrosis in the affected tissue, affecting $12.3\%$ of women [1,11]. This can cause pain and worsen symptoms. Lipedema seems to have an essential impact on mental health as it has been previously correlated to depressive symptoms and anxiety [1,12-14]. In recent years, there has been increasing research on the potential role of inflammation in developing psychiatric disorders, including ADHD. This is supported by the high co-occurrence of ADHD and inflammatory or autoimmune disorders and biomarkers and genetic associations with ADHD. These findings suggest that various underlying mechanisms, such as altered immune response and shared genetic and environmental factors, may be involved in the relationship between inflammation and ADHD. Higher levels of inflammation during early development may lead to the development of ADHD symptoms. Since ADHD has a strong genetic component, individuals with ADHD may have polymorphisms in genes related to inflammation. While some studies have identified such associations, there is no clear consensus on the specific genes involved. Overall, the evidence suggests that the immune system may play a role in the pathophysiology of ADHD [5]. Given the connection between inflammation and both lipedema and ADHD, as well as the overlap in symptoms between the two conditions, it is worth considering the possibility of concomitance between lipedema and ADHD. Some patients may also possess cognitive dysfunction concerning measures commonly impaired with ADHD, such as executive function and learning. Early diagnosis and intervention may explain why up to $60\%$ of individuals with ADHD experience partial remission of symptoms and improved cognitive function upon entering adulthood [15]. Conversely, individuals without an ADHD diagnosis but demonstrating decreased cognitive functioning may be less likely to exhibit improvement in response to the same therapeutic interventions or experience similar disease regression. Many studies on lipedema suggest that surgical treatment is the immediate solution and that removing modified fat tissue can significantly improve patient quality of life. Still, lipedema can be managed clinically [16]. However, some research indicates that lipedema may be connected to other diseases, which may mean that it is a result rather than a cause of these conditions. This highlights the importance of considering a more comprehensive range of factors when evaluating the effectiveness of lipedema surgery. The similarity between the population groups studied was demonstrated. There was a difference regarding weight and BMI, probably because lipedema can interfere with body weight. Dudek et al. [ 12]. also used a questionnaire to investigate a group of Polish women with suspected lipedema, estimating mean BMI at 30.8 (±7.1 SD) kg/m2, with $76.5\%$ classified as overweight ($26.5\%$) or obese ($50\%$). Another study observed that $67.5\%$ of women with lipedema had BMI greater than 25 kg/m2, with a mean BMI of 27 kg/m2. Elevated BMI makes diagnosis more difficult because of the complexity of differentiation from common obesity [1]. Although obesity has been implicated with ADHD [17], our study showed no significant difference when comparing ASRS groups, with an equal median (Mann-Whitney, $$p \leq 0.7652$$). Our study demonstrated a positive correlation between lipedema symptoms and ADHD symptoms, meaning that as self-reported lipedema symptoms increase, ASRS symptoms also tend to increase (Figure 1). ADHD exhibits a diversity of phenotypes, including inattention, impulsivity, or a combination of both. These specific features may not always be accounted for in surgical literature and may potentially influence surgical outcomes [18]. It is also worth noting that individuals with ADHD are at increased risk of comorbid mental health disorders, including mood and anxiety disorders [19]. Anxiety is reported in $61.3\%$ of lipedema patients and depression in $38.7\%$ [1]. This relationship is significant as recent studies have shown that poor cognitive functioning before surgery is closely linked to poor long-term weight loss outcomes following bariatric surgery and may have a similar impact on lipedema treatment [20,21]. Mocanu et al. [ 18]. found that ADHD patients have lower postoperative follow-up rates than non-ADHD patients following bariatric surgery. Implementing targeted strategies to improve clinical attendance for ADHD patients at risk may improve outcomes and reduce recidivism rates. Additionally, pharmacological treatment of ADHD may improve binge eating and impulsivity through reported improvements in anxiety, time management, and self-awareness, which may positively affect lipedema treatment. Understanding the potential overlap between lipedema and ADHD may aid in developing more effective exercise strategies for managing both conditions [22]. As a standard of care, lipedema surgery should only be considered for patients with controlled mental health conditions who can understand the risks and benefits of surgery and participate in necessary follow-up care. The current self-report measures for assessing the presence of ADHD and lipedema have limitations. Lipedema inflammation is cyclical [2], so it may influence ADHD more at certain times. The worst time to decide on surgical treatment may be during times of peak inflammation, as inflammation could significantly impact cognitive function at these times. The study was not designed to identify the intensity of inflammation at the testing time. The lipedema screening questionnaire and ASRS, available in the public domain, can provide a quick and cost-effective means of predicting the likelihood of a diagnosis but cannot give a definitive diagnosis. A previous study found that the ASRS can distinguish between individuals with previously diagnosed ADHD recruited from disability services [23]. Another study used the lipedema screening questionnaire to estimate the prevalence of the disease [1]. However, the literature on the relationship between ADHD and lipedema is not robust, which limits our understanding of more complex interactions between the two conditions. There may also be a discrepancy between self-reported and objective clinical analysis. Psychiatric disorders are mainly characterized by symptoms and not observable signs, which require interpreting the patient’s report and translating it into diagnostic terms. Self-reflection and self-evaluation can also be problematic and may lead to over- or underreporting of symptoms. Additionally, psychiatric symptoms or clusters of symptoms are often not specific to a particular disorder. Different patient assessment strategies, such as open, structured, or semi-structured interviews, can significantly influence the results, and even small changes in question-wording can affect the responses. While standardized interviews are considered the gold standard for psychiatric diagnosis, efforts to minimize these limitations can involve additional patient-related data, often improving diagnostic accuracy [5]. The sample of volunteers in this study consisted of women who registered on an online survey platform seeking information about lipedema. As a result, this population may not represent the general population, and there may be selection bias. It is possible that individuals with more severe or complex cases of lipedema or ADHD may have been more motivated to participate, which could have influenced the results. Compared to a previous population study that showed that $12.3\%$ of volunteers met the criteria for lipedema diagnosis, our study found a higher rate of $36.7\%$ [1]. Additionally, we did not clinically confirm the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV) criteria for ADHD or clinical lipedema diagnosis, which could also have affected the findings. Mental illnesses often exhibit a range of symptoms and can impact an individual’s functioning in various ways that were not necessarily captured in this study. These limitations suggest further research on the potential correlation between ADHD and lipedema. That way, we can better understand lipedema’s neurophysiological mechanisms. This includes exploring inflammation’s role in ADHD and lipedema and developing methods for better measuring low-grade chronic inflammation [2]. Incorporating mental health considerations into treatment approaches for lipedema may be beneficial [22], and routine screening for ADHD in lipedema patients may improve clinical treatment strategies. Additionally, developing educational materials on ADHD for medical professionals who treat lipedema could increase awareness of the potential overlap between these conditions and improve diagnosis and treatment. ## Conclusions In this study, we found a higher prevalence of self-reported symptoms of ADHD in patients with lipedema symptoms. This suggests a potential overlap between the two conditions and highlights the importance of screening for ADHD in patients with lipedema. Implementing targeted strategies to improve clinic attendance for individuals with ADHD may improve lipedema treatment outcomes and reduce variability in results. Furthermore, incorporating mental health considerations, including routine screening for ADHD, into treatment approaches for lipedema may be beneficial. While the limitations of self-reported symptoms should be considered, our study provides insights into the potential correlation between lipedema and ADHD. ## References 1. Amato AC, Amato FC, Amato JL, Benitti DA. Lipedema prevalence and risk factors in Brazil. J Vasc Bras (2022) 21 0 2. Amato ACM. Is lipedema a unique entity?. EC Clin Med Case Rep (2020) 2 1-7 3. Amato AC, Saucedo DZ, Santos KD, Benitti DA. Ultrasound criteria for lipedema diagnosis. Phlebology (2021) 36 651-658. PMID: 33853452 4. Leffa DT, Torres IL, Rohde LA. A review on the role of inflammation in attention-deficit/hyperactivity disorder. Neuroimmunomodulation (2018) 25 328-333. PMID: 29874674 5. Mattos P, Nazar BP, Tannock R. By the book: ADHD prevalence in medical students varies with analogous methods of addressing DSM items. Braz J Psychiatry (2018) 40 382-387. PMID: 29451590 6. Amato AC, Amato FC, Benitti DA, Amato LG. Development of a questionnaire and screening model for lipedema. J Vasc Bras (2020) 19 0 7. Mattos P, Segenreich D, Saboya E. Cross-cultural adaptation of the Adult Self-Report Scale to Portuguese for assessing attention-deficit/hyperactivity disorder (ADHD) in adults (Article in Portuguese). Arch Clin Psych (São Paulo) (2006) 4 8. Kessler RC, Adler L, Ames M. The World Health Organization Adult ADHD Self-Report Scale (ASRS): a short screening scale for use in the general population. Psychol Med (2005) 35 245-256. PMID: 15841682 9. Choudhry Z, Sengupta SM, Grizenko N, Harvey WJ, Fortier MÈ, Schmitz N, Joober R. Body weight and ADHD: examining the role of self-regulation. PLoS One (2013) 8 0 10. Song P, Zha M, Yang Q, Zhang Y, Li X, Rudan I. The prevalence of adult attention-deficit hyperactivity disorder: a global systematic review and meta-analysis. J Glob Health (2021) 11 4009 11. Herbst KL, Kahn LA, Iker E. Standard of care for lipedema in the United States. Phlebology (2021) 36 779-796. PMID: 34049453 12. Dudek JE, Białaszek W, Gabriel M. Quality of life, its factors, and sociodemographic characteristics of Polish women with lipedema. BMC Womens Health (2021) 21 27. PMID: 33446179 13. Romeijn JR, de Rooij MJ, Janssen L, Martens H. Exploration of patient characteristics and quality of life in patients with lipoedema using a survey. Dermatol Ther (Heidelb) (2018) 8 303-311. PMID: 29748843 14. Torre YS, Wadeea R, Rosas V, Herbst KL. Lipedema: friend and foe. Horm Mol Biol Clin Investig (2018) 33 15. Nielsen F, Georgiadou E, Bartsch M, Langenberg S, Müller A, de Zwaan M. Attention deficit hyperactivity disorder prevalence and correlates pre- and post-bariatric surgery: a comparative cross-sectional study. Obes Facts (2017) 10 1-11 16. Amato AC, Benitti DA. Lipedema can be treated non-surgically: a report of 5 cases. Am J Case Rep (2021) 22 0 17. Cortese S, Vincenzi B. Obesity and ADHD: clinical and neurobiological implications. Curr Top Behav Neurosci (2012) 9 199-218. PMID: 21845534 18. Mocanu V, Tavakoli I, MacDonald A, Dang JT, Switzer N, Birch DW, Karmali S. The impact of ADHD on outcomes following bariatric surgery: a systematic review and meta-analysis. Obes Surg (2019) 29 1403-1409. PMID: 30706311 19. Choi WS, Woo YS, Wang SM, Lim HK, Bahk WM. The prevalence of psychiatric comorbidities in adult ADHD compared with non-ADHD populations: a systematic literature review. PLoS One (2022) 17 0 20. Spitznagel MB, Alosco M, Strain G. Cognitive function predicts 24-month weight loss success after bariatric surgery. Surg Obes Relat Dis (2013) 9 765-770. PMID: 23816443 21. Spitznagel MB, Alosco M, Galioto R. The role of cognitive function in postoperative weight loss outcomes: 36-month follow-up. Obes Surg (2014) 24 1078-1084. PMID: 24570090 22. Amato AC. Exercise method for lipedema: guide for the personal trainer (Book in Portuguese). São Paulo (2022) 23. Gray S, Woltering S, Mawjee K, Tannock R. The Adult ADHD Self-Report Scale (ASRS): utility in college students with attention-deficit/hyperactivity disorder. PeerJ (2014) 2 0
--- title: 'Glymphatic system impairment in nonathlete older male adults who played contact sports in their youth associated with cognitive decline: A diffusion tensor image analysis along the perivascular space study' authors: - Yuichi Morita - Koji Kamagata - Christina Andica - Kaito Takabayashi - Junko Kikuta - Shohei Fujita - Thomas Samoyeau - Wataru Uchida - Yuya Saito - Hiroki Tabata - Hitoshi Naito - Yuki Someya - Hideyoshi Kaga - Yoshifumi Tamura - Mari Miyata - Toshiaki Akashi - Akihiko Wada - Toshiaki Taoka - Shinji Naganawa - Hirotaka Watada - Ryuzo Kawamori - Osamu Abe - Shigeki Aoki journal: Frontiers in Neurology year: 2023 pmcid: PMC9977161 doi: 10.3389/fneur.2023.1100736 license: CC BY 4.0 --- # Glymphatic system impairment in nonathlete older male adults who played contact sports in their youth associated with cognitive decline: A diffusion tensor image analysis along the perivascular space study ## Abstract ### Background and purpose Exposure to contact sports in youth causes brain health problems later in life. For instance, the repetitive head impacts in contact sports might contribute to glymphatic clearance impairment and cognitive decline. This study aimed to assess the effect of contact sports participation in youth on glymphatic function in old age and the relationship between glymphatic function and cognitive status using the analysis along the perivascular space (ALPS) index. ### Materials and methods A total of 52 Japanese older male subjects were included in the study, including 12 who played heavy-contact sports (mean age, 71.2 years), 15 who played semicontact sports (mean age, 73.1 years), and 25 who played noncontact sports (mean age, 71.3 years) in their youth. All brain diffusion-weighted images (DWIs) of the subjects were acquired using a 3T MRI scanner. The ALPS indices were calculated using a validated semiautomated pipeline. The ALPS indices from the left and right hemispheres were compared between groups using a general linear model, including age and years of education. Furthermore, partial Spearman's rank correlation tests were performed to assess the correlation between the ALPS indices and cognitive scores (Mini-Mental State Examination and the Japanese version of the Montreal Cognitive Assessment [MoCA-J]) after adjusting for age years of education and HbA1c. ### Results The left ALPS index was significantly lower in the heavy-contact and semicontact groups than that in the noncontact group. Although no significant differences were observed in the left ALPS index between the heavy-contact and semicontact groups and in the right ALPS index among groups, a trend toward lower was found in the right ALPS index in individuals with semicontact and heavy-contact compared to the noncontact group. Both sides' ALPS indices were significantly positively correlated with the MoCA-J scores. ### Conclusion The findings indicated the potential adverse effect of contact sports experience in youth on the glymphatic system function in old age associated with cognitive decline. ## 1. Introduction Contact sports, such as American football and soccer, involve physical contact between players, and these affect brain health [1]. Contact sports have been linked to neurocognitive changes due to repetitive head impacts (RHIs) (1–3). RHIs refer not only to mild traumatic brain injury and concussion but also to asymptomatic subconcussive trauma [4, 5]. RHIs in contact sports are quite frequent, with up to 3–70 head impacts per game that players were exposed to, depending on the sport [6, 7]. Previous animal studies have suggested that RHIs lead to neuroinflammation, synaptic changes, and glymphatic system dysfunction [8, 9]. Recently, glymphatic dysfunction has been considered one of the main causes of cognitive decline due to the accumulation of brain waste products [10, 11]. However, its involvement in older adults with contact sports participation in their youth is not yet fully understood. The glymphatic system is a brain waste clearance system [11, 12]. The underlying hypothesis is that CSF flows into the brain through the perivascular space around the arteries and enters the brain parenchyma through aquaporin-4 (AQP4) channels in the astrocyte endfeet. Then, CSF influx into the brain parenchyma promotes interstitial fluid (ISF). Finally, brain metabolic waste products are washed out through the perivascular space around the veins (11–13). Previous animal studies have shown reduced clearance of intrathecally injected gadolinium contrast agents and fluorescent tracers in the brain of RHI-injured rodents [8, 12, 14]. However, the tracer-based method used to assess the glymphatic system is invasive and thus not suitable for human studies. Taoka et al. proposed diffusion tensor imaging analysis along the perivascular space (DTI-ALPS) [15]. The analysis ALPS index is a potential indirect noninvasive indicator of the glymphatic system in humans by estimating the diffusivity of the perivascular space along the medullary veins at the level of the lateral ventricular body. Zhang et al. reported a strong correlation between the ALPS index and glymphatic function assessed using the intrathecal injection of gadolinium contrast agents in human brains [16]. Furthermore, a reduced ALPS index has been correlated with cognitive decline in older adults and patients with Alzheimer's or Parkinson's disease [15, 17, 18]. This study aimed to evaluate the effect of contact sports practice in youth on the glymphatic function in old age using the ALPS index and to study the relationship between glymphatic function and cognitive status. ## 2.1. Study participants A total of 52 community-dwelling older adults (66–83 years old) enrolled in the Healthy Brain Project by the Sportology Center of Juntendo University in Tokyo, Japan, from 2017 to 2018 [19], were included in this study. This study was approved by our Institutional Review Board. Written informed consent was obtained from all participants before evaluation. The inclusion criteria included nonathlete subjects with data of sports experience in their teenage and 20s, diffusion-weighted images (DWIs), cognitive scores (Mini-Mental State Examination [MMSE], and the Japanese version of the Montreal Cognitive Assessment [MoCA-J]). The exclusion criteria were subjects with cerebrovascular disease, severe cognitive decline (MMSE <23), and a history of severe traumatic brain injury or psychiatric or neurological disorders. Contact sports are sports in which players collide [20, 21]. To further understand the influence of the intensity of the collision in contact sports on the glymphatic system, study participants were categorized into heavy-contact and semicontact sports groups [20, 21]. Heavy-contact sports are sports with intense physical contact where the player is allowed to continuously and intentionally strike or tackle the opponent [20, 21]. Meanwhile, semicontact sports are sports with occasional physical contact, and intentional hitting or tackling is prohibited [20, 21]. For comparison, we also included age-matched individuals with experience in noncontact sports in their youth (noncontact group). Only male participants were included in this study to minimize the influence of sex differences. Noncontact sports are those with little or no physical contact [20, 21]. ## 2.2. Study participants' characteristics This study included data obtained from 52 older Japanese male adults, of whom 12 (71.2 ± 5.2 years) had experience with heavy-contact sports in their youth; 15 (73.1 ± 5.9 years) with semicontact sports; and age-matched 25 (71.3 ± 4.4 years) with noncontact sports. Demographic and clinical characteristics of the study participants in the noncontact, semicontact, and heavy-contact groups are summarized in Table 1. Table 1 \| Unnamed: 0 \| Noncontact Group \| Semicontact Group \| Heavy-Contact Group \| p-Values \| Semicontact versus Heavy-Contact \| Noncontact versus Semicontact \| Noncontact versus Heavy-Contact \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Number of subjects \| 25 \| 15 \| 12 \| \| \| \| \| \| Age (years) \| 71.3 ± 4.4 \| 73.1 ± 5.9 \| 71.2± 5.2 \| 0.67b \| \| \| \| \| Years of education (years) \| 15.4 ± 1.7 \| 14.9 ± 1.8 \| 14.1 ± 2.5 \| 0.19b \| \| \| \| \| BMI (kg/m2) \| 23.5 ± 2.9 \| 23.9 ± 1.9 \| 22.7 ± 3.7 \| 0.27b \| \| \| \| \| Sport experience during youth (years) \| 5.0 ± 4.1 \| 5.4 ± 2.4 \| 8.0 ± 2.3 \| 0.005b \| 0.06c \| 0.21c \| 0.002c \| \| Handedness (right, left, ambidextrous) \| 23, 0, 2 \| 18, 0, 0 \| 10, 1, 1 \| 0.23a \| \| \| \| \| MMSE \| 27.6 ± 1.5 \| 28.6 ± 1.3 \| 27.9 ± 1.7 \| 0.17b \| \| \| \| \| MoCA-J \| 25.4 ± 2.6 \| 25.0 ± 3.0 \| 24.5 ± 2.5 \| 0.16b \| \| \| \| \| Smoking history (Brinkman index) \| 272.2 ± 326.8 \| 767.7 ± 803.5 \| 592.6 ± 580 \| 0.15b \| \| \| \| \| Alcohol consumption (g/day) \| 13.0 ± 15.6 \| 33.2 ± 36.6 \| 33.3 ± 24.7 \| 0.10b \| \| \| \| \| Systolic blood pressure (mmHg) \| 137.2 ± 14.8 \| 136.5 ± 20.6 \| 137.9 ± 12.7 \| 0.91b \| \| \| \| \| Diastolic blood pressure (mmHg) \| 87.3 ± 8.1 \| 88.3 ± 12.7 \| 88.4 ± 8.6 \| 0.77b \| \| \| \| \| HbA1c (%) \| 6.0 ± 0.8 \| 5.9 ± 0.6 \| 5.5 ± 0.3 \| 0.07b \| \| \| \| \| Intake of carbonhydrates (g/day) \| 262.2 ± 63.6 \| 235.8 ± 76.9 \| 208.7± 63.9 \| 0.054b \| \| \| \| \| Current exercise time (Mets/week) \| 14.2 ± 10.5 \| 9.1 ± 10.8 \| 9.6± 17.0 \| 0.38b \| \| \| \| \| High-density lipoprotein cholesterol (mg/dL) \| 59.5 ± 12.5 \| 58,2 ± 10.6 \| 57.0± 16.9 \| 0.86b \| \| \| \| \| Low-density lipoprotein cholesterol (mg/dL) \| 116.6 ± 25.5 \| 113.6 ± 39.5 \| 107.5 ± 31.9 \| 0.37b \| \| \| \| \| Fazekas grade: \| \| \| \| \| \| \| \| \| Periventricular white matter (grade: 0/1/2) \| (1, 21, 4) \| (0, 10, 5) \| (0, 8, 4) \| 0.54a \| \| \| \| \| Deep and subcortical white matter (grade: 0/1/2/3) \| (0, 22, 4, 0) \| (0, 8, 5, 2) \| (0, 9,3, 0) \| 0.12a \| \| \| \| The heavy-contact and semicontact groups had significantly more years of sports experience than the noncontact group. No significant differences were observed in age, handedness, body mass index, years of education, systolic and diastolic blood pressures, high- and low-density lipoprotein cholesterol levels, HbA1c level, the Brinkman index, daily alcohol consumption, the Fazekas grade, the MMSE score, and the MoCA-J score among the three groups. The number of years of sports experience was 8.0, 5.4, and 5.0 years in the heavy-contact, semicontact, and noncontact groups, respectively. The subjects were not interviewed about their participation as varsity athletes. Meanwhile, those in the noncontact group had no contact sports experience. Data regarding the heavy-, semi-, and noncontact sports played by the study participants in their youth are shown in Table 2. Table 2 \| Unnamed: 0 \| Sports \| Number of participants \| \| --- \| --- \| --- \| \| Heavy-contact sports \| Rugby, judo, karate, boxing, kendo, and wrestling, soccer \| 12 \| \| Semicontact sports \| baseball, basketball, and handball \| 15 \| \| Noncontact sports \| Tennis, table tennis, track and field, skiing, archery, and orienteering \| 25 \| ## 2.3. MRI data acquisition All DWIs were acquired on a MAGNETOM Prisma 3T MRI scanner (Siemens, Erlangen, Germany) with a 64-channel head coil. Whole brain DWIs were acquired using multislice echo-planar imaging along 64 diffusion gradient directions in the anterior-posterior direction at a b-value = 1,000 s/mm2 with one nondiffusion-weighted ($b = 0$) volume using the following parameters: TR/TE = 3,$\frac{300}{70}$ ms, matrix size = 130 × 130, resolution = 1.8 mm × 1.8 mm, slice thickness = 1.8 mm, FOV = 229 mm × 229 mm, and acquisition time = 7 min 29 s. Furthermore, standard and reverse phase-encoded blipped images without diffusion weighting (blip-up and blip-down) were also acquired to correct magnetic susceptibility-induced distortions related to EPI acquisition [22]. ## 2.4. DWI data processing All DWIs were checked visually for severe artifacts in the axial, coronal, and sagittal planes. Noise and artifacts in DWIs were corrected using Marcenko–Pastur principal component analysis denoising [23] and degibbs correction using MRtrix (https://www.mrtrix.org/) [24]. Furthermore, the geometric distortions caused by eddy currents and motion-induced susceptibility [25] were corrected using the EDDY and TOPUP toolboxes, parts of the FMRIB Software Library (FSL; https://fsl.fmrib.ox.ac.uk/fsl/fslwiki) version 6.04 [26]. We obtained diffusivity maps for each study participant in the x-axis (right–left, Dxx), y-axis (anterior–posterior, Dyy), and z-axis (inferior–superior, Dzz) directions. We also generated fractional anisotropy (FA) maps for all subjects and registered them in the FMRIB58_FA standard space (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FMRIB58_FA) using FSL's linear image registration tool (http://www.fmrib.ox.ac.uk/fsl/fslwiki/FLIRT) and nonlinear registration tool (http://fsl.fmrib.ox.ac.uk/fsl~/fslwiki/FNIRT). ## 2.5. ALPS index calculation The ALPS index was calculated using a validated semiautomated pipeline [27]. One subject (a 68-year-old male control subject) with the smallest degree of warping was selected to place the region of interest (ROI). Using the color FA map of this subject, spherical ROIs measuring 5 mm in diameter were placed in the projection and association regions at the level of the lateral ventricular body in the left and right hemispheres (Figure 1) [27]. Then, the obtained ROIs were registered to the same FA template. Finally, the ROI's position on the FA images of each subject was manually checked. Since all ROIs were correctly positioned, no manual correction was performed. Figure 1: Region of interest (ROI) placement for calculating the along the perivascular space index. Spherical ROIs measuring 5 mm in diameter were placed in the areas of the projection (yellow) and association (pink) fibers. The values of x-, y-, and z-axis diffusivities in the ROIs were calculated for each individual. In the plane of the lateral ventricle, medullary vessels run in the right-left (x-axis) direction. Therefore, perivascular spaces are oriented along the x-axis. In this specific anatomical region, main white matter fibers run orthogonally to the x-axis (i.e., to the perivascular space direction), with projection fibers along the z-axis and association fibers along the y-axis. The ALPS index was then calculated as the ratio of the average x-axis diffusivity of the projection region (Dxxproj) and the average x-axis diffusivity of the association region (Dxxassoc) to the average y-axis diffusivity of the projection region (Dyyproj) and the average z-axis diffusivity of the association region (Dzzassoc), as follows: An ALPS index close to 1.0 indicates less diffusivity along the space around the perivascular space, while higher values indicate increased diffusivity. The left and right ALPS indices were calculated. ## 2.6. Statistical analysis Data normality was assessed using the Shapiro–Wilk test. Categorical and continuous data of participants' characteristics were compared using Fisher's exact test and the Kruskal–Wallis test, respectively. Left and right ALPS indices were compared among the heavy-contact, semicontact, and noncontact groups using a general linear model (GLM) analysis, including age and years of education and Hemoglobin A1c (HbA1c). Although no significant difference was found in HbA1c values among groups, the values were higher in the noncontact and semicontact groups. Furthermore, diabetes mellitus has been reported to cause changes in the function of the glymphatic system [28]. Therefore, in this study, we decided to include HbA1c as a confounding factor. The heavy-contact and semicontact groups had significantly longer years of sports experience than the noncontact group. Thus, as additional analyses, group comparisons and partial correlation analyses were performed, including years of sports experience as a confounding factor. Pairwise comparisons of groups were then performed using the Tukey–Kramer post hoc tests. In addition, we used Cohen's d to estimate the effect size among the three groups. Cohen's d of 0.1–0.3, 0.3–0.5, 0.5–0.7, and >0.7 were considered as small, medium, large, and very large effect sizes, respectively [29]. The associations between the left or right ALPS indices and the MMSE or MoCA-J scores after adjusting for age years of education and HbA1c were evaluated in the semicontact and heavy-contact groups combined using partial Spearman's rank correlation tests. In all analyses, a $p \leq 0.05$ was considered statistically significant. Due to the exploratory nature of this study, we did not perform corrections for multiple comparisons. Statistical analysis was performed using R version 4.12 (The R Foundation for Statistical Computing, Vienna, Austria). ## 3.1. Between-group differences After adjusting for age years of education and HbA1c, the heavy-contact ($$p \leq 0.005$$, Cohen's d = −1.03) and semicontact ($$p \leq 0.002$$, Cohen's d = −1.05) groups had significantly lower left ALPS index with very large effect sizes than the noncontact group (Figure 2). Meanwhile, a trend toward the lower right ALPS index was demonstrated in the heavy-contact ($$p \leq 0.23$$, Cohen's d = −0.65) and semicontact ($$p \leq 0.053$$, Cohen's d = −0.75) groups with very large and large effect sizes, respectively, compared to the noncontact group. As expected, effect sizes were larger in the heavy-contact group than in the semicontact group. No significant differences in left or right ALPS indices were observed between the heavy-contact and semicontact groups. Figure 2: Violin plots of left and right ALPS indices for noncontact (red), semicontact (green), and heavy-contact (blue) groups. Boxes indicate the interquartile range (75th [upper horizontal line] and 25th [lower horizontal line]), mean (bold black line), and median (black dot). Upper whiskers indicate the maximum value of the variable at a distance of 1.5 times the quartile range from the 75th percentile value. Lower whiskers indicate the distance to the 25th percentile value. Small dots indicate an outlier. Surrounding the boxes (shaded area) on each side is a rotated kernel density plot. The figure shows p-values after adjusting for age, years of education, HbA1c, and Cohen's d. Significant to P < 0.05. ALPS, analysis along the perivascular space.* In the additional group comparison analysis, consistent results were obtained after adjusting age, years of education, HbA1c, and years of sports experience. The heavy-contact ($$p \leq 0.03$$) and semicontact ($$p \leq 0.003$$) groups had significantly lower left ALPS index than the noncontact group. No significant differences were observed in the left ALPS index between the heavy-contact and semicontact groups and in the right ALPS index among groups. ## 3.2. Associations between cognitive performance and ALPS index Lower left ($$p \leq 0.003$$, $r = 0.59$) or right ($$p \leq 0.01$$, $r = 0.51$) ALPS indices in heavy-contact and semicontact groups combined were significantly correlated with worse MoCA-J scores after adjusting for age, years of education, and HbA1c (Figure 3). However, no significant correlation was observed between MMSE scores and left ($$p \leq 0.74$$, $r = 0.07$) or right ALPS ($$p \leq 0.56$$, $r = 0.13$) indices. Figure 3: Scatter plots show a significant (p < 0.05) association between left (A) or right (B) ALPS along the perivascular space ALPS indices and MoCA-J scores in the semicontact (yellow dots) and heavy-contact groups (blue dots) combined. Significant to P < 0.05. ALPS, analysis along the perivascular space; MoCA-J, Japanese version of the Montreal Cognitive Assessment.* The additional partial correlation analysis showed similar results when adjusted for age, years of education, HbA1c, and years of sports experience. The MoCA-J and ALPS indices showed significant correlations on both sides: left ($$p \leq 0.003$$, $r = 0.59$) and right ($$p \leq 0.009$$, $r = 0.53$), whereas MMSE and ALPS indices showed no significant correlation: left ($$p \leq 0.78$$, $r = 0.06$) and right ($$p \leq 0.39$$, $r = 0.19$). ## 4. Discussion We evaluated the effects of youth contact sports experiences on glymphatic system function in community-dwelling older adult males (66–83 years old) using the DTI-ALPS index. The study findings showed a significantly lower left ALPS index and a trend toward a lower right ALPS index in individuals with semicontact and heavy-contact sports experience in their youth than in those with noncontact sports experience, which are likely to be related to the impairment of glymphatic function (Figure 2). As expected, larger effect sizes were observed in individuals with heavy-contact sports experience than in those with semicontact sports experience, which indicates a more severe decrease in glymphatic function in those with heavy-contact sports experience. Furthermore, the partial correlation analyses showed significant associations between lower left or right ALPS indices and lower MoCA-J scores (Figure 3). The lower ALPS index reflects decreased diffusivity along the perivascular space of the deep medullary vein. Previous studies have reported that RHIs and mild head impacts cause glymphatic dysfunction in animal experiments using gadolinium contrast agents and fluorescent tracers [30, 31]. Ren et al. found decreased AQP4 expression in mice's perivascular space in the cerebral cortex and striatum after mild head impacts [32]. AQP4 plays an important role in facilitating the CSF and ISF exchange [10]. Therefore, reduced AQP4 expression due to RHIs might have also impaired CSF-ISF drainage toward the perivenous space leading to a reduced ALPS index. Furthermore, the risk of RHIs is increased in high-intensity contact sports, such as rugby [33, 34]. Given that larger effect sizes were observed in the heavy-contact group than in the semicontact group, we also speculate that higher intensity of contact sports might cause more severe impairment of the glymphatic system. Although the fluid dynamics of CSF and ISF have not been fully clarified, CSF/ISF dynamics impairment has been in many diseases, including Alzheimer's disease, Parkinson's disease, and stroke (15–17). Taoka et al. proposed the concept of central nervous system (CNS) interstitial fluidopathy, which would group pathologies associated with abnormal neurofluid dynamics [35, 36]. Arterial pulsatilities are an important driving force of CSF/ISF flow [37]. Several studies have also demonstrated the associations between RHIs and vascular pathology, which likely contributes to long-term detrimental effects on cerebrovascular functions [38, 39] and ISF-CSF exchange [40]. Taken together, the results of this study also indicate that CNS interstitial fluidopathy might be related to contact sports due to RHIs. Interestingly, the semicontact and heavy-contact groups showed a significantly reduced left ALPS index compared with the noncontact group. In contrast, there was only a trend toward a lower right ALPS index. One possible reason for the left–right difference observed in this study is that the left cerebral hemisphere may be more vulnerable to head impact than the right hemisphere. These findings are consistent with previous studies that examined subjects experiencing mild traumatic brain injury and reported white matter hypoperfusion, microstructural changes, and cortical thinning predominantly in the left hemisphere (41–43). In addition, as mentioned above, RHIs can cause glymphatic system dysfunction due to pathological changes in the cerebral arteries. The difference in the bifurcation of the right and left carotid arteries suggests that the left carotid is more directly susceptible to strong pulse pressure from the aortic arch and is more likely to experience severe plaque formation and intimal damage [27, 44]. Thus, this may have led to increased glymphatic dysfunction in the left cerebral hemisphere. The study findings showed a significant relationship between semicontact and heavy-contact sports-related glymphatic system dysfunction and MoCA-J scores after adjusting for age, years of education, and HbA1c. This further supports the negative effect of RHIs in youth on cognitive function later in life, possibly due to glymphatic dysfunction, using the ALPS index as an objective imaging indicator of cognitive function. In line with the study results, significant correlations were observed between lower ALPS index and poorer cognitive performance in healthy older adults and patients with Alzheimer's disease or cardiovascular disease [15, 16], and impaired glymphatic function was also observed. In this study, a significant positive correlation was observed between the ALPS index and the MoCA-J score, whereas an insignificant correlation was observed between the ALPS index and the MMSE score, and this might be due to the sensitivity of the MoCA-J score, which is greater than that of the MMSE score, to cognitive decline detection [45]. This study has some limitations. First, the study participants were only men. Meanwhile, female mice and rats showed a similar or more severe glymphatic dysfunction induced by RHI than males [14, 46]. It is possible that the glymphatic system may be similarly impaired in women who have experienced RHI. Therefore, future studies are needed to investigate this possibility. Second, some confounding factors that lead to cognitive and glymphatic impairment were not considered owing to the time lag between the experience of contact sport and image acquisition. Although there have been no studies on RHI-related long-term brain pathological in humans, some animal studies have reported long-term gliosis and pathological changes in cerebral white matter after RHIs [9, 47], supporting its chronic effect on the glymphatic system. Third, previous studies have shown the associations between blood pressure and HbA1c and glymphatic system function [27]. Therefore, we matched blood pressure among groups to minimize the effects of vascular risk factors in this study. In this study, HbA1c was not significant but tended to be higher in the noncontact group than that in the heavy-contact group, and diabetes mellitus is known to decrease glymphatic function [28, 48]. Nevertheless, a significant difference in the left ALPS index was consistently observed after HbA1c was included as a confounding factor. The subjects in the noncontact group had a greater carbohydrate intake than the subjects in the other groups, which is the reason for the higher trend in HbA1c values [49]. Furthermore, even when HbA1c was included as a covariate, the ALPS index tended to be lower in the semicontact and heavy-contact groups, suggesting that HbA1c had limited influence on the study's validity. Fourth, the ROIs for calculating ALPS indices include not only the medullary veins but also the surrounding cerebral white matter. Thus, it is impossible to exclusively evaluate the diffusivity of the perivascular space along the medullary veins. However, a strong correlation between the ALPS index and the functional assessment of the glymphatic system by intrathecal injection of gadolinium contrast agents in vivo was reported, which supports the use of the ALPS index [16]. Fifth, our cohort does not review the history of mild traumatic brain injuries (concussive head impacts). However, according to the trauma history of all the subjects, they have not experienced severe traumatic brain injuries requiring hospitalization. Although the sole effect of mild traumatic brain injury cannot be considered, this study analyzed the effects of RHIs, including mild traumatic brain injuries and subconcussive head impacts, excluding severe traumatic brain injury. Finally, this study lacked histopathological validation, and no direct evidence supports the changes in AQP4 expression. ## 5. Conclusion In summary, exposure to contact sports in youth may cause glymphatic dysfunction in old age. Furthermore, contact-sports-related glymphatic dysfunction is associated with cognitive decline. ## Data availability statement The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author. ## Ethics statement The studies involving human participants were reviewed and approved by Juntendo University Hospital Ethics Committee. The patients/participants provided their written informed consent to participate in this study. ## Author contributions YM, KK, CA, KT, JK, SF, TS, MM, TT, and SN contributed to the concept, design, and analysis of the study and drafted and revised the manuscript. WU, YSa, HT, HN, HK, YSo, YT, HW, and RK performed data and image acquisition, analyzed and interpreted the data, and revised the manuscript for intellectual content. TA, AW, OA, and SA supervised analysis, participated in study design, data interpretation, and drafted and revised the manuscript. All authors contributed to the study and approved the final manuscript. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Koerte IK, Nichols E, Tripodis Y, Schultz V, Lehner S, Igbinoba R. Impaired cognitive performance in youth athletes exposed to repetitive head impacts. J Neurotrauma. (2017) 34 2389-95. DOI: 10.1089/neu.2016.4960 2. Alosco ML, Mez J, Tripodis Y, Kiernan PT, Abdolmohammadi B, Murphy L. Age of first exposure to tackle football and chronic traumatic encephalopathy. Ann Neurol. (2018) 83 886-901. DOI: 10.1002/ana.25245 3. Montenigro PH, Alosco ML, Martin BM, Daneshvar DH, Mez J, Chaisson CE. Cumulative head impact exposure predicts later-life depression, apathy, executive dysfunction, and cognitive impairment in former high school and college football players. J Neurotrauma. (2017) 34 328-40. DOI: 10.1089/neu.2016.4413 4. Asselin PD, Gu Y, Merchant-Borna K, Abar B, Wright DW, Qiu X. Spatial regression analysis of mr diffusion reveals subject-specific white matter changes associated with repetitive head impacts in contact sports. Sci Rep. (2020) 10 13606. DOI: 10.1038/s41598-020-70604-y 5. McKee AC, Alosco ML, Huber BR. Repetitive head impacts and chronic traumatic encephalopathy. Neurosurg Clin N Am. (2016) 27 529-35. DOI: 10.1016/j.nec.2016.05.009 6. Crisco JJ, Fiore R, Beckwith JG, Chu JJ, Brolinson PG, Duma S. Frequency and location of head impact exposures in individual collegiate football players. J Athl Train. (2010) 45 549-59. DOI: 10.4085/1062-6050-45.6.549 7. Reynolds BB, Patrie J, Henry EJ, Goodkin HP, Broshek DK, Wintermark M. Comparative analysis of head impact in contact and collision sports. J Neurotrauma. (2017) 34 38-49. DOI: 10.1089/neu.2015.4308 8. Cai X, Harding IC, Sadaka AH, Colarusso B, Kulkarni P, Ebong E. Mild repetitive head impacts alter perivascular flow in the midbrain dopaminergic system in awake rats. Brain Commun. (2021) 3 fcab265. DOI: 10.1093/braincomms/fcab265 9. Xu X, Cowan M, Beraldo F, Schranz A, Mccunn P, Geremia N. Repetitive mild traumatic brain injury in mice triggers a slowly developing cascade of long-term and persistent behavioral deficits and pathological changes. Acta Neuropathol Commun (2021) 9 1. DOI: 10.1186/s40478-021-01161-2 10. Rasmussen MK, Mestre H, Nedergaard M. The glymphatic pathway in neurological disorders. The Lancet Neurology. (2018) 17 1016-24. DOI: 10.1016/S1474-4422(18)30318-1 11. Jessen NA, Munk ASF, Lundgaard I, Nedergaard M. The glymphatic system: a beginner's guide. Neurochem Res. (2015) 40 2583-99. DOI: 10.1007/s11064-015-1581-6 12. Iliff JJ, Wang M, Liao Y, Plogg BA, Peng W, Gundersen GA. A paravascular pathway facilitates csf flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid?. Sci Translat Med. (2012) 4 147ra11-ra1. DOI: 10.1126/scitranslmed.3003748 13. Naganawa S, Taoka T. The glymphatic system: a review of the challenges in visualizing its structure and function with mr imaging. Magn Reson Med. (2022) 22 182-94. DOI: 10.2463/mrms.rev.2020-0122 14. Christensen J, Wright DK, Yamakawa GR, Shultz SR, Mychasiuk R. Repetitive mild traumatic brain injury alters glymphatic clearance rates in limbic structures of adolescent female rats. Scientific Rep. (2020) 10 1. DOI: 10.1038/s41598-020-63022-7 15. Taoka T, Masutani Y, Kawai H, Nakane T, Matsuoka K, Yasuno F. Evaluation of glymphatic system activity with the diffusion mr technique: diffusion tensor image analysis along the perivascular space (Dti-Alps) in Alzheimer's disease cases. Jpn J Radiol. (2017) 35 172-8. DOI: 10.1007/s11604-017-0617-z 16. Zhang W, Zhou Y, Wang J, Gong X, Chen Z, Zhang X. Glymphatic clearance function in patients with cerebral small vessel disease. Neuroimage. (2021) 238 118257. DOI: 10.1016/j.neuroimage.2021.118257 17. Chen H-L, Chen P-C, Lu C-H, Tsai N-W, Yu C-C, Chou K-H. Associations among cognitive functions, plasma DNA, and diffusion tensor image along the perivascular space (Dti-Alps) in patients with Parkinson's disease. Oxid Med Cell Longev. (2021) 2021 1-10. DOI: 10.1155/2021/4034509 18. Siow TY, Toh CH, Hsu JL, Liu GH, Lee SH, Chen NH. Association of sleep, neuropsychological performance, and gray matter volume with glymphatic function in community-dwelling older adults. Neurology. (2022) 98 e829-e38. DOI: 10.1212/WNL.0000000000201080 19. Someya Y, Tamura Y, Kaga H, Nojiri S, Shimada K, Daida H. Skeletal muscle function and need for long-term care of urban elderly people in Japan (the Bunkyo Health Study): a prospective cohort study. BMJ Open. (2019) 9 e031584. DOI: 10.1136/bmjopen-2019-031584 20. Rice SG. Medical conditions affecting sports participation. Pediatrics. (2008) 121 841-8. DOI: 10.1542/peds.2008-0080 21. Maume DJ, Parrish M. Heavy-contact sport participation and early adolescent delinquency. Social Currents. (2021) 8 126-44. DOI: 10.1177/2329496520959300 22. Andersson JLR, Sotiropoulos SN. An integrated approach to correction for off-resonance effects and subject movement in diffusion mr imaging. Neuroimage. (2016) 125 1063-78. DOI: 10.1016/j.neuroimage.2015.10.019 23. Veraart J, Novikov DS, Christiaens D, Ades-Aron B, Sijbers J, Fieremans E. Denoising of diffusion mri using random matrix theory. Neuroimage. (2016) 142 394-406. DOI: 10.1016/j.neuroimage.2016.08.016 24. Perrone D, Aelterman J, PiŽurica A, Jeurissen B, Philips W, Leemans A. The effect of gibbs ringing artifacts on measures derived from diffusion MRI. Neuroimage. (2015) 120 441-55. DOI: 10.1016/j.neuroimage.2015.06.068 25. Andersson JL, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage. (2003) 20 870-88. DOI: 10.1016/S1053-8119(03)00336-7 26. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. Fsl. Neuroimage. (2012) 62 782-90. DOI: 10.1016/j.neuroimage.2011.09.015 27. Kikuta J, Kamagata K, Takabayashi K, Taoka T, Yokota H, Andica C. An investigation of water diffusivity changes along the perivascular space in elderly subjects with hypertension. Am J Neuroradiol. (2022) 43 48-55. DOI: 10.3174/ajnr.A7334 28. Yang G, Deng N, Liu Y, Gu Y, Yao X. Evaluation of glymphatic system using diffusion mr technique in T2dm cases. Front Hum Neurosci. (2020) 14 300. DOI: 10.3389/fnhum.2020.00300 29. Hopkins WG, Marshall SW, Batterham AM, Hanin J. Progressive statistics for studies in sports medicine and exercise science. Med Sci Sports Exerc. (2009) 41 3-12. DOI: 10.1249/MSS.0b013e31818cb278 30. Iliff JJ, Chen MJ, Plog BA, Zeppenfeld DM, Soltero M, Yang L. Impairment of glymphatic pathway function promotes tau pathology after traumatic brain injury. J Neurosci. (2014) 34 16180-93. DOI: 10.1523/JNEUROSCI.3020-14.2014 31. Li L, Chopp M, Ding G, Davoodi-Bojd E, Zhang L, Li Q. MRI detection of impairment of glymphatic function in rat after mild traumatic brain injury. Brain Res. (2020) 1747 147062. DOI: 10.1016/j.brainres.2020.147062 32. Ren Z, Iliff JJ, Yang L, Yang J, Chen X, Chen MJ. ‘Hit & run' model of closed-skull traumatic brain injury (Tbi) reveals complex patterns of post-traumatic Aqp4 dysregulation. J Cerebral Blood Flow Metabol. (2013) 33 834-45. DOI: 10.1038/jcbfm.2013.30 33. Nguyen JVK, Brennan JH, Mitra B, Willmott C. Frequency and magnitude of game-related head impacts in male contact sports athletes: a systematic review and meta-analysis. Sports Med. (2019) 49 1575-83. DOI: 10.1007/s40279-019-01135-4 34. Black SE, Follmer B, Mezzarane RA, Pearcey GEP, Sun Y, Zehr EP. Exposure to impacts across a competitive rugby season impairs balance and neuromuscular function in female rugby athletes. BMJ Open Sport Exer Med. (2020) 6 e000740. DOI: 10.1136/bmjsem-2020-000740 35. Taoka T, Naganawa S. Imaging for central nervous system (Cns) interstitial fluidopathy: disorders with impaired interstitial fluid dynamics. Jpn J Radiol. (2021) 39 1-14. DOI: 10.1007/s11604-020-01017-0 36. Agarwal N, Contarino C, Toro EF. Neurofluids: a holistic approach to their physiology, interactive dynamics and clinical implications for neurological diseases. Veins Lymphatics. (2019) 8 3. DOI: 10.4081/vl.2019.8470 37. Mestre H, Tithof J, Du T, Song W, Peng W, Sweeney AM. Flow of cerebrospinal fluid is driven by arterial pulsations and is reduced in hypertension. Nat Commun. (2018) 9 1. DOI: 10.1038/s41467-018-07318-3 38. Slobounov SM, Walter A, Breiter HC, Zhu DC, Bai X, Bream T. The effect of repetitive subconcussive collisions on brain integrity in collegiate football players over a single football season: a multi-modal neuroimaging study. Neuroimage Clin. (2017) 14 708-18. DOI: 10.1016/j.nicl.2017.03.006 39. Svaldi DO, Mccuen EC, Joshi C, Robinson ME, Nho Y, Hannemann R. Cerebrovascular reactivity changes in asymptomatic female athletes attributable to high school soccer participation. Brain Imaging Behav. (2017) 11 98-112. DOI: 10.1007/s11682-016-9509-6 40. Kress BT, Iliff JJ, Xia M, Wang M, Wei HS, Zeppenfeld D. Impairment of paravascular clearance pathways in the aging brain. Ann Neurol. (2014) 76 845-61. DOI: 10.1002/ana.24271 41. Cubon VA, Putukian M, Boyer C, Dettwiler A. A diffusion tensor imaging study on the white matter skeleton in individuals with sports-related concussion. J Neurotrauma. (2011) 28 189-201. DOI: 10.1089/neu.2010.1430 42. Hattori N, Swan Foster M, Stobbe G, Uomoto J, Minoshima S, Djang D. Dominant hemisphere's vulnerability may explain chronic fatigue after mild traumatic brain injury (Mtbi). J Nuclear Med. (2009) 50 369. DOI: 10.2967/jnumed.108.060368 43. Govindarajan KA, Narayana PA, Hasan KM, Wilde EA, Levin HS, Hunter JV. Cortical thickness in mild traumatic brain injury. J Neurotrauma. (2016) 33 1809-17. DOI: 10.1089/neu.2015.4253 44. Selwaness M, Van Den Bouwhuijsen Q, Van Onkelen RS, Hofman A, Franco OH, Van Der Lugt A. Atherosclerotic plaque in the left carotid artery is more vulnerable than in the right. Stroke. (2014) 45 3226-30. DOI: 10.1161/STROKEAHA.114.005202 45. Nasreddine ZS, Phillips NA, Bédirian VR, Charbonneau S, Whitehead V, Collin I. The montreal cognitive assessment, moca: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. (2005) 53 695-9. DOI: 10.1111/j.1532-5415.2005.53221.x 46. Ferguson SA, Mouzon BC, Lynch C, Lungmus C, Morin A, Crynen G. Negative impact of female sex on outcomes from repetitive mild traumatic brain injury in htau mice is age dependent: a chronic effects of neurotrauma consortium study. Front Aging Neurosci. (2017) 9 416. DOI: 10.3389/fnagi.2017.00416 47. Mcateer KM, Corrigan F, Thornton E, Turner RJ, Vink R. Short and long term behavioral and pathological changes in a novel rodent model of repetitive mild traumatic brain injury. PLoS ONE. (2016) 11 e0160220. DOI: 10.1371/journal.pone.0160220 48. Zhang L, Chopp M, Jiang Q, Zhang Z. Role of the glymphatic system in ageing and diabetes mellitus impaired cognitive function. Stroke Vascular Neurol. (2019) 4 90-2. DOI: 10.1136/svn-2018-000203 49. Haimoto H, Watanabe S, Komeda M, Wakai K. The impact of carbohydrate intake and its sources on hemoglobin A1c levels in japanese patients with type 2 diabetes not taking anti-diabetic medication. Diabetes Metab Syndr Obes. (2018) 11 53-64. DOI: 10.2147/DMSO.S154839
--- title: 'Community-Acquired Skin and Soft Tissue Infections: Epidemiology and Management in Patients Presenting to the Emergency Department of a Tertiary Care Hospital' journal: Cureus year: 2023 pmcid: PMC9977200 doi: 10.7759/cureus.34379 license: CC BY 3.0 --- # Community-Acquired Skin and Soft Tissue Infections: Epidemiology and Management in Patients Presenting to the Emergency Department of a Tertiary Care Hospital ## Abstract Background: Skin and soft tissue infections are one of the most common diseases presenting to the emergency department (ED). There is no study available on the management of Community-Acquired Skin and Soft Tissue Infections (CA-SSTIs) in our population recently. This study aims to describe the frequency and distribution of CA-SSTIs as well as their medical and surgical management among patients presenting to our ED. Methods: We conducted a descriptive cross-sectional study on patients presenting with CA-SSTIs to the ED of a tertiary care hospital in Peshawar, Pakistan. The primary objective was to estimate the frequency of common CA-SSTIs presenting to the ED and to assess the management of these infections in terms of diagnostic workup and treatment modalities used. The secondary objectives were to study the association of different baseline variables, diagnostic modalities, treatment modalities, and improvement with the surgical procedure performance for these infections. Descriptive statistics were obtained for quantitative variables like age. Frequencies and percentages were derived for categorical variables. The chi-square test was used to compare different CA-SSTIs in terms of categorical variables like diagnostic and treatment modalities. We divided the data into two groups based on the surgical procedure. A chi-square analysis was conducted to compare these two groups in terms of categorical variables. Results: Out of the 241 patients, $51.9\%$ were males and the mean age was 34.2 years. The most common CA-SSTIs were abscesses, infected ulcers, and cellulitis. Antibiotics were prescribed to $84.2\%$ of patients. Amoxicillin + Clavulanate was the most frequently prescribed antibiotic. Out of the total, 128 ($53.11\%$) patients received some type of surgical intervention. Surgical procedures were significantly associated with diabetes mellitus, heart disease, limitation of mobility, or recent antibiotic use. There was a significantly higher rate of prescription of any antibiotic and anti-methicillin-resistant Staphylococcus aureus (anti-MRSA) agents in the surgical procedure group. This group also saw a higher rate of oral antibiotics prescription, hospitalization, wound culture, and complete blood count. Conclusion: This study shows a higher frequency of purulent infections in our ED. Antibiotics were prescribed more frequently for all infections. Surgical procedures like incision and drainage were much lower even in purulent infections. Furthermore, beta-lactam antibiotics like Amoxicillin-Clavulanate were commonly prescribed. Linezolid was the only systemic anti-MRSA agent prescribed. We suggest physicians should prescribe antibiotics appropriate to the local antibiograms and the latest guidelines. ## Introduction Skin and soft tissue infections (SSTIs) are one of the most common diseases presented to the emergency department (ED) [1]. As of 2014, there were an estimated 29.7 SSTI-related emergency room visits per 1000 population in the ED of the United States [2]. Normally, many microorganisms colonize the skin without causing any harm. With any imbalance in the structural or functional protection of the skin, the pathogenic organisms spread in the layers of the skin, overgrow, and elicit acute or chronic inflammation [1]. This process is called infection. Some skin infections may result from a hematogenous spread of pathogens from a distant infection [1]. Most SSTIs are caused by streptococci and Staphylococcus aureus [3]. The United States food and drug authority classified SSTIs as uncomplicated and complicated infections [4,5]. Uncomplicated SSTIs include simple abscesses, cellulitis, impetigo, and furuncles. Complicated SSTIs include severe infections like necrotizing infections, infected burn wounds, infected open ulcers, and deep abscesses requiring major surgical intervention. It also includes infections in diabetic patients and immunocompromised patients [4]. Most simple SSTIs self-resolve. Large, complicated, and/or painful SSTIs require medical attention. Depending upon the severity, infections may need antibiotics and/or a surgical procedure like syringe aspiration, incision and drainage (I/D), debridement and drainage (D/D), or even amputation [6]. In an outpatient department and infectious disease setting, the focus is to provide specific antibiotics tailored to the sensitivity results. However, in the emergency room, physicians are mainly concerned with empiric therapy [7]. Based on the suspected source of infection, the SSTIs can be divided into two groups: Community-acquired: infections in non-hospitalized patients; and healthcare-associated: when an infection is acquired during or soon after hospitalization [8]. The healthcare-associated infections are considered a major complication, so they are frequently studied everywhere. Community-acquired skin and soft tissue infections (CA-SSTIs) are relatively less studied. There is no study available on the management of CA-SSTIs in our population. The main objective of our study was to describe the frequency and distribution of CA-SSTIs, as well as medical and surgical treatments employed for these infections in our ED. This can then be used for further study to improve our practices under the current guidelines. ## Materials and methods We conducted a descriptive cross-sectional study on patients presenting with CA-SSTIs to the ED of a tertiary care hospital in Peshawar, Pakistan. After approval from the hospital ethics and research board, we collected data in the ED from 1 September 2022 to 31 October 2022. Through a consecutive sampling technique, we included all the patients, of any age or gender, that presented with an active SSTI. We excluded the patients who did not consent to data collection, who visited for a follow-up of a past infection, and/or who, by definition, had a healthcare-associated infection. Any infection that presents 48 hours after hospital admission, within three days after discharge, or within 30 days of surgery, was called a healthcare-associated infection [9]. After informed verbal consent, we collected the required data on a pre-designed proforma (Appendices). This included the patient’s biodata, baseline variables, comorbid conditions, diagnostic tests, and the treatment modalities used. A telephonic follow-up was conducted regarding the disease status after one and two weeks of the visit. Data analysis We collected data from 264 patients based on an anticipated frequency of $3.18\%$ SSTIs in the ED and a $95\%$ level of confidence [2]. Twenty-three patients were excluded due to deficient or incorrect information. The final analysis was carried out on 241 patients in Statistical Product and Service Solutions (SPSS) (IBM SPSS Statistics for Windows, Version 25.0, Armonk, NY). Descriptive statistics were obtained for quantitative variables like age. Frequencies and percentages were derived for categorical variables including gender, age groups, clinical diagnosis, diagnostic tests, and treatment modalities used. The chi-square test was used to compare different SSTIs in terms of categorical variables like diagnostic modalities. A p-value of less than 0.05 was considered a statistically significant association. We further divided the data into two groups based on the surgical procedure done: the surgical procedure group (those who underwent any surgical procedure including syringe aspiration, I/D, D/D, or amputation for a CA-SSTI), and the non-surgical procedure group. A chi-square analysis was conducted to compare these two groups in terms of categorical variables. A p-value of less the 0.05 was considered a statistically significant association. ## Results Out of the 241 patients, $51.9\%$ were males and the mean ± SD age was 34.2 ± 18.1 (Table 1). The most common CA-SSTIs were abscesses ($35.3\%$), infected ulcers ($18.7\%$), and cellulitis ($12.9\%$) (Figure 1). Antibiotics were prescribed to $84.2\%$ of patients (Table 2). Amoxicillin + Clavulanate was the most common type of antibiotic prescribed followed by Linezolid, Moxifloxacin, and Cefoperazone + Sulbactam (Figures 2, 3). Out of the total, 128 ($53.11\%$) patients received some type of surgical intervention while 113 ($46.8\%$) did not receive surgical intervention. Surgical procedures were significantly more common in patients with diabetes mellitus, heart disease, limitation of mobility, or recent antibiotic use. However, there was no significant difference between the two groups for other variables (Table 3). Table 3 \| Variables \| Overall (n=241) \| Surgical Procedure Done (n=128) \| No Surgical Procedure (n=113) \| p-valuea \| \| --- \| --- \| --- \| --- \| --- \| \| \| % \| % \| % \| \| \| Gender \| Gender \| Gender \| Gender \| Gender \| \| Males \| 58.9% \| 59.4% \| 58.4% \| 0.879 \| \| Females \| 41.1% \| 40.6% \| 41.6% \| 0.879 \| \| Other baseline variables \| Other baseline variables \| Other baseline variables \| Other baseline variables \| Other baseline variables \| \| Pregnant \| 1.2% \| 0.8% \| 1.8% \| 0.49 \| \| Lactating \| 3.3% \| 1.6% \| 5.3% \| 0.105 \| \| Systemic infection \| 49% \| 37% \| 0.371% \| 0.371 \| \| Sepsis \| 3.3% \| 3.9% \| 2.7% \| 0.588 \| \| Co-morbidities \| Co-morbidities \| Co-morbidities \| Co-morbidities \| Co-morbidities \| \| Diabetic \| 30.3% \| 35.9% \| 23.9% \| 0.042 \| \| CKDb \| 0.8% \| 1.6% \| 0% \| 0.182 \| \| CLDc \| 0.8% \| 0.8% \| 0.8% \| 0.929 \| \| Heart disease \| 4.6% \| 7.8% \| 0.9% \| 0.01 \| \| Limitation of mobility \| 11.2% \| 15.6% \| 6.2% \| 0.021 \| \| Injection drug abuse \| 0.4% \| 0.8% \| 0% \| 0.346 \| \| Recurrent skin infections \| 11.6% \| 14.1% \| 8.8% \| 0.208 \| \| Recent antibiotic use \| 40.2% \| 47.7% \| 31.9% \| 0.013 \| With respect to management, there was a significantly higher rate of prescription of any antibiotic and anti-methicillin-resistant Staphylococcus aureus (anti-MRSA) antibiotics in the surgical procedure group. We also saw a higher rate of oral antibiotics prescription, hospitalization, wound culture, and complete blood count (CBC) in this group (Table 4). Although follow-up was included as a secondary objective of our study, only 129 patients responded to follow-up communication ($53.3\%$). Using a univariate chi-square analysis on these 129 cases showed no statistically significant difference in improvement between the surgical and non-surgical groups (Table 4). Table 4 \| Variables \| Surgical Procedure Done (n=128) \| No Surgical Procedure (n=113) \| p-valuea \| \| --- \| --- \| --- \| --- \| \| \| % \| % \| \| \| Category of Antibiotics \| Category of Antibiotics \| Category of Antibiotics \| Category of Antibiotics \| \| Any Antibiotics Prescribed \| 93.8 \| 73.5 \| < .001 \| \| Beta-lactam monotherapyb \| 50 \| 51.3 \| .837 \| \| Anti-MRSA monotherapyc \| 36.7 \| 13.3 \| < .001 \| \| Beta lactam + Anti MRSA \| 1.6 \| 0.9 \| .636 \| \| Other Antibioticsd \| 12.5 \| 12.4 \| .979 \| \| No Antibiotics \| 6.3 \| 26.5 \| < .001 \| \| Route of administration \| Route of administration \| Route of administration \| Route of administration \| \| Topical Antibiotics \| 7 \| 6.2 \| .795 \| \| Oral Antibiotics \| 80.5 \| 65.5 \| .009 \| \| Parenteral Antibiotics \| 15.6 \| 8 \| .068 \| \| Hospitalized \| 21.9 \| 1.8 \| < .001 \| \| Laboratory work-up \| Laboratory work-up \| Laboratory work-up \| Laboratory work-up \| \| Wound Culture \| 22.7 \| 5.3 \| < .001 \| \| CBCe \| 35.2 \| 15 \| < .001 \| \| Blood culture \| 0.8 \| 0.9 \| .929 \| \| Follow-up(n=129)f \| Follow-up(n=129)f \| Follow-up(n=129)f \| Follow-up(n=129)f \| \| One week improvement \| 36.7 \| 50 \| .136 \| \| Two weeks improvement \| 68.4 \| 74 \| .493 \| Finally, we compared different CA-SSTIs in terms of the proportions of different diagnostic and treatment modalities using a univariate chi-square. We found a statistically significant association of the different SSTIs with wound culture ($p \leq 0.001$), CBC ($p \leq 0.001$), blood culture ($$p \leq 0.014$$), and hospitalization ($p \leq 0.001$). There was no significant difference in the antibiotic prescription and surgical procedure for different CA-SSTIs (Table 5). Table 5 \| Unnamed: 0 \| Unnamed: 1 \| Wound Culture Ordered [p < 0.001] a \| CBCb [p < 0.001] \| Blood Culture [p = 0.014] \| Antibiotics Ordered [p = 0.53] \| Surgical Procedure [p = 0.24] \| Hospitalized [p < 0.001] \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Clinical Diagnosis \| Frequency (n) \| % \| % \| % \| % \| % \| % \| \| Abscess \| 85 \| 21.2% \| 24.7% \| 1.2% \| 84.7% \| 50.6% \| 8.2% \| \| Infected Ulcer \| 45 \| 17.8% \| 53.3% \| 0% \| 73.3% \| 64.4%% \| 24.4% \| \| Cellulitis \| 31 \| 0% \| 9.7% \| 0% \| 80.6% \| 38.7% \| 9.7% \| \| Paronychia \| 22 \| 0% \| 0% \| 0% \| 95.5% \| 45.5% \| 4.5% \| \| Traumatic wound infection \| 22 \| 0% \| 4.5% \| 0% \| 90.9% \| 59.1% \| 4.5% \| \| Infected cyst \| 13 \| 15.4% \| 30.8% \| 0% \| 100% \| 69.2% \| 7.7% \| \| Furuncle/Folliculitis \| 6 \| 16.7% \| 0% \| 0% \| 100% \| 33.3% \| 0% \| \| Carbuncle \| 5 \| 60% \| 80% \| 0% \| 100% \| 80% \| 40% \| \| Necrotizing fasciitis \| 5 \| 60% \| 100% \| 20% \| 80% \| 80% \| 80% \| \| Warts \| 5 \| 0% \| 0% \| 0% \| 40% \| 40% \| 0% \| \| Impetigo \| 1 \| 0% \| 0% \| 0% \| 100% \| 0% \| 0% \| \| Mastitis \| 1 \| 0% \| 0% \| 0% \| 100% \| 0% \| 0% \| \| Total \| 241 \| 14.5% \| 25.7% \| 0.8% \| 84.2% \| 53.1% \| 12.4% \| ## Discussion SSTIs are common presentations in ED [10]. Mistry et al. reported that $51\%$ of the SSTI patients treated in the ED were males and $49\%$ were females [6]. In addition, $57\%$ of their patients belonged to the 18-49 years age group. Our study showed $59\%$ male patients versus $41\%$ females. Most of our patients were in the 21-30 years age group ($26.6\%$). The most common CA-SSTIs were abscesses, infected ulcers, and cellulitis. Around half ($53.1\%$) of our patients received some sort of surgical procedure and $84\%$ received an antibiotic. The surgical procedure was strongly associated with diabetes, heart disease, limitation of mobility, and recent antibiotic use as these factors are associated with severe purulent infections. Among the purulent infections, a surgical procedure was performed in $50.6\%$ of abscesses, $80\%$ of carbuncles, $33\%$ of furuncles, $69\%$ of infected cysts, and $80\%$ of necrotizing fasciitis patients. The rate of antibiotic prescription was much higher in all these infections. Mistry et al. reported I/D in $27\%$ of patients presenting with an SSTI to the ED with an antibiotic prescription rate of $85\%$ overall. Like other hospitals, the instinctive practice of antibiotics prescription more than I/D exists in our setting as well. Diagnosis of these infections is mostly clinical [1]. Guidelines suggest that CBC and wound culture should be ordered for all complicated SSTIs, and for those with sepsis should be ordered for blood culture as well [11]. These tests can be avoided in uncomplicated infections [1]. In our study, wound culture was ordered in $14\%$ of patients, more commonly in the surgical procedure group ($22.7\%$). CBC was also strongly associated with surgical procedures ($p \leq 0.001$). Mistry et al. reported similar results with wound culture in $16\%$ of patients, more common in those receiving an I/D. However, this study reported a higher rate of CBC and blood culture in the non-surgical group. Kamath et al. reported wound culture in all the patients ($100\%$) undergoing I/D for an SSTI [11]. Although financial affordability is a huge limitation in our setting, wound culture is still strongly advisable in patients undergoing a surgical procedure for an SSTI. This will promote antibiotic stewardship and reduce antibiotic resistance. Our study reported a $12.4\%$ hospitalization, more commonly in necrotizing fasciitis, carbuncles, and infected ulcers. It was also strongly associated with surgical procedures ($21.9\%$). Hospitalization for SSTI is associated with larger lesions, fever, and comorbidities [12]. Mistry et al. also reported a strong association between hospitalization and I/D. In compliance with the guidelines, our emergency-based hospitalization is very selective and reserved only for severe conditions as reported [11]. In a large multicenter study from the United States, Fritz et al. reported that about $58.6\%$ of patients received anti-MRSA drugs and only $35.1\%$ beta-lactam drugs for ambulatory patients with SSTIs [10]. The most common antibiotics were trimethoprim-sulfamethoxazole (TMP-SMX), cephalexin, clindamycin, and doxycycline [10]. Antibiotics were prescribed to $84\%$ of our patients. The surgical procedure group was more likely to receive any antibiotics ($94\%$) and anti-MRSA antibiotics ($36.7\%$). In a large multicenter study from the United States, Mistry et al. reported no difference in the antibiotic prescription rate but showed that anti-MRSA drugs were more commonly prescribed to those undergoing an I/D. Guidelines by the Infectious Disease Society of America (IDSA) recommend I/D or D/D along with anti-MRSA agents as the first-line treatment for purulent infections [11]. Only $25\%$ of our patients received anti-MRSA drugs and $49\%$ received beta-lactam antibiotics. The most common antibiotics were Amoxicillin-Clavulanate and Linezolid. Although MRSA is a significant problem in our population [13], we are still relying on beta-lactam antibiotics mostly. This can be attributed to the lower awareness of the physicians on the updated local antibiogram and current international guidelines. According to a recent local study, highly sensitive anti-MRSA antibiotics include Fusidic acid, Teicoplanin, Chloramphenicol, Doxycycline, and Linezolid [14]. Another recent antibiogram from our population reported a high susceptibility of MRSA isolates to Vancomycin and Amikacin ($94.4\%$), followed by Teicoplanin, Doxycycline, Ciprofloxacin, Cefotaxime, Tigecycline, Clindamycin, and Linezolid [15]. While we have all these options available against MRSA infection, only Linezolid was prescribed as a systemic anti-MRSA agent. Although Linezolid is highly effective against MRSA and other staphylococci [16], the unchecked generous use of Linezolid as empiric therapy might result in the disastrous growth of extended drug-resistant staphylococci. Further study is required on the anti-MRSA antibiotics prescription in comparison to the local antibiogram. Awareness is required to promote the use of other first-line anti-MRSA drugs like clindamycin, TMP-SMX, doxycycline, vancomycin, etc. rather than only Linezolid. Although our study showed no difference in the outcomes in both groups, it is difficult to discuss this aspect due to the loss of follow-up. Limitations This prospective study gives a detailed insight into the burden of a prevalent complaint in our ED. It also describes in detail the current practices in our population. However, it has a few limitations: [1] loss of follow-up in about half the patients, [2] no microbiologic comparison, [3] a relatively small sample from only one center, and [4] multivariable regression analysis was not performed. Although a multivariate analysis is ideal to evaluate the association of multiple independent variables with a dependent variable, this type of analysis gives erroneous results in studies with a relatively small sample size [17]. ## Conclusions This study shows a higher frequency of purulent infections among the SSTIs presenting to our ED. Antibiotics were prescribed more frequently for all infections, though surgical procedures like incision and drainage were much lower even in purulent infections. Furthermore, beta-lactam antibiotics like Amoxicillin-Clavulanate were commonly prescribed. Linezolid was the only systemic anti-MRSA agent prescribed. We suggest physicians should prescribe antibiotics appropriate to the local antibiograms and the latest guidelines. We also recommend further studies on these infections in comparison to the microbiologic studies. ## References 1. Ramakrishnan K, Salinas RC, Higuita NIA. Skin and soft tissue infections. Am Fam Physician (2015) 92 474-483. PMID: 26371732 2. Morgan E, Hohmann S, Ridgway JP, Daum RS, David MZ. Decreasing incidence of skin and soft-tissue infections in 86 US Emergency Departments, 2009-2014. Clin Infect Dis (2019) 68 453-459. PMID: 29912305 3. Jeng A, Beheshti M, Li J, Nathan R. The role of beta-hemolytic streptococci in causing diffuse, nonculturable cellulitis: a prospective investigation. Medicine (Baltimore) (2010) 89 217-226. PMID: 20616661 4. May AK. Skin and soft tissue infections. Surg Clin North Am (2009) 89 403-0. PMID: 19281891 5. Sartelli M, Malangoni MA, May AK. World Society of Emergency Surgery (WSES) guidelines for management of skin and soft tissue infections. World J Emerg Surg (2014) 9 57. PMID: 25422671 6. Mistry RD, Shapiro DJ, Goyal MK, Zaoutis TE, Gerber JS, Liu C, Hersh AL. Clinical management of skin and soft tissue infections in the U.S. Emergency Departments. West J Emerg Med (2014) 15 491-498. PMID: 25035757 7. Abrahamian FM, Talan DA, Moran GJ. Management of skin and soft-tissue infections in the emergency department. Infect Dis Clin North Am (2008) 22 89-0. PMID: 18295685 8. Cardoso T, Almeida M, Friedman ND, Aragão I, Costa-Pereira A, Sarmento AE, Azevedo L. Classification of healthcare-associated infection: a systematic review 10 years after the first proposal. BMC Med (2014) 12 40. PMID: 24597462 9. Stevens DL. Treatments for skin and soft-tissue and surgical site infections due to MDR Gram-positive bacteria. J Infect (2009) 59 0-9 10. Fritz SA, Shapiro DJ, Hersh AL. National trends in incidence of purulent skin and soft tissue infections in patients presenting to ambulatory and emergency department settings, 2000-2015. Clin Infect Dis (2020) 70 2715-2718. PMID: 31605485 11. Kamath RS, Sudhakar D, Gardner JG, Hemmige V, Safar H, Musher DM. Guidelines vs actual management of skin and soft tissue infections in the emergency department. Open Forum Infect Dis (2018) 5 0 12. Talan DA, Salhi BA, Moran GJ, Mower WR, Hsieh YH, Krishnadasan A, Rothman RE. Factors associated with decision to hospitalize emergency department patients with skin and soft tissue infection. West J Emerg Med (2015) 16 89-97. PMID: 25671016 13. Ullah A, Qasim M, Rahman H. High frequency of methicillin-resistant Staphylococcus aureus in Peshawar Region of Pakistan. Springerplus (2016) 5 600. PMID: 27247896 14. Rehman LU, Khan AA, Afridi P, Rehman SU, Wajahat M, Khan F. Prevalence and antibiotic susceptibility of clinical Staphylococcus aureus isolates in various specimens collected from a tertiary care hospital, Hayatabad, Peshawar, Pakistan. Pak J Health Sci (2022) 3 105-110 15. Asif A, Asghar M, Khan HU. Antibiotic susceptibility pattern of clinical isolates of methicillin resistant Staphylococcus aureus in Peshawar, Pakistan. Ann Romanian Soc Cell Biol (2021) 25 20116-20131 16. Shariati A, Dadashi M, Chegini Z, van Belkum A, Mirzaii M, Khoramrooz SS, Darban-Sarokhalil D. The global prevalence of Daptomycin, Tigecycline, Quinupristin/Dalfopristin, and Linezolid-resistant Staphylococcus aureus and coagulase-negative staphylococci strains: a systematic review and meta-analysis. Antimicrob Resist Infect Control (2020) 9 56. PMID: 32321574 17. Nemes S, Jonasson JM, Genell A, Steineck G. Bias in odds ratios by logistic regression modelling and sample size. BMC Med Res Methodol (2009) 9 56. PMID: 19635144
--- title: Prevalence and factors associated with treatment and control of hypertension among adults with hypertension in Myanmar authors: - Ze Haung - Seo Ah Hong journal: International Health year: 2022 pmcid: PMC9977219 doi: 10.1093/inthealth/ihac047 license: CC BY 4.0 --- # Prevalence and factors associated with treatment and control of hypertension among adults with hypertension in Myanmar ## Abstract ### Background Due to a dearth in the number of studies conducted in low- and middle-income countries, this study aimed to identify the prevalence and determinants of the treatment and control of hypertension among patients with hypertension in Myanmar. ### Methods This community-based cross-sectional study was conducted among 410 adults who were registered for hypertensive treatment in health centers in Myitkyina Township, Kachin State, Myanmar. Multiple logistic regression was used to identify the associated factors. ### Results The prevalence of treatment and control of hypertension was $48.1\%$ and $20.5\%$, respectively. The factors associated with treatment were age (OR=2.60 for 46–60 y and OR=2.29 for 61–70 y compared with 30–45 y), ethnicity (OR=1.87), monthly family income (OR=1.90), comorbidity (OR=2.33), knowledge (OR=2.63) and adherence to physical activity (OR=1.86). Controlled hypertension was associated with age (OR=3.03 for 46–60 y and OR=2.27 for 61–70 y compared with 30–45 y), education (OR=1.81), comorbidity (OR=1.67) and adherence to medication (OR=3.45). ### Conclusions The prevalence of treated and controlled hypertension was relatively low in this study. To improve the prevalence of hypertension treatment and control in this study population, effective and culturally sensitive intervention programs under universal health coverage should be established with an emphasis on individuals with lower educational attainment and younger ages. ## Introduction Hypertension is a major topic of concern in public health globally, as it can lead to heart diseases, stroke and chronic kidney diseases.1 Hypertension is a major contributor to global mortality; it is estimated to be associated with $19.2\%$ of deaths (10.7 million) worldwide.2 The financial burden for uncontrolled cases of hypertension is too expensive to neglect. Uncontrolled hypertension is defined as systolic blood pressure ≥140 mmHg and diastolic blood pressure ≥90 mmHg.3 The annual global direct healthcare cost due to uncontrolled hypertension is estimated at US$372 billion, representing about $10\%$ of the world's overall healthcare expenditure.4 In fact, reducing blood pressure (BP) in patients with hypertension is highly beneficial to prevent the development of its complications and death. It is estimated that effective BP control could avert 0.77 million deaths among patients at medium to high cardiovascular disease (CVD) risk worldwide.5 Despite the importance of controlling hypertension, its treatment and control rates remain unacceptably low across the globe, particularly in low- and middle-income countries (LMICs).6 The most recent global estimates suggested that $36.9\%$ of those with hypertension receive treatment globally, while only $20\%$ achieve BP control in LMICs compared with $42\%$ in high-income countries (HICs).6 Studies also estimated that the rates of hypertension treatment and control have significantly improved in HICs, but they are still at a modest stage in LMICs.6 As a result, the risk of dying from hypertension is more than double in LMICs than in HICs.6 Pharmacological treatment is critical to control BP and to prevent the development of its complications. Randomized clinical trials have demonstrated that commonly available antihypertensive medicines significantly reduce the risk of CVD and all-cause mortality.7 Along with medical treatment, lifestyle modification, including self-care behaviors such as maintaining body weight, adopting a healthy diet, engaging in physical activity, avoiding smoking and moderating alcohol consumption, also play an important role in the prevention and control of hypertension.8 Studies revealed that self-care behavioral factors were associated with being treated and controlled.9,10 However, there are limited studies on the association of self-care behaviors with hypertension treatment and control in LMICs. In an attempt to improve the treatment and control rates of hypertension, understanding the factors associated with them might provide meaningful insights on how to better control hypertension. Sociodemographic factors such as age, gender, education, ethnicity, family income and body mass index (BMI) influence treatment and control.10–14 Likewise, knowledge on hypertension and self-efficacy in managing hypertension as psychological factors are also associated with being treated and controlled.15,16 Patients with hypertension necessitate regular follow-up and long-term treatment; therefore, accessibility to healthcare services is highly important for its effective management, especially in developing countries.17 Both qualitative and quantitative studies have underlined that the accessibility to healthcare services such as distance to health centers, perceived healthcare cost and the satisfaction derived from healthcare services influenced treated and controlled hypertension.17,18 Myanmar, an LMIC, is situated in Southeast Asia with diverse ethnicities, cultures, food habits and dietary patterns.19 There are seven states (Kachin, Kayah, Kayin, Mon, Rakhine, Shan and Chin) and seven regions (Ayeyarwady, Bago, Magway, Mandalay, Sagaing, Tanintharyi and Yangon) in Myanmar. Non-communicable diseases (NCDs) including hypertension are becoming prevalent and emerging as major public health concerns, and stroke and ischemic heart disease are leading causes of death in Myanmar.20 According to a nationwide study, the prevalence of hypertension in Myanmar was considerably high at $26.4\%$ in 2014,21 while the WHO estimated the prevalence of hypertension to be $23\%$ in 2018.22 Although the WHO's estimate was slightly lower than that of the nationwide survey, it was still the second highest prevalence among member countries of the Association of Southeast Asia Nations.22 With urbanization, adoption of a westernized lifestyle and economic development in Myanmar, some risk factors for NCDs such as smoking, alcohol consumption, unhealthy food habits, physical inactivity and obesity are on the increase.23 Despite the high prevalence of hypertension, limited information regarding the treatment and control of hypertension is available in Myanmar. Few studies have attempted to explore the treatment and control of hypertension in Myanmar.8,21 Additionally, since 2014, those kinds of studies have not been conducted in Myanmar or at least not in Kachin State. There is also little information on the associated factors for the treatment and control of hypertension. Therefore, this study aimed to describe the treatment and control rates of hypertension and their associated factors, such as sociodemographic, psychological, self-care behavioral and accessibility to healthcare services, among patients with hypertension in Myitkyina Township, Kachin State. Hence, the findings of this study will serve as baseline information for policymakers and health professionals to formulate effective interventions in the prevention and control of hypertension in Myanmar. It will help to achieve Myanmar's national strategic plan of reducing, to a reasonable extent, the unconditional probability of dying between the ages 30 to 70 y from major NCDs by $20\%$ from 2017 to 2025.24 The results here will also help provide a foundation to achieve the Sustainable Development Goals to reduce $25\%$ of NCD mortality by 2025 in Myanmar.22 ## Study design and participants This cross-sectional community-based study was carried out in Myitkyina Township, Kachin State, Myanmar during April and May 2019. Myitkyina is the capital city of Kachin State, the northernmost state of Myanmar. The ethnic make-up of Myitkyina includes Kachin, Shan and Burma-dominant ethnic groups. Adults aged 30–70 y who were registered for hypertension treatment in health centers in Myitkyina during the past year were included. The sample size was estimated using a CI of $95\%$, an acceptable error of $5\%$ and the prevalence of medication adherence is $50\%$.25 Based on the total estimated target population in the Township ($$n = 2567$$), this calculation determined a required sample size of 336. The sample size was increased to 402, which accounts for $10\%$ probable non-response. A multistage cluster random sampling method was used in this study. Stratified by rural and urban places of residence, one center among two urban health centers and four centers among six rural health centers were randomly selected. Each urban health center has their catchment areas (wards), and so do rural health centers (villages). Three wards of the selected urban health center and three villages of each of the four selected rural health centers were randomly selected. In total, three wards (urban areas) and 12 villages (rural areas) were selected. Lastly, adults aged 30 to 70 y and registered for hypertensive treatment in health centers were identified through outpatient medical records and were randomly selected. Those who were seriously ill, who had cognitive impairment and who were pregnant with gestational hypertension, were excluded. With the exclusion of 10 patients who did not meet the inclusion criteria or who were not available on the day of data collection, a total of 410 ($97.6\%$) patients voluntarily agreed to participate in the study. Ethical approval was obtained from the Committee for Research Ethics (Social Sciences), Mahidol University, Thailand (No. $\frac{2019}{070.0204}$) and Institutional Review Board, University of Public Health, Yangon, Myanmar (UPH-IRB-2019/Research/20). Prior to the study, we also obtained permission from Township Public Health Department and identified the sample and made appointments with participants in collaboration with local health personnel (public health practitioners and certified midwives) and community authorized persons. Trained researchers collected data via face-to-face interviews at the participants’ residences. The objectives and procedure of data collection were explained to the participants prior to the survey. They were also assured of the protection of their rights and agreement and were told that they could terminate participation at any time without prejudice. Written informed consent was obtained from each participant. All data were treated anonymously using study identification numbers. ## Measurement Treatment and control of hypertension Treatment of hypertension was assessed by the question ‘How many of the past 7 days did you take your blood pressure pills?’ The participants who responded that they took ≥1 d were counted as receiving treatment. Control of hypertension was defined as systolic BP <140 mmHg and diastolic BP <90 mmHg using the Seventh Report of the Joint National Committee (JNC 7) classification.8 BP was measured twice at an interval of 2 min in a sitting position after taking a rest for a minimum of 5 min using digital sphygmomanometers (OMRON HEM-8712). BP was determined by the average of those two measurements. Sociodemographic factors and accessibility to healthcare services The sociodemographic factors included were age, gender, marital status, occupation, ethnicity, education and monthly family income in Myanmar Kyats (MMK), place of residence, duration of hypertension, family history of hypertension, comorbidity and weight status. Weight status was defined using BMI. Each participant’s weight was measured by digital scales to the nearest 0.1 kg and height by wooden height measuring boards to the nearest 0.5 cm according to the WHO guidelines.26 BMI was classified into four categories according to the guidelines for Asian adults: underweight (BMI<18.5 kg/m2), normal weight (18.5 kg/m2≤BMI<23.0 kg/m2), overweight or obese (BMI≥23.0 kg/m2).27 In addition, distance and satisfaction to the nearest health center and perceived healthcare costs were also included. Psychological factors Knowledge of hypertension was measured by the Hypertension Knowledge Level Scale questionnaire.28 It was composed of 22 items; questions elicited information regarding patient knowledge on the definitions of hypertension, lifestyle, treatment, adherence to medication and complications of hypertension. Answer options included ‘True’, ‘False’ or ‘Don't know’. A correct answer received a score of 1 whereas the score was 0 for an incorrect answer or ‘Don't know’. The total score possible ranged from 0 to 22. Those with a total score of ≥18.0 were considered as having an adequate level of knowledge, while those with a total score of <18 were classified as having an inadequate level of knowledge.28 Perceived self-efficacy to manage hypertension was measured using the Perceived Self-efficacy to Manage Hypertension Scale questionnaire.29 It was composed of five items concerning the level of confidence in managing hypertension and the response options ranged from 1 (‘not confident at all’) to 10 (‘totally confident’). Of the total score (5 to 50), those having a mean score of ≥9 were classified as having a good level, while the scores of <9 were classified as a poor level of confidence. Cronbach's alpha was 0.68 for knowledge and 0.79 for self-efficacy in this study. Self-care behaviors Self-care behaviors were measured by Hypertension Self-Care Activity Level Effect questionnaires (H-scale) developed by Warren and Seymour to assess the self-care behaviors of patients with hypertension in healthcare settings and epidemiological surveys.30 It is a 29-item scale that has six subdomains, such as adherence to antihypertensive medication (two questions), healthy diet (11 questions), engagement in adequate physical activity (two questions), practicing proper weight management (10 questions), avoidance of tobacco use (two questions) and abstinence from harmful alcohol drinking (two questions). The cut-off points on this scale were according to the original study's classification.30 Details of these self-care behaviors in this study have been described elsewhere.31 ## Statistical analysis Descriptive statistics (frequency and %) were used to describe sample distributions. χ2 tests were conducted to examine the association between hypertension treatment and control variables, and independent variables. Variables that showed associations with the outcomes at a significance level of 0.1 were entered into a logistic regression model using the backward method to predict controlled and treated hypertension. The data were analyzed in SPSS version 23 (IBM, Armonk, NY, USA). ## Results The sociodemographic characteristics, accessibility to healthcare services, psychological factors and behavioral risk factors of patients with hypertension are presented in Table 1. The mean age of the population was 55.4 (SD=11.0) y, and more than two-thirds of the population were women ($76.6\%$), of Kachin ethnicity ($58.1\%$), living with a spouse/partner ($67.8\%$) and living in rural areas ($81.2\%$) (Table 1). The prevalence of hypertension treatment and control among all participants was $48.1\%$ and $20.5\%$, respectively. Table 1. \| Unnamed: 0 \| Unnamed: 1 \| Treatment \| Treatment.1 \| Control \| Control.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| n (%) \| n (%) \| p value \| n (%) \| p value \| \| Total sample \| 410 (100.0) \| 197 (48.1) \| \| 84 (20.5) \| \| \| Sociodemographic factors \| \| \| \| \| \| \| Age group, y \| \| \| \| \| \| \| 30–45 \| 85 (20.7) \| 29 (34.1) \| 0.0123 \| 8 (9.4) \| 0.0130 \| \| 46–60 \| 176 (42.9) \| 94 (53.4) \| \| 44 (25.0) \| \| \| 61–70 \| 149 (36.3) \| 74 (49.7) \| \| 32 (21.5) \| \| \| Gender \| \| \| \| \| \| \| Male \| 96 (23.4) \| 56 (58.3) \| 0.0212 \| 15 (15.6) \| 0.1774 \| \| Female \| 314 (76.6) \| 141 (44.9) \| \| 69 (22.0) \| \| \| Marital status \| \| \| \| \| \| \| Living without a partner \| 132 (32.2) \| 56 (42.4) \| 0.1162 \| 30 (22.7) \| 0.4388 \| \| Married/living with a partner \| 278 (67.8) \| 141 (50.7) \| \| 54 (19.4) \| \| \| Occupation \| \| \| \| \| \| \| Employed/self-employed \| 84 (20.5) \| 51 (60.7) \| 0.0074 \| 21 (25.0) \| 0.4971 \| \| Farmer \| 176 (42.9) \| 71 (40.3) \| \| 33 (18.8) \| \| \| Dependent/housewife/other \| 150 (36.6) \| 75 (50.0) \| \| 30 (20.0) \| \| \| Ethnicity \| \| \| \| \| \| \| Kachin \| 238 (58.1) \| 100 (42.0) \| 0.0040 \| 43 (18.1) \| 0.1532 \| \| Other \| 172 (42.0) \| 97 (56.4) \| \| 41 (23.8) \| \| \| Education \| \| \| \| \| \| \| Primary school – \| 241 (58.8) \| 103 (42.7) \| 0.0102 \| 39 (16.2) \| 0.0099 \| \| Secondary school + \| 169 (41.2) \| 94 (55.6) \| \| 45 (26.6) \| \| \| Monthly income (MMK) \| \| \| \| \| \| \| Low (<100 000) \| 180 (43.9) \| 65 (36.1) \| <0.0001 \| 28 (15.6) \| 0.0411 \| \| Middle (100 001–250 000) \| 102 (24.9) \| 54 (52.9) \| \| 21 (20.6) \| \| \| High (>250 000) \| 128 (31.2) \| 78 (60.9) \| \| 35 (27.3) \| \| \| Place of residence \| \| \| \| \| \| \| Urban \| 77 (18.8) \| 51 (66.2) \| 0.0004 \| 17 (22.1) \| 0.7013 \| \| Rural \| 333 (81.2) \| 146 (43.8) \| \| 67 (20.1) \| \| \| Duration of hypertension, y \| \| \| \| \| \| \| ≤3 \| 167 (40.7) \| 68 (40.7) \| 0.0138 \| 29 (17.4) \| 0.1941 \| \| >3 \| 243 (59.3) \| 129 (53.1) \| \| 55 (22.6) \| \| \| Family history of hypertension \| \| \| \| \| \| \| Yes \| 151 (36.8) \| 77 (51.0) \| 0.3622 \| 36 (23.8) \| 0.1990 \| \| No \| 259 (63.2) \| 120 (46.3) \| \| 48 (18.5) \| \| \| Comorbidity \| \| \| \| \| \| \| No \| 240 (58.5) \| 97 (40.4) \| 0.0002 \| 38 (15.8) \| 0.0055 \| \| Yes \| 170 (41.5) \| 100 (58.8) \| \| 46 (27.1) \| \| \| Body mass index, kg/m2 \| \| \| \| \| \| \| Overweight and obese (≥23.0) \| 266 (64.9) \| 137 (51.5) \| 0.0570 \| 50 (18.8) \| 0.2490 \| \| Normal and underweight (<23.0) \| 144 (35.1) \| 60 (41.7) \| \| 34 (23.6) \| \| Tables 1 and 2 show the association between independent variables and treatment and control of hypertension using χ2 tests. In univariate associations, factors associated with treatment were age ($$p \leq 0.0123$$), gender ($$p \leq 0.0212$$), occupation ($$p \leq 0.0074$$), ethnicity (0.0040), education (0.0102), monthly income ($p \leq 0.0001$), place of residence ($$p \leq 0.0004$$), duration of hypertension ($$p \leq 0.0138$$), family history ($$p \leq 0.3622$$), comorbidity ($$p \leq 0.0002$$), BMI ($$p \leq 0.0570$$) and knowledge ($p \leq 0.0001$), as well as adherence to medication ($p \leq 0.0001$), physical activity ($$p \leq 0.0030$$) and weight management ($$p \leq 0.0763$$). For controlled hypertension, they were age ($$p \leq 0.0130$$), education ($$p \leq 0099$$), monthly income ($$p \leq 0.0411$$), comorbidity ($$p \leq 0.0055$$), knowledge ($$p \leq 0.0809$$) and self-efficacy ($$p \leq 0.0153$$), as well as adherence to medication ($p \leq 0.0001$) and physical activity ($$p \leq 0.0218$$). Table 2. \| Unnamed: 0 \| Unnamed: 1 \| Treatment \| Treatment.1 \| Control \| Control.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| n (%) \| n (%) \| p value \| n (%) \| p value \| \| Accessibility to healthcare services \| \| \| \| \| \| \| Distance from nearest health center \| \| \| \| \| \| \| <1 mile (1.609344 km) \| 232 (56.6) \| 118 (50.9) \| 0.1455 \| 53 (22.8) \| 0.1727 \| \| 1–5 miles (1.609344–8.04672 km) \| 152 (37.1) \| 64 (42.1) \| \| 24 (15.8) \| \| \| >5 miles (8.04672 km) \| 26 (6.3) \| 15 (57.7) \| \| 7 (26.9) \| \| \| Satisfaction based on healthcare services \| \| \| \| \| \| \| Completely \| 324 (79.0) \| 150 (46.3) \| 0.1680 \| 62 (19.1) \| 0.1880 \| \| To some extent/not at all \| 86 (21.0) \| 47 (54.7) \| \| 22 (25.6) \| \| \| Perceived healthcare costs \| \| \| \| \| \| \| Cheap \| 308 (75.3) \| 150 (48.7) \| 0.7053 \| 64 (20.8) \| 0.8329 \| \| Expensive \| 101 (24.7) \| 47 (46.5) \| \| 20 (19.8) \| \| \| Psychological factors \| \| \| \| \| \| \| Knowledge on hypertension \| \| \| \| \| \| \| Inadequate \| 225 (54.9) \| 86 (38.2) \| <0.0001 \| 39 (17.3) \| 0.0809 \| \| Adequate \| 185 (45.1) \| 111 (60.0) \| \| 45 (24.3) \| \| \| Self-efficacy to manage hypertension \| \| \| \| \| \| \| Poor \| 331 (80.7) \| 156 (47.1) \| 0.4459 \| 60 (18.1) \| 0.0153 \| \| Good \| 79 (19.3) \| 41 (51.9) \| \| 24 (30.4) \| \| \| Self-care behaviors \| \| \| \| \| \| \| Adherence to medication \| \| \| \| \| \| \| Non-adherence (<14 scores) \| 311 (75.9) \| 98 (31.5) \| <0.0001 \| 44 (14.2) \| <0.0001 \| \| Adherence (14 scores) \| 99 (24.2) \| 99 (100.0) \| \| 40 (40.4) \| \| \| Adherence to healthy diet \| \| \| \| \| \| \| Low/middle diet quality (<52 scores) \| 402 (98.1) \| 192 (47.8) \| 0.4087 \| 83 (20.7) \| 0.5719 \| \| High diet quality (≥52 scores) \| 8 (2.0) \| 5 (62.5) \| \| 1 (12.5) \| \| \| Adherence to physical activity \| \| \| \| \| \| \| Non-adherence (<8 scores) \| 308 (75.1) \| 135 (43.8) \| 0.0030 \| 55 (17.9) \| 0.0218 \| \| Adherence (≥8 scores) \| 102 (24.9) \| 62 (60.8) \| \| 29 (28.4) \| \| \| Adherence to avoidance of tobacco use \| \| \| \| \| \| \| Non-adherence (≥1 scores) \| 204 (49.8) \| 101 (49.5) \| 0.5557 \| 44 (21.6) \| 0.5895 \| \| Adherence (0 score) \| 206 (50.2) \| 96 (46.6) \| \| 40 (19.4) \| \| \| Adherence to weight management \| \| \| \| \| \| \| Non adherence (<40 scores) \| 371 (90.5) \| 173 (46.6) \| 0.0763 \| 75 (20.2) \| 0.6737 \| \| Adherence (≥40 scores) \| 39 (9.5) \| 24 (61.5) \| \| 9 (23.1) \| \| \| Avoidance of harmful alcohol drinking \| \| \| \| \| \| \| No (>14 scores [M] and >7 [W]) \| 9 (2.2) \| 2 (22.2) \| 0.1169 \| 1 (11.1) \| 0.4810 \| \| Yes (≤14 scores [M] and ≤7 [W]) \| 401 (97.8) \| 195 (48.6) \| \| 83 (20.7) \| \| Factors having $p \leq 0.1$ in the bivariate analyses were employed in multiple logistic regression analyses (Table 3). The factors associated with treatment were age (OR=2.60, $95\%$ CI 1.45 to 4.67 for 46–60 y and OR=2.29, $95\%$ CI 1.24 to 4.21 for 61–70 y compared with 30–45 y), ethnicity (OR=1.87, $95\%$ CI 1.20 to 2.91), monthly family income (OR=1.90, $95\%$ CI 1.11 to 3.24 for middle and OR=2.56, $95\%$ CI 1.54 to 4.25 for high compared with low income), comorbidity (OR=2.33, $95\%$ CI 1.51 to 3.61), knowledge (OR=2.63, $95\%$ CI 1.71 to 4.06) and adherence to physical activity (OR=1.86, $95\%$ CI 1.12 to 3.08). Controlled hypertension was associated with age (OR=3.03, $95\%$ CI 1.32 to 6.99 for 46–60 y and OR=2.27, $95\%$ CI 0.96 to 5.39 for 61–70 compared with 30–45 y), education (OR=1.81, $95\%$ CI 1.08 to 3.02 for secondary school or higher compared with primary or less), comorbidity (OR=1.67, $95\%$ CI 1.00 to 2.79) and adherence to medication (OR=3.45, $95\%$ CI 2.03 to 5.88). Table 3. \| Unnamed: 0 \| Treatmenta \| Treatmenta.1 \| Controlb \| Controlb.1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| OR \| (95% CI) \| OR \| (95% CI) \| \| Sociodemographic factors \| \| \| \| \| \| Age group, y \| \| \| \| \| \| 46–60 vs 30–45 \| 2.60 \| (1.45 to 4.67) \| 3.03 \| (1.32 to 6.99) \| \| 61–70 vs 30–45 \| 2.29 \| (1.24 to 4.21) \| N.S \| \| \| Gender (female) \| N.S. \| \| - \| \| \| Occupation \| \| \| \| \| \| Farmer vs employed/self-employed \| N.S. \| \| - \| \| \| Dependent/housewife/other vs employed/self-employed \| N.S. \| \| - \| \| \| Ethnicity (other vs Kachin) \| 1.87 \| (1.20 to 2.91) \| - \| \| \| Education (secondary school + vs primary-) \| N.S. \| \| 1.81 \| (1.08 to 3.02) \| \| Monthly income \| \| \| \| \| \| Middle vs low \| 1.90 \| (1.11 to 3.24) \| N.S. \| \| \| High vs low \| 2.56 \| (1.54 to 4.25) \| N.S. \| \| \| Place of residence (urban) \| N.S. \| \| - \| \| \| Duration of hypertension (>3 y) \| N.S. \| \| - \| \| \| Comorbidity (yes) \| 2.33 \| (1.51 to 3.61) \| 1.67 \| (1.00 to 2.79) \| \| Weight status \| N.s. \| \| - \| \| \| Psychological factors \| \| \| \| \| \| Knowledge (adequate) \| 2.63 \| (1.71 to 4.06) \| N.S. \| \| \| Self-efficacy (good) \| - \| \| N.S. \| \| \| Self-care behavioral factors \| \| \| \| \| \| Adherence to medication (yes) \| N.S. \| \| 3.45 \| (2.03 to 5.88) \| \| Adherence to physical activity (yes) \| 1.86 \| (1.12 to 3.08) \| N.S. \| \| \| Adherence to weight management (yes) \| N.S. \| \| - \| \| ## Discussion The current study aimed to describe the prevalence of treated and controlled hypertension and their associated factors among patients with hypertension. This study revealed that less than half of the respondents ($48.1\%$) were treated. Hypertension treatment in this sample was higher than the national data in 2014, which estimated it at $34.9\%$, and a study in Yangon in the same year revealed $40.1\%$.21,32 In Myanmar, the WHO-recommended Prevention of Essential Non-communicable Diseases clinics project was only launched in 2017.33 The purpose of these clinics is to enhance the prevention, screening and treatment of NCDs including hypertension. The reason for the lower treatment rate in the Yangon study than in ours might be because the Yangon study was conducted before that project was launched. The hypertension control rate among all patients with hypertension in our study is also considerably low at $20.5\%$, while among treated patients it was $40.8\%$. However, the Yangon study stated that it was $45.3\%$ among treated patients with hypertension.32 The disparity in the control rates in our study and the Yangon study might be attributable to the fact that the calculation was performed based only on treated patients with hypertension in the Yangon study, but all the patients with hypertension were included in our study. Our study also found that several sociodemographic factors were associated with treated and controlled BP. Evidence supports that ethnic disparities may affect hypertension prevention, treatment and control.34 Our study also found that other ethnics (Burmese, Shan, Chinese and Nepalese) combined were more likely to be under treatment than in Kachin ethnics. Consistently, several studies have also documented that ethnicity is one of the associated factors for being treated.35,36 Therefore, a culturally sensitive hypertension control program should be developed to reduce ethnic disparities in the treatment of hypertension. Studies showed that individuals of an older age were more likely to be under treatment.12,13 Our study revealed that participants aged 46–60 and 61–70 y were more likely to be treated than those aged 30–45 y. This might be explained by participants being more sensitive to their health condition as they grew older and trying medicines to treat their diseases. This implies that younger individuals were less likely to be under treatment and they should be educated to seek treatment early to prevent hypertensive complications. Moreover, previous studies stated that older-aged individuals were more likely to have controlled BP.13,14 On the contrary, our study revealed that the respondents in middle age (aged 46–60 y) were more likely to have their BP controlled than younger individuals (aged 30–45 y), while no association was observed among older individuals (aged 61–70 y). One explanation may be that younger individuals had a lower frequency of control because they also had a low frequency of treatment. By contrast, while treatment was higher among older individuals, comorbid conditions or hypertensive complications could be influencing control. Moreover, our study also revealed that respondents with equal or more than secondary education level predicted controlled BP than those with equal or less than primary education level, which is supported by other studies.11,14 One potential explanation is that poor education attainment has been found to be associated with reduced levels of hypertension knowledge and poor compliance to hypertension management.37,38 These findings highlight the importance of educational attainment as a marker of better control of hypertension and point out that improving educational status may be a pathway for better hypertension management.39 Our study also showed that the likelihood of being treated was associated with higher family income, which is also consistent with other studies.40,41 Poor affordability of medicines for the prevention of cardiovascular complications was highlighted earlier in the Prospective Urban Rural Epidemiology study.42 In Myanmar, the most recent government health expenditure was only $4.79\%$ of the gross domestic product in 2018.43 This is still a very low percentage and a majority of health spending ($75\%$) is out-of-pocket spending by households.24 Hence, to minimize this affordability problem, Myanmar should establish a financial risk management program like universal healthcare coverage to reduce out-of-pocket expenditure on health.44 Moreover, comorbid participants had higher odds of being treated than those without comorbidity in our study, which is in line with other findings.10 It is possible that the respondents may have been more concerned about their diseases if they became comorbid. Thus, patients without comorbidity should be reminded not to take for granted the need to get treatment on time in order to prevent hypertensive complications. Our study stated that having comorbidity was one of the predictors of having controlled BP, which is concordant with a study in China.14 This might be explained by respondents with comorbidity being more aware of their health condition because they are comorbid and thus had better compliance to hypertension management and therefore their high BP was more likely to be controlled. As a psychological factor, there is another meaningful association between having knowledge about hypertension and being under treatment in our study. In accordance with previous studies,15,45 it was stated that those respondents with an adequate level of knowledge about hypertension were more likely to be under treatment than those with inadequate levels of knowledge. Several studies have revealed that knowledge gaps are important barriers to the effective prevention and treatment of hypertension.15,46 Knowledge has also been acknowledged as an important factor for adoption and sustained health behaviors.47 However, fewer than half of the participants reported having adequate knowledge about hypertension in our study. Hence, to improve patients’ behaviors concerning getting treatment, they should be educated through an effective education program. Several studies have identified the benefits of adopting self-care behaviors as aspects of lifestyle modifications in reducing high BP.8 As a self-care behavioral factor, our study also stated that adherence to engage adequate physical activity had higher odds of being treated. It seems that respondents who were more likely to engage in lifestyle modification also followed other positive lifestyle habits, including getting treatment.48 Furthermore, similar to other studies,9,49 this study also found that adherers to medication were more likely to have BP controlled than non-adherers. As hypertension is a chronic disease and needs long-term treatment, medication adherence is key to reducing high BP and preventing its complications.50 In addition, studies have stated that medication adherence was associated with lower rates of hospitalization and healthcare costs.51 Thus, patients with hypertension should be encouraged to constantly adhere to medication. This study had some limitations. As a cross-sectional study, the findings cannot be used to establish a conclusive cause-and-effect relationship between independent and dependent variables. In addition, the findings of this study cannot be generalized to the entire country. Another limitation is that we did not study the list of medicines that the respondents were taking. If we had studied it, we would have estimated the prevalence of resistant hypertension in this population. However, this study had several strengths. Firstly, participants were selected based on their residences by multistage random sampling techniques, although they were limited to those registered in health centers. Secondly, we measured the respondents’ BP according to a standard guideline in their residences, which may prevent the bias of white coat hypertension that may interfere with the interpretation of the condition of controlled hypertension.52,53 Thirdly, because most of the previous studies were conducted in big cities with the dominant ethnic group in Burma, this study was conducted among minority ethnic groups in the northern part of Myanmar. Thus, it provides some insights into the treatment and control of hypertension among ethnic minorities. ## Conclusions This study revealed that hypertension treatment and control among patients with hypertension were low in this multiethnic sample of adults from Myanmar. Moreover, there were associated factors with treatment and control that could be considered in the prevention and control of hypertension in this study population. Therefore, to improve treatment and control rates among patients with hypertension in this study area, effective and culturally sensitive intervention programs under a financial risk management program like universal health coverage should be established. That program should place an emphasis on individuals with lower educational attainment and younger age. For future studies, we recommend the study of other socioeconomic factors associated with low hypertension treatment and control, such as availability or accessibility to treatment, provision of free medicines, out-of-pocket expenses and costs of drugs, through qualitative and quantitative methods. ## Authors’ contributions Both authors equally contributed in conducting this study. ## Funding None. ## Competing interests The authors declare no conflicts of interest. ## Ethical approval Ethical approval was obtained from the Committee for Research Ethics (Social Sciences), Mahidol University, Thailand (No. $\frac{2019}{070.0204}$) and Institutional Review Board, University of Public Health, Yangon, Myanmar (UPH-IRB-2019/Research/20). ## Data availability The data underlying this article will be shared on reasonable request to the corresponding author. ## References 1. Vos T, Lim SS, Abbafati C. Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet (2020.0) 396 1204-22. PMID: 33069326 2. Forouzanfar MH, Liu P, Roth GA. Global burden of hypertension and systolic blood pressure of at least 110 to 115 mm Hg, 1990-2015. JAMA (2017.0) 317 165-82. PMID: 28097354 3. James PA, Oparil S, Carter BL. 2014 evidence-based guideline for the management of high blood pressure in adults: report from the panel members appointed to the Eighth Joint National Committee (JNC 8). JAMA (2014.0) 311 507-20. PMID: 24352797 4. Gaziano TA, Bitton A, Anand S. International Society of Hypertension. The global cost of nonoptimal blood pressure. J Hypertens (2009.0) 27 1472-7. PMID: 19474763 5. Nugent R, Brouwer E.. Benefits and costs of the non-communicable disease targets for the post-2015 development agenda (2015.0) 6. Mills KT, Bundy JD, Kelly TN. Global disparities of hypertension prevalence and control: a systematic analysis of population-based studies from 90 countries. Circulation (2016.0) 134 441-50. PMID: 27502908 7. Ettehad D, Emdin CA, Kiran A. Blood pressure lowering for prevention of cardiovascular disease and death: a systematic review and meta-analysis. Lancet (2016.0) 387 957-67. PMID: 26724178 8. Chobanian AV, Bakris GL, Black HR. National Heart, Lung, and Blood Institute Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure; National High Blood Pressure Education Program Coordinating Committee. The Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure: the JNC 7 report. JAMA (2003.0) 289 2560-71. PMID: 12748199 9. Menanga A, Edie S, Nkoke C. Factors associated with blood pressure control amongst adults with hypertension in Yaounde, Cameroon: a cross-sectional study. Cardiovasc Diagn Ther (2016.0) 6 439-45. PMID: 27747167 10. Ware LJ, Chidumwa G, Charlton K. Predictors of hypertension awareness, treatment and control in South Africa: results from the WHO-SAGE population survey (Wave 2). J Hum Hypertens (2019.0) 33 157-66. PMID: 30382179 11. Tian S, Dong G-H, Wang D. Factors associated with prevalence, awareness, treatment and control of hypertension in urban adults from 33 communities in China: the CHPSNE Study. Hypertens Res (2011.0) 34 1087-92. PMID: 21775998 12. Sanuade OA, Awuah RB, Kushitor M.. Hypertension awareness, treatment and control in Ghana: a cross-sectional study. Ethn Health (2020.0) 25 702-16. PMID: 29448808 13. Wang H, Zhang X, Zhang J. Factors associated with prevalence, awareness, treatment and control of hypertension among adults in Southern China: a community-based, cross-sectional survey. PloS One (2013.0) 8 e62469. PMID: 23671599 14. Jiang B, Hongmei L, Xiaojuan R. Hypertension detection, management, control and associated factors among residents accessing community health services in Beijing. Sci Rep (2014.0) 4 4845. PMID: 24784167 15. Li S-S, Zhou F, Lu Y-C. Hypertension related knowledge and behaviour associated with awareness, treatment and control of hypertension in a rural hypertensive population: a community based, cross-sectional survey. Blood Press (2016.0) 25 305-11. PMID: 27124657 16. Elder K, Ramamonjiarivelo Z, Wiltshire J. Trust, medication adherence, and hypertension control in Southern African American men. Am J Public Health (2012.0) 102 2242-5. PMID: 22698017 17. Legido-Quigley H, Naheed A, de Silva HA. Patients’ experiences on accessing health care services for management of hypertension in rural Bangladesh, Pakistan and Sri Lanka: a qualitative study. PLoS One (2019.0) 14 e0211100. PMID: 30682093 18. Niu H, Tian M, Ma A. Differences and determinants in access to essential public health services in China: a case study with hypertension people and under-sixes as target population. Chin Med J (2014.0) 127 1626-32. PMID: 24791865 19. Naing C, Aung K.. Prevalence and risk factors of hypertension in Myanmar: a systematic review and meta-analysis. Medicine (2014.0) 93 e100. PMID: 25380081 20. 20 Global Health Estimates 2020: Deaths by cause, age, sex, by country and by region. 2020. Available at https://www.who.int/data/gho/data/themes/mortality-and-global-health-estimates/gheleading -causes-of-death [accessed December 2, 2020].. (2020.0) 21. MEDBOX \| Report on National Survey of Diabetes Mellitus And Risk Factors for non-communicable diseases in Myanmar. (2014.0) 22. Noncommunicable diseases country profiles 2018 23. Bjertness MB, Htet AS, Meyer HE. Prevalence and determinants of hypertension in Myanmar - a nationwide cross-sectional study. BMC Public Health (2016.0) 16 590. PMID: 27430560 24. 24 Ministry of Health and Sports Myanmar . National strategic action plan for prevention and control of NCDs (2017-2021). 25. Han WP, Hong SA, Tiraphat S.. Factors related to medication adherence among essential hypertensive patients in tertiary hospitals in Yangon, Myanmar. J Public Health Dev (2015.0) 13 43-56 26. Physical status: the use of and interpretation of anthropometry. Report of a WHO Expert Committee;. (1995.0) 27. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. Lancet (2004.0) 363 157-63. PMID: 14726171 28. Baliz Erkoc S, Isikli B, Metintas S. Hypertension Knowledge-Level Scale (HK-LS): a study on development, validity and reliability. Int J Environ Res Public Health (2012.0) 9 1018-29. PMID: 22690180 29. Warren-Findlow J, Seymour RB, Huber LRB.. The association between self-efficacy and hypertension self-care activities among African American adults. J Community Health (2012.0) 37 15-24. PMID: 21547409 30. Warren-Findlow J, Seymour RB.. Prevalence rates of hypertension self-care activities among African Americans. J Natl Med Assoc (2011.0) 103 503-12. PMID: 21830634 31. Haung Z, Hong SA, Tejativaddhana P. Multiple self-care behaviors and associated factors in community-dwelling patients with hypertension in Myanmar. Nagoya J Med Sci (2020.0) 82 363-76. PMID: 32581415 32. Htet AS, Bjertness MB, Oo WM. Changes in prevalence, awareness, treatment and control of hypertension from 2004 to 2014 among 25-74-year-old citizens in the Yangon Region, Myanmar. BMC Public Health (2017.0) 17 847. PMID: 29073891 33. Aye LL, Tripathy JP, Maung Maung T. Experiences from the pilot implementation of the Package of Essential Non-communicable Disease Interventions (PEN) in Myanmar, 2017-18: a mixed method study. PLoS One (2020.0) 15 e0229081. PMID: 32069323 34. Bennett A, Parto P, Krim SR.. Hypertension and ethnicity. Curr Opin Cardiol (2016.0) 31 381-6. PMID: 27137760 35. Howard G, Prineas R, Moy C. Racial and geographic differences in awareness, treatment, and control of hypertension: the Reasons for Geographic and Racial Differences in Stroke study. Stroke. (2006.0) 37 1171-8. PMID: 16556884 36. Gu A, Yue Y, Desai RP. Racial and Ethnic Differences in Antihypertensive Medication Use and Blood Pressure Control Among US Adults With Hypertension: the National Health and Nutrition Examination Survey, 2003 to 2012. Circ Cardiovasc Qual Outcomes (2017.0) 10 e003166. PMID: 28096206 37. Sudharsanan N, Ali MK, McConnell M.. Hypertension knowledge and treatment initiation, adherence, and discontinuation among adults in Chennai, India: a cross-sectional study. BMJ Open (2021.0) 11 e040252 38. Abu H, Aboumatar H, Carson KA. Hypertension knowledge, heart healthy lifestyle practices and medication adherence among adults with hypertension. Eur J Pers Cent Healthc (2018.0) 6 108-14. PMID: 32405420 39. Marmot M, Friel S, Bell R. Closing the gap in a generation: health equity through action on the social determinants of health. Lancet (2008.0) 372 1661-9. PMID: 18994664 40. Lu J, Lu Y, Wang X. Prevalence, awareness, treatment, and control of hypertension in China: data from 1·7 million adults in a population-based screening study (China PEACE Million Persons Project). Lancet (2017.0) 390 2549-58. PMID: 29102084 41. Gupta R, Kaur M, Islam S. Association of household wealth index, educational status, and social capital with hypertension awareness, treatment, and control in South Asia. Am J Hypertens (2017.0) 30 373-81. PMID: 28096145 42. Khatib R, McKee M, Shannon H. Availability and affordability of cardiovascular disease medicines and their effect on use in high-income, middle-income, and low-income countries: an analysis of the PURE study data. Lancet (2016.0) 387 61-9. PMID: 26498706 43. 43 Current health expenditure (% of GDP) - Myanmar. The World Bank; 2018. Available at https://data.Worldbank.org/indicator/SH.XPD.CHEX.GD.ZS?locations = MM [accessed December 15, 2020].. (2018.0) 44. Olsen MH, Angell SY, Asma S. A call to action and a lifecourse strategy to address the global burden of raised blood pressure on current and future generations: the Lancet Commission on hypertension. Lancet. (2016.0) 388 2665-712. PMID: 27671667 45. Jiang X, Li W, Ma L. Knowledge on hypertension and the effect of management on hypertension in patients attending hospital clinics. Zhonghua Liu Xing Bing Xue Za Zhi (2002.0) 23 269-72. PMID: 12411071 46. Jolles EP, Clark AM, Braam B.. Getting the message across: opportunities and obstacles in effective communication in hypertension care. J Hypertens (2012.0) 30 1500-10. PMID: 22635137 47. Bandura A.. Human agency in social cognitive theory. Am Psychol (1989.0) 44 1175-84. PMID: 2782727 48. Dastan I, Erem A, Cetinkaya V.. Awareness, treatment, control of hypertension, and associated factors: results from a Turkish national study. Clin Exp Hypertens (2018.0) 40 90-8. PMID: 28686064 49. Cutler JA.. Combinations of lifestyle modification and drug treatment in management of mild-moderate hypertension: a review of randomized clinical trials. Clin Exp Hypertens (1993.0) 15 1193-204. PMID: 8268885 50. Corrao G, Parodi A, Nicotra F. Better compliance to antihypertensive medications reduces cardiovascular risk. J Hypertens (2011.0) 29 610-8. PMID: 21157368 51. Sokol MC, McGuigan KA, Verbrugge RR. Impact of medication adherence on hospitalization risk and healthcare cost. Med Care (2005.0) 43 521-30. PMID: 15908846 52. Whelton PK, Carey RM, Aronow WS. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA Guideline for the Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults: a Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension (2018.0) 71 e13-115. PMID: 29133356 53. Franklin SS, Thijs L, Hansen TW. White-coat hypertension: new insights from recent studies. Hypertension (2013.0) 62 982-7. PMID: 24041952
--- title: 'Association between urinary polycyclic aromatic hydrocarbon metabolites and diabetes mellitus among the US population: a cross-sectional study' authors: - Manthar Ali Mallah - Til Bahadur Basnet - Mukhtiar Ali - Fuwei Xie - Xiang Li - Feifei Feng - Wei Wang - Pingping Shang - Qiao Zhang journal: International Health year: 2022 pmcid: PMC9977221 doi: 10.1093/inthealth/ihac029 license: CC BY 4.0 --- # Association between urinary polycyclic aromatic hydrocarbon metabolites and diabetes mellitus among the US population: a cross-sectional study ## Abstract ### Background The primary aim of this study is to examine the association between urinary polycyclic aromatic hydrocarbons (PAHs) and diabetes mellitus (DM) among the US population. ### Methods We used data from the National Health and Nutritional Examination Survey 2003–16, which is a nationally representative population-based survey of the US non-institutionalized population. Logistic regression analysis was performed to evaluate the association between urinary PAHs and the prevalence of DM using odds ratios (ORs) and $95\%$ confidence intervals (CIs). ### Results The study sample including 13 792 individuals ≥18 y of age. The average ages of the three PAH tertiles were 42.56±19.67, 42.21±19.51 and 43.39±17.99 y. An increased risk of DM was found with increased odds for the second (OR 1.56 [$95\%$ CI 1.36 to 1.79]) and third tertile (OR 1.79 [$95\%$ CI 1.55 to 2.06)] of urinary PAH as compared with the first tertile. Similarly, higher chances of DM were observed in the second (men: OR 1.42 [$95\%$ CI 1.18 to 1.71]; women: OR 1.76 [$95\%$ CI 1.44 to 2.14]) and third tertile (men: OR 1.69 [$95\%$ CI 1.38 to 2.08]; women: OR 1.79 [$95\%$ CI 1.46 to 2.19]) of urinary PAHs as compared with the first tertile in both men and women. ### Conclusions A population-based cross-sectional study found a positive association between urinary PAHs and DM in the US population. ## Introduction Diabetes is a chronic disease that is one of the most common causes of disease burden and death worldwide.1,2 It is becoming more common, with the International Diabetes Federation (IDF) projecting that the number of diabetic patients will increase dramatically to 591.9 million by the year 2035.3 It may impose significant costs on the healthcare system. In 2015, the overall cost of diabetes-related healthcare expenditures was US\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}${\$}$\end{document}1.31 trillion, or 1.8 percent of worldwide gross domestic product.4 Similarly, by 2030 this expenditure is expected to increase to US\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}${\$}$\end{document}2.2 trillion.5 In addition to well-documented risk factors, including age, unhealthy dietary patterns, physical inactivity, smoking and obesity,6 recent findings have suggested that work-related and environmental factors like noise, air pollution, shift work and electromagnetic fields7 may have an influence on the progression of diabetes. In contrast, environmental contaminants are well known to be linked to a variety of chronic diseases.8 Polycyclic aromatic hydrocarbons (PAHs) are lipid-soluble contaminants produced by cigarette smoking, incomplete biomass, fossil fuels combustion, preparation of grilled and smoked foods, industrial procedures and volcanoes and forest fires.9 PAHs are widespread pollutants that may be found in air, water, soil and sediments10 due to their physicochemical characteristics, such as high melting and boiling points and low vapor pressure.11 Although inhalation is the most common route of PAH exposure, PAHs can be inhaled, absorbed via the skin or consumed in work-related and environmental situations.12 Nonetheless, based on the features stated above, PAHs are among the top 10 compounds on the priority list of hazardous materials.12 PAHs are converted in the body to monohydroxylated metabolites of PAH (OH-PAHs), which are mostly excreted in the urine in the first few hours after exposure.13 Urinary OH-PAHs measurement is a useful biomarker for determining recent PAH exposure via multiple pathways.14 Exposure to PAHs is likely influenced by non-modifiable risk factors, such as age, sex and race, as well as occupational risks, active smoking and/or passive smoking exposure. Being overweight or obese is a risk factor for diabetes15 and highly lipid-soluble PAHs.16 However, individuals with a low body mass index (BMI) are unlikely to be exposed to PAHs differently than those with a high BMI. The liver and kidney predominantly process PAHs after being eaten, breathed or absorbed via the skin and subsequently eliminated in bile and urine.15 Furthermore, PAHs have been detectable in virtually all internal organs, particularly those with large quantities of adipose tissue.9 Individual risk factors and comorbidities can operate synergistically with the lipophilic characteristics of PAHs, fluctuating with exposure length, exposure route and concentration to increase the severity of effects on the human body.9,15 Because PAHs are retained in adipose tissue until they are evacuated by regular bladder and gastrointestinal activities,15 PAHs may be more persistent in people with a higher BMI, which might impact their chances of developing diabetes compared with people with a lower BMI. Previous research has linked ambient air pollution to diabetes incidence, diabetes-related hospitalization and diabetes-related deaths, including a link between permanent organic contaminants and diabetes prevalence.17,18 In addition, many studies have examined the impacts of PAHs on people. These studies found that PAH exposure increases the risk of diseases such as cancer, DNA damage, cardiovascular diseases and metabolic syndrome through a variety of pathways.19,20 Several studies have looked at the possible relationship between urinary PAH metabolites and diabetes.15,21–25 Despite this, an association remains uncertain due to conflicting findings across research. Therefore the primary aim of this study is to examine the association between urinary PAHs and diabetes mellitus (DM) in the US population. ## Study population The National Health and Nutritional Examination Survey (NHANES), a population-based survey, is a nationally representative study that was utilized to compile the data from 2003 to 2016 for the present study. It consists of a series of surveys created by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC) to continually monitor the health status of the non-institutionalized civilian population in the USA.26 The NHANES program has included a series of surveys focusing on different demographic groups or health issues since its beginning in the early 1960s. A cross-sectional study was planned to determine the degree of the association between urinary PAHs and DM. We used data from the 2003–2004, 2005–2006, 2007–2008, 2009–2010, 2011–2012, 2013–2014 and 2015–2016 (seven) data cycles for our study. In all, 71 067 participants were included in this study throughout seven cycles, with 45 978 of them being ≥18 y of age. The urinary PAH metabolites were only examined in an NHANES subsample ($$n = 13$$ 792). Participants who missed information on PAHs and were <18 y of age were omitted from the final design. Consequently, 13 792 participants of the NHANES 2003–2016 were included in the final analyses (Figure 1). Figure 1.: Eligible participants and those included in the analyses of the association between urinary PAHs and DM among the US population. On the day of the physical examination, all participants completed the questionnaires and underwent a basic physical examination as well as provided blood and urine samples. Trained professionals collected data on demographic parameters, employment history, personal and family medical history and lifestyle behaviours, such as smoking and alcohol consumption, using structured questionnaires. The NHANES procedure was approved by the NCHS Institutional Review Committee and signed informed consent forms were acquired. ## Data collection A structured medical condition questionnaire was administered for a wide array of health conditions, including DM, during the personal interview. ‘ Has a doctor or other health professional ever informed you that you have diabetes mellitus?’ If a participant replied ‘yes’, she/he was classified as a DM case.27 Each participant's morning urine sample was collected in a sterile tube and were preserved at −20°C until they were utilized. Six urinary PAH metabolites were frequently available and tested in NHANES 2003–2016, including 1-hydroxynaphthalene, 2-hydroxynaphthalene, 2-hydroxyfluorene, 3-hydroxyfluorene, 1-hydroxyphenanthrene and 1-hydroxypyrene, using enzymatic hydrolysis of urine followed by removal, derivation and investigation using capillary gas chromatography and high-resolution mass spectrometry (GC-HRMS). The same unit, ng/L, was used to measure all of the metabolites.28 The lowest point in the calibration curve that has been identified to generate a signal:noise ratio (S:N) ≥3 was defined as the limit of detection (LOD) for urinary PAH metabolites.20 A structured questionnaire administered during a home interview collected sociodemographic information such as age (years), gender (men/women), marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous work activity, moderate work activity, vigorous recreational activities, moderate recreational activities, smoking and drinking.29 Data on anthropometric, physical and laboratory parameters were gathered during the medical centre assessment. Height and weight were recorded without shoes and in light indoor clothing. BMI was calculated as kilograms per square meter (kg/m2). Using BMI cut-off points, participants were classified into three groups: <25, 25–29.9 and ≥30 kg/m2. The fasting serum lipid profile, comprising total cholesterol (TC), triglycerides (TG), low-density lipoprotein (LDL) and high-density lipoprotein (HDL) was determined using a biochemical blood analyser. ## Statistical analyses For normally distributed continuous variables, an analysis of variance was performed to investigate variations in participant characteristics by tertiles of total PAHs. The χ2 test was used to compare the frequencies of the categorical variables. Urinary PAH metabolites concentration (ng/L) was adjusted by the corresponding urinary creatinine concentration (mg/dL), divided, and then multiplied by 0.01. Total PAHs was divided into tertile 1, tertile 2 and tertile 3. Logistic regression analysis with confounder adjusted odds ratios (ORs) and $95\%$ confidence intervals (CIs) was performed to evaluate the association between urinary PAHs and the prevalence of diabetes. Model 1 was unadjusted, model 2 was adjusted for age (years) and gender (men/women) and model 3 was adjusted for model 2 plus marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous work activity, moderate work activity, vigorous recreational activities, moderate recreational activities, smoking, drinking and BMI. All analyses were performed using SPSS version 25.0 (IBM, Armonk, NY, USA). Statistical significance was defined as a two-tailed p-value <0.05. ## Characteristics of participants The study sample consisted of 13 792 individuals ≥18 y of age; 6866 were men and 6926 were women. The mean age of all participants was 42.72±19.08. In our study, we analysed the participants’ characteristics for three different tertiles; the average ages of the three PAH tertiles were 42.56±19.67, 42.21±19.51 and 43.39±17.99 y, as shown in Table 1. The age, gender, marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous recreational activities, moderate recreational activities, smoking, drinking, BMI, HDL, TC, TG and diabetes were significantly different ($p \leq 0.05$) between the PAH tertiles (Table 1). No significant differences in vigorous work activity, moderate work activity and LDL were observed across the tertiles of urinary PAH ($p \leq 0.05$). Table 1. \| Unnamed: 0 \| Unnamed: 1 \| Ʃ PAHs (ng/g creatinine)0.01 \| Ʃ PAHs (ng/g creatinine)0.01.1 \| Ʃ PAHs (ng/g creatinine)0.01.2 \| Unnamed: 5 \| Unnamed: 6 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Characteristics \| All participants (N=13 792) \| ≤0.52 (n=4604) \| 0.53–1.29 (n=4594) \| ≥1.30 (n=4594) \| Statistic values \| P-Value \| \| Age (years), mean ±SD \| 42.72±19.08 \| 42.56±19.67 \| 42.21±19.51 \| 43.39±17.99 \| 4.647a \| 0.01 \| \| Age group (years), n (%) \| \| \| \| \| \| \| \| ≥18–39 \| 7090 (51.4) \| 2451 (53.2) \| 2437 (53.0) \| 2202 (47.9) \| 76.299b \| <0.001 \| \| 40–59 \| 3371 (24.4) \| 1003 (21.8) \| 1040 (22.6) \| 1328 (28.9) \| \| \| \| 60–≥80 \| 3331 (24.2) \| 1150 (25.0) \| 1117 (24.3) \| 1064 (23.2) \| \| \| \| Gender, n (%) \| \| \| \| \| \| \| \| Male \| 6866 (49.8) \| 2622 (57.0) \| 2084 (45.4) \| 2160 (47.0) \| 144.553b \| <0.001 \| \| Female \| 6926 (50.2) \| 1982 (43.0) \| 2510 (54.6) \| 2434 (53.0) \| \| \| \| Marital status, n (%) \| \| \| \| \| \| \| \| Married \| 6323 (50.4) \| 2281 (54.7) \| 2125 (52.3) \| 1917 (44.4) \| 113.721b \| <0.001 \| \| Divorced \| 2644 (21.1) \| 738 (17.7) \| 830 (20.4) \| 1076 (24.9) \| \| \| \| Single \| 3590 (28.6) \| 1151 (27.6) \| 1110 (27.3) \| 1329 (30.7) \| \| \| \| Ethnicity, n (%) \| \| \| \| \| \| \| \| Mexican American \| 2356 (17.1) \| 688 (149) \| 964 (21.0) \| 704 (15.3) \| 121.562b \| <0.001 \| \| Other Hispanic \| 1236 (9.0) \| 361 (7.8) \| 445 (9.7) \| 430 (9.4) \| \| \| \| Non-Hispanic white \| 5750 (41.7) \| 2041 (44.3) \| 1704 (37.1) \| 2005 (43.6) \| \| \| \| Non-Hispanic Black \| 3051 (22.1) \| 1019 (22.1) \| 981 (21.4) \| 1051 (22.9) \| \| \| \| Other race including multiracial \| 1399 (10.1) \| 495 (10.8) \| 500 (10.9) \| 404 (8.8) \| \| \| \| Education level, n (%) \| \| \| \| \| \| \| \| <9th grade \| 1418 (11.6) \| 379 (9.5) \| 527 (13.3) \| 512 (12.0) \| 437.26b \| <0.001 \| \| 9th–11th grade \| 1758 (14.4) \| 430 (10.7) \| 514 (13.0) \| 814 (19.1) \| \| \| \| High school graduate \| 2867 (23.4) \| 828 (20.7) \| 900 (22.7) \| 1139 (26.7) \| \| \| \| College degree \| 3508 (28.7) \| 1175 (29.4) \| 1099 (27.7) \| 1234 (29.0) \| \| \| \| College and above \| 2670 (21.8) \| 1187 (29.7) \| 925 (23.3) \| 558 (13.1) \| \| \| \| Work type, n (%) \| \| \| \| \| \| \| \| Employee of a private company \| 5447 (74.9) \| 1790 (72.0) \| 1843 (74.6) \| 1814 (74.6) \| 40.688b \| <0.001 \| \| Federal government employee \| 180 (2.5) \| 78 (3.1) \| 66 (2.7) \| 36 (1.6) \| \| \| \| State government employee \| 444 (6.1) \| 184 (7.4) \| 148 (6.0) \| 112 (4.8) \| \| \| \| Local government employee \| 445 (6.1) \| 170 (6.8) \| 153 (6.2) \| 122 (5.3) \| \| \| \| Self-employed \| 695 (9.6) \| 243 (9.8) \| 243 (9.8) \| 209 (9.0) \| \| \| \| Working without pay in farming \| 24 (0.3) \| 8 (0.3) \| 6 (0.2) \| 10 (0.4) \| \| \| \| Housing type, n (%) \| \| \| \| \| \| \| \| Own \| 3710 (56.4) \| 1170 (62.0) \| 1322 (57.0) \| 1218 (51.4) \| 51.829b \| <0.001 \| \| Rent \| 2859 (43.5) \| 717 (38.0) \| 993 (42.8) \| 1149 (48.5) \| \| \| \| People in household, n (%) \| \| \| \| \| \| \| \| 1–2 \| 4757 (39.3) \| 1602 (41.4) \| 1501 (36.5) \| 1654 (40.3) \| 30.306b \| <0.001 \| \| 3–4 \| 4276 (35.4) \| 1376 (35.5) \| 1494 (36.4) \| 1406 (34.2) \| \| \| \| 5–6 \| 1455 (12.0) \| 442 (11.4) \| 531 (12.9) \| 482 (11.7) \| \| \| \| ≥7 \| 1601 (13.2) \| 454 (11.7) \| 582 (14.2) \| 35.3 (13.8) \| \| \| \| Vigorous work activity, n (%) \| \| \| \| \| \| \| \| Yes \| 1134 (19.0) \| 326 (18.7) \| 401 (18.9) \| 407 (19.5) \| 3.446b \| 0.751 \| \| No \| 4823 (80.9) \| 1420 (81.3) \| 1725 (81.1) \| 1678 (80.4) \| \| \| \| Moderate work activity, n (%) \| \| \| \| \| \| \| \| Yes \| 2111 (35.4) \| 594 (34.0) \| 793 (37.3) \| 724 (34.7) \| 9.226b \| 0.161 \| \| No \| 3844 (64.5) \| 1151 (65.9) \| 1333 (62.7) \| 1360 (65.2) \| \| \| \| Vigorous recreational activities, n (%) \| \| \| \| \| \| \| \| Yes \| 1438 (24.1) \| 466 (26.7) \| 513 (24.1) \| 459 (22.0) \| 13.131b \| 0.011 \| \| No \| 4521 (75.9) \| 1281 (73.3) \| 1614 (75.9) \| 1626 (77.9) \| \| \| \| Moderate recreational activities, n (%) \| \| \| \| \| \| \| \| Yes \| 2391 (40.1) \| 745 (42.6) \| 857 (40.3) \| 789 (37.8) \| 10.44b \| 0.034 \| \| No \| 3567 (59.8) \| 1001 (57.3) \| 1269 (59.7) \| 1297 (62.2) \| \| \| \| Smoking status, n (%) \| \| \| \| \| \| \| \| Yes \| 2253 (38.3) \| 121 (9.1) \| 375 (24.8) \| 1757 (58.1) \| 1100.979b \| <0.001 \| \| No \| 3622 (61.7) \| 1214 (90.9) \| 1140 (75.2) \| 1268 (41.9) \| \| \| \| Drinking status, n (%) \| \| \| \| \| \| \| \| Yes \| 8204 (71.7) \| 2627 (70.1) \| 2589 (69.4) \| 2988 (75.3) \| 46.541b \| <0.001 \| \| No \| 3234 (28.3) \| 1118 (29.8) \| 1141 (30.6) \| 975 (24.6) \| \| \| \| BMI (kg/m2), n (%) \| \| \| \| \| \| \| \| <25 \| 4734 (34.8) \| 1604 (35.3) \| 1529 (33.7) \| 1601 (35.3) \| 9.604b \| 0.048 \| \| 25–29.9 \| 4220 (31.0) \| 1454 (32.0) \| 1391 (30.7) \| 1375 (30.3) \| \| \| \| ≥30 \| 4666 (34.3) \| 1492 (32.8) \| 1615 (35.6) \| 1559 (34.4) \| \| \| \| Cholesterol level, mean±SD \| \| \| \| \| \| \| \| HDL (mg/dL) \| 54.03±15.42 \| 53.6915.14 \| 56.0216.02 \| 52.57±15.04 \| 6.159a \| 0.002 \| \| LDL (mg/dL) \| 111.90±5.37 \| 112.16±34.71 \| 110.44±35.08 \| 113.11±36.34 \| 2.812a \| 0.06 \| \| TC (mg/dL) \| 190.86±42.27 \| 190.43±42.16 \| 189.71±41.46 \| 192.44±43.14 \| 4.856a \| 0.008 \| \| TG (mg/dL) \| 129.0±125.63 \| 125.83±103.83 \| 123.34±133.48 \| 138.28±137.99 \| 7.959a \| <0.001 \| \| Diabetes, n (%) \| \| \| \| \| \| \| \| Yes \| 1308 (10.8) \| 395 (10.2) \| 444 (10.8) \| 469 (11.4) \| 11.933b \| 0.003 \| \| No \| 10781 (89.2) \| 3479 (89.8) \| 3664 (89.2) \| 3638 (88.6 \| \| \| ## Association of urinary PAHs with DM Table 2 shows the associations between urinary PAHs and the increased prevalence of diabetes. The second and third tertiles were significantly associated with an increased prevalence of diabetes. A positive association was found for the second and third tertiles of urinary PAHs and the prevalence of diabetes (OR 1.56 [$95\%$ CI 1.36 to 1.79] and OR 1.79 [$95\%$ CI 1.55 to 2.06], respectively, with $p \leq 0.05.$ Similarly, men and women had a significantly positive association between the second and third tertiles of urinary PAHs with DM (men: OR 1.42 [$95\%$ CI 1.18 to 1.71], OR 1.76 [$95\%$ CI 1.44 to 2.14] and women: OR 1.69 [$95\%$ CI 1.38 to 2.08], OR 1.79 [$95\%$ CI 1.46 to 2.19]; model 1; Table 2). Moreover, after adjustment for age (years) and gender (men/women), an increased prevalence risk for diabetes was observed in both (second and third) tertiles. Likewise, in both men and women, a positive association was observed in the third tertiles of urinary PAHs with a prevalence of diabetes; however, no significant relationship was found in the second tertiles of both sexes (model 2; Table 2). Furthermore, after adjustment for age (years), gender (men, women), marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous work activity, moderate work activity, vigorous recreational activities, moderate recreational activities, smoking, drinking and BMI, there was no association observed between urinary PAH tertiles and the prevalence of diabetes. Additionally, a significant relationship was found among men and women (model 3; Table 2). Table 2.* \| Unnamed: 0 \| Diabetes (n=1308) \| Diabetes (n=1308).1 \| Diabetes (n=1308).2 \| Diabetes (n=1308).3 \| Diabetes (n=1308).4 \| Diabetes (n=1308).5 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| PAH quartiles \| Model 1 \| Model 1 \| Model 2 \| Model 2 \| Model 3 \| Model 3 \| \| All participants \| OR (95% CI) \| P-Value \| OR (95% CI) \| P-Value \| OR (95% CI) \| P-Value \| \| Tertile 1 (≥0.52) (n=4604) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.53–1.29) (n=4594) \| 1.56 (1.36 to 1.79) \| <0.001 \| 1.56 (1.36 to 1.79) \| <0.001 \| 1.44 (0.83 to 2.52) \| 0.195 \| \| Tertile 3 (≥1.30) (n=4594) \| 1.79 (1.55 to 2.06) \| <0.001 \| 1.79 (1.55 to 2.06) \| <0.001 \| 1.28 (0.76 to 2.19) \| 0.356 \| \| Men \| \| \| \| \| \| \| \| Tertile 1 (≥0.52) (n=2622) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.53–1.29) (n=2084) \| 1.42 (1.18 to 1.71) \| <0.001 \| 1.15 (0.94 to 1.39) \| 0.168 \| 1.59 (0.85 to 2.97) \| 0.149 \| \| Tertile 3 (≥1.30) (n=2160) \| 1.76 (1.44 to 2.14) \| <0.001 \| 1.36 (1.12 to 1.66) \| 0.003 \| 1.79 (0.96 to 3.37) \| 0.068 \| \| Women \| \| \| \| \| \| \| \| Tertile 1 (≥0.52) (n=1982) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.53–1.29) (n=2510) \| 1.69 (1.38 to 2.08) \| <0.001 \| 1.20 (0.97 to 1.49) \| 0.091 \| 1.09 (0.30 to 3.98) \| 0.889 \| \| Tertile 3 (≥1.30) (n=2434) \| 1.79 (1.46 to 2.19) \| <0.001 \| 1.10 (0.88 to 1.37) \| 0.387 \| 0.61 (1.86 to 2.01) \| 0.419 \| ## Association of urinary PAH metabolites with DM Table 3 shows the association between six urinary PAH metabolites and the prevalence of diabetes. We used the logistic regression method to analyse the association between PAH metabolites and diabetes. The results indicated that the increased diabetes prevalence was observed across the second and third tertiles of 3-hydroxyfluorene (OR 1.28 [$95\%$ CI 1.12 to 1.47] and OR 1.32 [$95\%$ CI 1.14 to 1.51]) and the second and third tertiles of 1-hydroxypyrene (OR 1.56 [$95\%$ CI 1.36 to 1.79] and OR 1.79 [$95\%$ CI 1.55 to 2.06]), with $p \leq 0.05$ (model 1; Table 3). Similarly, after adjustment for age (years) and gender (men/women), a positive association was observed between the second and third tertiles of 3-hydroxfluorene, the second and third tertiles of 1-hydroxypyrene and both tertiles of 1-hydroxyphenanthrene with diabetes, with $p \leq 0.05$ (model 2; Table 3). Additionally, after adjustment for marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous work activity, moderate work activity, vigorous recreational activities, moderate recreational activities, smoking, drinking and BMI in model 2 (and model 3), we did not find a statistically significant association between the levels of each PAH metabolite and the prevalence of diabetes, with $p \leq 0.05.$ Table 3. \| Unnamed: 0 \| Diabetes (n=1308) \| Diabetes (n=1308).1 \| Diabetes (n=1308).2 \| Diabetes (n=1308).3 \| Diabetes (n=1308).4 \| Diabetes (n=1308).5 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| PAH biomarkers \| Model-1 \| Model-1 \| Model 2 \| Model 2 \| Model 3 \| Model 3 \| \| \| OR (95% CI) \| P-Value \| OR (95% CI) \| P-Value \| OR (95% CI) \| P-Value \| \| 1-Hydroxynaphthalene \| \| \| \| \| \| \| \| Tertile 1 (≤0.1) (n=4129) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.11–0.39) (n=3977) \| 0.85 (0.73 to 0.98) \| 0.024 \| 1.02 (0.89 to 1.18) \| 0.808 \| 0.72 (0.41 to 1.26) \| 0.244 \| \| Tertile 3 (≥0.4) (n=3979) \| 0.78 (0.68 to 0.90) \| 0.001 \| 0.99 (0.86 to 1.16) \| 0.986 \| 1.04 (0.61 to 1.78) \| 0.886 \| \| 2-Hydroxynaphthalene \| \| \| \| \| \| \| \| Tertile 1 (≤0.28) (n=3738) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.29–0.73) (n=4144) \| 0.93 (0.81 to 1.08) \| 0.353 \| 0.78 (0.68 to 0.91) \| 0.001 \| 1.14 (0.60 to 2.18) \| 0.681 \| \| Tertile 3 (≥0.74) (n=4203) \| 0.94 (0.81 to 1.08) \| 0.378 \| 0.75 (0.64 to 0.87) \| <0.001 \| 1.09 (0.59 to 2.01) \| 0.783 \| \| 3-Hydroxyfluorene \| \| \| \| \| \| \| \| Tertile 1 (≤0.01) (n=4022) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.01–0.02) (n=4079) \| 1.28 (1.12 to 1.47) \| <0.001 \| 1.23 (1.06 to 1.42) \| 0.005 \| 1.17 (0.63 to 2.14) \| 0.623 \| \| Tertile 3 (≥0.2) (n=3984) \| 1.32 (1.14 to 1.51) \| <0.001 \| 1.19 (1.03 to 1.37) \| 0.02 \| 1.02 (0.59 to 1.76) \| 0.939 \| \| 2-Hydroxyfluorene \| \| \| \| \| \| \| \| Tertile 1 (≤0.02) (n=4071) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.03–0.03) (n=4037) \| 1.11 (0.96 to 1.27) \| 0.152 \| 1.13 (0.98 to 1.31) \| 0.088 \| 1.03 (0.58 to 1.84) \| 0.91 \| \| Tertile 3 (≥0.4) (n=3977) \| 1.09 (0.95 to 1.26) \| 0.192 \| 1.08 (0.94 to 1.25) \| 0.261 \| 1.19 (0.69 to 2.05) \| 0.535 \| \| 1-Hydroxypyrene \| \| \| \| \| \| \| \| Tertile 1 (≤0.01) (n=3715) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.02–0.02) (n=4159) \| 1.56 (1.36 to 1.79) \| <0.001 \| 1.18 (1.02 to 1.36) \| 0.025 \| 1.45 (0.83 to 2354) \| 0.191 \| \| Tertile 3 (≥0.03) (n=4211) \| 1.79 (1.55 to 2.06) \| <0.001 \| 1.22 (1.05 to 1.42) \| 0.009 \| 1.30 (0.76 to 2.23) \| 0.331 \| \| 1-Hydroxyphenanthrene \| \| \| \| \| \| \| \| Tertile 1 (≤0.01) (n=4135) \| Reference \| \| Reference \| \| Reference \| \| \| Tertile 2 (0.01–0.02) (n=4028) \| 1.08 (0.94 to 1.24) \| 0.266 \| 1.18 (1.02 to 1.36) \| 0.029 \| 1.36 (0.80 to 2.31) \| 0.255 \| \| Tertile 3 (≥0.03) (n=3922) \| 1.07 (0.93 to 1.23) \| 0.324 \| 1.24 (1.07 to 1.44) \| 0.004 \| 0.84 (0.50 to 1.41) \| 0.516 \| ## Stratification analysis of urinary PAHs and the prevalence of DM We performed subgroup analysis stratified by age (groups), gender (men/women), marital status (married, divorced, single), housing type (owned/rented), vigorous work activity (yes/no), moderate work activity (yes/no), vigorous recreational activities (yes/no), moderate recreational activities (yes/no), smoking status (yes/no), drinking status (yes/no) and BMI. In the age group ≥18–39 y, the second and third tertiles of urinary PAHs were significantly associated with the prevalence of diabetes (OR 1.54 [$95\%$ CI 1.15 to 2.07] and 1.32 [$95\%$ CI 0.99 to 1.76]), respectively, with $p \leq 0.05.$ A positive association was also observed between the second and third tertiles of urinary PAHs and diabetes in both men and women (men: OR 1.42 [$95\%$ CI 1.18 to 1.71] and 1.76 [$95\%$ CI 1.44 to 2.14]; women: 1.69 ($95\%$ CI 1.38 to 2.08) and 1.79 ($95\%$ CI 1.46 to 2.19), with $p \leq 0.05$, respectively]; Table 4). Table 4. \| Unnamed: 0 \| PAH tertile 2(n=4594) \| PAH tertile 2(n=4594).1 \| PAH tertile 3(n=4594) \| PAH tertile 3(n=4594).1 \| \| --- \| --- \| --- \| --- \| --- \| \| Variables \| OR (95% CI) \| P-Value \| OR (95% CI) \| P-Value \| \| Age group (years) \| \| \| \| \| \| ≤18–39 \| 1.54 (1.15 to 2.07) \| 0.004 \| 1.32 (0.99 to 1.76) \| 0.051 \| \| 40–59 \| 1.11 (0.84 to 1.48) \| 0.449 \| 1.11 (0.85 to 1.45) \| 0.451 \| \| 60–≥80 \| 1.14 (0.93 to 1.39) \| 0.205 \| 1.48 (1.17 to 1.87) \| 0.001 \| \| Gender \| \| \| \| \| \| Male (n=6864) \| 1.42 (1.18 to 1.71) \| <0.001 \| 1.76 (1.44 to 2.14) \| <0.001 \| \| Female (n=6924) \| 1.69 (1.38 to 2.08) \| <0.001 \| 1.79 (1.46 to 2.19) \| <0.001 \| \| Marital status \| \| \| \| \| \| Married \| 1.36 (1.13 to 1.63) \| 0.001 \| 1.51 (1.24 to 1.84) \| 0.001 \| \| Divorced \| 1.49 (1.15 to 1.94) \| 0.003 \| 1.96 (1.50 to 2.55) \| <0.001 \| \| Single \| 1.64 (1.11 to 2.43) \| 0.013 \| 1.25 (0.88 to 1.77) \| 0.212 \| \| Housing type \| \| \| \| \| \| Own \| 1.55 (1.22 to 1.97) \| <0.001 \| 1.59 (1.24 to 2.04) \| <0.001 \| \| Rent \| 1.39 (1.02 to 2.74) \| 0.036 \| 2.00 (1.46 to 2.74) \| <0.001 \| \| Vigorous work activity \| \| \| \| \| \| Yes \| 1.29 (0.78 to 2.12) \| 0.312 \| 1.15 (0.72 to 1.84) \| 0.567 \| \| No \| 1.59 (1.28 to 1.98) \| <0.001 \| 2.06 (1.64 to 2.59) \| <0.001 \| \| Moderate work activity \| \| \| \| \| \| Yes \| 1.32 (0.93 to 1.88) \| 0.123 \| 1.40 (0.98 to 1.99) \| 0.061 \| \| No \| 1.67 (1.31 to 2.13) \| <0.001 \| 2.11 (1.64 to 2.72) \| <0.001 \| \| Vigorous recreational activities \| \| \| \| \| \| Yes \| 1.83 (1.15 to 2.89) \| 0.01 \| 1.49 (0.96 to 2.33) \| 0.077 \| \| No \| 1.48 (1.18 to 1.84) \| 0.001 \| 1.94 (1.54 to 2.45) \| <0.001 \| \| Moderate recreational activities \| \| \| \| \| \| Yes \| 1.48 (1.08 to 2.03) \| 0.014 \| 2.01 (1.43 to 2.81) \| <0.001 \| \| No \| 1.58 (1.22 to 2.05) \| <0.001 \| 1.77 (1.37 to 2.29) \| <0.001 \| \| Smoking status \| \| \| \| \| \| Yes \| 1.05 (0.58 to 1.90) \| 0.857 \| 1.16 (0.68 to 1.98) \| 0.593 \| \| No \| 1.48 (1.18 to 1.86) \| 0.001 \| 2.34 (1.83 to 2.96) \| <0.001 \| \| Drinking status \| \| \| \| \| \| Yes \| 1.41 (1.18 to 1.69) \| 0.001 \| 1.62 (1.35 to 1.94) \| 0.001 \| \| No \| 1.74 (1.37 to 2.22) \| <0.001 \| 1.86 (1.43 to 2.43) \| <0.001 \| \| BMI (kg/m2) \| \| \| \| \| \| <25 \| 1.97 (1.37 to 2.86) \| <0.001 \| 1.78 (1.27 to 2.51) \| 0.001 \| \| 25–29.9 \| 1.11 (0.85 to 1.44) \| 0.444 \| 1.33 (1.01 to 1.75) \| 0.043 \| \| ≥30 \| 1.61 (1.34 to 1.95) \| <0.001 \| 1.59 (1.31 to 1.94) \| <0.001 \| ## Discussion Our findings demonstrate that increased levels of urinary PAHs were positively associated with DM in the US general population. In both sexes, the second and third tertiles of PAH levels compared with the first tertile were strongly linked with an increased OR for DM. Moreover, after adjustment for age and gender, the higher tertiles of urinary PAH levels were associated with DM, but interestingly, statistical significance was found in the higher tertile of urinary PAHs in males only. To our knowledge this is the first large-scale nationwide population-based epidemiological study showing the association between urinary PAHs and DM. Our study also found a positive association of 3-hydroxyfluorene and 1-hydroxypyrene with DM, after adjustment for age and gender. Thus our findings support prior research linking diabetes with PAH exposure21 and other persistent organic pollutants.30 The present study examined the association of urinary PAH values with the prevalence of diabetes and found significance in both men and women, even after adjustment for age. Nevertheless, after adjustment for confounding factors, the relationship was not confirmed. The majority of the confounding factors adjusted for in this study are DM risk factors, thus further adjustment for these covariates is required and should be undertaken carefully. Moreover, the findings of this study suggest that smoking and drinking habits, work type, exercise types and BMI may all influence the link between PAHs and the prevalence of DM. Furthermore, the findings of this research should be confirmed in a cohort or longitudinal studies. According to our findings, a higher OR for diabetes was observed among the individuals exposed to 1-hydroxynaphthalene, 2-hydroxynaphthalene and 2-hydroxyfluorene metabolites, but we did not find significant results. However, an insignificant relationship was observed for 1-hydroxynaphthalene, 2-hydroxynaphthalene and 2-hydroxyfluorene metabolites. Although 1-hydroxynaphthalene is commonly used as an indicator of major metabolites of PAH exposure,31 the true diagnostic usefulness of urinary 1-hydroxynaphthalene in low PAH exposure from urban air pollution and associated diseases is yet unknown.32 A study examined the relationship between serum biomarkers of cardiovascular disease and urinary 1-hydroxynaphthalene levels but did not investigate any significant link between serum biomarkers of inflammation and urinary 1-hydroxynaphthalene levels.33 Furthermore, other research investigated the relationship between urinary PAH metabolites and metabolic syndrome in non-diabetic individuals and found no significant relationship between urinary1-hydroxynaphthalene levels and metabolic syndrome and its elements.34 However, several investigations have found a link between 1-hydroxynaphthalene and inflammation and oxidative stress indicators35 as well as cardiometabolic disorders.36,37 PAHs are typically ingested through the lungs and can be removed by bronchial clearance.38 Impaired mucociliary clearance of contaminated particles may increase particle penetration into bronchial epithelial cells, where PAHs are oxidized.38 The majority of PAHs are eliminated from the body after a few hours of exposure. However, tiny quantities are known to be stored in body fat and the liver, which might lead to PAH bioaccumulation over time.39 PAHs were converted in the body to OH-PAHs and mostly excreted in the urine a few hours after exposure.13 Due to the short time to excretion of PAHs from the body, this study could not find an association between smoking and exercise modes. In a subgroup analysis, an association was found among non-smokers only. In contrast, the Korean National Environmental Health Survey, a cross-sectional nationwide biomonitoring survey of 6478 participants ≥19 y of age, found a positive association with smokers.40 Our results indicated that participants who performed vigorous and moderate work activity had no association with those who did not. Moreover, individuals who performed vigorous and moderate recreational activities had positive significance, as did those who did not perform vigorous and moderate recreational activities. The implicit explanation can be respiratory absorption of airborne contaminants increases with increased ventilation and diffusion capacity in the lungs during aerobic activity.41 Although no direct evidence of increased PAH exposure during exercise has been found in human research, prior epidemiologic studies have shown that routine exercise in polluted outdoor regions or indoor locations, such as schools, can increase air pollution exposure.42,43 As a result, increased urinary PAH levels in those who engage in regular physical activity might be due to increased intake of outdoor and indoor PAHs during exercise in the way described above. To verify these findings, more cohort and experimental research is needed. BMI was examined as a possible impact modulator between PAHs and diabetes. Effect modification is a biological phenomenon in which the effects of the same exposure vary depending on the characteristics of research participants.44 Assessing statistical interaction and effect modification is important for determining if a given characteristic's effects are synergistic or antagonistic with exposure, as well as who would benefit most from a particular intervention.44,45 In this study, heterogeneity of effects was detected as evidenced by the positive association between the highest quintiles of PAHs and diabetes among individuals of normal weight (BMI <25 kg/m2), overweight (BMI 25–29.9 kg/m2) and obese (BMI ≥30 kg/m2). There were no effects among overweight people (BMI 25–29.9 kg/m2) in the second quartile of PAHs. Obesity and insulin resistance are two major risk factors for type 2 diabetes.46 While the mechanisms linking PAHs to diabetes are unknown, they have the potential to impact insulin resistance and promote obesity in a variety of ways. First, PAHs disrupt the endocrine system's function and damage the function of β cells.34 Second, animal experiments have demonstrated that oxidative stress affects glucose metabolism and insulin resistance, which is more severe in obese people.47 Third, animal studies have illustrated that PAH exposure causes weight gain by affecting adipose tissue lipolysis.48 Finally, higher PAH levels are linked to increased systematic inflammatory activity, leading to insulin resistance.9 These data show that PAH exposure is linked to obesity and insulin resistance, which could be the cause of type 2 diabetes. Our research has multiple significant advantages. First, our study was the first to focus on the association between urinary PAHs and diabetes, the sample was a multi-ethnic sample of the USA, the NHANES laboratory and data collecting procedures are of excellent quality and we included the capacity to account for confounders from a general community without occupational PAH exposure. Second, we estimated individual PAH exposure using urinary PAH metabolites, indicating PAH exposure from various sources. Third, this study examined the relationship between urinary PAHs and diabetes in the general population of the USA, taking into account possible confounders like age, gender, marital status, ethnicity, education, work type, housing type, number of people in the household, vigorous work activity, moderate work activity, vigorous recreational activities, moderate recreational activities, smoking, drinking and BMI. However, the study has several limitations. First, due to the cross-sectional study design of the NHANES, the causal inferences regarding the association between PAHs and the risk of DM cannot be proven. Second, the NHANES does not collect data on dietary exposure to PAHs, thus the proportion of dietary exposure to PAHs and the proportion of dietary PAHs in total exposure cannot be determined in those who eat grilled or charred meat regularly, especially high-fat meats.49 Although eating habits were assumed to be reasonably homogeneous for individuals in the same community, we did not investigate the confounding effects of dietary patterns. Third, because the NHANES does not gather data on the type of diabetes, we cannot discriminate between type 1 and type 2 diabetes. However, based on the demographic distribution of the two DM phenotypes, we estimate that the bulk of diabetic individuals in our sample have type 2 diabetes. It is also conceivable that DM may potentially produce greater PAH concentrations in the body due to impaired renal functions, reverse causation that we address but that cannot be ruled out in prevalence data. ## Conclusions In conclusion, this population-based cross-sectional study found a positive association between urinary PAHs and DM prevalence in the US population. ## Authors’ contributions: MAM was responsible for conceptualization, methodology, software, formal analysis, writing the original draft and visualization. TBB was responsible for software, formal analysis and reviewing and editing the manuscript. MA was responsible for the investigation and reviewing and editing the manuscript. FX, and XL responsible for software, formal analysis and reviewing and editing the manuscript. FF was responsible for conceptualization and reviewing and editing the manuscript. WW was responsible for validation and reviewing and editing the manuscript. PS and QZ was responsible for the methodology, software, supervision, funding acquisition and reviewing and editing the manuscript. ## Acknowledgments: The authors would like to express gratitude to the NHANES team for providing the publicly available data. The authors thank members of the NCHS of the CDC and the participants who enrolled in the National Health and Nutrition Examination Survey. All the authors also thank and acknowledge their respective Universities and Institutes. ## Funding This study was funded by the scientific research program of innovation platform in State Tobacco Monopoly Administration (grant number 312021AW0420), China, and the general project of Natural Science Foundation of Henan Province (grant number 222300420535). ## Competing interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Data availability: Data are publicly available at https://www.cdc.gov/nchs/nhanes/index.htm. ## References 1. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2018.0) 392 1789-858. PMID: 30496104 2. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2018.0) 392 1736-88. PMID: 30496103 3. Guariguata L, Whiting DR, Hambleton I. Global estimates of diabetes prevalence for 2013 and projections for 2035. Diabetes Res Clin Pract (2014.0) 103 137-49. PMID: 24630390 4. Bommer C, Heesemann E, Sagalova V. The global economic burden of diabetes in adults aged 20–79 years: a cost-of-illness study. Lancet Diabetes Endocrinol (2017.0) 5 423-30. PMID: 28456416 5. Bommer C, Sagalova V, Heesemann E. Global economic burden of diabetes in adults: projections from 2015 to 2030. Diabetes Care (2018.0) 41 963-70. PMID: 29475843 6. Wu Y, Ding Y, Tanaka Y. Risk factors contributing to type 2 diabetes and recent advances in the treatment and prevention. Int J Med Sci (2014.0) 11 1185-200. PMID: 25249787 7. Liu F, Chen G, Huo W. Associations between long-term exposure to ambient air pollution and risk of type 2 diabetes mellitus: a systematic review and meta-analysis. Environ Pollut (2019.0) 252 1235-45. PMID: 31252121 8. Alhamdow A, Lindh C, Albin M. Early markers of cardiovascular disease are associated with occupational exposure to polycyclic aromatic hydrocarbons. Sci Rep (2017.0) 7 9426. PMID: 28842704 9. Abdel-Shafy HI, Mansour MSM. A review on polycyclic aromatic hydrocarbons: source, environmental impact, effect on human health and remediation. Egypt J Petrol (2016.0) 25 107-23 10. Kim KH, Jahan SA, Kabir E. A review of airborne polycyclic aromatic hydrocarbons (PAHs) and their human health effects. Environ Int (2013.0) 60 71-80. PMID: 24013021 11. Bandeira GC, Meneses HE. Handbook of polycyclic aromatic hydrocarbons: chemistry, occurrence and health issues. (2013.0) 12. Rengarajan T, Rajendran P, Nandakumar N. Exposure to polycyclic aromatic hydrocarbons with special focus on cancer. Asian Pac J Trop Biomed (2015.0) 5 182-9 13. Gunier RB, Reynolds P, Hurley SE. Estimating exposure to polycyclic aromatic hydrocarbons: a comparison of survey, biological monitoring, and geographic information system-based methods. Cancer Epidemiol Biomarkers Prev (2006.0) 15 1376-81. PMID: 16835339 14. Castaño-Vinyals G, D'Errico A, Malats N. Biomarkers of exposure to polycyclic aromatic hydrocarbons from environmental air pollution. Occup Environ Med (2004.0) 61 e12. PMID: 15031403 15. Stallings-Smith S, Mease A, Johnson TM. Exploring the association between polycyclic aromatic hydrocarbons and diabetes among adults in the United States. Environ Res (2018.0) 166 588-94. PMID: 29982146 16. Kim YH, Lee YS, Lee DH. Polycyclic aromatic hydrocarbons are associated with insulin receptor substrate 2 methylation in adipose tissues of Korean women. Environ Res (2016.0) 150 47-51. PMID: 27236571 17. Zanobetti A, Dominici F, Wang Y. A national case-crossover analysis of the short-term effect of PM2.5 on hospitalizations and mortality in subjects with diabetes and neurological disorders. Environ Health (2014.0) 13 38. PMID: 24886318 18. Raaschou-Nielsen O, Sørensen M, Ketzel M. Long-term exposure to traffic-related air pollution and diabetes-associated mortality: a cohort study. Diabetologia (2013.0) 56 36-46. PMID: 22918192 19. Hu C. Association of polycyclic aromatic hydrocarbons exposure with atherosclerotic cardiovascular disease risk: A role of mean platelet volume or club cell secretory protein. Environ Pollut (2018.0) 233 45-53. PMID: 29053997 20. Alshaarawy O, Elbaz HA, Andrew ME. The association of urinary polycyclic aromatic hydrocarbon biomarkers and cardiovascular disease in the US population. Environ Int (2016.0) 89–90 174-8 21. Alshaarawy O, Zhu M, Ducatman AM. Urinary polycyclic aromatic hydrocarbon biomarkers and diabetes mellitus. Occup Environ Med (2014.0) 71 437-41. PMID: 24638887 22. Ranjbar M, Rotondi MA, Ardern CI. Urinary biomarkers of polycyclic aromatic hydrocarbons are associated with cardiometabolic health risk. PLoS One (2015.0) 10 e0137536. PMID: 26340343 23. Hou J, Sun H, Zhou Y. Environmental exposure to polycyclic aromatic hydrocarbons, kitchen ventilation, fractional exhaled nitric oxide, and risk of diabetes among Chinese females. Indoor Air (2018.0) 28 383-93. PMID: 29444361 24. Hou J, Sun H, Xiao L. Combined effect of urinary monohydroxylated polycyclic aromatic hydrocarbons and impaired lung function on diabetes. Environ Res (2016.0) 148 467-74. PMID: 27136672 25. Hou J, Sun H, Huang X. Exposure to polycyclic aromatic hydrocarbons and central obesity enhanced risk for diabetes among individuals with poor lung function. Chemosphere (2017.0) 185 1136-43. PMID: 28764134 26. NHANES survey methods and analytic guidelines. Atlanta: Centres for Disease Control and Prevention; 2016 27. NHANES 2015–2016 questionnaire instruments – medical conditions. Atlanta: Centres for Disease Control and Prevention; 2016 28. NHANES 2015–2016 laboratory methods. Atlanta: Centres for Disease Control and Prevention; 2016 29. NHANES 2015–2016 questionnaire data. Atlanta: Centres for Disease Control and Prevention; 2016 30. Lee DH, Lee I-K, Song K. A strong dose-response relation between serum concentrations of persistent organic pollutants and diabetes: results from the National Health and Examination Survey 1999–2002. Diabetes Care (2006.0) 29 1638-44. PMID: 16801591 31. Hansen AM, Mathiesen L, Pedersen M. Urinary 1-hydroxypyrene (1-HP) in environmental and occupational studies—a review. Int J Hyg Environ Health (2008.0) 211 471-503. PMID: 18222724 32. Jeng HA, Pan C-H, Preedy VR, Patel VB. 1-Hydroxypyrene as a biomarker for environmental health. General methods in biomarker research and their applications (2015.0) 595-612 33. Clark JD, Serdar B, Lee DJ. Exposure to polycyclic aromatic hydrocarbons and serum inflammatory markers of cardiovascular disease. Environ Res (2012.0) 117 132-7. PMID: 22626472 34. Hu H, Kan H, Kearney GD. Associations between exposure to polycyclic aromatic hydrocarbons and glucose homeostasis as well as metabolic syndrome in nondiabetic adults. Sci Total Environ (2015.0) 505 56-64. PMID: 25310882 35. Brucker N, Moro AM, Charão MF. Biomarkers of occupational exposure to air pollution, inflammation and oxidative damage in taxi drivers. Sci Total Environ (2013.0) 463–4 884-93 36. Lee MS, Magari S, Christiani DC. Cardiac autonomic dysfunction from occupational exposure to polycyclic aromatic hydrocarbons. Occup Environ Med (2011.0) 68 474-8. PMID: 21172795 37. Shiue I. Are urinary polyaromatic hydrocarbons associated with adult hypertension, heart attack, and cancer? USA NHANES, 2011–2012. Environ Sci Pollut Res Int (2015.0) 22 16962-8. PMID: 26111752 38. Lodovici M, Bigagli E. Oxidative stress and air pollution exposure. J Toxicol (2011.0) 2011 487074. PMID: 21860622 39. Jin X, Hua Q, Liu Y. Organ and tissue-specific distribution of selected polycyclic aromatic hydrocarbons (PAHs) in ApoE-KO mouse. Environ Pollut (2021.0) 286 117219. PMID: 33984772 40. Nam YJ, Kim S-H. Association of urinary polycyclic aromatic hydrocarbons and diabetes in Korean adults: data from the Korean National Environmental Health Survey Cycle 2 (2012–2014). Diabetes Metab Syndr Obes (2020.0) 13 3993-4003. PMID: 33149638 41. Künzli N. The public health relevance of air pollution abatement. Eur Respir J (2002.0) 20 198-209. PMID: 12166570 42. Braniš M, Safránek J, Hytychová A. Indoor and outdoor sources of size-resolved mass concentration of particulate matter in a school gym-implications for exposure of exercising children. Environ Sci Pollut Res Int (2011.0) 18 598-609. PMID: 20972889 43. Slezakova K, Peixoto C, Carmo Pereira MD. (Ultra) Fine particle concentrations and exposure in different indoor and outdoor microenvironments during physical exercising. J Toxicol Environ Health A (2019.0) 82 591-602. PMID: 31288676 44. Corraini P, Olsen M, Pedersen L. Effect modification, interaction and mediation: an overview of theoretical insights for clinical investigators. Clin Epidemiol (2017.0) 9 331-8. PMID: 28652815 45. Celentano DD, Szklo M. In memoriam: Leon Gordis, MD, DrPH (2015.0) 46. Styskal J, Van Remmen H, Richardson A. Oxidative stress and diabetes: what can we learn about insulin resistance from antioxidant mutant mouse models?. Free Radic Biol Med (2012.0) 52 46-58. PMID: 22056908 47. Tangvarasittichai S. Oxidative stress, insulin resistance, dyslipidemia and type 2 diabetes mellitus. World J Diabetes (2015.0) 6 456-80. PMID: 25897356 48. Irigaray P, Ogier V, Jacquenet S. Benzo[a]pyrene impairs beta-adrenergic stimulation of adipose tissue lipolysis and causes weight gain in mice. A novel molecular mechanism of toxicity for a common food pollutant. FEBS J (2006.0) 273 1362-72. PMID: 16689925 49. Lee J-G, Kim S-Y, Moon J-S. Effects of grilling procedures on levels of polycyclic aromatic hydrocarbons in grilled meats. Food Chem (2016.0) 199 632-8. PMID: 26776018
--- title: Regulatory T cells suppress the formation of potent KLRK1 and IL-7R expressing effector CD8 T cells by limiting IL-2 authors: - Oksana Tsyklauri - Tereza Chadimova - Veronika Niederlova - Jirina Kovarova - Juraj Michalik - Iva Malatova - Sarka Janusova - Olha Ivashchenko - Helene Rossez - Ales Drobek - Hana Vecerova - Virginie Galati - Marek Kovar - Ondrej Stepanek journal: eLife year: 2023 pmcid: PMC9977273 doi: 10.7554/eLife.79342 license: CC BY 4.0 --- # Regulatory T cells suppress the formation of potent KLRK1 and IL-7R expressing effector CD8 T cells by limiting IL-2 ## Abstract Regulatory T cells (Tregs) are indispensable for maintaining self-tolerance by suppressing conventional T cells. On the other hand, Tregs promote tumor growth by inhibiting anticancer immunity. In this study, we identified that Tregs increase the quorum of self-reactive CD8+ T cells required for the induction of experimental autoimmune diabetes in mice. Their major suppression mechanism is limiting available IL-2, an essential T-cell cytokine. Specifically, Tregs inhibit the formation of a previously uncharacterized subset of antigen-stimulated KLRK1+ IL-7R+ (KILR) CD8+ effector T cells, which are distinct from conventional effector CD8+ T cells. KILR CD8+ T cells show superior cell-killing abilities in vivo. The administration of agonistic IL-2 immunocomplexes phenocopies the absence of Tregs, i.e., it induces KILR CD8+ T cells, promotes autoimmunity, and enhances antitumor responses in mice. Counterparts of KILR CD8+ T cells were found in the human blood, revealing them as a potential target for immunotherapy. ## eLife digest As well as protecting us from invading pathogens, like bacteria or viruses, our immune system can also identify dangerous cells of our own that may cause the body harm, such as cancer cells. Once detected, a population of immune cells called cytotoxic T cells launch into action to kill the potentially harmful cell. However, sometimes the immune system makes mistakes and attacks healthy cells which it misidentifies as being dangerous, leading to autoimmune diseases. Special immune cells called T regulatory lymphocytes, or ‘Tregs’, can suppress the activity of cytotoxic T cells, preventing them from hurting the body’s own cells. While this can have a positive impact and reduce the effects of autoimmunity, Tregs can also make the immune system less responsive to cancer cells and allow tumors to grow. But how Tregs alter the behavior of cytotoxic T cells during autoimmune diseases and cancer is poorly understood. While multiple mechanisms have been proposed, none of these have been tested in living animal models of these diseases. To address this, Tsyklauri et al. studied Tregs in laboratory mice which had been modified to have autoimmune diabetes, which is when the body attacks the cells responsible for producing insulin. The experiments revealed that Tregs take up a critical signaling molecule called IL-2 which cytotoxic T cells need to survive and multiply. As a result, there is less IL-2 molecules available in the environment, inhibiting the cytotoxic T cells’ activity. Furthermore, if Tregs are absent and there is an excess of IL-2, this causes cytotoxic T cells to transition into a previously unknown subset of T cells with superior killing abilities. Tsyklauri et al. were able to replicate these findings in two different groups of laboratory mice which had been modified to have cancer. This suggests that Tregs suppress the immune response to cancer cells and prevent autoimmunity using the same mechanism. In the future, this work could help researchers to develop therapies that alter the behavior of cytotoxic T cells and/or Tregs to either counteract autoimmune diseases, or help the body fight off cancer. ## Introduction Physiological immune responses aim at invading pathogens, but not at healthy tissues. There are two major principles of controlling self-reactive lymphocytes: clonal deletion (central tolerance) and suppression (peripheral tolerance). FOXP3+ regulatory T cells (Tregs) represent a major force of peripheral tolerance as they regulate homeostasis and immune responses of conventional T cells (Josefowicz and Rudensky, 2009). The absence of Tregs in FOXP3-deficient individuals leads to a severe autoimmune condition called immunodysregulation polyendocrinopathy enteropathy X-linked syndrome (Bennett et al., 2001; Brunkow et al., 2001; Wildin et al., 2001). On the other hand, Tregs inhibit antitumor immune responses (Togashi et al., 2019). Although most studies have focused on how Tregs suppress conventional CD4+ T cells, CD8+ T cells are also regulated by Tregs. At steady state, Tregs suppress the proliferation of CD8+ T cells (Chinen et al., 2016) and prevent the spontaneous differentiation of memory CD8+ T cells into effector cells (Laidlaw et al., 2015; Kalia et al., 2015). Moreover, Tregs suppress antigenic responses of CD8+ T cells (McNally et al., 2011; Kastenmuller et al., 2011; Pace et al., 2012). Multiple mechanisms of Treg-mediated suppression of CD8+ T cells have been proposed, such as reducing the expression of co-stimulatory molecules on antigen-presenting cells (APC) via CTLA4 (Kalia et al., 2015; Kastenmuller et al., 2011; Schildknecht et al., 2010), limiting the availability of IL-2 (McNally et al., 2011; Kastenmuller et al., 2011), production of anti-inflammatory cytokines IL-10 (Laidlaw et al., 2015) and TGFβ (Green et al., 2003), and limiting the production of CCL3, CCL4, and CCL5 chemokines by APCs (Pace et al., 2012). However, none of these studies investigated the interaction between Tregs and CD8+ T cells in an autoimmune model. Thus, it is still unclear which of these mechanisms is the most important for preventing self-reactive CD8+ T cells from inducing an autoimmune disease. A single study addressed how Tregs suppress CD8+ T-cell responses with various affinities to the cognate antigen (Pace et al., 2012). The authors concluded that Tregs inhibit CD8+ T-cell activation by low-affinity, but not high-affinity, antigens. In such a case, Tregs would not correct errors of central tolerance by suppressing highly self-reactive CD8+ T cells that escape negative selection in the thymus. Their tolerogenic role would be limited to increasing the antigen-affinity threshold in the periphery (King et al., 2012) by suppressing positively selected CD8+ T cells with intermediate to low affinity to self-antigens. In this study, we focused on Treg-mediated peripheral tolerance, which prevents self-reactive CD8+ T cells from inducing an autoimmune pathology. We found that Tregs increase the number of self-reactive CD8+ T cells required to trigger experimental diabetes across a wide range of their affinities to self-antigens. In addition, we observed that excessive IL-2 prevents Tregs from suppressing CD8+ T cells in vitro as well as during autoimmune and antitumor responses in vivo. Moreover, Treg depletion, as well as high IL-2 levels, induces the expansion of an unusual population of potent cytotoxic KLRK1+ IL-7R+ CD8+ T cells. Altogether, our data provide strong evidence that the major mechanism of Treg suppression of CD8+ T cells is limiting the availability of IL-2. ## Tregs increase the quorum of self-reactive CD8+ T cells for inducing an autoimmune pathology To study the potential role of FOXP3+ Tregs in the autoimmune response of self-reactive T cells, we employed a model of experimental autoimmune diabetes based on the transfer of ovalbumin (OVA)-specific OT-I T cells into double-transgenic mice expressing ovalbumin in pancreatic β-cells (Kurts et al., 1998) and diphtheria toxin receptor (DTR) in FOXP3+ Tregs. Most of the experiments were performed using RIP.OVA DEREG mice carrying a random insertion of a transgene encoding for DTR-GFP fusion protein under the control of Foxp3 promoter (Lahl et al., 2007), but some experiments were performed using Foxp3DTR RIP.OVA strain with DTR-GFP knocked-in into the 3' untranslated region of the Foxp3 locus (Kim et al., 2007). Subsequent priming of OT-I T cells induced the formation of effector T cells and eventual destruction of pancreatic β-cells. Administration of diphtheria toxin (DT) selectively depleted Tregs uncovering their role in this autoimmune pathology. The major advantages of this model are (i) known etiology – the pathology is triggered by self-reactive CD8+ T cells, (ii) control of key parameters (number of transferred OT-I T cells, affinity of the priming antigen, presence of Tregs, etc.), and (iii) the bypass of central tolerance, allowing us to study the mechanisms of peripheral tolerance separately from the central tolerance. The induction of diabetes in the majority of Treg-replete mice required 106 adoptively transferred OT-I T cells followed by priming with OVA peptide and LPS (Figure 1A–C). This revealed efficient mechanisms of peripheral tolerance preventing the self-reactive T cells from destroying the insulin-producing cells. After the depletion of Tregs, the number of transferred OT-I T cells, sufficient to induce diabetes in the vast majority of animals, dropped dramatically (Figure 1B and C, Figure 1—figure supplement 1A). We observed that increased glucose levels in the urine and blood are preceded by abnormal morphology of the pancreatic islets and loss of insulin production together with the infiltration of CD8+ and CD4+ T cells in the islets on day 4 post immunization (Figure 1D and E). Accordingly, the expansion of OT-I T cells in the spleen was enhanced in Treg-deficient mice compared with Treg-replete mice (Figure 1F, Figure 1—figure supplement 1B and C). When 106 OT-I T cells were transferred, $50\%$ of Treg-depleted mice developed diabetes, even though they were primed with LPS alone without the cognate peptide (Figure 1G, Figure 1—figure supplement 1D). This was possible because the small amount of endogenous OVA was sufficient to prime OT-I T cells in the absence of Tregs, as revealed by more robust expansion and CD44+ CD62L- effector T-cell formation in Foxp3DTR RIP.OVA than in Foxp3DTR mice (Figure 1H, Figure 1—figure supplement 1E). Overall, these results revealed that Tregs represent an important mechanism of peripheral tolerance that prevents the induction of diabetes by self-reactive CD8+ T cells. Figure 1.: Depletion of Tregs decreases the quorum of self-reactive CD8+ T cells required for diabetes induction.(A) Scheme of RIP. OVA diabetes model. OT-I T cells were transferred into Foxp3DTR RIP.OVA or DEREG RIP.OVA mice (Treg-depleted with DT) and RIP.OVA controls (Treg-replete). The next day, mice were immunized with OVA peptide and LPS. Urine glucose levels were monitored on a daily basis for 14 days. (B–H) Diabetes was induced in RIP.OVA mice as described in (A). (B) Percentage of diabetic mice is shown. Number of diabetic mice and total number of mice per group is indicated on top of each column. Number of transferred OT-I T cells is indicated for each group. (C) Glucose concentration in blood on day 7 post-immunization is shown. RIP.OVA n = 19, DEREG RIP.OVA n = 9, Foxp3DTR RIP.OVA n = 9. One mouse from Foxp3DTR RIP.OVA group died before the measurement (shown as 'x'). (D, E) Histological analysis of pancreas tissue sections from C57Bl/6J (n = 2), RIP.OVA (n = 4), and DEREG RIP.OVA (n = 4) mice on a day 4 post immunization. Representative images are shown. Two independent experiments were performed. Scale bar 200 μm. (D) Hematoxylin and eosin staining. A pancreatic islets is in the center of each image. (E) Immunofluorescence staining of nuclei (DAPI) and indicated markers with antibodies. (F) Diabetes was induced in RIP.OVA (Ly5.1/Ly5.2, n = 4) and Foxp3DTR RIP.OVA (Ly5.1, n = 7) mice using 0.5 × 106 OT-I T cells. Spleens were collected on day 5 and analyzed by flow cytometry. Percentage of OT-I T cells in CD8+ T-cell population is shown. (G, H) RIP.OVA (Ly5.1/Ly5.2), Foxp3DTR RIP.OVA (Ly5.1), DEREG RIP.OVA (Ly5.1/Ly5.2), and Foxp3DTR (Ly5.1) mice were treated as shown in (A), with the exception that on day 0, mice were stimulated with LPS only (without OVA peptide). (G) Percentage of diabetic mice is shown. Number of diabetic mice and total number of mice per group is indicated on top of each column. Number of transferred OT-I T cells is indicated for each group. (H) Diabetes was induced using 0.5 × 106 OT-I T cells. Spleens were collected on day 5 and analyzed by flow cytometry. Left: percentage of OT-I T cells in CD8+ T-cell population is shown. RIP.OVA n = 5, Foxp3DTR RIP.OVA n = 9, Foxp3DTR n = 8. Right: percentage of effector cells defined as CD44+ CD62L- in OT-I T-cell population is shown. RIP.OVA n = 4, Foxp3DTR RIP.OVA n = 9, Foxp3DTR n = 8. Statistical significance was calculated by Kruskal–Wallis test (p-value is shown in italics) with Dunn’s post-test (<0.05, *<0.01) for comparison of three groups (C, F), or two-tailed Mann–Whitney test for comparison of two groups (D, p-value shown in italics). Median is shown. To investigate how Tregs suppress priming of self-reactive T cells by antigens with various affinities, we applied a priming protocol based on the adoptive transfer of bone marrow-derived dendritic cells (DCs) loaded with OVA peptide (KD ~ 50 μM) or its variants recognized by OT-I T cells with lower affinity; Q4R7 (KD ~ 300 μM) or Q4H7 (KD ~ 850 μM) (Stepanek et al., 2014). In contrast to OVA peptide and LPS priming, as few as 103 transferred OT-I T cells were able to induce autoimmune diabetes in RIP.OVA mice upon DC-OVA priming (Figure 2A). When 3 × 102 OT-I T cells were transferred, all Treg-depleted mice, but only one third of Treg-replete mice, were diabetic (Figure 2A). Accordingly, the absence of Tregs increased the susceptibility to diabetes upon priming with DC-Q4R7, although as many as 3 × 104 OT-I T cells were required to induce diabetes in most Treg-depleted hosts (Figure 2B). Along this line, Treg-replete RIP.OVA mice were resistant to 106 OT-I T cells primed with DC-Q4H7, whereas most Treg-depleted mice manifested diabetes (Figure 2C). The expansion of OT-I T cells, their expression of IL-2Rα (CD25; a subunit of the high-affinity IL-2 receptor) and KLRG1 (a marker of short-lived effector cells), and the absolute number of KLRG1+ OT-I T cells in the spleen were greater in Treg-depleted mice than in Treg-replete RIP.OVA mice on day 6 post-immunization with DC-OVA, -Q4R7, or -Q4H7 (Figure 2D–F, Figure 2—figure supplement 1A–C). Please note that different numbers of OT-I cells were transferred for immunizations with different peptides in these experiments. DCs not loaded with any peptide did not induce diabetes with one exceptional mouse (Figure 2—figure supplement 1D and E). When we titrated the amount of transferred OT-I T cells, we observed that the hierarchy of the biological potencies of high-, intermediate-, and low-affinity antigens was preserved in Treg-deficient and Treg-replete hosts (Figure 2—figure supplement 1F). Collectively, these results showed that Tregs increase the quorum of self-reactive CD8+ T cells required for inducing autoimmune diabetes, irrespective of the affinity of the priming antigen. Figure 2.: Both high-affinity and low-affinity self-reactive CD8+ T cells are controlled by Tregs.(A–C) Treg-depleted DEREG+ RIP.OVA mice and control DEREG- RIP.OVA mice received indicated numbers of OT-I T cells (103 or 3 × 102 OT-I T cells in A, 3 × 104 OT-I T cells in B, 106 OT-I T cells in C). The next day, mice were immunized with DC loaded with an indicated peptide (OVA in A, Q4R7 in B, Q4H7 in C). Urine glucose level was monitored on a daily basis for 14 days. (A) Left: percentage of diabetic mice is shown. Number of diabetic mice and total number of mice per group is indicated on top of each column. Middle: survival curve. Number of mice per group is indicated. Right: blood glucose concentration on day 7 post-immunization. (B, C) Left: survival curve. Number of mice per group is indicated. Right: blood glucose concentration on day 7 post-immunization. (D–F) Diabetes was induced in DEREG- RIP.OVA and DEREG+ RIP.OVA mice similarly to (A–C). On day 6, spleens were collected and analyzed by flow cytometry. Percentage of OT-I T cells among CD8+ T cells, count of KLRG1+ OT-I T cells, and count of IL-2Rα+ OT-I T cells are shown. (D) Diabetes was induced using DC-OVA and 103 OT-I T cells. Left: DEREG- n = 9, DEREG+ n = 10. Middle: DEREG- n = 7, DEREG+ n = 9. Right: n = 5 mice per group. (E) Diabetes was induced using DC-Q4R7 and 3 × 104 OT-I T cells. Left: DEREG- n = 13, DEREG+ n = 10. Middle: DEREG- n = 10, DEREG+ n = 9. Right: DEREG- n = 6, DEREG+ n = 7. (F) Diabetes was induced using DC-Q4H7 and 106 OT-I T cells, n = 8 mice per group. (G, H) RIP.OVA mice were treated or not with IL-2/JES6 for five consecutive days. Two days after the last dose, the mice received 103 OT-I T cells, and the next day they were immunized with DC-OVA. Urine glucose level was monitored on a daily basis for 14 days. (G) Experimental scheme. (H) Left: survival curve. Right: blood glucose concentration on day 7 post-immunization. n = 12 mice per group. Statistical significance was calculated by log-rank (Mantel–Cox) test (survival) or two-tailed Mann–Whitney test (glucose concentration and flow cytometry analysis). p-value is shown in italics. Median is shown. To elucidate, whether the physiological activity of Tregs is a limiting factor of self-tolerance, we selectively expanded the Treg compartment prior to the diabetes induction using IL-2 and anti-IL-2 mAb JES6-1A12 (JES6) immunocomplexes (IL-2ic) (Polhill et al., 2012; Liu et al., 2010; Figure 2G). We observed that the expanded Tregs prevented DC-OVA-induced diabetes in most animals (Figure 2H), showing that enhancing the overall Treg activity increases self-tolerance and prevents the induction of the experimental diabetes. ## Tregs suppress self-reactive CD8+ T cells in the absence of conventional CD4+ T cells Tregs could inhibit self-reactive CD8+ T cells directly or via the suppressing a bystander help of CD4+ T cells. To address this question, we used T-cell-deficient Cd3e-/- RIP.OVA mice as hosts. First, we adoptively transferred 106 polyclonal OVA-tolerant CD8+ T cells isolated from RIP.OVA mice to replenish the CD8+ T-cell compartment in the Cd3e-/- RIP.OVA mice. One week later, we transferred conventional (GFP-) or Treg (GFP+) CD4+ T cells from DEREG RIP.OVA mice. A control group of mice did not receive any CD4+ T cells. Finally, we transferred OT-I T cells and subsequently primed them with DC-OVA (Figure 3A). We observed that conventional CD4+ T cells, albeit presumably tolerant to OVA, increased the incidence of experimental diabetes and accelerated its onset (Figure 3B and C, Figure 3—figure supplement 1A). In contrast, Tregs reduced the incidence of diabetes compared with mice receiving no CD4+ T cells (Figure 3B and C, Figure 3—figure supplement 1A). Conventional CD4+ T cells enhanced the expansion of OT-I T cells and increased the number of KLRG1+ OT-I T cells, whereas Tregs elicited the opposite effect (Figure 3D and E, Figure 3—figure supplement 1B and C), as revealed on day 8 post-immunization. The role of Tregs was even more apparent on day 14 post-immunization, which corresponded to the typical onset of diabetes in the absence of conventional CD4+ T cells (Figure 3F and G, Figure 3—figure supplement 1D and E). Figure 3.: Tregs suppress self-reactive CD8+ T cells in the absence of conventional CD4+ T cells.Cd3e-/- RIP.OVA mice received 106 polyclonal CD8+ T cells from Ly5.1 RIP.OVA mice. After 7 days, they received either 106 conventional CD4+ T cells (GFP-), or 0.4–1 × 106 Tregs (GFP+) from DEREG+ RIP.OVA mice, or no CD4+ T cell transfer (NT). Next, 250 OT-I T cells were adoptively transferred to all the recipients. Next day, mice were immunized with DC-OVA. (A) Experimental scheme. (B, C) Urine glucose level was monitored on a daily basis until day 21 post-immunization. (B) Survival curve. Number of mice is indicated. (C) Blood glucose concentration on day 8 and day 14 post-immunization is shown. Day 8: NT n = 28, Conv.CD4+ n = 25, Tregs n = 15. Day 14: NT n = 15, Tregs n = 8. (D–G) On day 8 or day 14, spleens were collected and analyzed by flow cytometry. Percentage of OT-I T cells among CD8+ T cells (D, F), and count of KLRG1+ OT-I T cells (E, G) are shown. Day 8: NT n = 25, Conv.CD4+ n = 15, Tregs n = 12. Day 14: NT n = 15, Tregs n = 7. Statistical significance was calculated by Kruskal–Wallis test (p-value is shown in italics) with Dunn’s post-test (<0.05, <0.01, <0.001) for comparison of three groups, or two-tailed Mann–Whitney test for comparison of two groups (p-value is shown in italics). Median is shown. Overall, these experiments showed that conventional CD4+ T cells provide a bystander help to self-reactive CD8+ T cells, whereas Tregs directly suppress the priming of CD8+ T cells. ## Tregs suppress self-reactive CD8+ T cells via limiting IL-2 One of the Treg-mediated effects observed in the previous experiments was the downregulation of IL-2Rα in CD8+ T cells. IL-2Rα expression on T cells can be induced by antigen stimulation or by IL-2 itself in a positive feedback loop (Sereti et al., 2000). Accordingly, downregulation of IL-2Rα in OT-I T cells in the presence of Tregs might be a consequence of the limited availability of IL-2. We observed significantly higher levels of IL-2Rα and an IL-2 signaling intermediate, pSTAT5 in OT-I T cells in Treg-deficient mice early after priming (Figure 4A and B, Figure 4—figure supplement 1A and B). These results were reproduced when congenic Ly5.1 OT-I T cells were used (Figure 4—figure supplement 1C–E). These results showed that Tregs limit IL-2 signaling in self-reactive CD8+ T cells in vivo. Figure 4.: Tregs maintain tolerance of CD8+ T cells via limiting IL-2.(A, B) OT-I T cells (5 × 104) were transferred into Treg-depleted DEREG+ RIP.OVA mice and control DEREG- RIP.OVA mice. The next day, mice were immunized with DC-OVA. On day 3 post-immunization, spleens were collected and analyzed by flow cytometry. OT-I T cells were identified as OVA-tetramer and TCRVα2 double positive. (A) IL-2Rα expression on OT-I T cells. Left: a representative experiment out of three in total. Right: n = 5 mice per group. Median is shown. (B) pSTAT5 expression in OT-I T cells. Left: a representative experiment out of three in total. Right: geometric mean fluorescence intensity (MFI) of anti-pSTAT5-Alexa Fluor 647 in OT-I T cells. Mean of MFI values for each group per experiment are shown as a gray dots. Lines connect data from corresponding experiments. DEREG- n = 7, DEREG+ n = 6. (C) In vitro proliferation assay. CTV-labeled CD8+ T lymphocytes were stimulated with plate-bounded anti-CD3ɛ-antibody in the presence or absence of Tregs and/or recombinant IL-2. After 72 hr, cells were analyzed by flow cytometry. Left: proliferation is indicated by the CTV dilution. Right: IL-2Rα expression on CD8+ T cells. A representative experiment out of four in total. (D–F) Treg-depleted DEREG+ RIP.OVA mice and control DEREG- RIP.OVA mice received OT-I T cells, followed by the immunization with OVA peptide and LPS. On days 0, 1, and 2 post-immunization, RIP.OVA mice received IL-2ic (IL-2/S4B6 or IL-2/JES6). (D) Experimental scheme. (E) Urine glucose level was monitored on a daily basis for 14 days. Percentage of diabetes-free mice is shown. Number of mice is indicated. Four mice from IL-2/JES6 group died before day 5 and were excluded from the analysis. (F) Blood glucose concentration on day 7 post-immunization. Number of mice is indicated in (E). (G, H) Diabetes was induced in Treg-depleted DEREG+ RIP.OVA mice (with DT) or Treg-replete DEREG+ RIP.OVA mice (without DT), which later received IL-2ic (IL-2/S4B6 or IL-2/JES6), or were left untreated (control) (scheme D). On day 5 post-immunization, spleens were collected and analyzed by flow cytometry. Four mice from IL-2/JES6 group died before the analysis (shown as 'x'). Control n = 6, IL-2/S4B6 n = 6, IL-2/JES6 n = 2, DEREG+ DT n=4. (G) Number of KLRG1+ OT-I T cells. Median is shown. (H) Number of Tregs (defined as GFP+ CD4+ T cells). Median is shown. Statistical significance was calculated by two-tailed Mann–Whitney test (comparison of two groups), Kruskal–Wallis test (comparison of four groups), or log-rank (Mantel–Cox) test (survival), p-value is shown in italics. Tregs efficiently suppressed proliferation and IL-2Rα expression in CD8+ T cells ex vivo (Figure 4C). An addition of excessive IL-2 abolished this suppression (Figure 4C). To address whether Tregs can suppress self-reactive CD8+ T cells in the excess of IL-2 in vivo, we performed the diabetic assay with IL-2ic provided during OT-I priming (Figure 4D). IL-2ic have markedly increased biological activity in comparison to IL-2 alone (Boyman et al., 2006). We used IL-2ic with JES6 antibody clone, which selectively promotes IL-2 signaling in T cells expressing high-affinity IL-2Rαβγ trimeric IL-2 receptor, or with the antibody clone S4B6, which promotes IL-2 signaling irrespective of IL-2Rα expression (Boyman et al., 2006; Spangler et al., 2015). Both clones of IL-2ic broke the peripheral tolerance and induced the onset of autoimmune diabetes similarly to the Treg-depletion (Figure 4E and F). IL-2ic promoted the formation of KLRG1+ effector OT-I T cells (Figure 4G), despite a dramatic IL-2-dependent expansion of Tregs at the same time (Figure 4H). Overall, these results showed that Tregs are unable to suppress CD8+ T cells in the excess of IL-2 signal, although Tregs themselves expand dramatically, which should augment their overall suppressive capacity with the exception of the limiting IL-2 mechanism. This supports the hypothesis that the major mechanism of Treg-mediated suppression of self-reactive CD8+ T cells is the reduction of IL-2 availability. This IL-2-dependent mechanism might be specific for the suppression of CD8+ T cells since antigen-activated CD8+ T cells were much more responsive to IL-2 than activated CD4+ T cells (Figure 4—figure supplement 1F). ## Tregs suppress formation of a unique subset of CD8+ effector T cells To study how Treg-mediated suppression alters the gene expression profiles of activated CD8+ T cells, we analyzed the transcriptomes of OT-I T cells primed in the presence or absence of Tregs on a single-cell level. IL-2-responsive genes were upregulated in OT-I T cells primed in the Treg-depleted mice, including those encoding the key cytotoxic molecules granzyme B and perforin, as well as IL-2Rα (Figure 5A, Figure 5—figure supplement 1A and B). Unsupervised clustering highlighted five distinct activated OT-I T cell subsets (Figure 5B). Based on their gene expression profiles (Figure 5—figure supplement 1C, Supplementary file 1) and comparison with CD8+ T-cell subsets formed during an antiviral response (Figure 5—figure supplement 1D), we identified clusters 0 and 1 as early effector T cells, cluster 2 as memory T cells precursors, and cluster 3 as precursors of exhausted T cells. Cluster 4, on the other hand, did not match to any established CD8+ T-cell subset. It was characterized by the expression of a natural killer (NK) cell activation receptor gene Klrk1 (alias Nkg2d), Gzmb (Figure 5C, Figure 5—figure supplement 1C) and by the overall similarity to the NK cell gene expression profile (Figure 5—figure supplement 1E). Interestingly, these cells occurred almost exclusively in the absence of Tregs (Figure 5D and E). This cluster could be further divided into two subsets. The larger of them expressed Il7r, Cd103, and NK cell markers Ifitm1-3 and Cd7, but not a typical effector T-cell marker Cd49d (Itga4), which encodes a subunit of the integrin VLA4 (Figure 5F–H, Figure 5—figure supplement 1C). We refer to these cells as KLRK1+ IL-7R+ (KILR) effector CD8+ T cells in this study, reflecting their co-expression of typical cytotoxic effector genes (Gzmb, Prf1) together with markers of NK cells (Klrk1) or T-cell memory (Il7r) genes. In the next step, we validated the scRNAseq results using flow cytometry. This confirmed that KLRK1+ IL-7R+ CD49d- CD8+ T cells are ~20-fold more frequent in the absence of Tregs during the priming than in Treg-replete conditions (Figure 5I). This experiment also confirmed that part of these KLRK1+ IL-7R+ CD49d- cells expressed CD103 (Figure 5—figure supplement 1F). Moreover, the gene expression profile of FACS-sorted populations of OT-I T cells primed in the presence or absence of Tregs was consistent with the original scRNAseq data (Figure 5—figure supplement 1G). Figure 5.: Tregs block the formation of KLRK1+ IL-7R+ cytotoxic T cells.(A–E) Ly5.1 OT-I T cells were adoptively transferred into Treg-depleted DEREG+ RIP.OVA and control DEREG- RIP.OVA mice (n = 3 mice per group). The next day, mice were immunized with DC-OVA. On day 3 post-immunization, spleens were isolated and OT-I T cells were FACS sorted as Ly5.1+ CD8+ cells, and analyzed via scRNAseq. (A) A heat map showing the relative expression of canonical IL-2-responsive genes. Each column represents one mouse. (B–D) UMAP projection of the individual OT-I T cells based on their gene expression profile. (B) The colors indicate individual clusters revealed by unsupervised clustering. (C) The intensity of the blue color indicates the level of Klrk1 expression in individual cells. (D) The origin of the cells (DEREG+ or DEREG- mice) is indicated. (E) The percentage of cells assigned to specific clusters is shown for individual mice. Statistical significance was calculated by unpaired Student’s t-test with Bonferroni correction, p-value is shown in italics. Median is shown. (F–H) Cluster 4 was reanalyzed separately. (F) UMAP projection showing two subclusters identified by unsupervised clustering. (G) The intensity of the blue color indicates the level of expression of Itga4 or Il7r in individual cells. (H) Projection of subclusters 0 and 1 on the original UMAP plot. (I) Ly5.1 OT-I T cells (5× 104) were adoptively transferred into Treg-depleted DEREG+ RIP.OVA mice (n = 7) or DEREG- RIP.OVA mice (n = 11). The next day, mice were immunized with DC-OVA. On day 3 post-immunization, spleens were collected and analyzed by flow cytometry. Left: a representative experiment out of three in total is shown. KLRK1+ subset of Ly5.1 OT-I T cells was divided into two gates based on their expression of CD49d and IL-7R. Percentage of KILR effector T cells, defined as KLRK1+ IL-7R+ CD49d- cells out of OT-I T cells is shown. Statistical significance was calculated by two-tailed Mann–Whitney test, p-value shown in italics. Median is shown. ## KILR effector CD8+ T cells have strong cytotoxic properties In the next step, we tested whether Treg depletion induces KILR effector CD8+ T cells in polyclonal mice. Although Treg deficiency increased the frequency of KILR effector CD8+ T cells, they were present at low numbers (Figure 6A, Figure 6—figure supplement 1A and B). We hypothesized that only the combination of antigenic and IL-2 signals can effectively induce KILR effector CD8+ T cells. To address this hypothesis, we directly treated the OT-I Rag2-/- mice with OVA and/or IL-2ic. Indeed, the combination of IL-2 administration and antigenic stimulation efficiently induced KILR effector CD8+ T cells in OT-I mice (Figure 6B–E), which were characterized by high GZMB and IL-7R expression, whereas the stimulation with IL-2 or antigen alone failed to induce the complete KILR phenotype. The combination of IL-2ic and OVA also induced the expression of IL-2Rα (Figure 6—figure supplement 1C). Figure 6.: KILR CD8+ T cells induced by the stimulation with the cognate antigen and high levels of IL-2 show superior cytotoxicity.(A) DEREG+ RIP.OVA mice were treated with DT in order to deplete Tregs on days 0 and 1. On day 3, spleens were collected and analyzed by flow cytometry. Percentage of KILR effector T cells defined as KLRK1+ CD49d- IL-7R+ among CD8+ T cells is shown. Untreated n = 6, DT n = 7. Median is shown. (B–E) OT-I Rag2-/- mice were treated with OVA peptide (single dose on day 0, n = 7) and/or IL-2/S4B6 (three doses, days 0, 1 and 2, n = 7 mice per group) or left untreated (n = 6). Spleens were collected and analyzed by flow cytometry on day 3. (B) Four populations of OT-I T cells were identified based on their expression of KLRK1 and CD49d. A representative experiment out of three in total. (C) Top: IL-7R expression on KLRK1+ CD49d- OT-I T cells from mice treated with OVA and/or IL-2/S4B6. Expression of IL-7R on naïve OT-I T cells is shown as a positive control. Representative staining. Bottom: number of KILR effector T cells, defined as KLRK1+ CD49d- IL-7R+ OT-I T cells. Median is shown. (D) IL-7R expression in CD49d+ KLRK1-, CD49d+ KLRK1+, CD49d- KLRK1+, and CD49d- KLRK1- OT-I T cells from mice treated with OVA + IL-2/S4B6. Top: a representative histogram. Bottom: percentage of IL-7R+ cells among OT-I T cells in indicated populations. Median is shown. (E) GZMB levels in OT-I T cells. Top: a representative histogram. Bottom: geometric mean fluorescence intensity (MFI) of anti-GZMB-eFluor 660 antibody on OT-I T cells. Obtained values were normalized to the average of MFI of untreated samples in each experiment (=1). Median is shown. (F–G) OT-I Rag2-/- and Ly5.1 OT-I Rag2-/- mice were treated with OVA peptide (day 0) and IL-2/S4B6 (three doses, days 0–2). On day 3, KILR (KLRK1+ CD49d-), naïve (KLRK1- CD49d-), and effector cells (KLRK1- CD49d+) were sorted. Recipient Cd3e-/- mice received a mix of KILR and naïve (n = 9), or KILR and effector (n = 11) congenic cells (1:1 ratio, 400 × 103 or 500 × 103 cells in total). On day 7, spleens of the recipient mice were analyzed by flow cytometry. Two independent experiments were performed. (F) Percentage of cells that kept their initial phenotype was determined. Top: cells sorted as KILR (magenta), naïve (blue), and effector cells (orange) fall into corresponding gates. Representative dot plot. Bottom: percentage of cells that fall into the KILR, effector, and naïve gate, after adoptive co-transfer. Median is shown. (G) Ratio of KILR cells to co-transferred control cells. Median is shown. (H) OT-I Rag2-/- mice were treated with OVA peptide (day 0), and/or IL-2/S4B6 (days 0, 1, and 2) or left untreated. Spleens were collected on day 3. KLRK1+ CD8+ or KLRK1- CD8+ cells were sorted and adoptively transferred into recipient RIP.OVA mice, which have received a mixture of target OVA-pulsed CTV-loaded and unpulsed CFSE-loaded splenocytes from Ly5.1 mice at ~1:1 ratio earlier on the same day. The next day, the spleens were analyzed for the presence of Ly5.1 donor cells by flow cytometry. Ratio of unpulsed (CFSE+) to OVA pulsed (CTV+) target cells was determined and normalized to control recipients which did not receive OT-I T cells (=1). KLRK1- (Untreated) n = 6, KLRK1- (OVA + IL-2/JES6) n=11, KLRK1+ (OVA + IL-2/JES6) n = 13, KLRK1+ (OVA peptide) n = 6. Four independent experiments. Median is shown. (I–L) Human CD8+ T-cell atlas was generated by integrating 14 scRNAseq data sets from blood of healthy donors. The gene expression data after the removal of MAIT cells were projected into a 2D UMAP plot. (I) The assignment of individual cells to clusters identified by unsupervised clustering. Individual clusters were matched to established CD8+ T cell subsets based on the expression of their signature markers (see Figure 6—figure supplement 1H). (J) The intensity of the blue color corresponds to the level of expression of indicated genes in individual cells. (K) The percentage of CD8+ T cells assigned to the KILR-like CD8+ T-cell cluster in individual donors (n = 14). Median is shown. (L) Left: clonally expanded T cells were identified based on their TCRαβ VDJ sequences. The intensity of the red color indicates the size of individual clones. Right: T cells with recovered complete TCRαβ VDJ information are shown in green. Statistical significance was calculated by two-tailed Mann–Whitney test (A, G), Wilcoxon matched-pairs signed rank test (F), or Kruskal–Wallis test (C, D, E, H) for multiple groups comparison (p-value is shown in italics) with Dunn’s post-test (<0.05, <0.01, <0.001). We addressed the stability of the KILR phenotype by co-transferring congenically marked KILR (CD49d- KLRK1+) and naïve (CD49d- KLRK1-), or KILR and effector (CD49d+ KLRK1-) OT-I T cells at 1:1 ratio into T-cell-deficient Cd3e-/- mice and analyzed the counts and phenotypes of these cells after 3 days (Figure 6—figure supplement 1D). We observed that all three subsets maintained the expression of their signature markers (Figure 6F), documenting the stability of the KILR phenotype. KILR cells slightly outcompeted effector T cells (Figure 6G, Figure 6—figure supplement 1E). This can be explained by higher frequency of apoptotic cells together with slightly reduced proliferation in effector than in KILR cells (Figure 6—figure supplement 1F and G). To assess the cytotoxic properties of these cells, we compared the ability of KLRK1+ and conventional KLRK1- effector T cells from OT-I mice treated with OVA + IL-2ic, and KLRK1+ T cells from mice treated only with OVA to kill splenocytes loaded with their cognate antigen in vivo (Figure 6—figure supplement 1H). KLRK1+ T cells induced by the combination of the antigen and IL-2ic showed the most potent cytotoxic activity on per cell basis (Figure 6H, Figure 6—figure supplement 1I and J). To identify putative human counterparts of KILR effector CD8+ T cells, we generated a human blood CD8+ T-cell atlas by integrating publicly available single cell transcriptomic datasets from healthy donors, because CD8+ T cell data from Treg-deficient patients are not available. After removing MAIT cells (Figure 6—figure supplement 1K, cluster 6), we identified naïve, memory, and effector T cells, and the population expressing some KILR effector CD8+ T-cell signature genes such as NK markers (KLRD1, IFITM3, CD7) and IL7R (Figure 6I and J, Figure 6—figure supplement 1L, Supplementary file 2). Similarly to mouse KILR effector CD8+ T cells, this human T-cell subset was enriched for NK signature genes (Figure 6—figure supplement 1M). Human KILR-like T cells constituted for ~1–$10\%$ of all CD8+ T cells (Figure 6K). These cells showed no signs of clonal expansion, but expressed rearranged αβTCR genes, ruling out the possibility that these cells were NK cells or another non-T cell subset (Figure 6L). Because these cells expressed high levels of IL2RB (Figure 6J), a subunit of IL-2 and IL-15 receptors, their gene expression profile was probably modulated by strong IL-2/IL-15 signals, which is in line with the origin of mouse KILR effector T cells. Although human KILR-like T cells expressed cytotoxic genes GNLY and GZMK, the expression of GZMA and GZMB was lower in these cells than in the conventional effector cells (Figure 6J). This probably reflects the fact that these T cells have not been stimulated by IL-2 together with their cognate antigen recently, which is required for the formation of the full KILR phenotype in mice (Figure 6B–E and H). Nevertheless, this previously uncharacterized population of human CD8+ T cells shows apparent similarities to mouse KILR effector CD8+ T cells induced by supra-physiological IL-2 levels. ## Strong IL-2 signal promotes anti-tumor CD8+ T-cell responses So far, we have documented that Tregs suppress self-reactive CD8+ T cells by limiting IL-2 signal in the context of the autoimmune pathology. We hypothesized that excessive IL-2 signaling may override Treg-mediated suppression of tumor-reactive CD8+ T cells as well. Therefore, we investigated the effect of IL-2/JES6 immunocomplexes which were previously considered tolerogenic, in a BCL1 leukemia model (Figure 7A). This IL-2ic slowed down the progression of the disease, when co-administrated with a chemotherapeutic drug doxorubicin (Dox) (Figure 7B, Figure 7—figure supplement 1A). This effect was dependent on CD8+ but not on CD4+ T cells, as revealed by antibody-mediated depletion of these subsets (Figure 7B). Depletion of CD4+ T cells even more improved the anti-tumor effect of Dox plus IL-2ic combinational treatment, presumably due to the depletion of Tregs, which could still limit IL-2 availability between the IL-2ic injections. The administration of IL-2ic without the chemotherapy showed no therapeutic effect (Figure 7B, Figure 7—figure supplement 1A), suggesting a synergy between the IL-2 signals and the presentation of antigens released from tumor cells undergoing an immunogenic cell death (Obeid et al., 2007). Figure 7.: A combined treatment of IL-2/JES6 immunocomplexes and chemotherapy hampers the tumor growth and induces KILR CD8+ T cells.(A–C) On day 0, BALB/C mice were inoculated with BCL1 leukemia cells. On days 11 and 24 post inoculation, mice received doxorubicin (Dox), with or without anti-CD4 or anti-CD8 depletion mAbs. On three consecutive days, mice were treated with IL-2/JES6. (A) Scheme of the experiment. (B) Survival curves. n = 16 mice per group in two independent experiments with the exception of “Dox + IL-2/JES6 + αCD8” conditions where three mice in the second experiment did not tolerate anti-CD8 administration for unknown reasons and were removed. (C) On day 30, spleens of mice from control (n = 7), Dox-treated (n = 8), and Dox + IL-2/JES6-treated (n = 6) groups were collected and analyzed by flow cytometry. Left: expression of KLRK1, CD49d, and IL-7R on CD44+ CD8+ T cells. Representative staining. Right: percentage of KLRK1+ CD49d-, KLRK1+ CD49d+, and KLRK1+ IL-7R+ cells out of CD44+ CD8+ T cells. Two independent experiments. (D–H) On day 0, C57Bl/6J mice were inoculated with B16F10 melanoma cells. On day 6, mice received doxrubicin (Dox). On three consecutive days, mice were treated with IL-2/JES6. (D) Scheme of the experiment showing also the eventual treatment with anti-CD4 or anti-CD8 depletion mAb relevant for Figure 7—figure supplement 1B. (E) Survival curves. n = 16 mice per group in two independent experiments. (F–H) On day 11, spleens and tumors of mice from control, Dox-treated, and Dox + IL-2/JES6-treated groups (n = 8 per group) were collected and analyzed by flow cytometry. Two independent experiments. (F) Top: expression of KLRK1 and granzyme B on splenic CD8+ T cells. Representative dot plot. Bottom: percentage of KLRK1+ GZMB B+ cells out of total CD8+ T cells. (G) Top: expression of KLRK1 and CD49d on intratumoral CD8+ T cells. Representative dot plot. Bottom: percentage of KLRK1+ CD49d- cells out of total CD8+ T cells. (H) IL-7R and GZMB expression on intratumoral KLRK1+ CD8+ T cells. Representative histograms and geometric mean fluorescence intensity (MFI) are shown. Statistical significance was calculated by log-rank (Mantel–Cox) test (survival B, E), or Kruskal–Wallis test (C, F–H) (p-value is shown in italics) with Dunn’s post-test (<0.05, <0.01, <0.001). Median is shown. We analyzed the effect of the combinatorial therapy on the formation of different effector subsets. Mice treated with Dox and IL-2ic, but not those treated with Dox alone or untreated, generated a large population of KLRK1+ T cells (Figure 7C). The frequencies of KLRK1+ CD49d+, KLRK1+ CD49d-, KRLK1+ IL-7R+, and KLRK1+ GZMB+ CD8+ T-cell subsets were significantly higher in the spleens of mice treated with Dox and IL-2ic than in the control mice (Figure 7C, Figure 7—figure supplement 1B).This analysis showed that the formation of KILR T cells and KLRK1+ T-cell in general correlates with the improved survival of leukemic mice. We observed very similar protective effects of combinational therapy of Dox plus IL-2/JES6 in a B16F10 melanoma model (Figure 7D and E, Figure 7—figure supplement 1C and D). The spleens of mice treated with Dox and IL-2ic had much higher frequencies of KLRK1+ GZMB+, KLRK1+ CD49d+, and KLRK1+ CD49d- CD8+ T cells than the untreated or Dox-only treated mice (Figure 7F, Figure 7—figure supplement 1E and F). A large proportion of CD8+ T cells infiltrating the tumors in the Dox plus IL-2ic treated mice had the KLRK1+ CD49d- phenotype, whereas these cells were relatively rare in the untreated and Dox-only treated mice. Moreover, KLRK1+ T cells in tumors of Dox + IL-2ic group expressed higher levels of IL-7R and GZMB than their counterparts in untreated or Dox-only treated mice (Figure 7H). Overall, the Dox + IL-2ic therapy enhanced the anti-tumor immune response in a CD8+ T cell-dependent manner and induced large numbers of KLRK1+ CD8+ T cells. A substantial part of these KLRK1+ CD8+ T cells, especially in the tumor, were bona fide KILR cells as shown by their high expression of IL-7R and GZMB and low expression of CD49d. Since IL-2/JES6 highly selectively stimulates IL-2Rα+ cells, represented mostly by Tregs in naïve unprimed mice, it was considered as an immunotherapeutic tool for the specific expansion of Tregs in vivo with a potential application in the treatment of autoimmune diseases and pre-transplantation care. On the contrary, our data shows that it promotes antitumor response of CD8+ T cells, particularly in the combination with the immunogenic chemotherapy (Obeid et al., 2007; Kim and Kin, 2021), and that this anti-tumor activity is not counteracted by Tregs. ## Discussion In this study, we investigated how Tregs prevent CD8+ T-cell mediated autoimmune pathology. Using a well-controlled system based on the transfer of CD8+ T cells specific for a pancreatic neo-self-antigen, we revealed that regulatory T cells substantially increase the quorum of self-reactive T cells (Bosch et al., 2017) required to induce the autoimmune pathology. We observed that Tregs do not alter the antigen-affinity discrimination as they suppress both high-affinity and low-affinity CD8+ T cells, which is in an apparent contrast with a previous study (Pace et al., 2012) concluding that Tregs suppress exclusively low-affinity T-cell responses. The most possible explanation is that the previous study focused on the role of Tregs in the CD8+ T-cell response to Listeria monocytogenes, where the presence or depletion of Tregs might regulate the kinetics of pathogen clearance and thus the dose of bacterial antigens. Contrary to that, the dose of the priming antigen was constant in our experimental diabetes. We have identified the reduction of available IL-2 as the major mechanism of Treg-mediated tolerance of CD8+ T cells. The evidence is that Treg depletion increases IL-2 signaling in these cells, i.e., pSTAT5 levels and expression of IL-2-responsive genes, and that the administration of IL-2 and IL-2ic mimics the absence of Tregs in vitro and in vivo, respectively. Strong exogenous IL-2 signals stimulate the expansion of Tregs and the expression of their effector molecules (Tomala and Kovar, 2016), which should enhance all their potential suppression mechanisms with the notable exception of IL-2 sequestration. However, as this exogenous IL-2 signal mimics the effect of Treg depletion in our model, mechanisms non-targeting IL-2 do not seem to contribute substantially to the suppression of self-reactive CD8+ T cells. This conclusion is in line with some previous studies of Treg-mediated suppression of effector CD8+ T cells (Kalia et al., 2015; McNally et al., 2011) and with the importance of IL-2Rα expression in self-reactive CD8+ T cells for the infiltration of pancreatic islets (Marchingo et al., 2014). It has been found that the biological activity of IL-2ic is dictated by the anti-IL-2 mAb clone (Boyman et al., 2006; Spangler et al., 2015). IL-2/JES6 immunocomplexes selectively stimulate cells expressing the high-affinity trimeric IL-2 receptor and were considered as an immunosuppressive agent acting via the expansion of Tregs (Tomala and Kovar, 2016). In contrast, our conclusion that IL-2 restriction is the major mechanism of Treg-mediated suppression of CD8+ T cells implies that the administration of IL-2/JES6 should release antigen-activated CD8+ T cells from the Treg control. Indeed, IL-2/JES6 in combination with chemotherapy significantly prolonged the survival of tumor-bearing mice in two different tumor models in a CD8+ T-cell-dependent manner. Accordingly, the administration of IL-2/JES6 after priming, but not before the transfer of self-reactive T cells, decreased the T-cell quorum for the induction of experimental diabetes. Collectively, these results reveal that a strong sustained IL-2 signal, even if selective for IL-2Rα+ cells, potentiates antigen-induced CD8+ T-cell antitumor immunity and autoimmunity despite its concomitant stimulatory effects on Tregs. Recently, a combinational therapy of in vitro expanded polyclonal Tregs and low-dose IL-2 was tested in patients with type I diabetes in a phase I clinical trial (Dong et al., 2021). The authors of the study did not observe a benefit of this intervention for the patients but they reported an expansion of Tregs along with activated NK and cytotoxic CD8+ T cells, which corresponds to our conclusions. Our finding that IL-2 restriction is the major mechanism of Treg-mediated regulation of self- and tumor-reactive CD8+ T cells does not exclude the involvement of additional mechanisms regulating other immune cell types, such as conventional CD4+ T cells. On the contrary, there is substantial evidence that Tregs suppress conventional CD4+ T cells largely via mechanisms not dependent on IL-2 sequestration. First, we show that antigen-stimulated CD4+ T cells are less sensitive to IL-2 signals than CD8+ T cells in vivo, which limits the potential impact of the IL-2 restriction on them. Second, it has been shown that Tregs lacking the high-affinity IL-2 receptor, implied in IL-2 sequestration, can still control the homeostasis of CD4+ T cells but not CD8+ T cells (Chinen et al., 2016). Third, it has been shown that Tregs suppress conventional CD4+ T cells by depleting their cognate peptide-MHCII from APC, which is not applicable for the suppression of MHCI-restricted CD8+ T cells (Akkaya et al., 2019). In contrast, a recent study proposed that self-reactive CD4+ T cells produce IL-2 locally attracting Tregs to their close proximity, which leads to the suppression of later phases of self-reactive T-cell activation at least partially via limiting IL-2 (Wong et al., 2021). Antigenic stimulation of CD8+ T cells in the presence of excessive IL-2, induced by depletion of Tregs or by administration of exogenous IL-2R agonists, leads to the formation of a previously uncharacterized subset of KILR effector CD8+ T cells. They express high levels of cytotoxic molecules, such as granzymes, but unlike conventional effector CD8+ T cells, they also express high levels of IL-7R. Moreover, they express several NK receptors, including KLRK1/NKG2D, which is involved in target cell killing (Prajapati et al., 2018). Since these cells show superior cytotoxic properties in vivo, suppression of their formation is probably a major mechanism of Treg-mediated self-tolerance. On the other hand, the existence of the KILR effector gene expression program suggests the possibility that these cells might arise naturally in specific immunological conditions, such as some type of infection. Indeed, it has been shown that the infection with lymphocytic choriomeningitis virus induces a transient drop in Treg numbers (Schorer et al., 2020), which might alleviate the Treg-mediated suppression of virus-specific CD8+ T cells in the early phase of the immune response. However, we could not find KILR effector CD8+ T cells in any scRNAseq dataset from the course of mouse infection, suggesting that such KILR effector-inducing conditions would be rather rare. On the other hand, we found increased frequencies of KILR cells in the spleens and tumors in mice with cancer treated with Dox and IL-2ic. The antigenic stimulation was probably provided by tumor-associated antigens released from cancer cells undergoing the immunogenic cell death induced by Dox (Obeid et al., 2007; Kim and Kin, 2021). The expression of IL-7Rα in KILR effector CD8+ T cells is striking as conventional effector T cells are characterized as IL-7R-, which is associated with their short lifespan (Joshi et al., 2007; Kaech et al., 2003). A previous study showed that antigenic stimulation together with activating antibodies to OX40 and 4-1BB TNF-family receptors induced IL-7R+ effector CD8+ T cells (Lee et al., 2007). These cells produced high levels of proinflammatory cytokines in an IL-7-dependent manner (Lee et al., 2007). Since 4-1BB signaling has been shown to enhance IL-2 production and IL-2Rα expression in T cells (Barsoumian et al., 2016; Oh et al., 2015), it is possible that those IL-7R+ effector cells were induced via a strong IL-2 signal. However, as the gene expression profile of OX$\frac{40}{4}$-1BB-induced IL-7R+ effector CD8+ T cells has not been analyzed (Lee et al., 2007), it is not clear whether or not these cells have other similarities to KILR effector CD8+ T cells. Our reanalysis of publicly available datasets of CD8+ T cells revealed KILR-like CD8+ T cells in the human peripheral blood. These cells show patterns of KILR T-cell gene expression, but they do not express high levels of cytotoxic molecules, probably because they lack recent antigenic/IL-2 stimulation. Therefore, future studies are needed to resolve their phenotype and function. In any case, the potent cytotoxic capacity and their expansion by IL-2 agonists make KILR effector CD8+ T cells a promising clinical target in cancer immunotherapy. ## Antibodies, peptides, and dyes Antibodies to the following antigens were purchased from BioLegend and used for flow cytometry: CD4 BV650 (RM4-5, #100545), CD4 Alexa Flour 700 (RM4-5, #100536), CD4 APC-Cy7 (GK15, #100414), CD4 APC (RM4-5, #100516), CD8a BV421 (53-6.7, #100738), CD8a PE (53-6.7, #100708), CD11c Alexa Flour 700 (N418, #117319), CD25 Alexa Fluor 700 (PC61, #102024), CD25 BV605 (PC61, #102036), CD25 BV650 (PC61, #102038), CD25 PE-Cy7 (PC61, #102016), CD44 BV 650 (IM7, #103049), CD45.1 Alexa Fluor 700 (A20, # 110723), CD45.1 BV650 (A20, #110735), CD45.1 FITC (A20, #110706), CD45.1 PerCP-Cy5.5 (A20, #110728), CD45.2 Alexa Fluor 700 (104, #109822), CD45.2 APC (104, #109814), CD49d APC (R1-2, #103622), CD49d PE-Cy7 (R1-2, #103618), CD62L FITC (MEL-14, #104406), CD80 PerCP-Cy5.5 (16-10A1, #104722), CD86 Alexa Fluor 700 (GL-1, #105024), CD103 PerCP-Cy5.5 (2 E7, #121416), CD127 PE-Cy7 (A7R34, #135013), TCRβ APC (H57-597, #109212), TCRβ PE (H57-597, #109208), and KLRG1 BV510 (2F1/KLRG1, #138421). Antibodies to TCR Vα2 FITC (B20.1, #553288), CD45.1 APC (A20, #558701), CD45.1 PE (A20, #553776), and CD49d PE (R1-2, #553157) were purchased from BD Pharmingen. Antibodies to Granzyme B eFluor660 (NGZB, #50-8898-82), FOXP3 PE-Cy7 (FJK-16s, #25-5773-82), FR4 PE/Dazzle594 (12A5, #125016), KLRK1 PE-eFluor610 (CX5, #61-5882-82), and MHC class II (I-A/I-E) FITC (M$\frac{5}{114.15.2}$, #11-5321-82) were purchased from eBioscience. Goat anti-rabbit IgG (H+L) secondary antibody conjugated to Alexa Fluor 647 from Invitrogen (#21245) was used following rabbit anti-mouse phospho-Stat5 (Tyr694) (D47E7) from Cell Signaling (#4322S). Anti-mouse CD3ε antibody (clone 145-2C11, #100302, BioLegend) was used for plate coating. Primary antibodies that were used for immunohistochemistry: CD8a (EPR21769, #217344, Abcam), CD4 Alexa Flour 647 (RM4-5, #100530), and insulin (polyclonal, # PA1-26938, Invitrogen). Secondary antibodies: goat anti-rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 555 (Invitrogen, #A32732), goat anti-guinea pig IgG (H+L) Highly Cross-Adsorbed Secondary Antibody (Invitrogen, #A-11073). Anti-CD4 (GK1.5, #BE0003-1) and anti-CD8α (53–6.7, #BE0004-1) depletion antibodies were purchased from Bioxcell (USA). Anti-mouse IL-2 mAb S4B6 (#BE0043-1) and anti-mouse IL-2 mAb JES6-1A12 (#BE0043) used for preparation of IL-2/S4B6 and IL-2/JES6 complexes, respectively, were purchased from Bioxcell (USA). OVA (SIINFEKL), Q4R7 (SIIQFERL), and Q4H7 (SIIRFEHL) peptides were purchased from Eurogentec or Peptides&Elephants. CFSE (#65-0850-84) and Cell Trace Violet (CTV) (#C34557), LIVE/DEAD near-IR (#L10119), and Hoechst 33258 (#H3569) dyes were purchased from Invitrogen. FITC Annexin V Apoptosis Detection Kit with 7-AAD (#640922) was purchased from BioLegend. ## IL-2/S4B6 and IL-2/JES6 complexes IL-2/S4B6 and IL-2/JES6 immunocomplexes were described previously (Tomala and Kovar, 2016). Complexes were prepared by mixing recombinant mouse IL-2 (Cat# 212-12, 100 µg/ml; PeproTech) with anti-IL-2 mAb at a molar ratio of 2:1 in PBS. After 15 min incubation at room temperature, the complexes were diluted in PBS into the desired concentration, frozen at –20°C, and thawed shortly before application. ## Mice All the mice had C57Bl/6J or BALB/C background. DEREG (RRID: MMRRC_032050-JAX) (Lahl et al., 2007), Foxp3DTR (RRID: IMSR_JAX:016958) (Kim et al., 2007), RIP.OVA (RRID: MGI:3789286) (Kurts et al., 1998), OT-I Rag2-/- (RRID:MGI:3783776, MGI:2174910) (Palmer et al., 2016; Shinkai et al., 1992), OT-II Rag2-/- (RRID: MGI:3762632, MGI:2174910) (Shinkai et al., 1992; Barnden et al., 1998), Ly5.1 (RRID: IMSR_JAX:002014) (Jang et al., 2018) strains were described previously. Mice were bred in specific-pathogen-free facilities (Institute of Molecular Genetics of the Czech Academy of Sciences, Prague; Department of Biomedicine, University Hospital, Basel) or in a conventional facility (Institute of Microbiology of the Czech Academy of Sciences, Prague). Mice were kept in the animal facility with a 12 hr of light–dark cycle with food and water ad libitum. Animal protocols were performed in accordance with the laws of the Czech Republic and Cantonal and Federal laws of Switzerland, and approved by the Czech Academy of Sciences (identification no. $\frac{11}{2016}$, $\frac{81}{2018}$, $\frac{15}{2019}$) or the Cantonal Veterinary Office of Baselstadt, Switzerland, respectively. Males and females were used for the experiments. At the start of the experiment, all mice were 6–11 weeks old. If possible, age- and sex-matched pairs of animals were used in the experimental groups. If possible, littermates were equally divided into the experimental groups. No randomization was performed when the experimental groups were based on the genotype of the mice; otherwise mice were assigned to experimental groups randomly (defined by their ID numbers) prior to the contact between the experimenter and the mice. The experiments were not blinded since no subjective scoring method was used. The estimation of the sample size was based on our previous experience. We typically aimed at having 12 animals per group in three independent experiments. The final number of mice per group depended on the number of available mice with required genotype, number of isolated T-cells for adoptive transfers, and unpredictable events (e.g., mouse death). In cancer experiments, we aimed at 16 mice per group in two independent experiments. ARRIVE Essential10 guidelines (https://arriveguidelines.org/) were followed. Besides the reported exclusion criteria described below, some very rare experiments were excluded, when we realized an unintended severe technical error/deviation from the protocol (a general pre-established criterium). ## RT-qPCR 2–10 × 104 OT-I T lymphocytes were FACS sorted as CD8+ CD45.1+ CD49d- KLRK1- cells (naïve), CD8+ CD45.1+ CD49d+ KLRK1- cells (antigen experienced), CD8+ CD45.1+ CD49d+ KLRK1+ cells (double positive), or CD8+ CD45.1+ CD49d- KLRK1+ cells (KILR). Total RNA was isolated by TRIzol LS (Invitrogen, #10296010) and in-column DNase digestion using RNA Clean & Concentrator Kit (Zymo Research), according to manufacturer's instructions. RNA was stored at –80°C or transcribed immediately using RevertAid reverse transcriptase (Thermo Fisher Scientific, #EP0442) with oligo(dT)18 primers according to the manufacturer’s instructions. RT-qPCR was carried out using LightCycler 480 SYBR green I master chemistry and a LightCycler 480 machine (Roche). All samples were measured in triplicates. Median CT values were normalized to a reference gene, Glyceraldehyde-3-Phosphate Dehydrogenase (Gapdh). The sequences of used primers are: Gapdh: F TGCACCACCAACTGCTTAGC, R GGCATGGACTGTGGTCATGAG mCd7: F TGGATGCCCAAGACGTACA, R TAAGATCCCTTCCAGGTGCC mIfitm1: F ATGCCTACTCCGTGAAGTCTAGG, R GACAACGATGACGACGATGGC mGzma: F AAAGGACTCCTGCAATGGGG, R ATCGGCGATCTCCACACTTC mGzmb: F GGGGCCCACAACATCAAAGA, R GGCCTTACTCTTCAGCTTTAGCA mGzmk: F AAGGATTCCTGCAAGGGTGA, R ATTCCAGGCTTTTTGGCGATG mKlrd1: F TCGGTGGAGACTGATGTCTG, R AACACAGCATTCAGAAACTTCC mIl7r: F AAAGCCAGAGCGCCTGGGTG, R CTGGGCAGGGCAGTTCAGGC mIl2ra: F AGAACACCACCGATTTCTGG, R GGCAGGAAGTCTCACTCTCG mCxcr6: F ACTGGGCTTCTCTTCTGATGC, R AAGCGTTTGTTCTCCTGGCT ## Enrichment of T lymphocytes T lymphocytes were enriched by negative selection using the Dynabeads Biotin Binder kit (Invitrogen, #11047), and biotinylated anti-CD19 (clone 1D3, ATCC# HB305), anti-CD4 (clone YTS177), or anti-CD8 antibodies (clone 2.43, ATCC# TIB-210), depending on the experimental setup. Antibodies were produced and biotinylated in house using (+)-Biotin N-hydroxysuccinimide ester (Sigma-Aldrich) in bicarbonate buffer. The excessive biotin was separated from the antibody using Sephadex G-25 (Sigma-Aldrich). For RT-qPCR experiments, CD8+ T cells were enriched using Dynabeads Untouched Mouse CD8 Cells Kit (Invitrogen, #11417D) according to the manufacturer’s instructions. ## Flow cytometry and cell sorting Live cells were stained with relevant antibodies on ice. LIVE/DEAD near-IR dye or Hoechst 33258 were used for discrimination of viable and dead cells. For intracellular staining of FOXP3 and Granzyme B, cells were fixed and permeabilized using Foxp3/Transcription Factor Staining Buffer Set (#00-5523-00, Invitrogen) according to the manufacturer’s instructions. Fixed cells were stained at room temperature for 1 hr. For intracellular staining of pSTAT5, splenic cells were fixed using Fixation/Permeabilization buffer (Foxp3/Transcription Factor Staining Buffer Set, #00-5523-00, Invitrogen) immediately after isolation at room temperature for 15 min, washed twice with permeabilization buffer (Foxp3/Transcription Factor Staining Buffer Set, #00-5523-00, Invitrogen), washed with PBS, and stained with anti-pSTAT5 antibody at room temperature overnight. The next day, cells were stained with secondary antibody at room temperature for 1 hr. Flow cytometry was carried out using an LSRII (BD Biosciences) or an Aurora (Cytek). Data were analyzed using FlowJo software (BD Biosciences). Cell sorting was performed on an Influx or an Aria machines (both BD Biosciences). ## Histological analysis of pancreas Pancreases were harvested from mice on day 4 post-immunization, mounted in OCT embedding compound (Sakura Finetek Tissue-Tek, #4583), and frozen at –80°C. 10-µm-thick tissue sections were cut using Cryostat (Leica Microsystems, CM1950) and mount on SuperFrost Plus Adhesion slides (Erpedia, #J1800AMNZ). Dry slides were stored at –80°C. For conventional light microscopy, tissue sections were fixed with acetone for 15 min and subjected to hematoxylin/eosin staining. DM6000-M microscope (Leica Microsystems) was used to acquire images. For immunofluorescence microscopy, tissue sections were fixed with $4\%$ PFA for 10 min, permeabilized with $0.1\%$ Triton X-100 for 10 min, and blocked using PBS/$5\%$ FBS, $5\%$ BSA for 1 hr. Next, samples were stained with the primary antibody mix (guinea pig anti-mouse insulin, rat anti-mouse CD4 Alexa Fluor 647, and rabbit anti-mouse CD8α) at 4°C overnight. After washing, the slides were incubated with the secondary antibodies for 1 hr, nuclei were stained with 5 µM DAPI solution (Invitrogen, D21490), and samples were mounted with ProLong Gold Antifade Mountant (Invitrogen, #P36930). Images were acquired using SP8 LIGHTNING confocal microscope (Leica Microsystems). ## ELISA The anti-idiotypic B1 mAb were produced by the use of B1 hybridoma via the conventional ascites producing approach in paraffin oil pre-treated BALB/C mice. It was purified by $45\%$ supercritical antisolvent precipitation (ammonium sulfate) followed by extensive dialysis against distilled water, centrifuged at 12,000 × g for removal of IgM, and purified by a protein A affinity chromatography. Next, it was biotinylated with Sulpho NHS-biotin reagent (Pierce) according to the manufacturer's protocol. Blood of mice was taken on days 28, 54, 94 (or 95, or 96) of B-cell leukemia/lymphoma experiment. Blood serum was separated and serially diluted (1:40–1:640) in PBS. Plate wells (Costar) were coated with 50 µl of diluted samples and incubated overnight at 4°C, followed by blocking with $1\%$ gelatin (200 µl per well, 2 hr at room temperature). Biotinylated anti-idiotypic B1 mAb was added (20 ng/ml) in a buffer containing $0.5\%$ gelatin, $3\%$ PEG, and $0.1\%$ tween, and plate was incubated for 2 hr at room temperature. Next, samples were conjugated with ExtrAvidin−Peroxidase (Sigma-Aldrich) for 1 hr at room temperature, and 3,3′,5,5′-tetramethylbenzidine substrate (Sigma-Aldrich) was added for 10 min in dark. Reaction was stopped with 50 µl of 2 M H2SO4, and absorbance at 450 nm was measured by a Biolisa spectrometer (Bioclin). ## Bone marrow-derived dendritic cells Bone marrow cells were seeded on 100 mm plates (tissue culture untreated) and maintained in DMEM (Sigma-Aldrich) containing $10\%$ FBS (GIBCO), 100 U/ml penicillin (BB Pharma), 100 mg/ml streptomycin (Sigma-Aldrich), 40 mg/ml gentamicin (Sandoz), and $2\%$ of supernatant from J558 cells producing GM-CSF for 10 days at $5\%$ CO2 at 37°C (Kralova et al., 2018). The cells were split every 2–3 days. On day 10, cells were incubated in the presence of 100 ng/ml LPS (Sigma-Aldrich) and 200 nM of indicated peptide for 3 hr at $5\%$ CO2 at 37°C. Next, plates were incubated with $0.02\%$ EDTA in PBS for 5 min at $5\%$ CO2 at 37°C and harvested. Washed and filtered cells were used for adoptive transfers. ## In vitro proliferation assay CD8+ T lymphocytes from WT mice were FACS sorted, labeled with CTV, and plated into an anti-CD3ε-antibody-coated 48-well plate in IMDM ($10\%$ FBS [Gibco], 100 U/ml penicillin [BB Pharma], 100 mg/ml streptomycin [Sigma-Aldrich], 40 mg/ml gentamicin [Sandoz]). Tregs, sorted as CD4+ GFP+ T lymphocytes from DEREG+ mice were added to corresponding wells of the plate in 1:1 ratio with conventional CD8+ T lymphocytes. Recombinant IL-2 (2 ng/ml) was added or not. Cells were incubated at 37°C, $5\%$ CO2 for 72 hr and analyzed by flow cytometry. ## In vivo proliferation assay OT-I CD8+ and OT-II CD4+ T cells were isolated from OT-I Rag2-/- and OT-II Rag2-/- mice, respectively, using MACS negative selection kits. On day 0, mixture of OT-I CD8+ (0.75 × 106 cells) and OT-II CD4+ (1.5 × 106 cells) CFSE labeled cells was adoptively transferred into recipient Ly5.1 mice. The next day, recipient mice were immunized with ovalbumin protein i.p. ( 75 μg/mouse), and after 6 hr, they received the first dose of IL-2ic (1.5 μg IL-2 equivalent/dose, i.p.). On days 2–4, mice received additional IL-2ic doses. On day 5 post-immunization, spleens of Ly5.1 mice were collected and used for flow cytometry analysis. ## Treg depletion In order to deplete Tregs, 0.25 µg of DT (#D0564, Sigma-Aldrich) was administered i.p. to DEREG+ RIP.OVA mice and control DEREG- RIP.OVA mice for two consecutive days. ## Model of autoimmune diabetes The model of autoimmune diabetes has been described previously (King et al., 2012). Briefly, an indicated number of OT-I T cells isolated from OT-I Rag2-/- mice were adoptively transferred into a recipient RIP.OVA mouse i.v. On the following day, the recipient mice were immunized with 106 of bone marrow-derived DCs loaded with indicated peptide (i.v.) or with 25 µg LPS + 50 µg OVA peptide in 200 µl PBS (i.p.). For the experiments using antigen-loaded DCs (Figure 2B and later), we included an internal control of two RIP.OVA mice (DEREG- and DEREG+) receiving 1000 OT-I T cells and 106 OVA-loaded DCs in each experiment. When these control mice did not develop diabetes, which happened sometimes probably because of the issues with DC culture, the whole experiment was excluded. We realized that ~$50\%$ of mice die when the diabetic protocol is combined with IL-2/JES6 immunocomplexes (Figure 4E–H). Alternatively, Cd3e-/- RIP.OVA recipient mice received 106 of polyclonal CD8+ T lymphocytes derived from RIP.OVA Ly5.1 donors 8 days prior to immunization. One day prior to immunization, Cd3e-/- RIP.OVA mice received 4–8 × 105 Tregs (sorted as CD4+ GFP+ TCRβ+) or 106 conventional CD4+ T lymphocytes (sorted as CD4+ GFP- TCRβ+) derived from DEREG RIP.OVA donors. If IL-2ic were used, 1.5 µg IL-2 equivalent/dose of IL-2/S4B6 or IL-2/JES6 were injected i.p. on days 0, 1, and 2 post-immunization. Alternatively, mice received five doses of IL-2/JES6 (2.5 µg IL-2 equivalent) on days −7,–6, −5,–4, –3 prior to the immunization. We removed these mice because they died before the first glucose measurement (a pre-established criterium). After two experiments with the same result, we did not repeat this assay anymore for ethical reasons. Urine glucose was monitored on a daily basis using test strips (GLUKOPHAN, Erba Lachema). Blood glucose was measured using Contour blood glucose meter (Bayer) on a specified day(s), depending on the experimental design. The animal was considered diabetic when the concentration of glucose in the urine reached ≥1000 mg/dl for two consecutive days. ## Induction of KILR effector T cells using IL-2ic On day 0, recipient mice (OT-I Rag2-/- or RIP.OVA) received 25 µg OVA peptide in 200 µl of PBS and/or 0.75 µg IL-2 equivalent/dose of IL-2/S4B6 in 250 µl PBS i.p. On days 1 and 2, mice received two more doses of IL-2/S4B6. On day 3, spleens were collected and used for flow cytometry analysis or FACS sort. ## Competitive adoptive transfer of KILR and control cells OT-I Rag2-/- and Ly5.1 OT-I Rag2-/- mice were immunized with OVA peptide (25 µg, single dose on day 0, i.p.) and/or IL-2/S4B6 (0.75 µg IL-2 equivalent/dose of IL-2/S4B6, three doses, days 0–2, i.p.). On day 3, three populations of CD8α+ T cells were FACS sorted: KILR (KLRK1+ CD49d-), naïve (KLRK1- CD49d-), or effector cells (KLRK1- CD49d+). KILR cells were mixed with naïve or effector cells (1:1, 500 × 103 cell/mouse in total in experiment 1, 400 × 103 cell/mouse in total in experiment 2), stained with CTV, and injected i.v. to recipient Cd3e-/- mice. On day 7, splenocytes of the recipient mice were analyzed by flow cytometry. ## In vivo killing assay In vivo CD8+ T cell killing assay was performed as described previously (Kim et al., 2014) with minor modifications. In short, OT-I Rag2-/- mice were immunized or not with OVA peptide (25 µg, single dose on day 0, i.p.) and/or IL-2/S4B6 (0.75 µg IL-2 equivalent/dose of IL-2/S4B6, three doses, days 0–2, i.p.). On day 3, FACS sorted KLRK1+ CD8α+ or KLRK1- CD8α+ cells were injected i.v. to recipient RIP.OVA mice, which had received a mixture of target cells (107 OVA-pulsed cells and 107 unpulsed target cells) earlier the same day. Target cells were prepared from spleens of Ly5.1 mice. OVA-pulsed cells were prepared via 1 hr incubation in RPMI-1640 containing $10\%$ FBS (GIBCO), 100 U/ml penicillin (BB Pharma), 100 mg/ml streptomycin (Sigma-Aldrich), 40 mg/ml gentamicin (Sandoz), and 2 µM OVA peptide at 37°C, $5\%$ CO2, followed by loading with CTV. Unpulsed target cells were incubated in parallel in the medium without OVA peptide and subsequently loaded with CFSE. On day 4, splenocytes of the recipient mice were analyzed by flow cytometry. Target cells were identified as CD45.1+ cells. Ratio of unpulsed (CFSE+) to OVA pulsed target cells (CTV+) was determined and normalized to those of recipients that did not receive OT-I T cells. ## Murine B-cell leukemia On day 0, BALB/C female mice were injected i.p. with 5 × 105 BCL1 cells (a BALB/c-derived leukemia cell line) (Slavin and Strober, 1978) in PBS. On days 11 and 24 post-inoculation, mice received doxorubicin (Adriblastina) (5 mg/kg in 250 µl PBS, i.v.), followed or not by anti-CD4 or anti-CD8α depletion mAbs (200 µg in 250 µl of PBS for both, i.p.). On days 12, 13, 14, 25, 26, and 27 IL-2/JES6 was administrated to mice (5 µg IL-2 equivalent/dose in 250 µl PBS, i.p.). Survival of mice was monitored from day 30 to day 100. On days 28, 54, and 94–96 blood serum of mice was used for ELISA for antibody against the idiotype of IgM expressed on the BCL1 cells. For the flow cytometry analysis, spleens were harvested on day 30. ## B16F10 melanoma Female C57Bl/6J mice were inoculated s.c. with 5 × 105 B16F10 melanoma cells (day 0). On day 6, mice received doxorubicin (Adriblastina) (8 mg/kg in 250 µl PBS, i.v.) followed or not by anti-CD8α or anti-CD4 depletion mAbs (200 ug in 250 µl of PBS, i.p.). On days 7–9, IL-2/JES6 was injected to mice (5 µg IL-2 equivalent/dose in 250 µl PBS, i.p.). Survival of mice was monitored on a daily basis. Tumor size was measured as the width and length using caliper every 2–4 days. Length of the tumor was determined as the longest diameter. The width was determined as longest diameter of the tumor perpendicularly to the length. The thickness of the tumor was arbitrary assigned to be a half of the tumor width. Tumor volume (mm3) was calculated as V = (L × W × W)/2, where V is tumor volume, W is tumor width, and L is tumor length. For the flow cytometry analysis, tumors and spleens were harvested on day 11. Tumor Dissociation Kit (Miltenyi Biotec, Cat# 130-096-730) has been used to prepare single-cell suspensions according to the manufacturer’s protocol. ## ScRNA sequencing Six-week-old female DEREG- RIP.OVA ($$n = 3$$) and DEREG+ RIP.OVA ($$n = 3$$) littermate mice were treated with DT (0.250 µg per mouse in 0.5 ml of PBS, i.p.) on days –2 and –1 prior to the immunization, followed by i.v. injection of 5 × 104 OT-I T cells in 200 µl of PBS on day –1. OT-I T cells were obtained from spleens and lymph nodes of Ly5.1 OT-I Rag2-/- mouse. On day 0, mice received Ly5.1 dendritic cells loaded with OVA (106 cells per mouse in 200 µl of PBS). On day 3 post-immunization, mice were sacrificed and spleens were collected. Erythrocytes were lysed with ACK buffer (2 min, on ice), cells were washed and resuspended in PBS/$2\%$ FBS. Next, cells were stained with LIVE/DEAD near-IR dye, anti-CD8a BV421, anti-CD45.1 PE, and anti-CD45.2 APC antibodies together with TotalSeq-C anti-mouse hashtag antibodies (anti-CD45 clone 30-F11, anti-H-2 clone M$\frac{1}{42}$, BioLegend, #155869 (MH5), #155871 (MH6), #155873 (MH7), #155875 (MH8), #155877 (MH9) and #155879 (MH10)). Viable CD8a+, CD45.1+, CD45.2- OT-I cells were FACS sorted. The individual samples were pooled together and with cells coming from an unrelated experiment. These unrelated cells were labeled with unique hashtag antibodies and removed during the analysis. The viability and concentration of cells after sort were measured using the TC20 Automated Cell Counter (#1450102, Bio-Rad). The viability of the cells pre-loading was >$90\%$. Cells were loaded onto a 10X Chromium machine (10X Genomics) aiming at the yield of 1000 cells per sample and processed with Feature Barcode technology for Cell Surface Protein protocol (#CG000186 Rev D) with the Chromium Single Cell 5’ Library & Gel Bead and Chromium Single Cell 5' Feature Barcode Library kits (10X Genomics, #PN-1000014, #PN-1000020, #PN-1000080, #PN-1000009, #PN-1000084). Resulting cDNA libraries were sequenced on a NovaSeq 6000 (Illumina) with the S1 Reagent Kit (100 or 300 cycles, Illumina, #20012865, #20012863). ## Analysis of scRNAseq data The raw scRNA data were mapped to Mouse Reference GRCm38 obtained from Ensembl database v102 (Howe et al., 2021) by 10X Genomics Cell Ranger 5.0.0 (Zheng et al., 2017). The same software was also used to create the employed mouse transcriptome reference. Default parameters were kept. The hashtag sequences were mapped using the hashtag sequence references by 10X Genomics Cell Ranger 5.0.0. The data were pre-processed using Seurat R package v4.0.3 (Haberman et al., 2014) on R v4.0.4 (https://www.r-project.org/). The cells with less than 200 transcripts were removed. A histogram of cell counts having a specified number of hashtag reads was computed for each hashtag. To detect the end of an initial slope of a histogram, a function based on descent along the gradient – moving to the neighboring point in the histogram as long as its associated value is lower – was used. Before its application, each histogram was averaged using the sliding window of size $$n = 5$$ points. The resulting limit was used to associate or not each cell with the respective sample. Cells marked by multiple hashtags were considered as doublets and excluded from further analysis. Read limits (lim) and the number of recovered cells (#) for individual hashtags wer MH5 (lim 19, #744), MH6 (lim 19, #778), MH7 (lim 24, #789), MH8 (lim 21, #849), MH9 (lim 35, #836), and MH10 (lim 25, #878). The lists of barcodes of each uniquely marked cell were created and used to separate read pairs where the start of first read R1 matches one of barcodes exactly from those with different barcodes and consequently either originating from the cells from unrelated experiment, from cell doublets or insufficiently marked cells. These reads were mapped again using 10X Genomics Cell Ranger 5.0.0 to generate data used further for downstream analysis. Cells with less than 200 transcripts and/or more than $15\%$ of transcripts mapping to mitochondrial genes were removed. Mitochondrial genes, TCRα and TCRβ V(D)J-genes, ribosomal genes and genes whose transcripts were detected in less than three cells were excluded. Log normalization (scale factor = 1 × 104), scaling, identification of variable features (2000 variable features), dimensional reduction (PCA and UMAP with top 50 and 15 principal components, respectively, 40 nearest neighbors for UMAP), identification of nearest neighbors in the reduced space (‘rann’ algorithm), and Louvain clustering (resolution = 0.8) were performed using the Seurat R package v4.0.3 (Haberman et al., 2014) on R v4.0.4 (https://www.r-project.org/). These steps allowed to identify the low-quality cells and contaminating cell types that were removed along with cells with more than $7.5\%$ of transcripts mapping to mitochondrial genes. In total, 4043 cells passed the QC steps. Afterward, all steps starting from and including log-normalization were repeated using slightly different parameters (800 variable features, 12 top principal components for both UMAP and PCA dimensional reductions and 30 nearest neighbors for UMAP, resolution = 0.3 for Louvain clustering; other parameters stayed the same). Cell cycle scores for S phase and G2/M phases were regressed out in the scaling step. Projection of clusters on a reference dataset of acute and chronic viral infection CD8+ T cells was done using the ProjecTILs R package v0.6.0 (Andreatta et al., 2021). NK cell signature was calculated for each cluster using the AddModuleScore function from the Seurat package with the default parameters. Signature genes were selected from the Molecular Signatures Database v7.5.1 (systematic name M4838 and M5669 for murine and human datasets, respectively). Signature genes that were not detected in our datasets were filtered out. For the separate analysis of KILR effector CD8+ T cells, cluster 4 was extracted from the original data and re-clustered again with adjusted parameters. The code for the cell filtration to hashtags, barcode extraction, and whole downstream analysis is accessible on GitHub (https://github.com/Lab-of-Adaptive-Immunity/Supereffectors_scRNAseq; Lab of Adaptive Immunity, 2023a, copy archived at swh:1:rev:7b0dec9507dd45ab4bb0619912b240f512c7798f). ## Gene set enrichment analysis Lists of IL-2-responsive genes were obtained from literature (Kovanen et al., 2005; Lin et al., 2012). Fold changes between DEREG+ and DEREG- samples were calculated with the Seurat R package v4.0.3 (Haberman et al., 2014). Gene set enrichment analysis (GSEA) analysis was performed using the fgsea R package v1.16.0 (Korotkevich, 2021). Genes with similar fold changes were ranked in a random order. ## Building and analysis of human CD8+ T-cell atlas The human CD8+ Atlas was built from 14 different data sets previously mapped to human genome reference GRCh38 by 10X Genomics and downloaded from their support site (https://support.10xgenomics.com/) either as raw feature matrices or, if not available for given data set, as filtered feature matrices. For data sets with available V(D)J information, their list was downloaded either as list of filtered annotated contigs or, if not available for given data set or it was originally processed by Cell Ranger 5.0.0 or higher, as list of all annotated contigs. The following data sets were used: CD8+ T cells of Healthy Donor 1, Single Cell Immune Profiling Dataset by Cell Ranger 3.0.2, 10X Genomics, (2019, May 9); CD8+ T cells of Healthy Donor 2, Single Cell Immune Profiling Dataset by Cell Ranger 3.0.2, 10X Genomics, (2019, May 9); CD8+ T cells of Healthy Donor 3, Single Cell Immune Profiling Dataset by Cell Ranger 3.0.2, 10X Genomics, (2019, May 9); CD8+ T cells of Healthy Donor 4, Single Cell Immune Profiling Dataset by Cell Ranger 3.0.2, 10X Genomics, (2019, May 9); 10k Human PBMCs with TotalSeq-B Human TBNK Antibody Cocktail, 3’ v3.1, Single Cell Gene Expression Dataset by Cell Ranger 6.0.0, 10X Genomics, (2021, March 31); 10k Human PBMCs Multiplexed, 2 CMOs - Inputs/Library, Single Cell Gene Expression Dataset by Cell Ranger 6.0.0, 10X Genomics, (2021,March 2); 5k Peripheral blood mononuclear cells (PBMCs) from a healthy donor (v3 chemistry), Single Cell Gene Expression Dataset by Cell Ranger 3.0.2, 10X Genomics, (2019, May 29); 10k PBMCs from a Healthy Donor – Gene Expression and Cell Surface Protein, Single Cell Gene Expression Dataset by Cell Ranger 3.0.0, 10X Genomics (2018, November 19); 8k PBMCs from a Healthy Donor, Single Cell Gene Expression Dataset by Cell Ranger 2.1.0, 10X Genomics, (2017, November 8); PBMCs of a Healthy Donor (v1), Single Cell Immune Profiling Dataset by Cell Ranger 3.1.0, 10X Genomics (2019, July 24); Human T cells from a Healthy Donor, 1k cells – multi (v2), Single Cell Immune Profiling Dataset by Cell Ranger 5.0.0, 10X Genomics (2020, November 19); Human PBMC from a Healthy Donor, 10k cells – multi (v2), Single Cell Immune Profiling Dataset by Cell Ranger 5.0.0, 10X Genomics, (2020, November 19); Human PBMC from a Healthy Donor, 1k cells (v2), Single Cell Immune Profiling Dataset by Cell Ranger 4.0.0, 10X Genomics (2020, August 25); and PBMCs of a healthy donor – 5' gene expression and cell surface protein, Single Cell Immune Profiling Dataset by Cell Ranger 3.0.0, 10X Genomics, (2018, November 19). Cells with less than 200 transcripts, above-average transcript count plus identified as doublets by scds R package v1.4.0 (Bais and Kostka, 2020) on R v4.0.4 (https://www.r-project.org/), and/or cells with more than $10\%$ transcripts mapping to mitochondrial genes were removed. The genes equivalent to those removed in our mouse data set plus V(D)J genes of TCRγ and TCRδ genes were excluded. Centered log-ratio normalization of cell surface protein counts for data sets that have them, log-normalization (scale factor = 1 × 104) and identification of variable features for each data set (2500 variable features) were performed using Seurat R package v4.0.0 (Haberman et al., 2014). The same package was used for the integration of all data sets and subsequent scaling, dimensional reduction (PCA and UMAP with top 20 principal components), nearest neighbors identification and Louvain clustering (resolution = 0.4) of the resulting data set. The newly emerging cluster of cells with below-average gene count and above-average proportion of transcripts mapping to mitochondrial genes was removed. In total, 140,564 cells passed the QC steps. Afterward, the previous steps starting from and including both normalizations were repeated using the same parameters. A cluster containing 5160 MAIT cells was identified using differential expression analysis and V(D)J information, which was kept for some analyses, but removed for others. OPTICs method from dbscan R package v1.1–6 (Hahsler, 2019) was applied (parameters minPts = 500 and eps_cl =.55 for data both with and without MAIT cells) to generate the final clustering. The code for building the atlas and its whole analysis is available on GitHub (https://github.com/Lab-of-Adaptive-Immunity/HS-CD8-Atlas; Lab of Adaptive Immunity, 2023b, copy archived at swh:1:rev:9b31b54fff516eba2a3ddb66449eb100db16521b). ## Statistical analysis The number of biological replicates (mice) is shown in the respective figure legends. The data are pooled from two or more independent experiments. Statistical analysis was performed using an appropriate test for given type of data using GraphPad Prism 5.0 or R. Quantitative data from mice were usually tested using nonparametric Mann–Whitney test or, for a comparison of more than two sample groups, Kruskal–Wallis test with Dunn’s posttest, if required. Unpaired Student’s t test was used only once for the statistical analysis of the size of cell clusters from the scRNAseq experiments, where the limited number of biological replicates ($$n = 3$$) did not allow a nonparametric test with reasonable statistical power. However, the results were subsequently confirmed using independent assays with more biological replicates. Mann–Whitney test with Bonferroni correction (adjusted p-value) for multiple comparisons was used for the statistical analysis of differentially expressed genes in the scRNAseq data. The survival/disease-free curves were analyzed by log-rank (Mantel–Cox) test. The statistical test is indicated for each experiment in figure legends. All tests were two-tailed. When applicable, the correction for multiple comparisons was applied. ## Funding Information This paper was supported by the following grants: ## Data availability All scRNA data analyzed in this study as well as the scripts used for the analysis are available without restrictions. The scRNAseq data generated in this study were deposited in the Gene Expression Omnibus (GSE183940). The following dataset was generated: TsyklauriO ChadimovaT NiederlovaV MichalikJ JanusovaS RossezH DrobekA VecerovaH GalantiV KovarM StepanekO 2022Regulatory T cells suppress the formation of super-effector CD8 T cells by limiting IL-2NCBI Gene Expression OmnibusGSE183940 The following previously published datasets were used: 10x Genomics 2019CD8+ T cells of Healthy Donor 110xGenomicscd-8-plus-t-cells-of-healthy-donor-1-1-standard-3-0-2 10x Genomics 2019CD8+ T cells of Healthy Donor 210xGenomicscd-8-plus-t-cells-of-healthy-donor-2-1-standard-3-0-2 10x Genomics 2019CD8+ T cells of Healthy Donor 310xGenomicscd-8-plus-t-cells-of-healthy-donor-3-1-standard-3-0-2 10x Genomics 2019CD8+ T cells of Healthy Donor 410xGenomicscd-8-plus-t-cells-of-healthy-donor-4-1-standard-3-0-2 10x Genomics 202110k Human PBMCs with TotalSeq-B Human TBNK Antibody Cocktail, 3' v3.110xGenomics10-k-human-pbm-cs-with-total-seq-b-human-tbnk-antibody-cocktail-3-v-3-1-3-1-standard-6-0-0 10x Genomics 202110k Human PBMCs Multiplexed, 2 CMOs - Inputs/Library10xGenomics10-k-human-pbm-cs-multiplexed-2-cm-os-3-1-standard-6-0-0 10x Genomics 20195k Peripheral blood mononuclear cells (PBMCs) from a healthy donor (v3 chemistry)10xGenomics5-k-peripheral-blood-mononuclear-cells-pbm-cs-from-a-healthy-donor-v-3-chemistry-3-1-standard-3-0-2 10x Genomics 201810k PBMCs from a Healthy Donor - Gene Expression with a Panel of TotalSeq-B Antibodies10xGenomics10-k-pbm-cs-from-a-healthy-donor-gene-expression-and-cell-surface-protein-3-standard-3-0-0 10x Genomics 20178k PBMCs from a Healthy Donor10xGenomics8-k-pbm-cs-from-a-healthy-donor-2-standard-2-1-0 10x Genomics 2019PBMCs of a Healthy Donor (v1)10xGenomicspbm-cs-of-a-healthy-donor-v-1-1-1-standard-3-1-0 10x Genomics 2020Human T cells from a Healthy Donor, 1k cells - multi (v2)10xGenomicshuman-t-cells-from-a-healthy-donor-1-k-cells-multi-v-2-2-standard-5-0-0 10x Genomics 2020Human PBMC from a Healthy Donor, 10k cells - multi (v2)10xGenomicshuman-pbmc-from-a-healthy-donor-10-k-cells-multi-v-2-2-standard-5-0-0 10x Genomics 2020Human PBMC from a Healthy Donor, 1k cells (v2)10xGenomicshuman-pbmc-from-a-healthy-donor-1-k-cells-v-2-2-standard-4-0-0 10x Genomics 2018PBMCs of a Healthy Donor - 5' Gene Expression with a Panel of TotalSeq-C Antibodies10xGenomicspbm-cs-of-a-healthy-donor-5-gene-expression-and-cell-surface-protein-1-standard-3-0-0 ## References 1. Akkaya B, Oya Y, Akkaya M, Al Souz JA, Holstein AH, Kabat J, Kamenyeva O, Dorward D, Glass D, Shevach EM. Regulatory T cells perform antigen specific suppression by depleting cognate peptide-MHC class II via trogocytosis. The Journal of Immunology (2019) 202. DOI: 10.4049/jimmunol.202.Supp.57.17 2. Andreatta M, Corria-Osorio J, Müller S, Cubas R, Coukos G, Carmona SJ. Interpretation of T cell states from single-cell transcriptomics data using reference atlases. Nature Communications (2021) 12. DOI: 10.1038/s41467-021-23324-4 3. Bais AS, Kostka D. Scds: computational annotation of doublets in single-cell RNA sequencing data. Bioinformatics (2020) 36 1150-1158. DOI: 10.1093/bioinformatics/btz698 4. Barnden MJ, Allison J, Heath WR, Carbone FR. Defective TCR expression in transgenic mice constructed using cDNA-based alpha- and beta-chain genes under the control of heterologous regulatory elements. Immunology and Cell Biology (1998) 76 34-40. DOI: 10.1046/j.1440-1711.1998.00709.x 5. Barsoumian HB, Yolcu ES, Shirwan H. 4-1Bb signaling in conventional T cells drives IL-2 production that overcomes CD4+CD25+FoxP3+ T regulatory cell suppression. PLOS ONE (2016) 11. DOI: 10.1371/journal.pone.0153088 6. Bennett CL, Christie J, Ramsdell F, Brunkow ME, Ferguson PJ, Whitesell L, Kelly TE, Saulsbury FT, Chance PF, Ochs HD. The immune dysregulation, polyendocrinopathy, enteropathy, X-linked syndrome (IPEX) is caused by mutations of FOXP3. Nature Genetics (2001) 27 20-21. DOI: 10.1038/83713 7. Bosch AJT, Bolinger B, Keck S, Stepanek O, Ozga AJ, Galati-Fournier V, Stein JV, Palmer E. A minimum number of autoimmune T cells to induce autoimmunity?. Cellular Immunology (2017) 316 21-31. DOI: 10.1016/j.cellimm.2017.03.002 8. Boyman O, Kovar M, Rubinstein MP, Surh CD, Sprent J. Selective stimulation of T cell subsets with antibody-cytokine immune complexes. Science (2006) 311 1924-1927. DOI: 10.1126/science.1122927 9. Brunkow ME, Jeffery EW, Hjerrild KA, Paeper B, Clark LB, Yasayko SA, Wilkinson JE, Galas D, Ziegler SF, Ramsdell F. Disruption of a new forkhead/winged-helix protein, scurfin, results in the fatal lymphoproliferative disorder of the scurfy mouse. Nature Genetics (2001) 27 68-73. DOI: 10.1038/83784 10. Chinen T, Kannan AK, Levine AG, Fan X, Klein U, Zheng Y, Gasteiger G, Feng Y, Fontenot JD, Rudensky AY. An essential role for the IL-2 receptor in Treg cell function. Nature Immunology (2016) 17 1322-1333. DOI: 10.1038/ni.3540 11. Dong S, Hiam-Galvez KJ, Mowery CT, Herold KC, Gitelman SE, Esensten JH, Liu W, Lares AP, Leinbach AS, Lee M, Nguyen V, Tamaki SJ, Tamaki W, Tamaki CM, Mehdizadeh M, Putnam AL, Spitzer MH, Ye CJ, Tang Q, Bluestone JA. The effect of low-dose IL-2 and treg adoptive cell therapy in patients with type 1 diabetes. JCI Insight (2021) 6. DOI: 10.1172/jci.insight.147474 12. Green EA, Gorelik L, McGregor CM, Tran EH, Flavell RA. Cd4+Cd25+ T regulatory cells control anti-islet CD8+ T cells through TGF-beta-TGF-beta receptor interactions in type 1 diabetes. PNAS (2003) 100 10878-10883. DOI: 10.1073/pnas.1834400100 13. Haberman Y, Tickle TL, Dexheimer PJ, Kim M-O, Tang D, Karns R, Baldassano RN, Noe JD, Rosh J, Markowitz J, Heyman MB, Griffiths AM, Crandall WV, Mack DR, Baker SS, Huttenhower C, Keljo DJ, Hyams JS, Kugathasan S, Walters TD, Aronow B, Xavier RJ, Gevers D, Denson LA. Pediatric crohn disease patients exhibit specific ileal transcriptome and microbiome signature. The Journal of Clinical Investigation (2014) 124 3617-3633. DOI: 10.1172/JCI75436 14. Hahsler M. Dbscan: fast density-based clustering with R. Journal of Statistical Software (2019) 91 1-30. DOI: 10.18637/jss.v091.i01 15. Howe KL, Achuthan P, Allen J, Allen J, Alvarez-Jarreta J, Amode MR, Armean IM, Azov AG, Bennett R, Bhai J, Billis K, Boddu S, Charkhchi M, Cummins C, Da Rin Fioretto L, Davidson C, Dodiya K, El Houdaigui B, Fatima R, Gall A, Garcia Giron C, Grego T, Guijarro-Clarke C, Haggerty L, Hemrom A, Hourlier T, Izuogu OG, Juettemann T, Kaikala V, Kay M, Lavidas I, Le T, Lemos D, Gonzalez Martinez J, Marugán JC, Maurel T, McMahon AC, Mohanan S, Moore B, Muffato M, Oheh DN, Paraschas D, Parker A, Parton A, Prosovetskaia I, Sakthivel MP, Salam AIA, Schmitt BM, Schuilenburg H, Sheppard D, Steed E, Szpak M, Szuba M, Taylor K, Thormann A, Threadgold G, Walts B, Winterbottom A, Chakiachvili M, Chaubal A, De Silva N, Flint B, Frankish A, Hunt SE, IIsley GR, Langridge N, Loveland JE, Martin FJ, Mudge JM, Morales J, Perry E, Ruffier M, Tate J, Thybert D, Trevanion SJ, Cunningham F, Yates AD, Zerbino DR, Flicek P. Ensembl 2021. Nucleic Acids Research (2021) 49 D884-D891. DOI: 10.1093/nar/gkaa942 16. Jang Y, Gerbec ZJ, Won T, Choi B, Podsiad A, B Moore B, Malarkannan S, Laouar Y. Cutting edge: check your mice-A point mutation in the ncr1 locus identified in CD45.1 congenic mice with consequences in mouse susceptibility to infection. Journal of Immunology (2018) 200 1982-1987. DOI: 10.4049/jimmunol.1701676 17. Josefowicz SZ, Rudensky A. Control of regulatory T cell lineage commitment and maintenance. Immunity (2009) 30 616-625. DOI: 10.1016/j.immuni.2009.04.009 18. Joshi NS, Cui W, Chandele A, Lee HK, Urso DR, Hagman J, Gapin L, Kaech SM. Inflammation directs memory precursor and short-lived effector CD8 (+) T cell fates via the graded expression of T-bet transcription factor. Immunity (2007) 27 281-295. DOI: 10.1016/j.immuni.2007.07.010 19. Kaech SM, Tan JT, Wherry EJ, Konieczny BT, Surh CD, Ahmed R. Selective expression of the interleukin 7 receptor identifies effector CD8 T cells that give rise to long-lived memory cells. Nature Immunology (2003) 4 1191-1198. DOI: 10.1038/ni1009 20. Kalia V, Penny LA, Yuzefpolskiy Y, Baumann FM, Sarkar S. Quiescence of memory CD8 (+) T cells is mediated by regulatory T cells through inhibitory receptor CTLA-4. Immunity (2015) 42 1116-1129. DOI: 10.1016/j.immuni.2015.05.023 21. Kastenmuller W, Gasteiger G, Subramanian N, Sparwasser T, Busch DH, Belkaid Y, Drexler I, Germain RN. Regulatory T cells selectively control CD8+ T cell effector pool size via IL-2 restriction. Journal of Immunology (2011) 187 3186-3197. DOI: 10.4049/jimmunol.1101649 22. Kim JM, Rasmussen JP, Rudensky AY. Regulatory T cells prevent catastrophic autoimmunity throughout the lifespan of mice. Nature Immunology (2007) 8 191-197. DOI: 10.1038/ni1428 23. Kim M, Ouyang W, Liao W, Zhang M, Li M. Murine in vivo CD8+ T cell killing assay. BIO-PROTOCOL (2014) 4. DOI: 10.21769/BioProtoc.1172 24. Kim R, Kin T. Current and future therapies for immunogenic cell death and related molecules to potentially cure primary breast cancer. Cancers (2021) 13. DOI: 10.3390/cancers13194756 25. King CG, Koehli S, Hausmann B, Schmaler M, Zehn D, Palmer E. T cell affinity regulates asymmetric division, effector cell differentiation, and tissue pathology. Immunity (2012) 37 709-720. DOI: 10.1016/j.immuni.2012.06.021 26. Korotkevich G. Fast Gene Set Enrichment Analysis. bioRxiv (2021). DOI: 10.1101/060012 27. Kovanen PE, Young L, Al-Shami A, Rovella V, Pise-Masison CA, Radonovich MF, Powell J, Fu J, Brady JN, Munson PJ, Leonard WJ. Global analysis of IL-2 target genes: identification of chromosomal clusters of expressed genes. International Immunology (2005) 17 1009-1021. DOI: 10.1093/intimm/dxh283 28. Kralova J, Glatzova D, Borna S, Brdicka T. Expression of fluorescent fusion proteins in murine bone marrow-derived dendritic cells and macrophages. Journal of Visualized Experiments (2018) 2018. DOI: 10.3791/58081-v 29. Kurts C, Miller JF, Subramaniam RM, Carbone FR, Heath WR. Major histocompatibility complex class I-restricted cross-presentation is biased towards high dose antigens and those released during cellular destruction. The Journal of Experimental Medicine (1998) 188 409-414. DOI: 10.1084/jem.188.2.409 30. Lab of Adaptive Immunity 2023aSupereffectors_scRNAseqswh:1:rev:7b0dec9507dd45ab4bb0619912b240f512c7798fSoftware Heritagehttps://archive.softwareheritage.org/swh:1:dir:a24b54a92579b7b10530c1119bc81ae22e433076;origin=https://github.com/Lab-of-Adaptive-Immunity/Supereffectors_scRNAseq;visit=swh:1:snp:2315a288b71fbc93e22b85ae7924b45146e00ab9;anchor=swh:1:rev:7b0dec9507dd45ab4bb0619912b240f512c7798f. Software Heritage (2023a) 31. Lab of Adaptive Immunity 2023bHS-CD8-atlasswh:1:rev:9b31b54fff516eba2a3ddb66449eb100db16521bSoftware Heritagehttps://archive.softwareheritage.org/swh:1:dir:1890b7a68107682d810bbd65ce12920c1a8747de;origin=https://github.com/Lab-of-Adaptive-Immunity/HS-CD8-Atlas;visit=swh:1:snp:f6b1a4ee7d7e0ba3e4e153f996f0a34d4f318945;anchor=swh:1:rev:9b31b54fff516eba2a3ddb66449eb100db16521b. Software Heritage (2023b) 32. Lahl K, Loddenkemper C, Drouin C, Freyer J, Arnason J, Eberl G, Hamann A, Wagner H, Huehn J, Sparwasser T. Selective depletion of Foxp3+ regulatory T cells induces a scurfy-like disease. The Journal of Experimental Medicine (2007) 204 57-63. DOI: 10.1084/jem.20061852 33. Laidlaw BJ, Cui W, Amezquita RA, Gray SM, Guan T, Lu Y, Kobayashi Y, Flavell RA, Kleinstein SH, Craft J, Kaech SM. Production of IL-10 by CD4 (+) regulatory T cells during the resolution of infection promotes the maturation of memory CD8 (+) T cells. Nature Immunology (2015) 16 871-879. DOI: 10.1038/ni.3224 34. Lee SJ, Rossi RJ, Lee SK, Croft M, Kwon BS, Mittler RS, Vella AT. Cd134 costimulation couples the CD137 pathway to induce production of supereffector CD8 T cells that become IL-7 dependent. Journal of Immunology (2007) 179 2203-2214. DOI: 10.4049/jimmunol.179.4.2203 35. Lin J-X, Li P, Liu D, Jin HT, He J, Ata Ur Rasheed M, Rochman Y, Wang L, Cui K, Liu C, Kelsall BL, Ahmed R, Leonard WJ. Critical role of STAT5 transcription factor tetramerization for cytokine responses and normal immune function. Immunity (2012) 36 586-599. DOI: 10.1016/j.immuni.2012.02.017 36. Liu R, Zhou Q, La Cava A, Campagnolo DI, Van Kaer L, Shi F-D. Expansion of regulatory T cells via IL-2/anti-IL-2 mAb complexes suppresses experimental myasthenia. European Journal of Immunology (2010) 40 1577-1589. DOI: 10.1002/eji.200939792 37. Marchingo JM, Kan A, Sutherland RM, Duffy KR, Wellard CJ, Belz GT, Lew AM, Dowling MR, Heinzel S, Hodgkin PD. Antigen affinity, costimulation, and cytokine inputs sum linearly to amplify T cell expansion. Science (2014) 346 1123-1127. DOI: 10.1126/science.1260044 38. McNally A, Hill GR, Sparwasser T, Thomas R, Steptoe RJ. Cd4+Cd25+ regulatory T cells control CD8+ T-cell effector differentiation by modulating IL-2 homeostasis. PNAS (2011) 108 7529-7534. DOI: 10.1073/pnas.1103782108 39. Obeid M, Tesniere A, Ghiringhelli F, Fimia GM, Apetoh L, Perfettini J-L, Castedo M, Mignot G, Panaretakis T, Casares N, Métivier D, Larochette N, van Endert P, Ciccosanti F, Piacentini M, Zitvogel L, Kroemer G. Calreticulin exposure dictates the immunogenicity of cancer cell death. Nature Medicine (2007) 13 54-61. DOI: 10.1038/nm1523 40. Oh HS, Choi BK, Kim YH, Lee DG, Hwang S, Lee MJ, Park SH, Bae Y-S, Kwon BS. 4-1Bb signaling enhances primary and secondary population expansion of CD8+ T cells by maximizing autocrine IL-2/IL-2 receptor signaling. PLOS ONE (2015) 10. DOI: 10.1371/journal.pone.0126765 41. Pace L, Tempez A, Arnold-Schrauf C, Lemaitre F, Bousso P, Fetler L, Sparwasser T, Amigorena S. Regulatory T cells increase the avidity of primary CD8+ T cell responses and promote memory. Science (2012) 338 532-536. DOI: 10.1126/science.1227049 42. Palmer E, Drobek A, Stepanek O. Opposing effects of actin signaling and LFA-1 on establishing the affinity threshold for inducing effector T-cell responses in mice. European Journal of Immunology (2016) 46 1887-1901. DOI: 10.1002/eji.201545909 43. Polhill T, Zhang GY, Hu M, Sawyer A, Zhou JJ, Saito M, Webster KE, Wang Y, Wang Y, Grey ST, Sprent J, Harris DCH, Alexander SI, Wang YM. IL-2/IL-2ab complexes induce regulatory T cell expansion and protect against proteinuric CKD. Journal of the American Society of Nephrology (2012) 23 1303-1308. DOI: 10.1681/ASN.2011111130 44. Prajapati K, Perez C, Rojas LBP, Burke B, Guevara-Patino JA. Functions of NKG2D in CD8+ T cells: an opportunity for immunotherapy. Cellular & Molecular Immunology (2018) 15 470-479. DOI: 10.1038/cmi.2017.161 45. Schildknecht A, Brauer S, Brenner C, Lahl K, Schild H, Sparwasser T, Probst HC, van den Broek M. Foxp3+ regulatory T cells essentially contribute to peripheral CD8+ T-cell tolerance induced by steady-state dendritic cells. PNAS (2010) 107 199-203. DOI: 10.1073/pnas.0910620107 46. Schorer M, Lambert K, Rakebrandt N, Rost F, Kao KC, Yermanos A, Spörri R, Oderbolz J, Raeber ME, Keller CW, Lünemann JD, Rogler G, Boyman O, Oxenius A, Joller N. Rapid expansion of treg cells protects from collateral colitis following a viral trigger. Nature Communications (2020) 11. DOI: 10.1038/s41467-020-15309-6 47. Sereti I, Gea-Banacloche J, Kan MY, Hallahan CW, Lane HC. Interleukin 2 leads to dose-dependent expression of the alpha chain of the IL-2 receptor on CD25-negative T lymphocytes in the absence of exogenous antigenic stimulation. Clinical Immunology (2000) 97 266-276. DOI: 10.1006/clim.2000.4929 48. Shinkai Y, Rathbun G, Lam KP, Oltz EM, Stewart V, Mendelsohn M, Charron J, Datta M, Young F, Stall AM. RAG-2-deficient mice lack mature lymphocytes owing to inability to initiate V (D) J rearrangement. Cell (1992) 68 855-867. DOI: 10.1016/0092-8674(92)90029-c 49. Slavin S, Strober S. Spontaneous murine B-cell leukaemia. Nature (1978) 272 624-626. DOI: 10.1038/272624a0 50. Spangler JB, Tomala J, Luca VC, Jude KM, Dong S, Ring AM, Votavova P, Pepper M, Kovar M, Garcia KC. Antibodies to interleukin-2 elicit selective T cell subset potentiation through distinct conformational mechanisms. Immunity (2015) 42 815-825. DOI: 10.1016/j.immuni.2015.04.015 51. Stepanek O, Prabhakar AS, Osswald C, King CG, Bulek A, Naeher D, Beaufils-Hugot M, Abanto ML, Galati V, Hausmann B, Lang R, Cole DK, Huseby ES, Sewell AK, Chakraborty AK, Palmer E. Coreceptor scanning by the T cell receptor provides a mechanism for T cell tolerance. Cell (2014) 159 333-345. DOI: 10.1016/j.cell.2014.08.042 52. Togashi Y, Shitara K, Nishikawa H. Regulatory T cells in cancer immunosuppression-implications for anticancer therapy. Nature Reviews. Clinical Oncology (2019) 16 356-371. DOI: 10.1038/s41571-019-0175-7 53. Tomala J, Kovar M. IL-2/anti-IL-2 mAb immunocomplexes: a renascence of IL-2 in cancer immunotherapy?. Oncoimmunology (2016) 5. DOI: 10.1080/2162402X.2015.1102829 54. Wildin RS, Ramsdell F, Peake J, Faravelli F, Casanova JL, Buist N, Levy-Lahad E, Mazzella M, Goulet O, Perroni L, Bricarelli FD, Byrne G, McEuen M, Proll S, Appleby M, Brunkow ME. X-Linked neonatal diabetes mellitus, enteropathy and endocrinopathy syndrome is the human equivalent of mouse scurfy. Nature Genetics (2001) 27 18-20. DOI: 10.1038/83707 55. Wong HS, Park K, Gola A, Baptista AP, Miller CH, Deep D, Lou M, Boyd LF, Rudensky AY, Savage PA, Altan-Bonnet G, Tsang JS, Germain RN. A local regulatory T cell feedback circuit maintains immune homeostasis by pruning self-activated T cells. Cell (2021) 184 3981-3997. DOI: 10.1016/j.cell.2021.05.028 56. Zheng GXY, Terry JM, Belgrader P, Ryvkin P, Bent ZW, Wilson R, Ziraldo SB, Wheeler TD, McDermott GP, Zhu J, Gregory MT, Shuga J, Montesclaros L, Underwood JG, Masquelier DA, Nishimura SY, Schnall-Levin M, Wyatt PW, Hindson CM, Bharadwaj R, Wong A, Ness KD, Beppu LW, Deeg HJ, McFarland C, Loeb KR, Valente WJ, Ericson NG, Stevens EA, Radich JP, Mikkelsen TS, Hindson BJ, Bielas JH. Massively parallel digital transcriptional profiling of single cells. Nature Communications (2017) 8. DOI: 10.1038/ncomms14049
--- title: Gasotransmitter modulation of hypoglossal motoneuron activity authors: - Brigitte M Browe - Ying-Jie Peng - Jayasri Nanduri - Nanduri R Prabhakar - Alfredo J Garcia journal: eLife year: 2023 pmcid: PMC9977277 doi: 10.7554/eLife.81978 license: CC BY 4.0 --- # Gasotransmitter modulation of hypoglossal motoneuron activity ## Abstract Obstructive sleep apnea (OSA) is characterized by sporadic collapse of the upper airway leading to periodic disruptions in breathing. Upper airway patency is governed by genioglossal nerve activity that originates from the hypoglossal motor nucleus. Mice with targeted deletion of the gene Hmox2, encoding the carbon monoxide (CO) producing enzyme, heme oxygenase-2 (HO-2), exhibit OSA, yet the contribution of central HO-2 dysregulation to the phenomenon is unknown. Using the rhythmic brainstem slice preparation that contains the preBötzinger complex (preBötC) and the hypoglossal nucleus, we tested the hypothesis that central HO-2 dysregulation weakens hypoglossal motoneuron output. Disrupting HO-2 activity increased the occurrence of subnetwork activity from the preBötC, which was associated with an increased irregularity of rhythmogenesis. These phenomena were also associated with the intermittent inability of the preBötC rhythm to drive output from the hypoglossal nucleus (i.e. transmission failures), and a reduction in the input-output relationship between the preBötC and the motor nucleus. HO-2 dysregulation reduced excitatory synaptic currents and intrinsic excitability in inspiratory hypoglossal neurons. Inhibiting activity of the CO-regulated H2S producing enzyme, cystathionine-γ-lyase (CSE), reduced transmission failures in HO-2 null brainstem slices, which also normalized excitatory synaptic currents and intrinsic excitability of hypoglossal motoneurons. These findings demonstrate a hitherto uncharacterized modulation of hypoglossal activity through mutual interaction of HO-2/CO and CSE/H2S, and support the potential importance of centrally derived gasotransmitter activity in regulating upper airway control. ## Introduction Sporadic airway collapse is a hallmark of obstructive sleep apnea (OSA), a prevalent breathing disorder estimated to affect nearly a billion people throughout the world Lyons et al., 2020; Malhotra et al., 2021. When left untreated, OSA predisposes the individual to a variety of diseases including hypertension Mehra, 2019; Yeghiazarians et al., 2021, diabetes Loffler et al., 2020; Hua, 2020, and cognitive decline Daulatzai, 2017; Bibbins-Domingo et al., 2017. Multiple factors contribute to the genesis of OSA including compromised pharyngeal anatomy Castro and Freeman, 2021; Genta et al., 2017, inadequate upper airway muscle function Neelapu et al., 2017; Vos et al., 2010; Kubin, 2016, low arousal threshold Eckert et al., 2013, and a hypersensitive chemoreflex (i.e. high loop gain) Nemati et al., 2011. Peng et al. recently reported that mice with deletion of the Hmox2 gene, which encodes the enzyme heme oxygenase 2 (HO-2), exhibit a high incidence of OSA Peng et al., 2017. OSA in HO-2 null mice was attributed, in part, to increased loop gain arising from the heightened carotid body chemoreflex Peng et al., 2017; Peng et al., 2018; Osman et al., 2018; Prabhakar and Semenza, 2012. While HO-2 activity produces several bioactive molecules Mancuso, 2004, the loss of HO-2 dependent carbon monoxide (CO) production was shown to be a primary driver of the enhanced carotid body chemoreflex and the subsequent OSA phenotype Peng et al., 2017. However, output from the hypoglossal motor pool can be influenced by multiple factors, including HO-2-mediated action from within the central nervous system itself. Loss of neuromuscular control over upper airway muscles has a key role in producing obstructive apneas Schwartz et al., 2008; Horner, 2001. Disrupting neuronal excitability in the hypoglossal nucleus that is responsible for genioglossal nerve activity increases the likelihood for the tongue to occlude the upper airway during inspiration. Such disruptions may involve changing the state-dependent balance between excitation and inhibition received by hypoglossal motoneurons Horner, 2009 and/or by directly modulating their intrinsic excitability Horton et al., 2017; Fleury Curado et al., 2017; Fleury Curado et al., 2018. It is, however, unknown how HO-2 signaling at the level of the preBötC and the hypoglossal nucleus influence these factors contributing to upper airway tone and patency. We tested the role of central HO-2 signaling in influencing hypoglossal motor output using a combination of electrophysiological, genetic, and pharmacological approaches in rhythmic medullary brainstem slice preparations. Dysregulated HO-2 activity increased the occurrence of subnetwork activity in the preBötC, which was associated with an increased cycle-to-cycle irregularity of rhythmogenesis. In hypoglossal motoneurons, excitatory synaptic drive currents and intrinsic excitability were reduced by HO-2 dysregulation. These phenomena also coincided with circuit level effects. HO-2 dysregulation diminished the input-output relationship and increased the likelihood of transmission failure between the preBötC activity and the hypoglossal nucleus. These effects of HO-2 dysregulation could be mimicked by exogenous H2S and mitigated by either pharmacological inhibition or genetic ablation of CSE. Together these observations indicate that centrally derived HO-2/CO and CSE/H2S signaling interact as important modulators of hypoglossal output contributing to upper airway tone. ## Hypoglossal neurons express hemeoxygenase-2 (HO-2) We assessed whether hypoglossal neurons express HO-2. Hypoglossal neurons showed positive immunohistochemical expression for HO-2 as indicated by co-localization of HO-2 with ChAT, a motoneuron marker (Figure 1A, $$n = 3$$). Figure 1.: Disruption of hemeoxygenase-2 (HO-2) impairs inspiratory activity from the hypoglossal nucleus and the preBötC.(A) HO-2 (red bottom left) expression co-localized to ChAT+ cells (green bottom right) of the hypoglossal nucleus (XIIn, overlay top, n=3). (B–F) Population recordings of rhythmic brain slices were made from ipsilateral preBötC and XIIn simultaneously, analyses were performed in baseline and during bath application of 20 µM ChrMP459. (B) Integrated traces of network activity in spontaneously rhythmic brainstem slices (n=34) recorded from XIIn (gray) and preBötC (black) before (top) and during (bottom) ChrMP459. Failed transmission events are highlighted (pink box) and subnetwork preBötC activity (#, defined as preBötC events with an integrated burst area ≤50% of the mean integrated burst area in baseline and are #) are evident in ChrMP459. Scale bar: 5 s. (C) Comparison of integrated amplitude irregularity score (IrSAMP) during baseline and in ChrMP459 from the XIIn (top) and the preBötC (bottom, n=34). Solid lines within violin plots illustrate IrSAMP from individual experiments. Thick dashed line illustrates mean IrSAMP. (D) Heat maps of I/O ratios from 25 consecutive cycles in baseline and ChrMP459. Each row reflects an individual experiment while each cell represents the I/O ratio for a given cycle. Gray boxes indicate non-events from recordings from slower rhythms where less than 25 cycles occurred during the analysis window. (E) Comparison of transmission from preBötC to XIIn between Baseline (gray) and ChrMP459 (purple). Solid lines within violin plots illustrate transmission from individual slices. Thick dashed line illustrates mean transmission value. (F) Distribution of transmitted (blue) and untransmitted (purple) preBötC bursts in ChrMP459. These values are expressed as a percentage of the total of preBötC events detected from all experiments and are binned. Bin interval = 0.1 intervals of the normalized integrated burst area. preBötC bursts were normalization to the mean integrated burst area during baseline. Statistical analysis for all comparisons via paired t-test; error bars: Standard Error Measurement (SEM); significance level P<0.05. ## Disrupting HO-2 function impairs hypoglossal inspiratory activity Two approaches were employed to assess the role of HO-2. First, using Cr(III) Mesoporphyrin IX chloride (ChrMP459, 20 µM) as a pharmacological inhibitor of HO Ding et al., 2008 in wild-type slices and second, with a genetic approach using brain slices from HO-2 null mice. Simultaneous extracellular field recordings were performed from preBötC and the hypoglossal nucleus in wild-type slices prior to and during ChrMP459 exposure ($$n = 34$$ slices). ChrMP459 promoted the emergence of subnetwork activity (i.e., integrated bursts ≤ $50\%$ of mean integrated burst amplitude during baseline; demarcated by hash symbol, Figure 1B) in the preBötC. This subnetwork preBötC activity commonly failed to drive corresponding output hypoglossal nucleus (highlighted in pink, Figure 1B) and was associated with an increase in the irregularity score of amplitude (IrSAMP) in the preBötC (Figure 1C, bottom; Baseline: 0.279±0.39, ChrMP459: 0.406±0.05, $$p \leq 0.021$$). However, ChrMP459 did not impact the IrSAMP in hypoglossal nucleus (Figure 1C, top; Baseline: 0.385±0.05, ChrMP459: 0.372±0.03, $$p \leq 0.701$$) nor affect the frequency or amplitude of the integrated bursts from either brainstem network (Table 1). Table 1. \| Unnamed: 0 \| Unnamed: 1 \| finst (Hz) \| finst (Hz).1 \| finst (Hz).2 \| Burst Amplitude (mV) \| Burst Amplitude (mV).1 \| Burst Amplitude (mV).2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Experiment \| Recording Location \| Control* (n) \| Dysregulated HO-2 (n) \| p-value \| Control* (n) \| Dysregulated HO-2 (n) \| p-value \| \| ChrMP459 \| preBötC \| 0.225±0.014 (34) \| 0.242±0.019 (34) \| 0.139 \| 0.078±0.011 (34) \| 0.08±0.012 (34) \| 0.626 \| \| HO-2 null slice \| preBötC \| 0.235±0.022 (18) \| 0.420±0.049 (11) \| 0.0006 \| 0.059±0.007 (18) \| 0.051±0.010 (11) \| 0.463 \| \| ChrMP459 \| XIIn \| 0.239±0.015 (34) \| 0.221±0.015 (34) \| 0.044 \| 0.047±0.008 (34) \| 0.046±0.009 (34) \| 0.804 \| \| HO-2 null slice \| XIIn \| 0.256±0.018 (18) \| 0.432±0.051 (11) \| 0.0006 \| 0.038±0.006 (18) \| 0.021±0.003 (11) \| 0.050 \| ChrMP459 caused a consistent reduction in the cycle-to-cycle input-output relationship between preBötC and the hypoglossal nucleus (Figure 1D) which was quantified by comparing input-output (I/O) ratios prior to and during ChrMP459 (Baseline: 1.06±0.04 vs. ChrMP459: 0.592±0.06; $p \leq 0.0001$). A reduction of input-output relationship between preBötC and the hypoglossal nucleus has been previously associated with increased transmission failure, which is defined by the inability of preBötC activity to produce hypoglossal output at the network level Garcia et al., 2016. The reduced I/O ratio in ChrMP459 was, indeed, associated with increased transmission failure (Figure 1E, Baseline: 94.130±$1.27\%$ vs. ChrMP459: 75.100 ± $3.43\%$, $p \leq 0.0001$). Examining the relationship between failed transmission and the burst area in the preBötC revealed that while the many transmission failures were associated with subnetwork preBötC activity (i.e. integrated burst areas ≤ $50\%$), these events were not restricted to subnetwork activity but rather, occurred across a range of integrated burst areas from the inspiratory network (Figure 1F). Thus, HO inhibition appeared to produce a generalized weakening in the relationship between preBötC and hypoglossal nucleus activity evident across the spectrum of different burst areas generated in the preBötC. As ChrMP459 cannot distinguish activities between heme oxygenase isoforms, we compared rhythmic activities in brain slices from wild-type mice ($$n = 18$$) and HO-2 null mice ($$n = 11$$) to assess the contribution of HO-2. Both the intermittent occurrence of subnetwork activity in the preBötC (demarcated by hash symbol, Figure 2A) and failed transmission events were observed in HO-2 null slices (highlighted in pink, Figure 2A). preBötC activity in HO-2 null mice was associated with an increased IrSAMP in both the preBötC (Figure 2B, WT: 0.323±0.02, HO-2 null: 0.419±0.03, p+0.014) and the hypoglossal nucleus (Figure 2C, WT: 0.269±0.03, HO-2 null: 0.534±0.12, $$p \leq 0.018$$). While the frequency was faster in HO-2 null rhythms relative to wild-type rhythms, the burst amplitude was not different in either preBötC or the hypoglossal (Table 1). Figure 2.: Genetic deletion of HO-2 reduces the I/O relationship between preBötC and the hypoglossal nucleus and uncouples of motor output from inspiratory rhythmogenesis.(A) Representative integrated traces of network rhythms in the preBötC and XIIn from wild-type (WT; left, n=18) and HO-2 null (right, n=11) slices. Failed transmissions (pink box) and subnetwork preBötC activity (#) are evident in HO-2 null slices. Scale bar: 4 s. (B) Comparison of IrSAMP in the preBötC of WT (blue) and HO-2 null (red) slices. (C) Comparison of IrSAMP in the XIIn of WT (blue) and HO-2 null (red) slices. (D) Heat maps of cycle-to-cycle I/O ratios from individual experiments performed in WT (left) and HO-2 null (right) slices. Gray boxes indicate non-events in recordings from slower rhythms where less than 25 events occurred during the analysis window. (E) Comparison of transmission from preBötC to XIIn between WT (blue) and HO-2 null (red) slices. Thin gray and purple lines illustrate transmission from individual slices. Thick dashed lines illustrate mean transmission value. (F) Distribution of transmitted (gray) and untransmitted (red) preBötC bursts in HO-2 null slices. These values are expressed as a percentage of the total of preBötC events detected from all experiments and are binned. Bin interval = 0.1 intervals of the normalized integrated burst area. preBötC bursts were normalization to the mean integrated burst area from each individual recording. Statistical analysis for all comparisons via unpaired t-test; error bars: SEM; significance level P<0.05. Relative to wild-type slices, the HO-2 null preparations had smaller cycle-to-cycle I/O ratios (Figure 2D; wild-type: 1.023±0.03 vs. HO-2 null: 0.799±0.07, $$p \leq 0.002$$); this was accompanied by a smaller percentage of transmission in HO-2 null slices (Figure 2E, wild-type: 98.210±$0.87\%$ vs. HO-2 null: 64.110 ± $6.00\%$, $p \leq 0.0001$). Similar to ChrMP459 experiments, while subnetwork activity led to many failed transmission events, failed transmission occurred across a range of burst areas from the preBötC (Figure 2F). These findings were consistent with ChrMP459 findings and illustrated that lost HO-2 activity is sufficient for promoting subnetwork preBötC activity, reducing input-output relationship between preBötC and the hypoglossal nucleus, and increasing transmission failures. Given these similarities and the limited availability of HO-2 null mice, several of the following studies were performed using the ChrMP459 in rhythmic wild-type brainstem slices. ## HO inhibition does not affect intermediate premotor neuron activity Intermediate premotor neurons relay drive from the preBötC to the hypoglossal nucleus Koizumi et al., 2013; Revill et al., 2015. Therefore, it was possible that HO inhibition impaired transmission of drive from the preBötC by perturbing activity from intermediate premotor neurons. To address this possibility, simultaneous extracellular recordings ($$n = 5$$) were made from the preBötC, the field of the ipsilateral premotor neurons, and the hypoglossal nucleus (Figure 3A, left panel). Baseline transmission from the preBötC to the premotor field and to the hypoglossal nucleus was reliable and consistent (Figure 3A, middle panel). In ChrMP459, intermittent failures of hypoglossal nucleus activity corresponding preBötC activity were evident yet during these failed cycles preBötC activity still produced detectable network activity in the field of intermediary premotor neurons (Figure 3A, right panel). Indeed, while neither the cycle-to-cycle I/O ratio nor transmission from the preBötC to the premotor field was affected by ChrMP459 (Figure 3B: left; I/O ratio: Baseline: 1.116±0.09 vs ChrMP 1.162±0.15, $$p \leq 0.816$$; right; Transmission: Baseline: 100.0±$0.0\%$ vs ChrMP 86.350 ± $11.81\%$, $$p \leq 0.312$$), the HO inhibitor reduced the cycle-to-cycle I/O ratio and the transmission between the premotor field and the hypoglossal nucleus (Figure 3C: left, I/O: Baseline: 1.102±0.18 vs ChrMP 0.401±0.09, $$p \leq 0.041$$; right, Transmission: Baseline 89.570±$4.34\%$ vs ChrMP 57.620 ± $13.69\%$, $$p \leq 0.041$$). Thus, these findings suggested that ChrMP459 potentially affected synaptic properties to the hypoglossal motoneurons and/or the intrinsic excitability of hypoglossal neurons. Figure 3.: While ChrMP459 does not change transmission from the preBötC to the premotor area, ChrMP459 increases transmission failure from the premotor area to the hypoglossal nucleus.(A) Diagram of medullary brain slice illustrating relative electrode placement for simultaneous triple extracellular recordings (n=5) from the XIIn (light gray, 1), premotor field (dark gray, 2), and preBötC (black, 3). Corresponding representative traces of integrated network activity in Baseline (left) and in 20 μM ChrMP459 (right). Failed transmission (pink box) from preBötC to XIIn and preBötC subnetwork burst activity (#) are evident in ChrMP459; scale bar: 5 s. (B) Heat maps of the cycle to cycle I/O ratio from individual slices (left) and transmission (right) between preBötC and the premotor field. (C) Heat maps of the cycle to cycle I/O ratio from individual slices (left) and transmission (right) between the premotor field and XIIn. Statistical analysis for all comparisons via paired t-test; error bars: SEM; significance level P<0.05. ## HO inhibition suppresses inspiratory drive currents and reduces excitability in hypoglossal neurons To assess the effect of HO inhibition on postsynaptic activity of the hypoglossal neurons, we performed patch clamp recordings from a total of 27 wild-type hypoglossal neurons exposed to ChrMP459. These hypoglossal neurons were disinhibited from fast inhibition using picrotoxin (50 μM) and strychnine (1 μM), which allowed us to focus on inspiratory-related fast glutamatergic drive. Of the 27 hypoglossal neurons, 19 received excitatory synaptic drive in-phase with the preBötC (i.e. inspiratory hypoglossal neurons). Peak inspiratory drive currents were reduced in ChrMP459 (Figure 4A, $$n = 19$$, Baseline: –142.90±22.82 pA vs. ChrMP459: –95.31±21.79 pA, $$p \leq 0.004$$). Reduced drive coincided with hypoglossal neurons generating fewer action potentials per preBötC burst in ChrMP459 (Figure 4B, $$n = 17$$, Baseline: 14.68±2.24 action potentials per burst vs. ChrMP459: 6.798±1.55 action potentials per burst, $p \leq 0.0001$). Injection of a depolarizing ramp current into hypoglossal neurons revealed that the HO inhibitor increased rheobase among inspiratory hypoglossal neurons (Figure 4C, $$n = 19$$, Baseline: 167.5±35.85 pA vs. ChrMP459: 338.0±82.50 pA, $$p \leq 0.007$$) yet decreased rheobase in non-inspiratory hypoglossal neurons (i.e. neurons not receiving drive during preBötC activity; Figure 4—figure supplement 1, $$n = 8$$, Baseline: 280.5±56.43 pA vs. ChrMP459: 228.2±47.96 pA, $$p \leq 0.0117$$). Figure 4.: Heme oxygenase inhibition reduces inspiratory drive currents in hypoglossal neurons.Whole cell patch clamp recordings were made from hypoglossal neurons in rhythmic brain slices while simultaneously recording ipsilateral preBötC activity in Baseline and in ChrMP459. Neurons were disinhibited from fast synaptic inhibition using 50 μM PTX and 1 μM Strychnine (DI). (A) (left) Representative voltage clamp recordings from a XIIn neuron (Vholding = –60 mV) aligned with corresponding integrated network activity from preBötC before (DI Baseline, top, gray) and after 20 μM ChrMP459 (DI ChrMP459, bottom, purple). Scale bar: 1 s x 10 pA. (middle) Magnification of highlighted (red dotted box) drive currents from DI Baseline (gray) and DI ChrMP459 (purple). Scale bar: 100 ms x 20 pA. (right). Comparison of XIIn inspiratory drive current magnitude distribution in DI Baseline (gray) and DI ChrMP459 (n=19, purple). Thin solid lines illustrate individual neuron response. Dashed black line illustrates mean drive current. (B) (left) Representative current clamp recordings from a spontaneously active XIIn neuron with the preBötC network rhythm in DI Baseline (top, gray) and DI ChrMP459 (bottom, purple); skipped transmission of action potentials in DI ChrMP459 are highlighted (pink box). Scale bar 2 s x 20 mV. (middle) Magnification of highlighted neuronal activity (red dashed box in trace, left). Scale bars: 100 msec x 25 mV. (right) Distribution of the average number of action potentials generated per inspiratory burst in DI Baseline (gray) and in DI ChrMP459 (n=17, purple). Thin solid lines illustrate individual neuron response. Dashed black line illustrates mean action potentials per drive. (C) (left) Representative trace of current clamp recording in response to ramp current injection during DI Baseline (gray trace) and in DI ChrMP459 (purple trace); scale bar: 500 ms. (right) Comparison of rheobase distributions found in inspiratory XIIn neurons during DI Baseline (gray) and in DI ChrMP459 (n=19, purple). Thin solid lines illustrate individual neuron response. Dashed black line illustrates mean Rheobase.Statistical analysis for all comparisons via paired t-test; error bars: SEM; significance level P<0.05. ## Disinhibition reduces occurrence of subnetwork activity and amplitude irregularities in the preBötC caused by HO inhibition Examining the preBötC rhythm under disinhibited conditions also revealed that ChrMP459 appeared to cause fewer subnetwork bursts in the disinhibited preBötC (Figure 5A). Indeed, the percentage of subnetwork preBötC activity was greater in ChrMP459 when synaptic inhibition was preserved (Figure 5B, ChrMP459, $$n = 34$$: 19.43±$3.37\%$ vs disinhibited ChrMP459, $$n = 17$$: 6.72 ± $3.13\%$, $$p \leq 0.02$$). Similarly, the IrSAMP during ChrMP459 was smaller in the disinhibited preBötC rhythm (Figure 5C, ChrMP459: 0.406±0.051 vs disinhibited ChrMP459: 0.215±0.04, $$p \leq 0.019$$). Figure 5.: Disinhibition reduces ChrMP459-induced subnetwork activity and amplitude irregularities in preBötC population activity.(A) Representative integrated trace of preBötC activity from slices with bath application of ChrMP459 (top, n=34) and disinhibited slices with ChrMP459 and 50 μM PTX and 1 μM Strychnine for fast synaptic inhibition (DI ChrMP459, bottom, n=17). Subnetwork preBötC activity (#) is identified in ChrMP459. Scale bar 5 sec. (B) Comparison of subnetwork activity in preBötC slices as a percentage of total bursts (subnetwork ≥%50 average burst area in baseline) in ChrMP459 (purple) and DI ChrMP459 (dark purple). Individual values represented by ◊, thick dashed line represents mean subnetwork activity. (C) Comparison of IrSAMP between ChrMP459 (purple, replotted from Figure 1C preBötC) and DI ChrMP459 (dark purple). Individual values represented by ◊, thick dashed line represents mean IrSAMP. Statistical analysis for all comparisons via paired t-test; error bars: SEM; significance level P<0.05. ## Elevated levels of H2S are observed in the hypoglossal nucleus of HO-2 null mice We next sought to determine the mechanism(s) by which inhibition of HO-2 affected hypoglossal neuron activity. Earlier studies Prabhakar, 2012; Morikawa et al., 2012 have reported that HO-2 is a negative regulator of CSE-dependent H2S production. To test this possibility, we first examined whether the hypoglossal neurons express CSE. In the wild-type hypoglossal nucleus, CSE is expressed in ChAT-positive hypoglossal neurons (Figure 6A, $$n = 3$$ mice). Homogenates made from hypoglossal and control tissue punches were prepared from wild-type and HO-2 mice to determine CSE-dependent H2S abundance. Relative to the wild-type hypoglossal nucleus (Figure 6B blue; $$n = 6$$, 60.58±6.37 nmol • mg–1 • h–1), H2S abundance was greater in the hypoglossal nucleus of HO-2 null mice (Figure 6B red; $$n = 6$$, 144.12±8.29 nmol • mg–1 • h–1, $p \leq 0.001$), but not different from the inferior olive brainstem region of HO-2 null mice (Figure 6B gray; $$n = 4$$, 56.10±2.88 nmol • mg–1 • h–1, $p \leq 0.05$). These findings show that the hypoglossal nucleus expresses CSE and suggests that HO-2 negatively regulates H2S production in the hypoglossal nucleus. Figure 6.: CSE-dependent H2S is produced in the hypoglossal nucleus and exogenous NaHS application uncouples hypoglossal nucleus activity from the preBötC.(A) CSE (red, bottom left) expression co-localizes with ChAT+ neurons (green, bottom right) in the XIIn (overlay, top). Scale bar 50 μm. (B) CSE-dependent H2S generation in pooled homogenates from WT and HO-2 null. Homogenates were prepared from tissue punches from the XIIn (red area in slice diagram) and inferior olive (gray area in slice diagram) at bregma between –7.20 mm and –7.90 mm. (WT: XIIn n=6; HO-2 null: XIIn n=6, inferior olive n=4). Each n in B represents a biological replicate consisting of the corresponding anatomical region pooled from two animals. (C) Integrated traces from XIIn (top) and preBötC (bottom) during Baseline (black), and in response to the H2S donor, NaHS, at 50 μM (dark gray) and 100 μM (light gray). NaHS application caused XIIn but not preBötC burst amplitude to diminish (blue dashed box) and in some cases, preBötC drive failed to produce activity in the XIIn (pink boxes). (D) Comparison of transmission from preBötC to XIIn after NaHS application at 10 μM, 50 μM and 100 μM. (E) I/O ratios for each NaHS concentration. (Baseline: n=9; 10 μM n=5; 50 μM n=6; 100 μM n=9). Statistical analysis for all comparisons via one-way ANOVA with Dunnett’s correction; error bars: SEM; significance level P<0.05. ## H2S mediates impaired transmission of inspiratory drive caused by disrupted HO-2 function If the impaired transmission of inspiratory drive to the hypoglossal nucleus by HO-2 dysregulation involves CSE-derived H2S then: [1] an H2S donor should mimic the effects of disrupted HO-2 activity; [2] CO administration should improve the input-output relationship in respiratory slices from HO-2 null mice and ChrM459 application; and, [3] CSE blockade should restore the transmission from the preBötC to the hypoglossal nucleus. The following experiments tested these possibilities. Wild-type brainstem slices exhibited a nearly 1:1 ratio of preBötC activity to hypoglossal nucleus output (Figure 6C, left). Application of NaHS, a H2S donor reduced this transmission from preBötC to hypoglossal output (Figure 6C, middle, right) in a dose-dependent manner (Figure 6D; 0 μM NaHS: $$n = 9$$, 100.0 ± $0.77\%$; 10 μM NaHS: $$n = 5$$, 90.14 ± $6.78\%$; 50 μM NaHS: $$n = 6$$, 84.18 ± $4.29\%$; 100 μM NaHS: $$n = 6$$, 81.26 ± $6.19\%$), which coincided with a reduction in I/O ratio by NaHS (Figure 6E; 0 μM NaHS: 1.055±0.03; 10 μM NaHS: 0.850±0.10; 50 μM NaHS: 0.816±0.09; 100 μM NaHS: 0.843±0.07). These findings demonstrated that increasing H2S abundance reduces hypoglossal neuronal activity consistent with findings from experiments using ChrM459 or HO-2 null mice. HO-2 dependent CO is known to inhibit CSE-dependent H2S production Prabhakar, 2012; Morikawa et al., 2012. Therefore, we sought to assess how the pharmacological CO donor, CORM-3 (20 μM), impacted activity in ChrMP459-treated wild type rhythmic slices (Figure 7A, $$n = 4$$) and rhythmic slices from HO-2 null mice ($$n = 4$$). Dysregulated HO-2 activity, caused by either pharmacological (ChrMP459) or genetic (HO-2 null mice) manipulation, is improved by CORM-3 as indicated by augmented I/O ratios (Figure 7B, $$n = 8$$; dysregulated HO-2: 0.705±0.09 vs. CORM-3: 1.05±0.07, $$p \leq 0.01$$) and improved transmission (Figure 7C, dysregulated HO-2: 75.05±$6.30\%$ vs. CORM-3: 94.21 ± $2.67\%$, $$p \leq 0.020$$). Figure 7.: HO-dependent transmission failures can be recovered with CO-donor CORM-3 and are not present in HO-2:CSE null transmission.(A) Representative integrated traces from preBotC and XIIn in WT slices during Baseline (left), in ChrMP459 alone (middle) and in ChrMP459 +20 µM CORM-3 (CORM-3, right). Both subnetwork (#) and failed transmissions (pink rectangle) are highlighted. (B) Heat maps of cycle-to-cycle I/O ratios during dysregulated HO-2 (n=8: n=4 HO-2 null and n=4 WT-ChrMP459) before and after CORM-3 application. Gray boxes indicate non-events in recordings from slower rhythms where less than 25 events occurred during the analysis window. (C) Comparison of transmission from preBötC to XIIn from dysregulated HO-2 slices before (red) and after (green) bath application of CORM-3. (D) Representative integrated traces from preBötC and XIIn in slices from HO-2:CSE null; scale bar 2 sec. (E) Heat map of cycle-to-cycle I/O ratio from preBötC to XIIn in HO-2:CSE null. The I/O ratio from HO-2:CSE null is greater than I/O ratios from HO-2 null (n=7, p=0.003). Gray boxes indicate non-events in recordings from slower rhythms where less than 25 events occurred during the analysis window. (F) Comparison of transmission from preBötC to XIIn in HO-2 null (red, n=7, subset replotted from Figure 2) and HO-2:CSE null (light blue, n=6). HO-2 null data used for comparisons in E and F are a subset of the data originally shown in Figure 2. Statistical analysis for B and C via paired t-test, analysis for E and F via unpaireed t-test; error bars: SEM; significance level P<0.05. To determine the involvement of CSE, inspiratory activity in the brainstem slice from HO-2:CSE double null mice appeared to be stable and consistent (Figure 7D). Quantification of simultaneous extracellular field recordings of preBötC activity and hypoglossal nucleus revealed a larger I/O ratio (Figure 7E, HO-2:CSE: 1.014±0.02, $$n = 6$$, $$p \leq 0.003$$) and near absence of transmission failures (Figure 7F, HO-2:CSE: 91.57 ± $3.20\%$, $$p \leq 0.0007$$) when compared to activity in HO-2 null slices. Similarly, in vivo L-PAG treatment, to acutely inhibit CSE activity, in HO-2 null mice improved transmission of preBötC activity to the hypoglossal nucleus in the rhythmic slice (Figure 8A, $$n = 6$$) as indicated by larger cycle-to-cycle I/O ratio (Figure 8B and L -PAG = 1.01± 0.03, $$p \leq 0.008$$) and greater transmission rates (Figure 8C and L -PAG 95.92 ± $2.18\%$, $p \leq 0.0001$) when compared to the respective metrics from untreated HO-2 null mice. Intermittent transmission failure was also evident in patch clamp recordings from untreated HO-2 null hypoglossal neurons (Figure 8D, left shaded cycles) but not in HO-2 null hypoglossal neurons treated with L-PAG (Figure 8D, right). These reduced transmission events correlated with smaller individual inspiratory drive currents in HO-2 null hypoglossal neurons when compared to corresponding inspiratory synaptic drive currents from L-PAG-treated HO-2 mice (Figure 8D–E, HO-2: –36.71±2.14 pA vs. L-PAG –194.3±82.73 pA, $$p \leq 0.0007$$). Together, these experiments implicated the involvement of CSE-dependent H2S signaling with the effects of disrupted HO-2 /CO signaling affecting synaptic drive and output from hypoglossal motoneurons. Figure 8.: Transmission failures in HO-2 null mice are rescued with CSE inhibitor L-propargylglycine.(A) Representative integrated traces of preBötC (bottom) and XIIn (top) in slices from HO-2 null mice treated with L- propargylglycine (L-PAG, 30 mg/kg, n=5). Scale bar 2 sec. (B) Heat map of cycle-to-cycle I/O ratio in rhythmic slices from HO-2 null mice treated with L-PAG (n=6). (C) Comparison of transmission between HO-2 null (n=7, subset replotted from Figure 2; red) and L-PAG slices (n=6, blue). (D) Representative voltage clamp recordings of inspiratory drive currents received by hypoglossal neurons from DI HO-2 null (n=8, left) and DI L-PAG (n=6, right). Neurons were disinhibited from fast synaptic inhibition using 50 μM PTX and 1 μM Strychnine. Scale bar 100 ms x 20 pA. Skipped transmission between preBötC (bottom) the XIIn neuron (top) occurs in untreated DI HO-2 null (highlighted pink boxes) but not in neurons from DI L-PAG. Subnetwork transmission to XIIn neuron in DI L-PAG (#). Magnified representative (red dashed box) drive potentials from DI HO-2 null and DI L-PAG. scale bars: 100 msec x 10 pA. (E) Comparison of average synaptic drive currents received by XIIn motoneurons from DI HO-2 null mice (n=8, red) produce smaller drive potentials when compared to DI L-PAG (n=6, blue). Statistical analysis for all comparisons via unpaired t-test; error bars: SEM; significance level P<0.05. ## Blockade of small conductance calcium-activated potassium channel (SKCa) activityrestores changes induced by HO-dysregulation in hypoglossal activity As our experiments implicated the involvement of H2S signaling, we next sought to determine how H2S sensitive ion channels may contribute to impairing hypoglossal neuron activity caused by HO-dysregulation. H2S has been shown to enhance activity of several different potassium channels, including SKCa and ATP-sensitive potassium channel (KATP) activities Mustafa et al., 2011. As both SKCa and KATP are important in the regulation of hypoglossal neuron excitability Haller et al., 2001; Lape and Nistri, 2000, we examined how blocking these channels affected hypoglossal activity during HO-dysregulation. At the network level, administration of the selective SKCa inhibitor, apamin (200 μM), increased the excitability of the hypoglossal neurons treated with ChrMP459. This enhanced activity exceeded the original baseline activity (i.e., prior to ChrMP459 administration) causing ectopic bursting in the hypoglossal nucleus (Figure 9—figure supplement 1A) making analysis of population transmission and I/O ratios unreliable. Therefore, we proceeded to resolve the effects of apamin on the influence of ChrMP459 at the level of individual hypoglossal inspiratory motoneurons. While in some hypoglossal neurons exposed to ChrMP459, apamin substantially increased drive currents (>100 pA; Figure 9A left), in others, apamin modestly increased the drive current (<100 pA; Figure 9A middle). Despite this variability, apamin increased inspiratory drive currents received by ChrMP459 treated hypoglossal neurons (Figure 9A, right, $$n = 6$$, ChrMP459: –85.77±38.54 pA vs. apamin: –219.97±97.76 pA, $$p \leq 0.031$$). Apamin also enhanced the number of action potential generated per preBötC burst during ChrMP459 (Figure 9B, $$n = 8$$, ChrMP-459: 12.57±3.68 action potential per burst vs. apamin 26.05±6.87 action potential per burst, $$p \leq 0.016$$). This was consistent with the ability of apamin to reduce rheobase in ChrMP459-treated inspiratory hypoglossal neurons (Figure 9C; $$n = 7$$, ChrMP459: 532.0±186.5 pA vs. Apamin: 307.09±79.62 pA, $$p \leq 0.016$$). Figure 9.: Apamin reverses changes to hypoglossal neurons’ intrinsic and synaptic excitability caused by HO dysregulation.(A) Representative synaptic drive current received by hypoglossal neurons in DI ChrMP459 (purple) and in DI Apamin (200 μM, gold). Left, depicts an example of apamin increasing the drive current >100 pA. Scale bars: 1 s x 50 pA., whereas middle, representative inspiratory drive current hypoglossal neuron in DI ChrMP459 (purple) and in DI Apamin (gold) where apamin increased the drive current by less than <100 pA Scale bars: 1 s x 50 pA. Comparison of inspiratory drive currents from hypoglossal neurons exposed to DI ChrMP459 (purple) to DI Apamin (n=6, gold). The effect of ChrMP459 on baseline disinhibited drive current for each of these neurons were reported in Figure 4A. (B). (left) Representative current clamp recordings from a spontaneously active XIIn neuron with the preBötC network rhythm during DI ChrMP459 (purple) and in DI Apamin (n=8, gold). Scale bars: 20 mV x 500 ms. (right) Comparison of action potentials generated per preBötC burst during DI ChrMP459 (purple) and DI Apamin (n=8, gold). The effect of DI ChrMP459 on baseline action potential generated per preBötC burst for each of these neurons were reported in Figure 4D. (C) (left) Representative traces of current clamp recordings in response to ramp current injection in DI (purple) and in DI Apamin (gold). Scale bar: 500 ms x 10 mV. (right) Rheobase comparison from inspiratory XIIn neurons during DI ChrMP459 (purple) and DI Apamin (n=7, gold). The effect of DI ChrMP459 on baseline rheobase for each of these neurons were reported in Figure 4F. Statistical analysis for all comparisons via paired t-test with Wilcoxon Correction; error bars: SEM; significance level P<0.05. To determine how blockade of KATP impacted hypoglossal activity during ChrMP459, we used the KATP channel blocker, tolbutamide (100 μM). Tolbutamide did not induce ectopic bursting in the hypoglossal nucleus during ChrMP459 (Figure 9—figure supplement 1B, $$n = 5$$). Furthermore, tolbutamide (100 μM), did not improve the rate of transmission of preBötC activity (Figure 9—figure supplement 1C, left, ChrMP459: 69.24 ± $6.00\%$ tolbutamide: 71.23 ± $6.86\%$, $$p \leq 0.83$$) but did increase the I/O ratio (Figure 9—figure supplement 1C, right, ChrMP459: 0.687±0.05 tolbutamide: 0.870±0.05, $$p \leq 0.037$$). Further, tolbutamide neither enhanced inspiratory drive currents in ChrMP459 (Figure 9—figure supplement 1D, $$n = 4$$, ChrMP459:–70.51±27.49 pA vs. tolbutamide: –83.06±36.29 pA, $$p \leq 0.375$$) nor increased the number of action potentials per preBötC burst in ChrMP459 (Figure 9—figure supplement 1E $$n = 4$$, ChrMP459: 7.063±2.08 action potential per burst vs. tolbutamide: 9.370±3.11 action potential per burst, $$p \leq 0.125$$). Moreover, tolbutamide did not affect rheobase of inspiratory hypoglossal neurons treated with ChrMP459 (Figure 9—figure supplement 1F, $$n = 7$$; ChrMP459: 221.02±74.80 pA vs. tolbutamide: 180.40±68.63 pA, $$p \leq 0.219$$). Thus, these results suggested that apamin could enhance activity of hypoglossal neurons during HO-inhibition; whereas, the efficacy of tolbutamide to impact activity during HO-inhibition was limited. ## Discussion Our study reveals a previously uncharacterized neuromodulatory interaction between HO-2 and CSE-derived H2S regulating activity from the hypoglossal nucleus. HO-2 dysregulation promoted subnetwork activity and irregularities in rhythmogenesis from the preBötC while also disturbing excitatory synaptic drive currents and intrinsic excitability of inspiratory hypoglossal neurons. Our investigations indicate that these phenomena contribute to a reduction in the input-output relationship and an increased propensity for transmission failure between the preBötC and the hypoglossal motor nucleus. Blocking CSE activity mitigated many effects caused by HO-2 dysregulation; whereas, using an H2S donor mimicked the impairments to the input-output relationship and intermittent failures observed with HO-2 dysregulation. Together these findings demonstrate a role for centrally-derived interactions between HO-2 and CSE activity in regulating motoneuron output responsible for maintaining upper airway patency. We used two approaches to dysregulate HO-2 activity in the rhythmic brainstem slice preparation. First, pharmacologically, by using the pan HO inhibitor, ChrMP459 in wild-type slices; and second, by performing experiments in slices from HO-2 null mice. Using these two approaches produced some empirical differences. The frequency of rhythmogenesis in the preBötC was faster and the hypoglossal IrSAMP was larger in HO-2 null networks when compared to wild-type networks. However, ChrMP459 did not impact these metrics. Despite these differences, both approaches produce similar effects on subnetwork preBötC activity, the I/O ratio, rate of transmission between preBötC and the hypoglossal nucleus and on synaptic drive currents in hypoglossal motoneurons (Table 2). Therefore, we considered these outcomes common to both approaches to represent the key effects caused by HO-2 dysregulation on inspiratory-related hypoglossal activity in the isolated brainstem slice and primarily focus on issues related to these phenomena throughout the remainder of the discussion. Table 2. \| Metric \| ChrMP459 (n) \| HO-2 null (n) \| p-value \| \| --- \| --- \| --- \| --- \| \| finst* (Hz) \| 0.24±0.019 (34) \| 0.42±0.049 (11) \| <0.0001 \| \| Burst Amplitude* (mV) \| 0.08±0.012 (34) \| 0.05±0.010 (11) \| 0.189 \| \| IrS AMP† (A.U.) \| 0.41±0.050 (34) \| 0.39±0.070 (11) \| 0.127 \| \| Subnetwork‡ (%) \| 19.43±3.373 (34) \| 16.70±4.898 (11) \| 0.670 \| \| I/O Ratio§ (A.U.) \| 0.59±0.064 (34) \| 0.79±0.074 (11) \| 0.096 \| \| Transmission¶ (%) \| 75.10±3.425 (34) \| 64.11±6.002 (11) \| 0.119 \| \| Synaptic Drive Current (pA) \| –95.31±21.790 (19) \| –36.71±2.141 (8) \| 0.097 \| Mice treated with intermittent hypoxia (IH) patterned after blood O2 profiles associated with sleep apnea also show failed transmission and a reduced input-output relationship between the preBötC and hypoglossal nucleus Garcia et al., 2016. These IH-dependent effects also correlated to cycle to cycle irregularities in rhythmogenesis and caused subnetwork preBötC activity that failed to produce measurable output from the motor nucleus Garcia et al., 2016. Like that of IH, HO-2 dysregulation increases cycle to cycle irregularity of preBötC burst amplitude and produces subnetwork activity in the preBötC that often failed to produce measurable hypoglossal output. The failure to generate measurable output from the motor pool also occurred with larger preBötC bursts, albeit in a smaller percentage of occurrence (Figure 1F and Figure 2F) suggesting the impact of HO-2 dysregulation was not restricted to effects on the preBötC. Simultaneous triple recordings from the preBötC, premotor field, and the hypoglossal nucleus demonstrated that in ChrMP459, premotor activity corresponding to the preBötC rhythm was reliable; whereas, hypoglossal activity often failed. These experiments suggested that HO-2 dysregulation produced deficits in synaptic physiology between intermediate premotor neurons and the hypoglossal nucleus and possibly affected the postsynaptic excitability of hypoglossal motoneurons. Patch clamp recordings demonstrated that ChrMP459 reduced postsynaptic action potential generation and intrinsic excitability of inspiratory hypoglossal neurons. Reduced motoneuron excitability also corresponded with HO-2 dysregulation mediated reduction in synaptic drive currents in hypoglossal neurons. Although these experiments did not resolve the contribution of changes in presynaptic or postsynaptic properties to suppressing synaptic drive, the phenomenon appeared independent of effects related to inhibition as these experiments were performed in the presence of blockers for fast GABAergic and glycinergic receptors. While HO-2 expression was not documented in either premotor neurons or the preBötC, HO-2 dysregulation impacted the respiratory network promoting subnetwork preBötC activity and increasing the irregularity of the rhythm. While the disinhibition experiments in ChrMP459 indicated that HO-2 dysregulation promotes a network state favoring synaptic inhibition to promote subnetwork activity (Figure 5A), disinhibition in HO-2 null slices neither improved the IrSAMP of the preBötC nor reduced the occurrence of subnetwork preBötC activity (data not shown). These divergent outcomes raise the possibility that long-term loss of HO-2 activity during development may have a broader impact on mechanisms of governing rhythmogenesis that cannot be corrected by acutely blocking synaptic inhibition in the brainstem network. Based on prior observations demonstrating that lost HO-2 dependent CO activity enhances CSE-dependent H2S production and leads to respiratory disturbances Peng et al., 2018, it was predicted that the provision of CO or blockade CSE activity would improve subnetwork activity and amplitude irregularities of rhythmogenesis from the preBötC. While using, the CO donor, CORM-3 and blocking CSE activity in HO-2 null slices, on average, reduced both subnetwork activity and the IrSAMP in the preBötC, both phenomena were still statistically similar to that observed in preBötC rhythms recorded from unmanipulated HO-2 null slices (Figure 8—figure supplement 1). These findings suggests that HO-2 dysregulation may also involve actions that are independent from CO and CSE / H2S activities. In addition to CO, biliverdin-bilirubin and ferrous iron are bioactive molecules generated from activity of heme oxygenases Snyder and Barañano, 2001. Bilirubin acts as antioxidant protecting neurons from oxidative injury Doré et al., 1999, whereas ferrous iron can promote an oxidative state and cause injury Gutteridge and Halliwell, 2018. These molecules generated from HO-2 activity may be important to maintaining redox state and stable rhythmogenesis from the preBötC during development. Indeed, reactive oxygen species have specific and different actions on rhythmogenesis Garcia et al., 2011 while the promote of a pro-oxidant state in the preBötC can lead to irregular rhythmogenesis Garcia et al., 2016. Further work is needed to resolve the potential role for different bioactive molecules derived from HO-2 activity on neurophysiology of the preBötC. Although we did not resolve potential differences in HO-2 expression among specific motoneurons that innervate different upper airway muscles, such as genioglossal neurons, divergent effects of ChrMP459 on non-inspiratory and inspiratory hypoglossal neuronal properties were observed. ChrMP459 decreased rheobase among non-inspiratory neurons, whereas in inspiratory hypoglossal neurons, it decreased the magnitude of drive currents, increased rheobase, and diminished the number of action potentials generated during preBötC bursting. While these findings illustrate the potential for HO-dysregulation to differentially impact inspiratory and non-inspiratory hypoglossal neurons, these findings also emphasize a need to further resolve how HO-2 activity impacts different hypoglossal neurons innervating various muscle groups of the upper airway. In the HO-2 null mouse, the incidence of OSA is absent with co-inhibition of CSE Peng et al., 2018, which is consistent with reports that CO generated by HO-2 inhibits CSE-dependent H2S production Prabhakar, 2012; Morikawa et al., 2012. After documenting CSE expression in hypoglossal neurons and demonstrating an increased abundance of H2S in the hypoglossal nucleus of HO-2 null mice, we demonstrated that a CO donor improves transmission and the input-output relationship between the preBötC and hypoglossal nucleus in HO-2 null slices (Figure 7A–C). Furthermore, using a H2S donor also increased transmission failures and reduced the I/O ratio similar to dysregulating HO-2 activity (Figure 6D–E). Endogenous H2S activity could originate from other H2S producing enzymes, such as cystathionone β-synthase (CBS) that is expressed primarily in astrocytes throughout the CNS Enokido et al., 2005; yet CBS inhibition appears to have limited impact on inspiratory activity from the hypoglossal nucleus da Silva et al., 2017. Additionally, our experiments manipulating CSE activity, by either treating HO-2 null mice with L-PAG or using HO-2:CSE null slices (Figures 7D–E,–8), improved the I/O and reduced transmission failure from preBötC to the hypoglossal nucleus. Larger synaptic drive currents were also observed in HO-2 null hypoglossal neurons after treatment with L-PAG. Thus, in contrast to that in the preBötC, where HO-2 dysregulation may involve actions that independent from CSE / H2S, the mutual interaction between HO-2/CO and CSE/H2S appears to have a major role in regulating hypoglossal output by both modulating excitatory synaptic drive currents received by hypoglossal motoneurons and impacting their intrinsic excitability. How might enhanced H2S signaling reduce excitatory synaptic currents and excitability of hypoglossal neurons? While it is possible that H2S may impact presynaptic release of glutamate from intermediate premotor neurons and/or postsynaptic receptor activity of hypoglossal neurons, the ChrMP459 mediated increase in rheobase among inspiratory hypoglossal neurons implicated the involvement of non-synaptic conductance(s) downstream of H2S-based signaling caused by perturbations in HO-2 activity. H2S can enhance both KATP and SKCa activities Mustafa et al., 2011. In the hypoglossal neurons, KATP is dynamically active causing periodic adjustment of neuronal excitability Haller et al., 2001 while SKCa also regulates excitability and firing properties of hypoglossal neurons Lape and Nistri, 2000. In ChrMP459, tolbutamide had a limited effect normalizing transmission as it improved the I/O ratio, but did not reduce transmission failure. Tolbutamide neither increased excitatory synaptic currents nor enhanced intrinsic excitability of hypoglossal neurons in ChrMP459. In contrast, apamin normalized synaptic drive currents and increased excitability of inspiratory hypoglossal neurons in ChrMP459. These results indicated that blockade of KATP was limited in countering the effects on HO-2 dysregulation in the hypoglossal nucleus, whereas that blockade of SKCa sufficiently mitigates many aspects of HO-2 dysregulation in hypoglossal neurons associated with preBötC activity. In addition, the occurrence of obstructive apnea in HO-2 null mice Peng et al., 2017, spontaneous hypertensive rats exhibit an increased incidence of apneas and hypopneas that are associated with reduced CO levels due to a reduction in HO-2 activity and increased H2S generation Peng et al., 2014. While in both cases, these effects have been linked to interactions between CO and H2S in the peripheral nervous system, our findings indicate that central HO-2 dysregulation may also contribute to the incidence of apneas. Loss of HO-2 activity causes irregular rhythmogenesis from the preBötC while also reducing synaptic drive and intrinsic excitability in hypoglossal motoneurons. In the hypoglossal nucleus, the impact of this dysregulation appears to be largely normalized by blocking CSE activity suggesting that the interaction between HO-2 dependent CO production and CSE-dependent H2S activity have an important role in regulating hypoglossal activity. Additionally, while enhanced loop gain in HO-2 null mice has been attributed to increased chemo reflex sensitivity regulated by the carotid bodies Peng et al., 2017; Peng et al., 2018; Osman et al., 2018; Prabhakar and Semenza, 2012, this study implicates a concurrent target for HO-2 dysregulation in the central respiratory circuit. The combination of increased chemoreflex sensitivity, imbalanced preBötC excitation/inhibition activity, and reduced hypoglossal motoneuron excitability caused by disturbed HO-2 activity could all contribute to the increased loop gain that perpetuates transmission failures in respiratory motor output and disrupt upper airway patency in HO-2 null mice and OSA patients. Moreover, this interaction may extend beyond apneas and may also be important for regulating the genioglossus and other muscle groups of the tongue during behaviors such as swallowing Fregosi and Ludlow, 2014 and vocalization Wei et al., 2022, particularly when considering the potential for rapid signaling via CO and H2S. Furthermore, should mutual interactions between these gasotransmitters and their respective enzymes exist in other motoneuron pools, our findings may be relevant to a variety of clinical conditions such as fentanyl-induced chest wall rigidity syndrome (i.e. wooden chest syndrome), amyotrophic lateral sclerosis, and spinal cord injury where upper airway control may be affected. Thus, while this study demonstrates the potential importance of central HO-2/CO and CSE/H2S interactions in regulating hypoglossal motoneurons, future work is needed to fully understand the role of gasotransmitters in the physiology of the hypoglossal nucleus and other motoneuron pools. ## Study approval In accordance with National Institutes of Health guidelines, all animal protocols were performed with the approval of the Institute of Animal Care and Use Committee at The University of Chicago (ACUP 72486, ACUP 71811). ## Experimental animals Experiments were performed using neonatal (postnatal day 6 to postnatal day 12) wild-type mice (C57BL/6; Charles River), HO-2 null mice (from S. H. Snyder), and HO-2:CSE double-null mice. HO-2:CSE double-null mice were created by crossing HO-2 null females with CSE null males (from R. Wang, Department of Biology, Laurentian University, Sudbury, ON, Canada). Tissues from both sexes were used. No sex-based differences were observed; therefore, data from both sexes were pooled for analysis. All litters were housed with their dam in ALAAC-approved facilities on a 12 hr / 12 hr light-dark cycle. ## Pharmacological agents Heme oxygenase activity was blocked using bath application of Chromium (III) Mesoporphyrin IX chloride (ChrMP459, 20 μM; Frontiers Sciences, Newark DE). A CO donor, CORM-3 (20 μM; Sigma-Aldrich St. Louis MO) was bath applied following ChrMP459 application. NaHS (10μM to 100 μM; Sigma-Aldrich), a H2S donor, was bath applied. In all patch clamp experiments, fast synaptic glycinergic and GABAergic inhibition was blocked by bath application of strychnine (1 μM; Sigma-Aldrich) and picrotoxin (50 μM; Sigma-Aldrich), respectively. Inhibition of CSE production was accomplished by in vivo L-propargylglycine L-PAG, 30 mg/kg (Sigma-Aldrich) administered (i.p. injection) 2.5–3 hrs prior to preparation of the rhythmic brainstem slice preparation. Inhibition of potassium channels SKCa and ATP-sensitive potassium channel (KATP) was via bath application of Apamin (200 μM; Sigma-Aldrich) and Tolbutamide (100 μM; Sigma-Aldrich), respectively. ## Measurement of H2S production Anaesthetized mice (urethane, 1.2 g•kg−1 i.p.) were rapidly euthanized by decapitation. Following rapid removal of the brainstem, the tissue was flash frozen in liquid N2. Flash frozen tissues were stored at –80 °C until coronal brainstem sections (300 μm thick) were cut with a cryostat at –20 °C and tissue punches of desired tissue were procured for immediate H2S measurements. The hypoglossal nucleus and control (inferior olive nucleus) brainstem tissue punches were made from the slices using a chilled micro-punch needle. Hypoglossal tissue from a single brainstem was not sufficient for effectively measuring H2S levels; therefore, we pooled bilateral micro punched tissue from two mice for each sample where H2S levels measured. H2S levels were determined as described previously Yuan et al., 2016. Briefly, cell homogenates from the pooled micro-punch tissue samples were prepared in 100 mM potassium phosphate buffer (pH 7.4). The enzyme reaction was carried out in sealed tubes. The assay mixture in a total volume of 500 μL contained (in final concentration): 100 mM potassium phosphate buffer (pH 7.4), 800 μM l-cysteine, 80 μM pyridoxal 5′-phosphate with or without L-PAG (20 µM) and cell homogenate (20 μg of protein), was incubated at 37 °C for 1 hr. At the end of the reaction, alkaline zinc acetate ($1\%$ mass / volume; 250 μL) and trichloroacetic acid ($10\%$ vol/vol) were sequentially added to trap H2S and stop the reaction, respectively. The zinc sulfide formed was reacted with acidic N,N-dimethyl-p-phenylenediamine sulfate (20 μM) and ferric chloride (30 μM) and the absorbance was measured at 670 nm using Shimadzu UV-VIS Spectrophotometer. L-PAG inhibitable H2S concentration was calculated from a standard curve and values are expressed as nanomoles of H2S formed per hour per mg of protein. ## Immunohistochemistry Anaesthetized mice (urethane, 1.2 g•kg−1 i.p.) were perfused transcardially with heparinized phosphate-buffered saline (PBS) for 20 min followed by $4\%$ paraformaldehyde in PBS. Brainstems were harvested, post-fixed in $4\%$ paraformaldehyde overnight, and cryoprotected in $30\%$ sucrose/PBS at 4 °C. Frozen tissues were serially sectioned at a thickness of 20 μm (coronal section) and stored at –80 °C. Sections were treated with $20\%$ normal goat serum, $0.1\%$ bovine serum albumin and $0.1\%$ Triton X-100 in PBS for 30 min and incubated with primary antibodies against choline acetyltransferase (ChAT, 1:100; Millipore; #AB144P), and HO-2 (1:200, Novus Biologicals; # NBP1-52849) or CSE (1:250; gift from SH Snyder, Johns Hopkins University) followed by Texas Red-conjugated goat anti-mouse IgG (HO-2 and CSE) or FITC-conjugated goat anti-rabbit IgG (1:250; Molecular Probes, ChAT). After rinsing with PBS, sections were mounted in Vecta shield containing DAPI (Vector Labs) and analyzed using a fluorescent microscope (Eclipse E600; Nikon). ## Brainstem slices for electrophysiology The isolated rhythmic brainstem slices were prepared as previously described Garcia et al., 2017. Briefly, anesthetized (1.5–$3\%$ isofluorane inhaled) animals were euthanized by rapid decapitation. Brainstems were dissected, isolated and placed into ice cold artificial cerebral spinal fluid (aCSF) (composition in mM: 118 NaCl, 25 NaHCO3, 1 NaH2PO4, 1 MgCl2, 3 KCl, 30 Glucose, 1.5 CaCl2, pH = 7.4) equilibrated with $95\%$ O2, $5\%$ CO2. The isolated brainstem was glued to an agar block (dorsal face to agar) with the rostral face up and submerged in aCSF equilibrated with carbogen. Serial cuts were made through the brainstem until the appearance of anatomical landmarks such as the narrowing of the fourth ventricle and the hypoglossal axons. The preBötC and XIIn was retained in a single transverse brainstem slice (thickness: 560±40 μm). The slice was transferred into the recording chamber (~6 mL volume) where it was continuously superfused with recirculating aCSF (flow rate: 12–15 mL/min). Prior to recording, extracellular KCl was raised to 8 mM and the spontaneous rhythm was allowed to stabilize prior to the start of every experiment. ## Electrophysiology Extracellular population activity was recorded with glass suction pipettes filled with aCSF. Pipettes were positioned over the ventral respiratory column containing the preBötC and over the ipsilateral medial dorsal column containing the hypoglossal nucleus. In some experiments, a third pipette was positioned between the preBötC and hypoglossal nucleus just lateral to the axon tract to record transmission through the premotor field containing intermediate premotor inspiratory neurons Koizumi et al., 2013; Revill et al., 2015. Extracellular population activity was recorded with glass suction pipettes filled with aCSF Garcia et al., 2016. The recorded signal was sampled at 5 kHz, amplified 10,000 X, with a lowpass filter of 10 kHz using an A-M instruments (A-M Systems, Sequim, WA, USA) extracellular amplifier. The signal was then rectified and integrated. Recordings were stored on a computer for posthoc analysis. All intracellular recordings were made using the Multiclamp 700B (Molecular Devices). Acquisition and post hoc analyses were performed using the Axon pCLAMP10 software suite (Molecular Devices). Whole cell patch clamp recordings of hypoglossal motoneurons were obtained using the blind-patch technique with a sample frequency of 40 kHz. Recordings were made with unpolished patch electrodes, pulled from borosilicated glass pipettes with a capillary filament Garcia et al., 2016. The electrodes had a resistance of 3.5–8 MΩ when filled with the whole cell patch clamp pipette solution containing (in mM): 140 K-gluc acid, 1 CaCl2, 10 EGTA, 2 MgCl2, 4 Na2-ATP, 10 HEPES. Neurons located at least two to three cell layers (about 75–250 μm) rostral from the caudal surface of the slice were recorded. The liquid junction potential was calculated to be –12 mV and was subtracted from the membrane potential. The series resistance was compensated and corrected throughout each experiment. In voltage clamp experiments, membrane potential was held at –60 mV. Current clamp experiments used a holding potential between 0 and –100 pA to establish the baseline resting membrane potential between –55 and –70 mV. In some cases, we determined rheobases using a ramp protocol in our current clamp recordings. This ramp protocol consisted of a hyperpolarizing step (–100 pA) succeeded by the injection of a ramping depolarizing current (122 pA/sec; peak current 600 pA). ## Statistical analyses Unless otherwise explicitly stated elsewhere, each n value represents an individual animal that served as a biological replicate for a given measurement. The irregularity score of amplitude (IrSAMP) was calculated as described in Garcia et al., 2016. Transmission was expressed as a percentage of the hypoglossal network bursts corresponding to the total network bursts from either the preBötC or the premotor field. Bursts were considered corresponding if initial start time of bursts were within 500–750ms of each other (corresponding time was maximized until only one hypoglossal burst per preBötC was detected). Mean I/O and transmission values for each slice were calculated using a 120 s window. This analysis window was taken at the end of each baseline or pharmacological agent phase (each phase duration = 600 s). The input-output (I/O) ratio for each inspiratory event (defined by a network burst in preBötC) was calculated as the ratio of preBötC burst area to corresponding hypoglossal burst area as previously described Garcia et al., 2016. Calculation of the I/O ratio, was performed using the following equation:IOn=∫BAXIIn / ∫BApreBo..tCn where IOn is the I/O ratio of the nth cycle, ∫BAXIIn is the integrated burst area in the hypoglossal nucleus of the nth cycle and preBo..tC the integrated burst area in the preBo..tC of the nth cycle. In cycles where preBo..tC did not correspond with hypoglossal output, ∫BAXIIn was assigned a value of 0. Prior to the calculation of the I/O ratio, each ∫BAXII was normalized to the mean hypoglossal integrated burst area of the analysis window, and each ∫BApreBo..tCn, was normalized to the mean preBo..tC integrated burst area of the analysis window. Heat maps were used to illustrate individual I/O ratios for up to 25 consecutive cycles in the analysis window for each experiment performed. To illustrate the cycle-to-cycle input-output relationships between networks, heat maps of I/O ratio values were plotted for each slice included in the experiment. Each row represents sequential cycles from a single slice experiment. As the rhythmic frequency across preparations varied, the number of events (i.e. cycle number) in the 120 s analysis window also varied; therefore, either the total number of cycles or up to 25 consecutive cycles from a given slice recording were plotted. Statistics were performed using Origin 8 Pro (OriginLab, RRID: SCR_014212) or Prism 6 (GraphPad Software; RRID: SCR_015807). In cases where the distribution of data appeared normal, comparisons between two groups were conducted using either paired or unpaired two-tailed t-tests as appropriate. In cases, where the distribution of individual data points did not appear normal, the Wilcoxon match-paired signed rank test was performed. A one-way ANOVA was performed followed by posthoc Dunnett’s test comparing experimental groups to control when a comparison of three or more groups. In plots where means are presented, the mean and error bars are presented, the error bars reflect the S.E.M. Differences were defined when the p-value was less than 0.05. ## Funding Information This paper was supported by the following grants: ## Data availability Numerical data used to generate figures is uploaded to Dryad. Source Data file names refer to current figure panels. The following dataset was generated: GarciaAJ BroweB 2022Numerical Data from: Gasotransmitter Modulation of Hypoglossal Motoneuron ActivityDryad Digital Repository10.5061/dryad.44j0zpchc ## References 1. Bibbins-Domingo K, Grossman DC, Curry SJ, Davidson KW, Epling JW, García FAR, Herzstein J, Kemper AR, Krist AH, Kurth AE, Landefeld CS, Mangione CM, Phillips WR, Phipps MG, Pignone MP, Silverstein M, Tseng CW. Screening for obstructive sleep apnea in adults: US preventive services task force recommendation statement*. JAMA* (2017) 317 407-414. DOI: 10.1001/jama.2016.20325 2. Castro D, Freeman LA. Oropharyngeal Airway (Internet) (2021) 3. da Silva GSF, Sabino JPJ, Rajani V, Alvares TS, Pagliardini S, Branco LGS, Funk GD. Excitatory modulation of the preBötzinger complex inspiratory rhythm generating network by endogenous hydrogen sulfide. Frontiers in Physiology (2017) 8. DOI: 10.3389/fphys.2017.00452 4. Daulatzai MA. Cerebral hypoperfusion and glucose hypometabolism: key pathophysiological modulators promote neurodegeneration, cognitive impairment, and Alzheimer’s disease. Journal of Neuroscience Research (2017) 95 943-972. DOI: 10.1002/jnr.23777 5. Ding Y, Li YL, Schultz HD. Downregulation of carbon monoxide as well as nitric oxide contributes to peripheral chemoreflex hypersensitivity in heart failure rabbits. Journal of Applied Physiology (2008) 105 14-23. DOI: 10.1152/japplphysiol.01345.2007 6. Doré S, Takahashi M, Ferris CD, Zakhary R, Hester LD, Guastella D, Snyder SH. Bilirubin, formed by activation of heme oxygenase-2, protects neurons against oxidative stress injury. PNAS (1999) 96 2445-2450. DOI: 10.1073/pnas.96.5.2445 7. Eckert DJ, White DP, Jordan AS, Malhotra A, Wellman A. Defining phenotypic causes of obstructive sleep apnea: identification of novel therapeutic targets. American Journal of Respiratory and Critical Care Medicine (2013) 188 996-1004. DOI: 10.1164/rccm.201303-0448OC 8. Enokido Y, Suzuki E, Iwasawa K, Namekata K, Okazawa H, Kimura H. Cystathionine beta-synthase, a key enzyme for homocysteine metabolism, is preferentially expressed in the radial glia/astrocyte lineage of developing mouse CNS. FASEB Journal (2005) 19 1854-1856. DOI: 10.1096/fj.05-3724fje 9. Fleury Curado T, Fishbein K, Pho H, Brennick M, Dergacheva O, Sennes LU, Pham LV, Ladenheim EE, Spencer R, Mendelowitz D, Schwartz AR, Polotsky VY. Chemogenetic stimulation of the hypoglossal neurons improves upper airway patency. Scientific Reports (2017) 7. DOI: 10.1038/srep44392 10. Fleury Curado TA, Pho H, Dergacheva O, Berger S, Lee R, Freire C, Asherov A, Sennes LU, Mendelowitz D, Schwartz AR, Polotsky VY. Silencing of hypoglossal motoneurons leads to sleep disordered breathing in lean mice. Frontiers in Neurology (2018) 9. DOI: 10.3389/fneur.2018.00962 11. Fregosi RF, Ludlow CL. Activation of upper airway muscles during breathing and swallowing. Journal of Applied Physiology (2014) 116 291-301. DOI: 10.1152/japplphysiol.00670.2013 12. Garcia AJ, Khan SA, Kumar GK, Prabhakar NR, Ramirez JM. Hydrogen peroxide differentially affects activity in the pre-Bötzinger complex and hippocampus. Journal of Neurophysiology (2011) 106 3045-3055. DOI: 10.1152/jn.00550.2010 13. Garcia AJ, Zanella S, Dashevskiy T, Khan SA, Khuu MA, Prabhakar NR, Ramirez JM. Chronic intermittent hypoxia alters local respiratory circuit function at the level of the prebötzinger complex. Frontiers in Neuroscience (2016) 10. DOI: 10.3389/fnins.2016.00004 14. Garcia AJ, Dashevskiy T, Khuu MA, Ramirez JM. Chronic intermittent hypoxia differentially impacts different states of inspiratory activity at the level of the prebötzinger complex. Frontiers in Physiology (2017) 8. DOI: 10.3389/fphys.2017.00571 15. Genta PR, Sands SA, Butler JP, Loring SH, Katz ES, Demko BG, Kezirian EJ, White DP, Wellman A. Airflow shape is associated with the pharyngeal structure causing OSA. Chest (2017) 152 537-546. DOI: 10.1016/j.chest.2017.06.017 16. Gutteridge JMC, Halliwell BMR. Mini-review: oxidative stress, redox stress or redox success?. Biochemical and Biophysical Research Communications (2018) 502 183-186. DOI: 10.1016/j.bbrc.2018.05.045 17. Haller M, Mironov SL, Karschin A, Richter DW. Dynamic activation of K (ATP) channels in rhythmically active neurons. The Journal of Physiology (2001) 537 69-81. DOI: 10.1111/j.1469-7793.2001.0069k.x 18. Horner RL. The neuropharmacology of upper airway motor control in the awake and asleep states: implications for obstructive sleep apnoea. Respiratory Research (2001) 2 286-294. DOI: 10.1186/rr71 19. Horner RL. Emerging principles and neural substrates underlying tonic sleep-state-dependent influences on respiratory motor activity. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences (2009) 364 2553-2564. DOI: 10.1098/rstb.2009.0065 20. Horton GA, Fraigne JJ, Torontali ZA, Snow MB, Lapierre JL, Liu H, Montandon G, Peever JH, Horner RL. Activation of the hypoglossal to tongue musculature motor pathway by remote control. Scientific Reports (2017) 7. DOI: 10.1038/srep45860 21. Hua F. New insights into diabetes mellitus and its complications: a narrative review. Annals of Translational Medicine (2020) 8. DOI: 10.21037/atm-20-7243 22. Koizumi H, Koshiya N, Chia JX, Cao F, Nugent J, Zhang R, Smith JC. Structural-Functional properties of identified excitatory and inhibitory interneurons within pre-botzinger complex respiratory microcircuits. The Journal of Neuroscience (2013) 33 2994-3009. DOI: 10.1523/JNEUROSCI.4427-12.2013 23. Kubin L. Neural control of the upper airway: respiratory and state-dependent mechanisms. Comprehensive Physiology (2016) 6 1801-1850. DOI: 10.1002/cphy.c160002 24. Lape R, Nistri A. Current and voltage clamp studies of the spike medium afterhyperpolarization of hypoglossal motoneurons in a rat brain stem slice preparation. Journal of Neurophysiology (2000) 83 2987-2995. DOI: 10.1152/jn.2000.83.5.2987 25. Loffler KA, Heeley E, Freed R, Meng R, Bittencourt LR, Gonzaga Carvalho CC, Chen R, Hlavac M, Liu Z, Lorenzi-Filho G, Luo Y, McArdle N, Mukherjee S, Yap HS, Zhang X, Palmer LJ, Anderson CS, McEvoy RD, Drager LF. Continuous positive airway pressure treatment, glycemia, and diabetes risk in obstructive sleep apnea and comorbid cardiovascular disease. Diabetes Care (2020) 43 1859-1867. DOI: 10.2337/dc19-2006 26. Lyons MM, Bhatt NY, Pack AI, Magalang UJ. Global burden of sleep-disordered breathing and its implications. Respirology (2020) 25 690-702. DOI: 10.1111/resp.13838 27. Malhotra A, Ayappa I, Ayas N, Collop N, Kirsch D, Mcardle N, Mehra R, Pack AI, Punjabi N, White DP, Gottlieb DJ. Metrics of sleep apnea severity: beyond the apnea-hypopnea index. Sleep (2021) 44. DOI: 10.1093/sleep/zsab030 28. Mancuso C. Heme oxygenase and its products in the nervous system. Antioxidants & Redox Signaling (2004) 6 878-887. DOI: 10.1089/ars.2004.6.878 29. Mehra R. Sleep apnea and the heart. Cleveland Clinic Journal of Medicine (2019) 86 10-18. DOI: 10.3949/ccjm.86.s1.03 30. Morikawa T, Kajimura M, Nakamura T, Hishiki T, Nakanishi T, Yukutake Y, Nagahata Y, Ishikawa M, Hattori K, Takenouchi T, Takahashi T, Ishii I, Matsubara K, Kabe Y, Uchiyama S, Nagata E, Gadalla MM, Snyder SH, Suematsu M. Hypoxic regulation of the cerebral microcirculation is mediated by a carbon monoxide-sensitive hydrogen sulfide pathway. PNAS (2012) 109 1293-1298. DOI: 10.1073/pnas.1119658109 31. Mustafa AK, Sikka G, Gazi SK, Steppan J, Jung SM, Bhunia AK, Barodka VM, Gazi FK, Barrow RK, Wang R, Amzel LM, Berkowitz DE, Snyder SH. Hydrogen sulfide as endothelium-derived hyperpolarizing factor sulfhydrates potassium channels. Circulation Research (2011) 109 1259-1268. DOI: 10.1161/CIRCRESAHA.111.240242 32. Neelapu BC, Kharbanda OP, Sardana HK, Balachandran R, Sardana V, Kapoor P, Gupta A, Vasamsetti S. Craniofacial and upper airway morphology in adult obstructive sleep apnea patients: a systematic review and meta-analysis of cephalometric studies. Sleep Medicine Reviews (2017) 31 79-90. DOI: 10.1016/j.smrv.2016.01.007 33. Nemati S, Edwards BA, Sands SA, Berger PJ, Wellman A, Verghese GC, Malhotra A, Butler JP. Model-based characterization of ventilatory stability using spontaneous breathing. Journal of Applied Physiology (2011) 111 55-67. DOI: 10.1152/japplphysiol.01358.2010 34. Osman AM, Carter SG, Carberry JC, Eckert DJ. Obstructive sleep apnea: current perspectives. Nature and Science of Sleep (2018) 10 21-34. DOI: 10.2147/NSS.S124657 35. Peng YJ, Makarenko VV, Nanduri J, Vasavda C, Raghuraman G, Yuan G, Gadalla MM, Kumar GK, Snyder SH, Prabhakar NR. Inherent variations in CO-H2S-mediated carotid body O2 sensing mediate hypertension and pulmonary edema. PNAS (2014) 111 1174-1179. DOI: 10.1073/pnas.1322172111 36. Peng YJ, Zhang X, Gridina A, Chupikova I, McCormick DL, Thomas RJ, Scammell TE, Kim G, Vasavda C, Nanduri J, Kumar GK, Semenza GL, Snyder SH, Prabhakar NR. Complementary roles of gasotransmitters CO and H2S in sleep apnea. PNAS (2017) 114 1413-1418. DOI: 10.1073/pnas.1620717114 37. Peng YJ, Zhang X, Nanduri J, Prabhakar NR. Therapeutic targeting of the carotid body for treating sleep apnea in a pre-clinical mouse model. Advances in Experimental Medicine and Biology (2018) 1071 109-114. DOI: 10.1007/978-3-319-91137-3_14 38. Prabhakar NR. Carbon monoxide (CO) and hydrogen sulfide (H (2) S) in hypoxic sensing by the carotid body. Respiratory Physiology & Neurobiology (2012) 184 165-169. DOI: 10.1016/j.resp.2012.05.022 39. Prabhakar NR, Semenza GL. Gaseous messengers in oxygen sensing. Journal of Molecular Medicine (2012) 90 265-272. DOI: 10.1007/s00109-012-0876-1 40. Revill AL, Vann NC, Akins VT, Kottick A, Gray PA, Del Negro CA, Funk GD. Dbx1 precursor cells are a source of inspiratory XII premotoneurons. eLife (2015) 4. DOI: 10.7554/eLife.12301 41. Schwartz AR, Patil SP, Laffan AM, Polotsky V, Schneider H, Smith PL. Obesity and obstructive sleep apnea: pathogenic mechanisms and therapeutic approaches. Proceedings of the American Thoracic Society (2008) 5 185-192. DOI: 10.1513/pats.200708-137MG 42. Snyder SH, Barañano DE. Heme oxygenase: a font of multiple messengers. Neuropsychopharmacology (2001) 25 294-298. DOI: 10.1016/S0893-133X(01)00275-5 43. Vos WG, De Backer WA, Verhulst SL. Correlation between the severity of sleep apnea and upper airway morphology in pediatric and adult patients. Current Opinion in Allergy and Clinical Immunology (2010) 10 26-33. DOI: 10.1097/ACI.0b013e328334f659 44. Wei XP, Collie M, Dempsey B, Fortin G, Yackle K. A novel reticular node in the brainstem synchronizes neonatal mouse crying with breathing. Neuron (2022) 110 644-657. DOI: 10.1016/j.neuron.2021.12.014 45. Yeghiazarians Y, Jneid H, Tietjens JR, Redline S, Brown DL, El-Sherif N, Mehra R, Bozkurt B, Ndumele CE, Somers VK. Obstructive sleep apnea and cardiovascular disease: a scientific statement from the American heart association. Circulation (2021) 144 e56-e67. DOI: 10.1161/CIR.0000000000000988 46. Yuan S, Pardue S, Shen X, Alexander JS, Orr AW, Kevil CG. Hydrogen sulfide metabolism regulates endothelial solute barrier function. Redox Biology (2016) 9 157-166. DOI: 10.1016/j.redox.2016.08.004
--- title: MDA5-dependent responses contribute to autoimmune diabetes progression and hindrance authors: - Samuel I. Blum - Jared P. Taylor - Jessie M. Barra - Ashley R. Burg - Qiao Shang - Shihong Qiu - Oren Shechter - Aleah R. Hayes - Todd J. Green - Aron M. Geurts - Yi-Guang Chen - Hubert M. Tse journal: JCI Insight year: 2023 pmcid: PMC9977297 doi: 10.1172/jci.insight.157929 license: CC BY 4.0 --- # MDA5-dependent responses contribute to autoimmune diabetes progression and hindrance ## Abstract Type 1 diabetes (T1D) is an autoimmune disease resulting in pancreatic β cell destruction. Coxsackievirus B3 (CVB3) infection and melanoma differentiation-associated protein 5–dependent (MDA5-dependent) antiviral responses are linked with T1D development. Mutations within IFIH1, coding for MDA5, are correlated with T1D susceptibility, but how these mutations contribute to T1D remains unclear. Utilizing nonobese diabetic (NOD) mice lacking Ifih1 expression (KO) or containing an in-frame deletion within the ATPase site of the helicase 1 domain of MDA5 (ΔHel1), we tested the hypothesis that partial or complete loss-of-function mutations in MDA5 would delay T1D by impairing proinflammatory pancreatic macrophage and T cell responses. Spontaneous T1D developed in female NOD and KO mice similarly, but was significantly delayed in ΔHel1 mice, which may be partly due to a concomitant increase in myeloid-derived suppressor cells. Interestingly, KO male mice had increased spontaneous T1D compared with NOD mice. Whereas NOD and KO mice developed CVB3-accelerated T1D, ΔHel1 mice were protected partly due to decreased type I IFNs, pancreatic infiltrating TNF+ macrophages, IFN-γ+CD4+ T cells, and perforin+CD8+ T cells. Furthermore, ΔHel1 MDA5 protein had reduced ATP hydrolysis compared with wild-type MDA5. Our results suggest that dampened MDA5 function delays T1D, yet loss of MDA5 promotes T1D. ## Introduction Type 1 diabetes (T1D) is a T cell–mediated autoimmune disease resulting in pancreatic β cell destruction [1]. A synergistic effect of genetics, the environment, and the immune system is proposed to induce T1D (2–5). Monozygotic twins have a ≈$30\%$–$50\%$ concordance rate for T1D, which suggests that the environment plays a major role in T1D development [6, 7]. One environmental factor associated with T1D is coxsackievirus B (CVB) infection [8, 9]. CVB viral RNA and/or virus particles have been detected in the blood, stool, and pancreatic islets of patients with recent-onset T1D (9–11). In the nonobese diabetic (NOD) mouse model, CVB infections accelerate T1D by inducing inflammatory pancreatic antiviral responses resulting in β cell destruction [12, 13]. The innate viral sensor melanoma differentiation-associated protein 5 (MDA5), encoded by the IFIH1 gene, detects dsRNA viral replication intermediates and initiates antiviral signaling [14, 15]. One of the key responses of MDA5 after binding its ligand is the synthesis of type I IFNs, such as IFN-α and IFN-β, to promote viral clearance and activation of macrophages, dendritic cells, and T cells (16–20). Although type I IFNs are crucial to antiviral responses, they have also been linked to early T1D development [21, 22]. In transgenic CD1 mice, where β cells constitutively express IFN-α, T1D onset occurs for $60\%$ of the mice by 10 weeks of age [23]. In contrast, loss of IFN-α and -β receptor subunit 1 (IFNAR1) expression in NOD female mice results in a significant delay in T1D development [24]. In patients with T1D, a type I IFN gene signature is detected in the blood prior to autoantibody development [21, 22], and GWAS have found genes associated with T1D that are involved in type I IFN synthesis and signaling, such as IFIH1 [25, 26]. Multiple single nucleotide polymorphisms (SNPs) within IFIH1 are associated with human T1D development. The A946T SNP (rs1990760), which results in an alanine-to-threonine change at amino acid 946, is associated with T1D risk and leads to increased IFN-α/β and IFN-stimulated gene production by human peripheral blood mononuclear cells [27, 28]. Mice carrying the A946T SNP are protected from a lethal viral challenge but at the cost of increased susceptibility to autoimmunity [27]. Conversely, CVB3 infection of human islets homozygous for the A946T SNP results in decreased type III IFN production and improved viral clearance [29]. These seemingly contradictory findings show that further studies are required to fully understand how mutations in IFIH1 result in an increased risk for developing T1D. In contrast, some IFIH1 SNPs are associated with protection from T1D, such as I923V (rs35667974), which results in an isoleucine-to-valine change at amino acid (AA) 923, and E627x (rs35744605), which results in a nonsense mutation and an early stop codon at AA 627 [30]. The I923V SNP results in reduced type I IFN synthesis and ATP hydrolysis and increased dsRNA dissociation [31]. The E627x SNP causes reduced MDA5 expression and reduced type I IFN synthesis [32]. Lincez et al. previously demonstrated that reduced MDA5 expression in NOD.MDA5+/– mice delays spontaneous and CVB type 4–accelerated (CVB4-accelerated) T1D partly due to enhanced regulatory T cells (Tregs) and reduced effector CD4+ T cells in the pancreatic lymph nodes (PLNs), which correlated with reduced pancreatic Ifna mRNA [13]. However, the role of MDA5 on macrophage and T cell responses within the pancreata during spontaneous and CVB-accelerated T1D remains unclear. To further investigate MDA5-dependent antiviral responses in T1D, we used zinc finger nuclease genomic editing to introduce mutations in the helicase 1 domain of MDA5 in NOD mice to recapitulate IFIH1 SNPs that cause reduced MDA5 expression and are associated with a delay in T1D progression [30]. We generated NOD mice with an in-frame 5 AA deletion in the helicase 1 domain of MDA5 (ΔHel1) and an out-of-frame deletion resulting in a premature stop codon in MDA5 (KO). Interestingly, the KO mutation does not lead to any detectable truncated MDA5 protein. The helicase 1 domain senses dsRNA, contains ATPase activity, and interacts with the caspase activation recruitment domain (CARD) to promote antiviral responses [30]. We used these 2 mouse models to explore the effects of mutations in MDA5 in spontaneous and CVB3-accelerated T1D. We hypothesized that partial or complete loss-of-function mutations in MDA5 would delay T1D onset by impairing proinflammatory pancreatic macrophage and T cell responses. Our results show that mutations in MDA5 can influence both spontaneous and CVB3-accelerated T1D. Interestingly, KO mice had no protection from spontaneous or CVB3-accelerated T1D. Male KO mice developed spontaneous T1D at a faster rate compared with NOD male mice. Conversely, ΔHel1 mice had a delay in spontaneous and CVB3-accelerated T1D, partly due to reductions in proinflammatory pancreatic macrophages and T cells. Furthermore, purified ΔHel1 MDA5 protein had reduced ATPase activity compared with wild-type (WT) MDA5 protein. Our data indicate that protection from T1D may be partially intrinsic to reduced MDA5 function and type I IFN synthesis. ## Ifih1ΔHel1 and Ifih1KO mutations affect T1D disease progression in NOD mice. To identify how the loss of MDA5 expression affected T1D development, NOD mice with mutations in Ifih1 were generated by genomic editing with zinc finger nucleases targeting the helicase 1 domain of MDA5. We generated NOD.Ifih1ΔHel1 (ΔHel1) mice with an in-frame deletion at AA 428–432 and NOD.Ifih1KO (KO) mice with an out-of-frame deletion at AA 425–436 resulting in the generation of a premature stop codon (Figure 1A). The effects of Ifih1ΔHel1 and Ifih1KO mutations on spontaneous autoimmune and virus-accelerated diabetes were assessed in male and female NOD, ΔHel1, and KO mice. Uninfected female NOD and KO mice developed T1D similarly, whereas T1D development in ΔHel1 mice was significantly ($P \leq 0.0001$) delayed (Figure 1B). Uninfected male ΔHel1 mice also exhibited significant delays in T1D compared with NOD ($P \leq 0.01$) and KO ($P \leq 0.0001$) mice (Figure 1C). Interestingly, uninfected male KO mice had significant ($P \leq 0.05$) acceleration of T1D compared with NOD mice (Figure 1C), highlighting the potentially novel role of MDA5 to promote or delay T1D. To evaluate the effects of the Ifih1 mutations on virus-accelerated T1D, we infected 12-week-old female and male NOD, ΔHel1, and KO mice with 100 PFU of CVB3/Woodruff and monitored for T1D. CVB3 infected female NOD and KO mice displayed a significant ($P \leq 0.05$) acceleration of T1D and became diabetic as early as 1 week postinfection (Figure 1B). However, female ΔHel1 mice were significantly ($P \leq 0.0001$) delayed from CVB3-accelerated T1D compared with infected NOD and KO mice (Figure 1B). CVB3-infected male NOD ($P \leq 0.01$) and KO ($P \leq 0.05$) mice also displayed a significant acceleration of T1D, whereas CVB3-infected male ΔHel1 mice were significantly ($P \leq 0.01$) delayed compared with infected NOD and KO mice (Figure 1C). To assess if the delay in spontaneous T1D observed in ΔHel1 mice was due to diminished immune responses, we performed an adoptive transfer with splenocytes from euglycemic female NOD, ΔHel1, and KO mice into NOD.Rag recipients. NOD and KO splenocytes induced T1D similarly in recipient mice, but the kinetics of disease transfer with ΔHel1 splenocytes was significantly ($P \leq 0.001$) delayed (Figure 1D). The Ifih1ΔHel1 mutation reduced diabetogenicity of immune cells, but the Ifih1KO mutation did not abrogate autoimmune responses. ## Ifih1 mutations dampen islet infiltration without hindering insulin secretion. To determine if the Ifih1ΔHel1 and Ifih1KO mutations affected glucose homeostasis and β cell function, intraperitoneal glucose tolerance test (IPGTT) and glucose-stimulated insulin secretion (GSIS) assays were performed on NOD, ΔHel1, and KO mice. No differences in glucose clearance were observed by IPGTT, as all mice returned to euglycemia by 120 minutes postinjection (Supplemental Figure 1, A–D; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.157929DS1). To validate these results, we performed a GSIS assay on islets, and no differences were observed in insulin secretion (Supplemental Figure 1, E and F). Furthermore, no changes were observed in the body weight of male and female NOD and ΔHel1 mice. Female KO mice had significantly ($P \leq 0.001$) reduced body weight compared with ΔHel1 mice but not NOD mice (Supplemental Figure 1G). The decreased body weight in female KO mice did not compromise their ability to thrive, as negative effects on health were not observed. Finally, no differences in body weight were observed in male mice (Supplemental Figure 1H). We next performed insulitis scoring on NOD, ΔHel1, and KO mice to assess differences in pancreatic islet infiltration. ΔHel1 and KO islets had a significant ($P \leq 0.05$ and $P \leq 0.01$, respectively) reduction in immune cell infiltration compared with NOD islets at 6 weeks of age. At 12, 16, and 20 weeks of age, insulitis scores from ΔHel1 mice were significantly ($P \leq 0.05$) reduced compared with NOD and KO mice (Figure 2, A and B). The Ifih1ΔHel1 mutation delayed T1D development partly due to reduced immune cell infiltration of islets without compromising β cell function. ## Mutations in Ifih1 lead to reduced pancreatic proinflammatory macrophage and T cell populations. Since CVB has a tropism for the pancreas and can induce macrophage and T cell infiltration, leading to the destruction of infected pancreatic exocrine and endocrine cells (33–35), we investigated if the Ifih1ΔHel1 and Ifih1KO mutations affected pancreatic macrophage and T cell populations. We analyzed pancreatic macrophages from both uninfected and CVB3-infected NOD, ΔHel1, and KO female mice by flow cytometry at 7 days postinfection. Pancreatic macrophage (F$\frac{4}{80}$+CD11b+) frequency and cell counts were unaltered between uninfected NOD, ΔHel1, and KO mice (Supplemental Figure 2, A and B). However, there was a significant increase in F$\frac{4}{80}$+CD11b+ macrophage cell counts following CVB3 infection compared with uninfected controls (Supplemental Figure 2B), while frequency remained unaltered (Supplemental Figure 2A). The activation status of macrophages was determined by MHC-II, CD80, and TNF expression. We observed no significant differences in frequency and cell count of activated CD80+F$\frac{4}{80}$+CD11b+ macrophages and MHC-II+F$\frac{4}{80}$+CD11b+ macrophages between all groups of uninfected and CVB3-infected mice (data not shown). However, there was a significant reduction in TNF+F$\frac{4}{80}$+CD11b+ macrophage frequency in uninfected ΔHel1 (≈1.6-fold, $P \leq 0.001$) and KO (≈1.3-fold, $P \leq 0.05$) mice compared with uninfected NOD mice (Figure 3, A and C). Uninfected ΔHel1 mice also had a ≈1.6-fold ($P \leq 0.05$) decrease in cell count compared with uninfected NOD, but no differences were observed between uninfected NOD and KO mice (Figure 3, B and C). At day 7 postinfection, pancreatic TNF+F$\frac{4}{80}$+CD11b+ macrophage cell counts from CVB3-infected ΔHel1 mice were significantly reduced when compared with NOD (≈2.5-fold, $P \leq 0.001$) and KO (≈2.0-fold, $P \leq 0.01$) mice (Figure 3, B and C). However, no differences in frequency were observed following CVB3 infection; this may be due to the large influx of macrophages into NOD and KO pancreata compared with ΔHel1 mice (Figure 3, A and C). Therefore, loss of MDA5 in KO mice did not alter inflammatory macrophages, while both uninfected and CVB3-infected ΔHel1 mice had a reduction in proinflammatory macrophages within the pancreata, which may partly explain the delay in both spontaneous and CVB3-accelerated T1D. Because the Ifih1ΔHel1 mutation decreased proinflammatory macrophage responses, we next examined the effect on pancreatic T cell effector responses from CVB3-infected NOD, ΔHel1, and KO mice at 7 days postinfection. Following CVB3 infection, KO mice had significantly fewer (≈1.8-fold, $P \leq 0.05$) CD4+ T cells within the pancreata compared with infected NOD and ΔHel1 mice (Supplemental Figure 2D) and significantly fewer (≈1.6-fold, $P \leq 0.05$) CD8+ T cells compared with infected NOD mice (Supplemental Figure 2F). However, no differences in the frequency or cell counts of total pancreas-infiltrating CD4+ or CD8+ T cells were observed between uninfected NOD, ΔHel1, or KO mice (Supplemental Figure 2, C–F). Furthermore, we observed no significant differences between activated CD69+CD4+ or CD69+CD8+ T cells between uninfected or CVB3-infected NOD, ΔHel1, and KO mice (data not shown). Conversely, the effector response of pancreatic CD4+ and CD8+ T cells was different in mice containing Ifih1ΔHel1 and Ifih1KO mutations. Uninfected NOD and KO mice had similar frequencies and cell counts of IFN-γ+CD4+ T cells (Figure 3, D–F) and following phorbol 12-myristate 13-acetate and ionomycin (PMA/I) stimulation (Supplemental Figure 3A). Conversely, ΔHel1 mice had a significant reduction in pancreatic IFN-γ+CD4+ T cell counts compared with NOD (≈2.4-fold, $P \leq 0.05$) and KO (≈2.3-fold, $P \leq 0.05$) mice (Figure 3, E and F). Similar decreases were also observed with IFN-γ+CD4+ T cells from ΔHel1 mice compared with NOD (≈2.3-fold, $P \leq 0.05$) and KO (≈2.1-fold, $P \leq 0.05$) mice following PMA/I stimulation (Supplemental Figure 3A). Even though there was no statistical difference in the frequency of IFN-γ+CD4+ T cells from ΔHel1 mice, the mean frequency of IFN-γ+CD4+ T cells from ΔHel1 mice was reduced compared with NOD and KO (Figure 3, D and F, and Supplemental Figure 3A). At day 7 postinfection with CVB3, there were no significant differences in the effector response of IFN-γ+CD4+ T cell frequencies between uninfected or CVB3-infected NOD, ΔHel1, and KO mice, but the mean frequency was reduced in ΔHel1 mice compared with CVB3-infected NOD and KO (Figure 3, D and F). With respect to cell numbers, CVB3-infected ΔHel1 (≈1.9-fold, $P \leq 0.001$) and KO (≈1.5-fold, $P \leq 0.05$) mice had significantly fewer pancreatic IFN-γ+CD4+ T cells compared with NOD mice (Figure 3, E and F). Uninfected ΔHel1 mice also had a significant reduction in pancreatic perforin+CD8+ T cell frequency (≈2.7-fold, $P \leq 0.05$) and cell count (≈3.9-fold, $P \leq 0.05$) compared with NOD, but no difference was observed compared to KO mice (Figure 3, G–I). Following CVB3 infection, we observed that KO mice had a significant increase (≈1.3-fold, $P \leq 0.01$) in perforin+CD8+ T cell frequency compared with CVB3-infected ΔHel1 mice, but no difference between CVB3-infected NOD and ΔHel1 mice was observed (Figure 3, G and I). CVB3-infected ΔHel1 (≈2.6-fold, $P \leq 0.0001$) and KO (≈1.7-fold, $P \leq 0.01$) mice had significantly fewer pancreatic perforin+CD8+ T cells compared with infected NOD mice (Figure 3, H and I). The discrepancy between frequency and cell count of IFN-γ+CD4+ and perforin+CD8+ T cells from CVB3-infected KO mice was due to significantly fewer total CD4+ and CD8+ T cells within the pancreata of KO mice compared with NOD mice. CVB3-infected ΔHel1 and KO mice had similar reductions in IFN-γ+CD4+ and perforin+CD8+ T cell counts, indicating that MDA5-dependent antiviral responses are necessary for efficient T cell effector responses. However, uninfected ΔHel1 mice had fewer IFN-γ+CD4+ and perforin+CD8+ T cells, which may explain the delay in spontaneous T1D development. Furthermore, Tregs play a critical role in peripheral tolerance and delaying T1D, but the frequency and cell counts of pancreatic CD25+FoxP3+CD4+ T cells were unaltered in NOD, ΔHel1, and KO mice (Supplemental Figure 3B). However, CD25+FoxP3+CD4+ T cell frequency was significantly (≈1.2-fold, $P \leq 0.05$) reduced in KO PLNs compared with ΔHel1 mice during T1D development and was unaltered compared to NOD (Supplemental Figure 3C). Reduced Tregs within the PLNs of KO mice may partly explain the inability of these mice to delay spontaneous T1D (Figure 1, B and C). ## Ifih1ΔHel1 mutation enhances myeloid-derived suppressor cell populations. One subset of innate immune cells that can regulate proinflammatory macrophages and T cells are myeloid-derived suppressor cells (MDSCs) [36, 37]. MDSCs are either neutrophil like (PMN-MDSCs) or monocyte like (M-MDSCs) and have potent immune suppressive function via arginase-1, nitric oxide synthase, reactive oxygen species, IL-10, TGF-β, IL-1β, and programmed cell death ligand 1 (38–40). These suppressor cells have been suggested to play a major role in preventing T1D. NOD mice adoptively transferred with MDSCs are protected from T1D development [41], and patients with T1D are reported to have reduced MDSC suppressive activity compared with healthy controls [42]. Studies in pancreatic cancer have suggested a link between MDA5, type I IFN signaling, and MDSC function, but type I IFNs can have divergent effects on MDSC function. Too much type I IFN signaling can result in dampened MDSC suppressor activity, whereas a complete loss of type I IFNs can result in impaired MDSC development (43–45). Previous studies have shown that poly(I:C) stimulation of MDA5 in MDSCs induces type I IFN synthesis, which dampens their suppressive capacity [45, 46]. However, a complete loss of IFNAR1 on MDSCs prevents their development and suppressive activity (43–45). These findings provide evidence that an optimal amount of type I IFN activity is necessary for MDSC development and function. Interestingly, reduced surface expression of IFNAR1 on MDSCs may increase their suppressive activity by reducing type I IFN signaling [43], suggesting that fine-tuning of type I IFN signaling may affect MDSC function. Given the importance of MDA5 and type I IFNs on MDSC suppressive function and development, we hypothesized that MDSC populations may be altered within ΔHel1 and KO mice. Since ΔHel1 mice have delayed T1D, but KO mice still develop T1D as do NOD mice, we investigated if MDSC populations were enhanced in ΔHel1 mice or dampened in KO mice. We analyzed MDSCs in the spleen, bone marrow, pancreata, and PLNs of 12-week-old NOD, ΔHel1, and KO female mice during spontaneous T1D. The frequency of PMN-MDSCs was not different (Figure 4A), but cell counts were significantly increased within ΔHel1 bone marrow compared with NOD (≈1.2-fold, $P \leq 0.001$) and KO mice (≈1.1-fold, $P \leq 0.05$) (Figure 4B). M-MDSC frequency was also increased in the ΔHel1 spleen (≈1.3-fold, $P \leq 0.05$), pancreata (≈1.8-fold, $P \leq 0.0001$), and PLNs (≈3.4-fold, $P \leq 0.001$) compared with NOD (Figure 4C). The overall cell count of ΔHel1 M-MDSCs was also increased within all organs but significantly increased in the bone marrow compared with NOD (≈1.4-fold, $P \leq 0.001$) and KO (≈1.3-fold, $P \leq 0.01$) mice (Figure 4D), whereas compared with KO mice, M-MDSC frequency was increased in the ΔHel1 spleen (≈1.3-fold, $P \leq 0.05$), pancreata (≈2.1-fold, $P \leq 0.0001$), and PLNs (≈1.5-fold, $$P \leq 0.066$$) (Figure 4C). These results suggest that the Ifih1ΔHel1 mutation enhances MDSC populations, which may partially explain the delay in T1D development (Figure 1). Since KO mice were not protected from T1D similar to NOD mice, loss of MDA5 may not promote MDSC development or may lead to MDSC deficiencies. We found no differences in PMN-MDSC or M-MDSC count or frequency in the spleen, bone marrow, pancreata, or PLNs of KO and NOD mice (Figure 4). These findings corroborate prior studies that type I IFN/IFNAR signaling is necessary for MDSC differentiation [43] and may partly explain how the loss of MDA5 expression in KO mice does not delay T1D (Figure 1). ## Ifih1ΔHel1 mutation leads to reduced MDA5 expression following MDA5-specific stimulation. To verify that the Ifih1KO mutation resulted in a truncated form of or loss in MDA5 expression, we used an MDA5 antibody with specificity for AA 1–205 of the CARD in MDA5. Western blot analysis of MDA5 was detected in bone marrow–derived macrophages (BMDMs) from NOD and ΔHel1 BMDMs stimulated with low–molecular weight poly(I:C), but MDA5 expression was absent in KO BMDMs (Supplemental Figure 4). To examine the effect of Ifih1 mutations on MDA5 expression in macrophages, we stimulated BMDMs from NOD, ΔHel1, and KO mice with lipopolysaccharide (LPS), transfected high–molecular weight (HMW) poly(I:C), or CVB3. Western blot analysis of MDA5 showed that stimulation of BMDMs from NOD mice with LPS, HMW poly(I:C), or CVB3 increased MDA5 expression (Figure 5, A and B). However, BMDMs from ΔHel1 mice had a significant reduction in MDA5 expression after stimulation with HMW poly(I:C) (≈2.0-fold; $P \leq 0.0001$) and CVB3 (≈4.8-fold; $P \leq 0.05$) compared with NOD, but no differences were observed following stimulation with LPS. BMDMs from KO mice had no detectable MDA5 protein expression before or after stimulation (Figure 5, A and B). Reduced MDA5 expression from ΔHel1 BMDMs, compared with NOD, suggest MDA5 responses and type I IFN synthesis may act as a positive feedback mechanism to further upregulate MDA5 expression [47, 48]. RIG-I expression was significantly downregulated in BMDMs from ΔHel1 (≈2.0-fold, $P \leq 0.05$ and ≈1.4-fold, $P \leq 0.0001$) and KO (≈3.1-fold, $P \leq 0.01$ and ≈1.4-fold, $P \leq 0.0001$) mice following stimulation with HMW poly(I:C) and CVB3, respectively, compared with BMDMs from NOD mice (Figure 5, A and C). See complete unedited blots in the supplemental material. We also detected p-STAT1 (Y701) expression and observed a significant downregulation in BMDMs from ΔHel1 (≈3.0-fold, $P \leq 0.01$ and ≈4.6-fold, $P \leq 0.0001$) and KO (≈2.4-fold, $P \leq 0.05$ and ≈5.9-fold, $P \leq 0.0001$) mice following stimulation with HMW poly(I:C) and CVB3, respectively, compared with NOD (Figure 5, A and D). Reduced RIG-I and p-STAT1 (Y701) expression in BMDMs from both ΔHel1 and KO mice following MDA5 stimulation, compared with NOD, indicates that decreased MDA5 expression dampens type I IFN–mediated responses [49]. ## ΔHel1 mice have improved viral clearance and reduced pancreatic IFN-α and IFN-β levels postinfection. To evaluate the effects of the Ifih1 mutations on antiviral responses, we infected 12-week-old female and male NOD, ΔHel1, and KO mice with 100 PFU of CVB3/Woodruff and monitored viral clearance and pancreatic type I IFN production. Following CVB3 infection, viral clearance was determined by pancreatic viral titer on days 1, 3, 7, 10, and 14 postinfection in NOD, ΔHel1, and KO mice. Peak CVB3 pancreatic viral titer was observed on day 3 postinfection within all mice, but ΔHel1 mice demonstrated a significant reduction on day 7 postinfection compared with both NOD (≈7,636-fold, $P \leq 0.01$) and KO (≈1,637-fold, $P \leq 0.05$) mice (Figure 6A). Corroborating the increase in viral titer, pancreatic IFN-α and IFN-β levels were maximal at day 3 postinfection and returned to basal levels by day 7 postinfection in all mice (Figure 6, B and C). At day 3 postinfection, CVB3-infected ΔHel1 mice had a significant ≈2.2-fold ($P \leq 0.0001$) reduction in pancreatic IFN-α (Figure 6D) and ≈3-fold ($P \leq 0.0001$) decrease in IFN-β (Figure 6E) compared with NOD mice. CVB3-infected KO mice had a significant ≈8.5-fold ($P \leq 0.0001$) and ≈3.9-fold ($P \leq 0.05$) reduction in pancreatic IFN-α compared with NOD and ΔHel1 mice, respectively (Figure 6D). KO IFN-β levels were also significantly reduced by ≈19.2-fold ($P \leq 0.0001$) compared with NOD mice, but no significant differences were observed between ΔHel1 and KO (Figure 6E). The delay in CVB3-accelerated T1D observed in ΔHel1 mice may be due to the optimal synthesis of type I IFNs necessary for viral clearance without inducing pancreatic inflammation and autoimmune activation. However, KO mice fail to produce robust levels of type I IFNs in response to CVB3 infection, which may contribute to CVB3-accelerated T1D without impairing viral clearance. ## Ifih1ΔHel1 mutation reduces MDA5-mediated ATP hydrolysis. MDA5 ATPase activity has been suggested to be a critical step required for MDA5 filament formation and disassembly, as well as its ability to interact with mitochondrial antiviral signaling protein (MAVS), which is required for downstream antiviral responses and type I IFN synthesis [31, 50, 51]. Since the Ifih1ΔHel1 mutation is within an ATPase motif of the helicase 1 domain of MDA5 and leads to reduced type I IFN synthesis (Figure 6, B–E), we hypothesized that reduced ΔHel1 immune responses and delayed T1D may be partly due to dampened ATPase activity in MDA5. We purified core WT and ΔHel1 MDA5 protein without CARDs to measure ATP hydrolysis. To determine the purity of our MDA5 samples, we separated our purified fractions by SDS-PAGE and stained the gel with GelCode Blue. Coomassie staining of WT and ΔHel1 MDA5 proteins revealed a prominent band around 83 kDa, with ΔHel1 MDA5 having a reduced molecular mass (Figure 7A), consistent with the predicted molecular weight, 83.0 and 82.6 kDa, respectively. Our protein samples were also probed for MDA5 by Western blot, and a specific band for MDA5 around 83 kDa was detected (Figure 7B). See complete unedited blots in the supplemental material. Utilizing our purified WT and ΔHel1 MDA5 protein in an ATPase assay, we observed that ΔHel1 MDA5 protein was functional, but had a significant (≈4.3-fold, $P \leq 0.0001$) reduction in ATP hydrolysis following poly(I:C) stimulation, compared with WT MDA5 protein (Figure 7C). Collectively, these findings indicate that dampened MDA5 ATPase activity in the ΔHel1 mouse may partly explain reduced proinflammatory immune cell responses and a delay in both spontaneous and CVB3-accelerated T1D. ## Discussion CVB, mumps, rubella, and cytomegalovirus infections are linked to T1D development (52–57). Many studies focused on the link between CVB and T1D indicate that CVB may accelerate T1D development by inducing proinflammatory MDA5-dependent antiviral responses and bystander activation of T cells (58–60). IFIH1 SNPs are associated with T1D [30], but how IFIH1 mutations affect diabetogenicity is poorly understood. To investigate the role of MDA5 in T1D, we generated an in-frame deletion within the helicase 1 domain of MDA5 (Ifih1ΔHel1) and an out-of-frame deletion (Ifih1KO) in NOD mice with zinc finger nuclease–mediated (ZFN-mediated) gene targeting [61]. The human T1D protective IFIH1 alleles are associated with lower MDA5 expression or activity [27, 32, 62, 63] To mimic this, we investigated if the complete absence of MDA5 expression or its reduced activity can similarly confer T1D protection. The Ifih1ΔHel1 mutation delayed both uninfected/spontaneous and CVB3-accelerated T1D, which was partly due to reductions in MDA5-mediated ATP hydrolysis, IFN-α/β synthesis, TNF+ macrophages, IFN-γ+CD4+ T cells, and perforin+CD8+ T cells in the pancreata. Therefore, decreased MDA5 function can reduce proinflammatory effector responses in T1D. However, the Ifih1KO mutation did not delay spontaneous or CVB3-accelerated T1D in female mice and, surprisingly, enhanced spontaneous T1D in male mice. Uninfected KO mice exhibited no reduction in pancreatic TNF+ macrophages, IFN-γ+CD4+ T cells, and perforin+CD8+ T cells during spontaneous T1D. While CVB3-infected KO mice had fewer pancreatic IFN-γ+CD4+ and perforin+CD8+ T cells compared with NOD mice, TNF+ macrophages were still elevated. However, during spontaneous T1D development, ΔHel1 mice had increased MDSC populations compared with NOD and KO, but no alteration in MDSCs was observed in KO mice compared to NOD. Failure to increase MDSCs in KO mice may partly explain their susceptibility to developing T1D similar to NOD mice. Our results demonstrated that MDA5-dependent responses can dictate T1D progression partly mediated by pancreas-infiltrating macrophages. Macrophages regulate inflammatory responses in the islet, facilitate T cell recruitment, and activate autoreactive T cells [64, 65]. NOD and KO mice had robust proinflammatory pancreatic macrophages during spontaneous and CVB3-accelerated T1D, but TNF+ macrophage populations were reduced in ΔHel1 mice. ΔHel1 pancreatic macrophages are inherently less inflammatory, possibly preventing β cell destruction and T1D. Effector CD4+ and CD8+ T cells mediate β cell destruction and T1D development [66]. During spontaneous T1D development, ΔHel1 mice had reduced numbers of pancreatic IFN-γ+CD4+ and perforin+CD8+ T cells compared with NOD and KO mice, which may partly explain the delay in T1D progression. However, both ΔHel1 and KO mice had reduced numbers of IFN-γ+CD4+ and perforin+CD8+ T cells compared with NOD following CVB3 infection. These findings provide evidence that MDA5-dependent type I IFN synthesis is necessary for maturing CD4+ and CD8+ T cell effector responses during CVB3-accelerated T1D. The decrease in type I IFN synthesis in the pancreata of CVB3-infected ΔHel1 mice may not be sufficient for maturing CD4+ and CD8+ T cell antiviral effector responses, thereby delaying CVB3-accelerated T1D. Interestingly, KO mice still developed spontaneous and CVB3-accelerated T1D but failed to produce heightened levels of type I IFNs within the pancreata following infection. The development of spontaneous T1D in KO mice may be due to an increase in proinflammatory pancreatic macrophages and effector T cells and, concomitantly, reduced Treg and/or MDSC populations. Our results with the KO mouse show that MDA5-mediated type I IFNs may be necessary to prevent T1D. Robust type I IFN synthesis during viral infections inhibits Treg activation and proliferation, but a complete loss of type I IFN signaling in Tregs impairs Treg FoxP3 expression and suppressor function [67, 68]. Although we observed no difference in pancreatic Treg populations between NOD, ΔHel1, and KO mice, there was a marked reduction in Tregs from the PLNs of KO mice compared with ΔHel1 (Supplemental Figure 3). Whether loss of MDA5 in KO mice impairs Treg suppressive function due to diminished type I IFN signaling needs to be determined. Another immunosuppressive immune cell type that may be influenced by MDA5-dependent signals is MDSCs. These cells produce immunomodulatory molecules that dampen inflammatory immune responses [38]. Impaired MDSC function has been linked to human T1D [42], and adoptive transfer of MDSCs can delay T1D in NOD mice [41]. In pancreatic cancer models, type I IFNs play an important role in regulating MDSC suppressor activity. Too much or too little type I IFN signaling is detrimental to MDSC suppressor function, but an intermediate amount of type I IFN signaling may result in optimal MDSC function (43–45). These reports support the differences we observed in uninfected/spontaneous and CVB3-accelerated T1D with ΔHel1 and KO mice and provide evidence that altered type I IFN signaling may affect MDSC function. We observed an increase in MDSCs in ΔHel1 mice within multiple organs, including the spleen, bone marrow, pancreata, and PLNs. These findings suggest that MDSCs in ΔHel1 mice may suppress inflammatory pancreatic macrophage and T cell responses, thereby delaying both spontaneous and CVB3-accelerated T1D. During spontaneous T1D, MDSC populations between NOD and KO mice were comparable and may partly explain why KO mice still develop autoimmune diabetes similar to NOD mice. It remains plausible that NOD mice lose MDSC suppressor function during spontaneous T1D and the absence of MDA5-dependent type I IFN synthesis can contribute to a loss in MDSC suppressor function as observed in our KO model. Future studies will examine if MDSCs or Tregs from ΔHel1 and KO mice exhibit enhanced or defective immunosuppressive function, respectively, in contrast to NOD mice. Our ΔHel1 mouse validated and expanded upon previous findings by Lincez et al. with NOD.MDA5+/– mice, which had delayed spontaneous and CVB4-accelerated T1D due to decreased MDA5 expression, had reduced Ifna mRNA, had a dampened CD4+ T cell effector response, and had a concomitant increase in Treg populations in the PLNs [13]. Although we observed no differences in Treg populations in NOD, ΔHel1, and KO mice, it is plausible that ΔHel1 mice may exhibit an increase in Treg suppressive function without affecting Treg numbers. Nevertheless, our 2 independent studies using different coxsackievirus strains and mouse models highlight the importance of MDA5-dependent responses in T1D. Mutations in IFIH1 are also associated with other autoimmune diseases, including multiple sclerosis (MS) [69, 70], systemic lupus erythematosus (SLE) [71, 72], and rheumatoid arthritis (RA) [73, 74]. Therefore, it is important to define how genetic mutations in IFIH1 contribute to MDA5 function, since this knowledge would apply not only to T1D but also to other autoimmune diseases. Patients with SLE and with an MDA5 gain-of-function mutation R779H (rs587777446) have increased IFN-α serum levels [75], which may be due to dysregulated helicase ATP hydrolysis and dsRNA binding [76]. The A946T and R843H SNPs within IFIH1 are associated with T1D risk. The A946T SNP results in increased MDA5 function and type I IFN synthesis, but the effect of the R843H SNP (rs3747517) remains unknown [30]. Conversely, some IFIH1 SNPs, such as E627x and I923V, are associated with protection [30]. The E627x SNP leads to reduced MDA5 expression and type I IFN synthesis due to a premature stop codon [32]. The I923V SNP also has reduced type I IFN synthesis, but this appears to be due to I923V MDA5 forming shorter filaments, having decreased ATP hydrolysis and enhanced dsRNA dissociation [27, 31]. The protective effect of E627x and I923V SNPs in IFIH1 appears to be due to either reduced MDA5 expression or reduced ATP hydrolysis and subsequently, diminished type I IFN synthesis. Gorman et al. showed that overexpression of the A946T SNP causes increased IFNB1 expression, while the I923V SNP leads to reduced expression of IFNB1 [27]. It remains unclear if the A946T SNP promotes T1D due to increased basal MDA5 activity [27] or altered response to ligands, such as self-dsRNA, CVB3, or endogenous retroelements (77–79). Nevertheless, MDA5 activity and signaling may be a key driving factor in T1D development by promoting type I IFN synthesis. Further investigation is warranted to determine whether mutations in MDA5 results in abnormal ligand binding and/or basal MDA5 activity leading to autoimmunity. ATPase activity has been shown to tightly regulate the stability of MDA5 during filament formation and disassembly in response to dsRNA [31] and is critical to prevent MDA5 binding to self-RNA [80]. MDA5 models indicate that regulation of MDA5 disassembly by ATP hydrolysis may be required for MDA5-dependent interaction with MAVS and subsequent antiviral signaling [31, 50, 51]. The Ifih1ΔHel1 mutation is located within an ATPase motif in the helicase 1 domain of MDA5. ΔHel1 MDA5 protein has functional ATPase activity following poly(I:C) stimulation, but compared with WT MDA5 protein, ATP hydrolysis was reduced, which may result in dampened signaling downstream of MDA5. A subsequent effect of reduced MDA5 function appears to be reduced type I IFN synthesis and immune cell activation in pancreata, leading to a delay in spontaneous and CVB3-accelerated T1D. Our results may parallel the phenotype of dampened ATP hydrolysis in the I923V SNP, thereby reducing type I IFN synthesis and delaying T1D development. Our study may provide the rationale for the development of small molecule inhibitors that target the ATPase motifs within the helicase domains of MDA5 to reduce type I IFN synthesis. This potentially novel therapeutic approach may dampen autoreactive T cell responses, decrease β cell damage, and delay T1D development similar to the protection observed in ΔHel1 mice. Future translational studies are warranted to assess whether alterations of the helicase 1 domain in human MDA5 can affect innate and adaptive immune responses and, subsequently, elicit immunoprotection against autoimmune diabetes. In addition to T1D, MDA5 inhibitors may also be useful in the treatment of other autoimmune diseases where exacerbated MDA5 responses may play a role, such as SLE [71, 72], MS [69, 70], or RA [73, 74]. ## Animals. NOD/ShiLtJ (NOD) and NOD.Rag mice were purchased from The Jackson Laboratory, while NOD.Ifih1ΔHel1 (ΔHel1) and NOD.Ifih1KO (KO) mice were provided by the Medical College of Wisconsin, Department of Pediatrics. All mice were bred and housed under pathogen-free conditions in the Research Support Building animal facility at The University of Alabama at Birmingham. ΔHel1 and KO mice were generated by ZFN-mediated gene targeting as described [61]. Constructs of the ZFN pairs specifically targeting exon 6 of the mouse Ifih1 gene were designed, assembled, and validated by MilliporeSigma (target sequence ATCTGGAGAGTGGAGAcgatgACGGTGTGCAGCTGTCAGG; ZFNs bind to each sequence shown in uppercase on opposite strands). mRNAs encoding ZFN pairs were prepared in injection buffer (1 mM Tris-HCl, 0.1 mM EDTA, pH 7.4) at a concentration between 5 and 10 ng/μL and injected into the pronucleus of fertilized NOD 1-cell embryos. Injected embryos were transferred to pseudopregnant CD-1 females (Charles River). Tail DNA was extracted and screened for ZFN-induced mutation by PCR amplification with forward (5′-TGGATTAAGTGGCGATACCC-3′) and reverse (5′-TTTTCAGGGAAGTGGAGCAC-3′) primers and standard sequencing. Identified founders were backcrossed to NOD mice followed by intercrossing to fix the mutation to homozygosity. Mice were maintained on a 12-hour light/12-hour dark cycle at 23°C and received standard lab chow and acidified water weekly. ## In vivo infections, diabetes incidence, and viral plaque assays. NOD, ΔHel1, and KO male and female mice at 12 weeks of age were infected by i.p. injection with 100 PFU CVB3/Woodruff in HBSS or HBSS control as published [58]. Diabetes incidence was monitored every other day by glucosuria (Diastix) and confirmed by 2 consecutive blood glucose readings ≥ 300 mg/dL with a Contour Next meter (Ascensia Diabetes Care) until 40 weeks of age. Pancreatic viral titers were performed as described [81]. ## Insulitis scoring. Pancreata from mice were fixed in $4\%$ (w/v) paraformaldehyde dissolved in phosphate buffer (0.12 M; pH 7.4), processed, and embedded in paraffin. Pancreata were sectioned, then stained with hematoxylin and eosin, and insulitis scoring was performed as we published [82]. ## IPGTT and GSIS. Mice were fasted for 15 hours, followed by an intraperitoneal injection of 2 g/kg body weight sterile filtered $20\%$ glucose solution in PBS. Blood glucose was measured at 0, 5, 15, 30, 60, 90, and 120 minutes postinjection as described above. GSIS was performed on islets from 12-week-old male NOD, ΔHel1, and KO mice as previously described [83]. ## Flow cytometry. Flow cytometry was performed on pancreatic cells from NOD, ΔHel1, and KO female mice at 12 weeks of age and following CVB3 infection as published [13]. Pancreata were harvested in 2 mL of RPMI 1640 (Invitrogen, Thermo Fisher Scientific) with 300 U/mL of collagenase type 4 (Worthington, LS004188), digested in a 37 °C water bath for 15 minutes with agitation every 5 minutes, and homogenized with a Dounce homogenizer. Samples were treated with GolgiPlug (BD Biosciences) with or without 100 ng/mL PMA and 1 μg/mL ionomycin, Fc receptors were blocked, and surface or intracellular flow cytometry was performed with fluorochrome-conjugated antibodies as previously described [84] (Supplemental Table 1A). For intracellular staining, cells were fixed and permeabilized with eBioscience (Thermo Fisher Scientific) FoxP3 transcription factor fix/perm overnight. Cells were analyzed with an Attune NxT flow cytometer (Thermo Fisher Scientific) with ≈1,000,000 events/sample and analyzed with FlowJo version 10.6.2 software. The flow cytometry gating scheme is shown in Supplemental Figures 5 and 6. The average viability of pancreatic samples digested with collagenase was about $65\%$–$80\%$, and any sample with less than $50\%$ viability was excluded from analysis. ## BMDMs. BMDMs were generated from NOD, ΔHel1, and KO male mice at 8–12 weeks of age as previously described [85]. BMDMs were stimulated with 1 μg/mL of transfected HMW poly(I:C) using LyoVec (InvivoGen), 1 μg/mL LPS (E. coli 055:B5), or 100 MOI CVB3/Woodruff. ## Western blot and Coomassie staining. MDA5, RIG-I, p-STAT1 (Y701), and STAT1 expression in untreated or poly(I:C)-, LPS-, or CVB3-stimulated BMDM whole-cell lysates was detected by Western blot analysis as previously described [86]. Purified proteins from size-exclusion chromatography were separated by SDS-PAGE, Coomassie stained with GelCode Blue (Thermo Fisher Scientific), and then probed for MDA5 by Western blot analysis. Proteins were detected by incubation with primary antibodies (Supplemental Table 1B) followed by an anti-rabbit IRDye $\frac{680}{800}$CW secondary antibody (LI-COR Biosciences), visualized on an Odyssey CLx Imager with Image Studio v4.0 software (LI-COR Biosciences) to calculate densitometry, and normalized to β-actin and unstimulated controls. ## Adoptive transfer of diabetes. Sixteen-week-old nondiabetic female NOD, ΔHel1, and KO spleens were resuspended at 108 cells/mL in HBSS. Ten million splenocytes were transferred intravenously into 10-week-old female NOD.Rag mice, which were monitored for diabetes as described above. ## IFN-α/β ELISA. Pancreata were collected in 1 mL of PBS containing $14.29\%$ of Protease Inhibitor Cocktail (Roche, MilliporeSigma; 11836153001), homogenized with an electric homogenizer, and centrifuged at 12,000 RCF for 10 minutes at 4°C. Supernatants were transferred to new tubes and frozen overnight. Samples were thawed, centrifuged (at 12,000 RCF for 10 minutes at 4°C), and used in an IFN-α and IFN-β ELISA (PBL Assay Science, 42120-2 and 42400-2) according to the manufacturer’s protocol. IFN-α and IFN-β levels were normalized to total protein measured by BCA assay (Thermo Fisher Scientific, 23227). Plates were read on a Synergy2 microplate reader, and the data were analyzed with Gen5 v.1.10 software (BioTek). ## Expression and purification of MDA5 protein. Mouse MDA5 (Ifih1) with CARDs removed (AA 304–1025) was subcloned into the pET His TEV LIC vector (Addgene) with an N-terminal hexa-histidine tag and a tobacco etch virus (TEV) protease cleavage site [50, 51]. The Hel2i L2 surface loop AA 646–663 was deleted to increase solubility [51]. As previously shown, deletion of the L2 loop does not alter MDA5 ATPase activity, type I IFN signaling, or dsRNA binding [27, 51, 87, 88]. The ΔHel1 mutation was introduced by deleting AA 428–432, using a Q5 Site-Directed Mutagenesis Kit (New England Biolabs), and both WT and ΔHel1 constructs were verified by DNA sequencing. ## E. coli BL21 (DE3) cells were transformed with MDA5 constructs and grown to OD600 0.6–0.8 absorbance units at 37°C, as described [51]. The temperature was reduced to 18°C, and protein expression was induced with 0.25 mM isopropyl-β-D-1-thiogalactopyranoside (Fisher Scientific) overnight. Cells were harvested by centrifugation (6,200 RCF for 15 minutes at 4°C) and resuspended in 20 mM Tris, 0.5 M NaCl, and 5 mM imidazole at pH 7.9, then lysed by ultrasonication on ice. Lysates were spun at 27,000g for 45 minutes at 4°C, and supernatant was loaded onto Ni-NTA agarose (Thermo Fisher Scientific). Protein was washed with 50 mM HEPES at pH 7.5, 0.15 M NaCl, $5\%$ glycerol, 20 mM imidazole, and 8 mM 2-mercaptoethanol (β-ME), and MDA5 was eluted with 50 mM HEPES pH 7.5, 0.15 M NaCl, $5\%$ glycerol, 0.3 M imidazole, and 8 mM β-ME. MDA5 was then purified on a heparin column (GE) (buffer A: 20 mM HEPES pH 7.5, 0.1 M NaCl, 2 mM dithiothreitol [DTT]; buffer B: 20 mM HEPES pH 7.5, 1 M NaCl, 2 mM DTT), then a Superdex 200 $\frac{10}{300}$ GL size-exclusion column (GE) by NaCl gradient in buffer 20 mM HEPES pH 7.5 and 2 mM DTT. Final protein was purified by size-exclusion chromatography on a Superdex 200 $\frac{10}{300}$ GL column in 20 mM HEPES pH 7.5, 0.5M NaCl, 5 mM MgCl2, and 2 mM DTT, similar to methods described [51]. ## ATPase activity assay. ATP hydrolysis by MDA5 was measured by the Malachite green assay (MilliporeSigma, MAK113-1KT). Purified MDA5 at 37.5 and 75.0 nM was incubated with 4 μg/mL of HMW poly(I:C) (InvivoGen) and 4 mM ATP (MilliporeSigma, A6419) for 30 minutes at room temperature in assay buffer according to the manufacturer’s protocol. As a positive control for ATP hydrolysis, 1,800 nM of myosin (MilliporeSigma, M0531) was incubated with or without 4 mM ATP. Malachite green reagent was added and developed for 30 minutes at room temperature. Plates were read with a Synergy2 microplate reader and analyzed with Gen5 v.1.10 software. ## Statistics. Data were analyzed using GraphPad Prism Version 8.0 statistical software. Statistical analysis between each experimental group was performed by 1-way ANOVA, 2-way ANOVA, or log-rank (Mantel-Cox) test, with Tukey’s multiple comparisons or uncorrected Fisher’s LSD as stated in the figure legends, with $P \leq 0.05$ considered significant. Error bars represent the SD of each data set. All experiments were performed independently at least 4 separate times with data obtained in a minimum of triplicate wells in each in vitro experiment. ## Study approval. All animal studies were approved by the University of Alabama at Birmingham Institutional Animal Use and Care Committee and performed per the University of Alabama at Birmingham Institutional Animal Use and Care Committee mouse guidelines and the NIH’s Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011). ## Author contributions SIB, JPT, ARB, AMG, SQ, TJG, and YGC designed the research studies, conducted experiments, acquired data, analyzed data, and wrote the manuscript. QS, OS, and ARH conducted experiments. JMB conducted experiments, acquired data, and analyzed data. HMT designed the research studies, analyzed data, and wrote the manuscript. HMT is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. ## 12/13/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Bluestone JA. Genetics, pathogenesis and clinical interventions in type 1 diabetes. Nature (2010) 464 1293-1300. DOI: 10.1038/nature08933 2. Blanter M. Genetic and environmental interaction in type 1 diabetes: a relationship between genetic risk alleles and molecular traits of enterovirus infection?. Curr Diab Rep (2019) 19. DOI: 10.1007/s11892-019-1192-8 3. Redondo MJ. Genetics of type 1 diabetes. Pediatr Diabetes (2018) 19 346-353. DOI: 10.1111/pedi.12597 4. Rewers M, Ludvigsson J. Environmental risk factors for type 1 diabetes. Lancet (2016) 387 2340-2348. DOI: 10.1016/S0140-6736(16)30507-4 5. Wallberg M, Cooke A. Immune mechanisms in type 1 diabetes. Trends Immunol (2013) 34 583-591. DOI: 10.1016/j.it.2013.08.005 6. Redondo MJ. Concordance for islet autoimmunity among monozygotic twins. N Engl J Med (2008) 359 2849-2850. DOI: 10.1056/NEJMc0805398 7. Redondo MJ. Genetic determination of islet cell autoimmunity in monozygotic twin, dizygotic twin, and non-twin siblings of patients with type 1 diabetes: prospective twin study. BMJ (1999) 318 698-702. DOI: 10.1136/bmj.318.7185.698 8. Elshebani A. Effects on isolated human pancreatic islet cells after infection with strains of enterovirus isolated at clinical presentation of type 1 diabetes. Virus Res (2007) 124 193-203. DOI: 10.1016/j.virusres.2006.11.004 9. Sioofy-Khojine AB. Molecular epidemiology of enteroviruses in young children at increased risk of type 1 diabetes. PLoS One (2018) 13. DOI: 10.1371/journal.pone.0201959 10. Andreoletti L. Detection of coxsackie B virus RNA sequences in whole blood samples from adult patients at the onset of type I diabetes mellitus. J Med Virol (1997) 52 121-127. DOI: 10.1002/(SICI)1096-9071(199706)52:2<121::AID-JMV1>3.0.CO;2-5 11. Busse N. Detection and localization of viral infection in the pancreas of patients with type 1 diabetes using short fluorescently-labelled oligonucleotide probes. Oncotarget (2017) 8 12620-12636. DOI: 10.18632/oncotarget.14896 12. Serreze DV. Diabetes acceleration or prevention by a coxsackievirus B4 infection: critical requirements for both interleukin-4 and gamma interferon. J Virol (2005) 79 1045-1052. DOI: 10.1128/JVI.79.2.1045-1052.2005 13. Lincez PJ. Reduced expression of the MDA5 Gene IFIH1 prevents autoimmune diabetes. Diabetes (2015) 64 2184-2193. DOI: 10.2337/db14-1223 14. Wang JP. MDA5 and MAVS mediate type I interferon responses to coxsackie B virus. J Virol (2010) 84 254-260. DOI: 10.1128/JVI.00631-09 15. Feng Q. MDA5 detects the double-stranded RNA replicative form in picornavirus-infected cells. Cell Rep (2012) 2 1187-1196. DOI: 10.1016/j.celrep.2012.10.005 16. Gitlin L. Essential role of mda-5 in type I IFN responses to polyriboinosinic:polyribocytidylic acid and encephalomyocarditis picornavirus. Proc Natl Acad Sci U S A (2006) 103 8459-8464. DOI: 10.1073/pnas.0603082103 17. Karimi Y. IFN-β signalling regulates RAW 264.7 macrophage activation, cytokine production, and killing activity. Innate Immun (2020) 26 172-182. DOI: 10.1177/1753425919878839 18. Ali S. Sources of type I interferons in infectious immunity: plasmacytoid dendritic cells not always in the driver’s seat. Front Immunol (2019) 10. DOI: 10.3389/fimmu.2019.00778 19. Longhi MP. Dendritic cells require a systemic type I interferon response to mature and induce CD4+ Th1 immunity with poly IC as adjuvant. J Exp Med (2009) 206 1589-1602. DOI: 10.1084/jem.20090247 20. Kolumam GA. Type I interferons act directly on CD8 T cells to allow clonal expansion and memory formation in response to viral infection. J Exp Med (2005) 202 637-650. DOI: 10.1084/jem.20050821 21. Ferreira RC. A type I interferon transcriptional signature precedes autoimmunity in children genetically at risk for type 1 diabetes. Diabetes (2014) 63 2538-2550. DOI: 10.2337/db13-1777 22. Kallionpaa H. Innate immune activity is detected prior to seroconversion in children with HLA-conferred type 1 diabetes susceptibility. Diabetes (2014) 63 2402-2414. DOI: 10.2337/db13-1775 23. Stewart TA. Induction of type I diabetes by interferon-alpha in transgenic mice. Science (1993) 260 1942-1946. DOI: 10.1126/science.8100367 24. Carrero JA. Type I and II interferon receptors differentially regulate type 1 diabetes susceptibility in male versus female NOD mice. Diabetes (2018) 67 1830-1835. DOI: 10.2337/db18-0331 25. Concannon P. Genetics of type 1A diabetes. N Engl J Med (2009) 360 1646-1654. DOI: 10.1056/NEJMra0808284 26. Todd JA. Etiology of type 1 diabetes. Immunity (2010) 32 457-467. DOI: 10.1016/j.immuni.2010.04.001 27. Gorman JA. The A946T variant of the RNA sensor IFIH1 mediates an interferon program that limits viral infection but increases the risk for autoimmunity. Nat Immunol (2017) 18 744-752. DOI: 10.1038/ni.3766 28. Schulte BM. Enterovirus exposure uniquely discriminates type 1 diabetes patients with a homozygous from a heterozygous melanoma differentiation-associated protein 5/interferon induced with helicase C domain 1 A946T genotype. Viral Immunol (2016) 29 389-397. DOI: 10.1089/vim.2015.0140 29. Domsgen E. et al. An IFIH1 gene polymorphism associated with risk for autoimmunity regulates canonical antiviral defence pathways in Coxsackievirus infected human pancreatic islets. Sci Rep (2016) 6. DOI: 10.1038/srep39378 30. Blum SI, Tse HM. Innate viral sensor MDA5 and coxsackievirus interplay in type 1 diabetes development. Microorganisms (2020) 8. DOI: 10.3390/microorganisms8070993 31. Peisley A. Cooperative assembly and dynamic disassembly of MDA5 filaments for viral dsRNA recognition. Proc Natl Acad Sci U S A (2011) 108 21010-21015. DOI: 10.1073/pnas.1113651108 32. Downes K. Reduced expression of IFIH1 is protective for type 1 diabetes. PLoS One (2010) 5. DOI: 10.1371/journal.pone.0012646 33. Mena I. Coxsackievirus infection of the pancreas: evaluation of receptor expression, pathogenesis, and immunopathology. Virology (2000) 271 276-288. DOI: 10.1006/viro.2000.0332 34. Kallewaard NL. Tissue-specific deletion of the coxsackievirus and adenovirus receptor protects mice from virus-induced pancreatitis and myocarditis. Cell Host Microbe (2009) 6 91-98. DOI: 10.1016/j.chom.2009.05.018 35. Bernard H. Coxsackievirus B type 4 infection in β cells downregulates the chaperone prefoldin URI to induce a MODY4-like diabetes via Pdx1 silencing. Cell Rep Med (2020) 1. DOI: 10.1016/j.xcrm.2020.100125 36. Sinha P. Cross-talk between myeloid-derived suppressor cells and macrophages subverts tumor immunity toward a type 2 response. J Immunol (2007) 179 977-983. DOI: 10.4049/jimmunol.179.2.977 37. Li H. Myeloid-derived suppressor cells suppress CD4. Mol Med Rep (2015) 12 3667-3673. DOI: 10.3892/mmr.2015.3791 38. Veglia F. Myeloid-derived suppressor cells in the era of increasing myeloid cell diversity. Nat Rev Immunol (2021) 21 485-498. DOI: 10.1038/s41577-020-00490-y 39. Mi Y. The emerging role of myeloid-derived suppressor cells in the glioma immune suppressive microenvironment. Front Immunol (2020) 11. DOI: 10.3389/fimmu.2020.00737 40. Corzo CA. Mechanism regulating reactive oxygen species in tumor-induced myeloid-derived suppressor cells. J Immunol (2009) 182 5693-5701. DOI: 10.4049/jimmunol.0900092 41. Yin B. Myeloid-derived suppressor cells prevent type 1 diabetes in murine models. J Immunol (2010) 185 5828-5834. DOI: 10.4049/jimmunol.0903636 42. Whitfield-Larry F. Myeloid-derived suppressor cells are increased in frequency but not maximally suppressive in peripheral blood of type 1 diabetes mellitus patients. Clin Immunol (2014) 153 156-164. DOI: 10.1016/j.clim.2014.04.006 43. Alicea-Torres K. Immune suppressive activity of myeloid-derived suppressor cells in cancer requires inactivation of the type I interferon pathway. Nat Commun (2021) 12. DOI: 10.1038/s41467-021-22033-2 44. Xiao W. IFNAR1 controls autocrine type I IFN regulation of PD-L1 expression in myeloid-derived suppressor cells. J Immunol (2018) 201 264-277. DOI: 10.4049/jimmunol.1800129 45. Metzger P. Correction to: immunostimulatory RNA leads to functional reprogramming of myeloid-derived suppressor cells in pancreatic cancer. J Immunother Cancer (2019) 7. DOI: 10.1186/s40425-019-0830-7 46. Duewell P. Targeted activation of melanoma differentiation-associated protein 5 (MDA5) for immunotherapy of pancreatic carcinoma. Oncoimmunology (2015) 4. DOI: 10.1080/2162402X.2015.1029698 47. Sato M. Positive feedback regulation of type I IFN genes by the IFN-inducible transcription factor IRF-7. FEBS Lett (1998) 441 106-110. DOI: 10.1016/S0014-5793(98)01514-2 48. Lukhele S. Type I interferon signaling, regulation and gene stimulation in chronic virus infection. Semin Immunol (2019) 43. DOI: 10.1016/j.smim.2019.05.001 49. Dempoya J. Double-stranded RNA induces biphasic STAT1 phosphorylation by both type I interferon (IFN)-dependent and type I IFN-independent pathways. J Virol (2012) 86 12760-12769. DOI: 10.1128/JVI.01881-12 50. Berke IC, Modis Y. MDA5 cooperatively forms dimers and ATP-sensitive filaments upon binding double-stranded RNA. EMBO J (2012) 31 1714-1726. DOI: 10.1038/emboj.2012.19 51. Yu Q. MDA5 disease variant M854K prevents ATP-dependent structural discrimination of viral and cellular RNA. Nat Commun (2021) 12. DOI: 10.1038/s41467-021-27062-5 52. Hyoty H. Mumps infections in the etiology of type 1 (insulin-dependent) diabetes. Diabetes Res (1988) 9 111-116. PMID: 3243043 53. Pak CY. Association of cytomegalovirus infection with autoimmune type 1 diabetes. Lancet (1988) 2 1-4. DOI: 10.1016/s0140-6736(88)92941-8 54. Forrest JM. High frequency of diabetes mellitus in young adults with congenital rubella. Lancet (1971) 2 332-334. DOI: 10.1016/s0140-6736(71)90057-2 55. Korkmaz HA, Ermis C. A case of immune-mediated type 1 diabetes mellitus due to congenital rubella ınfection. Ann Pediatr Endocrinol Metab (2019) 24 68-70. DOI: 10.6065/apem.2019.24.1.68 56. Oikarinen S. Virus antibody survey in different European populations indicates risk association between coxsackievirus B1 and type 1 diabetes. Diabetes (2014) 63 655-662. DOI: 10.2337/db13-0620 57. Laitinen OH. Coxsackievirus B1 is associated with induction of β-cell autoimmunity that portends type 1 diabetes. Diabetes (2014) 63 446-455. DOI: 10.2337/db13-0619 58. Burg AR. Superoxide production by NADPH oxidase intensifies macrophage antiviral responses during diabetogenic coxsackievirus infection. J Immunol (2018) 200 61-70. DOI: 10.4049/jimmunol.1700478 59. Horwitz MS. Coxsackieviral-mediated diabetes: induction requires antigen-presenting cells and is accompanied by phagocytosis of beta cells. Clin Immunol (2004) 110 134-144. DOI: 10.1016/j.clim.2003.09.014 60. Horwitz MS. Presented antigen from damaged pancreatic beta cells activates autoreactive T cells in virus-mediated autoimmune diabetes. J Clin Invest (2002) 109 79-87. DOI: 10.1172/JCI0211198 61. Chen YG. Gene targeting in NOD mouse embryos using zinc-finger nucleases. Diabetes (2014) 63 68-74. DOI: 10.2337/db13-0192 62. Shigemoto T. Identification of loss of function mutations in human genes encoding RIG-I and MDA5: implications for resistance to type I diabetes. J Biol Chem (2009) 284 13348-13354. DOI: 10.1074/jbc.M809449200 63. Chistiakov DA. Loss-of-function mutations E6 27X and I923V of IFIH1 are associated with lower poly(I:C)-induced interferon-β production in peripheral blood mononuclear cells of type 1 diabetes patients. Hum Immunol (2010) 71 1128-1134. DOI: 10.1016/j.humimm.2010.08.005 64. Chittezhath M. Islet macrophages are associated with islet vascular remodeling and compensatory hyperinsulinemia during diabetes. Am J Physiol Endocrinol Metab (2019) 317 E1108-E1120. DOI: 10.1152/ajpendo.00248.2019 65. Carrero JA. Resident macrophages of pancreatic islets have a seminal role in the initiation of autoimmune diabetes of NOD mice. Proc Natl Acad Sci U S A (2017) 114 E10418-E10427. DOI: 10.1073/pnas.1713543114 66. Roep BO. The role of T-cells in the pathogenesis of type 1 diabetes: from cause to cure. Diabetologia (2003) 46 305-321. DOI: 10.1007/s00125-003-1089-5 67. Srivastava S. Type I interferons directly inhibit regulatory T cells to allow optimal antiviral T cell responses during acute LCMV infection. J Exp Med (2014) 211 961-974. DOI: 10.1084/jem.20131556 68. Lee SE. Type I interferons maintain Foxp3 expression and T-regulatory cell functions under inflammatory conditions in mice. Gastroenterology (2012) 143 145-154. DOI: 10.1053/j.gastro.2012.03.042 69. Enevold C. Multiple sclerosis and polymorphisms of innate pattern recognition receptors TLR1-10, NOD1-2, DDX58, and IFIH1. J Neuroimmunol (2009) 212 125-131. DOI: 10.1016/j.jneuroim.2009.04.008 70. Martinez A. IFIH1-GCA-KCNH7 locus: influence on multiple sclerosis risk. Eur J Hum Genet (2008) 16 861-864. DOI: 10.1038/ejhg.2008.16 71. Zhang J. Autoimmune disease associated IFIH1 single nucleotide polymorphism related with IL-18 serum levels in Chinese systemic lupus erythematosus patients. Sci Rep (2018) 8. DOI: 10.1038/s41598-018-27782-7 72. Robinson T. Autoimmune disease risk variant of IFIH1 is associated with increased sensitivity to IFN-α and serologic autoimmunity in lupus patients. J Immunol (2011) 187 1298-1303. DOI: 10.4049/jimmunol.1100857 73. Martinez A. Association of the IFIH1-GCA-KCNH7 chromosomal region with rheumatoid arthritis. Ann Rheum Dis (2008) 67 137-138. DOI: 10.1136/ard.2007.073213 74. Cen H. Association of IFIH1 rs1990760 polymorphism with susceptibility to autoimmune diseases: a meta-analysis. Autoimmunity (2013) 46 455-462. DOI: 10.3109/08916934.2013.796937 75. Van Eyck L. Brief report: IFIH1 mutation causes systemic lupus erythematosus with selective IgA deficiency. Arthritis Rheumatol (2015) 67 1592-1597. DOI: 10.1002/art.39110 76. Oliveira L. Dysregulation of antiviral helicase pathways in systemic lupus erythematosus. Front Genet (2014) 5. PMID: 25505487 77. Weber F. Double-stranded RNA is produced by positive-strand RNA viruses and DNA viruses but not in detectable amounts by negative-strand RNA viruses. J Virol (2006) 80 5059-5064. DOI: 10.1128/JVI.80.10.5059-5064.2006 78. Dhir A. Mitochondrial double-stranded RNA triggers antiviral signalling in humans. Nature (2018) 560 238-242. DOI: 10.1038/s41586-018-0363-0 79. Levet S. Human endogenous retroviruses and type 1 diabetes. Curr Diab Rep (2019) 19. DOI: 10.1007/s11892-019-1256-9 80. Louber J. Kinetic discrimination of self/non-self RNA by the ATPase activity of RIG-I and MDA5. BMC Biol (2015) 13. DOI: 10.1186/s12915-015-0166-9 81. Horwitz MS. Protection from lethal coxsackievirus-induced pancreatitis by expression of gamma interferon. J Virol (1999) 73 1756-1766. DOI: 10.1128/JVI.73.3.1756-1766.1999 82. Tse HM. NADPH oxidase deficiency regulates Th lineage commitment and modulates autoimmunity. J Immunol (2010) 185 5247-5258. DOI: 10.4049/jimmunol.1001472 83. Kozlovskaya V. Ultrathin polymeric coatings based on hydrogen-bonded polyphenol for protection of pancreatic islet cells. Adv Funct Mater (2012) 22 3389-3398. DOI: 10.1002/adfm.201200138 84. Barra JM. Localized immunosuppression with tannic acid encapsulation delays islet allograft and autoimmune-mediated rejection. Diabetes (2020) 69 1948-1960. DOI: 10.2337/db20-0248 85. Tse HM. Activation of the mitogen-activated protein kinase signaling pathway is instrumental in determining the ability of. J Immunol (2002) 168 825-833. DOI: 10.4049/jimmunol.168.2.825 86. Padgett LE. Loss of NOX-derived superoxide exacerbates diabetogenic CD4 T-cell effector responses in type 1 diabetes. Diabetes (2015) 64 4171-4183. DOI: 10.2337/db15-0546 87. Wu B. Structural basis for dsRNA recognition, filament formation, and antiviral signal activation by MDA5. Cell (2013) 152 276-289. DOI: 10.1016/j.cell.2012.11.048 88. Yu Q. Cryo-EM structures of MDA5-dsRNA filaments at different stages of ATP hydrolysis. Mol Cell (2018) 72 999-1012. DOI: 10.1016/j.molcel.2018.10.012
--- title: Medium-chain fatty acids suppress lipotoxicity-induced hepatic fibrosis via the immunomodulating receptor GPR84 authors: - Ryuji Ohue-Kitano - Hazuki Nonaka - Akari Nishida - Yuki Masujima - Daisuke Takahashi - Takako Ikeda - Akiharu Uwamizu - Miyako Tanaka - Motoyuki Kohjima - Miki Igarashi - Hironori Katoh - Tomohiro Tanaka - Asuka Inoue - Takayoshi Suganami - Koji Hase - Yoshihiro Ogawa - Junken Aoki - Ikuo Kimura journal: JCI Insight year: 2023 pmcid: PMC9977302 doi: 10.1172/jci.insight.165469 license: CC BY 4.0 --- # Medium-chain fatty acids suppress lipotoxicity-induced hepatic fibrosis via the immunomodulating receptor GPR84 ## Abstract Medium-chain triglycerides (MCTs), which consist of medium-chain fatty acids (MCFAs), are unique forms of dietary fat with various health benefits. G protein–coupled 84 (GPR84) acts as a receptor for MCFAs (especially C10:0 and C12:0); however, GPR84 is still considered an orphan receptor, and the nutritional signaling of endogenous and dietary MCFAs via GPR84 remains unclear. Here, we showed that endogenous MCFA-mediated GPR84 signaling protected hepatic functions from diet-induced lipotoxicity. Under high-fat diet (HFD) conditions, GPR84-deficient mice exhibited nonalcoholic steatohepatitis (NASH) and the progression of hepatic fibrosis but not steatosis. With markedly increased hepatic MCFA levels under HFD, GPR84 suppressed lipotoxicity-induced macrophage overactivation. Thus, GPR84 is an immunomodulating receptor that suppresses excessive dietary fat intake–induced toxicity by sensing increases in MCFAs. Additionally, administering MCTs, MCFAs (C10:0 or C12:0, but not C8:0), or GPR84 agonists effectively improved NASH in mouse models. Therefore, exogenous GPR84 stimulation is a potential strategy for treating NASH. ## Introduction Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease worldwide (1–6). NAFLD includes a spectrum of well-defined stages, encompassing simple fatty liver (NAFL), which is a mostly benign condition, and nonalcoholic steatohepatitis (NASH). NASH progresses to cirrhosis and hepatocellular carcinoma (HCC) by activating inflammatory cascades and fibrogenesis [2, 3]. The major risk factors of NASH include metabolic disorders such as obesity, insulin resistance, glucose intolerance or type 2 diabetes, and dyslipidemia [4, 5]. Although the prevalence of NASH is rising in parallel with the global obesity pandemic, effective therapeutic strategies against the former are still in development [1, 6]. Patients have to undergo liver transplantation to prevent the progression of NASH. The crucial event involved in NAFLD progression is hepatic lipotoxicity resulting from an excessive free fatty acid (FFA) influx from the peripheral tissues, mainly the adipose tissue, to hepatocytes or from increased hepatic de novo lipogenesis (1–5). Hepatic lipotoxicity occurs when the capacity of hepatocytes to manage and export FFAs as triglycerides (TGs) is overwhelmed. FFAs act as energy sources and affect physiological functions such as hormone secretion, immune responses, and neurotransmission via the FFA-specific receptors FFAR1, FFAR4 (for long-chain fatty acids), FFAR2, and FFAR3 (for short-chain fatty acids) (7–12). Medium-chain fatty acids (MCFAs) also have a specific receptor — G protein–coupled receptor 84 (GPR84) [7, 13, 14]. However, GPR84 is still considered an orphan G protein–coupled receptor (GPCR) because of the low plasma levels of endogenous MCFAs [15]. Medium-chain triglycerides (MCTs), which consist of MCFAs, are unique forms of dietary fat exhibiting various health benefits [7, 16, 17]. MCTs are an appropriate dietary choice for individuals with high energy demands. In the elderly, MCTs counteract age-related decreased energy production, and in athletes, MCTs enhance performance. MCTs are also beneficial for individuals who have undergone major surgeries or experience stunted growth (18–20). GPR84 is coupled with the pertussis toxin–sensitive Gi/o protein and is predominantly expressed in the BM, lungs, and peripheral leukocytes [13, 14, 21]. Although some studies on GPR84-deficient mice have demonstrated that GPR84 plays an important role in immune and metabolic responses and may mediate the crosstalk between immune cells and adipocytes (22–25), comprehensive and integrated data bridging the gap between endogenous MCFAs and GPR84 are lacking, and the molecular mechanisms underlying these processes remain unclear. Here, we investigated the effects of molecular nutritional signaling by MCFAs on metabolic functions using GPR84-deficient mice, a model of high-fat diet–induced (HFD-induced) obesity, and a NASH mouse model. ## GPR84 deficiency accelerates chronic inflammation under HFD conditions. To study the role of GPR84 in the metabolic and immune systems, we generated Gpr84–/– mice (Supplemental Figure 1, A–C; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.165469DS1). HFD feeding in WT mice increases the levels of inflammatory cytokines, such as TNF-α, and long-term HFD exposure leads to chronic inflammation [26]. Therefore, we first compared the levels of plasma TNF-α as an inflammatory marker under short-term HFD feeding between WT and Gpr84–/– mice. Although plasma TNF-α levels were comparable between WT and Gpr84–/– mice under normal chow (NC) feeding, HFD feeding markedly increased plasma TNF-α levels in Gpr84–/– mice more than in WT mice (Figure 1A; $46.24\%$ increase). Moreover, the hepatic expression of the Tnf mRNA in Gpr84–/– mice was markedly higher than that in WT mice, whereas its expression in other tissues, such as the white adipose tissue (WAT) (in both mature adipocytes and stromal vascular fraction), muscle, small intestine, and colon, was comparable between WT and Gpr84–/– mice (Figure 1B and Supplemental Figure 2). RNA-Seq and a Gene Ontology (GO) enrichment analysis of liver from the HFD-fed Gpr84–/– mice revealed a relationship between the chemokine pathway and chronic inflammation (Supplemental Figure 3, A–C). Among the differentially expressed genes (DEGs), the expression of 59 inflammation-related genes was altered compared with that of the WT mice (Figure 1C). In particular, the hepatic mRNA expression of the fibrosis markers Col1a, Tgfb1, and Acta2 was considerably higher in Gpr84–/– mice than in WT mice (Figure 1D; Col1a: 4.03-fold increase, Tgfb1: 2.14-fold increase, and Acta2: 1.30-fold increase). The hepatic TG levels and mRNA expression of these fibrosis marker genes in the WAT were comparable between the groups (Supplemental Figure 4, A and B). Thus, GPR84-deficient mice exhibited chronic hepatic inflammation and fibrosis without the acceleration of hepatic fat accumulation, even under short-term HFD feeding. ## Long-term HFD-fed GPR84-deficient mice exhibit NASH. To determine how GPR84 deficiency affects the liver, we induced chronic inflammation and hepatic steatosis through long-term (12 weeks) feeding of an HFD to WT and Gpr84–/– mice. Liver weight was markedly lower in Gpr84–/– mice than in WT mice (Figure 2A), and the hepatic TG levels were comparable between them (Figure 2B). The hepatic levels of the inflammatory marker Tnf and the fibrosis markers Col1a, Tgfb1, and Acta2 were markedly elevated in Gpr84–/– mice compared with those in WT mice (Figure 2C; Tnf: 2.66-fold increase, Col1a: 8.46-fold increase, Tgfb1: 3.76-fold increase, and Acta2: 3.51-fold increase), whereas their levels in WAT were comparable (Supplemental Figure 4, C and D). Furthermore, HFD-fed Gpr84–/– mice showed increased numbers of F$\frac{4}{80}$-positive macrophages, levels of the fibrosis marker α–smooth muscle actin (α-SMA) (Figure 2D), and the macrophage marker genes Adgre1, Cd68, and Cd14 in the liver compared with HFD-fed WT mice (Figure 2E; Adgre1: 7.65-fold increase, Cd68: 6.87-fold increase, and Cd14: 6.75-fold increase). Consequently, the NAFLD activity score (NAS) for the livers of HFD-fed Gpr84–/– mice was higher than that for the livers of HFD-fed WT mice (Figure 2F). Thus, GPR84 deficiency accelerates the progression from HFD-induced hepatic steatosis to NASH. ## HFD feeding increases endogenous MCFA levels as GPR84 ligands in liver. GPR84 has been identified as a receptor for MCFAs and is coupled with the Gi/o protein, which decreases the intracellular cAMP concentration [14]. In HEK293 cells expressing mouse GPR84 (Supplemental Figure 5A), C9:0, C10:0, C11:0, C12:0, and C13:0 activated GPR84 in a dose-dependent manner, whereas such activation was not displayed by C6:0, C7:0, C8:0, and C14:0 or not observed in doxycycline-uninduced controls (Dox-uninduced controls; non-GPR84–expressing HEK293 cells) (Figure 3A and Supplemental Figure 5B). C10:0 was found to be the most potent agonist of GPR84, with an EC50 of 3.5 μM, and C12:0 was the second-most potent agonist, with an EC50 of 4.4 μM (Figure 3A). Next, we investigated the levels of endogenous MCFAs as GPR84 ligands after HFD feeding. As for the profiles of FFAs (C6:0–C14:0) including MCFAs, their levels were found to be elevated in the plasma and liver of HFD-fed mice compared with those in NC-fed mice (Figure 3B and Supplemental Figure 6A). The hepatic levels of C10:0 and C12:0 were markedly elevated in HFD-fed mice compared with those in NC-fed mice (Figure 3C). This increase in MCFA levels sufficiently activated GPR84 (Figure 3A). MCFAs were hardly detected in cecal contents under HFD conditions (Supplemental Figure 6, B and C). Comparison of the RNA-Seq data of the liver from NC- and HFD-fed mice showed that the 6 fatty acid synthesis and β-oxidation genes were coded as MCFA synthesis–related enzymes in 34 fatty acid synthesis and metabolism-related genes of DEGs (Supplemental Figure 6D). That is, acyl-CoA synthetase long-chain family member 1 (Acsl1) and acyl-CoA synthetase medium-chain family member 3 (Acsm3) code medium-chain acyl-CoA synthetase. Acyl-CoA dehydrogenase, long chain (Acadl), and acyl-CoA dehydrogenase, medium chain (Acadm), code medium-chain acyl-CoA dehydrogenase. Acyl-CoA thioesterase 11 (Acot11) and acyl-CoA thioesterase 13 (Acot13) code medium-chain acyl-CoA thioesterase. The hepatic mRNA expression levels of these MCFA synthesis-related enzymes were considerably higher in HFD-fed mice than in NC-fed mice (Figure 3D). Thus, HFD feeding increases the levels of endogenous MCFAs, which are GPR84 ligands, and accelerates fatty acid synthesis and β-oxidation in the liver. ## GPR84 suppresses BM-derived hepatic macrophages. We next investigated the molecular mechanisms underlying the protective activity of GPR84 against the progression of HFD-induced hepatic steatosis to fibrosis. The HFD increased not only hepatic endogenous MCFA production but also hepatic Gpr84 mRNA expression (Figure 4A). Hepatic Gpr84 was expressed in macrophages but not in hepatocytes, monocytes, stellate, or Kupffer cells (Figure 4B), and HFD feeding further accelerated its expression (Figure 4C). The population of macrophages in the livers of HFD-fed Gpr84–/– mice was higher than that in HFD-fed WT mice (Figure 4D). In contrast, the population of macrophages in the livers of NC-fed Gpr84–/– mice was comparable to that of macrophages in the livers of WT mice (Supplemental Figure 7A). Additionally, the population of Kupffer cells in the livers of both NC- and HFD-fed Gpr84–/– mice were also comparable to that of Kupffer cells in the livers of WT mice (Supplemental Figure 7, A and B). Moreover, Tnf mRNA expression was markedly higher in the hepatic macrophages of HFD-fed (versus NC-fed) Gpr84–/– mice than in HFD-fed WT mice (Figure 4D and Supplemental Figure 7C). Gpr84 was mainly expressed in the BM, which is the primary site of hematopoiesis (Supplemental Figure 7D). Hence, we further investigated the GPR84-mediated relationship between BM and hepatic macrophages. RNA-Seq and Gene Ontology (GO) enrichment analysis of the BM from HFD-fed Gpr84–/– mice showed that its expression profile was related to the macrophage-related chemokine pathway and chronic inflammation (Supplemental Figure 8, A–C). Additionally, the transplantation of Gpr84–/– mouse-derived BM into WT mice caused macrophage infiltration into the liver and NASH under HFD feeding as well as the hepatic phenotype of Gpr84–/– mice (Figure 4E). Thus, GPR84-positive BM-derived macrophages may prevent hepatic fibrosis. The mechanisms underlying this process in the liver were investigated under HFD feeding conditions using GPR84-deficient RAW264.7 macrophages. Saturated fatty acids, such as palmitic acid (C16:0), which are abundant in HFD, induce inflammation by activating macrophages [7, 27]. C16:0 stimulation upregulated the expression of the inflammatory marker Tnf and the macrophage infiltration marker CC chemokine ligand 2 (Ccl2) in RAW264.7 cells (Figure 4F and Supplemental Figure 9). C10:0 suppressed these effects and increased the expression of the antiinflammatory M2 macrophage marker arginase 1 (Arg-1) in a dose-dependent manner; while the effects of C10:0 were diminished in Gpr84–/– RAW264.7 cells (Figure 4F and Supplemental Figure 9). Furthermore, C16:0 administration increased the levels of intracellular MCFAs in the mouse hepatocyte cell line AML12 (Figure 4G). C16:0 stimulation in Gpr84–/– RAW264.7 cells cocultured with AML12 showed a marked increase in Tnf expression compared with that in RAW264.7 cells cocultured with AML12 (Figure 4H). Thus, MCFAs suppress lipotoxicity-induced macrophage activation via GPR84 in the liver. ## GPR84 activation by MCFAs improves NASH. Finally, we investigated whether GPR84 activation could suppress NASH progression in a NASH mouse model. A choline-deficient l-amino acid–defined HFD (CDAHFD) and CCl4 were used to establish NASH with rapidly progressive hepatic fibrosis in mice [28]. WT mice fed with the CDAHFD for 10 weeks exhibited signs of NASH and HCC (Figure 5A). Supplementation of dietary MCFAs (C8:0, C10:0, and C12:0) in CDAHFD-fed mice increased the plasma and hepatic levels of each MCFA (Supplemental Figure 10A). Interestingly, unlike in HFD-fed mice (Figure 3C), basal endogenous MCFA levels were comparable among NC-fed, CDAHFD-fed, and CCl4-administered mice (Supplemental Figure 10B). Although MCFA supplementation did not significantly change the liver and WAT weights, C10:0 and C12:0 supplementation in CDAHFD-fed WT mice effectively suppressed the signs of NASH and HCC (Figure 5, A and B). The hepatic TG levels were comparable between CDAHFD-fed WT and Gpr84–/– mice supplemented with dietary MCFAs (Figure 5C). The levels of the inflammatory marker Tnf, fibrosis markers Col1a, Tgfb1, and Acta2, and macrophage marker Adgre1 were also markedly decreased by C10:0 and C12:0, but not C8:0, supplementation in the livers of CDAHFD-fed WT mice. The effects of C10:0 were abolished in Gpr84–/– mice (Figure 5D). Consequently, the NAS decreased considerably after C10:0 and C12:0, but not C8:0, supplementation in WT mice, but not in Gpr84–/– mice (Figure 5E). Thus, MCFAs, except for C8:0, markedly suppressed NASH progression via GPR84. Furthermore, among the dietary MCT oils, which are sources of MCFAs, trioctanoin (TriC8) and tridecanoin (TriC10) supplementation increased the levels of MCFA C8:0 and C10:0 in the plasma and liver, respectively (Supplemental Figure 11A). Under TriC10 supplementation, but not TriC8, the levels of inflammatory, fibrosis, and macrophage markers markedly decreased without any changes in hepatic TG levels in CDAHFD-fed WT mice, but not Gpr84–/– mice (Supplemental Figure 11, B–D). The NAS markedly dropped after TriC10 supplementation (Figure 5F). Thus, GPR84 activation by dietary MCFAs (C10:0 and C12:0, but not C8:0) markedly improves NAFLD, thereby suppressing the progression of NAFL to NASH, but not to hepatic steatosis. ## GPR84 agonists are potential NASH therapeutic drugs. We confirmed that Gpr84 expression and NASH progression increased in human livers (Figure 6A). Therefore, GPR84-selective compounds may be potential therapeutic drugs. Embelin is a known GPR84 agonist [29]. In HEK293 cells expressing mouse GPR84, a tetracycline-controlled Tet-On gene expression system and TGF-α shedding assay [30] were used to confirm that embelin activated GPR84 in a dose-dependent manner (Figure 6B). Embelin, as well as C10:0, suppressed palmitate-induced increases in Tnf expression in a dose-dependent manner. The effects of embelin were abolished in Gpr84–/– RAW264.7 cells (Figure 6C). Hence, we administered GPR84-selective compounds in the NASH mouse model using embelin as the GPR84 agonist. Consequently, embelin markedly suppressed the levels of inflammatory, fibrosis, and macrophage markers, as well as the NAS, in both the CDAHFD-fed and CCl4-induced NASH mouse models (Figure 6, D–F; and Supplemental Figure 12, A and B). Thus, exogenous GPR84 stimulation markedly improved NAFLD. ## Discussion The exact contribution of endogenous MCFAs and the receptor GPR84 in controlling metabolic syndrome was previously unclear. Herein, MCFAs showed hepatoprotective activity against dietary fat–induced NASH progression. Under HFD feeding, NASH progression was observed in HFD-fed Gpr84–/– mice. In addition to saturated fatty acid excess–mediated macrophage activation under HFD feeding, macrophage-mediated phagocytosis of fat-accumulated hepatocytes further accelerated the inflammatory response. Thus, under lipotoxic conditions, endogenous MCFAs, which are released from hepatocytes along with long-chain fatty acids, suppressed the overactivation of macrophages via GPR84, thereby protecting hepatic functions. Metabolic disorders, such as obesity, insulin resistance, glucose intolerance, and type 2 diabetes, are significant risk factors of NASH [4, 5]. We recently reported that MCFA-stimulated GPR84 activation maintains glucose homeostasis by insulin regulation via glucagon-like peptide-1 (GLP-1) secretion [25]. GLP-1 also suppresses the proinflammatory and profibrotic phenotypes of macrophages, thereby suppressing NASH development [31, 32]. The regulation of GLP-1 secretion via GPR84 may thus be partly related to the suppression of NASH. Thus, GPR84 functions, including differentiation of macrophages from monocytes and filtration from BM to the liver, on other organs also may influence the NASH progression. Therefore, further studies using tissue-specific GPR84-deficient mice are needed to elucidate these metabolic mechanisms. Although it is known that MCFAs and MCTs have antiinflammatory effects and that GPR84 is coupled with inhibitory G proteins (Gi/o) [33, 34], recent in vitro studies have described GPR84 as a proinflammatory receptor (35–37). However, since these studies were conducted using only potent synthetic GPR84 agonists, the physiological activity of GPR84 remains unclear. In this study, we showed that endogenous or dietary MCFAs effectively suppress NASH progression through GPR84 as an antiinflammatory receptor, both in vivo and in vitro. Moreover, we confirmed the MCFA-GPR84–mediated antiinflammatory effects under lipotoxic conditions using blinded in vitro experiments. Previous in vitro studies have reported that GPR84 stimulation weakly promotes inflammation under normal or nonlipotoxic inflammatory conditions. In contrast, we showed that GPR84 stimulation suppresses inflammation under lipotoxicity-induced hyperinflammatory conditions. This contradiction may exemplify how FFARs, including GPR84, are optimal fine-tuning receptors for maintaining homeostasis by regulating biological processes and sensing nutritional states [7]. Therefore, we redefine GPR84 as an immunomodulating receptor, not simply a proinflammatory receptor. GPR84 antagonists weakly suppress NASH, and HFD feeding in Gpr84–/– mice weakly restores fibrosis, but not steatosis and inflammation (38–40). However, actual phase II clinical trials using the selective GPR84 antagonist GLPG1205 failed to demonstrate its efficacy [35, 41, 42]. Furthermore, another GPR84 antagonist (PBI-4547) also acts as a GPR40/GPR120 agonist [40]. Importantly, our results indicate that, although HFD feeding induced an increase in hepatic MCFA levels (as endogenous GPR84 ligands) (Figure 3C), CDAHFD feeding and CCl4 administration did not change hepatic MCFA levels compared with NC feeding (Supplemental Figure 10B). Furthermore, HFD-fed Gpr84–/– mice exhibited increased hepatic inflammation and fibrosis and progression to NASH compared with WT mice (Figure 2, C–E), whereas CDAHFD and CCl4 did not change the basal levels of inflammatory and fibrosis markers nor the NAS between WT and Gpr84–/– mice (Figures 5 and 6). Therefore, Simard et al. ’s HFD-fed mouse model [40] may not alter hepatic MCFA levels, and their methods, in which 10- to 14-week-old mice were fed an HFD for 14 weeks, might not be appropriate for establishing an HFD-induced metabolic syndrome mouse model. In comparison, our method involved mice aged 7 weeks that were fed an HFD for 5 weeks, or mice aged 4 weeks that were fed an HFD for 12 weeks. Although HFD induced obesity in our mouse model, neither our CDAHFD-fed or CCl4-administered mice nor Simard et al. ’s HFD-fed mice [40] exhibited an increase in BW. Further studies are needed to clarify the mechanism by which endogenous MCFAs are produced under HFD feeding and that of metabolic diseases and NASH progression. Nevertheless, our results indicate that exogenous GPR84 stimulation using dietary MCTs and a GPR84 agonist is effective in suppressing progression of NASH under low endogenous hepatic MCFA levels. To validate GPR84 as a therapeutic target, we suggest that GPR84 stimulation by GPR84 agonists may be a more effective strategy than developing a substitute GPR84 antagonist. In conclusion, GPR84 deficiency under excess dietary fat intake accelerates lipotoxicity-induced macrophage overactivation, thereby promoting hepatic fibrosis to NASH. In contrast, MCFA, MCT, and GPR84 agonist administration effectively improved NASH progression by suppressing hepatic fibrosis without influencing hepatic steatosis by fat accumulation. Hence, MCFAs, either endogenously synthesized or derived from dietary MCTs, may play important roles in recognizing nutrient excess and maintaining hepatic metabolic functions through GPR84 activation. Additionally, this study formally demonstrated that orphan GPCR GPR84 is a receptor for endogenous MCFAs. Collectively, GPR84 modulation may be an effective strategy for improving the progression of NASH and HCC. ## Animal study. C57BL/6J, Gpr84–/–, and congenic CD45.1 mice (Sankyo Lab Service) were housed under a 12-hour light/12-hour dark cycle and fed NC (CE-2, CLEA). Gpr84–/– mice with a C57BL/6J background were generated using the CRISPR/Cas9 system (Supplemental Figure 1, A–C). For short-term treatment, 7-week-old male mice were fed NC or an HFD with $60\%$ kcal fat (D12492, Research Diets) for 5 weeks. For long-term treatment, 4-week-old C57BL/6J male mice were fed an HFD for 12 weeks. At least 3 groups of littermates from each dam were analyzed in individual experiments. Chronic liver injury was induced by feeding the mice with CDAHFD containing $60\%$ kcal of fat and $0.1\%$ of methionine (A06071302, Research Diets) [43] or MCFA- or MCT-supplemented CDAHFD (Supplemental Table 1) for 10 weeks. MCT oils were purchased from the Nisshin OilliO Group. C57BL/6J male mice that were 7–8 weeks old were treated with CCl4 (0.6 mL/kg body weight, diluted in corn oil and injected i.p. every 3 days) for 8 weeks to induce hepatic fibrosis [44]. CDAHFD-fed or CCl4-treated mice were administered embelin (50 mg/kg body weight) through oral gavage once a day for 4 weeks. All efforts were made to minimize animal suffering. ## Human study. For the analysis of GPR84, TNF, and TGFB1 mRNA expression levels in human liver samples, a total of 53 liver samples were isolated from healthy participants, as well as patients with NAFL and NASH. ## Biochemical analyses. Hepatic levels of TGs were analyzed using commercial kits (LabAssay Triglyceride, FUJIFILM Wako). The levels of TNF-α were measured using the Mouse TNF-alpha Quantikine ELISA Kit (R&D Systems), following the manufacturer’s instructions. ## RNA isolation and quantitative reverse transcriptase PCR. Total RNA was extracted using an RNAiso Plus reagent (TAKARA). cDNA was transcribed using RNA as templates and Moloney murine leukemia virus reverse transcriptase (Invitrogen). Quantitative reverse transcriptase PCR analysis was performed using SYBR Premix Ex Taq II (TAKARA) and the StepOne real-time PCR system (Applied Biosystems) as described previously [10, 11]. The PCR protocol was as follows: 95°C for 30 seconds, followed by 40 cycles of 95°C for 5 seconds, 58°C for 30 seconds, and 72°C for 1 minute. Each sample was tested in duplicate for the average Ct value. Relative mRNA expression was calculated after normalization to the 18S rRNA reference gene using the 2-ΔΔCt method. Primer sequences for the targeted mouse genes were as follows: Gpr84, 5′-AGGTGACCCGTATGTGCTTC-3′ (forward) and 5′-GTTCATGGCTGCATAGAGCA-3′ (reverse); 18S, 5′-CTTAGAGGGACAAGTGGCG-3′ (forward) and 5′-ACGCTGAGCCAGTCAGTGTA-3′ (reverse); Col1α, 5′-CCTCAGGGTATTGCTGGACAAC-3′ (forward) and 5′-ACCACTTGATCCAGAAGGACCTT-3′ (reverse); Tgfb1, 5′-CCTGAGTGGCTGTCTTTTGACG-3′ (forward) and 5′-AGTGAGCGCTGAATCGAAAGC-3′ (reverse); Acta2, 5′-GTTCAGTGGTGCCTCTGTCA-3′ (forward) and 5′-ACTGGGACGACATGGAAAAG-3′ (reverse); Tnf, 5′-GGCAGGTCTACTTTGGAGTC-3′ (forward) and 5′-TCGAGGCTCCAGTGAATTCG-3′ (reverse); Adgre1, 5′-GATGTGGAGGATGGGAGATG-3′ (forward) and 5′-ACAGCAGGAAGGTGGCTATG-3′ (reverse); Cd68, 5′-TCCAAGATCCTCCACTGTTG-3′ (forward) and 5′-ATTTGAATTTGGGCTTGGAG-3′ (reverse); Cd14, 5′-GGCGCTCCGAGTTGTGACT-3′ (forward) and 5′-TACCTGCTTCAGCCCAGTGA-3′ (reverse); Ccl2, 5′-AATCTGAAGCTAATGCATCC-3′ (forward) and 5′-GTGTTGAATCTGGATTCACA-3′ (reverse); Arg1, 5′-AAAGCTGGTCTGCTGGAAAA-3′ (forward) and 5′-ACAGACCGTGGGTTCTTCAC-3′ (reverse); Acsl1, 5′-TGCCAGAGCTGATTGACATTC-3′ (forward) and 5′-GGCATACCAGAAGGTGGTGAG-3′ (reverse); Acsm3, 5′-CTTTGGCCCCAGCAGTAGATG-3′ (forward) and 5′-GGCTGTCACTGGCATATTTCAT-3′ (reverse); Acadl, 5′-TTTCCTCGGAGCATGACATTTT-3′ (forward) and 5′-GCCAGCTTTTTCCCAGACCT-3′ (reverse); Acadm, 5′-CCAGAGAGGAGATTATCCCCG-3′ (forward) and 5′-TACACCCATACGCCAACTCTT-3′ (reverse); Acot11, 5′-AGATCATGGCTTGGATGGAG-3′ (forward) and 5′-AAAGGCGTTATTCACGATGG-3′ (reverse); and Acot13, 5′-AGCAGCATGACCCAGAACCTA-3′ (forward) and 5′-GGAGCGTGCCCAGTTTATTAGTA-3′ (reverse). Primer sequences for the targeted human genes were as follows: GPR84, 5′-TTCAGCCCTTCTCTGTGGACA-3′ (forward) and 5′-TGCAGAAGGTGGCACCG-3′ (reverse); TNF, 5′-CACTAAGAATTCAAACTGGGGC-3′ (forward) and 5′- GAGGAAGGCCTAAGGTCCAC-3′ (reverse); TGFB1, 5′-CCCAGCATCTGCAAAGCTC-3′ (forward) and 5′-GTCAATGTACAGCTGCCGCA-3′ (reverse); and 18S, 5′-CGCCGCTAGAGGTGAAATC-3′ (forward) and 5′-CCAGTCGGCATCGTTTATGG-3′ (reverse). ## Histological analysis. The liver was excised and fixed overnight at 4°C in $4\%$ paraformaldehyde. The fixed tissues were embedded in O.C.T. compound (Sakura Finetek) and sectioned into 8 μm thick sections using a cryo-microtome (Leica). H&E staining was performed using standard techniques. The lipid contents in hepatocytes were visualized using Oil Red O staining. IHC analysis was performed using antibodies against F$\frac{4}{80}$ (1:1,000; catalog ab6640, Abcam) and α-SMA (1:300; catalog 19245, Cell Signaling Technology), and the nuclei were stained with DAPI (1:5,000; catalog 10236276001, Roche), as previously described [9]. Quantification of liver macrophage was quantified by counting F$\frac{4}{80}$ positive cells (green fluorescence), and total number of cells was counted based on the DAPI nuclear staining using BZ-X710 (Keyence). The sections were washed with PBS, blocked with $5\%$ BSA in PBS, and permeabilized with $0.1\%$ Triton X-100 (Sigma). Next, the sections were incubated with primary antibodies, followed by incubation with secondary antibodies conjugated with a fluorescent marker. Immunoreactive signals were developed using DAB staining with the Peroxidase Stain DAB Kit (Nacalai Tesque), and the sections were counterstained with Meyers hematoxylin (FUJIFILM Wako). A histopathological evaluation of NASH was performed based on the NAS and steatosis, lobular inflammation, and ballooning degeneration scores. Steatosis, lobular inflammation, and ballooning degeneration were scored on 0–3, 0–3, and 0–2 scales, respectively. Total NAS was scored as follows: 1–3, $\frac{4}{5}$, and 6–8. NAS is shown in Supplemental Table 2. ## RNA-Seq. RNA was extracted from the liver and BM of NC- and HFD-fed mice using an RNAiso Plus reagent (TAKARA) and RNeasy mini kit (QIAGEN). RNA-Seq libraries were generated with the TruSeq RNA Library Prep Kit (Illumina) and sequenced on an Illumina platform. Approximately 4 Gb paired-end reads of length 100 bp per sample were obtained. The RNA-Seq data were preprocessed using Trimmomatic to remove adapters or poor-quality reads [45]. The quality of the trimmed sequences was then assessed using FastQC [46]. The reads were aligned to the mouse reference genome (mm10) using HISAT2 [47] with the Bowtie2 aligner [48]. The aligned reads were assembled using StringTie [49]. The raw read counts were subjected to relative log expression normalization to obtain DEGs from all comparisons. The data were expressed as fold change using nbinomWaldTest with DESeq2. DEGs were identified based on the following 2 criteria: FDR-adjusted P value < 0.05 (using the Benjamini-Hochberg procedure) and \|log2 (fold change)\| > 0.5. A gene set enrichment analysis was performed using the Kyoto Encyclopedia of Genes and Genomes database (http://www.genome.jp/kegg/). The GO terms of molecular function, biological process, cellular component, and pathway were considered. ## Cell culture. All cell lines were cultured at 37°C with $5\%$ CO2. To generate Flp-In T-REx HEK293 cells (Invitrogen) expressing murine GPR84, HEK293 cells were transfected with a mixture of pcDNA5/FRT/TO-HA-mGPR84 and pOG44 using Lipofectamine reagent (Invitrogen) (Supplemental Figure 5A). The cells were cultured in DMEM supplemented with 10 μg/mL blasticidin S (Funakoshi), 100 μg/mL hygromycin B (Gibco), and $10\%$ FBS. For the localization analysis, the cells were fixed in $4\%$ paraformaldehyde in PBS for 10 minutes at room temperature then permeabilized with $0.2\%$ Triton X-100 in PBS for 30 minutes at room temperature. After washing with PBS, the cells were preincubated with $1\%$ BSA in PBS for 1 hour then probed with the primary anti-HA high-affinity antibodies (1:1,000; clone 3F10, Roche) in $1\%$ BSA/PBS for 1 hour at room temperature. The cells were washed twice with PBS, incubated with Alexa Fluor 488–conjugated secondary antibodies (1:200; catalog A11006, Invitrogen), and observed under a fluorescence microscope. For cAMP determination, GPR84-expressing HEK293 cells were seeded in 24-well plates (1 × 105 cells/well), cultured for 24 hours, and treated with or without doxycycline (10 μg/mL; Sigma) for 24 hours. The cells were treated with 2 μM forskolin (Sigma) and 500 μM of 3-isobutyl 1-methylxanthine (Sigma) to upregulate the cAMP levels. They were then stimulated with individual MCFAs (C6:0–C14:0; Nu-Chek Prep) or embelin (Cayman Chemical) for 10 minutes. The cAMP levels were determined using the cAMP ELISA kit (Cayman Chemical) following the manufacturer’s instructions. The TGF-α shedding assay was performed as described previously [50]. HEK293 cells were seeded in 6-well plates (2 × 105 cells/well) and cultured for 48 hours. Plasmid transfection was performed with a mixture of 500 ng AP-TGF-α–encoding plasmid and 200 ng GPR84-encoding plasmid with or without 100 ng Gαi3-encoding plasmid. After 1 day, the transfected cells were harvested by trypsinization, pelleted by centrifugation at 190g for 5 minutes at room temperature, and washed once with HBSS containing 5 mM HEPES (pH 7.4). After centrifugation, the cells were resuspended in the HEPES-containing HBSS. The cell suspension was seeded in a 96-well culture plate and incubated for 30 minutes at 37°C and $5\%$ CO2. The cells were treated with GPR84 ligands diluted in HBSS containing 5 mM HEPES (pH 7.4) and $0.01\%$ (w/v) BSA (fatty acid–free and protease-free grade; FUJIFILM Wako) for 1 hour. AP reaction solution (10 mM p-nitrophenyl phosphate, 120 mM Tris-HCl [pH 9.5], 40 mM NaCl, and 10 mM MgCl2) was dispensed into the cell plates. Absorbance at 405 nm of the plates was measured using a microplate reader (Multiskan GO, Thermo Fisher Scientific) before and after a 1-hour incubation period at room temperature. Ligand-induced AP-TGF-α release was calculated as described previously [50]. RAW264.7 cells (mouse macrophage cell line; ATCC) were cultured in DMEM supplemented with $1\%$ penicillin-streptomycin solution (Gibco) and $10\%$ FBS [51]. The GPR84-deficient RAW264.7 cells (RAW-KO cells) were generated using a CRISPR/Cas9-mediated homology-independent knockin system. sgRNA targeting Gpr84 (5′-ttcgtcccaagctccgaacc-3′) was designed based on a previous report [52] and cloned into the sgRNA expression vector peSpCAS9(1.1)-2xsgRNA (Addgene plasmid 80768). RAW264.7 cells were plated in 60 mm dishes (2.5 × 105 cells/dish) and cotransfected with the recombinant peSpCAS9(1.1)-2xsgRNA and pDonor-tBFPNLS-Neo (Addgene plasmid 80766) using Lipofectamine 2000 (Invitrogen). On day 2 after transfection, the cells were cultured in medium containing 250 μg/mL G418 (FUJIFILM Wako) to select the recombinant cells. At day 10 after selection, colonies grown from single cells were isolated. RAW264.7 and RAW-KO cells were stimulated with capric acid (C10:0; 0.01, 0.1, and 1 mM; Nu-Chek Prep) or embelin (0.1, 1, 10, 50, and 100 μM; Cayman Chemical) in the presence of palmitic acid (C16:0) for 12 hours. Before stimulation with these samples, the sample origin was blinded. The cells were then harvested to isolate their RNA. AML12 cells (mouse hepatocyte cell line; ATCC) were maintained in DMEM/HAM-F12 (1:1, 3.15 g/l-glucose) (Sigma) supplemented with $1\%$ penicillin-streptomycin solution, $10\%$ FBS, 0.005 mg/mL insulin, 0.005 mg/mL transferrin, and 40 ng/mL dexamethasone [53]. To measure the cellular MCFA contents, AML12 cells were treated with palmitic acid (C16:0) for 48 hours and harvested for liquid chromatography-mass spectrometry (LC-MS/MS) analysis. For coculture studies, long-chain fatty acid–stimulated AML12 cells were cocultured with RAW264.7 or RAW-KO cells for 72 hours and harvested for RNA isolation. ## MCFA determination. MCFA levels in the plasma, liver, adipose tissue, muscle, cecum, and NC and HFD samples were determined following a previously described protocol with modifications [12]. The samples containing an internal control (C19:0) were homogenized in methanol and mixed with chloroform and water to extract lipids. The samples were centrifuged at 2,000g and 17°C for 10 minutes. The supernatant containing MCFAs was collected and dried. The samples were resuspended with chloroform/methanol (1:3, v/v) and subjected to LC-MS/MS analysis using an ultra-performance LC system (UPLC, Waters) equipped with an Acquity UPLC system coupled to a Waters Xevo TQD mass spectrometer. The samples were separated on an ACQUITY UPLC BEH C18 column (2.1 × 150 mm, 1.7 μm; Waters) using a methanol gradient in 10 mM ammonium formate aqueous solution. ## Flow cytometry. To isolate hepatic mononuclear cells and Kupffer cells, the excised livers were cut into small pieces using a razor blade and subjected to enzymatic digestion in a digestion solution (3 mM CaCl2, 1 mg/mL collagenase I (FUJIFILM Wako), and $1.5\%$ BSA in HBSS) for 2 hours at 37°C. The cell suspension was passed through a 70 μm nylon mesh cell strainer (Corning). The cells were isolated using Percoll density gradient centrifugation [54]. Single-cell suspensions were blocked with an Fc receptor CD16/CD32 (clone 93, BioLegend) at 4°C for 10 minutes. For flow cytometric sorting, hepatic mononuclear cells and Kupffer cells were stained with Brilliant Violet (BV) 510–conjugated anti-CD45 (clone 30-F11, BioLegend), BV711-conjugated anti-Ly6C (clone HK1.4, BioLegend), Alexa Fluor 488–conjugated anti-F$\frac{4}{80}$ (clone BM8, BioLegend), PE-conjugated anti-CX3CR1 (clone SA011F11, BioLegend), and APC-conjugated anti-CD11b (clone M$\frac{1}{70}$, BioLegend) antibodies for 30 minutes at 4°C. The cells were then washed with FACS buffer (1× PBS containing $2\%$ FBS and 2 mM EDTA). For the transplantation studies, hepatic mononuclear cells were obtained using collagenase digestion and Percoll density gradient centrifugation. The samples were stained with PE-Cy7–conjugated anti-CD45.1 (clone A20, BD Biosciences), APC-Cy7–conjugated anti-CD45.2 (clone 104, BD Biosciences), PE-conjugated anti-Ly6C (clone HK1.4, BioLegend), FITC-conjugated anti-F$\frac{4}{80}$ (clone BM8, BioLegend), and APC-conjugated anti-CD11b (clone M$\frac{1}{70}$, BD Biosciences) antibodies. The cells were sorted using a FACSAria III cell sorter (BD Biosciences) and FACSMelody (BD Biosciences). The purity of the sorted cells was at least $95\%$. Flow cytometric data were analyzed using FlowJo v10 software (BD Biosciences). ## BM cell transplantation. C57BL/6J CD45.1 mice were lethally irradiated with a dose of 10 Gy. A total of 1 × 107 cells obtained from C57BL/6J or Gpr84–/– (CD45.2) mice were intravenously injected into the irradiated recipient mice. The mice were bred with water supplemented with 1 g/L neomycin and 1 g/L ampicillin for 2 weeks after transplantation. Mice with chimeric BM were fed the HFD (D12492 diet; Research Diets) for 8 weeks. The hepatocytes were isolated, and the proportion of lymphocytes and myeloid cells was calculated using flow cytometry. ## Data availability. The source data presented in Figures 1–6, Supplemental Figures 1–12, and Supplemental Tables 1 and 2, and RNA-Seq data have been deposited into the Dryad repository (https://doi.org/10.5061/dryad.m37pvmd36). ## Statistics. All values are presented as mean ± SEM. The violin plots depict the median, quartiles, and data range. The normality of the data was assessed by the Shapiro-Wilk test, followed by 2-tailed Student’s t test or the Mann-Whitney U test for statistical significance at 2 groups, whereas those between multiple groups (≥3 groups) were compared using 1-way ANOVA followed by Dunnett’s test or the Kruskal-Wallis test followed by the Dunn’s post hoc test. Differences were considered significant at $P \leq 0.05.$ The FDRs of RNA-Seq data were estimated using the Benjamini-Hochberg procedure. ## Study approval. All experimental procedures involving mice were performed according to the protocols approved by the Committee on the Ethics of Animal Experiments of the Kyoto University Animal Experimentation Committee (Lif-K21020) and the Tokyo University of Agriculture and Technology (permit 28–87). All mice were sacrificed under deep anesthesia using isoflurane. All studies were approved by the institutional review board of Kyushu University (approval 29-476, 2021-71) and performed in accordance with relevant guidelines. Written informed consent was obtained from patients at the time of recruitment, and their records were anonymized and deidentified. ## Author contributions ROK performed the experiments, interpreted data, and wrote the paper; HN performed the experiments, interpreted data, and wrote the paper; AN performed the experiments and interpreted data; YM performed the experiments; DT performed the experiments and interpreted data; TI performed the experiments; AU performed the experiments; MT interpreted data; MK performed the experiments; MI interpreted data; HK performed the experiments and interpreted data; TT interpreted data; AI interpreted data; TS interpreted data; KH interpreted data; YO interpreted data; and JA interpreted data. IK supervised the project, interpreted data, and wrote the paper; IK also had primary responsibility for the final content. All authors read and approved the final manuscript. ## 12/08/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Ferguson D, Finck BN. Emerging therapeutic approaches for the treatment of NAFLD and type 2 diabetes mellitus. Nat Rev Endocrinol (2021) 17 484-495. DOI: 10.1038/s41574-021-00507-z 2. Zhu C. Maladaptive regeneration - the reawakening of developmental pathways in NASH and fibrosis. Nat Rev Gastroenterol Hepatol (2021) 18 131-142. DOI: 10.1038/s41575-020-00365-6 3. Looma R. Mechanisms and disease consequences of nonalcoholic fatty liver disease. Cell (2021) 184 2537-2564. DOI: 10.1016/j.cell.2021.04.015 4. Targher G. The complex link between NAFLD and type 2 diabetes mellitus - mechanisms and treatments. Nat Rev Gastroenterol Hepatol (2021) 18 599-612. DOI: 10.1038/s41575-021-00448-y 5. Woo Baidal JA, Lavine JE. The intersection of nonalcoholic fatty liver disease and obesity. Sci Transl Med (2016) 8. PMID: 26819197 6. Vuppalanchi R. Therapeutic pipeline in nonalcoholic steatohepatitis. Nat Rev Gastroenterol Hepatol (2021) 18 373-392. DOI: 10.1038/s41575-020-00408-y 7. Kimura I. Free fatty acid receptors in health and disease. Physiol Rev (2020) 100 171-210. DOI: 10.1152/physrev.00041.2018 8. Kimura I. Short-chain fatty acids and ketones directly regulate sympathetic nervous system via G protein-coupled receptor 41 (GPR41). Proc Natl Acad Sci U S A (2011) 108 8030-8035. DOI: 10.1073/pnas.1016088108 9. Kimura I. The gut microbiota suppresses insulin-mediated fat accumulation via the short-chain fatty acid receptor GPR43. Nat Commun (2013) 4. DOI: 10.1038/ncomms2852 10. Kimura I. Maternal gut microbiota in pregnancy influences offspring metabolic phenotype in mice. Science (2020) 367. DOI: 10.1126/science.aaw8429 11. Miyamoto J. Ketone body receptor GPR43 regulates lipid metabolism under ketogenic conditions. Proc Natl Acad Sci U S A (2019) 116 23813-23821. DOI: 10.1073/pnas.1912573116 12. Miyamoto J. Gut microbiota confers host resistance to obesity by metabolizing dietary polyunsaturated fatty acids. Nat Commun (2019) 10. DOI: 10.1038/s41467-019-11978-0 13. Wojciechowicz ML, Ma’ayan A. GPR84: an immune response dial?. Nat Rev Drug Discov (2020) 19. DOI: 10.1038/d41573-020-00029-9 14. Wang J. Medium-chain fatty acids as ligands for orphan G protein-coupled receptor GPR84. J Biol Chem (2006) 281 34457-34464. DOI: 10.1074/jbc.M608019200 15. Luscombe VB. 20 years an orphan: is GPR84 a plausible medium-chain fatty acid-sensing receptor?. DNA Cell Biol (2020) 39 1926-1937. DOI: 10.1089/dna.2020.5846 16. Babayan VK. Medium chain fatty acid esters and their medical and nutritional applications. J Am Oil Chem Soc (1981) 58 49A-51A. DOI: 10.1007/BF02666072 17. St-Onge MP, Jones PJ. Physiological effects of medium-chain triglycerides: potential agents in the prevention of obesity. J Nutr (2002) 132 329-332. DOI: 10.1093/jn/132.3.329 18. Baba N. Enhanced thermogenesis and diminished deposition of fat in response to overfeeding with diets containing medium chain triglycerides. Am J Clin Nutr (1982) 35 678-682. DOI: 10.1093/ajcn/35.4.678 19. Bach AC, Babayan VK. Medium-chain triglycerides: an update. Am J Clin Nutr (1982) 36 950-962. DOI: 10.1093/ajcn/36.5.950 20. Kaunitz H. Nutritional properties of the triglycerides of medium chain-length. J Am Oil Chem Soc (1958) 35 10-13. DOI: 10.1007/BF02582731 21. Lattin JE. Expression analysis of G protein-coupled receptors in mouse macrophages. Immunome Res (2008) 4. DOI: 10.1186/1745-7580-4-5 22. Venkataraman C, Kuo F. The G-protein coupled receptor, GPR84 regulates IL-4 production by T lymphocytes in response to CD3 crosslinking. Immunol Lett (2005) 101 144-153. DOI: 10.1016/j.imlet.2005.05.010 23. Nagasaki H. Inflammatory changes in adipose tissue enhance expression of GPR84, a medium-chain fatty acid receptor: TNFα enhances GPR84 expression in adipocytes. FEBS Lett (2012) 586 368-372. DOI: 10.1016/j.febslet.2012.01.001 24. Du Toit E. Effect of GPR84 deletion on obesity and diabetes development in mice fed long chain or medium chain fatty acid rich diets. Eur J Nutr (2018) 57 1737-1746. DOI: 10.1007/s00394-017-1456-5 25. Nonaka H. Dietary medium-chain triglyceride decanoate affects glucose homeostasis through GPR84-mediated GLP-1 secretion in mice. Front Nutr (2022) 9. DOI: 10.3389/fnut.2022.848450 26. Duan Y. Inflammatory links between high fat diets and diseases. Front Immunol (2018) 9. DOI: 10.3389/fimmu.2018.02649 27. Korbecki J, Bajadak-Rusinek K. The effect of palmitic acid on inflammatory response in macrophages: an overview of molecular mechanisms. Inflamm Res (2019) 68 915-932. DOI: 10.1007/s00011-019-01273-5 28. Alsamman S. Targeting acid ceramidase inhibits YAP/TAZ signaling to reduce fibrosis in mice. Sci Transl Med (2020) 12. DOI: 10.1126/scitranslmed.aay8798 29. Gaidarov I. Embelin and its derivatives unravel the signaling, proinflammatory and antiatherogenic properties of GPR84 receptor. Pharmacol Res (2018) 131 185-198. DOI: 10.1016/j.phrs.2018.02.021 30. Inoue A. TGFα shedding assay: an accurate and versatile method for detecting GPCR activation. Nat Methods (2012) 9 1021-1029. DOI: 10.1038/nmeth.2172 31. Kawakubo M. Dipeptidyl peptidase-4 inhibition prevents nonalcoholic steatohepatitis-associated liver fibrosis and tumor development in mice independently of its anti-diabetic effects. Sci Rep (2020) 10. DOI: 10.1038/s41598-020-57935-6 32. Newsome PN. A placebo-controlled trial of subcutaneous semaglutide in nonalcoholic steatohepatitis. N Engl J Med (2021) 384 1113-1124. DOI: 10.1056/NEJMoa2028395 33. Wang B. Effects of long-chain and medium-chain fatty acids on apoptosis and oxidative stress in human liver cells with steatosis. J Food Sci (2016) 81 H794-H800. DOI: 10.1111/1750-3841.13210 34. Yang J. Effects of medium- and long-chain fatty acids on acetaminophen- or rifampicin-induced hepatocellular injury. Food Sci Nutr (2020) 8 3590-3601. DOI: 10.1002/fsn3.1641 35. Chen LH. Modulation of the G-protein-coupled receptor 84 (GPR84) by agonists and antagonists. J Med Chem (2020) 63 15399-15409. DOI: 10.1021/acs.jmedchem.0c01378 36. Recio C. Activation of the immune-metabolic receptor GPR84 enhances inflammation and phagocytosis in macrophages. Front Immunol (2018) 9. DOI: 10.3389/fimmu.2018.01419 37. Suzuki M. Medium-chain fatty acid-sensing receptor, GPR84, is a proinflammatory receptor. J Biol Chem (2013) 288 10684-10691. DOI: 10.1074/jbc.M112.420042 38. Grouix B. PBI-4050 reduces stellate cell activation and liver fibrosis through modulation of intracellular ATP levels and the liver kinase B1/AMP-activated protein kinase/mammalian target of rapamycin pathway. J Pharmacol Exp Ther (2018) 367 71-81. DOI: 10.1124/jpet.118.250068 39. Puengel T. The medium-chain fatty acid receptor GPR84 mediates myeloid cell infiltration promoting steatohepatitis and fibrosis. J Clin Med (2020) 9. DOI: 10.3390/jcm9041140 40. Simard JC. Fatty acid mimetic PBI-4547 restores metabolic homeostasis via GPR84 in mice with non-alcoholic fatty liver disease. Sci Rep (2020) 10. DOI: 10.1038/s41598-020-69675-8 41. Labéguère F. Discovery of 9-Cyclopropylethynyl-2-((. J Med Chem (2020) 63 13526-13545. DOI: 10.1021/acs.jmedchem.0c00272 42. Mahindra A. Investigating the structure-activity relationship of 1,2,4-Triazine G-protein-coupled receptor 84 (GPR84) antagonists. J Med Chem (2022) 65 11270-11290. DOI: 10.1021/acs.jmedchem.2c00804 43. Matsumoto M. An improved mouse model that rapidly develops fibrosis in non-alcoholic steatohepatitis. Int J Exp Pathol (2013) 94 93-103. DOI: 10.1111/iep.12008 44. Bangen JM. Targeting CCl. Hepatology (2017) 66 1242-1257. DOI: 10.1002/hep.29275 45. Bolger AM. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics (2014) 30 2114-2120. DOI: 10.1093/bioinformatics/btu170 46. Cock PJ. The Sanger FASTQ file format for sequences with quality scores, and the Solexa/Illumina FASTQ variants. Nucleic Acids Res (2010) 38 1767-1771. DOI: 10.1093/nar/gkp1137 47. Kim D. HISAT: a fast spliced aligner with low memory requirements. Nat Methods (2015) 12 357-360. DOI: 10.1038/nmeth.3317 48. Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods (2012) 9 357-359. DOI: 10.1038/nmeth.1923 49. Pertea M. StringTie enables improved reconstruction of a transcriptome from RNA-seq reads. Nat Biotechnol (2015) 33 290-295. DOI: 10.1038/nbt.3122 50. Inoue A. Illuminating G-protein-coupling selectivity of GPCRs. Cell (2019) 177 1933-1947.e25. DOI: 10.1016/j.cell.2019.04.044 51. Nakajima A. The short chain fatty acid receptor GPR43 regulates inflammatory signals in adipose tissue M2-type macrophages. PLoS One (2017) 12. DOI: 10.1371/journal.pone.0179696 52. Hayashima K. Role of ferritinophagy in cystine deprivation-induced cell death in glioblastoma cells. Biochem Biophys Res Commun (2021) 539 56-63. DOI: 10.1016/j.bbrc.2020.12.075 53. Wu JC. Establishment and characterization of differentiated, nontransformed hepatocyte cell lines derived from mice transgenic for transforming growth factor alpha. Proc Natl Acad Sci U S A (1994) 91 674-678. DOI: 10.1073/pnas.91.2.674 54. Lynch RW. An efficient method to isolate Kupffer cells eliminating endothelial cell contamination and selective bias. J Leukoc Biol (2018) 104 579-586. DOI: 10.1002/JLB.1TA0517-169R
--- title: Interfering with lipid metabolism through targeting CES1 sensitizes hepatocellular carcinoma for chemotherapy authors: - Gang Li - Xin Li - Iqbal Mahmud - Jazmin Ysaguirre - Baharan Fekry - Shuyue Wang - Bo Wei - Kristin L. Eckel-Mahan - Philip L. Lorenzi - Richard Lehner - Kai Sun journal: JCI Insight year: 2023 pmcid: PMC9977307 doi: 10.1172/jci.insight.163624 license: CC BY 4.0 --- # Interfering with lipid metabolism through targeting CES1 sensitizes hepatocellular carcinoma for chemotherapy ## Abstract Hepatocellular carcinoma (HCC) is the most common lethal form of liver cancer. Apart from surgical removal and transplantation, other treatments have not yet been well established for patients with HCC. In this study, we found that carboxylesterase 1 (CES1) is expressed at various levels in HCC. We further revealed that blockage of CES1 by pharmacological and genetical approaches leads to altered lipid profiles that are directly linked to impaired mitochondrial function. Mechanistically, lipidomic analyses indicated that lipid signaling molecules, including polyunsaturated fatty acids (PUFAs), which activate PPARα/γ, were dramatically reduced upon CES1 inhibition. As a result, the expression of SCD, a PPARα/γ target gene involved in tumor progression and chemoresistance, was significantly downregulated. Clinical analysis demonstrated a strong correlation between the protein levels of CES1 and SCD in HCC. Interference with lipid signaling by targeting the CES1-PPARα/γ-SCD axis sensitized HCC cells to cisplatin treatment. As a result, the growth of HCC xenograft tumors in NU/J mice was potently slowed by coadministration of cisplatin and CES1 inhibition. Our results, thus, suggest that CES1 is a promising therapeutic target for HCC treatment. ## Introduction Liver cancer is prevalent worldwide and is ranked as the third leading cause of cancer-related deaths [1]. Most adult liver cancers are hepatocellular carcinoma (HCC) and often have a poor prognosis, owing to the lack of effective therapies [2]. Currently, the most practical methods for HCC treatment are surgical and transplantation resection. However, the outcomes of surgical approaches are poor, with high recurrence rates [3]. Despite many years of dedicated studies, no other standard treatments have been formally established for HCC [4]. In this context, cisplatin has recently drawn great clinical attention because of its promising killing effect on advanced HCC. Cisplatin is a chemotherapeutic agent used to treat a wide range of human cancers [5, 6]. It exerts an antitumor effect mainly by interfering with genomic DNA replication, which induces DNA damage and apoptosis, thereby killing rapidly proliferating cancer cells. Unfortunately, while initial responsiveness is high, most HCC patients exhibit different degrees of drug insensitivity and chemoresistance upon treatment with cisplatin for prolonged periods [5]. Mechanistically, several lipid metabolic pathways have been identified as the key factors that induce cisplatin resistance in HCC [7]. Therefore, new combination regimens including cisplatin and interference with lipid metabolism might be an effective strategy to deal with chemoresistance in patients with advanced HCC (5–8). In addition to hepatitis infections and alcoholic injury, lipid disorder–related liver diseases, such as nonalcoholic fatty liver disease (NAFLD) and nonalcoholic steatohepatitis (NASH), have been directly linked to the development of HCC (9–14). Abnormal lipid metabolism is recognized as a pivotal factor that plays a critical role in HCC development and progression [15]. As the primary organ for lipid metabolism, the liver synthesizes fatty acids, which form triglycerides (TG) and other lipids via lipogenesis [16]. Different lipid species not only serve as the main energy source via β-oxidation in mitochondria, but also function as key building blocks for the growth of cancer cells. Furthermore, unique free fatty acids (FFAs) produced by lipolysis may act as a “third messenger” to trigger signaling pathways for HCC initiation, progression, and maintenance [17, 18]. A fine-tuned balance between lipid biosynthesis, desaturation, and metabolism is key to maintaining normal liver function, and disruptions of this balance can be the cause and consequence of fatty liver diseases and, hence, HCC (19–21). However, limited knowledge of the hepatic lipidome has prevented the development of related therapeutic agents to treat lipid disorder–induced HCC. The lipid components are assembled into lipid droplets in hepatocytes and other cell types as well. Lipid droplets serve as a major platform for dynamics of lipid metabolism. Many important structural proteins and enzymes, such as perilipins (PLIN1–PLIN5), CIDEA–C, ATGL, and CGI-58, are specifically located on the surface of lipid droplets. They tightly regulate the formation, growth, function, and turnover of the lipid droplets [22]. Given their key roles in lipid storage, membrane biosynthesis, lipid signaling, and inflammation in cells, lipid droplets have gradually been recognized as critical organelles in cancer cells [23]. Importantly, the components in the lipid droplets of the cancer cells exhibit unique features, including markedly increased levels of monounsaturated fatty acids (MUFA) and polyunsaturated fatty acids (PUFA). These special lipid molecules trigger multiple oncogenic signaling pathways that promote tumor growth (19, 23–25). A key enzyme that converts saturated fatty acids (SFAs) to MUFAs and PUFAs is stearoyl-CoA desaturase 1 (SCD1, referred to as SCD in humans) [26]. SCD is ubiquitously expressed in most cancer cells, and its levels are tightly associated with the aggressiveness of cancers [27]. Recent studies have highlighted its direct function in cancer cell stemness, proliferation, migration, and metastasis through regulation of lipid signaling pathways and membrane architecture (28–30). More importantly, SCD has also been linked to chemoresistance in certain types of cancers, including HCC [24, 28]. Based on previous findings, several specific inhibitors targeting SCD activity have been developed and are under preclinical tests to treat and/or deal with the chemoresistance of certain types of cancers [17, 31]. Nevertheless, the lipid signaling–driven pathways that trigger SCD activation in tumor cells remain unclear. Carboxylesterase 1 (CES1) belongs to a large mammalian serine esterase family [32]. CES1 is enriched in metabolically active tissues, including liver and white and brown adipose tissues [33, 34]. It catalyzes the hydrolysis of ester and thioester bonds in lipids both in vitro and in vivo and, hence, plays essential roles in lipid metabolism and whole-body energy homeostasis [33]. In rodents, its homolog is referred to as Ces1d or Ces3/TGH [32, 33]. While it has been considered to exert a catalytic function on lipids in the ER, our recent studies demonstrated that Ces1d directly targets lipid droplets, where it hydrolyzes TG and produces FFAs that promote energy expenditure [34, 35]. Particularly, in the liver, CES1 catabolizes lipids and promotes the assembly of apolipoproteins, thereby maintaining whole-body lipid metabolic homeostasis [33, 36, 37]. Intriguingly, even though artificially overexpressed CES1 was shown to exert a antiproliferative function in the liver cancer cell line Hep3B, its protein levels have long been considered undetectable in HCC and HCC-derived cell lines, probably because of the lack of high-affinity antibodies that specifically recognize endogenous CES1 (38–40). In this context, the bona fide function and regulation of CES1 in HCC per se remain to be elucidated. In this study, we used a recently established high-affinity anti-CES1 antibody to analyzed the protein levels of CES1 in an array of human liver tumor samples. The results revealed that the CES1 protein levels were detectable and varied among the samples. We further demonstrated that blockage of CES1 activity by a specific inhibitor WWL229 — which targets the active site of CES1 and, hence, inhibits its enzymatic activity [41] — or by genetic KO led to reprogrammed lipid metabolism. Consequently, mitochondrial function was impaired in response to the inhibition of CES1. Mechanistically, we found that key lipid signaling molecules that potentially trigger PPARα/γ transactivation, including multiple PUFAs, were significantly reduced when the activity of CES1 was blocked. As a result, the expression of SCD, a direct target of PPARα/γ, was dramatically downregulated. The lipid metabolism, whose interference was induced by reduced SCD, potently sensitized HCC cells to chemotherapeutic agents, such as cisplatin treatment. Our findings suggest that CES1 plays a role in regulation of HCC progression and chemoresistance; thus, they pinpoint it as a potential target for HCC therapy. ## CES1 is selectively expressed at different levels in human liver tumors. Previous studies have shown low to undetectable protein levels of CES1 in HCC and HCC-derived cell lines (38–40). However, with the newly developed high-affinity anti-CES1 antibodies, various levels of CES1 protein have been detected in different cancer cells, including HCC (www.proteinatlas.org/ENSG00000198848-CES1/pathology/liver+cancer#imid_19180094). Herein, we analyzed the levels of CES1 protein in an array of human liver cancer samples ($$n = 120$$) using immunofluorescence staining with the reported anti-CES1 antibody. The results revealed that the protein abundance of CES1 varied among different liver cancer patients (Figure 1A and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.163624DS1). Overall, the CES1 protein was detectable in most HCC samples ($$n = 110$$). Among them, some were higher, while others were lower than those in normal livers ($$n = 10$$) (Figure 1A). Quantitative analysis further indicated that the average levels of CES1 protein in HCC were significantly lower than those in normal livers and that the levels in hepatocholangiocarcinoma were even lower than those in HCC (Figure 1B). Intriguingly, the levels in grade 2 HCC were decreased, while the protein increased to almost the same levels as in the normal liver in grade 3 HCC (Figure 1C). Further analysis of the different stages of HCC showed that the levels reduced when the tumors developed to an advanced stage (Supplemental Figure 1B). Interestingly, when analyzing the levels of CES1 in different HCC cell lines, we found that HepG2 cells synthesized CES1 at a level that was similar to normal mouse and human livers, whereas the protein in SNU449 and Hep3B was undetectable (Figure 1D). We further analyzed CES1 expression in other cancer types based on the available databases (tnmplot.com, xena.ucsc.edu, and kmplot.com). First, we compared the expression levels of CES1 in normal and malignant human tissues (https://tnmplot.com/analysis/) [42]. The results suggest that the liver expressed the highest levels of CES1 among all the tissues. Intriguingly, the mRNA levels were significantly upregulated in malignant liver tissue (Supplemental Figure 1C) — a result that is different from the results on the protein abundance (Figure 1, A and B). We then analyzed the correlation between the protein levels of CES1 and survival probability in different cancers. For the analysis, the cutoff values to define the levels with “low” or “high” are the lower and upper quartiles of the CES1 expression. The results indicate that the correlation between CES1 and survival probability in pancancers is weak (Figure 1E). However, there remained a trend toward a negative correlation between the levels of CES1 and the survival probability in nonalcoholic, nonhepatitis virus–infected HCC patients (Figure 1F). In several other cancer types, including gastric adenocarcinoma, bladder carcinoma, and head and neck squamous cell carcinoma if measured in enough cancer populations, the levels of CES1 were negatively correlated with the survival rate (Figure 1, G–I), while in other cancers, if measured in enough cancer populations, the levels of CES1 were positively correlated with survival rate (Supplemental Figure 1, D–U). In summary, our experimental and metaanalysis results suggest that CES1 is expressed at various levels in HCC and many other cancer types. They might play divergent roles in different types of cancers. ## Blockage of CES1 activity alters the dynamics of lipid droplets in HepG2 cells. Given that we confirmed that liver cancer cells express varied levels of CES1, we sought to determine the role of CES1 in HepG2 cells, a well-characterized hepatoblastoma cell line that expresses high levels of CES1 (Figure 1D). Previously, we revealed that CES1 regulates the dynamics of lipid droplets by translocating onto their surfaces and digesting the lipid content in normal tissues [34, 35]. Herein, BODIPY staining revealed that there were more lipid droplets with significantly larger sizes in the HepG2 cells when treated with WWL229, a specific inhibitor of CES1 (Figure 2, A–C). Similar results were observed in CES1-KO (by a CRISPR deletion) HepG2 cells (Figure 2, A–C). Interestingly, when treating CES1-KO cells with WWL229, the number and morphology of the lipid droplets did not further change, suggesting the specific inhibitory effect of WWL229 on CES1 activity. The results are in line with previous findings [34]. To further characterize the changes in the lipid profiles upon CES1 inhibition, we performed untargeted lipidomic analysis using high-resolution mass spectrometry (MS) on cell lysates collected from WWL229-treated and CES1-KO HepG2 cells. The results show that TGs were globally increased, eventually leading to TG accumulation in WWL229-treated and CES1-KO cells (Figure 2D and Supplemental Figure 2A). Moreover, other components in the lipid droplets, such as stearyl esters, diradylglycerols, diacylglycerols, and alkyladiacylglycerols, were also increased or tended to increase (Figure 2E). In contrast, the levels of total FFA were significantly reduced in WWL229-treated and CES1-KO cells (Figure 2F). To determine whether loss of function of CES1 affects the dynamics of lipid droplets, we measured the levels of several key lipases and lipogenic enzymes that are related to lipid droplet dynamics. Western blotting results showed that lipid droplet–targeting lipolytic factors, such as ATGL, HSL, and CGI-58, did not significantly change (Supplemental Figure 2, B and C). Fatty acid synthetic and lipogenic factors, such as FASN and DGAT2, also did not change (Supplemental Figure 2, D and E). Intriguingly, while the total levels of the de novo lipogenic enzyme ACC1 were decreased, the ratio of phosphorylated ACC1/total ACC1 were slightly increased upon WWL229 treatment (Supplemental Figure 2, D and E). Similar results were observed in CES1-KO cells (Supplemental Figure 2, F–I), suggesting slowed progression of de novo lipogenesis upon loss of function of CES1 in the cells. Next, we determined whether other lipid droplet–associated factors were altered. Western blotting results revealed that PLIN2 and PLIN3 were slightly decreased, while other factors, such as PLIN5, CIDEA, and CIDEC, were not significantly changed upon WWL229 treatment (Figure 2, G and H). In contrast, while all other factors remained unchanged, CIDEA levels were significantly increased in CES1-KO cells (Figure 2, I and J). In summary, altered morphology and dynamics of lipid droplets were observed in response to the blockage of CES1. ## Blockage of CES1 activity leads to impaired mitochondrial function. CES1 hydrolyzes lipids to produce FFAs that serve as a fuel source for mitochondrial β-oxidation [34]. The ontology of the lipidomic profiles indicated that blockage of CES1 activity by either WWL229 or CES1 KO reduced the levels of lipid components involved in the plasma membrane, mitochondrial membrane, and endoplasmic reticulum (ER) membrane (Figure 3A). Interestingly, lipidomic analysis further revealed that the levels of acylcarnitine (16:0) were dramatically reduced, suggesting insufficient fatty acid β-oxidation (FAO) in WWL229-treated and CES1-KO cells (Figure 3B). The total levels of acylcarnitine also tended to be decreased in WWL229-treated and CES1-KO cells (Supplemental Figure 3A). In agreement with the lipidomic analysis, FAO activities were significantly reduced in WWL229-treated cells (Figure 3C), while they tended to be decreased in CES1-KO cells ($$P \leq 0.0573$$, Figure 3D). To determine whether loss of function of CES1 affects mitochondrial function, we monitored the oxygen consumption rate (OCR) by using a Seahorse XFe analyzer. The results revealed that key parameters, including the OCR of mitochondria, were massively decreased upon the addition of the modulators of respiration into WWL229-treated cells (Figure 3, E and F). Similar results were observed in CES1-KO cells (Figure 3, G and H). In line with the impaired mitochondrial function, quantitative PCR (qPCR) results revealed that the expression levels of mitochondrial oxidation enzymes, such as ACADS, ACADM, ACADVL, CPT2, and ECH1 were significantly downregulated in the WWL229-treated HepG2 cells (Figure 3I). Consistently, other mitochondrial biogenetic genes such as TFAM1 and NRF1 were also downregulated (Figure 3J). Similar results were observed in CES1-KO HepG2 cells, while reexpression of CES1 recovered the gene expression levels to various degrees (Figure 3, K and L) in the KO cells. Notably, the protein levels of the mitochondrial respiratory complexes (I–V) were not altered in response to the inhibition of CES1 (Supplemental Figure 3, B and C). In conclusion, blocking CES1 activity impairs the respiratory function of mitochondria, which may affect the progression of tumor growth. ## Blockage of CES1 activity causes reduced levels of SCD in HepG2 cells. SCD is a critical lipid-synthesizing enzyme. Its products, including MUFAs and multiple PUFAs, play essential roles in tumor cell proliferation and chemoresistance [28]. Increased SCD has been demonstrated to correlate with tumor aggressiveness and poor patient diagnosis [9]. Immunofluorescence analysis of an array of liver tumor samples with a specific anti-SCD antibody revealed that the protein levels of SCD varied among the different samples (Figure 4A). Importantly, we found a strong positive correlation between the protein levels of SCD (Figure 4A) and CES1 (Figure 1A and Figure 4B). Consistent with these results, qPCR analysis revealed that the expression levels of SCD were significantly downregulated upon WWL229 treatment in HepG2 cells (Figure 4C). Western blotting and immunofluorescence staining results further indicate that the protein levels of SCD were also reduced (Figure 4, D–F). Similarly, both the mRNA and protein levels were decreased in the CES1-KO cells, and reexpression of CES1 restored the levels of SCD in the KO cells (Figure 4, G–L). Interestingly, the rescuing effect of reexpressing CES1 was enhanced when cells were treated with FFAs such as oleic acid (OA) and palmitic acid (PA) (Supplemental Figure 4, A and B). The results of the lipidomic analysis are in line with the reduced levels of SCD upon blockage of CES1. Particularly, the lipid ontology of the global lipidomic profiles revealed that total PUFA levels were decreased in WWL229-treated and CES1-KO cells (Figure 4M). Notably, different species of PUFAs were reduced in WWL229-treated and CES1-KO cells at different levels (Figure 4M). Some MUFAs were also decreased in WWL229-treated and CES1-KO cells (Figure 4M). These results suggest that CES1 reprograms FFA saturation via regulation of SCD. ## PPARα/γ are involved in the CES1-mediated SCD regulation. Next, we sought to define the mechanisms by which CES1 downregulates SCD. PPARα/γ has been reported to directly regulate SCD gene expression [43]. Meanwhile, multiple PUFAs have been shown to activate the transcriptional activity of PPARα/γ [44]. Interestingly, when analyzing the levels of FFAs by liquid chromatography–tandem MS (LC-MS/MS), we found that the total PUFAs were significantly diminished in WWL229-treated and CES1-KO cells (Figure 4N and Supplemental Figure 4C). We further found that blocking the functions of PPARα and/or PPARγ using their specific siRNAs abolished the recovery effects of CES1 on SCD expression in the CES1-KO cells, suggesting that PPAR α/γ mediates the function of CES1 on regulation of SCD (Figure 4, O–Q; refer to Supplemental Figure 4, D and E, for the PPARα/γ knockdown efficiency and Supplemental Figure 4F for the protein levels of CES1 reexpression). Intriguingly, the liver-enriched nuclear receptor HNF4α was also reported to be activated by lipid signaling triggered by CES1 [35]. However, knockdown of HNF4α did not affect CES1-mediated SCD regulation in HepG2 cells (Supplemental Figure 4, G and H). Collectively, our results suggest that blockage of CES1 activity leads to downregulation of SCD. This effect is at least partially due to the diminished levels of PUFAs, which reduced PPARα/γ transcriptional activity on SCD expression. ## Blockage of CES1 activity induces ROS accumulation. Dysfunctional mitochondria and diminished SCD may lead to altered reactive oxygen species (ROS) levels [45, 46]. Consistently, flow cytometry analysis revealed that WWL229-treated HepG2 cells exhibited a shift from 2’,7’-dichlorodihydrofluorescein (DCF, the ROS detector) to the DCF+ population when compared with the control cells (Figure 5A). Quantitative analysis showed a marked increase in the DCF+ cells after WWL229 treatment (Figure 5B). Similar results were observed in CES1-KO cells (Figure 5, C and D). In agreement with an increase in ROS, antioxidative enzymes, such as SOD1, SOD2, GPX1, and CAT1, were downregulated in WWL229-treated cells (Figure 5E). A similar effect was found in CES1-KO cells, and reexpression of CES1 back to the cells restored the expression of the enzymes (Figure 5F). Both mitochondrial and SCD dysregulation leads to ER stress [47, 48]. We next determined the pathological changes in the ER in response to blockage of CES1 activity in HepG2 cells. Western blotting results showed that the BIP and XBP1s/u proteins were increased in WWL229-treated HepG2 cells, suggesting ER stress in the cells (Figure 5, G and H). Intriguingly, while we observed the same alteration for the XBP1 proteins, the BIP levels were not altered in the CES1-KO HepG2 cells, suggesting some unidentified compensatory effect in the permanent KO cells (Figure 5, I and J). In conclusion, blockage of CES1 accelerated ROS production, while mildly inducing ER stress in HepG2 cells. ## Blockage of CES1 activity sensitizes the HCC to cisplatin treatment. A recent study reported that abnormal TG catabolism by CES1 promotes aggressive colorectal tumor growth [49]. To determine whether loss of function of CES1 affects cell proliferation in HCC, we treated HepG2 cells with different doses of WWL229 and measured their viability using the MTT assay. Surprisingly, we only detected a mild effect on cell viability, even under the treatments of higher doses of WWL229 (20–50 μM) (Figure 6A). However, when cells were cotreated with the same doses of WWL229 and 20 μM cisplatin, a well-known anticancer agent, we found a synergistic effect of reduced cell survivability, even at relatively lower doses of WWL229 (0.5–5 μM) (Figure 6B). Moreover, when we chose a medium dose of WWL229 (50 μM) and cotreated the cells with different doses of cisplatin, we found that treatment with WWL229 markedly sensitized the cells to different doses of cisplatin (Figure 6C). A similar effect was observed in the CES1-KO cells (Figure 6D). Importantly, when we treated SNU449 cells (another liver cancer cell line that does not contain endogenous CES1) with cisplatin, we found a responsive cell killing effect. However, when we expressed a relatively low level of CES1 in them, the protein levels of SCD increased and the cells exhibited a resistance to cisplatin treatment (Figure 6, E and F). We further compared cell apoptosis in the groups by flow cytometry using annexin V+ cells. The results revealed that significantly more apoptotic cells were detected in the cotreated HepG2 cells than in the single treatment with either WWL229 (50 μM) or cisplatin (10 μM) (Figure 6, G and H). Consistently, more cleaved caspase 3, a marker of cell apoptosis, was detected in the cotreated cells (Supplemental Figure 5A). Similar results were observed in CES1-KO cells when treated with cisplatin (10 μM), while ablation of CES1 itself did not induce dramatic cell apoptosis in the CES1-KO cells (Figure 6, I and J, and Supplemental Figure 5B). Notably, while reexpression of CES1 efficiently reduced the cell apoptosis induced by cisplatin in CES1-KO cells, blocking the activity of SCD by its specific inhibitor MF348 significantly abolished the rescuing effect of CES1 reexpression (Figure 6, I and J), suggesting the key role of SCD in CES1-mediated HCC cell growth. ## Synergistic effect of CES1 inhibition and cisplatin treatment in HCC xenograft tumors. To further test the synergistic effect of loss of function of CES1 and treatment of cisplatin, we performed a tumor growth assay in HepG2 xenografted NU/J mice. The results indicate that, while single treatment of WWL229 (155 μmol/kg) did not have a significant effect and while and single treatment of cisplatin (10 μmol/kg) only slightly inhibited the growth of xenograft tumors, cotreatment with WWL229 and cisplatin significantly inhibited tumor growth (Figure 7A). In agreement with the dynamic changes of the tumor growth during the cotreatment, morphological examination of the xenograft tumors after the treatments showed that the sizes of the xenografts collected from the cotreated mice were dramatically smaller than those of the single-treated groups (Figure 7B; 1 sample in the cotreated samples shrank to an undetectable size, as shown by the dashed circle). A similar inhibitory effect on xenograft growth was observed in CES1-KO HepG2 cells when treated with cisplatin (10 μmol/kg), while ablation of CES1 itself did not affect the CES1-KO xenograft tumor growth (Figure 7, C and D, and Supplemental Figure 6, A and B). Consistently, we detected a decreased ratio of phosphorylated AKT and total AKT in CES1-KO xenografts, suggesting reduced tumor growth (Supplemental Figure 6, C and D). Notably, we did not find any differences in body weights between the groups during the entire treatment process (data not shown). Sorafenib is another chemotherapeutic agent used to treat advanced HCC [50, 51]. Interestingly, a correlation analysis using the KM plotter (kmplot.com) demonstrated a trend toward a negative association between the levels of CES1 and survival probability in HCC patients treated with sorafenib (Supplemental Figure 6E), suggesting a potential synergistic effect between CES1 inhibition and sorafenib treatment. However, more experimental and preclinical studies are needed to test this hypothesis. ## A working model for the role of CES1 in HCC growth. A working model is proposed based on our findings in this study. In the model, FFAs produced by CES1 fuel the mitochondria for β-oxidation and ATP production to support tumor growth. Meanwhile, some unique FFAs produced by CES1, such as multiple PUFAs, may function as signaling molecules for PPARα/γ activation. Upon activation, PPARα/γ binds to the SCD promoter and, hence, upregulates its expression. Upregulated SCD further promotes tumor growth by decreasing ER stress and increasing the levels of phosphorylated AKT and other pathways. Consequently, the enhanced mitochondrial function and increased levels of SCD induced by CES1 activation promote tumor growth and potential chemoresistance (Figure 8, left). In contrast, blockage of CES1 activity by pharmacological or genetic approaches impairs mitochondrial function, increases ROS production, and decreases the levels of SCD, thereby sensitizing HCC to chemotherapeutic agents, such as cisplatin and sorafenib (Figure 8, right). ## Discussion Lipid metabolism reprogramming has drawn considerable attention as an essential factor in tumor development and progression [17, 52]. CES1 is a key enzyme that plays an important role in lipid metabolism [33]. In humans, CES1 is abundantly expressed in the normal liver and adipose tissues, where it plays an important role in lipid droplet metabolism, lipoprotein assembly, and secretion [33, 36, 53]. However, for a long time, it was considered to be undetectable in HCC and human liver cancer cell lines [38, 39, 54, 55]. Although artificially transfected CES1 exerted antiproliferative effects in liver cancer cell lines [38], the bona fide role of CES1 in HCC per se has not yet been well characterized. In this study, we have demonstrated that CES1 was selectively expressed at various levels in different liver cancer samples and HCC cell lines. Intriguingly, while the mRNA levels of CES1 are upregulated consistently in all the liver tumors, its protein levels exhibit high heterogeneity among different types of HCC, suggesting profound posttranscriptional regulations of CES1. Of note, the trend toward a negative correlation between the levels of CES1 and the progression of HCC suggests a potential role of CES1 in NAFLD-induced HCC development. Further studies are needed to confirm the hypothesis. More importantly, for the first time to our knowledge, we report that, while blockage of CES1 only had a mild effect on inhibition of HCC growth and expansion, it potently sensitized HCC cells to chemotherapeutic agents, such as cisplatin. Mechanistically, we discovered that blockage of CES1 caused TG accumulation and, hence, induced the rewiring of lipid metabolism, which eventually led to impaired mitochondrial function. Moreover, we identified the CES1-PPARα/γ-SCD axis as a key modulator. Specifically, lack of unique FFAs — especially multiple species of PUFAs, due to blockage of CES1 — suppressed the transcriptional activity of PPARα/γ on SCD expression, which ultimately increased the sensitivity of HCC to the treatment of cisplatin. Emerging evidence has demonstrated that dysregulation of dynamics of lipid droplets in nonadipocytes is closely related to tumor cell adaptability and progression [56]. In particular, a slower turnover rate of lipid droplets by targeting lipolysis has been demonstrated to be detrimental to tumorous cells, providing a potential therapeutic opportunity to treat cancer [56, 57]. In that context, we and others have recently revealed that CES1 targets lipid droplets and hydrolyzes their surface lipid contents [34, 37, 58]. In this study, we found that blockage of CES1 by WWL229 or genetic KO induced accumulation of more lipid droplets with larger sizes in HCC. Although the levels of other lipid droplet–surrounding factors, including ATGL, HSL/phosphorylated HSL, GCI-58, and CIDEA–C, did not significantly change, the de novo lipogenic enzyme ACC decreased slightly, while the ratio of phosphorylated ACC/total ACC increased upon blockage of CES1, suggesting that lipogenesis was also affected by inhibition of CES1 in HCC. Functional mitochondria are important for cell growth. Specifically, mitochondria control diverse vital parameters, such as the generation of energy through oxidative phosphorylation, regulation of ROS and oxidative stress, and initiation of apoptosis in aggressively growing cancerous cells [59]. Enhanced mitochondrial metabolism alters cell redox status and increases ROS generation, which further stimulates tumor cell proliferation. Particularly, mitochondrial FAO plays multifaceted roles in proliferation, survival, stemness, and chemoresistance of the cancerous cells [60]. Accelerated lipolysis on lipid droplets significantly increases the levels of the FAO and enhances the function of mitochondria to promote cancer cell growth, which has been highlighted as a “lipolytic phenotype” [16, 60]. Herein, we found that blockage of CES1 activity leads to formation of larger lipid droplets and slowed lipolysis, as evidenced by lower levels of FFAs and glycerol in WWL229-treated and CES1-KO HepG2 cells. Dysregulation of lipolysis further reduced the production of total FFAs and a transport form of fatty acids acylcarnitine, thereby impairing the mitochondrial energy production function, as demonstrated by the significantly lower OCRs during the Seahorse assay. Moreover, the expression levels of FAO-related enzymes were also downregulated, probably due to the negative feedback from the lack of FFAs in the mitochondria. The metabolic rewiring in the mitochondria bears a potential to sensitize cells to chemotherapeutic drugs. Given the fact that CES1 inhibition also impairs the mitochondrial function in normal cells [34, 61], caution should be taken when considering the clinical implication of CES1 inhibitors in future. Tremendous studies have demonstrated that the lipogenic factor SCD is involved in cancer cell proliferation and metastasis [9]. SCD levels are positively correlated with cancer aggressiveness and chemoresistance [9]. Mechanistically, its direct products, such as numerous MUFAs and PUFAs, exert their impact on tumorigenesis via enhanced phosphorylation of AKT and decreased ER stress [9, 62]. Targeting SCD activity by its specific inhibitors results in tumor-specific apoptosis [63]. In addition, identification of novel signaling pathways that downregulate its expression might provide alternative strategies to treat or prevent SCD-associated malignant disease. In this context, we reported that blockage of CES1, either pharmacologically or genetically, reduced SCD levels in HCC. More importantly, our analysis on clinical data revealed a strong correlation between the levels of CES1 and SCD in patients with liver cancer. Lipidomic profiles indicated that blockage of CES1 led to decreased levels of numerous MUFAs and PUFA (n > 2 double bonds), which was consistent with the reduced levels of SCD. Intriguingly, not all species of MUFA were decreased in response to the inhibition of CES1, reflecting the complexity of the regulation of lipid synthesis and metabolism at multiple levels. We further detected increased ROS generation and ER stress — results that are also in agreement with the reduced levels of SCD in CES1-blocked HepG2 cells. Importantly, treatment with the SCD-specific inhibitor MF438 efficiently blocked the CES1 effect on HCC apoptosis, further highlighting the key role of SCD in CES1-mediated cancer cell growth. We further investigated the mechanisms underlying the regulation of SCD by CES1. Previous studies reported that SCD is a direct target of PPARα/γ [43, 44, 64, 65]. In this study, targeting PPARα/γ molecules by their specific siRNAs demonstrated their key role in CES1-mediated SCD regulation. To address how CES1 manipulates the transcriptional activation of PPARα/γ, we performed LC-MS/MS analysis on WWL229-treated and CES1-KO HepG2 cells and identified multiple PUFAs that were diminished in the cells. These PUFAs have been demonstrated to be the endogenous ligands for PPARα/γ activation [44]. Interestingly, another liver-specific nuclear receptor, HNF4, has also been reported to regulate SCD expression [66]. Indeed, we recently revealed that HNF4 is involved in the development of liver steatosis caused by loss of function of Ces1 in diet-induced obese mice [35]. However, in this study, we did not find evidence that supported the involvement of HNF4 in CES1-mediated regulation of SCD in HCC. The difference of HNF4 functions in normal hepatocytes and HCC demonstrate its cell type–specific regulation of lipid signaling. Notably, PUFAs have been demonstrated to regulate SCD at both the expression and protein levels [67], suggesting that, in addition to the regulation of expression through PPAR α/γ, decreased PUFAs might also affect the levels/activity of SCD via other profound mechanisms in HCC. Of note, in addition to the identified PPARα/γ regulation, other lipogenesis pathways, such as SREBP$\frac{1}{2}$ signaling, might be also involved in the regulation of SCD mediated by CES1. Further studies are needed to test these pathways. Unexpectedly, even though blocking CES1 activity impairs mitochondrial function and reduces SCD levels, both of which might vitally affect the growth and proliferation of tumor cells, we only observed a mild effect of cell apoptosis in WWL229-treated and CES1-KO HepG2 cells. However, when we combined the approaches of CES1 blockage and administration of the anticancer agent cisplatin, we detected a synergistic effect of cell apoptosis and tumor inhibition in HCC. Furthermore, the combination treatment significantly reduced HCC xenograft tumors in NU/J mice. Our findings are of clinical significance. As we know, emerging evidence has demonstrated the chemotherapeutic potential of cisplatin in the treatment of patients with advanced HCC. Unfortunately, despite a certain level of therapeutic efficacy, significant numbers of HCC patients have experienced insensitivity or resistance to cisplatin administration, eventually leading to therapeutic failure [7]. Rewiring of lipid metabolism has been recognized as a major cause of chemoresistance in HCC to cisplatin. Indeed, a better understanding of the key role of lipid metabolism in HCC has changed the concept about the cancer from a “genetic disease” to a “metabolic disease” [68]. To support this notion, several lipid metabolic enzymes, including SCD, ACSS2, ACC$\frac{1}{2}$, and alkylglyceronephosphate synthase (AGPS), have been reported to be directly involved in cisplatin resistance [69, 70]. Moreover, inhibition of the lipid synthesis enzyme fatty acid synthetase (FASN) efficiently reversed cisplatin resistance in cancer cells [71]. In our study, we found that blockage of CES1 led to alterations in cisplatin resistance–related factors, including PUFAs, SCD, and ACC1, thereby interfering with lipid metabolism and sensitizing HCC to cisplatin treatment. Our findings, thus, provide a strategy to deal with the chemoresistance of HCC in clinic. Further mechanistic studies are warranted to elucidate the association between lipid metabolism and DNA damage and repair induced by cisplatin in HCC. Importantly, our analysis of the clinical database further suggested that CES1 levels tended to have a negative association with the survival rate of sorafenib-treated HCC, suggesting that targeting CES1 might have the potential to sensitize HCC to a broad range of chemotherapeutic agents. In conclusion, HCC has been demonstrated to be naturally or adaptively resistant to chemotherapeutic agents, including cisplatin, thereby leading to uncontrolled tumor growth and metastasis during chemotherapy. Our findings demonstrate that targeting the CES1-PPARα/γ-SCD axis may sensitize HCC tumors to cisplatin and other anti-HCC drugs. Therefore, interfering with lipid metabolism by blocking CES1 activity has great potential for the treatment of HCC. ## Methods Supplemental Methods are available online with this article. ## Analysis of CES1 expression in human samples. CES1 expression analysis in normal human and tumor tissues was performed using TNMplot (https://tnmplot.com/analysis/). The significant difference between the normal and tumor tissues was analyzed using the Mann-Whitney U test, which was conducted using the web tool. Correlation analysis of CES1 expression and overall survival in all types of cancer was conducted using the tool in UCSC Xena browser (http://xena.ucsc.edu/) with the TCGA Pan-Cancer databases. Correlation analyses of CES1 expression and survival in different cancer types were performed using KM plotter (https://kmplot.com/analysis/). For the analysis, the cutoff values to define the levels with “low” or “high” are the lower and upper quartiles of the CES1 expression, respectively. ## Tissue array and protein level analysis. Liver carcinoma and normal tissue arrays were obtained from the US Biomax (no. BC03119b). The protein levels were analyzed by immunofluorescence staining with an anti-CES1 antibody. Briefly, deparaffinized slides containing the tissue arrays were permeabilized with $0.2\%$ Triton X-100 in 1× PBS for 10 minutes and incubated with sodium citrate buffer at 95°C for 30 minutes for antigen retrieval. After blocking with $5\%$ BSA for 1 hour, the slides were incubated with primary antibodies at 4°C overnight. The slides were then washed with 1× PBST 3 times ($0.1\%$ Tween-20 in PBS) and further incubated with Alexa Fluor 488–conjugated donkey anti–rabbit IgG (catalog 711-545-152, Jackson ImmunoResearch) at room temperature for 1 hour. After incubation, the slides were washed with 1× PBST for 3 times and mounted. The slides were imaged using a Cytation 5 imaging reader. Anti-CES1 antibody (catalog HPA012023, Sigma-Aldrich) and anti-SCD antibody (catalog HPA012107, Sigma-Aldrich) were used for immunofluorescence staining. ## Animals and in vivo xenograft model. NU/J mice (no. 002019) were purchased from The Jackson Laboratory and housed in an animal facility with a 12-hour light/dark cycle at room temperature (22°C ± 1°C). The animals had free access to water and regular chow diet. When mice were 10 weeks old, 5 × 106 HepG2 cells (WT and CES1 KO) were s.c. injected into the right flanks of the mice. Tumor volumes were measured with a caliper 3 times per week and calculated using the formulation of 0.5 × length × width2. When the volumes reached approximately 100 mm3, WWL229 (155 μmol/kg body weight) and cisplatin (10 μmol/kg body weight) were i.p. administered to the mice 3 times a week for a total of 2 weeks. Appropriate vehicles ($1\%$ dimethyl sulfoxide, $24\%$ polyethylene glycol 400, and $6\%$ Tween-80 in PBS for WWL229 cells and PBS for cisplatin cells) were administered to the control mice. After 16 days of drug treatment, the animals were sacrificed and the tumor tissues were collected for further analysis. ## Statistics. All data are presented as the mean ± SEM or mean ± SD. All statistical analyses were performed using GraphPad Prism 8. The unpaired 2-tailed Student’s t test or 2-tailed Mann-Whitney U test was used to compare the differences between the 2 groups in the meta-analysis. One-way ANOVA was used to compare the differences among multiple experimental groups. The Dunnett T3 test was applied for the post hoc test. There is a correction for multiple comparisons using statistical hypothesis testing. Two-way ANOVA followed by Tukey multiple-comparison test was used to for the tumor volume comparison. Pearson’s correlation was used to analyze the relationship between CES1 and SCD protein levels in the tissue array samples. Statistical significance was set at $P \leq 0.05.$ For the lipidomic analysis, raw peak intensity was represented by normalized Z scores, and pairwise P values were calculated using 2-tailed Student’s t test and 2-tailed unequal variations to convey the significant abundance in the treated groups compared with the controls. ## Study approvals. The protocol for the animal experiments was reviewed and approved by the Animal Welfare Committee of the University of Texas Health Science Center at Houston (Animal protocol no. AWC-21-0019). ## Author contributions KS, PLL, and RL conceptualized the research; KS, GL, XL, PLL, KLEM, and RL designed research studies; GL, XL, JY, BF, SW, and BW conducted experiments; GL, XL, IM, and BW acquired data; KS, GL, XL, IM, KLEM, and PLL analyzed data; KS, GL, and XL wrote the manuscript; and GL, XL, IM, JY, BF, BW, KLEM, PLL, and RL edited the manuscript. ## 12/06/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Sung H. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin (2021.0) 71 209-249. PMID: 33538338 2. Parkin DM. Global cancer statistics, 2002. CA Cancer J Clin (2005.0) 55 74-108. PMID: 15761078 3. Sherman M. Recurrence of hepatocellular carcinoma. N Engl J Med (2008.0) 359 2045-2047. PMID: 18923166 4. Ishikawa T. Future perspectives on the treatment of hepatocellular carcinoma with cisplatin. World J Hepatol (2009.0) 1 8-16. PMID: 21160960 5. Stordal B. A systematic review of platinum and taxane resistance from bench to clinic: an inverse relationship. Cancer Treat Rev (2007.0) 33 688-703. PMID: 17881133 6. Park J. Inhibition of endotrophin, a cleavage product of collagen VI, confers cisplatin sensitivity to tumours. EMBO Mol Med (2013.0) 5 935-948. PMID: 23629957 7. Wang L. The role of tumour metabolism in cisplatin resistance. Front Mol Biosci (2021.0) 8. PMID: 34250022 8. Furuse J. Systemic therapy for hepatocellular carcinoma: current status and future perspectives. Jpn J Clin Oncol (2021.0) 51 1363-1371. PMID: 34258616 9. Tracz-Gaszewska Z, Dobrzyn P. Stearoyl-CoA desaturase 1 as a therapeutic target for the treatment of cancer. Cancers (Basel) (2019.0) 11. PMID: 31284458 10. Budd J, Cusi K. Nonalcoholic fatty liver disease: what does the primary care physician need to know?. Am J Med (2020.0) 133 536-543. PMID: 32017891 11. Baffy G. Hepatocellular carcinoma in non-alcoholic fatty liver disease: an emerging menace. J Hepatol (2012.0) 56 1384-1391. PMID: 22326465 12. Kim M. The impact of endotrophin on the progression of chronic liver disease. Exp Mol Med (2020.0) 52 1766-1776. PMID: 33110211 13. Budd J, Cusi K. Role of agents for the treatment of diabetes in the management of nonalcoholic fatty liver disease. Curr Diab Rep (2020.0) 20. PMID: 33015726 14. Rui L, Lin JD. Reprogramming of hepatic metabolism and microenvironment in nonalcoholic steatohepatitis. Annu Rev Nutr (2022.0) 42 91-113. PMID: 35584814 15. Sangineto M. Lipid metabolism in development and progression of hepatocellular carcinoma. Cancers (Basel) (2020.0) 12. PMID: 32486341 16. Kalavalapalli S. Impact of exenatide on mitochondrial lipid metabolism in mice with nonalcoholic steatohepatitis. J Endocrinol (2019.0) 241 293-305. PMID: 31082799 17. Baenke F. Hooked on fat: the role of lipid synthesis in cancer metabolism and tumour development. Dis Model Mech (2013.0) 6 1353-1363. PMID: 24203995 18. Wang B, Tontonoz P. Liver X receptors in lipid signalling and membrane homeostasis. Nat Rev Endocrinol (2018.0) 14 452-463. PMID: 29904174 19. Alannan M. Targeting lipid metabolism in liver cancer. Biochemistry (2020.0) 59 3951-3964. PMID: 32930581 20. Wang S. Cannabinoid receptor 1 signaling in hepatocytes and stellate cells does not contribute to NAFLD. J Clin Invest (2021.0) 131. PMID: 34499619 21. Wang B, Tontonoz P. Phospholipid remodeling in physiology and disease. Annu Rev Physiol (2019.0) 81 165-188. PMID: 30379616 22. Olzmann JA, Carvalho P. Dynamics and functions of lipid droplets. Nat Rev Mol Cell Biol (2019.0) 20 137-155. PMID: 30523332 23. Cruz ALS. Lipid droplets: platforms with multiple functions in cancer hallmarks. Cell Death Dis (2020.0) 11. PMID: 32029741 24. Bansal S. Stearoyl-CoA desaturase plays an important role in proliferation and chemoresistance in human hepatocellular carcinoma. J Surg Res (2014.0) 186 29-38. PMID: 24135379 25. Cheng C. Lipid metabolism reprogramming and its potential targets in cancer. Cancer Commun (Lond) (2018.0) 38. PMID: 29784041 26. Flowers MT, Ntambi JM. Role of stearoyl-coenzyme A desaturase in regulating lipid metabolism. Curr Opin Lipidol (2008.0) 19 248-256. PMID: 18460915 27. Vivas-Garcia Y. Lineage-restricted regulation of SCD and fatty acid saturation by MITF controls melanoma phenotypic plasticity. Mol Cell (2020.0) 77 120-137. PMID: 31733993 28. Oatman N. Mechanisms of stearoyl CoA desaturase inhibitor sensitivity and acquired resistance in cancer. Sci Adv (2021.0) 7. PMID: 33568479 29. Kikuchi K, Tsukamoto H. Stearoyl-CoA desaturase and tumorigenesis. Chem Biol Interact (2020.0) 316. PMID: 31838050 30. Koundouros N, Poulogiannis G. Reprogramming of fatty acid metabolism in cancer. Br J Cancer (2020.0) 122 4-22. PMID: 31819192 31. Theodoropoulos PC. Discovery of tumor-specific irreversible inhibitors of stearoyl CoA desaturase. Nat Chem Biol (2016.0) 12 218-225. PMID: 26829472 32. Holmes RS. Recommended nomenclature for five mammalian carboxylesterase gene families: human, mouse, and rat genes and proteins. Mamm Genome (2010.0) 21 427-441. PMID: 20931200 33. Lian J. Carboxylesterases in lipid metabolism: from mouse to human. Protein Cell (2018.0) 9 178-195. PMID: 28677105 34. Yang L. A unique role of carboxylesterase 3 (Ces3) in β-adrenergic signaling-stimulated thermogenesis. Diabetes (2019.0) 68 1178-1196. PMID: 30862682 35. Li G. Adipose tissue-specific ablation of Ces1d causes metabolic dysregulation in mice. Life Sci Alliance (2022.0) 5. PMID: 35459739 36. Lian J. Ces1d deficiency protects against high-sucrose diet-induced hepatic triacylglycerol accumulation. J Lipid Res (2019.0) 60 880-891. PMID: 30737251 37. Lehner R, Vance DE. Cloning and expression of a cDNA encoding a hepatic microsomal lipase that mobilizes stored triacylglycerol. Biochem J (1999.0) 343 Pt 1 1-10. PMID: 10493905 38. Na K. Potential regulatory role of human-carboxylesterase-1 glycosylation in liver cancer cell growth. J Proteome Res (2020.0) 19 4867-4883. PMID: 33206527 39. Na K. Human plasma carboxylesterase 1, a novel serologic biomarker candidate for hepatocellular carcinoma. Proteomics (2009.0) 9 3989-3999. PMID: 19658107 40. Na K. Human liver carboxylesterase 1 outperforms alpha-fetoprotein as biomarker to discriminate hepatocellular carcinoma from other liver diseases in Korean patients. Int J Cancer (2013.0) 133 408-415. PMID: 23319432 41. Dominguez E. Integrated phenotypic and activity-based profiling links Ces3 to obesity and diabetes. Nat Chem Biol (2014.0) 10 113-121. PMID: 24362705 42. Lanczky A, Gyorffy B. Web-based survival analysis tool tailored for medical research (KMplot): development and implementation. J Med Internet Res (2021.0) 23. PMID: 34309564 43. Miller CW, Ntambi JM. Peroxisome proliferators induce mouse liver stearoyl-CoA desaturase 1 gene expression. Proc Natl Acad Sci U S A (1996.0) 93 9443-9448. PMID: 8790349 44. Pawlak M. Molecular mechanism of PPARα action and its impact on lipid metabolism, inflammation and fibrosis in non-alcoholic fatty liver disease. J Hepatol (2015.0) 62 720-733. PMID: 25450203 45. Shrestha R. Exploring the therapeutic potential of mitochondrial uncouplers in cancer. Mol Metab (2021.0) 51. PMID: 33781939 46. Wang MH. Exosome-mediated delivery of SCD-1 siRNA promoted the death of anaplastic thyroid carcinoma cells via regulating ROS level. Clin Transl Oncol (2022.0) 24 288-296. PMID: 34287816 47. Malhotra JD, Kaufman RJ. ER stress and its functional link to mitochondria: role in cell survival and death. Cold Spring Harb Perspect Biol (2011.0) 3. PMID: 21813400 48. Aljohani A. Hepatic stearoyl-CoA desaturase-1 deficiency-mediated activation of mTORC1- PGC-1α axis regulates ER stress during high-carbohydrate feeding. Sci Rep (2019.0) 9. PMID: 31673045 49. Capece D. Enhanced triacylglycerol catabolism by carboxylesterase 1 promotes aggressive colorectal carcinoma. J Clin Invest (2021.0) 131. PMID: 33878036 50. Llovet JM. Sorafenib in advanced hepatocellular carcinoma. N Engl J Med (2008.0) 359 378-390. PMID: 18650514 51. Roberts LR. Sorafenib in liver cancer--just the beginning. N Engl J Med (2008.0) 359 420-422. PMID: 18650519 52. Snaebjornsson MT. Greasing the wheels of the cancer machine: the role of lipid metabolism in cancer. Cell Metab (2020.0) 31 62-76. PMID: 31813823 53. Jones RD. Carboxylesterases are uniquely expressed among tissues and regulated by nuclear hormone receptors in the mouse. Drug Metab Dispos (2013.0) 41 40-49. PMID: 23011759 54. Megger DA. Proteomic differences between hepatocellular carcinoma and nontumorous liver tissue investigated by a combined gel-based and label-free quantitative proteomics study. Mol Cell Proteomics (2013.0) 12 2006-2020. PMID: 23462207 55. Mizuno H. Heterogeneous nuclear ribonucleoprotein A2/B1 in association with hTERT is a potential biomarker for hepatocellular carcinoma. Liver Int (2012.0) 32 1146-1155. PMID: 22372738 56. Petan T. Lipid droplets in cancer. Rev Physiol Biochem Pharmacol. DOI: 10.1007/112_2020_51 57. Huang J. Tumor-induced hyperlipidemia contributes to tumor growth. Cell Rep (2016.0) 15 336-348. PMID: 27050512 58. Soni KG. Carboxylesterase 3 (EC 3.1.1.1) is a major adipocyte lipase. J Biol Chem (2004.0) 279 40683-40689. PMID: 15220344 59. Wallace DC. Mitochondria and cancer. Nat Rev Cancer (2012.0) 12 685-698. PMID: 23001348 60. Ma Y. Fatty acid oxidation: an emerging facet of metabolic transformation in cancer. Cancer Lett (2018.0) 435 92-100. PMID: 30102953 61. Lian J. Genetic variation in human carboxylesterase CES1 confers resistance to hepatic steatosis. Biochim Biophys Acta Mol Cell Biol Lipids (2018.0) 1863 688-699. PMID: 29631096 62. Rogowski MP. SCD1 activity in muscle increases triglyceride PUFA content, exercise capacity, and PPARδ expression in mice. J Lipid Res (2013.0) 54 2636-2646. PMID: 23918045 63. Yu Y. Targeting a lipid desaturation enzyme, SCD1, selectively eliminates colon cancer stem cells through the suppression of Wnt and NOTCH signaling. Cells (2021.0) 10. PMID: 33430034 64. Hebbachi AM. Peroxisome proliferator-activated receptor alpha deficiency abolishes the response of lipogenic gene expression to re-feeding: restoration of the normal response by activation of liver X receptor alpha. J Biol Chem (2008.0) 283 4866-4876. PMID: 18079124 65. Ikeda J. PPARγ agonists attenuate palmitate-induced ER stress through up-regulation of SCD-1 in macrophages. PLoS One (2015.0) 10. PMID: 26061913 66. Odom DT. Control of pancreas and liver gene expression by HNF transcription factors. Science (2004.0) 303 1378-1381. PMID: 14988562 67. Ntambi JM. Regulation of stearoyl-CoA desaturase by polyunsaturated fatty acids and cholesterol. J Lipid Res (1999.0) 40 1549-1558. PMID: 10484602 68. Wishart DS. Is cancer a genetic disease or a metabolic disease?. EBioMedicine (2015.0) 2 478-479. PMID: 26288805 69. Jin L. MAST1 drives cisplatin resistance in human cancers by rewiring cRaf-independent MEK activation. Cancer Cell (2018.0) 34 315-330. PMID: 30033091 70. Zhu Y. Alkylglyceronephosphate synthase (AGPS) alters lipid signaling pathways and supports chemotherapy resistance of glioma and hepatic carcinoma cell lines. Asian Pac J Cancer Prev (2014.0) 15 3219-3226. PMID: 24815474 71. Bauerschlag DO. Fatty acid synthase overexpression: target for therapy and reversal of chemoresistance in ovarian cancer. J Transl Med (2015.0) 13. PMID: 25947066
--- title: LRP1 protects against excessive superior mesenteric artery remodeling by modulating angiotensin II–mediated signaling authors: - Jackie M. Zhang - Dianaly T. Au - Hisashi Sawada - Michael K. Franklin - Jessica J. Moorleghen - Deborah A. Howatt - Pengjun Wang - Brittany O. Aicher - Brian Hampton - Mary Migliorini - Fenge Ni - Adam E. Mullick - Mashhood M. Wani - Areck A. Ucuzian - Hong S. Lu - Selen C. Muratoglu - Alan Daugherty - Dudley K. Strickland journal: JCI Insight year: 2023 pmcid: PMC9977308 doi: 10.1172/jci.insight.164751 license: CC BY 4.0 --- # LRP1 protects against excessive superior mesenteric artery remodeling by modulating angiotensin II–mediated signaling ## Abstract Vascular smooth muscle cells (vSMCs) exert a critical role in sensing and maintaining vascular integrity. These cells abundantly express the low-density lipoprotein receptor–related protein 1 (LRP1), a large endocytic signaling receptor that recognizes numerous ligands, including apolipoprotein E–rich lipoproteins, proteases, and protease-inhibitor complexes. We observed the spontaneous formation of aneurysms in the superior mesenteric artery (SMA) of both male and female mice in which LRP1 was genetically deleted in vSMCs (smLRP1–/– mice). Quantitative proteomics revealed elevated abundance of several proteins in smLRP1–/– mice that are known to be induced by angiotensin II–mediated (AngII-mediated) signaling, suggesting that this pathway was dysregulated. Administration of losartan, an AngII type I receptor antagonist, or an angiotensinogen antisense oligonucleotide to reduce plasma angiotensinogen concentrations restored the normal SMA phenotype in smLRP1–/– mice and prevented aneurysm formation. Additionally, using a vascular injury model, we noted excessive vascular remodeling and neointima formation in smLRP1–/– mice that was restored by losartan administration. Together, these findings reveal that LRP1 regulates vascular integrity and remodeling of the SMA by attenuating excessive AngII-mediated signaling. ## Introduction Maintaining vascular integrity is essential for normal physiological function. Loss of integrity leads to formation of aortic aneurysms, which dilate abnormally and may eventually rupture, resulting in life-threatening events. Thoracic aortic aneurysms and dissections typically afflict the young and often result from underlying specific gene mutations [1, 2]. In contrast, abdominal aortic aneurysms classically affect older men with significant comorbidities. To date, no single genetic determinant has been identified that is sufficient to cause abdominal aortic aneurysms. Aneurysms are defined clinically as a permanent focal dilation and can occur in vessels outside of the aorta, such as the splanchnic arteries. This arterial bed includes the splenic, celiac, hepatic, superior mesenteric, and inferior mesenteric arteries. Splanchnic artery aneurysms occur with an estimated incidence of $0.1\%$–$2\%$ of the adult population [3], and currently, very little is known about the mechanism associated with their development. It is likely that aneurysms in these vessels occur via similar mechanisms that have been associated with aortic aneurysms. Recent studies in mice have identified a critical role for the low-density lipoprotein receptor–related protein 1 (LRP1) in protecting against aortic aneurysms (4–8). LRP1 is a large endocytic signaling receptor that contributes to vascular development [9], exerts a role in lipoprotein metabolism [10, 11], regulates protease concentrations [5, 12, 13], regulates inflammation [14], and attenuates the progression of atherosclerosis and aneurysm formation (4–8). Genetic studies in humans have revealed that the LRP1 gene is a susceptibility locus for aortic aneurysms and dissections (15–21). The mechanisms by which LRP1 protects the vasculature are not fully understood but may involve regulating platelet-derived growth factor receptor (PDGFR) activation (4, 22–24), modulating transforming growth factor-β (TGF-β) and connective tissue growth factor (CTGF) signaling (5, 7, 25–28), and regulating protease concentrations in the vessel wall [5]. Additionally, LRP1 regulates vascular smooth muscle cell (vSMC) contraction [6] and proliferation [6, 29]. In mice, genetic deletion of LRP1 in vSMCs (smLRP1–/–) results in development of spontaneous thoracic aneurysms, as well as abnormal medial wall thickening and degradation and fragmentation of the elastic laminae [5]. Interestingly, chronic infusion of angiotensin II (AngII) into smLRP1–/– mice results in pronounced superior mesenteric artery (SMA) medial thickening, neointimal formation, elastic fragmentation, a dramatically exacerbated dilatation, and a high rate of rupture [30]. AngII-mediated signaling has been studied frequently in the cardiovascular field as the renin-angiotensin system exerts a key role in regulating systemic vascular resistance and maintaining arterial structure [31, 32]. Angiotensinogen (AGT), a member of the serine protease inhibitor family, is predominantly secreted from the liver [33, 34] and cleaved by renin to produce angiotensin I, which is subsequently cleaved in a reaction catalyzed by angiotensin-converting enzyme to produce AngII. AngII signals via 2 receptors, AngII receptor type 1 (AGTR1) and AngII receptor type 2 (AGTR2), and increases neointimal hyperplasia development in response to vascular injury (35–38). Moreover, the renin-angiotensin system also contributes to development and progression of aortic aneurysms and dissections (39–42). Experimentally, AngII infusion into mice is used widely as a model for aortic aneurysms and dissections [43]. AngII-mediated signaling also upregulates LRP1 in vSMCs isolated from rat aorta [44] and upregulates several LRP1 ligands, including plasminogen activator inhibitor 1 [45], protease nexin 2 (serpine2) [46], TGF-β [47], CTGF [48], and matrix metalloproteinase 2 (MMP-2) [49]. We noted spontaneous and fully penetrant formation of SMA aneurysms in both male and female smLRP1–/– mice. The objective of the current investigation was to define molecular mechanisms by which LRP1 protects against aneurysm formation in this vessel bed. We also used a vascular injury model dependent upon AngII-mediated signaling to study the contribution of LRP1 in this process. Our results revealed that LRP1 maintains vascular wall integrity and regulates vascular remodeling in these arteries by attenuating AngII-mediated signaling. ## Spontaneous dilation and remodeling of the SMA in smLRP1–/– mice. Our prior studies have confirmed effective deletion of LRP1 from smooth muscle cells (SMCs) in smLRP1–/– mice [5, 30]. Micro-CT imaging of vasculature of LRP1+/+ (Figure 1, A and B) and smLRP1–/– mice (Figure 1C) revealed extensive dilatation of the SMAs in smLRP1–/– mice. Histological analyses of SMAs from LRP1+/+ (Figure 2A) and smLRP1–/– mice (Figure 2B) at 16 weeks of age revealed profound degradation of elastic laminae in smLRP1–/– mice. Morphometric measurements confirmed that remarkable thickening of the media (Figure 2C) and adventitia (Figure 2D) occurred as the mice aged. There was no notable neointima formation in the SMAs of smLRP1–/– mice. Ex vivo measurements using micro-CT imaging of the SMA lumen diameter revealed that the SMA lumen diameter increased significantly in smLRP1–/– mice at 24 and 64 weeks of age (Figure 2E). This was observed in both male and female smLRP1–/– mice (Figure 2F). There was no noticeable difference in the SMA lumen diameter between sexes regardless of genotype (Figure 2F). Upon ultrasound measurements of maximal lumen diameters of the SMA, at 20 weeks of age and beyond, enhanced vessel dilatation of the SMA upon was observed in smLRP1–/– when compared with LRP1+/+ mice (Figure 3A). The rate of SMA dilatation, when measured at 20 weeks (as baseline) over a subsequent 20-week period, revealed a 2-fold increase in the rate in smLRP1–/– when compared with LRP1+/+ mice (Figure 3B). ## Global proteomic analyses reveal activation of the AngII and TGF-β signaling pathways in SMAs of smLRP1–/– mice. To identify potential mechanisms by which LRP1 regulates vascular remodeling, we used quantitative proteomic analysis to characterize the molecular signatures that may be integral in contributing to the phenotype observed in the SMAs of smLRP1–/– mice. Principal component analysis revealed distinct clusters for LRP1+/+ versus smLRP1–/– proteomes (Figure 4A). Using a fold-change value of 2 and FDR < 0.01, proteomic analyses identified 2,465 total proteins, of which 809 were significantly altered in smLRP1–/– SMAs when compared with LRP1+/+ mice (Figure 4B). Mass spectrometry data supported a greater than 8-fold decrease in LRP1 abundance in the SMAs of smLRP1–/– mice (Figure 4C). Gene ontology enrichment analysis of upregulated proteins revealed major clusters in categories associated with extracellular matrix organization, actin filament organization, and collagen fibril organization (Figure 4D, left panel), while gene ontology analysis of downregulated proteins revealed major clusters in energy metabolism and membrane organization (Figure 4D, right panel). The intensity level of LRP1 ligands derived from the mass spectral analysis demonstrated that many LRP1 ligands were elevated in SMA tissue of smLRP1–/– mice relative to LRP1+/+ mice (Figure 4E). We also noted from the mass spectral analysis that integrin subunits and proteins involved in integrin function were downregulated in the SMAs of smLRP1–/– mice (Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.164751DS1). These include the α subunit of several integrins as well as talin 1 and kindlin-2, both of which interact with integrin cytoplasmic tails, leading to integrin activation [50]. Together, these observations imply that a major defect in the SMAs of smLRP1–/– mice involves integrin/matrix interactions, which are critical for normal vSMC function [51]. Casual analysis of the proteomic data [52] in Ingenuity Pathway Analysis (IPA) software revealed a high probability for the activation of a number of pathways known to influence vascular remodeling (Figure 4F) with a high degree of significance (Figure 4G). These pathways included the AngII-mediated (activation z score of 3.1; P value of overlap = 7.9 × 10–17) and TGF-β1 signaling pathways (z score = 2.0; P value of overlap = 6.3 × 10–23), both of which are associated with vascular remodeling [43, 53]. Interestingly, these data also predicted that the SMAD7 pathway was inhibited (z score = –3.5; P value of overlap = 1.4 × 10–5). SMAD7 functions as an inhibitor of TGF-β signaling by associating with the E3 ubiquitin ligase SMURF2, which triggers degradation of the TGF-β receptor type 1 [54, 55]. ## Global proteomic analyses predict inhibition of transcriptional programing that regulates SMC differentiation. Of additional interest, our proteomic data predicted that myocardin-mediated signaling was inhibited in the SMAs of smLRP1–/– mice (z score = –2.8; P value of overlap = 8.9 × 10–4). Myocardin is a nuclear protein expressed in SMCs that plays a crucial role in differentiation of SMCs [56]. The intensity levels of proteins derived from the mass spectral analysis that are regulated by myocardin are shown in Figure 4H and include several proteins, such as Myh11, which is associated with a mature contractile SMC phenotype. Interestingly, missense mutations in the MYH11 gene are associated with thoracic aortic aneurysms [57]. ## Global proteomic analyses predict deregulation of proteinases and proteinase inhibitors in the SMA. Proteomic data comparing the SMAs from smLRP1–/– mice with WT mice revealed that several proteases and protease inhibitors were dysregulated (Supplemental Figure 1, A and B). These included members of the ADAMTS family, MMPs, coagulation proteases, as well as members of the cathepsin family of proteinases. Interestingly, SERPINC1 (antithrombin III) was decreased in smLRP1–/– mice, suggesting excessive thrombosis in smLRP1–/– mice. ## Inhibition of AngII-mediated signaling restores the SMA phenotype. Since our proteomic data were consistent with AngII-mediated signaling being activated in smLRP1–/– mice, we designed experiments to further investigate the role of the AngII signaling pathway on SMA dilatation. To accomplish this, we elected to pharmacologically block this pathway by administering losartan, an AGTR1 antagonist that selectively blocks the binding of AngII to AGTR1. During administration, we monitored systolic and diastolic blood pressure, which was reduced in both LRP1+/+ and smLRP1–/– mice comparably (Supplemental Figure 2). Histological analysis, as well as morphometric measurements, revealed that AGTR1 blockade with losartan reduced elastic laminae degradation (Figure 5, A–D) and medial thickening in these mice at 16 weeks of age (Figure 5E). In addition, there was a significant decrease in the SMA lumen diameter as measured from micro-CT imaging in both male (Figure 5F) and female (Figure 5G) smLRP1–/– mice. These data revealed that losartan prevented formation of SMA aneurysms in smLRP1–/– mice and support results obtained from our proteomic data, revealing that AngII signaling exerts a critical role in SMA pathology in smLRP1–/– mice. ## Reduction in plasma AGT concentrations restores the SMA phenotype. To assess the contribution of plasma-derived AGT on SMA remodeling in smLRP1–/– mice, we employed AGT antisense oligonucleotide (ASO) to reduce plasma AGT concentrations. This approach significantly reduced plasma AGT concentrations in both LRP1+/+ and smLRP1–/– mice (Figure 6A). Twelve weeks following ASO administration, ultrasound measurements confirmed that the lumen diameter of the SMAs in smLRP1–/– mice was indistinguishable from those of LRP1+/+ mice (Figure 6B). Further, ex vivo measurements of SMA vessel width (Figure 6C and see Supplemental Figure 3 for examples) also revealed no difference in AGT ASO–administered smLRP1–/– and LRP1+/+ mice (Figure 6C). These results support the contribution of AGT-mediated signaling to the SMA phenotype. AGT has been reported to bind directly to LRP1 [58], and thus we considered the possibility that LRP1 expressed in vSMCs might regulate plasma AGT concentrations by binding this ligand and mediating its internalization and degradation. An ELISA supported equal concentrations of both AGT and renin in plasma of LRP1+/+ and smLRP1–/– mice (Supplemental Figure 4, A and B), indicating that LRP1 deficiency in SMCs did not affect AGT concentrations in plasma. ## LRP1 expression attenuates vascular remodeling upon injury by regulating AngII-mediated signaling. To further test the hypothesis that LRP1 attenuates AngII-mediated signaling, we used an established vascular injury model [59] that is known to be mediated by AngII-mediated signaling [38]. Initially, we examined the carotid arteries of LRP1+/+ and smLRP1–/– mice. Like the SMAs, extensive vascular remodeling occurred in the carotid arteries of smLRP1–/– mice, which resulted in extensive degradation of the elastic laminae (Figure 7, A–C), an increase in the total areas of the adventitia and media (Figure 7D), and medial and adventitial thickening in smLRP1–/– mice (Figure 7E). These data support that LRP1 deficiency also regulated the integrity of this vascular bed. In the vascular injury model, LRP1+/+ and smLRP1–/– mice at 12–16 weeks of age were subjected to ligation of the left common carotid artery. Four weeks following surgery, mice were euthanized, and the whole neck and head were dissected from each animal. Histological analysis of whole-neck sections by H&E staining (Figure 8A), EVG staining (Figure 8B), and Masson’s trichrome staining (Figure 8C) showed extensive vascular remodeling in smLRP1–/– mice with significant neointima formation ($P \leq 0.0001$, Figure 8D) compared with LRP1+/+ mice. Morphometric measurements verified significant increases in the adventitia and neointima in the carotid arteries of smLRP1–/– mice upon injury (Figure 8, D and E). To evaluate the impact of LRP1 deletion on AngII-mediated signaling in this model, LRP1+/+ and smLRP1–/– mice were subjected to carotid ligation without or with losartan (0.6 g/L) in their drinking water. Histological analyses of sections stained with EVG (Figure 9A) and morphometric measurements of the vessels demonstrated that losartan administration ablated neointima formation in smLRP1–/– mice (Figure 9, B and C). These results support a major role for LRP1 in regulating vascular remodeling by attenuating excessive AngII-mediated signaling. ## Discussion We investigated the spontaneous formation of SMA pathology that occurred in both male and female smLRP1–/– mice. The intrinsic characteristics of LRP1-deficient vSMCs resulted in SMA vessel wall architectures that exhibited hallmark characteristics of a damaged vessel wall undergoing vascular remodeling [60, 61]. These characteristics include disorganization and fragmentation of elastic fibers, medial thickening due to increased matrix deposition, significant adventitial thickening (Figure 2D), and higher abundance of extracellular matrix–degrading proteases. Quantitative global proteomics revealed that vSMCs in the SMAs of LRP1–/– mice have a phenotype in which contractile genes are downregulated, while extracellular matrix proteins are upregulated. Further, in smLRP1–/– mice, several myocardin-regulated proteins were downregulated. This is of interest, as conditional deletion of the Mycod in vSMCs in mice results in arterial aneurysms, dissections, and rupture [56]. The findings suggest that LRP1 preserves vascular integrity, at least in part, by promoting myocardin-mediated signaling, which is important for maintaining the contractile function of vSMCs. In summary, our studies reveal that deletion of LRP1 in SMCs results in (a) defective SMC differentiation, (b) defective matrix-SMC interactions via integrins, (c) an injury response associated with upregulation of AngII-targeted and TGF-β–targeted genes, and (d) deregulation of numerous proteinases known to be capable of degrading the matrix. Upstream regulator analysis [52] of our proteomic results from the SMAs of smLRP1–/– mice revealed dysregulation of the AngII- and TGF-β–mediated signaling pathways. These signaling pathways are of interest, as their excessive activation can result in aneurysm formation [43, 62, 63]. To test the hypothesis that dysregulation of AngII-mediated signaling events were causative for the SMA phenotype in smLRP1–/– vessels, we used the AGTR1 antagonist losartan. Clinically, losartan is used commonly to treat hypertension. However, losartan has gained traction as a potential drug to attenuate vascular remodeling in various disease states and is indicated for left ventricular hypertrophy in hypertension patients and nephropathy in type 2 diabetes patients [64, 65]. Most recently, losartan has been investigated in long-term clinical trials in patients with Marfan syndrome to improve overall survival by means of preventing aortic dissection and reducing aortic root dilation [66]. In addition to lowering blood pressure, losartan antagonizes the TGF-β signaling pathway, presumably through an AngII-based mechanism. In mouse models of Marfan syndrome or Loeys-Dietz syndrome, losartan attenuates vascular remodeling, prevents aortic aneurysms, and improves vessel wall structure [62, 67]. The results of our studies revealed that losartan was highly effective in restoring SMA integrity by reducing lumen diameter, medial thickening, and degradation of the elastic laminae. Losartan administration prevented SMA aneurysm formation in both male and female smLRP1–/– mice. To verify and extend the results obtained by AGTR1 receptor blockade, we also used ASO-mediated AGT-knockdown experiments. Reducing plasma AGT concentrations also restored the phenotype in smLRP1–/– mice. Together, these data provide convincing evidence that a major mechanism by which LRP1 regulates SMA remodeling is via attenuation of AngII-mediated signaling. Previously, Davis et al. [ 30] revealed that chronic AngII infusion into smLRP1–/– mice resulted in disproportionate SMA pathology and death from mesenteric rupture compared with their LRP1+/+ counterparts, which is consistent with our proteomic analysis. However, in their study, they also found that chronic infusion of norepinephrine to promote similar increases in hemodynamic pressure comparable to AngII infusion also produced SMA aneurysms, confounding the relationship between angiotensin and the mechanism(s) by which losartan influences SMA pathology in smLRP1–/– mice. Interestingly, without AngII infusion, the systolic and diastolic blood pressure of 1-year-old smLRP1–/– mice is significantly lower than their LRP1+/+ littermates [5], and as shown here, they still develop SMA pathology. Therefore, we conclude that the effects of losartan on SMA pathology seen in smLRP1–/– mice may partially be dependent upon reduction of blood pressure but are also exacerbated by AngII-mediated signaling events that are not solely associated with its elevating hemodynamic effects. We further tested the potential of LRP1 to modulate AngII-mediated signaling by using a well-characterized model of vascular remodeling [59] known to depend upon AngII-mediated signaling [38]. Our results revealed significant neointima formation and adventitial thickening in smLRP1–/– mice when compared with LRP1+/+ mice (Figure 7D), supporting that LRP1 protects against injury-induced vascular remodeling. These results concur with those from Basford et al., who used an endothelial denudation model to induce vascular remodeling [29]. Importantly, our data revealed that losartan completely blocked the excessive neointima formation noted in the smLRP1–/– mice upon vascular injury. These data provide additional supporting evidence that LRP1 exerts a role in attenuating AngII-mediated signaling events. Together, these results show that LRP1 exerts a critical role in regulating AngII-mediated signaling events, and in the absence of LRP1, the SMA is spontaneously remodeled in a process that is prevented by AGTR1 blockade or reduction of plasma AGT concentrations. We propose that LRP1 prevents excessive remodeling of the SMA by regulating SMC phenotype and by attenuating AngII-mediated signaling. Our studies raise the possibility that mutations in LRP1 may result in receptor defects that contribute to SMA pathology in human patients. In humans, genome-wide association studies, exome sequencing, and TaqMan single nucleotide polymorphism (SNP) genotyping assays have identified an association of LRP1 SNPs with aortic aneurysms (15–17, 20), aortic dissections [18, 19], and Marfan syndrome [68]. Interestingly, aortic, but not plasma, concentrations of soluble forms of LRP1 were significantly lower in patients with abdominal aortic aneurysm (AAA) compared with controls [69]. Chan et al. [ 70] also reported a significant reduction in LRP1 protein abundance in human AAA samples from a Chinese population and have recently demonstrated that translational inhibition by microRNA-205 is responsible for driving the lower abundance of LRP1 [71]. Furthermore, lower levels of LRP1 were speculated to result in accumulation of excess MMP-9, a well-documented protease that contributes to degradation of the extracellular matrix proteins, leading to AAA [72]. Based on our studies in mice, we hypothesize that rare variants in LRP1 might contribute to SMA pathology in patients. The process by which LRP1 attenuates AngII-mediated signaling is likely to involve multiple mechanisms as LRP1 is known to regulate the abundance of several important signaling molecules as well as matrix molecules [73] and affects multiple signaling events. Since the blood pressure reductions induced by losartan were not different between LRP1+/+ mice and smLRP1–/– littermates, we conclude that LRP1 most likely affected downstream signaling events mediated by AGTR1. Several studies have demonstrated a relationship between AngII and TGF-β signaling in vascular tissue and remodeling [74] as AngII-mediated signaling increases the production of TGF-β [75, 76]. Thus, in transgenic mice expressing mutant forms of cardiac troponin T, the interstitial fibrosis that is driven by TGF-β signaling was attenuated with losartan [77]. The role of TGF-β in vascular remodeling has been well established (27, 78–83). Additionally, aneurysms in AngII-infused ApoE–/– mice have also been associated with the increased expression of TGF-β in whole-genome expression analysis [84], suggesting a possible synergic effect between TGF-β and AngII signaling. Further, excessive TGF-β signaling was detected in mouse models of Marfan syndrome, and a TGF-β neutralizing antibody, as well as losartan, partially reversed vascular manifestations of Marfan’s syndrome [62]. These studies support the notion that LRP1 may affect vascular remodeling, in part, by attenuating TGF-β signaling pathways. This is strengthened by the findings that LRP1 binds to all forms of TGF-β [25, 27] and that LRP1 expressed in macrophages attenuates TGF-β signaling upon vascular injury in mice fed a Western diet [27]. Further, liver-specific deletion of LPR1 in mice accelerates liver disease progression in mouse models by increasing sensitivity of profibrotic gene expression to promote steatohepatitis [85]. It is also possible that the LRP1-mediated effect could be independent of TGF-β signaling, as AngII appears capable of activating the Smad pathway independent of TGF-β signaling [86]. In addition, administration of a TGF-β neutralizing antibody in AngII-infused normocholesterolemic mice disrupts their resistance to aneurysm formation, implying a seemingly controversial protective effect of TGF-β instead [87]. Further, increased AngII-mediated and insulin-like growth factor 1–mediated signaling, independent of TGF-β signaling, is thought to drive a form of inherited nonsyndromic thoracic aortic aneurysms associated with missense mutations in the MYH11 gene [57]. Interestingly, our proteomic data revealed a 7-fold decrease in the protein levels of MYH11 in the SMAs of smLRP1–/– mice. Given the complexity of the multiple interactions of the AngII signaling pathway and the expansiveness of our proteomic analysis, additional studies are warranted to determine the role of LRP1 in these other signaling pathways and elucidate potential signaling crosstalk of LRP1 with the renin-angiotensin pathway. In summary, our studies have demonstrated a critical role for LRP1 in maintaining an appropriate vSMC phenotype and in attenuating excessive AngII-mediated signaling events in the SMA. Given that little is known about mechanisms associated with splanchnic artery aneurysms in humans, our studies raise the possibility that LRP1 may play a critical role in regulating the integrity of this vasculature in humans as well, and it will be important in future studies to determine if LRP1 missense mutations are associated with splanchnic artery aneurysms. ## Animals. All mice were weaned at 3–4 weeks of age, maintained on a 12-hour light/12-hour dark cycle, fed a standard laboratory rodent diet ($4\%$ wt/wt fat; Envigo 2018SX), and given standard drinking water ad libitum. Mice that received drugs were provided losartan (0.6 g/L) dissolved in drinking water or ASO via subcutaneous injections. Embryonic deletion of Lrp1 in vSMCs was achieved by crossing transgenic mice expressing Cre recombinase under the control of an SM22 SMC-specific promoter with mice expressing loxP sites flanking the Lrp1 gene (provided by J Herz, University of Texas Southwestern Medical Center, Dallas, Texas, USA). The resulting offspring, Lrp1fl/fl SM22-Cre–/– (LRP1+/+) and Lrp1fl/fl SM22-Cre+/– (smLRP1–/–), were used in experimental studies with LRP1+/+ littermates serving as controls. ## Ultrasonography. SMAs were scanned using a Vevo 3100 ultrasound system with an MS550 transducer (FUJIFILM VisualSonics Inc.). Mice were placed on a heated platform (37°C) to avoid hypothermia and anesthetized with isoflurane (1–$2\%$ vol/vol) to adjust the heart rate between 400 and 550 beats/minute. Color Doppler was used to confirm the pulsatile flow of the abdominal aorta. Then the probe was moved from the diaphragm caudally to visualize the SMA. A cine loop of the SMA was captured to define the maximum dilation of the SMA. Maximal luminal diameters were measured on the captured images using Vevo LAB 3.1.1 software (FUJIFILM VisualSonics Inc.). ## Microfil injection. Mice were euthanized by an overdose of ketamine and xylazine cocktail (90 and 10 mg/kg, respectively). The thoracic cavity was cut open, and the right atrium was nicked to allow the exit of blood flow. Saline (10 mL) was perfused through the left ventricle using a pressure-controlled peristatic pump (PS/200, Living Systems Instrumentation) at physiological pressure. Directly after perfusion, the right atrium was sealed, and Microfil (Flow Tech, Inc.) was injected through the same catheter at physiological pressure. Once Microfil was visualized in the arterioles surrounding the small intestine, the pump was stopped, the catheter was clamped shut to prevent backflow of Microfil into the thoracic cavity, and the animal was set aside to allow the compound to harden (~90 minutes). ## Micro-CT scanning and 3D reconstruction. After Microfil perfusion, animals were scanned using a Skyscan 1276 micro-CT (Bruker), and images were acquired with a pixel size of 20 μm at 2,016 × 1,344 resolution. CT scans were reconstructed using the NRecon program (Bruker) to adjust for beam hardening and ring artifacts. Image sets were saved as DICOM or BMP files (~1,200–1,500 images/animal). Reconstruction in 3D was performed using the 3D Slicer program. All bone and vasculatures not of interest were removed using the scissors tool within the program to display the aorta and its major branches. To visualize SMAs, all the other vasculatures were removed using the scissors tool. ## AGT ASO experiments. AGT ASOs were provided by Ionis Pharmaceuticals. PBS alone (control) or AGT ASO (80 mg/kg) was injected subcutaneously on days 1 and 4 in male LRP1+/+ and smLRP1–/– littermates when they were 6 weeks of age. Subsequently, either PBS or AGT ASO (40 mg/kg) was injected once every week for 11 weeks. At termination (18 weeks of age), plasma was collected to measure AGT concentrations. ## Plasma AGT and renin measurements. Plasma AGT concentrations were measured using a mouse AGT ELISA kit (ab245718; Abcam). Plasma renin concentrations were measured using an ELISA kit (IB59131, Immuno-Biological Laboratories Co., Ltd.) in which the angiotensin I product was determined after incubation of plasma with recombinant mouse AGT at 37°C for 1 hour. ## SMA tissue collection. SMA tissue was collected for proteomic and histological analysis. Mice that were designated for proteomic quantification were euthanized by CO2 asphyxiation, and the SMAs were collected. The adventitia and periadventitial fat from the SMAs were removed in cold PBS; then the SMA was immediately snap-frozen and stored at –80°C until analysis. For histological analysis, SMAs were collected after micro-CT images were acquired. Residual Microfil contrast reagent was removed from the SMA. The tissue was then fixed again in $4\%$ paraformaldehyde overnight, then placed in $70\%$ ethanol solution in preparation for decalcification, sectioning, and staining. Tissue cross sections of 5 μm thickness were sliced and stained by H&E or EVG. Morphometric measurements were performed using EVOS FL Auto Imaging System software (Invitrogen, Thermo Fisher Scientific) and ImageJ (NIH). All measurements were performed while blinded to the sample identification. ## Carotid artery ligation. Ligation of the left common carotid artery was performed on male LRP1+/+ and smLRP1–/– mice at 12–16 weeks of age. Mice were placed in an induction chamber and anesthetized with $3\%$ vaporized isoflurane (Fluriso; VetOne 502017) in oxygen flowing at 1 L/min. Sedated mice were laid supine on a heating pad and maintained on $2.5\%$ vaporized isoflurane in oxygen via nose cone. The neck area was administered a depilatory (Nair) to remove hair, disinfected with alternating $7.5\%$ povidone-iodine (Betadine Surgical Scrub; Purdue Pharma NDC 67618-151-16) and $70\%$ isopropyl alcohol (Webcol Alcohol Preps; Covidien 5033), and an incision was made from the sternum to the area just below the chin. The underlying fascia and glandular tissues were separated, and the exposed muscle layer was dissected carefully and retracted. The left common carotid artery was separated from the surrounding fascia and adjacent vagus nerve, and the isolated vessel was permanently ligated proximal to the carotid bifurcation using a sterilized 4-0 silk suture to fully obstruct blood flow. The incision was sutured using a 4-0 PDO absorbable monofilament suture (AD Surgical M-D430T17), and animals were weighed and administered 0.05 mg/kg buprenorphine hydrochloride (Buprenex; Reckitt Benckiser NDC 12496-0757-5) diluted in $0.9\%$ sodium chloride injection, USP (Hospira NDC 0409-4888-02), via subcutaneous injection before returning to a cage placed on a heating pad. Animals were monitored for recovery from anesthesia and ambulatory movements. Two additional injections of 0.05 mg/kg Buprenex were administered within 24 hours of surgery at ≥6-hour intervals. Immediately following surgery, animals were given water ad libitum supplemented with or without losartan potassium (0.6 g/L; Aurobindo Pharma Limited NDC 65862-201-99) for 4 weeks. During administration, animals were monitored for changes in appearance, activity, and food and water intake, and body weights were recorded twice per week. Four weeks postsurgery, animals were euthanized by CO2 asphyxiation, the right common carotid artery and ligated left common carotid artery were dissected, and the adventitia was removed from each tissue. All tissues were frozen immediately on dry ice and stored at ≤–70°C for protein analyses. For histological analysis, the whole neck and head were prepared as described below. ## Blood pressure measurements. Blood pressure measurements in mice were obtained using the CODA High Throughput Non-Invasive Blood Pressure System (Kent Scientific Corporation CODA-HT4). Blood pressure measurements were recorded in LRP1+/+ and smLRP1–/– mice after weaning at 3–4 weeks of age per the protocol detailed in Daugherty et al. [ 88]. Noninvasive measurements of systolic and diastolic blood pressures were averaged over approximately 15 recorded cycles. Measurements were repeated if the standard deviation was greater than 30 mmHg. Blood pressures were taken daily for 2 weeks to allow mice to acclimate to the device. The remaining measurements were taken 3 times each week for the remaining length of the experiment. ## Carotid artery histology and vessel morphometry. The whole neck and head were dissected from LRP1+/+ and smLRP1–/– mice subjected to left carotid artery ligation with and without losartan. Samples were then skinned and fixed in $10\%$ buffered formalin phosphate (fixative solution; Thermo Fisher Scientific SF100-20) for 3 days, with fixative solution exchanged for fresh fixative solution once per day. After 3 days of fixation, samples were placed in $70\%$ ethanol solution and transferred to the Center for Vascular and Inflammatory Diseases Histology Core at the University of Maryland School of Maryland or shipped to Histoserv, Inc. for decalcification, sectioning, and staining. Whole-neck serial cross sections of 5 μm thickness were sliced starting from the carotid bifurcation to the area inferior to the lesion apex. The apex of the lesion area was identified by analyzing serial sections at 100 μm intervals by H&E, EVG, and Masson’s trichrome staining. Morphometric measurements were performed using EVOS FL Auto Imaging System software (Invitrogen, Thermo Fisher Scientific). All measurements were performed while blinded to the sample identification. ## Global quantification of protein expression. SMA tissue from 14-week-old WT and smLRP1–/– mice was rinsed in PBS to remove blood, frozen with liquid nitrogen in a tissueTUBE TT05M (Covaris catalog 520071), and impact pulverized with a cryoPREP CP01 (Covaris catalog 500230). Fractured tissue was transferred to a 1 mL milliTUBE containing an AFA fiber (Covaris catalog 520135) in 200 μL of 50 mM HEPES pH 8.5, 150 mM NaCl, and $2\%$ Triton X-114 and sonicated with an M220 Focused-Ultrasonicator (Covaris catalog 500295). Sonication parameters were temperature 15°C, peak power 75 W, duty factor 26, cycles/burst = 1,000, and duration 600 seconds. Extracted proteins were clarified of insoluble material by centrifugation at 15,000g for 20 minutes at 4°C. Protein concentrations were determined with the Micro BCA colorimetric assay (Pierce, Thermo Fisher Scientific) with the addition of SDS to a final concentration of $1\%$ in the assay solvent to prevent detergent clouding. Aliquots containing approximately 5 μg of protein were processed using the SP3 protocol as described [89] with some modifications. Briefly, the sample aliquots were brought to 50 μL volume, and disulfide bonds were reduced and alkylated simultaneously with 10 mM TCEP, 40 mM 2-chloroacetamide in 50 mM HEPES pH 8.5, and $1\%$ sodium deoxycholate at 70°C for 10 minutes, then cooled on ice. Proteins were precipitated and captured following addition of 10 μL of a washed 10 μg/μL suspension of SpeedBeads (Cytiva) and 400 μL of ethanol. After shaking for 10 minutes at room temperature, the beads were magnetically captured and washed 3 times with 200 μL of $80\%$ ethanol in water. Proteins were digested on the beads in 50 μL of 50 mM HEPES pH 8.5, $1\%$ sodium deoxycholate, and 10 ng/μL trypsin (Promega) overnight at room temperature with shaking sufficient to maintain the beads in suspension. The digest was diluted 10-fold with $80\%$ acetonitrile and $1\%$ formic acid, then separated from the beads magnetically, and the resulting peptides were captured on 2 mm discs of Empore Cation (CDS Analytical) fitted into 1,000 μL pipette tips (Sartorious catalog 791000). Detergents and other contaminants were removed by washing the tips serially with 1) ethyl acetate; 2) $80\%$ acetonitrile, $1\%$ formic acid; and 3) $10\%$ acetonitrile, $0.2\%$ formic acid. Peptides were eluted directly into injection vials with freshly prepared $80\%$ acetonitrile and $5\%$ ammonium hydroxide and immediately dried down in a centrifugal vacuum evaporator. One-fifth of the recovered peptides from each sample was subsequently analyzed by liquid chromatography-tandem mass spectrometry. In-house capillary columns were constructed from 360 μm OD and 100 μm internal diameter × 30 cm fused silica tubing (Molex) with laser-pulled tips (Sutter Instruments) and were packed with Reprosil-PUR 3 μm C18-AQ (Dr. Maisch GmBH). Solvents A and B consisted of $0.1\%$ formic acid in water and $80\%$ acetonitrile with $0.1\%$ formic acid, respectively. A 180-minute linear gradient from $2\%$ to $35\%$ solvent B was used for chromatographic separation. Peptides were analyzed with an Orbitrap Elite (Thermo Fisher Scientific) mass spectrometer using nano-electrospray ionization with an applied voltage of 1,800 V. MS1 spectra were acquired at a resolution of 120,000, and the 15 most abundant precursor ions were selected for fragmentation by higher energy collision dissociation. MS2 spectra were acquired at a resolution of 15,000. Dynamic exclusion parameters were list size of 500, mass window of ±7 ppm, and duration of 1 minute. Automatic gain control settings were MS1 target 1 × 106, maximum inject time 100 ms; MS2 target 4 × 104, maximum inject time 100 ms. ## Mass spectrometry data analysis. Spectrum matching and protein identification and validation were performed with MSFragger [90], and quantification of protein intensities with matching between runs was performed with IonQuant [91] as components of the FragPipe analysis pipeline using the default settings of each module. The protein database used for the search was the Mus musculus reviewed sequence database downloaded from UniProt on March 8, 2022. The results were subsequently processed to filter out common contaminants, decoy hits from the reverse database, and protein groups identified by a single peptide. The data were filtered as follows: (a) binary expression of a protein (i.e., protein exclusively identified in either LRP1+/+ or smLRP1–/–) was only considered relevant if all LRP1+/+ samples or all smLRP1–/– samples expressed the protein. The missing values were imputed with the minimum intensity value for each specific data set; (b) for samples expressed in both LRP1+/+ and smLRP1–/– tissue, the filtering process required 2 or more proteins to be detected in both the LRP1+/+ and smLRP1–/– samples. False discovery analysis was performed using the Benjamini, Krieger, and Yekutieli method [92] using GraphPad Prism 9.0 software. Causal analysis of proteomic data was performed [52] in IPA upstream analysis software (QIAGEN). For IPA, the binary values were imputed using local minimum intensities. Enrichment analyses for gene ontology (biological process) were performed using clusterProfiler 4.2.2 R package on R 4.1.0. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [93] partner repository with the data set identifier PXD038236. ## Statistics. Prism 9.0 (GraphPad Software) was used for statistical analysis. Normality was determined on data sets using the Shapiro-Wilk test. To compare variables between 2 groups, an unpaired 2-tailed Student’s t test was used for normally distributed variables. When the effect of 2 variables was analyzed in data sets containing normally distributed variables, a 2-way ANOVA with Tukey’s post hoc test was used. To compare more than 2 groups in which normality was not met, the variables were analyzed by a 1-way ANOVA on ranks (Kruskal-Wallis nonparametric test with Dunn’s multiple-comparison post hoc test). All results are presented as mean ± SEM, with P values shown above bars. A P ≤ 0.05 was set as the threshold for significance. ## Study approval. All animal studies were approved by the Institutional Animal Care and Use Committee of the University of Maryland School of Medicine or the University of Kentucky. ## Author contributions JZ, DTA, MF, BOA, BH, MM, FN, AEM, MMW, MF, JJM, DAH, and PW conducted experiments and acquired and analyzed data. JZ, DTA, HS, BH, MM, FN, AAU, HSL, SCM, AD, HS, HSL, AD, and DKS analyzed data. JZ, DTA, HSL, HS, AD, and DKS wrote and edited the manuscript. ## 12/06/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Milewicz DM. Genetic basis of thoracic aortic aneurysms and dissections: focus on smooth muscle cell contractile dysfunction. Annu Rev Genomics Hum Genet (2008) 9 283-302. DOI: 10.1146/annurev.genom.8.080706.092303 2. Lindsay ME, Dietz HC. Lessons on the pathogenesis of aneurysm from heritable conditions. Nature (2011) 473 308-316. DOI: 10.1038/nature10145 3. Corey MR. The natural history of splanchnic artery aneurysms and outcomes after operative intervention. J Vasc Surg (2016) 63 949-957. DOI: 10.1016/j.jvs.2015.10.066 4. Boucher P. LRP: role in vascular wall integrity and protection from atherosclerosis. Science (2003) 300 329-332. DOI: 10.1126/science.1082095 5. Muratoglu SC. LRP1 protects the vasculature by regulating levels of connective tissue growth factor and HtrA1. Arterioscler Thromb Vasc Biol (2013) 33 2137-2146. DOI: 10.1161/ATVBAHA.113.301893 6. Au DT. LRP1 (low-density lipoprotein receptor-related protein 1) regulates smooth muscle contractility by modulating Ca. Arterioscler Thromb Vasc Biol (2018) 38 2651-2664. DOI: 10.1161/ATVBAHA.118.311197 7. Boucher P. LRP1 functions as an atheroprotective integrator of TGFβ and PDGF signals in the vascular wall: implications for Marfan syndrome. PLoS One (2007) 2. DOI: 10.1371/journal.pone.0000448 8. Zhou L. LRP1 regulates architecture of the vascular wall by controlling PDGFRbeta-dependent phosphatidylinositol 3-kinase activation. PLoS One (2009) 4. DOI: 10.1371/journal.pone.0006922 9. Nakajima C. The lipoprotein receptor LRP1 modulates sphingosine-1-phosphate signaling and is essential for vascular development. Development (2014) 141 4513-4525. DOI: 10.1242/dev.109124 10. Willnow TE. Inhibition of hepatic chylomicron remnant uptake by gene transfer of a receptor antagonist. Science (1994) 264 1471-1474. DOI: 10.1126/science.7515194 11. Rohlmann A. Inducible inactivation of hepatic LRP gene by cre-mediated recombination confirms role of LRP in clearance of chylomicron remnants. J Clin Invest (1998) 101 689-695. DOI: 10.1172/JCI1240 12. Strickland DK, Kounnas MZ. Mechanisms of cellular uptake of thrombin-antithrombin II complexes role of the low-density lipoprotein receptor-related protein as a serpin-enzyme complex receptor. Trends Cardiovasc Med (1997) 7 9-16. DOI: 10.1016/S1050-1738(96)00120-X 13. Arai AL. High-affinity binding of LDL receptor — related protein 1 to matrix metalloprotease 1 requires protease: inhibitor complex formation. Biochemistry (2020) 59 2922-2933. DOI: 10.1021/acs.biochem.0c00442 14. Mantuano E. LDL receptor-related protein-1 regulates NF κ B and microRNA-155 in macrophages to control the inflammatory response. Proc Natl Acad Sci U S A (2016) 113 1369-1374. DOI: 10.1073/pnas.1515480113 15. Bown MJ. Abdominal aortic aneurysm is associated with a variant in low-density lipoprotein receptor-related protein 1. Am J Hum Genet (2011) 89 619-627. DOI: 10.1016/j.ajhg.2011.10.002 16. Bradley DT. A variant in LDLR is associated with abdominal aortic aneurysm. Circ Cardiovasc Genet (2013) 6 498-504. DOI: 10.1161/CIRCGENETICS.113.000165 17. Galora S. Association of rs1466535 LRP1 but not rs3019885 SLC30A8 and rs6674171 TDRD10 gene polymorphisms with abdominal aortic aneurysm in Italian patients. J Vasc Surg (2015) 61 787-792. DOI: 10.1016/j.jvs.2013.10.090 18. Debette S. Common variation in PHACTR1 is associated with susceptibility to cervical artery dissection. Nat Genet (2015) 47 78-83. DOI: 10.1038/ng.3154 19. Guo D. Genetic variants in LRP1 and ULK4 are associated with acute aortic dissections. Am J Hum Genet (2016) 99 762-769. DOI: 10.1016/j.ajhg.2016.06.034 20. Jones GT. Meta-analysis of genome-wide association studies for abdominal aortic aneurysm identifies four new disease-specific risk loci. Circ Res (2017) 120 341-353. DOI: 10.1161/CIRCRESAHA.116.308765 21. Erhart P. Genetic variation in LRP1 associates with stanford type B aortic dissection risk and clinical outcome. J Cardiovasc Dev Dis (2022) 9. PMID: 35050224 22. Muratoglu SC. Low density lipoprotein receptor-related protein 1 (LRP1) forms a signaling complex with platelet-derived growth factor receptor-beta in endosomes and regulates activation of the MAPK pathway. J Biol Chem (2010) 285 14308-14317. DOI: 10.1074/jbc.M109.046672 23. Newton CS. Platelet-derived growth factor receptor-beta (PDGFR-beta) activation promotes its association with the low density lipoprotein receptor-related protein (LRP). Evidence for co-receptor function. J Biol Chem (2005) 280 27872-27878. DOI: 10.1074/jbc.M505410200 24. Loukinova E. Platelet-derived growth factor (PDGF)-induced tyrosine phosphorylation of the low density lipoprotein receptor-related protein (LRP). Evidence for integrated co-receptor function betwenn LRP and the PDGF. J Biol Chem (2002) 277 15499-15506. DOI: 10.1074/jbc.M200427200 25. Huang SS. Cellular growth inhibition by IGFBP-3 and TGF-β 1 requires LRP-1. FASEB J (2003) 17 2068-2081. DOI: 10.1096/fj.03-0256com 26. Cabello-Verrugio C, Brandan E. A novel modulatory mechanism of transforming growth factor-beta signaling through decorin and LRP-1. J Biol Chem (2007) 282 18842-18850. DOI: 10.1074/jbc.M700243200 27. Muratoglu SC. Macrophage LRP1 suppresses neo-intima formation during vascular remodeling by modulating the TGF-β signaling pathway. PLoS One (2011) 6. DOI: 10.1371/journal.pone.0028846 28. Tseng WF. LRP-1/TbetaR-V mediates TGF-beta1-induced growth inhibition in CHO cells. FEBS Lett (2004) 562 71-78. PMID: 15044004 29. Basford JE. Smooth muscle LDL receptor-related protein-1 inactivation reduces vascular reactivity and promotes injury-induced neointima formation. Arterioscler Thromb Vasc Biol (2009) 29 1772-1778. DOI: 10.1161/ATVBAHA.109.194357 30. Davis FM. Smooth muscle cell deletion of low-density lipoprotein receptor-related protein 1 augments angiotensin II-induced superior mesenteric arterial and ascending aortic aneurysms. Arterioscler Thromb Vasc Biol (2015) 35 155-162. DOI: 10.1161/ATVBAHA.114.304683 31. Mehta PK, Griendling KK. Angiotensin II cell signaling: physiological and pathological effects in the cardiovascular system. Am J Physiol Cell Physiol (2007) 292 C82-C97. DOI: 10.1152/ajpcell.00287.2006 32. Yu C, Jeremy RW. Angiotensin, transforming growth factor β and aortic dilatation in Marfan syndrome: of mice and humans. Int J Cardiol Heart Vasc (2018) 18 71-80. PMID: 29876507 33. Lu H. Angiotensinogen exerts effects independent of angiotensin II. Arterioscler Thromb Vasc Biol (2016) 36 256-265. DOI: 10.1161/ATVBAHA.115.306740 34. Wu C. Cys18-Cys137 disulfide bond in mouse angiotensinogen does not affect AngII-dependent functions in vivo. Hypertension (2015) 65 800-805. DOI: 10.1161/HYPERTENSIONAHA.115.05166 35. Moon MC. Perivascular delivery of losartan with surgical fibrin glue prevents neointimal hyperplasia after arterial injury. J Vasc Surg (2004) 40 130-137. DOI: 10.1016/j.jvs.2004.02.031 36. Li Y. Valsartan decreases neointimal hyperplasia in balloon-injured rat aortic arteries by upregulating HO-1 and inhibiting angiotensin II type 1 receptor. Life Sci (2014) 110 70-76. DOI: 10.1016/j.lfs.2014.06.021 37. Huckle WR. Effects of subtype-selective and balanced angiotensin II receptor antagonists in a porcine coronary artery model of vascular restenosis. Circulation (1996) 93 1009-1019. DOI: 10.1161/01.CIR.93.5.1009 38. Powell JS. Inhibitors of angiotensin-converting enzyme prevent myointimal proliferation after vascular injury. Science (1989) 245 186-188. DOI: 10.1126/science.2526370 39. Daemen MJAP. Angiotensin II induces smooth muscle cell proliferation in the normal and injured rat arterial wall. Circ Res (1991) 68 450-456. DOI: 10.1161/01.RES.68.2.450 40. Laporte S, Escher E. Neointima formation after vascular injury is angiotensin II mediated. Biochem Biophys Res Commun (1992) 187 1510-1516. DOI: 10.1016/0006-291X(92)90473-X 41. Clozel J-P. Influence of the status of the renin-angiotensin system on the effect of cilazapril on neointima formation after vascular injury in rats. Circulation (1993) 88 1222-1227. DOI: 10.1161/01.CIR.88.3.1222 42. Janiak P. Role of the renin-angiotensin system in neointima formation after injury in rabbits. Hypertension (1994) 24 671-678. DOI: 10.1161/01.HYP.24.6.671 43. Daugherty A. Angiotensin II promotes atherosclerotic lesions and aneurysms in apolipoprotein E-deficient mice. J Clin Invest (2000) 105 1605-1612. DOI: 10.1172/JCI7818 44. Sendra J. Angiotensin II upregulates LDL receptor-related protein (LRP1) expression in the vascular wall: a new pro-atherogenic mechanism of hypertension. Cardiovasc Res (2008) 78 581-589. DOI: 10.1093/cvr/cvn043 45. Vaughan DE. Angiotensin II regulates the expression of plasminogen activator inhibitor-1 in cultured endothelial cells. A potential link between the renin-angiotensin system and thrombosis. J Clin Invest (1995) 95 995-1001. DOI: 10.1172/JCI117809 46. Huo KG. miR-431-5p knockdown protects against angiotensin II-induced hypertension and vascular injury. Hypertension (2019) 73 1007-1017. DOI: 10.1161/HYPERTENSIONAHA.119.12619 47. Subramanian V. Regulation of peroxisome proliferator-activated receptor-γ by angiotensin II via transforming growth factor-β1-activated p38 mitogen-activated protein kinase in aortic smooth muscle cells. Arterioscler Thromb Vasc Biol (2012) 32 397-405. DOI: 10.1161/ATVBAHA.111.239897 48. Iwanciw D. Induction of connective tissue growth factor by angiotensin II: integration of signaling pathways. Arterioscler Thromb Vasc Biol (2003) 23 1782-1787. DOI: 10.1161/01.ATV.0000092913.60428.E6 49. Patel VB. Angiotensin-converting enzyme 2 is a critical determinant of angiotensin II-induced loss of vascular smooth muscle cells and adverse vascular remodeling. Hypertension (2014) 64 157-164. DOI: 10.1161/HYPERTENSIONAHA.114.03388 50. Bledzka K. Spatial coordination of kindlin-2 with talin head domain in interaction with integrin β cytoplasmic tails. J Biol Chem (2012) 287 24585-24594. DOI: 10.1074/jbc.M111.336743 51. Karimi A, Milewicz DM. Structure of the elastin-contractile units in the thoracic aorta and how genes that cause thoracic aortic aneurysms and dissections disrupt this structure. Can J Cardiol (2016) 32 26-34. DOI: 10.1016/j.cjca.2015.11.004 52. Krämer A. Causal analysis approaches in ingenuity pathway analysis. Bioinformatics (2014) 30 523-530. DOI: 10.1093/bioinformatics/btt703 53. Neptune ER. Dysregulation of TGF-beta activation contributes to pathogenesis in Marfan syndrome. Nat Genet (2003) 33 407-411. DOI: 10.1038/ng1116 54. Kavsak P. Smad7 binds to Smurf2 to form an E3 ubiquitin ligase that targets the TGFβ receptor for degradation. Mol Cell (2000) 6 1365-1375. DOI: 10.1016/S1097-2765(00)00134-9 55. Hayashi H. The MAD-related protein Smad7 associates with the TGFbeta receptor and functions as an antagonist of TGFbeta signaling. Cell (1997) 89 1165-1173. DOI: 10.1016/S0092-8674(00)80303-7 56. Du KL. Myocardin is a critical serum response factor cofactor in the transcriptional program regulating smooth muscle cell differentiation. Mol Cell Biol (2003) 23 2425-2437. DOI: 10.1128/MCB.23.7.2425-2437.2003 57. Pannu H. MYH11 mutations result in a distinct vascular pathology driven by insulin-like growth factor 1 and angiotensin II. Hum Mol Genet (2007) 16 2453-2462. DOI: 10.1093/hmg/ddm201 58. Rong J. Loss of hepatic angiotensinogen attenuates sepsis-induced myocardial dysfunction. Circ Res (2021) 547 547-564. PMID: 34238019 59. Kumar A, Lindner V. Remodeling with neointima formation in the mouse carotid artery after cessation of blood flow. Arterioscler Thromb Vasc Biol (1997) 17 2238-2244. DOI: 10.1161/01.ATV.17.10.2238 60. De Meyer GR, Bult H. Mechanisms of neointima formation — lessons from experimental models. Vasc Med (1997) 2 179-189. DOI: 10.1177/1358863X9700200304 61. Renna NF. Pathophysiology of vascular remodeling in hypertension. Int J Hypertens (2013) 2013. PMID: 23970958 62. Habashi JP. Losartan, an AT1 antagonist, prevents aortic aneurysm in a mouse model of Marfan syndrome. Science (2006) 312 117-121. DOI: 10.1126/science.1124287 63. Holm TM. Noncanonical TGFβ signaling contributes to aortic aneurysm progression in Marfan syndrome mice. Science (2011) 332 358-361. DOI: 10.1126/science.1192149 64. Devereux RB. Regression of hypertensive left ventricular hypertrophy by losartan compared with atenolol: the Losartan Intervention for Endpoint Reduction in Hypertension (LIFE) trial. Circulation (2004) 110 1456-1462. DOI: 10.1161/01.CIR.0000141573.44737.5A 65. Brenner BM. Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy. N Engl J Med (2001) 345 861-869. DOI: 10.1056/NEJMoa011161 66. Van Andel MM. Long-term clinical outcomes of losartan in patients with Marfan syndrome: follow-up of the multicentre randomized controlled COMPARE trial. Eur Heart J (2020) 41 4181-4187. DOI: 10.1093/eurheartj/ehaa377 67. Gallo EM. Angiotensin II-dependent TGF-β signaling contributes to Loeys-Dietz syndrome vascular pathogenesis. J Clin Invest (2014) 124 448-460. DOI: 10.1172/JCI69666 68. Li G. Exome sequencing identified new mutations in a Marfan syndrome family. Diagn Pathol (2014) 9. DOI: 10.1186/1746-1596-9-25 69. Moxon JV. Plasma low-density lipoprotein receptor-related protein 1 concentration is not associated with human abdominal aortic aneurysm presence. Eur J Vasc Endovasc Surg (2015) 50 466-473. DOI: 10.1016/j.ejvs.2015.06.023 70. Chan CYT. A pilot study on low-density lipoprotein receptor-related protein-1 in Chinese patients with abdominal aortic aneurysm. Eur J Vasc Endovasc Surg (2013) 46 549-556. DOI: 10.1016/j.ejvs.2013.08.006 71. Chan CYT. Clearance of matrix metalloproteinase-9 is dependent on low-density lipoprotein receptor-related protein-1 expression downregulated by microRNA-205 in human abdominal aortic aneurysm. J Vasc Surg (2017) 65 509-520. DOI: 10.1016/j.jvs.2015.10.065 72. Sakalihasan N. Activated forms of MMP2 and MMP9 in abdominal aortic aneurysms. J Vasc Surg (1996) 24 127-133. DOI: 10.1016/S0741-5214(96)70153-2 73. Yamamoto K. A top-down approach to uncover the hidden ligandome of low-density lipoprotein receptor-related protein 1 in cartilage. Matrix Biol (2022) 112 190-218. DOI: 10.1016/j.matbio.2022.08.007 74. Rosenkranz S. TGF-beta1 and angiotensin networking in cardiac remodeling. Cardiovasc Res (2004) 63 423-432. DOI: 10.1016/j.cardiores.2004.04.030 75. Campbell SE, Katwa LC. Angiotensin II stimulated expression of transforming growth factor-beta1 in cardiac fibroblasts and myofibroblasts. J Mol Cell Cardiol (1997) 29 1947-1958. DOI: 10.1006/jmcc.1997.0435 76. Bataller R. NADPH oxidase signal transduces angiotensin II in hepatic stellate cells and is critical in hepatic fibrosis. J Clin Invest (2003) 112 1383-1394. DOI: 10.1172/JCI18212 77. Lim DS. Angiotensin II blockade reverses myocardial fibrosis in a transgenic mouse model of human hypertrophic cardiomyopathy. Circulation (2001) 103 789-791. DOI: 10.1161/01.CIR.103.6.789 78. Majesky MW. Production of transforming growth factor beta 1 during repair of arterial injury. J Clin Invest (1991) 88 904-910. DOI: 10.1172/JCI115393 79. Nabel EG. Direct transfer of transforming growth factor beta 1 gene into arteries stimulates fibrocellular hyperplasia. Proc Natl Acad Sci U S A (1993) 90 10759-10763. DOI: 10.1073/pnas.90.22.10759 80. Smith JD. Soluble transforming growth factor-beta type II receptor inhibits negative remodeling, fibroblast transdifferentiation, and intimal lesion formation but not endothelial growth. Circ Res (1999) 84 1212-1222. DOI: 10.1161/01.RES.84.10.1212 81. Tsai S. TGF-beta through Smad3 signaling stimulates vascular smooth muscle cell proliferation and neointimal formation. Am J Physiol Heart Circ Physiol (2009) 297 H540-H549. DOI: 10.1152/ajpheart.91478.2007 82. Suwanabol PA et al. Transforming growth factor-β increases vascular smooth muscle cell proliferation through the Smad3 and extracellular signal-regulated kinase mitogen-activated protein kinases pathways. J Vasc Surg (2012) 56 446-454. DOI: 10.1016/j.jvs.2011.12.038 83. Gao F. Disruption of TGF-β signaling in smooth muscle cell prevents flow-induced vascular remodeling. Biochem Biophys Res Commun (2014) 454 245-250. DOI: 10.1016/j.bbrc.2014.10.092 84. Rush C. Whole genome expression analysis within the angiotensin II-apolipoprotein E deficient mouse model of abdominal aortic aneurysm. BMC Genomics (2009) 10. DOI: 10.1186/1471-2164-10-298 85. Hamlin AN. Low-density lipoprotein receptor–related protein-1 dysfunction synergizes with dietary cholesterol to accelerate steatohepatitis progression. J Biol Chem (2018) 293 9674-9684. DOI: 10.1074/jbc.RA118.001952 86. Rodríguez-Vita J et al. Angiotensin II activates the Smad pathway in vascular smooth muscle cells by a transforming growth factor-β–independent mechanism. Circulation (2005) 111 2509-2517. DOI: 10.1161/01.CIR.0000165133.84978.E2 87. Wang Y. TGF-β activity protects against inflammatory aortic aneurysm progression and complications in angiotensin II–infused mice. J Clin Invest (2010) 120 422-432. DOI: 10.1172/JCI38136 88. Daugherty A. Measuring blood pressure in mice using volume pressure recording, a tail-cuff method. J Vis Exp (2009). PMID: 19488026 89. Hughes CS. Single-pot, solid-phase-enhanced sample preparation for proteomics experiments. Nat Protoc (2019) 14 68-85. DOI: 10.1038/s41596-018-0082-x 90. Kong AT. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry-based proteomics. Nat Methods (2017) 14 513-520. DOI: 10.1038/nmeth.4256 91. Yu F. IonQuant enables accurate and sensitive label-free quantification with FDR-controlled match-between-runs. Mol Cell Proteomics (2021) 20. DOI: 10.1016/j.mcpro.2021.100077 92. Benjamini Y. Adaptive linear step-up procedures that control false discovery rate. Biometrika (2006) 93 491-507. DOI: 10.1093/biomet/93.3.491 93. Deutsch EW. The ProteomeXchange consortium in 2020: enabling “big data” approaches in proteomics. Nucleic Acids Res (2020) 48 D1145-D1152. PMID: 31686107
--- title: Insulin-like growth factor 2 mRNA-binding protein 3 promotes kidney injury by regulating β-catenin signaling authors: - Dongyan Song - Jingyue Shang - Yinyi Long - Menghua Zhong - Li Li - Jiongcheng Chen - Yadie Xiang - Huishi Tan - Haili Zhu - Xue Hong - Fan Fan Hou - Haiyan Fu - Youhua Liu journal: JCI Insight year: 2023 pmcid: PMC9977311 doi: 10.1172/jci.insight.162060 license: CC BY 4.0 --- # Insulin-like growth factor 2 mRNA-binding protein 3 promotes kidney injury by regulating β-catenin signaling ## Abstract Wnt/β-catenin is a developmental signaling pathway that plays a crucial role in driving kidney fibrosis after injury. Activation of β-catenin is presumed to be regulated through the posttranslational protein modification. Little is known about whether β-catenin is also subjected to regulation at the posttranscriptional mRNA level. Here, we report that insulin-like growth factor 2 mRNA-binding protein 3 (IGF2BP3) plays a pivotal role in regulating β-catenin. IGF2BP3 was upregulated in renal tubular epithelium of various animal models and patients with chronic kidney disease. IGF2BP3 not only was a direct downstream target of Wnt/β-catenin but also was obligatory for transducing Wnt signal. In vitro, overexpression of IGF2BP3 in kidney tubular cells induced fibrotic responses, whereas knockdown of endogenous IGF2BP3 prevented the expression of injury and fibrosis markers in tubular cells after Wnt3a stimulation. In vivo, exogenous IGF2BP3 promoted β-catenin activation and aggravated kidney fibrosis, while knockdown of IGF2BP3 ameliorated renal fibrotic lesions after obstructive injury. RNA immunoprecipitation and mRNA stability assays revealed that IGF2BP3 directly bound to β-catenin mRNA and stabilized it against degradation. Furthermore, knockdown of IGF2BP3 in tubular cells accelerated β-catenin mRNA degradation in vitro. These studies demonstrate that IGF2BP3 promotes β-catenin signaling and drives kidney fibrosis, which may be mediated through stabilizing β-catenin mRNA. Our findings uncover a previously underappreciated dimension of the complex regulation of Wnt/β-catenin signaling and suggest a potential target for therapeutic intervention of fibrotic kidney diseases. ## Introduction Chronic kidney disease (CKD), characterized by progressive tissue fibrosis and gradual loss of kidney function, is becoming a major public health problem worldwide [1]. CKD is highly prevalent and associated with increased risk of progression to end-stage renal disease, a devastating condition with high morbidity and mortality [2]. Extensive studies show that kidney fibrosis is the common outcome of all kinds of CKD, regardless of the initial causes [3]. The pathophysiology of kidney fibrosis is complex and involves many types of cells, in which the role of tubular epithelial cells is particularly intriguing, because they are the primary targets of kidney injury in most circumstances (4–6). In response to various insults, tubular cells undergo different changes, such as partial epithelial-mesenchymal transition, cell cycle arrest, cellular senescence, or metabolic reprogramming (7–13). These responses are controlled by several key signal cascades, including Wnt/β-catenin signaling. Wnt/β-catenin is an evolutionarily conserved signaling pathway that is activated after tissue injury and plays a crucial role in driving tissue fibrosis [14]. As a master transcriptional regulator, β-catenin drives kidney fibrogenesis by inducing a variety of fibrosis-related downstream targets, such as Snail1, fibronectin, matrix metalloproteinase-7 (MMP-7), plasminogen activator inhibitor-1 (PAI-1), and components of the renin-angiotensin system [15]. Consistently, inhibition of β-catenin ameliorates renal fibrotic lesions in various models of CKD (16–18). These findings underscore that hyperactive β-catenin could be the causative culprit behind kidney damage and disease after injury. It is widely accepted that the regulation of β-catenin is primarily controlled at the posttranslational level. In quiescent state, β-catenin is constitutively destructed by a phosphorylation-triggered, ubiquitin-mediated protein degradation. Upon Wnt activation, β-catenin is dephosphorylated, leading to its stabilization, accumulation, and nuclear translocation [19, 20]. However, whether β-catenin is also subjected to regulation at the mRNA level is largely unknown. Furthermore, what controls β-catenin mRNA regulation and how important it is in the pathogenesis of CKD remain enigmatic. Insulin-like growth factor 2 mRNA-binding protein 3 (IGF2BP3) belongs to a highly conserved family of RNA-binding proteins, which also includes IGF2BP1 and IGF2BP2 (21–26). As RNA-binding proteins, IGF2BPs regulate many biological processes, such as embryonic development, tumor formation, and pathogenesis of human diseases, by interacting with various target RNAs, thereby controlling their stability, storage, nuclear export, and subcellular localization and degradation and affecting gene expression output [25, 27, 28]. IGF2BPs share a high degree of homology in their sequences, but each of them exhibits uniqueness in terms of expression pattern, target specificity, and injury responses [22, 25]. For example, IGF2BP2 has been implicated as a candidate gene involved in type 2 diabetes and has diverse functions in cell metabolism [29, 30], whereas IGF2BP3 is associated with liver fibrosis through controlling transformation of hepatic stellate cells into myofibroblasts [31]. IGF2BP3 has been shown to be a prognostic and diagnostic biomarker and therapeutic target for renal cell carcinoma (32–35). However, its role in the pathogenesis of kidney fibrosis remains unknown. In this study, we show that IGF2BP3 is induced in the kidneys of animal models and patients with CKD. Overexpression of IGF2BP3 impairs tubular cell integrity and promotes kidney fibrosis, possibly by stabilizing β-catenin mRNA. These studies identify IGF2BP3 as a major player in kidney fibrogenesis. ## Induction of IGF2BP3 in various animal models of CKD. We first assessed the expression of IGF2BP3 in various well-established animal models of CKD induced by unilateral ureteral obstruction (UUO), ischemia/reperfusion injury (IRI), adriamycin (ADR), and angiotensin II (Ang II), respectively. These models represent diverse etiologies that lead to renal failure and fibrotic lesions. As shown in Figure 1, A–H, IGF2BP3 protein was markedly induced at 7 days after UUO, 11 days after IRI, 2 weeks after ADR, and 4 weeks after chronic Ang II infusion. These results suggest that IGF2BP3 induction is a common pathologic finding in CKD, regardless of the initial causes. We next examined IGF2BP3 localization in the fibrotic kidneys by using immunohistochemical staining. IGF2BP3 was undetectable in normal kidney but markedly upregulated in CKD (Figure 1I). Strong IGF2BP3 staining was predominantly localized in renal tubular epithelium, whereas glomeruli were essentially negative (Figure 1I). Judging from the staining pattern, IGF2BP3 appeared primarily present in the cytoplasm of tubular epithelial cells. To verify this, we assessed its abundance in the fibrotic kidneys after nuclear and cytoplasmic fractionation. As shown in Figure 1J, IGF2BP3 was upregulated in the cytoplasm, but not in the nuclei, after IRI. In contrast, β-catenin protein was mainly upregulated in the nuclei in the same setting (Figure 1J). Similar induction of renal IGF2BP3 mRNA was evident in the fibrotic kidneys, as demonstrated by quantitative real-time polymerase chain reaction (qRT-PCR) (Figure 1, K–M). We also assessed the expression of other members of IGF2BP family proteins. IGF2BP2 was also induced in the kidneys after UUO and IRI, although to a lesser extent (Supplemental Figure 1, A, C, D, and F; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.162060DS1). IGF2BP2 localized in renal tubular epithelium as well (Supplemental Figure 1G). The expression of IGF2BP1 was CKD content dependent, and it was induced in the kidney after IRI but not in the fibrotic kidney after UUO (Supplemental Figure 1). Based on these observations, we chose to focus on IGF2BP3 in this study as its induction was most robust and common in all CKD tested. We found that IGF2BP3 was mainly upregulated in CKD, and little or no induction of IGF2BP3 was observed in animal models of acute kidney injury (AKI) at 3 days after cisplatin or 1 day after IRI (Supplemental Figure 2), suggesting that it could be relevant to kidney fibrogenesis but not to renal repair and regeneration. ## IGF2BP3 is upregulated and associated with kidney dysfunction and fibrosis in human CKD. To study the clinical relevance of these findings, we investigated IGF2BP3 expression in human kidney biopsies from patients with CKD. As shown in Figure 2, A and B, IGF2BP3 protein was barely detectable in 5 cases of control, nontumor kidney tissue specimens. However, IGF2BP3 protein was readily observed in all 25 biopsy specimens from patients with different CKDs, albeit with varying staining intensities. The demographic and clinical data of the patients with CKD are presented as Supplemental Table 1. It appeared that IGF2BP3 was predominantly induced in renal tubular epithelium, whereas its glomerular staining was not seen or was minimal (Figure 2A). Representative micrographs of IGF2BP3 staining in kidney biopsy specimens from patients with DN, CTIN, LN, IgAN, FSGS, and MN are presented in Figure 2A. The relative levels of IGF2BP3 protein in control and CKD groups are shown in Figure 2B. We further assessed fibrotic lesions in the kidney biopsies of patients with CKD after Masson’s trichrome staining and analyzed the relationship between IGF2BP3 and fibrotic lesions and kidney function in 25 cases of CKD. As shown in Figure 2C, IGF2BP3 protein was correlated with the severity of fibrotic lesions in these patients. Similarly, IGF2BP3 was also closely correlated with serum creatinine levels (Figure 2D). Consistently, IGF2BP3 was inversely associated with estimated glomerular filtration rate (eGFR) (Figure 2E). ## IGF2BP3 is a downstream target of Wnt/β-catenin signaling in vitro and in vivo. We next investigated the potential mechanism underlying IGF2BP3 induction in diseased kidneys. Because activation of Wnt/β-catenin is a common finding in virtually all CKD, this prompted us to examine its role in regulating IGF2BP3 expression. We first analyzed the regulatory regions of the IGF2BP3 promoter by a bioinformatics approach. As shown in Figure 3A, there were putative TBSs in the promoter region of human, mouse, and rat IGF2BP3 genes, which were perfectly matched with TBS consensus sequences (Figure 3A). To ascertain the regulation of IGF2BP3 by Wnt/β-catenin, HKC-8 cells were incubated with Wnt3a for various periods. As shown in Figure 3, B–D, Wnt3a induced active β-catenin and IGF2BP3 expression in a time-dependent manner. Similarly, Wnt3a also dose-dependently induced active β-catenin and IGF2BP3 expression (Figure 3, E–G). The subcellular localization of IGF2BP3 was assessed by immunostaining. IGF2BP3 predominantly localized in the cytoplasm of HKC-8 cells (Figure 3H). This result was further verified by immunoblotting of cytosolic and nuclear fractionations (Figure 3I). After transfection with N-terminus–truncated, constitutively activated β-catenin expression vector (pDel-β-cat), a substantial upregulation of IGF2BP3 was observed only in the cytosolic fraction (Figure 3I). To further validate the involvement of Wnt/β-catenin in regulating IGF2BP3, we utilized a specific small molecule (ICG-001) that selectively inhibited β-catenin–mediated gene transcription in HKC-8 cells. As shown in Figure 3, J and K, ICG-001 abolished IGF2BP3 expression induced by activated β-catenin. Similarly, in a mouse model of UUO, ICG-001 also largely abolished β-catenin activation and IGF2BP3 induction in vivo, as shown by Western blotting and immunostaining (Figure 3, L–Q). ## Overexpression of IGF2BP3 activates β-catenin signaling in vitro. To delineate the role of IGF2BP3 in tubular cell biology, we examined the effect of its overexpression on kidney tubular epithelial cells. To this end, HKC-8 cells were infected with either control lentivirus (LV-Ctrl) or IGF2BP3-expressing lentivirus (LV-IGF2BP3) overnight and then incubated for 2 days as indicated. As shown in Figure 4, A and B, infection with LV-IGF2BP3 resulted in marked induction of IGF2BP3 or Flag-tag. Interestingly, compared with the LV-Ctrl, overexpression of IGF2BP3 induced the expression of fibronectin, vimentin, collagen I, and α–smooth muscle actin (α-SMA) (Figure 4, A and C–F). Immunostaining for fibronectin gave rise to similar results (Figure 4G). We found that overexpression of IGF2BP3 induced β-catenin and active β-catenin (Figure 4, H–J), suggesting its role in activating β-catenin. As shown in Figure 4, H and K–M, several downstream targets of β-catenin, such as PAI-1, MMP-7, and Snail1, were also upregulated after overexpression of IGF2BP3. Furthermore, IGF2BP3 activated β-catenin–mediated gene transcription in the TOPFlash reporter luciferase assay in HEK293T cells (Figure 4N). However, we found that IGF2BP3 did not affect TGF-β signaling, as overexpression of IGF2BP3 showed little effect on the expression of TGF-β1, active TGF-β1, TGF-β receptor I, TGF-β receptor II, and total and active Smad2 and Smad3 (Supplemental Figure 3). Therefore, these results indicate that overexpression of IGF2BP3 induces profibrotic responses in tubular epithelial cells primarily by activating β-catenin signaling. We also examined the effect of IGF2BP3 on cell cycle arrest and cellular senescence, as they contribute to the pathogenesis of kidney injury and fibrosis. As shown in Figure 4, O–R, compared with the LV-Ctrl, overexpression of IGF2BP3 induced the expression of kidney injury molecule-1 (KIM-1) (a tubular injury marker), phosphorylated histone H3 (p-H3) (a G2/M arrest–related marker), and p16INK4A (a cellular senescence–related marker). To clarify whether these effects are mediated through IGF2BP3 regulation of β-catenin, HKC-8 cells were infected with either control lentivirus or IGF2BP3 lentivirus overnight, then incubated with ICG-001. As shown in Supplemental Figure 4, ICG-001 completely blocked IGF2BP3-induced upregulation of active β-catenin, fibronectin, vimentin, KIM-1, and p-H3. These data suggest that β-catenin plays a critical role in mediating the effects of IGF2BP3 on tubular injury, cell cycle arrest, and fibrotic response. ## IGF2BP3 is required for Wnt/β-catenin signaling in vitro. We wondered whether IGF2BP3 induction is required for Wnt/β-catenin to elicit its profibrotic action. To test this, we transfected HKC-8 cells with control or IGF2BP3-specific siRNA in the absence or presence of Wnt3a. As shown in Figure 5, A and B, transfection of IGF2BP3-specific siRNA knocked down IGF2BP3 induced by Wnt3a in HKC-8 cells. Consistently, knockdown of IGF2BP3 abrogated the Wnt3a-triggered β-catenin activation and its downstream Snail1, fibronectin, α-SMA, KIM-1, p53, and p16INK4A (Figure 5, A and C–J). Immunostaining also showed that depletion of IGF2BP3 abolished Wnt3a-induced fibronectin deposition (Figure 5K). Therefore, IGF2BP3 is required for Wnt3a/β-catenin activation and its downstream genes’ expression. On the flip side, we found that β-catenin was required for IGF2BP3 induction and function, as ICG-001 abolished the induction of IGF2BP3, activation of β-catenin, and upregulation of MMP-7, fibronectin, p-H3, and p16INK4A after Wnt3a stimulation (Figure 5, L–R). Together, IGF2BP3 induction and β-catenin activation form a reciprocal feed-forward loop (Figure 5S). ## Overexpression of IGF2BP3 aggravates kidney fibrosis and promotes β-catenin signaling in vivo. To explore the role of IGF2BP3 in vivo, normal mice were injected with control adeno-associated viral vector (AAV-Ctrl) or IGF2BP3 adeno-associated viral vector (AAV-BP3) into 6 sites of the left renal cortex for 12 weeks, as intraparenchymal AAV injections result in robust but relatively local transduction [36]. Immunostaining and Western blotting of whole-kidney homogenates revealed that injections of AAV-BP3 vectors induced renal tubular expression of Flag-tagged IGF2BP3 (Supplemental Figure 5). However, overexpression of IGF2BP3 by intrarenal injections of AAV-BP3 did not trigger fibrotic response in normal mice (Supplemental Figure 5). To further investigate the role of IGF2BP3 in the pathogenesis of CKD, we used a mouse model of UUO, a widely used model with robust kidney fibrosis. To this end, mice were injected with AAV-Ctrl or AAV-BP3 at 12 weeks prior to UUO (Figure 6A). Western blotting of whole-kidney homogenates revealed that injections of AAV-BP3 vectors induced renal expression of Flag-tagged IGF2BP3 (Figure 6, B and C). This result was verified by immunostaining for Flag and IGF2BP3, respectively (Figure 6D and Supplemental Figure 6, A and B). Both IGF2BP3 and Flag staining were predominantly localized in cortical tubular epithelium of mouse kidneys (Figure 6D). We found that overexpression of IGF2BP3 promoted the expression of several fibrosis-related proteins and aggravated kidney fibrotic lesions. As shown in Figure 6, E–H, exogenous IGF2BP3 could upregulate renal vimentin, fibronectin, and α-SMA expression at 7 days after UUO. This result was further verified by immunostaining (Figure 6I and Supplemental Figure 6, C–E). Furthermore, overexpression of IGF2BP3 promoted collagen deposition in UUO kidneys, as shown by Sirius red staining (Figure 6, I and J). We also investigated the effect of IGF2BP3 on β-catenin signaling in vivo. As shown in Figure 6, K and L, renal expression of β-catenin, active β-catenin, MMP-7, Snail1, p-H3, p53, and p16INK4A was upregulated in UUO mice, and overexpression of IGF2BP3 augmented these inductions. ## Knockdown of IGF2BP3 ameliorates kidney fibrosis and inhibits β-catenin signaling in vivo. To further confirm the role of IGF2BP3 in CKD, we sought to knock down IGF2BP3 by using an shRNA-mediated inhibition approach in vivo. Mice were injected intravenously with either Ctrl-shR or IGF2BP3-shR plasmids at 2 days after UUO (Figure 7A). As illustrated in Figure 7, B and C, renal expression of IGF2BP3 was inhibited after intravenous injection of IGF2BP3-shR. We next assessed the effects of IGF2BP3 depletion on fibrotic lesions after UUO. As shown in Figure 7, D and E, knockdown of IGF2BP3 reduced collagen deposition, as shown by Sirius red staining. Western blotting showed that knockdown of IGF2BP3 inhibited renal expression of vimentin, fibronectin, α-SMA, and collagen I, and restored, at least partially, renal E-cadherin expression in UUO mice (Figure 7F and Supplemental Figure 7, A–E). Immunostaining for E-cadherin, vimentin, fibronectin, and α-SMA proteins gave rise to similar results (Figure 7G and Supplemental Figure 7, F–I). We also assessed the effects of IGF2BP3 depletion on β-catenin signaling. As shown in Figure 7H and Supplemental Figure 8, A–G, renal expression of β-catenin, active β-catenin, PAI-1, MMP-7, Snail1, p-H3, and p16INK4A was upregulated in UUO mice, whereas knockdown of IGF2BP3 abolished the induction of these proteins. Immunostaining for β-catenin produced similar results (Figure 7G and Supplemental Figure 7J). Knockdown of IGF2BP3 also reduced renal mRNA levels of fibronectin and β-catenin after UUO (Supplemental Figure 8, H and I). Therefore, depletion of IGF2BP3 ameliorates renal fibrosis and inhibits β-catenin activation after UUO. ## IGF2BP3 binds to and stabilizes β-catenin mRNA. We further investigated the potential mechanisms underlying IGF2BP3 regulation of β-catenin. As an RNA-binding protein, IGF2BP3 action might be related to its interaction with target RNAs [21, 22, 28]. Accordingly, we tested whether IGF2BP3 directly binds to β-catenin mRNA by performing an RNA immunoprecipitation (RIP) assay. As shown in Figure 8, A–C, β-catenin mRNA was detected in the immunocomplexes precipitated by anti-IGF2BP3 antibody in HKC-8 cells, suggesting a direct interaction between IGF2BP3 and β-catenin mRNA. In addition, knockdown of IGF2BP3 decreased the steady-state levels of β-catenin mRNA and protein in HKC-8 cells, respectively (Figure 8, D–F). Furthermore, knockdown of IGF2BP3 reduced the mRNA stability of β-catenin, as its steady-state level rapidly declined in IGF2BP3-depleted HKC-8 cells when new RNA transcription was inhibited by treatment with actinomycin D (Figure 8G). We found that IGF2BP3 and β-catenin mRNA and protein were colocalized in the tubular epithelial cells of the fibrotic kidney. As shown in Figure 8H, IGF2BP3 protein was colocalized with β-catenin mRNA by ISH on serial sections. Similarly, IGF2BP3 was colocalized with β-catenin protein by immunostaining (Figure 8I). These results suggest that IGF2BP3 promotes β-catenin expression by binding to and stabilizing its mRNA. As depicted in Figure 8J, IGF2BP3 binds to β-catenin mRNA, leading to its stabilization. This results in increased β-catenin protein expression and its subsequent activation, which in turn induces its downstream target genes, including IGF2BP3. ## Discussion As the principal mediator of canonical Wnt signaling, β-catenin plays a central role in promoting kidney fibrosis by controlling the expression of a wide variety of fibrosis-related genes [37]. The regulation of β-catenin is primarily controlled at the posttranslational level via protein modifications such as phosphorylation and ubiquitination [19]. In this study, we show a potentially novel mechanism of β-catenin regulation possibly through stabilization of its mRNA by IGF2BP3, an RNA-binding protein. We show that IGF2BP3 is induced predominantly in kidney tubular epithelium in all CKD models tested, including UUO, IRI, ADR, and Ang II infusion, as well as in human kidney biopsies of patients with various CKDs such as DN, CTIN, LN, IgAN, FSGS, and MN. Overexpression of IGF2BP3 impairs kidney integrity and drives renal fibrosis via activation of β-catenin. Interestingly, IGF2BP3 itself is a direct downstream target of Wnt/β-catenin, but it is also obligatory for Wnt signal transduction. Mechanistically, IGF2BP3 directly binds to β-catenin mRNA and increases its stability. Collectively, IGF2BP3 and β-catenin constitute a reciprocal feed-forward activation loop, which plays a central role in renal fibrogenesis. These studies provide potentially novel insights into the critical role and molecular mechanism of IGF2BP3 in kidney fibrogenesis and underscore the importance of mRNA regulation in β-catenin signaling. These findings expand our understanding of the complex regulation of Wnt/β-catenin and offer a potential new target for therapeutic intervention of fibrotic CKD. IGF2BPs, like other RNA-binding proteins, are a group of proteins that bind to the single- or double-stranded RNA in cells and participate in forming ribonucleoprotein complexes [25]. IGF2BPs contain various structural motifs, such as RNA recognition motif, RNA-binding domain, and K-homology domains [25]. They establish highly dynamic interactions with coding and noncoding RNAs and play a specific role in dictating the entire RNA life cycle from alternative splicing to nuclear export, storage, stabilization, subcellular localization, and degradation [25, 27]. As such, IGF2BPs are one of the major posttranscriptional regulators of gene expression for fine-tuning protein production. Although the role of IGF2BPs in tumorigenesis is increasingly recognized, the present study represents what we believe is the first comprehensive investigation of IGF2BPs in the pathogenesis of CKD in animal models and in humans. Among these IGF2BP proteins, IGF2BP3 is identified as the one with the most robust and consistent upregulation in diseased kidneys (Figure 1). The induction of IGF2BP1 appears disease specific, which occurs in the kidney after IRI but not UUO, while IGF2BP2 is only moderately induced in both UUO and IRI (Supplemental Figure 1). These observations prompted us to select IGF2BP3 for subsequent investigation in detail. It worthwhile to note that there is a high degree of similarity among these IGF2BPs, suggesting that they could share many biological functions. Consistent with this view, approximately $55\%$–$70\%$ of the recognized target RNAs are shared among 3 IGF2BPs [22]. Of interest, IGF2BP3 expression was not changed in the mouse model of AKI induced by cisplatin and only slightly increased after IRI (Supplemental Figure 2), suggesting that it is more relevant to kidney fibrogenesis than renal repair and regeneration after injury. Regarding the trigger and mechanism governing IGF2BP3 induction in CKD, the present study has illustrated that IGF2BP3 is a downstream target of Wnt/β-catenin (Figure 3). IGF2BP3 is expressed during embryogenesis but largely silent in adult tissues. It is reexpressed in numerous tumors [21, 22]. We show here that IGF2BP3 is induced in various animal models and human CKD, which is localized in the cytoplasm of tubular epithelial cells, consistent with previous studies in renal cell carcinoma [32, 34]. The expression pattern of IGF2BP3 is reminiscent of β-catenin in diseased kidney, raising the possibility that β-catenin may control its expression. Indeed, there were putative TBSs in the regulatory region of human, mouse, and rat IGF2BP3 genes. Human Wnt3a or constitutively active β-catenin induced IGF2BP3 expression. Furthermore, inhibition of β-catenin signaling by ICG-001, a small molecule that selectively inhibits β-catenin–mediated gene transcription, blocked IGF2BP3 expression both in vitro and in vivo. The finding that IGF2BP3 is a downstream target of Wnt/β-catenin is also consistent with earlier reports that IGF2BP1 and IGF2BP2 expression is correlated with β-catenin (38–41). One interesting finding of the present study is that IGF2BP3 elicits its profibrotic actions by activating β-catenin, suggesting that the interplay between IGF2BP3 and β-catenin is bidirectional. In fact, IGF2BP3 is absolutely required for proper Wnt signal transduction in kidney tubular cells, because knockdown of IGF2BP3 abolished β-catenin activation and its target genes’ expression in response to Wnt3a stimulation (Figure 5). As an RNA-binding protein, the action of IGF2BP3 relies on its interaction with its target RNAs [21, 22]. Indeed, RIP showed that IGF2BP3 directly binds to and stabilizes β-catenin mRNA in kidney tubular cells. Knockdown of IGF2BP3 decreased the stability and accelerated the degradation of β-catenin mRNA (Figure 8). Furthermore, IGF2BP3 is colocalized with β-catenin mRNA and protein in renal tubular epithelium of the diseased kidney (Figure 8). Taken together, IGF2BP3 and β-catenin create a reciprocal activation loop that synergistically promotes the pathogenesis of renal fibrosis. Our studies also uncover that regulation of mRNA stability could be an important and clinically relevant controlling mechanism of β-catenin signaling in kidney fibrosis. Notably, ectopic expression of IGF2BP3 in normal mice does not cause appreciable kidney injury, suggesting that it is required, but not sufficient, to initiate kidney damage under physiological conditions in vivo. The present study, however, leaves several questions unanswered. For example, the exact molecular mechanisms by which IGF2BP3 recognizes and regulates β-catenin mRNA remain elusive. Earlier studies have shown that IGF2BPs bind to their target RNAs at the 5′-untranslated region (UTR), the 3′-UTR, or coding regions by recognizing specific RNA motifs [42, 43]. The binding sites between IGF2BP3 and β-catenin mRNA in the setting of CKD remain unknown. Furthermore, a recent study has provided compelling evidence showing that IGF2BPs are identified as RNA N6-methyladenosine (m6A) readers and promote the stability of their mRNA targets in an m6A-dependent manner [28]. However, whether IGF2BP3-mediated β-catenin mRNA stabilization is also m6A-dependent needs to be elucidated. In addition, as IGF2BP3 can potentially bind to many RNA transcripts, the possibility also exists that it may promote kidney fibrosis by interacting with other RNA targets beyond β-catenin [28]. In this regard, however, IGF2BP3 does not affect the activity of TGF-β signaling (Supplemental Figure 3). The present study also has several limitations. For example, there is a lack of genetically modified mouse models such as IGF2BP3 knockout or conditional knockout mutants for strengthening the conclusion. Furthermore, only the UUO model was used in the investigation of the impact of different levels of IGF2BP3 on kidney fibrosis in vivo. Clearly, more studies are warranted in the future. In summary, the present study demonstrates that IGF2BP3, an RNA-binding protein, is upregulated in animal models and patients with CKD and critically involved in kidney fibrogenesis by activating β-catenin signaling. These findings underscore that regulation of β-catenin at the mRNA level, a process that was largely overlooked in the past, is an underappreciated dimension of the complex regulation of Wnt signaling. Although more studies are needed, our studies could provide a potential target for therapeutic intervention of fibrotic CKD. ## Animal models. Male C57BL/6 mice and BALB/c mice, weighing about 20–25 g, were obtained from the Experimental Animal Center of Southern Medical University in Guangzhou, China. For the UUO model, C57BL/6 mice were used as described previously [44]. Briefly, UUO was carried out under general anesthesia by ligating the left ureter via 4-0 silk after an abdominal midline incision. At 7 days after UUO, groups of mice were sacrificed, and kidney tissues were collected for subsequent analyses. To investigate renal expression of IGF2BP3 in different AKI and CKD models, mouse models of AKI including IRI and cisplatin injury and CKD including IRI, ADR nephropathy, and chronic Ang II infusion were used. For ischemic AKI mouse models, C57BL/6 mice were subjected to bilateral renal IRI by an established protocol as described previously [45]. Briefly, IRI was carried out by clamping bilateral renal pedicles for 30 minutes using microaneurysm clamps. During the ischemic period, body temperature was maintained at 37.5°C by using a temperature-controlled heating system. Mice were sacrificed at 24 hours after IRI and serum and kidney tissues collected for various analyses. For cisplatin injury, C57BL/6 mice were subjected to a single intraperitoneal injection of cisplatin (MilliporeSigma) at a dose of 20 mg/kg as described elsewhere [45]. Mice were sacrificed at 3 days after cisplatin injection, and serum and kidney samples were collected for various analyses. For ischemic CKD mouse models, C57BL/6 mice were subjected to unilateral renal IRI by an established protocol as described previously [7]. Briefly, IRI was carried out by clamping renal pedicles of the left kidney for 35 minutes using microaneurysm clamps. During the ischemic period, body temperature was maintained at 37.5°C by using a temperature-controlled heating system. At day 10, the contralateral, intact kidney was removed. At day 11 after IRI, groups of mice were sacrificed, and serum and kidney tissues were collected for various analyses. For the ADR model, BALB/c mice were administered with ADR at 10 mg/kg (doxorubicin hydrochloride; MilliporeSigma) by intravenous injection [46]. At 2 weeks after ADR injection, mice were euthanized, and kidney tissues were collected for various analyses. For the Ang II infusion model, C57BL/6 mice were implanted with osmotic minipumps (Model 2ML4; Alzet) subcutaneously for chronic administration of Ang II at 0.75 mg/kg/d [47]. At 4 weeks after Ang II infusion, mice were euthanized and kidney tissues collected for various analyses. ## Human kidney biopsy samples. Human kidney biopsy samples were obtained from diagnostic renal biopsies performed at the Nanfang Hospital, Southern Medical University, with written informed consent from the patients. Human normal kidney controls were obtained from nontumor renal tissues of patients who had renal cell carcinoma and underwent nephrectomy. Paraffin-embedded human kidney biopsy sections (3 μm) were prepared using a standard procedure and used for immunohistochemical staining. Quantification was assessed by a computer-aided point-counting technique. ## Cell culture and treatment. Human kidney proximal tubular cells (HKC-8) were provided by Lorraine C. Racusen (Johns Hopkins University, Baltimore, Maryland, USA). HKC-8 cells were grown in DMEM/Ham’s F12 medium supplemented with $10\%$ fetal bovine serum (FBS). HEK293T cells were obtained from the American Type Culture Collection and cultured in DMEM supplemented with $10\%$ FBS. These cells were cultured at 37°C in an atmosphere containing $5\%$ CO2. Serum-starved HKC-8 cells were treated by human recombinant Wnt3a protein (5036-WN-010; R&D Systems, Bio-Techne) at varying dosages in the serum-free medium for various periods as indicated. In some experiments, HKC-8 cells were pretreated with ICG-001 (HY-14428; MedChemExpress) at 10 μM for 1 hours, followed by incubation with vehicle or Wnt3a for an additional 2 days. Cells were then collected and subjected to various analyses. ## Lentivirus infection, plasmid transfection, and siRNA inhibition. The LV was used to express IGF2BP3 gene by infecting cells. The recombinant lentivirus vector expressing IGF2BP3, designated as LV-IGF2BP3, was constructed by Hanbio Biotechnology Co. Briefly, the full length of IGF2BP3 was inserted into a pHBLV-CMV-MCS-3xflag-EF1-ZsGreen-T2A-PURO vector. After validation by sequencing, lentivirus was packed, purified, and titrated. HKC-8 cells were infected with either LV-Ctrl (1.5 × 108 transduction units [TU]/mL) or LV-IGF2BP3 (3 × 108 TU/mL) overnight, then incubated for 2 days as indicated. The expression of the Flag-IGF2BP3 was assessed by Western blot analysis in HKC-8 cells. In some experiments, HKC-8 cells were transfected with pDel-β-cat using the Lipofectamine 2000 reagent (Invitrogen, Thermo Fisher Scientific) as previously reported [47]. The empty vector pcDNA3.1 (Invitrogen) was used as a mock transfection control. For knockdown of endogenous IGF2BP3 expression, HKC-8 cells were transfected with either control siRNA or IGF2BP3-specific siRNA. At 6 hours after transfection, cells were treated with or without Wnt3a (100 ng/mL) for another 2 days. The expression of the relevant proteins was assessed by Western blot analysis and immunofluorescence staining in HKC-8 cells. The sequences of siRNA used are described in Supplemental Table 2. ## Nuclear and cytoplasmic fractionation. Nuclear and cytoplasmic fractionation was performed with a commercial kit (BB-3102-50T; BestBio) according to the procedures specified by the manufacturer. ## Western blot analysis. Whole-kidney homogenates and cells were prepared with RIPA buffer containing $1\%$ NP-40, $0.1\%$ SDS, 100 μg/mL PMSF, and $1\%$ Halt Protease and Phosphatase Inhibitor Single-Use Cocktail (78442; Thermo Fisher Scientific) in PBS on ice. The supernatants were collected after centrifugation at 13,000g at 4°C for 15 minutes, and concentration was determined with BCA protein assay (K813-5000-1; BioVision). Gel electrophoresis was performed on reduced, denatured samples. After blotting onto PVDF microporous membranes (IPVH00010; MilliporeSigma) and blocking with $5\%$ milk, membranes were incubated with primary antibodies and HRP-conjugated secondary antibodies. The protein bands were visualized by SuperEnhanced chemiluminescence detection reagents (P1010; Applygen Technologies Inc.) and Kodak x-ray film. The primary and secondary antibodies used are listed in Supplemental Table 3. Relative protein levels of Western blots were quantified with densitometries, analyzed by ImageJ software (NIH), and reported after normalizing to the loading controls. ## qRT-PCR. Total RNA isolation was carried out using the TRIzol RNA Isolation Reagent (Life Technologies, Thermo Fisher Scientific). The first-strand cDNA synthesis was performed by using 2 μg of RNA in 20 μL of reaction buffer using a Reverse Transcription System kit (Promega). qRT-PCR was performed on ABI PRISM 7000 Sequence Detection System (Applied Biosystems, Thermo Fisher Scientific). The PCR reaction mixture was in a 25 μL volume including 12.5 μL 2× SYBR Green PCR Master Mix (Applied Biosystems, Thermo Fisher Scientific), 5 μL diluted reverse transcription product (1:10), and 0.5 μM sense and antisense primer sets. The mRNA levels of different genes were calculated after normalization with β-actin or GAPDH. The sequences of the PCR primer pairs used are described in Supplemental Table 4. ## Renal expression of IGF2BP3 via AAV9 vector. AAV9 vector was used as a gene transfer vehicle for overexpression of IGF2BP3 in the kidney. The recombinant AAV vector expressing IGF2BP3, designated as AAV-BP3, was constructed by Hanbio Biotechnology Co. Briefly, the full length of IGF2BP3 cDNA was inserted into a pHBAAV-CMV-MCS-3xflag-T2A-ZsGreen vector, and the accuracy of the inserted IGF2BP3 was validated by sequencing. AAV9 was packed, purified, and titrated. Male C57BL/6 mice were anesthetized and injected with AAV-Ctrl (1.9 × 1012 viral genomes [vg]/mL) or AAV-BP3 (1.7 × 1012 vg/mL) into 6 sites (10 μL at each site) of the left renal cortex with a glass micropipette [36]. Pilot experiments showed that exogenous IGF2BP3 expression was highest at 12 weeks after injection. Based on this information, UUO was performed at 12 weeks after intraparenchymal microinjection of AAV. Mice were divided into 4 groups ($$n = 5$$ per group): (i) sham+AAV-Ctrl, (ii) UUO+AAV-Ctrl, (iii) UUO+AAV-BP3, and (iv) sham+AAV-BP3. The detailed experimental design is presented in Figure 6A. Mice were sacrificed at 7 days after UUO, and kidney tissues were collected for various analyses. ## Knockdown of IGF2BP3 expression in vivo. Knockdown of IGF2BP3 expression in vivo was carried out by an shR-based inhibition. Briefly, IGF2BP3-specific shR expression plasmid (pLVX-IGF2BP3-shR) and empty vector (pLVX-shR) were administered into mice via tail vein injection, using an established hydrodynamics-based gene transfer approach, as described previously [7]. The sequence of mouse IGF2BP3 siRNA is listed in Supplemental Table 2. Plasmid injection was carried out at 2 days after UUO. Male C57BL/6 mice were divided into 3 groups ($$n = 5$$ per group): (i) sham, (ii) UUO+Ctrl-shR, and (iii) UUO+IGF2BP3-shR. The detailed experimental design is shown in Figure 7A. Mice were sacrificed 7 days after UUO, and the kidney tissues were collected for various analyses. ## Inhibition of β-catenin signaling in vivo. For assessing the effect of β-catenin signaling on IGF2BP3 expression in vivo, mice were divided into 3 groups ($$n = 5$$ per group): (i) sham, (ii) UUO injected with vehicle, and (iii) UUO injected with ICG-001. Mice were injected intraperitoneally with vehicle or ICG-001 at 5 mg/kg body weight once a day beginning from day 3 after UUO. Mice were sacrificed 7 days after UUO, and the kidney tissues were collected for various analyses. ## Determination of Scr levels. Scr levels were determined by an automatic chemistry analyzer (AU480; Beckman Coulter). The levels of Scr were expressed as milligrams per deciliter. ## Histology and immunohistochemical and immunofluorescence staining. Paraffin-embedded mouse and human kidney sections (3 μm thickness) were prepared by a routine procedure [48, 49]. The sections were stained with Sirius red or Masson’s trichrome, according to the manufacturer’s protocols, respectively. Immunohistochemical staining was performed using the established protocol as described previously [50]. Quantification of fibrotic lesions and positive protein area was assessed by a computer-aided point-counting technique. Briefly, about 10 random, nonoverlapping 40× cortical images were taken using an upright Olympus light microscope for each mouse and human. Integrated optical density of fibrotic lesions, positive protein area, and the whole field area were quantified using Image-Pro Plus (Media Cybernetics). The percentage of fibrotic lesions or positive protein area over the whole field area was calculated, respectively. The average value from 10 images for each mouse and human was used as the final value. HKC-8 cells cultured on coverslips were fixed with $4\%$ paraformaldehyde for 15 minutes at room temperature, then immersed in $0.2\%$ Triton X-100 for 10 minutes. After blocking with $10\%$ donkey serum for 1 hour, slides were immunostained with the specified antibodies, then stained with Cy2- or Cy3-conjugated secondary antibodies. Antibodies used are summarized in Supplemental Table 3. Nuclei were stained with DAPI (C1006; Beyotime Biotechnology) for 10 minutes. Images were captured under fluorescence microscopy (Leica SP8, Leica Microsystems). ## TOPFlash luciferase reporter assay. HEK293T cells were infected with either LV-Ctrl or LV-IGF2BP3 overnight and then incubated for 2 days as indicated. The HEK293T cells were then transfected with the TOPFlash luciferase reporter plasmid (0.9 μg, Addgene plasmid 12456) using Lipofectamine 2000 reagent. A fixed amount (0.1 μg) of internal control reporter Renilla reniformis luciferase driven under a thymidine kinase promoter (pRL-TK, Promega) was also cotransfected for normalizing the transfection efficiency. The luciferase assay was performed using the Dual-Luciferase Reporter Assay System kit according to the manufacturer’s protocols (E1910; Promega). Relative luciferase activity (arbitrary units) was calculated as fold-induction over controls after normalizing the transfection efficiency. ## RIP. HKC-8 cells at $80\%$~$90\%$ confluence were harvested by trypsinization and subjected to RIP. Briefly, RIP was carried out by using the RIP-Assay Kit (RN1001; MBL International) according to the manufacturer’s recommendations with normal rabbit IgG/RIP-certified IGF2BP3 antibody and protein A/G plus-agarose beads (Santa Cruz Biotechnology). Western blotting of a portion of washed RIP samples was performed to confirm specificity of IP. Following RNA isolation, RIP and input RNA were quantified. The IGF2BP3 RIP usually yielded more than 500 ng of RNA while IgG control was less than 50 ng. qRT-PCR was performed as previously described with the primers of CTNNB1 and 18S rRNA, using equal volumes of RIP eluate for IgG and IGF2BP3 for cDNA production. 18S rRNA was used for normalization. ## CTNNB1 mRNA stability. For assessing the stability of CTNNB1 mRNA, control and IGF2BP3-depleted HKC-8 cells were incubated with actinomycin D (HY-17559; MedChemExpress) at 5 μg/mL. Cells were then collected at different time points as indicated, and RNA was isolated for qRT-PCR. β-Actin was used for normalization. ## ISH. CTNNB1 probes (5′-ACTCAAGCTGATTTGATGGAGTTGGACATG-3′, 5′-GGGTTCAGATGATATAAATGTGGTCACCTG-3′, and 5′-TGCCTCCAGGTGACAGCAATCAGCTGGCCT-3′) were purchased from Boster. Paraffin-embedded kidney sections (3 μm) were fixed with $4\%$ paraformaldehyde, and ISH was performed with a commercial kit (Boster) according to the procedures specified by the manufacturer. ## Statistics. All data examined were expressed as mean ± SEM. Statistical analyses of the data were performed using SPSS 19.0. Comparisons between groups were made by 2-tailed t test or 1-way ANOVA followed by the Student-Newman-Keuls test. Spearman’s (nonparametric) correlation analysis was used to assess the relationship between IGF2BP3 area and other variables. $P \leq 0.05$ was considered significant. ## Study approval. All animal studies were approved by the Animal Ethics Committee at the Nanfang Hospital, Southern Medical University. Studies involving human samples, for which written informed consent was provided for use, were approved by the Medical Ethics Committee at the Nanfang Hospital, Southern Medical University. All animal studies were in accordance with the US Public Health Service Policy on Humane Care and Use of Laboratory Animals. ## Author contributions HF and Y Liu conceived the study. DS, HF, and Y Liu designed the experiments; DS, JS, Y Long, MZ, LL, JC, YX, HT, HZ, and XH performed the experiments; DS, HF, FFH, and Y Liu analyzed the data; FFH provided human kidney specimens; DS and Y Liu created the figures; DS and Y Liu wrote the manuscript; and all authors approved the final version of the manuscript. ## 12/15/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Jha V. Chronic kidney disease: global dimension and perspectives. Lancet (2013) 382 260-272. DOI: 10.1016/S0140-6736(13)60687-X 2. Romagnani P. Chronic kidney disease. Nat Rev Dis Primers (2017) 3. DOI: 10.1038/nrdp.2017.88 3. Humphreys BD. Mechanisms of renal fibrosis. Annu Rev Physiol (2018) 80 309-326. DOI: 10.1146/annurev-physiol-022516-034227 4. Liu Y. Cellular and molecular mechanisms of renal fibrosis. Nat Rev Nephrol (2011) 7 684-696. DOI: 10.1038/nrneph.2011.149 5. Duffield JS. Cellular and molecular mechanisms in kidney fibrosis. J Clin Invest (2014) 124 2299-2306. DOI: 10.1172/JCI72267 6. Liu BC. Renal tubule injury: a driving force toward chronic kidney disease. Kidney Int (2018) 93 568-579. DOI: 10.1016/j.kint.2017.09.033 7. Zhu H. Tenascin-C promotes acute kidney injury to chronic kidney disease progression by impairing tubular integrity via αvβ6 integrin signaling. Kidney Int (2020) 97 1017-1031. DOI: 10.1016/j.kint.2020.01.026 8. Lovisa S. Partial epithelial-to-mesenchymal transition and other new mechanisms of kidney fibrosis. Trends Endocrinol Metab (2016) 27 681-695. DOI: 10.1016/j.tem.2016.06.004 9. Lovisa S. Epithelial-to-mesenchymal transition induces cell cycle arrest and parenchymal damage in renal fibrosis. Nat Med (2015) 21 998-1009. DOI: 10.1038/nm.3902 10. Yang L. Epithelial cell cycle arrest in G2/M mediates kidney fibrosis after injury. Nat Med (2010) 16 535-543. DOI: 10.1038/nm.2144 11. Luo C. Wnt9a promotes renal fibrosis by accelerating cellular senescence in tubular epithelial cells. J Am Soc Nephrol (2018) 29 1238-1256. DOI: 10.1681/ASN.2017050574 12. Xu J. Cellular senescence in kidney fibrosis: pathologic significance and therapeutic strategies. Front Pharmacol (2020) 11. DOI: 10.3389/fphar.2020.601325 13. Kang HM. Defective fatty acid oxidation in renal tubular epithelial cells has a key role in kidney fibrosis development. Nat Med (2015) 21 37-46. DOI: 10.1038/nm.3762 14. Zuo Y, Liu Y. New insights into the role and mechanism of Wnt/β-catenin signalling in kidney fibrosis. Nephrology (carlton) (2018) 23 Suppl 4 38-43. PMID: 30298654 15. Zhou L, Liu Y. Wnt/β-catenin signalling and podocyte dysfunction in proteinuric kidney disease. Nat Rev Nephrol (2015) 11 535-545. DOI: 10.1038/nrneph.2015.88 16. Hao S. Targeted inhibition of β-catenin/CBP signaling ameliorates renal interstitial fibrosis. J Am Soc Nephrol (2011) 22 1642-1653. DOI: 10.1681/ASN.2010101079 17. Zhou L. Multiple genes of the renin-angiotensin system are novel targets of Wnt/β-catenin signaling. J Am Soc Nephrol (2015) 26 107-120. DOI: 10.1681/ASN.2014010085 18. Zhao Y. Wnt/β-catenin signaling mediates both heart and kidney injury in type 2 cardiorenal syndrome. Kidney Int (2019) 95 815-829. DOI: 10.1016/j.kint.2018.11.021 19. Schunk SJ. WNT-β-catenin signalling — a versatile player in kidney injury and repair. Nat Rev Nephrol (2021) 17 172-184. DOI: 10.1038/s41581-020-00343-w 20. Kahn M. Can we safely target the WNT pathway?. Nat Rev Drug Discov (2014) 13 513-532. DOI: 10.1038/nrd4233 21. Lederer M. The role of the oncofetal IGF2 mRNA-binding protein 3 (igf2bp3) in cancer. Semin Cancer Biol (2014) 29 3-12. PMID: 25068994 22. Mancarella C, Scotlandi K. IGF2BP3 from physiology to cancer: novel discoveries, unsolved issues, and future perspectives. Front Cell Dev Biol (2019) 7. PMID: 32010687 23. Huang X. Insulin-like growth factor 2 mRNA-binding protein 1 (IGF2BP1) in cancer. J Hematol Oncol (2018) 11. DOI: 10.1186/s13045-018-0628-y 24. Cao J. The roles of insulin-like growth factor 2 mRNA-binding protein 2 in cancer and cancer stem cells. Stem Cells Int (2018) 2018. PMID: 29736175 25. Bell JL. Insulin-like growth factor 2 mRNA-binding proteins (IGF2BPs): post-transcriptional drivers of cancer progression?. Cell Mol Life Sci (2013) 70 2657-2675. DOI: 10.1007/s00018-012-1186-z 26. Degrauwe N. IMPs: an RNA-binding protein family that provides a link between stem cell maintenance in normal development and cancer. Genes Dev (2016) 30 2459-2474. DOI: 10.1101/gad.287540.116 27. Pereira B. RNA-binding proteins in cancer: old players and new actors. Trends Cancer (2017) 3 506-528. DOI: 10.1016/j.trecan.2017.05.003 28. Huang H. Recognition of RNA N. Nat Cell Biol (2018) 20 285-295. DOI: 10.1038/s41556-018-0045-z 29. Christiansen J. IGF2 mRNA-binding protein 2: biological function and putative role in type 2 diabetes. J Mol Endocrinol (2009) 43 187-195. DOI: 10.1677/JME-09-0016 30. Dai N. The diverse functions of IMP2/IGF2BP2 in metabolism. Trends Endocrinol Metab (2020) 31 670-679. DOI: 10.1016/j.tem.2020.05.007 31. Wang S. RNA binding proteins control transdifferentiation of hepatic stellate cells into myofibroblasts. Cell Physiol Biochem (2018) 48 1215-1229. DOI: 10.1159/000491987 32. Jiang Z. Analysis of RNA-binding protein IMP3 to predict metastasis and prognosis of renal-cell carcinoma: a retrospective study. Lancet Oncol (2006) 7 556-564. DOI: 10.1016/S1470-2045(06)70732-X 33. Hoffmann NE. External validation of IMP3 expression as an independent prognostic marker for metastatic progression and death for patients with clear cell renal cell carcinoma. Cancer (2008) 112 1471-1479. DOI: 10.1002/cncr.23296 34. Tschirdewahn S. Circulating and tissue IMP3 levels are correlated with poor survival in renal cell carcinoma. Int J Cancer (2019) 145 531-539. DOI: 10.1002/ijc.32124 35. Gu Y. DMDRMR-mediated regulation of m. Cancer Res (2021) 81 923-934. DOI: 10.1158/0008-5472.CAN-20-1619 36. Liu T. MicroRNA-493 targets STMN-1 and promotes hypoxia-induced epithelial cell cycle arrest in G. FASEB J (2019) 33 1565-1577. DOI: 10.1096/fj.201701355RR 37. Tan RJ. Wnt/β-catenin signaling and kidney fibrosis. Kidney Int Suppl (2011) (2014) 4 84-90. DOI: 10.1038/kisup.2014.16 38. Noubissi FK. CRD-BP mediates stabilization of betaTrCP1 and c-myc mRNA in response to beta-catenin signalling. Nature (2006) 441 898-901. DOI: 10.1038/nature04839 39. Gu W. Feedback regulation between zipcode binding protein 1 and beta-catenin mRNAs in breast cancer cells. Mol Cell Biol (2008) 28 4963-4974. DOI: 10.1128/MCB.00266-08 40. Huang Q. A novel circular RNA, circXPO1, promotes lung adenocarcinoma progression by interacting with IGF2BP1. Cell Death Dis (2020) 11. DOI: 10.1038/s41419-020-03237-8 41. Xing M. Overexpression of p62/IMP2 can promote cell migration in hepatocellular carcinoma via activation of the Wnt/β-catenin pathway. Cancers (Basel) (2019) 12. DOI: 10.3390/cancers12010007 42. Hafner M. Transcriptome-wide identification of RNA-binding protein and microRNA target sites by PAR-CLIP. Cell (2010) 141 129-141. DOI: 10.1016/j.cell.2010.03.009 43. Schneider T. Combinatorial recognition of clustered RNA elements by the multidomain RNA-binding protein IMP3. Nat Commun (2019) 10. DOI: 10.1038/s41467-019-09769-8 44. Fu H. Tenascin-C is a major component of the fibrogenic niche in kidney fibrosis. J Am Soc Nephrol (2017) 28 785-801. DOI: 10.1681/ASN.2016020165 45. Hu C. Matrix metalloproteinase-10 protects against acute kidney injury by augmenting epidermal growth factor receptor signaling. Cell Death Dis (2021) 12. DOI: 10.1038/s41419-020-03301-3 46. Chen Q. MicroRNA-466o-3p mediates β-catenin-induced podocyte injury by targeting Wilms tumor 1. FASEB J (2020) 34 14424-14439. DOI: 10.1096/fj.202000464R 47. Li Z. (Pro)renin receptor is an amplifier of Wnt/β-catenin signaling in kidney injury and fibrosis. J Am Soc Nephrol (2017) 28 2393-2408. DOI: 10.1681/ASN.2016070811 48. Yuan Q. A Klotho-derived peptide protects against kidney fibrosis by targeting TGF-β signaling. Nat Commun (2022) 13. DOI: 10.1038/s41467-022-28096-z 49. Mo H. CXCR4 induces podocyte injury and proteinuria by activating β-catenin signaling. Theranostics (2022) 12 767-781. DOI: 10.7150/thno.65948 50. Li L. Fibrillin-1-enriched microenvironment drives endothelial injury and vascular rarefaction in chronic kidney disease. Sci Adv (2021) 7. DOI: 10.1126/sciadv.abc7170
--- title: MRAP2 regulates energy homeostasis by promoting primary cilia localization of MC4R authors: - Adelaide Bernard - Irene Ojeda Naharros - Xinyu Yue - Francois Mifsud - Abbey Blake - Florence Bourgain-Guglielmetti - Jordi Ciprin - Sumei Zhang - Erin McDaid - Kellan Kim - Maxence V. Nachury - Jeremy F. Reiter - Christian Vaisse journal: JCI Insight year: 2023 pmcid: PMC9977312 doi: 10.1172/jci.insight.155900 license: CC BY 4.0 --- # MRAP2 regulates energy homeostasis by promoting primary cilia localization of MC4R ## Abstract The G protein–coupled receptor melanocortin-4 receptor (MC4R) and its associated protein melanocortin receptor–associated protein 2 (MRAP2) are essential for the regulation of food intake and body weight in humans. MC4R localizes and functions at the neuronal primary cilium, a microtubule-based organelle that senses and relays extracellular signals. Here, we demonstrate that MRAP2 is critical for the weight-regulating function of MC4R neurons and the ciliary localization of MC4R. More generally, our study also reveals that GPCR localization to primary cilia can require specific accessory proteins that may not be present in heterologous cell culture systems. Our findings further demonstrate that targeting of MC4R to neuronal primary cilia is essential for the control of long-term energy homeostasis and suggest that genetic disruption of MC4R ciliary localization may frequently underlie inherited forms of obesity. ## Introduction The regulation of food intake and energy expenditure is dependent on the genetic, molecular, and cellular integrity of the central melanocortin system, a network of hypothalamic neurons that integrate peripheral information about energy status and regulate long-term energy homeostasis, thereby preventing obesity [1]. This system comprises the neuropeptides α-MSH and AGRP, produced by independent populations of neurons in the arcuate nucleus, which are sensitive to the adipocyte-secreted hormone leptin, as well as the receptor for these neuropeptides, the melanocortin-4 receptor (MC4R). MC4R, 1 of 5 members of the Gαs-coupled melanocortin receptor family [2], is found in multiple brain regions, but its expression in the paraventricular nucleus (PVN) of the hypothalamus is both necessary and sufficient for the regulation of food intake and body weight [3, 4]. Underscoring the essential role of this receptor in the maintenance of energy homeostasis, heterozygous loss-of-function mutations in MC4R are the most common cause of monogenic obesity in humans (5–7), and the MC4R locus displays the second strongest association with obesity among the common variants influencing BMI (8–10). Mc4r+/– and Mc4r–/– mice recapitulate the obesity phenotypes observed in humans [11]. The central importance of MC4R in energy homeostasis has made it a major target for the pharmacotherapy of obesity. However, little is known about the molecular and cellular pathways underlying the maintenance of long-term energy homeostasis by MC4R-expressing neurons. Recently, we reported that MC4R localizes and functions at the neuronal primary cilium [12, 13], a cellular organelle that projects from the surface of most mammalian cell types and functions as an antenna to sense extracellular signals [14]. Inherited mutations that disrupt ciliary structure or function cause ciliopathies (15–17), disorders characterized by pleiotropic clinical features that can include hyperphagia and severe obesity [18], such as in Alström or in Bardet-Biedl syndrome (BBS). In adult mice, genetic ablation of neuronal primary cilia also causes obesity [19]. MC4R physically interacts with melanocortin receptor–associated protein 2 (MRAP2), a member of the MRAP family composed of single-pass transmembrane proteins that interact with melanocortin receptors [20] and other GPCRs [21, 22]. In nonciliated heterologous systems, MRAP2 binds to MC4R and increases ligand sensitivity, as well as MC4R-mediated generation of cAMP (23–25). In humans, MRAP2 variants were found in patients with obesity (23, 26–28). Mrap2–/– mice develop severe obesity, although they lack the early-onset hyperphagia of Mc4r–/– mice [23], which brings into question the extent to which MRAP2 qualitatively and quantitatively interacts with MC4R in vivo. Indeed, MRAP2 has also been suggested to interact with other GPCRs such as the ghrelin receptor and the prokineticin receptor [21, 22]. Here, we find that the central mechanism of MRAP2-associated obesity is the critical role for MRAP2 in targeting MC4R to cilia. ## MC4R neurons require MRAP2 to regulate energy homeostasis. Germinal deletion of MRAP2, in Mrap2–/– mice, leads to obesity, although to a lesser extent than that observed in Mc4r–/– mice [23]. To determine whether MRAP2 is essential for MC4R neurons to control food intake and body weight, we deleted MRAP2 from MC4R-expressing cells. We obtained mice bearing an Mrap2-KO–first (“tm1a”) allele and confirmed that, as previously reported [20, 23, 29], homozygous mutant mice (hereafter referred to as Mrap2–/–) (Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.155900DS1) were obese and hyperphagic at 12 weeks of age but did not differ from their WT littermates at 4 weeks of age (Supplemental Figure 1, B and C). From these mice, we generated an Mrap2-floxed allele (“tm1c” allele, hereafter referred to as Mrap2fl) (Supplemental Figure 1A). Mrap2fl/fl mice weighed the same as their WT littermates (Supplemental Figure 1D). To specifically delete MRAP2 in MC4R-expressing neurons throughout the brain, we generated Mc4r-t2a-Cre [13] Mrap2fl/fl mice (hereafter referred to as Mc4rt2aCre/t2aCre Mrap2fl/fl). Mc4rt2aCre/t2aCre Mrap2fl/fl mice fed ad libitum regular chow developed early-onset obesity with significantly higher body weight (Figure 1, A–F) and fat mass (Figure 1, K–R), associated with hyperphagia (Figure 1, S–V) compared with their Mc4rt2aCre/t2aCre Mrap2+/+ littermates. This phenotype was apparent at 4 weeks of age (Figure 1, C, E, K, M, O, Q, S, and U) and was present both in females and in males (Figure 1). No differences in energy expenditure (EE) or activity were observed between groups, both in females and in males (Supplemental Figure 2). Together, these data demonstrate that MRAP2 is essential in MC4R-expressing neurons to regulate food intake and body weight. ## MRAP2 promotes MC4R targeting to the primary cilium in heterologous cells. The interaction between MRAP2 and MC4R, as well as the cellular localization of MRAP2, have been previously studied in nonciliated cells (23–26, 30). Since MC4R localizes to the primary cilium, and MRAP2 has been reported to interact with MC4R in vitro [23], we tested whether MRAP2 colocalizes with MC4R at the primary cilium following transient transfection in murine ciliated inner medullary collecting duct 3 (IMCD3) cells. We found that both MRAP2 and MC4R colocalized with the ciliary shaft labeled by acetylated tubulin (AcTub; Supplemental Figure 3B) and that ciliary localization of MC4R was strongly increased when MRAP2 and MC4R were cotransfected (Figure 2A, top panel). We then determined whether MRAP2 localization is a common feature of members of the MRAP family and found that the MRAP1, a paralog of MRAP2 and an essential accessory factor for the functional expression of the MC2R/ACTH receptor, did not localize to the primary cilium (Figure 2, A and B, and Supplemental Figure 3, A and B). We further tested to what extent MRAP2 promotes MC4R ciliary localization and found that MRAP2 increased MC4R enrichment at the primary cilium, whereas MRAP1 had no effect (Figure 2, A and B, top panel). Finally, we assessed the specificity of the interaction between MRAP2 and MC4R by systematically testing the ciliary enrichment of all 5 melanocortin receptor family members in the presence or absence of MRAPs ($$n = 30$$; Figure 2B and Supplemental Figure 3C). Although MRAP2 promoted the ciliary localization of other melanocortin receptors, the greatest effect was observed on MC4R (Figure 2B and Supplemental Figure 3C). MRAP1 did not affect ciliary enrichment of any of the receptors (Figure 2B). Thus, MRAP2 specifically promotes MC4R localization to the primary cilia in heterologous cells. ## MRAP2 colocalizes with MC4R at the primary cilium of hypothalamic neurons in vivo. MRAP2 subcellular localization was assessed by immunofluorescence, using a commercial antibody, the specificity of which was initially confirmed in vitro in transfected cells (Supplemental Figure 4A) and in vivo compared with MRAP2-KO sections (Supplemental Figure 4B). We first examined the subcellular localization of MRAP2 in a transgenic reporter mouse in which all primary cilia are labeled (Arl13b-GFPtg) [31]. In these mice, MRAP2 was found to colocalize with some but not all primary cilia in the PVN of the hypothalamus (Figure 3, A–C). To further determine whether MRAP2 colocalizes with MC4R at primary cilia, we used a mouse line in which GFP was fused to the C-terminus of MC4R at the endogenous locus [12] (hereafter referred to as Mc4rgfp), allowing for the assessment of the subcellular localization of the endogenous MC4R. In the PVN of Mc4rgfp mice, most MC4R localized at primary cilia (Supplemental Figure 5), and remarkably, MRAP2 localized exclusively to the cilium of these cells (Figure 3, D–F). Since MC4R is widely expressed throughout the brain, we sought to determine whether MC4R colocalizes with MRAP2 at the cilium in other brain regions. We found that, although MRAP2 expression is broader than MC4R’s, MC4R always colocalizes with MRAP2 at the primary cilium in other nuclei where MC4R is expressed (Supplemental Figure 6). ## MRAP2 is required for MC4R localization to primary cilia in vivo. Since MRAP2 enhances MC4R localization at primary cilia in vitro, and MC4R colocalizes with MRAP2 at primary cilia in vivo, we tested whether loss of MRAP2 compromises MC4R localization at primary cilia in vivo. We generated Mrap2+/+ Mc4rgfp and Mrap2–/– Mc4rgfp mice to compare the ciliary localization of MC4R in the presence and absence of MRAP2. In the PVN of Mrap2+/+ Mc4rgfp mice, MC4R mainly localized to cilia (Figure 4A). Remarkably, in Mrap2–/– Mc4rgfp mice, MC4R was rarely found at primary cilia and, in some cells, was detectable in the neuronal cell bodies (Figure 4B). Of note, we confirmed that no MRAP2 staining was detected in Mrap2–/– Mc4rgfp mice — in particular, in the few cells in which MC4R localized to primary cilia — further confirming the specificity of the anti-MRAP2 antibody (Supplemental Figure 4B). The normalized intensity of MC4R at cilia (defined by ADCY3 immunostaining) was decreased in Mrap2 mutants compared with WT (Figure 4C), and MC4R was no longer enriched in cilia in the absence of MRAP2 (Figure 4D). Thus, MRAP2 is necessary for MC4R enrichment at primary cilia both in vitro and in vivo. Lack of MC4R localization to primary cilia in Mrap2–/– Mc4rgfp mice could be secondary, resulting from developmental effects. We therefore determined whether acute deletion of Mrap2 in the PVN of adult mice would affect MC4R ciliary localization. We injected Mrap2fl/fl Mc4rgfp mice unilaterally with an adeno associated virus (AAV) encoding mCherry-IRES-Cre. Mrap2 was deleted from the infected PVN by Cre-mediated recombination, and the contralateral uninfected PVN served as an internal control (Figure 5, A and B). The brains were harvested 3 weeks following the AAV injections and were analyzed. While MRAP2 localized to cilia in the contralateral control PVN, MRAP2 staining was not detected in the mCherry-Cre–injected side, confirming deletion of the gene ($$n = 4$$; Figure 5, B and C). Cilia, identified by ADCY3 staining, were present in the infected PVN (Figure 5D), confirming that MRAP2 is dispensable for PVN primary cilia maintenance. Remarkably, MC4R localized to cilia in control PVNs but not in the PVNs from which MRAP2 had been removed (Figure 5, D–F), confirming that MRAP2 is essential for MC4R ciliary localization. To further assess the specificity of MRAP2 in mediating the localization of MC4R to primary cilia, we tested whether MRAP2 contributes to the ciliary localization of other GPCRs. Specifically, we assessed the localization of the somatostatin receptor 3 (SSTR3) in the presence or absence of MRAP2. SSTR3 is a cilia-localized GPCR that is also expressed in the PVN [32], including a subset of MRAP2-expressing neurons (Supplemental Figure 7, D and E). Interestingly, the absence of MRAP2 did not affect ciliary localization of SSTR3 (Supplemental Figure 8). ## Discussion Ablation of primary cilium by conditionally knocking out Ift88 in adult mice leads to obesity — either when deleted ubiquitously, specifically in neurons, or only in the PVN — directly implicating PVN cilia in the control of feeding behavior [13, 19]. The composition of the primary ciliary membrane is different from that of the surrounding plasma membrane, as it is enriched for proteins involved in specific forms of signaling [33]. We previously identified MC4R as one of a select subset of GPCRs that localizes to cilia. Moreover, antagonizing Gαs signaling specifically at the primary cilium of MC4R-expressing PVN neurons also leads to obesity [13], suggesting that not only does MC4R localize to cilia, but it functions in cilia to mediate energy homeostasis. Importantly, human obesity-associated MC4R mutations affecting its third intracellular loop (a domain previously implicated in the ciliary localization of other GPCRs; ref. 34) impair MC4R ciliary localization without affecting its trafficking to the cell membrane or its ability to couple to G proteins [12]. In the present study, we demonstrate that a ciliary GPCR accessory protein, MRAP2, restrains feeding by acting in MC4R-expressing neurons to direct its associated GPCR, MC4R, to cilia. These studies provide multiple concordant lines of evidence demonstrating that ciliary localization is critical for MC4R function in humans and mice. Since MC4R and MRAP2 are essential for suppressing feeding behavior and since MC4R fails to localize to primary cilia in Mrap2–/– mice, it is likely that the failure of MC4R to localize to cilia accounts for the obesity observed in MRAP2-deficient mice and humans. Previous in vitro studies in unciliated cells have reported conflicting effects of MRAP2 on MC4R function: one study reported that MRAP2 does not affect MC4R activity [30], another suggested that MRAP2 may inhibit MC4R activity [24], and others reported that MRAP2 increases its response to ligand [23, 25]. MRAP2 was also reported to either positively or negatively regulate MC4R activity depending on their relative concentrations [26]. Similarly, MRAP2 has been reported to either decrease [24] or modestly increase [25] MC4R cell-surface expression. It will be of interest to repeat these assays in ciliated cells. The ciliary targeting of class A GPCRs such as SSTR3, D1R, MCHR1, and HTR6 depends on interactors such as Tubby family members and the IFT-A complex (32, 35–37). In contrast, the dependence of MC4R on MRAP2 for ciliary trafficking reveals a more specific requirement. For example, we found that, although MRAP2 and SSTR3 colocalize at primary cilia of some PVN neurons, SSTR3 does not require MRAP2 for ciliary localization. Therefore, MRAP2 may be a ciliary trafficking chaperone for MC4R, rather than a component of the general ciliary trafficking machinery. Ciliary accessory proteins such as MRAP2 may represent a site of regulation for the ciliary localization and function of their associated receptors. Although MRAP2 is not required generally for ciliary GPCR localization, it may associate with other GPCRs beyond MC4R, including GHSR1a [22] and PKR1 [21, 21]. Indeed, association of MRAP2 with receptors other than MC4R could also explain the differences in weight phenotypes caused by removing Mrap2 globally or specifically in MC4R-expressing neurons. Specifically, we found that inactivating Mrap2 specifically in MC4R-expressing neurons causes, like Mc4R loss-of-function, early-onset hyperphagia and obesity, whereas germline inactivation of Mrap2 caused a milder phenotype, with late-onset hyperphagia and obesity [23, 29]. Therefore, whether MRAP2 also promotes the ciliary function of orexigenic receptors should be assessed. The role of primary cilia in metabolism (12, 18, 38–40) has motivated screens in heterologous cell culture systems to identify ciliary GPCRs that control energy homeostasis. Previously described ciliary GPCRs include the melanin-concentrating hormone receptor 1 (MCHR1) [34] and the neuropeptide y receptor 2 (NPY2R) [41, 42], which have also been shown to be implicated in energy homeostasis. However, the necessity of an accessory protein like MRAP2 for ciliary trafficking of specific GPCRs was not considered, suggesting that these screens may have missed a number of ciliary GPCRs, including MC4R [41, 42]. Whether other ciliary GPCRs use accessory proteins for ciliary trafficking is an open question but is hinted at by studies demonstrating that proteins, such as rhodopsin, localize robustly to cilia in their native cell types but less well in heterologous ciliated cells [43]. Our findings, therefore, suggest that it will be essential to consider the role of accessory proteins in future work both trying to uncover new ciliary receptor and working with these receptors in vitro to get as close as possible to physiological conditions. Our study reveals that MRAP2 is required for the localization of MC4R to the primary cilia and the function of MC4R neurons. Within the context of the emerging connection between primary cilia and energy homeostasis (12, 18, 38–40), the observations described here further suggests that all genes controlling localization of MC4R to primary cilia are candidate genes for human obesity. ## Expression plasmids. MC1R-GFP, MC2R-GFP, MC3R-GFP, and MC5R-GFP expression constructs were constructed as previously described for MC4R-GFP [44]. Plasmids encoding MRAP1-FLAG and MRAP2-FLAG were obtained from Patricia M Hinkle (University of Rochester Medical Center, Rochester, New York, USA) [30]. ## Ciliary expression of MCRs and MRAPs in cultured cells. All IMCD3 cells were generated from a parental mouse IMCD-Flpln line (Thermo Fisher Scientific). IMCD3 cells were transfected using X-tremeGENE 9 DNA Transfection Reagent (06365809001, Roche). The transfection reagent was diluted in OptiMEM (Invitrogen) and incubated at room temperature for 5 minutes. Then, the mixture was added to the diluted plasmids in a 6:1 ratio (6 μL transfection reagent to 1 μg DNA) and incubated at room temperature for 20 minutes. In total, 50,000 cells in suspension were added to the transfection mixture in a 24-well plate. Transfected cells were switched to starvation media after 24 hours and fixed 16 hours afterward. Double plasmid transfections were performed by diluting equal mass of each vector. Three independent transfections were carried out per condition. ## Cell imaging. In total, 50,000 cells were seeded for transfection on acid-washed 12 mm #1.5 cover glass (Fisherbrand, Thermo Fisher Scientific) in a 24-well plate. Starved cells were fixed in phosphate buffered saline (PBS) containing $4\%$ paraformaldehyde (Electron Microscopy Sciences) for 15 minutes at room temperature and permeabilized in ice-cold $100\%$ methanol (Thermo Fisher Scientific) for 5 minutes. Cells were then further permeabilized in PBS containing $0.1\%$ Triton X-100 (BP151-500, Thermo Fisher Scientific), $5\%$ normal donkey serum (017-000-121, Jackson ImmunoResearch), and $3\%$ BSA (BP1605-100, Thermo Fisher Scientific) for 30 minutes. Permeabilized cells were incubated with specified antibodies (i.e., mouse IgG2b monoclonal anti–acetylated tubulin antibody [T6793, Sigma-Aldrich, 1:1,000] and mouse IgG1 monoclonal anti–FLAG M2 antibody [F1804, Sigma-Aldrich, 1:500]) for 1 hour, washed with PBS, and incubated with dye-coupled secondary antibodies (Jackson ImmunoResearch) for 30 minutes. Cells were then washed with PBS, stained with Hoechst DNA dye, and washed with PBS before mounting with Fluoromount G (Electron Microscopy Sciences). The anti-MRAP2 antibody was validated in a stable clonal cell line expressing MC4R-3xNeonGreen and MRAP2-3xFlag. The 2 proteins are produced in stoichiometric amount as they are encoded within a single mRNA and separated by a self-cleaving T2A peptide (pEFαΔTATA:MRAP2-3xFLAG-T2A-IgK-HA-Halo-MC4R-3xmNeonGreen). The parental line, IMCD3 FlpIn, was used as negative control. Both cell lines were stained as mentioned above with rabbit anti-MRAP2 (17259-1-AP, Proteintech, 1:50) and mouse anti–acetylated tubulin (6-11-B, T6793, Sigma, 1:1,000). MC4R-3xNeonGreen was detected by endogenous fluorescence. Cells were imaged in a widefield fluorescence DeltaVision microscope (Applied Precision) equipped with a PlanApo 60×/1.40NA objective lens (Olympus), a pco.edge 4.2 sCMOS camera, a solid state illumination module (Insight), and a Quad polycroic (Chroma). Z stacks with 0.2 μm separation between planes were acquired using SoftWoRx. The illumination settings were: 140 μW 390 nm wavelength for 0.15 seconds to image Hoechst, 222 μW 475 nm wavelength for 0.3 seconds to image Alexa Fluor 488–stained MCRs, 123 μW 543 nm wavelength for 0.3 seconds to image Cy3-stained acetylated tubulin, and 115 μW 632 nm wavelength for 0.15 seconds to image Cy5-stained FLAG-labeled MRAPs. Images were flat field corrected, background subtracted, and maximally projected using Fiji. Ciliary intensity measurements were also taken in Fiji as previously published [45]. To compensate for the cell-to-cell differences in expression driven by the transient nature of the transfection, we normalized the ciliary intensity to the intensity at the cytoplasm using an ROI of equal area to that of the cilium. ## Animals. Mice were housed in a barrier facility and maintained on a 12:12 light cycle (from 7 a.m. to 7 p.m.) at an ambient temperature of 23°C ± 2°C and relative humidity $50\%$–$70\%$. Mice were fed with rodent diet 5058 (Lab Diet) and group housed up to 5. Experiments were performed with weight-matched littermates. ## Descriptions on mouse lines used. The Mc4rtm1(egfp)Vai mice have an EGFP tag inserted in frame at the C-terminus of the endogenous Mc4r locus (“Mc4rgfp”; ref. 12). The Mc4rtm2(t2a Cre)Vai mice have a -t2a-Cre sequence inserted in frame at the C-terminus of the endogenous Mc4r locus (MC4Rt2aCre; ref.13). Mice carrying a Flip recombinase (Tg[ACTFLPe]9205Dym) express a FLP1 recombinase gene under the direction of the human ACTB promoter (The Jackson Laboratory). Arl13B-GFPtg transgenic mice ubiquitously expresses a GFP-tagged version of the ciliary protein ARL13B, under the control of the CAG promoter (Markus Delling, UCSF, San Francisco; ref. 31). ## EUCOMM MRAP2 mice. EUCOMM Mrap2-KO–first allele (“tm1a”) mice carry an frt-flanked β-gal gene and neo cassette preventing widespread expression of the Mrap2 gene (EUCOMM tm1a allele or Mrap2–/–, C57BL/6-Mrap2tm1a(EUCOMM)Wtsi/Jsbgj, The Jackson Laboratory). When mice harboring this allele are crossed into an actin-FLpE background, the frt-flanked cassette is excised and MRAP2 WT function is restored (EUCOMM tm1c allele or Mrap2fl/fl). After Flip-mediated excision, a loxP-flanked Exon 4 remains, which allows for MC4R cell-specific deletion when crossed to Mc4r-t2a-Cre–knock-in mice (Mc4rt2aCre/t2aCre Mrap2tm1c/tmc1 or Mc4rt2aCre/t2aCre Mrap2fl/fl). We obtained Mc4rt2aCre/t2aCre Mrap2fl/fl and Mc4rt2aCre/t2aCre Mrap2+/+ littermates by crossing Mc4rt2aCre/t2aCre Mrap2fll+ mice. All mice were maintained on a mixed background. ## Stereotaxic AAV-injection surgeries Eight-week-old Mrap2fl/fl Mc4rgfp females mice ($$n = 4$$) were injected unilaterally with pAAV-Ef1a-mCherry- IRES-CRE (Addgene, 55632-AAV8; http://n2t.net/addgene:55632; RRID:Addgene_55632). Animals were anesthetized with an initial flow of $4\%$ isoflurane (Dechra), maintained under anesthesia using $2\%$ isoflurane, and kept at 30°C–37°C using a custom heating pad. The surgery was performed using aseptic and stereotaxic techniques. Briefly, the animals were put into a stereotaxic frame (KOPF Model 1900), the scalp was opened, the planarity of the skull was adjusted, and a hole was drilled (PVN coordinates: AP = –0.8, ML = –0.2, DV = –5.3). A volume of 300 nL was injected at a rate of 0.1 μL/min. Animals were given preoperative analgesic (buprenorphine, 0.3 mg/kg, Covetrus) and postoperative antiinflammatory meloxicam (5 mg/Kg, Pivetal) and were allowed to recover at least 10 days, during which time they were single housed and handled frequently. The mice were calorie restricted at $75\%$ for 2 weeks prior to perfusion. ## Brain imaging Mice were perfused transcardially with PBS, followed by $4\%$ paraformaldehyde fixation solution. Brains were dissected and postfixed in fixation solution at 4°C overnight, soaked in $30\%$ sucrose solution overnight, embedded in OCT (Tissue-Tek), frozen, and cut into 20–35 μm coronal sections before being stored at –80°C until staining. After washing, sections were blocked for 1 hour in $50\%$ serum (goat, MilliporeSigma, S26; donkey, MilliporeSigma, D9663), $50\%$ antibody buffer ($1.125\%$ NaCl [MilliporeSigma, S5886], $0.75\%$ Tris base (Thermo Fisher Scientific, BP152), $1\%$ BSA (Thermo Fisher Scientific, BP1600), 1.8 % L-Lysine (Alfa Aesar, J62225), and $0.04\%$ sodium azide (Millipore Sigma), followed by incubation with primary antibody overnight at 4°C (chicken anti-GFP, Abcam, ab13970, 1:250; rabbit anti-Adcy3, Santa Cruz Biotechnology, sc-588, 1:500; rabbit anti-MRAP2, Proteintech, 17259-1-AP, 1:200; and goat anti-SSTR3 [M-18], Santa Cruz Biotechnology, sc-11617, 1:200). After washing, sections were incubated with secondary antibodies for 1 hour at room temperature (goat anti–chicken Alexa Fluor 488 [Invitrogen, A11039]; goat anti–rabbit Alexa Fluor 633 [Invitrogen, A21070] or 555 [Invitrogen, A21429]; donkey anti–chicken Alexa Fluor 488 [Invitrogen, A78948]; donkey anti–rabbit Alexa Fluor 647 [Invitrogen, A-31573]; donkey anti–goat Alexa Fluor 555 [Invitrogen, A21432]; all 1:500), washed, and stained with Hoechst (1:5,000; Invitrogen, H3570). They were then washed and mounted with Prolong Diamond antifade Mountant (Invitrogen, P36970). The anti-MRAP2 antibody was validated in vitro in transfected cells (Supplemental Figure 4A) and in vivo on brain sections from Mrap2 WT mice compared with MRAP2-KO sections (Supplemental Figure 4B). In Figure 5, the immunofluorescence stainings were performed on brain sections from mice that were calorie restricted for a week at $75\%$ of baseline food intake on regular chow and fasted for 24 hours prior to perfusion. In Supplemental Figure 4 (validation of MRAP2 antibody), the staining for MRAP2 and GFP was performed on brain sections from mice expressing MC4R-GFP, either WT or KO for MRAP2 (MRAP21a/1a). In Figures 3 and 4, and in Supplemental Figures 4 and 5, images were taken of brains from P6 pups, and from P8 pups in Supplemental Figures 6–8. The brain of 1 P8 mouse was assessed in Supplemental Figure 6. The sex of P6 and P8 mice could not be determined because the anogenital distance is hardly measurable at such a young postnatal age. Of note, direct colocalization of MRAP2 with ADCY3 could not be assessed in vivo, as the primary antibodies to detect both proteins are produced in rabbits. We therefore performed immunofluorescence staining on 2 separate sets of slides to demonstrate that MC4R colocalizes with ADCY3 and that MC4R then colocalizes with MRAP2. MRAP2 localization to primary cilia was also confirmed by colocalization with arl13-GFP (Figure 3, A–C). ## Microscopy The images used for the assessment of MC4R-GFP at the cilium of MRAP2 WT versus KO were acquired with a Leica SP5 (Z stack, 0.5 μm steps). The images used to assess the colocalization of MC4R-GFP and MRAP2 in P6 mice were acquired with a Leica SP8 with resonant scanner. The SSTR3 imaging and MRAP2 immunofluorescence in MC4R-GFP adult mice were acquired with a Nikon W1 wide field-of-view spinning disk confocal with Andor Zyla sCMOS camera (Z stack, 0.26 μm steps). Images in Supplemental Figure 4 were acquired on a CSU-W1/SoRa spinning disk microscope, with dual camera Hamamatsu ORCA-FusionBT (Z stack, 0.2 μm steps). ## Image processing Images were processed with Fiji. Maximal intensity Z projections are from at least 20 slices over 15–20 μm. Quantified slices were matched for the number of slices projected and settings. ## Quantification of ciliary localization in cultured cells and hypothalamus sections Matched Z stack maximum projections were analyzed in Fiji. Relative ciliary enrichment was calculated as follows: each primary cilium was manually defined by a segmented line following ADCY3+ (in vivo) or ACTB+ (in vitro) signal. This 2D space was then used to the pixel intensity of the other channels (integrated density [IntDen]). Ciliary intensity of MC4R-GFP was then calculated as the IntDen of MC4R-GFP in the cilium, subtracting adjacent background (measured as IntDen of same defined area near the cilium). To calculate relative cilia enrichment, the IntDen (cilium) was divided by the IntDen (cell body), measured in the closest cell body (as defined by the presence of a Hoechst+ nucleus). Enrichment > 1, therefore, indicates higher localization of the receptor at the primary cilium compared with the cell body. Figure 2 details an in vitro experiment in which data are from 3 different replicates and are represented as a superplots, and $$n = 20$$–40 ciliated cells per condition were imaged and analyzed. Figure 4 indicates an in vivo experiment in which 60–90 cilia per mouse were quantified. ## Mouse metabolism studies For experiments presented in Figure 1, mice were weaned at 4 weeks of age, were single housed, and their food intake was measured manually every 24 hours for 4 consecutive days and averaged. Food intake data were excluded if the mouse lost a significant amount of weight because of single-housing stress. The animals were then housed in groups of 5, and their weight was measured weekly until 12 weeks of age. Food intake was again assessed as described at 12 weeks of age. Body composition was assessed by EchoMRI at 4 and 12 weeks of age. EE was measured by Comprehensive Lab Animal Monitoring System (CLAMS) at 8 weeks of age (Columbus Instruments). Mice were tested over 96 continuous hours, and the data from the last 48 hours were analyzed. EE is expressed in terms of kcal per hour and was calculated using the Lusk equation: EE = (3.815 + 1.232 × respiratory exchange ratio) × VO2. It was analyzed with CalR app software [46]. The interaction between the effect of body weight and genotype was quantified in a multivariate linear regression model adjusted for sex using the lm() function of R software [47]. ## Data availability The data that support the findings of this study are available from the corresponding author upon request. ## Statistics Sample sizes were chosen based upon the estimated effect size drawn from previous publications and from the performed experiments. Data distributions were assumed to be normal, but this was not formally tested. All tests used are indicated in the figure legends. We analyzed all data using Prism 7.0 (GraphPad Software), except for EE data that was analyzed using the lm() function of R software [47]. P ≤ 0.05 was considered significant.. ## Study approval All animal procedures were approved by the IACUC of the UCSF. ## Author contributions A Bernard, JFR, MVN, and CV designed the research study and wrote the manuscript. A Bernard, ION, and XY conducted experiments and acquired and analyzed data. JC, EM, FBG, A Blake, KK, and SZ conducted experiments. FM analyzed data. ## 01/24/2023 Electronic publication ## References 1. Krashes MJ. Melanocortin-4 receptor-regulated energy homeostasis. Nat Neurosci (2016.0) 19 206-219. DOI: 10.1038/nn.4202 2. Abdel-Malek ZA. Melanocortin receptors: their functions and regulation by physiological agonists and antagonists. Cell Mol Life Sci (2001.0) 58 434-441. DOI: 10.1007/PL00000868 3. Balthasar N. Divergence of melanocortin pathways in the control of food intake and energy expenditure. Cell (2005.0) 123 493-505. DOI: 10.1016/j.cell.2005.08.035 4. Shah BP. MC4R-expressing glutamatergic neurons in the paraventricular hypothalamus regulate feeding and are synaptically connected to the parabrachial nucleus. Proc Natl Acad Sci U S A (2014.0) 111 13193-13198. DOI: 10.1073/pnas.1407843111 5. Lubrano-Berthelier C. Melanocortin 4 receptor mutations in a large cohort of severely obese adults: prevalence, functional classification, genotype-phenotype relationship, and lack of association with binge eating. J Clin Endocrinol Metab (2006.0) 91 1811-1818. DOI: 10.1210/jc.2005-1411 6. Vaisse C. Melanocortin-4 receptor mutations are a frequent and heterogeneous cause of morbid obesity. J Clin Invest (2000.0) 106 253-262. DOI: 10.1172/JCI9238 7. Vaisse C. A frameshift mutation in human MC4R is associated with a dominant form of obesity. Nat Genet (1998.0) 20 113-114. DOI: 10.1038/2407 8. Benzinou M. Common nonsynonymous variants in PCSK1 confer risk of obesity. Nat Genet (2008.0) 40 943-945. DOI: 10.1038/ng.177 9. Loos RJF. Common variants near MC4R are associated with fat mass, weight and risk of obesity. Nat Genet (2008.0) 40 768-775. DOI: 10.1038/ng.140 10. Willer CJ. Six new loci associated with body mass index highlight a neuronal influence on body weight regulation. Nat Genet (2009.0) 41 25-34. DOI: 10.1038/ng.287 11. Huszar D. Targeted disruption of the melanocortin-4 receptor results in obesity in mice. Cell (1997.0) 88 131-141. DOI: 10.1016/S0092-8674(00)81865-6 12. Siljee JE. Subcellular localization of MC4R with ADCY3 at neuronal primary cilia underlies a common pathway for genetic predisposition to obesity. Nat Genet (2018.0) 50 180-185. DOI: 10.1038/s41588-017-0020-9 13. Wang Y. Melanocortin 4 receptor signals at the neuronal primary cilium to control food intake and body weight. J Clin Invest (2021.0) 131. DOI: 10.1172/JCI142064 14. Goetz SC, Anderson KV. The primary cilium: a signalling centre during vertebrate development. Nat Rev Genet (2010.0) 11 331-344. DOI: 10.1038/nrg2774 15. Reiter JF, Leroux MR. Genes and molecular pathways underpinning ciliopathies. Nat Rev Mol Cell Biol (2017.0) 18 533-547. DOI: 10.1038/nrm.2017.60 16. Green JA, Mykytyn K. Neuronal ciliary signaling in homeostasis and disease. Cell Mol Life Sci (2010.0) 67 3287-3297. DOI: 10.1007/s00018-010-0425-4 17. Hildebrandt F. Ciliopathies. N Engl J Med (2011.0) 364 1533-1543. DOI: 10.1056/NEJMra1010172 18. Vaisse C. Cilia and obesity. Cold Spring Harb Perspect Biol (2017.0) 9. DOI: 10.1101/cshperspect.a028217 19. Davenport J. Disruption of intraflagellar transport in adult mice leads to obesity and slow-onset cystic kidney disease. Curr Biol (2007.0) 17 1586-1594. DOI: 10.1016/j.cub.2007.08.034 20. Rouault AAJ. Melanocortin receptor accessory proteins (MRAPs): functions in the melanocortin system and beyond. Biochim Biophys Acta Mol Basis Dis (2017.0) 1863 2462-2467. DOI: 10.1016/j.bbadis.2017.05.008 21. Chaly AL. The melanocortin receptor accessory protein 2 promotes food intake through inhibition of the prokineticin receptor-1. Elife (2016.0) 5. DOI: 10.7554/eLife.12397 22. Srisai D. MRAP2 regulates ghrelin receptor signaling and hunger sensing. Nat Commun (2017.0) 8. DOI: 10.1038/s41467-017-00747-6 23. Asai M. Loss of function of the melanocortin 2 receptor accessory protein 2 is associated with mammalian obesity. Science (2013.0) 341 275-278. DOI: 10.1126/science.1233000 24. Chan LF. MRAP and MRAP2 are bidirectional regulators of the melanocortin receptor family. Proc Natl Acad Sci U S A (2009.0) 106 6146-6151. DOI: 10.1073/pnas.0809918106 25. Sebag JA. Developmental control of the melanocortin-4 receptor by MRAP2 proteins in zebrafish. Science (2013.0) 341 278-281. DOI: 10.1126/science.1232995 26. Schonnop L. Decreased melanocortin-4 receptor function conferred by an infrequent variant at the human melanocortin receptor accessory protein 2 gene. Obesity (Silver Spring) (2016.0) 24 1976-1982. DOI: 10.1002/oby.21576 27. Geets E. Copy number variation (CNV) analysis and mutation analysis of the 6q14.1-6q16.3 genes SIM1 and MRAP2 in Prader Willi like patients. Mol Genet Metab (2016.0) 117 383-388. DOI: 10.1016/j.ymgme.2016.01.003 28. Baron M. Loss-of-function mutations in MRAP2 are pathogenic in hyperphagic obesity with hyperglycemia and hypertension. Nat Med (2019.0) 25 1733-1738. DOI: 10.1038/s41591-019-0622-0 29. Novoselova TV. Loss of Mrap2 is associated with Sim1 deficiency and increased circulating cholesterol. J Endocrinol (2016.0) 230 13-26. DOI: 10.1530/JOE-16-0057 30. Sebag JA, Hinkle PM. Regulation of G protein-coupled receptor signaling: specific dominant-negative effects of melanocortin 2 receptor accessory protein 2. Sci Signal (2010.0) 3. PMID: 20371771 31. Delling M. Primary cilia are specialized calcium signaling organelles. Nature (2013.0) 504 311-314. DOI: 10.1038/nature12833 32. Handel M. Selective targeting of somatostatin receptor 3 to neuronal cilia. Neuroscience (1999.0) 89 909-926. DOI: 10.1016/S0306-4522(98)00354-6 33. Berbari NF. The primary cilium as a complex signaling center. Curr Biol (2009.0) 19 R526-R535. DOI: 10.1016/j.cub.2009.05.025 34. Berbari NF. Identification of ciliary localization sequences within the third intracellular loop of G protein-coupled receptors. Mol Biol Cell (2008.0) 19 1540-1547. DOI: 10.1091/mbc.e07-09-0942 35. Berbari NF. Bardet-Biedl syndrome proteins are required for the localization of G protein-coupled receptors to primary cilia. Proc Natl Acad Sci U S A (2008.0) 105 4242-4246. DOI: 10.1073/pnas.0711027105 36. Sun X. Tubby is required for trafficking G protein-coupled receptors to neuronal cilia. Cilia (2012.0) 1. DOI: 10.1186/2046-2530-1-21 37. Mukhopadhyay S. TULP3 bridges the IFT-A complex and membrane phosphoinositides to promote trafficking of G protein-coupled receptors into primary cilia. Genes Dev (2010.0) 24 2180-2193. DOI: 10.1101/gad.1966210 38. Saeed S. Loss-of-function mutations in ADCY3 cause monogenic severe obesity. Nat Genet (2018.0) 50 175-179. DOI: 10.1038/s41588-017-0023-6 39. Oh EC. Metabolic regulation and energy homeostasis through the primary Cilium. Cell Metab (2015.0) 21 21-31. DOI: 10.1016/j.cmet.2014.11.019 40. Song DK. Primary cilia as a signaling platform for control of energy metabolism. Diabetes Metab J (2018.0) 42 117-127. DOI: 10.4093/dmj.2018.42.2.117 41. Loktev AV, Jackson PK. Neuropeptide Y family receptors traffic via the Bardet-Biedl syndrome pathway to signal in neuronal primary cilia. Cell Rep (2013.0) 5 1316-1329. DOI: 10.1016/j.celrep.2013.11.011 42. Omori Y. Identification of G protein-coupled receptors (GPCRs) in primary cilia and their possible involvement in body weight control. PLoS One (2015.0) 10. DOI: 10.1371/journal.pone.0128422 43. Geneva II. Untangling ciliary access and enrichment of two rhodopsin-like receptors using quantitative fluorescence microscopy reveals cell-specific sorting pathways. Mol Biol Cell (2017.0) 28 554-566. DOI: 10.1091/mbc.e16-07-0549 44. Ersoy BA. Mechanism of N-terminal modulation of activity at the melanocortin-4 receptor GPCR. Nat Chem Biol (2012.0) 8 725-730. DOI: 10.1038/nchembio.1008 45. Nager AR. An actin network dispatches ciliary GPCRs into extracellular vesicles to modulate signaling. Cell (2017.0) 168 252-263. PMID: 28017328 46. Mina AI. CalR: a web-based analysis tool for indirect calorimetry experiments. Cell Metab (2018.0) 28 656-666. DOI: 10.1016/j.cmet.2018.06.019 47. 47R core team. A language and environment for statistical computing. https://www.r-project.org/ Accessed December 2, 2022
--- title: Hepatocyte-derived DPP4 regulates portal GLP-1 bioactivity, modulates glucose production, and when absent influences NAFLD progression authors: - Natasha A. Trzaskalski - Branka Vulesevic - My-Anh Nguyen - Natasha Jeraj - Evgenia Fadzeyeva - Nadya M. Morrow - Cassandra A.A. Locatelli - Nicole Travis - Antonio A. Hanson - Julia R.C. Nunes - Conor O’Dwyer - Jelske N. van der Veen - Ilka Lorenzen-Schmidt - Rick Seymour - Serena M. Pulente - Andrew C. Clément - Angela M. Crawley - René L. Jacobs - Mary-Anne Doyle - Curtis L. Cooper - Kyoung-Han Kim - Morgan D. Fullerton - Erin E. Mulvihill journal: JCI Insight year: 2023 pmcid: PMC9977314 doi: 10.1172/jci.insight.154314 license: CC BY 4.0 --- # Hepatocyte-derived DPP4 regulates portal GLP-1 bioactivity, modulates glucose production, and when absent influences NAFLD progression ## Abstract Elevated circulating dipeptidyl peptidase-4 (DPP4) is a biomarker for liver disease, but its involvement in gluconeogenesis and metabolic associated fatty liver disease progression remains unclear. Here, we identified that DPP4 in hepatocytes but not TEK receptor tyrosine kinase–positive endothelial cells regulates the local bioactivity of incretin hormones and gluconeogenesis. However, the complete absence of DPP4 (Dpp4–/–) in aged mice with metabolic syndrome accelerates liver fibrosis without altering dyslipidemia and steatosis. Analysis of transcripts from the livers of Dpp4–/– mice displayed enrichment for inflammasome, p53, and senescence programs compared with littermate controls. High-fat, high-cholesterol feeding decreased Dpp4 expression in F$\frac{4}{80}$+ cells, with only minor changes in immune signaling. Moreover, in a lean mouse model of severe nonalcoholic fatty liver disease, phosphatidylethanolamine N-methyltransferase mice, we observed a 4-fold increase in circulating DPP4, in contrast with previous findings connecting DPP4 release and obesity. Last, we evaluated DPP4 levels in patients with hepatitis C infection with dysglycemia (Homeostatic Model Assessment of Insulin Resistance > 2) who underwent direct antiviral treatment (with/without ribavirin). DPP4 protein levels decreased with viral clearance; DPP4 activity levels were reduced at long-term follow-up in ribavirin-treated patients; but metabolic factors did not improve. These data suggest elevations in DPP4 during hepatitis C infection are not primarily regulated by metabolic disturbances. ## Introduction Type 2 diabetes (T2D) is a metabolic disease characterized by the development of hyperglycemia. Dysregulated islet hormone secretion and an inability to overcome peripheral insulin resistance are central components [1, 2]. Given the increased appreciation of the reciprocal nature of dyslipidemia, obesity, dysglycemia and liver disease, including nonalcoholic fatty liver disease (NAFLD) and traditional risk factors, a new classification of metabolic (dysfunction) associated fatty liver disease (MAFLD) has emerged [3]. The secretion of incretin hormones from enteroendocrine cells in the gut epithelium, including glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide (GIP), potentiates postprandial insulin secretion, a phenomenon known as the incretin effect. GLP-1 and GIP are central in coordinating nutrient intake, nutrient disposal, and satiety (4–6). In patients with T2D, there is a defect in the incretin-mediated potentiation of insulin secretion [7]. Additionally, individuals with NAFLD exhibit an impaired incretin effect independent of diabetes, displaying fasting hyperglucagonemia [8] and increased hepatic gluconeogenesis [9]. The bioactivity and action of endogenous incretin hormones are limited through proteolytic cleavage and inactivation by the serine protease dipeptidyl peptidase-4 (DPP4) and renal elimination [10, 11]. Enzymatically active DPP4 is present in both membrane-bound and circulating forms [12]. Plasma DPP4 levels are elevated in several settings associated with metabolic dysfunction, such as obesity [13, 14]; chronic liver disease, including NAFLD and hepatitis C infection (HCV) [12, 15, 16]; as well as type 1 diabetes [17] and T2D [18, 19]. Increased DPP4 expression in the liver positively correlates with the degree of steatosis and NAFLD [20, 21]. Studies in mice using several tissue-specific targeting strategies have confirmed that the elevation of circulating DPP4 in obesity is liver derived (22–24), suggesting that DPP4 produced in the liver may primarily contribute to the progression of NAFLD [25]. Systemic inhibition of DPP4 decreases blood glucose by reducing hepatic glucose production (HGP) in patients with T2D [26]. Given the success of incretin-based drugs in treating both diabetes and obesity, and their potential for treating NAFLD, further dissection of the regulation of hepatic metabolic pathways by DPP4 is warranted (27–33). Here, we evaluated the role of hepatic DPP4 on incretin bioactivity within the portal vein (PV), hepatic glucose metabolism, and chronic liver disease progression in mice. We additionally examined circulating DPP4 levels and the mRNA abundances of sheddases in a mouse model of severe metabolic liver disease without obesity, phosphatidylethanolamine N-methyltransferase (Pemt–/–) mice. We also evaluated circulating DPP4 levels and substrates in a patient population undergoing treatment for HCV. ## Reduced HGP in high-fat, high-cholesterol–fed Dpp4–/– mice is due to loss of hepatocyte-derived DPP4. In patients with T2D, DPP4 inhibitors contribute to glucose homeostasis by decreasing hepatic gluconeogenesis [26, 34]. To validate this effect in mice, we performed hyperinsulinemic-euglycemic clamps in Dpp4+/+ (wild-type, WT) and Dpp4–/– (global Dpp4 deletion) mice fed a high-fat, high-cholesterol (HFHC) diet for 12 weeks. Glucose infusion rates (GIRs, Figure 1A) and glucose disposal rates (GDRs, Figure 1B) were indistinguishable between Dpp4–/– mice and their WT littermate controls. Although basal levels of HGP were unchanged, insulin-stimulated HGP was significantly lower in HFHC-fed Dpp4–/– (Figure 1C), resulting in a significantly greater suppression of HGP by insulin in Dpp4–/– mice (Figure 1D). Body weight and fasting glucose levels were unchanged (Supplemental Figure 1, A and B; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.154314DS1). Although the mRNA abundance of hepatic Gck, the enzyme that phosphorylates glucose to produce glucose-6-phosphate, was significantly upregulated in Dpp4–/– mice compared with livers of control mice (Figure 1E), transcript levels of the gluconeogenic enzymes, Pck and G6p, were unchanged (Figure 1E). mRNA levels of hepatic Gsk3β, a protein kinase that phosphorylates and inhibits glycogen synthase, were also elevated in Dpp4–/– mice compared with Dpp4+/+ (Figure 1E). Hepatic mRNA expression of Pygl, Gcgr, Igf1, and Igf1r was indistinguishable between Dpp4–/– and Dpp4+/+ mice (Figure 1, E and F). To determine if the reduction in HGP in Dpp4–/– mice was mediated through the actions of hepatocyte-derived DPP4, we then measured HGP in HFHC-fed, hepatocyte-specific Dpp4-knockout mice (Dpp4hep–/–). Similar to Dpp4–/– mice, HFHC-fed control mice (Dpp4GFP) and Dpp4hep–/– mice did not differ in GIR (Figure 1G) or GDR (Figure 1H) or in body weight or fasting glucose (Supplemental Figure 1, C and D). In addition, as seen in Dpp4–/– mice, basal levels of HGP were unchanged, and Dpp4hep–/– mice showed lower HGP under clamp conditions and greater insulin suppression of HGP (Figure 1, I and J). Akin to Dpp4–/– mice, mRNA expression of hepatic glucose utilization or gluconeogenic enzymes was unchanged except Gck, which trended ($$P \leq 0.09$$) upward in Dpp4hep–/– mice (Figure 1K). The gene expression of Gcgr, Igf1, and Igf1r was comparable between genotypes (Figure 1L). ## Deletion of Dpp4 in hepatocytes lowers portal DPP4 activity and increases portal concentrations of bioactive GLP-1 and GIP. We evaluated glucose excursion after intraperitoneal injection of pyruvate to further evaluate the cell-specific roles of DPP4 in regulating HGP. Consistent with our previous work [22], Dpp4hep–/– mice had significantly reduced Dpp4 mRNA expression in whole liver extracts (Supplemental Figure 2A); fasted DPP4 activity in plasma was decreased $50\%$ in 20-week-old Dpp4GFP and Dpp4hep–/– mice fed an HFHC diet for 4 weeks (Supplemental Figure 2B); but DPP4 plasma concentration was unchanged (Supplemental Figure 2C). Accordant with the clamp data, Dpp4hep–/– mice had significantly reduced glucose excursion AUC after injection of pyruvate compared with littermate Dpp4GFP controls (Figure 2A). Next, we examined whether reduced HGP in Dpp4hep–/– mice is mediated by DPP4’s action on GLP-1 receptor (GLP-1R) signaling. The decrease in plasma glucose following pyruvate injection was abrogated with the administration of exendin-9-39, a compound that blocks signaling through the GLP-1R [35], 15 minutes before injection of pyruvate (Figure 2B). Dpp4hep–/– mice showed no change compared to Dpp4GFP in an arginine tolerance test, indicating that islet responsivity to acute depolarization was normal (Supplemental Figure 2D). In addition, in response to arginine, mice lacking hepatocyte Dpp4 did not demonstrate any differences in blood glucose, active GLP-1, insulin, or glucagon (Supplemental Figure 2, E–G). GLP-1 has been reported to circulate through the portal circulation and the lymphatics to enter systemic circulation at the thoracic duct [36, 37]. To evaluate how DPP4 in hepatocytes may influence the bioactivity of incretin hormones at each of these sites (i.e., portal or systemic concentration), we administered oral glucose to Dpp4hep–/– and control mice and sampled PV blood via cannulation followed by systemic blood 15 minutes later by cardiac puncture (CP). DPP4 activity but not protein level within the PV was significantly reduced in Dpp4hep–/– mice compared with Dpp4GFP mice (Figure 2, C and D). In addition, active GLP-1 and active GIP levels were elevated approximately 4-fold in the local portal circulation in Dpp4hep–/– mice compared with Dpp4GFP mice (Figure 2, E and F). On the other hand, no significant differences in active GLP-1 and GIP were noted in plasma isolated immediately in the same mice by CP (Figure 2, E and F). Despite the increase in circulating incretin, plasma insulin and glucagon levels were unchanged in local hepatic and systemic circulations (Figure 2, G and H). To investigate the role of DPP4 in hepatocytes versus other sites that contribute to circulating DPP4, including TEK receptor tyrosine kinase–positive (Tie2+) endothelial cells (ECs) and immune cells, we used Dpp4EC–/– mice. Previously, we have reported that deletion of Dpp4 from the Tie2+ ECs (Dpp4EC–/–) of high-fat–fed mice results in increased systemic concentrations of GLP-1 and improved glucose excursion but does not affect HGP [38]. As expected, in contrast to Dpp4hep–/–, Dpp4EC–/– mice had no decrease in hepatic Dpp4 mRNA expression (Supplemental Figure 2H), and, consistent with previous reports [38], systemic fasting plasma DPP4 activity and DPP4 protein levels (Supplemental Figure 2, I and J) were significantly decreased in Dpp4EC–/– mice compared with controls. Furthermore, a pyruvate tolerance test showed no change in HGP in HFHC-fed Dpp4EC+/+ and Dpp4EC–/– mice previously administered saline or exendin-9-39 (Figure 2, I and J), consistent with our previous hyperglycemic-euglycemic clamp analyses [38]. To directly determine how DPP4 in these different tissue settings governs incretin bioactivity and HGP, we measured the portal concentrations of DPP4 and incretins and also examined pyruvate tolerance in HFHC-fed Dpp4EC+/+ and Dpp4EC–/– mice. Although portal DPP4 activity was unchanged (Figure 2K), DPP4 protein levels in the PV were significantly decreased in Dpp4EC–/– mice (Figure 2L). Furthermore, levels of active GLP-1 and active GIP were unchanged in the portal circulation in Dpp4EC–/– mice, but systemic concentrations of incretins were increased 2.5-fold (GLP-1) or trended 2-fold higher (GIP) in samples isolated by CP (Figure 2, M and N). Levels of insulin and glucagon at either sampling site (Figure 2, O and P) were unaltered. Together, these results suggest that the reduction in HGP observed in HFHC-fed Dpp4–/– mice is driven by the loss of Dpp4 in hepatocytes and not Tie2+ EC populations. Furthermore, this improved suppression of HGP is associated with elevated incretin concentrations within the PV, not the systemic circulation. ## Dyslipidemia and liver steatosis are unaffected by the genetic elimination of Dpp4 in aged, HFHC-fed mice. Insulin resistance, de novo lipogenesis, and dysregulated blood glucose are key hallmarks in the progression of NAFLD as part of the multiple-hit hypothesis [39]. To determine whether the reduction in HGP observed following the deletion of Dpp4 from the whole animal or liver only can prevent NAFLD progression, we fed aged 6-month-old mice either a standard laboratory diet (SLD) or an HFHC diet for 24 weeks. In a small subset of aged mice ($$n = 2$$/group), we performed similar PV cannulations and CPs and observed trends consistent with those shown in Figure 2 (Supplemental Table 1). Analysis of Dpp4 mRNA in whole liver extracts revealed elimination in all expected genotypes (Figure 3A). DPP4 activity in plasma and liver was absent from Dpp4–/– on SLD and HFHC diet and reduced by $60\%$ and $75\%$ in Dpp4hep–/– mice compared with controls (Figure 3, B and C). Agreeing with our previous results [22], deletion of hepatocyte Dpp4 led to sustained reductions in plasma and hepatic DPP4 activities, with little change in circulating DPP4 protein levels (Figure 3D). Deletion of hepatocyte-specific DPP4 in these aged mice did not affect glucose tolerance as only whole-body deletion of DPP4 increased systemic active GLP-1 and led to reduced blood glucose excursion during oral glucose tolerance test (Supplemental Figure 3, A and B), consistent with previous work [22, 23]. Glycogen concentrations were unchanged in all settings (Supplemental Figure 3C). In both SLD-fed and HFHC-fed Dpp4–/– mice, lack of Dpp4 did not affect HDL, LDL, or total cholesterol levels in plasma (Figure 3, E–G). However, fasting plasma triglycerides were significantly reduced in Dpp4–/– mice (Figure 3H). Biochemical measurement and histochemical analysis with Oil Red O staining in livers revealed no differences in neutral lipid concentrations between genotypes (Figure 3, I–M). Hepatic gene expression analysis also revealed that sterol regulatory element–binding transcription factor 1 (Srebf1) mRNA expression was significantly upregulated in HFHC-fed Dpp4–/– mice but unchanged in SLD-fed mice and HFHC-fed Dpp4hep–/– mice, compared with controls (Figure 3N). Microsomal triglyceride transfer protein (Mttp) mRNA expression was significantly decreased in SLD-fed Dpp4–/– mice compared with controls but unchanged in all genotypes under HFHC feeding (Figure 3O). In contrast, hepatic Forkhead box protein O1 (FoxO1) expression was unchanged between all genotypes under both diets (Figure 3P). Taken together, these data suggest that hepatic lipid accumulation is largely unaffected by whole-body or hepatocyte-specific elimination of Dpp4 gene in aged, SLD- or HFHC-fed mice. ## Systemic, not hepatic, loss of Dpp4 in aged mice increases hepatic fibrosis. Soluble, circulating DPP4 has been shown to be a marker of liver fibrosis [40]. Thus, we examined if systemic and hepatic loss of Dpp4 in mice affected liver fibrosis. Liver damage markers, such as alanine aminotransferase (ALT), aspartate transaminase (AST), and alkaline phosphatase, were not significantly different between all groups and their respective controls (Figure 4, A–C). Liver size normalized to tibia length was also unchanged (data not shown). To our surprise, mRNA levels of fibrosis markers, including Col1a1, Col3a1, Mmp2, Mmp11, Des, and Ddr2, were elevated in HFHC-fed Dpp4–/– mouse livers (Figure 4, D–I), suggesting worsening fibrosis in the Dpp4–/– mice. However, this elevated gene expression was not observed in Dpp4hep–/– mouse livers. No changes of expression were noted in fibrosis and hepatic stellate cell activation factors, Mmp9, Gfap and Vim, between any of the genotypes (Supplemental Figure 4, A–C) while both Lrat and Pcdh7 were significantly reduced in SLD-fed Dpp4–/– mice compared with controls (Supplemental Figure 4, D and E). Consistent with gene expression, visualization of collagen with Picrosirius red staining revealed that HFHC-fed Dpp4–/– mice had elevated fibrotic area in the liver, whereas it was relatively unchanged in Dpp4GFP and Dpp4hep–/– mice (Figure 4, J–M). Supporting increased fibrosis, blinded meta-analysis of histological data in viral hepatitis (METAVIR) histopathological scoring of liver samples revealed a shift of $33\%$ in Dpp4–/– mice to level 4 relative to littermate controls, while no Dpp4hep–/– mice were scored in this range (Figure 4N). Overall, these data suggest that systemic, not hepatic, loss of Dpp4 increases liver fibrosis. ## Global, but not hepatic, loss of DPP4 increases expression of genes associated with adaptive immunity, inflammasome, and senescence-associated genes and pathways. To gain molecular insights into the inflammatory responses in the liver mediated by loss of Dpp4, we conducted NanoString mRNA analysis on liver tissue using an immunology panel of over 500 immune-related genes. All Dpp4hep–/– mice were confirmed by quantitative real-time PCR (qRT-PCR) with primers specific for the recombined Dpp4 flox sites (Figure 3A), given the location of the flox deletion site toward the C-terminal end of the Dpp4 transcript and modest reduction in gene expression detected by NanoString probes (Supplemental Figure 5A). Supervised hierarchical clustering analysis of differentially expressed genes in Dpp4–/– mice revealed a distinct cluster of genes that were upregulated in SLD- and HFHC-fed Dpp4–/– mice compared with Dpp4+/+ and Dpp4GFP versus Dpp4hep–/– (Supplemental Figure 5, B and C). We identified differentially expressed genes that were distinct and overlapping among SLD-fed Dpp4–/–, HFHC-fed Dpp4–/–, and HFHC-fed Dpp4hep–/– livers (Supplemental Figure 5D). Pathway analysis in each comparison was performed, showing that only cytokine signaling in SLD-fed Dpp4–/– mice was significantly upregulated compared with Dpp4+/+ mice (Figure 5A). Notably, 18 pathways, including adaptive and innate immune pathways, inflammasome, Toll-like receptor signaling, oxidative stress, and TGF-β signaling, were all significantly upregulated in HFHC-fed Dpp4–/– compared with Dpp4+/+ mice (Figure 5B). Consistent with no difference in liver fibrosis (Figure 4H), all pathways were indistinguishable between HFHC-fed Dpp4GFP and Dpp4hep–/– mouse livers (Figure 5C), suggesting that hepatocyte DPP4 was not influencing the immunological response. We validated these results by immunostaining for the top differentially expressed transcript, Marco, in mice fed the HFHC diet. Consistent with the NanoString analysis, Marco staining was significantly reduced in the HFHC-fed Dpp4–/– mouse livers compared with controls (Supplemental Figure 6, A–G). Additionally, Marco expression in Dpp4GFP mice exhibited a spread of high- and low-expressing livers, which was recapitulated with immunostaining (Supplemental Figure 6, C, D, F, and G). When we probed gene expression within the inflammasome pathway, whose activation is a contributing factor in the initial progression of NAFLD [41], we found that all genes (App, Bcl2, Nfkb1, Nfkb2, and Rela) were significantly upregulated in HFHC-fed Dpp4–/– mice (Figure 5D). In contrast, only App and Rela were significantly upregulated in SLD-fed Dpp4–/– mice compared with controls, and Rela was significantly downregulated in Dpp4hep–/– mice (Figure 5D). Similarly, 19 of 28 NF-κB signaling pathway genes were significantly upregulated in HFHC-fed Dpp4–/– mice, whereas many genes were unchanged, or downregulated, in SLD-fed Dpp4–/– mice and HFHC-fed Dpp4hep–/– mice, compared with respective controls (Figure 5E). To further complement this analysis, we analyzed mRNA expression of known chemokine substrates of DPP4 in whole liver extracts. Consistent with its role in NAFLD progression [42], Ip-10 (Cxcl10) gene expression was significantly upregulated in HFHC-fed Dpp4–/– mice compared with controls but unchanged in SLD-fed mice and HFHC-fed Dpp4hep–/– mice compared with controls (Figure 5F). Expression of the gene regulated on activation, T cell expressed, and secreted (RANTES; Ccl5), which is associated with severe liver fibrosis [43], was also significantly upregulated in HFHC-fed Dpp4–/– mice (Figure 5G), whereas no changes in Mcp-1 (Ccl2) or Eotaxin (Ccl11) gene expression were noted (Figure 5, H and I). In contrast, gene expression of Ip-10, RANTES, Mcp-1, and Eotaxin were unchanged between HFHC-fed Dpp4GFP and Dpp4hep–/– mice (Figure 5, F–I). Cellular senescence has been identified to be involved in the transition from liver steatosis to a more severe phenotype involving hepatocyte ballooning and elevated fibrosis [44]. DPP4 has been identified on the surface of senescent cells, preferentially sensitizing them to cytotoxicity by NK cells [45]. Therefore, we were prompted to evaluate gene expression of senescence-associated secretory phenotype (SASP) factors [46] in our models. Our liver NanoString analysis identified increased expression of Trp53 in both SLD-fed and HFHC-fed, aged Dpp4–/– compared with their respective controls (Figure 5, J and K), which was unchanged in Dpp4hep–/– mice compared with Dpp4GFP mice (Figure 5L). We additionally measured genes associated with p53 signaling [47, 48]. The mRNA level of Ankrd1 was increased in SLD- and HFHC-fed Dpp4–/–, but unchanged in Dpp4hep–/–, compared with controls (Figure 5M). Cdkn1a expression was unchanged across both diets, and all genotypes (Figure 5N), while Cdkn2a was significantly decreased in HFHC-fed Dpp4–/– mice only (Figure 5O). When we analyzed protein levels of chemokine and cytokine SASPs, 8 weeks after starting the diet, CXCL1 levels were significantly increased in SLD-fed Dpp4–/– mice (Supplemental Figure 7A), while IL-6 was significantly increased in HFHC-fed Dpp4–/– mice (Supplemental Figure 7B). Other plasma cytokines, including IL-1β, IFN-γ, IL-10, and IL-2, were unchanged (Supplemental Figure 7, C–H). However, in HFHC-fed mice, IL-4 was significantly decreased in Dpp4–/– mice compared with controls (Supplemental Figure 7I), while it was increased in Dpp4hep–/– as was IL-5 at both 8 weeks and endpoint (Supplemental Figure 7, I and G). Few other significant changes were noted in plasma or within liver tissue at endpoint (Supplemental Figure 7, J–Y). ## Immune-related genes are upregulated in F4/80+ cells of SLD-fed Dpp4–/– mice, but HFHC feeding reduces DPP4 expression in F4/80+ cells. Roles of both liver-resident macrophages and recruited monocyte-derived macrophages in NAFLD progression [49] and liver fibrosis [50] have been established. Additionally, DPP4 is known to be upregulated when macrophages are polarized with proinflammatory stimuli and implicated in macrophage polarization and activation to mediate inflammation [51]. We therefore probed if liver-resident macrophages were critical in driving the increased inflammation in livers of mice with global Dpp4 deletion. We isolated F$\frac{4}{80}$+ cells from the liver and conducted NanoString mRNA analysis using the same immunology panel described above. Surprisingly, a large cluster of immune-related genes were upregulated in F$\frac{4}{80}$+ cells of SLD-fed Dpp4–/– mice compared with controls. However, these same genes were not differentially expressed in HFHC-fed Dpp4–/– mice (Supplemental Figure 8, A and C) whereas in HFHC-fed Dpp4hep–/– mice, 2 distinct clusters were revealed to be significantly altered compared with controls (Supplemental Figure 8, B and C). Pathway analysis revealed significant increases in NF-κB signaling, adaptive and innate immune system, and cytokine signaling (Figure 6A) in SLD-fed Dpp4–/– mice compared with controls. However, no pathways were significantly altered in HFHC-fed Dpp4–/– (Figure 6B) and Dpp4hep–/– mice (Figure 6C). In the isolated F$\frac{4}{80}$+ cells, unexpectedly, Dpp4 was downregulated in HFHC-fed Dpp4+/+ mice compared with SLD-fed Dpp4+/+ mice. Its expression was unchanged between HFHC-fed Dpp4GFP and Dpp4hep–/– mice, verifying deletion was restricted to hepatocytes (Figure 6D). To further understand potential differences in F$\frac{4}{80}$+ cells’ composition within the liver, we assessed differences in the abundance of transcripts associated with characterized populations. We found Adgre1 was upregulated in HFHC-fed Dpp4–/– mice (Figure 6E), and Ccr2 trended toward increase in HFHC-fed Dpp4–/– mice compared with controls. These markers remained unchanged in SLD-fed Dpp4–/– and Dpp4hep–/– mice versus their respective controls (Figure 6F). However, no differences in F$\frac{4}{80}$ immunostaining were noted (Supplemental Figure 9, A–E) between the HFHC-fed groups. Additionally, no changes were observed in macrophage polarization and population markers, Itgax (Figure 6G), Mrc1 (Figure 6H), Cd163 (Figure 6I), Arg1 (Figure 6J), and Clec4f (CLC4F) (Figure 6K), between genotypes and their respective controls. Consistent with these results, CLEC4F immunostaining revealed no differences between HFHC-fed groups (Supplemental Figure 9, A–D and F). Trp53 was upregulated in SLD-fed Dpp4–/– and downregulated in HFHC-fed Dpp4hep–/–, while Cxcr4 was upregulated in both (Figure 6L). SLD-fed Dpp4–/– mice had significant changes in many components of the NF-κB signaling pathway, but few of these patterns were observed in HFHC-fed Dpp4–/– or Dpp4hep–/– mice (Figure 6M). These data reveal an unexpected, complex relationship between DPP4 and liver-resident F$\frac{4}{80}$+ cells’ immunological profiles associated with diet composition. ## Liver-specific insults affect circulating DPP4 protein concentrations. Given its strong correlation with adipose tissue accumulation and obesity in both humans and mice, soluble, circulating DPP4 was initially characterized as an adipokine [13, 22]. Recent studies with adipocyte-specific targeting of DPP4 have determined that although adipocytes shed a small amount of DPP4 [22, 52], hepatocytes account for the significant elevation in DPP4 observed in high-fat diet feeding and metabolic dysregulation [22, 24]. Further, liver DPP4 expression is elevated in NAFLD [20, 21]. Comprehensive studies in cultured hepatocytes have determined that a combination of leptin and palmitic acid stimulates a 6-fold increase in Dpp4 mRNA expression [53]. To test the necessity of adiposity and peripheral insulin resistance in vivo for elevated enzymatically active, circulating DPP4, we assessed plasma DPP4 activity in HFHC-fed Pemt–/– mice, which are a lean model of hepatomegaly and hepatic steatosis due to disruption in de novo synthesis of choline [54] (Supplemental Figure 10A). Pemt–/– mice do not develop obesity with high-fat feeding, retain insulin sensitivity, and have lower leptin concentrations compared with littermate controls [54, 55]. Notably, systemic DPP4 activity was increased 4-fold relative to controls (Supplemental Figure 10B), suggesting dysregulation of hepatic lipid pathways is related to increased DPP4 activity, independent of the development of obesity. Unexpectedly, Dpp4 mRNA level in the liver was unchanged (Supplemental Figure 10C). However, mRNA expression of candidate sheddases was significantly increased, including Mmp9, Mmp2, and Adamst, but not Adam17 (Supplemental Figure 10, D–G). In addition to steatosis, other liver-specific insults associated with elevated circulating DPP4 include HCV [56]. Chronic HCV’s association with metabolic disease has been established [57]. We have recently shown that in patients with HCV treated with paritaprevir/ritonavir/ombitasvir/dasabuvir (PrOD), with or without ribavirin, fasting glucose, insulin, and Homeostatic Model Assessment of Insulin Resistance (HOMA-IR) are unchanged during treatment and follow-up after treatment [58]. We now sought to investigate whether elevated DPP4 levels are reversed following successful therapy for viral clearance in patients with metabolic dysfunction (HOMA-IR > 2) that persists through the treatment and follow-up period. Plasma AST (Figure 7A) and ALT (Figure 7B) significantly decreased in all treatment groups and were maintained at decreased levels during follow-up. Not surprisingly, given the lack of change in metabolic disease parameters, high variability and no significant changes were observed in C-reactive protein concentrations (Figure 7C). Consistent with other studies utilizing IFN-α treatment [59, 60], DPP4 concentration in plasma significantly decreased within all PrOD regimens from baseline to 12 weeks posttreatment (Figure 7D). In comparison, DPP4 activity substantially decreased with ribavirin treatment from baseline to follow-up in ribavirin-treated patients (Figure 7E). A similar trend to DPP4 protein was observed with known DPP4 substrates IP-10 (Figure 7F) and macrophage inflammatory protein 1α (MIP-1α) (Figure 7G). However, Eotaxin, another substrate, was unaffected (Figure 7H). Soluble intracellular adhesion molecule 1 (sICAM-1) was significantly decreased with treatment (Figure 7I). Taken together, DPP4 concentration and activity decrease with HCV treatment and viral clearance. However, this occurs independently of changes in metabolic parameters. ## Discussion Potentiation of GLP-1 action through receptor agonists has demonstrated efficacy in treating metabolic disease [61, 62]; however, our knowledge of the effects of DPP4 elimination to potentiate endogenous GLP-1 action within the context of chronic liver disease progression is limited. The present data demonstrate that eliminating hepatocyte DPP4 in HFHC-fed mice decreased DPP4 activity and increased intact incretins in the portal circulation and reduced HGP. These data are also consistent with patients in which DPP4 inhibitors (DPP4is) decrease HGP and are associated with reductions in glucagon [26, 34]. In contrast, in HFHC-fed Dpp4EC–/–, we report increased levels of active GLP-1 in the systemic circulation and no effect on HGP as assessed by hyperinsulinemic-euglycemic clamp [38]. Studies in mice have also demonstrated that intact GLP-1R signaling within the portal circulation is integral to glucose sensing [63]. This is interesting given that postprandial GLP-1 levels measured in the lymph are 5–6 times higher relative to sampling performed in portal plasma [36]. Consistent with our data identifying different regulation of incretin bioactivity and modulation of glucose metabolism with DPP4 in hepatocytes versus Tie2+ cells, recent reports have determined that elevated levels of active portal GLP-1 are disconnected from the classic definition of the incretin effect as increased portal circulation of GLP-1 does not potentiate nutrient-stimulated insulin secretion [64, 65]. In studies performed in rats, samples taken 20 minutes following a high-fat diet meal showed elevated GLP-1 in the lymph collected from the mesenteric lymph duct rather than the PV, while no difference was observed after a low-fat meal [66]. Circulating DPP4 activity is lower in lymph than in plasma [36, 66]. However, lymphatic ECs have been reported to express DPP4, and modulation of levels with siRNA affects migration and function [67]. Therefore, our current study is consistent with a model where, during HFHC diet–induced metabolic dysregulation, the deletion of Dpp4 within Tie2+ cells increases the abundance of GLP-1 delivered to the systemic circulation, enabling the incretin effect, and improves oral glucose tolerance but does not affect HGP. This is in contrast to Dpp4 in hepatocytes, which when deleted in HFHC-fed mice, increases GLP-1 bioactivity within the portal circulation and improves insulin-mediated suppression of HGP. Our data align with results from Hif1αhep–/– mice, demonstrating that elevation in DPP4 through activation of hepatocyte HIF-1α reduces active GLP-1 in the portal circulation [53]. Additionally, Baumeier et al. [ 25] demonstrated that hepatic Dpp4 overexpression results in decreased active, glucose-stimulated GLP-1 in the vena cava after liver passage. Recent studies have documented that GLP-1R engagement in the portal circulation is reduced under high-fat diet–feeding conditions [68], suggesting together with our data that multiple mechanisms converge to control GLP-1 action in the hepatic portal circulation, which may contribute to glucose dysregulation in mice upon high-fat feeding. Elevated concentrations of plasma DPP4 are associated with liver disease severity and fibrosis [69]. Consistent with this, hepatocyte-specific overexpression of DPP4 in mice results in increased hepatic steatosis, liver enzymes, and markers of inflammation [25]. In the current study, we report DPP4 was elevated 4-fold in Pemt–/– mice, which have both hepatomegaly and nonalcoholic steatohepatitis [54], demonstrating that in addition to obesity, liver-specific insults can induce the release of DPP4. Surprisingly, however, complete genetic elimination of Dpp4 or hepatocyte-specific elimination in aged mice resulted in no changes in liver enzymes and the degree of steatosis. Consistent with our results, using liver-specific knockdown of DPP4 via therapeutic siRNA, in obese and diabetic db/db mice, no effect on liver enzymes or glucose tolerance is observed [23]. Both Varin et al. and Ghorpade et al. reported modest impact on the liver with long-term targeting strategies [22, 24], suggesting that acute treatments targeting DPP4 more readily influence mouse lipid metabolism than long-term deletion of DPP4, which may be prone to metabolic adaptation. In mice, DPP4is decrease liver fibrosis [70, 71]; however, this was not recapitulated by genetically eliminating Dpp4 as Picrosirius red staining revealed a trend toward increased fibrosis, and expression of Col1a1 and Col3a1 was increased in Dpp4–/– mice. Chronic liver inflammation and immune reactions often precede fibrosis [72]; therefore, we assessed mRNA expression of immune-related genes. Under HFHC-fed conditions, livers of Dpp4–/– mice but not Dpp4hep–/– mice exhibited upregulation of transcripts associated with the inflammasome and markers of NF-κB signaling, pathways characterized to be activated in NAFLD [73, 74]. This was surprising given that DPP4is have been reported to suppress NF-κB activation [75, 76]. These results support important differences obtained by enzymatic inhibition versus complete deletion of the DPP4 protein. Cellular senescence has been proposed as a key factor in NAFLD progression [77]. Further, DPP4 has been identified as a surface protein that is enriched in senescent cells [45], and modulation of DPP4 in presenescent WI-38 cells [45] or in vascular endothelial cells [78] reduces markers of senescence. Additionally, senescence induced by glucocorticoid treatment, known to increase DPP4 transcriptionally [79], can be modulated by inhibition of DPP4 activity [80]. In our study, lifelong deletion of DPP4 in aged mice led to upregulation of Trp53, a component of the senescent machinery [81]. Additionally, the expression of Ankrd1, Il1a, Ccl2, and Ccl3 was increased in HFHC-fed Dpp4–/– mice. A molecular link between p53 and DPP4 has been established in which p53 antagonizes ferroptosis by blocking DPP4 activity [82]. However, the mechanistic link between DPP4, p53, cellular senescence, and liver fibrosis could not be deduced from this study and warrants further investigation. Elevated DPP4 is also observed in other chronic liver diseases, such as HCV, in addition to NAFLD [15]. Consistent with data from patients who received IFN treatment [59], plasma DPP4 concentrations, liver enzymes, and cytokine levels were reduced after resolution of infection with direct-acting antiviral therapy with or without ribavirin. DPP4 activity varied between treatment groups at baseline but overall remained decreased from baseline at follow-up in all. Additionally, plasma AST, ALT, IP-10, MIP-1α, and sICAM-1 all decreased with treatment. However, these changes were not associated with changes in glucose regulation [58]. We acknowledge several shortcomings to our study, including that the effects of exendin-9-39 cannot exclude the potential for signaling of glucagon through the GLP-1R (83–86), in addition to active GLP-1 [87]. Additionally, while mice remained unrestrained during the hyperinsulinemic-euglycemic clamps, they were sampled by the tail vein, which may contribute differential physiological responses to stress across genotypes. In addition, many of our inferences are dependent on transcript levels rather than direct protein quantitation due to the availability and reliability of relevant antibodies. In summary, we have identified hepatocyte-derived Dpp4 as a key factor in regulating the bioactivity of GLP-1 in the PV and extended these findings to demonstrate that its elimination results in a reduction in HGP. However, despite the elevation in GLP-1 and improvements in hepatic insulin sensitivity, we demonstrate a disconnect as markers of lipid metabolism, fibrosis, and inflammation were unchanged or worsened in mice. ## Animals. All studies were performed according to protocols approved by the University of Ottawa Animal Care Committee and in accordance with guidelines of the Canadian Council on Animal Care. Male mice were housed under a 12-hour light/12-hour dark cycle and maintained on SLD (Harlan Teklad) or HFHC diet (TD.88137, Envigo-Teklad Custom Diets). Whole-body Dpp4–/– mice, on a C57BL/6 background, have been described [38, 88]. To generate Dpp4hep–/– mice, Dpp4fl/fl adult mice, provided by Merck Research Laboratories [38], were i.v. injected with 1.5 × 1011 genome copies per mouse of AAV8.TBG.pi.egfp.wpre.bgh (AAV-GFP; control virus, 105535-AAV8, Dpp4GFP) or AAV8.TBG.PI.CRE.rBG (AAV-Cre; 107787-AAV8, Dpp4hep–/–) prior to the onset of HFHC diet feeding. Both AAV constructs were obtained from the University of Pennsylvania Vector Core Lab as a gift from James M. Wilson (Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA). B6.Cg-Tg(Tek-cre) 1Ywa/J mice were obtained from The Jackson Laboratory (Strain 008863, RRID: IMSR_JAX:008863) and bred with Dpp4fl/fl to generate Dpp4EC–/– mice. Experiments in Pemt+/+ and Pemt−/− mice, provided by LKS Centre for Health Research Innovation [89], were approved by the University of Alberta’s Institutional Animal Care Committee. Mice were fed an HFHC diet (F3282; Bio-Serv) for 3 weeks and fasted for 12 hours before sacrifice and tissue collection. Hyperinsulinemic-euglycemic clamps were performed in 12-week-old Dpp4+/+, Dpp4–/–, Dpp4fl/fl, and Dpp4hep–/– mice subjected to 12–16 weeks of HFHC diet. Experiments were performed in 18- to 20-week-old mice after 5 weeks of HFHC diet to validate our HGP results. Last, to assess NAFLD progression, experiments were performed in 16- to 28-week-old mice fed an HFHC diet for 24 weeks, with sacrifice and tissue collection at 40–52 weeks. Eight mice died prematurely (1 Dpp4hep–/–, 2 Dpp4GFP, 1 Dpp4–/–, and 4 WT mice), and 1 Dpp4GFP mouse was omitted from analysis due to the presence of a very enlarged spleen. Aged, HFHC-fed Dpp4hep–/– and Dpp4GFP mice were assessed for portal hormone concentrations ($$n = 2$$), with averages presented in Supplemental Table 1. Mice were genotyped from genomic DNA (gDNA) isolated from tail samples, and DNA recombination was confirmed in gDNA isolated from the liver. After DNA extraction and amplification, the PCR product was loaded on $1\%$ agarose gels (UltraPure Agarose, Invitrogen, Thermo Fisher Scientific), then separated for 30 minutes at 150 V (Owl Easycast B2 Separating System), and the bands were visualized under a blue light transilluminator (UVP GelDoc-It2 Imager). ## Hyperinsulinemic-euglycemic clamp. Hyperinsulinemic-euglycemic clamps were performed as previously described [90]. Briefly, a catheter was surgically placed into the right jugular vein, and mice were allowed to recover for 5 days. All mice regained their presurgical weights following surgery. The catheter was made accessible through an adaptor port implanted in the dorsal subscapular region. On the day of the clamp, mice were fasted for 5 hours and then infused with d-[3-3H]-glucose (PerkinElmer) for 1 hour to evaluate basal glucose disposal. Human insulin (10 mU/kg/min, NovoRapid, Novo Nordisk) infusate containing d-[3-3H]-glucose was then administered, and blood glucose levels were titrated with $50\%$ dextrose to achieve and maintain euglycemia. Mice were not physically restrained and were free to move around their cage, with blood samples taken via the tail vein. Basal and clamped rates of glucose disposal and HGP were calculated as described [90]. All tissues were rapidly dissected, snap-frozen in liquid nitrogen, and stored at −80°C for later analyses. ## PV and CP cannulation. Mice were given an oral bolus of glucose (2 g/kg) and subsequently anesthetized with $4\%$ isoflurane. Once fully unconscious, the abdomen was disinfected, and the mouse was placed on a heated surgical platform. A sagittal incision was made through the skin and fascia of the lower abdomen. A lateral transverse incision was made through the muscular layer to expose the abdominal contents. The PV was gently exposed by moving the intestines laterally toward the left body wall. Once exposed, the PV was cannulated with a 20-gauge butterfly needle 15 minutes after the glucose bolus, and the needle was withdrawn to collect blood. Approximately 1 minute later, the beating heart was exposed by cutting through the diaphragm and thorax. An 18-gauge needle was used to collect blood from the right ventricle. Blood was aliquoted into 2 EDTA-coated capillary microvette tubes, one with $10\%$ TED (5,000 KIU/mL Trasylol, 1.2 mg/mL EDTA, and 0.1 nmol/L Diprotin A) (vol/vol) and one without. Plasma was isolated after centrifugation (13,523g, 10 minutes, 4°C) and stored at −80°C for later analyses. ## Pyruvate, arginine, and glucose tolerance tests. After a 16-hour fast, mice were intraperitoneally injected with either saline or exendin-9-39 (24 nmol/kg body weight; ref. 91; Bachem), 15 minutes prior to injection, with 2 g/kg body weight pyruvate in sterile $0.9\%$ saline. After a 4-hour fast, mice were intraperitoneally injected with 2 g/kg body weight arginine in sterile $0.9\%$ saline. For glucose tolerance tests, mice were fasted for 5 hours and given glucose in PBS (2 g/kg body weight) in sterile $0.9\%$ saline. Blood for glucose measurements (Glucometer, MediCure Canada) was obtained from the tail vein before pyruvate injection and at 15, 30, 45, 60, and 90 minutes after pyruvate injection, or before arginine injection and at 15 and 30 minutes after arginine injection. ## Blood and tissue collection. All blood samples were collected in EDTA-coated capillary microvette tubes, and plasma was isolated after centrifugation (13,523g, 10 minutes, 4°C). During metabolic tolerance tests, blood was taken via tail vein. For terminal studies, mice were sacrificed by CO2 inhalation, and blood was obtained by CP. For measurement of plasma active GLP-1 (Meso Scale Diagnostics) and active GIP (Crystal Chem), blood was mixed with $10\%$ TED (vol/vol) and plasma stored at –80°C until further analysis. Plasma insulin (Alpco Diagnostics) and glucagon (Crystal Chem) levels were determined as per manufacturer’s instructions. Analysis for plasma ALT, AST, alkaline phosphatase, TG, cholesterol, LDL, and HDL was performed by the Pathology core at The Centre for Phenogenomics. The Beckman Coulter AU480 clinical chemistry analyzer was used in combination with appropriate reagents (ALT, AST, TG, cholesterol, LDL, and HDL), calibrators (Beckman Coulter Lyophilized Chemistry Calibrator levels 1 and 2), and quality control materials (Bio-Rad Liquid Assayed Multiqual levels 1 and 3). DPP4 activity was assessed using fluorometric assay (substrate: 10 mM H-Gly-Pro-AMC HBr, Bachem catalog I-1225; standard: AMC, Bachem catalog Q-1025). DPP4 protein level was measured using DPPIV/CD26 DuoSet ELISA kit (DY954; R&D Systems, Bio-Techne) following the manufacturer’s instructions. ## Picrosirius red and Oil Red O staining. Liver tissue was fixed in $4\%$ paraformaldehyde (PFA) and routinely processed for paraffin-embedding and cross-sectioned to obtain 5 μm–thick sections. The slides were incubated with a $0.1\%$ Picrosirius red solution and mounted with Permount Mounting Medium (Thermo Fisher Scientific). Collagen accumulation was determined by the number of red-stained pixels using ImageJ (NIH). Accumulation of fat droplets in the liver was visualized using Lipid (Oil Red O) Staining Kit as per manufacturer protocol (BioVision). A pathologist assessed the Picrosirius red sections and provided a METAVIR score following a protocol blinded to genotype. ## Immunofluorescence staining. Liver tissue was fixed in $4\%$ PFA and routinely processed for paraffin-embedding and cross-sectioned to obtain 5 μm–thick sections. Sections were dewaxed with toluene, then rehydrated with graded washes of ethanol, ending with water. Antigen retrieval was performed using sodium citrate buffer (0.1 M, pH 6, with $0.05\%$ Tween-20) in a microwave, then washed with de-ionized water and PBS on a shaker. Sections were blocked with $10\%$ donkey serum (catalog D9663 MilliporeSigma) for 30 minutes at room temperature. Antibodies against MARCO (Abcam, clone EPR22944-66, catalog ab259264, 1:200), F$\frac{4}{80}$ (Invitrogen, Thermo Fisher Scientific, clone CI:A3-1, catalog MA1-91124, 1:250), and CLEC4F (R&D Systems, Bio-Techne, catalog AF2784, 1:200) were incubated overnight at 4°C. Slides were washed with PBS-Tween, then incubated with secondary antibodies donkey anti-rabbit IgG (H+L) highly cross-absorbed Alexa Fluor Plus 647 (A32795, 1:500), donkey anti-rat IgG (H+L) highly cross-absorbed Alexa Fluor Plus 647 (A48272, 1:500), and donkey anti-goat IgG (H+L) cross-absorbed Alexa Fluor 555 (A-21432, 1:500) for 45 minutes at room temperature. Slides were washed and nuclei were stained using DAPI (Thermo Fisher Scientific 62248, 1:2,000) for 5 minutes at room temperature, washed, and mounted (Abcam, ab104135). Images were obtained on a Zeiss AxioImager Z1 epifluorescence microscope with 20× or 63× oil immersion objective and analyzed using Zeiss ZEN Blue microscopy software. ## Liver TG and cholesterol content. Total liver lipids were extracted using a modified Folch method [92]. For SLD- and HFHC-fed mice, a 100 mg or 50 mg piece of liver tissue was homogenized in 4 mL chloroform/methanol (2:1, v/v) and processed as described previously [93]. Lipids were quantified using Infinity Cholesterol or Triglyceride reagent (both Thermo Fisher Scientific) at 540 nm. ## Hepatic F4/80+ cell isolation. Fresh mouse livers were enzymatically digested using the components of a liver dissociation kit (kit 130-105-807, Miltenyi Biotec), and the gentleMACS Dissociators were used for the mechanical dissociation steps as previously described [94]. Hepatic F$\frac{4}{80}$+ cells were isolated from dissociated liver samples with Anti-F$\frac{4}{80}$ MicroBeads UltraPure (mouse, Miltenyi Biotec) as per manufacturer’s protocol and flash-frozen in liquid nitrogen before being stored at –80°C for mRNA extraction. ## RNA isolation, cDNA, and qRT-PCR. Frozen liver and isolated hepatic F$\frac{4}{80}$+ cells were homogenized with Tri Reagent (MilliporeSigma) using a TissueLyser II system (QIAGEN), and total RNA was extracted using manufacturer’s protocol. Reverse transcription was performed with the Applied Biosystems (Thermo Fisher Scientific) High-Capacity cDNA Reverse Transcription Kit. cDNA was subsequently used to assess mRNA expression by qRT-PCR (QuantStudio 5 System, Thermo Fisher Scientific) with TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific, 4444557) and TaqMan Gene Expression Assays (Thermo Fisher Scientific). The specific gene expression assays used are listed (Supplemental Table 1). Quantification of transcript levels was performed by the standard curve method, and expression levels for each gene were normalized to Actb (β-actin). ## NanoString mRNA analysis. NanoString Technologies nCounter Mouse Immunology Panel (catalog XT-CSO-MIM1-12) was used where 100 ng RNA was incubated with reporter and capture probes, consisting of 547 immunology-related mouse genes and 14 internal reference controls, for 16 hours at 65°C. Following hybridization, unbound probes were removed. According to the manufacturer’s instructions, assays were performed and quantified on the nCounter system, sample preparation station, and digital analyzer (NanoString Technologies). Raw gene expression data were analyzed using NanoString’s software, nSolver v4.0.70, with the Advanced Analysis Module v2.0.115 with background subtraction. Genes with counts below a threshold of 20 were excluded from subsequent analysis. Data normalization was performed on background-subtracted samples using internal positive controls and selected housekeeping genes. Housekeeping genes were selected based on those that were consistent in all analyses across genotypes and diets: Sdha, Eef1g, Gapdh, Hprt, Polr2a, Rpl19, Oaz1, Tbp, and Tubb5 for liver tissue and Rpl19, Ppia, Oaz1, Eef1g, Sdha, Pol2a, Gusb, Tubb5, Gapdh, and Hprt for F$\frac{4}{80}$+ cells. Differential gene expression analyses were performed using nSolver, which applies several multivariate linear regression models to identify significant genes (mixture negative binomial, simplified negative binomial, or log-linear model). Raw mRNA counts were log2-transformed, and significance was determined using 2-tailed t test. Statistically significant differentially expressed genes were identified as those with a $P \leq 0.05.$ Ratios of log2-normalized transcript count data were generated for SLD Dpp4–/– mice versus baseline SLD WT mice, HFHC-fed Dpp4–/– mice versus baseline HFHC-fed WT mice, and HFHC-fed Dpp4hep–/– mice versus baseline HFHC-fed Dpp4GFP mice. The NanoString data have been deposited in NCBI’s Gene Expression Omnibus (GEO) and are accessible through GEO Series accession number GSE218767. Pathway scores generated from nSolver Advanced Analysis were standardized by Z-scaling. ClustVis [95] was used to perform supervised hierarchical clustering analysis and principal component analysis of log2-transformed transcript count data and Z-scaled pathway scores. ## Human studies. Participants were recruited between July 2015 and April 2016 from The Ottawa Hospital Viral Hepatitis Program (Ottawa, Canada). All participants were 18 years or older, planned to initiate HCV antiviral treatment, and provided signed informed consent documents to participate in a single-center, open-label study (ClinicalTrials.gov Identifier: NCT02734173), which was approved by The Ottawa Health Science Network Research Ethics Board (REB 2015-0305). Three groups of patients were examined for this study: noncirrhotic genotype 1a–infected participants receiving standard therapy plus ribavirin, noncirrhotic genotype 1b–infected participants receiving standard therapy, and compensated cirrhotic genotype 1a– or 1b–infected participants dosed with standard therapy plus ribavirin; all had a HOMA-IR ≥ 2 [58]. Patients were treated for 12 weeks with ribavirin and direct-acting antivirals, after which they achieved sustained virologic response, and they were followed up for an additional 36 weeks. Inclusion and exclusion criteria, as well as methods for treatment, have been described [58]. Plasma measurements were conducted by Laboratory Services at The Ottawa Hospital, as standard clinical procedure. Additional blood samples were treated with $1\%$ Triton X-100 and $0.3\%$ tributyl phosphate and incubated at 37°C for 1 hour to destroy any virus. The concentrations of circulating factors in treated plasma were quantified using multiplexing immunobead assays analyzed using Meso Scale Diagnostics as described above. Plasma DPP4 concentration was quantified using the R-PLEX Human DPPIV Antibody Set (Meso Scale Diagnostics, catalog F21YC) [96]. To remove the variance in trait values attributed to sex and age differences for results described in Figure 7, a linear regression model compared the retrieved residuals (adjusted trait values) using the unpaired t test function available in R (version 4.0.5). Data are expressed as mean ± SD, and $P \leq 0.05$ was considered statistically significant. ## Statistics. All data were plotted and statistical analyses were performed using GraphPad Prism (version 8.4.3). Data are expressed as mean ± SEM; human data are expressed as mean ± SD. Statistical differences between groups were evaluated by 2-way ANOVA with Tukey’s multiple-comparison post hoc test when analyzing time course data. All other differences were evaluated by a 2-tailed unpaired t test with Welch’s correction. $P \leq 0.05$ was considered statistically significant. ## Study approval. Animals were cared for in accordance with the Canadian Guide to the Care and Use of Laboratory Animals (Canadian Council on Animal Care, 2020). Experimental procedures were approved under AUP#2909 and AUP#2029 by the University of Ottawa Animal Care and Veterinary Service (Ottawa, Ontario, Canada). Experiments in Pemt+/+ and Pemt−/− were approved by the University of Alberta’s Institutional Animal Care Committee (Edmonton, Alberta, Canada). All participants were 18 years or older, planned to initiate HCV antiviral treatment, and provided written informed consent to participate in a single-center, open-label study (ClinicalTrials.gov Identifier: NCT02734173), which was approved by The Ottawa Health Science Network Research Ethics Board (REB 2015-0305, Ottawa, Ontario, Canada). ## Author contributions MAD, MDF, and EEM conceptualized the study; KHK, MDF, and EEM developed methodology; NAT performed formal analysis, NAT, BV, MAN, NJ, EF, NMM, CAAL, NT, AAH, JRCN, CO, JNVDV, ILS, RS, SMP, ACC, AMC, RLJ, MAD, CLC, KHK, MDF, and EEM investigated; MDF and EEM provided resources; NAT, BV, and EEM wrote the original draft; all authors reviewed and edited the draft; KHK, MDF, EEM, and RLJ supervised the project; and EEM performed project administration and funding acquisition. ## 12/06/2022 In-Press Preview ## 01/24/2023 Electronic publication ## References 1. Haedersdal S. The role of glucagon in the pathophysiology and treatment of type 2 diabetes. Mayo Clin Proc (2018) 93 217-239. DOI: 10.1016/j.mayocp.2017.12.003 2. Petersen MC, Shulman GI. Mechanisms of insulin action and insulin resistance. Physiol Rev (2018) 98 2133-2223. DOI: 10.1152/physrev.00063.2017 3. Eslam M. MAFLD: a consensus-driven proposed nomenclature for metabolic associated fatty liver disease. Gastroenterology (2020) 158 1999-2014. DOI: 10.1053/j.gastro.2019.11.312 4. McLean BA. Revisiting the complexity of GLP-1 action from sites of synthesis to receptor activation. Endocr Rev (2021) 42 101-132. DOI: 10.1210/endrev/bnaa032 5. El K, Campbell JE. The role of GIP in α-cells and glucagon secretion. Peptides (2020) 125. DOI: 10.1016/j.peptides.2019.170213 6. El K. GIP mediates the incretin effect and glucose tolerance by dual actions on α cells and β cells. Sci Adv (2021) 7. DOI: 10.1126/sciadv.abf1948 7. Nauck M. Reduced incretin effect in type 2 (non-insulin-dependent) diabetes. Diabetologia (1986) 29 46-52. DOI: 10.1007/BF02427280 8. Junker AE. Diabetic and nondiabetic patients with nonalcoholic fatty liver disease have an impaired incretin effect and fasting hyperglucagonaemia. J Intern Med (2016) 279 485-493. DOI: 10.1111/joim.12462 9. Sunny NE. Excessive hepatic mitochondrial TCA cycle and gluconeogenesis in humans with nonalcoholic fatty liver disease. Cell Metab (2011) 14 804-810. DOI: 10.1016/j.cmet.2011.11.004 10. Deacon CF. Dipeptidyl peptidase 4 inhibitors in the treatment of type 2 diabetes mellitus. Nat Rev Endocrinol (2020) 16 642-653. DOI: 10.1038/s41574-020-0399-8 11. Mulvihill EE. Dipeptidyl peptidase inhibitor therapy in type 2 diabetes: control of the incretin axis and regulation of postprandial glucose and lipid metabolism. Peptides (2018) 100 158-164. DOI: 10.1016/j.peptides.2017.11.023 12. Mulvihill EE, Drucker DJ. Pharmacology, physiology and mechanisms of action of dipeptidyl peptidase-4 inhibitors. Endocr Rev (2014) 35 992-1019. DOI: 10.1210/er.2014-1035 13. Lamers D. Dipeptidyl peptidase 4 is a novel adipokine potentially linking obesity to the metabolic syndrome. Diabetes (2011) 60 1917-1925. DOI: 10.2337/db10-1707 14. Sarkar J. Increased plasma dipeptidyl peptidase-4 (DPP4) activity is an obesity-independent parameter for glycemic deregulation in type 2 diabetes patients. Front Endocrinol (lausanne) (2019) 10. DOI: 10.3389/fendo.2019.00505 15. Grungreiff K. Plasma concentrations of zinc, copper, interleukin-6 and interferon-γ, and plasma dipeptidyl peptidase IV activity in chronic hepatitis C. Mol Med Rep (2009) 2 63-68. DOI: 10.3892/mmr_00000062 16. Itou M. Altered expression of glucagon-like peptide-1 and dipeptidyl peptidase IV in patients with HCV-related glucose intolerance. J Gastroenterol Hepatol (2008) 23 244-251. DOI: 10.1111/j.1440-1746.2007.05183.x 17. Varga T. Higher serum DPP-4 enzyme activity and decreased lymphocyte CD26 expression in type 1 diabetes. Pathol Oncol Res (2011) 17 925-930. DOI: 10.1007/s12253-011-9404-9 18. Mannucci E. Hyperglycaemia increases dipeptidyl peptidase IV activity in diabetes mellitus. Diabetologia (2005) 48 1168-1172. DOI: 10.1007/s00125-005-1749-8 19. Ryskjaer J. Plasma dipeptidyl peptidase-IV activity in patients with type-2 diabetes mellitus correlates positively with HbAlc levels, but is not acutely affected by food intake. Eur J Endocrinol (2006) 155 485-493. DOI: 10.1530/eje.1.02221 20. Miyazaki M. Increased hepatic expression of dipeptidyl peptidase-4 in non-alcoholic fatty liver disease and its association with insulin resistance and glucose metabolism. Mol Med Rep (2012) 5 729-733. DOI: 10.3892/mmr.2011.707 21. Baumeier C. Hepatic DPP4 DNA methylation associates with fatty liver. Diabetes (2017) 66 25-35. DOI: 10.2337/db15-1716 22. Varin EM. Circulating levels of soluble dipeptidyl peptidase-4 are dissociated from inflammation and induced by enzymatic DPP4 inhibition. Cell Metab (2019) 29 320-334. DOI: 10.1016/j.cmet.2018.10.001 23. Gorgens SW. A siRNA mediated hepatic dpp4 knockdown affects lipid, but not glucose metabolism in diabetic mice. PLoS One (2019) 14. DOI: 10.1371/journal.pone.0225835 24. Ghorpade DS. Hepatocyte-secreted DPP4 in obesity promotes adipose inflammation and insulin resistance. Nature (2018) 555 673-677. DOI: 10.1038/nature26138 25. Baumeier C. Elevated hepatic DPP4 activity promotes insulin resistance and non-alcoholic fatty liver disease. Mol Metab (2017) 6 1254-1263. DOI: 10.1016/j.molmet.2017.07.016 26. Balas B. The dipeptidyl peptidase IV inhibitor vildagliptin suppresses endogenous glucose production and enhances islet function after single-dose administration in type 2 diabetic patients. J Clin Endocrinol Metab (2007) 92 1249-1255. DOI: 10.1210/jc.2006-1882 27. Carris NW, Dietrich EA. Semaglutide for weight loss. Ann Pharmacother (2022) 56 224-226. DOI: 10.1177/10600280211027204 28. Da BL, Satapathy SK. Semaglutide or placebo for nonalcoholic steatohepatitis. N Engl J Med (2021) 385. DOI: 10.1056/NEJMc2106921 29. Meier JJ. Efficacy of semaglutide in a subcutaneous and an oral formulation. Front Endocrinol (lausanne) (2021) 12. DOI: 10.3389/fendo.2021.645617 30. Slawson DC. Once-weekly semaglutide is an effective adjunct for weight loss in adults without diabetes who are overweight or obese. Am Fam Physician (2021) 104 96-97. PMID: 34264618 31. Srivastava G, Kumar RB. Once-weekly semaglutide in adults with overweight or obesity. N Engl J Med (2021) 385. DOI: 10.1056/NEJMc2106918 32. Frias JP. The sustained effects of a dual GIP/GLP-1 receptor agonist, NNC0090-2746, in patients with type 2 diabetes. Cell Metab (2017) 26 343-352. DOI: 10.1016/j.cmet.2017.07.011 33. Jall S. Monomeric GLP-1/GIP/glucagon triagonism corrects obesity, hepatosteatosis, and dyslipidemia in female mice. Mol Metab (2017) 6 440-446. DOI: 10.1016/j.molmet.2017.02.002 34. Muscelli E. Mechanisms for the antihyperglycemic effect of sitagliptin in patients with type 2 diabetes. J Clin Endocrinol Metab (2012) 97 2818-2826. DOI: 10.1210/jc.2012-1205 35. Serre V. Exendin-(9-39) is an inverse agonist of the murine glucagon-like peptide-1 receptor: implications for basal intracellular cyclic adenosine 3′,5′-monophosphate levels and β-cell glucose competence. Endocrinology (1998) 139 4448-4454. DOI: 10.1210/endo.139.11.6295 36. D’Alessio D. Fasting and postprandial concentrations of GLP-1 in intestinal lymph and portal plasma: evidence for selective release of GLP-1 in the lymph system. Am J Physiol Regul Integr Comp Physiol (2007) 293 R2163-R2169. DOI: 10.1152/ajpregu.00911.2006 37. Qin X. GLP-1 reduces intestinal lymph flow, triglyceride absorption, and apolipoprotein production in rats. Am J Physiol Gastrointest Liver Physiol (2005) 288 G943-G949. DOI: 10.1152/ajpgi.00303.2004 38. Mulvihill EE. Cellular sites and mechanisms linking reduction of dipeptidyl peptidase-4 activity to control of incretin hormone action and glucose homeostasis. Cell Metab (2017) 25 152-165. DOI: 10.1016/j.cmet.2016.10.007 39. Buzzetti E. The multiple-hit pathogenesis of non-alcoholic fatty liver disease (NAFLD). Metabolism (2016) 65 1038-1048. DOI: 10.1016/j.metabol.2015.12.012 40. Williams KH. Circulating dipeptidyl peptidase-4 activity correlates with measures of hepatocyte apoptosis and fibrosis in non-alcoholic fatty liver disease in type 2 diabetes mellitus and obesity: a dual cohort cross-sectional study. J Diabetes (2015) 7 809-819. DOI: 10.1111/1753-0407.12237 41. Wree A. NLRP3 inflammasome activation results in hepatocyte pyroptosis, liver inflammation, and fibrosis in mice. Hepatology (2014) 59 898-910. DOI: 10.1002/hep.26592 42. Wada N. Serum-inducible protein (IP)-10 is a disease progression-related marker for non-alcoholic fatty liver disease. Hepatol Int (2017) 11 115-124. DOI: 10.1007/s12072-016-9773-y 43. Berres ML. Antagonism of the chemokine Ccl5 ameliorates experimental liver fibrosis in mice. J Clin Invest (2010) 120 4129-4140. DOI: 10.1172/JCI41732 44. Papatheodoridi AM. The role of senescence in the development of nonalcoholic fatty liver disease and progression to nonalcoholic steatohepatitis. Hepatology (2020) 71 363-374. DOI: 10.1002/hep.30834 45. Kim KM. Identification of senescent cell surface targetable protein DPP4. Genes Dev (2017) 31 1529-1534. DOI: 10.1101/gad.302570.117 46. Lopes-Paciencia S. The senescence-associated secretory phenotype and its regulation. Cytokine (2019) 117 15-22. DOI: 10.1016/j.cyto.2019.01.013 47. Medeiros Tavares Marques JC. Identification of new genes associated to senescent and tumorigenic phenotypes in mesenchymal stem cells. Sci Rep (2017) 7. DOI: 10.1038/s41598-017-16224-5 48. el-Deiry WS. WAF1, a potential mediator of p53 tumor suppression. Cell (1993) 75 817-825. DOI: 10.1016/0092-8674(93)90500-P 49. Kazankov K. The role of macrophages in nonalcoholic fatty liver disease and nonalcoholic steatohepatitis. Nat Rev Gastroenterol Hepatol (2019) 16 145-159. DOI: 10.1038/s41575-018-0082-x 50. Weiskirchen R. Organ and tissue fibrosis: molecular signals, cellular mechanisms and translational implications. Mol Aspects Med (2019) 65 2-15. DOI: 10.1016/j.mam.2018.06.003 51. Zhong J. A potential role for dendritic cell/macrophage-expressing DPP4 in obesity-induced visceral inflammation. Diabetes (2013) 62 149-157. DOI: 10.2337/db12-0230 52. Romacho T. DPP4 deletion in adipose tissue improves hepatic insulin sensitivity in diet-induced obesity. Am J Physiol Endocrinol Metab (2020) 318 E590-E599. DOI: 10.1152/ajpendo.00323.2019 53. Lee YS. Hepatocyte-specific HIF-1α ablation improves obesity-induced glucose intolerance by reducing first-pass GLP-1 degradation. Sci Adv (2019) 5. DOI: 10.1126/sciadv.aaw4176 54. Jacobs RL. Impaired de novo choline synthesis explains why phosphatidylethanolamine N-methyltransferase-deficient mice are protected from diet-induced obesity. J Biol Chem (2010) 285 22403-22413. DOI: 10.1074/jbc.M110.108514 55. Gao X. Decreased lipogenesis in white adipose tissue contributes to the resistance to high fat diet-induced obesity in phosphatidylethanolamine N-methyltransferase-deficient mice. Biochim Biophys Acta (2015) 1851 152-162. DOI: 10.1016/j.bbalip.2014.11.006 56. Itou M. Dipeptidyl peptidase-4: a key player in chronic liver disease. World J Gastroenterol (2013) 19 2298-2306. DOI: 10.3748/wjg.v19.i15.2298 57. Chang ML. Metabolic alterations and hepatitis C: from bench to bedside. World J Gastroenterol (2016) 22 1461-1476. DOI: 10.3748/wjg.v22.i4.1461 58. Doyle MA. Hepatitis C direct acting antivirals and ribavirin modify lipid but not glucose parameters. Cells (2019) 8. DOI: 10.3390/cells8030252 59. Maes M. Treatment with interferon-alpha (IFN alpha) of hepatitis C patients induces lower serum dipeptidyl peptidase IV activity, which is related to IFN alpha-induced depressive and anxiety symptoms and immune activation. Mol Psychiatry (2001) 6 475-480. DOI: 10.1038/sj.mp.4000872 60. Meissner EG. Dynamic changes of post-translationally modified forms of CXCL10 and soluble DPP4 in HCV subjects receiving interferon-free therapy. PLoS One (2015) 10. DOI: 10.1371/journal.pone.0133236 61. Baggio LL, Drucker DJ. Glucagon-like peptide-1 receptor co-agonists for treating metabolic disease. Mol Metab (2021) 46. DOI: 10.1016/j.molmet.2020.101090 62. Drucker DJ. Mechanisms of action and therapeutic application of glucagon-like peptide-1. Cell Metab (2018) 27 740-756. DOI: 10.1016/j.cmet.2018.03.001 63. Burcelin R. Glucose competence of the hepatoportal vein sensor requires the presence of an activated glucagon-like peptide-1 receptor. Diabetes (2001) 50 1720-1728. DOI: 10.2337/diabetes.50.8.1720 64. Aulinger BA. Rapid hepatic metabolism blunts the endocrine action of portally infused GLP-1 in male rats. Am J Physiol Endocrinol Metab (2020) 318 E189-E197. DOI: 10.1152/ajpendo.00298.2019 65. Trzaskalski NA. Dipeptidyl peptidase-4 at the interface between inflammation and metabolism. Clin Med Insights Endocrinol Diabetes (2020) 13. DOI: 10.1177/1179551420912972 66. Jejelava N. Intestinal lymph as a readout of meal-induced GLP-1 release in an unrestrained rat model. Am J Physiol Regul Integr Comp Physiol (2018) 314 R724-R733. DOI: 10.1152/ajpregu.00120.2017 67. Shin JW. Lymphatic-specific expression of dipeptidyl peptidase IV and its dual role in lymphatic endothelial function. Exp Cell Res (2008) 314 3048-3056. DOI: 10.1016/j.yexcr.2008.07.024 68. Malbert CH. Glucose sensing mediated by portal glucagon-like peptide 1 receptor is markedly impaired in insulin-resistant obese animals. Diabetes (2021) 70 99-110. DOI: 10.2337/db20-0361 69. Barchetta I. Circulating dipeptidyl peptidase-4 is independently associated with the presence and severity of NAFLD/NASH in individuals with and without obesity and metabolic disease. J Endocrinol Invest (2021) 44 979-988. DOI: 10.1007/s40618-020-01392-5 70. Kawakubo M. Dipeptidyl peptidase-4 inhibition prevents nonalcoholic steatohepatitis-associated liver fibrosis and tumor development in mice independently of its anti-diabetic effects. Sci Rep (2020) 10. DOI: 10.1038/s41598-020-57935-6 71. Kaji K. Dipeptidyl peptidase-4 inhibitor attenuates hepatic fibrosis via suppression of activated hepatic stellate cell in rats. J Gastroenterol (2014) 49 481-491. DOI: 10.1007/s00535-013-0783-4 72. Koyama Y, Brenner DA. Liver inflammation and fibrosis. J Clin Invest (2017) 127 55-64. DOI: 10.1172/JCI88881 73. Guo Y. NF- κ B/HDAC1/SREBP1c pathway mediates the inflammation signal in progression of hepatic steatosis. Acta Pharm Sin B (2020) 10 825-836. DOI: 10.1016/j.apsb.2020.02.005 74. Dela Pena A. NF-kappaB activation, rather than TNF, mediates hepatic inflammation in a murine dietary model of steatohepatitis. Gastroenterology (2005) 129 1663-1674. DOI: 10.1053/j.gastro.2005.09.004 75. Zeng Y. The DPP-4 inhibitor sitagliptin attenuates the progress of atherosclerosis in apolipoprotein-E-knockout mice via AMPK- and MAPK-dependent mechanisms. Cardiovasc Diabetol (2014) 13. DOI: 10.1186/1475-2840-13-32 76. Shinjo T. DPP-IV inhibitor anagliptin exerts anti-inflammatory effects on macrophages, adipocytes, and mouse livers by suppressing NF-κB activation. Am J Physiol Endocrinol Metab (2015) 309 E214-E223. DOI: 10.1152/ajpendo.00553.2014 77. Meijnikman AS. Evaluating causality of cellular senescence in non-alcoholic fatty liver disease. JHEP Rep (2021) 3. DOI: 10.1016/j.jhepr.2021.100301 78. Chen Z. Dipeptidyl peptidase-4 inhibition improves endothelial senescence by activating AMPK/SIRT1/Nrf2 signaling pathway. Biochem Pharmacol (2020) 177. DOI: 10.1016/j.bcp.2020.113951 79. Diaz-Jimenez D. Glucocorticoids mobilize macrophages by transcriptionally up-regulating the exopeptidase DPP4. J Biol Chem (2020) 295 3213-3227. DOI: 10.1074/jbc.RA119.010894 80. Wang T. Targeting cellular senescence prevents glucocorticoid-induced bone loss through modulation of the DPP4-GLP-1 axis. Signal Transduct Target Ther (2021) 6. DOI: 10.1038/s41392-021-00528-0 81. Rufini A. Senescence and aging: the critical roles of p53. Oncogene (2013) 32 5129-5143. DOI: 10.1038/onc.2012.640 82. Xie Y. The tumor suppressor p53 limits ferroptosis by blocking DPP4 activity. Cell Rep (2017) 20 1692-1704. DOI: 10.1016/j.celrep.2017.07.055 83. Svendsen B. Insulin secretion depends on intra-islet glucagon signaling. Cell Rep (2018) 25 1127-1134. DOI: 10.1016/j.celrep.2018.10.018 84. Capozzi ME. β cell tone is defined by proglucagon peptides through cAMP signaling. JCI Insight (2019) 4. DOI: 10.1172/jci.insight.126742 85. Zhu L. Intra-islet glucagon signaling is critical for maintaining glucose homeostasis. JCI Insight (2019) 5. DOI: 10.1172/jci.insight.127994 86. Song Y. Gut-proglucagon-derived peptides are essential for regulating glucose homeostasis in mice. Cell Metab (2019) 30 976-986. DOI: 10.1016/j.cmet.2019.08.009 87. Chepurny OG. Nonconventional glucagon and GLP-1 receptor agonist and antagonist interplay at the GLP-1 receptor revealed in high-throughput FRET assays for cAMP. J Biol Chem (2019) 294 3514-3531. DOI: 10.1074/jbc.RA118.005682 88. Marguet D. Enhanced insulin secretion and improved glucose tolerance in mice lacking CD26. Proc Natl Acad Sci U S A (2000) 97 6874-6879. DOI: 10.1073/pnas.120069197 89. Walkey CJ. Disruption of the murine gene encoding phosphatidylethanolamine. Proc Natl Acad Sci U S A (1997) 94 12880-12885. DOI: 10.1073/pnas.94.24.12880 90. Fullerton MD. Single phosphorylation sites in Acc1 and Acc2 regulate lipid homeostasis and the insulin-sensitizing effects of metformin. Nat Med (2013) 19 1649-1654. DOI: 10.1038/nm.3372 91. Chambers AP. The role of pancreatic preproglucagon in glucose homeostasis in mice. Cell Metab (2017) 25 927-934. DOI: 10.1016/j.cmet.2017.02.008 92. Folch J. A simple method for the isolation and purification of total lipides from animal tissues. J Biol Chem (1957) 226 497-509. DOI: 10.1016/S0021-9258(18)64849-5 93. Patel R. LXRβ is required for glucocorticoid-induced hyperglycemia and hepatosteatosis in mice. J Clin Invest (2011) 121 431-441. DOI: 10.1172/JCI41681 94. Rao X. Oxidized LDL upregulates macrophage DPP4 expression via TLR4/TRIF/CD36 pathways. EBioMedicine (2019) 41 50-61. DOI: 10.1016/j.ebiom.2019.01.065 95. Metsalu T, Vilo J. ClustVis: a web tool for visualizing clustering of multivariate data using principal component analysis and heatmap. Nucleic Acids Res (2015) 43 W566-W570. DOI: 10.1093/nar/gkv468 96. Baggio LL. Plasma levels of DPP4 activity and sDPP4 are dissociated from inflammation in mice and humans. Nat Commun (2020) 11. DOI: 10.1038/s41467-020-17556-z
--- title: Inhibition of indoleamine dioxygenase leads to better control of tuberculosis adjunctive to chemotherapy authors: - Bindu Singh - Chivonne Moodley - Dhiraj K. Singh - Ruby A. Escobedo - Riti Sharan - Garima Arora - Shashank R. Ganatra - Vinay Shivanna - Olga Gonzalez - Shannan Hall-Ursone - Edward J. Dick - Deepak Kaushal - Xavier Alvarez - Smriti Mehra journal: JCI Insight year: 2023 pmcid: PMC9977315 doi: 10.1172/jci.insight.163101 license: CC BY 4.0 --- # Inhibition of indoleamine dioxygenase leads to better control of tuberculosis adjunctive to chemotherapy ## Abstract The expression of indoleamine 2,3-dioxygenase (IDO), a robust immunosuppressant, is significantly induced in macaque tuberculosis (TB) granulomas, where it is expressed on IFN-responsive macrophages and myeloid-derived suppressor cells. IDO expression is also highly induced in human TB granulomas, and products of its activity are detected in patients with TB. In vivo blockade of IDO activity resulted in the reorganization of the granuloma with substantially greater T cells being recruited to the core of the lesions. This correlated with better immune control of TB and reduced lung M. tuberculosis burdens. To study if the IDO blockade strategy can be translated to a bona fide host-directed therapy in the clinical setting of TB, we studied the effect of IDO inhibitor 1-methyl-d-tryptophan adjunctive to suboptimal anti-TB chemotherapy. While two-thirds of controls and one-third of chemotherapy-treated animals progressed to active TB, inhibition of IDO adjunctive to the same therapy protected macaques from TB, as measured by clinical, radiological, and microbiological attributes. Although chemotherapy improved proliferative T cell responses, adjunctive inhibition of IDO further enhanced the recruitment of effector T cells to the lung. These results strongly suggest the possibility that IDO inhibition can be attempted adjunctive to anti-TB chemotherapy in clinical trials. ## Mycobacterium tuberculosis (M. tuberculosis), an important intracellular pathogen, causes approximately 1.8 million deaths every year through tuberculosis (TB) (1). Infection of human lungs with M. tuberculosis is characterized by a robust innate and adaptive immune response, which eventually leads to the formation of a pathological lesion called the granuloma, a hallmark of TB that influences the outcome of the infection [1]. It is believed that the granuloma locally helps contain the infection, though the specific mechanisms by which the granuloma exerts immune control of M. tuberculosis, i.e., the spatial understanding of the granuloma function, have not been completely understood [2]. The architecture and composition of the granuloma can, however, directly influence both the phenotype of the pathogen as well as the host immune response, thus affecting the disease outcome in many different ways [3]. Single cell–based approaches are now being used to study gene and protein expression to better understand the TB lung (4–6). These techniques leverage both opportunistically available human granuloma samples as well as those from experimentally infected animal models [4, 7] and provide a much more detailed picture of granuloma gene expression, cellular composition, and function. Multiplexed imaging of the human TB granuloma recently revealed the highly immunosuppressed nature of the granuloma microenvironment [7] — human TB granulomas are depleted for IFN-γ+ cells but instead enriched for TGF-β, regulatory T cells (Tregs), and indoleamine 2,3-dioxygenase+ (IDO+) programmed cell death ligand 1+ myeloid cells. IDO is one of the most abundant proteins present in human TB granulomas [7]. IDO catabolizes the essential amino acid tryptophan (Trp) to kynurenine (Kyn) [8] and exerts robust direct and indirect immunosuppressive effects on T cell activation [9]. Several studies in nonhuman primates have also reported that the TB granuloma is an immunosuppressive environment. Thus, while newly formed TB granulomas from rhesus macaques express a robust proinflammatory gene signature, this is rapidly reprogrammed by the expression of genes involved in tissue remodeling [10], coincident with the development of necrosis and hypoxia in these macaque lesions [11]. This causes the pathogen to alter its in vivo phenotype dependent on the DosR regulon for intragranulomatous persistence [11, 12]. Gideon et al. showed that an exceedingly small number of T cells from granulomas derived from cynomolgus macaques infected with M. tuberculosis respond with cytokine production after stimulation with M. tuberculosis–specific antigens, and few “multifunctional” T cells were observed [13]. Our work in both rhesus and cynomolgus macaques showed that after M. tuberculosis infection, the expression of IDO is significantly induced in the myeloid layer of nonhuman primate TB granulomas [10, 14]. Expression of IDO in response to M. tuberculosis infection is not limited to primate hosts but can be observed in murine models as well as in macrophages in vitro [15, 16]. The induction of IDO expression in the lung occurs in a manner directly proportional to the burden of M. tuberculosis [16]. We have since identified that IDO is primarily expressed on IFN-responsive, inflammatory, interstitial macrophages in the lungs of M. tuberculosis–infected macaques [4], as well as on immunosuppressive myeloid-derived suppressor cells (MDSCs) [17]. A compendium of studies have since shown that IDO expression is highly induced in the human TB granuloma environment, and products of IDO-mediated Trp catabolism are detected in the plasma, sera, and urine of patients with active TB, including multidrug-resistant tuberculosis (MDR-TB) as well as TB/HIV, in cohorts from various regions of the world (18–20). Taken together, these results from animal models of TB as well as patients unequivocally show that the expression of IDO, a potent immunosuppressor of T cell activity, is induced in macrophages infected with M. tuberculosis and that this expression can be clearly observed in granulomas, which are a highly immunosuppressed environment. In the case of many intracellular pathogenic organisms, e.g., Chlamydia, Leishmania, Coxiella and Listeria, host-mediated catabolism of the essential amino acid Trp, initiated by the activation of the rate-limiting enzyme, IDO, represents an effective means of innate immune control (21–24). Granuloma-resident M. tuberculosis is, however, able to synthesize Trp [12, 25, 26]. Unfortunately, therefore, the host’s strategy to deplete Trp is ineffective during M. tuberculosis infection, and furthermore, downstream Kyn metabolites of this pathway impair phagosome/lysosome fusion and autophagy [27], processes that serve to kill intracellular M. tuberculosis. Furthermore, downstream metabolites of the IDO pathway serve to impair the function of CD4+ T cells, via expanding Tregs and MDSCs [17, 28]. Coupled with the lack of Trp for proliferating T cells, these mechanisms create an immunosuppressive environment conducive to the persistence of M. tuberculosis. Thus, the IDO pathway is ineffective and actually deleterious during M. tuberculosis infection. Since IDO is a powerful suppressant of T cell function, this ineffectiveness of the IDO pathway to control M. tuberculosis in vivo, coupled with the strong induction of IDO in TB granulomas, together suggest that blockade of IDO activity in vivo may also serve as an attractive host-directed therapy (HDT) target for TB. Our group has developed a macaque model of M. tuberculosis infection via the natural, aerosol route of exposure [11, 29, 30]. Based on the choice of strain and dose of infection, macaques either develop immune control of M. tuberculosis similar to latent tuberculosis infection (LTBI) [11, 31] or progress to pulmonary TB [29, 30]. Coinfection of controllers with simian immunodeficiency virus SIVmac239 results in the reactivation of LTBI (31–34). Furthermore, in our model, reactivation strongly correlates with the presence of chronic immune activation in lungs [35]. We therefore investigated the ability of an IDO inhibitor to provide antimicrobial activity as well as enhance adaptive and innate immunity in a macaque model of active TB. In vivo blockade of IDO activity, using monotherapy with 1-methyl-d-tryptophan (D1MT), during M. tuberculosis infection leading to TB was indeed beneficial to the host [16]. However, in real life, IDO inhibitors are unlikely to be used as monotherapy; furthermore, our model of active TB, while beneficial for the evaluation of vaccines and therapeutics, requires exposure to a nonphysiological high dose of M. tuberculosis for infection and may not represent a real-life situation. We therefore devised an experiment where we tested the effectiveness of D1MT to adjunctively enhance the effectiveness of a deliberately suboptimal chemotherapeutic regimen against TB, in a model where macaques were infected with a dose of drug-sensitive M. tuberculosis CDC1551, a low-virulence strain, such that about $50\%$ of the animals develop active TB [11]. Our results suggest that IDO inhibition can indeed improve immune responses and adjunctively enhance the chemotherapeutic potential of anti-TB therapy. ## M. tuberculosis infection and inclusion of animals in different groups of treatment. A total of 18 Indian origin rhesus macaques (RMs) were infected with approximately 25–50 CFU M. tuberculosis CDC1551 (Figure 1A). We have previously shown that this dose of infection with a low-virulence strain of M. tuberculosis results in the progression of approximately $50\%$ of the RMs to active TB over 2–4 months. The experimental design illustration (Figure 1A) includes details about the infection, treatment groups, the period of treatment, and procedures performed. As indicated, PET/CT scans were performed at a preinfection time point, as well as at weeks 6, 12, and 18 (or earlier if necessary, at the endpoint). RMs were assigned to 1 of 3 groups based on the week 6 PET/CT scores. The 3 groups were untreated control group receiving no treatment following M. tuberculosis infection; ME treatment group receiving moxifloxacin and ethambutol (M, 10 mg/kg; E, 20 mg/kg) regimen following M. tuberculosis infection between weeks 7 and 19, i.e., for 12 weeks; and ME/D1MT treatment group receiving moxifloxacin and ethambutol (M, 10 mg/kg; E, 20 mg/kg) regimen for 12 weeks as described above along with concurrent treatment with 45 mg/kg IDO inhibitor D1MT daily, at the beginning of the chemotherapy, between weeks 7 and 11. Because of our choice of the dose of M. tuberculosis for aerosol infection, some animals developed disease, as measured by high PET/CT scores (scores > 3), while others showed signs of TB (scores 2–3), and yet others had minimal evidence of TB (scores 1) (Figure 1, B and C). This is unlike aerosol infection with a high dose of M. tuberculosis, where all animals rapidly develop active TB [16], or with a very low dose of M. tuberculosis, where nearly all animals develop LTBI [33, 34]. PET/CT scores generated in a blinded fashion from macaques imaged at week 6, before the initiation of any therapy, are shown in Figure 1C. The macaques were assigned to 1 of the 3 groups such that each group had an even distribution of scores and that no bias was introduced due to heterogeneity in disease levels. As is evident from the graph shown in Figure 1C, there were no differences in the PET/CT scores of the macaques in the 3 groups before initiation of treatment. ## Treatment with ME (alone as well as in combination with D1MT) results in improvement of clinical parameters. Both groups of RMs that received ME treatment (ME and ME/D1MT) harbored significantly lower clinical levels of TB disease as measured by serum C-reactive protein (CRP) levels (Figure 1D) and percentage change in body weights (Figure 1E). Serum CRP levels in the macaques of the ME/D1MT group were found to be significantly reduced after 12 weeks of treatment as compared with untreated group (Figure 1D). As is evident from the line graph shown in Figure 1E, upon the treatment initiation, all animals initially lost weight. Upon the initiation of treatment, the ME/D1MT group started gaining weight, whereas the ME-only and the untreated group continued to lose weight. Around week 12 (after 5 weeks of ME treatment), the ME group also started gaining weight but to a lesser extent as compared with the ME/D1MT group. However, the control group continued to exhibit weight loss till the endpoint, such that on an average every RM in this group had lost an average of more than 0.5 kg in body weight compared with preinfection levels (Figure 1E). The graph depicting the change in weight at the endpoint (or week 19) is shown in Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.163101DS1 No notable differences were observed in change in temperature among the 3 groups (Supplemental Figure 1B). ## Inclusion of D1MT adjunctive to ME results in the inhibition of IDO enzymatic activity in a model of heterogeneous TB progression. We measured IDO enzymatic activity by detecting Kyn (the major end product of the IDO pathway) by immunofluorescence staining in BAL-derived cells from the 3 groups of RMs collected at pretreatment time point (week 7) and after D1MT and ME treatment for 4 weeks (week 11). Representative images of BAL cells stained with Kyn antibody are shown in Figure 2, A and B. Quantification of Kyn-positive cells in BAL revealed that before the initiation of treatment (i.e., at week 7), all the macaques had high levels of Kyn in BAL, which accounted for the presence of $65\%$–$80\%$ Kyn-positive cells, which was comparable among the 3 groups. However, after 4 weeks of treatment, both the ME/D1MT and the ME groups exhibited a significant reduction in the percentage of Kyn-positive cells as compared with the untreated group as shown in Figure 2C. The reduction was more evident in the ME/D1MT group, with the presence of only approximately $26\%$ Kyn-positive cells compared with approximately $48\%$ in the ME group. Inclusion of D1MT for 4 weeks, therefore, resulted in significant reduction of IDO activity. Although the differences between the 2 treated groups were highly significant, the percentages of Kyn-positive cells were also significantly reduced in both treatment groups as compared with the untreated one. We also calculated the ratio of Kyn-positive cells pretreatment versus posttreatment, where we observed significantly high values in the ME/D1MT group as compared with untreated as well as ME-only groups as depicted in Figure 2D. IDO levels in lung (at endpoint) and BAL cells (before and after D1MT treatment) were also assessed by quantitative reverse transcription PCR (qRT-PCR). We could see some reduction in IDO levels in lungs (Supplemental Figure 2A), though the difference was statistically not significant. However, no marked differences were observed in BAL cells before and after D1MT treatment (Supplemental Figure 2E). The levels of a few other gene transcripts, including IDO2, IFN-γ, and IFN-β, were also quantified by qRT-PCR, with no notable differences (Supplemental Figure 2, B–D and F–H). We next performed multi-labeled immunohistochemistry by staining lung tissues for IDO expression in macrophages, using anti-IDO and anti-CD68 antibodies. The representative confocal images from untreated, ME only–treated, and ME/D1MT-treated macaques are shown in Figure 2E (20× original magnification) and Figure 2F (63× original magnification). The third panel of Figure 2E shows the IDO expression in granuloma present in the ME/D1MT-treated animals, and the fourth panel depicts IDO expression in a resolved granuloma. The latter was a more common occurrence in ME/D1MT-treated RMs. The images were analyzed using HALO analysis software in a blinded manner to quantify the percentage of IDO-expressing cells in lungs (Figure 2G) as well as IDO-expressing macrophages (Figure 2H) and other cell types (Figure 2I). We observed significant decrease in IDO-positive cells in the ME/D1MT-treated group as compared with the untreated group (Figure 2G). However, no notable difference was observed in the percentages of IDO-expressing cells in the ME/D1MT group in comparison to the ME-only group. Interestingly, no significant differences were observed between the ME and untreated groups. The ME/D1MT group had the least bacillary burden and pathology of all 3 groups tested and was therefore likely to harbor less cellular infiltration. This could explain why statistically significant differences were not obtained between the 2 treatment groups. However, no significant differences were observed in the fraction of IDO-expressing macrophages in the lungs of the 3 groups of RMs (Figure 2H), but IDO-expressing cell types other than macrophages were significantly higher in the untreated group in comparison with the ME/D1MT group (Figure 2I). ## Inclusion of D1MT adjunctive to ME treatment results in better control of M. tuberculosis infection with complete clearance of bacilli. M. tuberculosis burden longitudinally assessed in BAL at various points of the study timeline (Figure 3A), and at the endpoint in the same sample (Figure 3B), showed no marked differences among the 3 groups. A more prominent decrease in the ME/D1MT group was also observed at the endpoint relative to the ME group (Figure 3B). Significant reduction was observed in M. tuberculosis levels at the endpoint in lungs in the ME/D1MT group versus untreated group, with no significant reduction in the ME/D1MT group as compared with ME only (Figure 3C). M. tuberculosis burden assessment in lung granulomas also showed a significant reduction in the treated groups relative to the untreated group (Figure 3D). We also determined CFUs in mesenteric lymph node, liver, spleen, and kidney. We observed that the macaques in the ME/D1MT group had completely sterile organs, with few detectable CFUs in untreated and ME groups, but no statistical differences among the 3 groups were found (Supplemental Figure 1). CFU results from the lung-draining lymph nodes also depicted a significant decrease in bacterial load in the ME/D1MT but not in the ME treatment group relative to the untreated one (Figure 3E). These data suggest that D1MT in adjunct to ME therapy is more efficient in reducing/sterilizing M. tuberculosis than the suboptimal therapy with ME alone, whereas ME alone, while reducing M. tuberculosis burdens in lungs and granulomas of RMs, fails to sterilize the tissues completely (Figure 3, B–E). The extent of sterile lobes in the lungs (Figure 3F) and the number of individual granulomas derived from the lungs that were sterile (Figure 3G) were significantly higher in ME/D1MT-treated lungs and granulomas relative to the ME-only group. H&E staining (Figure 3H) and quantification of affected lung area (marked by inflammation and presence of lesions) depicted a marked decrease in percentage lung involvement in the ME/D1MT group with respect to the untreated group as well as reduced lung involvement when compared with the ME-only group (Figure 3I). Analysis of gross lung pathology at the endpoint was congruent with these results, with the greatest percentage of pathology being observed in the control group, followed by the ME group and then the ME/D1MT group, which exhibited the least amount of lung pathology (Supplemental Figure 1, D–F). The representative H&E images from each macaque are depicted in Supplemental Figure 3. ## Validation of the superior effectiveness of D1MT in controlling M. tuberculosis infection adjunctive to ME in RMs, by PET/CT radiology. TB pathogenesis and efficacy of ME and ME/D1MT prophylaxis regimens were examined using PET/CT scans [36]. We used PET/CT as the primary correlate of progression of M. tuberculosis infection in the lungs of RMs. As described earlier, week 6 PET/CT images and scores were used to assign animals to different groups so that each group contained RMs with comparable disease progression at that time point. We also performed PET/CT imaging at week 12 (which was at the end of the D1MT treatment in the ME/D1MT group and one-third of the way in the 12-week ME treatment group). Finally, PET/CT imaging/analysis was performed at week 18 or endpoint, if earlier. All the macaques ($\frac{18}{18}$) in the study had clear lungs prior to infection and focal nodular lung opacities at week 6 (pretreatment). The average PET/CT score in each group was 2.67, 2.33, and 2.2, respectively, for the control, ME, and ME/D1MT groups. All 18 animals had mild to moderate lymph node enlargement by 6 weeks after aerosol M. tuberculosis infection. The 18F-fluorodeoxyglucose (FDG) scans performed at either 12 or 18 weeks postinfection or at endpoint, if earlier, clearly revealed both the presence of persistent infection in the controls (Figure 4A) and the partial effectiveness of the ME regimen (Figure 4B) at the completion of the treatment; furthermore, these results exhibit the enhanced effectiveness of the ME regimen when D1MT was included in the treatment for the first 4 weeks (Figure 4C). Scans in the treated groups displayed no new lung lesions, while the previously reported lung lesions were resolved, although to a higher level in the ME/D1MT over ME group, i.e., no increase in lesion volume and no increase in FDG uptake (Figure 4D). In contrast, control animals displayed an increase in the size of lung lesions and increased FDG standardized uptake value (SUV) (Figure 4, D and E). All control animals showed TB and further progression of lung TB pathology including involvement of multiple lung lobes, with consolidation, lobar collapse, cavitary lesions, and massive mediastinal lymph node enlargement. The number and volume of lung lesions (Figure 4D), mean SUV (Figure 4E), and total lung activity (Figure 4F) of control animals was each higher compared with 2 treatment groups after treatment completion. Importantly, these values were lower for the ME/D1MT group relative to the ME-only group, clearly suggesting an advantage due to IDO blockade (Figure 4, D–F); however, the differences were not statistically significant between the 2 treatment groups. Our results suggest that the extent of TB pathology and disease increased over time in control, untreated RMs (Figure 4A), as compared with ME only–treated (Figure 4B) and ME/D1MT-treated groups (Figure 4C), and that these groups of animals harbored different levels of pulmonary disease as measured by the various radiology attributes (Figure 4, D–F). M. tuberculosis infection led to development of varying degrees of active TB in untreated animals, as demonstrated by the presence of numerous granulomatous lesions by CT scans (Figure 4A) and the increased volume of FDG lesions (Figure 4D). All untreated, M. tuberculosis–infected animals had substantial evidence of granulomatous lesions (Figure 4A). Furthermore, the RMs in the 2 treatment groups did not demonstrate the presence of a significant number of lesions at week 12/endpoint or 18/endpoint (Figure 4, B and C). RMs in the control group showed gradual progression in TB pathology over time with multiple new lung lesions and an increase in size of previously emergent nodular lung lesions. ## Superior control of M. tuberculosis infection by ME adjunctive to D1MT is accompanied by improved T cell immune responses. We have shown earlier that D1MT treatment of RMs improved granuloma-specific immune responses. Here, we studied if overall T cell activation, proliferation, and recruitment as well as their antigen specificity was improved in the ME/D1MT relative to ME and control groups. Figure 5 shows the percentages of total CD3+ cells, CD4+ cells, CD8+ cells, CD4+ effector T cells, CD4+ memory T cells, CD4+ Ki67+ T cells, CD8+ effector T cells, CD8+ memory T cells, and CD8+Ki67+ T cells in BAL (Figure 5, A–I) and PBMCs (Figure 5, J–R) of the 3 groups at various time points of the study. It was only at the endpoint (Supplemental Figure 5) that we observed higher percentages of CD4+ effector T cells in BAL (Figure 5D and Supplemental Figure 5D) of ME/D1MT macaques in comparison with ME only–treated and untreated groups, with no changes observed in CD3+ T cells (Figure 5, A and J), CD4+ T cells (Figure 5, B and K), CD8+ T cells (Figure 5, C and L), CD4+ memory T cells (Figure 5, E and N), CD8+ effector T cells (Figure 5, G and P), and CD8+ memory T cells (Figure 5, H and Q). Both the treated groups exhibited significantly higher percentages of proliferative CD4+Ki67+ T cells (Figure 5F and Supplemental Figure 5F) and CD8+Ki67+ T cells (Figure 5I and Supplemental Figure 5I) in BAL relative to the untreated group. Flow cytometry analysis of T cells in lungs at the endpoint showed no significant changes in the overall percentages of CD3+, CD4+, and CD8+ cells; CD4+ effector T cells; CD4+ memory T cells; CD8+ effector T cells; CD8+ memory T cells; CD4+Ki67+ T cells; and CD8+Ki67+ T cells (Figure 6, A–I) among the 3 groups. ## Discussion HDTs are an exciting new area of research in the field of TB (37–42). HDTs seek to modulate specific host immune pathways, including those that affect inflammation and immunopathology, to limit M. tuberculosis infection, persistence, reactivation, dissemination, and resulting pathology (37–42). Interest in HDT for TB is driven by the length of conventional TB therapy regimens and the desire to shorten them to increase compliance, thus reducing the incidence of MDR-TB. The concept of treating TB with adjunctive HDT also incorporates increasing knowledge that productive immune responses are subverted during pulmonary TB. A number of preclinical studies have highlighted promising candidates that either increase the effectiveness of the host to kill M. tuberculosis or reduce the destructive nature of an overexuberant host response, thus enhancing the effectiveness of pathogen-directed chemotherapy (16, 43–45). These approaches include arachidonic acid pathway modulators, NSAIDs [46], phosphodiesterase inhibitors [47], tyrosine kinase inhibitors (e.g., imatinib) [48, 49], antidiabetic drugs (e.g., Metformin) [50], and statins [51]. Arachidonic acid pathway modulators provide a delicate balance in eicosanoid levels, enhancing M. tuberculosis control. NSAIDs interrupt the formation of proinflammatory and immunosuppressive mediators, such as prostaglandins and leukotrienes. Phosphodiesterase inhibitors reduce inflammation by increasing intracellular cAMP. Tyrosine kinase inhibitors are reported to reduce bacillary burden by promoting myelopoiesis, phagosome maturation, acidification, and autophagy. The immunomodulatory effects of antidiabetic drugs promote macrophage autophagy via the AMP kinase/mTOR loop. Statins have displayed control of lipid levels by targeting HMG-CoA reductase. ## M. tuberculosis has strong adjuvant properties and promotes a robust Th1 response, resulting in chronic, local granulomatous inflammation [52, 53]. That M. tuberculosis is deliberately immunogenic is counterintuitive, as the resulting response could eliminate the pathogen [53]. Hence, M. tuberculosis invokes novel local mechanisms to potentiate its survival in the face of this immune stress to complete its life cycle, e.g., modulating TCR signaling [54] in a TLR-dependent manner [55, 56], phagolysosomal fusion [57], apoptosis [58], and IFN-γ [59] or TNF-α signaling [60]. IDO catabolizes Trp [61] to starve pathogens of an essential amino acid [62]. This strategy is, however, ineffective in restricting M. tuberculosis, which can synthesize Trp de novo [26]. IDO blocks T cell proliferation downstream of IFN-γ, as Trp is essential for rapidly dividing effector cells [62]. By reducing local Trp levels, IDO inhibits Th1 functions and generates Tregs, causing immunosuppression. This regulatory role of IDO is well studied in the immune escape of cancers and pregnancy, associated with poor prognosis [63], and linked to bacteremia [64]. A decade ago we discovered that the expression of IDO is induced to very high levels in macaque TB granulomas [10, 14]. Abundant data in both animal models as well as human samples since then strongly suggest that IDO is a key molecule that governs immunosuppression in the granuloma. The expression of IDO is highly induced in murine or primate macrophages [16], B6 [15] or Kramnik [16] mice, and macaques [16] upon M. tuberculosis infection. IDO expression is lowered in active TB (ATB) animals on chemotherapy or during nonpathogenic infection, is not induced in LTBI, and correlates with M. tuberculosis burden [16]. IDO is expressed exclusively on myeloid cells in the inner ring of the granuloma [16]. Single-cell RNA-Seq revealed that the majority of IDO transcript expression takes place on IFN-responsive, inflammatory, interstitial macrophages [4], as well as on immunosuppressive MDSCs [17], in the lungs of M. tuberculosis–infected macaques. IDO expression is not just a feature of TB granulomas in animal models. Single cell–resolution multiplexed ion beam imaging–TOF has revealed that IDO is one of the most highly expressed proteins in granulomas derived from human patients with TB [7]. Furthermore, products of IDO-mediated Trp catabolism are detected in the plasma, sera, and urine of patients with ATB, including MDR-TB as well as TB/HIV, in cohorts from various regions of the world (18–20), correlating with prognosis and inversely with treatment [65]. Inhibition of IDO signaling in M. tuberculosis–infected macaques improved clinical signs, bacterial burden, and lung pathology as a function of inhibition of IDO enzymatic activity [16]. Taken together, these results from animal models of TB as well as patients unequivocally suggest that inhibition of IDO adjunctive to anti-TB chemotherapy is a viable HDT for patients with TB. In the current study, we designed experiments to directly address if the addition of D1MT, an IDO inhibitor, adjunctive to TB chemotherapy improves the clearance of M. tuberculosis from the lungs of infected RMs and lowers the risk of TB disease. RMs closely represent several aspects of human TB, including ATB disease with high bacterial loads and pathology in the lungs, dissemination of M. tuberculosis to extrathoracic regions, and systemic inflammation; LTBI characterized by a lack of overt disease by microbiologic or radiologic measures but with immunological response to M. tuberculosis antigens; and HIV coinfection–mediated reactivation TB [31, 32, 35]. Our model has also been utilized to study vaccine efficacy and mechanisms of protection as well as modeling antiretroviral [33, 34] and TB [66] therapies. We expose RMs to infectious aerosols of M. tuberculosis, thus mimicking the natural route of infection in humans. There are 3 typical models in our lab, where RMs are exposed to M. tuberculosis strain CDC1551, which has somewhat lower pathogenicity than the Erdman strain [67]. RMs infected with 100–200 CFU of M. tuberculosis CDC1551 invariably develop ATB with a $100\%$ progression to euthanasia within 3 months of infection. On the other hand, RMs exposed to very low doses of M. tuberculosis (~5–10 CFU) largely develop asymptomatic LTBI [33, 34]. Exposure of RMs to approximately 25 CFU M. tuberculosis results in some RMs developing disease and others exhibiting control of infection [11]. Treatment of nonhuman primates infected with drug-sensitive M. tuberculosis with frontline anti-TB chemotherapeutic regimen isoniazid + rifampin + pyrazinamide + ethambutol sterilizes lungs in macaques [68]. To study the effectiveness of D1MT as an anti-TB HDT adjunctive to chemotherapy, we decided to develop a model of suboptimal anti-TB chemotherapy. Since HDTs are most needed in populations with drug-resistant TB, most such cases involve resistance to isoniazid (H), and moxifloxacin is used to replace it, we treated 2 groups of M. tuberculosis–infected (~25 CFU) animals with ME. Human equivalent doses of this regimen controlled M. tuberculosis infection but did not result in complete sterilization. This allowed us to study the impact of including D1MT as an adjunctive therapy in the second of the 2 treatment groups. Since IDO is a checkpoint inhibitor, modulation of its activity could lead to overexuberant immune responses and an uncontrolled pathology. We have earlier shown in a model of ATB that 4 to 5 weeks of treatment with D1MT is sufficient to reorganize the granuloma, modulate immune responses, and effect a reduction of M. tuberculosis burdens in RMs. We therefore used a treatment plan where 1 group was treated with ME at a time when many infected animals exhibited ATB (ME treatment group). The other group was similarly treated with ME but was only treated with D1MT for the initial 4 weeks (ME/D1MT treatment group). Our results show that while suboptimal therapy with ME reduces disease measures, inclusion of D1MT for only 4 weeks at the initiation of chemotherapy further enhances sterilization of M. tuberculosis while substantially improving T cell responses. The results from our current study preclinically establish IDO inhibition, using D1MT, an approved, safe molecule currently in clinical trials, as a leading HDT strategy for TB. We have now shown its effectiveness in controlling progression of M. tuberculosis infection in 2 macaque studies, one in the setting of ATB and another in the setting of controlled progression and adjunctive to treatment. Further studies are necessary to elucidate if the mechanisms by which inhibition of IDO results in control of M. tuberculosis infection in both instances are shared and involve greater granuloma performance due to increased access of T cells to lesion core regions. It may also be important to further test the effectiveness of IDO inhibition in a different species of macaque, e.g., the cynomolgus macaque species, and in the setting of HIV coinfection. It would also be useful to test the effectiveness of D1MT in improving granuloma performance in the setting of LTBI. Fast-tracking this and other novel HDT strategies for TB in the clinical space may significantly improve treatment of TB. ## Animals, infection, sampling, and euthanasia. This study included 18 Indian origin RMs (Macaca mulatta) from 2 studies. Data were included from our recently completed studies, wherein specific pathogen–free, mycobacteria-naive Indian origin RMs were enrolled to the protocol after being obtained from a colony maintained at the Tulane National Primate Research Center (TNPRC, $$n = 9$$), the Southwest National Primate Research Center (SNPRC) ($$n = 6$$), or the Caribbean Primate Research Center (CPRC, $$n = 3$$) (Supplemental Table 1). All macaques were infected with an intermediate dose of approximately 25–50 CFU M. tuberculosis CDC1551 (BEI Resources, catalog NR13649) via aerosol as described before (14, 29–31). Tuberculin skin test was performed at weeks 3 and 5 after M. tuberculosis infection to confirm the infection. RMs were monitored for CRP, percentage changes in body weight and body temperature, and CBC weekly through the study period. A total of 12 macaques were then treated with the ME (M, 10 mg/kg; E, 20 mg/kg) regimen, beginning week 7 postinfection, for 12 weeks. A total of 6 of these RMs were also treated with D1MT daily (45 mg/kg, Sigma-Aldrich), as described earlier, for 4 weeks (week 7–11). This group that was also treated with D1MT is referred to as the ME/D1MT group while the group that only received TB chemotherapy is referred to as the ME group. A total of 6 RMs remained naive of all treatments during the protocol. The study demographics are presented in Supplemental Table 1. The assignment of macaques into 3 experimental groups was as follows. The D1MT/ME group included 2 macaques obtained from the TNPRC, 2 from the SNPRC, and 1 from the CPRC. The ME-only group included 2 macaques each from the SNPRC, TNPRC, and CPRC. The untreated group included 4 macaques from the TNPRC and 1 each from the SNPRC and CPRC. The macaque that received partial treatment was obtained from the TNPRC (please see Supplemental Table 1 for more details). This macaque was initially included in the ME/D1MT treatment group but rapidly progressed to ATB, and a decision was made by the veterinarian to euthanize it 2 days later. ## PET/CT. Three sequential PET/CT scans were performed, using Mediso’s LFER150 PET/CT scanner, at 6, 12, and $\frac{18}{19}$ weeks after M. tuberculosis infection with the last scan prior to necropsy. PET/CT scanning was essentially performed as described earlier [66]. Briefly, we performed FDG PET/CT scans for each anesthetized macaque using the breath-hold technique [69, 70]. Animals were anesthetized and intubated under supervision of a veterinarian as per approved IACUC protocols. All the animals received an intravenous injection of 5 mCi of FDG [71] in the right arm, procured from Cardinal Health radio pharmacy. Single- and a double–field of view CT scans were performed using breath-hold as described [72]. The single–field of view (single-FOV) CT scan was performed with bread-hold as described previously [34] to obtain a clear reconstructed image of the lung; the 2-FOV scan was used for the reconstruction of the PET as the material map. Two FOV PET scans were acquired after a 45-minute FDG uptake period. Images were visualized using Interview Fusion 3.03 (Mediso) and reconstructed using Nucline nanoScan LFER 1.07 (Mediso) with parameters as described [73]. 3D image analysis was performed using VivoQuant 4.0 (Invicro) [74] to calculate the SUV in the M. tuberculosis lesions observed in the lung. ## Microbiological evaluation. Mycobacterial burden in BAL was measured throughout the study period as previously described [32]. Viable M. tuberculosis burden was also measured at necropsy in BAL, lung, spleen, bronchial lymph node, mesenteric lymph node, liver, kidney, and individual granulomas collected at necropsy from each macaque [32, 75]. ## Pathology evaluation. This was performed as previously described [34]. Briefly, the RMs were anesthetized for necropsy, and lung lobes, spleen, liver, kidney, bronchial lymph nodes, BAL, and blood were collected. Tissues were fixed in $10\%$ neutral-buffered formalin, paraffin-embedded, sectioned at 5 μm thickness, and stained with H&E using standard methods. Stereology scores were prepared by a board-certified veterinary pathologist based on the percentage of multiple lung tissue sections affected. ## Confocal microscopy. To validate various findings, multilabel immunohistochemistry was performed on M. tuberculosis–infected and M. tuberculosis–infected, ME- and ME/D1MT-treated RM lungs at necropsy as described [74]. The lung sections were stained for CD68 (macrophages) and IDO-1. DAPI was used for nuclear staining. The slides were scanned using Zeiss Axio Scan Z1, and quantification was done using HALO software (Indica Labs). Kyn staining was performed on BAL cells from pretreatment and post–D1MT treatment time points, to quantify it before and after IDO inhibition. Briefly, BAL cells collected biweekly were concentrated on a slide in monolayer using Cytospin 4 centrifuge at 1,000g, 5 minutes, room temperature (Thermo Fisher Scientific). Images were captured using Zeiss LSM-800 confocal microscope, and ImageJ (Fiji) was utilized to quantify Kyn-positive cells in BAL. Antibodies used in immunofluorescence and immunohistochemistry are listed in Supplemental Table 2. ## qRT-PCR. RNA was isolated from lung (obtained at endpoint) and BAL cells (from before and after D1MT treatment time point) using Direct-zol RNA Miniprep Kit (Zymo Research, catalog R2051) and quantified with Quant iT RNA HS Assay Kit (Molecular Probes, Thermo Fisher Scientific, catalog Q32852). Subsequently, RNA samples were reverse-transcribed to cDNA by utilizing iScript Advanced cDNA Synthesis Kit (Bio-Rad, catalog 1725038). cDNA obtained in this step was employed for measuring the levels of IDO1, IDO2, IFN-γ, and IFN-β1 genes by using TaqMan Gene Expression Assays [76] designed for RMs (Applied Biosystems, Thermo Fisher Scientific) with the following assay IDs: Rh02841203_m1 (IDO1), Rh04390839_m1 (IDO2), Rh02621721_m1 (IFN-γ), Rh02621721_m1 (IFN-γ), Rh03648734_s1 (IFN-β1), and Rh02621745_g1 (GAPDH). Fold expression of these target genes was computed relative to GAPDH using ΔΔCt method (or 2-ΔΔCt). ## Flow cytometry. High-parameter flow cytometry was performed on BAL cells and PBMCs at preinfection, at pretreatment, during treatment, at posttreatment, and at necropsy as previously described [17, 34, 66, 74, 77]. The single cells prepared from lung, BAL, PBMCs, and other tissues were stained with surface and intracellular markers to study T cell phenotypes. Tissues obtained at necropsy were digested using Liberase and DNase (both Sigma-Aldrich), filtered, and subjected to RBC lysis (ACK Lysis Buffer, Gibco). The cells were then counted and used for staining for flow cytometry. The cells were first stained with extracellular/surface antibodies: CD3, CD4, CD8, CD45, CD28, and CD95 for 25 minutes at room temperature, followed by the Fixable Viability Stain 575V (BD Biosciences). The cells were then fixed and permeabilized using Fixation/Permeabilization Kit (BD Biosciences) for 30 minutes at 4°C. Subsequently, the cells were stained with intracellular antibody (Ki67) to study the T cell proliferation. Cells were then washed and acquired on a BD FACSSymphony flow cytometer. Analysis was performed using FlowJo (v10.5.3) using previously published gating strategies (Supplemental Figure 4) (30, 32–34). The details of antibodies used in flow cytometry experiments are provided (Supplemental Table 3). ## Statistics. Statistical analysis was performed using 1-way ANOVA with Tukey’s correction, 2-way ANOVA with Tukey’s multiple-comparison test, contingency χ2 (and Fisher’s exact) test, and Kruskal-Wallis test as applicable using GraphPad Prism (version 9). A P value of less than 0.05 was considered statistically significant. Data are represented as mean or mean ± SEM, as applicable. ## Study approval. All infected macaques were housed under Animal Biosafety Level 3 facilities at the SNPRC, Texas Biomedical Research Institute, where they were treated according the standards recommended by the Association for Assessment and Accreditation of Laboratory Animal Care International and the NIH Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011). The study procedures were approved by the Animal Care and Use Committee of the Texas Biomedical Research Institute. ## Author contributions SM designed the study. BS, CM, DKS, RAE, RS, GA, XA, and SRG researched. BS, CM, DKS, and XA analyzed data. VS, OG, EJD, and SHU provided veterinary medicine and pathology analysis. SM and DK provided funding. SM wrote the initial draft. SM, XA, DK, BS, CM, and DKS edited the manuscript. All authors contributed to the manuscript. ## 01/24/2023 Electronic publication ## References 1. Russell DG. Who puts the tubercle in tuberculosis?. Nat Rev Microbiol (2007) 5 39-47. DOI: 10.1038/nrmicro1538 2. Veatch AV, Kaushal D. Opening Pandora’s box: mechanisms of Mycobacterium tuberculosis resuscitation. Trends Microbiol (2018) 26 145-157. DOI: 10.1016/j.tim.2017.08.001 3. Kaufmann SH. Mycobacterium tuberculosis and the host response. J Exp Med (2005) 201 1693-1697. DOI: 10.1084/jem.20050842 4. Esaulova E. The immune landscape in tuberculosis reveals populations linked to disease and latency. Cell Host Microbe (2020) 29 165-178. PMID: 33340449 5. Akter S. Mycobacterium tuberculosis infection drives a type I IFN signature in lung lymphocytes. Cell Rep (2022) 39. DOI: 10.1016/j.celrep.2022.110983 6. Gideon HP. Multimodal profiling of lung granulomas in macaques reveals cellular correlates of tuberculosis control. Immunity (2022) 55 827-846. DOI: 10.1016/j.immuni.2022.04.004 7. McCaffrey EF. The immunoregulatory landscape of human tuberculosis granulomas. Nat Immunol (2022) 23 318-329. DOI: 10.1038/s41590-021-01121-x 8. Mellor AL, Munn DH. IDO expression by dendritic cells: tolerance and tryptophan catabolism. Nat Rev Immunol (2004) 4 762-774. DOI: 10.1038/nri1457 9. Munn DH. Inhibition of T cell proliferation by macrophage tryptophan catabolism. J Exp Med (1999) 189 1363-1372. DOI: 10.1084/jem.189.9.1363 10. Mehra S. Transcriptional reprogramming in nonhuman primate (rhesus macaque) tuberculosis granulomas. PLoS One (2010) 5. DOI: 10.1371/journal.pone.0012266 11. Mehra S. The DosR regulon modulates adaptive immunity and is essential for M. tuberculosis persistence. Am J Respir Crit Care Med (2015) 191 1185-1196. DOI: 10.1164/rccm.201408-1502OC 12. Hudock TA. Hypoxia sensing and persistence genes are expressed during the intragranulomatous survival of Mycobacterium tuberculosis. Am J Respir Cell Mol Biol (2017) 56 637-647. DOI: 10.1165/rcmb.2016-0239OC 13. Gideon HP. Variability in tuberculosis granuloma T cell responses exists, but a balance of pro- and anti-inflammatory cytokines is associated with sterilization. PLoS Pathog (2015) 11. DOI: 10.1371/journal.ppat.1004603 14. Mehra S. Granuloma correlates of protection against tuberculosis and mechanisms of immune modulation by Mycobacterium tuberculosis. J Infect Dis (2013) 207 1115-1127. DOI: 10.1093/infdis/jis778 15. Blumenthal A. M. tuberculosis induces potent activation of IDO-1, but this is not essential for the immunological control of infection. PLoS One (2012) 7. DOI: 10.1371/journal.pone.0037314 16. Gautam US. In vivo inhibition of tryptophan catabolism reorganizes the tuberculoma and augments immune-mediated control of Mycobacterium tuberculosis. Proc Natl Acad Sci U S A (2018) 115 E62-E71. PMID: 29255022 17. Singh B. Myeloid-derived suppressor cells mediate T cell dysfunction in nonhuman primate TB granulomas. mBio (2021) 12. DOI: 10.1128/mbio.03189-21 18. Isa F. Mass Spectrometric identification of urinary biomarkers of pulmonary tuberculosis. EBioMedicine (2018) 31 157-165. DOI: 10.1016/j.ebiom.2018.04.014 19. Shi W. Plasma indoleamine 2,3-dioxygenase activity as a potential biomarker for early diagnosis of multidrug-resistant tuberculosis in tuberculosis patients. Infect Drug Resist (2019) 12 1265-1276. DOI: 10.2147/IDR.S202369 20. Collins JM. Tryptophan catabolism reflects disease activity in human tuberculosis. JCI Insight (2020) 5. DOI: 10.1172/jci.insight.137131 21. Popov A. Indoleamine 2,3-dioxygenase-expressing dendritic cells form suppurative granulomas following Listeria monocytogenes infection. J Clin Invest (2006) 116 3160-3170. DOI: 10.1172/JCI28996 22. Ganesan S, Roy CR. Host cell depletion of tryptophan by IFNγ-induced Indoleamine 2,3-dioxygenase 1 (IDO1) inhibits lysosomal replication of Coxiella burnetii. PLoS Pathog (2019) 15. DOI: 10.1371/journal.ppat.1007955 23. Makala LH. Leishmania major attenuates host immunity by stimulating local indoleamine 2,3-dioxygenase expression. J Infect Dis (2011) 203 715-725. DOI: 10.1093/infdis/jiq095 24. Nelson DE. Chlamydial IFN-gamma immune evasion is linked to host infection tropism. Proc Natl Acad Sci U S A (2005) 102 10658-10663. DOI: 10.1073/pnas.0504198102 25. Zhang YJ, Rubin EJ. Feast or famine: the host-pathogen battle over amino acids. Cell Microbiol (2013) 15 1079-1087. DOI: 10.1111/cmi.12140 26. Zhang YJ. Tryptophan biosynthesis protects mycobacteria from CD4 T-cell-mediated killing. Cell (2013) 155 1296-1308. DOI: 10.1016/j.cell.2013.10.045 27. Kondrikov D. Kynurenine inhibits autophagy and promotes senescence in aged bone marrow mesenchymal stem cells through the aryl hydrocarbon receptor pathway. Exp Gerontol (2020) 130. DOI: 10.1016/j.exger.2019.110805 28. Holmgaard RB. Tumor-expressed IDO recruits and activates MDSCs in a Treg-dependent manner. Cell Rep (2015) 13 412-424. DOI: 10.1016/j.celrep.2015.08.077 29. Mehra S. The Mycobacterium tuberculosis stress response factor SigH is required for bacterial burden as well as immunopathology in primate lungs. J Infect Dis (2012) 205 1203-1213. DOI: 10.1093/infdis/jis102 30. Kaushal D. Mucosal vaccination with attenuated Mycobacterium tuberculosis induces strong central memory responses and protects against tuberculosis. Nat Commun (2015) 6. DOI: 10.1038/ncomms9533 31. Mehra S. Reactivation of latent tuberculosis in rhesus macaques by coinfection with simian immunodeficiency virus. J Med Primatol (2011) 40 233-243. DOI: 10.1111/j.1600-0684.2011.00485.x 32. Foreman TW. CD4+ T-cell-independent mechanisms suppress reactivation of latent tuberculosis in a macaque model of HIV coinfection. Proc Natl Acad Sci U S A (2016) 113 E5636-E5644. PMID: 27601645 33. Ganatra SR. Anti-retroviral therapy does not reduce tuberculosis reactivation in a tuberculosis-HIV co-infection model. J Clin Invest (2020) 130 5171-5179. DOI: 10.1172/JCI136502 34. Sharan R. Antiretroviral therapy timing impacts latent tuberculosis infection reactivation in a Mycobacterium tuberculosis/SIV coinfection model. J Clin Invest (2022) 132. DOI: 10.1172/JCI153090 35. Bucsan AN. Mechanisms of reactivation of latent tuberculosis infection due to SIV co-infection. J Clin Invest (2019) 129 5254-5260. DOI: 10.1172/JCI125810 36. Lin PL. PET CT identifies reactivation risk in cynomolgus macaques with latent M. tuberculosis. PLoS Pathog (2016) 12. DOI: 10.1371/journal.ppat.1005739 37. Wallis RS, Hafner R. Advancing host-directed therapy for tuberculosis. Nat Rev Immunol (2015) 15 255-263. DOI: 10.1038/nri3813 38. Zumla AI. New antituberculosis drugs, regimens, and adjunct therapies: needs, advances, and future prospects. Lancet Infect Dis (2014) 14 327-340. DOI: 10.1016/S1473-3099(13)70328-1 39. Zumla A. Inflammation and tuberculosis: host-directed therapies. J Intern Med (2015) 277 373-387. DOI: 10.1111/joim.12256 40. Zumla A. Advances in the development of new tuberculosis drugs and treatment regimens. Nat Rev Drug Discov (2013) 12 388-404. DOI: 10.1038/nrd4001 41. Zumla A. Host-directed therapies for tuberculous pericarditis. Int J Infect Dis (2015) 32 30-31. DOI: 10.1016/j.ijid.2014.11.017 42. Kaufmann SHE. Progress in tuberculosis vaccine development and host-directed therapies—a state of the art review. Lancet Respir Med (2014) 2 301-320. DOI: 10.1016/S2213-2600(14)70033-5 43. Beever A. L-GSH supplementation in conjunction with rifampicin augments the treatment response to Mycobacterium tuberculosis in a diabetic mouse model. Front Pharmacol (2022) 13. DOI: 10.3389/fphar.2022.879729 44. Jiang Q. Glutamine is required for M1-like polarization of macrophages in response to Mycobacterium tuberculosis infection. mBio (2022) 13. DOI: 10.1128/mbio.01274-22 45. Dutta NK. Statin adjunctive therapy shortens the duration of TB treatment in mice. J Antimicrob Chemother (2016) 71 1570-1577. DOI: 10.1093/jac/dkw014 46. Maitra A. Repurposing drugs for treatment of tuberculosis: a role for non-steroidal anti-inflammatory drugs. Br Med Bull (2016) 118 138-148. DOI: 10.1093/bmb/ldw019 47. Maiga M. Successful shortening of tuberculosis treatment using adjuvant host-directed therapy with FDA-approved phosphodiesterase inhibitors in the mouse model. PLoS One (2012) 7. DOI: 10.1371/journal.pone.0030749 48. Vilcheze C. Enhanced respiration prevents drug tolerance and drug resistance in. Proc Natl Acad Sci U S A (2017) 114 4495-4500. DOI: 10.1073/pnas.1704376114 49. Napier RJ. Low doses of imatinib induce myelopoiesis and enhance host anti-microbial immunity. PLoS Pathog (2015) 11. DOI: 10.1371/journal.ppat.1004770 50. Lee YJ. The effect of metformin on culture conversion in tuberculosis patients with diabetes mellitus. Korean J Intern Med (2018) 33 933-940. DOI: 10.3904/kjim.2017.249 51. Liao JK. Effects of statins on 3-hydroxy-3-methylglutaryl coenzyme a reductase inhibition beyond low-density lipoprotein cholesterol. Am J Cardiol (2005) 96 24F-33F. PMID: 16126020 52. Tian X. The upregulation of LAG-3 on T cells defines a subpopulation with functional exhaustion and correlates with disease progression in HIV-infected subjects. J Immunol (2015) 194 3873-3882. DOI: 10.4049/jimmunol.1402176 53. Russell DG. The evolutionary pressures that have molded Mycobacterium tuberculosis into an infectious adjuvant. Curr Opin Microbiol (2013) 16 78-84. DOI: 10.1016/j.mib.2012.11.007 54. Mahon RN. Mycobacterium tuberculosis ManLAM inhibits T-cell-receptor signaling by interference with ZAP-70, Lck and LAT phosphorylation. Cell Immunol (2012) 275 98-105. PMID: 22507872 55. Mahon RN. Mycobacterium tuberculosis cell wall glycolipids directly inhibit CD4+ T-cell activation by interfering with proximal T-cell-receptor signaling. Infect Immun (2009) 77 4574-4583. DOI: 10.1128/IAI.00222-09 56. Lancioni CL. Infect Immun (2011) 79 663-673. DOI: 10.1128/IAI.00806-10 57. Deretic V, Fratti RA. Mycobacterium tuberculosis phagosome. Mol Microbiol (1999) 31 1603-1609. DOI: 10.1046/j.1365-2958.1999.01279.x 58. Briken V, Miller JL. Living on the edge: inhibition of host cell apoptosis by Mycobacterium tuberculosis. Future Microbiol (2008) 3 415-422. DOI: 10.2217/17460913.3.4.415 59. Pai RK. Inhibition of IFN-gamma-induced class II transactivator expression by a 19-kDa lipoprotein from Mycobacterium tuberculosis: a potential mechanism for immune evasion. J Immunol (2003) 171 175-184. DOI: 10.4049/jimmunol.171.1.175 60. Dutta NK. The stress-response factor SigH modulates the interaction between Mycobacterium tuberculosis and host phagocytes. PLoS One (2012) 7. DOI: 10.1371/journal.pone.0028958 61. Taylor MW, Feng GS. Relationship between interferon-gamma, indoleamine 2,3-dioxygenase, and tryptophan catabolism. FASEB J (1991) 5 2516-2522. DOI: 10.1096/fasebj.5.11.1907934 62. Thomas SM. IFN-gamma-mediated antimicrobial response. Indoleamine 2,3-dioxygenase-deficient mutant host cells no longer inhibit intracellular chlamydia spp. or toxoplasma growth. J Immunol (1993) 150 5529-5534. DOI: 10.4049/jimmunol.150.12.5529 63. Munn DH. Prevention of allogeneic fetal rejection by tryptophan catabolism. Science (1998) 281 1191-1193. DOI: 10.1126/science.281.5380.1191 64. Tattevin P. Enhanced indoleamine 2,3-dioxygenase activity in patients with severe sepsis and septic shock. J Infect Dis (2010) 201 956-966. DOI: 10.1086/650996 65. Almeida AS. Tuberculosis is associated with a down-modulatory lung immune response that impairs Th1-type immunity. J Immunol (2009) 183 718-731. DOI: 10.4049/jimmunol.0801212 66. Sharan R. Isoniazid and rifapentine treatment effectively reduces persistent M. tuberculosis infection in macaque lungs. J Clin Invest (2022) 132. DOI: 10.1172/JCI161564 67. Bucsan AN. Response to hypoxia and the ensuing dysregulation of inflammation impacts Mycobacterium tuberculosis pathogenicity. Am J Respir Crit Care Med (2022) 206 94-104. DOI: 10.1164/rccm.202112-2747OC 68. Lin PL. Metronidazole prevents reactivation of latent Mycobacterium tuberculosis infection in macaques. Proc Natl Acad Sci U S A (2012) 109 14188-14193. DOI: 10.1073/pnas.1121497109 69. Darrah PA. Prevention of tuberculosis in macaques after intravenous BCG immunization. Nature (2020) 577 95-102. DOI: 10.1038/s41586-019-1817-8 70. Dijkman K. Prevention of tuberculosis infection and disease by local BCG in repeatedly exposed rhesus macaques. Nat Med (2019) 25 255-262. DOI: 10.1038/s41591-018-0319-9 71. Stammes MA. Recommendations for standardizing thorax PET-CT in non-human primates by recent experience from macaque studies. Animals (Basel) (2021) 11. PMID: 33467761 72. Mattila JT. Positron emission tomography imaging of macaques with tuberculosis identifies temporal changes in granuloma glucose metabolism and integrin α4β1-expressing immune cells. J Immunol (2017) 199 806-815. DOI: 10.4049/jimmunol.1700231 73. Sakai S. Functional inactivation of pulmonary MAIT cells following 5-OP-RU treatment of non-human primates. Mucosal Immunol (2021) 14 1055-1066. DOI: 10.1038/s41385-021-00425-3 74. Singh DK. Responses to acute infection with SARS-CoV-2 in the lungs of rhesus macaques, baboons and marmosets. Nat Microbiol (2021) 6 73-86. DOI: 10.1038/s41564-020-00841-4 75. Bucsan AN. Mechanisms of reactivation of latent tuberculosis infection due to SIV coinfection. J Clin Invest (2019) 129 5254-5260. DOI: 10.1172/JCI125810 76. Mayer-Barber KD. Host-directed therapy of tuberculosis based on interleukin-1 and type I interferon crosstalk. Nature (2014) 511 99-103. DOI: 10.1038/nature13489 77. Sharan R. Characterizing early T cell responses in nonhuman primate model of tuberculosis. Front Immunol (2021) 12. DOI: 10.3389/fimmu.2021.706723
--- title: The kidney drug transporter OAT1 regulates gut microbiome–dependent host metabolism authors: - Jeffry C. Granados - Vladimir Ermakov - Koustav Maity - David R. Vera - Geoffrey Chang - Sanjay K. Nigam journal: JCI Insight year: 2023 pmcid: PMC9977316 doi: 10.1172/jci.insight.160437 license: CC BY 4.0 --- # The kidney drug transporter OAT1 regulates gut microbiome–dependent host metabolism ## Abstract Organic anion transporter 1 (OAT1/SLC22A6, NKT) is a multispecific drug transporter in the kidney with numerous substrates, including pharmaceuticals, endogenous metabolites, natural products, and uremic toxins. Here, we show that OAT1 regulates levels of gut microbiome–derived metabolites. We depleted the gut microbiome of Oat1-KO and WT mice and performed metabolomics to analyze the effects of genotype (KO versus WT) and microbiome depletion. OAT1 is an in vivo intermediary between the host and the microbes, with 40 of the 162 metabolites dependent on the gut microbiome also impacted by loss of Oat1. Chemoinformatic analysis revealed that the altered metabolites (e.g., indoxyl sulfate, p-cresol sulfate, deoxycholate) had more ring structures and sulfate groups. This indicates a pathway from gut microbes to liver phase II metabolism, to renal OAT1–mediated transport. The idea that multiple gut-derived metabolites directly interact with OAT1 was confirmed by in vitro transport and magnetic bead binding assays. We show that gut microbiome–derived metabolites dependent on OAT1 are impacted in a chronic kidney disease (CKD) model and human drug-metabolite interactions. Consistent with the Remote Sensing and Signaling Theory, our results support the view that drug transporters (e.g., OAT1, OAT3, OATP1B1, OATP1B3, MRP2, MRP4, ABCG2) play a central role in regulating gut microbe–dependent metabolism, as well as interorganismal communication between the host and microbiome. ## Introduction Organic anion transporter 1 (OAT1/SLC22A6, originally described as NKT) is a multispecific drug transporter localized to the basolateral membrane of the kidney proximal tubule. OAT1 is involved in the uptake of multiple classes of drugs (e.g., antibiotics, antivirals, NSAIDs, diuretics), endogenous metabolites, toxins, antioxidants, and natural products from the blood into the proximal tubule cell, where they can then be excreted into the urine by apical efflux transporters, reintroduced to the bloodstream, or metabolized by the cell (1–10). While OAT1 favors the transport of organic anions, it can also handle several structurally different small molecules, including some cations and zwitterions [11]. To date, the overwhelming majority of research interest in OAT1 has been related to its role in the clearance of drugs, as the US Food and Drug Administration (FDA) and other global regulatory agencies have recommended that novel drug entities be tested for OAT1 interaction due to potential drug-drug interactions (DDIs) occurring at the site of the transporter [12, 13]. Despite the pharmaceutical relevance of this transporter, recent studies have highlighted a further role of OAT1 and other drug transporters in endogenous metabolism in the context of several observations (14–17). Many of the metabolites that have received clinical attention— such as 4-ethylphenyl sulfate, p-cresol sulfate, and indoxyl sulfate — are organic anions and well-known OAT1 substrates and have been associated with the gut microbiome. The gut microbiome plays an important role in the endogenous host metabolism by producing several important metabolites. While these metabolites are generated within the host, they are often the products of complex interactions between the host and the commensal microbes residing in the gut. Gut microbiome–derived metabolites include short-chain fatty acids, secondary bile acids, aromatic amino acid derivatives, polyamines, and several others (18–20). The full repertoire of gut microbiome–derived metabolites and how they are generated remains incomplete, but it is clear that these metabolites play important signaling roles in both healthy and disease states [21]. For example, gut bacterial metabolites have been shown to influence the immune system [22]. Furthermore, inflammatory bowel disease (IBD), cardiovascular disease, and chronic kidney disease (CKD) are associated with microbial dysbiosis, which can lead to abnormal levels of serum metabolites (23–27). CKD, in particular, is associated with increases in circulating gut-derived uremic toxins due to diminished glomerular and tubular renal function (28–30). OAT1 is known to be critical for the transport of many of these metabolites into the proximal tubule. These gut microbiome–derived metabolites include indoxyl sulfate, p-cresol sulfate, hippurate, and other metabolites and signaling molecules that are organic anions, suggesting that there may be an important role for OAT1 in the mediation of host-microbiome interaction [3, 28, 31]. Considering the wide array of substrates transported by OAT1, the competition between drugs, toxins, and gut-derived metabolites at the site of the transporter could also lead to drug-metabolite interactions (DMI) —especially in patients with CKD, who are likely to be taking multiple drugs to treat symptoms associated with comorbidities. The Remote Sensing and Signaling Theory (RSST) addresses how OAT1, along with other solute carrier (SLC) and ATP-binding cassette (ABC) “drug” transporters, plays a major role in homeostasis of many small molecules, including rate-limiting metabolites, signaling molecules, antioxidants, gut microbe–derived products, vitamins, and cofactors [32]. These other SLC and ABC transporters include OAT3 (SLC22A8), organic anion transporting polypeptide 1B1 (OATP1B1, SLCO1B1), OATP1B3 (SLCO1B3), multidrug resistance protein 2 (MRP2, ABCC2), MRP4 (ABCC4), and ABC G subfamily 2 (ABCG2, BCRP). The RSST proposes a complex adaptive system of drug transporters, drug metabolizing enzymes, nuclear receptors, and kinases that regulates endogenous metabolism through transport, metabolism, and conjugation of small molecules with signaling roles between remote organs (e.g., gut, liver, kidney, brain) and multiorganismal systems (e.g., gut microbe–host, mother-fetus) [33, 34]. Drug transporters in the SLC22, SLC organic (SLCO), and ATP-binding cassette C subfamily (ABCC) families and their multispecificity (ability to handle structurally different organic anions) are central to the RSST and have been identified as important hubs in a cross-tissue coexpression network, suggesting a principal role in endogenous metabolism at multiple scales (organism to organ to organelle) [14, 32, 35]. The RSST has mainly been explored through the lens of interorgan communication (e.g., organ crosstalk) — but an important, understudied aspect is interorganismal communication between the host and the bacterial species in the gut, which is possibly mediated by the role of kidney OAT1 and other drug transporters in modulating gut microbiome host interactions [36]. While OAT1 and OAT3 in the kidney are hypothesized to be central to interorganismal communication via gut-derived metabolites, the hepatic transporters, OATP1B1 and OATP1B3, along with various ABC transporters, such as ABCC2, ABCC4, and ABCG2, are also thought to contribute to the transport of gut microbiome derived compounds [37, 38]. Furthermore, drug-metabolizing enzymes are central to the metabolism and conjugation of these compounds along the gut-liver-kidney axis (39–41). Nevertheless, much of the data supporting these notions are from in vitro rather than in vivo studies. In this work, we focused on the in vivo role of a single multispecific kidney “drug” transporter, OAT1, in the regulation of gut-derived metabolites and the metabolic pathways involving these molecules. We first established the efficacy of the Oat1-KO mouse model by demonstrating in vivo alterations in the handling of a well-studied OAT1 substrate, as evidenced by changes in levels in the blood and urine. We then focused on depleting the gut microbiome of these mice and their WT counterparts. Gnotobiotic mice have frequently been used as a model to understand the impact of gut microbes, but while they are generally healthy, they are difficult to compare with WT mice due to differences in their genetic backgrounds, as well as issues related to development, immune defects, and energy metabolism [42, 43]. This is a key concern in the context of OAT1, since OAT1 is an α-ketoglutarate antiporter and is, thus, directly linked to aerobic metabolism, upon which the kidney proximal tubule almost exclusively depends [31, 44]. Therefore, like many in the field, we chose to employ antibiotic treatment to deplete the gut microbes [45]. We then assessed the impact of loss of the Oat1 gene (Oat1-KO versus WT) and microbiome depletion on biochemical pathways, and we applied chemoinformatics approaches to characterize the altered metabolites. To support our in vivo findings, we performed in vitro transport assays and employed a magnetic bead binding assay to evaluate mechanistic relationships between gut-derived metabolites and OAT1. Furthermore, we established clinical and disease relevance of our results by showing that the gut microbiome–derived metabolites that are OAT1 dependent are significantly affected in a clear example of human DMI and in a rodent CKD model. Our results indicate that OAT1 plays a surprisingly important role in the handling of a number of gut-derived metabolites and, consistent with the RSST, mediates interorganismal communication between the host and gut microbes, in large part by regulating the circulating levels of these compounds (Figure 1). ## Clearance of OAT1-interacting compound altered in vivo in KO mice. We first characterized our Oat1-KO mice and their WT counterparts by measuring the levels of Tc-99m mercaptoacetyl-triglycine (MAG3) in the urine (bladder) and the blood. Tc-99m MAG3 is a probe compound used in the assessment of renal function that is nearly entirely eliminated by tubular secretion. Previous studies have demonstrated that Tc-99m MAG3 is a rat OAT1 (rOAT1) substrate in vitro and that its uptake is inhibited by classic OAT1 inhibitors, such as PAH and probenecid [46]. Clinically, results in humans have shown that MAG3 levels in the blood were elevated following treatment with PAH and probenecid [47]. These observations were supported by assays using HEK293 cells expressing human OAT1, which showed that the protein is involved in the uptake of Tc-99m MAG3 [47]. We evaluated whether the Oat1-KO and WT mice had different clearance patterns with this well-established OAT1 substrate (Figure 2A). Tc-99m MAG3 was administered to the mice via tail-vein injection, and its levels were monitored over the course of 30 minutes using a γ camera. We mainly focused on the bladder, as Tc-99m MAG3 quickly passes through the kidneys into the urine. We found that the WT bladders reached their maximal levels of Tc-99m MAG3 more quickly than the Oat1-KO mice following direct injection of the probe compound (Figure 2B). These results were further supported by postmortem γ counts scaled to weight, which showed that, for 4 of 5 pairs, the bladder levels of Tc-99m MAG3 were higher in the WT mice (Figure 2C) and the blood levels were higher in the KO mice (Figure 2D). Given that OAT1 is considered the rate-limiting step for excretion of many organic anions into the urine, our results support the usefulness of the KO mice as in vivo models for analyzing OAT1-related function. ## Gut microbiome was depleted in Oat1-KO and WT mice. Previous in vivo and in vitro experiments have shown that OAT1 has several putative gut-derived substrates, such as indoxyl sulfate, p-cresol sulfate, and hippurate [17, 31, 48]. While these metabolites are useful in understanding a potential role for OAT1 in regulating circulating levels of gut microbe–derived metabolites, the results were collated from multiple past experiments performed under a variety of conditions and not designed to evaluate the in vivo contribution of gut microbiome and renal OAT1 to host systemic metabolism. In this work, we depleted the gut microbiome in both WT and KO mice through the administration of an antibiotic cocktail (ampicillin, vancomycin, neomycin, and metronidazole [AVNM]). The AVNM antibiotic cocktail has been established as an effective method of depleting the gut microbiome and — in contrast to germ-free mice, which can develop metabolic problems that can lead to obesity [43] — seemed less likely to confound the essential role of OAT1 in kidney aerobic metabolism [31, 44] and the tendency to hepatic steatosis seen in approximately 24 month old Oat1-KO mice [16]. The AVNM cocktail was administered through a vehicle control in the drinking water for 4 weeks [45]. Following the administration of the cocktail, depletion of the gut microbiome was confirmed via metagenomic analysis of the feces, which showed a significant decrease in the number of operational taxonomic units (OTUs) for AVNM-treated mouse groups (Figure 3A). The global metabolic profiles of all animals were separable by linear discriminant analysis (Figure 3B), and several well-established gut-derived metabolites were significantly decreased in the serum of AVNM-treated animals, regardless of genetic background (Figure 3, C–F). Furthermore, quantitative PCR (qPCR) using 16S primers for Eubacteria also showed a significant decrease in gut microbes (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.160437DS1). ## Loss of Oat1 and microbiome depletion significantly affect the levels of over 200 metabolites. Since OAT1 is localized to the basolateral (blood-facing) side of the proximal tubule, its absence directly affects the circulating levels of metabolites in the serum. To identify these compounds, we performed a 2-way ANOVA to determine the individual impact of genotype (Oat1-KO versus WT), where a metabolite was considered altered if it had an FDR-corrected P value below 0.05. Although metabolomics of the Oat1-KO has been previously performed, this is the first time to our knowledge that nearly 1,000 compounds were measured, and here — apart from including the effects of microbiome depletion — we also focus on all altered metabolites, not just elevated metabolites [3, 16, 17, 31, 49]. Global metabolic profiling detected a total of 964 metabolites in the volume-adjusted serum samples collected from these mice. Based on the Metabolon grouping of these metabolites, this analysis covered 10 biochemical superpathways (e.g., Lipid, Amino Acid, Xenobiotic) and 109 biochemical subpathways (e.g., Primary Bile Acid Metabolism, Tryptophan Metabolism, Benzoate Metabolism), with each metabolite belonging to 1 superpathway and 1 subpathway. We identified 103 significantly altered metabolites due to the absence of OAT1 (Supplemental Table 1), including several metabolites that are known to directly interact with OAT1, such as pyridoxate, indoxyl sulfate, and p-cresol sulfate [50]. We then performed an enrichment analysis of these metabolites and found that over 20 subpathways were altered, with enrichment values of 1 or greater, indicating an outsized effect. Benzoate Metabolism and Fatty Acid Metabolism (Acyl Glycine) were among the most significantly affected pathways (Figure 4A). Having established depletion of the gut microbiome with AVNM treatment (by decreased OTUs and qPCR), we then analyzed how this impacted the serum metabolome of the microbiome-depleted mice. We found that 162 metabolites were significantly altered in the serum of the microbiome-depleted mice (Supplemental Table 2). Thus, microbe depletion has a direct impact on over 100 metabolites, including several metabolites that have previously been established as gut derived, like cinnamoylglycine, indolepropionate, and others [51]. Subpathway analysis revealed that Benzoate Metabolism, Phospholipid Metabolism, and Tyrosine Metabolism were among the most altered subpathways, and over 20 subpathways had enrichment values of 1 or greater (Figure 4B). Given that there is no current consensus on the full range of commensal gut bacteria–derived metabolites that enter the host circulation, we interpreted these metabolites to be products of gut microbiome–associated metabolism with the understanding that, for some metabolites, the levels in the serum may be due to complex interactions between bacterial species themselves and, upon entry into the host circulation, indirect effects on host metabolic pathways that likely include complex feedback and/or feedforward loops, with secondary bile acid metabolism being a good example [52]. ## Deoxycholate levels depends on an interaction between OAT1 and the gut microbiome. We then explored the statistical interaction between the 2 independent variables: loss of Oat1 and microbiome depletion. Only 3 metabolites (2-amino–p-cresol sulfate, deoxycholate, docosahexaenoylcarnitine [C22:6]) were impacted by the interaction between the variables, which implies that genotype and treatment, together, influence few metabolites compared with the individual effects of genotype versus treatment (Figure 4C). C22:6 and 2-amino–p-cresol sulfate are poorly characterized, but deoxycholate has a well-established signaling role, suggesting an important role for OAT1, together with gut microbes, in regulation of this important bile acid signaling molecule (53–55). ## Pathways and chemoinformatics analyses of the 40 metabolites affected by both loss of Oat1 and microbiome depletion. We then aimed to identify the overlap between the 103 metabolites altered by loss of Oat1 and the 162 metabolites affected by microbiome depletion. We found 40 metabolites (Figure 5A) that satisfied both criteria and found that some subpathways, particularly Benzoate Metabolism with 11 compounds and Food Component/Plant with 5 compounds, were markedly affected in the overlap (Figure 5B). These 40 compounds could be separated into 4 distinct groups: Group 1 (elevated by loss of Oat1 and elevated by microbiome depletion); Group 2 (elevated by loss of Oat1 and decreased by microbiome depletion); Group 3 (decreased by loss of Oat1 and elevated by microbiome depletion); and Group 4 (decreased by loss of Oat1 and decreased by microbiome depletion) (Figure 5C). The metabolites we were most interested in were those that were in Group 2 (elevated in the Oat1-KO mice and decreased due to microbiome depletion), as these are likely OAT1 substrates that are generated by the gut microbiome. Of the 40 metabolites, 22 fell into this group, including indoxyl sulfate and p-cresol sulfate. We were also interested in the 9 metabolites in Group 1 (increased in the Oat1-KO mice and increased due to microbiome depletion), as microbiome depletion can also lead to increases in specific metabolites by reducing the species that metabolize these compounds. Finally, the last 2 groups were more difficult to interpret from the OAT1 perspective, as there is no clear renal physiological mechanism for their decreases in circulation; nonetheless, there were 7 metabolites in Group 3 (decreased in the Oat1-KO mice and decreased by microbiome depletion) and 5 metabolites in Group 4 (decreased in the Oat1-KO mice and increased by microbiome depletion). We then aimed to structurally characterize the 31 metabolites with known chemical structures. Chemoinformatics analyses can shed light on sets of molecular properties that help define particular groups of metabolites altered by a biological experiment. In this case, for example, we were most interested in Group 2 (the metabolites that were both elevated due to loss of OAT1 and decreased after microbiome depletion), as these were not only the largest group, but also most likely to be gut microbe–derived organic anion metabolites transported by OAT1 in vivo. This could yield a kind of “signature” of metabolites that originate in the gut microbiome and then follow the gut-kidney or gut-liver-kidney axes to OAT1 in the renal proximal tubule cells. To this end, we first calculated molecular properties for the 783 compounds with valid chemical structures. We performed linear discriminant analysis of the 31 compounds (Group 1, 5 metabolites with structures; Group 2, 17 metabolites with structures; Group 3, 3 metabolites with structures; Group 4, 6 metabolites with structures) that had available chemical structures and observed clear separation between the groups (Figure 5D). We then analyzed the weights of the top 2 linear discriminant analysis axes, which together explained over $95\%$ of the variance. Among the most influential variables were number of sulfate groups and number of aromatic bonds (Figure 5, E and F). Again, we were most interested in the 17 compounds with chemical structures that were elevated by loss of Oat1 and decreased by microbiome depletion (Group 2), and these compounds tended to have a higher number of aromatic bonds. The gut microbiome is known to handle a number of aromatic compounds, such as tryptophan and tyrosine derivatives. With respect to sulfation, 8 of the 17 metabolites in Group 2 featured a sulfate group. This was especially interesting because only 1 sulfated metabolite was present in the other 3 groups combined (Figure 5F). Nevertheless, this is consistent with the notion that certain diet-derived compounds are metabolized by the gut microbiome before being “tagged” via sulfation by the host for excretion, primarily through the urine [56, 57]. ## Gut-derived metabolites interact with human OAT1 in cell-based transport assays. While the Oat1-KO mouse model is critical for establishing potential in vivo OAT1 substrates, the complex physiology in vivo could lead to alterations caused by a factor other than loss of OAT1 function. To evaluate a mechanistic interaction between metabolites and OAT1, we performed in vitro cell-based transport assays. Competitive inhibition assays were carried out for deoxycholate, indolepropionate, 4-hydroxycinnamate, 2-hydroxyphenylacetic acid, and 5-hydroxyindoleacetate, which are all thought to be gut-derived metabolites. In the statistical analysis of the serum metabolome, deoxycholate was affected by genotype-treatment interaction, 2-hydroxyphenylacetic acid was affected by genotype, and indolepropionate was affected by treatment. Although 4-hydroxycinnamate and 5-hydroxyindoleacetate were not impacted by either variable, for experimental evaluation, they were included because they are derived from cinnamate and indole, respectively. Each metabolite showed comparable inhibition when OAT1 was treated with probenecid, the prototypical inhibitor of OAT1 activity. The relatively low IC50 values (<115 μM) suggest that these metabolites interact with OAT1 with high affinity (Figure 6). ## A magnetic bead binding assay shows direct physical OAT1 interaction with gut-derived metabolites. To further evaluate our results, we also employed a potentially novel magnetic bead binding assay using 6-carboxyfluorescein to analyze 20 gut-derived metabolites, a number of which were measured in the serum metabolomics (Figure 7A). The strength of this is that the assay requires relatively low amounts of OAT1 protein compared with those used for other methods (e.g., fluorescence polarization technique). Using this method, we surveyed metabolites mainly known to derive from tryptophan and tyrosine, some of which were measured in our in vivo experiments and have been evaluated in transport assays (Table 1). Overall, we found that 15 of the metabolites resulted in a significant shift, further supporting a direct interaction between the gut microbe–derived metabolites and OAT1(Figure 7B). Taken together with the cell-based in vitro transport assays that also support a direct interaction of the gut-derived metabolites with OAT1, as well as the fact that many of the interacting metabolites were from Group 2 (elevated in Oat1-KO mice and decreased by microbiome depletion), the data support the in vivo involvement of OAT1 in the regulation of this group of gut microbe–derived metabolites. ## The in vivo OAT1-dependent, gut microbe–derived metabolites overlap with those impacted by a CKD model (5/6 nephrectomy). We then aimed to contextualize our findings by comparing our results to metabolomics data previously generated by our group in a rodent $\frac{5}{6}$ nephrectomy model of CKD [28]. This model is thought to capture aspects of progressive CKD, and renal capacity is dramatically reduced over time [28]. In these experiments, plasma was collected from animals who had undergone a $\frac{5}{6}$ nephrectomy and their healthy controls, and the relative levels of hundreds of metabolites were measured. In the comparison between the nephrectomized and healthy animals, many of the elevated metabolites have been shown to include numerous uremic solutes or uremic toxins [28]. However, the nature of their diminished clearance remains unclear, since this model of renal insufficiency impacts both tubular and glomerular function. By comparing the metabolites elevated in the $\frac{5}{6}$ nephrectomy with the 40 metabolites that are gut derived and altered in the Oat1-KO mice, we were able to identify metabolites that are likely impacted by diminished proximal tubule function, as that is where OAT1 is primarily expressed. Thus, 7 metabolites (indoxyl sulfate, p-cresol sulfate, phenylacetylglycine, 4-ethylphenyl sulfate, 3-methylhistidine, N-acetylserine, 2-isopropylmalate) are likely uremic solutes or uremic toxins transported by OAT1 and produced by the gut microbiome (Table 2). ## Gut-derived metabolites transported by OAT1 are involved in human DMI. Having established that 40 metabolites are likely OAT1 mediated and produced by the gut microbiome in a mouse model, we then aimed to understand the clinical relevance of these results. A recent metabolomics study by our group analyzed the plasma and urine of healthy volunteers before and after probenecid treatment and identified dozens of unique short-term DMI [1]. While that study did not concern itself with gut microbe–derived metabolites, we were able to reanalyze that data in the context of the new data in this study. Thus, we performed an overlap of the metabolites implicated in the present Oat1-KO microbe-depleted mouse study with those significantly elevated in the plasma and significantly decreased in the urine of humans treated with probenecid (Figure 8, Supplemental Table 3, and ref. 1). Over a quarter of the OAT1-transported gut-derived metabolites from the current mouse study (11 of 40) were significantly elevated in the plasma of the probenecid treated humans, including indoxyl sulfate, p-cresol sulfate, and other compounds. When we compared the 40 metabolites with those significantly decreased in the urine of probenecid treated humans, we found that half the metabolites (20 of 40) were present in both lists. Interestingly, 4-ethylphenyl sulfate, a compound associated with autism, was present in this overlap, along with others [58]. When all 3 lists were overlapped, 8 metabolites were present, with most being sulfated organic anions (Figure 8). These gut microbiome–derived metabolites are dependent on OAT1 and are also involved in DMI. ## Discussion The loss of Oat1 and the depletion of the gut microbiome (Figure 3), separately and together, have major effects on systemic metabolism (Figure 4 and Supplemental Tables 1 and 2). The genetic KO of Oat1 primarily leads to elevated metabolites, presumably because they are no longer able to enter the proximal tubule and must remain in the blood, whereas the depletion of the gut microbiome mainly leads to lower circulating levels of gut-derived metabolites by eliminating the species that synthesize or modify the compounds. Both conditions together — KO of Oat1 and gut microbe depletion — provide perhaps the deepest glimpse to date of how the renal organic anion transport system works together with the gut microbiome to regulate systemic levels of many well-known metabolites and signaling molecules (Figure 1). Most impressive were the effects of gut microbiome depletion on metabolites elevated due to loss of Oat1 (compared with the WT with a normal microbiome). Many of these metabolites elevated in the Oat1-KO mice were significantly, and sometimes markedly, decreased after gut microbiome depletion (Figure 5). Based on this in vivo data, and the evidence presented here for a direct in vitro interaction between a number of these metabolites and OAT1 in transport and binding assays (Figures 6 and 7), it is highly likely that the largest fraction of these metabolites were derived from the gut microbes and are regulated by OAT1, which is important, given the important roles of these metabolites play in the immune response and other key physiological systems [22]. That said, it should be noted that there were also instances of metabolites significantly decreased by loss of Oat1, as well as metabolites significantly increased by microbiome depletion. The interpretation of these changes is less clear but may be due to indirect effects such as elevation in the KO of another metabolite transported by OAT1 that inhibits the synthesis of the decreased metabolite. Overall, we identified 40 metabolites (31 with chemical structures) (Supplemental Table 3) that were significantly impacted by both loss of Oat1 and gut microbiome depletion, with 22 of those compounds being elevated due to KO and decreased due to antibiotic treatment. These included derivatives of tryptophan and tyrosine. Chemoinformatics analyses of this group of metabolites (e.g., elevated in the Oat1-KO mice and decreased after gut microbiome depletion) revealed that these metabolites tended to have aromatic rings and more sulfate groups, which is generally consistent with known molecular properties of OAT1 substrates (Figure 5) [59, 60]. The high fraction of sulfated metabolites is particularly interesting and likely indicative of the interaction of gut-derived metabolites with sulfotransferases in the liver before transport into the kidney proximal tubule by OAT1 — examples, similar to that of indoxyl sulfate, of the conjunction of remote interorganismal communication and organ crosstalk [61]. Indeed, 13 of the 22 putative gut-derived OAT1 substrates in Group 2 are sulfated compounds. However, since other OATs, such as OAT3 [62], have a strong preference for steroid sulfates, it may be the context in which the sulfated compounds are presented that determines OAT1 interaction. Our data suggest that a 1- or 2-ringed structure with a sulfate may be preferred by OAT1. The strongest candidates for in vivo OAT1-transported gut microbiome–derived metabolites would seem to be those that are (a) elevated in the plasma of Oat1-KO mouse; (b) decreased by gut microbiome depletion; and (c) shown to directly interact with the transporter. Thus, to further analyze the in vivo metabolomics results, a number of the identified metabolites (and others that have been suggested to be gut derived) were tested in vitro for interaction with human OAT1 overexpressed in cells. These compounds displayed IC50 values in transport assays that indicate a strong interaction with the transporter. Additional support was gained from a magnetic bead binding assay demonstrating that a fluorescent prototypical OAT1 substrate was displaced by a number of gut-derived metabolites (Table 1). While the transport assays are traditionally used to determine interaction, the magnetic bead binding assay provides further context for the nature of the interaction and allows for more rapid screening of small molecules. Our results are also highly relevant to disease and clinical settings. We found that many of the metabolites implicated in our study were elevated in $\frac{5}{6}$ nephrectomy rodent models of CKD, indicating that these gut-derived and OAT1-mediated metabolites can be altered in the setting of diminished renal function [28]. Among the metabolites present in both studies were indoxyl sulfate, p-cresol sulfate, and 4-ethylphenyl sulfate, further supporting the view that OAT1 and the gut microbiome are jointly involved in the generation/handling of uremic toxins. We also addressed the important clinical issue of DMI. A previous study by our group analyzed the DMI caused by the drug probenecid [1]. While that clinical study did not focus on the gut microbiome, given the results from the present mouse study, we investigated whether DMI with an OAT-inhibiting drug (probenecid) had a major impact on the disposition of gut-derived metabolites. Indeed, 8 metabolites were elevated in the plasma of probenecid-treated humans, decreased in the urine of probenecid-treated humans, and altered by both loss of Oat1 and microbiome depletion (Figure 8). Furthermore, 20 of the 40 metabolites were decreased in the urine, while 11 of the 40 were elevated in the plasma. OAT1 is a major transporter of antibiotics, antivirals, NSAIDs, diuretics, and other common drugs [63]. Our analysis suggests that gut microbiome–dependent DMI at the level of OAT1 could be quite widespread [64]. This requires further study. Taken together, our results indicate that OAT1 is a crucial intermediary between the host and the microbes, with 40 metabolites of the 162 metabolites decreased by gut microbiome depletion being presumably influenced by OAT1. This suggests that as much as $25\%$ of microbiome-influenced metabolism may be modulated by OAT1. However, we must also note that the metabolomics platform we used is biased toward compounds that have already been identified and are likely relevant in clinical or research settings. Consistent with previous work, we too observed dozens of metabolites decreased by microbe depletion that were the products of hepatic metabolism [20, 31]. The communication between the host and the gut microbes is complex, but it is clear that these 2 entities have coevolved over time to develop a symbiotic relationship from the perspective of metabolism. This is evidenced by the gut-derived metabolites, which cannot be produced by the host alone, that have important signaling roles, such as nuclear receptor and G-protein–coupled receptor activation [65, 66]. Among the implicated metabolites, deoxycholate, indoxyl sulfate, indolepropionate, and others have been shown to have important signaling roles (67–70). While the signaling roles of gut-derived metabolites with respect to target proteins is an important field of research, our results indicate that much more attention needs to be paid to the proteins that regulate their levels in biofluids, as well as in tissues and along organ axes (e.g., gut-liver-kidney, gut-brain). These proteins, which include OAT1, are important avenues by which the host and its gut microbes interact, as they control the bioavailability of signaling molecules. SLC22 family members (e.g., OATs, organic cation transporters [OCTs], organic cation and zwitterion transporters [OCTNs]) and other multispecific drug transport proteins (e.g., ABCG2, ABCC2) are hubs in a recently proposed Remote Sensing and Signaling Network [32]. The Remote Sensing and Signaling Theory (RSST) emphasizes the importance of the adaptive network of multispecific transporters, enzymes, and nuclear receptors — working together with oligo-specific and monospecific proteins — in the optimization of the levels of numerous metabolites in cells, tissues, organs, and bodily fluids, such as blood, cerebrospinal fluid, and urine (10, 14, 32–34, 41). These proteins have been extensively studied from the pharmaceutical perspective, but their ability to handle structurally diverse molecules is perhaps most important from the perspective of endogenous and gut-derived metabolites. The theory mainly serves as a framework to describe interorgan communication, such as in the gut-liver-kidney axis (where most drug transporters and drug metabolizing enzymes are highly expressed), to maintain and reestablish homeostasis of key metabolites, signaling molecules, antioxidants, and other small molecules with “high informational content” [10]. While we focused on the individual role of OAT1 in this work, it is likely that the combined network of transporters and enzymes, including cytochrome P450s (CYPs), sulfotransferases (SULTs), uridine 5’-diphospho-glucuronosyltransferases (UGTs), ABCCs, and SLCOs, also contribute to the handling of gut-derived products. Indeed, it has been shown that the SLCOs transport the gut-derived secondary bile acids [71]. One aspect of the RSST that has received less attention is interorganismal communication between the host and the gut microbes [72]. In-depth study is likely to have important clinical ramifications — for instance, in understanding the role of gut microbe–derived uremic toxins in the aberrant metabolism of CKD [28, 30, 61]. If we treat the gut microbiome as an independent organ, it can be considered to express thousands of transporters and enzymes, many of which exhibit broad substrate specificity and are commonly involved in movement of metabolites, and the synthesis of small molecules as well as their hydrolysis, reduction, or removal of conjugated groups [73]. It is established that the substrates and products of the enzymatic reactions occurring in bacterial species overlap with those of the transporters and enzymes in the host, enabling interorganismal communication via multi-, oligo-, and monospecific enzymes that are part of the Remote Sensing and Signaling Network. Furthermore, there is evidence that the gut microbiome and its metabolites can impact transporter and enzyme expression. In renal disease, it has been shown that gut-derived metabolites have an impact on the expression of several drug-metabolizing enzymes in the kidney [74]. The presence of gut microbes has been shown to alter the expression of hepatic DME genes in mice [75]. Indoxyl sulfate, an important gut-derived uremic toxin, has also been shown to regulate the level of OAT1 expression through AHR activation and, along with other aspects of this pathway, has been interpreted as an experimentally verified example of the RSST [68, 76]. Establishing that the communication between the host and the gut microbes is so strongly mediated by renal OAT1 opens up many questions. For example, it has been shown that several drugs — including statins and ACE inhibitors — have an impact on the composition of the gut microbiome and, by implication, the levels and composition of gut microbe–derived metabolites in the host. However, the mechanisms are unclear. Since these and other drugs impacting the gut microbiota are OAT1 substrates, it is possible that these changes could alter competition of the drug with metabolites at the level of the transporter. There is now good evidence in humans to support this kind of DMI at the level of OAT1 [1]. In summary, our studies indicate that multispecific transporters and enzymes combine with the gut microbiome to regulate circulating levels of key metabolites, including those with signaling capabilities. Since these effects need not be limited to OAT1 in the kidney, it is worthwhile to perform similar analyses with multispecific transporters (e.g., SLCO and ABCC families) expressed in the kidney, liver, intestine, and other organs. A much more complex and clinically actionable picture of the regulation of microbiome-dependent host metabolism is likely to emerge. This can be particularly useful for studying DMI. ## Animals. Adult male KO and WT mice were housed in a 12-hour light-dark cycle and allowed ad libitum access to food and water. Oat1-KO mice were generated and maintained as previously described [49]. Feces were collected from mice weekly in sterile 2 mL centrifuge tubes, flash frozen, and stored at –80°C. Animals were sacrificed by CO2 inhalation, and blood samples were extracted from mice by cardiac puncture. Serum was extracted, and samples were flash frozen and stored at –80°C until further analysis. ## Tc-99m MAG3 imaging. Live-imaging experiments were performed with the In Vivo Imaging Shared Resource at the Moore’s Cancer Center at the UCSD. Mice were transported from the vivarium to the In Vivo Imaging Shared Resource. In total, 100 μCi of Tc-99m MAG3 was injected by tail vein into mice prior to imaging. Adult, male KO ($$n = 5$$) and WT mice ($$n = 5$$) were imaged in 5 separate pairs containing 1 KO mouse and 1 WT mouse each. Mice were initially weighed and anesthetized ($2\%$ isoflurane, 200 mL/min flow of O2). Mice were placed on their backs on the high-resolution γ imager (γ imager; BioSpace) fitted with a high-resolution, low-energy collimator. For 30 minutes, time-activity curves were collected from the kidneys, liver, and bladder. Following imaging, blood, urine, kidneys, liver, and spleen were isolated and weighed from each mouse for radioactive assessment using a γ counter. ## Microbiome depletion protocol. Over a 4-week period, mice were given a 125 mL antibiotic cocktail or vehicle control in place of drinking water. The cocktail consisted of 1 mg/mL of neomycin sulfate (Thermo Fisher Scientific, BP-2669-25), 1 mg/mL of ampicillin (Sigma-Aldrich, A9518-100G), 1 mg/mL of metronidazole (Alfa Aesar, H60258), 0.5 mg/mL of vancomycin (Alfa Aesar, J62790), and 3.75 mg/mL of Kool Aid grape drink powder (Kraft-Heinz Foods Company). The Kool Aid encouraged consumption of the cocktail. Antibiotic cocktails were replenished every other day on Monday, Wednesday, and Friday. New solutions were passed through a 0.22 μm cellulose acetate sterilizing filter (Corning, 430517). Bottles of antibiotic cocktail were also wrapped in foil to prevent light damage. The weights of mice and the amount of antibiotic cocktail consumed were monitored over the treatment period as markers of consumption. Following a 1-week decrease in both weight and consumption, mice returned to near their original weights (Supplemental Figure 1). ## Assessment of gut microbiome depletion. For 16S variable region sequencing, murine fecal samples from week 0 and week 4 (end of treatment time point) were extracted using the MagMax Microbiome Ultra Nucleic Acid Isolation Kit (Thermo Fisher Scientific, A42357) according to the manufacturer’s instructions. Variable (V4) regions of 16S SSU rRNA were amplified using 515F-806R primers according to the protocol described in http://earthmicrobiome.ucsd.edu/protocols-and-standards/16s/ 16S sequencing was performed by the Institute for Genomic Medicine (IGM) UCSD. Resulting files were analyzed using the web-based Qiita tool [77]. For the qPCR, feces total RNA was extracted using the RNeasy PowerMicrobiome Kit (Qiagen, 26000-50) according to the manufacturer’s instructions. A cDNA library of the RNA extracted was created using the SUPERSCRIPT III kit (Invitrogen, 18080-044) with random hexamers (Invitrogen, 48190011) as primers according to the guidelines of the manufacturer. RNA and cDNA were quantified using a Nanodrop 1000 (Thermo Fisher Scientific, 2353-30-0010) and were subsequently used to load equal amounts of cDNA for the qPCR performed. Each well in the qPCR plate contained 20 ng of cDNA from fecal RNA. Duplicates of each sample containing Eubacteria primers were utilized, targeting the universal 16S rRNA gene that captures a majority of bacteria [78]. 1.KAPA SYBR FAST Universal kit was utilized with an accompanying protocol (Roche, KK4608). A default 16S metagenomics workflow was run in Qiita under Qiime version 1.9. Raw reads were demultiplexed and trimmed, and OTUs were closed-reference picked using SortMeRNA v2.1 with a $97\%$ sequence similarity minimum. OTUs were assigned taxonomies from the GreenGenes 16S rRNA database, version 13_8, and tabulated into feature and reference tables. All analyses of the feature and reference tables were performed in the Qiita platform and graphed using the Python package, Seaborn. ## Metabolomics analysis of WT and KO mice (untreated and treated). Serum samples were shipped on dry ice to Metabolon Inc. for preparation and metabolomics. Per Metabolon Inc., proteins were removed from the serum and 5 fractions were generated for different mass spectrometry methods. Each sample passed quality control compared with well-characterized controls and was analyzed by ultrahigh performance liquid chromatography–tandem mass spectroscopy (UPLC-MS/MS). Peaks were identified and associated with defined compounds based on retention time, mass/charge ratio, and MS/MS spectral data. Quantification of peaks was performed using AUC. ## Physicochemical analysis of metabolites. Two-dimensional chemical structures were obtained from 783 of the measured metabolites by their Pubchem IDs. Seventy-seven 1-dimensional descriptors were calculated for each metabolite using ICM Molsoft-Pro. These molecular properties were then trimmed to eliminate heavily correlated features using a Spearman’s correlation cutoff of 0.90. Linear discriminant analysis was then applied and visualized using the Python packages Seaborn and scikit-learn. Visualizations of the chemical structures were performed using RDKit. ## In vitro OAT1 transport assays. Human embryonic kidney (HEK) cells stably overexpressing human OAT1 (SOLVO Biotechnology) were used for in vitro inhibition assays. Cells were maintained in DMEM (Thermo Fisher Scientific, catalog 11965092) supplemented with $10\%$ FBS (Thermo Fisher Scientific, catalog 26140079), $1\%$ penicillin/streptomycin (Thermo Fisher Scientific, catalog 15140122), and blasticidin (InvivoGen, catalog ant-bl-1), a selective marker for OAT1 expression. Cells were tested for mycoplasma contamination, and no contamination was observed. Prior to functional assays, cells were plated in 96-well plates and grown for 24 hours or until confluent in media without blasticidin. Metabolites were added at either 1 mM or 2 mM, and serial dilution was performed down all columns. A fixed concentration of 10 μM 6-carboxyfluorescein (6-CF) was introduced to each well for 10 minutes. Cells were rinsed 3 times in ice-cold PBS, and the fluorescence was measured using a fluorescence plate reader. IC50 values were calculated using GraphPad Prism 9. Controls were carried out using probenecid, an established inhibitor of OAT1 function. ## Magnetic bead binding assay. The OAT1 gene was cloned into a third-generation lentiviral vector system. The full-length protein was expressed as GFP fusion in HEK293 cells to check for protein expression. OAT1 protein was solubilized using n-Dodecyl-β-D-Maltoside (bDDM) detergent and then purified by immobilization on magnetic beads coupled to anti-flag antibody (MedChemExpress, HY-K0207). We used magnetic beads (5 μm) as the basis for a binding assay to screen small molecule compounds competing with a well-established OAT1 substrate, 6-CF, which was used as a fluorescence tracer. Loss of fluorescence when challenged with another substrate at a given concentration indicated potential competition for the same binding site. Multiple compounds and concentrations were assayed in a 96-well format using flow cytometry, gating directly on the scattering of the beads. Candidate compounds were initially screened at 3–10 times the concentration of 6-CF (kept constant at 6 μM, for example). Fluorescence measurements were conducted using a Novacyte flow cytometer (Agilent) with a sampler that reads 1 sample well at a time at regular time intervals. Since these magnetic beads are denser than water, we used $50\%$ glycerol to reduce bead sedimentation. Loss of fluorescence when challenged with candidate compounds at different concentrations was normalized against the maximum fluorescence due to 6-CF binding to OAT1, which was also measured periodically between every set of 10 samples. These check-point measurements (evaluating 6-CF binding to OAT1, in this case) spaced in time allowed us to monitor and to correct fluorescence due to bead sedimentation over time. Using this approach, we screened 20 compounds and categorized them (as OAT1 binder versus nonbinder) based on their competitive efficiency against 6 μM 6-CF, selecting for loss of fluorescence signal compared with the relative error in repeated measurements of 6 μM 6-CF binding to OAT1 alone. ## 5/6 Nephrectomy model. In the metabolomics data from $\frac{5}{6}$ nephrectomy model previously described by us [28], 1 kidney and $\frac{2}{3}$ of the other were removed to model diminished renal function. A sham operation was performed on the healthy controls. After 2 weeks, plasma samples were collected from the animals following sacrifice. The samples were metabolically profiled, and comparisons between $\frac{5}{6}$ nephrectomized and healthy, nonnephrectomized animals were analyzed. We then compared these metabolites with those in Supplemental Table 3. ## Human DMI. As previously described by us, plasma and urine samples were collected from 20 healthy participants before and 5 hours after an oral dose of probenecid [1]. These samples were metabolically profiled, and pre- and postcomparisons were analyzed to determine compounds that were significantly altered. We analyzed compounds that were elevated in the plasma, those that were decreased in the urine, and those that satisfied both criteria. We then compared these metabolites to those in Supplemental Table 3. ## Statistics. Scaled intensity for each metabolite was normalized to volume, and missing values were imputed with the lowest value for the compound. For all fold-change calculations, scaled intensities were averaged and compared with each other. Statistical comparisons were performed using the Python module, statsmodels (https://www.statsmodels.org/stable/index.html) and were made between groups by 2-way ANOVA following log transformation and FDR correction, with $P \leq 0.05$ being considered significant. Enrichment for each superpathway and subpathway was calculated using the number of total metabolites measured and the number of metabolites in each respective superpathway and subpathway, as previously described [16]. ## Study approval. All experimental protocols were approved by the UCSD IACUC, and the animals were handled in accordance with the Institutional Guidelines on the Use of Live Animals for Research. All the experiments described here follow the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines. ## Author contributions JCG and SKN wrote the manuscript. JCG, VE, and KM conducted experiments and acquired data. JCG and KM analyzed data. SKN, DRV and GC provided reagents and resources. SKN conceived of the project. SKN and GC designed various aspects of the research studies. All authors reviewed the manuscript. ## 01/24/2023 Electronic publication ## References 1. Granados JC. Blockade of organic anion transport in humans after treatment with the drug probenecid leads to major metabolic alterations in plasma and urine. Clin Pharmacol Ther (2022.0) 112 653-664. DOI: 10.1002/cpt.2630 2. Wang X. From the cover: identification of natural products as inhibitors of human organic anion transporters (OAT1 and OAT3) and their protective effect on mercury-induced toxicity. Toxicol Sci (2018.0) 161 321-334. DOI: 10.1093/toxsci/kfx216 3. Wu W. Key role for the organic anion transporters, OAT1 and OAT3, in the in vivo handling of uremic toxins and solutes. Sci Rep (2017.0) 7. DOI: 10.1038/s41598-017-04949-2 4. Lu H. Interactions of 172 plant extracts with human organic anion transporter 1 (SLC22A6) and 3 (SLC22A8): a study on herb-drug interactions. PeerJ (2017.0) 5. PMID: 28560096 5. Liu HC. An organic anion transporter 1 (OAT1)-centered metabolic network. J Biol Chem (2016.0) 291 19474-19486. DOI: 10.1074/jbc.M116.745216 6. Burckhardt G. Drug transport by Organic Anion Transporters (OATs). Pharmacol Ther (2012.0) 136 106-130. DOI: 10.1016/j.pharmthera.2012.07.010 7. Torres AM. Deletion of multispecific organic anion transporter Oat1/Slc22a6 protects against mercury-induced kidney injury. J Biol Chem (2011.0) 286 26391-26395. DOI: 10.1074/jbc.M111.249292 8. Lopez-Nieto CE. Molecular cloning and characterization of NKT, a gene product related to the organic cation transporter family that is almost exclusively expressed in the kidney. J Biol Chem (1997.0) 272 6471-6478. DOI: 10.1074/jbc.272.10.6471 9. Zhang J. Regulation of organic anion transporters: role in physiology, pathophysiology, and drug elimination. Pharmacol Ther (2021.0) 217. DOI: 10.1016/j.pharmthera.2020.107647 10. Nigam SK, Granados JC. A biological basis for pharmacokinetics: the remote sensing and signaling theory. Clin Pharmacol Ther (2022.0) 112 456-460. DOI: 10.1002/cpt.2634 11. Ahn SY. Interaction of organic cations with organic anion transporters. J Biol Chem (2009.0) 284 31422-31430. DOI: 10.1074/jbc.M109.024489 12. Fromm MF. Transporters and drug-drug interactions: important determinants of drug disposition and effects. Pharmacol Rev (2015.0) 238 949-966. PMID: 23686349 13. Giacomini KM. Membrane transporters in drug development. Nat Rev Drug Discov (2010.0) 9 215-236. DOI: 10.1038/nrd3028 14. Nigam SK. What do drug transporters really do?. Nat Rev Drug Discov (2015.0) 14 29-44. DOI: 10.1038/nrd4461 15. Liang Y. The physiological role of drug transporters. Protein Cell (2015.0) 6 334-350. DOI: 10.1007/s13238-015-0148-2 16. Granados JC. A key role for the transporter OAT1 in systemic lipid metabolism. J Biol Chem (2021.0) 296. DOI: 10.1016/j.jbc.2021.100603 17. Granados JC. Coordinate regulation of systemic and kidney tryptophan metabolism by the drug transporters OAT1 and OAT3. J Biol Chem (2021.0) 296. DOI: 10.1016/j.jbc.2021.100575 18. Krautkramer KA. Gut microbial metabolites as multi-kingdom intermediates. Nat Rev Microbiol (2020.0) 19 77-94. PMID: 32968241 19. Martinez KB. Microbial metabolites in health and disease: navigating the unknown in search of function. J Biol Chem (2017.0) 292 8553-8559. DOI: 10.1074/jbc.R116.752899 20. Wikoff WR. Metabolomics analysis reveals large effects of gut microflora on mammalian blood metabolites. Proc Natl Acad Sci U S A (2009.0) 106 3698-3703. DOI: 10.1073/pnas.0812874106 21. Sharon G. Specialized metabolites from the microbiome in health and disease. Cell Metab (2014.0) 20 719-730. DOI: 10.1016/j.cmet.2014.10.016 22. Hosseinkhani F. The contribution of gut bacterial metabolites in the human immune signaling pathway of non-communicable diseases. Gut Microbes (2021.0) 13 1-22. PMID: 33590776 23. Tang WHW. Gut microbiota in cardiovascular health and disease. Circ Res (2017.0) 120 1183-1196. DOI: 10.1161/CIRCRESAHA.117.309715 24. Wang ZN, Zhao YZ. Gut microbiota derived metabolites in cardiovascular health and disease. Protein Cell (2018.0) 9 416-431. DOI: 10.1007/s13238-018-0549-0 25. Glassner KL. The microbiome and inflammatory bowel disease. J Allergy Clin Immunol (2020.0) 145 16-27. DOI: 10.1016/j.jaci.2019.11.003 26. Nallu A. Gut microbiome in chronic kidney disease: challenges and opportunities. Transl Res (2017.0) 179 24-37. DOI: 10.1016/j.trsl.2016.04.007 27. Hobby GP. Chronic kidney disease and the gut microbiome. Am J Physiol Renal Physiol (2019.0) 316 F1211-F1217. DOI: 10.1152/ajprenal.00298.2018 28. Bush KT. Gut-derived uremic toxin handling in vivo requires OAT-mediated tubular secretion in chronic kidney disease. JCI Insight (2020.0) 5. DOI: 10.1172/jci.insight.133817 29. Duranton F. Normal and pathologic concentrations of uremic toxins. J Am Soc Nephrol (2012.0) 23 1258-1270. DOI: 10.1681/ASN.2011121175 30. Nigam SK, Bush KT. Uraemic syndrome of chronic kidney disease: altered remote sensing and signalling. Nat Rev Nephrol (2019.0) 15 301-316. DOI: 10.1038/s41581-019-0111-1 31. Wikoff WR. Untargeted metabolomics identifies enterobiome metabolites and putative uremic toxins as substrates of organic anion transporter 1 (Oat1). J Proteome Res (2011.0) 10 2842-2851. DOI: 10.1021/pr200093w 32. Rosenthal SB. A network of SLC and ABC transporter and DME genes involved in remote sensing and signaling in the gut-liver-kidney axis. Sci Rep (2019.0) 9. DOI: 10.1038/s41598-019-47798-x 33. Ahn SY, Nigam SK. Toward a systems level understanding of organic anion and other multispecific drug transporters: a remote sensing and signaling hypothesis. Mol Pharmacol (2009.0) 76 481-490. DOI: 10.1124/mol.109.056564 34. Wu W. Remote communication through solute carriers and ATP binding cassette drug transporter pathways: an update on the remote sensing and signaling hypothesis. Mol Pharmacol (2011.0) 79 795-805. DOI: 10.1124/mol.110.070607 35. Nigam SK, Granados JC. OAT, OATP, and MRP drug transporters and the remote sensing and signaling theory. Annu Rev Pharmacol Toxicol. DOI: 10.1146/annurev-pharmtox-030322-084058 36. Byndloss MX, Baumler AJ. The germ-organ theory of non-communicable diseases. Nat Rev Microbiol (2018.0) 16 103-110. DOI: 10.1038/nrmicro.2017.158 37. Suga T. Preference of conjugated bile acids over unconjugated bile acids as substrates for OATP1B1 and OATP1B3. PLoS One (2017.0) 12. DOI: 10.1371/journal.pone.0169719 38. Toyohara T. SLCO4C1 transporter eliminates uremic toxins and attenuates hypertension and renal inflammation. J Am Soc Nephrol (2009.0) 20 2546-2555. DOI: 10.1681/ASN.2009070696 39. Swanson HI. Drug metabolism by the host and gut microbiota: a partnership or rivalry?. Drug Metab Dispos (2015.0) 43 1499-1504. DOI: 10.1124/dmd.115.065714 40. Nichols RG. Interplay between the host, the human microbiome, and drug metabolism. Hum Genomics (2019.0) 13. PMID: 31186074 41. Nigam SK. The systems biology of drug metabolizing enzymes and transporters: relevance to quantitative systems pharmacology. Clin Pharmacol Ther (2020.0) 108 40-53. DOI: 10.1002/cpt.1818 42. Kennedy EA. Mouse microbiota models: comparing germ-free mice and antibiotics treatment as tools for modifying gut bacteria. Front Physiol (2018.0) 9. DOI: 10.3389/fphys.2018.01534 43. Rosenbaum M. The gut microbiota in human energy homeostasis and obesity. Trends Endocrinol Metab (2015.0) 26 493-501. DOI: 10.1016/j.tem.2015.07.002 44. Vriend J. Organic anion transporters 1 and 3 influence cellular energy metabolism in renal proximal tubule cells. Biol Chem (2019.0) 400 1347-1358. DOI: 10.1515/hsz-2018-0446 45. Tirelle P. Comparison of different modes of antibiotic delivery on gut microbiota depletion efficiency and body composition in mouse. BMC Microbiol (2020.0) 20. DOI: 10.1186/s12866-020-02018-9 46. Shikano N. Transport of 99mTc-MAG3 via rat renal organic anion transporter 1. J Nucl Med (2004.0) 45 80-85. PMID: 14734677 47. Takahara N. Drugs interacting with organic anion transporter-1 affect uptake of Tc-99m-mercaptoacetyl-triglycine (MAG3) in the human kidney: therapeutic drug interaction in Tc-99m-MAG3 diagnosis of renal function and possible application of Tc-99m-MAG3 for drug development. Nucl Med Biol (2013.0) 40 643-650. DOI: 10.1016/j.nucmedbio.2013.03.006 48. Hsueh CH. Identification and quantitative assessment of uremic solutes as inhibitors of renal organic anion transporters, OAT1 and OAT3. Mol Pharm (2016.0) 13 3130-3140. DOI: 10.1021/acs.molpharmaceut.6b00332 49. Eraly SA. Decreased renal organic anion secretion and plasma accumulation of endogenous organic anions in OAT1 knock-out mice. J Biol Chem (2006.0) 281 5072-5083. DOI: 10.1074/jbc.M508050200 50. Shen H. Evidence for the validity of pyridoxic acid (PDA) as a plasma-based endogenous probe for OAT1 and OAT3 function in healthy subjects. J Pharmacol Exp Ther (2019.0) 368 136-145. DOI: 10.1124/jpet.118.252643 51. Krautkramer KA. Gut microbial metabolites as multi-kingdom intermediates. Nat Rev Microbiol (2021.0) 19 77-94. DOI: 10.1038/s41579-020-0438-4 52. Visconti A. Interplay between the human gut microbiome and host metabolism. Nat Commun (2019.0) 10. PMID: 31582752 53. Yu Y. Deoxycholic acid activates colonic afferent nerves via 5-HT3 receptor-dependent and -independent mechanisms. Am J Physiol Gastrointest Liver Physiol (2019.0) 317 G275-G284. DOI: 10.1152/ajpgi.00016.2019 54. Liu X. The membrane bile acid receptor TGR5 drives cell growth and migration via activation of the JAK2/STAT3 signaling pathway in non-small cell lung cancer. Cancer Lett (2018.0) 412 194-207. DOI: 10.1016/j.canlet.2017.10.017 55. Jensen DD. The bile acid receptor TGR5 does not interact with β-arrestins or traffic to endosomes but transmits sustained signals from plasma membrane rafts. J Biol Chem (2013.0) 288 22942-22960. DOI: 10.1074/jbc.M113.455774 56. Correia MSP. Comparative dietary sulfated metabolome analysis reveals unknown metabolic interactions of the gut microbiome and the human host. Free Radic Biol Med (2020.0) 160 745-754. DOI: 10.1016/j.freeradbiomed.2020.09.006 57. Koppel N. Chemical transformation of xenobiotics by the human gut microbiota. Science (2017.0) 356. DOI: 10.1126/science.aag2770 58. Needham BD. Plasma and fecal metabolite profiles in autism spectrum disorder. Biol Psychiatry (2021.0) 89 451-462. DOI: 10.1016/j.biopsych.2020.09.025 59. Nigam AK. Unique metabolite preferences of the drug transporters OAT1 and OAT3 analyzed by machine learning. J Biol Chem (2020.0) 295 1829-1842. DOI: 10.1074/jbc.RA119.010729 60. Liu HC. Molecular properties of drugs interacting with SLC22 transporters OAT1, OAT3, OCT1, and OCT2: a machine-learning approach. J Pharmacol Exp Ther (2016.0) 359 215-229. DOI: 10.1124/jpet.116.232660 61. Lowenstein J, Nigam SK. Uremic toxins in organ crosstalk. Front Med (lausanne) (2021.0) 8. PMID: 33937275 62. Bush KT. The drug transporter OAT3 (SLC22A8) and endogenous metabolite communication via the gut-liver-kidney axis. J Biol Chem (2017.0) 292 15789-15803. DOI: 10.1074/jbc.M117.796516 63. Nigam SK. The organic anion transporter (OAT) family: a systems biology perspective. Physiol Rev (2015.0) 95 83-123. DOI: 10.1152/physrev.00025.2013 64. Nigam SK. Handling of drugs, metabolites, and uremic toxins by kidney proximal tubule drug transporters. Clin J Am Soc Nephrol (2015.0) 10 2039-2049. DOI: 10.2215/CJN.02440314 65. Tan JK. Metabolite-sensing g protein-coupled receptors-facilitators of diet-related immune regulation. Annu Rev Immunol (2017.0) 35 371-402. DOI: 10.1146/annurev-immunol-051116-052235 66. Colosimo DA. Mapping interactions of microbial metabolites with human G-protein-coupled receptors. Cell Host Microbe (2019.0) 26 273-282. DOI: 10.1016/j.chom.2019.07.002 67. Graboski AL, Redinbo MR. Gut-derived protein-bound uremic toxins. Toxins (Basel) (2020.0) 12. DOI: 10.3390/toxins12090590 68. Jansen J. Remote sensing and signaling in kidney proximal tubules stimulates gut microbiome-derived organic anion secretion. Proc Natl Acad Sci U S A (2019.0) 116 16105-16110. DOI: 10.1073/pnas.1821809116 69. Schroeder JC. The uremic toxin 3-indoxyl sulfate is a potent endogenous agonist for the human aryl hydrocarbon receptor. Biochemistry (2010.0) 49 393-400. DOI: 10.1021/bi901786x 70. Yasuda H. Involvement of membrane-type bile acid receptor M-BAR/TGR5 in bile acid-induced activation of epidermal growth factor receptor and mitogen-activated protein kinases in gastric carcinoma cells. Biochem Biophys Res Commun (2007.0) 354 154-159. DOI: 10.1016/j.bbrc.2006.12.168 71. Van de Steeg E. Organic anion transporting polypeptide 1a/1b-knockout mice provide insights into hepatic handling of bilirubin, bile acids, and drugs. J Clin Invest (2010.0) 120 2942-2952. DOI: 10.1172/JCI42168 72. Jamshidi N, Nigam SK. Drug transporters OAT1 and OAT3 have specific effects on multiple organs and gut microbiome as revealed by contextualized metabolic network reconstructions. Sci Rep (2022.0) 12. DOI: 10.1038/s41598-022-21091-w 73. Clarke G. Gut reactions: breaking down xenobiotic-microbiome interactions. Pharmacol Rev (2019.0) 71 198-224. DOI: 10.1124/pr.118.015768 74. Prokopienko AJ, Nolin TD. Microbiota-derived uremic retention solutes: perpetrators of altered nonrenal drug clearance in kidney disease. Expert Rev Clin Pharmacol (2018.0) 11 71-82. DOI: 10.1080/17512433.2018.1378095 75. Selwyn FP. RNA-seq quantification of hepatic drug processing genes in germ-free mice. Drug Metab Dispos (2015.0) 43 1572-1580. DOI: 10.1124/dmd.115.063545 76. Granados JC. AHR is a master regulator of diverse pathways in endogenous metabolism. Sci Rep (2022.0) 12. DOI: 10.1038/s41598-022-20572-2 77. Gonzalez A. Qiita: rapid, web-enabled microbiome meta-analysis. Nat Methods (2018.0) 15 796-798. DOI: 10.1038/s41592-018-0141-9 78. Robertson SJ. Nod1 and Nod2 signaling does not alter the composition of intestinal bacterial communities at homeostasis. Gut Microbes (2013.0) 4 222-231. DOI: 10.4161/gmic.24373 79. Kaler G. Structural variation governs substrate specificity for organic anion transporter (OAT) homologs. Potential remote sensing by OAT family members. J Biol Chem (2007.0) 282 23841-23853. DOI: 10.1074/jbc.M703467200
--- title: Ablation of Tumor Necrosis Factor Alpha Receptor 1 Signaling Blunts Steatohepatitis in Peroxisome Proliferator Activated Receptor α-Deficient Mice authors: - Ian N. Hines - Jamie Milton - Michael Kremer - Michael D. Wheeler journal: Medical research archives year: 2022 pmcid: PMC9977327 doi: 10.18103/mra.v10i9.3082 license: CC BY 4.0 --- # Ablation of Tumor Necrosis Factor Alpha Receptor 1 Signaling Blunts Steatohepatitis in Peroxisome Proliferator Activated Receptor α-Deficient Mice ## Abstract Tumor necrosis factor -alpha (TNFα) is strongly associated with fatty liver disease (i.e, hepatosteatosis). Cytokine production has been thought of as a consequence of hepatic lipid accumulation which becomes a critical factor in the development of chronic liver pathologies as well as insulin resistance. The purpose of this study was to test the hypothesis that TNFα directly regulates lipid metabolism in liver in the mutant peroxisome-proliferator activated receptor-alpha (PPARα−/−) mouse model with robust hepatic lipid accumulation. At 10 weeks of age, TNFα and TNF receptor 1 expression are increased in livers of PPARα−/− mice compared to wild type. PPARα−/− mice were then crossed with mice lacking the receptor for TNFα receptor 1 (TNFR1−/−). Wild type, PPARα−/−, TNFR1−/−, PPARα−/− x TNFR1−/− mice were housed on ad-libitum standard chow diet for up to 40 weeks. Increases in hepatic lipid and liver injury and metabolic disruption associated with PPARα ablation were largely blunted when PPARα−/− mice were crossed with TNFR1−/− mice. These data support the hypothesis that TNFR1 signaling is critical for accumulation of lipid in liver. Therapies that reduce pro-inflammatory responses, namely TNFα, could have important clinical implications to reduce hepatosteatosis and progression of severe liver disease. ## INTRODUCTION Non-alcoholic fatty liver disease (NAFLD) is a rapidly growing cause of liver damage, dysfunction, and failure in westernized countries 1. Estimates within the United States alone predict greater than $22\%$ of the population suffers from some form of this spectrum disorder. Indeed, hepatocellular steatosis, or the accumulation of lipid within hepatocytes, was once thought to be a benign pathology with little effect on cell or tissue function 2. Indeed, NAFLD is characterized as a progressive pathology beginning with simple steatosis and evolving, over time, to steatohepatitis (NASH) with significant inflammatory cell infiltrate and hepatocellular damage, and ultimately in a smaller percentage of patients (~$20\%$ of patients with NASH) to tissue scarring or fibrosis and cirrhosis 3. Key experimental studies have defined a number of factors which contribute to NAFLD disease induction and evolution 4. From these, it is clear that early activation of hepatic macrophages, the Kupffer cell, by gut-derived factors including endotoxin promote Tlr4 dependent expression of a variety of inflammatory cytokines and promote inflammatory cell infiltrate, hepatocellular mitochondrial dysfunction, and lipid metabolism disruption 5 6 7. Indeed, Miura and others, using the choline deficient diet model of NAFLD, demonstrate the profound importance of macrophages and their recruitment through inflammatory chemokine receptors, specifically CCR2, to promote lipid accumulation and progression towards fibrogenesis 8. Moreover, loss of Tlr4 was shown to limit lipid accumulation in a similar model of fatty liver in mice 5. Together, it is clear that inflammation and inflammatory cytokines derived from or initiated by resident immune cells propagate hepatocellular injury and disruption of critical metabolic programs leading to lipid accumulation. While it is still not fully understood how inflammatory cytokines impact hepatic metabolism or likewise, how altered hepatic metabolism affects inflammatory responses, a growing body of experimental data would implicate certain pro-inflammatory cytokines as potential initiators and / or propagators of non-alcoholic steatohepatitis (NASH) 9. Increased lipid accumulation itself promotes hepatocyte dysfunction, hepatocellular oxidant stress, and activation of hepatic innate immunity 10. Moreover, lipid accumulation depletes antioxidants and increases oxidative stress through lipid peroxidation, both of which may promote DNA damage and hepatic carcinogenesis 2,10,11. One potential hypothesis is the lipid accumulation within liver, as with the case of high-fat diet, hyper-caloric diet or ethanol containing diet, activates inflammatory cascades resulting in cytokine release that propagates further inflammatory cell recruitment and hepatocellular damage. Tumor necrosis factor alpha (TNFa) is a key cytokine produced by a variety of inflammatory cells including macrophages as well as by hepatocytes during periods of stress 12 13. Signaling primarily through two TNF receptors (TNFR1 and TNFR2), TNFa is associated with a variety of cellular responses from cell proliferation and differentiation to apoptosis and regulation of the immune response 14. Within the liver, TNFa is well appreciated for its ability to promote inflammation through propagation of macrophage function while its production promotes the regenerative response following hepatectomy 14 15 16. Intriguingly, TNFa is also associated with mitochondrial dysfunction 7. Recent studies have highlighted its capacity to decrease mitochondrial respiration in aging platelets as well as in neuronal cells in vitro 17 18. Similarly, very early work with isolated hepatocytes showed reduced mitochondrial respiration in response to TNFa exposure in vitro, data which was later extended showing a direct nuclear factor kappa B (NFkB) mediated, reactive oxygen species dependent uncoupling of complexes I and III 19 20. Likewise, TNFα also remains a clear link between innate immunity and obesity and metabolism, particularly within the liver 21. To this point, mice lacking the TNF receptor 1 show resistance to diet-induced insulin resistance 22. In summary, TNFa plays an important role in hepatic inflammatory responses and contributes to altered metabolic function both within the liver as well as peripherally. Mice lacking the gene for peroxisome proliferator-activated receptor a (PPARα−/−) develop robust fatty liver disease within 20 weeks of age and severe liver disease within 40–60 weeks of age 23. Mechanistically, PPARα is a transcriptional regulator of acyl CoA oxidase and thus is a critical regulator of hepatic fatty acid oxidation 24. Since the accumulation of lipid is related to an impairment in lipid oxidation in PPARα−/− mice, these mutant mice are a useful model to investigate the effects of lipid accumulation in liver in contrast to dietary models where nutrient composition, intake, or the bio-active effect of nutrients on gut microbiota, metabolism or immune response play a role. Importantly, PPARα−/− mice also exhibit an increase in TNFα expression in liver and this observation may suggest that TNFα pathways may be activated in response to lipid accumulation or possible causal in the lipid accumulation that leads to subsequent liver injury. ## Objective: To better understand the role of TNF receptor dependent signaling in the development and progression of fatty liver disease, PPARα-deficient mice were crossed with mice lacking TNFR1 deficient mice. Using this well-defined genetic model of dysregulated lipid metabolism and consequent fatty liver development which is independent of dietary variables, this study will determine the influence of TNF signaling directly on metabolic responses in this paradigm within the murine liver. Information gathered here will better define the factors responsible for fatty liver development and provide possible therapeutic targets to treat or prevent NAFLD. ## Animals and treatment. Male C57Bl/6J wild type mice or TNFα receptor 1-deficient (TNFR1−/−; C57Bl/6- Tnfrsf1atm1Imx/J) mice were purchased from Jackson laboratories (Bar Harbor, ME) while peroxisome proliferator activated receptor alpha-deficient mice (PPARα−/−, B6.129S4-Pparatm1Gonz N12) on a C57Bl/6 background were obtained from Taconic (Hudson, NY). The PPARα and TNFR1 double (DKO) deficient mice were generated by cross-breeding in house. The genotypes of all animals were verified by standard PCR procedures using primer sequences from the suppliers. Each strain was maintained through established breeding protocols and kept within AALAC approved facilities and guidelines. The procedures for the care and treatment of mice were followed according to those set by East Carolina University Institutional Animal Care and Use Committee guidelines. Male mice of each genotype were then fed a standard lab diet (Prolab RMH 3000, LabDiet. St. Louis, MO) for up to 40 weeks of age. Following feeding, the mice were sacrificed and serum and tissue collected for further analysis of routine parameters of liver injury and lipid accumulation. ## Live animal monitoring. NMR-MRI (EchoMRI, Houston TX) analyses were performed for body composition measurements of fat, lean, free water and total water masses in live mice. ## Body mass index. Body weight was measured for each mouse prior to euthanasia. Crown-rump length defined as the distance between the crown of the skull and a point located in the middle of a line between the two caput femoris was measured. The BMI was then calculated as the body weight (g)/[crown-rump length (mm)]2. ## Measurements of Serum Parameters. Serum levels of alanine aminotransferase and triglycerides were measured by spectrophotometric analysis (Sigma-Aldrich, St. Louis MO). Clinical chemistry was performed by the University of North Carolina Clinical Chemistry laboratory. Serum glucose levels were determined using a glucose analyzer (Beckman, Fullerton, California) while insulin, leptin, and adiponectin were measured via radioimmunoassay as previously described 25. ## Histopathology and Immunohistochemistry. Tissue was fixed in $4\%$ phosphate buffered formalin for 24 hours and subsequently embedded in paraffin. Tissue sections were prepared (7μm thick) and subjected to routine hematoxylin and eosin staining. ## Real time Reverse Transcriptase Polymerase Chain Reaction. Total RNA from liver was isolated using the Trizol reagent (Gibco/ ThermoFisher Scientific, Grand Island NY) according to the manufacturer’s recommendations. Total RNA (1μg) was used to synthesize cDNA using the High-Capacity cDNA Reverse Transcription kit from Applied Biosystems. For quantification of message expression, cDNA was amplified using specific primer sequences for murine TNFα (F- 5’-AGCCCACGTAGCAAACCACCAA-3’; R- 5’-ACACCCATTCCCTTCACAGAGCAAT-3’), TNF receptor 1 (F- 5- ’-GCCCGAAGTCTACTCCATCATTTG-3’; R- 5’GGCTGGGGAGGGGGCTGGAGTTAG-3’), and b actin (F - 5’-AGGTGTGCACCTTTTATTGGTCTCAA-3’; R - 5’-TGTAGTAAGGTTTGGTCTCCCT-3’) in the presence of Taq polymerase and Sybr Green using a kit from Applied Biosystems/ ThermoFisher Scientific (Grand Island NY) using a standard PCR protocol (95°C for 10s, 57°C for 15s, and 72°C for 20s, total of 40 cycles. β-actin message expression was used as the house keeping gene and for quantification of relative expression levels using the comparative cT method of quantification. ## Statistical Analysis Data are presented as mean ± standard error of the mean (SEM) of 3 or more animals per group. Data were analyzed using non-parametric Student’s t-Test where significance was set at $p \leq 0.05.$ ## RESULTS It is documented elsewhere that ablation of PPARα results in age-dependent lipid accumulation within liver 26. Here, it is demonstrated that PPARα−/− mice (PKO) also exhibit increased mRNA levels of TNFα at 10 weeks of age. Compared to wild type mice, the level of mRNA for TNFα was increased 2.8-fold in livers of PPARα−/− mice at this early timepoint (Figure 1). The levels of mRNA for TNFR1p55 in liver were not, however, significantly elevated in PKO mice when compared to their wild type controls at 10 weeks of age. The increase in TNFα expression in liver precedes the known accumulation of hepatic lipid associated with the ablation of PPARα. These findings support the hypothesis that the TNFα pathway plays a causal role in the liver pathology associated with PPARα ablation. To address the role of TNFα in a genetic model of hepatic steatosis, TNFR1−/− mice (TKO) were crossed with mice lacking peroxisome proliferator activated receptor alpha (PKO) to generate double mutant PPARα−/− TNFR1−/− mice (PTKO). Mice were housed in 12-hour light-dark cycles and were provided standard chow diet ad libitum. A significant increase in body weight was observed in PKO mice compared to that of wild type (WT) mice (Figure 2) at 40 weeks of age. The weight gain in both TKO and PTKO mice was not significantly different from WT mice. Body fat composition of each strain was determined by magnetic resonance. Lean muscle mass was not significantly different among strains. The fat mass percentage of WT mice was 18.8±$3.0\%$ while body fat content in PKO mice at 40 weeks of age was 24.6±$1.8\%$. Importantly, the body fat percentage in TKO and PTKO mice at similar ages was 13.0 and $13.8\%$, respectively (Figure 2). The changes in body fat composition correlated with BMI, which was determined at sacrifice. The increase in BMI observed in PKO compared to WT mice was significantly blunted in PTKO mice. Loss of PPARa is well appreciated to result in hepatosteatosis. Livers from PKO mice at 10 weeks of age had evidence of very mild fat accumulation where wild type had exhibited normal histology, but at 40 weeks of age, PKO mice exhibited severe fat accumulation and mild inflammation which was not observed in similarly aged WT controls. Hepatosteatosis was not observed after 10 weeks or 40 weeks in TKO mice. Importantly, hepatosteatosis was absent in PTKO mice after 10 and 40 weeks (Figure 3). Serum ALT levels were measured at 10 and 40 weeks of age in all groups. Loss of PPARa led to an increased serum ALT levels at 40 weeks of age when compared to similarly aged wild type mice. This increase was blunted in PTKO mice. Liver weight was measured at euthanasia and liver to body weight ratio was determined. Liver to body weight ratio, indicative of hepatosteatosis, was significantly increased in PKO compared to WT mice at 40 weeks of age (Figure 3). Like body weight and fat mass, the liver to body weight ratio was similar to WT in both TKO and PTKO mice. Liver triglyceride levels were significantly elevated (approximately 2-fold) at 40 weeks in PKO mice, compared to the level in WT mice at similar time points. In TKO mice, liver triglyceride levels were similar to that of WT animals at both 10 and 40 weeks. In PTKO mice, liver triglyceride levels were only mildly elevated after 40 weeks. Importantly, the liver triglyceride levels were significantly blunted in PTKO mice compared to PKO mice (Figure 3). These data support the hypothesis that fatty liver due to loss of PPARα is dependent upon TNFR1 receptor expression. To further understand the impact of TNFa on the systemic metabolic response in PKO mice, serum levels of glucose, triglycerides, insulin, leptin, and adiponectin were measured (Figure 4). Following 40 weeks of chow feeding, wild type mice had serum glucose levels of 297.33 ± 6.96 mg/dL. Interestingly, loss of PPARa was associated with a reduction, although not significant, in blood glucose levels when compared to wild type controls. Absence of TNFaR1 was also associated with a reduction, although not significant, in blood glucose levels when compared to controls. Absence of both TNFaR1 and PPARa showed the most consistent reduction in blood glucose levels when compared to similarly aged wild type mice. Serum triglycerides were also examined. Following 40 weeks of chow feeding, wild type mice presented with serum TGs at 79.66 ± 21.16 mg/dL. Loss of PPARa did not alter serum triglyceride levels at this age when compared to wild type controls (84.33 ± 8.74 mg/dL). In the absence of TNFaR1, serum levels of triglycerides were significantly reduced compared to wild type controls (41.66 ± 2.14 mg/dL). Importantly, when TNFaR1 was absent in PPARa deficient mice, serum triglycerides were also significantly reduced compared to wild type controls and PPARa deficient mice. Selected metabolic hormone levels were also measured. Following 40 weeks of chow diet feeding, insulin levels were mildly elevated in PKO mice when compared to similarly aged wild type mice. However, when TNFa receptor was absent in PKO mice, serum insulin levels were consistent with wild type control levels. Loss of TNFa receptor alone did not alter insulin levels after 40 weeks of chow feeding when compared to wild type controls (Figure 4). Evaluation of serum leptin levels revealed a large increase in PKO mice when compared to wild type mice at 40 weeks of age. This increase was completely abrogated in PTKO mice suggesting a role for TNFa signaling in this increase. Loss of TNFa receptor alone did not alter serum leptin levels when compared to similarly aged wild type controls. Finally, serum adiponectin levels were measured and no differences were noted in the three mutants when compared to similarly aged wild type controls (Figure 4). ## DISCUSSION A number of pro-inflammatory cytokines and chemokines including TNFα are up-regulated in fatty livers 27. Likewise, innate immune responses activated within fatty livers have great potential for amplification 28 29 5 30. Once cytokine production is initiated, these pro-inflammatory cytokines propel the progression from steatosis to steatohepatitis. The evidence in favor of the role of TNFα in fatty liver disease is overwhelming, although direct evidence from animal models has been mixed 31 32 33 34 35. Moreover, the mechanism by which TNFα contributes to hepatic lipid metabolism is a gap in our understanding. TNFα is a potent inflammatory mediator derived from a variety of cell types which interacts with a wide range of signaling pathway. Importantly, we have demonstrated here and previously in an ethanol-diet model of steatosis/steatohepatitis that TNFα is crucial for the accumulation of lipid in the liver 15. Here, we used the TNFR1−/− deficient mice to determine its role in the general mechanisms of fatty liver disease in the genetically obese PPARα−/− mice. Importantly, these data suggest that hepatic inflammation may precede and promote the accumulation of lipid, affirming the notion that cytokines drive and potentiate hepatic metabolic dysfunction as well as the pro-inflammatory cascade. There is some experimental evidence that TNFα is increased in response to the accumulation of lipid as well as a growing body of evidence that TNFa is also a regulator of hepatic lipid metabolism. Diehl and others demonstrated decreased lipid accumulation in TNFα-deficient mice fed a high calorie diet 32. This was an important finding since there is much evidence that high fat diet- similar to an ethanol containing diet- causes an increase in TNFα production in liver. Much work has also investigated c-jun N-terminal kinase (JNK), a primary downstream target of TNFR1 in hepatocytes, as a regulator of hepatic lipid accumulation 36 37. Indeed, JNK null mice were resistant to both methionine-choline deficient diet induced hepatosteatosis and high-fat diet induced fatty liver and secondary tissue injury and hepatocellular apoptosis resulting from this lipid overaccumulation 36. A central question remains whether TNFα is a consequence of lipid accumulation that follows metabolic derangement for example due to ethanol, high fat or hyper-caloric diet or whether TNF is a driver of metabolic changes that result in fat accumulation and subsequent liver inflammation and injury. In either case, the hypothesis is that TNFα through TNFR1 pathways exacerbates liver inflammation and blunts lipid metabolic pathways, which further perpetuates the development of liver injury. The notion of a “feed forward” cycle involving lipid metabolism and pro-inflammatory cytokines is not novel. Fiengold reported that TNF suppressed lipid metabolism including an increase in serum triglyceride levels and a decrease in hepatic fatty acid oxidation, in bile acid synthesis, and in high-density lipoprotein levels 21. These effects of TNFα were through the suppressed expression of nuclear hormone receptors retinoid X receptor alpha (RXRa), PPARa, PPARg, and liver X receptor alpha (LXRa), as well as coactivators peroxisome proliferator-activated receptor gamma co-activator 1 alpha (PGC-1a) and PGC-1b 21. This observation as well as others led to the notion of a “two-hit” model of liver injury, where metabolic alterations were coupled with the pro-inflammatory responses to propagate a futile and deleterious cycle of liver injury ultimately leading to irreversible pathology. Studies of the effects and interactions of PPARa deficiency and TNFa in the regulation of systemic metabolic were less conclusive. Consistent with previous reports, loss of PPARa led to a reduction, albeit not significant, in fasting glucose levels, a response which was furthered by the loss of TNFa signaling in these mice 23. This correlated with increased levels of insulin secretion in these mice when compared to wild type controls. The concomitant loss of TNFa and PPARa returned serum insulin levels to that seen in wild type controls. These findings are somewhat consistent with previous studies which demonstrated a role for TNFa in the development of insulin insensitivity 22. The differences likely lie within the models utilized, where high fat, high calorie diets likely magnify the baseline inflammatory response increasing the influence of factors such as TNFa and others in the metabolic processes. Similarly, loss of PPARa did not significantly alter serum triglyceride levels when compared to wild type controls. This is also consistent with previous reports showing no significant alterations in serum triglyceride levels in PPARa deficient mice on a standard chow diet 38. Importantly, our data again support a role for TNF signaling directly in the regulation of systemic metabolic responses. Absence of TNFa alone correlated with a significant reduction in serum TGs when compared to wild type controls. Moreover, this effect was consistent in PPARa deficient mice where PTKO mice TG levels were significantly reduced when compared to wild type mice. Early studies correlated increased TNFa levels with increased body mass index, fasting glucose levels and circulating triglycerides 12. Importantly, our data suggest that alterations in hepatic metabolic function and lipid accumulation does not, in the current model system, correlate with systemic alterations in lipid homeostasis, that rather, inflammatory factors including TNFa more likely influence these processes. One interesting feature of our data is that ablation of PPARa, a regulator of hepatic lipid peroxidation, caused a large increase in serum leptin levels when compared to wild type controls, a response which was completely abrogated by deletion of TNFaR1 in these same mice. Previous studies have linked PPARa to leptin production whereby activation with exogenous ligand, gemfibrozil, decreased leptin secretion in diet-induced obese rats 33. Leptin is well appreciated for its influence on energy balance and a multitude of secondary effects which alter a variety of physiological processes 29. Specifically and intriguingly, leptin has also been associated with the induction of pro-inflammatory cytokines by a variety of cells including macrophages. The initiator of the inflammatory response, particularly in this relatively simple model, is proposed to be lipid accumulation itself however secondary factors including leptin may promote or amplify this inflammatory cascade. Likewise, local TNF production at sites of leptin production, likely the white adipose tissue, appear to initiate this inflammo-endocrine cascade. Further study is needed to define the importance of leptin production in this genetic-induced obesity model. ## Conclusion These data highlight the importance of TNF receptor signaling in PPARα-deficient mice to facilitate lipid accumulation. TNFα, as well as other downstream pro-inflammatory cytokines, are increased as a result of PPARα ablation in mice. This effect is also observed in diet-induced hepatic steatosis. Data presented here supports the hypothesis that TNFα through its TNF receptors is a critical factor in the development of fatty liver, suggesting that TNF and the pro-inflammatory response are not merely a consequence of lipid accumulation but a major driver of the changes in lipid metabolism leading to the accumulation of lipid in liver. These findings have important implications for the role of TNF and pro-inflammatory cytokines in diet-induced fatty liver disease, not just in genetic models of steatohepatitis. Therapeutic or perhaps nutritional strategies to reduce the pro-inflammatory response represent potential early interventions for non-alcoholic fatty liver disease. ## References 1. Chiang DJ, Pritchard MT, Nagy LE. Obesity, diabetes mellitus, and liver fibrosis. Am J Physiol Gastrointest Liver Physiol (2011) 300 697. DOI: 10.1152/ajpgi.00426.2010 2. Tilg H, Adolph TE, Dudek M, Knolle P. Non-alcoholic fatty liver disease: The interplay between metabolism, microbes and immunity. Nat Metab (2021) 3 1596-1607. DOI: 10.1038/s42255-021-00501-9 3. Mitra S, De A, Chowdhury A. Epidemiology of non-alcoholic and alcoholic fatty liver diseases. Translational gastroenterology and hepatology (2020) 5 16. DOI: 10.21037/tgh.2019.09.08 4. Maurice J, Manousou P. Non-alcoholic fatty liver disease. Clin Med (Lond) (2018) 18 245-250. DOI: 10.7861/clinmedicine.18-3-245 5. Rivera CA, Adegboyega P, van Rooijen N, Tagalicud A, Allman M, Wallace M. Toll-like receptor-4 signaling and kupffer cells play pivotal roles in the pathogenesis of non-alcoholic steatohepatitis. J Hepatol (2007) 47 571-579. PMID: 17644211 6. Miyoshi H, Moriya K, Tsutsumi T. Pathogenesis of lipid metabolism disorder in hepatitis C: Polyunsaturated fatty acids counteract lipid alterations induced by the core protein. J Hepatol (2011) 54 432-438. DOI: 10.1016/j.jhep.2010.07.039 7. Perez-Carreras M, Del Hoyo P, Martin MA. Defective hepatic mitochondrial respiratory chain in patients with nonalcoholic steatohepatitis. Hepatology (2003) 38 999-1007. DOI: 10.1053/jhep.2003.50398 8. Miura K, Yang L, van Rooijen N, Ohnishi H, Seki E. Hepatic recruitment of macrophages promotes nonalcoholic steatohepatitis through CCR2. Am J Physiol Gastrointest Liver Physiol (2012) 302 1310. DOI: 10.1152/ajpgi.00365.2011 9. Liu TF, Brown CM, El Gazzar M. Fueling the flame: Bioenergy couples metabolism and inflammation. J Leukoc Biol (2012) 92 499-507. DOI: 10.1189/jlb.0212078 10. Chen Z, Tian R, She Z, Cai J, Li H. Role of oxidative stress in the pathogenesis of nonalcoholic fatty liver disease. Free Radic Biol Med (2020) 152 116-141. PMID: 32156524 11. Rada P, Gonzalez-Rodriguez A, Garcia-Monzon C, Valverde AM. Understanding lipotoxicity in NAFLD pathogenesis: Is CD36 a key driver?. Cell Death Dis (2020) 11. DOI: 10.1038/s41419-020-03003-w 12. Ventre J, Doebber T, Wu M. Targeted disruption of the tumor necrosis factor-alpha gene: Metabolic consequences in obese and nonobese mice. Diabetes (1997) 46 1526-1531. DOI: 10.2337/diab.46.9.1526 13. Spencer NY, Zhou W, Li Q. Hepatocytes produce TNF-alpha following hypoxia-reoxygenation and liver ischemia-reperfusion in a NADPH oxidase- and c-src-dependent manner. Am J Physiol Gastrointest Liver Physiol (2013) 305 84. DOI: 10.1152/ajpgi.00430.2012 14. Holbrook J, Lara-Reyna S, Jarosz-Griffiths H, McDermott M. Tumour necrosis factor signalling in health and disease. F1000Res (2019) 8. DOI: 10.12688/f1000research.17023.1 15. Yin M, Wheeler MD, Kono H. Essential role of tumor necrosis factor alpha in alcohol-induced liver injury in mice. Gastroenterology (1999) 117 942-952. PMID: 10500078 16. Beutler B, Greenwald D, Hulmes JD. Identity of tumour necrosis factor and the macrophage-secreted factor cachectin. Nature (2014) 316 552-554. DOI: 10.1038/316552a0 17. Davizon-Castillo P, McMahon B, Aguila S. TNF-alpha-driven inflammation and mitochondrial dysfunction define the platelet hyperreactivity of aging. Blood (2019) 134 727-740. DOI: 10.1182/blood.2019000200 18. Russell AE, Doll DN, Sarkar SN, Simpkins JW. TNF-alpha and beyond: Rapid mitochondrial dysfunction mediates TNF-alpha-induced neurotoxicity. J Clin Cell Immunol (2016) 7. DOI: 10.4172/2155-9899.1000467 19. Stadler J, Bentz BG, Harbrecht BG. Tumor necrosis factor alpha inhibits hepatocyte mitochondrial respiration. Ann Surg (1992) 216 539-546. DOI: 10.1097/00000658-199211000-00003 20. Kastl L, Sauer SW, Ruppert T. TNF-alpha mediates mitochondrial uncoupling and enhances ROS-dependent cell migration via NF-kappaB activation in liver cells. FEBS Lett (2014) 588 175-183. DOI: 10.1016/j.febslet.2013.11.033 21. Kim MS, Sweeney TR, Shigenaga JK. Tumor necrosis factor and interleukin 1 decrease RXRalpha, PPARalpha, PPARgamma, LXRalpha, and the coactivators SRC-1, PGC-1alpha, and PGC-1beta in liver cells. Metabolism (2007) 56 267-279. PMID: 17224343 22. Uysal KT, Wiesbrock SM, Marino MW, Hotamisligil GS. Protection from obesity-induced insulin resistance in mice lacking TNF-alpha function. Nature (1997) 389 610-614. DOI: 10.1038/39335 23. Wang Y, Nakajima T, Gonzalez FJ, Tanaka N. PPARs as metabolic regulators in the liver: Lessons from liver-specific PPAR-null mice. International Journal of Molecular Sciences (2020) 21. DOI: 10.3390/ijms21062061 24. Dreyer C, Keller H, Mahfoudi A, Laudet V, Krey G, Wahli W. Positive regulation of the peroxisomal beta-oxidation pathway by fatty acids through activation of peroxisome proliferator-activated receptors (PPAR). Biol Cell (1993) 77 67-76. DOI: 10.1016/s0248-4900(05)80176-5 25. Neschen S, Morino K, Hammond LE. Prevention of hepatic steatosis and hepatic insulin resistance in mitochondrial acyl-CoA:Glycerol-sn-3-phosphate acyltransferase 1 knockout mice. Cell Metab (2005) 2 55-65. PMID: 16054099 26. Perpinan E, Perez-Del-Pulgar S, Londono MC. Cirrhosis hampers early and rapid normalization of natural killer cell phenotype and function in hepatitis C patients undergoing interferon-free therapy. Front Immunol (2020) 11 129. DOI: 10.3389/fimmu.2020.00129 27. Cai D, Yuan M, Frantz DF. Local and systemic insulin resistance resulting from hepatic activation of IKK-beta and NF-kappaB. Nat Med (2005) 11 183-190. DOI: 10.1038/nm1166 28. Arkan MC, Hevener AL, Greten FR. IKK-beta links inflammation to obesity-induced insulin resistance. Nat Med (2005) 11 191-198. PMID: 15685170 29. Hong GQ, Cai D, Gong JP, Lai X. Innate immune cells and their interaction with T cells in hepatocellular carcinoma. Oncol Lett (2021) 21 57. DOI: 10.3892/ol.2020.12319 30. Lee JY, Plakidas A, Lee WH. Differential modulation of toll-like receptors by fatty acids: Preferential inhibition by n-3 polyunsaturated fatty acids. J Lipid Res (2003) 44 479-486. DOI: 10.1194/jlr.M200361-JLR200 31. dela Peña A, Leclercq I, Field J, George J, Jones B, Farrell G. NF-κB activation, rather than TNF, mediates hepatic inflammation in a murine dietary model of steatohepatitis. Gastroenterology (2005) 129 1663-1674. DOI: 10.1053/j.gastro.2005.09.004 32. Li Z, Yang S, Lin H. Probiotics and antibodies to TNF inhibit inflammatory activity and improve nonalcoholic fatty liver disease. Hepatology (2003) 37 343-350. DOI: 10.1053/jhep.2003.50048 33. Deng QG, She H, Cheng JH. Steatohepatitis induced by intragastric overfeeding in mice. Hepatology (2005) 42 905-914. DOI: 10.1002/hep.20877 34. Garcia-Ruiz I, Rodriguez-Juan C, Diaz-Sanjuan T. Uric acid and anti-TNF antibody improve mitochondrial dysfunction in ob/ob mice. Hepatology (2006) 44 581-591. DOI: 10.1002/hep.21313 35. Memon RA, Grunfeld C, Feingold KR. TNF-alpha is not the cause of fatty liver disease in obese diabetic mice. Nat Med (2001) 7 2-3. DOI: 10.1038/83316 36. Czaja MJ. JNK regulation of hepatic manifestations of the metabolic syndrome. Trends Endocrinol Metab (2010) 21 707-713. DOI: 10.1016/j.tem.2010.08.010 37. Seki E, Brenner DA, Karin M. A liver full of JNK: Signaling in regulation of cell function and disease pathogenesis, and clinical approaches. Gastroenterology (2012) 143 307-320. DOI: 10.1053/j.gastro.2012.06.004 38. Abdelmegeed MA, Yoo SH, Henderson LE, Gonzalez FJ, Woodcroft KJ, Song BJ. PPARalpha expression protects male mice from high fat-induced nonalcoholic fatty liver. J Nutr (2011) 141 603-610. DOI: 10.3945/jn.110.135210
--- title: 'Chronic kidney disease - determinants of progression and cardiovascular risk. PROGREDIR cohort study: design and methods' authors: - Maria Alice Muniz Domingos - Alessandra Carvalho Goulart - Paulo Andrade Lotufo - Isabela Judith Martins Benseñor - Silvia Maria de Oliveira Titan journal: São Paulo Medical Journal year: 2017 pmcid: PMC9977340 doi: 10.1590/1516-3180.2016.0272261116 license: CC BY 4.0 --- # Chronic kidney disease - determinants of progression and cardiovascular risk. PROGREDIR cohort study: design and methods ## ABSTRACT ### CONTEXT AND OBJECTIVE: Chronic kidney disease (CKD) has become an important public health issue. The socioeconomic burden of renal replacement therapy (RRT) is very high, as is CKD-related cardiovascular mortality and morbidity. Preventive and therapeutic measures only have modest impact and more research is needed. Few cohort studies have been conducted on populations with CKD. Our aim was to establish a cohort that would include more advanced forms of CKD (stages 3 and 4). Data collection was focused on renal and cardiovascular parameters. ### DESIGN AND SETTING: Prospective cohort study; São Paulo, Brazil. ### METHODS: Recruitment took place in Hospital das Clínicas, São Paulo, from March 2012 to December 2013. Data relating to medical history, food-frequency questionnaire, anthropometry, laboratory work-up, calcium score, echocardiography, carotid intimal-medial thickness, pulse-wave velocity, retinography and heart rate variability were collected. A biobank including serum, plasma, post-oral glucose tolerance test serum and plasma, urine (morning and 24-hour urine) and DNA was established. ### RESULTS: 454 participants ($60\%$ men and $50\%$ diabetics) of mean age 68 years were enrolled. Their mean estimated glomerular filtration rate-CKD Epidemiology Collaboration was 38 ml/min/1.73 m2. Follow-up is ongoing and the main outcomes are the start of RRT, cardiovascular events and death. ### CONCLUSIONS: The PROGREDIR cohort is a promising prospective study that will allow better understanding of CKD determinants and validation of candidate biomarkers for the risks of CKD progression and mortality. ## CONTEXTO E OBJETIVO: A doença renal crônica (DRC) tornou-se um problema de saúde pública. A carga socioeconômica da terapia renal substitutiva é muito elevada, assim como a morbimortalidade cardiovascular associada à DRC. Medidas terapêuticas e preventivas têm impacto parcial e novos estudos são necessários. Há poucos estudos de coorte em populações com DRC. Nosso objetivo foi criar uma coorte que contemplasse formas mais avançadas de DRC (estágios 3 e 4). A coleta de dados foi centrada em parâmetros renais e cardiovasculares. ## TIPO DE ESTUDO E LOCAL: Estudo de coorte prospectivo; São Paulo, Brasil. ## MÉTODOS: O recrutamento ocorreu entre março de 2012 e dezembro de 2013, no Hospital das Clínicas, em São Paulo. Foram coletados dados de história médica, questionário de frequência alimentar, antropometria, exames laboratoriais, escore de cálcio, ecocardiografia, espessura de camada médio-intimal de carótidas, velocidade de onda de pulso, retinografia e variabilidade de frequência cardíaca. Um biobanco incluindo soro, plasma, soro e plasma pós-teste oral de tolerância à glicose, urina (manhã e 24 horas) e DNA foi estabelecido. ## RESULTADOS: 454 participantes ($60\%$ homens e $50\%$ diabéticos) com idade média de 68 anos foram recrutados. A taxa média de filtração glomerular estimada-Colaboração da Epidemiologia para DRC foi de 38,4 ml/min/1,73 m2. O seguimento está em andamento e os desfechos principais são: início de terapia renal substitutiva, eventos cardiovasculares e óbito. ## CONCLUSÃO: A coorte PROGREDIR é um estudo prospectivo promissor que permitirá melhor compreensão dos determinantes de DRC e a validação de biomarcadores candidatos para o risco de progressão de DRC e de mortalidade. ## INTRODUCTION Chronic kidney disease (CKD) has become an important public health issue worldwide. Increasing prevalence of obesity and diabetes mellitus and today’s high life expectancy, particularly among patients with atherosclerosis, are all contributory factors. In addition, CKD progression is still a major challenge, with few new specific therapeutic measures available. The socioeconomic burden on individuals who need renal replacement therapy (RRT) is very high and comes together with CKD-related high cardiovascular mortality and morbidity, with incidence that may in fact even exceed the figures for RRT.1,2,3,4,5,6,7 In the United States, according to the Annual Report of the United States Renal Data System (USRDS),8 the prevalence of CKD stages 1-4 was around $14\%$ in the general population and the incidence of end-stage renal disease (ESRD) was 353 cases per million/year in 2012. The prevalence of cardiovascular disease reached $69.8\%$ among CKD patients versus $34.8\%$ among individuals without CKD and the adjusted mortality rates for CKD patients was 76 deaths per 1000 patients, compared with 52 deaths per 1,000 individuals without CKD in 2012. Medicare expenses relating to CKD reach US$ 1700, 3500 and 12,700 per person-year for CKD patients with stages 2, 3 and 4, respectively.9 Overall, CKD accounts for $6.7\%$ of total Medicare costs.9 In Brazil, there were 100,397 patients on dialysis at the end of 2013, with incidence of 170 cases per million/year and an estimated mortality rate of $17.9\%$ per year.10 In 2013, 5,433 kidney transplantations were performed in Brazil, mostly using public resources. Preventive measures are highly necessary, and the search for new biomarkers and new therapeutic strategies is intense. While several studies on general populations and cardiovascular cohorts have yielded important contributions towards CKD knowledge, more specific cohorts focusing on CKD progression instead of CKD incidence are necessary within nephrology. In response to this need, several countries like the United States (CRIC study), Germany (GCKD), Canada (CanPREDDICT), Japan, Australia and Uruguay, among others, have ongoing CKD cohort studies.11 Along the same lines, the PROGREDIR cohort study was designed to enable better understanding of the determinants of CKD progression and CKD-related mortality, with particular emphasis on mineral metabolism as a cardiovascular risk factor. The cohort comprises people with CKD stages 3 and 4 in São Paulo, Brazil. The cohort was established and baseline data were collected in 2012-2013. Prospective data on hard outcomes such as the start of renal replacement therapy, cardiovascular events and death are currently being gathered. The PROGREDIR cohort is funded by the Research Support Foundation of the State of São Paulo (Fundação de Amparo à Pesquisa do Estado de São Paulo, FAPESP; 2011-17341-0), São Paulo, Brazil. ## OBJECTIVE The aim of this study was to establish a CKD cohort that would include participants with more advanced forms of this disease (CKD stages 3 and 4), with data collection focused on renal and cardiovascular parameters. ## Study population and recruitment Patients attending the outpatient service of Hospital das Clínicas, São Paulo, a public university facility providing quaternary-level care for patients with chronic diseases, were invited to participate in this study. Initially, from the outpatient records, all patients aged ≥ 30 years and at least two measurements of creatinine (with a minimum interval of 3 months) ≥ 1.6 mg/dl for men and ≥ 1.4 mg/dl for women were considered potential candidates. Patients attending oncology, psychiatry, human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS), viral hepatitis and glomerulonephritis services were excluded. The remaining candidates were then contacted by phone and were invited to participate if they did not meet any exclusion criteria. The exclusion criteria checked by the interviewer were: hospitalization within the last six months, acute myocardial infarction within the last six months, autoimmune diseases, pregnancy, psychiatric diseases, ongoing chemotherapy or immunosuppressive therapy, ongoing RRT, glomerulonephritis, HIV/AIDS infection, hepatitis B or C and any organ transplantation. Recruitment took place between March 2012 and December 2013, and 454 participants were enrolled. The study was approved by two local ethics committees and written informed consent was obtained from all participants. ## Sample size estimation The sample size was calculated using an estimate of the annual incidence of end-stage renal disease (ESRD) of $2\%$ and an annual rate of cardiovascular events of 2-$3.5\%$ among diabetic nephropathy patients.12 By assuming a difference in event rate incidence of $3\%$ between exposed and non-exposed subjects, a sample size of 500 was estimated, with an alpha error of 0.05 and a power of $80\%$. ## Baseline examination and data collection The baseline assessment lasted approximately six hours and was performed on a single-day visit to our study center. Data collection included all the variables depicted in Table 1. Sex and self-declared race were also registered. Anthropometry was performed first, with the participants wearing light clothes, following standard techniques.13 Blood pressure (BP) was measured using a validated oscillometric device (Omron HEM 705CPINT). Three measurements were made at one-minute intervals. The mean of the last two BP measurements was used as the definition for high blood pressure. Overnight fasting blood samples and 24-hour and spot urine samples were collected. A standard 75-g oral glucose tolerance test was administered to all participants without known diabetes. Urine and blood aliquots were prepared and stored at -180 °C in nitrogen. DNA extraction was performed and the material was stored at -80 °C. Baseline laboratory measurements were made using conventional techniques (Table 2). Table 1:Baseline assessments in PROGREDIR cohort study Table 2:Baseline laboratory tests Interviews were conducted by trained personal under strict quality control. Data on medical history, socioeconomic variables, family history, medication use, physical activity, smoking and alcohol consumption were obtained. A food-frequency questionnaire was also administered. The food list was defined on the basis of the dietary intake of the Brazilian population and the reproducibility and validity of the questionnaire has been measured elsewhere.14 Conventional 12-lead electrocardiograms (ECG) were performed using a digital device (Atria 6100, Burdick, Cardiac Science Corporation, USA) with automated readings of heart rate; P wave, QRS complex and T wave duration, amplitude and axis; and QT, QTc and QT dispersion. All precordial electrodes were positioned after identifying the location for the V4 electrode with a square. The Electrocardiogram Reading Center (ERC) at the Heart Institute of the University of São Paulo (INCOR) provided all ECG readings. For heart rate variability determinations, a 10-minute continuous ECG was obtained from a single lead (usually D2) using a digital electrocardiograph (Micromed, Brazil) at a frequency of 250 Hz, with subjects in the supine position. Computer software (WinCardio) was used to generate time series of RR intervals that were sent to the central cardiovascular physiology laboratory (IC-ES). All readings were made through computer software that eliminated artifacts and selected RR intervals lasting 0.5 to 2.0 seconds. Temporal and spectral analyses of heart rate variability (HRV) were then performed using an autoregressive model to identify very low-frequency (VLF, 0 to 0.04 Hz), low-frequency (LF, 0.04 to 0.1 Hz) and high-frequency spectral bands (HF, 0.1 to 0.4 Hz). Transthoracic echocardiography was performed on all participants using a device (Aplio XG; Toshiba Corporation, Tokyo, Japan) with a 2.5 MHz sector transducer. All examinations were performed by the same echocardiographer. The readings consisted of qualitative analysis of echocardiographic findings and measurements of quantitative parameters such as: left ventricular (LV) geometry and size, left atrial size, LV systolic and diastolic function, segmental LV dysfunction, valvular heart disease and pericardial appearance. Cardiac mass was calculated using the Devereaux formula.15 Measurement of the anterior-posterior diameter of the right lobe of the liver was performed by means of ultrasound for quantitative assessment of nonalcoholic fatty liver disease (NAFLD). Liver images were obtained using standard equipment (Toshiba SSA-770A Aplio, Japan) and a broadband convex transducer (PVT-375BT) with a central frequency of 3.5 MHz (2.5-5.5 MHz).16 The carotid-femoral pulse-wave velocity (PWV) was measured using a validated automated device (Complior, Artech Medicale, France), with the subject in the supine position in a temperature-controlled room (20-24 °C). First, BP was measured in the right arm with the subject in the supine position using an oscillometric device (HRM Onrom 705 CP). The distance from the sternal furcula to the right femoral pulse was determined using a measuring tape regardless of abdominal curvature. Pulse sensors were positioned in the right carotid and femoral arteries so that pulse waves were recorded and viewed on a computer screen. Computer software that could adequately detect and record pulse waves was used. PWV was calculated by dividing the distance from the sterna furcula to the femoral pulse by the difference between the rise delays of the carotid and femoral pulses. A subject’s PWV was the arithmetic average of readings obtained in ten consecutive cardiac cycles at a regular heart rate. Carotid intimal-media thickness (IMT) was assessed in all patients in a standardized manner using a device (Aplio XG, Toshiba) with a 7.5 MHz linear transducer. The technique used for IMT measurement was as previously published.17 IMT was measured in the outer wall of a predefined carotid segment of 1 cm in length from 1 cm below the carotid bifurcation, during three cardiac cycles. We considered the images acquired to be valid if they clearly showed three reference points on both sides: anatomical guides for the common carotid arteries;interfaces between the lumen and the far wall of the vessel; andinterfaces between the media and adventitia layers of the far wall of the vessel. We used the MIA software to standardize the readings and interpret the carotid scans as previously described. IMT was then defined as the mean of the right and left carotid measurements. To determine the coronary artery calcium score, the participants underwent non-contrast computed tomography. The scans were performed using a 64-slice detector computed tomography scanner (Philips Brilliance, Philips, Netherlands). After scout images had been produced, each patient also underwent an ECG-gated prospective calcium score examination with a tube potential of 120 kV and a tube current adjusted to body habitus. Images were reconstructed at 2.5 mm slice thickness using standard filtered back projection. The coronary artery calcium score was expressed in terms of Agatston units and the percentiles were evaluated in a blinded manner by an experienced cardiologist using semi-automated software (Calcium Scoring, Philips Workstation). Coronary calcium scores were not obtained for participants who reported that they had been fitted with coronary stents, since the stent material greatly overestimates the calcium scores. Retinography was performed using a nonmydriatic retinograph (CR-1, Canon, Japan) with a 10-megapixel digital camera (Canon EOS 40 D). The subjects underwent natural dilation of their pupils through resting in a darkened room for about four minutes, and for each eye two 45° fundus images were obtained: one centered on the optical disk and the second on the maculae. Our institution’s central retinography laboratory (IC-RS) developed standardized image acquisition and reading protocols, and DICOM images (approximately 30 MB) and JPEG images (approximately 3 MB) were acquired. The JPEG images were recorded on CD/DVD at the study sites and were mailed to the central retinography laboratory. ## Follow-up The participants are being contacted again annually, for telephone interviews that include questions on hospitalizations, need for RRT and self-rated health. The main clinical endpoints investigated are death, acute myocardial infarction, unstable angina pectoris, cardiac revascularization, heart failure, stroke and RRT. Any cardiovascular and renal clinical events that are reported are then investigated and classified in line with the study protocol, by a panel of physicians that has received training in accordance with international classification criteria.18 In the event of the participant’s death, information regarding this event is sought. Surveillance of clinical events is also conducted through state databases such as the Mortality Registry and the São Paulo State Registry of Dialysis and Transplantation. ## RESULTS Over the two-year recruitment period, 454 participants were enrolled. Table 3 shows the main clinical and laboratory parameters at baseline. The population recruited mainly had CKD in stages 3 and 4, with a mean estimated glomerular filtration rate-CKD Epidemiology Collaboration (eGFR-CKDEPI) of 38.4 (± 14.6) ml/min/1.73 m2. The albuminuria range was wide, with similar frequencies of normoalbuminuria ($35\%$), microalbuminuria ($31\%$) and macroalbuminuria ($34\%$). The participants’ median age was 67 years; $63\%$ were men; $60\%$ were current or past smokers; $45\%$ self-reported diabetes; and $32\%$ reported having had previous myocardial infarction. Coronary artery calcification scores were also high, with more than half of the cohort presenting an Agatston score above 100. Follow-up is ongoing. Up to the present date, i.e. over the first three years of follow-up, event rates have been high, with a 5-$7\%$ mortality rate per year and 2-$3\%$ incidence of ESRD and non-fatal cardiovascular events per year. With this event rate, from 2017 onwards, survival analysis will be started, focusing on biomarkers for mineral metabolism. Table 3:Baseline clinical and laboratory profile of 454 participants in the PROGREDIR cohortcalculated for participants without known diabetes. SD = standard deviation; IQR = interquartile range; RBP = retinol-binding protein; eGFR-CKDEPI = estimated glomerular filtration rate-Chronic Kidney Disease Epidemiology Collaboration; OGTT = oral glucose tolerance test; HOMA-IR = homeostasis model assessment as an index of insulin resistance; LDL = low-density lipoprotein; HDL = high-density lipoprotein. ## DISCUSSION The PROGREDIR cohort was designed specifically to address CKD progression among patients with moderate to advanced disease. Over a two-year recruitment period, we were able to enroll 454 participants, and thus nearly reached the estimated sample size. The baseline characteristics of these participants were in accordance with the profile expected from the inclusion and exclusion criteria: older age, predominance of men and high rates of diabetes and previous cardiovascular disease. In the PROGREDIR cohort, we avoided overrepresentation of glomerulonephritis and other specific kidney diseases such as those relating to HIV, hepatitis C and lupus. Transplantation patients (any organ) were also not included. This decision was mostly related to the fact that PROGREDIR was designed to be a cohort of general CKD cases and not to address mechanisms relating to specific systemic or primary diseases. The eligible participants were originally from a quaternary-level hospital, which might have yielded an excessive number of glomerulonephritis cases if exclusion criteria had not been applied. Other CKD cohort studies have applied similar inclusion and exclusion criteria and have ended up with recruited populations compatible with the profile observed in PROGREDIR.19,20,21 One important accomplishment was to have nearly equal representation of normoalbuminuria, microalbuminuria and macroalbuminuria subpopulations in the baseline profile of the cohort. The prevalence of and interest in normoalbuminuric CKD is increasing,22,23,24 since it is now known that 30-$45\%$ of diabetic patients may in fact present CKD and normoalbuminuria. It is currently of interest not only to understand the determinants of CKD progression in the normoalbuminuric CKD population, but also to compare the performance of traditional and new risk factors in normoalbuminuric and albuminuric populations, in order to test whether the results can be generalized to a broad spectrum of diseases. Baseline data were collected in this study in accordance with the study design, covering traditional cardiovascular risk factors and biomarkers for CKD. Surrogate measurements of atherosclerosis and hypertension such as coronary calcium score, cardiac hypertrophy, IMT, PWV and retinography were made, and these will allow understanding and stratification of baseline CKD among the participants. The biobank is wide-ranging and kept under strict quality control, thus providing a source for reliable future measurements. Follow-up is ongoing and a high event rate is being observed. Follow-up data collection is being centered on three major clinical events: death, non-fatal cardiovascular events and starting of RRT. These events are of particular importance, since CKD is known to be a very important cardiovascular risk factor that makes a significant contribution to high rates of morbidity and mortality.2 Focusing data* collection only on renal events would lead to selection bias, because a significant proportion of the participants might experience cardiovascular events prior to renal events. Thus, to fully address the impact of CKD biomarkers and measurements, it is very important to account for their impact both on renal events such as mortality and on fatal and non-fatal cardiovascular events. Now that the cohort has been established, the PROGREDIR study can be used for research investigation in two major ways. First, it can be used to test the performance of candidate biomarkers for CKD progression. The current need to promote discovery and validation of biomarkers in CKD is highlighted by the recent launch of a CKD Biomarkers Consortium (BioCon)25 by the National Institute of Diabetes and Digestive and Kidney Diseases in the United States. Similar approaches are being used by European countries.26 Secondly, the cohort can be used to test high throughput technologies, which are an innovative approach that may provide new insights on the mechanisms and pathways of complex diseases, as well as enabling identification of novel biomarkers for diseases. As a first step, untargeted metabolomic assessments are currently being performed on baseline serum and the data thus obtained will be analyzed in relation to renal function and clinical events. Additionally, to contribute towards improvement of scientific knowledge on CKD, the PROGREDIR study will also serve the purpose of being a national data source in which biomarkers can be replicated and validated. Racial factors are known to have an important effect on the risk of diseases, and this has recently been very well illustrated by the discovery of the higher risk attributable to the APOL1 gene in the African-American population.27 In this regard, it is very important that national datasets should be available, so that the performance of candidate biomarkers can be tested on the Brazilian population, which is known to be highly admixed. ## CONCLUSION In conclusion, the PROGREDIR cohort recruitment and baseline data collection were successfully implemented. In addition to being a national dataset, the PROGREDIR cohort provides promising prospective study material that will allow better understanding of CKD determinants and validation of candidate biomarkers for CKD progression and mortality risk. ## References 1. Anavekar NS, McMurray JJ, Velazquez EJ. Relation between renal dysfunction and cardiovascular outcomes after myocardial infarction. N Engl J Med (2004) 351 1285-1295. PMID: 15385655 2. Astor BC, Hallan SI, Miller ER, Yeung E, Coresh J. Glomerular filtration rate, albuminuria, and risk of cardiovascular and all-cause mortality in the US population. Am J Epidemiol (2008) 167 1226-1234. PMID: 18385206 3. Rahman M, Pressel S, Davis BR. Cardiovascular outcomes in high-risk hypertensive patients stratified by baseline glomerular filtration rate. Ann Intern Med (2006) 144 172-180. PMID: 16461961 4. Wen CP, Cheng TY, Tsai MK. All-cause mortality attributable to chronic kidney disease: a prospective cohort study based on 462 293 adults in Taiwan. Lancet (2008) 371 2173-2182. PMID: 18586172 5. Di Angelantonio E, Chowdhury R, Sarwar N. Chronic kidney disease and risk of major cardiovascular disease and non-vascular mortality: prospective population based cohort study. BMJ (2010) 341 c4986-c4986. PMID: 20884698 6. Meisinger C, Döring A, Löwel H. Chronic kidney disease and risk of incident myocardial infarction and all-cause and cardiovascular disease mortality in middle-aged men and women from the general population. Eur Heart J (2006) 27 1245-1250. PMID: 16611670 7. Astor BC, Hallan SI, Miller ER, Yeung E, Coresh J. Glomerular filtration rate, albuminuria, and risk of cardiovascular and all-cause mortality in the US population. Am J Epidemiol (2008) 167 1226-1234. PMID: 18385206 8. 8. United States Renal Data System, 2014 Annual Data Report: Epidemiology of Kidney Disease in the United States Bethesda, MD National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases 2014. United States Renal Data System, 2014 Annual Data Report: Epidemiology of Kidney Disease in the United States (2014) 9. Foley RN, Collins AJ. End-stage renal disease in the United States an update from the United States Renal Data System. J Am Soc Nephrol (2007) 18 2644-2648. PMID: 17656472 10. 10 Censo da Sociedade Brasileira de Nefrologia Censo de diálise 2013 Available from: http://arquivos.sbn.org.br/pdf/censo_2013-14-05.pdf Accessed in 2016 (Dec 28). Censo de diálise (2013) 11. Dienemann T, Fujii N, Orlandi P. International Network of Chronic Kidney Disease cohort studies (iNET-CKD): a global network of chronic kidney disease cohorts. BMC Nephrol (2016) 17 121-121. PMID: 27590182 12. Adler AI, Stevens RJ, Manley SE. Development and progression of nephropathy in type 2 diabetes: the United Kingdom Prospective Diabetes Study (UKPDS 64). Kidney Int (2003) 63 225-232. PMID: 12472787 13. Lohman TG, Roche AF, Martorell R. Anthropometric standardization reference manual (1988) 14. Mannato LW, Pereira TS, Velasquez-Melendez G. Comparison of a short version of the Food Frequency Questionnaire with its long version--a cross-sectional analysis in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Sao Paulo Med J (2015) 133 414-420. PMID: 26648430 15. Devereux RB, Alonso DR, Lutas EM. Echocardiographic assessment of left ventricular hypertrophy: comparison to necropsy findings. Am J Cardiol (1986) 57 450-458. PMID: 2936235 16. Goulart AC, Oliveira IR, Alencar AP. Diagnostic accuracy of a noninvasive hepatic ultrasound score for non-alcoholic fatty liver disease (NAFLD) in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Sao Paulo Med J (2015) 133 115-124. PMID: 26018881 17. Santos IS, Bittencourt MS, Oliveira IR. Carotid intima-media thickness value distributions in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Atherosclerosis (2014) 237 227-235. PMID: 25244507 18. Barreto SM, Ladeira RM, Bastos MSCBO. Estratégias de identificação, investigação e classificação de desfechos incidentes no ELSA-Brasil [ELSA-Brasil strategies for outcome identification, investigation and ascertainment]. Rev Saúde Pública (2013) 47 79-86. PMID: 24346724 19. Feldman HI, Appel LJ, Chertow GM. The Chronic Renal Insufficiency Cohort (CRIC) Study: Design and Methods. J Am Soc Nephrol (2003) 14 S148-S153. PMID: 12819321 20. Levin A, Rigatto C, Brendan B. Cohort profile Canadian study of prediction of death, dialysis and interim cardiovascular events (CanPREDDICT). BMC Nephrol (2013) 14 121-121. PMID: 23758910 21. Eckardt KU, Bärthlein B, Baid-Agrawal S. The German Chronic Kidney Disease (GCKD) study: design and methods. Nephrol Dial Transplant (2012) 27 1454-1460. PMID: 21862458 22. Kramer HJ, Nguyen QD, Curhan G, Hsu CY. Renal insufficiency in the absence of albuminuria and retinopathy among adults with type 2 diabetes mellitus. JAMA (2003) 289 3273-3277. PMID: 12824208 23. Afkarian M, Zelnick LR, Hall YN. Clinical Manifestations of Kidney Disease among US Adults with Diabetes, 1988-2014. JAMA (2016) 316 602-610. PMID: 27532915 24. Garg AX, Kiberd BA, Clark WF, Haynes RB, Clase CM. Albuminuria and renal insufficiency prevalence guides population screening: results from the NHANES III. Kidney Int (2002) 61 2165-2175. PMID: 12028457 25. Hsu CY, Ballard S, Batlle D. Cross-Disciplinary Biomarkers Research: Lessons Learned by the CKD Biomarkers Consortium. Clin J Am Soc Nephrol (2015) 10 894-902. PMID: 25739849 26. Brück K, Stel VS, Gambaro G. CKD Prevalence Varies across the European General Population. J Am Soc Nephrol (2016) 27 2135-2147. PMID: 26701975 27. Genovese G, Friedman DJ, Pollak MR. APOL1 variants and kidney disease in people of recent African ancestry. Nat Rev Nephrol (2013) 9 240-244. PMID: 23438974
--- title: Promotive and risk factors for children’s mental health—Finnish municipal policymakers’ and leading officeholders’ views authors: - Outi Savolainen - Marjorita Sormunen - Hannele Turunen journal: Health Promotion International year: 2023 pmcid: PMC9977352 doi: 10.1093/heapro/daac111 license: CC BY 4.0 --- # Promotive and risk factors for children’s mental health—Finnish municipal policymakers’ and leading officeholders’ views ## Abstract Findings on children’s mental health promotion at the policy level are scarce, and the perceptions of the municipal administration on factors affecting children’s mental health have not been reported. This study describes the perspectives of policymakers and leading officeholders on promotive and risk factors for children’s mental health in a socioecological context. The perspectives of Finnish policymakers ($$n = 15$$) and officeholders ($$n = 10$$) in municipalities were examined using semi-structured interviews. The data were analyzed using inductive content analysis and were categorized according to the five levels of a socioecological model of health promotion: public policy, community, organizational, interpersonal and individual levels. The public policy level emerged strongly in the findings, specifically strategic planning and implementation challenges related to the promotion of children’s mental health in the municipality and state administration. At the community level, environmental factors promoting children’s mental health as well as risk factors were described. The organizational level consisted of support, requirements and development needs in children’s services. The importance of family and close networks at the interpersonal level, as well as the individual basis of mental health, were also evident. The integration and better collaboration of child and family services, the use of child rights impact assessment in political decision-making, and financial support from the state could contribute to improving strategic planning to support children’s mental health at the municipal level. ## INTRODUCTION The basis of mental health (MH) is formed during childhood and adolescence. At best, the childhood environment supports MH and provides an opportunity for the positive development of mental resources (Vorma et al., 2020). The goals of MH promotion are to strengthen mental well-being by creating supportive living conditions and environments and to reduce factors that are harmful to MH (Klemera et al., 2017; Min et al., 2017). In addition to genetic factors (Taylor et al., 2017), there are a multiplicity of risk factors that can predispose children to negative outcomes—for example, pre- and peri-natal factors, such as young parental age (Oerlemans et al., 2016) and maternal postnatal depression (Netsi et al., 2018); individual factors, such as female gender (Merikukka et al., 2018); family- and parenting-related factors, such as parental alcohol abuse (O’Hara et al., 2019; Raitasalo et al., 2019; Ukkola et al., 2020) and environmental factors, such as seasonal changes (Wiens et al., 2016). However, there are also promotive factors that can buffer these risks, such as healthy lifestyles (Bertoni, 2015; Mikkelsen et al., 2017; Tamana et al., 2019), positive relationships (Auersperg et al., 2019), pupil participation in health promotion measures in the school context (Griebler et al., 2017) and proximity to nature (Wiens et al., 2016). In Finland, as in many other European countries [e.g. (Pereira et al., 2022)], municipalities have the overall responsibility for promoting the MH of children. Ways to promote MH include the prevention of bullying and discrimination, reducing poverty in families with children, making the strengthening of MH part of the culture of early childhood education and care (ECEC) and schools, and making work life more family friendly. The municipality needs to ensure that the conditions for well-being are taken into account in the activities of all administrative sectors and to organize appropriate social and health services (Health Care Act of Finland, $\frac{1326}{2010}$; Local Government Act of Finland, $\frac{410}{2015}$). The Finnish National Child Strategy (Parliamentary National Child Strategy Committee, 2021), published in February 2021, emphasizes services that promote MH and treat MH disorders. In addition, the realization of the rights of the child requires the reform of services used by children (United Nations, 1989) and emphasizes the roles and rights of the child in administration and services. These efforts include considering and caring for the best interests of the child and supporting parents in their child-rearing tasks, including child protection and childcare services. Moreover, professionals in different fields need appropriate education, training and professional assistance to be able to meet the needs of parents and families (Sormunen et al., 2011; Noonan et al., 2019). Collaboration between different sectors is further emphasized in the healing and rebuilding needed to address the harmful consequences of the current COVID-19 pandemic (Guido et al., 2020; Thualagant et al., 2022), which has profoundly affected the lives of children and young people by increasing MH problems (Ministry of Social Affairs and Health, 2021; Wennberg et al., 2020). Thus, such actions are necessary to secure children’s rights. Despite the growing body of literature related to the factors affecting children’s MH and MH promotion in different settings, such as homes, schools, workplaces and communities, throughout their lives (Sharma et al., 2017), findings on factors affecting children’s MH and MH promotion at the municipal policy level are scarce (Jenkins et al., 2020). In general, there is a gap in the knowledge regarding the effects of political decision-makers and leading officeholders in municipal administrations on the well-being of children. The socioecological model (SEM) of health promotion (Stokols, 1996) is useful in explaining this complex phenomenon. The model covers both risk and promotive factors at five levels of the socioecological environment (i.e. individual, interpersonal, organizational, community and public policy) and is therefore appropriate for examining public health challenges too complex to be adequately understood and addressed from a single level, such as the MH problems of children. Consequently, the aim of the current study is to describe the perceptions of political decision-makers and leading officeholders on promotive and risk factors for children’s MH in a socioecological context. ## MATERIALS AND METHODS Finland is a Nordic welfare society characterized by a public sector that offers citizens welfare services and a social safety net. Like all Nordic countries, Finland is a parliamentary democracy, with legislation requiring a majority in parliament (The Nordic Council and the Nordic Council of Ministers, 2021). Municipal administration, in turn, is based on the self-government of municipalities. The political decision-making power of the municipality is exercised by the municipal council. In addition, the municipality is required to have a municipal board and an inspection board. The council may appoint several committees, such as urban environmental, culture and leisure, education, social services and health committees. The municipal board and committees implement the decisions made by the council (Local Government Act of Finland, $\frac{410}{2015}$). Municipalities have a dual management system, in which local authority is characterized by division into political and professional management. Political management consists of decision-makers elected to municipal councils, boards and committees. Professional management, in turn, consists of leading officeholders who act as professional representatives of the administration and participate extensively in the various stages of the decision-making process in their area of administration. The officeholder organization manages the operational activities of the municipality (Constitution of Finland, $\frac{731}{1999}$; Local Government Act of Finland, $\frac{410}{2015}$). In this paper, the term policymaker refers to political decision-makers from municipal decision-making bodies, the term officeholders refers to the directors and leaders of different service sectors, and the term municipal administration refers to a management system consisting of political and professional management. Using qualitative research methodology, open-ended semi-structured individual interviews were conducted in three municipalities in Finland. Purposive sampling was used to identify potential participants from sectors of growth and learning, well-being promotion, environment and social and health services, who would provide focused information on children’s MH promotion from the perspective of municipal administration (Curtis et al., 2000). Policymakers’ and leading officeholders’ contact information was obtained from the public websites of the municipalities and municipal administrative services. First, the prospective participants received an information letter describing the study by e-mail. Altogether, 20 policymakers and 12 officeholders were contacted in person to ask about their willingness to participate in the study. Subsequently, 15 policymakers and 10 leading officeholders were interviewed by the first author. In later interviews, there were high degrees of repetition of the answers, and the amount of new information decreased. In the last two interviews, the views of the participants did not yield any new information. Therefore, data saturation was achieved by a total of 25, interviews and no more participants were recruited. The interview guide (Table 1) was developed based on previous literature and the SEM of health promotion (Stokols, 1996). A pilot interview confirmed the feasibility of the interview guide. The interview topics were provided to the participants 1 week before their interviews to familiarize them. The interviews, which were conducted in Finnish by telephone between September and November 2019, lasted between 21 and 58 min and were recorded by the interviewer and transcribed by a professional transcription company; omissions in speech, word repetitions, missed syllables and single pronouns were omitted. Ethical approval for the study was obtained from the ethics committee of the university (statement $\frac{5}{2019}$, 17.4.2019). In addition, a research permit was obtained separately from each municipal administration, and written informed consent was obtained from all participants. Table 1: \| Topic \| Subtopic \| \| --- \| --- \| \| Background information \| Municipality (working/serving as a municipal politician) \| \| Background information \| Profession \| \| Background information \| Role (officeholder/policymaker) \| \| Background information \| Year of birth \| \| Background information \| Work experience/experience serving as a municipal politician (years) \| \| Mental health and mental health promotion of children in general \| Definition of mental health and mental health promotion \| \| Mental health and mental health promotion of children in general \| The role of the policymaker/officeholder in promoting the mental health of children \| \| Factors affecting the mental health of children at different levels of the socioecological environment \| Individual characteristics, development, diseases, etc., of the child (individual level) \| \| Factors affecting the mental health of children at different levels of the socioecological environment \| The child’s immediate environment, family, friends, etc. (interpersonal level) \| \| Factors affecting the mental health of children at different levels of the socioecological environment \| Early childhood education and care, basic education and primary health care (organizational level) \| \| Factors affecting the mental health of children at different levels of the socioecological environment \| Family support, inclusion, justice, physical environment, leisure activities, etc. (community level) \| \| Factors affecting the mental health of children at different levels of the socioecological environment \| Local decision-making and resources, national mental health promotion structures (public policy level) \| \| Mental health promotion of children at the municipality \| Early intervention in children’s mental health symptoms \| \| Mental health promotion of children at the municipality \| Children’s mental health services \| The data were analyzed using inductive content analysis (Elo and Kyngäs, 2008; Vaismoradi et al., 2013). The transcribed texts were read several times to obtain an overall impression, and the data were reviewed for their content. Meaning units were sentences or phases (Graneheim and Lundman, 2004), and they were chosen in line with the purpose of the study. Meaning units derived were condensed and coded for the identified categories according to the five levels of the socioecological environment (i.e. individual, interpersonal, organizational, community and public policy). In the analysis, the similarities and differences between the meaning units were compared, and categories and subcategories were created based on the comparison. An initial reading of the transcripts and preliminary coding were conducted in Finnish by the first author. The final analysis was validated by all authors. There was an ongoing dialog among authors during the final analysis process. By using Finnish in the analysis, the authors sought to ensure that the voices of the participants were accurately understood and represented. The English translation process was conducted during the preparation of this paper, and all authors participated in it. ## General information Participants ($$n = 25$$) consisted of policymakers ($$n = 15$$) and officeholders ($$n = 10$$), including both males ($$n = 13$$) and females ($$n = 12$$), aged 32–75 years. Leading officeholdersʼ work experience in their current position was 7.6 years, and municipal politicians had 9.1 years of experience. Policymakers were from municipal councils, municipal governments and committees. Officeholders were from different service sectors as part of the city’s service organization, including social and health services, growth and learning, well-being promotion and environment. Policymakers and officeholders described promotive and risk factors related to children’s MH at the public policy, community, organizational, interpersonal and individual levels of the socioecological environment. Five main categories emerged from the descriptions of participants: (i) strategic planning and implementation challenges related to the promotion of children’s MH in the municipality and state administration; (ii) the contradictory roles of the environmental, building, culture and sport sectors in children’s MH promotion; (iii) support, requirements and development needs in children’s services; (iv) the importance of family and close networks and (v) the individual basis of MH. The main categories with subcategories are presented in Tables 2 and 3. ## Promotive and risk factors at the public policy and community levels Political decision-makers and officeholders reported various factors that were associated with strategic planning and implementation challenges related to the promotion of children’s MH in the municipality and state administration and referred to the public policy level (Table 2). Municipal administration played a personal, political and practical role in children’s MH promotion. This role varied depending on whether the participant was a policymaker or officeholder. Municipal decision-making was felt to be largely based on pursuing the interests of one’s own political group. Policymakers stated that opinions, values and political perspectives influenced the municipality’s MH promotion policies. The role of officeholders in children’s MH promotion, in turn, was related to local management, including directing resources, maintenance, control and information from child and family service entities, and various collaboration groups. Officeholders described being involved in the various stages of the decision-making process as professional representatives of the administration. Participants saw the financial and human resources of the municipalities as poor or underused from the perspective of the welfare state in organizing services for children, parents and families. This meant that although the statutory basic services were properly organized, the focus was on addressing problems as they arose. In addition, interviewees felt that models of teamwork and multidisciplinary work were lacking and that the problem was operating by sector. Additional resources were sought for the affairs of individual sectors; the service package was not considered as a whole. Regarding human resources, the officeholders’ view was that person-years have been added to the provision of services for children and young people and that human resources are in place in the service structures. The policymakers saw the situation differently. They described a shortage of school curators and psychologists in student welfare services and a shortage of family counseling staff in social work. Municipal strategies, programs and plans were largely described as good and taking children into account, but structures and resources were not always seen to match plans, and the system was fragmented and siloed. The integration and collaboration of child and family services was one way to improve the implementation of the plans through a holistic view of the child and the family that these services offer. Participants described the important role of the central government in promoting children’s MH, both positively and negatively. Financial support for municipal services for families with children and young people, early support, and positive discrimination as an example were forms of support received from the state. In turn, measures taken by the state that negatively affected the well-being of families and thus for the MH of children were measures that increased the poverty of families with children, the reduction of municipal subsidies, cuts in MH services and restriction of the subjective right to day care. Generally, participants were not satisfied with the promotion of children’s MH at the national level. Both national and international documents have addressed child well-being, but the existence of these documents was perceived as a separate matter from the actual guidance. According to participants, activities were governed by laws and regulations, but legislation requires reform and tightening. Local decision-making values services and allocates resources. With such guidance, the promotion of children’s MH was not seen to be sufficiently effective in municipalities. The contradictory role of the environmental, building, culture, and sport sectors in children’s MH promotion included factors related to the community level. Several participants referred to the size and structure of the municipality. The growth environment in small rural municipalities was perceived to be safer and closer to nature and to have fewer urban problems, such as crime and drug use. However, there were fewer leisure services, and there were fewer social and health care services in these municipalities, while the distances to these services were longer. Participants described that child-friendliness should also be considered in environmental planning. This meant, for example, a stimulating daycare center environment. However, child-friendliness was not seen to be realized in the best possible way. For example, there were few places for youth to go. In addition, municipalities, parishes, third sectors and associations organized different hobbies for children. In some municipalities, leisure activities were limited, but there were other problems associated with the hobbies, such as fees that were too expensive, excessive competition and long distances to travel to the services. ## Promotive and risk factors at the organizational, interpersonal and individual levels Support, requirements, and development needs in children’s services refer to factors at the organizational level (Table 3). According to political decision-makers and officeholders, early intervention in children’s MH symptoms requires small group sizes and sufficient trained staff in the ECEC. Although participants emphasized the basics, such as dining, outdoor activities and a regular daily rhythm, and suitable facilities, such as spacious enough buildings, the importance of collaboration among sectors working with children, as well as collaboration with parents/guardians, also became clear. Actions to support children’s MH at school were described somewhat similarly to those in ECEC. Multi-professional collaborations and collaborations between home and school were even more emphasized than in ECEC. In addition, the role of schools in the promotion of inclusion and equality, the prevention of exclusion and the prevention of bullying were perceived as crucial. Participants described that actions to support children’s MH in social and health care included actions targeted toward children, such as mapping the child’s situation, identifying risk factors and supporting well-being. However, the child’s MH was also promoted through parental support. Social security in particular was perceived to work with the whole family to support the well-being of the children. Child MH services in primary health care and child psychiatry functioned well for the most part, well but the problem consisted of long queues and long geographical distances to services. Developmental needs in social and health care were discussed, which could also contribute to solving the problems outlined above in child MH services. A decentralized service network and processes would help to perceive the family situations as a whole and could facilitate access to services. The development of services was seen as having the potential to reduce the fear of stigma associated with MH services and to contribute to the success of early intervention in cases where symptoms are observed in the child. At the interpersonal level, the importance of family and close networks was emphasized. In the case of children in particular, different family and home conditions have a large role. Caring, supporting and safe growing conditions and an intact home were felt to support the child’s development. Challenging life situations of parents/guardians, substance abuse, MH problems and disadvantages, in addition to poor parenting skills of parents/guardians, were the largest risk factors at the interpersonal level. Participants were also concerned about the intergenerational transmission of problems. Parents need support in challenging life situations, but they do not always have their own networks, such as close relatives. In this case, the child’s social networks were also narrowed, which could have contributed to the emergence of experiences of externalities. Strengthening friendships was considered important because being left alone, in addition to bullying, was felt to have a great impact on children’s MH. The premise for the MH and MH challenges faced by the child reflected the individual basis of MH. Policymakers and officeholders emphasized caring for the basic needs of the child. In addition to practical issues, such as treatment of physical illnesses, good self-esteem and inclusion in peer groups were mentioned as factors supporting the child’s MH as well as the right to be a child. Participants also mentioned various MH challenges faced by children. Although MH symptoms are due to genetic factors in some children, most of the risks are related to broader contemporary challenges, such as social media, or excessive demands related to, for example, school success or personal appearance. ## DISCUSSION Several important findings related to the views of municipal political decision-makers and officeholders on the factors affecting children’s MH in a socioecological context arise from this study. The key findings at the public policy level were related to strategic planning and implementation challenges in the municipality and state administration. Participants expressed concern that the promotion of children’s MH was considered in the strategies, programs and plans of the municipalities, but the plans may not have been realized in practice. This has also been confirmed in a previous study (Savolainen et al., 2021). One of the main reasons for this lack of realization was probably that services for children, parents and families were perceived as both dysfunctional and inadequately resourced. Policymakers in particular saw human resources as inadequate. Trained, competent and well-meaning personnel improve the quality and continuity of services and support the rights and well-being of the child (Noonan et al., 2019). At present, personnel targets in social and health services are not fully realized, personnel are overburdened and turnover is high, and reducing turnover would help the child establish safe adult contact (Parliamentary National Child Strategy Committee, 2021). From the global perspective, the situation in Finland regarding resources in organizing services for children is reasonable. However, interviewees stated that resources are used to deal with problems instead of for preventive services; this situation indicates a need for development considering that previous studies have reported the effectiveness of MH promotion as well as prevention of childhood mental illness (Raval et al., 2019; Weare and Nind, 2011). In addition, participants perceived that the integration of fragmented services would allow a more holistic view of the whole family. In Finland, attention has been given to intact service packages in recent child strategy work. The strategy considers that the child- and family-oriented nature of services, along with accessibility and low-threshold services, should be developed with the help of the family center model (Parliamentary National Child Strategy Committee, 2021). Inconsistently, the state has reduced municipal subsidies, despite the fact that municipalities need financial support for development work. Budget cuts from efforts to support well-being are short-sighted because the cost of prevention shifts to repairing the harm, as seen from the financial cuts made during the economic depression of the 1990s in Finland, which increased the need for children’s MH services (Ristikari et al., 2016). Thus, the child’s right to the best possible health care and adequate social and health services must be secured when planning possible budget cuts (United Nations, 1989; Constitution of Finland, $\frac{731}{1999}$). The roles of policymakers and leading officeholders in the MH promotion of children followed the traditional division. Policymakers described their role as ideological decision-makers who act as representatives of residents, especially their own constituents. Officeholders, in turn, identify themselves as local managers, including professional representations of the administration. In this way, officeholders were widely involved in the various stages of the decision-making process. Consequently, the MH promotion of children in the municipalities was influenced by the success of this dual management system. However, the turnover of policymakers is affected by the municipal elections held every 4 years (Local Government Act of Finland, $\frac{410}{2015}$). As a result, the support of political parties may also change, which would change the municipality’s strategy and priorities in terms of children’s well-being. As officeholders also have a great influence in directing and implementing the MH promotion-related activities of the municipality, it should be ensured that they develop effective and long-term policies that address children’s health needs. However, previous studies have found that children and families with children are poorly reflected in political discourse and goal setting (Benning et al., 2020; Hiilamo et al., 2021; Pereira et al., 2022). The participants descriptions of the roles of the environmental, building, culture and sport sectors in children’s MH promotion appeared to be contradictory. Participants highlighted the strengths of small rural municipalities, such as proximity to nature, which has been found to increase well-being (Wiens et al., 2016). However, children in small municipalities may be in an unequal position due to the reduced access to service provision and the long travel distances to services. Social inequality sharpens differences between children’s starting points, which can have far-reaching consequences (Vorma et al., 2020). Thus, a growth environment that supports the positive development of mental resources should be invested in, for example, by developing the services of small municipalities to equal those of larger municipalities and by investing in e-health services that make social and health services more accessible to all (Risling et al., 2017; Wynn et al., 2020). Several times, participants raised the importance of multi-professional and sectoral collaboration when they described support, requirements and development needs in children’s services. However, there is a need to develop the necessary knowledge and skills to promote effective collaboration across sectors (Tamminen et al., 2022). Previous studies have found various factors, such as organizations’ different practices, lack of time, employees’ working habits and data protection, that prevent the transmission of information, which may pose a challenge to collaboration (Anderson et al., 2017; Hoffman et al., 2016). In addition, problems have been identified in the past in connections between basic and special services and between MH and other social and health services (Greene et al., 2016). Alongside collaboration between professionals, participants also raised the importance of collaboration with parents. Such collaboration contributes to supporting parents in their upbringing work, as they play such an important role in children’s well-being (Sormunen et al., 2011). This is a priority, especially for parents who have, for example, health-related problems, because parental health problems can also increase children’s subsequent MH problems and high-risk health behavior (Remes et al., 2019). In addition, participants were concerned that the problems in the families, such as social disadvantages, were inter-generationally inherited. In socially disadvantaged families, preventive social work and preventive forms of child protection are important means of preventing the intergenerational transfer of problems (Vauhkonen et al., 2017). According to participants, the basis of MH is built individually. Municipalities should consider the individual needs of every child in addition to universal measures. The best outcomes in terms of promoting children’s MH are achieved by incorporating both universal and targeted approaches (Weare and Nind, 2011). This means that the promotion of children’s MH and the planning of related services should, in principle, be based on knowledge of how children’s well-being is structured and what kind of support families with children need from society (Pekkanen et al., 2020). The results of this study highlight negative issues. Although, when compared internationally, the general levels of well-being and population health have continuously improved in Finland, the distribution of health and well-being in the population is increasingly unequal (OECD, 2019). The focus should be more on children who are in a worse situation. Several risk and promotive factors affect the overall MH of children. Thus, MH promotion should include both decreasing risk factors and increasing the number of protective factors. Not every factor that affects well-being can be fixed, such as possible genetic factors (Taylor et al., 2017), but much can be done, such as taking the well-being of children into account in the activities of all administrative sectors and organizing social and health services that support the health of children and families (Vorma et al., 2020). This study has some limitations that need to be acknowledged and addressed. First, the results combine the responses of two different groups with different roles in municipal decision-making—policymakers and officeholders. Considering municipal policymakers and officeholders separately could have deepened the analysis of the views of a single group. However, there was a desire to obtain an overall picture, as both groups are involved in the decision-making process and the combined perspectives better answer the research question. Second, it is possible that participants with a particular interest in the topic were selected for interviews. This may have influenced some of the participants’ answers. Despite this limitation, both supportive and risk factors emerged equally from the results. The trustworthiness of the study results was assessed through credibility, dependability and transferability (Graneheim and Lundman, 2004). Previous research on the promotion of children’s MH from the perspective of municipal policymakers and leading officeholders is scarce. Thus, the data were appropriate to collect through interviews. Participants were from different service sectors and from municipalities with different populations. This diversity supported the credibility of the study. The authors had previous experience in conducting qualitative research, and they had an open dialog within the research team, which strengthened the research’s dependability. In this paper, every stage of the study is described accurately, including a distinct description of the context, selection and characteristics of participants, data collection and process of analysis. Additionally, appropriate quotations are presented to help with the transferability (Elo and Kyngäs, 2008). ## CONCLUSIONS This study describes the perceptions of municipal political decision-makers and leading officeholders on promotive and risk factors for children’s MH in a socioecological context. In general, the results highlight the need to promote children’s MH in municipalities. Most children in Finland are doing well, but support structures should target at the most vulnerable children to avoid the polarization of society and to support their parents, families and the children themselves. Specific, areas for development include the integration and better collaboration of child and family services and steering boards and the use of child rights impact assessments in political decision-making. Development requires action from the state, including legislative reform in addition to financial support. ## Funding This work was supported by the Finnish Cultural Foundation, North Savo Regional Fund (9.4.2019), the OLVI Foundation (4.6.2019) and the Doctoral Programme in Health Sciences of the University of Eastern Finland. Recipient: First author. ## Ethical Approval Ethical approval for the study was obtained from the Ethics Committee of the university (statement $\frac{5}{2019}$, 17.4.2019). In addition, a research permit was obtained separately from each municipal administration, and written informed consent was obtained from all participants. ## References 1. Anderson J. K., Howarth E., Vainre M., Jones P. B., Humphrey A.. A scoping literature review of service-level barriers for access and engagement with mental health services for children and young people.. Children and Youth Services Review (2017) 77 164-176 2. Auersperg F., Vlasak T., Ponocny I., Barth A.. Long-term effects of parental divorce on mental health—a meta-analysis.. Journal of Psychiatric Research (2019) 119 107-115. PMID: 31622869 3. Benning T. J., Ashby G. B., Chapp C. B.. Frequency and specificity of pediatric health policy discussions in political campaigns.. JAMA Pediatrics (2020) 174 795-796. PMID: 32478807 4. Bertoni M.. Hungry today, unhappy tomorrow? Childhood hunger and subjective wellbeing later in life.. Journal of Health Economics (2015) 40 40-53. PMID: 25576999 5. Curtis S., Gesler W., Smith G., Washburn S.. Approaches to sampling and case selection in qualitative research: examples in the geography of health.. Social Science and Medicine (2000) 50 1001-1014. PMID: 10714922 6. Elo S., Kyngäs H.. The qualitative content analysis process.. Journal of Advanced Nursing (2008) 62 107-115. PMID: 18352969 7. Graneheim U. H., Lundman B.. Qualitative content analysis in nursing research: concepts, procedures and measures to achieve trustworthiness.. Nurse Education Today (2004) 24 105-112. PMID: 14769454 8. Greene C., Ford J., Ward-Zimmerman B., Honigfeld L., Pidano A.. Strengthening the coordination of pediatric mental health and medical care: piloting a collaborative model for freestanding practices.. Child and Youth Care Forum (2016) 45 729-744. PMID: 27833396 9. Griebler U., Rojatz D., Simovska V., Forster R.. Effects of student participation in school health promotion: a systematic review.. Health Promotion International (2017) 32 195-206. PMID: 24395957 10. Guido C. A., Amedeo I., Avenoso F., Bruni J., Zicari A. M., Loffredo L.. Risk factors and mental health promotion strategies in children during COVID-19.. Frontiers in Public Health (2020) 8 580720. PMID: 33178664 11. Hiilamo A., Jalonen H., Helve O., Kaila M., Ristikari T., Virtanen P.. (2021) 12. Hoffman J. A., Bunger A. C., Robertson H. A., Cao Y., West K. Y.. Child welfare caseworkers' perspectives on the challenges of addressing mental health problems in early childhood.. Children and Youth Services Review (2016) 65 148-155 13. Jenkins E., Haines-Saah R., McGuinness L., Hirani S., Boakye-Yiadom N., Halsall T.. Assessing the impacts of the Agenda Gap intervention for youth mental health promotion through policy engagement: a study protocol.. International Journal of Mental Health Systems (2020) 14 1-58. PMID: 31921334 14. Klemera E., Brooks F. M., Chester K. L., Magnusson J., Spencer N.. Self-harm in adolescence: protective health assets in the family, school and community.. International Journal of Public Health (2017) 62 631-638. PMID: 27658811 15. Merikukka M., Ristikari T., Tuulio-Henriksson A., Gissler M., Laaksonen M.. Childhood determinants for early psychiatric disability pension: a 10-year follow-up study of the 1987 Finnish Birth Cohort.. International Journal of Social Psychiatry (2018) 64 715-25. PMID: 30394811 16. Mikkelsen K., Stojanovska L., Polenakovic M., Bosevski M., Apostolopoulos V.. Exercise and mental health.. Maturitas (2017) 106 48-56. PMID: 29150166 17. Min K., Kim H., Kim H., Min J.. Parks and green areas and the risk for depression and suicidal indicators.. International Journal of Public Health (2017) 62 647-656. PMID: 28337512 18. Ministry of Finance. (2015) Local Government Act of Finland 410.. Local Government Act of Finland 410 (2015) 19. Ministry of Justice. (1999) Constitution of Finland 731.. Constitution of Finland 731 (1999) 20. Ministry of Social Affairs and Health. (2010) Health Care Act of Finland 1326.. Health Care Act of Finland 1326 (2010) 21. Ministry of Social Affairs and Health. (2021) Children, youth and the COVID-19. Assessment of the materialisation of the rights of the child and proposals for post-crisis measures presented by a working group on the National Child Strategy, Publication 2. Finnish Government, Helsinki. https://julkaisut.valtioneuvosto.fi/handle/10024/162647 (last accessed 11 November 2021).. (2021) 22. Netsi E., Pearson R. M., Murray L., Cooper P., Craske M. G., Stein A.. Association of persistent and severe postnatal depression with child outcomes.. JAMA Psychiatry (2018) 75 247-253. PMID: 29387878 23. Noonan M., Galvin R., Jomeen J., Doody O.. Public health nurses’ perinatal mental health training needs: a cross sectional survey.. Journal of Advanced Nursing (2019) 75 2535-2547. PMID: 30937923 24. OECD. (2019) Health for Everyone? Social Inequalities in Health and Health Systems, OECD Health Policy Studies. OECD Publishing, Paris. 10.1787/3c8385d0-en (last accessed 26 October 2021).. Health for Everyone? Social Inequalities in Health and Health Systems, OECD Health Policy Studies (2019). DOI: 10.1787/3c8385d0-en 25. Oerlemans A. M., Burmanje M. J., Franke B., Buitelaar J. K., Hartman C. A., Rommelse N. N. J.. Identifying unique versus shared pre- and perinatal risk factors for ASD and ADHD using a simplex-multiplex stratification.. Journal of Abnormal Child Psychology (2016) 44 923-935. PMID: 26466830 26. O’Hara K. L., Sandler I. N., Wolchik S. A., Tein J.. Coping in context: the effects of long-term relations between interparental conflict and coping on the development of child psychopathology following parental divorce.. Development and Psychopathology (2019) 31 1695-1713. PMID: 31535609 27. Parliamentary National Child Strategy Committee. (2021) National Child Strategy, Committee Report. Finnish Government, Helsinki.. National Child Strategy, Committee Report (2021) 28. Pekkanen N., Saaristo V., Ståhl T.. Hyvinvoinnin ja terveyden edistämisen rakenteet, prosessit ja toimenpiteet kunnissa sekä alueilla, Tutkimuksesta tiiviisti 5 (2020) 29. Pereira M., Padez C., Nogueira H.. Municipal health promotion programs: is childhood obesity a concern at local level in Portugal?. Health Promotion International, (2022) 37. DOI: 10.1093/heapro/daab044 30. Raitasalo K., Holmila M., Jääskeläinen M., Santalahti P.. The effect of the severity of parental alcohol abuse on mental and behavioural disorders in children.. European Child and Adolescent Psychiatry (2019) 28 913-922. PMID: 30430262 31. Raval G., Montanez E., Meyer D., Berger-Jenkins E.. School-based mental health promotion and prevention program “Turn 2 Us” reduces mental health risk behaviors in urban, minority youth.. Journal of School Health (2019) 89 662-668. PMID: 31169913 32. Remes H., Moustgaard H., Kestilä L. M., Martikainen P.. Parental education and adolescent health problems due to violence, self-harm and substance use: what is the role of parental health problems?. Journal of Epidemiology and Community Health (2019) 73 225-231. PMID: 30635438 33. Risling T., Martinez J., Young J., Thorp-Froslie N.. Evaluating patient empowerment in association with ehealth technology: scoping review.. Journal of Medical Internet Research (2017) 19 e329. PMID: 28963090 34. Ristikari T., Törmäkangas L., Lappi A., Haapakorva P., Kiilakoski T., Merikukka M.. Finland as a Growth Environment for Young People, 25-Year Follow up of Those Born in Finland in 1987 (2016) 35. Savolainen O., Turunen H., Sormunen M.. Municipal strategies and meeting minutes’ descriptions of the promotion of children’s mental health: a document analysis.. Scandinavian Journal of Public Health (2021) 49 519-528. PMID: 34636264 36. Sharma A., Sharma S. D., Sharma M.. Mental health promotion: a narrative review of emerging trends.. Current Opinion in Psychiatry (2017) 30 339-345. PMID: 28661906 37. Sormunen M., Tossavainen K., Turunen H.. Home-school collaboration in the view of fourth grade pupils, parents, teachers, and principals in the Finnish education system. The School Community Journal (2011) 21 11-21 38. Stokols D.. Translating social ecological theory into guidelines for community health promotion.. American Journal of Health Promotion (1996) 10 282-298. PMID: 10159709 39. Tamana S. K., Ezeugwu V., Chikuma J., Lefebvre D. L., Azad M. B., Moraes T. J.. Screen-time is associated with inattention problems in preschoolers: results from the CHILD birth cohort study.. PLoS One (2019) 14 e0213995. PMID: 30995220 40. Tamminen N., Solin P., Barry M. M., Kannas L., Kettunen T.. Intersectoral partnerships and competencies for mental health promotion: a Delphi-based qualitative study in Finland.. Health Promotion International (2022) 37. DOI: 10.1093/heapro/daab096 41. Taylor M. K., Beckerley S. E., Henniger N. E., Hernández L. M., Larson G. E., Granger D. A.. A genetic risk factor for major depression and suicidal ideation is mitigated by physical activity.. Psychiatry Research (2017) 249 304-306. PMID: 28142104 42. The Nordic Council and the Nordic Council of Ministers. (2021) Nordic co-operation.https://www.norden.org/en (last accessed 11 November 2021).. (2021) 43. Thualagant N., Simonsen N., Sarvimäki A., Stenbock-Hult B., Olafsdottir H. S., Fosse E.. Nordic responses to covid-19 from a health promotion perspective.. Health Promotion International (2022) 1-13. DOI: 10.1093/heapro/daab211 44. Ukkola A., Metsämuuronen J., Paananen M.. (2020) 45. ).. (1989) 46. Vaismoradi M., Turunen H., Bondas T.. Content analysis and thematic analysis: implications for conducting a qualitative descriptive study.. Nursing and Health Sciences (2013) 15 398-405. PMID: 23480423 47. Vauhkonen T., Kallio J., Kauppinen T. M., Erola J.. Intergenerational accumulation of social disadvantages across generations in young adulthood.. Research in Social Stratification and Mobility (2017) 48 42-52 48. Vorma H., Rotko T., Larivaara M., Kosloff A.. (2020) 49. Weare K., Nind M.. Mental health promotion and problem prevention in schools: what does the evidence say?. Health Promotion International (2011) 26 129-169 50. Wennberg M., Luukkonen T., Haila K.. (2020) 51. Wiens V., Kyngäs H., Pölkki T.. The meaning of seasonal changes, nature, and animals for adolescent girls' wellbeing in northern Finland: a qualitative descriptive study.. International Journal of Qualitative Studies on Health and Well-Being (2016) 11 30160-14. PMID: 26905401 52. Wynn R., Oyeyemi S. O., Budrionis A., Marco-Ruiz L., Yigzaw K. Y., Bellika J. G.. Electronic health use in a representative sample of 18,497 respondents in Norway (The Seventh Tromsø Study—part 1): Population-Based Questionnaire Study.. JMIR Medical Informatics (2020) 8 e13106. PMID: 32134395
--- title: Plasma protein levels of young healthy pigs as indicators of disease resilience authors: - Yulu Chen - Steven Lonergan - Kyu-Sang Lim - Jian Cheng - Austin M Putz - Michael K Dyck - PigGen Canada - Frederic Fortin - John C S Harding - Graham S Plastow - Jack C M Dekkers journal: Journal of Animal Science year: 2023 pmcid: PMC9977353 doi: 10.1093/jas/skad014 license: CC BY 4.0 --- # Plasma protein levels of young healthy pigs as indicators of disease resilience ## Abstract Selection for disease resilience, which refers to the ability of an animal to maintain performance when exposed to disease, can reduce the impact of infectious diseases. However, direct selection for disease resilience is challenging because nucleus herds must maintain a high health status. A possible solution is indirect selection of indicators of disease resilience. To search for such indicators, we conducted phenotypic and genetic quantitative analyses of the abundances of 377 proteins in plasma samples from 912 young and visually healthy pigs and their relationships with performance and subsequent disease resilience after natural exposure to a polymicrobial disease challenge. Abundances of 100 proteins were significantly heritable (false discovery rate (FDR) <0.10). The abundance of some proteins was or tended to be genetically correlated (rg) with disease resilience, including complement system proteins (rg = −0.24, FDR = 0.001) and IgG heavy chain proteins (rg = −0.68, FDR = 0.22). Gene set enrichment analyses (FDR < 0.2) based on phenotypic and genetic associations of protein abundances with subsequent disease resilience revealed many pathways related to the immune system that were unfavorably associated with subsequent disease resilience, especially the innate immune system. It was not possible to determine whether the observed levels of these proteins reflected baseline levels in these young and visually healthy pigs or were the result of a response to environmental disturbances that the pigs were exposed to before sample collection. Nevertheless, results show that, under these conditions, the abundance of proteins in some immune-related pathways can be used as phenotypic and genetic predictors of disease resilience and have the potential for use in pig breeding and management. Plasma protein levels of young healthy pigs are promising as biomarkers to select for disease resilience. ## Introduction Infectious diseases directly take their toll on the swine industry by increasing mortality, reducing productivity and animal welfare, and are among the main obstacles and sources of losses in pork production (Morgan and Prakash, 2006). For example, in 2013, Holtkamp et al. [ 2013] estimated porcine reproductive and respiratory syndrome to cost the U.S. swine industry around $664 million annually and this disease continues to be a major problem in the swine industry in North America and globally. For the past three decades, goals for pig breeding have focused primarily on lean meat growth, feed efficiency, carcass quality, and reproduction. In recent years, however, the goal of pig breeding has expanded to include health-related traits such as disease resilience (Mellencamp et al., 2008), which refers to the ability of an animal to maintain relatively undiminished performance when exposed to diseases (Albers et al., 1987; Bisset and Morris, 1996; Nguyen-Ba et al., 2020). Incorporating health traits into breeding programs can reduce economic losses due to disease, increase production efficiency, reduce the use of antibiotics, and improve animal welfare. Disease pressures on commercial pig farms are complex, so selective breeding of pigs that are more resilient across common diseases can be a practical way to improve productivity (Putz et al., 2019). In most cases, direct selection for disease resilience in breeding programs is not feasible because nucleus breeding stock is by necessity kept in high-health bio-secure environments. Quantifying disease resilience to polymicrobial infectious agents in commercial farms is also challenging because it is a manifestation of many biological processes, is simultaneously affected by multiple factors, and collecting individual animal data in commercial farms is complex. A promising alternative is an indirect selection based on a suitable indicator trait of disease resilience that can be measured on healthy animals, i.e. in the nucleus, preferably at a young age. For an indicator trait for disease resilience to be effective, it needs to be measurable in a nucleus environment, heritable, and genetically correlated with disease resilience. Proteins are downstream products of the genome that drive many cellular processes (Lesk, 2001). The blood proteome is a complex composite of proteins, including classical blood proteins and proteins secreted or leaked from tissues, including hormones, cytokines, adipokines, chemokines, and growth factors that coordinate biological processes associated with health or diseases (Lin et al., 2008). The blood proteome, thereby, provides a window into the current state of the body, including health (Anderson et al., 2004). By applying machine learning to large-scale and deep plasma proteome data, Williams et al. [ 2019] demonstrated that plasma proteome patterns can be a comprehensive predictor of human life span and health status, including future age-related disease risks. In addition, colocalization of DNA variants that are associated with diseases with the genomic location of genes for intermediate phenotypes such as blood protein levels can be used to identify drug targets and disease-translational biomarkers (Kastenmuller et al., 2015; Suhre et al., 2021). Previous studies have shown that the abundance of some proteins in blood is heritable or regulated by genetic factors, including in humans (Johansson et al., 2013; Liu et al., 2015), mice (Holdt et al., 2013), dairy cattle (Cecchinato et al., 2018), and pigs (Clapperton et al., 2009; Reyer et al., 2019; Ballester et al., 2020). In pigs, some studies have shown that the blood proteome changes with disease status (te Pas et al., 2013; Muk et al., 2019) and that the levels of some plasma proteins, such as alpha-acid glycoproteins, are genetically negatively correlated with average daily gain (from −0.72 to −0.53) (Clapperton et al., 2009). Here, we integrated the population-level plasma proteome of young, visually healthy pigs with whole-genome single-nucleotide polymorphism (SNP) genotype data and extensive phenotypes measured before and after their exposure to a polymicrobial natural disease challenge. The overall objective was to explore whether the plasma proteome of young, healthy piglets can be early indicators for disease resilience by [1] investigating the genetic basis of the plasma proteome of young, healthy pigs, and [2] identifying phenotypic and genetic associations of the plasma proteome of young, healthy pigs with their subsequent disease resilience phenotypes, as well as the biological basis behind these associations. ## Ethical statement The project protocol was approved by the Animal Protection Committee of the Centre de Recherche en Sciences Animales de Deschambault (15PO283) and the Animal Care and Use Committee of the University of Alberta (AUP00002227) and carried out following Canadian Council on Animal Care guidelines (CCAC; https://ccac.ca/en/guidelines-and-policies/fundamental-principles.html). The Quebec Provincial Centre for Population Development and pastoralists and project veterinarians provided comprehensive oversight of animal care. Pigs in this project were humanely euthanized when humane intervention points were exceeded or response to treatment was inadequate. Following CCAC guidelines, electrocution was used to euthanize pigs during the nursery period, while a captive cranial bolt was used during the finisher period. According to standard approved industry protocols, pigs that reached slaughter weight were stunned by electrocution at a commercial slaughter facility, followed by exsanguination. ## Natural disease challenge model This study used data from the polymicrobial natural disease challenge model (NDCM) described by Putz et al. [ 2019], which was established in 2015 and continued until 2021 at the CDPQ in Québec, Canada, to study the genetic control of disease resilience in grow-finish pigs. The NCDM was designed to simulate the disease pressure in a commercial farm with poor health and was established by bringing naturally infected pigs into a nursery and finisher barn and maintained by continuous flow, with older batches of pigs exposing incoming batches to diseases by nose-to-nose contact. Every 3 weeks, a batch of 60 or 75 weaned, Large White by Landrace crossbred barrows was provided from a bio-secure multiplier farm in Canada from one of the seven members of PigGen Canada (http://piggencanada.org/), in rotation. One rotation of seven batches (one batch per company) was referred to as a cycle, for a total of seven cycles. Further details are in Putz et al. [ 2019]. Performance (prior to challenge) and resilience (after challenge) phenotypes were collected on all pigs over three growth phases, as described by Putz et al. [ 2019]: [1] quarantine nursery phase (on average 19 days, starting at ~21 days of age); [2] challenge nursery phase (on average 28 days, starting at ~40 days of age); and [3] finisher phase (on average 100 days, starting at ~70 days of age). Detailed performance and resilience phenotypes were available on 3,205 pigs from cycles 1 to 7, as described by Putz et al. [ 2019] and Cheng et al. [ 2020] and included: average daily gain in the quarantine nursery (qNurADG), in the challenge nursery (cNurADG), and in the finisher (cFinADG); the number of individual parenteral antibiotic treatments provided in the challenge nursery, adjusted to 27 days (cNurTRT), in the finisher, adjusted to 100 days (cFinTRT), and from birth, adjusted to a standard 180 days of age at slaughter (AllTRT); mortality (0 for pigs that survived; 1 for pigs that died) in the challenge nursery (cNurMOR), in the finisher (cFinMOR), and across the challenge nursery and finisher phases (AllMOR); subjective health scores (HS) assigned by trained personnel on a 1–5 scale based on clinical signs (1 = severe clinical signs to 5 = perfect health, see Cheng et al. [ 2020]) at 5 and 19 days after entry into the quarantine nursery (qNurHS1 and qNurHS2), at 3 weeks after entry into the challenge nursery (cNurHS), and at 6 weeks after entry into the finisher (cFinHS); average daily feed intake (ADFI), feed conversion ratio (FCR), and residual feed intake (RFI) in the finisher; and carcass weight (CWT), carcass back fat (CBF), carcass loin depth (CLD), dressing percentage (DRS), and lean yield (LYD) at slaughter. Incomplete phenotypes for cFinADG, cFinTRT, AllTRT, ADFI, and FCR for pigs that died in the finisher were imputed and expanded as described by Cheng et al. [ 2020], to put them on the same scale as those of pigs that survived to slaughter. This resulted in two data sets for these traits: survivor data, which only included phenotypes on pigs that survived to slaughter, and expanded data, which also included imputed data on selected pigs that died in the finisher (see Cheng et al. [ 2020] for details). Because of the limited number of animals receiving low health scores (see Cheng et al. [ 2020] for details) and because the scale of scores may not be linear, pigs with an HS less or equal to 4 were assigned a score of 4. Detailed statistics and estimates of genetic parameters of these performance and resilience phenotypes are in Cheng et al. [ 2020]. ## Genotyping and quality control All animals were genotyped for 658,692 SNPs using a 650 k Affymetrix Axiom Porcine Genotyping Array by Delta Genomics (Edmonton AB, Canada). Raw data were processed separately for each cycle by Delta Genomics, using default settings of the Axiom Analysis Suite (quality control thresholds: call rate for marker > 0.10; call rate for individual > 0.10; minor allele frequency > 0.05), as described by Putz et al. [ 2019]. After quality control, 435,172 SNPs on 3,205 pigs remained for further analysis. ## Protein abundance measurement Whole blood samples were collected ~5 days after entering the quarantine nursery (~26 days of age) into K2 ethylenediaminetetraacetic acid (EDTA) tubes (BD Vacutainer Blood Collection Tubes, United States). Multiple studies have shown that physiological indicators at this time may also be affected by weaning and transportation (Zhu et al., 2012; Buchet et al., 2017; Montagne et al., 2022), so we regard pigs at this time as visually healthy weaned pigs and we use “healthy” to refer to “visually healthy weaned” in the remainder. After centrifugation (2,000 g at 4 °C for 10 min) the plasma layer was aliquoted by transfer pipette into Thermo Scientific Nunc barcoded tubes. Immediately after processing, samples were frozen at −80 °C until subsequent analysis. The available samples were processed in two groups (groups 1 and 2), in December 2018 and November 2019. Samples in group 1 came from cycles 4 and 5, while samples in group 2 came from cycles 4 to 7. The protein content in each blood sample was determined (Bradford, 1979) using pre-mixed reagents (Bio-Rad Laboratories, Hercules, CA) and adjusted to 10 µg/µL. Protein abundances, quality, and profiles were evaluated using $15\%$ SDS–PAGE gels and Colloidal Coomassie blue staining ($1.7\%$ ammonium sulfate, $30\%$ methanol, $3\%$ phosphoric acid, and $0.1\%$ Coomassie G-250) (Cruzen et al., 2015). Samples were stored at −80 °C until labeling and analysis. The Thermo Scientific TMT Mass Tag Labeling Kits and Reagents protocol (11-plex) was used to identify and quantify the abundance of individual proteins in each plasma sample (Thompson et al., 2003). For each sample, 25 μg was diluted with 50 mM Tris (pH 8) to a concentration of 0.5 μg/μL. Then, 5 μL of 0.1 M DTT was added to reach a final concentration of 5 mM, and samples were mixed and incubated at 37 °C for 30 min. The alkylation process was conducted by adding 1.5 μL of 1M iodoacetamide to a final concentration of 15 mM, after which samples were mixed and incubated in the dark at room temperature for 30 min. Then, 400 μL of 50 mM Tris–HCl (pH 8) was added to dilute samples, which reduced the concentration of urea for optimal trypsin activity. Trypsin was added to each sample at a 1 μg trypsin:50 μg sample ratio and incubated overnight at 37 ºC. The digestion process was stopped by adding formic acid (5 μL) to a final concentration of $1\%$. Samples were then centrifuged at 14,000–16,000 × g for 10 min using a benchtop microcentrifuge to remove particulate material, after which the samples were desalted using the Microspin column (SEM SS18V) and dried down using the SpeedVac. The tryptic peptide samples were reconstituted in 100 μL with 50 mM triethyl ammonium bicarbonate), mixed with 0.2 mg (10 μL) of the corresponding labeling reagent, and incubated for 1 h. The samples were then quenched with 8 μL of $5\%$ hydroxylamine (50 μL hydroxylamine in 450 μL 100 mM TEAB) and incubated for 15 min. The TMT 11-plex system can conduct quantitation using high-resolution MS for 11 samples simultaneously, which we refer to as a run. Within a run, each of the 11 samples had a unique labeling tag (126, 127N, 127C, 128N, 128C, 129N, 129C, 130N, 130C, 131N, and 131C). For each of the two groups of samples a reference sample was created by pooling 10 µg from each sample in the group, which was labelled with tag 131C in each run. For each run, the 11 labelled peptide samples were mixed, then eluted in 55 μL $5\%$ acetonitrile and $0.1\%$ formic acid, and dried by vacuum. Peptides were separated by liquid chromatography (Thermo Scientific EASY nLC-1200 coupled to a Thermo Scientific Nanospray FlexIon source) through a pulled glass emitter 75 µm × 20 cm (Agilent capillary, part #16-2644-5). The tip of the emitter was packed with C18 packing material (Agilent Zorbax Chromatography Packing, SB-C18, 5 µm, part #8220966-922), while the remainder of the column was packed with UChrom C18 3 micron material from nanoLCMS Solutions (part #80002). In detail, buffer A was $0.1\%$ formic acid in the water and buffer B was $0.1\%$ formic acid in $80\%$ acetonitrile/water. The gradient was comprised of an increase from $0\%$ to $35\%$ B in 210 min, followed by an increase to $70\%$ B in 20 min, then an increase to $100\%$ B in 5 min. The flow rate for the equilibration and separation was 300 nL/min. ESI voltage was at 2.65 kV in positive polarity mode. Subsequently, the peptides were fragmented for analysis by MS/MS using a Thermo Scientific Q Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer with an HCD fragmentation cell (Waltham, MA) using a Full MS/DD-MS2 (TOPN) method. Full MS scans were run with a scan range of 400–2000 m/z at a resolution of 70,000, with an AGC target of 1e6 and a maximum IT of 80 ms. The top 20 MS2 scans were run at a resolution of 35,000 with an AGC target of 1e5 and a maximum IT of 50 ms. The isolation window was set to 1.2 m/z and the fixed first mass was set to 110.0 m/z. The NCE setting was at 32. The acquired raw data were analyzed using the Proteome Discoverer software (version 2.4; Thermo Fisher Scientific, San Jose, CA, USA), separately for the groups 1 and 2 samples. Each raw file was searched against a Sus scrofa FASTA file database (UniProtKB database) on Sequest HT and Mascot search engines. Both searches were performed with a static modification of carbamidomethyl (Cys) and TMT label (Lys and N-termini), along with dynamic modifications of oxidation (Met) and deamidation (Asn, Gln). For both search engines, peptide spectral matches were validated at an FDR of $1\%$ under the percolator of the Proteome Discoverer software. The precursor mass tolerance was set at 10 ppm and the fragment mass tolerance at 0.02 Da for both search engines. The identified proteins were required to have at least one peptide sequence detected. The plasma samples were processed in two groups, resulting in some differences in the proteins detected for the two groups. Samples in groups 1 and 2 were, respectively, evaluated in 41 and 51 TMT 11-plex runs. Proteins detected also differed between runs but were confounded within a run, i.e. if one sample in one run had a missing value for a protein, then all samples in that run had a missing value for this protein. As a result, the percent of missingness increased significantly when all samples were combined, which was also observed by Bramer et al. [ 2021] and Brenes et al. [ 2019]. Missingness was, however, not associated with the level of abundance of the protein and could be considered random (see Supplementary Figure S1). The processed datasets from the two groups were then merged and proteins that were identified in more than 20 runs were used for statistical analysis. ## Statistical analyses To improve normality, the protein abundance data were transformed using the logarithm function with base 2. In Supplementary Figure S1, there was no relationship between the frequency of missing values for a protein vs. the average of the log2 protein abundance for the samples in runs for which that protein was not missing. Hence, the missing data were assumed to be missing at random in subsequent analyses. For each filtered protein, the following mixed linear model [1] was used to estimate the residual adjusted for systematic effects for each protein abundance observation: where yijklm is the log2 transformed protein abundance; Batchi and Tagj are the fixed effects of batch ($i = 1$, 2, …, 50) and TMT tag ($j = 1$, 2, …, 10), respectively; EntryAgeijklm is the covariate of age at entry into the quarantine nursery; ref(group 1)ijklm and ref (group 2)ijklm are the log2 transformed abundances of the reference sample for run m for groups 1 and 2, respectively, which were fitted as covariates (ref(group 1)ijklm was set equal to 0 for runs for group 2, and vice versa); Plexl is the random effect of run, assumed to be distributed N(0, Iσpl2), where σpl2 is the run variance; and eijklm is a random residual, which was assumed to be distributed N(0, Iσe2), where σe2 is the residual variance. For each protein, residuals that were outliers based on the 1.5 × inter-quartile range rule were removed (around $3\%$ of observations). ## Phenotypic associations with performance and resilience Associations of the abundance of each protein with each of the recorded performance and resilience traits were analyzed using the following mixed linear model, separately for each protein: where yijkm is the phenotype for a performance or resilience trait for a pig with proteome data; Proteinijkm is the protein residual from model [1], which was fitted as a covariate, one protein at a time; litterijk is the random effect of common litter environmental effects, assumed to be distributed N(0, Iσl2), where σl2 is the litter variance, and all other effects are as described for model [1], except that Penj refers to the pen corresponding to the time period of yijk, as described in Cheng et al. [ 2020]. For carcass traits, slaughter date was added as a fixed effect, as well as the covariates of age and weight at slaughter. For categorical traits (i.e. health scores and mortality), a reverse mixed linear model was used to analyze the association between protein abundance and the trait phenotype because logistic regression analyses of these traits failed to converge in some cases. In the reverse mixed linear model analyses, the protein residual from model [1] was used as the response variable and the binary phenotype ($\frac{0}{1}$) was fitted as a covariate in a mixed linear model [2]. The resulting estimates of the regression coefficient of the binary trait on protein abundance were then converted to regression coefficients of protein abundance on the binary trait by multiplying the estimate by the ratio of the variances of the residuals of the binary trait and of protein abundance. The Benjamini and Hochberg [1995] method was used to estimate the FDR of P-values across all recorded phenotypes and tests. To ensure sufficient proteins were available for downstream analyses, estimates with FDR less than 0.35 considered statistically significant. These analyses were implemented in the R packages lme4 (Bates et al., 2014) and car (Fox and Weisberg, 2019). ## Heritability of protein abundance A model similar to model [1] was used to estimate the variance components and genetic parameters for abundance of each protein but with the addition of random litter effects (litterikm, assumed distributed N (0, Iσl2), where σl2 is the litter environmental variance) and random animal additive genetic effects (aijkmu), which were assumed to be distributed N(0, Gσa2), where G is the genomic relationship matrix and σa2 is the additive genetic variance. Matrix G was created based on the SNP genotype data using the PreGSf90 software of BLUPF90 (Misztal et al., 2002). This matrix was computed separately for pigs from each company and then combined, with the relationship between pigs from different companies set to 0 to focus on pooled within-company variances, as described by Cheng et al. [ 2020]. Variance components were estimated using restricted maximum likelihood using the ASReml 4.0 software (Gilmour et al., 2014). Estimates of heritability and of the proportion of variance due to the litter effects were calculated as a proportion of phenotype variance, which was the sum of estimates of σe2, σl2, and σa2. A likelihood ratio test based on the difference in likelihood between models with and without additive genetic effects was used to determine the significance from zero estimates of heritability and of the proportion of variance due to litter effects. The P-value for this test was obtained by comparison to a Chi-square distribution with one degree of freedom, with the resulting P-value divided by 2 because the maximum likelihood estimates were restricted to be positive (Visscher, 2006). Variance ratios with FDR (Benjamini and Hochberg, 1995) less than 0.10 across proteins were considered significant. Figure 3 shows a histogram of estimates of heritability and of the proportion of variance due to litter effects for the abundance of each protein in the blood. Heritability estimates were significant (FDR < 0.10) for 100 of the 377 proteins, ranging from 0.17 to close to 1. The top four heritable proteins, with estimates close to 1, were A0A480TLF3, A0A480P4D2, A0A286ZKB4, and A0A4X1T2W4. Litter effects were significant (FDR < 0.10) for only two proteins. Figure 3.: Histograms of (a) heritability estimates and (b) proportions of variance due to litter effects for the abundance of each protein in the blood of young healthy pigs. The green color indicates estimates with FDR less than 0.10. ## Genetic correlations of protein abundance with performance and resilience traits Genetic correlations of the abundance of proteins with each performance and resilience trait were estimated using bivariate models in ASReml 4.0. Genetic correlations were only estimated for proteins whose abundance had an heritability estimate larger than 0.05 because bivariate analyses often fail to converge or lead to estimates of genetic correlations with extremely large SE if one of the traits has a low heritability. Genetic correlations were estimated using the proteome data on the 912 animals and performance and resilience data from pigs in all 50 batches (3205 animals), with the full genomic relationship matrix, created as described by Cheng et al. [ 2020]. The model used for proteome abundance was the same as that used to estimate heritability. For the performance and resilience traits, model [2] was used but with protein residual removed as a covariate and an animal additive genetic effect added as a random effect, as described in Cheng et al. [ 2020]. To determine significance of genetic correlation estimates from zero, a likelihood ratio test (Visscher, 2006) for models with and without the genetic correlation set to 0 was used, with P-values obtained by comparison to a Chi-square distribution with one degree of freedom. The Benjamini and Hochberg [1995] method was used to estimate the FDR of P-values across all phenotypic traits. To ensure sufficient proteins were available for downstream analyses, estimates of genetic correlations with FDR less than 0.35 were considered to be statistically significant. Histograms of estimates of genetic correlations of the abundance of 193 proteins that had heritability estimates greater than 0.05 with performance and resilience traits are shown in Figure 4. The abundance of five proteins had significant non-zero genetic correlations with is trait at the liberally chosen threshold of FDR < 0.35 (Table 4). No proteins were significant for more than one trait. For all but four traits (qNurHS2, cNurMOR, cFinMOR, and AllMOR), the mean of the genetic correlation estimates across proteins markedly deviated from zero. For example, the mean of genetic correlation estimates across proteins was positive with mortality during the different phases and negative with qNurHS2. This might be because these binary traits were treated as continuous variables in these analyses. A heatmap of the signed −log10 of the P-values of estimates of genetic correlations for proteins with heritability estimates higher than 0.05 is shown in Figure 5. Since many estimates had large standard errors, the signed −log10(P-value) rather than the genetic correlation estimate was used to represent the strength of the genetic relationships. Plots of the signed −log10(P-value) against the estimate of the genetic correlation for each trait are in Supplementary Figure S3. Patterns of the signed −log10(P-value) were consistent for mortality in the different phases (Figure 5), as well as for feed conversion ratio (FCR) and residual feed intake (RFI). The patterns of genetic correlation estimates for the survivor and expanded data were highly similar, except for AllTRT. Figure 5.: Heat map of the signed −log10(P-value) for estimates of the genetic correlation of protein abundance in blood of young healthy pigs with subsequent performance and disease resilience phenotypes for the survivor and expanded (exp) data sets. For trait abbreviations, see Table 1. The colors for the traits indicate the time period that the trait measured, where green represents quarantine nursery, orange the challenge nursery, red the finisher, and blue slaughter. Red/blue of the heat map value indicates that an increase in expression of that protein is favorable/unfavorably genetically correlated with the performance or disease resilience phenotype. Colors on the dendrogram identified different clusters based on ward.D clustering. ## Gene set enrichment analyses The GSEAPreranked tool of the GSEA_4.1.0 software (Subramanian et al., 2005) was used to perform gene set enrichment analysis of estimates of phenotypic associations of protein abundance with performance and resilience phenotypes from model [1]. For these analyses, the Gene Ontology (GO) biological process library and the REACTOME pathway library based on UniProt protein ID were used as protein annotation databases in separate analyses. The GO biological process library was built based on the current pig GOA (https://www.ebi.ac.uk/GOA/pig_release, released on 14 August 2020), with 10,627 GO biological processes. The REACTOME pathway library was downloaded from https://reactome.org/download-data (accessed on August 2020), with 1,562 pathways for swine. The format of the two libraries was adjusted to the GSEA GMT format. The GSEA analyses were conducted separately for each resilience trait, with the proteins ranked based on regression coefficient estimates of the phenotype on protein abundance residuals from model [1], and corresponding estimates from the reverse linear mixed models for categorical traits. For regression coefficients to be comparable across traits and proteins, they were scaled by multiplying the estimate by the ratio of the SD of protein abundance residuals and the SD of the resilience trait phenotype, such that they were expressed in units of (SD) of the resilience trait per SD of protein expression residual. The GSEA software was run using the UniProt protein ID’s of the libraries as “gene sets,” with the following settings: number of permutations = 1,000; no collapse; enrichment statistic = weighted; max size for excluding larger sets = 500; min size for exclude smaller sets = 1. The FDR and normalized enrichment scores for each set and trait were obtained from the GSEA output. A similar procedure was used for GSEA of estimates of genetic correlations between proteins and performance and resilience traits, except that proteins were ranked by the signed −log10 of the P-value of the likelihood ratio test for the estimate of the genetic correlation, with the sign reversed if the genetic correlation estimate was negative to provide a direction to the enrichment associations. REACTOME pathway or GO terms with FDR below a chosen threshold for at least one resilience trait (chosen to ensure sufficient terms were available for subsequent analyses) were clustered using the ward. D method (Ward Jr, 1963), separately for the phenotypic and genetic enrichment analyses. This clustering was based on the signed −log10(FDR) of enrichment of these terms or pathways for each trait, where the sign was based on whether an increase in abundance of core proteins in the REACTOME pathway or GO terms were associated with a favorable (+) or an unfavorable (−) change in the trait phenotype based on the estimate of the corresponding regression coefficient or genetic correlation. For this purpose, the signs of the estimates for TRT, MORT, FCR, RFI, and CBF were reversed because lower values are favorable for these traits. The R package ComplexHeatmap (Gu et al., 2016) was used to visualize the GSEA results. ## Results This study used data and samples from the polymicrobial natural disease challenge model (NDCM) for grow-finish pigs described by Putz et al. [ 2019]. Performance and resilience phenotypes collected are summarized in Table 1. Proteome data were obtained on plasma samples collected in a quarantine nursery on 912 pigs, prior to their entry into the disease challenge, as illustrated in Figure 1. The plasma samples were processed in two groups, with details in Table 2. After filtering, 377 proteins that were identified in more than 20 runs were used for further analyses. Distributions of raw protein abundances, of log2 transformed protein abundances, and of residuals of log2 protein abundances adjusted for nuisance effects (see later) are shown in Supplemental Figure S2 for randomly selected proteins. Distributions of the residuals of log2 abundances were close to normal. ## Phenotypic associations of protein abundance with performance and resilience phenotypes Associations of the abundance of individual proteins in plasma of young, healthy pigs with their concurrent and subsequent performance and resilience phenotypes were analyzed and are presented in Figure 2 as a heatmap of the signed −log10 of the P-value of the phenotypic association. For a given protein and trait, the signed −log10 (P-value) was highly correlated with the corresponding scaled regression coefficients of protein abundance on trait phenotype, as shown in Supplementary Figure S3. Table 3 shows proteins that were significantly associated with at least one performance or resilience phenotype at an FDR less than 0.35 across all proteins and analyzed phenotypes. Abundance of several proteins had relatively strong associations with some traits, especially with traits that were recorded during the phase when the samples for proteome analysis were collected, e.g. average daily gain (ADG; see Table 1 for abbreviations) in the quarantine nursery and subjective health scores (HS) taken at two time points in the quarantine nursery (qNurHS1 and qNurHS2). These three phenotypes had relatively similar association patterns with protein abundances (Figure 2). Health score recorded in the challenge nursery had similar association patterns with protein abundances as the first health score in the quarantine nursery (Figure 2). The number of health treatments a pig received during the different phases, i.e. in the challenge nursery (cNurTRT), the finisher (cFinTRT), and across the challenge nursery and finisher (AllTRT), also showed similar association patterns to each other (Figure 2), recognizing that AllTRT has a part-whole relationship with cNurTRT and cFinTRT. Associations with mortality (MOR) during the different periods were rather weak and inconsistent. For traits in the finisher, association patterns for analyses based on data of pigs that survived to slaughter (survivor data) and data that included selected pigs that died (expanded data, see Table 1) were very similar. ## Gene set enrichment analysis of phenotypic associations To overcome the limitation of low statistical power to detect associations of the plasma proteome with performance and resilience traits for individual proteins, gene set enrichment analyses based on GO terms and REACTOME pathways were used to evaluate patterns in associations across proteins for each recorded phenotype. For these analyses, proteins were ranked based on the signed −log10(P-value) of their estimated association with the trait. The level of significance and direction of the enrichment of GO terms and REACTOME pathways that were significantly enriched for at least one trait are shown in Figure 6 for all recorded traits. For this purpose, liberal significance thresholds based on FDR were chosen to allow a meaningful number of terms or pathways to be included in the subsequent clustering analyses. Figure 6.: Heat map of the signed −log10 (FDR) for gene set enrichment analyses with proteins ranked based on the magnitude of the phenotypic association of their abundance with subsequent performance and disease resilience phenotypes for the survivor and expanded (exp) data sets. For trait abbreviations, see Table 1. Red/blue = an increase in expression of core enrichment proteins in this set was associated with better/poor performance. (a) REACTOME pathways (n = 72) that were significantly (FDR < 0.2) enriched among proteins ranked based on the magnitude of the association of their abundance with at least one phenotype trait. (b) GO Biological process (n = 63) that were significantly (FDR < 0.35) enriched among proteins ranked based on the magnitude of the association of their abundance with at least one phenotype trait. Colors on the dendrogram identified different clusters based on ward.D clustering. The 72 REACTOME pathways with FDR < 0.2 for enrichment of phenotypic associations for at least one trait separated into nine clear clusters, of which four (1-green, 3-red, 4-light blue, and 5-pink) were related to the immune system (Figure 6a). These clusters were unfavorably associated with disease resilience phenotypes, especially with health scores and the number of treatments. The REACTOME pathways in cluster 4 (light blue), which included the complement cascade process, had strong unfavorable associations with disease resilience in the finisher period, except with mortality. The REACTOME pathways in clusters 6 (grey), 8 (purple), and 9 (light brown) were primarily associated with metabolism and were favorably associated with phenotypes recorded in the quarantine and challenge nursery but unfavorably associated with phenotypes recorded in the finisher. The 63 GO biological processes (BP) with FDR < 0.35 based on enrichment in phenotypic associations for at least one trait were separated into seven clusters (Figure 6b). The grey cluster contained several GO BPs related to immune response, such as complement activation, and was favorably associated with phenotypes recorded in the quarantine nursery but unfavorably associated with phenotypes recorded in the challenge nursery and finisher, especially with cFinTRT. The 32 immune-related REACTOME pathways that were significantly (FDR < 0.2) enriched in phenotypic associations with disease resilience traits included 35 proteins, and their relationships and associations with all recorded phenotypes are illustrated as a chord diagram in Figure 7a. Among the significantly enriched immune-related pathways, pathways related to the innate immune system were found to be the most abundant [23], then pathways related to the adaptive immune system [5], and to cytokine signaling [4]. All these proteins are involved in the innate immune system and most are also involved in the adaptive immune system and in cytokine signaling. Figure 7.: Chord diagrams of the relationships between proteins and enriched immune-related terms for the survivor data set. For trait abbreviations, see Table 1. The left half of the circle represents proteins and the right half immune-related terms that were enriched among proteins based on their phenotypic or genetic associations with performance and resilience phenotypes. Each layer of the concentric circles represents one numbered phenotype trait. (a) Chord diagram based on estimates of phenotypic associations of 35 proteins and 32 enriched REACTOME pathways (innate immune system: 23; adaptive immune system: 5; cytokine signaling: 4). (b) Chord diagram based on genetic correlation estimates of 23 proteins and 21 enriched REACTOME pathways (innate immune system: 15; adaptive immune system: 5; cytokine signaling: 1) is shown on the right side. Red/blue for the left halves indicates the scaled phenotypic regression coefficients (a) and genetic correlations (b) between protein abundance and phenotypic traits. Red/blue for the right halves indicates the signed −log10 (FDR) to enrich REACTOME pathways. ## Gene set enrichment analysis of genetic correlations Figure 8 shows gene set enrichment results for estimates of genetic correlations for proteins with heritability estimates greater than 0.05. The proteins were ranked by the signed −log10(P-value) of their estimate of the genetic correlation with a trait. In total, 50 REACTOME pathways were significantly (FDR < 0.2) enriched, which separated into eight clusters (Figure 8a). The first three clusters (light green, yellow, and pink) were related to the immune system and had similar genetic relationships with the recorded phenotypes; they were unfavorably associated with disease resilience and performance traits, except with AllTRT (exp), but favorably associated with carcass traits. The REACTOME pathways in the green cluster were related to developmental biology and had low favorable genetic correlations with performance and resilience traits, except with CBF and CLC. The light blue cluster included pathways related to the transport of small molecules and showed favorable genetic correlations with carcass traits. The REACTOME pathways in the last three clusters included metabolism of proteins and signal transduction and showed unfavorable genetic correlations with disease resilience, except with AllTRT in the expanded data. Figure 8.: Heat map of the signed −log10 (FDR) for gene set enrichment analyses with proteins ranked based on magnitude of the phenotypic association of their abundance with subsequent performance and disease resilience phenotypes for the survivor and expanded (exp) data sets. For trait abbreviations, see Table 1. Red/blue = an increase in expression of core enrichment proteins in this set was genetically correlated with better/poor performance. (a) REACTOME pathways (n = 50) that were significantly (FDR < 0.2) enriched among proteins ranked based on the magnitude of the genetic correlation of their abundance with at least one phenotype trait. (b) GO Biological process (n = 49) that were significantly (FDR < 0.2) enriched among proteins ranked based on the magnitude of the genetic correlation of their abundance with at least one phenotype trait. Colors on the dendrogram identified different clusters based on ward.D clustering. The 21 immune-related REACTOME pathways that were significantly (FDR < 0.2) enriched in genetic correlations with disease resilience traits included 15 innate immune system pathways, 5 adaptive immune system pathways, and 1 cytokine signaling pathway. In total, 23 proteins were involved in these pathways (Table 5), and their relationships and genetic correlations with all recorded phenotypes are illustrated as a chord diagram in Figure 7b. All proteins and immune-related pathways shown in Figure 7b overlapped with those identified based on phenotypic associations in Figure 7a. Based on genetic correlations, almost all significantly enriched immune-related pathways were unfavorably correlated with the recorded phenotypes, but the direction of estimates of genetic correlations of the abundance of the proteins involved in these immune-related pathways with the phenotypic traits was not consistent. Table 5. \| Protein ID \| Protein name \| Gene name \| No. of Immune system \| No. of Innate immune system \| No. of adaptive immune system \| No.of cytokine signaling \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| A0A286ZKA5 \| C1q domain-containing protein \| C1QB \| 6 \| 6 \| 0 \| 0 \| \| A0A286ZSJ7 \| Complement C1q subcomponent subunit C (complement C1q subcomponent subunit C isoform 1) \| C1QC \| 6 \| 6 \| 0 \| 0 \| \| A0A286ZTC4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZVT2 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZWN6 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZXN4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287A6Q0 \| Vitamin K-dependent protein S \| PROS1 \| 3 \| 3 \| 0 \| 0 \| \| A0A287A8V0 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AA42 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AG13 \| Apolipoprotein B-100 \| APOB \| 2 \| 2 \| 0 \| 0 \| \| A0A287AKL0 \| Serum amyloid A protein \| \| 2 \| 2 \| 0 \| 0 \| \| A0A287ASW4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AXC8 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287B3W7 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287BFU6 \| Uncharacterized protein \| C1RL \| 6 \| 6 \| 0 \| 0 \| \| A0A287BH90 \| Complement component C9 \| C9 \| 3 \| 3 \| 0 \| 0 \| \| A7YX24 \| Gamma-synuclein \| SNCG \| 6 \| 6 \| 0 \| 0 \| \| F1RL06 \| Ig-like domain-containing protein \| LOC100523213 \| 20 \| 14 \| 5 \| 1 \| \| F1RX35 \| Fibrinogen C-terminal domain-containing protein \| FGG \| 2 \| 2 \| 0 \| 0 \| \| F1S788 \| MACPF domain-containing protein \| C8A \| 3 \| 3 \| 0 \| 0 \| \| F1STC2 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| F1STC5 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| I3L728 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| Figure 8b shows 49 GO BPs that were significantly enriched at FDR < 0.2 for at least one of the recorded phenotypes, which were separated into five clusters. None of the enriched GO BP related to immune response, but one was related to response to virus and response to the stimulus. ## Plasma proteome of young healthy pigs The overall objective of this study was to explore the biological and genetic basis of the abundance of proteins in plasma from young, healthy pigs and investigate their associations with and potential as phenotypic and genetic indicators for disease resilience. Compared with previous studies on plasma proteins in pigs, this is the first study to explore the plasma proteome of young, healthy pigs on a large-scale. Previous studies focused on the genetic basis of specific proteins in the plasma of pigs. For example, Clapperton et al. [ 2009] studied the genetic basis of acute-phase proteins in the plasma of pigs raised in specific-pathogen-free and non-specific-pathogen-free environments, including alpha-acid glycoproteins, C-reactive protein, haptoglobin, and transthyretin. Ballester et al. [ 2020] investigated the genetic parameters of immunoglobulins and acute phase proteins (C-reactive protein and haptoglobin) in plasma from 8-week-old healthy Duroc piglets. The second novelty of this study is that this is the first to investigate phenotypic and genetic associations between the plasma proteome measured in a healthy condition with subsequent performance and resilience phenotypes under a disease challenge. The phenotypic and genetic associations were used to evaluate whether plasma protein levels measured in healthy piglets can be used as phenotypic or genetic indicators or predictors of disease resilience traits following concurrent exposure to multiple diseases, reflecting a severe disease challenge in a commercial environment. ## Genetic basis of the proteome in plasma from young healthy pigs This study is the first to comprehensively investigate the genetic basis of the plasma proteome in young, healthy pigs by combining proteome abundance data with genome-wide SNP data. Abundances of nearly a quarter of the proteins [100] were found to be heritable (FDR < 0.10), with heritability estimates ranging from 0.17 to 1 (Supplemental Table 1). This reflects a strong impact of genetics on the plasma protein profile of young, healthy pigs. Litter effects were significant (FDR < 0.10) for only two proteins (L8B0W9 and L8AXL9, both are IgG heavy chains), explaining $18\%$ and $21\%$ of the phenotypic variance, respectively. The top four heritable proteins were the complement system proteins, with estimates of heritability close to 1, including complement factor H isoform a (A0A480TLF3, $71.7\%$ missing), complement factor I isoform 1 preproprotein (A0A480P4D2, $63.0\%$ missing), complement component C9 (A0A4X1T2W4, $46.7\%$ missing), and complement C5a anaphylatoxin (A0A286ZKB4, $65.2\%$ missing). The IgG heavy chain proteins were also highly heritable, with estimates up to 0.79. In humans, using the mass spectrometry method, Johansson et al. [ 2013] found that the abundances of nearly one-fifth of the more than 1,000 peptides identified in plasma from around 1,000 individuals (at least 15 years of age) were significantly heritable (FDR < 0.05), with estimates ranging from 0.08 to 0.43. Liu et al. [ 2015] applied the SWATH mass spectrometry method to quantify the abundance of 342 proteins in plasma samples from 232 humans (between 38 and 78 years of age) and found that the abundance of 67 proteins had significant heritability (unadjusted P-value < 0.05), with estimates ranging from 0.21 to 0.66. In our study, estimates of heritability of the abundance of significantly heritable proteins in plasma from healthy piglets ranged from 0.17 to 1, i.e. higher than obtained in these previous studies in humans. The difference may be due to differences in the species, in age and uniformity of age when samples were collected, in health status, and the number of samples. Interestingly, consistent with our finding of high heritability estimates of abundances of proteins of the complement system, Johansson et al. [ 2013] found the abundance of a peptide from the complement three protein in human plasma, coded by the C3 gene, to have the highest heritability estimate (0.43). Also, in the study by Liu et al. [ 2015], abundance of complement factor H-related proteins one and three had among the highest heritability estimates, around 0.6, and abundances of other complement proteins, such as C3/C5 convertase (B9TSR8) and C4a anaphylatoxin (B0LFE9), also had higher estimates of heritability, around 0.55. Consistent with our results, Liu et al. [ 2015] found that abundance of the IgG heavy chain protein had a high heritability (0.65) in human plasma. In a pig study, Ballester et al. [ 2020] estimated the heritability of abundance of IgG in plasma to be 0.65, while the estimate of the abundance of the acute phase protein haptoglobin was 0.40. The latter protein was also detected in our study, with an estimate of heritability of its abundance of 0.32. In another pig study, Clapperton et al. [ 2009] found the heritability of the abundance of haptoglobin in serum to be 0.23 in large white pigs raised in non-specific-pathogen-free environments. Our estimate of heritability for abundance of haptoglobin was between the estimates obtained in these two studies. In addition to sampling errors, differences may be due to differences in breeds used, age, health status, sample size, and our use of plasma rather than serum. Monocyte differentiation antigen CD14 was another innate immune-related protein that was identified in our study. The CD14 protein acts as a co-receptor with the toll-like receptor TLR-4 and MD-2 to detect bacterial lipopolysaccharide (Kitchens, 2000; Tapping and Tobias, 2000). The estimate of heritability of abundance of the CD14 protein was 0.17 in our study, which was lower than the estimate of 0.33 in the blood of adult humans (Reiner et al., 2013). Rao et al. [ 2015] found that the heritability of the abundance of apolipoprotein B-containing lipoproteins in blood of adult humans was 0.87, compared to our estimate of 0.44 for the heritability of the abundance of apolipoprotein B coded by the APOB gene. Although different types of apolipoproteins were detected and measured, estimates of heritability of the abundance of apolipoprotein were relatively high in both these studies. The main function of apolipoproteins is to transport lipids to cells in various tissues and their concentration in blood is an important indicator of cardiovascular diseases. Mice lacking apolipoprotein were shown to be more susceptible to bacterial infections (Peterson et al., 2008). ## Phenotypic associations of the proteome in plasma of young healthy pigs with performance and resilience phenotypes In the phenotypic association studies between the abundance of specific proteins in plasma with concurrent and subsequent performance and resilience phenotypes, only a few proteins were identified to be significant after multiple test corrections (FDR < 0.35) because of limited statistical power (Table 3). To overcome this, gene set enrichment analyses were used, borrowing information across proteins involved in the same pathway or biological process. This identified many significantly enriched REACTOME pathways and GO biological processes and led to identification of more potentially relevant proteins (Figure 6), as discussed in the following. In Figure 6, the direction of the identified association of an enriched biological term with a phenotype was indicated to be favorable or unfavorable. A favorable direction of an association indicates that an increase in the abundance of core proteins associated with this term (not necessarily all proteins) was correlated with better performance or resilience. For the phenotypic association studies, immune-related REACTOME pathways (1-green, 3-red, 4-light blue, and 5-pink) and the metabolism of protein REACTOME pathway (6-grey) were unfavorably associated with performance and resilience phenotypes, especially for health scores and the number of health treatments. Immune-related GO biological processes (8-grey) were also unfavorably associated with phenotypes measured in the challenge nursery and finisher, especially with the number of treatments in the finisher. The direction of these results was opposite to expectations, as it is commonly understood that an increase in the abundance of immune-related proteins after exposure is associated with higher disease resilience. For example, Boehmer et al. [ 2011] found that the content of antimicrobial peptides and acute-phase proteins in the alveolar fluid of bovines increased after experimental induction of pneumonia with Mannheimia haemolytica. However, in our study, protein abundance was measured before exposure to disease, which may explain why a greater abundance of core proteins associated with immune-related REACTOME pathways and GO biological processes was associated with lower disease resilience after exposure to disease. A blood transcriptomics study of these same pigs by Lim et al. [ 2021] also suggested that piglets with greater expression of immune-related genes in whole blood before exposure tended to be less resilient after exposure to disease. The immune system pathways that were associated with subsequent disease resilience clustered into three categories based on their biological function, as shown in Figure 7a, i.e. the innate immune system, the adaptive immune system, and cytokine signaling. The host’s innate and adaptive immune mechanisms work together to eliminate pathogens. Innate immunity represents general, nonspecific immunity, and is the first line of defence against non-autogenous pathogens through physical, chemical, and cellular mechanisms. Adaptive immunity, also called acquired immunity, occurs when the animal is exposed to a specific pathogen or is administered a vaccine, and can provide protection from that pathogen with subsequent exposure. Cytokines are a group of signaling proteins that are secreted by immune-related cells and can mediate and regulate the immune system. The proteins involved in these three immune system components are shown in Figure 7a, with details of the proteins provided in Table 6. The chord diagram displays the inter-relationships between proteins and pathways that were significantly enriched with phenotypes by gene set enrichment and shows that the number of significant pathways was largest for the innate immune system. This was as expected because protein levels were measured before the piglets were exposed to the disease challenge. However, several adaptive immune system pathways were significantly associated with subsequent disease resilience based on gene set enrichment, possibly reflecting exposure to minor pathogens or stress from weaning, transportation, mixing, and exposure to a new environment and diet, as also suggested by Lim et al. [ 2021] based on blood transcriptome analyses of these same pigs. Table 6 \| Uniprot ID \| Protein name \| Gene name \| No of immune system \| No. of innate immune system \| No. of adaptive immune system \| No. of cytokine signaling \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| A0A286ZLR0 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZSJ7 \| Complement C1q subcomponent subunit C (complement C1q subcomponent subunit C isoform 1) \| C1QC \| 6 \| 6 \| 0 \| 0 \| \| A0A286ZTC4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZVT2 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZWN6 \| Immunoglobulin-like domain \| \| 20 \| 14 \| 5 \| 1 \| \| A0A286ZXN4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287A6Q0 \| Vitamin K-dependent protein S \| PROS1 \| 3 \| 3 \| 0 \| 0 \| \| A0A287A8V0 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AA42 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AG13 \| Apolipoprotein B-100 \| APOB \| 2 \| 2 \| 0 \| 0 \| \| A0A287ASW4 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A0A287AXC8 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| A2SW51 \| Monocyte differentiation antigen CD14 (myeloid cell-specific leucine-rich glycoprotein) \| CD14 \| 8 \| 8 \| 0 \| 0 \| \| F1RL06 \| Ig-like domain-containing protein \| LOC100523213 \| 20 \| 14 \| 5 \| 1 \| \| F1S788 \| MACPF domain-containing protein \| C8A \| 4 \| 4 \| 0 \| 0 \| \| F1SMJ1 \| Complement component C7 \| C7 \| 4 \| 4 \| 0 \| 0 \| \| F1SS24 \| Fibronectin \| FN1 \| 5 \| 1 \| 0 \| 4 \| \| F1STC2 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| F1STC5 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| I3L728 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| \| P82460 \| Thioredoxin (Trx) \| TXN \| 1 \| 1 \| 0 \| 0 \| \| A0A286ZKA5 \| C1q domain-containing protein \| C1QB \| 6 \| 6 \| 0 \| 0 \| \| A0A286ZMS2 \| Complement C1s subcomponent \| \| 6 \| 6 \| 0 \| 0 \| \| A0A287BH90 \| Complement component C9 \| C9 \| 4 \| 4 \| 0 \| 0 \| \| A0A287A113 \| Spectrin beta chain \| SPTB \| 3 \| 0 \| 0 \| 3 \| \| A7YX24 \| Gamma-synuclein \| SNCG \| 6 \| 6 \| 0 \| 0 \| \| A0A287AFQ4 \| Lipocln_cytosolic_FA-bd_dom domain-containing protein \| C8G \| 4 \| 4 \| 0 \| 0 \| \| A0A287AQ20 \| Complement factor I isoform 1 preproprotein \| CFI \| 3 \| 3 \| 0 \| 0 \| \| F1RQW7 \| C3/C5 convertase (EC 3.4.21.43) (Complement C2, C2a fragment, C2b fragment) \| C2 \| 4 \| 4 \| 0 \| 0 \| \| A0A287BFU6 \| Complement C1r subcomponent \| \| 6 \| 6 \| 0 \| 0 \| \| F1RX35 \| Fibrinogen C-terminal domain-containing protein \| FGG \| 2 \| 2 \| 0 \| 0 \| \| F1S0J2 \| Uncharacterized protein \| C4BPA \| 3 \| 3 \| 0 \| 0 \| \| P26234 \| Vinculin (Metavinculin) \| VCL \| 5 \| 1 \| 0 \| 4 \| \| A0A287AKL0 \| Serum amyloid A protein \| \| 9 \| 8 \| 0 \| 1 \| \| A0A287B3W7 \| Ig-like domain-containing protein \| \| 20 \| 14 \| 5 \| 1 \| Based on the phenotypic association studies, the proteins involved in the significant immune-related pathways fell into two main classes: Ig-like domain-containing proteins and complement system proteins. Ig-like domain-containing proteins participate in all three components of the immune system (innate, adaptive, and cytokine signaling), while complement system proteins only participate in the innate immune system. Abundance of one of the Ig-like domain-containing proteins, F1STC5, as well as its related immune pathways, were found to be significantly unfavorably associated with a health score in the finisher, which suggests that piglets that had a greater abundance of the immune-related protein in plasma before exposure tended to be less resilient following exposure to disease. Transcriptome studies on these same pigs by Lim et al. [ 2021] reached the same conclusion. The complement system plays an important role in the innate immune system and builds a functional bridge between the innate and adaptive immune systems (Dunkelberger and Song, 2010). The complement system is a complex network of plasma and membrane-associated serum proteins. After being activated by pathogens, various complement components have bacteriolytic and cytolytic immune activity, lyse cells, mediate inflammation, regulate phagocytosis and immune response, and clear immune-related complexes (Sarma and Ward, 2011). In addition to Ig-like domain-containing proteins and complement system proteins, monocyte differentiation antigen CD14 and apolipoprotein B-100 were also among the proteins involved in significantly enriched immune-related pathways that were associated with disease resilience (Table 6). However, as individual proteins, these proteins were not significant in the phenotypic association analysis, illustrating the ability of gene set enrichment to identify additional proteins that are potentially relevant. In addition to the proteins referred to above that were identified by gene set enrichment, several other proteins were found to be significantly associated with at least one trait after multiple test corrections (Table 3), most with ADG in the quarantine nursery. Among them, three proteins had moderate heritability (0.3–0.6), which were two uncharacterized proteins (A0A4X1U6L2 and A0A4X1U6T3) and Complement C4 gamma chain (A0A4X1VBD2). Complement C4 is a type of anaphylatoxin that plays an important role in immune response and host defence (Fritzinger et al., 1992). Anaphylatoxin can result in a local inflammatory response by triggering the release of substances by endothelial cells, phagocytes, or mast cells (Gennaro et al., 1986). Studies in humans have shown that complement C4 protein deficiency is related to systemic lupus erythematosus (Hauptmann et al., 1974) and type I diabetes mellitus (Dawkins et al., 1983; Mijovic et al., 1987). In a hepatic fibrosis study, Yang et al. [ 2011] found that complement component 4A in the serum of humans is a biomarker of hepatic fibrosis, with the grade of fibrosis increasing as the level of complement component 4A decreases. Complement C4 has not been studied much in pigs. Combined with results of the aforementioned studies on this protein in humans, the lower level of Complement C4 protein in plasma from pigs may indicate weak immune ability of pigs, which can negatively affect ADG when pigs are subjected to certain stressors such as infection, transport, mixing, and/or adaptation to solid feed post-weaning. In general, it is impossible to determine whether the levels of proteins whose abundance during the quarantine nursery were associated with performance and disease resilience reflect base-line levels in young, healthy pigs, or are the result of response to environmental disturbances during or before quarantine nursery period. Gene set enrichment analyses of blood transcriptome data on these same pigs also identified both immune and stress response-related GO terms to be enriched among genes whose increased expression was unfavorably associated with both pre-and post-challenge traits (Lim et al., 2021). However, that study also identified GO terms related to protein localization and viral gene expression that were enriched among genes that were associated with reduced performance and health traits after but not before the challenge. ## Genetic associations of the proteome in plasma of young healthy pigs with performance and resilience phenotypes This is the first study to estimate genetic correlations of the plasma proteome of young healthy pigs with their concurrent and subsequent performance and disease resilience phenotypes. Although for the latter, phenotypes on over 3,200 pigs were used through genomic relationships with the 912 pigs that had proteome data, only five plasma proteins were identified to have significant non-zero genetic correlations with at least one recorded phenotype after multiple test corrections, because of the high standard errors that are typically associated with estimates of genetic correlations (Table 4). None of these proteins overlapped with the proteins that were found to have significant phenotypic associations with these same traits (Table 3). Genetic correlations were not estimated for proteins with estimates of heritability less than 0.05 because standard errors of estimates of genetic correlations would even be larger for these proteins. As for the phenotypic association studies, gene set enrichment analyses were also applied to the genetic correlation estimates, with proteins ranked based on the signed −log10(P-value) of the genetic correlation estimates with the recorded phenotypes. The pathways and GO biological processes that were significantly enriched based on genetic correlations were similar to those identified in the phenotypic association enrichment analyses. The three immune-related clusters (1-light green, 2-yellow, and 3-pink in Figure 8a) had similar patterns of genetic correlations between the enriched pathways and phenotype traits. The enriched immune-related pathways and metabolism of protein pathways were genetically unfavorably associated with performance and disease resilience traits, especially for traits measured during the quarantine and challenging nursery period, consistent with what we found based on phenotypic associations. However, in contrast to the phenotypic associations, the significantly enriched immune-related pathways were genetically favorably associated with carcass traits. The reason or relevance of this is not clear. All proteins that had significant genetic correlations with one or more phenotypes had moderate to high heritability estimates (0.27–1, Table 4). Among them, the abundance of IgG heavy chain proteins had a strong negative genetic correlation (−0.68, L8B0V2) with ADG in the challenge nursery and a positive genetic correlation (0.85, L8AXM5) with mortality in the challenge nursery. Estimates of the litter effects for these two IgG heavy chain proteins were close to zero and not significant, which means the levels of IgG heavy chain proteins were not related to passive immunity. Further, abundance of complement component C9 was negatively genetically correlated with FCR in the expanded data (−0.24) and also had a very high heritability estimate (h2 = 1). C9 protein is one of the members of the complement system and plays an important role in innate immunity (Lint et al., 1980). When the complement system is activated, the C9 protein is polymerized and forms pores in the target cell membranes, causing cell lysis and death (Dudkina et al., 2016). A deficiency in C9 results in an inability to assemble the membrane attack complex, with a subsequent increase in susceptibility to infection (Zaoutis and Chiang, 2007). Combining the biological functions of the C9 protein and these previous studies, our result that an increase in the abundance of C9 protein (lower susceptibility to infection) in the plasma of young, healthy pigs is genetically associated with better feed efficiency under challenge is plausible. Estimating the genetic correlation between protein abundance and phenotypes, as in our study, is considered the “gold standard” for identifying intermediate phenotypes, such as transcript or protein abundance, that is genetically correlated with a target phenotype (Gusev et al., 2016). Other studies have used so-called transcriptome-wide (Gusev et al., 2016) or proteome-wide association studies (Wingo et al., 2021) to identify such intermediate phenotypes. In TWAS or PWAS, a data set with the intermediate–omics phenotypes and whole-genome SNP genotypes are used to develop genomic predictions for each intermediate–omics phenotype, which are then used to predict the intermediate–omics phenotypes in a data set on individuals that have SNP genotypes and phenotypes for the target trait but no intermediate–omics phenotypes. Then, genetic associations between the intermediate–omics phenotypes and the target phenotype are identified based on the correlation between the genomic prediction for the intermediate–omics phenotype and the outcome phenotype. Although the resulting correlations are the result of genetics that affect both the intermediate and the target phenotype, they are not directly comparable to genetic correlations, as the latter quantify correlations between true genetic values for pairs of traits. In the present study, we used a large data set with SNP genotypes and target phenotypes, of which a subset also had proteomics data as the intermediate phenotypes. For such a data set, direct estimation of genetic correlations using maximum likelihood is well accepted to be the more powerful method to identify genetically correlated traits. ## Conclusions This study combined quantitative analysis of population-level plasma proteome abundance data from blood collected on young, healthy pigs with phenotypes related to performance and disease resilience before and after their exposure to a polymicrobial natural disease challenge. Our results provide novel evidence that there is a genetic basis to differences in the plasma proteome of young, healthy pigs and that some of these differences are associated with performance and disease resilience following exposure, both phenotypically and genetically. Abundances in plasma prior to the disease challenge of proteins in pathways that are related to the immune system, especially the innate immune system, were unfavorably associated with performance and disease resilience after exposure to pathogens, both phenotypically and genetically. These results imply that pigs with unfavorable genetics for disease resilience either have higher base-line immune responses or produce a greater immune response to environmental disturbances that they are exposed to as young, visually healthy pigs. Regardless, our results show that the abundance of proteins in plasma from young, visually healthy pigs have potential as biomarkers for disease resilience and could be incorporated into breeding programs to improve selection for disease resilience. ## Conflict of Interest Statement No conflict of interest, financial, or otherwise is declared by the author(s). Funding organizations were not involved in the execution of the project or the interpretation of results. ## Author’ Contributions GP, JH, JD, MD, PGC, FF conceived and designed the experiment; PGC supplied the animals to the NDCM; FF, JH, GP, MD, JD managed the NDCM; CA, SL measured plasma proteomes in samples, GP directed the genotyping; AP, JC conducted the pre-process of genotype; AP developed measures of resilience; YC performed, KSL consulted on, and JD oversaw the statistical analyses and interpretation of results; YC wrote the manuscript with input from JD. All authors approved the final manuscript. ## References 1. Albers G. A., Gray G. D., Piper L. R., Barker J. S., Le Jambre L. F., Barger I. A.. The genetics of resistance and resilience to. Int. J. Parasitol (1987) 17 1355-1363. DOI: 10.1016/0020-7519(87)90103-2 2. Anderson N. L., Polanski M., Pieper R., Gatlin T., Tirumalai R. S., Conrads T. P., Veenstra T. D., Adkins J. N., Pounds J. G., Fagan R.. The human plasma proteome: a nonredundant list developed by combination of four separate sources. Mol. Cell. Proteomics (2004) 3 311-326. DOI: 10.1074/mcp.M300127-MCP200 3. Ballester M., Ramayo-Caldas Y., Gonzalez-Rodriguez O., Pascual M., Reixach J., Diaz M., Blanc F., Lopez-Serrano S., Tibau J., Quintanilla R.. Genetic parameters and associated genomic regions for global immunocompetence and other health-related traits in pigs. Sci. Rep (2020) 10 18462. DOI: 10.1038/s41598-020-75417-7 4. Bates D., Mächler M., Bolker B., Walker S.. Fitting linear mixed-effects models using lme4.. arXiv (2014). DOI: 10.18637/jss.v067.i01 5. Benjamini Y., Hochberg Y.. Controlling the false discovery rate: a practical and powerful approach to multiple testing.. J. R. Stat. Soc. Series B: Stat. Methodol (1995) 57 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x 6. Bisset S. A., Morris C. A.. Feasibility and implications of breeding sheep for resilience to nematode challenge. Int. J. Parasitol (1996) 26 857-868. DOI: 10.1016/s0020-7519(96)80056-7 7. Boehmer J. L., DeGrasse J. A., Lancaster V. A., McFarland M. A., Callahan J. H., Ward J. L.. Evaluation of protein expression in bovine bronchoalveolar fluid following challenge with Mannheimia haemolytica. Proteomics (2011) 11 3685-3697. DOI: 10.1002/pmic.201000710 8. Bradford M. M.. A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding.. Anal Biochem (1979) 72 248-254 9. Bramer L. M., Irvahn J., Piehowski P. D., Rodland K. D., Webb-Robertson B. M.. A review of imputation strategies for isobaric labeling-based shotgun proteomics. J. Proteome Res (2021) 20 1-13. DOI: 10.1021/acs.jproteome.0c00123 10. Brenes A., Hukelmann J., Bensaddek D., Lamond A. I.. Multibatch TMT reveals false positives, batch effects and missing values. Mol. Cell. Proteomics (2019) 18 1967-1980. DOI: 10.1074/mcp.ra119.001472 11. Buchet A., Belloc C., Leblanc-Maridor M., Merlot E.. Effects of age and weaning conditions on blood indicators of oxidative status in pigs. PLoS One (2017) 12 e0178487. DOI: 10.1371/journal.pone.0178487 12. Cecchinato A., Bobbo T., Ruegg P. L., Gallo L., Bittante G., Pegolo S.. Genetic variation in serum protein pattern and blood β-hydroxybutyrate and their relationships with udder health traits, protein profile, and cheese-making properties in Holstein cows. J. Dairy Sci (2018) 101 11108-11119. DOI: 10.3168/jds.2018-14907 13. Cheng J., Putz A. M., Harding J. C. S., Dyck M. K., Fortin F., Plastow G. S., Canada P., Dekkers J. C. M.. Genetic analysis of disease resilience in wean-to-finish pigs from a natural disease challenge model. J. Anim. Sci (2020) 98. DOI: 10.1093/jas/skaa244 14. Clapperton M., Diack A. B., Matika O., Glass E. J., Gladney C. D., Mellencamp M. A., Hoste A., Bishop S. C.. Traits associated with innate and adaptive immunity in pigs: heritability and associations with performance under different health status conditions. Genet. Sel. Evol (2009) 41 54. DOI: 10.1186/1297-9686-41-54 15. Cruzen S. M., Pearce S. C., Baumgard L. H., Gabler N. K., Huff-Lonergan E., Lonergan S. M.. Proteomic changes to the sarcoplasmic fraction of predominantly red or white muscle following acute heat stress. J. Proteomics (2015) 128 141-153. DOI: 10.1016/j.jprot.2015.07.032 16. Dawkins R. L., Uko G., Christiansen F. T., Kay P. H.. Low C4 concentrations in insulin dependent diabetes mellitus. Br. Med. J. (Clin. Res. Ed.) (1983) 287 839. DOI: 10.1136/bmj.287.6395.839-b 17. Dudkina N. V., Spicer B. A., Reboul C. F., Conroy P. J., Lukoyanova N., Elmlund H., Law R. H., Ekkel S. M., Kondos S. C., Goode R. J.. Structure of the poly-C9 component of the complement membrane attack complex. Nat. Commun (2016) 7 10588. DOI: 10.1038/ncomms10588 18. Dunkelberger J. R., Song W. C.. Complement and its role in innate and adaptive immune responses. Cell Res (2010) 20 34-50. DOI: 10.1038/cr.2009.139 19. Fox J., Weisberg S.. An R companion to applied regression (2019) 20. Fritzinger D. C., Petrella E. C., Connelly M. B., Bredehorst R., Vogel C. W.. Primary structure of cobra complement component C3. J. Immunol (1992) 149 3554-3562. DOI: 10.4049/jimmunol.149.11.3554 21. Gennaro R., Simonic T., Negri A., Mottola C., Secchi C., Ronchi S., Romeo D.. C5a fragment of bovine complement. Purification, bioassays, amino-acid sequence and other structural studies. Eur. J. Biochem (1986) 155 77-86. DOI: 10.1111/j.1432-1033.1986.tb09460.x 22. Gilmour A., Gogel B., Cullis B., Welham S., Thompson R., Butler D., Cherry M., Collins D., Dutkowski G., Harding S.. ASReml user guide release 4.1 structural specification (2014) 5 23. Gu Z., Eils R., Schlesner M.. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics (2016) 32 2847-2849. DOI: 10.1093/bioinformatics/btw313 24. Gusev A., Ko A., Shi H., Bhatia G., Chung W., Penninx B. W., Jansen R., de Geus E. J., Boomsma D. I., Wright F. A.. Integrative approaches for large-scale transcriptome-wide association studies. Nat. Genet (2016) 48 245-252. DOI: 10.1038/ng.3506 25. Hauptmann G., Grosshans E., Heid E.. Lupus erythematosus syndrome and complete deficiency of the fourth component of complement.. Boll. Ist. Sieroter. Milan (1974) 53 228 26. Holdt L. M., von Delft A., Nicolaou A., Baumann S., Kostrzewa M., Thiery J., Teupser D.. Quantitative trait loci mapping of the mouse plasma proteome (pQTL). Genetics (2013) 193 601-608. DOI: 10.1534/genetics.112.143354 27. Holtkamp D. J., Kliebenstein J. B., Neumann E. J., Zimmerman J. J., Rotto H. F., Yoder T. K., Wang C., Yeske P. E., Mowrer C. L., Haley C. A.. Assessment of the economic impact of porcine reproductive and respiratory syndrome virus on United States pork producers.. J. Swine Health Prod (2013) 21 72-84. DOI: 10.2460/javma.2005.227.385 28. Johansson A., Enroth S., Palmblad M., Deelder A. M., Bergquist J., Gyllensten U.. Identification of genetic variants influencing the human plasma proteome. Proc. Natl. Acad. Sci. U. S. A (2013) 110 4673-4678. DOI: 10.1073/pnas.1217238110 29. Kastenmuller G., Raffler J., Gieger C., Suhre K.. Genetics of human metabolism: an update.. Hum. Mol. Genet (2015) 24 R93-R101. DOI: 10.1093/hmg/ddv263 30. Kitchens R. L.. Role of CD14 in cellular recognition of bacterial lipopolysaccharides. Chem. Immunol (2000) 74 61-82. DOI: 10.1159/000058750 31. Lesk A. M.. Introduction to protein architecture: the structural biology of proteins (2001) 32. Lim K. S., Cheng J., Putz A., Dong Q., Bai X., Beiki H., Tuggle C. K., Dyck M. K., Canada P. G., Fortin F.. Quantitative analysis of the blood transcriptome of young healthy pigs and its relationship with subsequent disease resilience. BMC Genomics (2021) 22 614. DOI: 10.1186/s12864-021-07912-8 33. Lin H., Lee E., Hestir K., Leo C., Huang M., Bosch E., Halenbeck R., Wu G., Zhou A., Behrens D.. Discovery of a cytokine and its receptor by functional screening of the extracellular proteome. Science (2008) 320 807-811. DOI: 10.1126/science.1154370 34. Lint T. F., Zeitz H. J., Gewurz H.. Inherited deficiency of the ninth component of complement in man. J. Immunol (1980) 125 2252-2257. DOI: 10.4049/jimmunol.125.5.2252 35. Liu Y., Buil A., Collins B. C., Gillet L. C., Blum L. C., Cheng L. Y., Vitek O., Mouritsen J., Lachance G., Spector T. D.. Quantitative variability of 342 plasma proteins in a human twin population. Mol. Syst. Biol (2015) 11 786. DOI: 10.15252/msb.20145728 36. Mellencamp M., Galina-Pantoja L., Gladney C., Torremorell M.. Improving pig health through genomics: a view from the industry, Animal Genomics for Animal Health No. 132. (2008) 35-41 37. Mijovic C. H., Fletcher J. A., Bradwell A. R., Barnett A. H.. Low C4 levels in type 1 (insulin-dependent) diabetes. Diabetologia (1987) 30 824. DOI: 10.1007/BF00275752 38. Misztal I., Tsuruta S., Strabel T., Auvray B., Druet T., Lee D.. (2002) 39. Montagne L., Gilbert H., Muller N., Le Floc’h N.. Physiological response to the weaning in two pig lines divergently selected for residual feed intake. J. Anim. Physiol. Anim. Nutr (2022) 106 802-812. DOI: 10.1111/jpn.13622 40. Morgan N., Prakash A.. International livestock markets and the impact of animal disease. Rev. Sci. Tech (2006) 25 517-528. DOI: 10.20506/rst.25.2.1685 41. Muk T., Stensballe A., Pankratova S., Nguyen D. N., Brunse A., Sangild P. T., Jiang P. P.. Rapid proteome changes in plasma and cerebrospinal fluid following bacterial infection in preterm newborn pigs. Front. Immunol (2019) 10 2651. DOI: 10.3389/fimmu.2019.02651 42. Nguyen-Ba H., Van Milgen J., Taghipoor M.. A procedure to quantify the feed intake response of growing pigs to perturbations. Animal (2020) 14 253-260. DOI: 10.1017/s1751731119001976 43. Peterson M. M., Mack J. L., Hall P. R., Alsup A. A., Alexander S. M., Sully E. K., Sawires Y. S., Cheung A. L., Otto M., Gresham H. D.. Apolipoprotein B is an innate barrier against invasive. Cell Host Microbe (2008) 4 555-566. DOI: 10.1016/j.chom.2008.10.001 44. Putz A. M., Harding J., Dyck M. K., Fortin F., Plastow G. S., Dekkers J.. Novel resilience phenotypes using feed intake data from a natural disease challenge model in wean-to-finish pigs.. Front. Genet (2019) 660. DOI: 10.3389/fgene.2018.00660 45. Rao F., Schork A. J., Maihofer A. X., Nievergelt C. M., Marcovina S. M., Miller E. R., Witztum J. L., O’Connor D. T., Tsimikas S.. Heritability of biomarkers of oxidized lipoproteins: twin pair study. Arterioscler. Thromb. Vasc. Biol (2015) 35 1704-1711. DOI: 10.1161/atvbaha.115.305306 46. Reiner A. P., Lange E. M., Jenny N. S., Chaves P. H., Ellis J., Li J., Walston J., Lange L. A., Cushman M., Tracy R. P.. Soluble CD14: genomewide association analysis and relationship to cardiovascular risk and mortality in older adults. Arterioscler. Thromb. Vasc. Biol (2013) 33 158-164. DOI: 10.1161/atvbaha.112.300421 47. Reyer H., Oster M., Wittenburg D., Murani E., Ponsuksili S., Wimmers K.. Genetic contribution to variation in blood calcium, phosphorus, and alkaline phosphatase activity in pigs. Front. Genet (2019) 10 590. DOI: 10.3389/fgene.2019.00590 48. Sarma J. V., Ward P. A.. The complement system. Cell Tissue Res (2011) 343 227-235. DOI: 10.1007/s00441-010-1034-0 49. Subramanian A., Tamayo P., Mootha V. K., Mukherjee S., Ebert B. L., Gillette M. A., Paulovich A., Pomeroy S. L., Golub T. R., Lander E. S.. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A (2005) 102 15545-15550. DOI: 10.1073/pnas.0506580102 50. Suhre K., McCarthy M. I., Schwenk J. M.. Genetics meets proteomics: perspectives for large population-based studies. Nat. Rev. Genet (2021) 22 19-37. DOI: 10.1038/s41576-020-0268-2 51. Tapping R. I., Tobias P. S.. Soluble CD14-mediated cellular responses to lipopolysaccharide. Chem. Immunol (2000) 74 108-121. PMID: 10608084 52. te Pas M. F., Koopmans S. J., Kruijt L., Calus M. P., Smits M. A.. Plasma proteome profiles associated with diet-induced metabolic syndrome and the early onset of metabolic syndrome in a pig model. PLoS One (2013) 8 e73087. DOI: 10.1371/journal.pone.0073087 53. Thompson A., Schafer J., Kuhn K., Kienle S., Schwarz J., Schmidt G., Neumann T., Johnstone R., Mohammed A. K., Hamon C.. Tandem mass tags: a novel quantification strategy for comparative analysis of complex protein mixtures by MS/MS. Anal. Chem (2003) 75 1895-1904. DOI: 10.1021/ac0262560 54. Visscher P. M.. A note on the asymptotic distribution of likelihood ratio tests to test variance components. Twin Res. Hum. Genet (2006) 9 490-495. DOI: 10.1375/twin.9.4.490 55. Ward J. H.. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc (1963) 58 236-244. DOI: 10.1080/01621459.1963.10500845 56. Williams S. A., Kivimaki M., Langenberg C., Hingorani A. D., Casas J. P., Bouchard C., Jonasson C., Sarzynski M. A., Shipley M. J., Alexander L.. Plasma protein patterns as comprehensive indicators of health. Nat. Med (2019) 25 1851-1857. DOI: 10.1038/s41591-019-0665-2 57. Wingo A. P., Liu Y., Gerasimov E. S., Gockley J., Logsdon B. A., Duong D. M., Dammer E. B., Robins C., Beach T. G., Reiman E. M.. Integrating human brain proteomes with genome-wide association data implicates new proteins in Alzheimer’s disease pathogenesis. Nat. Genet (2021) 53 143-146. DOI: 10.1038/s41588-020-00773-z 58. Yang L., Rudser K. D., Higgins L., Rosen H. R., Zaman A., Corless C. L., David L., Gourley G. R.. Novel biomarker candidates to predict hepatic fibrosis in hepatitis C identified by serum proteomics. Dig. Dis. Sci (2011) 56 3305-3315. DOI: 10.1007/s10620-011-1745-4 59. Zaoutis L. B., Chiang V. W.. Comprehensive pediatric hospital medicine (2007) 60. Zhu L., Zhao K., Chen X., Xu J.. Impact of weaning and an antioxidant blend on intestinal barrier function and antioxidant status in pigs. J. Anim. Sci (2012) 90 2581-2589. DOI: 10.2527/jas.2011-4444
--- title: Median eminence blood flow influences food intake by regulating ghrelin access to the metabolic brain authors: - Nicola Romanò - Chrystel Lafont - Pauline Campos - Anne Guillou - Tatiana Fiordelisio - David J. Hodson - Patrice Mollard - Marie Schaeffer journal: JCI Insight year: 2023 pmcid: PMC9977422 doi: 10.1172/jci.insight.165763 license: CC BY 4.0 --- # Median eminence blood flow influences food intake by regulating ghrelin access to the metabolic brain ## Abstract Central integration of peripheral appetite-regulating signals ensures maintenance of energy homeostasis. Thus, plasticity of circulating molecule access to neuronal circuits involved in feeding behavior plays a key role in the adaptive response to metabolic changes. However, the mechanisms involved remain poorly understood despite their relevance for therapeutic development. Here, we investigated the role of median eminence mural cells, including smooth muscle cells and pericytes, in modulating gut hormone effects on orexigenic/anorexigenic circuits. We found that conditional activation of median eminence vascular cells impinged on local blood flow velocity and altered ghrelin-stimulated food intake by delaying ghrelin access to target neurons. Thus, activation of median eminence vascular cells modulates food intake in response to peripheral ghrelin by reducing local blood flow velocity and access to the metabolic brain. ## Introduction Adaptation of food intake to energy requirements is dependent on the ability of different brain centers to rapidly sense and integrate changes in blood levels of peripheral molecules. Although the regulation of communications between peripheral circulation and central integration compartments may play an integral part in controlling feeding behavior and energy balance [1], the mechanisms involved in this process remain poorly understood. Many appetite-regulating peptides exclusively produced in the periphery are able to act on specific target sites in the brain to regulate food intake [2]. The hypothalamic arcuate nucleus (ARH) constitutes the primary integration center [3, 4], where circulating factors act on functionally opposing neuronal populations to coordinate homeostatic responses and metabolism [4, 5]. Thus, both access rate and distribution patterns of peripheral gut hormones into the ARH constitute key regulators of food intake. Although neurons are usually protected by the blood-brain barrier (BBB), the ARH is located in close proximity to the capillary plexus of the median eminence (ME), a key circumventricular organ (CVO). While ME tanycytes can actively transport blood-borne signals to regulate food intake at relatively slow kinetics [6], nutrient-dependent regulation of fast molecule access to the metabolic brain implicates fenestrated endothelial capillary loops (7–9). ME capillaries allow passive extravasation into the ARH of small circulating hormones (≤40 kDa in size) [7] and undergo profound remodeling in response to fasting-induced decrease in blood glucose in mice [8]. Such remodeling involves vascular endothelial growth factor A–dependent (VEGFA-dependent) increases in ARH capillary fenestration and number of capillary loops, which promote rapid delivery of circulating appetite-regulating hormones (e.g., ghrelin) [8]. Of note, VEGFA-dependent mechanisms are likewise involved in remodeling of the ME vasculature according to seasonal changes in energy requirements to aid central-peripheral communications [10]. Thus, plasticity of hormone uptake mechanisms in the ARH plays a key role in the adaptive response to metabolic changes. As such, understanding the mechanisms that regulate molecule access to the key neuronal determinants of feeding behavior in the ARH, and potentially other CVOs, may be important for therapeutic development. Among other mechanisms that influence access rate and distribution of peripheral gut hormones into the ARH, the role of capillary blood flow velocity at the level of the ME remains unclear. The ME microvasculature is composed of arterioles and capillaries, lined by contractile smooth muscle cells and pericytes, respectively [11]. Pericytes have attracted increasing interest due to their role in regulating capillary blood flow (12–14), BBB integrity [15], and vascular permeability to circulating leptin [16]. In addition, pericytes are implicated in the sustained blood flow decrease in the brain following ischemia, stroke [17], and development of pathologies such as cerebral arteriopathy [18, 19] and diabetic retinopathies [20]. As such, we hypothesized that modulation of ME blood flow might be able to modulate food intake through regulation of peripheral molecules’ access to the ARH, thus providing an additional level of regulation in gut-brain crosstalk. Here, using selective optogenetic manipulation of ME mural cells and ghrelin injection as a food intake trigger in mice, we demonstrate that short-lived changes in ME blood flow are able to alter peptide hormone access to the ARH as well as food intake. ## Optogenetic activation of ME NG2-positive cells reduces local blood flow velocity through vessel constriction. Neuroglial 2 (NG2) is a proteoglycan that is involved in cell adhesion, communication, and migration [21]. NG2 is mainly expressed in perivascular cells in the ME but is essentially absent in endothelial cells [22, 23] (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.165763DS1). NG2-positive cell coverage at the level of the ME was assessed using mice expressing DsRed under the control of the NG2 promoter (NG2DsRed mice), which present homogeneous labeling of both cytoplasmic and cell protrusions in NG2-positive cells [12, 24, 25]. NG2DsRed cells lined the dense capillary network of the ME (Figure 1A), as well as the fenestrated capillary loops projecting within the ME (Figure 1B) [7]. NG2DsRed cells in the ME also expressed both the platelet-derived growth factor receptor–β (PDGFR-β), a marker commonly used to identify pericytes [15], and the contractile smooth muscle actin protein (α-SMA) (Figure 1, C and D). Of note, arteriolar smooth muscle cells also express α-SMA and NG2 (26–28). To allow precise functional interrogation of NG2-positive cell activity, the blue light–sensitive cation channel ChR2 was expressed in a Cre-dependent manner in NG2-Cre animals. As expected, a similar anatomical distribution of NG2-positive cells was seen at the ME level in NG2-Cre ROSA26-ChR2-tdTomato (NG2-ChR2) mice (Figure 2A). tdTomato labeling was, however, limited to the plasma membrane of NG2-positive cells in these mice. Blood flow was imaged after removal of the jaw bone in terminally anesthetized mice as reported [7], whereas laser stimulation was performed using optic fibers pointed at the level of the ME (Figure 2B). Vascular plasma was labeled using fluorescent dextran (Figure 2C) and blood flow recorded over 1 hour of laser stimulation in vivo. Optogenetic stimulation was performed using a 1 Hz stimulation with 50 ms pulses of 473 nm laser light. Light stimulation resulted in a prompt decrease in blood flow in NG2-ChR2 mice (Supplemental Video 1). After an initial large drop in flow velocity, blood flow showed a slight increase, without reaching the velocity observed under unstimulated conditions. In addition, the overall blood flow remained significantly reduced during 1 hour of laser stimulation ($P \leq 0.05$, 2-way ANOVA) (Figure 2D). Blood flow was not affected by laser stimulation in control littermates lacking ChR2 (Figure 2D). Notably, the effects of optogenetic stimulation were readily reversible, with red blood cell (RBC) velocity returning to baseline levels after switching off the laser (Figure 2D). The decrease in blood flow was associated with contraction of vessels in NG2-ChR2 mice (ranging from ~$5\%$ to $40\%$ reduction in vessel diameter, mean ~$10\%$; Figure 2, E–G, and Supplemental Video 2), with a stronger effect on larger vessels than small capillaries. Thus, conditional activation of ME NG2-positive cells impinges on local blood flow velocity, presumably through vessel constriction. ## Optogenetic activation of ME NG2-positive cells reduces and delays ghrelin-stimulated food intake. To determine whether activation of NG2-positive cells at the ME modulates peripheral hormone sensing in the metabolic brain, optical fibers were implanted above the ME in either NG2-ChR2 mice or control littermates for optogenetic stimulation in freely moving animals. Ghrelin-stimulated hourly food intake was measured multiple times in individual mice, before and after fiber implantation, with or without laser stimulation, in randomized order between different animals (Figure 3A). Ghrelin injection resulted in a rapid and significant increase in food intake in both NG2-ChR2 and control mice (Supplemental Figure 2A), and total food intake was proportional to the number of eating episodes (Supplemental Figure 2B). Laser stimulation reduced the total number of ghrelin-stimulated eating episodes in NG2-ChR2 mice but not in control animals (Figure 3, B and C, $P \leq 0.05$, mixed-effect model). Furthermore, laser stimulation induced a significant delay (~3 minutes) in the lag to first eating episode after ghrelin injection in NG2-ChR2 animals compared with control animals (Figure 3D, $P \leq 0.01$, mixed-effect model). In contrast, laser stimulation of NG2-ChR2 mice under basal conditions (i.e., without ghrelin stimulation) did not modify feeding behavior compared to control littermates (Supplemental Figure 3). ## Optogenetic activation of ME NG2-positive cells impinges on peripheral molecule access to the metabolic brain. To quantify the effects of ME NG2-positive cell activation on peripheral molecule access to the metabolic brain, first we used a fluorescent ghrelin analog that is internalized by target neurons in the ARH upon growth hormone secretagogue receptor (GHSR1a) binding [7]. Fluorescent ghrelin was injected into the tail vein following 30 minutes of light stimulation (50 ms, 1 Hz pulses), before sacrificing animals and quantifying the total number of ghrelin-positive neurons in the ME/arcuate region (5–10 minutes postinjection) (Figure 4A). Light stimulation significantly reduced the number of neurons labeled with fluorescent ghrelin in NG2-ChR2 animals compared with their control littermates (~$30\%$ reduction; Figure 4, B and C; $P \leq 0.05$, Mann-Whitney test). Second, we assessed the diffusion of a fixable fluorescent inert sugar (10 kDa dextran-FITC) within the size range of most peptide hormones. Fluorescent dextran was injected into the tail vein following 30 minutes of light stimulation (50 ms, 1 Hz pulses), before sacrificing animals and quantifying molecule diffusion area in the ME/arcuate region (5–10 minutes postinjection). Light stimulation significantly reduced the diffusion area of fluorescent dextran in NG2-ChR2 animals compared with their control littermates (Figure 4, D and E, $P \leq 0.001$, Mann-Whitney test). Thus, activation of ME NG2-positive cells modulates food intake in response to peripheral ghrelin by reducing local blood flow velocity and access to target sites, rather than relying on an active uptake mechanism. More generally, activation of ME NG2-positive cells modulates access of circulating molecules to the ME/ARH. To test effects of ME NG2-positive cell manipulation on neuron functional activation in the metabolic brain, we injected ghrelin following 30 minutes of light stimulation (50 ms, 1 Hz pulses), sacrificed animals 2 hours postinjection, and quantified c-Fos–positive nuclei in ME/ARH slices (Supplemental Figure 4). Light stimulation did not significantly modify the number of c-Fos–labeled nuclei in NG2-ChR2 animals compared to their control littermates (Supplemental Figure 4, 1-way ANOVA). ## Discussion Rapid sensing and integration of changes in circulating appetite-modifying peptide levels by the metabolic brain ensure maintenance of energy homeostasis. Key neuronal determinants of feeding behavior in the ARH are strategically located in close apposition to the ME, a CVO, which dynamically regulates molecule access to the metabolic brain in a nutrient-dependent manner [29]. Since mechanisms involved in peripheral hormone access to the ARH may be important for developing new antiobesity therapies, we sought to investigate the role of contractile vascular cells in modulating molecule access to the metabolic brain. To do so, we combined optogenetic manipulation of ME NG2-positive cells with ghrelin-stimulated food intake. NG2-positive cell activation in the ME induced a local reduction in blood flow velocity, decreased and delayed ghrelin-stimulated food intake, and delayed ghrelin access to the ARH. We thus demonstrate that activation of NG2-positive mural cells in the ME alters food intake in response to ghrelin injection in the periphery, through a reduction of blood flow velocity, which delays ghrelin access to target neurons in the ARH. NG2 is expressed by different cell types in the ARH. First, pericytes wrap around nonfenestrated brain endothelial cells, contributing to the endothelial barrier properties of the central nervous system (CNS) (16, 30–32). CNS pericytes are a heterogeneous cell population [33], commonly identified by the NG2 and PDGFR-β markers [34]. In addition, α-SMA labels contractile pericytes on capillaries [12, 27, 28]. Second, NG2 is expressed in morphologically distinct arteriolar smooth muscle cells, also expressing α-SMA [14, 35]. Finally, NG2 is expressed in oligodendrocyte precursor cells (OPCs), also called NG2-glia [36]. Therefore, optogenetic stimulation of NG2-ChR2–expressing cells using a 200 μm diameter fiber optic aimed at the ventral ME likely targets OPCs, arteriolar α-SMA–positive cells, and pericytes and does not target endothelial cells, which do not express (or express very low levels of) NG2 [23]. It has been shown that pericytes are electrically responsive, enabling optogenetic control of their activity [14, 27, 28, 35]. In accordance with a role for pericytes in blood flow regulation described previously in the cerebellum [13] and the cortex [12], manipulation of NG2-positive cell activity through optogenetic stimulation in vivo was able to modify blood flow in the ME. Very small ($5\%$–$10\%$) and sometimes unnoticeable variations in vessel diameter were able to lead to great changes in capillary blood flow (~$50\%$). Since capillary diameters are usually similar to RBC width, minor vessel constrictions resulting from pericyte and/or arteriolar α-SMA–positive cell activity can lead to relatively large increases in friction and resistance to RBC flow. In addition, it has been suggested that contraction of pericyte longitudinal processes could alter capillary blood flow without changes in diameter, through stiffening of the vascular wall [27]. It is, however, unclear how activation of OPCs might alter blood flow. Optogenetic stimulation of ME NG2-positive cells delayed the time of the first ghrelin-stimulated eating episode by approximately 3 minutes, which was accompanied by an overall decrease in food intake over the 1-hour experiment time. This decrease in food intake could not solely be explained by the delay in ghrelin-stimulated feeding. Indeed, the $5\%$ reduction in time during which mice ate over an hour (corresponding to the 3-minute delay) is unlikely to be entirely responsible for the $30\%$ reduction in number of feeding episodes observed in laser-stimulated NG2-ChR2 compared with control mice. Although reduced blood flow in the ME may delay ghrelin access to the metabolic brain and thereby increase the lag to first eating episode, alteration of ghrelin delivery to the ARH may also lead to prolonged effects on GHSR1a signaling, altering overall food intake during the 1-hour period after ghrelin injection. Indeed, previous data on ghrelin-GHSR1a stimulation demonstrated that the most robust response was observed when the triggering stimulus was delivered as a bolus rather than an infusion [37, 38]. Although 2 hours of ghrelin infusion was unable to induce a change in c-Fos activation in the present study, we note that longer time scales might be necessary to invoke ARH neuron activation [39], as well as allow for other slower receptor-mediated transport mechanisms or transcytosis mechanisms to be triggered [40]. Further studies with high spatiotemporal resolution are required to analyze early changes in neuronal activation, e.g., fiber photometry or 2-photon imaging of calcium activity or voltage dynamics in defined ARH neuronal populations. Together, these observations suggest that vascular cells modulate the specific temporal pattern of ghrelin signaling in the ARH, through fine control of the concentration and length of exposure to the hormone, mediated by changes in blood flow. Although reduced blood flow might locally impact oxygen and nutrient delivery, gasses and small molecules are highly diffusible and rely less on vessel fenestration for diffusion [41]. Consistent with this, laser stimulation of NG2-ChR2 mice under basal conditions (i.e., without ghrelin stimulation) did not modify feeding behavior compared to control littermates, suggesting neurons receive sufficient oxygen and nutrient supply through vessels unaffected by laser stimulation. Further (challenging) experiments measuring local in vivo oxygen tension would help to clarify this issue. Ghrelin was used in the present study since it is an acute and rapid trigger of the feeding response in the fasted state (within minutes) [37]. While it would be of interest to study the impact of ME blood flow on the entry of other hormones into the ARH, inhibition of food intake by leptin or glucagon-like peptide 1 receptor agonist injection, for instance, may require a much longer observation time and continuous laser stimulation, since feeding episodes occur mainly during the nocturnal phase and are relatively spread out. In addition, while fenestrated capillaries of ME provide a route of entry for leptin [8], effects on the ARH are largely mediated by leptin transported from the cerebrospinal fluid through tanycytes in a receptor-mediated process [42]. Thus, coexistence of different transport mechanisms acting in conjunction and in relay may add to the complexity of analyzing leptin effects. Together with the recently described role for pericytes and OPCs in the modulation of leptin signaling in the hypothalamus [16, 36], our data support a central role for vascular mural cells, in addition to tanycytes, in modulating feeding behavior and possibly as a general gatekeeper to modulate hormone entrance to the brain. Consistent with an effect of reduced blood flow in the ME on altered ghrelin access to the metabolic brain, and GHSR1a-ligand complex internalization by endocytosis in a time-dependent manner [38], fewer (~$30\%$) ghrelin-positive neurons could be observed in the ME/arcuate region in laser-stimulated NG2-ChR2 mice compared with control mice 5–10 minutes postinjection of a bioactive fluorescent ghrelin. However, the location of labeled neurons was still restricted to ARH regions, where the main ghrelin-sensitive food intake–regulating neurons reside [4, 5], as previously shown [7]. Reduced/delayed activation of ARH neurons may contribute to the lasting effects of laser stimulation on food intake over the 1-hour experiment since the number of ghrelin-positive neurons was still reduced in NG2-ChR2 compared with control mice 5–10 minutes following bioactive fluorescent ghrelin injection (i.e., a time point beyond the lag to first ghrelin-stimulated feeding episode). We cannot exclude that changes in vascular permeability contribute to the reduction in ghrelin-labeled cell number, since pericytes are able to modulate vessel permeability [16, 43]. However, it is unlikely that changes in permeability would further affect extravasation since ghrelin already has a very high extravasation rate in the ME because of its low molecular weight [7]. In addition, control of vessel permeability by pericytes has been described in vessels outside the ARH, and it is not clear whether modulation of pericyte activity could reduce permeability in fenestrated vessels. Supporting a direct role of pericytes in the control of molecule extravasation into the ME/ARH, the extent of diffusion of fixable 10 kDa dextran-FITC was reduced in laser-stimulated NG2-ChR2 mice versus control mice 5–10 minutes postinjection [7]. However, further experiments are warranted to differentiate effects of reduced blood flow and vascular permeability on molecule diffusion. In summary, we have shown that in vivo optogenetic activation of ME NG2-positive cells modulates ghrelin-stimulated food intake by reducing local blood flow velocity and ghrelin access to the target site, thus unveiling a new angle of metabolic regulation with important implications for therapeutic design. Future studies are warranted to assess ME vascular mural cells’ function in the context of metabolic diseases to evaluate their potential as a therapeutic target. ## Mice. C57BL/6 mice were purchased from Janvier-SAS. NG2DsRed, NG2-Cre, and ROSA26-ChR2 (H134R)-tdTomato mice [44] on a C57BL/6 background were sourced from The Jackson Laboratory. In all experiments, 8- to 15-week-old mice were used. ROSA26-ChR2-Tomato mice were crossed to NG2-Cre mice to allow specific expression of light-sensitive ion channels in NG2-expressing cells, a marker commonly used to identify pericytes [25], to impose a firing tempo by depolarization, opening of voltage-activated calcium channels, and pericyte contraction using laser flashes (473 nm) [12]. Controls consisted of a combination of littermate WT, NG2-Cre, and ROSA26-ChR2-Tomato mice, as results obtained in these 3 genotypes were not different. ## Intravital imaging of blood flow. To analyze blood flow dynamics in vivo in response to pericyte stimulation in the ME, mouse ME was exposed by surgery as described [7, 45]. Briefly, animals were anesthetized by injection of ketamine/xylazine ($\frac{0.1}{0.02}$ mg/g), then placed on a heating pad. Heart rate was monitored continuously and respiration was controlled by tracheotomy. The ventral side of the brain was exposed by drilling a hole in the palate bone and superfused continuously with NaCl $0.9\%$ solution. Vessels were labeled by i.v. injection of fluorescent dextran molecules (labeled with D2, 150 kDa, 25 mg/mL in NaCl $0.9\%$, 100 μL/20 g body weight), and fluorescence was captured using an epi-fluorescence microscope fitted with a fast sCmos camera (ORCA Flash4.0, Hamamatsu) and a long working distance objective (2 cm, Mitutuyo, M Plan Apo ×20, NA 0.4) [45]. Simultaneously, pericyte electrical activity was controlled in vivo using computer-controlled light flashes (50 ms, 1 Hz, 473 nm). Light was delivered to the tissue using optic fiber of 200 μm diameter placed at a maximum distance of 0.5 mm. ## Image data analysis. Single Z-plane 1-minute movies were acquired every 5 minutes during 60 minutes. Blood flow in basal conditions was measured for 10 minutes before switching on the laser. A movie was then acquired every 5 minutes during 40 minutes of continuous laser stimulation (50 ms flashes at a frequency of 1 Hz). Acquisition rate was set to 150 frame/s. Movies were stabilized using ImageJ (NIH). Measurements were performed in at least 3 independent experiments per condition. The displacement of RBC shadows was visualized by contrast and quantified using previously described image analysis methods to calculate RBC velocity [45]. RBC velocities were measured in 5–10 vessels per movie in at least 3 different mice per condition. Changes in vessel diameter induced by optogenetic stimulation of pericytes were measured using ImageJ (line scan function). Five to 10 vessels were measured prior to laser stimulation and 60 seconds after switching on the continuous laser stimulation in at least 3 independent movies from 3 different mice per condition. ## Fabrication of optic fibers. Optic fiber implants were built by gluing a multimodal optic fiber (Thorlabs) with 200 μm core into a ceramic ferrule using high-temperature curing epoxy (Precision Fiber Products) as previously described [46]. Fibers were polished and tested before implantation, to ensure optimal light transmission. ## Implantation of optic fibers. Mice were anesthetized with 10 μL/g body weight of a mix of $1\%$ ketamine and $0.1\%$ xylazine in $0.9\%$ NaCl. The head of the mouse was then blocked onto a stereotaxic frame and sterilized with ethanol and $10\%$ betadine. Optical gel (Lacrigel, Europhta) was used to prevent dehydration of the eyes. A sagittal incision was made through the skin to expose the cranium, and then the head was positioned so that the sagittal suture lay horizontally in the rostro-caudal axis. After performing a small craniotomy using a dentist’s drill at the implantation site, 3 jeweler’s screws (Plastics One) were fixed to the skull to improve stability of the head cap. The optic fiber was then lowered through the brain to reach the stereotaxic coordinates (relative to bregma) –1.3 mm rostro-caudal, 0 mm medio-lateral, 5.7 mm ventral and finally fixed to the skull with Dentalon dental acrylic (Phymep). The mouse was left to recover in its cage for at least a week before initiating the experiments. Postoperatory analgesia was provided as an i.m. injection of ketoprofen. After surgery, animals were housed separately and checked daily. To minimize intra-animal variability, repeated analyses were made on the same mice, leaving at least 1 week between each experiment. ## In vivo peptide treatment. Single-housed mice were attached to the stimulating laser through an optic fiber attached to a swivel and left to acclimate 3–4 hours. Ghrelin (rat, mouse, Phoenix Pharmaceuticals) was then injected i.p. ( 1 μg/g mouse), and food intake was measured over the course of 1 hour. Number of feeding episodes (manually counted as number of bites taken from food pellets) and total amount of food consumed (measured by weighing food pellets before ghrelin injection and 1 hour after) were monitored. To assess in vivo diffusion of fluorescent molecules, either fixable 10 kDa dextran-FITC (MilliporeSigma) or bioactive fluorescent ghrelin derivative (coupled to a red fluorescent probe) developed by Cisbio Bioassays in collaboration with the Institut des Biomolécules Max Mousseron, Montpellier, France [7, 47], was used. Fixable 10 kDa dextran-FITC (3 mg/animal) or red ghrelin (25 nmoles/animal) was injected i.v. into the tail vein of either control (WT, NG2-Cre, or ROSA26-ChR2-Tomato mice) or NG2-ChR2 mice that were implanted with optic fibers at least 1 week before, fasted for 24 hours, left to habituate with optic fibers 3–4 hours, and subjected to 30 minutes of continuous laser stimulation (50 ms, 1 Hz, 200 μm fiber, 473 nm). Laser stimulation was continued and animals were sacrificed 5–10 minutes after ghrelin injection. Images were acquired using the same parameters between the different groups, and brightness and contrast were adjusted to the same levels between all images. Areas of diffusion of fixable 10 kDa dextran-FITC, corresponding to areas in which green fluorescent signal was detectable, were measured using ImageJ in image projections of the same thickness, or numbers of ghrelin-positive cells were counted, respectively. To test for c-Fos upregulation in vivo, 0.9 % NaCl or commercial rat ghrelin (25 nmol/mouse, rat, mouse; Phoenix Pharmaceuticals) was injected in either control (WT, NG2-Cre, or ROSA26-ChR2-Tomato mice) or NG2-ChR2 mice that were implanted with optic fibers at least 1 week before, left to habituate with optic fibers 3–4 hours, and subjected to 30 minutes of continuous laser stimulation (50 ms, 1 Hz, 200 μm fiber, 473 nm). Laser stimulation was continued and animals were sacrificed 2 hours minutes after ghrelin injection. c-Fos–positive nuclei were counted using ImageJ in 20 μm projection of ME/ARH slices. ## Confocal imaging. Terminally anesthetized mice were perfused via the heart with 10 mL of PBS followed by 30 mL of $4\%$ paraformaldehyde solution. In some experiments, vessels were labeled using FITC-labeled lectin (400 μg per mouse, Vector Laboratories) diluted in the perfusate (PBS). Brains were collected and prepared for confocal imaging [7]. Antibodies used were anti–PDGFR-β (1:200, R&D Systems, Bio-Techne; reference AF1042), anti–α-SMA (1:200, Abcam, reference ab5694), and anti–c-Fos (1:1,000, Santa Cruz Biotechnology, reference sc-166940). Nuclei were labeled using DAPI (MilliporeSigma). One to 4 slices were randomly selected from more than 3 animals/group. Images were acquired using a Zeiss LSM 780 confocal microscope and analyzed using Imaris (Bitplane) and ImageJ (NIH). ## Single-cell RNA-Seq data analysis. Data from Campbell et al. 2017 [23] were imported in R 4.1.3 from the scRNAseq Bioconductor package version 2.7.2, then converted to a Seurat object, for plotting using the Seurat package, version 4.1.1. The cell labels were imported from the metadata and renamed according to what was reported in the original article. ## Statistics. Values are represented as mean ± SEM. Statistical tests were performed using GraphPad Prism. Normality was tested using D’Agostino-Pearson test, and comparisons were made using either unpaired 2-tailed Student’s t test, or 2-tailed Mann-Whitney U test, as appropriate. Multiple comparisons were made using 1-way or 2-way ANOVA followed by Bonferroni’s post hoc test. P values were considered significant at $P \leq 0.05.$ Feeding data analysis was performed with R; mixed-effect models were generated using either food intake or delay as outcomes, the presence of implant and laser as fixed factors, and the animal as a random factor. ## Study approval. Animal studies were conducted according to the European guidelines for animal welfare and were approved by the Institutional Animal Care and Use Committee (CEEA-LR-12167, Région Languedoc-Roussillon, Montpellier, France). ## Author contributions NR and MS designed experiments; NR, CL, PC, AG, TF, and MS performed experiments; NR, DJH, and MS analyzed data; and NR, DJH, PM, and MS wrote the manuscript. ## 12/27/2022 In-Press Preview ## 02/08/2023 Electronic publication ## References 1. Myers MG. How is the hungry brain like a sieve?. Cell Metab (2013.0) 17 467-468. DOI: 10.1016/j.cmet.2013.03.007 2. Waterson MJ, Horvath TL. Neuronal regulation of energy homeostasis: beyond the hypothalamus and feeding. Cell Metab (2015.0) 22 962-970. DOI: 10.1016/j.cmet.2015.09.026 3. Myers MG, Olson DP. Central nervous system control of metabolism. Nature (2012.0) 491 357-363. DOI: 10.1038/nature11705 4. Morton GJ. Neurobiology of food intake in health and disease. Nat Rev Neurosci (2014.0) 15 367-378. PMID: 24840801 5. Morton GJ. Central nervous system control of food intake and body weight. Nature (2006.0) 443 289-295. DOI: 10.1038/nature05026 6. Duquenne M. Leptin brain entry via a tanycytic LepR-EGFR shuttle controls lipid metabolism and pancreas function. Nat Metab (2021.0) 3 1071-1090. DOI: 10.1038/s42255-021-00432-5 7. Schaeffer M. Rapid sensing of circulating ghrelin by hypothalamic appetite-modifying neurons. Proc Natl Acad Sci U S A (2013.0) 110 1512-1517. DOI: 10.1073/pnas.1212137110 8. Langlet F. Tanycytic VEGF-A boosts blood-hypothalamus barrier plasticity and access of metabolic signals to the arcuate nucleus in response to fasting. Cell Metab (2013.0) 17 607-617. DOI: 10.1016/j.cmet.2013.03.004 9. Clarke IJ. Hypothalamus as an endocrine organ. Compr Physiol (2015.0) 5 217-253. PMID: 25589270 10. Castle-Miller J. Mechanisms regulating angiogenesis underlie seasonal control of pituitary function. Proc Natl Acad Sci U S A (2017.0) 114 E2514-E2523. PMID: 28270617 11. Dore-Duffy P, Cleary K. Morphology and properties of pericytes. Methods Mol Biol (2011.0) 686 49-68. PMID: 21082366 12. Hall CN. Capillary pericytes regulate cerebral blood flow in health and disease. Nature (2014.0) 508 55-60. DOI: 10.1038/nature13165 13. Peppiatt CM. Bidirectional control of CNS capillary diameter by pericytes. Nature (2006.0) 443 700-704. DOI: 10.1038/nature05193 14. Michau A. Metabolic stress impairs pericyte response to optogenetic stimulation in pancreatic islets. Front Endocrinol (Lausanne) (2022.0) 13. DOI: 10.3389/fendo.2022.918733 15. Winkler EA. Central nervous system pericytes in health and disease. Nat Neurosci (2011.0) 14 1398-1405. DOI: 10.1038/nn.2946 16. Butiaeva LI. Leptin receptor-expressing pericytes mediate access of hypothalamic feeding centers to circulating leptin. Cell Metab (2021.0) 33 1433-1448. DOI: 10.1016/j.cmet.2021.05.017 17. Yemisci M. Pericyte contraction induced by oxidative-nitrative stress impairs capillary reflow despite successful opening of an occluded cerebral artery. Nat Med (2009.0) 15 1031-1037. DOI: 10.1038/nm.2022 18. Bell RD. Pericytes control key neurovascular functions and neuronal phenotype in the adult brain and during brain aging. Neuron (2010.0) 68 409-427. DOI: 10.1016/j.neuron.2010.09.043 19. Dziewulska D, Lewandowska E. Pericytes as a new target for pathological processes in CADASIL. Neuropathology (2012.0) 32 515-521. DOI: 10.1111/j.1440-1789.2011.01290.x 20. Geraldes P. Activation of PKC-delta and SHP-1 by hyperglycemia causes vascular cell apoptosis and diabetic retinopathy. Nat Med (2009.0) 15 1298-1306. DOI: 10.1038/nm.2052 21. Karram K. NG2-expressing cells in the nervous system: role of the proteoglycan in migration and glial-neuron interaction. J Anat (2005.0) 207 735-744. DOI: 10.1111/j.1469-7580.2005.00461.x 22. Kirkwood PM. Single-cell RNA sequencing redefines the mesenchymal cell landscape of mouse endometrium. FASEB J (2021.0) 35. PMID: 33710643 23. Campbell JN. A molecular census of arcuate hypothalamus and median eminence cell types. Nat Neurosci (2017.0) 20 484-496. DOI: 10.1038/nn.4495 24. Virgintino D. An intimate interplay between precocious, migrating pericytes and endothelial cells governs human fetal brain angiogenesis. Angiogenesis (2007.0) 10 35-45. DOI: 10.1007/s10456-006-9061-x 25. Milesi S. Redistribution of PDGFRβ cells and NG2DsRed pericytes at the cerebrovasculature after status epilepticus. Neurobiol Dis (2014.0) 71 151-158. DOI: 10.1016/j.nbd.2014.07.010 26. Hall CN. Capillary pericytes regulate cerebral blood flow in health and disease. Nature (2014.0) 508 55-60. PMID: 24670647 27. Attwell D. What is a pericyte?. J Cereb Blood Flow Metab (2016.0) 36 451-455. DOI: 10.1177/0271678X15610340 28. Hill RA. Regional blood flow in the normal and ischemic brain is controlled by arteriolar smooth muscle cell contractility and not by capillary pericytes. Neuron (2015.0) 87 95-110. DOI: 10.1016/j.neuron.2015.06.001 29. Schaeffer M. The blood-brain barrier as a regulator of the gut-brain axis. Front Horm Res (2014.0) 42 29-49. DOI: 10.1159/000358313 30. Trost A. Brain and retinal pericytes: origin, function and role. Front Cell Neurosci (2016.0) 10. PMID: 26869887 31. Haddad-Tóvolli R. Development and function of the blood-brain barrier in the context of metabolic control. Front Neurosci (2017.0) 11. DOI: 10.3389/fnins.2017.00224 32. Hayden MR. Attenuation of endocrine-exocrine pancreatic communication in type 2 diabetes: pancreatic extracellular matrix ultrastructural abnormalities. J Cardiometab Syndr (2008.0) 3 234-243. DOI: 10.1111/j.1559-4572.2008.00024.x 33. Etchevers HC. The cephalic neural crest provides pericytes and smooth muscle cells to all blood vessels of the face and forebrain. Development (2001.0) 128 1059-1068. DOI: 10.1242/dev.128.7.1059 34. Trost A. Neural crest origin of retinal and choroidal pericytes. Invest Ophthalmol Vis Sci (2013.0) 54 7910-7921. DOI: 10.1167/iovs.13-12946 35. 35Tamayo A, et al. Pericyte control of pancreatic islet blood flow impacts glucose homeostasis [preprint]. 10.1101/2021.08.24.457539 Posted on bioRxiv August 25, 2021. DOI: 10.1101/2021.08.24.457539 36. Kohnke S. Nutritional regulation of oligodendrocyte differentiation regulates perineuronal net remodeling in the median eminence. Cell Rep (2021.0) 36. DOI: 10.1016/j.celrep.2021.109362 37. Tolle V. In vivo and in vitro effects of ghrelin/motilin-related peptide on growth hormone secretion in the rat. Neuroendocrinology (2001.0) 73 54-61. DOI: 10.1159/000054620 38. Yin Y. The growth hormone secretagogue receptor: its intracellular signaling and regulation. Int J Mol Sci (2014.0) 15 4837-4855. DOI: 10.3390/ijms15034837 39. Kovács KJ. c-Fos as a transcription factor: a stressful (re)view from a functional map. Neurochem Int (1998.0) 33 287-297. DOI: 10.1016/S0197-0186(98)00023-0 40. Edwards A, Abizaid A. Clarifying the ghrelin system’s ability to regulate feeding behaviours despite enigmatic spatial separation of the GHSR and its endogenous ligand. Int J Mol Sci (2017.0) 18. DOI: 10.3390/ijms18040859 41. Kadry H. A blood-brain barrier overview on structure, function, impairment, and biomarkers of integrity. Fluids Barriers CNS (2020.0) 17. DOI: 10.1186/s12987-020-00230-3 42. Balland E. Hypothalamic tanycytes are an ERK-gated conduit for leptin into the brain. Cell Metab (2014.0) 19 293-301. DOI: 10.1016/j.cmet.2013.12.015 43. Armulik A. Pericytes regulate the blood-brain barrier. Nature (2010.0) 468 557-561. DOI: 10.1038/nature09522 44. Madisen L. A toolbox of Cre-dependent optogenetic transgenic mice for light-induced activation and silencing. Nat Neurosci (2012.0) 15 793-802. DOI: 10.1038/nn.3078 45. Lafont C. Cellular in vivo imaging reveals coordinated regulation of pituitary microcirculation and GH cell network function. Proc Natl Acad Sci U S A (2010.0) 107 4465-4470. DOI: 10.1073/pnas.0902599107 46. Sparta DR. Construction of implantable optical fibers for long-term optogenetic manipulation of neural circuits. Nat Protoc (2011.0) 7 12-23. PMID: 22157972 47. Leyris JP. Homogeneous time-resolved fluorescence-based assay to screen for ligands targeting the growth hormone secretagogue receptor type 1a. Anal Biochem (2011.0) 408 253-262. DOI: 10.1016/j.ab.2010.09.030
--- title: TRPM7 kinase is required for insulin production and compensatory islet responses during obesity authors: - Noushafarin Khajavi - Andreas Beck - Klea Riçku - Philipp Beyerle - Katharina Jacob - Sabrina F. Syamsul - Anouar Belkacemi - Peter S. Reinach - Pascale C.F. Schreier - Houssein Salah - Tanja Popp - Aaron Novikoff - Andreas Breit - Vladimir Chubanov - Timo D. Müller - Susanna Zierler - Thomas Gudermann journal: JCI Insight year: 2023 pmcid: PMC9977431 doi: 10.1172/jci.insight.163397 license: CC BY 4.0 --- # TRPM7 kinase is required for insulin production and compensatory islet responses during obesity ## Abstract Most overweight individuals do not develop diabetes due to compensatory islet responses to restore glucose homeostasis. Therefore, regulatory pathways that promote β cell compensation are potential targets for treatment of diabetes. The transient receptor potential cation channel subfamily M member 7 protein (TRPM7), harboring a cation channel and a serine/threonine kinase, has been implicated in controlling cell growth and proliferation. Here, we report that selective deletion of Trpm7 in β cells disrupted insulin secretion and led to progressive glucose intolerance. We indicate that the diminished insulinotropic response in β cell–specific Trpm7-knockout mice was caused by decreased insulin production because of impaired enzymatic activity of this protein. Accordingly, high-fat–fed mice with a genetic loss of TRPM7 kinase activity displayed a marked glucose intolerance accompanied by hyperglycemia. These detrimental glucoregulatory effects were engendered by reduced compensatory β cell responses because of mitigated protein kinase B (AKT)/ERK signaling. Collectively, our data identify TRPM7 kinase as a potentially novel regulator of insulin synthesis, β cell dynamics, and glucose homeostasis under obesogenic diet. ## Introduction In obese individuals, a combination of environmental and genetic factors may lead to insulin resistance. The onset of insulin resistance is initially offset by enhanced production and secretion of insulin. However, prolonged demand for elevated levels of circulating insulin may eventually result in β cell exhaustion, progressive β cell dysfunction, and development of type 2 diabetes (T2D) [1]. Identification of signaling molecules that enhance islet compensatory responses in insulin-resistant states may blaze the trail for novel therapeutic approaches to prevent the progression of insulin resistance to T2D. Transient receptor potential cation channel subfamily M member 7 (TRPM7) is a ubiquitously expressed membrane protein, consisting of a divalent cation-selective channel linked to a protein kinase domain. The channel moiety of TRPM7 has been implicated in cellular and systemic Mg2+ homeostasis [2, 3]. Mg2+ plays a key role in maintaining β cell health, and while Mg2+ deficiency impairs insulin secretion and promotes insulin resistance [4], its supplementation improves β cell function [5]. The kinase moiety of TRPM7 belongs to the atypical α-kinase family [6] and has been implicated in controlling numerous cellular processes, such as proliferation, growth, migration, apoptosis, differentiation, and exocytosis [7]. α-Kinases are structurally and evolutionarily unrelated to conventional eukaryotic protein kinases, yet they share common sequence motifs and the position of key amino acid residues essential for catalysis [8, 9]. The remaining dissimilarities of α-kinases to conventional protein kinases are of potential interest for selective pharmacological targeting [10]. In mice, TRPM7 is a central regulator of embryogenesis and organogenesis, with genetic inactivation of TRPM7 causing early embryonic lethality. It has been suggested that localized increases in the concentration of divalent cations due to transmembrane ion flux through the TRPM7 channel trigger kinase activity engaging signaling pathways that are of fundamental relevance in early development [11]. There is strong evidence that increases in TRPM7 activity are required to elicit expression of key cell cycle genes in various cell types [12, 13]. In hepatic stellate cells, TRPM7 regulates cell proliferation via phosphatidylinositol-3-kinase (PI3K) and ERK$\frac{1}{2}$ signaling pathways [14]. In lymphocytes, TRPM7 ablation arrests cell proliferation with a high percentage of arrested cells accumulating at the beginning of the cell cycle, suggesting a potential involvement of TRPM7 in processes orchestrating exit from the quiescence/G0 phase of the cell cycle [15]. Notably, TRPM7 inactivation in pancreatic adenocarcinoma cells decreases proliferation and arrests cells in the G0/G1 phases of the cell division cycle [16]. The kinase moiety of TRPM7 has been suggested to regulate gene transcription through histone modifications. TRPM7 kinase has been reported to be cleaved from the channel domain in a cell type–specific fashion. It would subsequently translocate to the nucleus and bind to components of chromatin-remodeling complexes [17]. Furthermore, TRPM7 plays a role in the regulation of Ca2+ signaling in various cell types (18–21). In osteoblasts, the TRPM7 channel modulates cell migration by facilitating Ca2+ oscillations [18]. TRPM7 has been shown to maintain the Ca2+ content of intracellular stores in resting cells. In splenocytes and B lymphocytes, the TRPM7 channel and its kinase moieties regulate store-operated Ca2+ entry (SOCE) [19, 20]. Dysfunctional SOCE in β cells contributes to the pathogenesis of diabetes and has been reported to disrupt glucose-stimulated Ca2+ oscillations in β cells [22]. TRPM7 is highly expressed in human and murine pancreatic β cells [23, 24]. Recent studies revealed that TRPM7 contributes to pancreatic endocrine development and β cell proliferation through modulating intracellular Mg2+ levels. Furthermore, it has been suggested that TRPM7 channels augment β cell glucose-stimulated Ca2+ influx in pancreatic β cells [21]. However, it is still not known how changes in TRPM7 kinase–linked activity regulate glucose metabolism and Ca2+ signaling in pancreatic β cells. The present study was designed to clarify the role of TRPM7 in maintaining β cell function under physiological and metabolically challenged conditions. To decipher the function of TRPM7 in glucose homeostasis, we generated a mouse model with a selective deletion of Trpm7 in β cells (βTrpm7-KO mice). Initially, within 4 weeks, these mice exhibited no overt changes in glucose metabolism following recombination. However, a latent induction of glucose intolerance was evident over 28 weeks, suggesting that TRPM7 disruption leads to progressive β cell dysfunction. In-depth metabolic analysis of TRPM7 kinase–dead mice (Trpm7tm1.1Mkma C56BL/6; herein Trpm7R/R mice) demonstrated that the kinase moiety of TRPM7 is the key player in this scenario. Specifically, our data identify TRPM7 kinase as a crucial cellular component involved in the preservation of glucometabolic islet function under conditions of diet-induced obesity. ## Trpm7 deletion in β cells impairs insulin secretion and glucose metabolism. To assess the requirement of β cell TRPM7 for insulin secretion and glucose homeostasis, we generated tamoxifen-inducible β cell–specific Trpm7-KO mice (βTrpm7-KO mice) by crossing Trpm7fl/fl with MIP-Cre/ERT mice. Pancreatic islets isolated from βTrpm7-KO mice revealed efficient β cell–targeted recombination after tamoxifen administration. Notably, no Cre-mediated recombination was detected in any region of the brain or liver in βTrpm7-KO mice (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.163397DS1). We monitored the metabolic phenotype of βTrpm7-KO mice on regular chow diet within 28 weeks. While wild-type and βTrpm7-KO mice had similar body weight (Figure 1A), the transgenic mice exhibited progressive glucose intolerance starting 16 weeks postrecombination (Figure 1, B–D). Moreover, βTrpm7-KO mice displayed elevated fed blood glucose relative to control littermates after 28 weeks of tamoxifen-induced recombination (Figure 1E). Plasma insulin levels (Figure 1F) and insulin sensitivity (Figure 1G) remained unchanged in both genotypes. Glucose-induced insulin secretion (GIIS) was severely diminished in isolated islets from βTrpm7-KO mice after 28 weeks of tamoxifen-induced recombination. Augmentation of glucose concentrations from 2.8 to 20 mM increased insulin release about 5-fold in WT islets, whereas only a 2.5-fold increase of basal insulin exocytosis was observed in the βTrpm7-KO islets (Figure 1H). Collectively, these data show that tissue-specific TRPM7 disruption in β cells progressively impairs glucose metabolism, which we hypothesized to be attributable to loss of β cell identity and impaired cell cycle regulation. ## TRPM7 kinase disruption reduces glucose tolerance and GIIS. To decipher whether the impaired glucose metabolism is linked to the TRPM7 channel or kinase moiety, we took advantage of a mouse model harboring a single point mutation at the active site of the enzyme (Trpm7R/R). We monitored the metabolic phenotype of Trpm7R/R mice within 28 weeks. When fed a standard chow diet, Trpm7R/R mice showed no difference in body weight compared to their WT controls (Figure 2, A and B). However, Trpm7R/R mice displayed glucose intolerance at 8 weeks of age that became more prominent in 28-week-old mice (Figure 2, C and D, and Supplemental Figure 2A). Although the concentrations of blood glucose remained unchanged in fasted and fed 8-week-old Trpm7R/R mice relative to controls (Figure 2E), an elevated fed blood glucose was detected in Trpm7R/R mice at 28 weeks of age (Figure 2F). Insulin sensitivity and plasma insulin levels remained unaffected within this period (Supplemental Figure 2, B–E). This observation pointed to a putative role of the TRPM7 kinase in sustaining islet function and modulating insulin secretion. To test this hypothesis, we investigated GIIS using isolated islets from Trpm7R/R mice and control littermates. In accord with the impaired glucose tolerance observed in vivo, TRPM7 kinase disruption reduced maximal insulin secretion capacity in isolated islets (Figure 2G and Supplemental Figure 2F). Augmenting the glucose concentration from 2.8 to 20 mM enhanced insulin exocytosis about 4-fold in WT islets, whereas insulin exocytosis only rose 3-fold in isolated islets from 8-week-old Trpm7R/R mice (Figure 2G) and 2.2-fold at 28 weeks of age (Supplemental Figure 2F). Membrane depolarization of β cells by exposure to either 25 mM KCl or 300 μM tolbutamide increased insulin secretion to about 7- and 3.5-fold, respectively, while insulin exocytosis only increased 6.5- and 2-fold in isolated islets of Trpm7R/R mice at 8 weeks of age. Interestingly, basal insulin secretion was significantly lower in Trpm7R/R islets relative to WT controls (Figure 2G and Supplemental Figure 2F). ## TRPM7 kinase inactivation has no effect on glucose-induced Ca2+ responses and TRPM7-mediated ion currents. β Cells display a characteristic Ca2+ oscillation pattern in response to high glucose concentration, thus regulating the exocytosis of insulin. Hence, we next asked whether impaired Ca2+ responses are attributable to declines in glucose-induced rises in insulin secretion in Trpm7R/R islets. Exposure to glucose initially induced a rapid increase in [Ca2+]i followed by continuous oscillations in a subset of WT islet cells. Trpm7R/R islets responded similarly to 20 mM glucose in terms of [Ca2+]i transients when compared to WT islets (Figure 3, A and B). The responses to tolbutamide (Figure 3, A and C) and KCl (Supplemental Figure 3, A and B) under identical conditions were also very similar to one another, supporting that canonical KATP signaling is not affected by loss of TRPM7 kinase function. Neither the Ca2+ oscillation frequency nor the average amplitude of the oscillatory response was different from one another in both genotypes (Supplemental Figure 4, A–C). Taken together, these data demonstrate that TRPM7 kinase disruption does not affect glucose-induced Ca2+ responses in β cells. As TRPM7 channels are activated by depletion of intracellular Mg2+ [2], we determined the effects of intracellular Mg2+ removal on the underlying whole-cell currents in pancreatic islet cells isolated from WT and Trpm7R/R mice (Figure 4, A–G). Upon break-in by the patch pipette, most islet cells revealed huge outward currents, probably mediated by a voltage-dependent potassium efflux, which rapidly vanished after infusion of the cesium-based Mg2+-free pipette solution and washout of the cytosolic potassium content (Figure 4A). However, within 600 seconds, small currents with a current-voltage relationship reminiscent of TRPM7 activation developed in both the WT (Figure 4B, black trace) and Trpm7R/R (Figure 4C, purple trace) islet cells. To verify the dependency of these small outward currents on TRPM7 activation, we applied DVF (buffered by EDTA), resulting in substantially increased in- and outward currents in islet cells of both genotypes as a hallmark of TRPM7 activation (Figure 4, A–C). Since removal of external divalent cations increases TRPM7’s permeability to monovalent cations, both in- and outward currents increased and the current-voltage relationship switched from slightly outward rectification to a nearly linear shape (Figure 4, B and C, blue traces). Figure 4, D and E, show the current-voltage relations of the basal current–subtracted net TRPM7 current, extracted after 600 seconds (Figure 4D) and the basal current–subtracted net current measured in DVFs (Figure 4E) in islet cells from WT (black traces) and Trpm7R/R mice (purple). The amplitudes of net TRPM7 currents induced by depletion of free intracellular Mg2+ at +80 mV in the presence (Figure 4, D and F) and absence of extracellular divalent cations (Figure 4, E and G) in islet cells isolated from WT and Trpm7R/R mice were not significantly different, suggesting that in islet cells TRPM7-induced currents are not affected by the presence or absence of its kinase activity. ## Genetic loss of TRPM7 kinase does not alter the pancreatic islet cytoarchitecture. Next, we asked whether loss of TRPM7 kinase function affects pancreatic islet development. We observed that islet density (Figure 5, A and B) as well as islet size distribution (Figure 5C) were similar in both genotypes. Islet size ranged between >45 and <420 μm in both genotypes, with the highest frequency found between 61 and 180 μm (Figure 5C). In both genotypes, approximately $60\%$–$70\%$ of the endocrine cells within the islets were insulin-positive β cells, localized in the central part of the islets, while glucagon-positive α cells were located at the islet periphery surrounding β cells (Figure 5, A and D). Moreover, the ratio of β to α cells remained unchanged in both genotypes (Figure 5E). ## TRPM7 kinase disruption attenuates insulin biosynthesis. Next, we measured the insulin content in islets of both genotypes and noted a remarkable decrease of the insulin content in Trpm7R/R islets relative to WT controls (Figure 6A). While GIIS was also decreased in Trpm7R/R islets (see Figure 2G), the GIIS normalized to the content was not significantly different between WT and Trpm7R/R. These results indicate that the TRPM7 kinase plays a role in insulin biosynthesis rather than insulin release (Figure 6B). Western blot analysis supported the declines in insulin levels (Figure 6, C and D). To dissect the cellular mechanism underlying the reduced insulin content of Trpm7R/R islets, we performed real-time quantitative reverse transcription PCR (qRT-PCR) analyses and identified reduced expression of key genes involved in insulin production in Trpm7R/R islets relative to those in WT cells, including Ins2, pancreatic and duodenal homeobox 1 (Pdx1), and V-maf musculoaponeurotic fibrosarcoma oncogene homolog A (MafA) (Figure 6E). The reduced expression of PDX1 protein in Trpm7R/R islets was supported by Western blotting and immunohistochemistry (Figure 6, F–I). In addition, we demonstrated a remarkable decrease in insulin content as well as expression levels of PDX1 protein in isolated islets from βTrpm7 KO after 28 weeks of tamoxifen-induced recombination (Supplemental Figure 5, A–C). These data suggest that the reduced insulin content of Trpm7R/R and βTrpm7-KO islets may be caused by decreased insulin synthesis secondary to a reduction of Pdx1 expression. ## TRPM7 kinase inactivation impairs HFD-induced β cell mass expansion. Pdx1 is implicated in compensatory β cell mass expansion in response to diet-induced insulin resistance [25]. Therefore, we examined whether TRPM7 kinase is required for maintenance of pancreatic islet function under HFD feeding. Although cumulative food intake (Figure 7A) was comparable in both genotypes, Trpm7R/R mice underwent pronounced increases in body weight and blood glucose levels (Figure 7, B and C) and a significant reduction in plasma insulin in the fed state relative to WT littermates (Figure 7D). Of note, during HFD feeding, glucose tolerance was severely impaired in Trpm7R/R mice relative to WT littermates (Figure 7E). Both groups of mice showed similar reductions in blood glucose levels in an insulin tolerance test (Figure 7F). Interestingly, the number of islets per pancreatic section (Figure 8A) as well as the frequency distribution of larger (301–360 μm) islets were reduced, while the distribution of small (0–60 μm) islets was increased in Trpm7R/R pancreatic slices (Figure 8B). However, total pancreatic weight did not differ between high-fat–fed Trpm7R/R and control littermates (Figure 8C). As normal islet compensation involves an expansion of β cell mass achieved by β cell hypertrophy and proliferation [26, 27], we examined β cell size, proliferation, and survival rates in response to HFD. Notably, Trpm7R/R islets displayed a significant decrease in β cell size relative to control littermates (Figure 8D). Ki67 was used as a proliferation marker. The abundance of Ki67-positive β cells was markedly reduced in Trpm7R/R islets compared with control littermates (Figure 8, E and F). Thus, these results suggest that loss of TRPM7 kinase restrains β cell proliferation in response to HFD. Importantly, we did not detect any TUNEL-positive, i.e., apoptotic, β cells in WT and Trpm7R/R islets (Supplemental Figure 6A). Furthermore, to investigate whether reduced β cell proliferation in Trpm7R/R mice is associated with impaired SOCE, we monitored passive Ca2+ release in response to the sarcoplasmic/endoplasmic reticulum calcium ATPase (SERCA) pump blocker cyclopiazonic acid followed by SOCE after Ca2+ re-addition. However, there was no significant difference both in passive Ca2+ release and in subsequent SOCE between β cells derived from Trpm7R/R and WT genotypes (Supplemental Figure 6, B–D). ## TRPM7 kinase inactivation reduces expression levels of genes involved in insulin production and cell cycle regulation. We investigated the gene expression profile by RNA sequencing using RNA prepared from isolated islets of high-fat–fed Trpm7R/R and WT mice. We identified upregulation of 382 genes and downregulation of 1,615 genes in Trpm7R/R islets. Interestingly, genes critical to insulin production were found to be downregulated, including Ins2, MafA, Pdx1, Cpe, and Nkx6-1. We also observed downmodulation of genes involved in cell cycle regulation, such as cyclin-dependent kinase 4 (Cdk4) and Ccnd2, and several proliferation markers, including Cirp, Mll5, and Pimerg (Supplemental Table 1). No changes were observed in the expression levels of genes involved in glucose sensing and exocytosis (Supplemental Table 2). A volcano plot (Figure 8G) and heatmaps (Supplemental Figure 7, A and B) illustrate the differential expression of genes involved in insulin biosynthesis, cell cycle progression, and cellular proliferation. A summary of the genes involved in insulin production and cell cycle regulation that were downregulated in islets lacking TRPM7 kinase is shown in Supplemental Figure 7, C and D. ## TRPM7 regulates β cell function via protein kinase B (AKT) and ERK1/2 signaling pathways. TRPM7 kinase regulates TGF-β/SMAD signaling [28]. This signaling pathway has been implicated in various cellular processes, including proliferation, differentiation, apoptosis, and cell migration [29]. To gain insight into the connection between TRPM7 kinase and signaling cascades triggering compensatory islet responses to an HFD, we employed a bead-based Bio-Plex assay to simultaneously measure the phosphorylation status of multiple TGF-β/SMAD signaling proteins in the same sample. We examined the phosphorylation status of SMAD2 (Ser465/Ser467), SMAD3 (Ser423/Ser425), AKT (Ser473), and ERK$\frac{1}{2}$ (Thr185/Tyr187) as well as total protein levels of SMAD4 in isolated islets from Trpm7R/R and WT mice on an HFD for 16 weeks. A slight decrease was detected in the phosphorylation levels of SMAD2 and total SMAD4. However, the differences did not reach statistical significance. Trpm7R/R mice displayed a significant reduction of ERK-dependent signaling under HFD feeding. Interestingly, our results demonstrated a $40\%$ reduction of AKT phosphorylation in Trpm7R/R mice fed with an HFD (Figure 8H). Quantification of SMAD3 phosphorylation status was below the detection limit in both genotypes and excluded from the study. Next, we asked whether reduced ERK$\frac{1}{2}$ and AKT phosphorylation in Trpm7R/R islets is attributable to the dysfunctional TRPM7 in β cells. To address this question, we measured the phosphorylation status of ERK$\frac{1}{2}$ and AKT in isolated islets from βTrpm7-KO mice and control littermates on 16 weeks of an HFD. Notably, βTrpm7-KO mice demonstrated a significant decrease in both ERK- and AKT-dependent signaling (Supplemental Figure 8, A and B), supporting the role of TRPM7 in modulating these signaling cascades in β cells. ## TRPM7 overexpression induces GIIS in MIN6 cells. To test whether enhanced expression of Trpm7 in β cells induces insulin secretion, we used cultured MIN6 mouse insulinoma cells as an in vitro model. We transiently transfected MIN6 cells with Trpm7 WT or Trpm7R/R plasmids or with empty vector (Supplemental Figure 9A). The MIN6 cells treated with the Trpm7R/R plasmid had a GIIS response similar in magnitude to that of MIN6 cells treated with an empty vector. In contrast, MIN6 cells transfected with the Trpm7 WT plasmid displayed a pronounced increase in GIIS (Supplemental Figure 9B). This response pattern is in good agreement with the data obtained with isolated islets from βTrpm7-KO and Trpm7R/R mice. Next, we investigated the effect of Trpm7 overexpression in MIN6 cells on the expression levels of key signaling proteins. Western blotting studies showed that treatment of MIN6 cells with Trpm7 WT plasmid led to a significant increase in phosphorylated forms of ERK$\frac{1}{2}$ and AKT (Ser473), as compared with cells treated with empty vectors. The expression levels of total ERK and AKT remained essentially unchanged after Trpm7 WT overexpression (Supplemental Figure 9, C–E). ## Discussion We report here that TRPM7 regulates glucose homeostasis and compensatory pancreatic islet responses via its kinase moiety. Impaired glucose tolerance and GIIS were comparable among βTrpm7-KO and Trpm7R/R genotypes. Importantly, mice from both genotypes developed age-dependent rises in dysfunctional glucose metabolism and declines in GIIS. Reduced insulin secretion in response to various insulin secretagogues in Trpm7R/R suggests a salient role of this α-kinase in maintaining β cell function. Intracellular Ca2+ transients are the final trigger for insulin exocytosis. Studies with Trpm7R/R pancreatic islets showed that glucose and high KCl-induced Ca2+ responses remained unaffected, suggesting that the reduction of GIIS in Trpm7R/R mice is not caused by impaired Ca2+ signaling in β cells. In addition, TRPM7-like currents in Trpm7R/R islets were comparable to those of WT cells, demonstrating that the channel moiety of TRPM7 remains intact in this mouse model. Interestingly, we found a pronounced reduction of insulin content accompanied by reduced Pdx1 transcript and protein levels in Trpm7R/R and βTrpm7-KO mice. In humans, mutations of Pdx1 are strongly associated with diabetes [30]. Previous studies have demonstrated a correlation between low PDX1 levels and β cell dysfunction [31, 32], because PDX1 directly regulates the expression of the insulin gene and other components of the GIIS pathway, including MafA [33, 34]. Therefore, we suggest that Pdx1 downregulation might suppress insulin production in Trpm7R/R and βTrpm7-KO mice. Importantly, Pdx1 and MafA are the key β cell markers and major transcription factors to maintain β cell identity. Altered identity of β cells has been proposed as an underlying mechanism of diabetes progression in patients [35, 36]. Furthermore, previous studies linked overexpression of PDX1 to the upregulation of several cell cycle genes and increases in β cell proliferation [37, 38]. Therefore, we attribute the age-dependent progressive impairment in glucose metabolism in Trpm7R/R and βTrpm7-KO mice to the gradual loss of β cell identity and reduced β cell proliferation due to Pdx1 downregulation. Accumulating evidence has underscored the pivotal role of PDX1 in β cell expansion and survival in response to an HFD challenge [25]. Therefore, we set out to define the role of TRPM7 kinase in β cell survival and compensatory hypertrophy in response to an HFD. Thus, an obesogenic diet resulted in increased body weight, hyperglycemia, reduced insulin levels, and glucose intolerance in Trpm7R/R mice relative to WT controls. These phenotypic changes were not caused by severe insulin resistance in Trpm7R/R mice, because insulin tolerance was comparable in both genotypes. It is worth mentioning that high-fat–fed Trpm7R/R mice became more severely hyperglycemic than control littermates after 16 weeks of obesogenic diet, especially in the fed state. Collectively, our data are compatible with the notion that the pronounced impairment of glucose homeostasis in Trpm7R/R mice is attributable to impaired compensatory β cell mass expansion and proliferation in response to obesogenic diet. Glucose intolerance observed in the Trpm7R/R mice on the HFD was consistent with the metabolic phenotype of Pdx1+/– animals [39]. Like our findings in Trpm7R/R mice, high-fat feeding induced a similar weight gain in Pdx1+/– animals relative to WT controls [25]. Taken together, we suggest that reduced Pdx1 expression in Trpm7R/R mice dampens both insulin production and compensatory β cell mass expansion, entailing compromised glucose tolerance in high-fat–fed Trpm7R/R mice. Prior to our work, Altman et al. [ 21] observed that β cell proliferation induced by 2-week HFD was significantly reduced in TRPM7 deficient β cells. In agreement with our finding, TRPM7 disruption in β cells did not alter glucose tolerance within 4 weeks of recombination. However, the authors suggested that reduced proliferation observed after exposure to obesogenic diet was mediated by impeded Mg2+ influx into β cells during proliferation [21]. In this context, it is worth mentioning that TRPM7 channel and kinase activities are mutually interdependent, in that the kinase functionality requires the influx of Mg2+ through the channel pore [40]. Therefore, we put forward an alternative explanation and suggest that hampered β cell proliferation may be attributable to reduced kinase activity that results from declines in Mg2+ levels. Furthermore, our histological studies of Trpm7R/R mice pancreas do not agree with the developmental changes Altman et al. showed after TRPM7 inactivation [21]. This difference might point to the role of the channel moiety of TRPM7 in early events influencing pancreatic endocrine development. RNA-Seq studies with RNA isolated from high-fat–fed Trpm7R/R and control islets demonstrated that TRPM7 kinase deficiency downregulated the genes involved in insulin biosynthesis, cell cycle progression, and proliferation. Notably, TRPM7 kinase disruption engenders reduced expression of 2 early G1/S phase molecules, cyclin D2 and Cdk4 in high-fat–fed mice. It has previously been demonstrated that mice lacking Cdk4 exhibit islet deformity and a reduced size of islets accompanied by diminished insulin production, whereas activation of the CDK4 pathway resulted in β cell hyperplasia [41]. Interestingly, reduced pancreas size in Cdk4-deficient mice is thought to result from impaired mesenchymal development and decreased numbers of PDX1+ pancreatic progenitor cells [42]. Moreover, CDK4 enhances β cell replication within adult islets and activates progenitor cells within adult pancreatic ductal epithelium in response to partial pancreatectomy [43]. We suggest that reduced islet size, and impaired β cell proliferation in high-fat–fed Trpm7R/R mice, might be attributable at least partially to reduced CDK4 expression in pancreatic islets. In addition, we noted that the transcript levels of several other proliferation markers, including Cirp, Mll5, and Pimerg, were substantially reduced in high-fat–fed TRPM7 Trpm7R/R mice. Interestingly, CIRP activation occurs downstream of various stress stimuli and is known to regulate cell survival and cell proliferation, particularly during stress [44]. In mouse models of diet-induced obesity, high-fat feeding has been linked to ER stress in β cells, resulting in the inability to trigger an appropriate unfolded protein response (UPR), potentially leading to β cell apoptosis [45]. Previous studies show that Pdx1+/– β cells are more susceptible to ER stress under high-fat feeding [25]. PDX1 plays a crucial role in the regulation of genes involved in ER function, including disulfide bond formation, protein folding, and the UPR. Here, we show the downregulation of several ER-related genes in Trpm7R/R islets, including genes encoding enzymes critical for disulfide bond formation in the ER (Pdia4 and Pdia6), ER chaperone (Hspa5), and mediators of UPR pathways (Atf4), which are direct transcriptional targets of PDX1 [25]. Although it has been suggested that Pdx1 deficiency promotes ER stress–associated cell death, we did not detect apoptosis in WT and Trpm7R/R islets, even when challenged by high-fat feeding. This observation is fully in line with a recent study by Barella et al., who did not detect any apoptotic β cells in β-barr1-KO mice in the presence of severely impaired Pdx1 expression [46]. Furthermore, Altman et al. reported that TRPM7 KO has no effect on β cell apoptosis [21]. Previous studies demonstrated that knockdown of Pdx1 in rat insulinoma cells (INS-1) results in a reduced SERCA2b expression and decreased ER Ca2+ levels [47]. Importantly, TRPM7 kinase deficiency has been shown to suppress SOCE in T- cells and B lymphocytes [19, 20]. Nevertheless, we found that both SOCE and ER Ca2+ storage were unaffected in Trpm7R/R islets (Supplemental Figure 6, B and C), ruling out a major impediment of this pathway in pancreatic β cells from Trpm7R/R mice. Phosphorylation of PDX1 is required for its nuclear translocation and binding to target promoters [48]. PDX1 phosphorylation occurs in response to PI3K/AKT signaling [49] and ERK$\frac{1}{2}$ [50]. Blocking the PI3K/AKT pathway in pancreatic β cells reduces insulin content and insulin secretion [51]. Overexpressing Akt1 in pancreatic β cells increases β cell mass, proliferation, and cell size, which leads to improved glucose tolerance and insulin secretion [52, 53]. A recent study reported that the improvements in glucose tolerance, β cell proliferation, and β cell mass induced by enhanced AKT signaling were blunted in PDX1-deficient mice [49]. Furthermore, FOXO1 is an established upstream regulator of PDX1. FOXO1 acts as a repressor of FOXA2, which is known to activate the Pdx1 promoter. Haploinsufficiency of FOXO1 reverses β cell failure in Irs2–/– mice through partial restoration of β cell proliferation and increased expression of Pdx1 [54]. In pancreatic cancer cells, inhibition of PI3K/AKT and MAPK/ERK pathways activates FOXO transcription factors, leading to cell cycle arrest and apoptosis [55]. Moreover, in pancreatic β cells, MAPKs ERK$\frac{1}{2}$ have been shown to be the major expressed forms of ERKs, playing an essential role in mediating cell proliferation [56, 57]. TRPM7 is a known regulator of the PI3K/AKT, SMAD, and ERK$\frac{1}{2}$ signaling pathways [29, 58]. Our data suggest that TRPM7 kinase might directly or indirectly phosphorylate AKT and ERK$\frac{1}{2.}$ Activation of the AKT and ERK$\frac{1}{2}$ pathways enhances PDX1 transcriptional activity, leading to compensatory β cell hypertrophy and proliferation. However, it is worth mentioning that AKT also induces proliferation of β cells through direct regulation of cyclin D1, cyclin D2, and CDK4 levels [59]. Our results do not exclude the possibility that TRPM7 kinase might be involved in β cell cycle regulation and proliferation in a PDX1-independent manner. To further corroborate the concept that TRPM7 kinase regulates AKT/ERK signaling, we transfected MIN6 cells with Trpm7 WT and Trpm7R/R. We found that overexpression of Trpm7 WT in MIN6 cells enhances phosphorylation of ERK$\frac{1}{2}$ and AKT and leads to increases in insulin secretion. These results further support the notion that the detrimental glucoregulatory effects in Trpm7R/R mice are due to mitigated AKT/ERK signaling. Obesity is a leading pathogenic factor for developing insulin resistance. Insulin resistance in obese individuals triggers a compensatory response in pancreatic islets. In this study, we provide evidence that TRPM7 kinase regulates insulin production and elicits an appropriate compensatory islet response to an obesogenic diet. Furthermore, the results from this study point to a potential link between TRPM7 kinase activity and the expression of critical genes required for insulin biosynthesis and cell cycle regulation. Therefore, we identify TRPM7 kinase as a critical cellular gatekeeper to preserve and improve β cell function under metabolically challenging circumstances. ## Mouse strains and genotyping procedures. MIP-Cre/ERT and Trpm7tm1Clph (Trpm7fl/fl) mice were obtained from The Jackson Laboratory. Trpm7tm1.1Mkma C56BL/6 (K1646R, Trpm7R/R) mice were provided by Masayuki Matsushita (Okayama University Medical School, Okayama, Japan). Trpm7fl/fl mice [60] and Trpm7R/R mice [61] were reported previously. Mice were backcrossed to C57BL/6 (≥6 generations). Mice were housed in ventilated cages at the animal facility of the Walther Straub Institute of Pharmacology and Toxicology, LMU Munich, Munich, Germany. Trpm7fl/fl and MIP-Cre/ERT mice were bred to obtain age- and sex-matched homozygous Trpm7fl/fl MIP-Cre/ERT mice. To induce Cre activity in β cells of adult mice, 8-week-old male Trpm7fl/fl MIP-Cre/ERT mice were injected i.p. with tamoxifen in corn oil (2 mg/d/mouse for 5 consecutive days). Negative controls were Trpm7fl/fl MIP-Cre/ERT mice, which received just injections of corn oil. Heterozygous K1646R animals were bred to obtain age- and sex-matched homozygous WT and homozygous Trpm7R/R mice. For genotyping, DNA was extracted from ear fragments using the Mouse Direct PCR Kit (Biotool). DNA samples were analyzed by PCR using a set of allele-specific oligonucleotides (Metabion). Sequence information is provided in Supplemental Table 2. Genotyping of Trpm7fl/fl and Trpm7R/R mice was performed as previously described [3]. Inheritance of MIP-Cre/ERT transgene was determined by PCR using the following conditions: 94°C for 2 minutes followed by 94°C for 15 seconds, 60°C for 15 seconds, 72°C for 10 seconds, last 3 steps repeated for 30 cycles, and 72°C for 2 minutes. Male and female mice were fed chow diet or diabetogenic diet (Research Diets, D12451), containing $45\%$ kcal from fat, beginning at 8 weeks of age. Mice were single- or group-housed on a 12-hour light/12-hour dark cycle at 22°C with free access to food and water. Mice were maintained under these conditions for a maximum of 36 weeks. ## Characterization of glucose homeostasis. For the determination of glucose tolerance, 8- or 24-week-old mice (male and female) were fasted overnight (16 hours). Basal blood glucose was sampled, and glucose was administered as an intraperitoneal (i.p.) injection at a dose of 2 g/kg body weight ($20\%$ w/v d-glucose from MilliporeSigma in $0.9\%$ w/v saline). Blood samples were obtained from the tail vein. Blood glucose levels were measured by glucometer (TheraSense FreeStyle) before (0 minutes) and at 15, 30, 60, and 120 minutes after injection. For the determination of insulin tolerance, mice were fasted for 4 hours at the onset of the light cycle and injected intraperitoneally with 0.75 units of insulin/kg body weight. Blood glucose levels were measured by glucometer (TheraSense FreeStyle) before (0 minutes) and at 15, 30, 60, and 120 minutes after injection. For investigation of blood parameters, blood was collected after euthanasia using EDTA-coated microvette tubes (Sarstedt), immediately cooled on ice, centrifuged at 2,000g and 4°C for 10 minutes, and plasma stored at −80°C. Plasma insulin was quantified by an insulin ELISA kit (ALPCO). ## Islet isolation and determination of insulin secretion. Islets were isolated from 8- to 36-week-old male and female mice. Isolation of pancreatic islets was performed as previously described [62]. In brief, pancreas was perfused by injection of 3 mM Collagenase-P (Roche) (0.3 mg/mL) in HBSS containing 25 mM HEPES and $0.5\%$ (w/v) BSA into the common bile duct. Isolated islets were recovered for 48 hours in RPMI 1640 (Thermo Fisher Scientific) in humidified $5\%$ CO2, at 37°C. After this period, islets were used for functional assessments. Before determination of insulin secretion, islets were equilibrated for 1 hour in KRB buffer (115 mM NaCl, 4.5 mM KCl, 1.2 mM KH2PO4, 2.6 mM CaCl2, 1 mM MgCl2, 10 mM HEPES, 20 mM NaHCO3, $0.2\%$ w/v BSA, pH 7.4) with 2.8 mM glucose. Determination of insulin secretion from the islets was performed in 12-well plates containing 60 μL KRB (8 islets/well, 5 independent experiments performed in triplicate). After 1 hour preincubation in KRB with 2.8 mM glucose, islets were incubated for 1 hour in 20 mM glucose, 25 mM KCl, or 300 μM tolbutamide. Released insulin was measured in the supernatant using an insulin ELISA kit. Insulin content was determined from groups of 10 islets lysed in the protein extraction reagent M-PER (Thermo Fisher Scientific), using insulin ELISA kit. ## Calcium imaging. Islets were loaded with 4 μM fluo-4 AM (Invitrogen) for 2 hours at room temperature in extracellular buffer containing 138 mM NaCl, 5.6 mM KCl, 2.6 mM CaCl2, 1 mM MgCl2, and 5 mM HEPES, pH 7.4. Changes in [Ca2+]i were recorded by laser-scanning confocal microscopy using an LSM 510 Meta system (Zeiss) using a water immersion objective (63×/NA1.2). Individual cells were selected as “regions of interest” with the LSM software, and their calcium responses to the different stimuli were measured as alterations in fluo-4 emission intensity at 500–550 nm upon excitation with the 488 nm line of an argon laser. Eight-bit 512 × 512 pixel images were acquired every 5 seconds. Calculation of calcium oscillation frequency and amplitude is described in detail in Supplemental Methods. ## Electrophysiological recordings. Whole-cell membrane currents were recorded using an EPC-9 amplifier (HEKA Electronics). Patch pipettes were pulled from glass capillaries GB150T-8P (Science Products) at a vertical Puller (PC-10, Narishige) and had resistances of 3 to 4 MΩ when filled with internal solution. The internal solution (0 Mg) consisted of (in mM) 120 Cs-glutamate, 8 NaCl, 10 HEPES, and 10 Cs-EDTA to chelate internal divalents (pH adjusted to 7.2 with CsOH). The extracellular solution contained (in mM) 140 NaCl, 2 MgCl2, 1 CaCl2, 10 HEPES, and 10 glucose (pH adjusted to 7.2 with NaOH). DVF (CaCl2 and MgCl2 were omitted from the external solution and 5 mM Na-EDTA was added) was directly applied onto the patch-clamped cell via an air pressure–driven (MPCU, Lorenz Meßgerätebau) application pipette. All solutions revealed an osmolality of 290 to 310 mOsm. Every 2 seconds voltage ramps of 50 ms duration spanning from –100 mV to +100 mV were applied from a holding potential of 0 mV using the PatchMaster software (HEKA Electronics). All voltages were corrected for a liquid junction potential of 10 mV, and currents were filtered at 2.9 kHz and digitized at 100 μs intervals. Before each voltage ramp, capacitive currents and series resistance were determined and corrected by the EPC9 automatic capacitance compensation. Inward and outward currents at –80 and +80 mV were extracted from each individual ramp current recording, and amplitudes were plotted versus time. IVs were extracted at indicated time points. To obtain the net developing current (Inet), basic currents (Imin) were subtracted from single IVs. All currents were normalized to the initial size, i.e., capacitance of the cell to obtain current densities (pA/pF). ## Morphological analysis. Standard hematoxylin and eosin staining on 10 μm cryosections of islets and immunofluorescence staining of whole islets were performed to assess pancreatic islet morphology. Antibodies and their working dilutions are listed in Supplemental Table 3. Digital imaging fluorescence microscopy of the pancreas was performed using a scanning platform (MetaSystems) with an Imager Z.2 microscope (Carl Zeiss MicroImaging, Inc.). Quantitative image analysis of islet morphology was performed using ImageJ (NIH). Size of β cells size was measured by imaging randomly selected cells at 400× and determined as mean individual β cell cross-sectional area for at least 5 islets per animal using ImageJ software. For the mean individual β cell cross-sectional area, the insulin-positive area of each islet was divided by the number of nuclei within the insulin-positive area. Investigators followed a blinded protocol during analysis. ## Western blot. Western blot analysis was performed as previously described [63]. A total of 20 μg of protein was loaded, resolved on $8\%$–$12\%$ Tris-HCl SDS-PAGE, and blotted onto a nitrocellulose membrane (Amersham Biosciences). Membranes were blocked for 1 hour using $5\%$ BSA or nonfat dried milk diluted in Tris-buffered saline with $0.1\%$ Tween 20 detergent at room temperature and incubated with primary antibodies (Supplemental Table 3) at 4°C for 16 hours. After washing, membranes were incubated with HRP-conjugated secondary antibodies (Supplemental Table 3) for 1 hour at room temperature. Immunobound antibody was visualized with an enhanced chemiluminescence kit (GE Healthcare Europe). ChemiDoc MP Imaging System (Bio-Rad) was used for chemiluminescence detection. For the loading control, membranes were stripped and incubated with an antibody against ERK2 or histone H3 for approximately 16 hours at 4°C. ## RNA isolation. RNA was extracted from pancreatic islets using the RNeasy Mini Kit (QIAGEN), following the manufacturer’s instructions. cDNA was prepared using QuantiTect Reverse Transcription Kit (QIAGEN), according to the manufacturer’s protocol. ## qRT-PCR. Real-time PCR was performed in triplicate with a Bio-Rad iCycler by cycling 40 times using the following conditions: 95°C for 10 seconds, 60°C for 45 seconds. Primers were designed using Primer3 as above and tested for linear amplification using serial dilutions of cDNA before use on experimental samples. Sequence information is provided in Supplemental Table 4. ## Measurement of β cell proliferation and apoptosis. Pancreatic slices were prepared from Trpm7R/R and control littermates after 16 weeks of chow or HFD. To study β cell proliferation, pancreatic islets were costained for insulin and Ki67. Ki67-insulin double-positive cells were counted and divided by the total number of insulin-positive cells per pancreatic section. To investigate β cell apoptosis, ApopTag Red In Situ Apoptosis Detection Kit was used according to the manufacturer’s (Merck) instructions. TUNEL-insulin double-positive cells were counted and divided by the total number of insulin-positive cells per pancreatic section. ## RNA-Seq studies. RNA-Seq data have been uploaded to the NCBI’s Gene Expression Omnibus under the accession number GSE218030 (https://www.ncbi.nlm.nih.gov/geo). Total RNA was extracted from isolated pancreatic islets of Trpm7R/R and their control littermates, which had been maintained on an HFD for 16 weeks. Template amplification and clustering were performed on the NovaSeq 6000 (Illumina) applying the exclusion amplification (ExAmp) chemistry. The ExAmp workflow is a proprietary Illumina method and ensures that only single DNA templates are bound within single wells of the patterned NovaSeq flow cells and are almost instantaneously amplified. Cluster generation and sequencing were operated under the control of the NovaSeq Control Software v1.6.0. The P value of a pairwise comparison was derived from the Wald test. To control the false-positive rate, FDR-corrected [64] as well as Bonferroni-corrected P values were calculated, where FDR is the proportion of false-positive hits among all positive hits. A gene or transcript is classified as upregulated or downregulated in a specific comparison if its FDR-corrected P value is ≤0.05 and its fold change is ≥2. ## Bio-Plex Pro cell signaling assay. Murine pancreatic islets were washed and lysed in MILLIPLEX MAP Lysis Buffer (Merck). Protein content was measured using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, catalog 23225). Samples were stored at −80°C. Collected samples were processed and assayed according to manufacturer’s instructions specific for MILLIPLEX MAP TGFβ Signaling Pathway Magnetic Bead 6-Plex kit (Merck, catalog 48-614MAG), and MILLIPLEX MAP β-Tubulin Total Magnetic Bead MAPmate (Merck, catalog 46-713MAG). ## Cell culture. Mouse WT and kinase-dead TRPM7 in pIRES-EGFP vector were reported previously [65]. MIN6 cells were provided by Per-Olof Berggren and Barbara Leibiger, Karolinska Institutet, Stockholm, Sweden. MIN6 cells were grown at 37°C and $5\%$ CO2 in DMEM (MilliporeSigma) supplemented with $10\%$ FBS (Thermo Fisher Scientific), 100 U/mL penicillin and 100 μg/mL streptomycin (MilliporeSigma), and 75 μM β-Mercaptoethanol (Gibco). Cells with approximately $60\%$ confluence in 96-well plates or 6 cm dishes were transiently transfected by 0.1 or 2 μg cDNAs, respectively. TurboFect was used as a transfection reagent (Thermo Fisher Scientific). GIIS was measured 48 hours after transfection in 96-well plates. Cells were harvested 48 hours after transfection from 6 cm dishes for Western blotting. ## Statistics. Data were expressed as mean ± SEM. P value less than 0.05 was considered significant. Graph presentations, curve fittings, statistics, and P values were obtained by Prism software (version 9.0.1; GraphPad). For comparison of 2 groups, P values were calculated by the unpaired 2-tailed Student’s t test for parametric or Mann-Whitney test for nonparametric distribution. For 3 or more groups, 1-way ANOVA with Bonferroni’s multiple comparison were used for parametrically distributed data. Glucose and insulin tolerance tests were compared using 2-way ANOVA with Bonferroni’s multiple comparison. ## Study approval. All animal experiments were performed in accordance with the EU Animal Welfare Act and were approved by the District Government of Upper Bavaria, Munich, Germany, on animal care (permit no. 55.2-2532.Vet_02-19-035). ## Author contributions NK designed and conducted experiments, analyzed and interpreted data, prepared figures, and wrote the manuscript. A Beck, KR, PB, KJ, SFS, A Belkacemi, PCFS, HS, and TP conducted experiments, analyzed and interpreted data, and edited the manuscript. PSR, AN, A Breit, VC, TDM, and SZ interpreted data and edited the manuscript. TG directed the project, designed experiments, interpreted data, and edited the manuscript. ## 12/27/2022 In-Press Preview ## 02/08/2023 Electronic publication ## References 1. Prentki M, Nolan CJ. Islet beta cell failure in type 2 diabetes. JCI Insight (2006) 116 1802-1812. DOI: 10.1172/JCI29103 2. Schmitz C. Regulation of vertebrate cellular Mg2+ homeostasis by TRPM7. Cell (2003) 114 191-200. DOI: 10.1016/S0092-8674(03)00556-7 3. Mittermeier L. TRPM7 is the central gatekeeper of intestinal mineral absorption essential for postnatal survival. Proc Natl Acad Sci U S A (2019) 116 4706-4715. DOI: 10.1073/pnas.1810633116 4. Pham PC. Hypomagnesemia in patients with type 2 diabetes. Clin J Am Soc Nephrol (2007) 2 366-373. DOI: 10.2215/CJN.02960906 5. Rodríguez-Morán M, Guerrero-Romero F. Oral magnesium supplementation improves insulin sensitivity and metabolic control in type 2 diabetic subjects: a randomized double-blind controlled trial. Diabetes Care (2003) 26 1147-1152. DOI: 10.2337/diacare.26.4.1147 6. Ryazanov AG. Elongation factor-2 kinase and its newly discovered relatives. FEBS Lett (2002) 514 26-29. DOI: 10.1016/S0014-5793(02)02299-8 7. Zierler S. TRPM channels as potential therapeutic targets against pro-inflammatory diseases. Cell Calcium (2017) 67 105-115. DOI: 10.1016/j.ceca.2017.05.002 8. Drennan D, Ryazanov AG. Alpha-kinases: analysis of the family and comparison with conventional protein kinases. Prog Biophys Mol Biol (2004) 85 1-32. DOI: 10.1016/S0079-6107(03)00060-9 9. Riazanova LV. [Novel type of signaling molecules: protein kinases covalently linked to ion channels]. Mol Biol (Mosk) (2001) 35 321-332. PMID: 11357414 10. Yamaguchi H. Crystal structure of the atypical protein kinase domain of a TRP channel with phosphotransferase activity. Mol Cell (2001) 7 1047-1057. DOI: 10.1016/S1097-2765(01)00256-8 11. Jin J. The channel kinase, TRPM7, is required for early embryonic development. Proc Natl Acad Sci U S A (2012) 109 E225-E233. DOI: 10.1073/pnas.1120033109 12. Yu Y. TRPM7 is involved in angiotensin II induced cardiac fibrosis development by mediating calcium and magnesium influx. Cell Calcium (2014) 55 252-260. DOI: 10.1016/j.ceca.2014.02.019 13. Sahni J, Scharenberg AM. TRPM7 ion channels are required for sustained phosphoinositide 3-kinase signaling in lymphocytes. Cell Metab (2008) 8 84-93. DOI: 10.1016/j.cmet.2008.06.002 14. Fang L. TRPM7 channel regulates PDGF-BB-induced proliferation of hepatic stellate cells via PI3K and ERK pathways. Toxicol Appl Pharmacol (2013) 272 713-725. DOI: 10.1016/j.taap.2013.08.009 15. Sahni J. TRPM7 regulates quiescent/proliferative metabolic transitions in lymphocytes. Cell Cycle (2010) 9 3565-3574. DOI: 10.4161/cc.9.17.12798 16. Yee NS. Transient receptor potential ion channel Trpm7 regulates exocrine pancreatic epithelial proliferation by Mg2+-sensitive Socs3a signaling in development and cancer. Dis Model Mech (2011) 4 240-254. DOI: 10.1242/dmm.004564 17. Krapivinsky G. The TRPM7 chanzyme is cleaved to release a chromatin-modifying kinase. Cell (2014) 157 1061-1072. DOI: 10.1016/j.cell.2014.03.046 18. Abed E, Moreau R. Importance of melastatin-like transient receptor potential 7 and magnesium in the stimulation of osteoblast proliferation and migration by platelet-derived growth factor. Am J Physiol Cell Physiol (2009) 297 C360-C368. DOI: 10.1152/ajpcell.00614.2008 19. Beesetty P. Inactivation of TRPM7 kinase in mice results in enlarged spleens, reduced T-cell proliferation and diminished store-operated calcium entry. Sci Rep (2018) 8. DOI: 10.1038/s41598-018-21004-w 20. Faouzi M. The TRPM7 channel kinase regulates store-operated calcium entry. J Physiol (2017) 595 3165-3180. DOI: 10.1113/JP274006 21. Altman MK. TRPM7 is a crucial regulator of pancreatic endocrine development and high-fat-diet-induced β-cell proliferation. Development (2021) 148. DOI: 10.1242/dev.194928 22. Kono T. Impaired store-operated calcium entry and STIM1 loss lead to reduced insulin secretion and increased endoplasmic reticulum stress in the diabetic β-cell. Diabetes (2018) 67 2293-2304. DOI: 10.2337/db17-1351 23. Islam MS. Molecular regulations and functions of the transient receptor potential channels of the islets of Langerhans and insulinoma cells. Cells (2020) 9. DOI: 10.3390/cells9030685 24. Gommers LMM. Low extracellular magnesium does not impair glucose-stimulated insulin secretion. PLoS One (2019) 14. DOI: 10.1371/journal.pone.0217925 25. Sachdeva MM. Pdx1 (MODY4) regulates pancreatic beta cell susceptibility to ER stress. Proc Natl Acad Sci U S A (2009) 106 19090-19095. DOI: 10.1073/pnas.0904849106 26. Wang P. Diabetes mellitus--advances and challenges in human β-cell proliferation. Nat Rev Endocrinol (2015) 11 201-212. DOI: 10.1038/nrendo.2015.9 27. Golson ML. High fat diet regulation of β-cell proliferation and β-cell mass. Open Endocrinol J (2010) 4. DOI: 10.2174/1874216501004010066 28. Romagnani A. TRPM7 kinase activity is essential for T cell colonization and alloreactivity in the gut. Nat Commun (2017) 8. DOI: 10.1038/s41467-017-01960-z 29. Zhang Y. TGF-β family signaling in the control of cell proliferation and survival. Cold Spring Harb Perspect Biol (2017) 9. DOI: 10.1101/cshperspect.a022145 30. Melloul D. Regulation of pdx-1 gene expression. Diabetes (2002) 51 Suppl 3 S320-S325. DOI: 10.2337/diabetes.51.2007.S320 31. Seufert J. Differential expression of the insulin gene transcriptional repressor CCAAT/enhancer-binding protein beta and transactivator islet duodenum homeobox-1 in rat pancreatic beta cells during the development of diabetes mellitus. J Clin Invest (1998) 101 2528-2539. DOI: 10.1172/JCI2401 32. Ahn YB. Changes in gene expression in beta cells after islet isolation and transplantation using laser-capture microdissection. Diabetologia (2007) 50 334-342. DOI: 10.1007/s00125-006-0536-5 33. Chakrabarti SK. Quantitative assessment of gene targeting in vitro and in vivo by the pancreatic transcription factor, Pdx1. Importance of chromatin structure in directing promoter binding. J Biol Chem (2002) 277 13286-13293. DOI: 10.1074/jbc.M111857200 34. Ohlsson H. IPF1, a homeodomain-containing transactivator of the insulin gene. EMBO J (1993) 12 4251-4259. DOI: 10.1002/j.1460-2075.1993.tb06109.x 35. Moin ASM, Butler AE. Alterations in beta cell identity in type 1 and type 2 diabetes. Curr Diab Rep (2019) 19. DOI: 10.1007/s11892-019-1194-6 36. Guo S. Inactivation of specific β cell transcription factors in type 2 diabetes. J Clin Invest (2013) 123 3305-3316. DOI: 10.1172/JCI65390 37. Hayes HL. Pdx-1 activates islet α- and β-cell proliferation via a mechanism regulated by transient receptor potential cation channels 3 and 6 and extracellular signal-regulated kinases 1 and 2. Mol Cell Biol (2013) 33 4017-4029. DOI: 10.1128/MCB.00469-13 38. Reers C. Impaired islet turnover in human donor pancreata with aging. Eur J Endocrinol (2009) 160 185-191. DOI: 10.1530/EJE-08-0596 39. Kulkarni RN. PDX-1 haploinsufficiency limits the compensatory islet hyperplasia that occurs in response to insulin resistance. J Clin Invest (2004) 114 828-836. DOI: 10.1172/JCI21845 40. Chubanov V. Role of kinase-coupled TRP channels in mineral homeostasis. Pharmacol Ther (2018) 184 159-176. DOI: 10.1016/j.pharmthera.2017.11.003 41. Rane SG. Loss of Cdk4 expression causes insulin-deficient diabetes and Cdk4 activation results in beta-islet cell hyperplasia. Nat Genet (1999) 22 44-52. DOI: 10.1038/8751 42. Kim SY, Rane SG. The Cdk4-E2f1 pathway regulates early pancreas development by targeting Pdx1+ progenitors and Ngn3+ endocrine precursors. Development (2011) 138 1903-1912. DOI: 10.1242/dev.061481 43. Lee JH. Cdk4 regulates recruitment of quiescent beta-cells and ductal epithelial progenitors to reconstitute beta-cell mass. PLoS One (2010) 5. DOI: 10.1371/journal.pone.0008653 44. Masuda T. Cold-inducible RNA-binding protein (Cirp) interacts with Dyrk1b/Mirk and promotes proliferation of immature male germ cells in mice. Proc Natl Acad Sci U S A (2012) 109 10885-10890. DOI: 10.1073/pnas.1121524109 45. Laybutt DR. Endoplasmic reticulum stress contributes to beta cell apoptosis in type 2 diabetes. Diabetologia (2007) 50 752-763. DOI: 10.1007/s00125-006-0590-z 46. Barella LF. β-Arrestin-1 is required for adaptive β-cell mass expansion during obesity. Nat Commun (2021) 12. DOI: 10.1038/s41467-021-23656-1 47. Johnson JS. Pancreatic and duodenal homeobox protein 1 (Pdx-1) maintains endoplasmic reticulum calcium levels through transcriptional regulation of sarco-endoplasmic reticulum calcium ATPase 2b (SERCA2b) in the islet β cell. J Biol Chem (2014) 289 32798-32810. DOI: 10.1074/jbc.M114.575191 48. Elrick LJ, Docherty K. Phosphorylation-dependent nucleocytoplasmic shuttling of pancreatic duodenal homeobox-1. Diabetes (2001) 50 2244-2252. DOI: 10.2337/diabetes.50.10.2244 49. Jara MA. Pancreatic and duodenal homeobox-1 (PDX1) contributes to β-cell mass expansion and proliferation induced by Akt/PKB pathway. Islets (2020) 12 32-40. DOI: 10.1080/19382014.2020.1762471 50. Khoo S. Regulation of insulin gene transcription by ERK1 and ERK2 in pancreatic beta cells. J Biol Chem (2003) 278 32969-32977. DOI: 10.1074/jbc.M301198200 51. Huang X. The PI3K/AKT pathway in obesity and type 2 diabetes. Int J Biol Sci (2018) 14 1483-1496. DOI: 10.7150/ijbs.27173 52. Bernal-Mizrachi E. Islet beta cell expression of constitutively active Akt1/PKB alpha induces striking hypertrophy, hyperplasia, and hyperinsulinemia. J Clin Invest (2001) 108 1631-1638. DOI: 10.1172/JCI200113785 53. Tuttle RL. Regulation of pancreatic beta-cell growth and survival by the serine/threonine protein kinase Akt1/PKBalpha. Nat Med (2001) 7 1133-1137. DOI: 10.1038/nm1001-1133 54. Kitamura T. The forkhead transcription factor Foxo1 links insulin signaling to Pdx1 regulation of pancreatic beta cell growth. J Clin Invest (2002) 110 1839-1847. DOI: 10.1172/JCI200216857 55. Roy SK. Inhibition of PI3K/AKT and MAPK/ERK pathways causes activation of FOXO transcription factor, leading to cell cycle arrest and apoptosis in pancreatic cancer. J Mol Signal (2010) 5. DOI: 10.1186/1750-2187-5-10 56. Kim TK. Triiodothyronine induces proliferation of pancreatic β-cells through the MAPK/ERK pathway. Exp Clin Endocrinol Diabetes (2014) 122 240-245. DOI: 10.1055/s-0034-1367060 57. Stewart AF. Human β-cell proliferation and intracellular signaling: part 3. Diabetes (2015) 64 1872-1885. DOI: 10.2337/db14-1843 58. Nadolni W. TRPM7 kinase is essential for neutrophil recruitment and function via regulation of Akt/mTOR signaling. Front Immunol (2020) 11. DOI: 10.3389/fimmu.2020.606893 59. Fatrai S. Akt induces beta-cell proliferation by regulating cyclin D1, cyclin D2, and p21 levels and cyclin-dependent kinase-4 activity. Diabetes (2006) 55 318-325. DOI: 10.2337/diabetes.55.02.06.db05-0757 60. Jin J. Deletion of Trpm7 disrupts embryonic development and thymopoiesis without altering Mg2+ homeostasis. Science (2008) 322 756-760. DOI: 10.1126/science.1163493 61. Kaitsuka T. Inactivation of TRPM7 kinase activity does not impair its channel function in mice. Sci Rep (2014) 4. DOI: 10.1038/srep05718 62. Khajavi N. An incretin-based tri-agonist promotes superior insulin secretion from murine pancreatic islets via PLC activation. Cell Signal (2018) 51 13-22. DOI: 10.1016/j.cellsig.2018.07.006 63. Khajavi N. L-carnitine reduces in human conjunctival epithelial cells hypertonic-induced shrinkage through interacting with TRPV1 channels. Cell Physiol Biochem (2014) 34 790-803. DOI: 10.1159/000363043 64. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol (1995) 57 289-300 65. Schmidt E. Structural mechanism of TRPM7 channel regulation by intracellular magnesium. Cell Mol Life Sci (2022) 79. DOI: 10.1007/s00018-022-04192-7
--- title: Mitochondrial and NAD+ metabolism predict recovery from acute kidney injury in a diverse mouse population authors: - Jean-David Morel - Maroun Bou Sleiman - Terytty Yang Li - Giacomo von Alvensleben - Alexis M. Bachmann - Dina Hofer - Ellen Broeckx - Jing Ying Ma - Vinicius Carreira - Tao Chen - Nabil Azhar - Romer A. Gonzalez-Villalobos - Matthew Breyer - Dermot Reilly - Shannon Mullican - Johan Auwerx journal: JCI Insight year: 2023 pmcid: PMC9977436 doi: 10.1172/jci.insight.164626 license: CC BY 4.0 --- # Mitochondrial and NAD+ metabolism predict recovery from acute kidney injury in a diverse mouse population ## Abstract Acute kidney failure and chronic kidney disease are global health issues steadily rising in incidence and prevalence. Animal models on a single genetic background have so far failed to recapitulate the clinical presentation of human nephropathies. Here, we used a simple model of folic acid–induced kidney injury in 7 highly diverse mouse strains. We measured plasma and urine parameters, as well as renal histopathology and mRNA expression data, at 1, 2, and 6 weeks after injury, covering the early recovery and long-term remission. We observed an extensive strain-specific response ranging from complete resistance of the CAST/EiJ to high sensitivity of the C57BL/6J, DBA/2J, and PWK/PhJ strains. In susceptible strains, the severe early kidney injury was accompanied by the induction of mitochondrial stress response (MSR) genes and the attenuation of NAD+ synthesis pathways. This is associated with delayed healing and a prolonged inflammatory and adaptive immune response 6 weeks after insult, heralding a transition to chronic kidney disease. Through a thorough comparison of the transcriptomic response in mouse and human disease, we show that critical metabolic gene alterations were shared across species, and we highlight the PWK/PhJ strain as an emergent model of transition from acute kidney injury to chronic disease. ## Introduction Kidney disease is a global burden on our health system. A large fraction of the population — more than $20\%$ for acute kidney injury (AKI) [1] and for $10\%$ for chronic kidney disease (CKD) [2] — suffer from kidney diseases. Despite being largely preventable, kidney disease is both a direct cause of morbidity and mortality as well as an important risk factor for various cardiovascular diseases [1]. AKI is mainly triggered by environmental factors — hypovolemic shock (e.g., infection, bleeding, vascular surgery), heart failure, and exposure to nephrotoxins (e.g., antibiotics, chemotherapy, contrast agents) [3]. Delayed graft failure, which is common in deceased donor transplants, is also a form of AKI of the transplanted kidney [3]. Although AKI can be reversible, a portion of AKI patients, especially those suffering from repeated injuries, do not recover fully and eventually develop CKD [1]. The severity of AKI can be influenced by genetics, from highly penetrant monogenic diseases to complex and multifactorial polygenic diseases [4, 5]. Likewise in animal models, different genetic backgrounds affect the onset and progression of kidney diseases induced either by a single-gene mutation [6] or the response to toxins [7]. The mechanisms and genetic determinants underpinning the severity of AKI and whether subjects recover from AKI or progress to CKD are poorly understood [8]. There exists no treatment for AKI, to date, and a mechanistic understanding of the disease and how it can progress to kidney fibrosis and CKD are urgently needed to formulate therapeutic strategies. In this study, we characterized the response to kidney injury in 7 mouse inbred strains known as the founders of the Collaborative Cross (CC), Diversity Outbred (DO), and BXD populations [9] (Figure 1A).The DO population has previously been used to investigate the genetic causes behind variations in glomerular filtration rate (GFR) assessed through Cystatin C [10], and researchers found associations with type I IFN genes and inflammatory pathways, but there was no induction of overt kidney disease. In a cross between X-Linked Alport Syndrome mice and DO mice, authors found associations between the X-linked gene Fmn1, albuminuria, and GFR disruption [11]. The purpose of our study was to establish a baseline of the responses of the founder strains of the DO population to AKI, to later define a suitable genetic cross to find the underlying genetic cause of susceptibility to AKI and transition to CKD. Rodent models of AKI consisting of various chemical (e.g., cisplatin, folic acid [FA]) or surgical (ischemia reperfusion injury [IRI] and colon ligation puncture [CLP]) injuries that impact on the kidney [12] often fail to reproduce the full features of human diseases [13]. Among AKI models, FA-induced nephropathy is a good compromise between simplicity and relevance to human kidney diseases [14], and it is frequently used in preclinical drug development [15, 16]. In rodents, yellow crystals of FA are detected in the kidney as early as 30 minutes after injection with a toxic dose of FA (440 mg/kg); then, those crystals are fully cleared over approximately 4 days [14]. Such high concentrations of FA lead to kidney injury, both through crystal toxicity and direct nephrotoxicity [17, 18], causing an increase of serum creatinine, blood urea nitrogen (BUN), and urine glucose levels [14]. In this study, we followed the evolution of injury and remission after the expected clearance of crystals 1, 2, and 6 weeks after a i.p. FA injection in 5 domesticated (C57BL/6J, DBA/2J, A/J, 129S1/SvlmJ, WSB/EiJ) and 2 wild-derived (CAST/EiJ, PWK/PhJ) inbred mouse strains (Figure 1A). After a short pilot experiment, we chose a mild (125 mg/kg) dose of FA because higher doses caused severe toxicity, leading to death of some strains. We measured plasma and urine biomarkers, as well as histological kidney features, and determined the renal transcript profiles. Through this approach, we determined that the C57BL/6J, DBA/2J, and PWK/PhJ mouse strains are the most susceptible strains to AKI and possibly to the transition to CKD. In addition, we observed that modulation of immune and mitochondrial pathways, notably NAD+ metabolism, during the initial recovery are predictive of long-term remission. We provide access to all data in this study through public repositories, as well as an interactive app, enabling users to explore and analyze all traits and gene expression changes observed in this study at www.systems-genetics.org/CC_founders_AKI This resource will help guide the selection of relevant strains for specific kidney disease modeling, thereby improving the translational potential of further research. ## Mouse strains exhibit a wide range of responses to AKI, from full resistance to chronic disease. Kidney injury was induced by injecting a single dose of folic acid (FA) i.p. ( 125 mg/kg in 0.3M of sodium bicarbonate) in 8-week-old male mice ($$n = 6$$ per strain and condition for a total of 210 animals). Mice were sacrificed 1, 2, or 6 weeks after dosing, in order to capture the initial recovery and long-term remission phases (Figure 1B). In accordance with the 3R principles (replace, reduce, refine), control mice were only sacrificed at 2 or 6 weeks. Therefore, throughout the manuscript, week 1 FA-treated mice were always compared with week 2 control mice. We performed principal component analysis (PCA) on the phenotype data of all mice to explore the relationship between strains and conditions as a function of time (Figure 1C). Control mice of the wild-derived WSB/EiJ and CAST/EiJ are separated from the other strains on the first component, which explains $26.64\%$ of the variance. The WSB/EiJ — and, to a lesser extent, CAST/EiJ — have high baseline urinary albumin and creatinine and low FGF-21 (Figure 1C, online app; www.systems-genetics.org/CC_founders_AKI). After 1 week of FA injection, the measured phenomes of PWK/PhJ, DBA/2J, and C57BL/6J mice shift toward the upper right quadrant of the PCA plot. This shift is associated with an increase in plasma BUN and creatinine, kidney weight, and fibrosis measured by quantification of Sirius red staining in kidney sections. At weeks 2 and 6, the FA mice from all strains gradually clustered back with the control mice, except for PWK/PhJ, C57BL/6J, and DBA/2J in the lower right quadrant characterized by high circulating Fgf-21 and kidney fibrosis (Figure 1C, online app), suggestive of a slow recovery or persistent disease at week 6. Body weight (BW) was measured throughout the study as a general readout of disease severity and its subsequent recovery. It decreased after 1 week of FA in most strains (Figure 1D). The A/J and 129S1/SvlmJ strains were most prone to BW loss, while the wild-derived strain CAST/EiJ lost almost no weight upon FA injury. Consistently, food intake decreased in laboratory mouse strains but not in wild-derived mice (Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.164626DS1). All strains recovered normal BW at week 2, except for the DBA/2J and 129S1/SvlmJ strains (Figure 1D). Kidney weight increased in the acute phase in some responsive strains (C57BL/6J, A/J, and PWK/PhJ; Figure 1D), as previously reported [14]; then, it was reduced below the initial weight during the recovery phase at 6 weeks (Figure 1D). Spleen size increased by $40\%$ in the PWK/PhJ strain, reflecting immune activation, and this remained the case even after 6 weeks, perhaps suggesting the onset of chronic inflammation. Other strains (CAST/EiJ, A/J, and WSB/EiJ) transiently gained spleen weight at week 1, but this was resolved before week 6 (Figure 1D and Supplemental Figure 1B). We observed no major weight or visual aspect changes in other organs (Supplemental Figure 1B). Plasma and urine parameters were collected to assess the severity of kidney injury (Figure 1D and Supplemental Figure 2). Glycemia decreased upon FA in the strains that had the greatest change in BW (A/J, 129S1/SvlmJ, WSB/EiJ, and PWK/PhJ) and was maintained only in CAST/EiJ. Plasma markers of kidney injury — namely creatinine, BUN, and TIMP-1 — increased in all strains except for the highly resistant CAST/EiJ that only showed a transient increase in TIMP-1. While creatinine and TIMP-1 resolved itself in most strains, BUN remained high in 2 strains even at 6 weeks. Urine creatinine and albumin levels, and the respective ratios, were less responsive than plasma levels (Figure 1D and Supplemental Figure 2), possibly because there was a difference in urinary output, which we unfortunately did not measure and could partially compensate for high urinary protein [19]. Urine creatinine levels were only significantly affected by FA in PWK/PhJ (week 6). Urine albumin levels were affected in 129S1/SvlmJ (increased at week 1), CAST/EiJ (increased at week 1 and week 2), and C57BL/6J (increased at week 6). In addition, the urine albumin/creatinine (Alb/Cru) ratio indicated that 129S1/SvlmJ (week 1), DBA/2J, CAST/EiJ (week 2), and C57BL/6J (week 6) are the most sensitive. The creatinine urine/plasma (Cru/p) ratio seems to be consistent with the plasma parameters and reflects a large decrease in C57BL/6J, DBA/2J, A/J, and PWK/PhJ. The circulating factors GDF-15 and FGF-21 are known to have a prognostic and even predictive value in CKD in humans [20, 21], but these markers are not exclusive to kidney injuries and are elevated in a wide range of conditions involving mitochondrial stress and injury, giving them the designation of mitokines [22]. Both mitokines were elevated in most strains at week 1 (Figure 1D), but there was some heterogeneity between strains, with the C57BL/6J and PWK/PhJ strains showing a far stronger induction of GDF-15, the same strains that retain elevated kidney injury markers at week 6. This result may indicate that mitochondrial stress is a predictor of the susceptibility to AKI. Overall, these results allow us to classify the strains into 3 groups, based on their response to FA-mediated kidney injury: Group I indicates resistant strains, including CAST/EiJ, a wild-derived strain from the Mus musculus castaneus subspecies, which shows only a minimal response to injury. Group II indicates recovering strains (A/J, 129SvlmJ, WSB/EiJ). These strains show a variable initial response to injury but subsequently recover almost fully at 6 weeks. Group III indicates sensitive strains (C57BL/6J, PWK/PhJ, and — to an extent — DBA2/J). These strains retain strong markers of kidney injury and/or inflammation at 6 weeks, indicating a very slow recovery or a transition to chronic disease. ## Kidney histopathology recapitulates strain-specific responses to kidney injury. We performed H&E and Sirius red staining of parallel kidney sections of each animal and focused on the same region of the cortex and medulla in each staining (Figure 2A). The medulla and cortex were scored by a pathologist in a double-blinded manner. Representative examples of images corresponding to a severity from 0 to 4 are shown with the corresponding score (Figure 2A), along with the full quantification (Figure 2B). The histological severity scores in both medulla and cortex closely followed the trends that were observed in the biochemical parameters of kidney injury. The CAST/EiJ (Group I) strain showed no significant difference with controls at any time, suggesting resistance to injury. Group II strains (A/J, 129SvlmJ, WSB/EiJ) displayed mostly tubule degeneration, interstitial inflammation, and fibrosis but no tubule dilation at week 1 and had no significant histological markers from week 2 onwards, indicating a fast recovery from injury (Figure 2B). At week 1, sensitive strains (Group III: C57BL/6J, PWK/PhJ, DBA2/J) showed tubule dilation, tubule degeneration, interstitial inflammation, and fibrosis (Figure 2B); although variability was high and there were high average scores, these data did not always reach statistical significance. Group III strains retained tubule degeneration and inflammation (PWK/PhJ and DBA/2J) or fibrosis (C57BL/6J and PWK/PhJ) even after 6 weeks (Figure 2B). Notably, we detected no hyaline casts or glomerulopathy in any of the strains — changes that are not expected in AKI and help explain the mild changes observed in the Alb/Cru ratio. ## The kidney transcriptomic response to AKI is qualitatively similar across strains. To gain insight into the molecular pathways associated with the strain differences in kidney injury severity, we profiled the kidney transcriptomes of all 202 mice using RNA-Seq analysis. Through PCAs of the expression levels (Figure 3A) and the fold changes compared with control (Figure 3B), we observed that the principal component 1 (PC1) separated strains according to treatment and time, while the second PC2 separated strains from each other. PC2 is driven by mouse strain differences, which are greatest in different subspecies’ PWK/PhJ (Mus musculus musculus) and CAST/EiJ (Mus musculus castaneus) (Figure 1A). Despite the differences in injury severity and recovery between strains, their transcriptomic response at week 1 was remarkably similar. At week 1, approximately 1,300 genes were differentially expressed in the same direction across all 7 strains (Figure 3C). A further ~2,800 genes were shared by all strains but the resistant CAST/EiJ. This similarity ended at weeks 2 and 6 when several strains no longer had any significant differentially expressed genes (Figure 3D), showing that all strains but the resistant CAST/EiJ reacted similarly to injury but that their responses diverged in the subsequent remission phase. ## The intensity of the kidney transcriptomic response parallels the severity of clinical injury. To better understand the transcriptomic response to FA, we performed a PCA of the expression fold changes between treatment and controls of the different strains at different time points (Figure 3B). Interestingly, the first PC capturing $27\%$ of variation faithfully traces the trajectory of the strains as they react to the injury and recover from it to varying extents, mirroring the histological, biochemical, and clinical phenotypes (Figure 3B). C57BL/6J, DBA/2J, A/J, and PWK/PhJ are the most responsive strains related to the transcripts that are consistently changed in the population. 129S1/SvlmJ and WSB/EiJ were less responsive, whereas CAST/EiJ was again resistant to changes in transcript levels induced by FA kidney injury. The sensitive strains in Group III (C57BL/6J and PWK/PhJ) retained a strong transcriptomic injury footprint at week 6. It is worth noting that the transcriptional effect of FA on PWK/PhJ at week 6 (Figure 3, B and D) was larger in magnitude than the acute effect on the CAST/EiJ mice, showing again the extensive range of severity in these strains. The overall transcriptomic response was mirrored by an increase in markers of tubular and kidney injury (Lcn2, Hacr1, Pdgfb; Figure 3E) and a decrease in markers of tubular identity (Slc9a3 and Lrp2) and slit diaphragm of podocytes (Nphs1, Nphs2), a critical component of the glomerular filter of the kidney. The circulating markers observed in the plasma (Gdf-15, Fgf-21, and Timp1) were also expressed higher at the transcript level, confirming their renal origin. Expression of transcripts encoding for different collagens (Col1a1, Col1a2, and many others) was higher across all strains, indicating fibrogenesis, while inflammatory (Il1b, Il6, Tnf) and antiinflammatory (Il10) cytokines increased together with the main components of the inflammasome, Nlrp3, and Casp1. The resistant CAST/EiJ strain showed increased transcript levels encoding for kidney injury markers Lcn2 and Havcr1 (Figure 3E) in the wake of mildly induced inflammatory mediators and collagens, indicating that it does suffer an initial injury. However, tubular and slit diaphragm proteins and, notably, the hallmarks of mitochondrial stress, Fgf21 and Gdf15, were not induced, and Fgf21 was significantly reduced (Figure 3E). Conversely in the Group III strains, PWK/PhJ, DBA/2J, and C57BL/6J, most markers of kidney injury and identity were normalized at week 6; however, these sensitive strains retained a robust expression of proinflammatory cytokines, inflammasome components, and collagens, indicating a transition from acute injury to chronic inflammation and fibrosis. Deficiency in EGF and TGF-α signaling through EGFR has been shown to be an important modifier of kidney structure and function and a risk factor in renal diseases [23]. The baseline expression of Egf was near constant across strains but declined strongly in all strains except the CAST/EiJ upon injury (Supplemental Figure 3A). Conversely, Tgfa expression was lower and varied across strains at baseline but did not correlate with injury (Supplemental Figure 3A). Despite these changes, the downstream targets of the pathway were not significantly enriched upon injury in any of the strains (Supplemental Figure 3B), suggesting that this pathway is not a determinant of sensitivity in our model. ## Folate metabolism and transport do not explain strain susceptibility to injury. The enzyme Dhfr can metabolize low doses of FA to the nontoxic 5,6,7,8-tetrahydrofolic acid. Dhfr activity is low in humans but effective in rodents [24]. Basal Dhfr expression was lowest in the sensitive PWK/PhJ strain but remained high in C57BL/6J, which is also susceptible to injury (Supplemental Figure 4A). Upon FA, Dhfr expression was reduced in most strains, but the reduction was lowest in the resistant CAST/EiJ strain and highest in the sensitive C57BL/6J and PWK/PhJ strains (Supplemental Figure 4B). However, this may be a consequence of glomerular injury, rather than a cause. It is unclear how much direct catabolism of FA by Dhfr compares with urinary excretion at high doses of FA. To have a more complete picture of folate metabolism across our strains, we measured the expression of proteins involved in mitochondrial folate transport (Supplemental Figure 4C), high-affinity FA transport [25] (Supplemental Figure 4D), folate metabolism downstream of DHFR [26] (Supplemental Figure 4E), and low-affinity folate transporters [25] (Supplemental Figure 4F). All of these genes present a diverse pattern of expression across strains, consistent with the diversity between the represented mouse strains and subspecies. However, these variations have no consistent correlation with the severity of the injury seen in each strain, suggesting that they are not responsible for the observed difference in injury susceptibility or speed of recovery. ## The kidney transcriptome shows an association between early mitochondrial stress and later chronic inflammation. Given that some of the observed transcriptional differences may be due to changes in the proportions of different cell populations, we performed cell type deconvolution using single-cell data from another study [27] (Figure 4A). We detected an infiltration of immune cells in all strains, except the CAST/EiJ, while the amount of proximal and distal tubule cells was reduced in all strains following injury. We also found an expansion of the proportion of podocytes across all strains (although, once again, greatest in Group III strains), which, to our knowledge, has never been reported before and may be an artifact of the deconvolution procedure, indicating that further inquiry would be required to ascertain this. We performed a gene set enrichment analysis (GSEA) of the transcriptomic response to kidney injury by comparing either FA with control in each strain (Figure 4B) or by comparing the response in each strain with the response in every other strain (Figure 4C). The comparison with control (Figure 4B) again highlighted how universal the transcriptomic response to AKI was. In all strains and at nearly all time points, even including the resistant CAST/EiJ, we observed an induction of fibrosis and collagen-related gene sets, an induction of immune-related gene sets notably the proinflammatory IFN- and IL-1–mediated responses as well as a reduction in mitochondria-related gene sets. Targets of the critical immune-related NF-κB and Stat families of transcription factors were among the most enriched. A more complete selection of gene sets is also included (Supplemental Figure 5). GSEA uses the ranking of genes rather than their absolute values, and this ranking is similar in all strains, including the resistant CAST/EiJ, showing that the differences between strains are matters of scale, rather than completely different responses. To better highlight the differences between strains, we performed a differential expression analysis and GSEA on the comparison between each strain and every other strain (Figure 4C). From this, we could infer that the sensitive strains (Group III) differed from recovering strains (Group II) primarily by an early (week 1) downregulation of mitochondria-related gene sets, followed by a late (week 3) activation of immune related gene sets (Figure 4C). While the overall trend of loss of mitochondrial content and increased inflammation was present in all strains (Figure 5A), their evolution over time strongly differed between Group II and Group III strains. Group II strains lose little mitochondrial gene expression early and then have reduced adaptive immunity later, while sensitive Group III showed a strongly repressed early mitochondrial gene expression and later develop persistent T cell–and B cell–mediated adaptive immunity (most notably the PWK/PhJ). This enhanced expression of adaptive immune response genes, characterized by B and T cell markers, as well as inflammasome activation and Il1b, Il6, and Tnf expression (Figure 3E), has all the features of the low-grade inflammation characteristic of CKD [28, 29]. ## Mitochondrial stress responses differ between resistant and sensitive strains. The transcript levels of the mitokine, Gdf-15, were robustly induced in most strains except in CAST/EiJ (Figure 1D and Figure 3E), and this may reflect differences in the induction of the mitochondrial stress response (MSR). To this end, we compared the induction of gene sets involved in the MSR against a wide range of gene sets representative of stress responses in other cellular compartments (Figure 5B). Each stress response was defined by a set of confirmed targets, drawn from overlapped ChIP-Seq and RNA-Seq data sets (Supplemental Table 1). Expression of gene sets indicative of the heat shock and oxidative stress responses were not significantly increased in any of the strains and time points, while transcripts for ER and integrated stress responses (ISR) were induced in most of them, although the induction did not always reach statistical significance (Figure 5B). MSR gene sets increased most at weeks 1 and 2 in all strains but the CAST/EiJ, and this difference between strains was highly significant (adjusted $P \leq 0.0001$). The gene sets reflecting the ISR were also not significantly enriched in CAST/EiJ mice, likely because it is triggered downstream of both ER and mitochondrial stresses [30], and neither the ER stress or MSR were induced in this strain. IFN-stimulated genes (ISGs) were also activated across all strains (Figure 5B). This immune response is also known to be activated through the cGAS-STING pathway by mitochondrial DNA (mtDNA) released by damaged or stressed mitochondria [31, 32] or released after an injury [33]. To assess whether the MSRs that we observed were also present at the protein level, we performed Western blot analysis of key mitochondrial oxidative stress components — mediators of the MSR, ISR, and ER stress — and downstream targets at the early time point, 1 week after kidney injury (Figure 5C) and also performed quantification (Figure 5D). All strains except the CAST/EiJ strain had reduced expression of mitochondrial electron chain (complexes I–V) and chaperone proteins (HSPA9, HSPD,1 and LONP1), and this loss was most pronounced in the PWK/PhJ strain (Figure 5, C and D). The central pathway of the ISR proceeds from EIF2α phosphorylation, leading to translation blockade and preferential translation of select proteins, most notably the translation factor ATF4 [34]. Both EIF2α phosphorylation and ATF4 were induced in most strains, testifying to the activation of the ISR (Figure 5, C and D). The mediators of the MSR, ATF5, ASNS, and TRIB3 were strongly increased [35], while BIP/GRP78, an indicator of ER stress, was only modestly increased (Figure 5, C and D). Interestingly, the mtDNA/nuclear DNA ratio significantly increased across strains (1-way ANOVA, $P \leq 0.001$), although it was only individually significant in the PWK/PhJ strain (Figure 5E) and there was no increase in the CAST/EiJ. The basal levels of mitochondrial proteins (Figure 5C) and mtDNA/nuclear DNA ratio (Figure 5D) showed little variation between strains at baseline, indicating that the mitochondrial quantity before injury was similar at baseline. This increase in mtDNA in damaged or stressed mitochondria is consistent with several studies that found accumulation of mtDNA in damaged or dysfunctional mitochondria (36–39). Across all these measurements, the PWK/PhJ strain was notable for the more pronounced loss of mitochondrial proteins and robust activation of the MSR, while the CAST/EiJ was almost nonresponsive. The CAST/EiJ strain exhibited a slight increase in BIP/GRP78, reflecting the induction of ER stress genes (Figure 5, C and D). The lack of induction of mitochondrial stress, but not other cellular stress pathways, is a unique feature of the CAST/EiJ and may play a role in its resistance to injury. ## Early loss of NAD+ synthesis and salvage, and increased consumption of NAD+, are early events that predict disease severity. The coenzyme nicotinamide adenine dinucleotide (NAD+) is critical for mitochondrial function and interventions on NAD+ have been shown by us and others to have wide-ranging beneficial effects on mitochondrial-related diseases (reviewed in ref. 40). In addition, several recent studies have implicated NAD+ metabolism in both chronic and acute forms of kidney disease — AKI (41–44), IRI [45] and diabetic kidney disease [46]. Furthermore, the central mediator of the ISR DDIT3/CHOP — which is downstream of both MSR and ER stress responses — was implicated as a potential regulator of NAD+ synthesis in AKI through the critical synthesis enzyme QPRT [43]. Given the strong differences observed in the MSR and ISR pathways in our study, we assessed the transcript levels of the critical enzymes involved in NAD+ de novo synthesis (Figure 6A and Supplemental Figure 6A), salvage (Figure 6B and Supplemental Figure 6A), and consumption (Figure 6C and Supplemental Figure 6A). While the CAST/EiJ strain undergoes almost no variation in expression of NAD+-related transcripts, the PWK/PhJ and, to a lesser extent, the C57BL/6J strain underwent a significant loss of NAD+ biosynthesis genes accompanied by a large increase in the NAD-consuming genes Cd38 and Parp1. Other strains presented an intermediate phenotype (Supplemental Figure 6A), and NAD+ synthesis genes were mostly restored at week 2, except in the PWK/PhJ and C57BL/6J strains, while only the PWK/PhJ strain retained reduced Qprt and increase Cd38 at week 6 (Supplemental Figure 6A). We then assessed NAD+ levels by high-performance liquid chromatography–mass spectrometry (HPLC-MS) and found that it was reduced at week 1, with a 3-fold reduction in the PWK/PhJ strain and lesser reductions in other strains except the CAST/EiJ and A/J strains (Figure 6D). NAD+ levels remained lower at weeks 2 and 6 but were only significant in a few strains (Supplemental Figure 6B). Baseline NAD+ levels (Figure 6D) and NAD+-related genes (Figure 6, A–C, and Supplemental Figure 6A) varied strongly across strains, but these baseline differences before injury had no obvious relation to resistance to injury. However, NAD+ levels upon injury across strains negatively correlated with biochemical (Figure 6, E–G, and Supplemental Figure 7A) and histological (Figure 6, H–J, and Supplemental Figure 7A) markers of disease severity, and many of these correlations were maintained even at weeks 2 and 6 (Supplemental Figure 7, B and C), indicating that early loss of NAD+ may be a biomarker of disease severity and duration and that the PWK/PhJ strain may be a suitable model for the role of NAD+ in human kidney diseases. ## The mouse transcriptomic response to FA matches human nephropathies. To address the relevance of these results to human pathologies, we used a set of gene signatures from analyses of a range of different human nephropathies, representing the transcriptomic response in both acute and chronic forms of kidney disease (Supplemental Table 2). For each human disease, we first examined the degree of overlap between differentially expressed genes in humans and those in our study in every strain and time point; we then performed pathway overrepresentation analyses to understand which pathways were shared between mice and humans (Figure 7A). At week 1, there was a strong overlap between FA-treated mice and human diseases representative of either CKD or AKI (Figure 7B), which was surprising considering that the AKI and CKD data sets we used had little overlap with each other (Supplemental Figure 9A); this overlap suggests that the mice at week 1 after FA may already have a phenotype somewhere between CKD and AKI. The overlap was strongest in PWK/PhJ and C57BL/6J mice, with — most often — between $40\%$ and $50\%$ of shared DEGs between mice and humans (Figure 7B). This validates our model of low-dose FA injury and subsequent recovery as highly representative of human disease. At week 6, the similarity between mouse DEGs and human AKI DEGs was mostly lost, while the PWK/PhJ and C57BL/6J strains retained a strong similarity to human CKD (Figure 7B), indicating that they may have transitioned from AKI to CKD. ## Critical pathways are shared between human kidney diseases and the FA response in C57BL/6J and PWK/PhJ strains. To assess which biological pathways were most shared between our mouse model and humans, we performed an overrepresentation analysis for each of the mouse condition–human disease pairs (Figure 7A and Supplemental Figure 8). Among the 5 most represented pathways, strains “regulation of immune system process” (green), representing immune activation, and “collagen-containing extracellular matrix” (orange), representing fibrogenesis, were consistently upregulated in mice and humans (Figure 7C), highlighting that the same genes are implicated in these pathway activations across species. Immune activation, but not fibrogenesis transcripts, was maintained at week 6 in the C57BL/6J and PWK/PhJ strains (Figure 7C). Among shared downregulated genes, “mitochondrion”, and “organic acid metabolic process” represented up to 3 quarters of shared downregulated transcripts between mice and human CKD at weeks 1 and 2, highlighting the absolutely central role of mitochondria and metabolism in kidney disease (Figure 7D). This loss of metabolic genes was stronger in CKD than AKI-related diseases, suggesting that our model may be especially suited to modelling metabolic alterations in CKD. An unexpected finding was a strong enrichment for downregulated targets of the transcription factor HNF1 (red; (Figure 7D), which is present in most strains at week 1 but is strongly increased in the PWK/PhJ strain at week 6. The HNF1 family of transcription factors consists of 2 members, HNF1α and HNF1β, which form heterodimers to regulate lipid and metabolic genes [47]. Mutations in Hnf1β cause renal cysts and renal function decline in both humans and mice [48, 49]. Similarly, Hnf1a variants were picked up in a recent genome-wide association studies (GWAS) for kidney function with > 1.2 million patients [50]. Our data, however, point to downregulation of HNF1 targets as a shared feature in many kidney diseases, and the PWK/PhJ strain may be especially suited to study this. The expression of Hnf1a and Hfn1b genes was reduced in both C57BL/6J and PWK/PhJ strains at week 1, and a trend toward downregulation was still present at week 6 in the PWK/PhJ (Supplemental Figure 9B). This comparison with humans highlights the validity of our mouse model of acute injury and transition to chronic disease, and it points to the PWK/PhJ strain as a promising model to study the role of both NAD+ and HNF1 in human kidney disease and the AKI-to-CKD transition. ## A web resource on kidney disease. The phenotypic traits and transcriptome data collected in this study can be explored with an online, interactive interface (www.systems-genetics.org/CC_founders_AKI). This resource enables researchers to examine the individual variation of FA-induced injury in all mouse strains and to choose an appropriate mouse model. ## Discussion In the clinic, the onset of AKI is often unpredictable; no one can anticipate a hemorrhagic shock, severe burn, or adverse reaction to antibiotics, but with the increased penetrance of personalized medicine, knowing one’s genetic predisposition to AKI and whether one will fully recover from this injury and avoid development of progressive fibrosis or CKD may become a reality in the near future. Our study was designed to emulate differences in genetic susceptibility on the clinical progression from AKI to interstitial fibrosis and CKD through a diverse panel of mouse strains. Across the 7 mouse strains, whether we consider the plasma and urine markers, organ weight changes (Figure 1), or histological markers (Figure 2), we observed a continuum of severity from “highly susceptible” (C57BL/6J, PWK/PhJ, DBA2/J) over “recovers well” (A/J, 129SvlmJ, WSB/EiJ) to “fully resistant” (CAST/EiJ). Mitochondrial defects are one of the earliest events observed in human AKI [51], and defects in mitochondria have been reported in FA nephropathy mouse models and have been proposed as either potential biomarkers [42, 52] or therapeutic targets [42, 53]. A distinguishing feature of the resistant CAST/EiJ strain compared with more susceptible strains was the absence of the induction of the MSR. This difference in MSR in the CAST/EiJ was confirmed at the protein level and was supported by a lack of GDF-15 induction, a mitokine known to signal mitochondrial stress and influence CKD progression [20, 54] (Figure 1D and Figure 3E). FA can directly be taken up by mitochondria and cause mitochondrial damage, and the kidney toxicity of FA was shown to be prevented by N-acetyl-cysteine before administration [55]. This direct effect of FA on mitochondria can be a potential confounder in our model. However, the mitochondrial defects we observed strongly overlap decreased mitochondrial genes in human CKD, which suggests that this model remains relevant to human disease. In our comparison with human disease, the CAST/EiJ strain shared an early induction of immunity and fibrogenesis (Figure 7C) but lacked the concomitant reduction in mitochondrial and metabolism transcripts (Figure 7D). The critical metabolic coenzyme NAD+ is increasingly recognized as a central modulator of both mitochondrial-related diseases [40] and acute and CKD (41–46). GWAS have also revealed different kidney susceptibility associated with NAD+-related genes. The rate-limiting NAD+ salvage enzyme NAMPT was associated with serum creatinine and eGFR in one UK biobank–based GWAS [56], and SIRT-1 was associated with BUN in a cross-population study [57] (accessed through the NHGRI-EBI GWAS catalog; ref. 58), highlighting the importance of this pathway in humans. Although we cannot conclude on whether loss of NAD+ was the cause or the consequence of kidney damage, NAD+ levels at an early time point were indicative of both disease severity and duration. Our results further suggest that MSR genes, mitokines such as GDF-15, and metabolic-related transcription factors such as HNF1 may be of particular interest as both biomarkers and targets in early AKI. The association between early mitochondrial stress and prolonged inflammation evident at later stages after the initial insult may be linked to the release of mitochondrial DAMPs [33] and the activation of cGAS-STING and IFN signaling feeding into the inflammatory NF-κB pathway — features that were also evident in our susceptible strains. Genetic studies of the susceptibility to AKI point to a complex multigenic environment, but inflammatory genes — chiefly TNFa, Il6, STAT1 and NFKB-1 — are among the most important genetic determinants [5]. In mouse strains, Tnf-a and Il6 were highly correlated with the severity of the injury (Figure 3E), as were enrichments of NF-κB and Stat targets (Figure 4C), highlighting similar trends across species. Of note, such immune regulation genes were shared between the PWK/PhJ strain and most forms of human CKD at week 6 (Figure 7B). The PWK/PhJ strain was particularly notable for its prolonged and intense upregulation of the Nlrp3 inflammasome, which remains strongly upregulated at week 6 (Figure 3E). In mice, Nlrp3 inflammasome activation in macrophages is thought to be sufficient to trigger chronic inflammation [59], and its strong upregulation specifically in the PWK/PhJ strain may be a further indicator of a transition toward chronic disease. Our results point particularly to early mitochondrial stress and reduced NAD+ content, as well as the mitokine GDF-15, as potential translatable biomarkers of severe AKI, which may be of interest in humans. Our collection of phenotypic and molecular traits can be explored in an online resource (www.systems-genetics.org/CC_founders_AKI). This resource can help researchers select the ideal animal model to study particular aspects of AKI, but it also showcases the importance of examining the diversity of outcomes present across mice before attempting translation to humans, as findings originating from a single mouse strain often translate poorly to humans [7]. From resistance to kidney injury in the CAST/EiJ strain (in humans, only $12\%$ of surgery patients develop AKI; ref. 60), to reversible AKI (Group II strains) and transition to chronic disease and inflammation (Group III strains, and chiefly the PWK/PhJ strain), the continuum of responses to AKI and development of kidney fibrosis following AKI in this study is a good basis for understanding which mechanisms lead to disease progression and guide therapeutic efforts in the right direction. In addition, by looking at the extreme phenotypes of the resistant CAST/EiJ and sensitive PWK/PhJ strain, our results make the case that mitochondrial and metabolic biomarkers such as NAD+ are critical determinants of recovery after AKI or progression to CKD. ## Choice of mouse models. This study used 7 domesticated (C57BL/6J, DBA/2J, A/J, 129S1/SvlmJ, and WSB/EiJ) or wild-derived (CAST/EiJ and PWK/PhJ) inbred mouse strains drawn from founders of the well-characterized BXD and CC panels, which are well known for their diversity in genetics, as well as in molecular and cardiometabolic phenotypes [61]. The CC founder strains NOD/ShiLtJ and NZO/HlLtJ were excluded because they naturally develop diabetes and other symptoms in the absence of injury [61] (NOD, diabetes and immune defects; NZO, severe obesity and diabetes), and this causes hyperfiltration and could interfere with the conclusions of the study. ## Mouse handling. Mouse strains were imported from Charles River and bred at the EPFL animal facility for more than 2 generations before incorporation into the study. The mice were fed a chow diet (Harlan 2018; $6\%$ kCal of fat, $20\%$ kCal of protein, and $74\%$ kCal of carbohydrates). Mice were housed at 2–4 animals per cage under 12-hour light/dark cycle, with ad libitum access to food and water at all times. BW was measured weekly from 8 weeks of age until killing. We examined 7 mouse inbred strains. At 9 weeks of age, 3 groups of mice were treated with 125 mg/kg of FA in 3M sodium bicarbonate, and 2 groups were treated with bicarbonate alone (vehicle controls). Strains were entered into each group randomly and were then observed daily to monitor their health. The mice were scored weekly according to BW/food intake, coat condition, movement, and signs of pain. Mice scoring above 1 were monitored 3 times a week, and mice scoring above 2 were monitored daily. Mice with a score of 3 were sacrificed. ## Sacrifices and sample collection. Mice were sacrificed at week 1, 2, and 6 (FA treated) or weeks 2 and 6 (controls). The sacrifices took place from 8:30 a.m. until 12 p.m., with isoflurane anesthesia followed by a complete blood draw (~1 mL) from the vena cava and by perfusion with phosphate-buffered saline. Half of the blood was placed into lithium-heparin–coated (LiHep-coated) tubes and the other half in EDTA-coated tubes. Then, both were shaken and stored on ice, followed immediately by collection of the kidneys, liver, heart, spleen, gastrocnemius, and epididymal white adipose tissue. The LiHep blood taken for plasma analysis was also centrifuged at 3,600g at 4°C for 10 minutes at 4°C before being flash-frozen in liquid nitrogen. The left kidney and other organs were flash-frozen in liquid nitrogen and stored at –80°C until processing. The right kidney was split according to the schematic Supplemental Figure 10, and parts were stored in formalin or OCT for histological analysis. ## Urine and plasma biochemistry. Plasma from LiHep-coated tubes was stored at –80°C prior to analysis. Plasma parameters were measured 2 times on diluted samples (1:1 ratio of plasma to diluent) using Dimension Xpand Plus (Siemens Healthcare Diagnostics AG). The biochemical tests were performed according to the manufacturer kit for each parameters: enzymatic creatinine (Siemens Healthcare, DF270B), glucose (Siemens Healthcare, DF40), transaminase ASAT (Siemens Healthcare, DF41A), transaminase ALT (Siemens Healthcare, DF143), and urea nitrogen (Siemens Healthcare, DF21). Plasma levels of TIMP-1, FGF-21, and GDF-15 were measured using the Mouse Premixed Multi-Analyte Kit (LXSAMSM, R&D Systems) in a Luminex 200 system following manufacturer’s instructions. ## Histopathology. With more than 48 hours of formation fixation, the selected middle cross sections of the right kidneys from all mice were then rinsed in $70\%$ ethanol, trimmed, and processed with a conventional paraffin-embedding technique. Paraffin-embedded specimens were then blocked and sliced using rotary microtome at 5 μm thickness; then, sliced sections were stained either with H&E or Picrosirius red (PSR) using internal protocols. All slides were then evaluated by an experienced pathologist in a double-blinded manner. A $20\%$ gradient scoring method based on percentage of affected region of lesions was applied as the semiquantitative analysis assay (Grade 1, < $20\%$; Grade 2, $21\%$–$40\%$; Grade 3, $41\%$–$60\%$; Grade 4, $61\%$–$80\%$; Grade 5 > $81\%$). ## Quantitative kidney tissue section image analysis for collagen content. Automated tissue section–based quantification of PSR histochemical staining (surrogate for collagen) was performed using image analysis algorithms in Visiopharm (version 2020.08.0.8126, Visiopharm). Cortex, medulla, and renal papilla were manually annotated based on morphological criteria from whole transversal kidney sections. Pelvis and adjacent connective tissues were excluded from the analysis. ## RNA extraction. For mRNA, liver tissues were crushed in liquid nitrogen, and 10 mg of tissues were suspended in TRIzol (Invitrogen) and homogenized with stainless steel beads using a TissueLyser II (Qiagen) at 30 Hz for 2 minutes. RNA was extracted and purified using Direct-zol-96 RNA kits (Zymo Research). mRNA concentration was measured for all samples. All samples passed a quality check of purity (NanoDrop) and fragmentation (FragmentAnalyzer). ## RNA-Seq and mapping. RNA libraries were prepared for sequencing using SMARTER mRNA-Seq Library Prep Kit standard protocols. RNA-Seq was performed on a BGISEQ-500. FastQC (default parameters) was used to verify the quality of the mapping. No low-quality reads were present, and no trimming was needed. The STAR aligner [62] was used for mapping the RNA-Seq data to the C57BL/6J reference genome and determining gene counts. We did not use distinct genomes for each strain due to various genome-quality differences between mouse strains that could create bigger artefacts than mapping all strains on the same reference genome, in terms of mapping efficiency and gene count estimation. ## Comparison with human data sets. Human CKD signatures were downloaded from ref. 63, and AKI data sets were downloaded from the GEO database as indicated on Supplemental Figure 2 [64, 65]. For CKD, we used the DEGs provided [63], whereas for AKI data sets, we downloaded the data from raw data from GEO and performed differential gene expression using the limma R package and the voom method [66]. For all data sets, we selected the DEGs with \|fold change\| > 0.5 and FDR < 0.05. Genes were only considered overlapping if they varied in the same direction in both mice and humans. The overlap percentage was calculated as “number overlapped genes”/”maximum possible overlap”, where the maximum possible overlap is the number of human and mouse DEGs — whichever is smallest. For each set of common genes between mice and humans, we performed a gene set overrepresentation analysis using the clusterprofiler R package [67] and represented it as an interactive visualization (Supplemental Figure 8). Five of the most commonly overrepresented gene sets were selected and were represented as pie charts, where the sections of the pie represent the proportion of genes in a given gene set. ## Western blots. Frozen kidney samples, pooled from 4 randomly picked mouse kidneys for each condition, were lysed by mechanical homogenization with RIPA buffer containing inhibitors for protease (catalog 78430, Thermo Fisher Scientific) and for phosphatase (catalog 78428, Thermo Fisher Scientific). The concentration of extracted protein was then determined and normalized using the DC Protein Assay Reagents (catalog 5000116, Bio-Rad). Subsequently, the lysates were analyzed by SDS–PAGE and Western blots using the following antibodies: Total OXPHOS Cocktail (catalog ab110413, Abcam, 1:1,000), P–Eif-2α (catalog 3597, CST, 1:500), Eif-2α (catalog 9722, CST, 1:1,000), Atf5 (catalog ab60126, Abcam, 1:1,000), Atf4 (catalog 11815, CST, 1:1,000), Asns (catalog sc-365809, Santa Cruz Biotechnology Inc., 1:1,000), Hspa9 (catalog ABIN361739, Antibodies Online, 1:1,000), Hspd1 (catalog sc-59567, Santa Cruz Biotechnology Inc., 1:1,000), Lonp1 (catalog HPA002192, MilliporeSigma, 1:1,000), Trib3 (catalog 66702-1, Proteintech, 1:1,000), BIP/Grp78 (catalog ab21685, Abcam, 1:1,000), Gapdh (catalog sc-365062, Santa Cruz Biotechnology Inc., 1:1,000), Tubulin (catalog T5168, MilliporeSigma, 1:2,000), and HRP-labeled anti-rabbit (catalog 7074, CST, 1:5,000) and anti-mouse (catalog 7076, CST, 1:5,000) secondary antibodies. See complete unedited blots in the supplemental material. ## MtDNA/nuclear DNA ratio. The mtDNA/nuclear DNA ratio was measured as described in ref. 68. Briefly, DNA was extracted from crushed kidney by isopropanol precipitation in 0.3M sodium acetate, followed by washing in $70\%$ ethanol. Quantitative PCR (qPCR) was performed on a Lightcycler 480 II (Roche) with the following primers: 16rRNA Fwd (5′–3′): CCGCAAGGGAAAGATGAAAGAC, Rev: TCGTTTGGTTTCGGGGTTTC; ND1 Fwd: CTAGCAGAAACAAACCGGGC, Rev: CCGGCTGCGTATTCTACGTT; HK2 Fwd: GCCAGCCTCTCCTGATTTTAGTGT, Rev: GGGAACACAAAAGACCTCTTCTGG; and UCP2 Fwd: CTACAGATGTGGTAAAGGTCCGC, Rev: GCAATGGTCTTGTAGGCTTCG. The relative mtDNA/nuclear DNA ratio was computed by the ΔΔCt method. ## Kidney NAD+ measurements. NAD+ was extracted using the acidic extraction method and was analyzed by HPLC-MS as described [69]. Briefly, approximately 10 mg of frozen crushed kidney was used for NAD+ extraction in $10\%$ perchloric acid and neutralized in 3M K2CO3 on ice. After final centrifugation (50,000g at 4°C for 5 minutes), the supernatant was filtered and the internal standard (NAD+-C13) was added and loaded onto a column (Kinetex 2.6 µm EVO C18 100 Å, LC Column 150 × 2.1 mm). HPLC was run for 1 minute at a flow rate of 300 mL/min with $100\%$ buffer A (methanol/H2O, 80/$20\%$ v/v). Then, a linear gradient to $100\%$ buffer B (H2O + 5 mM ammonium acetate) was performed (at 1–6 minutes). Buffer B ($100\%$) was maintained for 3 minutes (at 6–9 minutes), and then a linear gradient back to $100\%$ buffer A (at 9–13 minutes) started. Buffer A was then maintained at $100\%$ until the end (at 13–18 minutes). NAD+ eluted as a sharp peak at 3.3 minutes and was quantified on the basis of the peak area ratio between NAD+ and the internal standard and normalized to tissue weight and protein content. ## Interactive data visualization and availability. All metabolic traits, mitochondrial activity, and transcriptome data collected in this study can be explored with an online, interactive interface at www.systems-genetics.org/CC_founders_AKI RNA-Seq data from this study have been submitted to GEO database (GSE222570; https://0-www-ncbi-nlm-nih-gov.brum.beds.ac.uk/geo/query/acc.cgi?acc=GSE222570), and phenotype data from this study is available through the online app. ## Statistics. All the bioinformatics and statistical analyses were performed in R 3.5.2 and Rstudio Pro. All performed statistical tests were 2-sided. When needed, P values were corrected for multiple testing with the Benjamin-Hochberg FDR. Histology scores were compared using Wilcoxon test due to being nonlinear values. mtDNA/nuclear DNA ratios were compared with 1-way ANOVA. Differential gene expression was done using the limma R package and the voom method [66]. Cell type deconvolution was performed using MuSiC [70]. GSEA was done using the GSEA method of the clusterprofiler R package [67]. Genes were ranked according the signed –log10 (Benjamin-Hochberg–adjusted P value) obtained by looking at the treatment effect in female or male mice in each strain separately. We selected the gene sets from the biological process of the gene ontology that had the highest significance levels in the overall response to the diet in each sex. Plots used the ggplot2 [71] and plotly [72] R packages. ## Study approval. Mouse experiments were approved by the Swiss cantonal veterinary authorities of Vaud under license 30759. All human data used are anonymized and publicly available. ## Author contributions JDM, MBS, SM, MB, DR, RAGV, and JA conceived and designed the project. DH performed animal experiments together with technicians and animal facility personnel. TYL and JDM performed laboratory experiments. EB prepared histology samples, JYM and VC performed PSR quantitation, and TC completed histopathology analysis. JDM, MBS, AMB, and NA analyzed the data. MBS and GVA created the web interface. JDM and JA wrote the manuscript with input from all the authors. ## 02/08/2023 Electronic publication ## References 1. Hoste EAJ. Global epidemiology and outcomes of acute kidney injury. Nat Rev Nephrol (2018.0) 14 607-625. DOI: 10.1038/s41581-018-0052-0 2. Bikbov B. Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2020.0) 395 709-733. DOI: 10.1016/S0140-6736(20)30045-3 3. Makris K, Spanou L. Acute kidney injury: definition, pathophysiology and clinical phenotypes. Clin Biochem Rev (2016.0) 37 85-98. PMID: 28303073 4. Obrador GT. Genetic and environmental risk factors for chronic kidney disease. Kidney Int Suppl (2017.0) 7 88-106. DOI: 10.1016/j.kisu.2017.07.004 5. Ortega-Loubon C. Genetic susceptibility to acute kidney injury. J Clin Med (2021.0) 10. DOI: 10.3390/jcm10143039 6. Hudkins KL. BTBR Ob/Ob mutant mice model progressive diabetic nephropathy. J Am Soc Nephrol (2010.0) 21 1533-1542. DOI: 10.1681/ASN.2009121290 7. Nadeau JH, Auwerx J. The virtuous cycle of human genetics and mouse models in drug discovery. Nat Rev Drug Discov (2019.0) 18 255-272. DOI: 10.1038/s41573-018-0009-9 8. Guzzi F. Molecular mechanisms of the acute kidney injury to chronic kidney disease transition: an updated view. Int J Mol Sci (2019.0) 20. DOI: 10.3390/ijms20194941 9. Saul MC. High-diversity mouse populations for complex traits. Trends Genet (2019.0) 35 501-514. DOI: 10.1016/j.tig.2019.04.003 10. Huda MN. Dissecting the genetic architecture of cystatin C in diversity outbred mice. G3 (Bethesda) (2020.0) 10 2529-2541. DOI: 10.1534/g3.120.401275 11. Takemon Y. Uncovering modifier genes of X-linked alport syndrome using a novel multiparent mouse model. J Am Soc Nephrol (2021.0) 32 1961-1973. DOI: 10.1681/ASN.2020060777 12. Rabe M, Schaefer F. Non-transgenic mouse models of kidney disease. Nephron (2016.0) 133 53-61. DOI: 10.1159/000445171 13. Bao Y-W. Kidney disease models: tools to identify mechanisms and potential therapeutic targets. Zool Res (2018.0) 39 72-86. DOI: 10.24272/j.issn.2095-8137.2017.055 14. Fink M. Experimental folic acid nephropathy. Pathology (1987.0) 19 143-149. DOI: 10.3109/00313028709077125 15. Newbury LJ. Inhibition of Kirsten-Ras reduces fibrosis and protects against renal dysfunction in a mouse model of chronic folic acid nephropathy. Sci Rep (2019.0) 9. DOI: 10.1038/s41598-019-50422-7 16. Shi Y. RIPK3 blockade attenuates kidney fibrosis in a folic acid model of renal injury. FASEB J (2020.0) 34 10286-10298. DOI: 10.1096/fj.201902544RR 17. Huguenin ME. An evaluation of the role of tubular obstruction in folic acid-induced acute renal failure in the rat. A micropuncture study. Nephron (1978.0) 22 41-54. DOI: 10.1159/000181422 18. Mulay SR. Novel insights into crystal-induced kidney injury. Kidney Dis (Basel) (2018.0) 4 49-57. DOI: 10.1159/000487671 19. Drickamer LC. Rates of urine excretion by house mouse (Mus domesticus): differences by age, sex, social status, and reproductive condition. J Chem Ecol (1995.0) 21 1481-1493. DOI: 10.1007/BF02035147 20. Abulizi P. Growth differentiation factor-15 deficiency augments inflammatory response and exacerbates septic heart and renal injury induced by lipopolysaccharide. Sci Rep (2017.0) 7. DOI: 10.1038/s41598-017-00902-5 21. Suassuna PG de A. Fibroblast growth factor 21 in chronic kidney disease. J Nephrol (2019.0) 32 365-377. DOI: 10.1007/s40620-018-0550-y 22. Quirós PM. Mitonuclear communication in homeostasis and stress. Nat Rev Mol Cell Biol (2016.0) 17 213-226. DOI: 10.1038/nrm.2016.23 23. Tang J. Role of epidermal growth factor receptor in acute and chronic kidney injury. Kidney Int (2013.0) 83 804-810. DOI: 10.1038/ki.2012.435 24. Bailey SW, Ayling JE. The extremely slow and variable activity of dihydrofolate reductase in human liver and its implications for high folic acid intake. Proc Natl Acad Sci U S A (2009.0) 106 15424-15429. DOI: 10.1073/pnas.0902072106 25. Samodelov SL. Renal reabsorption of folates: pharmacological and toxicological snapshots. Nutrients (2019.0) 11. DOI: 10.3390/nu11102353 26. Nazki FH. Folate: metabolism, genes, polymorphisms and the associated diseases. Gene (2014.0) 533 11-20. DOI: 10.1016/j.gene.2013.09.063 27. Park J. Single-cell transcriptomics of the mouse kidney reveals potential cellular targets of kidney disease. Science (2018.0) 360 758-763. DOI: 10.1126/science.aar2131 28. Mihai S. Inflammation-related mechanisms in chronic kidney disease prediction, progression, and outcome. J Immunol Res (2018.0) 2018. DOI: 10.1155/2018/2180373 29. Furman D. Chronic inflammation in the etiology of disease across the life span. Nat Med (2019.0) 25 1822-1832. DOI: 10.1038/s41591-019-0675-0 30. Quirós PM. Multi-omics analysis identifies ATF4 as a key regulator of the mitochondrial stress response in mammals. J Cell Biol (2017.0) 216 2027-2045. DOI: 10.1083/jcb.201702058 31. West AP. Mitochondrial DNA stress primes the antiviral innate immune response. Nature (2015.0) 520 553-557. DOI: 10.1038/nature14156 32. Patoli D. Inhibition of mitophagy drives macrophage activation and antibacterial defense during sepsis. J Clin Invest (2020.0) 130 5858-5874. DOI: 10.1172/JCI130996 33. Zhang Q. Circulating mitochondrial DAMPs cause inflammatory responses to injury. Nature (2010.0) 464 104-107. DOI: 10.1038/nature08780 34. Pakos-Zebrucka K. The integrated stress response. EMBO Rep (2016.0) 17 1374-1395. DOI: 10.15252/embr.201642195 35. Sorrentino V. Repairing mitochondrial dysfunction in disease. Annu Rev Pharmacol Toxicol (2018.0) 58 353-389. DOI: 10.1146/annurev-pharmtox-010716-104908 36. Ryu D. Urolithin A induces mitophagy and prolongs lifespan in C. elegans and increases muscle function in rodents. Nat Med (2016.0) 22 879-888. DOI: 10.1038/nm.4132 37. Yang Q. LONP-1 and ATFS-1 sustain deleterious heteroplasmy by promoting mtDNA replication in dysfunctional mitochondria. Nat Cell Biol (2022.0) 24 181-193. DOI: 10.1038/s41556-021-00840-5 38. Palikaras K. Coordination of mitophagy and mitochondrial biogenesis during ageing in C. elegans. Nature (2015.0) 521 525-528. DOI: 10.1038/nature14300 39. Zhang Q. The memory of neuronal mitochondrial stress is inherited transgenerationally via elevated mitochondrial DNA levels. Nat Cell Biol (2021.0) 23 870-880. DOI: 10.1038/s41556-021-00724-8 40. Lee CF. Targeting NAD. Sci Rep (2019.0) 9. DOI: 10.1038/s41598-019-39419-4 41. Katsyuba E. De novo NAD. Nature (2018.0) 563 354-359. DOI: 10.1038/s41586-018-0645-6 42. Poyan Mehr A. De novo NAD. Nat Med (2018.0) 24 1351-1359. DOI: 10.1038/s41591-018-0138-z 43. Bignon Y. Cell stress response impairs de novo NAD+ biosynthesis in the kidney. JCI Insight (2021.0) 7. DOI: 10.1172/jci.insight.153019 44. Tran MT. PGC1α drives NAD biosynthesis linking oxidative metabolism to renal protection. Nature (2016.0) 531 528-532. DOI: 10.1038/nature17184 45. Morevati M. Effect of NAD+ boosting on kidney ischemia-reperfusion injury. PLoS One (2021.0) 16. DOI: 10.1371/journal.pone.0252554 46. Xu J. NAD. Front Med (lausanne) (2021.0) 8. DOI: 10.3389/fmed.2021.703076 47. Tan J. HNF1α controls liver lipid metabolism and insulin resistance via negatively regulating the SOCS-3-STAT3 signaling pathway. J Diabetes Res (2019.0) 2019. DOI: 10.1155/2019/5483946 48. Verhave JC. Hepatocyte nuclear factor 1β–associated kidney disease: more than renal cysts and diabetes. J Am Soc Nephrol (2016.0) 27 345-353. DOI: 10.1681/ASN.2015050544 49. Ferrè S, Igarashi P. New insights into the role of HNF-1β in kidney (patho)physiology. Pediatr Nephrol (2019.0) 34 1325-1335. DOI: 10.1007/s00467-018-3990-7 50. Stanzick KJ. Discovery and prioritization of variants and genes for kidney function in >1.2 million individuals. Nat Commun (2021.0) 12. DOI: 10.1038/s41467-021-24491-0 51. Ralto KM, Parikh SM. Mitochondria in acute kidney injury. Semin Nephrol (2016.0) 36 8-16. DOI: 10.1016/j.semnephrol.2016.01.005 52. Stallons LJ. Suppressed mitochondrial biogenesis in folic acid-induced acute kidney injury and early fibrosis. Toxicol Lett (2014.0) 224 326-332. DOI: 10.1016/j.toxlet.2013.11.014 53. Li Y. Targeting mitochondria and metabolism in acute kidney injury. J Clin Med (2021.0) 10. DOI: 10.3390/jcm10173991 54. Nair V. Growth differentiation factor-15 and risk of CKD progression. J Am Soc Nephrol (2017.0) 28 2233-2240. DOI: 10.1681/ASN.2016080919 55. Aparicio-Trejo OE. Chronic impairment of mitochondrial bioenergetics and β-oxidation promotes experimental AKI-to-CKD transition induced by folic acid. Free Radic Biol Med (2020.0) 154 18-32. DOI: 10.1016/j.freeradbiomed.2020.04.016 56. Sinnott-Armstrong N. Genetics of 35 blood and urine biomarkers in the UK Biobank. Nat Genet (2021.0) 53 185-194. DOI: 10.1038/s41588-020-00757-z 57. Sakaue S. A cross-population atlas of genetic associations for 220 human phenotypes. Nat Genet (2021.0) 53 1415-1424. DOI: 10.1038/s41588-021-00931-x 58. Buniello A. The NHGRI-EBI GWAS catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res (2019.0) 47 D1005-D1012. DOI: 10.1093/nar/gky1120 59. Frising UC. Nlrp3 inflammasome activation in macrophages suffices for inducing autoinflammation in mice. EMBO Rep (2022.0) 23. DOI: 10.15252/embr.202154339 60. Lei VJ. Risk stratification for postoperative acute kidney injury in major noncardiac surgery using preoperative and intraoperative data. JAMA Netw Open (2019.0) 2. DOI: 10.1001/jamanetworkopen.2019.16921 61. Kollmus H. A comprehensive and comparative phenotypic analysis of the collaborative founder strains identifies new and known phenotypes. Mamm Genome (2020.0) 31 30-48. DOI: 10.1007/s00335-020-09827-3 62. Dobin A. STAR: ultrafast universal RNA-seq aligner. Bioinformatics (2013.0) 29 15-21. DOI: 10.1093/bioinformatics/bts635 63. Tajti F. A functional landscape of CKD entities from public transcriptomic data. Kidney Int Rep (2020.0) 5 211-224. DOI: 10.1016/j.ekir.2019.11.005 64. Janosevic D. The orchestrated cellular and molecular responses of the kidney to endotoxin define a precise sepsis timeline. Elife (2021.0) 10. DOI: 10.7554/eLife.62270 65. Park M. RNA-Seq identifies condition-specific biological signatures of ischemia-reperfusion injury in the human kidney. BMC Nephrol (2020.0) 21. DOI: 10.1186/s12882-020-02025-y 66. Law CW. voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol (2014.0) 15. DOI: 10.1186/gb-2014-15-2-r29 67. Yu G. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS (2012.0) 16 284-287. DOI: 10.1089/omi.2011.0118 68. Quiros PM. Analysis of mtDNA/nDNA ratio in mice. Curr Protoc Mouse Biol (2017.0) 7 47-54. DOI: 10.1002/cpmo.21 69. Yoshino J, Imai S. Accurate measurement of nicotinamide adenine dinucleotide (NAD. Methods Mol Biol (2013.0) 1077 203-215. DOI: 10.1007/978-1-62703-637-5_14 70. Wang X. Bulk tissue cell type deconvolution with multi-subject single-cell expression reference. Nat Commun (2019.0) 10. DOI: 10.1038/s41467-018-08023-x 71. 71Villanueva RAM, Chen ZJ. ggplot2: Elegant Graphics for Data Analysis (2nd ed.). https://www.tandfonline.com/doi/abs/10.1080/15366367.2019.1565254?journalCode=hmes20 Accessed January 11, 2023 72. Li R, Bilal U. Interactive web-based data visualization with R, plotly, and shiny (Carson Sievert). Biometrics (2021.0) 77 776-777. DOI: 10.1111/biom.13474
--- title: Low nephron endowment increases susceptibility to renal stress and chronic kidney disease authors: - Pamela I. Good - Ling Li - Holly A. Hurst - Ileana Serrano Herrera - Katherine Xu - Meenakshi Rao - David A. Bateman - Qais Al-Awqati - Vivette D. D’Agati - Frank Costantini - Fangming Lin journal: JCI Insight year: 2023 pmcid: PMC9977438 doi: 10.1172/jci.insight.161316 license: CC BY 4.0 --- # Low nephron endowment increases susceptibility to renal stress and chronic kidney disease ## Abstract Preterm birth results in low nephron endowment and increased risk of acute kidney injury (AKI) and chronic kidney disease (CKD). To understand the pathogenesis of AKI and CKD in preterm humans, we generated potentially novel mouse models with a $30\%$–$70\%$ reduction in nephron number by inhibiting or deleting Ret tyrosine kinase in the developing ureteric bud. These mice developed glomerular and tubular hypertrophy, followed by the transition to CKD, recapitulating the renal pathological changes seen in humans born preterm. We injected neonatal mice with gentamicin, a ubiquitous nephrotoxic exposure in preterm infants, and detected more severe proximal tubular injury in mice with low nephron number compared with controls with normal nephron number. Mice with low nephron number had reduced proliferative repair with more rapid development of CKD. Furthermore, mice had more profound inflammation with highly elevated levels of MCP-1 and CXCL10, produced in part by damaged proximal tubules. Our study directly links low nephron endowment with postnatal renal hypertrophy, which in this model is maladaptive and results in CKD. Underdeveloped kidneys are more susceptible to gentamicin-induced AKI, suggesting that AKI in the setting of low nephron number is more severe and further increases the risk of CKD in this vulnerable population. ## Introduction The global rate of preterm birth (birth before 37 weeks gestation) is roughly $11\%$, with 15 million premature births each year [1, 2]. With advances in neonatal care, more preterm infants are surviving to adulthood, albeit with long-term health consequences [3]. Preterm birth is a risk factor for acute kidney injury (AKI), which is associated with increased morbidity and mortality in the neonatal period [4]. AKI increases the risk for developing CKD [5, 6], and survivors of prematurity have increased incidence of chronic kidney disease (CKD) later in life (7–9). Nephrogenesis is not complete until 34–36 weeks gestation, and $60\%$ of nephrons are formed during the third trimester [10]. Human autopsy studies show that, at birth, preterm infants have reduced nephron number. Studies have also suggested that, while there may be a limited period of postnatal nephrogenesis, a significant number of glomeruli appear abnormal, suggesting that ex utero nephrogenesis is perturbed [11, 12]. This results in low functional nephron mass in humans born preterm, particularly those born extremely preterm (birth before 28 weeks gestation). Brenner et al. were the first to hypothesize that low nephron endowment in humans increases an individual’s risk for hypertension and CKD later in life (13–17). Using an adult rat model of $\frac{5}{6}$ nephrectomy, Brenner and others showed that remnant nephrons underwent hyperfiltration to increase the single-nephron glomerular filtration rate (GFR). These altered hemodynamics led to maladaptive changes resulting in eventual glomerular sclerosis, nephron drop out, and renal failure [18, 19]. Less drastic reduction in nephron number, by removing 1 kidney, does not result in progressive renal failure; otherwise, kidney donation for transplantation would not be acceptable. However, studies in adult rodents do not recapitulate the reduced nephron mass seen in preterm humans because, in adults, hyperfiltration occurs abruptly after surgical reduction of nephron mass and not during the critical period of postnatal renal growth and maturation. Several models of congenitally reduced nephron number exist. One model uses restriction in maternal protein and calorie intake during gestation (20–23); however, this leads to generalized epigenetic responses and affects multiple organs in offspring (24–27). Another model resembling premature kidneys involves cesarean delivery 1–2 days prior to natural birth in mice and results in reduced nephron endowment with evidence of CKD manifested by albuminuria, hypertension, and lower GFR 5 weeks later. While this study recapitulates human premature birth, the nephron deficits are mild ($20\%$), and this likely resembles a late preterm human gestation [28]. To understand the pathogenesis of AKI and CKD in the growing population of humans born preterm, we generated potentially novel mouse models of congenitally low nephron number. Since glial cell–derived neurotrophic growth factor (GDNF)/Ret signaling plays a critical role in ureteric bud (UB) branching morphogenesis and nephron induction, we used chemical or genetic approaches to manipulate Ret tyrosine kinase expression or activity during the late stage of kidney development. We generated a cohort of mice with $30\%$–$70\%$ reduction in nephron number. These models recapitulate the likely spectrum of nephron endowment at birth in humans born preterm [10]. We showed that mice with congenitally reduced nephron number undergo postnatal compensatory glomerular and tubular hypertrophy. By 6–12 weeks, they begin to show signs of developing CKD. Neonatal mice with low nephron number are more susceptible to gentamicin-induced AKI, with more severe injury and a unique and exaggerated inflammatory response originating in damaged proximal tubular cells. There is incomplete tubular repair and the accelerated emergence of a CKD phenotype. Overall, our study shows that mice with low nephron number that simulate preterm human kidneys develop compensatory hypertrophy, which becomes maladaptive with resultant CKD even in the absence of prior AKI. Additionally, these kidneys are at high risk of AKI, highlighting the vulnerability of underdeveloped kidneys. ## Inhibition or deletion of ret tyrosine kinase reduces nephron number. Nephrogenesis depends on reciprocal interaction between the UB and metanephric mesenchyme (MM). In the embryonic kidney, Ret is expressed in the UB, and its ligand, GDNF, is secreted by the surrounding MM. This interaction results in UB branching, which induces cells in the MM to condense around UB tips, transition to renal progenitor cells, and ultimately form the nephron (29–33). Genetic manipulations such as noninducible knocking-out of Ret or GDNF in mice results in renal agenesis or severe hypodysplasia [34, 35], limiting our ability to use these tools to generate mice with low nephron number and simulate renal consequences of human preterm birth. A mouse line with a floxed Ret allele that is engineered with a single amino acid substitution (V805A) at the ATP binding site of the Ret protein has no discernable abnormalities at the baseline [36, 37] but increases the receptor’s sensitivity to a small-molecule ATP competitive inhibitor, NA-PP1, by approximately 1,000-fold over WT tyrosine kinases (IC50 of nM versus μM) (38–41). We refer to this strain as Retflox-V805A. We injected pregnant females with vehicle or NA-PP1 at 32.25 mg/kg, 50 mg/kg, or 62.5 mg/kg once a day i.p. starting E16.5 for 3 consecutive days. The small-molecule inhibitor–based approach generated viable offspring. Examination of more than 85 mice showed no evidence of hydronephrosis (Figure 1A), suggesting no obstruction of the collecting system and lower urinary tract. Although kidneys exposed to NA-PP1 were smaller at birth, Masson’s trichrome staining revealed no fibrosis (Figure 1A). While glomerular number (Nglom), quantified using established acid maceration methods [42], was 11,790 ± 998.9 per kidney in vehicle-treated mice, exposing offspring to NA-PP1 in utero resulted in $30\%$–$50\%$ reduction of Nglom with a mean glomerular number of 8,250 ± 2,323; 6,293 ± 1,479; and 6,520 ± 1,884 in pups exposed to 32.25 mg/kg, 50 mg/kg, or 62.5 mg/kg, respectively. All doses of NA-PP1 yielded a significant reduction in Nglom compared with vehicle-exposed controls ($$n = 13$$–33 per group, $P \leq 0.001$; Figure 1B). To confirm that Ret activity was decreased, we performed immunoblot analyses using an anti-Ret antibody that recognizes phospho-Ret (upper band,175 kb) and Ret protein (lower band,155 kb) in E17.5 kidney homogenates from mice exposed to NA-PP1 or vehicle. The results show that kidneys exposed to NA-PP1 had reduced phospho-Ret as well as Ret proteins, which reflects that inhibition of Ret tyrosine kinase activity results in fewer Ret-containing UB tips (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.161316DS1). We focused on analysis of mice born to mothers treated with 50 mg/kg of NA-PP1 (lowest dose that resulted in roughly $50\%$ nephron reduction). Whole-mount staining with calbindin followed by 3-D image reconstruction of kidneys on day of birth (P1) showed reduced UB branching with truncated UB tips (Figure 1C), suggesting that Ret tyrosine kinase inhibition reduced nephron induction. Immunostaining of 2-week-old kidneys exposed to NA-PP1 in utero (50 mg/kg) revealed expression for markers of proximal tubules (Lotus Tetragonolobus Lectin [LTA]), thick ascending limb (Tamm-Horsfall protein [THP]), distal tubules (thiazide-sensitive Na-Cl cotransporter [TSC]), and collecting ducts (Dolichos biflorus agglutinin [DBA]). All proximal tubules contained brush border and stained positive for LTA. While there appeared to be less LTA fluorescence signals in NA-PP1–exposed sections, this is likely due to nephron number deficits. NA-PP1–exposed mice had patent glomerular capillary tufts and peritubular capillaries (Figure 1D). These results indicate that inhibiting Ret tyrosine kinase activity with 50 mg/kg of NA-PP1 beginning E16.5 did not prevent formation of these renal structural components. Kidney/body weight ratio was significantly lower in mice exposed to NA-PP1 compared with vehicle-exposed mice (Figure 1E). Because Ret signaling is required for neural crest cell migration to the gut mesenchyme (from E9 to E14) [43, 44] and may play a role in the enteric neuron survival [36], we examined distal colons for the number and organization of neurons. Treatment on or after E16.5 resulted in no significant differences in the number of neurons in the distal colon. The neuronal organization was similar to vehicle-exposed controls (Supplemental Figure 1). NA-PP1–exposed mice had adequate postnatal weight gain, suggesting sufficient gastrointestinal function. To circumvent potential effects of systemic Ret tyrosine kinase inhibition, we deleted Ret specifically in the developing kidney by taking advantage of the UB-specific expression of Hoxb7-rtTA combined with the tet-O-Cre system [35, 45, 46]. Since the engineered Ret allele (Retflox-V805A) is floxed, treating pregnant Hoxb7-rtTA;tet-O-Cre;Retflox-V805A mice with doxycycline (Dox) is expected to delete Ret in the UB of offspring carrying the same transgenes (named RetUB del mice), thus reducing UB branching and decreasing nephron induction. Our breeding strategy yielded littermate controls with the genotype of tet-O-Cre;Retflox-V805A lacking rtTA and unable to activate tet-O-Cre. Previous work shows that HoxB7-rtTA mice expressed high levels of rtTA and that induction of Cre activity is achieved after 1-day treatment with Dox in the drinking water [45]. We treated pregnant females with Dox beginning at E15.5, E16.5, or E17.5 through delivery. Analysis of adult offspring kidneys showed significant reduction of Nglom to 3,977 ± 1,377; 5,440 ± 1,333; or 8,888 ± 1,657 with Dox treatment beginning E15.5, E16.5, or E17.5, respectively, compared with 13,800 ± 1,069 in littermate controls ($$n = 10$$–47/group, $P \leq 0.05$ for all comparisons) (Figure 2A). No hydronephrosis was observed with Ret deletion on E16.5 (Figure 2B). Kidneys contained patent glomerular capillary tufts and peritubular capillaries and expressed markers for proximal tubules, thick ascending limbs, distal tubules, and collecting ducts (Supplemental Figure 2). Furthermore, there were no morphological changes suggestive of renal dysplasia, supporting the utility of these mice as a model of congenitally low nephron endowment. We chose mice exposed to Dox starting E16.5 for detailed analysis. First, we detected no differences in the mean glomerular number between control females (average Nglom = 13,594) and control males (average Nglom = 14,212, $$P \leq 0.35$$, $$n = 6$$–12). There were no sex-related differences in the mean glomerular number in RetUB del mice, with the average Nglom of 5,522 in females and average Nglom of 5,338 in males ($$P \leq 0.65$$, $$n = 21$$–26). At P1, whole-mount kidneys from mice exposed to Dox staring at E16.5 showed truncated UB branching with qualitatively decreased UB tips and reduced Six2-expressing cap mesenchyme (Figure 2C), suggesting that reduced GDNF/Ret signaling in the developing kidney attenuated UB branching and nephron induction. The kidneys had fewer glomeruli (Figure 2D) and were smaller, with significantly reduced kidney/body weight ratios compared with controls (Figure 2E). No interstitial fibrosis was identified at early time points (up to P10; data not shown). This cohort of RetUB del mice provided a range of $35\%$–$70\%$ nephron reduction, simulating underdeveloped kidneys in humans born preterm [47]. ## Mice with low nephron number develop glomerular and tubular hypertrophy. Given the increased workload of individual nephrons in mice with congenitally low nephron number, we next assessed for postnatal renal adaptation to low nephron endowment. We chose kidneys exposed to Dox starting E16.5 ($40\%$–$60\%$ reduction of Nglom) for quantification of glomerular size and tubular diameter over the course of renal growth and maturation. While no significant differences were detected between RetUB del mice and controls at 2 weeks of age (5 mean difference of 9.4 µm2; $95\%$ CI, –150.2 to 303.0; $$P \leq 0.61$$), RetUB del mice developed glomerular enlargement over the course of next 4 weeks (Figure 3A). By 6 weeks of age, the mean glomerular surface area in RetUB del mice was 1,332 μm2 ($95\%$ CI, 6,72.9–2,140.7 μm2) or $61.1\%$ ($95\%$ CI, 30.9–98.3 µm2) larger than that of controls ($$P \leq 0.002$$). While glomerular surface area increased between 2 and 6 weeks in both groups, it increased at an accelerated rate in mice with low nephron number, and the difference in growth trajectories was significant (excess growth of 1,272.49 μm2; $95\%$ CI, 1,000.1–1,606.9 µm2; $$P \leq 0.0011$$). Tubular hypertrophy also occurred between 2 and 6 weeks of age (Figure 3B). Since studies in hypertrophic kidneys of experimental animals indicate that the most prominent tubular hypertrophy occurs in the proximal tubules [48, 49], we measured tubular diameter in the S3 segment (pars recta) of proximal tubules, localized to the outer stripe of the outer medulla. At 2 weeks, there were no differences between RetUB del and controls (1.37 μm; $95\%$ CI, –1.15 to 3.88 µm; $$P \leq 0.317$$) (Figure 3B); however, by 6 weeks, the mean diameter in RetUB del mice was 7.81 μm ($95\%$ CI, 3.61–12.01 µm) larger than that of controls ($$P \leq 0.007$$). While proximal tubular diameter increased in both groups between 2 and 6 weeks of age, the increase was greater in the RetUB del group (excess growth of 6.44 μm; $95\%$ CI, 1.85–11.04 µm; $$P \leq 0.02$$). Interestingly, glomerular and tubular size were highly correlated (Figure 3C) with a correlation coefficient of $R = 0.917$ ($95\%$ CI, 0.680–0.981). The degree of glomerular hypertrophy is paralleled in the proximal tubules; more hypertrophy in both nephron components was expected in mice with the lowest nephron number within the RetUB del group, suggesting that glomerular hyperfiltration and tubular hypertrophy are interrelated. Proximal tubular brush border appeared exuberant in RetUB del mice as early as 2 weeks of age. At 6 weeks, immunostaining of NK-ATPase showed more elaborate basolateral membrane in-folding, suggesting a possible increase in epithelial transport activity. Immunostaining of Lamp1, a membrane protein located on late endosomes and lysosomes, revealed that RetUB del mice have a more robust endolysosomal system (Figure 3D), with increased integrated density of Lamp1 fluorescence signals (mean 224 × 103 in controls versus mean 305 × 103 in RetUB del mice; $$P \leq 0.03$$) per number of nuclei in the cortical region where S1 and S2 segments of the proximal tubule are located. These adaptive changes likely reflect increased glomerular and tubular function at the single-nephron level. It is interesting to note that glomerular obsolescence with collapsed capillary loops were observed in the subcapsular region, where the newest generation of glomeruli are located (Supplemental Figure 3). This glomerular phenotype is reminiscent of human kidney autopsies showing that glomeruli in premature infants are structurally abnormal [11, 12]. In contrast, the glomeruli located in the deeper cortex and corticomedullary junction were larger but morphologically similar to controls (data not shown). Next, we tested whether $50\%$ nephron reduction in adult mice resulted in similar glomerular and tubular adaptation. We performed right nephrectomy or sham operation in 8-week-old Retflox-V805A mice (no NA-PP1 or Dox exposure) and analyzed kidneys 4 weeks later. We did not detect differences in the size of glomeruli. Although the tubular diameter was slightly larger in nephrectomized mice compared with sham-operated mice, the degree of hypertrophy was much less than that of mice with congenitally reduced nephron number ($10\%$ larger after nephrectomy versus $26\%$ larger in mice with congenitally low nephron number). In addition, there was no evidence of interstitial fibrosis, inflammation, or tubular atrophy (Supplemental Figure 4). Overall, our results suggest that our newly generated mouse models of low nephron endowment simulate postnatal renal compensatory hypertrophy in humans born with low nephron number, with hypertrophy and excess growth occurring between 2 and 6 weeks of age. Importantly, kidneys with congenital reduction of nephron number by $40\%$–$60\%$ had unique adaptive changes during the period of postnatal growth and maturation, whereas a similar nephron reduction in adults does not result in such adaptation. ## Adult RetUB del mice develop CKD. Low nephron endowment predisposes humans to renal disease later in life [13, 16]. In a rodent model of $\frac{5}{6}$ nephrectomy ($83\%$ nephron reduction), hypertrophy and hyperfiltration at the single-nephron level alter renal hemodynamics, leading to maladaptive changes such as glomerular sclerosis, nephron dropout, and renal failure [18, 19]. However, this occurs after the critical window of postnatal growth and maturation, and it may cause necrosis and inflammation at the incision margins. To test whether compensatory nephron hypertrophy due to congenital nephron deficits becomes maladaptive and contributes to the development of CKD, we performed serial examinations of RetUB del mice that were exposed to Dox (starting at E16.5) for up to 9 months. Fibrotic foci started to appear — especially in areas of glomerular obsolescence — at 6 weeks (Figure 4A), and there was significantly higher collagen I expression compared with age-matched controls (Figure 4, A and B). In areas of increased collagen I deposition, CD45+ inflammatory cells also accumulated as shown in Figure 4A. Tubular atrophy and interstitial inflammation emerged by 12 weeks (Supplemental Figure 3). Serum creatinine (sCr) was higher at 12 weeks (control 0.1 mg/dL versus RetUB del 0.16 mg/dL, $$P \leq 0.02$$, $$n = 11$$), and urinary albumin excretion increased (mean urine albumin/urine creatinine [Ualb:cr] 15.5 versus 5.98 mg/g in controls, $$P \leq 0.02$$, $$n = 15$$) (Figure 4C). By 9 months, glomerular changes secondary to hyperfiltration, which included perihilar hyalinosis and segmental luminal obliteration by endocapillary foam cells, appeared (Figure 4D). These changes have been shown to be early signs of developing focal segmental glomerular sclerosis (FSGS) [50]. In summary, we found that in mice with $30\%$–$70\%$ congenitally reduced nephron number, compensatory tubular and glomerular hypertrophy occurred between 2 and 6 weeks of age, and by 6–12 weeks of age, mice exhibit pathologic changes characteristic of CKD. This differs from adult mice with $50\%$ reduced nephron number, as nephrectomized adults have no inflammation or morphologic changes resembling CKD. In mice with congenitally low nephron number, increased workload at the single-nephron level and compensatory hypertrophy appear to activate cellular stress and inflammatory pathways. The end point of this process is a steady decline in renal structure and function, even in the absence of additional stressors and injuries. ## RetUB del mice have more severe injury and accelerated development of CKD after neonatal AKI. Next, we tested whether adverse environmental exposures accelerated the development of CKD in mice with congenitally reduced nephron number. Premature infants are at increased risk of AKI in the neonatal period, with an incidence of $18\%$–$48\%$ in infants born at fewer than 36 weeks, and with higher rates of AKI associated with earlier gestational age [4]. AKI is often multifactorial, and risk factors include exposure to nephrotoxins. Gentamicin is one of the most ubiquitous exposures, as preterm infants are at high risk of infection with bacteria susceptible to aminoglycosides [51, 52]. Gentamicin enters proximal tubular cells by binding to the brush border [53, 54], followed by receptor-mediated endocytosis [55, 56]. Gentamicin accumulates in lysosomes, causing swelling and rupture [57], releasing gentamicin and lysosomal enzymes into the cytosol, and causing further cell injury. However, it is not clear whether kidneys with low nephron endowment are more susceptible to gentamicin-induced injury. We injected neonatal RetUB del and littermate controls with gentamicin (100 mg/kg) or saline (4 μL/g) s.c. once a day for 7 days from P3 to P9. Kidneys were harvested 1 day after completing 7 days of treatment (P10). Of note, gentamicin total dose was lower in mice with low nephron number, given the weight-based dosing, as mice with low nephron number are slightly smaller than their littermate controls (Supplemental Figure 5). Gentamicin injection led to proximal tubular injury in both control and RetUB del mice. PAS-stained sections revealed proximal tubular vacuolization, which was more severe in RetUB del mice compared with littermate controls (Figure 5A). Immunostaining of Lamp1 showed that, under basal conditions, there is a similar pattern of Lamp1 expression in late endosomes and lysosomes under the brush border of proximal tubules in saline-injected controls and RetUB del mice, albeit with more endosomes and lysosomes in mice with low nephron number, as described in Figure 3D. However, after gentamicin exposure, RetUB del mice have more severely enlarged membrane vesicles in the proximal tubules, suggesting more endosome and lysosome swelling and engorgement after gentamicin exposure in mice with low nephron number (Figure 5A). To determine whether there was a difference in lysosome size between control and RetUB del mice after gentamicin exposure, we performed blinded human observer–based comparisons of Lamp1-expressing vesicle size. Given that lysosomal size varies among S1–S3 segments of the proximal tubules [58], we focused on the renal cortex where S1 and S2 segments were localized. We obtained 50 randomized images in each experimental group and performed 100 comparisons between groups. Our observer identified gentamicin-exposed RetUB del kidneys as having larger lysosomes than gentamicin-exposed controls in $85\%$ of comparisons, which corresponded to a χ2 value of 24.5 and $P \leq 0.001.$ Further examination with electron microscopy revealed that, after gentamicin exposure, RetUB del and littermate controls had proximal tubular injury with vacuolated cytoplasm filled with endosomes, phagosomes, and lysosomes (Figure 5B). Early cell death was apparent, and areas of interstitial edema and inflammatory cell infiltrates were identified. Mitochondria were morphologically similar between normal and low nephron number groups (data not shown); however, there was mitochondrial rarefication in both groups, which could be secondary to damage and rupture or could be due to marginalization in the setting of profound lysosomal swelling and engorgement. While glomeruli were mostly intact, there were short segments of foot process effacement (Supplemental Figure 6). The more severe vacuolization corresponded to higher expression of kidney injury molecule 1 (Kim1) and increased inflammation with CD45+ cell infiltration in RetUB del mice (Figure 5C). While quantification of Kim1 and CD45 infiltrates showed no difference between saline-injected control and RetUB del mice, gentamicin injection resulted in a significant increase in Kim1 expression in RetUB del mice compared with controls, with a mean Kim1 integrated density of 511 × 103 in RetUB del compared with 76.25 × 103 in controls ($$P \leq 0.018$$, 1-way ANOVA with Tukey’s test for multiple comparisons). Although CD45+ cells were increased in both groups, RetUB del kidneys had significantly more infiltrates than control kidneys, with a mean CD45 integrated density of 1,039 × 103 in RetUB del mice compared with 264 × 103 in controls ($$P \leq 0.019$$, 1-way ANOVA with Tukey’s test for multiple comparisons) (Figure 5D). These results indicate that low nephron endowment increases susceptibility to gentamicin-induced injury in the proximal tubules, likely due to increased proximal tubular endocytosis and uptake of gentamicin, given the robust-appearance of endosomes and lysosomes in RetUB del mice (Figure 3D and Figure 5A). Studies on renal ischemic injury indicate that focal epithelial cell death and injury triggers tubular repair by proliferation of surviving cells [59, 60]. We quantified proliferating cells by the expression of phospho-histone 3 (pH3) in the cortex and the outer strip of the outer medulla (S1–S3 segments of proximal tubule). While quantification of the number of pH3+ proliferating cells showed no difference in saline-injected control and RetUB del kidneys, gentamicin injection led to a significant increase in pH3+ cells in control kidneys, but not in RetUB del kidneys (Figure 6, A and B). This may be due to more severe epithelial injury resulting in fewer competent cells able to enter a proliferative state in the RetUB del kidneys. The low proliferative repair is associated with a more accelerated CKD phenotype, higher injury scores, and persistent inflammation 4 weeks after gentamicin injection (Figure 6, C and D). Our results indicate more severe long-term renal adverse effects following neonatal AKI in the setting of low nephron endowment. Given the persistence of inflammatory infiltrates in RetUB del mice, we reasoned that there may be an initial exaggerated inflammatory response to gentamicin in these mice. ## RetUB del mice have a unique inflammatory response to gentamicin-induced AKI. The importance of intact proximal tubular structure and function on the integrity of interstitial compartment has been well demonstrated [61, 62]. Tubular epithelial damage and the resulting molecular response can trigger interstitial inflammation. Cytokines are major players in the complex interplay between damaged tubules and inflammatory and immune cells [63]. To identify inflammatory mediators in the injured kidney, we used a commercially available Proteome Profiler Mouse Cytokine Array Kit (R&D, ARY006) for simultaneous detection of 40 cytokines, chemokines, and acute phase reactants in mouse kidney homogenates 1 day following the completion of gentamicin or saline injection. Under basal conditions (saline-exposed, age-matched mice), there were no differences in cytokine expression between RetUB del mice and controls, suggesting that there is no inflammatory response associated with low nephron number in mice at P10. In contrast, we detected the expression of 11 cytokines in the kidneys of control and RetUB del mice following gentamicin injection. Among them, Timp-1, MCP-1, CXCL10, and IL-1ra were significantly higher in gentamicin-exposed RetUB del kidneys compared with gentamicin-exposed littermates with normal nephron number (Figure 7). Given that Timp-1, MCP-1, and CXCL10 have all been implicated in renal inflammatory diseases (64–79), we focused on these cytokines for further analysis. We found that, while both control and RetUB del mice had significantly increased expression of Timp-1 and MCP-1 after gentamicin, the increase was much greater in RetUB del mice. Interestingly, CXCL10 was uniquely elevated in RetUB del mice after gentamicin (Figure 7C). Quantitative PCR (qPCR) analysis from kidney homogenates showed increased expression of mRNA for Timp-1, MCP-1, and CXCL10 in gentamicin-exposed RetUB del mice, confirming their expression in cells of renal origin and/or intrarenal infiltration (Figure 8A). To determine the location of cytokine producing cells, we performed RNAscope using probes to Timp-1, MCP-1, and CXCL10, and we colabeled proximal tubules using antibody to CD13 (ACD, RNA Protein Codetection Assay). We discovered that, while all 3 cytokines were produced in interstitial cells (data not shown), MCP-1 and CXCL10 were also produced by the damaged proximal tubular epithelial cells (Figure 8B). Although there was detectable MCP-1 and CXCL10 mRNA signal in tubules of both control and RetUB del mice, there was more signal detected in mice with low nephron number, which is consistent with the higher protein and mRNA levels shown by the cytokine array and qPCR analysis. In summary, mice with congenitally reduced nephron number develop glomerular and proximal tubular hypertrophy by 6 weeks of age, which — while initially adaptive — becomes maladaptive with the emergence of a CKD phenotype by 6–12 weeks of age. Increased cellular stress secondary to nephron hypertrophy may accelerate the development of CKD in mice with low nephron endowment. As neonates, RetUB del mice are more susceptible to gentamicin-induced AKI, with more severe injury, a profound and unique inflammatory response, and incomplete repair with rapid progression to CKD. Our studies highlight the vulnerability of the kidney with low nephron endowment, and mice generated in this study are useful for the study of pathogenesis of kidney disease in humans born preterm. ## Discussion We generated mouse models of $30\%$–$70\%$ nephron reduction that simulate human premature kidneys by manipulating Ret tyrosine kinase activity or expression during late gestation. While early delivery would be an ideal experimental design to study prematurity related human kidney disease, delivering mice more than 2 days before natural birth does not yield viable animals [80]. Mice delivered shortly prior to natural delivery results in mice with a $20\%$ nephron deficit [28]; however, this does not recapitulate the spectrum of nephron number seen in humans born preterm — particularly, the significant nephron reduction seen in extremely preterm infants [10]. Although Ret signaling is not known to play a role in renal growth and maturation beyond the period of nephrogenesis, interfering with Ret signaling could result in subtle phenotypic changes not recognized in this study. While deleting Ret in our genetic model is permanent, in our chemical model, Ret tyrosine kinase activity is only transiently inhibited. Therefore, if Ret were necessary for further renal growth and development, kidneys in the chemical model would not be affected. In addition, UB tips that have Ret deletion are often replaced by overgrown WT cells [81], so in the case of Ret deletion, any potential role of Ret in renal growth and maturation after nephrogenesis is unlikely to be affected by deletion of Ret in a fraction of fetal UB cells. Future analysis including detailed cell type and gene expression profiling may reveal subtle phenotypic changes. It is also important to note that, while humans with hypomorphic Ret mutations may have low nephron endowment, the mutations can cause other developmental defects such as Hirschsprung’s disease, renal dysplasia, and urinary tract anomalies [82, 83] because Ret signaling is affected from the beginning of organogenesis. In contrast, perturbation of Ret signaling late in gestation causes no significant disturbance of the enteric nervous system, and there is no evidence of renal dysplasia or hydronephrosis, likely because treatment begins after the critical period of Ret-dependent development of urinary tract and enteric nervous system. Of the 2 models generated, RetUB del mice generated by UB-specific deletion of Ret are more consistent in the range of nephron reduction. Interestingly, we found that the loss of Six2+ progenitor niches was patchy, suggesting that Ret deletion following Dox administration may not be uniform. It is possible that cells in which Ret is successfully deleted do not form branching tips [81], leaving areas with no surrounding Six2+ cells; areas in which Ret is not deleted continue to branch, and Six2+ niches form, resulting in a patchy appearance of Six2+ progenitor niches. Our study directly links low nephron number with compensatory glomerular and proximal tubular hypertrophy. While several human studies have shown an inverse correlation between nephron number and glomerular size [84, 85], no studies have demonstrated that this is an acquired process, nor have any studies outlined a timeline for the development of compensatory hypertrophy. We show that mice with low nephron number have normal glomerular size at 2 weeks of age but that they embark on an accelerated growth trajectory between 2 and 6 weeks of age. Human autopsy studies corroborate the clinical relevance of this model. In studies of humans born preterm who developed chronic kidney disease later in life, glomeruli appeared enlarged [86]. In addition, we have shown that hypertrophy extends beyond glomeruli to tubules. We found that proximal tubules are hypertrophied with a direct correlation between glomerular surface area and mean tubular diameter. This is in agreement with the early report by Oliver et al. showing a close relationship between the size of the glomerulus and the proximal tubules in kidneys where small glomeruli are connected to small tubules, normal glomeruli to normal tubules, and larger glomeruli to larger tubules [48], suggesting that nephron hypertrophy is linked to absorptive workload that is in parallel with glomerular filtration. Compensatory hypertrophy in kidneys with low nephron number is thought to be due to increased single-nephron GFR. Brenner et al. [ 13] first hypothesized that, in the setting of increased single-nephron GFR, there is an adaptive compensatory hypertrophy that becomes maladaptive. In this study, we show that maladaptive changes emerge by adolescence in mice with congenitally reduced nephron number. By 12 weeks, mice develop evidence of decreased renal function, as shown by increased sCr and urine albumin excretion. By 6–12 weeks, they have tubular atrophy, interstitial fibrosis, and inflammatory cell infiltrates, suggestive of the development of CKD. By 9 months of age, there are glomerular changes of evolving FSGS. These features are reminiscent of clinical findings in young adults with a history of preterm birth and low birthweight who were found to have proteinuria and elevated creatinine and who underwent renal biopsies. These individuals were found to have oligomeganephronia, glomerulosclerosis, mild tubular atrophy, and patchy fibrosis [86]. In contrast to congenitally low nephron number, uninephrectomy in adult mice did not result in the same degree of renal hypertrophy, and there were no changes to suggest CKD during the 4-week study period. This is in agreement with other reports showing less compensatory growth when nephrectomy occurred in adult rodents [87]. One previous study demonstrated that neonatal nephron loss in rats (removal of 1 kidney day 1 after birth, when nephrogenesis is ongoing) resulted in compensatory renal growth 4 weeks later with larger glomerular perimeters and more cells per glomerular cross section [88]. Conventional wisdom suggests that the progressive increase in metabolic demand during somatic growth drives this postnatal growth. Indeed, human autopsy studies in White American males showed that marked glomerular hypertrophy in kidneys with low nephron number was closely associated with high body surface area [89], supporting metabolic demand as a driving force for renal compensatory growth. However, in neonatal nephrectomized rats [88] and in our model of congenitally low nephron number, early adaptive growth was associated with the development of renal pathologic changes of CKD. At this time, the cellular and molecular mechanisms of glomerular and tubular hypertrophy, as well as the cellular basis for the transition from compensatory hypertrophy to maladaptive changes resembling CKD, are not fully understood. Future studies with single-nucleus RNA-Seq in conjunction with spatial transcriptomics may reveal transcriptional programs regulating renal hypertrophy and the development of CKD. We used RetUB del mice to develop a model of preterm neonatal AKI using gentamicin as a nephrotoxic exposure in the first few days of life. Mice with congenitally decreased nephron number experience more severe AKI after gentamicin and have a higher level of the proinflammatory cytokines/chemokines MCP-1, CXCL10, and Timp-1. While MCP-1 and Timp-1 were elevated in controls exposed to gentamicin, CXCL10 was only elevated in mice with low nephron number exposed to gentamicin. In addition, MCP-1 and Timp-1 levels were much higher in RetUB del mice treated with gentamicin compared with controls under the same treatment conditions. To determine the origin of MCP-1, CXCL10, and Timp-1 after injury, RNAscope was performed in conjunction with immunostaining with antibody to CD13 to identify proximal tubular cells. We found that, while all 3 cytokines were produced in interstitial cells, MCP-1 and CXCL10 were also produced in the damaged proximal tubules. MCP-1 and CXCL10 were detected in both control and RetUB del mice exposed to gentamicin; however, there were higher mRNA levels in mice with low nephron number as confirmed by qPCR analysis. MCP-1 is a monocyte chemotactic factor that binds CCR2 on monocytes and promotes monocyte mobilization from bone marrow, recruitment to local tissues, and differentiation into macrophages [73, 90]. MCP-1 has been implicated in inflammation and kidney diseases [73, 75, 79, 91]. CXCL10 is also increased after injury in RetUB del kidneys. CXCL10 is a chemokine induced by IFN-γ. It binds the chemokine receptor CXCR3 on CD4+ and CD8+ lymphocytes, recruits T cells to inflammatory sites, and promotes effector activity. It also is a chemoattractant for macrophages, monocytes, and NK cells [92, 93]. CXCL10 has been implicated in renal transplant rejection, and increased urinary CXCL10 is associated with tubulointerstitial inflammation and transplant rejection [64, 77, 94, 95]. Timp-1 is a metalloproteinase inhibitor that also has many functions, including cell proliferation and growth, apoptosis, angiogenesis, and inhibition of smooth muscle cell migration [96]. Timp-1 has also been explored as a biomarker of human AKI [68]. Elevated levels of MCP-1, CXCL10, and Timp-1 in RetUB del mice after AKI suggests a more severe inflammatory response in mice with low nephron number, yet the functional impact of this dysregulated cytokine response remains untested in this study. Understanding cytokine function in this context will be an important next step to validate the role of inflammation after AKI in mice with low nephron number. The presence of such cytokines in human studies and the fact that they are well-known chemoattractants raises the possibility of therapies aimed at blocking the inflammatory loop in human preterm AKI. The reason for increased inflammation and more severe injury with poor repair after gentamicin in mice with low nephron number is not fully understood. Our basic understanding of tubular repair is largely obtained from animal models of acute ischemia reperfusion injury (IRI) when kidneys are deprived of blood flow, typically for 15–45 minutes, and then reperfused. This type of insult causes the most severe injury to the S3 segments localized to the outer stripe of outer medulla, although other segments of the nephron and collecting ducts sustain injury as well. Cell injury, death, and detachment from the tubular basement membrane typically occur in a focal and mosaic fashion even within the same tubular segment. Genetic labeling and lineage-tracing studies indicate that a subset of surviving tubular cells responded to injury with transcriptional activation and reenter the proliferative state to cover the denuded area and redifferentiate to establish intercellular junctions with neighboring cells [59, 60, 97, 98]. In contrast, the source of cells for tubular repair after gentamicin-induced injury is less studied. More than 50 years ago, electron microscopic examinations in rats injected with gentamicin (40 mg/kg, daily for 14 days) suggested that regenerating cells appeared to originate from residual epithelial cells [99]. Our current study indicates that gentamicin causes more generalized and profound tubular injury along the entire length of proximal tubules in mice with low nephron endowment. It is conceivable that there are fewer cells with the capacity to activate reparative program and reenter the cell cycle for tubular repair. The reduced repair, prolonged injury, and more severe interstitial inflammation can all account for the accelerated CKD phenotype in underdeveloped kidneys injured by gentamicin. It is possible that reduced renal mass results in higher exposure to gentamicin per cell, causing increased toxicity in mice with low nephron number. We attempted to achieve a similar volume of distribution of gentamicin in mice with normal and low nephron number by giving the drug dose based on body weight. Whether toxicity is due to innate characteristics or responses to injury in kidneys with low nephron mass needs to be further explored. In this study, we chose the dose of gentamicin by extrapolating from studies of adult rodents with gentamicin-induced AKI [100, 101] and this dose is higher than the dose given to preterm infants per kilogram of body weight. Therefore, caution must be taken when interpreting these results. In the clinical setting, while gentamicin doses are lower, infants also experience concurrent renal insults such as infections, hemodynamic instability, and/or other nephrotoxic exposures, which can all contribute to renal injury or worsen the nephrotoxic effects of a lower dose of gentamicin. Further understanding of cellular and molecular mechanisms of injury and cell death — and, more importantly, tubular stress resistance and repair — following gentamicin use is essential in order to develop measures to reduce its renal toxicity in premature infants. This study highlights the vulnerability of the preterm kidney. In our model of low nephron endowment, the risk of maladaptive changes and the development of CKD — in the absence of episodes of acute injury — is high. In addition, AKI from a commonly prescribed medication, gentamicin, further exacerbates renal injury with incomplete repair and accelerates the development of CKD. Given that up to $50\%$ of extremely preterm infants will experience AKI in the neonatal period and a vast majority will be treated with gentamicin, this population is at extremely high risk for future renal dysfunction. A recent publication by the AKI!NOW Steering Committee [102] highlights the importance of clinical care of patients who have recovered from critical illness and AKI. It emphasizes the need for ongoing research to improve outcomes in vulnerable populations, including infants discharged from the neonatal intensive care unit, as these children will grow into adulthood facing increased risk of CKD. ## Methods Supplemental Methods are available online with this article. ## Animal studies. Retflox-V805A (The Jackson Laboratory, 028548) female and male mice were bred for timed pregnancy. To inhibit Ret tyrosine kinase activity, pregnant female mice were injected i.p. with NA-PP1 (Medchem express, HY-13941/CS-1804) or vehicle (cremaphor/saline/ethanol in 1:7:2 ratio) from E16.5 through E18.5 at the dose of 32.25 mg/kg, 50 mg/kg, and 62.5 mg/kg, respectively. Pups delivered spontaneously E19.5. NA-PP1 was prepared using published methods [37]. Both male and female offspring were used for experiments. To delete Ret in the UB, Tet-O-Cre [103] and Hoxb7rtTA (The Jackson Laboratory, 036718) mice were crossed into Retflox-V805A mice. Tet-O-Cre;Hoxb7rtTA;Retflox-V805A mice were bred for timed pregnancy. Pregnant females were given Dox (Henry Schein, 1315046; 2 mg/mL) dissolved in drinking water beginning E15.5, E16.5, or E17.5 through delivery. Several breeders were heterozygous for Hoxb7rtTA and yielded littermate controls without Ret deletion (Tet-O-Cre; Retflox-V805A). Offspring were genotyped through Transnetyx core service. Pups with the genotype of Tet-O-Cre;Hoxb7rtTA;Ret Retflox-V805A who were exposed to Dox in utero with resultant Ret deletion in the UB were named RetUB del. ## Neonatal AKI with gentamicin. AKI was induced by injecting gentamicin (100 mg/kg, Henry Schein, 54894) or saline (4 μL/g) s.c. P3-P9. Kidneys were harvested 1 day or 1 month after completing injections. ## Quantification of glomerular number. Entire kidneys were harvested, decapsulated, cut into 2 mm3 pieces, and incubated in 5 mL of 6N Hydrochloric acid at 37°C for 35 minutes with gentle shaking, followed by pipetting to dissociate glomeruli from the surrounding tissue. Digestion was terminated with 25 mL of sterile water and a number of glomeruli in 1 mL of digested kidney was counted in duplicate (or triplicate, if duplicates differed by > $10\%$) [42]. ## Quantification of glomerular and tubular size. Kidneys were harvested at 2, 6, or 12 weeks. Formalin-fixed, paraffin-embedded kidneys were sectioned at 4 μm thickness. PAS-stained sections were scanned at 200× magnification using Aperio ImageScope from Leica Biosystems, and image analysis was performed using QuPath software (v0.2.2). Glomerular surface area was measured manually by a blinded investigator by outlining the entire glomerular basement membrane area of the glomerular globe, excluding Bowman’s capsule, and using QuPath software to measure surface area in the measured plane. All glomeruli in the scanned kidney section were measured (range 50–200 glomeruli per kidney) to reduce sampling bias. The narrowest proximal tubular profiles in the outer strip of the outer medulla were selected, and diameter was manually measured. In total, 50 tubules in half kidney sections and 100 tubules in entire kidney sections were measured. All measurements were performed by an investigator blinded to mouse genotypes and treatment. Tubular size and glomerular surface area were analyzed using mixed-effects regression models accounting for within-subject correlations of tubular diameter or log glomerular surface area. All mean values reported for glomerular surface are geometric means. ## Electron microscopy examination. Transmission EM studies were performed using the standard method, and images were acquired with a JEOL JEM-1011 electron microscopy equipped with a Gatan digital camera. ## RNAscope. Kidneys were fixed in formalin for 24 hours followed by paraffin embedding and sectioning at 8 μm thickness. RNAscope was performed according to manufacturer’s instructions (ACD, RNAscope with IHC codetection kit), using a 15-minute target retrieval and protease plus for protease digestion. Fluorophores were diluted 1:1,500 (Akoya Biosciences). Sections were colabeled using anti-CD13 antibody (1:800). Samples were imaged on a Zeiss Axio Observer CSU-X spinning disc confocal microscope with a 63×/1.4 NA objective. ## Statistics. The number of animals is indicated for each experiment. All ELISA and urinary creatinine tests were performed in duplicate. All qPCR was performed in triplicate. Data are presented as the mean with $95\%$ CI or SD. Welch’s 2-tailed t test was used to determine the statistical significance between 2 groups. Comparisons of multiple means were performed with 1-way ANOVA followed by Tukey’s test for multiple comparisons. The size of Lamp1 positive objects was analyzed using blinded, observer-based comparisons between groups. Data was analyzed with χ2 test. For all studies, a P value of less than 0.05 was considered significant. To determine the mean glomerular surface area and mean tubular diameter, we used mixed effects regression models to account for the multiple measurements within each mouse. Glomerular surface area has an approximate log-normal distribution; thus, the dependent variable in models involving glomerular surface area is its natural logarithm. Mean values of glomerular surface that we present are predicted geometric means generated by exponentiation of model results. We used similar mixed-effects models to generate predicted arithmetic means for tubular diameter. Statistical analysis was performed with Prism software (version 8.1.2) and R (R Core Team 2021; R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org/). A P value of less than 0.05 was considered significant. ## Study approval. All procedures involving mice were conducted according to the NIH’s Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011) and approved by the IACUC of Columbia University. ## Author contributions PIG designed and performed the majority of experiments, analyzed data, and wrote the manuscript. LL, HAH, ISH, and KX helped with the experiments and acquired and analyzed data. DAB performed the majority of statistical analyses. MR advised on studies using NA-PP1 and the analysis of enteric nervous system. FC and QAA advised on study design and data interpretation and edited the manuscript. VDD provided expertise and insightful discussions of mouse kidney histopathology and edited the manuscript. FL designed and directed the study, performed experiments, and edited the manuscript. ## 01/10/2023 In-Press Preview ## 02/08/2023 Electronic publication ## References 1. Purisch SE, Gyamfi-Bannerman C. Epidemiology of preterm birth. Semin Perinatol (2017.0) 41 387-391. DOI: 10.1053/j.semperi.2017.07.009 2. Walani SR. Global burden of preterm birth. Int J Gynaecol Obstet (2020.0) 150 31-33. DOI: 10.1002/ijgo.13195 3. Crump C. An overview of adult health outcomes after preterm birth. Early Hum Dev (2020.0) 150 105187-105187. DOI: 10.1016/j.earlhumdev.2020.105187 4. Jetton JG. Incidence and outcomes of neonatal acute kidney injury (AWAKEN): a multicentre, multinational, observational cohort study. Lancet Child Adolesc Health (2017.0) 1 184-194. DOI: 10.1016/S2352-4642(17)30069-X 5. Chawla LS. Acute kidney injury and chronic kidney disease as interconnected syndromes. N Engl J Med (2014.0) 371 58-66. DOI: 10.1056/NEJMra1214243 6. Coca SG. Chronic kidney disease after acute kidney injury: a systematic review and meta-analysis. Kidney Int (2012.0) 81 442-448. DOI: 10.1038/ki.2011.379 7. Horie A. Long-term renal follow up of preterm neonates born before 35 weeks of gestation. Pediatr Int (2019.0) 61 1244-1249. DOI: 10.1111/ped.14004 8. Hsu CW. Prenatal risk factors for childhood CKD. J Am Soc Nephrol (2014.0) 25 2105-2111. DOI: 10.1681/ASN.2013060582 9. Crump C. Preterm birth and risk of chronic kidney disease from childhood into mid-adulthood: national cohort study. BMJ (2019.0) 365. DOI: 10.1053/j.semperi.2017.07.009 10. Hinchliffe SA. Human intrauterine renal growth expressed in absolute number of glomeruli assessed by the disector method and Cavalieri principle. Lab Invest (1991.0) 64 777-784. PMID: 2046329 11. Sutherland MR. Accelerated maturation and abnormal morphology in the preterm neonatal kidney. J Am Soc Nephrol (2011.0) 22 1365-1374. DOI: 10.1681/ASN.2010121266 12. Rodríguez MM. Histomorphometric analysis of postnatal glomerulogenesis in extremely preterm infants. Pediatr Dev Pathol (2003.0) 7 17-25. DOI: 10.1007/s10024-003-3029-2 13. Brenner BM. Glomeruli and blood pressure. Less of one, more the other?. Am J Hypertens (1988.0) 1 335-347. DOI: 10.1093/ajh/1.4.335 14. Luyckx VA, Brenner BM. Birth weight, malnutrition and kidney-associated outcomes--a global concern. Nat Rev Nephrol (2015.0) 11 135-149. DOI: 10.1038/nrneph.2014.251 15. Luyckx VA. Low nephron number and its clinical consequences. Rambam Maimonides Med J (2011.0) 2. DOI: 10.5041/RMMJ.10061 16. Luyckx VA. Effect of fetal and child health on kidney development and long-term risk of hypertension and kidney disease. Lancet (2013.0) 382 273-283. DOI: 10.1016/S0140-6736(13)60311-6 17. Hostetter HS, Brenner BM. Fewer nephrons at birth: a missing link in the etiology of essential hypertension?. Am J Kidney Dis (1995.0) 26 91-98. DOI: 10.1016/0272-6386(95)90161-2 18. Hostetter TH. Hyperfiltration in remnant nephrons: a potentially adverse response to renal ablation. Am J Physiol (1981.0) 241 F85-F93. DOI: 10.1152/ajprenal.1981.241.1.F85 19. Shimamura T, Morrison AB. A progressive glomerulosclerosis occurring in partial five-sixths nephrectomized rats. Am J Pathol (1975.0) 79 95-106. PMID: 164779 20. Schreuder MF. Glomerular number and function are influenced by spontaneous and induced low birth weight in rats. J Am Soc Nephrol (2005.0) 16 2913-2919. DOI: 10.1681/ASN.2004100875 21. Merlet-Bénichou C. Intrauterine growth retardation leads to a permanent nephron deficit in the rat. Pediatr Nephrol (1994.0) 8 175-180. DOI: 10.1007/BF00865473 22. Gilbert JS. Maternal nutrient restriction in sheep: hypertension and decreased nephron number in offspring at 9 months of age. J Physiol (2005.0) 565 137-147. DOI: 10.1113/jphysiol.2005.084202 23. Woods LL. Programming of adult blood pressure by maternal protein restriction: role of nephrogenesis. Kidney Int (2004.0) 65 1339-1348. DOI: 10.1111/j.1523-1755.2004.00511.x 24. Chan SY. Maternal nutrient deprivation induces sex-specific changes in thyroid hormone receptor and deiodinase expression in the fetal guinea pig brain. J Physiol (2005.0) 566 467-480. DOI: 10.1113/jphysiol.2005.084673 25. Zelko IN. Maternal undernutrition during pregnancy alters the epigenetic landscape and the expression of endothelial function genes in male progeny. Nutr Res (2019.0) 61 53-63. DOI: 10.1016/j.nutres.2018.10.005 26. Trahair JF. Restriction of nutrition in utero selectively inhibits gastrointestinal growth in fetal sheep. J Nutr (1997.0) 127 637-641. DOI: 10.1093/jn/127.4.637 27. Carney EW. The effects of feed restriction during in utero and postnatal development in rats. Toxicol Sci (2004.0) 82 237-249. DOI: 10.1093/toxsci/kfh249 28. Stelloh C. Prematurity in mice leads to reduction in nephron number, hypertension, and proteinuria. Transl Res (2012.0) 159 80-89. DOI: 10.1016/j.trsl.2011.10.004 29. Costantini F, Kopan R. Patterning a complex organ: branching morphogenesis and nephron segmentation in kidney development. Dev Cell (2010.0) 18 698-712. DOI: 10.1016/j.devcel.2010.04.008 30. Costantini F, Shakya R. GDNF/Ret signaling and the development of the kidney. Bioessays (2006.0) 28 117-127. DOI: 10.1002/bies.20357 31. Dressler GR. Advances in early kidney specification, development and patterning. Development (2009.0) 136 3863-3874. DOI: 10.1242/dev.034876 32. Sampogna RV. Developmental programming of branching morphogenesis in the kidney. J Am Soc Nephrol (2015.0) 26 2414-2422. DOI: 10.1681/ASN.2014090886 33. 33Potter E, ed. Normal and Abnormal Development of the Kidney. Year Book Medical Publisher; 1972. 34. Uesaka TK. To bud or not to bud: the RET perspective in CAKUT. Pediatr Nephrol (2014.0) 29 597-608. DOI: 10.1007/s00467-013-2606-5 35. Costantini F. GDNF/Ret signaling and renal branching morphogenesis: from mesenchymal signals to epithelial cell behaviors. Organogenesis (2010.0) 6 252-262. DOI: 10.4161/org.6.4.12680 36. Uesaka T. Diminished Ret expression compromises neuronal survival in the colon and causes intestinal aganglionosis in mice. J Clin Invest (2008.0) 118 1890-1898. DOI: 10.1172/JCI34425 37. Savitt J. The in vivo response of stem and other undifferentiated spermatogonia to the reversible inhibition of glial cell line-derived neurotrophic factor signaling in the adult. Stem Cells (2012.0) 30 732-740. DOI: 10.1002/stem.1028 38. Bishop AC. A chemical switch for inhibitor-sensitive alleles of any protein kinase. Nature (2000.0) 407 395-401. DOI: 10.1038/35030148 39. Bishop AC, Shokat KM. Acquisition of inhibitor-sensitive protein kinases through protein design. Pharmacol Ther (1999.0) 82 337-346. DOI: 10.1016/s0163-7258(98)00060-6 40. Bishop AC. Design of allele-specific inhibitors to probe protein kinase signaling. Curr Biol (1998.0) 8 257-266. DOI: 10.1016/S0960-9822(98)70198-8 41. Endo S. A single amino-acid change in ERK1/2 makes the enzyme susceptible to PP1 derivatives. Biochem Biophys Res Commun (2006.0) 341 261-265. DOI: 10.1016/j.bbrc.2005.12.179 42. MacKay K. Glomerulosclerosis and renal cysts in mice transgenic for the early region of SV40. Kidney Int (1987.0) 32 827-837. DOI: 10.1038/ki.1987.283 43. Taraviras S. Signalling by the RET receptor tyrosine kinase and its role in the development of the mammalian enteric nervous system. Development (1999.0) 126 2785-2797. DOI: 10.1242/dev.126.12.2785 44. Natarajan D. Requirement of signalling by receptor tyrosine kinase RET for the directed migration of enteric nervous system progenitor cells during mammalian embryogenesis. Development (2002.0) 129 5151-5160. DOI: 10.1242/dev.129.22.5151 45. Shakya R. The role of GDNF in patterning the excretory system. Dev Biol (2005.0) 283 70-84. DOI: 10.1016/j.ydbio.2005.04.008 46. Stieger K. In vivo gene regulation using tetracycline-regulatable systems. Adv Drug Deliv Rev (2009.0) 61 527-541. DOI: 10.1016/j.addr.2008.12.016 47. Hughson M. Glomerular number and size in autopsy kidneys: the relationship to birth weight. Kidney Int (2003.0) 63 2113-2122. DOI: 10.1046/j.1523-1755.2003.00018.x 48. 48Oliver JR, ed. Architecture of the Kidney in Chronic Bright’s Disease. Paul B. Hoeber; 1939. 49. Fine L. The biology of renal hypertrophy. Kidney Int (1986.0) 29 619-634. DOI: 10.1038/ki.1986.45 50. 50Jennette JC, et al, eds. Heptinstall’s Pathology of the Kidney. Lippincott Williams & Wilkins; 2014. 51. Hsieh EM. Medication use in the neonatal intensive care unit. Am J Perinatol (2013.0) 31 811-821. DOI: 10.1055/s-0033-1361933 52. Puopolo KM. Management of neonates born at ≤34 6/7 weeks’ gestation with suspected or proven early-onset bacterial sepsis. Pediatrics (2018.0) 142. DOI: 10.1542/peds.2018-2896 53. Beauchamp D. Subcellular distribution of gentamicin in proximal tubular cells, determined by immunogold labeling. Antimicrob Agents Chemother (1991.0) 35 2173-2179. DOI: 10.1128/AAC.35.11.2173 54. Sastrasinh M. Identification of the aminoglycoside binding site in rat renal brush border membranes. J Pharmacol Exp Ther (1982.0) 222 350-358. PMID: 7097555 55. Dagil R. Gentamicin binds to the megalin receptor as a competitive inhibitor using the common ligand binding motif of complement type repeats: insight from the nmr structure of the 10th complement type repeat domain alone and in complex with gentamicin. J Biol Chem (2013.0) 288 4424-4435. DOI: 10.1074/jbc.M112.434159 56. Schmitz C. Megalin deficiency offers protection from renal aminoglycoside accumulation. J Biol Chem (2002.0) 277 618-622. DOI: 10.1074/jbc.M109959200 57. Regec AL. Effect of gentamicin on the lysosomal system of cultured human proximal tubular cells. Endocytotic activity, lysosomal pH and membrane fragility. Biochem Pharmacol (1989.0) 38 2527-2534. DOI: 10.1016/0006-2952(89)90098-1 58. Schuh CD. Combined structural and functional imaging of the kidney reveals major axial differences in proximal tubule endocytosis. J Am Soc Nephrol (2018.0) 29 2696-2712. DOI: 10.1681/ASN.2018050522 59. Lin F. Intrarenal cells, not bone marrow-derived cells, are the major source for regeneration in postischemic kidney. J Clin Invest (2005.0) 115 1756-1764. DOI: 10.1172/JCI23015 60. Humphreys BD. Intrinsic epithelial cells repair the kidney after injury. Cell Stem Cell (2008.0) 2 284-291. DOI: 10.1016/j.stem.2008.01.014 61. Grgic I. The origin of interstitial myofibroblasts in chronic kidney disease. Pediatr Nephrol (2012.0) 27 183-193. DOI: 10.1007/s00467-011-1772-6 62. Grgic I. Targeted proximal tubule injury triggers interstitial fibrosis and glomerulosclerosis. Kidney Int (2012.0) 82 172-183. DOI: 10.1038/ki.2012.20 63. Rabb H. Inflammation in AKI: current understanding, key questions, and knowledge gaps. J Am Soc Nephrol (2016.0) 27 371-379. DOI: 10.1681/ASN.2015030261 64. Blydt-Hansen TD. Elevated urinary CXCL10-to-creatinine ratio is associated with subclinical and clinical rejection in pediatric renal transplantation. Transplantation (2015.0) 99 797-804. DOI: 10.1097/TP.0000000000000419 65. Hasegawa H. Antagonist of monocyte chemoattractant protein 1 ameliorates the initiation and progression of lupus nephritis and renal vasculitis in MRL/lpr mice. Arthritis Rheum (2003.0) 48 2555-2566. DOI: 10.1002/art.11231 66. Shimizu S. Anti-monocyte chemoattractant protein-1 gene therapy attenuates nephritis in MRL/lpr mice. Rheumatology (Oxford) (2004.0) 43 1121-1128. DOI: 10.1093/rheumatology/keh277 67. De Lema GP. Chemokine receptor Ccr2 deficiency reduces renal disease and prolongs survival in MRL/lpr lupus-prone mice. J Am Soc Nephrol (2005.0) 16 3592-3601. DOI: 10.1681/ASN.2005040426 68. Bojic S. Diagnostic value of matrix metalloproteinase-9 and tissue inhibitor of matrix metalloproteinase-1 in sepsis-associated acute kidney injury. Tohoku J Exp Med (2015.0) 237 103-109. DOI: 10.1620/tjem.237.103 69. Engelmyer E. Differential mRNA expression of renal cortical tissue inhibitor of metalloproteinase-1, -2, and -3 in experimental hydronephrosis. J Am Soc Nephrol (1995.0) 5 1675-1683. DOI: 10.1681/ASN.V591675 70. Janssens P. Enhanced MCP-1 release in early autosomal dominant polycystic kidney disease. Kidney Int Rep (2021.0) 6 1687-1698. DOI: 10.1016/j.ekir.2021.03.893 71. Tang WW. Extracellular matrix accumulation in immune-mediated tubulointerstitial injury. Kidney Int (1994.0) 45 1077-1084. DOI: 10.1038/ki.1994.144 72. Ninichuk V. Late onset of Ccl2 blockade with the Spiegelmer mNOX-E36-3’PEG prevents glomerulosclerosis and improves glomerular filtration rate in db/db mice. Am J Pathol (2008.0) 172 628-637. DOI: 10.2353/ajpath.2008.070601 73. Haller H. Monocyte chemoattractant protein-1 and the kidney. Curr Opin Nephrol Hypertens (2016.0) 25 42-49. DOI: 10.1097/MNH.0000000000000186 74. Chow FY. Monocyte chemoattractant protein-1 promotes the development of diabetic renal injury in streptozotocin-treated mice. Kidney Int (2006.0) 69 73-80. DOI: 10.1038/sj.ki.5000014 75. Zoja C. Renal expression of monocyte chemoattractant protein-1 in lupus autoimmune mice. J Am Soc Nephrol (1997.0) 8 720-729. DOI: 10.1681/ASN.V85720 76. Gao J. Role of chemokine (C-X-C Motif) ligand 10 (cxcl10) in renal diseases. Mediators Inflamm (2020.0) 2020. DOI: 10.1155/2020/6194864 77. Hirt-Minkowski P. Six-month urinary CCL2 and CXCL10 levels predict long-term renal allograft outcome. Transplantation (2016.0) 100 1988-1996. DOI: 10.1097/TP.0000000000001304 78. Kulkarni O. Spiegelmer inhibition of CCL2/MCP-1 ameliorates lupus nephritis in MRL-(Fas)lpr mice. J Am Soc Nephrol (2007.0) 18 2350-2358. DOI: 10.1681/ASN.2006121348 79. Marks SD. Urinary monocyte chemoattractant protein-1 correlates with disease activity in lupus nephritis. Pediatr Nephrol (2010.0) 25 2283-2288. DOI: 10.1007/s00467-010-1605-z 80. Ramirez SC. Perinatal breathing patterns and survival in mice born prematurely and at term. Front Physiol (2019.0) 10. DOI: 10.3389/fphys.2019.01113 81. Riccio P. Ret and Etv4 promote directed movements of progenitor cells during renal branching morphogenesis. PLoS Biol (2016.0) 14. DOI: 10.1371/journal.pbio.1002382 82. Zhang Z. A common RET variant is associated with reduced newborn kidney size and function. J Am Soc Nephrol (2008.0) 19 2027-2034. DOI: 10.1681/ASN.2007101098 83. Romeo G. Point mutations affecting the tyrosine kinase domain of the RET proto-oncogene in Hirschsprung’s disease. Nature (1994.0) 367 377-378. DOI: 10.1038/367377a0 84. Mañalich R. Relationship between weight at birth and the number and size of renal glomeruli in humans: a histomorphometric study. Kidney Int (2000.0) 58 770-773. DOI: 10.1046/j.1523-1755.2000.00225.x 85. Hoy WE. Nephron number, hypertension, renal disease, and renal failure. J Am Soc Nephrol (2005.0) 16 2557-2564. DOI: 10.1681/ASN.2005020172 86. Hodgin JB. Very low birth weight is a risk factor for secondary focal segmental glomerulosclerosis. Clin J Am Soc Nephrol (2009.0) 4 71-76. DOI: 10.2215/CJN.01700408 87. Celsi G. Influence of age on compensatory renal growth in rats. Pediatr Res (1986.0) 20 347-350. DOI: 10.1203/00006450-198604000-00018 88. Menendez-Castro C. Neonatal nephron loss during active nephrogenesis — detrimental impact with long-term renal consequences. Sci Rep (2018.0) 8. DOI: 10.1038/s41598-018-22733-8 89. Puelles VG. Glomerular hypertrophy in subjects with low nephron number: contributions of sex, body size and race. Nephrol Dial Transplant (2014.0) 29 1686-1695. DOI: 10.1093/ndt/gfu088 90. Deshmane SL. Monocyte chemoattractant protein-1 (MCP-1): an overview. J Interferon Cytokine Res (2009.0) 29 313-326. DOI: 10.1089/jir.2008.0027 91. Rovin BH. Urine chemokines as biomarkers of human systemic lupus erythematosus activity. J Am Soc Nephrol (2005.0) 16 467-473. DOI: 10.1681/ASN.2004080658 92. Taub DD. Recombinant human interferon-inducible protein 10 is a chemoattractant for human monocytes and T lymphocytes and promotes T cell adhesion to endothelial cells. J Exp Med (1993.0) 177 1809-1814. DOI: 10.1084/jem.177.6.1809 93. Park J-W. Primary hepatocytes from mice treated with IL-2/IL-12 produce T cell chemoattractant activity that is dependent on monokine induced by IFN-gamma (Mig) and chemokine responsive to gamma-2 (Crg-2). J Immunol (2001.0) 166 3763-3770. DOI: 10.4049/jimmunol.166.6.3763 94. Ho J. Urinary CXCL10 chemokine is associated with alloimmune and virus compartment-specific renal allograft inflammation. Transplantation (2018.0) 102 521-529. DOI: 10.1097/TP.0000000000001931 95. Merhi B. Role for urinary biomarkers in diagnosis of acute rejection in the transplanted kidney. World J Transplant (2015.0) 5 251-260. DOI: 10.5500/wjt.v5.i4.251 96. Gomez DE. Tissue inhibitors of metalloproteinases: structure, regulation and biological functions. Eur J Cell Biol (1997.0) 74 111-122. PMID: 9352216 97. Kumar S. Sox9 activation highlights a cellular pathway of renal repair in the acutely injured mammalian kidney. Cell Rep (2015.0) 12 1325-1338. DOI: 10.1016/j.celrep.2015.07.034 98. Berger K. Origin of regenerating tubular cells after acute kidney injury. Proc Natl Acad Sci U S A (2014.0) 111 1533-1538. DOI: 10.1073/pnas.1316177111 99. Houghton DC. A light and electron microscopic analysis of gentamicin nephrotoxicity in rats. Am J Pathol (1976.0) 82 589-612. PMID: 1258978 100. Huang H. Gentamicin-induced acute kidney injury in an animal model involves programmed necrosis of the collecting duct. J Am Soc Nephrol (2020.0) 31 2097-2115. DOI: 10.1681/ASN.2019020204 101. Morales AI. Metformin prevents experimental gentamicin-induced nephropathy by a mitochondria-dependent pathway. Kidney Int (2010.0) 77 861-869. DOI: 10.1038/ki.2010.11 102. Liu KD. AKI!Now initiative: recommendations for awareness, recognition, and management of AKI. Clin J Am Soc Nephrol (2020.0) 15 1838-1847. DOI: 10.2215/CJN.15611219 103. Eremina V. VEGF inhibition and renal thrombotic microangiopathy. N Engl J Med (2008.0) 358 1129-1136. DOI: 10.1056/NEJMoa0707330
--- title: Acute high-fat diet impairs macrophage-supported intestinal damage resolution authors: - Andrea A. Hill - Myunghoo Kim - Daniel F. Zegarra-Ruiz - Lin-Chun Chang - Kendra Norwood - Adrien Assié - Wan-Jung H. Wu - Michael C. Renfroe - Hyo Wong Song - Angela M. Major - Buck S. Samuel - Joseph M. Hyser - Randy S. Longman - Gretchen E. Diehl journal: JCI Insight year: 2023 pmcid: PMC9977439 doi: 10.1172/jci.insight.164489 license: CC BY 4.0 --- # Acute high-fat diet impairs macrophage-supported intestinal damage resolution ## Abstract Chronic exposure to high-fat diets (HFD) worsens intestinal disease pathology, but acute effects of HFD in tissue damage remain unclear. Here, we used short-term HFD feeding in a model of intestinal injury and found sustained damage with increased cecal dead neutrophil accumulation, along with dietary lipid accumulation. Neutrophil depletion rescued enhanced pathology. Macrophages from HFD-treated mice showed reduced capacity to engulf dead neutrophils. Macrophage clearance of dead neutrophils activates critical barrier repair and antiinflammatory pathways, including IL-10, which was lost after acute HFD feeding and intestinal injury. IL-10 overexpression restored intestinal repair after HFD feeding and intestinal injury. Macrophage exposure to lipids from the HFD prevented tethering and uptake of apoptotic cells and Il10 induction. Milk fat globule-EGF factor 8 (MFGE8) is a bridging molecule that facilitates macrophage uptake of dead cells. MFGE8 also facilitates lipid uptake, and we demonstrate that dietary lipids interfere with MFGE8-mediated macrophage apoptotic neutrophil uptake and subsequent Il10 production. Our findings demonstrate that HFD promotes intestinal pathology by interfering with macrophage clearance of dead neutrophils, leading to unresolved tissue damage. ## Introduction Within the intestine, a single layer of epithelial cells separates internal tissues from luminal contents. This forms a selective barrier, allowing for nutrient absorption while excluding harmful substances and intestinal microbes. Injury to this barrier increases entry of luminal contents into the tissue. Under normal conditions, microbial entry into tissue after damage is sensed by intestinal cells, including macrophages, which recruit neutrophils and other immune cells to clear infection and orchestrates tissue repair [1, 2]. Macrophages and epithelial cells secrete chemokines, such as CXCL1 and CXCL2, to recruit CXCR2-expressing neutrophils and support microbial clearance from the tissue (3–5). After microbial clearance, neutrophil effector mechanisms must be rapidly contained to prevent collateral tissue damage [6]. Neutrophil apoptosis and subsequent phagocytosis by macrophages activate key signals for tissue healing and for dampening inflammation [6, 7]. A functional switch of macrophages from an antimicrobial, proinflammatory response to a pro–tissue repair, antiinflammatory response promotes barrier repair [6, 7]. This includes macrophage production of antiinflammatory prorepair cytokine IL-10, which supports repair of intestinal damage (8–11). One inducer of this switch is macrophage phagocytosis of dead neutrophils [6, 12, 13]. Defects in macrophage dead cell clearance results in impaired prorepair responses and lost IL-10 production and is linked to chronic inflammatory diseases such as atherosclerosis, cardiovascular disease, chronic obstructive pulmonary disease (COPD), and autoimmune diseases, including type 1 diabetes. Furthermore, enhanced or unresolved intestinal barrier damage underlies disease pathology in inflammatory bowel disease (IBD), including loss of IL-10 and increased tissue neutrophil accumulation [5, 14]. Identifying mechanisms that support repair and understanding how these pathways are dysregulated in disease are key to improving resolution of tissue damage in IBD. The intestinal tract is exposed to a variety of environmental factors with the potential to disrupt tissue repair and enhance disease pathology. One such factor is dietary lipids, with increased fat intake as a risk factor for chronic inflammatory diseases, including IBD [15]. Chronic exposure to a high-fat diet (HFD) suppresses fat-associated macrophage antiinflammatory functions, including IL-10 production, and promotes proinflammatory responses, supporting local adipose tissue expansion and systemic inflammation (16–18). HFD increases susceptibility to intestinal damage and inflammation in colitis models (19–22). However, whether dysregulated repair mechanisms in early stages of HFD exposure sensitizes to the development or progression of intestinal disease is unclear. HFD-induced altered macrophage functions is driven by macrophage lipid uptake (16–18). Studies suggest that dead cells and lipids are taken up through a shared pathway involving bridging molecule milk fat globule-EGF factor 8 (MFGE8). The discoidin domains of MFGE8 bind externalized phosphatidyl serine (PS) on the membrane of dead cells and induces uptake through macrophage αVβ3 [7, 23, 24]. Similarly, the MFGE8 discoidin domains also bind dietary lipids, and this binding induces lipid uptake into adipocytes and promotes HFD-induced obesity [25]. Disruption of MFGE8 binding to αVβ3 or loss of MFGE8 impairs apoptotic neutrophil and lipid uptake, highlighting the importance of this pathway in uptake of both and suggesting that this pathway is important in HFD regulation of macrophage function [24, 25]. In this study, we sought to determine the impact of short-term HFD feeding in intestinal injury. Using in vivo and in vitro studies, we demonstrate that HFD impairs resolution after intestinal injury by decreasing macrophage clearance of apoptotic neutrophils. We find this effect is driven by the direct interference of dietary lipids on MFGE8-mediated macrophage uptake of apoptotic neutrophils and subsequent IL-10 production. These findings demonstrate a previously unidentified mechanism by which dietary lipids, a risk factor for intestinal disease, directly interfere with homeostatic processes required to resolve tissue injury. The breakdown of intestinal tissue repair responses in the early stages of HFD feeding may set the stage for enhanced inflammation seen after chronic HFD exposure. ## Acute HFD feeding impairs intestinal epithelial repair after injury. In intestinal disease, long-term exposure to HFD is associated with increased pathology. However, whether the initial effects of HFD on damage repair sets the stage for increased pathology seen after chronic HFD feeding is unclear. To understand whether short-term exposure to HFD altered resolution after intestinal damage, we fed mice with a HFD or low-fat diet (LFD) for 1 week before exposure to dextran sodium sulfate (DSS), which causes intestinal epithelial damage in the cecum and colon [26, 27]. Mice remained on their respective diets before, during, and after DSS treatment. We chose this timeframe in order to allow us to assess the impact of HFD on intestinal damage prior to the induction of metabolic and systemic inflammatory effects caused by chronic HFD feeding [28]. Without DSS exposure, mice in both groups had comparable body weight with normal glucose and insulin tolerance, as compared with mice fed HFD for 8 weeks who had decreased glucose but normal insulin tolerance (Figure 1A and Supplemental Figure 1, A and B; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.164489DS1). After DSS treatment, mice exposed to HFD or LFD initially lost a comparable amount of weight. While LFD-fed mice recovered initial weight, HFD-fed mice showed sustained weight loss (Figure 1A). By histopathological analysis, we found that acute HFD feeding alone did not increase tissue damage or inflammation in the cecum, terminal ileum, or colon (Figure 1, B and C, and Supplemental Figure 1, C–E). At day 5 of DSS treatment, we found increased pathology in the cecum, terminal ileum, proximal, and distal colon with equivalent colitis scores in both HFD- and LFD-treated animals (Figure 1, B and C, and Supplemental Figure 1, C–E), indicating that diet was not amplifying initial damage or inflammation, as is seen after chronic exposure to HFD [22, 29, 30]. In LFD- and HFD-fed mice, by 4 days after DSS treatment (day 9), the damage was partially resolved in the terminal ileum and proximal and distal colon (Supplemental Figure 1, C–E) with reduced tissue infiltration of immune cells and restitution of the epithelial barrier. In the cecum of LFD-fed mice, we observed a similar resolution of damage (Figure 1, B and C). In contrast, in HFD-fed mice, we observed prolonged tissue damage and cellular infiltration (Figure 1, B and C). HFD consumption is associated with microbiota compositional changes, which are suggested to be a mechanism for HFD-driven intestinal inflammation [31]. Changes in microbiota composition with outgrowth of specific taxa are also associated with increased inflammation and can support pathology after DSS treatment (32–34). Increased damage to intestinal tissue can also cause shifts in intestinal microbe composition [33]. To determine if microbiota changes occurred after acute HFD and DSS treatment and whether these changes could account for sustained cecal damage, we performed 16S rRNA-Seq to assess cecal microbial composition. At the phylum level (Supplemental Figure 1F), only Proteobacteria and Actinobacteria were increased in the cecum of HFD-fed mice before DSS treatment, with similar levels between diet groups after DSS. Firmicutes decreased in both groups after DSS treatment, with a greater decrease in HFD-fed mice. Bacteroidetes increased similarly in both groups after DSS treatment. There were minimal changes at the family level. In HFD-fed mice, Desulfovibrionaceae, Clostridiales, and Lachnospiraceae were increased before DSS treatment. After DSS treatment, Desulfovibrionaceae, Clostridiales, and Lachnospiraceae levels were equivalent between diet groups (Supplemental Figure 1G). After DSS treatment, we found increased Bacteroidaceae in HFD-treated mice (Supplemental Figure 1G). In LFD-treated mice, we found increased Erysipelotrichaceae that deceased after DSS treatment but was still elevated as compared with HFD-fed mice (Supplemental Figure 1H). After DSS treatment, Bacteroidales and Porphyromonadaceae increased in the cecum of LFD-fed mice (Supplemental Figure 1H). Sutterellaceae was elevated in both diet groups after DSS treatment (Supplemental Figure 1I). Increased levels of Erysipelotrichaceae, Bacteroidales, and Porphyromonadaceae are associated with enhanced disease in patients with IBD and mouse models of intestinal inflammation (35–37), while Bacteroidaceae, is reduced in patients with IBD [38]. To confirm that altered microbes were not sufficient to impede repair after DSS treatment, we transplanted cecal microbiota contents from LFD- and HFD-fed mice into antibiotic-treated mice, which we then treated with DSS. Colonization of mice with cecal contents from HFD-treated mice did not phenocopy HFD feeding itself, as mice did not have sustained weight loss as compared with mice colonized with cecal microbes from LFD-fed mice (Supplemental Figure 1J). Since we did not find expansion of microbes associated with enhanced intestinal pathology after HFD feeding or sustained disease after transplant of cecal microbes from HFD-fed mice, we do not think HFD-driven changes in microbial composition underlies lack of resolution of cecal damage after short-term HFD feeding. To understand why mice on HFD had impaired damage resolution only in the cecum, we asked if lipids specifically accumulated in the cecum. To assess lipid content, we used BODIPY, which stains neutral lipids. BODIPY staining intensity was similar between the cecum and colon of LFD-fed mice (Supplemental Figure 1K). In HFD-fed mice, there was increased BODIPY staining in the cecum, as compared with the colon. When comparing between diet groups, HFD-fed mice has increased lipids in the cecum but not colon, as compared with LFD-fed mice (Supplemental Figure 1K), suggesting that increased cecal lipids are associated with lack of tissue repair. Based on these findings, we focused the remainder of our study on the cecum. One of the first steps in repairing intestinal epithelial damage is proliferation of intestinal stem cells (ISC) followed by differentiation to repopulate lost cell types [39]. Chronic HFD exposure alone can alter colonic ISC proliferation, resulting in increased ISC numbers [40]. We next assessed if ISC proliferation in response to damage was different in LFD- versus HFD-fed mice. Before DSS and at day 5 of DSS treatment, we found similar epithelial proliferation in the cecum of HFD- and LFD-fed mice (Figure 1, D and E). By day 9, proliferation in the cecum decreased to pre-DSS levels in the LFD-treated mice. However, it remained elevated in the HFD-treated mice (Figure 1, D and E). These findings demonstrate that sustained ISC proliferation maintained increased numbers of ISC in cecal crypts but did not support the reestablishment of the epithelial barrier (Figure 1, B and C). In order to form a tight seal between repopulated epithelial cells in repair of intestinal damage, tight junctions must be reestablished. This paracellular barrier prevents microbial translocation into the tissue [41]. Expression of tight junction proteins occludin (Ocln) and zonula occluden-1 (ZO1) — which, together, are the major tight-junction proteins [42] — did not differ in the cecum between LFD- and HFD-fed mice before DSS treatment (Figure 2, A and B). By day 7, we found upregulated expression of Ocln and Tjp1 in the cecum of LFD, but not HFD treated mice (Figure 2, A and B). Next, we performed immunofluorescence (IF) staining to determine differences in OCLN and ZO1 cecal protein expression in the epithelium. OCLN, but not ZO1, protein was decreased in the epithelium of HFD-fed mice at day 0 prior to DSS treatment (Figure 2, C–F). Similar effects on OCLN expression in the distal ileum adjacent to the cecum have been seen in long-term HFD feeding [20]. Previous studies demonstrate that OCLN and ZO1 expression decreases after DSS treatment [43]. By day 9 after DSS treatment, OCLN was reduced in both diet groups, with lower expression in HFD-fed mice as compared with LFD-fed mice (Figure 2, C and D). ZO1 expression in LFD-fed mice at day 9 after DSS was comparable with mice before DSS treatment. In contrast, ZO1 expression was reduced in HFD-fed mice after DSS treatment (Figure 2, E and F). Restoration of the intestinal barrier includes the repopulation of goblet cells and mucus, which aid in decreasing inflammatory microbial-epithelial interactions [44]. We performed Alcian blue/PAS (ABPAS) staining to assess goblet cell numbers and found decreased goblet cells at day 9 after DSS in HFD-fed mice as compared with LFD-fed mice (Supplemental Figure 1L). We next performed FISH for bacteria and MUC2 staining for mucus in the cecum of LFD- and HFD-fed mice before and after DSS treatment to assess these interactions. The mucus barrier in the cecum was equivalent in both diet groups prior to DSS treatment and was similarly disrupted at day 5 of DSS, with extensive interactions between intestinal microbes and the cecal epithelium in both groups (Figure 3, A–C). While mucus production was elevated in LFD-treated mice by day 9, mucus production remained low in the cecum of HFD-treated mice (Figure 3, A and B). While there was good separation by day 9 of microbes from the epithelium in the LFD-treated mice, there was little separation between microbes and epithelial cells in the cecum of HFD-treated mice (Figure 3, A and C). These increased interactions with intestinal microbes likely support sustained tissue damage and inflammation in HFD-fed mice. ## Continued neutrophil accumulation limits damage repair after DSS in HFD-treated mice. During intestinal damage, interactions of luminal microbes with the epithelium and translocation into damaged tissue induces neutrophil recruitment into both the lamina propria and intestinal lumen [5], enabling microbial clearance. However, continued recruitment and increased neutrophil numbers can be pathological to the tissue (45–47). Neutrophils are short-lived cells that die by apoptosis to prevent the release of cytotoxic intracellular components [48]. One mechanism that aids in induction of tissue repair is the clearance of dead neutrophils from the tissue to prevent secondary necrosis, which amplifies tissue pathology and limits intestinal barrier repair [5, 7]. Defects in apoptotic neutrophil clearance is associated with impaired tissue repair [6]. By H&E staining, we observed neutrophils in the cecum and lumen of both LFD- and HFD-fed mice after DSS treatment, and it remained elevated in HFD-fed mice (Supplemental Figure 2A). Since dead neutrophil accumulation can amplify tissue damage, we next costained for apoptotic cells using TUNEL and neutrophil marker Ly6G to assess neutrophil numbers and viability in the tissue. Prior to DSS treatment, we found no TUNEL or Ly6G staining in the cecum of either group (Figure 4, A–C). We saw similarly increased numbers of neutrophils and dead cells (the majority of which were neutrophils) in the cecum of HFD- and LFD-fed mice at day 5 of DSS (Figure 4, A–D). At day 9, while neutrophils, dead cells, and dead neutrophil numbers decreased in the cecum of LFD-treated mice, they further increased in the cecum of HFD-treated mice (Figure 4, A–D). In inflammation resolution, neutrophil-recruiting chemokine expression is reduced. We examined gene expression of neutrophil chemoattractant proteins CXCL1 and CXCL2 and chemokine receptor CXCR2 in LFD- and HFD-fed mice before and after DSS treatment. At day 7 after DSS treatment, expression of Cxcl1, Cxcl2, and Cxcr2 were increased in the cecum of both groups and were further elevated in HFD-fed mice (Figure 5, A–C). Continued recruitment and accumulation of dead neutrophils correlated with the enhanced pathology we observe in the cecum of HFD-fed DSS-treated mice. To investigate whether neutrophils contributed to reduced tissue repair after acute HFD feeding, we depleted neutrophils in HFD-fed mice starting at day 4 of DSS treatment after the induction of damage. As compared with isotype control antibody–treated mice, neutrophil depletion resulted in improved body weight, improved histopathology, decreased epithelial proliferation, and restored goblet cell numbers (Figure 5, D–J). Neutrophil depletion resulted in the restoration of the mucus barrier and reduced mucosal-associated microbes (Figure 6, A–C). Depletion of neutrophils also resulted in decreased numbers of TUNEL+ cells within the tissue (Figure 6, D and E). Collectively, these data suggest that continued neutrophil recruitment and accumulation of dead neutrophils supports defective resolution after injury in HFD-fed mice. ## Acute HFD feeding impairs intestinal macrophage clearance of dead neutrophils. Macrophage recruitment into the tissue is key to recovery from injury, including clearance of apoptotic cells [49, 50]. We first tested if normal numbers of macrophages were recruited into the tissue. By IF staining and flow cytometric analysis, we confirmed equivalent macrophage recruitment into the tissue after DSS treatment in both HFD- and LFD-fed mice (Supplemental Figure 2, B–G). Macrophage clearance of apoptotic neutrophils is one mechanism that supports antiinflammatory signaling and tissue repair [6, 7]. To identify whether tethering of dead neutrophils, the first step of neutrophil uptake, differed in macrophages from LFD- and HFD-fed mice, we exposed flow-sorted intestinal macrophages from DSS-treated HFD- and LFD-fed mice to TAMRA-labeled dead neutrophils and measured tethering by fluorescence microscopy; we saw reduced tethering of dead neutrophils by macrophages from HFD-treated mice (Figure 7, A and B). We hypothesized that lipid components from the HFD were directly interfering with macrophage uptake of dead neutrophils. The major lipids in the HFD are oleic acid ($50\%$) and palmitic acid ($49\%$; Nu-Chek Prep, N-16-A). To test if these lipids interfered with macrophage uptake of neutrophils, we treated bone marrow–derived macrophages (BMDMs) with oleic acid before incubation with TAMRA-labeled dead neutrophils and assessed neutrophil engulfment by fluorescence microscopy. After pretreatment with oleic acid, we found that BMDMs had a decreased capacity to engulf dead neutrophils with a decreased proportion containing a single neutrophil and none containing more than 1 cell (Figure 7, C–E). In order to engulf dead cells, macrophages must first tether to the exposed phosphatidylserine (PS) on the dead cell membrane. Next, to assess macrophage tethering of dead neutrophils, we performed our uptake assays at 4°C to decrease phagocytosis. While control-treated BMDMs tethered dead neutrophils, oleic acid treatment decreased the ability of BMDMs to tether dead neutrophils (Figure 7F). We found similar results when BMDMs were exposed to palmitic acid, with decreased uptake and tethering (Figure 7, G–J). Using immortalized BMDMs (iBMDMs), we found that this effect was limited to dead neutrophils, as uptake of beads or dead thymocytes was not impacted by oleic acid treatment (Supplemental Figure 2, H–K). These findings demonstrate that lipid components of the HFD directly impair macrophage uptake of dead neutrophils and likely contribute to apoptotic neutrophil accumulation in HFD-fed mice after injury. ## Acute HFD treatment limits macrophage IL-10 production after intestinal injury. Macrophage clearance of apoptotic neutrophils regulates pathways important for barrier repair. This includes decreased expression of neutrophil chemoattractant proteins (Figure 5, A–C) [7]. Macrophage uptake of dead cells also induces expression of antiinflammatory and tissue repair factors, including the antiinflammatory cytokine IL-10 [12, 51]. IL-10 supports epithelial barrier repair [8, 9], regulates T cell responses to the intestinal microbiota and pathogens [52], and increases macrophage phagocytic capacity. While we found Il10 gene expression upregulated in the cecum of LFD-fed mice after DSS treatment, Il10 was not upregulated in the cecum of DSS-treated mice acutely exposed to HFD (Figure 8A). In the intestine, we and others find macrophages are the major source of IL-10 (8, 9, 52–54). After DSS exposure, Il10 expression by cecal macrophages from mice on HFD was reduced as compared with macrophages from LFD-fed mice (Figure 8B). Total tissue and macrophage expression of proinflammatory genes such as Tnf were not altered after HFD treatment (Supplemental Figure 3, A and B). These findings demonstrate that Il10 expression is lost in cecal macrophages in HFD-treated mice in response to tissue injury. To determine if HFD lipids directly limited macrophage Il10 expression in response to dead cells, we pretreated BMDMs with oleic or palmitic acid before exposure to dead neutrophils. Exposure to dead neutrophils increased Il10 expression. However, ll10 was not upregulated after lipid pretreatment (Figure 8C). Together, these findings demonstrate that dietary lipids can inhibit macrophage production of antiinflammatory signals needed to support intestinal tissue repair. ## IL-10 overexpression rescues tissue repair defects in HFD-fed mice. To understand if IL-10 expression was sufficient to protect mice from intestinal damage in the presence of HFD, we overexpressed IL-10 in vivo (Supplemental Figure 3C) and found it protected HFD-treated mice from increased weight loss after DSS treatment (Figure 8D). We also observed improved tissue histology, reduced epithelial proliferation, and increased expression of gap junction proteins (Figure 8, E–J). We also found restored goblet cell numbers (Figure 8, K and L). IL-10 overexpression restored mucus production with reduced microbial interactions with the epithelium (Figure 9, A–C) alongside reduced accumulation of dead cells (Figure 9, D and E). These findings demonstrate that, as seen under normal dietary conditions, IL-10 is sufficient to normalize barrier repair in HFD-treated mice after intestinal injury, suggesting that loss of this response in macrophages exposed to dead neutrophils contributes to defective cecal tissue repair in mice fed HFD. The importance of IL-10 signaling in epithelial cells in the repair of intestinal damage has been shown by many other groups [8, 9]. To determine whether IL-10 signaling induces repair in HFD-fed mice with intestinal injury, we overexpressed IL-10 in HFD-fed mice lacking IL-10 receptor α (IL-10Rα) on macrophages (Supplemental Figure 3D). IL-10Rα on macrophages is dispensable, as IL-10 overexpression rescued HFD-treated mice lacking macrophage IL-10Rα after DSS treatment (Supplemental Figure 3, E–I). We then overexpressed IL-10 in HFD-fed mice lacking the IL-10Rα on epithelial cells (Supplemental Figure 3J). In contrast, IL-10 overexpression did not rescue pathology in HFD-fed mice lacking IL-10Rα on epithelial cells with increased body weight loss and histopathology in the presence of exogenous IL-10 (Supplemental Figure 3, K and L). By histology, we observed very few crypt structures, indicating this signaling pathway may also be critical for stem cell renewal. Excess epithelial proliferation, loss of goblet cells, dead cell accumulation, and the mucus barrier were also not rescued by IL-10 overexpression in HFD DSS-treated mice where epithelial cells lack IL-10Rα (Supplemental Figure 3, M–P). These data demonstrate that IL-10 signaling on intestinal epithelial cells supports repair of cecal damage in HFD-fed mice. ## Dietary lipids interfere with MFGE8-mediated macrophage uptake of apoptotic neutrophils. We sought to identify a potential mechanism of dietary lipid interference with macrophage clearance of apoptotic neutrophils. Previously described pathways for apoptotic neutrophil uptake into macrophages are identified as also important for cellular uptake of dietary lipids [25]. The bridging molecule MFGE8 and its receptor αVβ3 facilitate apoptotic neutrophil uptake into macrophages [7, 23, 24]. The MFGE8 discoidin domain binds to externalized PS on dead neutrophils and the Gly-Arg-Gly-Asp-Asn-Pro (RGD) motif of MFGE8 binds to αVβ3 [24]. Recent studies show that dietary lipid uptake into adipocytes is facilitated by MFGE8 and αVβ3 with lipid binding to the discoidin domains of MFGE8 [25]. Together, these studies demonstrate a shared mechanism of apoptotic neutrophil and lipid uptake. Due to the dual role of αVβ3 and MFGE8 in PS and dietary lipid uptake into cells, we next asked whether αVβ3 and MFGE8 expression was altered in LFD- and HFD-fed mice in response to intestinal injury, as loss of MFGE8 or αVβ3 can impair apoptotic neutrophil and lipid uptake (23–25). By flow cytometry, we found similar surface expression of αVβ3 on cecal macrophages from LFD- and HFD-fed mice after DSS treatment (Supplemental Figure 4A). No change in expression was seen for other PS receptors, including Axl and Mertk [55] (Supplemental Figure 4, B–E). In the intestine, MFGE8 is constitutively expressed, with increased expression after intestinal damage that returns to baseline after tissue healing [56]. Loss of MFGE8 results in delayed recovery after DSS treatment, but treatment with recombinant MFGE8 attenuates damage in colitis models [56, 57]. We used IF staining to assess MFGE8 levels in LFD- and HFD-fed mice after DSS treatment [56]. We found similar expression of MFGE8 in both groups before DSS treatment, with similar increases in response to DSS (Figure 10, A and B). However, by day 9, while expression decreased in the cecum of LFD-fed mice, MFGE8 remained elevated in HFD-fed mice (Figure 10, A and B). These findings demonstrate that cecal MFGE8 expression in HFD corresponded with impaired healing of damage. To test if MFGE8 interactions with αVβ3 were required for uptake of both apoptotic cells and lipids, we pretreated BMDMs with RGD peptide, which blocks MFGE8 interaction with its receptor [25]. This pretreatment reduced uptake of both apoptotic neutrophils and lipids (Supplemental Figure 4, F and G). This response was specific for apoptotic neutrophils and lipids, since RGD treatment of BMDMs did not reduce uptake of apoptotic thymocytes (Supplemental Figure 4H), which is Mertk dependent [58]. These data demonstrate that macrophage uptake of both apoptotic neutrophils and dietary lipids requires MFGE8 and αVβ3. Increased apoptotic neutrophils and MFGE8 expression in HFD-fed mice after injury suggested that MFGE8 may not facilitate apoptotic neutrophil clearance in the presence of dietary lipids. To test if there was competition between these 2 MFGE8 ligands, we first assessed whether macrophage uptake of apoptotic neutrophils or dietary lipids was enhanced by MFGE8. We exposed BMDMs to TAMRA-labeled apoptotic neutrophils or oleic acid in the presence or absence of recombinant mouse MFGE8 (rmMFGE8). The addition of rmMFGE8 increased uptake of both TAMRA-labeled apoptotic neutrophils and lipids, as assessed by BODIPY staining (Figure 10, C–F). Since apoptotic neutrophils and oleic acid uptake both depend on MFGE8 and αVβ3, we asked if oleic acid inhibited uptake of apoptotic neutrophils. Coexposure of macrophages to apoptotic neutrophils and oleic acid resulted in decreased uptake of apoptotic neutrophils, with no impact on oleic acid uptake (Figure 10, C–F). Since we saw increased MFGE8 expression in the cecum of HFD-fed mice treated with DSS, we asked if rmMFGE8 could rescue neutrophil uptake in the presence of oleic acid. We treated BMDMs with apoptotic neutrophils, oleic acid, and rmMFGE8 and found that BMDM lipid uptake — but not apoptotic neutrophils — was increased (Figure 10, C–F). The findings suggest that MFGE8 favors uptake of lipids over apoptotic neutrophils, resulting in decreased macrophage uptake of dead cells when lipids are present. We then assessed whether MFGE8 mediated macrophage IL-10 responses to apoptotic neutrophils. We exposed control-treated BMDMs, in the presence or absence of rmMFGE8, to apoptotic neutrophils and assessed IL-10 gene expression. As expected, IL-10 expression was increased in response to apoptotic neutrophils in control-treated BMDMs (Figure 10G). rmMFGE8 further increased Il10 expression in BMDMs in response to apoptotic neutrophils (Figure 10G). In the presence of oleic acid, Il10 expression was impaired in macrophages exposed to apoptotic neutrophils alone or in combination with rmMFGE8 (Figure 10G). These findings demonstrate that macrophage Il10 in response to apoptotic neutrophils is dependent on MFGE8 and αVβ3 and is impaired in the presence of dietary lipids. Taken together, our findings suggest that dietary lipids interfere with MFGE8 binding to apoptotic neutrophils, limiting their uptake with neutrophil accumulation pathological to the tissue. Furthermore, neutrophil uptake is required for induction of the antiinflammatory and prorepair cytokine IL-10. By interfering with MFGE8-mediated macrophage uptake of apoptotic neutrophils, exposure to dietary lipids sustains intestinal injury. ## Discussion In this study, we demonstrate that acute exposure to HFD impairs recovery after intestinal damage, and we identified loss of a key resolving pathway after acute HFD treatment. We found increased weight loss and extended duration of tissue damage in the cecum, but not the distal colon, of mice acutely fed a HFD and exposed to DSS. This corresponds to the region of lipid accumulation and supports a direct effect of dietary lipids in preventing intestinal repair. Previous studies suggest that increased lipid consumption results in lipid deposition in the cecum and colon due to insufficient absorption in the small intestine [59, 60]. Slow movement in the cecum could result in increased accumulation in cecum relative to the colon. The introduction of damage to this region could further prevent transport of lipids to the colon, resulting in increased lipid retention in the cecum, further amplifying defects in damage resolution. In parallel, we also found accumulation of dead neutrophils within the cecum lamina propria and lumen of DSS-treated HFD-fed mice. Depletion of neutrophils in HFD-treated mice rescues from enhanced intestinal pathology. While neutrophils are important for clearance of microbes that penetrate the tissue after injury, excess neutrophils in tissues can also drive inflammation through their production of inflammatory mediators [61]. Due to their short life-span, neutrophils then undergo cell death. If dead neutrophils are not cleared, they can undergo secondary necrosis, further amplifying tissue inflammation [7]. Uptake of dead neutrophils activates tissue repair and antiinflammatory pathways in macrophages, including downregulation of chemokines that recruit neutrophils into tissue [6, 62, 63]. We found that intestinal macrophages from HFD-fed mice had a decreased capacity to uptake dead neutrophils. We show that in vitro treatment of macrophages with oleic and palmitic acids, the primary lipids found in the HFD, limited macrophage uptake of dead neutrophils. Our results support direct interference by excess dietary lipids with a critical intestinal macrophage function necessary to promote the resolution of tissue damage. Continued intestinal neutrophil accumulation is a feature of IBD [5], and our studies suggest that defects in macrophage clearance of apoptotic neutrophils may serve as a contributing factor, especially in the context of HFD feeding. Uptake and clearance of dead cells activate macrophage production of the antiinflammatory cytokine IL-10 [12, 13]. Loss of IL-10 or IL-10 signaling corresponds with early-onset, severe human intestinal disease and with development of colitis in mouse models (64–66). IL-10 supplementation protects against DSS-induced colitis [67]. While IL-10 is normally induced after DSS treatment, we do not find it upregulated in HFD-treated mice. Others have demonstrated that, in obesity, adipose tissue macrophages express decreased levels of IL-10 [18]. Macrophage pretreatment with oleic and palmitic acids limits Il10 expression after exposure to dead neutrophils. Overexpression of IL-10 rescued barrier repair defects and enhanced pathology in acute HFD-treated mice. Others have demonstrated that IL-10 directly supports epithelial proliferation in vitro [8]. Importantly, overexpression of IL-10 does not protect against enhanced pathology after DSS treatment of HFD-fed mice, if epithelial cells lack IL-10Rα. We identified loss of IL-10 signaling downstream of dead neutrophil uptake after acute exposure to HFD. In addition to its role in uptake, MFGE8 also suppresses neutrophil recruitment by downregulating neutrophil Cxcr2 expression [68]. We found elevated expression of Cxcr2 in the cecum of HFD-fed mice after injury. Identifying whether dietary lipids also interfere with MFGE8-mediated downregulation of neutrophil Cxcr2 expression would further provide insight into an additional mechanism by which dietary lipids promote increased inflammation. MFGE8 levels are decreased in patients with ulcerative colitis (UC) compared with healthy controls and are associated with increased disease activity, highlighting the importance of MFGE8 in intestinal disease [69]. However, increased MFGE8 levels in HFD-treated mice exposed to DSS does not improve tissue repair. Our study suggests that alterations in MFGE8 function may contribute to HFD-associated intestinal disease. We identified dietary lipid interference of MFGE8-mediated uptake of apoptotic neutrophils by macrophages as a mechanism by which dietary lipids impair resolution of intestinal injury. The 2 C-terminal discoidin domains of MFGE8 allow it to act as an opsin-like molecule that bind to the externalized PS on dead cells, while its N-terminal EGF domain binds to αvβ3 [24]. Our work suggests that lipids outcompete PS for binding to the MFGE8 discoidin domains, preventing macrophage uptake of apoptotic neutrophils. This effect is specific for apoptotic neutrophils, as we found that lipids did not alter apoptotic thymocyte uptake, which others show depends on Mertk [58]. MFGE8 likely has further roles in intestinal repair, since treatment with recombinant MFGE8 promotes epithelial migration to support wound closure in in vitro models [70]. Lipid interference with MFGE8 binding to damaged epithelium may further contribute to lost repair. Better characterization of MFGE8 effector functions would allow for identification of ways to increase tissue repair and could be important therapeutically in resolving tissue damage in diet-associated diseases, such as atherosclerosis and intestinal disease. We found that oleic acid, an unsaturated lipid, and palmitic acid, a saturated lipid, which both comprise the majority of the HFD, impair macrophage uptake of apoptotic neutrophils and Il10 expression in response to apoptotic neutrophils. Dietary lipids consist of additional unsaturated lipids, including linoleic and palmitoleic acid and saturated lipids like stearic acid. Prior studies demonstrate that various lipid classes can impact macrophage anti- and proinflammatory functions, including cytokine production [71, 72]. Understanding how various lipid classes, individual lipids, and the amounts of these lipids alter macrophage responses to injury, including uptake of apoptotic neutrophils, and intracellular pathways involved in downstream responses to injury, would provide better understanding of how the types of lipids we consume can prevent or predispose an individual to intestinal disease development. In future studies, it will be important to determine the impact of individual lipids on shared downstream pathways of macrophage breakdown of apoptotic neutrophils and dietary lipids on additional macrophage functions. Recent data demonstrate that, in the steady state, intestinal macrophage and DC uptake of dead epithelial cells induces a homeostatic transcriptional program that promotes Tregs [73]. It remains to be determined if the same PS receptors are utilized in this homeostatic pathway and, furthermore, if dietary lipids also interfere with dead cell clearance at steady state. Defects in similar clearance pathways are observed in several inflammatory disorders, and it will be important to understand whether dietary lipids interfere with dead neutrophil clearance in nonintestinal sites. While diets high in fat have long been associated with human disease, these effects have been thought to be secondary to the expansion of adipose tissue, which supports systemic inflammation. Our results indicate that HFD not only increase inflammation after chronic exposure [74], but they also directly interfere with intestinal barrier repair. Unresolved damage leads to continued neutrophil recruitment into the tissue and increased microbial interaction with the epithelium, further potentiating tissue damage and limiting tissue repair. We have identified a mechanism of direct interference by dietary lipids in antiinflammatory and repair processes leading to sustained pathology, which may be of importance to human intestinal disease. Further study of this pathway may lead to new therapeutic targets to attenuate intestinal and other inflammatory diseases. ## Experimental animals Male C57BL/6J (stock no. 000664), CX3CR1-GFP/+ (stock no. 005582), CX3CR1-CreERT2 (stock no. 021160), and Lgr5-EGFP-IRES-creERT2 (stock no. 008875) mice were purchased from The Jackson Laboratory. IL-10 receptor α conditional (IL-10Rαfl/fl) mice were from Werner Muller (University of Manchester, Manchester, England) [75]. Mice were kept and bred under standard specific pathogen–free (SPF) conditions at the Baylor College of Medicine or Memorial Sloan Kettering Cancer Center animal facility. All lines were backcrossed for at least 12 generations to the C57BL/6J background. All mouse experiments were performed with at least 4 mice per group in male mice between 6 and 8 weeks of age. Multiple experiments were combined to assess statistical significance. Littermate controls were used for each experiment, and mice were randomly assigned to experimental groups. ## Acute diet feeding and intestinal injury Mice were fed $10\%$ kcal LFD (Research Diets, D12450B) or $60\%$ kcal HFD (Research Diets, D12492) ad libitum for 1 week prior to treatment with $2\%$ DSS (Thermo Fisher Scientific, AAJ1448922) in drinking water for 5 days, followed by plain drinking water. ## Glucose tolerance test and insulin tolerance test Glucose tolerance test (GTT) and insulin tolerance test (ITT) in mice fed LFD and HFD for 2 and 8 weeks were performed by the Mouse Metabolic Phenotyping Center at Baylor College of Medicine. ## GTT. After a 6-hour overnight fast, 1.5 g/Kg body weight of glucose was given i.p. to each mouse. Blood was collected from tail vein at 0, 15, 30, 60, and 120 minutes, and glucose levels were checked using a glucometer (Life Scan). ## ITT. In total, 1 U/kg body weight insulin (HUMULIN R) was injected i.p. to mice after a 4-hour fast. Blood glucose was measured as above. ## Cecal microbiota transplant Mice were treated with a single dose of 20 mg/mL streptomycin (MilliporeSigma, S9137-25G) by per os (P.O.) with 1 g/1 L ampicillin (Fisher BioReagent, Thermo Fisher Scientific; BP176025) in drinking water for 2 weeks [76]. Mice were switched to autoclaved water for 2 days before transplant of PBS-resuspended cecal content from mice fed HFD or LFD for 7 days. ## Histology Ileum, cecum, proximal colon, and distal colon were fixed in Carnoy’s fixation for 1–2 days before being placed in methanol prior to paraffin embedding. Samples were deparaffinized, cut into 4 μM sections, and stained with hematoxylin or ABPAS. Images were taken with a Nikon Ti Eclipse or Leica microscope. Sections from 4–6 mice were used for blinded colitis scoring, according to established criteria [77, 78]. The number of ABPAS+ goblet cells were counted per 10 villi for 5–9 mice per group. A score of 1 refers to mild mucosal inflammation. A score of 2 refers to inflammatory cell infiltrate in the mucosa and submucosa. A score of 3 consists of inflammatory infiltrates plus focal ulcerations. Focal ulceration with mucosa and submucosa inflammatory infiltrate comprises a score of 4. A score of 5 consists of extensive focal ulceration, in addition to mucosa and submucosa inflammatory infiltrate. A score of 6 is indicative of transmural inflammation and extensive ulcerations. ## IF tissue staining Sections were fixed, embedded, deparaffinized, and cut as described above. Sections were permeabilized, blocked, and stained overnight at 4°C with the following primary antibodies at a 1:100 dilution: anti-Ki67 (polyclonal, Novus Biologicals, catalog NB110-89719), anti-OCLN (EPR20992, Abcam, catalog ab216327), anti-ZO1 (EPR19945-296, Abcam, catalog ab221547), anti-MUC2 (polyclonal, Cloud Clone, catalog PAA705Mu02), anti-Ly6g (1A8, BioXCell, catalog BE0075-1), anti-F$\frac{4}{80}$ (CI:A3-1, Abcam, catalog ab6640), and anti-MFGE8 (18A2-G10, MBL, catalog D199-3). Sections were washed, stained with the following secondary antibodies at a 1:200 dilution at room temperature for 1 hour: anti–rabbit NorthernLights-557 (R&D systems, catalog NL007), anti–rat Alexa Fluor 488 (Cell Signaling, catalog 4416), and anti–rabbit Alexa Fluor 488 (Cell Signaling, catalog 4412). This was followed by staining with DAPI (Sigma-Aldrich, catalog D9542) and mounting using Aqua Mount (Polysciences, catalog 18606-100) anti-fade mounting media before being cover slipped. In Situ Cell Death Detection Kit TMR red (TUNEL) (MilliporeSigma, catalog 12156792910) staining was performed according to manufacturer’s instructions prior to IF staining. FISH staining was performed prior to IF staining as described in ref. 79 using the following UNI519 universal primer-probe sequence: /5Alex594N/GTATTACCGCGGCTGCTG (Integrated DNA Technologies [IDT]). Images were taken on Nikon Ti Eclipse microscope using 20× and 100× objectives, and images were processed using FIJI. ## DNA preparation and 16S (v4) rRNA-Seq Cecal DNA extraction was performed using a Promega Maxwell RSC PureFood GMO and Authentication Kit (AS1600) following manufacturer’s instructions. The library was generated following the protocol from the Earth Microbiome Project (http://press.igsb.anl.gov/earthmicrobiome/protocols-and-standards/16s/). Library quality and size verification was performed using PerkinElmer LabChip GXII instrument with DNA 1K Reagent Kit (CLS760673). Libraries were normalized to 2 nM and pooled using the same volume across all libraries. Pooled libraries were sequenced on the Illumina MiSeq with paired-end 250 using MiSeq Reagent Kit v2, 500-cycles (MS-102-2003). Demultiplexed raw reads were processed to generate an operational taxonomic unit (OTU) table using USEARCH version 11.0.667 [80]. Taxonomic classification of OTU representative sequences was performed using usearch-sintax, an implementation of the SINTAX algorithm, version 16 of the Ribosomal Database Project (RDP) Training Set [81]. The α diversity estimation was performed using the phyloseq R package [82]. DNA preparation, sequencing, and analysis were performed by the JRI Microbiome Core at Weill Cornell Medicine.16S sequencing data are openly available in NCBI under BioProject ID PRJNA904807. ## Ki67 and ABPAS. Three images were taken per mouse per group. For each image, Ki67+ or ABPAS+ cells per 10 crypts were counted and the average used for quantification. ## MUC2 and FISH. Three images from 4 mice per group were used to quantify MUC2 intensity and bacterial encroachment. For MUC2 intensity, ImageJ (NIH) was used to set a threshold and mask for each image, and pixel intensity was measured using ImageJ measuring tool. Bacterial encroachment was measured as the distance between the closets bacteria to the intestinal epithelium using the ImageJ measuring tool. ## Total TUNEL+, total Ly6G+, and total Ly6G+TUNEL+ cells. Four images were taken per mouse per group. Using ImageJ counting tool and the red (TUNEL) and blue fluorescent (DAPI) channels, the total number of cells positive for both DAPI and TUNEL were used to quantify the total number of TUNEL+ cells per image and the average count per mouse was used. Using an overlay of the red (TUNEL) and green channel (Ly6G), the number of double-positive cells were counted based on overlapping green and red fluorescence intensity using the ImageJ counting tool. Cells with only green fluorescence were considered live Ly6G+ neutrophils (Figure 4A). ## MFGE8 intensity. ImageJ was used to set a threshold and mask for each image, and pixel intensity was measured using ImageJ measuring tool. ## Occludin and ZO1. ImageJ was used to set a threshold and to mask an epithelial region for each image, and pixel intensity was measured using ImageJ measuring tool. ## BMDMs BMDMs were differentiated from 8-week-old male and female C57BL/6J mice as previously described [83]. Single-cell suspension of BM cells was cultured for 6 days in $50\%$ DMEM (Corning, 10-017-CV) supplemented with $20\%$ FBS (Corning, 35-10-CV), $30\%$ L cell (ATCC CRL-2648) media, 2 mM glutamine (Hyclone, SH30034.01), 1 mM pyruvate (Thermo Fisher Scientific, 11360-070), 1 unit/mL pen/strep (Hyclone, SV30010), and 55 μM β-mercaptoethanol (Thermo Fisher Scientific, 21985023). Confirmation of macrophage differentiation was assessed by flow cytometry as described below. All assays were performed in DMEM supplemented with $10\%$ FBS, 1 unit/mL pen/strep, and 1 mM HEPES (MilliporeSigma, H0887-100ml). iBMDMs from Jonathan C. Kagan (Harvard Medical School, Boston, Massachusetts, USA) were cultured as described [84]. ## Dead cells Neutrophils were isolated from BM using a density gradient (Histopaque 1077 and 1099) as previously described [85] and incubated for 24 hours at 37°C in DMEM containing $1\%$ FBS. Thymocytes were treated with 1 mm staurosporine (Enzo, ALX-380-014-M001) for 4 hours at 37°C in DMEM containing $10\%$ FBS. Death was assessed by trypan blue staining (Lonza, 17-942E). For phagocytosis assays, dead neutrophils and thymocytes were stained with TAMRA (Thermo Fisher Scientific, C1171) according to manufacturer’s instructions. ## Gene expression RNA from whole cecum; 0.5 inches of the terminal ileum, proximal colon, and distal colon; or sorted macrophages were isolated using Trizol (Invitrogen, 15596018) according to manufacturer’s protocol. cDNA was synthesized using iScript reverse transcription kit (Bio-Rad, 1708841). Real-time quantitative PCR (qPCR) was performed using SYBR Green Supermix (Bio-Rad, 1725124) using a CFX384 Touch real-time PCR machine. Thermocycling program was 95°C for 2 minutes, followed by 40 cycles at 95°C for 15 seconds, 60°C for 30 seconds, and 72°C for 30 seconds. The following primers were used: mIL-10 forward (F): 5′-CCAGCTGGACAACATACTGCT-3′, mIL-10 reverse (R): 5′-AACCCCACAAGAGTTCTTTCAAA-3′; mGAPDH F: 5′-AATGTGTCCGTCGTGGATCT-3′, mGAPDH R: 5′-CATCGAAGGTGGAAGAGTGG-3′; mTNF F: 5′-TGGGAGTAGACAAGGTACAACCC-3′, mTNF R: 5′-CATCTTCTCAAAATTCGAGTGACA-3′; mOccludin F: 5′-TCAGGGAATATCCACCTATCACCTCAG-3′, mOccludin R: 5′-CATCAGCAGCAGCCATGTACTCTTCAC-3′; mZO1 F: 5′-AGGACACCAAAGCATGTGAG-3′, mZO1 R: 5′-GGCATTCCTGCTGGTTACA-3′; mCXCL1 F: 5′-TGAGCTGCGCTGTCAGTGCCT-3′, mCXCL1 R: 5′-AGAAGCCAGCGTTCACCAGA-3′; mCXCL2 F: 5′-GAGCTTGAGTGTGACGCCCCCAGG-3′, mCXCL2 R: 5′-GTTAGCCTTGCCTTTGTTCAGTATC-3′; and mCXCR2 F: 5′-TCT-GGC-ATG-CCC-TCT-ATT-CTG-3′, mCXCR2 R: 5′-AAG-GTA-ACC-TCC-TTC-ACG-TA-3′. Relative expression of target genes was determined using the ΔΔCT method with GAPDH used as an internal control. ## Phagocytosis and lipid uptake assay Sorted intestinal macrophages isolated from LFD- and HFD-fed mice were incubated at a 1:2 ratio with TAMRA-loaded dead neutrophils for 1 hour in FACS tubes and cytospun before immunostaining. BMDMs were plated at 1 × 106 cells in cover glass MakTek dish (MakTek Corporation, P35-1.5-14-C) or 1.5 × 106 per well in 24-well tissue culture plates and incubated at a 1:2 ratio with TAMRA-labeled dead neutrophils, dead thymocytes, or FITC-labeled latex beads according to manufacturer’s instructions (Cayman Chemical, 500290) for 1 hour before RNA isolation or immunostaining. Using the ImageJ counting tool and automated cell counting described below, macrophages (F480, green) (Figure 7) that stained positive for TAMRA (apoptotic neutrophils) were used to quantify phagocytosis per image, and the average count and percentage was used for quantification. IL-10 gene expression was assessed as described above. ## Lipid treatment of BMDMs and BODIPY staining Fatty acids were dissolved in ethanol [17]. BMDMs were treated with 400 μm oleic acid (Nu-Chek Prep, U-46-A) or equivalent amount of solvent (ethanol) alone or with TAMRA-labeled dead neutrophils or thymocytes for 1 hour [17]. To assess lipid uptake, BMDMs were exposed to 1 μm of BODIPY $\frac{493}{503}$ (Invitrogen, D3922) for 30 minutes at room temperature after immunostaining. BODIPY intensity was measured using ImageJ. ## Recombinant MFGE8 and RGD peptide treatments Apoptotic neutrophils or oleic acid were preincubated with 2 μg/mL of rmMFEG8 (R&D, 2805-MF-050) for 1 hour prior to addition to BMDMs. BMDMs were pretreated with 2 μg/mL of RGD peptide (Enzo Life Sciences, BML-P700) for 1 hour prior to exposure to apoptotic neutrophils or oleic acid. ## Automated cell counting Microscopy pictures were processed with Fiji/ImageJ v$\frac{.2.3.0}{1.53}$f [86], and a custom macro was written to measure the fluorescence of single cells in each picture. Briefly, the macro detects single cells, extracts them from the main pictures, and measure the fluorescence signal for each individual channel. We first performed a background correction using the ballpoint background correction. Then, using representative pictures from an oleic acid–treated well, we set fluorescence thresholds for the detection of green (BODIPY), red (TAMRA), blue (DAPI), and magenta (F480) fluorescence (Figure 10). For cell detection, we used the magenta fluorescence channel to detect the cell silhouettes on every microscopy picture. Then we extracted individual silhouettes as rectangular frames from the main picture and removed the regions outside of the magenta area, to only leave cell-related fluorescence information. For each frame, we quantified the cell area (magenta signal), the number and size of cell nuclei (blue signal), red signals, and green signals. Finally, we summarized the overall fluorescence data in R v.4.1.2 (https://www.R-project.org/) and the tidyverse package [87]. ImageJ macro and R script are available online at https://bitbucket.org/the-samuel-lab/mcalester-2022/ ## Neutrophil depletion Mice were injected i.p. with 400 μg of IgG2a isotype control (BioXCell, BE0089) or anti-Ly6g (BioXCell, BE0075-1) at day 4 of DSS treatment and continued every other day until the completion of the experiment. ## Lamina propria cell isolation Isolation of lamina propria cells was performed as previously described [52, 88]. In brief, indicated intestinal tissue was placed in PBS and was cut open, and luminal contents were removed. Intestine were cut in 1 cm sections and then treated with 1 mM DTT (MilliporeSigma, DN25) and 30 mM EDTA (Invitrogen, AM9261), followed by in 30 mM EDTA, both for 10 minutes at 37°C to remove mucus and epithelial cells. Tissue was then digested in 200 U/mL collagenase 8 (Sigma-Aldrich, C-2139) and 150 μg/mL DNase (MilliporeSigma, DN25) in RPMI supplemented with $10\%$ FBS while shaking at 37°C for 1 hour, followed by separation on a $40\%$/$80\%$ Percoll (Cytiva, 17-0891-01) gradient. ## Flow cytometry and FACS Flow cytometry and analysis were performed with an LSR II (BD Biosciences) and FlowJo software (Tree Star Inc.). Dead cells were excluded using the Live/Dead fixable aqua dead cell stain kit (Invitrogen). Macrophage populations were sorted on a FACSAria Cell sorter (BD Biosciences). Pooled samples from 5–7 CX3CR1-GFP/+ mice were used to obtain $$n = 1$$ for a total “n” equaling 4–5 for each group for gene expression and statistical analysis. Sorted macrophages from 1 mouse per group equaling $$n = 4$$–5 per group were treated with TAMRA-labeled dead neutrophils. The following antibodies were used for flow staining and or sorting: MHCII (M$\frac{5}{114.15.2}$, BioLegend, catalog 107620), CD11b (M$\frac{1}{70}$, BioLegend, catalog 101226), CD11c (N418, BioLegend, catalog 117317), Ly6C (AL-21, BD Pharmingen, catalog560525), CD45 (30-F11, BioLegend, catalog103149), and DAPI (Sigma-Aldrich, catalog D9542). We identified each population as follows: monocyte-derived macrophages (CX3CR1hiCD11b+MHCII+Ly6c–Tim4–), monocytes (CX3CR1+CD11b+Ly6C+), and conventional dendritic cells (CD11c+MHCII+CD103+). ## Overexpression of IL-10 Plasmid DNA expression of control and IL-10 (InVivoGen, puno1-mil10) were delivered i.v. at 10 μg DNA/mouse diluted in TransIT-EE Hydrodynamic Delivery solution (Mirus, MIR 5340) at 0.1 mL/g body weight 1 day after start of DSS treatment [89, 90]. ## 4-hydroxy tamoxifen (4OHT) administration 4OHT (MillipporeSigma, 68392-35-8) was resuspended to 20 mg/mL in ETOH with heating to 37°C. 4OHT was diluted in corn oil (MilliporeSigma, 8001-30-7), and mice were injected i.p. with 0.2 mg every 3 days (CX3CR1-CreERT2 and control) starting on day 0 of DSS treatment or on 2 sequential days 1 week before the start of DSS treatment (LGR5-CreERT2 and control). ## Statistics One-way ANOVA with Tukey’s post hoc test or unpaired 2-tailed Student’s t test was performed using a $95\%$ CI. All data are presented as mean ± SEM. All analyses were performed using GraphPad Prism version 8.0. Differences were considered to be significant at P values of less than 0.05. ## Study approval All experiments were performed in accordance with approved protocols by the IACUC at Baylor College of Medicine and Memorial Sloan Kettering Cancer Center. ## Author contributions GED and AAH designed experiments and wrote the manuscript with input from all coauthors. AAH, MK, DFZR, and GED performed, designed, and analyzed the experiments. AMM, LCC, HWS, MCR, WJHW, KN, and JMH performed experiments. RSL supervised 16s sequencing and designed experiments. AAH, AA, and BSS designed the automated counting script. AAH is the primary lead on the project and is listed first as a co–first author. ## 12/20/2022 In-Press Preview ## 02/08/2023 Electronic publication ## References 1. Soehnlein O, Lindbom L. Phagocyte partnership during the onset and resolution of inflammation. Nat Rev Immunol (2010) 10 427-439. DOI: 10.1038/nri2779 2. Grainger JR, Grencis RK. Neutrophils worm their way into macrophage long-term memory. Nat Immunol (2014) 15 902-904. DOI: 10.1038/ni.2990 3. Kim ND, Luster AD. The role of tissue resident cells in neutrophil recruitment. Trends Immunol (2015) 36 547-555. DOI: 10.1016/j.it.2015.07.007 4. Szabady RL, McCormick BA. Control of neutrophil inflammation at mucosal surfaces by secreted epithelial products. Front Immunol (2013) 4. PMID: 23914188 5. Fournier BM, Parkos CA. The role of neutrophils during intestinal inflammation. Mucosal Immunol (2012) 5 354-366. DOI: 10.1038/mi.2012.24 6. Korns D. Modulation of macrophage efferocytosis in inflammation. Front Immunol (2011) 2. PMID: 22566847 7. Greenlee-Wacker MC. Clearance of apoptotic neutrophils and resolution of inflammation. Immunol Rev (2016) 273 357-370. DOI: 10.1111/imr.12453 8. Quiros M. Macrophage-derived IL-10 mediates mucosal repair by epithelial WISP-1 signaling. J Clin Invest (2017) 127 3510-3520. DOI: 10.1172/JCI90229 9. Morhardt TL. IL-10 produced by macrophages regulates epithelial integrity in the small intestine. Sci Rep (2019) 9. DOI: 10.1038/s41598-018-38125-x 10. Moore KW. Interleukin-10 and the interleukin-10 receptor. Ann Rev Immunol (2001) 19 683-765. DOI: 10.1146/annurev.immunol.19.1.683 11. Ouyang W. Regulation and functions of the IL-10 family of cytokines in inflammation and disease. Annu Rev Immunol (2011) 29 71-109. DOI: 10.1146/annurev-immunol-031210-101312 12. Chung EY. Interleukin-10 expression in macrophages during phagocytosis of apoptotic cells is mediated by homeodomain proteins Pbx1 and Prep-1. Immunity (2007) 27 952-964. DOI: 10.1016/j.immuni.2007.11.014 13. Voll RE. Immunosuppressive effects of apoptotic cells. Nature (1997) 390 350-351. DOI: 10.1038/37022 14. Ouyang W, O’Garra A. IL-10 Family cytokines IL-10 and IL-22: from basic science to clinical translation. Immunity (2019) 50 871-891. DOI: 10.1016/j.immuni.2019.03.020 15. Ananthakrishnan AN. Epidemiology and risk factors for IBD. Nat Rev Gastroenterol Hepatol (2015) 12 205-217. DOI: 10.1038/nrgastro.2015.34 16. Kratz M. Metabolic dysfunction drives a mechanistically distinct proinflammatory phenotype in adipose tissue macrophages. Cell Metab (2014) 20 614-625. DOI: 10.1016/j.cmet.2014.08.010 17. Hill AA. Activation of NF-κB drives the enhanced survival of adipose tissue macrophages in an obesogenic environment. Mol Metab (2015) 4 665-677. DOI: 10.1016/j.molmet.2015.07.005 18. Lumeng CN. Obesity induces a phenotypic switch in adipose tissue macrophage polarization. J Clin Invest (2007) 117 175-184. DOI: 10.1172/JCI29881 19. Park C. Obesity modulates intestinal intraepithelial T cell persistence, CD103 and CCR9 expression, and outcome in dextran sulfate sodium-induced colitis. J Immunol (2019) 203 3427-3435. DOI: 10.4049/jimmunol.1900082 20. Lee J-C. Obesogenic diet-induced gut barrier dysfunction and pathobiont expansion aggravate experimental colitis. PLoS One (2017) 12. DOI: 10.1371/journal.pone.0187515 21. Ma X. High-fat diet modulates non-CD1d-restricted natural killer T cells and regulatory T cells in mouse colon and exacerbates experimental colitis. Clin Exp Immunol (2008) 151 130-138. PMID: 17991290 22. Wang X. Interleukin-22 alleviates metabolic disorders and restores mucosal immunity in diabetes. Nature (2015) 514 237-241. PMID: 25119041 23. Savill J. Thrombospondin cooperates with CD36 and the vitronectin receptor in macrophage recognition of neutrophils undergoing apoptosis. J Clin Invest (1992) 90 1513-1522. DOI: 10.1172/JCI116019 24. Hanayama R. Identification of a factor that links apoptotic cells to phagocytes. Nature (2002) 417 182-187. DOI: 10.1038/417182a 25. Khalifeh-Soltani A. Mfge8 promotes obesity by mediating the uptake of dietary fats and serum fatty acids. Nat Med (2014) 20 175-183. DOI: 10.1038/nm.3450 26. Chassaing B. Dextran sulfate sodium (dss)-induced colitis in mice. Curr Protoc Immunol (2014) 104 15.25.1-15.25.14. PMID: 24510619 27. Okayasu I. A novel method in the induction of reliable experimental acute and chronic ulcerative colitis in mice. Gastroenterology (1990) 98 694-702. DOI: 10.1016/0016-5085(90)90290-H 28. Hotamisligil GS. Foundations of immunometabolism and implications for metabolic health and disease. Immunity (2017) 47 406-420. DOI: 10.1016/j.immuni.2017.08.009 29. Luck H. Regulation of obesity-related insulin resistance with gut anti-inflammatory agents. Cell Metab (2015) 21 527-542. DOI: 10.1016/j.cmet.2015.03.001 30. Garidou L. The gut microbiota regulates intestinal CD4 T cells expressing RORγt and controls metabolic disease. Cell Metab (2015) 22 100-112. DOI: 10.1016/j.cmet.2015.06.001 31. Turnbaugh PJ. The effect of diet on the human gut microbiome: a metagenomic analysis in humanized gnotobiotic mice. Sci Transl Med (2009) 1. PMID: 20368178 32. Berry D. Phylotype-level 16S rRNA analysis reveals new bacterial indicators of health state in acute murine colitis. ISME J (2012) 6 2091-2106. DOI: 10.1038/ismej.2012.39 33. Schwab C. Longitudinal study of murine microbiota activity and interactions with the host during acute inflammation and recovery. ISME J (2014) 8 1101-1114. DOI: 10.1038/ismej.2013.223 34. Berry D. Intestinal microbiota signatures associated with inflammation history in mice experiencing recurring colitis. Front Microbiol (2015) 6. PMID: 26697002 35. Palm NW. Immunoglobulin A coating identifies colitogenic bacteria in inflammatory bowel disease. Cell (2014) 158 1000-1010. DOI: 10.1016/j.cell.2014.08.006 36. Dinh DM. Intestinal microbiota, microbial translocation, and systemic inflammation in chronic HIV infection. J Infect Dis (2015) 211 19-27. DOI: 10.1093/infdis/jiu409 37. Loh G, Blaut M. Role of commensal gut bacteria in inflammatory bowel diseases. Gut Microbes (2012) 3 544-555. DOI: 10.4161/gmic.22156 38. Alam MT. Microbial imbalance in inflammatory bowel disease patients at different taxonomic levels. Gut Pathog (2020) 12. DOI: 10.1186/s13099-019-0341-6 39. Barker N. Adult intestinal stem cells: critical drivers of epithelial homeostasis and regeneration. Nat Rev Mol Cell Biol (2014) 15 19-33. PMID: 24326621 40. Beyaz S. High-fat diet enhances stemness and tumorigenicity of intestinal progenitors. Nature (2016) 531 53-58. DOI: 10.1038/nature17173 41. Lee SH. Intestinal permeability regulation by tight junction: implication on inflammatory bowel diseases. Intest Res (2015) 13 11-18. DOI: 10.5217/ir.2015.13.1.11 42. Landy J. Tight junctions in inflammatory bowel diseases and inflammatory bowel disease associated colorectal cancer. World J Gastroenterol (2016) 22 3117-3126. DOI: 10.3748/wjg.v22.i11.3117 43. Woo S-H. Osteopontin protects colonic mucosa from dextran sodium sulfate-induced acute colitis in mice by regulating junctional distribution of occludin. Dig Dis Sci (2019) 64 421-431. DOI: 10.1007/s10620-018-5246-6 44. Grondin JA. Mucins in intestinal mucosal defense and inflammation: learning from clinical and experimental studies. Front Immunol (2020) 11. DOI: 10.3389/fimmu.2020.02054 45. Kruger P. Neutrophils: between host defence, immune modulation, and tissue injury. PLoS Pathog (2015) 11. DOI: 10.1371/journal.ppat.1004651 46. Ravichandran KS. Beginnings of a good apoptotic meal: the find-me and eat-me signaling pathways. Immunity (2011) 35 445-455. DOI: 10.1016/j.immuni.2011.09.004 47. Majno G, Joris I. Apoptosis, oncosis, and necrosis. An overview of cell death. Am J Pathol (1995) 146 3-15. PMID: 7856735 48. Soehnlein O. Neutrophils as protagonists and targets in chronic inflammation. Nat Rev Immunol (2017) 17 248-261. DOI: 10.1038/nri.2017.10 49. Peiser L. Scavenger receptors in innate immunity. Curr Opin Immunol (2002) 14 123-128. DOI: 10.1016/S0952-7915(01)00307-7 50. Wynn TA, Vannella KM. Macrophages in tissue repair, regeneration, and fibrosis. Immunity (2016) 44 450-462. DOI: 10.1016/j.immuni.2016.02.015 51. Blander JM. The many ways tissue phagocytes respond to dying cells. Immunol Rev (2017) 277 158-173. DOI: 10.1111/imr.12537 52. Kim M. Critical role for the microbiota in CX. Immunity (2018) 49 151-163. DOI: 10.1016/j.immuni.2018.05.009 53. Denning TL. Lamina propria macrophages and dendritic cells differentially induce regulatory and interleukin 17-producing T cell responses. Nat Immunol (2007) 8 1086-1094. DOI: 10.1038/ni1511 54. Murai M. Interleukin 10 acts on regulatory T cells to maintain expression of the transcription factor Foxp3 and suppressive function in mice with colitis. Nat Immunol (2009) 10 1178-1184. DOI: 10.1038/ni.1791 55. Bosurgi L. Macrophage function in tissue repair and remodeling requires IL-4 or IL-13 with apoptotic cells. Science (2017) 356 1072-1076. DOI: 10.1126/science.aai8132 56. Chogle A. Milk fat globule-EGF factor 8 is a critical protein for healing of dextran sodium sulfate-induced acute colitis in mice. Mol Med (2011) 17 502-507. PMID: 21308148 57. Zhang Y. Recombinant human MFG-E8 ameliorates colon damage in DSS- and TNBS-induced colitis in mice. Lab Invest (2015) 95 480-490. DOI: 10.1038/labinvest.2015.32 58. Seitz HM. Macrophages and dendritic cells use different Axl/Mertk/Tyro3 receptors in clearance of apoptotic cells. J Immunol (2007) 178 5635-5642. DOI: 10.4049/jimmunol.178.9.5635 59. Petit V. Chronic high-fat diet affects intestinal fat absorption and postprandial triglyceride levels in the mouse. J Lipid Res (2007) 48 278-287. DOI: 10.1194/jlr.M600283-JLR200 60. Snipes RL. Limited fat absorption in the large intestine of mice. A morphological study. Acta Anat (Basel) (1977) 99 435-439. DOI: 10.1159/000144866 61. Bian Z. Regulation of the inflammatory response: enhancing neutrophil infiltration under chronic inflammatory conditions. J Immunol (2012) 188 844-853. DOI: 10.4049/jimmunol.1101736 62. Elliott MR. Efferocytosis signaling in the regulation of macrophage inflammatory responses. J Immunol (2017) 198 1387-1394. DOI: 10.4049/jimmunol.1601520 63. Köröskényi K. Involvement of adenosine A2A receptors in engulfment-dependent apoptotic cell suppression of inflammation. J Immunol (2011) 186 7144-7155. DOI: 10.4049/jimmunol.1002284 64. Franke A. Sequence variants in IL10, ARPC2 and multiple other loci contribute to ulcerative colitis susceptibility. Nat Genet (2008) 40 1319-1323. DOI: 10.1038/ng.221 65. Franke A. Genome-wide meta-analysis increases to 71 the number of confirmed Crohn’s disease susceptibility loci. Nat Genet (2010) 42 1118-1125. DOI: 10.1038/ng.717 66. Kühn R. Interleukin-10-deficient mice develop chronic enterocolitis. Cell (1993) 75 263-274. DOI: 10.1016/0092-8674(93)80068-P 67. Cardoso A. The dynamics of interleukin-10-afforded protection during dextran sulfate sodium-induced colitis. Front Immunol (2018) 9. DOI: 10.3389/fimmu.2018.00400 68. Aziz M. Milk fat globule-epidermal growth factor-factor 8 attenuates neutrophil infiltration in acute lung injury via modulation of CXCR2. J Immunol (2012) 189 393-402. DOI: 10.4049/jimmunol.1200262 69. Zhao Q. Milk fat globule-epidermal growth factor 8 is decreased in intestinal epithelium of ulcerative colitis patients and thereby causes increased apoptosis and impaired wound healing. Mol Med (2012) 18 497-506. DOI: 10.2119/molmed.2011.00369 70. Bu H-F. Milk fat globule-EGF factor 8/lactadherin plays a crucial role in maintenance and repair of murine intestinal epithelium. J Clin Invest (2007) 117 3673-3683. PMID: 18008006 71. Batista-Gonzalez A. New insights on the role of lipid metabolism in the metabolic reprogramming of macrophages. Front Immunol (2020) 10. DOI: 10.3389/fimmu.2019.02993 72. Davanso MR. Impact of dietary fatty acids on macrophage lipid metabolism, signaling and function. Immunometabolism (2020) 2 73. Cummings RJ. Different tissue phagocytes sample apoptotic cells to direct distinct homeostasis programs. Nature (2016) 539 565-569. DOI: 10.1038/nature20138 74. Neurath MF. New targets for mucosal healing and therapy in inflammatory bowel diseases. Mucosal Immunol (2014) 7 6-19. DOI: 10.1038/mi.2013.73 75. Pils MC. Monocytes/macrophages and/or neutrophils are the target of IL-10 in the LPS endotoxemia model. Eur J Immunol (2010) 40 443-448. DOI: 10.1002/eji.200939592 76. Wu W-JH. Interleukin-1β secretion induced by mucosa-associated gut commensal bacteria promotes intestinal barrier repair. Gut Microbes (2022) 14. DOI: 10.1080/19490976.2021.2014772 77. Kim JJ. Investigating intestinal inflammation in DSS-induced model of IBD. J Vis Exp (2012). PMID: 22331082 78. Erben U. A guide to histomorphological evaluation of intestinal inflammation in mouse models. Int J Clin Exp Pathol (2014) 7 4557-4576. PMID: 25197329 79. Langendijk PS. Quantitative fluorescence in situ hybridization of Bifidobacterium spp. with genus-specific 16S rRNA-targeted probes and its application in fecal samples. Appl Environ Microbiol (1995) 61 3069-3075. DOI: 10.1128/aem.61.8.3069-3075.1995 80. Edgar RC. Search and clustering orders of magnitude faster than BLAST. Bioinformatics (2010) 26 2460-2461. DOI: 10.1093/bioinformatics/btq461 81. Cole JR. Ribosomal Database Project: data and tools for high throughput rRNA analysis. Nucleic Acids Res (2014) 42 D633-D642. DOI: 10.1093/nar/gkt1244 82. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One (2013) 8. DOI: 10.1371/journal.pone.0061217 83. Trouplin V. Bone marrow-derived macrophage production. J Vis Exp (2013). PMID: 24300014 84. Evavold CL. The pore-forming protein gasdermin D regulates interleukin-1 secretion from living macrophages. Immunity (2018) 48 35-44. DOI: 10.1016/j.immuni.2017.11.013 85. Swamydas M, Lionakis MS. Isolation, purification and labeling of mouse bone marrow neutrophils for functional studies and adoptive transfer experiments. J Vis Exp (2013). PMID: 23892876 86. Schindelin J. Fiji: an open-source platform for biological-image analysis. Nat Methods (2012) 9 676-682. DOI: 10.1038/nmeth.2019 87. Wickham H. Welcome to the Tidyverse. J Open Source Softw (2019) 4. DOI: 10.21105/joss.01686 88. Diehl GE. Microbiota restricts trafficking of bacteria to mesenteric lymph nodes by CX(3)CR1(hi) cells. Nature (2013) 494 116-120. DOI: 10.1038/nature11809 89. Longman RS. CX. J Exp Med (2014) 211 1571-1583. DOI: 10.1084/jem.20140678 90. Qiu J. The aryl hydrocarbon receptor regulates gut immunity through modulation of innate lymphoid cells. Immunity (2012) 36 92-104. DOI: 10.1016/j.immuni.2011.11.011

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.