protein_name
stringlengths 4
4
| structure_title
stringlengths 3
322
| main_text
stringlengths 0
302k
|
|---|---|---|
3M2H
|
Crystallographic and Single Crystal Spectral Analysis of the Peroxidase Ferryl Intermediate
|
Crystallographic and Single Crystal Spectral Analysis of the
Peroxidase Ferryl Intermediate
Yergalem T. Meharenna, Tzanko Doukovb, Huiying Lia, S. Michael Soltisb,*, and Thomas L.
Poulosa,*
aDepartments of Molecular Biology and Biochemistry, Pharmaceutical Sciences, and Chemistry,
University of California, Irvine, California 92697-3900
bMacromolecular Crystallographic Group, The Stanford Synchrotron Radiation Lightsource,
SLAC, Stanford University, Stanford, California 94025
Abstract
The ferryl (Fe(IV)O) intermediate is important in many heme enzymes and thus the precise nature
of the Fe(IV)-O bond is critical in understanding enzymatic mechanisms. The 1.40 Å crystal
structure of cytochrome c peroxidase Compound I has been solved as a function of x-ray dose
while monitoring the visible spectrum. The Fe-O bond increases linearly from 1.73 Å in the low x-
ray dose structure to 1.90 Å in the high dose structure. The low dose structure correlates well with
a Fe(IV)=O bond while we postulate that the high dose structure is the cryo-trapped Fe(III)-OH
species previously thought to be Fe(IV)-OH.
The ferryl, Fe(IV)O, species is a critically important intermediate in a number of
metalloproteins and especially heme enzymes. The high redox potential enables Fe(IV)O to
serve as a potent oxidant utilized by several heme enzymes including cytochromes P450,
nitric oxide synthase (NOS), cytochrome oxidase, and peroxidases. Since the ferryl
intermediate is quite stable in peroxidases, most of what we know about Fe(IV)O in heme
enzymes derives from studies with peroxidases.
In most heme peroxidases one H2O2 oxidizing equivalent is used to oxidize Fe(III) to
Fe(IV)O and the second is used to oxidize an organic group to give Fe(IV)R.+ (1) and this
activated intermediate is called Compound I. In most heme peroxidases such as horse radish
peroxidase (HRP) R is the porphyrin (2) although in yeast cytochome c peroxidase (CCP) R
is the active site Trp191 (3). A majority of studies find that the Fe(IV)-O bond is short,
somewhat less than 1.7 Å, thus indicating a Fe(IV)=O bond as opposed to a Fe(IV)-OH
bond (4). An empirical formula called Badger’s rule relates the calculated Fe-O bond with
the calculated vibrational frequency (5) and the experimental frequencies and EXAFS bond
distances fit very well to these plots (5) further supporting a Fe(IV)=O double bond.
However, a majority of x-ray crystal structures are distinct outliers giving distances closer to
1.8-1.9 Å (4, 6) with one exception being the HRP Compound I structure (7). These
differences are not trivial since the longer bond predicts that the ferryl species should be
protonated to give Fe(IV)-OH, while the shorter bond gives Fe(IV)=O. The chemistry of
each of these species is quite different (8) and knowing the correct structure is essential if
we are to understand details of heme enzyme mechanisms.
*To whom correspondences should be addressed. T.L.P.: [email protected]; phone (494) 824-7020; FAX, (949) 824-3280.
SUPPORTING INFORMATION AVAILABLE
Experimental details and Tables 1S and 2S . This material is available free of charge at http://pubs.acs.org.
NIH Public Access
Author Manuscript
Biochemistry. Author manuscript; available in PMC 2011 October 27.
Published in final edited form as:
Biochemistry. 2010 April 13; 49(14): 2984–2986. doi:10.1021/bi100238r.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
A serious problem encountered at high intensity synchrotron x-ray sources is rapid reduction
of metal centers, particularly high potential metal centers such as Fe(IV). As a result great
care must be taken to minimize reduction and the redox state should be verified during data
collection (for example with UV/VIS spectroscopy). We recently found that crystals of the
CCP N184R mutant diffract unusually well (9) and thus might provide an opportunity to
obtain a low x-ray dose Compound I structure but at sufficiently high resolution to resolve
the discrepancies between crystal structures and solution studies. Here we present single
crystal spectroscopy together with a composite data collection strategy that has allowed the
Fe-O bond distance to be measured as a function of x-ray dose.
Fig. 1A shows the single crystal spectrum of CCP Compound I as a function of x-ray dose.
Before data collection the spectrum in the 500-700 nm region is identical to the solution
spectrum of Compound I. After extensive x-ray exposure (inset to Fig. 1A) the spectrum
clearly is no longer that of Compound I nor is this similar to the Fe(III) high spin solution
spectrum of CCP. The nature of this species will be discussed further on. Fig. 1B shows the
estimated percentage of Compound I remaining in the crystal as a function of x-ray exposure
as monitored by changes in the visible spectrum. Based on this plot ~90% of Compound I
remains after receiving an estimated x-ray dose of 0.035 MGy (calculations were performed
using RADDOSE (10)) or just ~0.1% of the theoretical radiation damage limit for protein
crystals, ≈30 MGy (11). Therefore, a data collection strategy for obtaining predominantly
Compound I was employed using multiple crystals, none of which received more than 0.035
MGy.
With this maximum dose, we estimate that the resulting “integrated” structure is comprised
of ~90% Compound I. Crystallographic data collection was carried out at 65 K on SSRL
BL9-2 (~4×1011 photons/s at 13.0 KeV). Nearly 100 crystals were mounted and indexed in
an automated fashion. Exposures used for indexing were attenuated by 99% and did not
significantly contribute to reduction of Compound I. For each crystal, data collections were
carried out in 15 separate runs. Run 1 consisted of 5° of data, representing the first 0.035
MGy of x-ray exposure. Then the same 5° of scanning angle were recollected 12 more times
giving runs 2 through 13 with increased x-ray dose. In run 14 a full 120° of data were
collected in order to fully reduce the crystal followed by run 15 which again repeated the
same 5° representing the highest x-ray dose. The same 15-run data collection protocol was
adopted for similarly sized crystals and the scanning angles were chosen to optimize the
completeness of the data. Each composite data set was assembled by merging 5° of data
with identical run numbers from 19 crystals. A total of 15 structures at 1.40 Å resolution
were refined providing a picture of the structural changes associated with increasing x-ray
dose (Table S1).
In Fig. 2A we compare the structures of the low dose (set 1) and the ferric resting state 1.06
Å structure of the N184R mutant (3E2O) (9). In the ferric resting state a water molecule is
positioned ≈ 2.0 Å from the heme iron while in the low dose data set the Fe-O oxygen
distance is 1.73 Å. In both structures a water molecule is within H-bonding distance of the
Fe-linked oxygen. In the ferric state the heme iron is displaced from the porphyrin plane by
0.18 Å toward the proximal His ligand while in Compound I the iron is displaced by 0.07 Å
in the opposite direction toward the distal pocket. Thus the net movement of the iron is ≈
0.25 Å relative to the porphyrin plane owing to the oxidation of the iron from Fe(III) to
Fe(IV). Note that the water molecules in the distal pocket, including the one closest to the
iron, are located in nearly the same position relative to the heme while the His-Fe bond
increases from 2.07 Å to 2.12 Å upon oxidation to Fe(IV). Thus, the short Fe-O bond in the
Compound I structure is due in large part to motion of the iron. As in our previous work on
peroxide treated CCP (12) Arg48 in the distal pocket forms a 2.78 Å H-bond with the iron
linked O atom.
Meharenn et al.
Page 2
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
We next compare the set 1 (low dose, Fig. 2C) and set 15 (high dose, Fig. 2D) structures. At
the 4.0 σ contour level the electron density between the Fe and O atoms is not continuous in
set 15 and the Fe-O bond length has increased from 1.73 Å to 1.90 Å. The local water
structure remains largely unchanged. The changes owing to x-ray induced reduction are
highlighted by examining a Fo(low dose)-Fo(high dose) electron density difference map
contoured at ±5σ (Fig. 2B). This map clearly shows that the iron is positioned quite
differently in each structure and is closer toward the distal pocket in the low dose structure.
In addition the His-Fe bond decreases from 2.12 Å to 2.07 Å upon photo reduction again
due to motion of the iron back into the porphyrin plane. The only other notable feature in the
Fo(low dose)-Fo(high dose) difference map is around the carbonyl O atom of the heme
ligand, His175. This group is slightly less than 0.1 Å closer to Trp191 in the low dose
structure and may reflect a local tightening of the structure around the Trp191 cation radical
that provides additional electrostatic stability. The various heme parameter distances are
provided in Table S2.
The structures of set 1 through set 13 next were used to assess how the Fe-O bond changes
as a function of x-ray dose and the results are shown in Fig. 3. The fit to a simple straight
line equation is remarkably good and extrapolates to zero dose at a Fe-O bond distance of
1.72 Å. Raman data (13) coupled with Badger’s rule (4) gives a Fe-O bond of 1.68 Å.
Therefore, the low dose Compound I crystal structure agrees within 0.04 Å with the Raman
data and the ferryl center in CCP Compound I can best be described as Fe(IV)=O and not
Fe(IV)-OH.
The nature of the ferryl center after extensive x-ray exposure is intriguing: the short Fe-O
bond (1.90 Å) compared to the ≈ 2.0 - 2.3 Å observed in Fe(III) high spin peroxidase
structures and the total lack of similarity between the high dose spectrum (Fig. 1) and the
solution spectrum of Fe(III) CCP shows that the high dose structure is not that of Fe(III)
high spin CCP. The spectrum is similar to that of HRP Fe(II) in both the crystal and solution
except in HRP there is no ligand coordinated to the iron (7). Since we clearly see a ligand
coordinated to the iron in the high dose structure we very likely have trapped either Fe(II)-
OH or Fe(III)-OH. Unfortunately we cannot compare single crystal and solution spectra
since formation of Fe(III)-OH, and presumably Fe(II)-OH, requires an increase in pH and
CCP is not stable above pH 8.0.
Our first goal in this study was to further develop the necessary methods and protocols
required to obtain x-ray structures of high potential intermediates in metalloproteins. This
requires isomorphous crystals that diffract well in order to have sufficient resolution to
obtain the level of accuracy required for estimating subtle bond parameter differences (7).
Coupling data collection with on-line single crystal spectroscopy to monitor the redox state
is also essential. Our second goal was to obtain a very low dose x-ray structure of CCP
Compound I at high resolution in order to reconcile the long standing differences observed
in the Fe(IV)-O bond distance between most available x-ray structures and other biophysical
techniques. The low dose CCP Compound I structure agrees within 0.04 Å of previous
experimental estimates indicating that the ferryl species in Compound I is Fe(IV)=O and not
Fe(IV)-OH. It should be noted that from the perspective of the heme, CCP Compound I is
equivalent to HRP Compound II since both contain Fe(IV) with no porphyrin radical. Thus
it is likely that other crystal structures where the Fe(IV)-O bond in Compound II was
estimated to be 1.8 Å (7, 14) or longer may also have a significant amount of a reduced iron
species.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Meharenn et al.
Page 3
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Acknowledgments
We thank Aina Cohen, John Kovarick and Michael Hollenbeck for their contribution to the design and
implementation of the single-crystal microspectrophotometer. Portions of this research were carried out at the
Stanford Synchrotron Radiation Lightsource, a national user facility operated by Stanford University on behalf of
the U.S. Department of Energy, Office of Basic Energy Sciences. The SSRL Structural Molecular Biology Program
is supported by the Department of Energy, Office of Biological and Environmental Research, and by the National
Institutes of Health, National Center for Research Resources, Biomedical Technology Program, and the National
Institute of General Medical Sciences. Work at UCI was supported by NIH grant GM42614 (TLP).
References
1. Poulos TL, Kraut J. J Biol Chem. 1980; 255:8199–8205. [PubMed: 6251047]
2. Dolphin D, Forman A, Borg DC, Fajer J, Felton RH. Proc Natl Acad Sci USA. 1971; 68:614–618.
[PubMed: 5276770]
3. Sivaraja M, Goodin DB, Smith M, Hoffman BM. Science. 1989; 245:738–740. [PubMed: 2549632]
4. Behan RK, Green MT. J Inorg Biochem. 2006; 100:448–459. [PubMed: 16500711]
5. Green MT. J Am Chem Soc. 2006; 128:1902–1906. [PubMed: 16464091]
6. Hersleth HP, Hsiao YW, Ryde U, Gorbitz CH, Andersson KK. Chem Biodivers. 2008; 5:2067–
2089. [PubMed: 18972498]
7. Berglund GI, Carlsson GH, Smith AT, Szoke H, Henriksen A, Hajdu J. Nature. 2002; 417:463–468.
[PubMed: 12024218]
8. Green MT, Dawson JH, Gray HB. Science. 2004; 304:1653–1656. [PubMed: 15192224]
9. Meharenna YT, Oertel P, Bhaskar B, Poulos TL. Biochemistry. 2008; 47:10324–10332. [PubMed:
18771292]
10. Paithankar KS, Owen RL, Garman EF. J Synchr Radiat. 2009; 16:152–162.
11. Owen RL, Rudino-Pinera E, Garman EF. Proc Natl Acad Sci U S A. 2006; 103:4912–4917.
[PubMed: 16549763]
12. Bonagura CA, Bhaskar B, Shimizu H, Li H, Sundaramoorthy M, McRee D, Goodin DB, Poulos
TL. Biochemistry. 2003; 42:5600–5608. [PubMed: 12741816]
13. Reczek CM, Sitter AJ, Terner J. J Molec Struc. 1989; 214:27–41.
14. Hersleth HP, Dalhus B, H GC, A KK. J Biol Inorg Chem. 2002; 7:299–304. [PubMed: 11935353]
Meharenn et al.
Page 4
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 1.
Single crystal spectra of CCP Compound I as a function of x-ray dose. Prior to x-ray
exposure the spectrum is identical to the solution spectrum of Compound I. The estimated
percentage of Compound I remaining in the crystal as a function of x-ray dose in panel B
was based on the decrease in the absorbance peak at 634 nm.
Meharenn et al.
Page 5
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 2.
A) Superposition of the low dose structure (red) on the Fe(III) structure (cyan). Note that the
iron is displaced below the plane of the heme in the Fe(III) structure and above the plane of
the heme in the low dose structure; B) Fo(low dose)-Fo(high dose) electron density
difference map using phases obtained from the low dose structure. The map is contoured at
-5.0σ (green) and +5.0σ (blue); C and D) 2Fo-Fc electron density maps contoured at 4.0σ for
the dose data set 1 (panel C) and high dose data set 15 (panel D). Oxygen and water
molecules are represented by the small spheres.
Meharenn et al.
Page 6
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 3.
Plot of the Fe-O distance as a function of x-ray dose. Each of the 13 structures was refined
exactly the same way using the same starting structure and two different protocols. In the
first the distances between the Fe and N atoms (4 pyrrole and 1 His closed circles) were
restrained while in the second protocol no restraints were applied (open circles). At no time
were restraints imposed on the Fe-O distance. The estimated error in the Fe-O bond distance
is ≈0.017Å (see Supporting Information).
Meharenn et al.
Page 7
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
|
3M2I
|
Crystallographic and Single Crystal Spectral Analysis of the Peroxidase Ferryl Intermediate
|
Crystallographic and Single Crystal Spectral Analysis of the
Peroxidase Ferryl Intermediate
Yergalem T. Meharenna, Tzanko Doukovb, Huiying Lia, S. Michael Soltisb,*, and Thomas L.
Poulosa,*
aDepartments of Molecular Biology and Biochemistry, Pharmaceutical Sciences, and Chemistry,
University of California, Irvine, California 92697-3900
bMacromolecular Crystallographic Group, The Stanford Synchrotron Radiation Lightsource,
SLAC, Stanford University, Stanford, California 94025
Abstract
The ferryl (Fe(IV)O) intermediate is important in many heme enzymes and thus the precise nature
of the Fe(IV)-O bond is critical in understanding enzymatic mechanisms. The 1.40 Å crystal
structure of cytochrome c peroxidase Compound I has been solved as a function of x-ray dose
while monitoring the visible spectrum. The Fe-O bond increases linearly from 1.73 Å in the low x-
ray dose structure to 1.90 Å in the high dose structure. The low dose structure correlates well with
a Fe(IV)=O bond while we postulate that the high dose structure is the cryo-trapped Fe(III)-OH
species previously thought to be Fe(IV)-OH.
The ferryl, Fe(IV)O, species is a critically important intermediate in a number of
metalloproteins and especially heme enzymes. The high redox potential enables Fe(IV)O to
serve as a potent oxidant utilized by several heme enzymes including cytochromes P450,
nitric oxide synthase (NOS), cytochrome oxidase, and peroxidases. Since the ferryl
intermediate is quite stable in peroxidases, most of what we know about Fe(IV)O in heme
enzymes derives from studies with peroxidases.
In most heme peroxidases one H2O2 oxidizing equivalent is used to oxidize Fe(III) to
Fe(IV)O and the second is used to oxidize an organic group to give Fe(IV)R.+ (1) and this
activated intermediate is called Compound I. In most heme peroxidases such as horse radish
peroxidase (HRP) R is the porphyrin (2) although in yeast cytochome c peroxidase (CCP) R
is the active site Trp191 (3). A majority of studies find that the Fe(IV)-O bond is short,
somewhat less than 1.7 Å, thus indicating a Fe(IV)=O bond as opposed to a Fe(IV)-OH
bond (4). An empirical formula called Badger’s rule relates the calculated Fe-O bond with
the calculated vibrational frequency (5) and the experimental frequencies and EXAFS bond
distances fit very well to these plots (5) further supporting a Fe(IV)=O double bond.
However, a majority of x-ray crystal structures are distinct outliers giving distances closer to
1.8-1.9 Å (4, 6) with one exception being the HRP Compound I structure (7). These
differences are not trivial since the longer bond predicts that the ferryl species should be
protonated to give Fe(IV)-OH, while the shorter bond gives Fe(IV)=O. The chemistry of
each of these species is quite different (8) and knowing the correct structure is essential if
we are to understand details of heme enzyme mechanisms.
*To whom correspondences should be addressed. T.L.P.: [email protected]; phone (494) 824-7020; FAX, (949) 824-3280.
SUPPORTING INFORMATION AVAILABLE
Experimental details and Tables 1S and 2S . This material is available free of charge at http://pubs.acs.org.
NIH Public Access
Author Manuscript
Biochemistry. Author manuscript; available in PMC 2011 October 27.
Published in final edited form as:
Biochemistry. 2010 April 13; 49(14): 2984–2986. doi:10.1021/bi100238r.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
A serious problem encountered at high intensity synchrotron x-ray sources is rapid reduction
of metal centers, particularly high potential metal centers such as Fe(IV). As a result great
care must be taken to minimize reduction and the redox state should be verified during data
collection (for example with UV/VIS spectroscopy). We recently found that crystals of the
CCP N184R mutant diffract unusually well (9) and thus might provide an opportunity to
obtain a low x-ray dose Compound I structure but at sufficiently high resolution to resolve
the discrepancies between crystal structures and solution studies. Here we present single
crystal spectroscopy together with a composite data collection strategy that has allowed the
Fe-O bond distance to be measured as a function of x-ray dose.
Fig. 1A shows the single crystal spectrum of CCP Compound I as a function of x-ray dose.
Before data collection the spectrum in the 500-700 nm region is identical to the solution
spectrum of Compound I. After extensive x-ray exposure (inset to Fig. 1A) the spectrum
clearly is no longer that of Compound I nor is this similar to the Fe(III) high spin solution
spectrum of CCP. The nature of this species will be discussed further on. Fig. 1B shows the
estimated percentage of Compound I remaining in the crystal as a function of x-ray exposure
as monitored by changes in the visible spectrum. Based on this plot ~90% of Compound I
remains after receiving an estimated x-ray dose of 0.035 MGy (calculations were performed
using RADDOSE (10)) or just ~0.1% of the theoretical radiation damage limit for protein
crystals, ≈30 MGy (11). Therefore, a data collection strategy for obtaining predominantly
Compound I was employed using multiple crystals, none of which received more than 0.035
MGy.
With this maximum dose, we estimate that the resulting “integrated” structure is comprised
of ~90% Compound I. Crystallographic data collection was carried out at 65 K on SSRL
BL9-2 (~4×1011 photons/s at 13.0 KeV). Nearly 100 crystals were mounted and indexed in
an automated fashion. Exposures used for indexing were attenuated by 99% and did not
significantly contribute to reduction of Compound I. For each crystal, data collections were
carried out in 15 separate runs. Run 1 consisted of 5° of data, representing the first 0.035
MGy of x-ray exposure. Then the same 5° of scanning angle were recollected 12 more times
giving runs 2 through 13 with increased x-ray dose. In run 14 a full 120° of data were
collected in order to fully reduce the crystal followed by run 15 which again repeated the
same 5° representing the highest x-ray dose. The same 15-run data collection protocol was
adopted for similarly sized crystals and the scanning angles were chosen to optimize the
completeness of the data. Each composite data set was assembled by merging 5° of data
with identical run numbers from 19 crystals. A total of 15 structures at 1.40 Å resolution
were refined providing a picture of the structural changes associated with increasing x-ray
dose (Table S1).
In Fig. 2A we compare the structures of the low dose (set 1) and the ferric resting state 1.06
Å structure of the N184R mutant (3E2O) (9). In the ferric resting state a water molecule is
positioned ≈ 2.0 Å from the heme iron while in the low dose data set the Fe-O oxygen
distance is 1.73 Å. In both structures a water molecule is within H-bonding distance of the
Fe-linked oxygen. In the ferric state the heme iron is displaced from the porphyrin plane by
0.18 Å toward the proximal His ligand while in Compound I the iron is displaced by 0.07 Å
in the opposite direction toward the distal pocket. Thus the net movement of the iron is ≈
0.25 Å relative to the porphyrin plane owing to the oxidation of the iron from Fe(III) to
Fe(IV). Note that the water molecules in the distal pocket, including the one closest to the
iron, are located in nearly the same position relative to the heme while the His-Fe bond
increases from 2.07 Å to 2.12 Å upon oxidation to Fe(IV). Thus, the short Fe-O bond in the
Compound I structure is due in large part to motion of the iron. As in our previous work on
peroxide treated CCP (12) Arg48 in the distal pocket forms a 2.78 Å H-bond with the iron
linked O atom.
Meharenn et al.
Page 2
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
We next compare the set 1 (low dose, Fig. 2C) and set 15 (high dose, Fig. 2D) structures. At
the 4.0 σ contour level the electron density between the Fe and O atoms is not continuous in
set 15 and the Fe-O bond length has increased from 1.73 Å to 1.90 Å. The local water
structure remains largely unchanged. The changes owing to x-ray induced reduction are
highlighted by examining a Fo(low dose)-Fo(high dose) electron density difference map
contoured at ±5σ (Fig. 2B). This map clearly shows that the iron is positioned quite
differently in each structure and is closer toward the distal pocket in the low dose structure.
In addition the His-Fe bond decreases from 2.12 Å to 2.07 Å upon photo reduction again
due to motion of the iron back into the porphyrin plane. The only other notable feature in the
Fo(low dose)-Fo(high dose) difference map is around the carbonyl O atom of the heme
ligand, His175. This group is slightly less than 0.1 Å closer to Trp191 in the low dose
structure and may reflect a local tightening of the structure around the Trp191 cation radical
that provides additional electrostatic stability. The various heme parameter distances are
provided in Table S2.
The structures of set 1 through set 13 next were used to assess how the Fe-O bond changes
as a function of x-ray dose and the results are shown in Fig. 3. The fit to a simple straight
line equation is remarkably good and extrapolates to zero dose at a Fe-O bond distance of
1.72 Å. Raman data (13) coupled with Badger’s rule (4) gives a Fe-O bond of 1.68 Å.
Therefore, the low dose Compound I crystal structure agrees within 0.04 Å with the Raman
data and the ferryl center in CCP Compound I can best be described as Fe(IV)=O and not
Fe(IV)-OH.
The nature of the ferryl center after extensive x-ray exposure is intriguing: the short Fe-O
bond (1.90 Å) compared to the ≈ 2.0 - 2.3 Å observed in Fe(III) high spin peroxidase
structures and the total lack of similarity between the high dose spectrum (Fig. 1) and the
solution spectrum of Fe(III) CCP shows that the high dose structure is not that of Fe(III)
high spin CCP. The spectrum is similar to that of HRP Fe(II) in both the crystal and solution
except in HRP there is no ligand coordinated to the iron (7). Since we clearly see a ligand
coordinated to the iron in the high dose structure we very likely have trapped either Fe(II)-
OH or Fe(III)-OH. Unfortunately we cannot compare single crystal and solution spectra
since formation of Fe(III)-OH, and presumably Fe(II)-OH, requires an increase in pH and
CCP is not stable above pH 8.0.
Our first goal in this study was to further develop the necessary methods and protocols
required to obtain x-ray structures of high potential intermediates in metalloproteins. This
requires isomorphous crystals that diffract well in order to have sufficient resolution to
obtain the level of accuracy required for estimating subtle bond parameter differences (7).
Coupling data collection with on-line single crystal spectroscopy to monitor the redox state
is also essential. Our second goal was to obtain a very low dose x-ray structure of CCP
Compound I at high resolution in order to reconcile the long standing differences observed
in the Fe(IV)-O bond distance between most available x-ray structures and other biophysical
techniques. The low dose CCP Compound I structure agrees within 0.04 Å of previous
experimental estimates indicating that the ferryl species in Compound I is Fe(IV)=O and not
Fe(IV)-OH. It should be noted that from the perspective of the heme, CCP Compound I is
equivalent to HRP Compound II since both contain Fe(IV) with no porphyrin radical. Thus
it is likely that other crystal structures where the Fe(IV)-O bond in Compound II was
estimated to be 1.8 Å (7, 14) or longer may also have a significant amount of a reduced iron
species.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Meharenn et al.
Page 3
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Acknowledgments
We thank Aina Cohen, John Kovarick and Michael Hollenbeck for their contribution to the design and
implementation of the single-crystal microspectrophotometer. Portions of this research were carried out at the
Stanford Synchrotron Radiation Lightsource, a national user facility operated by Stanford University on behalf of
the U.S. Department of Energy, Office of Basic Energy Sciences. The SSRL Structural Molecular Biology Program
is supported by the Department of Energy, Office of Biological and Environmental Research, and by the National
Institutes of Health, National Center for Research Resources, Biomedical Technology Program, and the National
Institute of General Medical Sciences. Work at UCI was supported by NIH grant GM42614 (TLP).
References
1. Poulos TL, Kraut J. J Biol Chem. 1980; 255:8199–8205. [PubMed: 6251047]
2. Dolphin D, Forman A, Borg DC, Fajer J, Felton RH. Proc Natl Acad Sci USA. 1971; 68:614–618.
[PubMed: 5276770]
3. Sivaraja M, Goodin DB, Smith M, Hoffman BM. Science. 1989; 245:738–740. [PubMed: 2549632]
4. Behan RK, Green MT. J Inorg Biochem. 2006; 100:448–459. [PubMed: 16500711]
5. Green MT. J Am Chem Soc. 2006; 128:1902–1906. [PubMed: 16464091]
6. Hersleth HP, Hsiao YW, Ryde U, Gorbitz CH, Andersson KK. Chem Biodivers. 2008; 5:2067–
2089. [PubMed: 18972498]
7. Berglund GI, Carlsson GH, Smith AT, Szoke H, Henriksen A, Hajdu J. Nature. 2002; 417:463–468.
[PubMed: 12024218]
8. Green MT, Dawson JH, Gray HB. Science. 2004; 304:1653–1656. [PubMed: 15192224]
9. Meharenna YT, Oertel P, Bhaskar B, Poulos TL. Biochemistry. 2008; 47:10324–10332. [PubMed:
18771292]
10. Paithankar KS, Owen RL, Garman EF. J Synchr Radiat. 2009; 16:152–162.
11. Owen RL, Rudino-Pinera E, Garman EF. Proc Natl Acad Sci U S A. 2006; 103:4912–4917.
[PubMed: 16549763]
12. Bonagura CA, Bhaskar B, Shimizu H, Li H, Sundaramoorthy M, McRee D, Goodin DB, Poulos
TL. Biochemistry. 2003; 42:5600–5608. [PubMed: 12741816]
13. Reczek CM, Sitter AJ, Terner J. J Molec Struc. 1989; 214:27–41.
14. Hersleth HP, Dalhus B, H GC, A KK. J Biol Inorg Chem. 2002; 7:299–304. [PubMed: 11935353]
Meharenn et al.
Page 4
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 1.
Single crystal spectra of CCP Compound I as a function of x-ray dose. Prior to x-ray
exposure the spectrum is identical to the solution spectrum of Compound I. The estimated
percentage of Compound I remaining in the crystal as a function of x-ray dose in panel B
was based on the decrease in the absorbance peak at 634 nm.
Meharenn et al.
Page 5
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 2.
A) Superposition of the low dose structure (red) on the Fe(III) structure (cyan). Note that the
iron is displaced below the plane of the heme in the Fe(III) structure and above the plane of
the heme in the low dose structure; B) Fo(low dose)-Fo(high dose) electron density
difference map using phases obtained from the low dose structure. The map is contoured at
-5.0σ (green) and +5.0σ (blue); C and D) 2Fo-Fc electron density maps contoured at 4.0σ for
the dose data set 1 (panel C) and high dose data set 15 (panel D). Oxygen and water
molecules are represented by the small spheres.
Meharenn et al.
Page 6
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 3.
Plot of the Fe-O distance as a function of x-ray dose. Each of the 13 structures was refined
exactly the same way using the same starting structure and two different protocols. In the
first the distances between the Fe and N atoms (4 pyrrole and 1 His closed circles) were
restrained while in the second protocol no restraints were applied (open circles). At no time
were restraints imposed on the Fe-O distance. The estimated error in the Fe-O bond distance
is ≈0.017Å (see Supporting Information).
Meharenn et al.
Page 7
Biochemistry. Author manuscript; available in PMC 2011 October 27.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
|
3M2K
|
Crystal Structure of fluorescein-labeled Class A -beta lactamase PenP in complex with cefotaxime
|
RESEARCH ARTICLE
Open Access
Structural studies of the mechanism for
biosensing antibiotics in a fluorescein-
labeled β-lactamase
Wai-Ting Wong, Ho-Wah Au, Hong-Kin Yap, Yun-Chung Leung*, Kwok-Yin Wong* and Yanxiang Zhao*
Abstract
Background: β-lactamase conjugated with environment-sensitive fluorescein molecule to residue 166 on the
Ω-loop near its catalytic site is a highly effective biosensor for β-lactam antibiotics. Yet the molecular mechanism of
such fluorescence-based biosensing is not well understood.
Results: Here we report the crystal structure of a Class A β-lactamase PenP from Bacillus licheniformis 749/C with
fluorescein conjugated at residue 166 after E166C mutation, both in apo form (PenP-E166Cf) and in covalent
complex form with cefotaxime (PenP-E166Cf-cefotaxime), to illustrate its biosensing mechanism. In the apo
structure the fluorescein molecule partially occupies the antibiotic binding site and is highly dynamic. In the PenP-
E166Cf-cefatoxime complex structure the binding and subsequent acylation of cefotaxime to PenP displaces
fluorescein from its original location to avoid steric clash. Such displacement causes the well-folded Ω-loop to
become fully flexible and the conjugated fluorescein molecule to relocate to a more solvent exposed environment,
hence enhancing its fluorescence emission. Furthermore, the fully flexible Ω-loop enables the narrow-spectrum
PenP enzyme to bind cefotaxime in a mode that resembles the extended-spectrum β-lactamase.
Conclusions: Our structural studies indicate the biosensing mechanism of a fluorescein-labelled β-lactamase. Such
findings confirm our previous proposal based on molecular modelling and provide useful information for the
rational design of β-lactamase-based biosensor to detect the wide spectrum of β-lactam antibiotics. The
observation of increased Ω-loop flexibility upon conjugation of fluorophore may have the potential to serve as a
screening tool for novel β-lactamase inhibitors that target the Ω-loop and not the active site.
Background
β-Lactamase is one of the major mechanisms of antibio-
tic resistance in bacteria. Enzymes of this family deacti-
vate β-lactam antibiotics by hydrolyzing the conserved
β-lactam moiety in the antibiotics and rendering them
ineffective to bind to their target proteins, the penicillin-
binding proteins (PBPs), which are essential for bacterial
cell wall synthesis and survival [1,2]. Detailed mechanis-
tic studies of these enzymes over the past decades have
revealed a conserved mechanism of β-lactam hydrolysis
that consists of two steps, the acylation step in which
the β-lactam ring is “opened” and acylated to the side
chain hydroxyl group of Ser70 through nucleophilic
attack to form the enzyme-substrate acyl adduct ES*;
followed by the deacylation step in which the ES* inter-
mediate is hydrolyzed and released as E + P facilitated
by Glu166 (residue numbering according to the most
conserved Class A β-lactamases) [3].
The substrate profile of a β-lactamase in hydrolyzing
diverse β-lactam antibiotics is strongly influenced by a
structural element termed Ω-loop, a short stretch of
residues on the surface of the β-lactamase structure that
forms part of the outer part of the antibiotic binding
site [4-9]. For narrow-spectrum β-lactamases such as
the PenP used in this study and the clinically significant
TEM-1 or SHV-1 enzymes, Ω-loop is tightly packed
onto the enzyme active site through hydrophobic and
electrostatic interactions with residues lining the
* Correspondence: [email protected]; [email protected];
[email protected]
Department of Applied Biology and Chemical Technology, Central
Laboratory of the Institute of Molecular Technology for Drug Discovery and
Synthesis, The Hong Kong Polytechnic University, Hung Hom, Hong Hong,
China
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
© 2011 Wong et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
catalytic site, posing as steric hindrance for binding of
second- or third-generation antibiotics with bulky side
chains attached onto the β-lactam nucleus. Many
mutant strains of TEM- and SHV-like β-lactamases
overcome this inefficiency and broaden their hydrolytic
profile by acquiring mutations in the Ω-loop region to
render this region more flexible to accommodate large-
sized antibiotics [4-10]. Many extended-spectrum
β-lactamases have significantly extended Ω-loop, result-
ing in an enlarged active site that readily binds to and
hydrolyzes almost all antibiotics [11-13].
Exploiting the proximity of Ω-loop to the antibiotic
binding site and its structural flexibility, we have suc-
cessfully converted a β-lactamase PenPC from Bacillus
cereus 569/H into a biosensor for β-lactam antibiotics
by mutating the catalytically critical residue Glu166 on
the Ω-loop to cysteine and conjugating an environment-
sensitive fluorescein molecule to its reactive side chain
thiol group to form PenPC-E166Cf as reported in pre-
vious studies [14-16]. Fluorescein is an environment-
sensitive fluorophore with suppressed fluorescence in a
hydrophobic environment but fluoresces strongly in a
polar aqueous environment [17]. The mutation of
Glu166 to cysteine severely reduces the efficiency of the
deacylation step of β-lactamase catalysis, rendering the
enzyme to stall at the acylation step and form a stable
ES* acyl adduct that enhances the fluorescence emission
of the conjugated fluorescein [16]. We have speculated
that the fluorescein molecule is positioned near the cat-
alytic site so that the binding and subsequent acylation
of β-lactam antibiotics would displace it to a more polar
environment, enhancing its fluorescence intensity [16].
Here we report structural studies of fluorescein-
conjugated PenP β-lactamase from Bacillus licheniformis
749/C to validate our proposed biosensing mechanism.
The structural findings suggest an important role of Ω-
loop in the biosensing process, which will help the
rational design of improved biosensors for β-lactam
detection as well as for novel antibiotics discovery.
Results and Discussion
The biosensing profile of PenP-E166Cf
The biosensing profile of fluorescein conjugated PenP
(PenP-E166Cf) for detecting β-lactam antibiotics have
never been reported before. In our previous study, a
highly similar enzyme, PenPC from Bacillus cereus 569/
H with 58% amino acid sequence identity to PenP, was
successfully engineered into a biosensor using the same
design scheme (PenPC-E166Cf) [15,16]. We chose to
work with PenP in this study for the advantage of its
easy propensity for crystallization, which would enable
structural studies to understand its biosensing mechan-
ism at atomic resolution. PenPC, on the other hand, has
poor thermal stability and is difficult to crystallize.
Because of the high sequence similarity between these
two proteins, as well as the general sequence conserva-
tion among all Class A β-lactamase enzymes we expect
that PenP can serve as a good model system to under-
stand the biosensing mechanism of fluorescein-based
biosensing.
Indeed the biosensing profile of PenP-E166Cf is highly
similar to that of PenPC-E166Cf. The conjugation of
fluorescein to the mutated Cys166 residue through thiol
linkage is highly efficient for PenP. The ESI-MS profile
confirmed that over 90% of PenP was labelled by the
fluorophore and converted to PenP-E166Cf, with little
unlabelled PenP remaining (Figure 1a). The fluorescence
scanning spectrum of PenP-E166Cf shows an increase of
~25% in emitted intensity when the antibiotic cefotaxime
is present at 10 μM concentration (Figure 1b). A variety
of β-lactam antibiotics, including the first-, second- and
third-generation compounds with diverse chemical struc-
tures in addition to the conserved β-lactam core, induce
significant fluorescence enhancement in PenP-E166Cf at
concentration as low as 1 μM (Figure 1c). Lastly the
time-dependent spectra of PenP-E166Cf in the presence
of cefotaxime at different concentrations ranging from
0.01 μM to 10 μM shows that PenP-E166Cf can detect
cefotaxime at concentration as low as 0.01 μM and the
fluorescence response is saturated at 1 μM (Figure 1d).
The structure of PenP-E166Cf in apo form
PenP-E166Cf readily crystallized in the form of clustered
needles. These crystals were tinted in bright yellow col-
our, indicating the presence of fluorescein (data not
shown YZ). To confirm that fluorescein remaining con-
jugated to the protein in the crystal form we harvested
and thoroughly washed these yellow-coloured crystals
and analyzed the dissolved crystals on SDS-PAGE gel
under both visible and UV light. A band corresponding
to PenP (~30.5 kDa) is clearly visible under both condi-
tions, confirming that the crystals are indeed of PenP-
E166Cf (data not shown YZ).
The structure of PenP-E166Cf was solved by molecu-
lar replacement using the known structure of PenP
(PDB ID 4BLM) as search model. Two molecules of
PenP-E166Cf are found in each asymmetric unit. Struc-
ture rebuilding and refinement were done in CCP4 pro-
gram [18]. The overall structure of PenP-E166Cf is
largely identical to that of the wild-type unlabeled PenP.
The RMSD of all 4011 protein atoms between the
labeled and wild-type structures is just ~1.5 Å. For main
chain atoms, the RMSD is only 0.8 Å. Key residues lin-
ing the catalytic site, including Ser70 and mutated
Cys166 are virtually identical between the labeled and
wild-type structures (Figure 2a). In summary the conju-
gation of fluorescein to PenP does not alter its overall
structural folding.
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 2 of 8
The fluorescein molecule was modeled onto the PenP
structure after careful inspection of the fo-fc and 2fo-fc
electron density map. These maps are not of high qual-
ity at regions around the fluorescein conjugation site,
with only pieces of discontinuous density visible at 2.0 s
contour level in the fo-fc map (Figure 2a). We tried our
best to fit fluorescein into these pieces of electron den-
sity, particularly matching the melaimide group to a
piece of electron density near the thiol side chain of
Cys166, as well as matching the xanthene group at the
end of the fluorescein molecule to a large piece of elec-
tron density near the catalytic site (Figure 2a). This
modeled structure is stable after rounds of structural
refinement, showing good electron density for the
Ω-loop residues and the fluorescein molecule at 1.0 s
contour level in the 2fo-fc map, suggesting that our
fitting is reasonable (Figure 2b). However, no electron
density was visible for the benzoic group in the
30053 Da
(a)
(d)
PenP(29608Da)+fluorescein(427Da)+water(18Da)=30053Da
(c)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
505
525
545
565
585
Wavelength (nm)
Relative Fluorescence Intensity
E166Cf only
10-5M cefotaxime
(b)
10-5
10-6
10-7
5x10-8
10-8
E166Cf only
Figure 1 Biosensing of b-lactam antibiotics by fluorescein-labelled PenP. (a) De-convoluted ESI mass spectrum of PenP-E166Cf. The add-up
at the bottom confirms the correct mass of the labelled protein. (b) Fluorescence scanning spectra of PenP-E166Cf in the presence of 10-5M
cefotaxime in 50 mM phosphate buffer (pH 7.0). (c) Change in fluorescence emission of PenP-E166Cf after incubation with different antibiotics
(cefotaxime, ceftriaxone, ceftazidime, cephaloridine, cephalothin, cefoxitin, cefuroxime, penicillin G and ampicillin) at 10-6 M for 100 s. (d) Time-
dependent fluorescence spectra in the presence of different concentrations (1 × 10-8 M - 1 × 10-5 M) of cefotaxime monitored at 515 nm.
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 3 of 8
mid-region of the fluorophore molecule, indicating that
this region is more disordered as compared to other
parts of the fluorophore molecule.
In our PenP-E166Cf structure the fluorescein molecule
partially occupies the outer edge of the antibiotic binding
region and is in close contact with several residues at the
catalytic site. The maleimide moiety near the thiol link-
age site is inserted into the catalytic core, located within
2.5 Å from the side chain of Ser70 on one side and 3.5 Å
away from Ω-loop on the other side. The xanthene group
near the other end of the fluorescein molecule extends
toward the solvent (Figure 2b), loosely packed against
β-strand B3 that forms part of the extended substrate
binding area involved in coordinating antibiotics as
shown in the extended-spectrum class A β-lactamase,
Toho-1, in complex with cefotaxime, cephalothin, and
benzylpenicillin [19]. No specific interactions were
observed between fluorescein and the protein. Total sol-
vent accessible area is 188 Å2, 33% of the total surface
area, indicating that fluorescein is partially packed against
the PenP molecule and not fully solvent exposed. The
fluorescein molecule is highly dynamic, as reflected by
the poor electron density map as well as high average
temperature factor (~72.3). In contrast, the rest of the
structure shows excellent electron density and low aver-
age temperature factor (~23.5) that is typical of the 2.2 Å
data set. The Ω-loop, on which the fluorescein molecule
is conjugated, was little affected by the dynamic fluoro-
phore and adopts the same conformation as that of the
unlabelled PenP (Figure 2b).
The structure of PenP-E166Cf in complex with cefotaxime
We chose to determine the PenP-E166Cf-cefotaxime
structure, using cefotaxime as a representative of the
many β-lactam antibiotics because of its positive fluores-
cence response induced in PenP-E166Cf as well as its
chemical structure that contains functional groups typi-
cal of both second- and third-generation antibiotics.
Cefotaxime was soaked into the PenP-E166Cf crystals
by incubating the crystals in the reservoir solution with
0.01 M cefotaxime added for 20 minutes. The PenP-
E166C structure, without the conjugated fluorescein
molecule, was used as the starting model for structure
determination. After initial rounds of refinement both
the fo-fc and 2fo-fc electron density maps were carefully
inspected for evidence of cefotaxime and fluorescein, as
well as for any structural changes on PenP.
The cefotaxime was clearly visible in fo-fc electron
density map as covalently bonded through its carbonyl
carbon atom C7 to the Og atom of Ser70, which repre-
sents the acylated ES* adduct (Figure 3a). But we could
not identify any electron density in either fo-fc or 2fo-fc
:-loop
fluorescein
Ser70
Cys166
2.5 Å
(b)
Ser70
Cys166
:-loop
(a)
Figure 2 Crystal structure of PenP-E166Cf. (a) The fo-fc omit map of fluorescein-5-maleimide contoured at 2.0 s. (b) The 2fo-fc map of
Phe165 to Asn170 and fluorescein-5-maleimide contoured at 1.0 s. Side chains of Phe165 to Asn170 and Ser70 are shown in cpk cylinder
model. Fluorescein is shown in green cylinder model.
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 4 of 8
map that would be accountable for fluorescein molecule
around the location seen in PenP-E166Cf or anywhere
nearby. Furthermore, the fo-fc map showed strong nega-
tive signal for a large segment of Ω-loop (residues 164
to 174) and the 2fo-fc map showed no electron density
for this region at all, indicating this region became
highly
disordered
upon
acylation
of
cefotaxime
(Figure 3b). Based on these observations we did not
include fluorescein molecule or the disordered region of
Ω-loop in our final refined structure of PenP-E166Cf-
cefotaxime.
The overall structure folding of fluorescein-labeled
and cefotaxime-bound PenP is nearly identical to that of
the wild-type unlabeled PenP and the fluorescein-labeled
PenP-E166Cf. From the calculation result by the CCP4
program, it was found that the B factor of Glu163,
(c)
:-loop
GC1
Toho-1
PenP-E166Cf
(c)
:-loop
GC1
Toho-1
PenP-E166Cf
Ser70
cefotaxime
Ω-loop
Cys166
Ser70
cefotaxime
fluorescein
:-loop
Cys166
Ser70
cefotaxime
fluorescein
:-loop
(a)
(b)
(c)
Figure 3 Crystal structure of PenP-E166Cf-cefotaxime. (a) The fo-fc map of cefotaxime in PenP-E166Cf-cefotaxime complex contoured at 2.0
s. The light blue dash line represents the disordered Arg164 to Pro174 due to the poor electron density. (b) Comparison of PenP-E166Cf-
cefotaxime complex with apo PenP-E166Cf structure. The two structures are superimposed by main chain atoms. Key residues including Cys166,
Ser70 and cefotaxime are also shown in cpk cylinder model. (c) Comparison of binding mode of cefotaxime in PenP-E166Cf with that of Toho-1
and GC-1. PenP-E166Cf, Toho-1 and GC1 are superimposed by aligning on overall main chain atoms. Cefotaxime is in cylinder model colored in
cpk (PenP-E166Cf-cefotaxime), golden (Toho-1) and red (GC1) respectively.
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 5 of 8
Gly175, Glu176 on Ω-loop, which are next to the disor-
dered region, is significantly higher (~65 Å2) than other
parts of the protein (~20 Å2). The refinement statistic
for this set of crystal structure has different values from
that of the apo PenP-E166Cf structure due to the cefo-
taxime and the difference in Ω-loop.
To investigate why the binding and acylation of cefotax-
ime causes the Ω-loop and the conjugated fluorescein
molecule to become highly flexible and structural disor-
dered, we superposed the PenP-E166Cf structure onto the
PenP-E166Cf-cefotaxime complex structure. Fluorescein is
seen as occupying a site that partially overlaps with the
acylated cefotaxime; particularly the benzoic group of
fluorescein molecule is in direct steric clash with the
7-amino substituent of cefotaxime (Figure 3b). Thus the
binding and acylation of cefotaxime to PenP would dis-
place fluorescein from its original position to avoid steric
clash. It is likely that the Ω-loop, in order to accommodate
such displacement, loses its well-folded structure and
becomes highly flexible. As a consequence the fluorescein
molecule conjugated to the flexible Ω-loop becomes fully
exposed to the polar aqueous environment, leading to
enhanced fluorescence. Thus our structural findings con-
firmed our initial proposal of a biosensing mechanism
based on displacement of fluorescein [15,16].
To understand the impact of conjugated fluorescein
molecule on the substrate binding kinetics of PenP we
compared the PenP-E166Cf-cefotaxime structure to two
other β-lactamase structures in complex with cefotaxime,
including the narrow-spectrum Toho-1 and the
extended-spectrum GC1 [19,20]. In Toho-1 structure the
methoxyimino side chain points away from the active site
and is solvent-exposed (Figure 3c). Such an orientation
packs the methoxyimino side chain tightly against the
thiozolyl ring, leading to a distorted configuration of the
cephem nucleus that is catalytically incompetent for dea-
cylation [19]. In GC1 structure the transition analog of
cefotaxime binds to GC1 in a fully extended conforma-
tion, with oxyimino group inserted to active site and
extended away from the thiozolyl ring (Figure 3c). This
conformation is regarded as catalytically competent to
facilitate deacylation because the distortion on the
cephem nucleus is released [20]. Importantly, the binding
mode of cefotaxime in our PenP-E166Cf-cefotaxime
structure closely resembles that of GC1 (Figure 3c), sug-
gesting that with its Ω-loop fully flexible the naturally
narrow-spectrum PenP can accommodate cefotaxime in
a manner that resembles the extended-spectrum GC1.
Conclusions
Our structural studies indicate the molecular mechan-
ism how fluorescein-labeled β-lactamase detects β-lac-
tam antibiotics. The conjugated fluorescein molecule is
located near the catalytic site and partially occupies the
antibiotic binding region. The binding and acylation of
β-lactam antibiotics such as cefotaxime would expel the
fluorescein molecule from its original position and leads
to increased flexibility of the Ω-loop, to which the fluor-
ophore is linked. As a result, the fluorophore is relo-
cated from its original position with partial solvent
exposure to become fully solvent exposed, leading to
enhanced fluorescence emission. These findings confirm
our previous proposal based on structural modeling.
Furthermore the Ω-loop demonstrates the propensity
of becoming highly flexible and unstructured if its tight
packing against the catalytic site is disturbed. Such
increased flexibility enables PenP to bind and acylate
cefotaxime, a naturally poor substrate, in a manner that
resembles the extended-spectrum cefotaxime-resistant
β-lactamases. This finding could be valuable in the
future design of novel antibiotics that resist the binding
or hydrolysis by β-lactamases.
Methods
Protein expression and purification
Two constructs of PenP protein were used for our
experiments, the maltose binding protein (MBP)-fusion
construct for time-dependent fluorescence measure-
ments and the His6-tagged construct for crystallization
and structural studies, as well as scanning fluorescence
spectra. The MBP fusion has been shown not to inter-
fere with fluorescence measurements in our previous
studies (data not shown). The MBP-fusion construct
was cloned into pMAL-c2X vector (NEB). The His6-
tagged PenP enzyme was cloned into a modified pRset-
A vector (Invitrogen) with a TEV protease cleavage site
upstream of the PenP gene. The E166C mutation was
constructed using QuikChange Site-Directed Mutagen-
esis Kits (Strategene).
The MBP-fusion construct was expressed in E. coli
strain BL21 (DE3) at 37°C for overnight after induction
by 300 μM IPTG when A600 reached 0.5-0.7. The har-
vested cells were centrifuged and lysed by sonication.
The supernatant after sonication was passed through
amylose affinity chromatography. The eluted fractions
were pooled and buffer exchanged to 20 mM ammo-
nium bicarbonate. The protein was freeze-dried for sto-
rage afterwards.
The His6-tagged PenP protein was expressed in E. coli
strain BL21 (DE3) at 37°C for overnight after induction
by 200 μM IPTG when A600 reached 0.8-1.2. The har-
vested cells were centrifuged and the supernatant was
passed through Nickel affinity chromatography, followed
by DEAE anion exchange chromatography. The frac-
tions containing the target protein were pooled and con-
centrated by Amicon® Ultra-15 Centrifugal Filter
Devices (Millipore NMWL = 10,000). The His6-tag was
cleaved by adding the TEV protease in 1:20 molar ratio
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 6 of 8
to the concentrated PenP-E166C protein (2 mg/ml). The
mixture was incubated at 30°C for 6 hours and was
further purified by Nickel affinity chromatography to
remove uncleaved protein.
Fluorescein labeling of PenP-E166C to form PenP-E166Cf
A ten-fold molar excess of fluorescein, with concentra-
tion of 20 mM, was dissolved in DMF (Dimethyl forma-
mide) and added to the concentrated PenP-E166C
protein solution drop by drop. The labelling reaction
was allowed to proceed in darkness with stirring for
1 hour, and then dialysed against 50 mM potassium phos-
phate buffer (pH 7.0) at 4°C for several times in order to
remove excess fluorescein. The labelled PenP-E166Cf pro-
tein was concentrated to less than 1 ml and further puri-
fied by Superdex™75 gel filtration column (GE
Healthcare). The running buffer contains 20 mM Tris-
HCl, 50 mM NaCl, pH 7.5. The target fractions were
pooled and concentrated by Amicon Ultra to 25 mg/ml.
The labelling efficiency was confirmed by ESI-MS.
Fluorescence spectra of PenP-E166Cf for antibiotic
detection
Fluorescence profile of PenP-E166Cf alone, as well as in
presence of various β-lactams were measured using Per-
kin-Elmer LS50B spectrofluorimeter. Both scanning
spectra and time-dependent spectra were measured. Dif-
ferent β-lactam antibiotics, including cefotaxime, cef-
triaxone, ceftazidime, cephaloridine, cephalothin,
cefoxitin, cefuroxime, penicillin G, and ampicillin, were
incubated with PenP-E166Cf for 100 s at 1 μM to allow
sufficient acylation of the antibiotic to form ES* adduct.
The product after acylation was subjected to fluores-
cence measurement as previously described [15].
Crystallization, structure determination and refinement
Crystals of PenP-E166Cf were grown by hanging-drop
vapour diffusion method after mixing 1 μl of protein
and 1 μl of reservoir solution containing 25% (w/v) PEG
4000, 0.1 M Hepes pH 7.2, 0.4 M NH4Acetate and 0.2
M K2HPO4. Small crystals in the form of clustered nee-
dles appeared readily. For data collection, single crystals
were obtained after separating them from the clustered
needles. Crystals were harvested and cryoprotected in its
reservoir solution supplemented with 20% ethylene gly-
col for one minute prior to flash freeze and data collec-
tion on the Rigaku MicroMax™-007HF x-ray machine.
For PenP-E166Cf-cefotaxime data set, crystals were
soaked in its growth solution added with 0.01 M of
cefotaxime for 15 minutes and then mounted to the
x-ray machine. Data were integrated and scaled by Crys-
talClear™1.3.5 SP2 (Rigaku Inc.).
The crystals belong to the monoclinic group P21 with
cell parameter: a = 43.43 Å, b = 92.3 Å, c = 66.43 Å and
β = 104°. The PenP-E166Cf crystals diffracted to 2.15 Å
resolution, while the PenP-E166Cf-cefotaxime crystal
diffracted to 2.8 Å. Both structures were determined by
molecular replacement using PenP structure as the
search model (PDB ID 4BLM) [21]. The program
COOT was used for inspection of electron density maps
and model building [22]. There are two molecules per
asymmetric unit. The fluorescein and cefotaxime mole-
cules were built by PRODRG [23] and appended to the
PenP structure for refinement. Structure determination
and refinement of PenP-E166Cf and PenP-E166Cf-
cefotaxime were done using the CCP4 program suite
[18]. A summary of the crystallographic data and refine-
ment statistics are given in Table 1. The coordinates
and structure factors from this study have been
Table 1 X-ray data-collection and structure refinement
statistics.
E166Cf
E166Cf+cefotaxime
PDB code
3M2J
3M2K
Data collection
Space group
P21
P21
Unit cell parameters (Å)
a
43.3
43.5
b
92.3
91.4
c
66.3
66.1
b
104.82
104.52
Resolution range (Å)
52-2.15
(2.24-2.15)
45-2.80
(2.95-2.80)
No. of total reflections
79750
40611
No. of unique reflections
29537
12412
I/s
7.1 (2.7)
6.3 (2.4)
Completeness (%)
97.0 (99.5)
99.8 (99.9)
Rmerge (%)
9.7 (27.1)
11.8 (32.0)
Structure refinement
Resolution (Å)
50.0-2.20
45.0-2.80
Rcryst/Rfree (%)
20.0/23.2
21.2/27.7
r.m.s.d. bonds (Å)/angles (°)
0.018/1.784
0.010/1.672
No. of reflections
Working set
24217
11749
Test set
1291
647
No. of atoms
Protein atoms
4011
3706
Water molecules
254
29
Average B-factor (Å2)
Main chain
24.7
16.96
Ligand molecules
48.4
42.46
Water
32.7
10.7
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 7 of 8
deposited into Protein Data Bank (PDB) under accession
codes 3M2J (PenP-E166Cf apo structure) and 3M2K
(PenP-E166Cf-cefotaxime).
Acknowledgements
This work was supported by the Research Grants Council (PolyU 5463/05 M,
PolyU 5017/06P, PolyU 5641/08 M, and PolyU 5639/09M), the Area of
Excellence Fund of the University Grants Committee (AoE/P-10/01) and the
Research Committee of the Hong Kong Polytechnic University. We thank
Shanghai Synchrotron Radiation Facility (SSRF) for access to beam time. Mr.
C.H. Cheng is acknowledged for technical assistance with in-house x-ray
crystallography facility.
Authors’ contributions
WTW performed experiments, analyzed data and drafted manuscript. HWA
and HKY assisted in experiments. YXZ, KYW and YCL designed project,
analyzed data and drafted manuscript. All authors read and approved the
final manuscript.
Received: 21 September 2010 Accepted: 28 March 2011
Published: 28 March 2011
References
1.
Fisher JF, Mobashery S: Three decades of the class A beta-lactamase acyl-
enzyme. Curr Protein Pept Sci 2009, 10(5):401-407.
2.
Neu HC: The crisis in antibiotic resistance. Science 1992,
257(5073):1064-1073.
3.
Strynadka NC, Adachi H, Jensen SE, Johns K, Sielecki A, Betzel C, Sutoh K,
James MN: Molecular structure of the acyl-enzyme intermediate in beta-
lactam hydrolysis at 1.7 A resolution. Nature 1992, 359(6397):700-705.
4.
Palzkill T, Botstein D: Identification of amino acid substitutions that alter
the substrate specificity of TEM-1 beta-lactamase. J Bacteriol 1992,
174(16):5237-5243.
5.
Palzkill T, Le QQ, Venkatachalam KV, LaRocco M, Ocera H: Evolution of
antibiotic resistance: several different amino acid substitutions in an
active site loop alter the substrate profile of beta-lactamase. Mol
Microbiol 1994, 12(2):217-229.
6.
Petrosino J, Cantu C, Palzkill T: beta-Lactamases: protein evolution in real
time. Trends Microbiol 1998, 6(8):323-327.
7.
Petrosino JF, Palzkill T: Systematic mutagenesis of the active site omega
loop of TEM-1 beta-lactamase. J Bacteriol 1996, 178(7):1821-1828.
8.
Trehan I, Beadle BM, Shoichet BK: Inhibition of AmpC beta-lactamase
through a destabilizing interaction in the active site. Biochemistry 2001,
40(27):7992-7999.
9.
Banerjee S, Pieper U, Kapadia G, Pannell LK, Herzberg O: Role of the
omega-loop in the activity, substrate specificity, and structure of class A
beta-lactamase. Biochemistry 1998, 37(10):3286-3296.
10.
Knox JR: Extended-spectrum and inhibitor-resistant TEM-type beta-
lactamases: mutations, specificity, and three-dimensional structure.
Antimicrob Agents Chemother 1995, 39(12):2593-2601.
11.
Powers RA, Caselli E, Focia PJ, Prati F, Shoichet BK: Structures of
ceftazidime and its transition-state analogue in complex with AmpC
beta-lactamase: implications for resistance mutations and inhibitor
design. Biochemistry 2001, 40(31):9207-9214.
12.
Nukaga M, Kumar S, Nukaga K, Pratt RF, Knox JR: Hydrolysis of third-
generation cephalosporins by class C beta-lactamases. Structures of a
transition state analog of cefotoxamine in wild-type and extended
spectrum enzymes. J Biol Chem 2004, 279(10):9344-9352.
13.
Crichlow GV, Kuzin AP, Nukaga M, Mayama K, Sawai T, Knox JR: Structure
of the extended-spectrum class C beta-lactamase of Enterobacter
cloacae GC1, a natural mutant with a tandem tripeptide insertion.
Biochemistry 1999, 38(32):10256-10261.
14.
Chan PH, Chan KC, Liu HB, Chung WH, Leung YC, Wong KY: Fluorescein-
labeled beta-lactamase mutant for high-throughput screening of
bacterial beta-lactamases against beta-lactam antibiotics. Anal Chem
2005, 77(16):5268-5276.
15.
Chan PH, Liu HB, Chen YW, Chan KC, Tsang CW, Leung YC, Wong KY:
Rational design of a novel fluorescent biosensor for beta-lactam
antibiotics from a class A beta-lactamase. J Am Chem Soc 2004,
126(13):4074-4075.
16.
Chan PH, So PK, Ma DL, Zhao Y, Lai TS, Chung WH, Chan KC, Yiu KF,
Chan HW, Siu FM, et al: Fluorophore-labeled beta-lactamase as a
biosensor for beta-lactam antibiotics: a study of the biosensing process.
J Am Chem Soc 2008, 130(20):6351-6361.
17.
Klonis N, Clayton AH, Voss EW Jr, Sawyer WH: Spectral properties of
fluorescein in solvent-water mixtures: applications as a probe of
hydrogen bonding environments in biological systems. Photochem
Photobiol 1998, 67(5):500-510.
18.
The CCP4 suite: programs for protein crystallography. Acta Crystallogr D
Biol Crystallogr 1994, 50(Pt 5):760-763.
19.
Shimamura T, Ibuka A, Fushinobu S, Wakagi T, Ishiguro M, Ishii Y,
Matsuzawa H: Acyl-intermediate structures of the extended-spectrum
class A beta-lactamase, Toho-1, in complex with cefotaxime,
cephalothin, and benzylpenicillin. J Biol Chem 2002, 277(48):46601-46608.
20.
Crichlow GV, Nukaga M, Doppalapudi VR, Buynak JD, Knox JR: Inhibition of
class C beta-lactamases: structure of a reaction intermediate with a
cephem sulfone. Biochemistry 2001, 40(21):6233-6239.
21.
Knox JR, Moews PC: Beta-lactamase of Bacillus licheniformis 749/C.
Refinement at 2 A resolution and analysis of hydration. J Mol Biol 1991,
220(2):435-455.
22.
Emsley P, Cowtan K: Coot: model-building tools for molecular graphics.
Acta Crystallogr D Biol Crystallogr 2004, 60(Pt 12 Pt 1):2126-2132.
23.
Schuttelkopf AW, van Aalten DM: PRODRG: a tool for high-throughput
crystallography of protein-ligand complexes. Acta Crystallogr D Biol
Crystallogr 2004, 60(Pt 8):1355-1363.
doi:10.1186/1472-6807-11-15
Cite this article as: Wong et al.: Structural studies of the mechanism for
biosensing antibiotics in a fluorescein-labeled β-lactamase. BMC
Structural Biology 2011 11:15.
Submit your next manuscript to BioMed Central
and take full advantage of:
• Convenient online submission
• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution
Submit your manuscript at
www.biomedcentral.com/submit
Wong et al. BMC Structural Biology 2011, 11:15
http://www.biomedcentral.com/1472-6807/11/15
Page 8 of 8
|
3M2L
|
Crystal structure of the M113F mutant of alpha-hemolysin
|
Molecular bases of cyclodextrin adapter interactions
with engineered protein nanopores
Arijit Banerjeea, Ellina Mikhailovaa, Stephen Cheleya, Li-Qun Gub,2, Michelle Montoyac,3, Yasuo Nagaokaa,4,
Eric Gouauxd, and Hagan Bayleya,1
aDepartment of Chemistry, University of Oxford, Oxford, OX1 3TA, United Kingdom; bDepartment of Medical Biochemistry & Genetics, Texas A&M
University System Health Science Center, College Station, TX 77843-1114; cDepartment of Biochemistry and Molecular Biophysics and Howard Hughes
Medical Institute, Columbia University, New York, NY 10032; and dVollum Institute and Howard Hughes Medical Institute, Oregon Health and Science
University, Portland, OR 97239
Edited by Gregory A. Petsko, Brandeis University, Waltham, MA, and approved March 3, 2010 (received for review December 15, 2009)
Engineered protein pores have several potential applications in
biotechnology: as sensor elements in stochastic detection and
ultrarapid DNA sequencing, as nanoreactors to observe single-
molecule chemistry, and in the construction of nano- and micro-
devices. One important class of pores contains molecular adapters,
which provide internal binding sites for small molecules. Mutants
of the α-hemolysin (αHL) pore that bind the adapter β-cyclodextrin
(βCD) ∼104 times more tightly than the wild type have been ob-
tained. We now use single-channel electrical recording, protein en-
gineering including unnatural amino acid mutagenesis, and high-
resolution x-ray crystallography to provide definitive structural in-
formation on these engineered protein nanopores in unparalleled
detail.
alpha-hemolysin ∣single molecule ∣stochastic sensing ∣structure ∣
unnatural amino acid
M
any research groups have used protein engineering to
obtain enzymes and antibodies with new properties suited
for specific tasks (1–6). Fewer groups have taken on the difficult
problem of engineering membrane proteins (7). We have engi-
neered the α-hemolysin protein pore, mindful of several potential
applications in biotechnology, including its ability to act as a de-
tector in stochastic sensing (8) and ultrarapid DNA sequencing
(9), to serve as a nanoreactor for the observation of single-
molecule chemistry (10) and to act as a component for the con-
struction of nano- and microdevices (11).
An important breakthrough in this area, which enabled the sto-
chastic sensing of organic molecules including the detection of
DNA bases in the form of nucleoside monophosphates (12, 13),
was the discovery of internal molecular adapters, a form of non-
covalent protein modification (14). Most useful have been cyclo-
dextrin (CD) adapters, which have until now been used in the
absence of detailed structural information about how they work.
The present paper is a definitive investigation, which provides
such information through the application of a wide variety of
technical approaches: single-channel electrical recording, protein
engineering including unnatural amino acid mutagenesis, and
x-ray crystallography. The studies employing mutagenesis show
that the striking interactions seen in the crystal structures also
occur in individual pores in lipid bilayers.
We reveal that the tight-binding αHL mutants (15) M113N7
and M113F7 bind βCD in different orientations within the hep-
tameric pore. In the case of M113N7, the top (primary hydroxyls)
of the CD ring faces the trans entrance of the pore. In the case of
M113F7, the bottom (secondary hydroxyls) of the CD ring faces
the trans entrance, while the top of the ring is bonded to the pore
through remarkable CH-π interactions. Another tight-binding
mutant, M113V7, can bind the CD in both orientations. These
results illustrate the exquisite level of engineering that can be
achieved with protein nanopores, which is, for example, far be-
yond what is possible with solid-state pores. The work also pro-
vides information valuable for the design of new binding sites
within the lumen of the αHL pore or within other β-barrel pro-
teins. Our results will be of interest to others exploring the inter-
actions of CDs with the αHL pore (16, 17), including groups
involved in computational studies (18, 19). In addition CDs bind
to a variety of other pores, including porins (20, 21) and connex-
ins (22), and are being tested in vivo as blockers of the anthrax
protective antigen pore (23, 24). The CD adapter concept has
also been incorporated into other formats, e.g., with glass nano-
pores (25), and artificial pores based on CDs have been made by
several groups (26–28). Our work is pertinent to these studies.
Results
Kinetics and Thermodynamics of the Interactions of βCD with αHL
Pores Containing Met, Phe and Asn at Position 113. We showed earlier
that position 113 in the αHL pore (Fig. 1A) is critical for the bind-
ing of βCD (14). Subsequently, residue 113, which is Met in the
WT protein, was changed to each of the remaining 19 naturally
occurring amino acids by site-directed mutagenesis (15). We
found that 11 of these mutants, expressed as homoheptamers,
bound βCD with a similar affinity and with similar kinetics to
the WT homoheptamer. Two mutants (P, W) bound βCD about
10 times more strongly than the WT homoheptamer, while six of
them (V, H, Y, D, N, F) bound with high affinity, i.e., with a Kd
value 103 to 104 times lower than the WT.
Remarkably, the side chains of the latter six amino acids bear
little resemblance to one another, and this issue is addressed in the
present paper. We first examined the two amino acids with the
most disparate side chains (Fand N) by making heteromeric pores
containing WT (Met-113), M113F, and M113N subunits. Three
series of heteroheptamers were produced: WT7−nM113Nn,
WT7−nM113Fn, and M113F7−nM113Nn. The heteroheptamers
were separated by SDS-polyacrylamide gel electrophoresis aided
by an oligoaspartate (D8) tail on the first of the two types of sub-
unit (Fig. 1B) (29). All 21 combinations of WT, M113F, and
M113N subunits formed αHL pores that interacted with βCD as
shown by single-channel current recordings, which revealed the
extent of block by βCD (Fig. S1), the association and dissociation
Author contributions: A.B., S.C., E.G., and H.B. designed research; A.B., E.M., S.C., L.-Q.G.,
M.M., and Y.N. performed research; A.B., E.G., and H.B. analyzed data; and A.B. and
H.B. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
1To whom correspondence should be addressed. E-mail: [email protected].
2Present address: Department of Biological Engineering and Dalton Cardiovascular
Research Center, University of Missouri, Columbia, MO 65211.
3Present address: Nature Structural & Molecular Biology, 75 Varick Street, 9th Floor, New
York NY 10013-1917.
4Present address: Department of Biotechnology, Faculty of Engineering, Kansai University,
3-3-35 Yamate-cho, Suita, Osaka 564-8680, Japan.
This article contains supporting information online at www.pnas.org/cgi/content/full/
0914229107/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.0914229107
PNAS ∣May 4, 2010 ∣vol. 107 ∣no. 18 ∣8165–8170
BIOCHEMISTRY
rate constants for βCD (kon and koff), and (from the latter) the
equilibrium dissociation constant for βCD (Kd ¼ koff∕kon) (15).
The kon values for βCD for the 21 combinations of subunits
were all similar at ∼5 × 105 M−1 s−1 (Fig. 1C, Upper). By contrast,
the koff values differed widely, ranging from ∼5 × 10−2 s−1 to
∼103 s−1. For WT7−nM113Nn and WT7−nM113Fn, the koff values
decreased as M113N or M113F subunits were added. In the case
of M113N, there was a steep drop in the value of koff after the
fifth subunit had been incorporated. In the case of M113F,
the decrease in the value of koff occurred less precipitously as the
M113F subunits were added (Fig. 1C, Lower). Intriguingly, with
M113F7−nM113Nn, koff first increased as M113N subunits were
added to M113F7 until n ¼ 4 (M113F3M113N4) and then de-
creased for larger values of n (Fig. 1C, Lower). We recognize that
there is more than one permutation of heteromers containing two
to five mutant subunits (Fig. 1B), but we have ignored this fact
here because no significant differences in the properties of indi-
vidual heteromers were observed. For example, 42 recordings
were made of WT5M113N2, which has three permutations.
Because, kon showed little variation with subunit composition,
the variation in Kd was similar to the variation in koff (Fig. 1C).
While these studies were in progress, the crystal structures of
βCD complexed to M113N7 (Fig. 2B) and M113F7 (Fig. 2C) were
solved (Table S1) (30). High-resolution structures could be
obtained because the CD and the αHL pore have the same C7
symmetry. In the case of M113N7, βCD is bound with the second-
ary hydroxyl face “upward” (Fig. 2B). In each glucose unit of the
βCD, the 2-hydroxyl is hydrogen bonded to the side-chain amide
of an Asn-113 (the residue introduced by mutagenesis) and the
3-hydroxyl is hydrogen bonded to the ϵ-amino group of Lys-147.
In the case of M113F7, two βCDs are bound to the αHL pore
(Fig. 2C). It is the top βCD in the structure that concerns us, be-
cause it is in contact with the Phe-113 residues introduced by mu-
tagenesis. It is immediately apparent that the top βCD in M113F7
is in the opposite orientation to the βCD in M113N7 with each
6-hydroxyl group in a CH-π bonding interaction (31–35) with a
Phe-113 side chain. The opposite orientations of the βCDs in
M113N7 and M113F7 immediately explain why heteromers
formed from similar numbers of M113N and M113F subunits
(e.g., M113N4M113F3) bind βCD weakly (see also Discussion).
Unnatural Amino Acid Mutagenesis. To further explore the range of
noncovalent interactions that are available when βCD binds to
the αHL pore, five unnatural amino acids (Fig. 3A and Fig. S2)
were incorporated at position 113, by using the in vitro nonsense
codon suppression method (36). In particular, we had noted that
M113V7 containing the β-branched Val binds βCD tightly (15),
and therefore we compared cyclopropylglycine (Cpg) and cyclo-
propylalanine (Cpa). We also further examined the means by
which M113F7 binds βCD tightly, by comparing the properties of
4-fluorophenylalanine (f1F), pentafluorophenylalanine (f5F),
and cyclohexylalanine (Cha) at position 113.
The five homomeric pores all produced single-channel cur-
rents with unitary conductance values in the range expected
for properly assembled heptamers (Fig. S3). All five bound βCD
(Fig. 3B, Level 2), either tightly (f1F, Cpg) or weakly (f5F, Cha,
Cpa) as described in detail below. During the long βCD binding
events, additional current spikes were seen (Fig. 3B). Similar
Fig. 1.
Binding of βCD by heteromeric αHL pores formed by WT, M113F and M113N subunits. (A) Crystal structure of WT-αHL (61) showing residue 113 (Met,
yellow). Left panel, side view and right panel, top view. (B) Separation of 35S-labeled αHL heteroheptamers by SDS-polyacrylamide electrophoresis. The
separation of the M113F7-nM113Nn heteromers is shown as detected by autoradiography of a dried gel. The M113F subunits carried a D8 tail. Lane 1,
M113N7; lane 2, M113F7−nM113Nn (the heteromers formed from several preparations made with differing ratios of M113F and M113N subunits were mixed
to give roughly equal amounts of each subunit combination); lane 3, M113F7. A diagram of the eight different combinations of subunits and their permuta-
tions is shown to the right of the autoradiogram. The various permutations are not separated by electrophoresis. (C) Kinetics of the interaction of βCD with
single heteromeric αHL pores as determined by bilayer recording. Values of kon were calculated by using kon ¼ 1∕ðτon½βCDÞ, where τon is the mean interevent
interval. Values of koff were determined by using koff ¼ 1∕τoff, where τoff is the mean dwell time of βCD in the pore. Values of Kd were calculated by using
Kd ¼ koff∕kon. Each point represents the mean s:d: for three or more determinations. Where they cannot be seen, the s.d. values lie within the symbol. Black
squares, WT7−nM113Nn; gray squares, M113F7−nM113Nn; empty squares, M113F7−nWTn.
8166
∣
www.pnas.org/cgi/doi/10.1073/pnas.0914229107
Banerjee et al.
events had been observed previously with certain Met-113 repla-
cement mutants and may represent movement of the βCD at its
binding site (e.g., rotation about axes perpendicular to the C7
axis) (15). The additional current spikes were more prevalent
for M113V7 and M113Cpg7, which may take part in more con-
formationally labile interactions with βCD, compared with say
M113F7 (Fig. S4).
Interactions of βCD with Homoheptamers Bearing Aromatic Residues
at Position 113. To further understand the nature of the binding of
βCD to aromatic side chains, we examined the kinetics of βCD
binding to the homoheptamers containing f1F or f5F at position
113, M113f1F7 and M113f5F7 (Fig. 3C). For both mutants, the
value of kon was very similar to that of WT7, but the values of koff
and therefore Kd for M113f1F7 differed dramatically from WT7
and were close to the values for the tight-binding mutant M113F7
(Table S2A). By contrast, koff and Kd for M113f5F7 were similar
to the values for WT7 (Table S2A).
To determine whether M113f1F7 binds βCD in the same orien-
tation as M113F7 (Fig. 2C), we made heteromers of the M113f1F
subunit with M113N or M113F and examined M113F4M113f1F3
and M113N4M113f1F3. M113F4M113f1F3 binds βCD as strongly
as either M113F7 or M113f1F7, but M113N4M113f1F3 binds
βCD weakly with a similar affinity to WT7 (Fig. 3D and
Table S3). Therefore, it is reasonable to infer that M113F7
and M113f1F7 bind βCD in the same orientation with the 6-
hydroxyl groups of the CD in proximity to the aromatic rings
on the protein.
Cyclohexylalanine (Cha) was used to replace the aromatic side
chains with a roughly isosteric hydrophobic group. Again the va-
lue of kon for βCD was little changed, but koff for M113Cha7 had
an intermediate value of 42 6 s−1. Therefore, M113Cha7 binds
βCD more weakly than M113F7 but distinctly more strongly than
the WT7 pore (Table S2A and Fig. 3C).
Interactions of βCD with Homoheptamers Bearing Hydrophobic Resi-
dues at Position 113. M113V7 binds βCD very strongly, and there-
fore we compared αHL pores with Cpg or Cpa at position 113.
Cpg is roughly isosteric with Val, and like Val has a β-branched
side chain. Gratifyingly, M113Cpg7 has a kon value similar to the
other αHL pores, and koff and Kd values close to those of
M113V7 (Table S2B and Fig. 3E). Cyclopropylalanine (Cpa), with
an additional methylene group compared to Cpg, is roughly
isosteric with Leu, a weak binder, and M113Cpa7 also binds
βCD weakly with kon, koff and Kd values similar to those of
WT7 (Table S2B and Fig. 3E). M113I7 and M113T7, which are
β-branched, are also weak binders, but Ile and Thr are less closely
related to Val than Cpg.
To determine whether M113V7 binds βCD in the same orien-
tation as M113F7 or M113N7 (Fig. 2), we made heteromers of
M113V and the M113N or M113F subunits. M113V3M113F4,
M113V4M113F3, M113V3M113N4, and M113V4M113N3 were
examined in detail. All four heteroheptamers bound βCD more
weakly than M113V7, M113F7 or M113N7 (Fig. 3F and Table S4),
suggesting that Val at position 113 interacts with βCD strongly but
in a different manner to either Phe or Asn. Each heteromer
exhibited a range of Kd values, perhaps reflecting the various pos-
sible permutations of the two different subunits around the cen-
tral axis of the heptamer, although this heterogeneity was not
seen for heteromers made from WT, M113F and M113N (Fig. 1).
Discussion
Soon after we discovered that βCD binds to the WT-αHL pore for
around a millisecond, we found a mutant pore, M113N7, that re-
leases βCD ∼104 times more slowly (14). This prompted us to
examine all 19 mutants in which residue 113 is replaced by a nat-
ural amino acid, with the surprising result that a collection of ami-
no acids with structurally unrelated side chains (V, H, Y, D, N, F)
are tight binders (15). Here, we have examined the nature of the
binding interactions more closely by single-channel electrical re-
cording, protein engineering including unnatural amino acid mu-
tagenesis, and high-resolution x-ray crystallography, and we
provide the first definitive structural information on an engi-
neered protein nanopore.
We find that βCD can bind tightly to the αHL pore in three
different ways depending on the residue at 113, as exemplified
by Asn, Phe, and Val. Because Asn and Phe have quite different
side chains, we first compared the ability of M113N and M113F
subunits to take part in binding the CD. The examination of het-
eromeric proteins containing WT (Met-113), M113N and M113F
subunits showed that the replacement of WT subunits in WT7
with M113N or M113F subunits led to increased affinity for
βCD. The more M113N or M113F subunits that were added, the
tighter binding became. By contrast, when subunits in M113N7
were replaced with M113F subunits, binding became weaker,
reaching a minimum at three to four M113F subunits, and then
increasing in strength with five M113F subunits or more (Fig. 1C).
Parallel structural studies (30) revealed the basis of the “oppos-
ing” effects of the M113N and M113F subunits. βCD binds to
M113N7 in the opposite orientation to that in which it binds
to M113F7. In M113N7, the secondary hydroxyls in the βCD ring
are hydrogen bonded to Lys-147 and Asn-113 (Fig. 2A). By con-
trast, βCD interacts with M113F7 through its primary hydroxyl
face (Fig. 2B).
It seemed likely that M113V7, bound βCD in yet another way,
and this was examined by forming heteromers between M113V
and M113N or M113F. The presence of three or four subunits
of either M113N or M113F greatly decreases the affinity of
the pore for βCD (Fig. 3F), with an average koff of 7.3 × 101 s−1,
indicating that a third binding mode is indeed operating
Fig. 2.
X-ray structures of M113N and
M113F homoheptamers with βCD bound.
(A) Side view of heptameric αHL. βCD binds
in the blue highlighted region. (B) βCD
bound to M113N7 (dotted lines indicate hy-
drogen bonding). The side chains of Lys-147
are in pale brown and the side chains of Asn-
113 in yellow. (C) βCD bound to M113F7
(dotted lines indicate CH-π bonding). The
side chains of Phe-113 are in yellow. The sec-
ond βCD in the M113F7 · ðβCDÞ2 structure is
hydrogen bonded to the top βCD in a head-
to-head arrangement and has no apparent
interactions with the protein. For both (B)
and (C), four β strands were omitted from
the barrel to give a better view.
Banerjee et al.
PNAS
∣
May 4, 2010
∣
vol. 107
∣
no. 18
∣
8167
BIOCHEMISTRY
(Table S4). In summary, the three groups of tight-binding mutants
comprise αHL pores incorporating, at position 113: (i) the hydro-
gen-bonding amino acids N, D (the latter would have to be largely
in the protonated form), and possibly H; (ii) the aromatics F, Y,
f1F, and possibly H, and more weakly W; (iii) the β-branched ami-
no acids V, Cpg. There may be yet other means by which CDs can
bind to the αHL pore. For example, we earlier found that hepta-
6-sulfato-βCD can bind tightly to αHL pores containing the
N139Q mutation (37). Presumably, this CD is bound at a site low-
er down in the β barrel in a fashion that includes hydrogen bond-
ing to the Gln at position 139. While the various mutants
exhibited widely different koff values, the value of kon was almost
invariant and averaged ∼2.3 × 105 M−1 s−1 (Table S2) (15). Ap-
parently, transport to the binding site is rate limiting, through
a route unaffected by mutagenesis.
koff increased precipitously with the addition of WTsubunits to
M113N7 (Fig. 1C). Crystal structures of M113N7 show that resi-
dues
111,
113,
and
147
are
reorganized
by
compari-
son with WT7 and then undergo a more limited rearrangement
when βCD binds (Fig. S5). For example, the side chain of
Lys-147 shifts position to form a bifurcated hydrogen bond with
a 3-hydroxyl group of βCD and the side chain carbonyl of an Asn-
113 (Fig. S6). Therefore, the side chains of residues 111, 113, and
147 might be in a variety of conformations in WT7−nM113Nn het-
eromers and offer less well preorganized binding sites for βCD
than they do in M113N7. Further, the intramolecular hydrogen
bonds of the secondary hydroxyls in βCD (38) must be disrupted
upon binding as both hydroxyls on each glucose ring form hydro-
gen bonds to the mutant subunits (Fig. 2B). Because the hydrogen
bonds that are broken in βCD are arranged in a circle, the break-
age of bonds involving a single glucose (three bonds in all) will be
relatively more disruptive than those involving adjoining glucose
residues or the entire circle. The overall binding cooperativity in
M113N7 could be attributed to enthalpic cooperativity outweigh-
ing entropic penalties to binding (39). Positive cooperativity has
been observed previously in fairly rigid model systems (40).
By contrast with M113N7, there is little movement of side
chains in ðM113FÞ7 by comparison with WT7 and little move-
ment, including Phe-113, upon binding βCD (Fig. S7A). Further,
the crystal structure of M113F7 · βCD suggests that each Phe re-
sidue interacts independently with the βCD through what appear
to be CH-π interactions (Fig. S7B). These interactions are ex-
pected to be weak and not strongly directional and hence offer
less enthalpic cooperativity, as supported by the B-factors (crys-
tallographic temperatures factors) at the primary βCD binding
site, which are between ∼40 and 50. Positive cooperativity is ob-
served, but it is less pronounced than in the case of M113N7
(Table S5). In the case of M113N7, the B-factors of the residues
that bind βCD are in the 20s implying that the βCD is more rigidly
held than it is in M113F7.
The binding of sugars to aromatic residues in proteins can in-
clude CH-π bonding (41) or OH-π bonding or a finely balanced
complement of both (42, 43). However, we have dismissed the
possibility of an OH-π interaction between Phe-113 and the
6-hydroxyl groups of βCD as the distance between the center
of the phenyl rings to the nearest hydroxyl oxygen is higher
(5.2 0.65 Å, n ¼ 7) than that expected for a favorable OH-π
interaction (33). While we propose that βCD binds to Phe-113
through a C-6 CH-π interaction (Fig. S7B), the distances between
the center of the Phe-113 ring and the nearest C-6 of βCD ob-
served in the M113F7 · βCD structure (4.66 0.24 Å, n ¼ 7)
are in the upper range of the expected distance for a strong inter-
action, which is ∼4.5 Å (33). The angle between the normal to the
aromatic rings and the line connecting the C-6 atoms to the aro-
matic midpoint is 8.0 5.6°, which is well within the expected
range (44). The measurements with M113f5F7 argue against a
hydrophobic interaction between Phe residues at position 113
and the βCD ring. In f5F, the hydrophobicity of the phenyl ring
is significantly increased (45) yet M113f5F7 binds βCD weakly,
like WT7 (Fig. 3C and Table S2A).
By contrast with F, f1F, Y and N, homomeric αHL pores with
f5F and W at position 113 bound βCD relatively weakly (Fig. 3C
and Table S2A). In the case of f5F, the powerful electron with-
drawing action of the five fluorine atoms leaves a highly increased
positive charge at the center of the ring (46, 47), mitigating
against a hydrogen-bonding interaction. The electron-rich Trp
Fig. 3.
Properties of pores containing natural and unnatural amino acid sub-
stitutions at position 113. The data were recorded at þ40 mV in 1.0 M NaCl,
10 mM sodium phosphate, pH 7.5. (A) Unnatural amino acids used in this
study: 4-fluorophenylalanine, f1F; pentafluorophenylalanine, f5F; cyclohex-
ylalanine, Cha; cyclopropylglycine, Cpg; cyclopropylalanine, Cpa. (B) Repre-
sentative
current
traces
from
single
homoheptameric
αHL
pores,
containing unnatural amino acids at position 113, in the presence of βCD.
βCD (40 μM final) was added to the trans chamber. Level 1, open pore current;
level 2, pore occupied by βCD. The broken line indicates zero current. (C) In-
teraction of βCD with homomeric αHL pores containing aromatic amino acids
at position 113. Kd values for the interaction between βCD and the αHL pore
were calculated by using Kd ¼ koff∕kon. Each column represents the mean
s:d: for 10 or more determinations: dark gray, natural amino acids; light gray,
unnatural amino acids. Data adapted from Gu and colleagues (15) are
marked (*). (D) Representative current traces from single-channel recordings
of βCD binding to M113F4M113f1F3 and M113N4M113f1F3. βCD (40 μM final)
was added to the trans chamber. The broken line indicates zero current. (E)
Interaction of βCD with homomeric αHL pores containing hydrophobic amino
acids at position 113. Kd values for the interaction between βCD and the αHL
pore were calculated by using Kd ¼ koff∕kon. Each column represents the
mean s:d: for ten or more determinations: dark gray, natural amino acids;
light gray, unnatural amino acids. Data adapted from Gu and colleagues (15)
are marked (*). (F) koff values for βCD from heteroheptamers formed with
M113F and M113V subunits and with M113N and M113V subunits. βCD
(40 μM final) was added to the trans chamber. The kon values for βCD for
all these mutants are similar, at ∼3 × 105 M−1 s−1. Empty square: average
koff values for the mutant (bar is s:d). Filled square: M113V3M113F4; filled
circle: M113V4M113F3; filled upright triangle: M113V3M113N4; filled in-
verted triangle: M113V4M113N3.
8168
∣
www.pnas.org/cgi/doi/10.1073/pnas.0914229107
Banerjee et al.
ring (44, 46, 47) should favor hydrogen bonding, but here we can-
not make a direct comparison with the crystal structure of
M113F7 as the indole ring is far larger than benzene. It is possible
that it cannot become oriented in the same manner and that it is
misaligned for hydrogen bonding.
Our experiments suggest that M113V7 and M113Cpg7 bind
βCD in a third way. In heteromers with M113V, both M113F
and M113N reduce the affinity of the pore for βCD suggesting that
neither the CH-π interaction with Phe-113 nor the hydrogen-
bonding interactions with Asn-113 and Lys-147 are compatible
with binding to Val. Close interactions of Val with glucose rings
have been noted previously (48). Therefore, we propose that the
Val side-chain interacts with the side of the glucose ring. This in-
teraction might occur in one or both orientations of the CD
ring (Fig. 4).
Conclusion
We provide structural information on engineered protein nano-
pores and describe three distinct ways in which βCD can bind
within the lumen of mutant αHL pores in atomic detail. Our re-
sults will be useful in several areas of basic science and biotech-
nology. By using host molecules lodged within the αHL pore,
host-guest interactions can be investigated in fine detail at the
single-molecule level (17, 49). The present work will now permit
us to examine binding events at a known face of a CD. The work
also provides information for designing new binding sites within
the lumen of the αHL pore (37) or within other β barrel proteins
(21, 50) and for using molecular design to devise ways in which to
covalently attach CDs within pores (13, 51). These areas impact
practical applications of nanopore technology including stochas-
tic sensing (8), single-molecule DNA sequencing (9, 12, 13, 52),
the use of nanoreactors for the observation of single-molecule
chemistry (10), and the construction of nano- and microdevices
(11, 53), as well as the design of CDs as therapeutic agents
(23, 24).
Methods
Full details of the experimental procedures can be found in SI Appendix.
Materials
L-Amino acids were obtained as follows: 4-fluorophenylalanine (f1F) (Fluka);
pentafluorophenylalanine (f5F) (PepTech Corp.); cyclopropylglycine (Cpg) (Ty-
ger); cyclopropylalanine (Cpa) (Tyger). 4-N-benzoyl-5′-O-(4,4′-dimethoxytri-
tyl)-2′-deoxycytidine-3′-[(2-cyanoethyl)-(N,N-diisopropyl)]-phosphoramidite
and bis(2-cyanoethyl)-N,N-diisopropylphosphoramidite for the synthesis of
pdCpA were purchased from Glen Research and Toronto Research Chemicals,
respectively.
Preparation
of
NVOC-Protected
Aminoacyl-pdCpA.
NVOC-protected
aminoacyl-pdCpAs were prepared as reported previously by reacting the
dinucleotide pdCpA with N-protected, carboxylic acid-activated, amino
acids (54–56).
Preparation of NVOC-Protected Aminoacyl-tRNA. NVOC-protected aminoacyl-
pdCpAs were ligated enzymatically with a truncated tRNA, prepared by using
methods described elsewhere (57, 58).
Genetic Constructs and Mutagenesis. All new αHL constructs were verified by
DNA sequencing. Details of each construct can be found in SI Appendix.
Synthesis, Assembly, and Purification of Mutant αHL pores. αHL monomers (WT
and mutants) were prepared in vitro by coupled transcription and translation
(IVTT) and assembled into homoheptamers on rabbit red blood cell
membranes followed by purification by SDS–PAGE as described earlier
(59). Heteroheptamers were prepared by mixing the two required DNAs
(one encoding an αHL with a D8 tail) before IVTT and then oligomerizing
the mixed translation products on rabbit red blood cell membranes. Pores
with the desired combinations of subunits were purified by SDS–PAGE (59).
Synthesis, Assembly, and Purification of αHL Mutants Containing Unnatural Ami-
no Acids. αHL polypeptides containing unnatural amino acids were synthe-
sized by IVTT in the presence of rabbit red blood cell membranes. The
plasmid with a stop codon (TAG) at position 113 was used. Deprotected ami-
noacyl-tRNAs (SI Appendix) were added to the IVTT mixtures. For heterohep-
tamers with subunits containing unnatural amino acids in combination with
M113N or M113F, monomers were first made, which were then coassembled
on rabbit red blood cell membranes and subsequently purified by SDS–PAGE.
Single-Channel Current Recordings in Planar Lipid Bilayers. (15, 60) Recordings
were made with 1.0 M NaCl, 10 mM sodium phosphate, pH 7.5, in both cham-
bers, at an applied potential of þ40 mV. Data were recorded at 22 2°C. The
bilayer was formed from 1,2-diphytanoyl-sn-glycero-phosphocholine (Avanti
Polar Lipids). Proteins were added to the cis chamber, and βCD to the trans
chamber. Single-channel currents were recorded with an Axopatch 200B
patch-clamp amplifier (Axon Instruments) and filtered at 2 kHz with a
built-in 4-pole low-pass Bessel Filter. The data were acquired at a sampling
rate of 10 kHz. For mutants that bind βCD strongly, the data were acquired
for at least 30 min and for weak-binding mutants for at least 10 min.
Kinetic Data Analysis. Current amplitude and dwell-time histograms were
made by using ClampFit 9.0. The mean dwell times, τoff, were determined
by fitting the dwell-time histograms to single exponentials. Values of kon
and koff were obtained by using the mean dwell times and mean interevent
intervals, as described previously (15, 60). This analysis assumes a binary in-
teraction, which was supported in all cases examined by the finding of only
one major blockade level and a single exponential distribution of dwell
times (τoff).
Fig. 4.
Molecular model showing the three classes of interaction between
the αHL pore and βCD identified in this work. The model identifies the region
of βCD responsible for each interaction (H atoms interacting with Phe-113 or
Asn-113 and Lys-147: gray). The first class of interaction is with aromatic
residues and involves the seven -CH2OH groups of the βCD. The second class
is typified by the interactions with Asn at position 113, which involve hydro-
gen-bonds to the secondary 2-hydroxyls of the βCD. Structural studies show
that this interaction is supported by hydrogen bonding between Lys-147 and
the secondary 3-hydroxyls of the βCD. Structural studies and experiments
with heteromers suggest that the βCD in M113F7 is in the opposite orienta-
tion to the βCD in M113N7, in support of the model shown here. As the inter-
action with Val is hydrophobic, it is not directional and βCD may not bind at
the same position inside the β barrel as it does in M113F7 or M113N7.
Banerjee et al.
PNAS
∣
May 4, 2010
∣
vol. 107
∣
no. 18
∣
8169
BIOCHEMISTRY
Protein Crystallography. Details can be found in SI Appendix. Protein Data
Bank: The coordinates and structure factors of the described structures have
been deposited with accession codes 3M2L ðM113F7Þ, 3M3R ðM113F7 · βCDÞ,
3M4D ðM113N7Þ, 3M4E ðM113N7 · βCDÞ.
ACKNOWLEDGMENTS. We thank Dennis Dougherty for the plasmid pTHG73.
This work was funded by a Royal Society Wolfson Research Merit Award
(to H.B.), the Medical Research Council (H.B.), the National Institutes of
Health (H.B.), and the Howard Hughes Medical Institute (E.G.).
1. Lu Y, Yeung N, Sieracki N, Marshall NM (2009) Design of functional metalloproteins.
Nature 460:855–862.
2. Gebauer M, Skerra A (2009) Engineered protein scaffolds as next-generation antibody
therapeutics. Curr Opin Chem Biol 13:245–255.
3. Gronwall C, Stahl S (2009) Engineered affinity proteins—Generation and applications.
J Biotechnol 140:254–269.
4. Arnold U (2009) Incorporation of non-natural modules into proteins: Structural
features beyond the genetic code. Biotechnol Lett 31:1129–1139.
5. Tracewell CA, Arnold FH (2009) Directed enzyme evolution: Climbing fitness peaks one
amino acid at a time. Curr Opin Chem Biol 13:3–9.
6. Fruk L, Kuo CH, Torres E, Niemeyer CM (2009) Apoenzyme reconstitution as a chemical
tool for structural enzymology and biotechnology. Angew Chem Int Ed Engl
48:1550–1574.
7. Bayley H, Jayasinghe L (2004) Functional engineered channels and pores. Mol Membr
Biol 21:209–220.
8. Bayley H, Cremer PS (2001) Stochastic sensors inspired by biology. Nature 413:226–230.
9. Branton D, et al. (2008) The potential and challenges of nanopore sequencing. Nature
Biotechnol 26:1146–1153.
10. Bayley H, Luchian T, Shin S-H, Steffensen MB (2008) Single Molecules and Nanotech-
nology, eds R Rigler and H Vogel (Springer, Heidelberg), pp 251–277.
11. Maglia G, et al. (2009) Droplet networks with incorporated protein diodes show
collective properties. Nat Nanotechnol 4:437–440.
12. Astier Y, Braha O, Bayley H (2006) Toward single molecule DNA sequencing: Direct
identification of ribonucleoside and deoxyribonucleoside 5'-monophosphates by
using an engineered protein nanopore equipped with a molecular adapter. J Am
Chem Soc 128:1705–1710.
13. Clarke J, et al. (2009) Continuous base identification for single-molecule nanopore
DNA sequencing. Nature Nanotechnol 4:265–270.
14. Gu L-Q, Braha O, Conlan S, Cheley S, Bayley H (1999) Stochastic sensing of organic
analytes by a pore-forming protein containing a molecular adapter. Nature
398:686–690.
15. Gu L-Q, Cheley S, Bayley H (2001) Prolonged residence time of a noncovalent
molecular adapter, β-cyclodextrin, within the lumen of mutant α-hemolysin pores.
J Gen Physiol 118:481–494.
16. Ervin EN, Kawano R, White RJ, White HS (2008) Simultaneous alternating and direct
current readout of protein ion channel blocking events using glass nanopore
membranes. Anal Chem 80:2069–2076.
17. Gurnev PA, Harries D, Parsegian VA, Bezrukov SM (2009) The dynamic side of the Hof-
meister effect: A single-molecule nanopore study of specific complex formation.
ChemPhysChem 10:1445–1449.
18. Mamonova T, Kurnikova M (2006) Structure and energetics of channel-forming
protein-polysaccharide complexes inferred via computational statistical thermody-
namics. J Phys Chem B 110:25091–25100.
19. Egwolf B, Luo Y, Walters DE, Roux B (2010) Ion selectivity of alpha-hemolysin with
beta-cyclodextrin adapter. II. Multi-ion effects studied with grand canonical monte
carlo/brownian dynamics simulations. J Phys Chem B 114:2901–2909.
20. Orlik F, et al. (2003) CymA of Klebsiella oxytoca outer membrane: Binding of cyclodex-
trins and study of the current noise of the open membrane. Biophys J 85:876–885.
21. Chen M, Khalid S, Sansom MSP, Bayley H (2008) OmpG: Engineering a quiet pore for
biosensing. Proc Natl Acad Sci USA 105:6272–6277.
22. Locke D, Koreen IV, Liu JY, Harris AL (2004) Reversible pore block of connexin channels
by cyclodextrins. J Biol Chem 279:22883–22892.
23. Karginov VA, Nestorovich EM, Moayeri M, Leppla SH, Bezrukov SM (2005) Blocking
anthrax lethal toxin at the protective antigen channel by using structure-inspired drug
design. Proc Natl Acad Sci USA 102:15075–15080.
24. Moayeri M, Robinson TM, Leppla SH, Karginov VA (2008) In vivo efficacy of beta-
cyclodextrin derivatives against anthrax lethal toxin. Antimicrob Agents Chemother
52:2239–2241.
25. Gao C, Ding S, Tan Q, Gu LQ (2009) Method of creating a nanopore-terminated probe
for single-molecule enantiomer discrimination. Anal Chem 81:80–86.
26. Pregel MJ, Jullien L, Lehn J-M (1992) Towards artificial ion channels: Transport of alkali
metal ions across liposomal membranes by “bouquet” molecules. Angew Chem Int
Edit 31:1637–1639.
27. Bacri L, Benkhaled A, Guegan P, Auvray L (2005) Ionic channel behavior of modified
cyclodextrins inserted in lipid membranes. Langmuir 21:5842–5846.
28. Jog PV, Gin MS (2008) A light-gated synthetic ion channel. Org Lett 10:3693–3696.
29. Howorka S, Cheley S, Bayley H (2001) Sequence-specific detection of individual DNA
strands using engineered nanopores. Nat Biotechnol 19:636–639.
30. Montoya M (2004) Insights into Membrane Association and Bioengineering of a Pore-
Forming Toxin: Structural Studies of Staphylococcal α-Hemolysin (Columbia University,
New York).
31. Steiner T (2002) The hydrogen bond in the solid state. Angew Chem Int Ed 41:49–76.
32. Steiner T (2002) Hydrogen bonds from water molecules to aromatic acceptors in very
high-resolution protein crystal structures. Biophys Chem 95:195–201.
33. Steiner T, Koellner G (2001) Hydrogen bonds with pi-acceptors in proteins: Frequencies
and role in stabilizing local 3D structures. J Mol Biol 305:535–557.
34. Brandl M, Weiss MS, Jabs A, Sühnel J, Hilgenfeld R (2001) CH...π-interactions in
proteins. J Mol Biol 307:357–377.
35. Weiss MS, Brandl M, Suhnel J, Pal D, Hilgenfeld R (2001) More hydrogen bonds for the
(structural) biologist. Trends Biochem Sci 26:521–523.
36. Wang L, Xie J, Schultz PG (2006) Expanding the genetic code. Annu Rev Biophys
Biomol Struct 35:225–249.
37. Gu L-Q, Cheley S, Bayley H (2001) Capture of a single molecule in a nanocavity. Science
291:636–640.
38. Saenger W, et al. (1998) Structures of the common cyclodextrins and their larger
analogues—Beyond the doughnut. Chem Rev 98:1787–1802.
39. Hunter CA, Tomas S (2003) Cooperativity, partially bound states, and enthalpy-entropy
compensation. Chem Biol 10:1023–1032.
40. Bisson AP, Hunter CA (1996) Cooperativity in the assembly of zipper complexes. Chem
Commun 1723–1724.
41. del Carmen Fernandez-Alonso M, Canada FJ, Jimenez-Barbero J, Cuevas G (2005)
Molecular recognition of saccharides by proteins. Insights on the origin of the
carbohydrate-aromatic interactions. J Am Chem Soc 127:7379–7386.
42. Jimenez-Barbero J, Asensio JL, Canada FJ, Poveda A (1999) Free and protein-bound
carbohydrate structures. Curr Opin Struct Biol 9:549–555.
43. Stanca-Kaposta EC, et al. (2007) Carbohydrate molecular recognition: A spectroscopic
investigation
of
carbohydrate-aromatic
interactions.
Phys
Chem
Chem
Phys
9:4444–4451.
44. Brandl M, Weiss MS, Jabs A, Suhnel J, Hilgenfeld R (2001) CH...π-interactions in
proteins. J Mol Biol 307:357–377.
45. Woll MG, Hadley EB, Mecozzi S, Gellman SH (2006) Stabilizing and destabilizing effects
of phenylalanine –>F5-phenylalanine mutations on the folding of a small protein.
J Am Chem Soc 128:15932–15933.
46. Mecozzi S, West AP, Jr, Dougherty DA (1996) Cation-pi interactions in aromatics
of biological and medicinal interest: Electrostatic potential surfaces as a useful
qualitative guide. Proc Natl Acad Sci USA 93:10566–10571.
47. Dougherty DA (2008) Cys-loop neuroreceptors: Structure to the rescue?. Chem Rev
108:1642–1653.
48. Hondoh H, et al. (2008) Substrate recognition mechanism of alpha-1,6-glucosidic
linkage hydrolyzing enzyme, dextran glucosidase from Streptococcus mutans.
J Mol Biol 378:913–922.
49. Kang XF, Cheley S, Guan X, Bayley H (2006) Stochastic detection of enantiomers. J Am
Chem Soc 128:10684–10685.
50. Chen M, Li QH, Bayley H (2008) Orientation of the monomeric porin OmpG in planar
lipid bilayers. ChemBioChem 9:3029–3036.
51. Wu H-C, Astier Y, Maglia G, Mikhailova E, Bayley H (2007) Protein nanopores with
covalently attached molecular adapters. J Am Chem Soc 129:16142–16148.
52. Bayley H (2006) Sequencing single molecules of DNA. Curr Opin Chem Biol 10:628–637.
53. Astier Y, Bayley H, Howorka S (2005) Protein components for nanodevices. Curr Opin
Chem Biol 9:576–584.
54. Robertson SA, Noren CJ, Anthony-Cahill SJ, Griffith MC, Schultz PG (1989) The use of
5'-phospho-2 deoxyribocytidylylriboadenosine as a facile route to chemical aminoacy-
lation of tRNA. Nucleic Acids Res 17:9649–9660.
55. Ellman J, Mendel D, Anthony-Cahill S, Noren CJ, Schultz PG (1991) Biosynthetic method
for introducing unnatural amino acids site-specifically into proteins. Method Enzymol
202:301–336.
56. Kearney PC, et al. (1996) Dose-response relations for unnatural amino acids at the
agonist binding site of the nicotinic acetylcholine receptor: Tests with novel side chains
and with several agonists. Mol Pharmacol 50:1401–1412.
57. England TE, Bruce AG, Uhlenbeck OC (1980) Specific labeling of 3′ termini of RNA with
T4 RNA ligase. Method Enzymol 65:65–74.
58. Nowak MW, et al. (1998) In vivo incorporation of unnatural amino acids into ion
channels in xenopus oocyte expression system. Method Enzymol 293:504–529.
59. Cheley S, Braha O, Lu X, Conlan S, Bayley H (1999) A functional protein pore with a
“retro” transmembrane domain. Protein Sci 8:1257–1267.
60. Gu L-Q, et al. (2000) Reversal of charge selectivity in transmembrane protein pores by
using non-covalent molecular adapters. Proc Natl Acad Sci USA 97:3959–3964.
61. Song L, et al. (1996) Structure of staphylococcal α-hemolysin, a heptameric transmem-
brane pore. Science 274:1859–1865.
8170
∣
www.pnas.org/cgi/doi/10.1073/pnas.0914229107
Banerjee et al.
|
3M2R
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry using Coenzyme B Analogues
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry
using Coenzyme B Analogues,†,‡
Peder E. Cedervall§,#, Mishtu Dey||,#,⊥, Arwen R. Pearson§,⊥, Stephen W. Ragsdale||, and
Carrie M. Wilmot*,||,§
§ Department of Biochemistry, Molecular Biology and Biophysics, University of Minnesota,
Minneapolis, Minnesota 55455
|| Department of Biological Chemistry, University of Michigan, Ann Arbor, Michigan 48109
Abstract
Methyl-coenzyme M reductase (MCR) catalyzes the final and rate-limiting step in methane
biogenesis; the reduction of methyl-coenzyme M (methyl-SCoM) by coenzyme B (CoBSH) to
methane and a heterodisulfide (CoBS-SCoM). Crystallographic studies show that the active site is
deeply buried within the enzyme, and contains a highly reduced nickel-tetrapyrrole, coenzyme
F430. Methyl-SCoM must enter the active site prior to CoBSH, as species derived from analogues
of methyl-SCoM are always observed bound to the F430 nickel in the deepest part of the 30 Å long
substrate channel that leads from the protein surface to the active site. The seven-carbon
mercaptoalkanoyl chain of CoBSH binds within a 16 Å predominantly hydrophobic part of the
channel close to F430, with the CoBSH thiolate lying closest to the nickel at a distance of 8.8 Å. It
has previously been suggested that binding of CoBSH initiates catalysis by inducing a
conformational change that moves methyl-SCoM closer to the nickel promoting cleavage of the C-
S bond of methyl-SCoM. In order to better understand the structural role of CoBSH early in the
MCR mechanism, we have determined crystal structures of MCR in complex with four different
CoBSH analogues; pentanoyl-, hexanoyl-, octanoyl- and nonanoyl- derivatives of CoBSH
(CoB5SH, CoB6SH, CoB8SH and CoB9SH respectively). The data presented here reveal that the
shorter CoB5SH mercaptoalkanoyl chain overlays with that of CoBSH, but terminates two units
short of the CoBSH thiolate position. In contrast, the mercaptoalkanoyl chain of CoB6SH adopts a
different conformation, such that its thiolate is coincident with the position of the CoBSH thiolate.
†This work was supported by a Department of Energy grant DE-FG02-08ER15931 to S.W.R and supplement to C.M.W. and a
Minnesota Partnership for Biotechnology and Medical Genomics grant SPAP-05-0013-P-FY06.
‡Co-ordinates and structure factors have been deposited in the Protein Data Bank as entries 3m1v (MCRHSCoM); 3m2r
(MCRCoB5SH); 3m2u (MCRCoB6SH); 3m2v (MCRCoB8SH); 3m30 (MCRCoB9SH).
*Address correspondence to: Carrie M. Wilmot, Tel: (612) 624-2406, [email protected].
⊥Current addresses: Mishtu Dey, Department of Chemistry, Massachusetts Institute of Technology, 77 Massachusetts Ave.,
Cambridge, Massachusetts 02139; Arwen Pearson, Astbury Centre for Structural Molecular Biology, Institute for Molecular and
Cellular Biology, Astbury Building. Leeds, LS2 9JT, U.K.
#These authors contributed equally to this work.
Supporting Informational material is available free of charge via the Internet at http://pubs.acs.org and contains the following:
MATERIAL AND METHODS for single crystal UV-visible microspectrophotometry, X-ray photoreduction experiment and X-ray
crystallography of the MCR-heterodisulfide product complex (MCRCoBSH + methyl-SCoM); RESULTS AND DISCUSSION for
redox changes and MCRCoBSH + methyl-SCoM crystal structure; Table S1, X-ray Data Collection, Processing and Refinement
Statistics for MCRCoBSH + methyl-SCoM; Figure S1, use of Fo-Fc electron density in modelling MCRCoB8SH; Figure S2,
illustration of electron density quality of MCRCoB5SH; Figure S3, solution and single crystal UV-visible spectra; Figure S4,
modeling of CoB6SH and CoBSH into the electron density of MCRCoB6SH; Figure S5, partially occupied HSCoM in MCRCoB6SH;
Figure S6, the active site and substrate channel of MCRCoBSH + methyl-SCoM; Figure S7, alternative conformation of Valα482 in
MCRHSCoM; Figure S8, the two conformations of the gly-rich loop in MCRHSCoM; Figure S9, propagation of conformational
changes in MCRHSCoM; Figure S10, the two conformations of CoB9SH in MCRCoB9SH; Figure S11, EPR spectra of MCRred1
sample; Scheme S1, scheme of the characterized forms of MCR.
NIH Public Access
Author Manuscript
Biochemistry. Author manuscript; available in PMC 2011 September 7.
Published in final edited form as:
Biochemistry. 2010 September 7; 49(35): 7683–7693. doi:10.1021/bi100458d.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
This is consistent with the observation that CoB6SH is a slow substrate. A labile water in the
substrate channel was found to be a sensitive indicator for the presence of CoBSH and HSCoM.
The longer CoB8SH and CoB9SH analogues can be accommodated in the active site through
exclusion of this water. These analogues react with Ni(III)-methyl; a proposed MCR catalytic
intermediate of methanogenesis. The CoB8SH thiolate is 2.6 Å closer to the nickel than that of
CoBSH, but the additional carbon of CoB9SH only decreases the nickel thiolate distance a further
0.3 Å. Although the analogues did not induce any structural changes in the substrate channel, the
thiolates appeared to preferentially bind at two distinct positions in the channel; one being the
previously observed CoBSH thiolate position, and the other being at a hydrophobic annulus of
residues that lines the channel proximal to the nickel.
INTRODUCTION
Methanogenic archaea are organisms that under strict anaerobic conditions derive energy by
reducing compounds such as carbon dioxide, methylamine, acetate and methanol, to
methane (1, 2). The global production of methane by these organisms is estimated at one
billion tons annually. Microbially produced methane is not only a potential source of
renewable energy but also a potent greenhouse gas, and as such study of this process has
environmental ramifications. In these microorganisms, methyl-coenzyme M reductase
(MCR)1 is the enzyme that catalyzes the final step in methanogenesis, in which the
substrates methyl-coenzyme M (methyl-SCoM, 2-(methylthio)ethanesulfonate) and
coenzyme B (CoBSH, N-7-mercaptoheptanoylthreonine phosphate) are converted to
methane and a heterodisulfide (CoBS-SCoM) (Scheme 1) (3).
MCR is a 300 kDa protein with six subunits arranged in a α2β2γ2 oligomer (4). The known
crystal structures show that MCR has two active sites approximately 50 Å apart that are
deeply buried within the enzyme (5). The active site pocket is comprised of residues from
subunits α, α′, β and γ, with a 30 Å long substrate channel leading to the enzyme surface
(Figure 1). At the heart of the active site pocket is coenzyme F430, which is a highly reduced
nickel-containing tetrapyrrole (6–8). Currently sixteen distinct enzymatic and complexed
states of MCR have been spectroscopically characterized (Supporting Information, Scheme
S1) (9–31). In the resting active state of the enzyme, denoted MCRred1, the redox-active
nickel of F430 is present in the Ni(I) state (9, 16, 32). MCR is extremely oxygen sensitive
and upon oxygen exposure the enzyme enters an inactive Ni(II) state, denoted MCRred1-silent
(6). In this state it cannot be converted back to the active Ni(I) form by any known reducing
agent making this a challenging system to study. Additional complications involve the tight
association of coenzymes to purified MCR that are not easily displaced as demonstrated by
X-ray crystallographic and kinetic studies (5, 33–35).
Despite the fact that MCR has been studied for decades, no true catalytic intermediate has
been observed, and the actual mechanism remains elusive. Currently three general
mechanistic schemes for the enzymatic reaction have been proposed, each of which posit
different chemistry to initiate catalysis. Mechanism I involves Ni(I) acting as a nucleophile
in an SN2-type reaction that generates Ni(III)-methyl as an intermediate (Scheme 2A) (35–
38). Mechanism II starts with methyl-SCoM undergoing homolytic cleavage at the Ni(I) to
generate a methyl radical and a Ni(II)-SCoM species (Scheme 2B) (39–41). A more recently
proposed mechanism III suggests protonation of coenzyme F430 promotes reductive
cleavage of the methyl-SCoM thioether bond (42).
1Abbreviations: MCR, Methyl-coenzyme M reductase; methyl-SCoM, methyl-coenzyme M; CoBSH, coenzyme B; HSCoM,
coenzyme M; CoBS-SCoM, heterodisulfide of coenzyme B and coenzyme M; APS, Advanced Photon Source; ASU, asymmetric unit;
BPS, bromopropanesulfonate.
Cedervall et al.
Page 2
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Due to the stringent requirement to exclude O2, the available MCR crystal structures are all
in the inactive Ni(II) state. All but one contain CoBSH and HSCoM (demethylated methyl-
SCoM, an inhibitor and substrate analogue) in the active site (PDB codes 1hbn, 1hbo, 1hbu,
1e6y, 1e6v) (5, 33, 34). Another crystal structure has bound heterodisulfide product, CoBS-
SCoM (MCRsilent, PDB code 1hbm, Scheme 1 and Supporting Information, Scheme S1) (5,
33). All these structures reveal that both substrates access the active site through the same
channel (Figure 1). The binding site of HSCoM (and presumably methyl-SCoM) is more
deeply buried within the enzyme, and so it must enter prior to CoBSH for productive
chemistry to occur. As binding of CoBSH in the absence of co-substrate would be
inhibitory, it was suggested that a conformational change upon methyl-SCoM binding might
lower the Kd for CoBSH, and thus promote an ordered mechanism. Furthermore it has been
suggested that CoBSH binding induces a conformational change that brings the methyl-
SCoM substrate into closer proximity to the nickel, and this promotes C-S bond cleavage.
To investigate the proposed structural role of CoBSH in initiating catalysis, we have solved
the X-ray crystal structures of MCR in complex with four different CoBSH analogues.
CoBSH has a heptanoyl moiety linked to the thiol group, and the analogues are pentanoyl-,
hexanoyl-, octanoyl- or nonanoyl-containing derivatives of CoBSH (CoB5SH, CoB6SH,
CoB8SH and CoB9SH respectively; Figure 2) (3, 35, 43–47). In addition, we present a
structure in which the substrate channel predominantly lacks either CoBSH or
heterodisulfide product.
MATERIALS AND METHODS
Materials
The organism Methanothemobacter marburgensis (catalog OCM82) was obtained from the
Oregon Collection of Methanogens (Portland, OR). All buffers and media reagents were
obtained from Sigma-Aldrich (St. Louis, MO). The gases N2 (99.98%), H2/CO2 (80%/20%),
and ultra high purity H2 (99.999%) were obtained from Cryogenic Gases (Grand Rapids,
MI). Ti(III) citrate solutions were prepared from a stock solution of 246 mM Ti(III) citrate,
which was synthesized by adding sodium citrate to Ti(III) trichloride (30 wt % solution in 2
N hydrochloric acid) (Acros Organics, Morris Plains, NJ) under anaerobic conditions and
adjusting pH to 7.0 with sodium bicarbonate (48). The concentration of Ti(III) citrate was
determined by titrating against a solution of methyl viologen.
Synthesis of methyl-SCoM, CoB5SH, CoB6SH, CoBSH, CoB8SH, and CoB9SH
Methyl-SCoM was prepared from HSCoM and methyl iodide (49). The homodisulfides,
CoB5S-SCoB5, CoB6S-SCoB6, CoBS-SCoB, CoB8S-SCoB8, CoB9S-SCoB9 were prepared
as described from 5-bromovaleric acid, 6-bromohexanoic acid (Sigma-Aldrich, St. Louis,
MO), 7-bromoheptanoic acid (Karl Industries, Aurora, OH), 8-bromooctanoic acid, and 9-
bromononanoic acid (Matrix Scientific, Columbia, SC), respectively (43, 46). The free thiol
forms of CoB5SH, CoB6SH, CoBSH, CoB8SH, and CoB9SH were generated by the
reduction of the homodisulfides as previously described (45). The purity of the CoBSH
analogues was determined by 1H NMR spectroscopy. All compounds synthesized were
stored in a Vacuum Atmospheres chamber maintained at an oxygen level below 1 ppm, as
monitored continually with an oxygen analyzer (model 317, Teledyne Analytical
Instruments, City of Industry, CA) until use.
M. marburgensis Growth and MCRred1 Purification
Buffer preparations and all manipulations were performed under strict anaerobic conditions
in a Vacuum Atmospheres chamber maintained at an oxygen level below 1 ppm, as
monitored continually with an oxygen analyzer (model 317, Teledyne Analytical
Instruments, City of Industry, CA). MCRred1 was isolated from M. marburgensis cultured on
Cedervall et al.
Page 3
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
H2/CO2 (80%/20%) at 65 °C in a 14-L fermentor (New Brunswick Scientific Co., Inc. New
Brunswick, NJ). Culture media were prepared as previously described (20, 24). MCRred1
was generated in vivo and purified as described previously (20). The purification procedure
routinely generates 65–75% MCRred1 as determined by UV-visible and EPR spectroscopy.
Spectroscopy of MCR
UV-visible spectra of the Ni(I)-containing MCRred1 were recorded in the anaerobic chamber
using a spectrophotometer (Model USB4000-UV-VIS, Ocean Optics, Dunedin, FL). EPR
spectra were recorded on a Bruker EMX spectrometer (Bruker Biospin Corp., Billerica,
MA), equipped with an Oxford ITC4 temperature controller, a Hewlett-Packard Model 5340
automatic frequency counter and Bruker gaussmeter. The EPR spectroscopic parameters
included: temperature, 70 K; microwave power, 10 mW; microwave frequency, 9.43 GHz;
receiver gain, 2 × 104; modulation amplitude, 10.0 G; modulation frequency, 100 kHz.
Double integrations of the EPR spectra were performed and referenced to a 1 mM copper
perchlorate standard. The NMR data were acquired at 298 K on a Bruker Avance DRX 500
MHz instrument equipped with a TXI cryoprobe.
Preparation of MCRred1 for Crystallization
All crystallization experiments were performed in the anaerobic chamber in which MCR
was purified unless otherwise noted. MCRred1 was prepared in 50 mM Tris, pH 7.6 and
excess HSCoM was removed by buffer exchange using an Amicon Ultra centrifuge filter
with a 50 kDa cut-off membrane (Millipore). Typically, 2–3 ml of MCRred1 was exchanged
with 10–15 ml of 50 mM Tris, pH 7.6. The enzyme was concentrated to 500–600 μl, and
this process was repeated three times. The fraction of MCRred1 in the purified MCR sample
was calculated from the UV-visible spectrum using extinction coefficients of 27.0
mM−1cm−1 at 385 nm for Ni(I)-MCRred1, and 9.15 mM−1cm−1 at 420 nm for Ni(II)-
MCRred1-silent (20). The amount of MCRred1 in samples used for crystallization was
determined to be ~80% and the concentration of total enzyme used was in the range of about
120–150 μM (~32–40 mg/ml). All crystallization experiments were performed anaerobically
by incubating 2.0 μl of enzyme solution in 50 mM Tris, pH 7.5 and 2.0 μl of reservoir
solution (100 mM Hepes-Na, pH 7.3/7.5/8.0; 150 mM magnesium acetate (Mg(CH3COO)2),
and 20/22% (w/v) PEG 400) in a sitting drop over 1 ml reservoir solution at 9 ºC. Triangular
and rectangular prismatic crystals with a bright yellowish-green color confirmed the
presence of nickel coenzyme F430. The crystals grew to a size of approximately 100–200 μm
in 4–5 days. CoBSH-depleted crystals were obtained by incubating 2 μl of a reaction
mixture containing 139 μM MCRred1 and 13 mM HSCoM with 2 μl of reservoir solution
(100 mM Hepes-Na pH 7.5, 150 mM magnesium acetate (Mg(CH3COO)2), 22% PEG 400).
Crystals of MCR complexed with the CoBSH analogues were grown by co-crystallization.
The CoB5SH co-crystals were obtained by incubating 2μl enzyme solution containing 124
μM MCRred1, 10 mM methyl-SCoM and 1 mM CoB5SH with 2μl of reservoir solution (100
mM Hepes-Na, pH 7.5, 150 mM (Mg(CH3COO)2), 22 % PEG 400). The crystals with
bound CoB6SH and CoB9SH were obtained by co-crystallization of 1 mM of analogue with
142 μM MCRred1 and equilibrated with 2μl of reservoir solution (100 mM Hepes-Na, pH
7.5, 150 mM Mg(CH3COO)2, 20 % PEG 400 for CoB6SH and 100 mM Hepes-Na, pH 7.3,
150 mM Mg(CH3COO)2), 22 % PEG 400 for CoB9SH). Crystals were cryoprotected in
reservoir solution containing 25 % (v/v) PEG 400 by soaking for 2–5 minutes before
cryocooling in liquid nitrogen in the anaerobic chamber. Crystals of CoB8SH bound to MCR
were obtained by incubating 2 μl of a mixture of 119 μM MCRred1 and 1 mM CoB8SH with
2 μl of reservoir solution (100 mM Hepes-Na, pH 7.3, 150 mM Mg(CH3COO)2), 20 % PEG
400). Before cryoprotection, the crystals were soaked for 5–10 minutes in a 100 mM
solution of methyl iodide2. The methyl iodide solution used for soaking was prepared by
adding a concentrated stock of methanolic solution of methyl iodide to the reservoir
Cedervall et al.
Page 4
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
solution. Soaked crystals were quickly cryoprotected as described above and cryocooled in
the anaerobic chamber.
X-ray Diffraction Data Collection, Processing and Refinement
X-ray diffraction data were collected at 100 K on a ADSC Quantum-315 detector at the APS
Beamline 14-BM-C (BioCARS). The wavelength of X-rays was 0.979 Å. Data were
processed using HKL2000 (50). As in the previous X-ray crystallographic studies, the
crystals belong to the monoclinic space group P21 (a = 82 Å, b = 118 Å, c = 122 Å, β = 92°),
with one MCR molecule (two active sites) per asymmetric unit (5, 33). For refinement,
REFMAC in the Collaborative Computational Project Number 4 (CCP4) program suite was
used (51). A random sample of 5 % of the data across all resolution shells was chosen to
check refinement progress through calculation of an Rfree. The same reflections were used to
calculate Rfree for all structures, thus preventing bias due to high structural identity. The
remaining reflections were used in refinement (Rwork). Model building was done using the
Crystallographic Object-Oriented Toolkit (COOT) (52). The diffraction data and their
models are designated as MCRCoBXSH, where X is the number of carbons in the alkanoyl
portion of the analogue. Library files in CCP4 for F430 and CoBSH were incorrect, and these
were modified in Monomer Library Sketcher in the CCP4 program suite by comparison with
schematic drawings from Grabarse et al. (33). Co-ordinate and CCP4 library files for the
different CoBSH analogues were created in Monomer Library Sketcher. The general model
building and refinement strategy for all structures was as follows. It was clear from the
electron density in the substrate channel and at the active site that mixtures of species were
present in all datasets. These could be visualized with Fo-Fc and Fo-Fo difference electron
density maps (Supporting Information, Figure S1). The known positions of CoBSH and
HSCoM from the published Ni(II)-MCR crystal structures (PDB codes 1hbn, 1hbo, 1hbu
(33)) were used as guides to determine which species could be present in each dataset, and
these were then simultaneously modeled into the electron density. By alteration of their
relative occupancies (in 10% increments) followed by refinement, the ratio of occupancy
between different species was determined using the assumption that the average B-factors
for all molecular species bound should be similar to that of F430 and adjacent well-ordered
protein atoms within the active site and substrate channel. The combinations of modeled
ligands were constantly reassessed throughout refinement based on the remaining difference
electron density. This included test refinements of different ligand combinations during the
latter stages, thus using the optimized phases to check whether a different combination of
ligands could also explain the electron density. Sensible chemical structures and
interactions, along with keeping the combined occupancies of sterically mutually exclusive
species ≤ 100%, were maintained throughout refinement. The model was finally accepted
when the difference electron density map was minimal and the B-factors for the models
converged.
In practice the first structure refined was that of MCRCoB5SH. Initial phases were generated
by difference Fourier using a previously determined crystal structure (PDB code 1mro (5))
but with all non-bonded molecules, including water, removed from the model except F430.
Initial rigid body refinement followed by restrained refinement of MCRCoB5SH reduced the
Rwork to 26.5%. After model building and subsequent rounds of restrained refinement the
Rwork was 14.3% (Rfree 16.6%). Of the five structures only the CoB5SH analogue is
completely coincident with CoBSH, and so particular care had to be used in teasing apart the
ratios of the two species in modeling the MCRCoB5SH electron density. This was done by
2Methyl iodide was added with the intention of creating a Ni(III)-methyl species, which CoB8SH stabilizes. This was not achieved,
but the diffraction quality of this crystal was significantly better than crystals co-crystallized with CoB8SH alone, and so has been
included in this study.
Cedervall et al.
Page 5
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
initially refining 100% CoBSH in the substrate channel. Positive Fo-Fc difference density
located at the carbon where the shorter CoB5SH thiol might be expected to be indicated the
presence of a more electron-rich species than carbon, which is consistent with the presence
of the CoB5SH sulfur. The refinement converged at a model containing a 50:50 mix of
CoBSH and the analogue. However, positive Fo-Fc difference density was still present at the
position of the CoBSH thiol, therefore a water molecule was added to the CoB5SH model at
50% occupancy and upon refinement this accounted for the electron density. An illustration
of the electron density quality from this structure is shown in Supporting Information,
Figure S2. An HSCoM-, CoBSH- and CoBSH analogue-free version of the refined
MCRCoB5SH structure was used as the starting model to generate initial phases for the four
other structures. After the initial round of restrained refinement the Rwork for these structures
were reduced to 14.5–15.6 %.
RESULTS AND DISCUSSION
Crystal Structures of MCR
Five crystal structures were determined, four of which are in complex with CoBSH
analogues differing in the number of carbons atoms in the alkanoyl portion of the molecule.
CoBSH is an N-7-mercaptoheptanoyl-containing molecule, whereas the four CoBSH
analogues contain N-5-mercaptopentanoyl-, N-6-mercaptohexanoyl-, N-8-mercaptooctanoyl-
or N-9-mercaptononanoyl-moieties (Figure 2). The corresponding crystal structures are
designated as MCRCoBXSH, where X is the number of carbons in the alkanoyl portion of the
analogue. The other crystal structure is of MCRred1c-silent (MCR in the Ni(II) state in
complex with HSCoM, designated here as MCRHSCoM) that is CoBSH-depleted. The
datasets have resolutions in the range from 1.35 – 1.8 Å. Although the crystallizations were
set up with the MCR solution predominantly in the Ni(I)-MCRred1 state, by the time X-ray
diffraction data were collected they had been oxidized to the Ni(II)-MCRred1-silent state
(Supporting Information). Following data collection there was no evidence for
photoreduction of the Ni(II) back to Ni(I) in any of the crystals, as assessed by single crystal
UV-visible microspectrophotometry (Supporting Information and Figure S3). Attempts to
photoreduce the crystals using different wavelengths and temperatures were unsuccessful
(Supporting Information).
Overall, the resulting structures are very similar to each other and to the previously
published structures of MCR, with differences mainly localized to the active site and
substrate channel. The two active sites in the ASU were refined independently. Unless
otherwise stated there was no difference between them. All five datasets contain a mixture
of species bound to the enzyme. There is always a background of CoBSH and HSCoM,
which co-purify with MCR and cannot be fully removed by extensive buffer exchange or by
the addition of a CoBSH analogue. HSCoM is added during purification of MCR, as it
stabilizes the resting active Ni(I) state (unpublished data), and this leads to HSCoM
occupancies between 50–100% amongst the structures (Table 1). In contrast CoBSH, which
is not added during purification, has occupancies ranging from 30–50%. As these
confounding species have all been described at high occupancy in other crystallographic
studies, the structural data of interest could be isolated (5, 33). In each case, the additional
electron density could be explained by inclusion of the appropriate CoBXSH model used in
that experiment at 50% or higher occupancy. The resulting models, along with 2Fo-Fc
electron density, are shown in Figure 3. The Rwork for the final structures range from 13.0 to
15.0 % (Rfree 15.5 to 19.5 %). The X-ray data collection, processing, refinement and model
building statistics are given in Table 1.
Cedervall et al.
Page 6
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Analogues shorter than CoBSH; CoB5SH and CoB6SH
CoB5SH is two methylene groups shorter than CoBSH, the MCR substrate. The
MCRCoB5SH structure is to 1.35 Å resolution. As expected the pentanoyl chain follows the
path of the CoBSH heptanoyl carbons down the substrate channel, and thus its thiol is
positioned in the same place as the second carbon preceding the CoBSH thiol (Figure 3A
and 4). There are no published MCR kinetic studies using CoB5SH, but as it binds in the
substrate channel, it is likely to be an inhibitor.
CoB6SH is one methylene shorter than CoBSH, and is a slow substrate of MCR. In this case
the 1.4 Å resolution electron density of MCRCoB6SH indicates that the analogue
unexpectedly binds in the substrate channel such that its thiol is virtually in the same
position as the thiol of CoBSH (Figure 3B and 4). The hexanoyl chain is oriented so that it
takes a shorter route down the substrate channel between carbons 2 and 5 (with the carbonyl
carbon labeled as carbon 1) than CoBSH (Figure 4, Supporting Information, Figure S4).
This short-cut is not seen in any of the other CoBXSH complex crystal structures, but
presumably arises because this CoB6SH binding conformer is energetically more favorable,
although it is not clear from the structure why this might be the case. CoB6SH binds very
tightly to MCR, with an apparent Ki value of 0.1 μM (3).
Water structure in the absence of HSCoM
The electron density for the MCRCoB6SH crystal structure only supported the modeling of 50
% bound HSCoM. In the fraction of MCR molecules where HSCoM is absent, the HSCoM
binding site is occupied by a network of four water molecules (Supporting Information,
Figure S5). Two waters are positioned close to the absent sulfonate oxygen positions of
HSCoM. Based on the presence of positive difference electron density, a third water was
modeled ligated to the Ni, and refined to a distance of ~2 Å (2.0 Å and 2.1 Å in the two
active sites of the ASU) with no distance restraint imposed between the Ni and water. This
water is in a similar position as the Ni coordinating sulfonate oxygen of the heterodisulfide
product in MCRsilent (Supporting Information, Figure S6 and PDB codes 3m32, 1hbm) (5,
33). The fourth water was in the vicinity of the expected position of a bridging water (W1)
seen in other structures (Figure 1, 3A and 3C).
Water structure in the absence of CoBSH
The 1.45 Å resolution electron density obtained for MCRHSCoM indicates that the substrate
channel contains only 30 % CoBSH. Nine ordered waters (W1–W9), along with an acetate
ion from the crystallization solution occupy the channel, with the acetate positioned where
the phosphothreonine linkage of CoBSH would be (Figure 3C). Presumably 1–2 further
waters would replace the acetate under physiological conditions. Other than W3 and W7, the
waters form hydrogen bonds with protein (Figure 5). One water (W2) occupies the same site
as the CoBSH thiol. Presumably due to the loss of favorable interactions that exist when
CoBSH is present, the hydrophobic side-chain of Valα482 adopts a second conformation
modeled at 60 % occupancy (Supporting Information, Figure S7).
Position of the “bridging” water, W1
The equivalent of W1 has previously been observed in MCRred1-silent and MCRox1-silent
crystal structures where, in the presence of CoBSH and HSCoM, it is sited equidistant (3.2
Å) between the two coenzyme thiols (PDB codes: 1hbn, 1hbo, 1hbu) and thus been termed
the “bridging water” (Figure 1) (5, 33). However, in the MCRHSCoM structure, due to the
presence of the more polarized W2 water, W1 is displaced away from HSCoM to maximize
the hydrogen bond interaction with W2 (2.9 Å to W2; 3.5 Å to HSCoM thiol, Figure 5). In
the MCRCoB5SH structure that also contained W2, the electron density indicated that this
Cedervall et al.
Page 7
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
repositioning of W1 towards W2 also occurred. In contrast, the MCRCoB6SH structure
contained 100% thiol at the CoBSH position, but a partial occupancy of HSCoM (50%). In
this case the electron density for W1 indicated it had moved towards the nickel to form an
optimal hydrogen bond with a Ni-ligating water that was only present in the absence of
HSCoM (3.7 Å to CoB(6)SH thiol; 3.0 Å to Ni-ligating water, Supporting Information,
Figure S5). In all structures reported here, W1 (if present) appears to be a sensitive indicator
of the relative electronegativity of the Ni-ligated atom to that occupying the position of the
CoBSH thiol, and was a useful check in the crystallographic modeling and refinement
process.
Flexibility in the substrate channel: Alternative protein conformers
The binding site of HSCoM (and presumably methyl-SCoM) is more deeply buried within
the enzyme, and so it must enter prior to CoBSH for productive chemistry to occur. As
binding of CoBSH in the absence of co-substrate would be inhibitory, it was suggested that
a conformational change upon methyl-SCoM binding might lower the Kd for CoBSH, and
thus promote an ordered mechanism. Compared to the 1hbn MCRox1-silent and 1hbu
MCRred1-silent crystal structures, which both have full occupancy HSCoM, the lower
occupancy of HSCoM in the 1hbo MCRred1-silent structure was associated with significantly
greater flexibility within the channel, and the ability to model a second conformation of a
Gly-rich amino acid stretch that formed part of the CoBSH channel. This suggested that
methyl-SCoM binding might cause the channel to become more ordered, increasing the
affinity of MCR for CoBSH by conformational restriction rather than a switch mechanism
where the structure reorganizes from one well-defined conformer to another (33). In the
MCRHSCoM data containing 30 % CoBSH and 100% HSCoM, the Fo-Fc difference electron
density map at one of the two independent active sites in the ASU contained positive peaks
that suggested the presence of an alternate conformation also involving this part of the
polypeptide (Supporting Information, Figure S8). Using this as a guide, a similar second
conformation involving seven contiguous amino acid residues of the same Gly-rich amino
acid stretch (β366–372) could be modeled and refined at 20 % occupancy leaving no
residual difference density. Parts of the α′ subunit (α′111–129 and α′237–242) that are in
close proximity to this stretch of amino acids also exhibit second conformations, with the
main-chain carbonyl of α′243 in van der Waals contact with the B ring of F430 tetrapyrrole
(Supporting Information, Figure S9). Modeling these at 20% occupancy accounted for the
weak positive Fo-Fc difference electron density peaks observed in these areas. The evidence
of alternate conformers in these areas lends support to the proposal that increased flexibility
in the substrate channel propagates through the protein (33).
The MCRCoB6SH crystal structure contains 50 % CoB6SH, 50% CoBSH and 50% HSCoM.
In this case there is no evidence of an alternate loop conformation in either active site of the
ASU. However, as CoBSH and CoB6SH combined are at 100% occupancy, it is not
surprising their favorable interactions with the substrate channel would reduce
conformational disorder, despite the partial occupancy of HSCoM.
Analogues longer than CoBSH; CoB8SH and CoB9SH
Both analogues could be accommodated in the MCR substrate channel (Figures 3D and 3E).
The electron density supported final models containing 50% CoB8SH for MCRCoB8SH (1.8
Å resolution) and 60 % CoB9SH for MCRCoB9SH (1.45 Å resolution). The phosphate head-
groups are in identical positions to those of CoBSH, CoB5SH, CoB6SH (Figure 4) (5, 33).
Both analogues follow the crystallographically observed chain path of bound CoBSH, with
the extra atoms displacing the W1 water and placing the thiols closer to the nickel (Figure
6). CoB9SH does have a second conformer that deviates from the CoBSH path, but the thiol
position for this conformer and the CoBSH-tracking conformer are identical (Figure 3E and
Cedervall et al.
Page 8
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Supporting Information, Figure S10). Interestingly, the thiol of CoB8SH is not coincident
with the CoB9SH carbon that precedes the CoB9SH thiol (Figure 6). CoB8SH is an MCR
inhibitor with an apparent Ki of 15 μM (3). CoB9SH has never been tested for inhibition of
MCR-catalyzed methane formation, but it is reasonable to assume that it would be an
inhibitor.
CoBXSH thiol-to-nickel spatial relationship
The position of CoBSH in previous crystal structures poses a conundrum (5, 33). In all the
proposed catalytic mechanisms, CoBSH must interact with species generated at the nickel.
Perplexingly, the sulfur of the CoBSH substrate is 8.8 Å from the Ni(II) in the MCRox1-silent
and both MCRred1-silent crystal structures, and 6.4 Å from the thiol of the substrate analogue
HSCoM (Figure 1). Modeling studies demonstrated that the addition of a methyl group to
HSCoM did not bridge this gap (35, 45, 53). Therefore, a conformational change has been
postulated that would enable CoBSH to penetrate deeper into the substrate channel, and thus
approach closer to any nickelbound species. The heterodisulfide product in the MCRsilent
crystal structure has the CoBSH portion in virtually the same place as in MCRox1-silent,
giving no clue to possible structural changes that might occur to facilitate CoBSH reacting
with nickel-associated intermediates (5, 33).
Trigonometry suggests that if the alkanoyl chain of CoBSH or its analogue is in an extended
conformation, each additional unit in the chain would lead to the thiol moving ~1.2 Å
towards the Ni. Until this study there have been no crystal structures of CoBSH analogues in
complex with MCR, so mechanistic studies using different chain length analogues of
CoBSH assumed that shorter analogues would trace the observed path of CoBSH, and
longer analogues would penetrate about ~1.2 Å deeper per additional chain unit into the
channel. In the case of the shortest analogue CoB5SH, it does indeed follow the path of
CoBSH, with the thiol of CoB5SH being 2.8 Å away from the thiol position of CoBSH.
However, due to the conformation CoBSH adopts when bound in the substrate channel, the
difference in the S-Ni distance is small; the CoB5SH thiol being only 0.5 Å farther from the
Ni than CoBSH (8.8 Å for CoBSH vs. 9.3 Å for CoB5SH) (Table 2). This is due to the
alkanoyl chain of CoBSH not being in an extended conformation from carbons 4 to 6
(carbon 1 is the carbonyl carbon). CoB6SH on the other hand adopts a conformation that
places its thiol in virtually the same position as the thiol of CoBSH (Figure 4 and Table 2).
This is consistent with CoB6SH being a substrate. However, the kcat is 1000-fold lower than
for CoBSH (3, 35) although its Km value (180 μM, Dey & Ragsdale, in preparation) is
similar to that of CoBSH (Km = 75 μM (3)). The reason for this may be that the shorter
alkanoyl chain may not enable the analogue thiol to approach the nickel close enough for
efficient catalysis, and thus explain why CoB6SH is such a poor substrate.
In the case of the longer CoBXSH analogues, the sulfur of CoB8SH is 2.6 Å closer to the Ni
ion of F430 than that of CoBSH, and 2.5 Å closer to the thiol of HSCoM (Figure 6 and Table
2). The CoB9SH molecule follows the path of CoBSH, and reaches only a little further into
the substrate channel than CoB8SH, with the CoB9SH thiol positioned 2.9 Å closer to the Ni
than the thiol of CoBSH (Figure 6 and Table 2). This is only 0.3 Å closer than the distance
observed for the CoB8SH thiol, even though they are non-coincident. The distance to the
thiol of HSCoM is 2.6 Å closer than that of the substrate, CoBSH; only 0.1 Å closer than the
CoB8SH thiol. The two analogue thiols are above an annular hydrophobic aromatic
environment created by Pheα330, Tyrα333, Pheα443, Pheβ361 and Tyrβ367 that lies
between them and F430 (Figure 6). As a result, penetrating further into the channel may be
energetically unfavorable, consistent with the small difference in relative distances between
the CoB8SH/CoB9SH thiols and the HSCoM thiol/F430 nickel. The annulus is proposed to
be catalytically important in positioning methyl-SCoM and stabilizing the methane product,
Cedervall et al.
Page 9
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
and the tyrosines have been proposed to be proton donors associated with mechanism II
(Scheme 2B) (5, 33).
Thus, there appear to be three preferential distances for thiols (including that of HSCoM)
within the MCR substrate channel; HSCoM at 2.4 Å, CoB6SH/CoBSH at 8.7 – 8.8 Å and
CoB8SH/CoB9SH at 5.9 – 6.2 Å from the nickel of F430 (Table 2).
Recent ENDOR, high-field continuous and pulse EPR work has identified changes in nickel
co-ordination when CoBSH is added to MCRred1c (active Ni(I)-MCRred1 + HSCoM) (14,
15, 18, 31). This generates up to 50% MCRred2 which is comprised of two distinct nickel co-
ordination geometries; an axial MCRred2a formally assigned as a Ni(III)-hydride, and a
rhombic MCRred2r in which the thiol of HSCoM is a Ni(I) ligand (Supporting Information,
Scheme S1). Formation of MCRred2 could also be induced by addition of CoBSH substrate
analogues CoBS-CH3 and CoBS-CF3, which have a chain length one unit longer than
substrate CoBSH (18, 53). The CoBS-CF3 enabled 19F-ENDOR studies to be performed,
and demonstrated that following CoBS-CF3 addition the remaining MCRred1 species had
Ni(I)-19F distances of 6.2 – 7.7 Å. This distance range agreed with a CoBS-CF3 model
created using the CoBSH position observed in the MCRox1-silent crystal structure (53).
However, in the MCRred2 species the Ni(I)-19F distances had shortened indicating a
movement of ~ 2 Å towards the nickel (MCRred2a Ni(I)-19 F, 4.0 – 5.5 Å; MCRred2r
Ni(I)-19F, 4.5 – 5.7 Å). In the case of the CoB8SH analogue (the closest equivalent to CoBS-
CF3 used in this study) the thiol to Ni(II) distance lies between the distance ranges observed
in the CoBS-CF3 studies, and so the fluorine(s) of the CoBS-CF3 in MCRred2 might
penetrate a little further into the hydrophobic annulus in the MCRred2 species. As the
alkanoyl chain of CoBSH is not fully extended it could easily undergo a similar
conformation change to that observed in the MCRred2 state.
Reaction of MCR Ni(III)-alkyl species’ with CoB8SH and CoB9SH
The two longer CoBXSH analogues have been shown to undergo alkylation when reacted
with MCRPS, a [Ni(III)-alkyl ↔ Ni(II)-alkylsulfonate radical] formed from reaction of
Ni(I)-MCRred1 with bromopropanesulfonate (BPS) (Supporting Information, Scheme S1)
(20, 23, 30, 45, 54). BPS is a substrate of MCRred1 in a reaction that involves a rapid
CoBSH-independent nucleophilic attack by Ni(I) on BPS to displace bromide and generate
MCRPS at a rate ~60-fold faster than generation of methane from CoBSH and methyl-
HSCoM (20, 45). Certain thiols can eliminate the propylsulfonate to yield a thioether
product and regenerate MCRred1, although at a rate 1000-fold slower than methane
formation (45). Both CoB8SH and CoB9SH can react with MCRPS to regenerate MCRred1,
but CoBSH cannot. The overall second-order rate constant for the reactivation of MCR by
CoB8SH is 160 M−1s−1, whereas for CoB9SH the reaction is more sluggish (12 M−1s−1).
CoB9SH might be expected to be closer to the proximal Ni ligand. It was therefore proposed
that this caused steric interference and explained why CoB9SH was a poorer reactivator of
MCR than CoB8SH. Our study has shown that the thiols of these two analogues are placed
such that they are approximately the same distance (~3.7 Å) from the thiol of HSCoM
ligated to the Ni atom (Table 2). The Ni(II)-HSCoM bond is 2.4 Å, whereas an Ni(III)-alkyl
bond is expected to be ~2 Å (24, 33), thus indicating that a conformational change is
required to effect the nucleophilic attack of the CoB8SH and CoB9SH thiols on an alkyl-
bound species. It would thus appear that a conformational change, such as observed in
MCRred2, is required for this chemistry also (53).
A Ni(III)-alkyl species is akin to the first intermediate in mechanism I of MCR-catalyzed
methane formation, Ni(III)-methyl (MCRMe, Supporting Information, Scheme S1, Scheme
2A) (11, 27). MCRMe has been shown to be capable of generating MCRred1 and methyl-
SCoM upon addition of HSCoM (which is the reverse of mechanism I, step 1, Scheme 2A);
Cedervall et al.
Page 10
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
similar chemistry to the observed formation of a thioether product from the Ni(III)-alkyl.
Further addition of CoBSH following HSCoM treatment of MCRMe led to methane and
heterodisulfide formation, the natural products of methanogenesis. Although this lends
credence to mechanism I, it should be noted that like MCRPS, MCRMe in these experiments
was generated artificially. However, the MCRCoB9SH crystal structure demonstrates that the
two additional methylene units in the alkanoyl chain c.f. CoBSH, do not necessarily translate
into direct interaction of the thiol with the nickel proximal ligand. However, this could
represent the favorable position for a CoBSH thiol interacting with the methyl group of
methyl-SCoM. Just as the alkanoyl chain of CoB6SH has a more extended conformation
than CoBSH in the substrate channel, CoBSH could also adopt a more extended
conformation so that its thiol was in a similar position as the thiol of CoB8SH, priming it for
reaction with a nickel bound species.
If a significant conformational change is required early in MCR-catalyzed chemistry, which
would be a requirement of mechanism I, catalysis may well involve a rearrangement of the
aromatic amino acid annulus due to the presence of the methyl of methyl-SCoM, and this
might enable deeper penetration of CoBSH into MCR (Figure 6). All the crystal structures in
this study, and those solved previously, are of the inactive Ni(II)-MCR, which disfavors
close approach to the nickel in the absence of Ni(I)-bound methyl-SCoM, even in the case of
CoB9SH.
Conclusion
The goal of this study was to induce structural changes within the substrate channel and
active site of MCR using analogues of coenzyme CoBSH. It was hoped that this would shed
light on the nature of conformational changes that have been proposed to occur in MCR
catalysis. We have shown that that the CoBXSH analogues do not lead to any significant
conformational changes within the context of inactive Ni(II)-MCR. Therefore, it may be that
methyl-SCoM is the key coenzyme, in combination with a nickel oxidation state of 1 (and
3), that triggers a conformational change bringing the thiol of CoBSH closer to the nickel.
Thus, the crystal structure of the Ni(I)-methyl-SCoM/MCR complex may be required to
structurally define the conformational changes required for MCR-mediated chemistry.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Acknowledgments
X-ray data were collected at beam-line 14-BM-C and photoreduction studies at 14-ID-B, BioCARS at the
Advanced Photon Source (APS), Argonne National Laboratory, Argonne, IL. We thank Vukica Srajer and Yu-
Sheng Chen for valuable assistance during data collection. Use of the Advanced Photon Source was supported by
the U.S. Department of Energy, Basic Energy Sciences, Office of Science under Contract No. DE-
AC02-06CH11357. Use of the BioCARS Sector 14 was supported by the National Institutes of Health, National
Center for Research Resources, under grant number RR007707. Computer resources were provided by the Basic
Sciences Computing Laboratory of the University of Minnesota Supercomputing Institute, and we thank Can
Ergenekan for his support. We also thank Ed Hoeffner at the Kahlert Structural Biology Laboratory (KSBL) at the
University of Minnesota. Use of the KSBL was supported by a Minnesota Partnership for Biotechnology and a
Medical Genomics Grant SPAP-05-0013-P-FY06.
References
1. Thauer RK. Biochemistry of methanogenesis: a tribute to Marjory Stephenson. Microbiology. 1998;
144:2377–2406. [PubMed: 9782487]
2. Thauer RK, Shima S. Methane as fuel for anaerobic microorganisms. Ann N Y Acad Sci. 2008;
1125:158–170. [PubMed: 18096853]
Cedervall et al.
Page 11
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
3. Ellermann J, Hedderich R, Bocher R, Thauer RK. The final step in methane formation.
Investigations with highly purified methyl-CoM reductase (component C) from Methanobacterium
thermoautotrophicum (strain Marburg). Eur J Biochem. 1988; 172:669–677. [PubMed: 3350018]
4. Ellefson WL, Wolfe RS. Component C of the methylreductase system of Methanobacterium. J Biol
Chem. 1981; 256:4259–4262. [PubMed: 6783657]
5. Ermler U, Grabarse W, Shima S, Goubeaud M, Thauer RK. Crystal structure of methyl-coenzyme
M reductase: the key enzyme of biological methane formation. Science. 1997; 278:1457–1462.
[PubMed: 9367957]
6. Diekert G, Gilles HH, Jaenchen R, Thauer RK. Incorporation of 8 succinate per mol nickel into
factors F430 by Methanobacterium thermoautotrophicum. Arch Microbiol. 1980; 128:256–262.
[PubMed: 7212929]
7. Diekert G, Jaenchen R, Thauer RK. Biosynthetic evidence for a nickel tetrapyrrole structure of
factor F430 from Methanobacterium thermoautotrophicum. FEBS Letters. 1980; 119:118–120.
[PubMed: 7428919]
8. Whitman WB, Wolfe RS. Presence of nickel in Factor F430 from Methanobacterium bryantii.
Biochem Biophys Res Comm. 1980; 92:1196–1201. [PubMed: 7370029]
9. Albracht SPJ, Ankel-Fuchs D, Böcher R, Ellermann J, Moll J, van der Zwann JW, Thauer RK. Five
new EPR signals assigned to nickel in methyl-coenzyme M reductase from Methanobacterium
thermoautotrophicum, strain Marburg. Biochim Biophys Acta. 1988; 955:86–102.
10. Dey M, Kunz RC, Lyons DM, Ragsdale SW. Characterization of alkyl-nickel adducts generated by
reaction of methyl-coenzyme m reductase with brominated acids. Biochemistry. 2007; 46:11969–
11978. [PubMed: 17902704]
11. Dey M, Telser J, Kunz RC, Lees NS, Ragsdale SW, Hoffman BM. Biochemical and spectroscopic
studies of the electronic structure and reactivity of a methyl-Ni species formed on methyl-
coenzyme M reductase. J Am Chem Soc. 2007; 129:11030–11032. [PubMed: 17711283]
12. Duin EC, Cosper NJ, Mahlert F, Thauer RK, Scott RA. Coordination and geometry of the nickel
atom in active methyl-coenzyme M reductase from Methanothermobacter marburgensis as
detected by X-ray absorption spectroscopy. J Biol Inorg Chem. 2003; 8:141–148. [PubMed:
12459909]
13. Duin EC, Signor L, Piskorski R, Mahlert F, Clay MD, Goenrich M, Thauer RK, Jaun B, Johnson
MK. Spectroscopic investigation of the nickel-containing porphinoid cofactor F(430). Comparison
of the free cofactor in the (+)1, (+)2 and (+)3 oxidation states with the cofactor bound to methyl-
coenzyme M reductase in the silent, red and ox forms. J Biol Inorg Chem. 2004; 9:563–576.
[PubMed: 15160314]
14. Finazzo C, Harmer J, Bauer C, Jaun B, Duin EC, Mahlert F, Goenrich M, Thauer RK, Van
Doorslaer S, Schweiger A. Coenzyme B induced coordination of coenzyme M via its thiol group
to Ni(I) of F430 in active methyl-coenzyme M reductase. J Am Chem Soc. 2003; 125:4988–4989.
[PubMed: 12708843]
15. Finazzo C, Harmer J, Jaun B, Duin EC, Mahlert F, Thauer RK, Van Doorslaer S, Schweiger A.
Characterization of the MCRred2 form of methyl-coenzyme M reductase: a pulse EPR and
ENDOR study. J Biol Inorg Chem. 2003; 8:586–593. [PubMed: 12624730]
16. Goubeaud M, Schreiner G, Thauer RK. Purified methyl-coenzyme-M reductase is activated when
the enzyme-bound coenzyme F430 is reduced to the nickel(I) oxidation state by titanium(III)
citrate. Eur J Biochem. 1997; 243:110–114. [PubMed: 9030728]
17. Harmer J, Finazzo C, Piskorski R, Bauer C, Jaun B, Duin EC, Goenrich M, Thauer RK, Van
Doorslaer S, Schweiger A. Spin density and coenzyme M coordination geometry of the ox1 form
of methyl-coenzyme M reductase: a pulse EPR study. J Am Chem Soc. 2005; 127:17744–17755.
[PubMed: 16351103]
18. Harmer J, Finazzo C, Piskorski R, Ebner S, Duin EC, Goenrich M, Thauer RK, Reiher M,
Schweiger A, Hinderberger D, Jaun B. A nickel hydride complex in the active site of methyl-
coenzyme m reductase: implications for the catalytic cycle. J Am Chem Soc. 2008; 130:10907–
10920. [PubMed: 18652465]
Cedervall et al.
Page 12
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
19. Hinderberger D, Ebner S, Mayr S, Jaun B, Reiher M, Goenrich M, Thauer RK, Harmer J.
Coordination and binding geometry of methyl-coenzyme M in the red1m state of methyl-
coenzyme M reductase. J Biol Inorg Chem. 2008; 13:1275–1289. [PubMed: 18712421]
20. Kunz RC, Horng YC, Ragsdale SW. Spectroscopic and kinetic studies of the reaction of
bromopropanesulfonate with methyl-coenzyme M reductase. J Biol Chem. 2006; 281:34663–
34676. [PubMed: 16966321]
21. Mahlert F, Bauer C, Jaun B, Thauer RK, Duin EC. The nickel enzyme methyl-coenzyme M
reductase from methanogenic archaea: In vitro induction of the nickel-based MCR-ox EPR signals
from MCR-red2. J Biol Inorg Chem. 2002; 7:500–513. [PubMed: 11941508]
22. Mahlert F, Grabarse W, Kahnt J, Thauer RK, Duin EC. The nickel enzyme methyl-coenzyme M
reductase from methanogenic archaea: in vitro interconversions among the EPR detectable MCR-
red1 and MCR-red2 states. J Biol Inorg Chem. 2002; 7:101–112. [PubMed: 11862546]
23. Rospert S, Voges M, Berkessel A, Albracht SP, Thauer RK. Substrate-analogue-induced changes
in the nickel-EPR spectrum of active methyl-coenzyme-M reductase from Methanobacterium
thermoautotrophicum. Eur J Biochem. 1992; 210:101–107. [PubMed: 1332856]
24. Sarangi R, Dey M, Ragsdale SW. Geometric and electronic structures of the Ni(I) and methyl-
Ni(III) intermediates of methyl-coenzyme M reductase. Biochemistry. 2009; 48:3146–3156.
[PubMed: 19243132]
25. Tang Q, Carrington PE, Horng YC, Maroney MJ, Ragsdale SW, Bocian DF. X-ray absorption and
resonance Raman studies of methyl-coenzyme M reductase indicating that ligand exchange and
macrocycle reduction accompany reductive activation. J Am Chem Soc. 2002; 124:13242–13256.
[PubMed: 12405853]
26. Telser J, Davydov R, Horng YC, Ragsdale SW, Hoffman BM. Cryoreduction of methyl-coenzyme
M reductase: EPR characterization of forms, MCR(ox1) and MCR (red1). J Am Chem Soc. 2001;
123:5853–5860. [PubMed: 11414817]
27. Yang N, Reiher M, Wang M, Harmer J, Duin EC. Formation of a nickel-methyl species in methyl-
coenzyme M reductase, an enzyme catalyzing methane formation. J Am Chem Soc. 2007;
129:11028–11029. [PubMed: 17711279]
28. Albracht SPJ, Ankelfuchs D, Vanderzwaan JW, Fontijn RD, Thauer RK. A New Electron-
Paramagnetic-Res Signal of Nickel in Methanobacterium-Thermoautotrophicum. Biochim
Biophys Acta. 1986; 870:50–57.
29. Telser J, Horng YC, Becker DF, Hoffman BM, Ragsdale SW. On the assignment of nickel
oxidation states of the Ox1, Ox2 forms of methyl-coenzyme M reductase. J Am Chem Soc. 2000;
122:182–183.
30. Hinderberger D, Piskorski RR, Goenrich M, Thauer RK, Schweiger A, Harmer J, Jaun B. A nickel-
alkyl bond in an inactivated state of the enzyme catalyzing methane formation. Angewandte
Chemie-International Ed. 2006; 45:3602–3607.
31. Kern DI, Goenrich M, Jaun B, Thauer RK, Harmer J, Hinderberger D. Two sub-states of the red2
state of methyl-coenzyme M reductase revealed by high-field EPR spectroscopy. J Biol Inorg
Chem. 2007; 12:1097–1105. [PubMed: 17690920]
32. Becker DF, Ragsdale SW. Activation of methyl-SCoM reductase to high specific activity after
treatment of whole cells with sodium sulfide. Biochemistry. 1998; 37:2639–2647. [PubMed:
9485414]
33. Grabarse W, Mahlert F, Duin EC, Goubeaud M, Shima S, Thauer RK, Lamzin V, Ermler U. On
the mechanism of biological methane formation: structural evidence for conformational changes in
methyl-coenzyme M reductase upon substrate binding. J Mol Biol. 2001; 309:315–330. [PubMed:
11491299]
34. Grabarse W, Mahlert F, Shima S, Thauer RK, Ermler U. Comparison of three methyl-coenzyme M
reductases from phylogenetically distant organisms: unusual amino acid modification,
conservation and adaptation. J Mol Biol. 2000; 303:329–344. [PubMed: 11023796]
35. Horng YC, Becker DF, Ragsdale SW. Mechanistic studies of methane biogenesis by methyl-
coenzyme M reductase: evidence that coenzyme B participates in cleaving the C-S bond of
methyl-coenzyme M. Biochemistry. 2001; 40:12875–12885. [PubMed: 11669624]
Cedervall et al.
Page 13
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
36. Berkessel A. Methyl-Coenzyme-M Reductase - Model Studies on Pentadentate Nickel-Complexes
and a Hypothetical Mechanism. Bioorg Chem. 1991; 19:101–115.
37. Jaun B. Coenzyme-F430 from Methanogenic Bacteria - Oxidation of F430 Pentamethyl Ester to
the Ni(Iii) Form. Helvetica Chimica Acta. 1990; 73:2209–2217.
38. Signor L, Knuppe C, Hug R, Schweizer B, Pfaltz A, Jaun B. Methane formation by reaction of a
methyl thioether with a photo-excited nickel thiolate - A process mimicking methanogenesis in
archaea. Chemistry-a European Journal. 2000; 6:3508–3516.
39. Chen SL, Pelmenschikov V, Blomberg MR, Siegbahn PE. Is there a Ni-methyl intermediate in the
mechanism of methyl-coenzyme M reductase? J Am Chem Soc. 2009; 131:9912–9913. [PubMed:
19569621]
40. Pelmenschikov V, Blomberg MRA, Siegbahn PEM, Crabtree RH. A mechanism from quantum
chemical studies for methane formation in methanogenesis. J Am Chem Soc. 2002; 124:4039–
4049. [PubMed: 11942842]
41. Pelmenschikov V, Siegbahn PE. Catalysis by methyl-coenzyme M reductase: a theoretical study
for heterodisulfide product formation. J Biol Inorg Chem. 2003; 8:653–662. [PubMed: 12728361]
42. Duin EC, McKee ML. A new mechanism for methane production from methyl-coenzyme M
reductase as derived from density functional calculations. J Phys Chem. 2008; B 112:2466–2482.
43. Bobik TA, Wolfe RS. Physiological importance of the heterodisulfide of coenzyme M and 7-
mercaptoheptanoylthreonine phosphate in the reduction of carbon dioxide to methane in
Methanobacterium. Proc Natl Acad Sci U S A. 1988; 85:60–63. [PubMed: 3124103]
44. Goenrich M, Duin EC, Mahlert F, Thauer RK. Temperature dependence of methyl-coenzyme M
reductase activity and of the formation of the methyl-coenzyme M reductase red2 state induced by
coenzyme B. J Biol Inorg Chem. 2005; 10:333–342. [PubMed: 15846525]
45. Kunz RC, Dey M, Ragsdale SW. Characterization of the Thioether Product Formed from the
Thiolytic Cleavage of the Alkyl-Nickel Bond in Methyl-Coenzyme M Reductase. Biochemistry.
2008; 47:2661–2667. [PubMed: 18220418]
46. Noll KM, Donnelly MI, Wolfe RS. Synthesis of 7-mercaptoheptanoylthreonine phosphate and its
activity in the methylcoenzyme M methylreductase system. J Biol Chem. 1987; 262:513–515.
[PubMed: 3100513]
47. Olson KD, McMahon CW, Wolfe RS. Photoactivation of the 2-(methylthio)ethanesulfonic acid
reductase from Methanobacterium. Proc Natl Acad Sci U S A. 1991; 88:4099–4103. [PubMed:
1903534]
48. Zehnder AJ, Wuhrmann K. Titanium (III) citrate as a nontoxic oxidation-reduction buffering
system for the culture of obligate anaerobes. Science. 1976; 194:1165–1166. [PubMed: 793008]
49. Gunsalus RP, Romesser JA, Wolfe RS. Preparation of coenzyme M analogues and their activity in
the methyl coenzyme M reductase system of Methanobacterium thermoautotrophicum.
Biochemistry. 1978; 17:2374–2377. [PubMed: 98178]
50. Otwinowski Z, Minor W. Processing of X-ray Diffraction Data Collected in Oscillation Mode.
Methods in Enzymology: Macromolecular Crystallography, part A. 1997; 276:307–326.
51. The CCP4 suite: programs for protein crystallography. Acta Crystallogr D Biol Crystallogr. 1994;
50:760–763. [PubMed: 15299374]
52. Emsley P, Cowtan K. Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol
Crystallogr. 2004; 60:2126–2132. [PubMed: 15572765]
53. Ebner S, Jaun B, Goenrich M, Thauer RK, Harmer J. Binding of coenzyme B induces a major
conformational change in the active site of methyl-coenzyme M reductase. J Am Chem Soc. 2010;
132:567–575. [PubMed: 20014831]
54. Goenrich M, Mahlert F, Duin EC, Bauer C, Jaun B, Thauer RK. Probing the reactivity of Ni in the
active site of methyl-coenzyme M reductase with substrate analogues. J Biol Inorg Chem. 2004;
9:691–705. [PubMed: 15365904]
Cedervall et al.
Page 14
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 1.
The active site and substrate channel in the MCRox1-silent crystal structure (PDB code 1hbn)
(9). Coenzyme F430, CoBSH and HSCoM are drawn as stick colored by atom (carbon: dark
grey). The nickel is displayed as a green sphere, water as a red sphere. Interactions are
drawn as dashed lines, and the corresponding distance is indicted in Angstroms (Å). The
path of the substrate channel was defined in the absence of F430, CoBSH, HSCoM and
water, with the surface closest to the viewer cut away. The figure was generated using
PyMOL (http://www.pymol.org).
Cedervall et al.
Page 15
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 2.
Drawing of CoBSH analogues; (A) N-5-mercaptopentanoylthreonine phosphate (CoB5SH);
(B) N-6-mercaptohexanoylthreonine phosphate (CoB6SH); (C) N-8-
mercaptooctanoylthreonine phosphate (CoB8SH); (D) N-9-mercaptononanoylthreonine
phosphate (CoB9SH).
Cedervall et al.
Page 16
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 3.
The active sites and substrate channels of the MCR crystal structures.; (A) MCRCoB5SH; (B)
MCRCoB6SH; (C) MCRHSCoM; (D) MCRCoB8SH; (E) MCRCoB9SH. 2Fo-Fc electron density
map around the CoBXSH analogues, waters in the CoBSH binding part of the channel and
the acetate ion (contoured at 1σ) is shown as a blue mesh. The protein is drawn as cartoon.
CoBXSH and acetate are drawn as stick and colored by atom (carbon: CoB5SH orange;
CoB6SH pale yellow; CoB8SH light blue; CoB9SH magenta; acetate white). Coenzyme F430
and HSCoM are drawn as stick colored by atom (carbon: F430 dark grey; HSCoM medium
grey). The nickel is displayed as a green sphere, and waters as red spheres. The figure was
generated using PyMOL (http://www.pymol.org/).
Cedervall et al.
Page 17
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 4.
Overlay of CoBSH (from PDB code 1hbn) and the different CoBSH analogues. CoBXSH are
drawn as stick with the thiol represented by a sphere, and colored CoB5SH orange; CoB6SH
pale yellow; CoBSH light green; CoB8SH light blue; CoB9SH magenta. The protein is
drawn as cartoon. Coenzyme F430 and HSCoM are drawn as stick colored by atom (carbon:
F430 dark grey; HSCoM medium grey). The nickel is displayed as a green sphere. The figure
was generated using PyMOL (http://www.pymol.org/).
Cedervall et al.
Page 18
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 5.
Hydrogen bonding diagram for the water structure modeled in MCRHSCoM. The water
molecules are named as in Figure 3C (W1–W9); WA, WB and WC are water molecules that
are present in all structures (i.e. in concert with the substrate CoBSH and the CoBSH
analogues). Interactions between surrounding residues and the water molecules are drawn as
dashed lines, and the corresponding distance is indicated in Angstroms (Å).
Cedervall et al.
Page 19
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Figure 6.
Stereo image of the annulus of aromatic amino acids distal of coenzyme F430. The protein is
drawn as cartoon with the side-chains of the aromatic residues drawn as white stick.
CoBSH, (from PDB code 1hbn (9)), CoB8SH and CoB9SH are drawn as stick with the thiols
represented by spheres, and colored CoBSH light green; CoB8SH light blue; CoB9SH
magenta. Coenzyme F430 and HSCoM are drawn as stick colored by atom (carbon: F430
dark grey; HSCoM medium grey). The nickel is displayed as a green sphere. The figure was
generated using PyMOL (http://www.pymol.org/).
Cedervall et al.
Page 20
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Scheme 1.
Reaction catalyzed by methyl-coenzyme M reductase
Cedervall et al.
Page 21
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Scheme 2.
Two of the proposed catalytic mechanisms for methyl-coenzyme M reductase; (A)
mechanism I; (B) mechanism II.
Cedervall et al.
Page 22
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Cedervall et al.
Page 23
Table 1
X-ray Data Collection, Processing and Refinement Statistics
Data collection and processing statistics
Name of data set
MCRCoB5SH
MCRCoB6SH
MCRHSCoM
MCRCoB8SH
MCRCoB9SH
Measured reflections
1969388
2427498
1440665
1160543
1425506
Unique reflections
553755
446253
405349
211803
401701
Resolution (Å) a
50.0–1.30 (1.35–1.30)
50.0–1.40 (1.45–1.40)
50.0–1.45 (1.50–1.45)
50.0–1.80 (1.86–1.80)
50.0–1.45 (1.50–1.45)
Completeness (%) a
97.1 (78.1)
99.9 (100.0)
99.5 (99.7)
99.8 (100.0)
98.1 (95.4)
R-sym (%) a,b
5.5 (32.9)
7.3 (44.7)
6.2 (44.0)
8.4 (47.7)
5.6 (42.5)
I/σI a
22.3 (3.6)
20.4 (4.0)
20.2 (3.2)
21.8 (3.9)
24.3 (3.2)
Space group
P21
P21
P21
P21
P21
Refinement and model building statistics
Resolution (Å) a
20.49–1.30 (1.33–1.30)
19.89–1.40 (1.44–1.40)
20.15–1.45 (1.49–1.45)
19.93–1.80 (1.84–1.80)
20.07–1.45 (1.48–1.45)
No. of reflection in working set a
525817 (30239)
423854 (25833)
384868 (25791)
201128 (11193)
381474 (23611)
No. of reflection in test set a
27777 (1576)
22348 (1331)
20362 (1319)
10625 (557)
20163 (1210)
R-work (%) c
14.32
13.04
13.47
14.95
13.58
R-free (%) d
16.56
15.53
16.22
19.54
16.44
ESU (Å) R-work/R-free
0.044/0.046
0.049/0.051
0.056/0.059
0.121/0.119
0.057/0.060
No. protein atoms
20087
19960
20265
19750
20036
No. coenzyme atoms
218
220
180
224
272
No. ligand atoms
37
62
52
26
49
No. water molecules
2443
2352
2516
1893
2432
RMS
bond lengths (Å)
0.033
0.033
0.032
0.028
0.032
bond angles (deg.)
2.693
2.625
2.468
2.059
2.549
Ramachandran plot (%)
favored
97.8
97.5
97.6
97.2
97.7
allowed
2.1
2.4
2.3
2.7
2.1
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Cedervall et al.
Page 24
disallowed
0.1
0.1
0.1
0.1
0.1
Average B-factor (Å2)
protein
12.42
13.35
12.12
17.22
12.73
coenzymes
8.20
9.24
7.25
11.24
8.27
ligands
31.95
35.48
28.29
33.76
32.92
waters
22.95
24.89
23.85
26.79
24.09
over all
13.54
14.57
13.40
18.02
13.93
Occupancy of HSCoM per active site (%)e
90/90
50/50
100/100
90/90
90/85
Occupancy of CoBSH per active site (%) e
50/50
50/50
30/30
50/50
40/40
CoBSH analogue, occupancy per active site (%) e
CoB5SH, 50/50
CoB6SH, 50/50
CoB8SH, 50/50
CoB9SH, 60/60
Other molecule, occupancy per active site (%) e
Acetate, 70/70
aValues in brackets correspond to the highest resolution shell.
bR-sym = ΣhklΣNj=1|Ihkl−Ihkl(j)|/ΣhklN*Ihkl, sum over all reflections and all observations N, with Ihkl(j) intensity of the jth observation of reflection hkl and Ihkl mean intensity of the reflection hkl.
cR-work = Σ||Fo| − |Fc||/Σ|Fo|, where |Fo| = observed structure factor amplitude and |Fc| = calculated structure factor amplitude.
dR-free, R-factor based on 5% of the data excluded from refinement.
eOccupancy of model in each of the two crystallographically independent active sites in the ASU
Biochemistry. Author manuscript; available in PMC 2011 September 7.
NIH-PA Author Manuscript
NIH-PA Author Manuscript
NIH-PA Author Manuscript
Cedervall et al.
Page 25
Table 2
Distances from analogue thiols.
CoBXS - SCoM distance (Å)
CoBXS - Ni distance (Å)
CoB5SH
7.11/7.11a
9.30/9.30
CoB6SH
6.26/6.26
8.70/8.70
CoB7SH (substrate) b
6.37/6.39
8.73/8.77
CoB8SH
3.75/3.78
6.16/6.17
CoB9SH
3.71/3.68
5.96/5.91
aDistances in the two crystallographically independent active sites in the ASU
bDistances in the 1.16 Å resolution MCRox1-silent structure (PDB code 1hbn)(33)
Biochemistry. Author manuscript; available in PMC 2011 September 7.
|
3M2U
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry using Coenzyme B Analogues
| "Structural Insight into Methyl-Coenzyme M Reductase Chemistry\nusing Coenzyme B Analogues,†,‡\n(...TRUNCATED)
|
3M2V
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry using Coenzyme B Analogues
| "Structural Insight into Methyl-Coenzyme M Reductase Chemistry\nusing Coenzyme B Analogues,†,‡\n(...TRUNCATED)
|
3M30
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry using Coenzyme B Analogues
| "Structural Insight into Methyl-Coenzyme M Reductase Chemistry\nusing Coenzyme B Analogues,†,‡\n(...TRUNCATED)
|
3M31
|
Structure of the C150A/C295A mutant of S. cerevisiae Ero1p
| "Steps in reductive activation of the\ndisulfide-generating enzyme Ero1p\nNimrod Heldman,1 Ohad Vons(...TRUNCATED)
|
3M32
|
Structural Insight into Methyl-Coenzyme M Reductase Chemistry using Coenzyme B Analogues
| "Structural Insight into Methyl-Coenzyme M Reductase Chemistry\nusing Coenzyme B Analogues,†,‡\n(...TRUNCATED)
|
End of preview. Expand
in Data Studio
primary_citation_with_pmcid.jsonl
This dataset links PDB protein structures with their corresponding primary ciatation text content.
Format
Each line is a JSON object:
{
"protein_name": "9HCG",
"structure_title": "Mouse mitoribosome large subunit assembly intermediate bound to NSUN4, MTERF4, and mt-RNAs",
"main_text": "..."
}
- Downloads last month
- 12