Abstract
FeS clusters are versatile cofactors of a variety of proteins, but the mechanisms of their biosynthesis are still unknown. The cystine C-S lyase from Synechocystis has been identified as a participant in ferredoxin FeS cluster formation. Herein, we report on the crystal structure of the lyase and of a complex with the reaction products of cystine cleavage at 1.8- and 1.55-Å resolution, respectively. The sulfur-containing product was unequivocally identified as cysteine persulfide. The reactive persulfide group is fixed by a hydrogen bond to His-114 in the center of a hydrophobic pocket and is thereby shielded from the solvent. Binding and stabilization of the cysteine persulfide represent an alternative to the generation of a protein-bound persulfide by NifS-like proteins and point to the general importance of persulfidic compounds for FeS cluster assembly.
Iron sulfur (FeS) clusters are found ubiquitously as functional units in numerous proteins that play pivotal roles in redox, catalytic, or regulatory processes such as oxidative phosphorylation, photosynthesis, nitrogen fixation, or environmental sensing (1, 2). FeS clusters now rank with cofactors as hemes and flavins in pervasive occurrence and multiplicity of function. Despite the continuing interest in FeS proteins, little is known about the mechanisms by which the FeS moieties are formed and introduced into the proteins in vivo.
In Azotobacter vinelandii, genetic studies on the biosynthesis of the nitrogenase metalloclusters unveiled NifS as an essential factor for the assembly of the [4Fe-4S] cluster of the iron protein subunit (3). NifS is a pyridoxal 5′-phosphate (PLP)-dependent l-cysteine desulfurase producing alanine and elemental sulfur (4) and seems to play a general role in the mobilization of sulfur for FeS cluster biosynthesis as deduced from its widespread appearance (5–7). The putative active intermediate for the delivery of S0 sulfur is an enzyme-bound persulfide that is formed by nucleophilic attack of an active-site cysteine on the PLP-bound substrate cysteine (8).
Recently, a distinct l-cysteine/l-cystine C-S lyase (named C-DES) has been characterized by Kessler and coworkers (9, 10). C-DES was isolated selectively from a Synechocystis PCC 6714 lysate by its capacity to direct FeS cluster insertion into Synechocystis apoferredoxin in vitro. C-DES shares limited sequence homology to the NifS family of proteins (10–12). Remarkably, none of the three NifS-like proteins encoded by the Synechocystis genome (13) showed up in the holoferredoxin formation assay. Contrary to the NifS family, C-DES is missing the largely conserved active-site cysteine leading to the observed insensitivity toward thiol alkylating agents. Furthermore, C-DES produces ammonium pyruvate and sulfide from cysteine instead of alanine and sulfur and has a strong substrate preference for cystine over cysteine. By using desaminocystine as an analogue of the cystine substrate, formation of a substrate-based persulfide could be demonstrated (10). A common theme of sulfur mobilization for the purpose of FeS cluster assembly therefore seems to be the occurrence of a persulfidic intermediate. This intermediate can be observed directly in the x-ray structure of the product complexed form of C-DES presented in this report.
Materials and Methods
Crystallization, Data Collection, and Multiwavelength Anomalous Diffraction Phasing of C-DES.
Crystallization of C-DES was carried out with an N-terminal modified derivative in which the 11 N-terminal residues of the wild-type enzyme MADPVNLIPDR were exchanged by the octapeptide MTMITPSL. For consistency, the numbering of the wild-type protein was transferred to the modified C-DES and is used throughout the article. The N-terminal variant of C-DES was expressed in Escherichia coli as reported (10). Purification of the recombinant protein was accomplished via a three-step procedure that will be published elsewhere. Yellow plate-like crystals were grown at 20°C by using the sitting drop vapor diffusion method. A solution (4 μl) containing 10 mg/ml C-DES in 10 mM 3- (N-morpholino)propanesulfonic acid/NaOH, 10 μM PLP (pH 6.5) was mixed with 2 μl of precipitant solution comprising 27% (vol/vol) polyethylene glycol (molecular weight = 8,000), 0.1 M (NH4)2SO4, and 0.1 M citrate (pH 6.0) and equilibrated against 500 μl of a corresponding reservoir solution. Crystals appeared within 4 days and belonged to space group P212121 with cell constants a = 62.4 Å, b = 65.4 Å, and c = 170.1 Å, containing two monomers per asymmetric unit (solvent content 33%). The structure of the complex with the products of β-cleavage was obtained by using the same conditions as with unliganded C-DES but with 10 mM of cysteine added and yielded orange-colored crystals.
To avoid crystal damage by cooling with nitrogen, C-DES crystals were transferred to a cryobuffer containing 20% (vol/vol) polyethylene glycol (molecular weight = 8,000), 0.1 M (NH4)2SO4, 0.1 M citrate (pH 6.0), and 12% (vol/vol) 2-methyl-2,4-pentanediol. Heavy atom derivative screens were performed by soaking crystals in 40 μl of crystallization buffer containing appropriate amounts of heavy atoms. Data were collected initially at 100 K on our in-house MarResearch (Hamburg, Germany) image plate with graphite monochromated CuKα radiation (λ = 1.5418 Å) from a Rigaku (Tokyo) RU 200 rotating anode operating at 120 mA and 50 kV. For a 2-day soak with 5 mM PtCl4, difference Patterson analyses performed with the ccp4 program suite (14) yielded four well occupied Pt sites.
Diffraction data for the multiwavelength anomalous diffraction experiment (data sets Ptλ1–3), the native (data set E), and the product-complexed C-DES (data set EP) form were measured at cryotemperatures from single crystals at the wiggler beamline BW7A at DORIS (Deutsches Elektronen Synchrotron Hamburg, Germany). For multiwavelength anomalous diffraction phasing, x-ray data were collected on a 300-mm MarResearch imaging plate detector at three wavelengths: λ1 = 0.89 Å as a remote data set, λ2 = 1.0722 Å, and λ3 = 1.0728 Å as maximal f′′ and minimal f′ values of the K shell absorption edge of platinum to maximize anomalous and dispersive contributions. All diffraction data were processed and scaled with the programs denzo and scalepack (15). Refinement of heavy atom parameters and phase calculation were done with sharp (16) leading to a figure of merit of 0.55 for data between 20.0 and 2.35 Å (Table 1). Cycles of solvent flattening (n = 130) were performed with solomon (17), assuming a solvent content of 33%, and resulted in an electron density map of excellent quality in which the complete model of C-DES was built in.
Table 1.
Data collection and phasing | Ptλ1 | Ptλ2 | Ptλ3 | E | EP |
---|---|---|---|---|---|
Wavelength, Å | 0.89 | 1.0722 | 1.0728 | 0.89 | 0.89 |
Resolution, Å | 25.0–2.35 | 25.0–2.35 | 25.0–2.35 | 25.0–1.80 | 25.0–1.55 |
Completeness, % | 95.8 | 99.0 | 93.5 | 93.8 | 93.7 |
Rmerge* | 2.7 | 2.9 | 3.0 | 4.7 | 5.2 |
I/σ(I) | 25.2 | 27.7 | 25.6 | 19.1 | 18.5 |
Last shell | 2.40–2.35 | 2.40–2.35 | 2.40–2.35 | 1.83–1.80 | 1.57–1.55 |
Completeness, % | 92.1 | 98.7 | 92.3 | 89.7 | 87.3 |
Rmerge* | 7.2 | 8.0 | 7.9 | 30.6 | 36.5 |
I/σ(I) | 6.3 | 7.4 | 6.5 | 3.2 | 2.8 |
Phasing (15.0–2.35 Å) | |||||
Phasing power (iso)† | — | 1.21 | 1.72 | ||
Phasing power (ano)‡ | 2.90 | 3.30 | 1.86 | ||
Rcullis (iso)§ | — | 0.75 | 0.68 |
Refinement | |||||
---|---|---|---|---|---|
Data set | Unique reflections | Active atoms | Rcryst/Rfree, %∥ | Average B value, Å2 | rms deviation: bonds, Å/angles, degrees/bonded Bs, Å2 |
E | 61,410 | 6,579 | 19.8 /24.9 | 27.4 | 0.012/1.60/2.43 |
EP | 95,716 | 6,680 | 21.4/25.9 | 25.4 | 0.011/1.58/2.46 |
*Σ|I − 〈I〉|/ΣI.
†Σn|FH|/Σn|E|, where |FH| is the calculated structure factor amplitude of the heavy atom structure and E is the residual lack of closure.
‡Σn|FH"|/Σn|E|, where |FH"| is the anomalous contribution amplitude.
§Σhkl||FPH ± FP| − FH(calc)|/Σhkl|FPH ± FP| for centric reflections and isomorphous contributions.
∥Σ∥Fobs| − |Fcalc∥/Σ|Fobs|, Rfree is the R value calculated with 5% of the data that were not used for the refinement.
Model Building and Refinement.
Conventional crystallographic refinement was carried out in alternating cycles of model building with the program o (18), and energy restrained crystallographic refinement was based on maximum likelihood algorithms with cns (19) by using the parameters of Engh and Huber (20). Bulk solvent, overall anisotropic B-factor corrections, and noncrystallographic restraints were introduced depending on the behavior of the free R factor index. When the refinement proceeded smoothly to an R value of 28%, solvent molecules were introduced at stereochemically reasonable positions with high difference electron densities. During the whole refinement, the R factor decreased from 41.2 to 19.8% (Rfree from 42.3 to 24.9%) for all data between 25.0- and 1.8-Å resolution (Table 1). A similar refinement procedure was used for the EP complex. Several rounds of simulated annealing, positional, and B-factor optimization with cns were alternated with manual refitting with o. The PLP-aminoacrylate derivative and the bound cysteine persulfide were included only in the last round of refinement.
Results and Discussion
Overall Fold.
We solved the crystal structure of an N-terminal-modified, catalytically competent C-DES by multiwavelength anomalous diffraction by using a PtCl4 heavy atom derivative. The fold of C-DES (Fig. 1A) is related to that of other PLP-dependent enzymes of the aminotransferase family (21, 22). Each monomer is constructed by two distinct domains, a small domain comprising the N- and C-terminal parts of the polypeptide chain (residues 1–52 and 298–393) and the large domain formed by the intervening residues containing the cofactor binding site (residues 53–297). The large PLP-binding domain consists of a central, mainly parallel seven-stranded β-sheet (+a, −g, +f, +e, +d, +b, +c) that is surrounded by 10 helices resulting in an overall αβα architecture. The large domain contains furthermore an antiparallel two-stranded β-sheet (+h, −i) that is spatially removed from the bulk of the domain and packed against the second subunit in the active dimer. PLP is bound covalently to Lys-223 at the C-terminal end of the seven-stranded β-sheet in the interdomain cleft. The small domain folds into an αβ sandwich. In the center of the domain, a parallel two-stranded β-sheet (N-C) is combined with an antiparallel four-stranded β-sheet (A-B-E-D). The connecting helices are packed against the face opposite to the large domain. β-strand N and helices 1 and 2 originate from the N-terminal segment.
Although C-DES behaved as a monomer in gel filtration experiments (9), the crystal structure reveals the presence of an α2 dimer. As compared with related PLP enzymes, the highly polar contact surface is remarkably large. Much of the solvent-accessible surface area (18.4%; 2,950 Å2) of each subunit is buried on dimerization. Most of the intersubunit contacts are clustered around the active sites of the individual monomers (Fig. 1B) involving helices 2, 3, 5, 6, 11, and 12 and strands h and i and their associated loops. The key elements in dimer formation are helix 2, interacting in a two-helix bundle with its counterpart in the other monomer, and the exposed loop structure contributing residues 242–281. This extended loop acts as a molecular dimerization clamp that protrudes to the active site of the neighboring subunit where it interacts with the cofactor's phosphate group.
Because no structure of a NifS-like protein was available up to now, the DALI algorithm (25) was used in the search for structural homologues; 8-amino-7-oxononanoate synthase, serine hydroxymethyltransferase, phosphoserine aminotransferase, and aspartate aminotransferase were identified as the most similar structures in the database with z scores of 27.9, 27.7, 27.6, and 25.2, respectively. The multiple structural alignment of the four cited enzymes with C-DES resulted in 310 ± 10 Cαs, aligned with an rms deviation of 3.5 ± 0.4 Å, and indicated that none of the deposited PLP enzyme structures have structural equivalents to C-DES residues 45–60, bridging helix 2 and helix 3, and to residues 248–274, forming the molecular dimerization clamp. Both of these C-DES characteristic segments form part of the active site wall in the neighboring subunit and carry catalytically important residues.
Active Site.
The active site of C-DES is located near the center of the monomer–monomer interface. Contrary to other PLP enzymes, the active-site crevice is composed equally of residues from both subunits in the dimer. The PLP is bound tightly to the protein by several specific interactions (Fig. 2), most notably the Schiff-base linkage to Lys-223. The phosphate group is anchored by a total of seven hydrogen bonds of which two originate from the neighboring subunit (Trp-251* and Thr-276*). Charge stabilization of the phosphate is provided by the imidazole side chain of His-222, preceding the PLP-binding lysine, and the positive macrodipole of helix 5 (residues 88–99). The pyridine ring is sandwiched between His-114 and Ala-199; its nitrogen forms a hydrogen bond/salt bridge to Asp-197, and the hydroxyl group at C3 is located close to Gln-200. All these interactions determine the electronic distribution in the delocalized π-system of the cofactor and ensure its electron-sink character capable of stabilizing the obligatory developing carbanionic intermediate.
The substrate spectrum of C-DES is confined to compounds with at least one l-cysteinyl moiety and various S substituents of different size and electronegativity (10). Accordingly, the substrate-binding pocket of C-DES is constructed as a highly polar canyon, 28 Å × 15 Å wide and 10 Å deep, that accommodates amino acid substrates of various lengths. The general electrostatic architecture of the active-site crevice is complementary to the bipolar cystine molecule with a central hydrophobic region enclosed by two highly polar areas. The polar regions, termed distal and proximal with respect to the location of the cofactor, are built up separately by the individual monomers in the dimer, whereas the hydrophobic pocket is wedged between them as a hybrid of residues from both subunits. The wall of the distal half of the pocket is lined primarily with polar residues showing a preponderance of basic residues (Arg-68*, Arg-252*, and Lys-270* besides Asn-57*, Gln-61*, Thr-276*, and Tyr-279*); the central hydrophobic area is constructed around His-114 by residues Phe-52*, Trp-251*, Tyr-256*, Pro-115, and Trp-168, and the proximal half is built up by the N-terminal loop Phe-25–Gly-26–Gly-27–Gln-28, Gln-200, and the two arginines Arg-360 and Arg-369. The pronounced hydrogen-bonding network that protrudes from the proximal PLP-half of the crevice via residues 25–28 into the distal half creates a sticky polar surface by which caught substrates are funneled to the PLP cofactor.
Formation and Stabilization of Cysteine Persulfide.
Crystals of C-DES containing a stable PLP aminoacrylate adduct were obtained by cocrystallizing C-DES and cysteine at pH 6.1 under aerobic conditions, i.e., in the presence of a small fraction of cystine. Obviously, reverse transaldimination becomes rate-limiting at low pH, which allows this intermediate to be studied in the crystals as well as in solution, where the adduct is characterized by a broad absorption maximum at 490 nm (T.C., unpublished results). The modeled PLP-aminoacrylate fits perfectly to the observed density and indicates an extension of the PLP π-electron system to Cβ, which is consistent with the low-energy absorption at 490 nm. Most notably, the elimination product of cystine β-cleavage, cysteine persulfide, was identified in the C-DES active site (Fig. 2a).
Its α-carboxylate group is fixed by the guanidinium group of Arg-360, and its persulfidic tail is protruding to His-114. The short distance between Sδ and His-114–NE1 (2.71 Å) suggests a strong hydrogen bond, by which the labile persulfide group is dragged into a small hydrophobic pocket built up by Trp-251*, Tyr-256*, Pro-115, and Trp-168. Identification of the cystine β-cleavage products from a cocrystallization experiment with only partially oxidized cysteine corroborates the previously reported strong preference of C-DES to bind cystine instead of cysteine (10). Furthermore, the tertiary structure of the complexed enzyme implies that a large-scale positional rearrangement resulting in an active-site closure does not occur during the catalytic cycle. Only the side chains of the four active-site residues His-114, Pro-115, Trp-168, and Arg-360 reorient on product binding (Fig. 2b).
The active-site structures of native and product-complexed C-DES suggest a molecular reaction mechanism for the generation of cysteine persulfide (Fig. 3). In the Michaelis complex, the α- and the distal carboxylate groups form double hydrogen-bonded ion pairs with Arg-369 and Arg-360, respectively, which are both optimized by coplanar arrangement. Additional anchoring of the α-carboxylate group is achieved by polar interactions with Gly-27 and Gln-200. Furthermore, the aromatic rings of Phe-52* and Trp-251* are in van-der-Waals contact to the substrate disulfide bond and enclose it within the pocket. Once the substrate is bound in proper conformation, transaldimination should be initiated by His-114, which is the most probable candidate to deprotonate the substrate's amino group for nucleophilic attack of the internal aldimine. The later steps are catalyzed by Lys-223 whose ɛ-amino group can approach Cα, Sγ, or C4′ leading to Cα proton abstraction, protonation of the leaving group, or reverse transaldimination, respectively.
Obviously, the C-DES-catalyzed formation of cysteine-persulfide represents an alternative pathway of S0 production as compared with the sulfuration of an active-site cysteine in NifS proteins. The fact that C-DES stabilizes this labile species strikingly indicates the functional significance of the cysteine persulfide product as an activated sulfur source. However, the acceptor of the activated sulfur during FeS cluster synthesis and the mode of sulfur transfer into the FeS cluster have to await identification. At least the remarkable size of the C-DES active-site groove gives the impression of a docking site for another molecule (Fig. 1B). This molecule may be apoferredoxin during the in vitro synthesis of the [2Fe-2S] cluster, although biochemical proof for such an interaction is lacking up to now. Alternatively, some other acceptor molecule could be involved whose necessity may be bypassed in the in vitro system: comparison with the NifS/IscS/Nfs1p-dependent pathways in Azotobacter (7) and yeast (26–28) would suggest some IscU- or IscA-related polypeptide. Another possibility would be the uncharacterized ligand to the postulated cluster precursor that is exported from mitochondria in yeast (27). However, no IscU homologue is found in the Synechocystis genome, a finding that excludes this candidate. It should be noted that none of the three iscS-like genes and neither of the two iscA-like genes of Synechocystis are organized in an iscSUA operon as found for several other prokaryotes (7). Therefore, the FeS cluster biosynthesis in Synechocystis and probably also in chloroplasts may use some variant pathway.
Acknowledgments
We thank E. Pohl for excellent assistance during collection of the multiwavelength anomalous diffraction data at the EMBL beamline BW7A. D.K. is grateful to J. Knappe for constant interest and support, to A. Becker for a quality assessment of the first crystals obtained, to I. Leibrecht for technical assistance, and to the Deutsche Forschungsgemeinschaft for support by a Habilitandenstipendium. C.S. acknowledges financial support by Boehringer Ingelheim Fonds.
Abbreviations
- C-DES
cystine C-S lyase
- PLP
pyridoxal 5′-phosphate
Footnotes
Data deposition: The atomic coordinates have been deposited in the Protein Data Bank, www.rcsb.org [PDB ID codes ELQ (native c-DES) and ELU (product-complexed c-DES)].
References
- 1.Beinert H, Holm R H, Münck E. Science. 1997;277:653–659. doi: 10.1126/science.277.5326.653. [DOI] [PubMed] [Google Scholar]
- 2.Johnson M K. Curr Opin Chem Biol. 1998;2:173–181. doi: 10.1016/s1367-5931(98)80058-6. [DOI] [PubMed] [Google Scholar]
- 3.Jacobson M R, Cash V L, Weiss M C, Laird N F, Newton W E, Dean D R. Mol Gen Genet. 1989;219:49–57. doi: 10.1007/BF00261156. [DOI] [PubMed] [Google Scholar]
- 4.Zheng L, White R H, Cash V L, Jack R F, Dean D R. Proc Natl Acad Sci USA. 1993;90:2754–2758. doi: 10.1073/pnas.90.7.2754. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Mihara H, Kurihara T, Yoshimura T, Soda K, Esaki N. J Biol Chem. 1997;272:22417–22424. doi: 10.1074/jbc.272.36.22417. [DOI] [PubMed] [Google Scholar]
- 6.Nakai Y, Yoshihara Y, Hayashi H, Kagamiyama H. FEBS Lett. 1998;433:143–148. doi: 10.1016/s0014-5793(98)00897-7. [DOI] [PubMed] [Google Scholar]
- 7.Zheng L, Cash V L, Flint D H, Dean D R. J Biol Chem. 1998;273:13264–13272. doi: 10.1074/jbc.273.21.13264. [DOI] [PubMed] [Google Scholar]
- 8.Zheng L, White R H, Cash V L, Dean D R. Biochemistry. 1994;33:4714–4720. doi: 10.1021/bi00181a031. [DOI] [PubMed] [Google Scholar]
- 9.Leibrecht I, Kessler D. J Biol Chem. 1997;272:10442–10447. doi: 10.1074/jbc.272.16.10442. [DOI] [PubMed] [Google Scholar]
- 10.Lang T, Kessler D. J Biol Chem. 1999;274:189–195. doi: 10.1074/jbc.274.1.189. [DOI] [PubMed] [Google Scholar]
- 11.Ouzounis C, Sander C. FEBS Lett. 1993;322:159–164. doi: 10.1016/0014-5793(93)81559-i. [DOI] [PubMed] [Google Scholar]
- 12.Mehta P K, Christen P. Eur J Biochem. 1993;211:373–376. doi: 10.1111/j.1432-1033.1993.tb19907.x. [DOI] [PubMed] [Google Scholar]
- 13.Kaneko T, Sato S, Kotani H, Tanaka A, Asamizu E, Nakamura Y, Miyajima N, Hirosawa M, Sugiura M, Sasamoto S, et al. DNA Res. 1996;3:109–136. doi: 10.1093/dnares/3.3.109. [DOI] [PubMed] [Google Scholar]
- 14.Collaborative Computational Project No. 4. Acta Crystallogr D. 1994;50:760–763. [Google Scholar]
- 15.Otwinowski Z, Minor W. Methods Enzymol. 1997;276:307–326. doi: 10.1016/S0076-6879(97)76066-X. [DOI] [PubMed] [Google Scholar]
- 16.La Fortelle E D, Irwin J J, Bricogne G. Crystallogr Comput. 1997;7:1–9. [Google Scholar]
- 17.Abrahams J P, Leslie A G W. Acta Crystallogr D. 1996;52:30–42. doi: 10.1107/S0907444995008754. [DOI] [PubMed] [Google Scholar]
- 18.Jones T A, Zou J Y, Cowan S W, Kjeldgaard M. Acta Crystallogr A. 1991;47:110–119. doi: 10.1107/s0108767390010224. [DOI] [PubMed] [Google Scholar]
- 19.Brunger A T, Adams P D, Clore G M, DeLano W L, Gros P, Grosse-Kunstleve R W, Jiang J S, Kuszewski J, Nilges M, Pannu N S, et al. Acta Crystallogr D. 1998;54:905–921. doi: 10.1107/s0907444998003254. [DOI] [PubMed] [Google Scholar]
- 20.Engh R A, Huber R. Acta Crystallogr A. 1991;47:392–400. [Google Scholar]
- 21.John R A. Biochim Biophys Acta. 1995;1248:81–96. doi: 10.1016/0167-4838(95)00025-p. [DOI] [PubMed] [Google Scholar]
- 22.Jansonius J N. Curr Opin Struct Biol. 1998;8:759–769. doi: 10.1016/s0959-440x(98)80096-1. [DOI] [PubMed] [Google Scholar]
- 23.Kraulis P J. J Appl Crystallogr. 1991;24:946–950. [Google Scholar]
- 24.Merritt E A, Murphy M E P. Acta Crystallogr D. 1994;50:869–873. doi: 10.1107/S0907444994006396. [DOI] [PubMed] [Google Scholar]
- 25.Holm L, Sander C. J Mol Biol. 1993;233:123–138. doi: 10.1006/jmbi.1993.1489. [DOI] [PubMed] [Google Scholar]
- 26.Strain J, Lorenz C R, Bode J, Garland S, Smolen G A, Tall D T, Vickery L E, Culotta V C. J Biol Chem. 1998;273:31138–31144. doi: 10.1074/jbc.273.47.31138. [DOI] [PubMed] [Google Scholar]
- 27.Kispal G, Csere P, Prohl C, Lill R. EMBO J. 1999;18:3981–3989. doi: 10.1093/emboj/18.14.3981. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Schilke B, Voisine C, Beinert H, Craig E. Proc Natl Acad Sci USA. 1999;96:10206–10211. doi: 10.1073/pnas.96.18.10206. [DOI] [PMC free article] [PubMed] [Google Scholar]