X-ray Structure of the [FeFe]-Hydrogenase Maturase HydE from Thermotoga maritima*

Maturation of the [FeFe]-hydrogenase active site depends on at least the expression of three gene products called HydE, HydF, and HydG. We have solved the high resolution structure of recombinant, reconstituted S-adenosine-l-methionine-dependent HydE from Thermotoga maritima. Besides the conserved [Fe4S4] cluster involved in the radical-based reaction, this HydE was reported to have a second [Fe4S4] cluster coordinated by three Cys residues. However, in our crystals, depending on the reconstitution and soaking conditions, this second cluster is either a [Fe2S2] center, with water occupying the fourth ligand site or is absent. We have carried out site-directed mutagenesis studies on the related HydE from Clostridium acetobutylicum, along with in silico docking and crystal soaking experiments, to define the active site region and three anion-binding sites inside a large, positive cavity, one of which binds SCN- with high affinity. Although the overall triose-phosphate isomerase-barrel structure of HydE is very similar to that of biotin synthase, the residues that line the internal cavity are significantly different in the two enzymes.

Two classes of microbial enzymes, the [NiFe]-and [FeFe]hydrogenases, catalyze the reversible oxidation of molecular hydrogen according to the reaction, H 2 N 2H ϩ ϩ 2e Ϫ . A third class, the [Fe] (formally " [FeS] cluster-free")-hydrogenases, heterolytically split hydrogen into H Ϫ and H ϩ and transfer the former to their substrate (1). Although these proteins are phylogenetically unrelated, they share the presence of one or two iron ions coordinated by CN Ϫ and/or CO ligands (2-4) (Scheme 1). Overexpression of hydrogenases in heterologous systems leads to inactive proteins due to the absence of properly assembled active sites (5,6). This observation emphasizes the requirement of maturation machineries in order to obtain active enzymes for each of the three classes. Nothing is known about the assembly of the active site FeCO 2 unit in [Fe]-hydrogenases for which a structure has been recently reported. 3 The maturation system of [NiFe]-hydrogenase is by far the most studied and the best understood, although many aspects of the process remain still unresolved (for an extensive review, see Böck et al. (7)). It has been established that the carbamoylphosphate is a precursor that reacts with HypE, forming a -Cys-S-CN intermediate, which, in turn, transfers CN Ϫ to the prospective active site iron ion. Carbon isotopic labeling experiments indicate a different route for active site CO synthesis (8), but its nature has not yet been elucidated (9,10). Once the FeCO(CN) 2 unit is put together, the GTPase HypB mediates nickel binding to the conserved CXXC motif of the C-terminal region of the large subunit, a specific nickel-containing endopeptidase cleaves an amino acid stretch at the C-terminal end, and the active site is fully assembled, deeply buried in the [NiFe]-hydrogenase structure (7).
It has been known for over 20 years that overexpression of the structural Desulfovibrio vulgaris [FeFe]-hydrogenase genes in Escherichia coli, which lacks an endogenous [FeFe]-hydrogenase, produces inactive enzyme, devoid of the active site [FeFe] center (5) (Scheme 1B). Indeed, very recently, three genes, named hydE, hydF, and hydG, have been implicated in [FeFe]-hydrogenase maturation (11) and identified as necessary, and possibly sufficient, to produce active Clostridium acetobutylicum [FeFe]-hydrogenase in E. coli (11)(12)(13). The products of these three maturase genes are thought to be able to synthesize and/or incorporate CO, CN Ϫ , and the small active site bridging molecule, possibly a dithiomethylamine (DTN) 4 (14 -16), and to participate in the correct assembly and insertion of the active site [FeFe] center into the [FeFe]-hydrogenase (11). Amino acid sequence analyses and functional tests show that HydF is a [Fe 4 S 4 ] cluster-containing GTPase (17), which, by analogy with HypB in [NiFe]-hydrogenases (7), may be involved in the active site [FeFe] subcluster insertion (Scheme 1B). The specific functions of HydE and HydG in hydrogenase maturation also remain to be established. Both belong to the recently identified superfamily of S-adenosyl-L-methionine (AdoMet) radical enzymes (18,19). Typically, AdoMet radical enzymes use an invariant reduced [Fe 4 S 4 ] cluster to cleave the cofactor and produce a highly reactive and versatile 5Ј-deoxyadenosyl radical species (5Ј-Ado ⅐ ) (20). Especially relevant to the synthesis of DTN is the sulfur insertion to normally inert carbon atoms, as catalyzed by BioB (biotin synthase), LipA (lipoate synthase), and MiaB (21)(22)(23). These three AdoMet radical proteins are thought to use an additional [FeS] cluster as the sulfur donor (24). Prior to their identification as [FeFe]-hydrogenase maturases, hydE genes were annotated as coding for biotin synthases (and they still are in most data bases), because their respective protein sequences share over 40% similarity. Because HydE from Thermotoga maritima (TmHydE) has a second [FeS] cluster (25), it was tempting to suggest that it is involved in the biosynthesis of the small bridging active site putative DTN (Scheme 1B). However, as will be further discussed below, the cysteine residues that coordinate the second [FeS] cluster are not conserved in all hydE gene sequences (7). Alternatively, DTN could be synthesized by HydG, which has also been proposed to bind an additional [FeS] cluster (25), necessary for hydrogenase maturation (12).
Here, we report two high resolution x-ray structures of TmHydE, with and without an additional [Fe 2 S 2 ] cluster, respectively. By monitoring the effect on recombinant hydrogenase activity of amino acid substitutions in HydE, we have identified several putative key functional residues in the latter. In addition, through molecular docking in silico and small ions binding in the crystals, we have defined a minimal substratebinding region including a plausible carboxylate recognition site. Finally, we have identified two additional anion-binding sites, one found halfway between the active site and the bottom of the protein ␤-barrel and another one near it that binds thiocyanate with very high affinity.

EXPERIMENTAL PROCEDURES
Crystallization-The protein was purified, and the [FeS] clusters were subsequently reconstituted as previously described (25). Prior to crystallization, the protein was incubated with 2 mM AdoMet (an 8-fold molar excess) and 6 mM CHAPSO. Hanging drops were prepared by mixing 2 l of a TmHydE solution at 10 mg/ml in 20 mM Tris, pH 8.0, 200 mM NaCl with 2 l of a solution containing 30% polyethylene glycol 4000, 100 mM Tris, pH 8.0, and 200 mM lithium sulfate and were kept in an anaerobic glove box (Jacomex) at 20°C. Crystals that grew in about 2 weeks were cryoprotected using a solution containing 18% glycerol, 6 mM CHAPSO, 50 mM Tris, pH 8, 100 mM lithium sulfate, 30% polyethylene glycol. Anaerobic flash cooling of the crystals was carried out inside the glove box as previously described (26).
Data Collection and Phasing-All of the x-ray data sets were collected at the European Synchrotron Radiation Facility ( Table 1). The structure was solved by the single wavelength anomalous dispersion (SAD) method using the iron and sulfur atoms present in the native protein. In order to minimize the anomalous signal errors, a total of 290 frames with ⌬ ϭ 1°were collected at ϭ 1.71 Å. Data were processed using MOSFLM (27), scaled with SCALA (4), and reduced using TRUNCATE (28). Initial anomalous scattering atoms positions were obtained from the F3 data set using the SAD method with SHELXD (29). The resulting model corresponded to four iron atoms and one sulfur atom. These sites were then refined using SHARP (30); the final anomalous scattering model consisted of six iron and 21 sulfur atoms. Subsequent refinement, phasing, and solvent flattening gave maps that were of good enough quality to allow the structure determination. Subsequent data sets were processed with XDS (31), and special care was taken to keep the same set of reflections for the R free calculation in all cases.
Structure Determination-The structure of TmHydE was solved by the SAD method using x-ray data collected from native crystals at the high energy side of the iron absorption edge (despite the significant amino acid sequence similarities between TmHydE and BioB from E. coli, no molecular replacement solution was found when using the latter as a search model). Iron atom positions were determined from Patterson maps using SHELXD (29) and subsequently refined with SHARP (30). The resulting phases, extended to 1.77 Å resolution, were of good enough quality to build the initial model.
Model Building and Refinement-Ten cycles of automated building with ARP/wARP (32) and refinement using REFMAC (33) at 1.35 Å resolution resulted in a model containing 312 of a total of 346 residues placed in density with the correct sequence. After several cycles of manual model building using Coot (34), the model was completed by adding residues 2-17, 297-306, and 346 -347, the FeS clusters, the AdoMet/ AdoHCys, the Cl Ϫ ions, five detergent molecules, one glycerol, over 400 water molecules, and the modeling of alternative conformations of several side chains. This model was refined using REFMAC (33), generating the final atomic coordinate set. The relative occupancies of the second [FeS] cluster atoms were determined based on B factor refinement and comparison with the relative occupancies of the alternative conformations of the side chain of Cys 322 .
Soaking Experiments-Soaks of TmHydE crystals with different small molecules were carried out using cryosolutions containing 50 mM dithiothreitol, and 5-100 mM concentration of one of the following molecules: formamide, formate, thiocya- nate, cyanate, cyanide, carbamoylphosphate, phosphate, cysteine, pyruvate, asparagine, succinamate, maleamate, glycine, acetate, and lactate. These molecules were considered to be either plausible substrates, products, precursors, or their fragments. Dimethylamine could be a precursor of the small bridging active site molecule DTN, whereas propanedithiol corresponds to an analog of it. Glycine has been proposed to be a precursor for both CO and CN Ϫ ligands (35), and carbamoylphosphate is the source of CN Ϫ in [NiFe]-hydrogenases (36). Carbamate and phosphate both correspond to fragments of carbamoylphosphate. Formamide, cyanate, thiocyanate, and cyanide are putative substrate, intermediate, and product in CN Ϫ synthesis. Succinamate, maleamate, acetate, lactate, pyruvate, and asparagine correspond to biologically relevant ligands or their fragments, selected from the docking experiment. Crystals were soaked in these solutions for no more than 5 min prior to flash cooling. As an alternative to crystal soaking, in order to test if diffusion of the molecule is the rate-limiting step, some of these molecules were tested for binding by co-crystallization.
HydE Mutagenesis and [FeFe]-Hydrogenase Maturation in E. coli-Since no in vitro test is available to check the effect of point mutations on TmHydE activity, all of the tests were performed on the C. acetobutylicum HydE protein, using an in vivo assay that measures the relative amount of active hydrogenase produced. TmHydE and C. acetobutylicum HydE share 52% identity (180 amino acids) and 71% similarity (246 amino acids) scattered throughout the sequences (supplemental Fig. 1). Thus, point mutation effects on C. acetobutylicum HydE can be reasonably extrapolated to TmHydE. For expression of active [FeFe]-hydrogenase in E. coli, the pET Duet plasmid (Novagen), containing C. acetobutylicum hydA1 in MCS I and C. acetobutylicum hydE in MCS II, and the pCDF Duet plasmid (Novagen), containing C. acetobutylicum hydF in MCS I and C. acetobutylicum hydG in MCS II were transformed into E. coli strain BL-21 (DE3) (Novagen) with co-selection for Ap r (pET Duet) and Sm r (pCDF Duet) colonies. Site-directed mutations were generated in hydE using the XL site-directed mutagenesis kit (Stratagene) according to the manufacturer's instructions. Both the hydA and hydE genes were sequenced (Davis Sequencing, LLC) to confirm the presence of the desired mutation in hydE and to ensure that no unintended PCR-induced mutations were present. Transformed cells were grown overnight in LB medium (Sigma) and antibiotics. Cells were then diluted 100:1 in fresh LB medium containing antibiotics and grown with constant shaking at 37°C to an optical density at 600 nm of 0.5-0.7. Isopropyl 1-thio-␤-D-galactopyranoside was added to a final concentration of 1.0 mM, and cultures were shaken at 100 rpm at room temperature for 2 h. Cultures were then transferred to 130-ml serum vials, which were sealed with rubber septa and flushed with argon at room temperature overnight.
[FeFe]-hydrogenase activity was then measured by adding 1.0 ml of anaerobic 2ϫ whole cell reaction buffer (50 mM potassium phosphate, pH 7.0, 5 mM methyl viologen, 5 mM dithionite, 3 mM NaOH) to an argon-purged serum vial (13 ml) sealed with a rubber septum, followed by the addition of 1.0 ml of induced E. coli cells. The reaction mixture was incubated at 37°C, and headspace gas was removed at various times with gas-tight syringes to assay H 2 levels using gas chromatography (Hewlett Packard 5820).
Anomalous Scattering Labeling-In order to probe the putative anion-binding sites, crystals were flash-cooled within 5 min from a solution containing 100 mM NaBr. X-ray diffraction data were subsequently collected at a slightly higher energy than the absorption edge for bromine. In another experiment, the protein was co-crystallized with NaI, and the crystal was flashcooled within 5 min in a cryoprotecting solution containing NaCl. The fraction of remaining bound I Ϫ was determined by measuring the residual x-ray anomalous signal at ϭ 0.933 Å. The relative occupancy was then refined, applying B factors similar to those of surrounding atoms.
Docking of Small Molecules-The docking experiment was carried out using the software GLIDE (37) from the SCHRO-DINGER package. The ligand library was built from the ZINC data base (38) using Ϫ2 Յ net charge Յ ϩ1 and 20 Da Յ mass Յ 150 Da as selection criteria. This led to a subset of about 20,000 molecules that was subsequently treated with LIGPREP in order to build all of the alternative diastereoisomers of the selected molecules at pH 7 Ϯ 2. The receptor grid was prepared using the protein preparation wizard on the protein coordinates. Special care was taken to describe the cofactor correctly. The less occupied side chain alternative positions as well as all of the ions, detergent, and water molecules were deleted from the file.

RESULTS
Structure-The final 1.35 Å resolution structure, which was determined by the SAD method, contains 347 of the 348 protein residues and 426 solvent and 5 detergent molecules and has excellent refinement statistics and stereochemistry ( Table 1). As previously suggested by amino acid sequence analyses (11,19), TmHydE has a distorted triose-phosphate isomerase (TIM)-barrel fold ( Fig. 1).
Comparison with Biotin Synthase and Other AdoMet Radical Proteins-Overall, HydE is significantly similar to E. coli BioB (supplemental Fig. 2); the superposition of both enzymes gives a root mean square deviation of 2.5 Å for 292 of 348 C␣ atoms. As expected, the patches of conserved residues in HydE that differ from those from BioB are mainly found at the internal face of the ␤-strands that define the barrel (19). They are likely to be responsible for both substrate binding and reactivity. The two additional ␣-helices at the N-terminal end of the TIM barrel in the TmHydE structure (residues 17-25 and 32-47) have counterparts in E. coli BioB (residues 9 -17 and 20 -35) as previously suggested by sequence comparisons (19) (supplemental Fig. 2). The two additional helices contact the TIM-barrel helix 6 that in both proteins is composed of several invariant hydrophobic side chains and either one arginine or lysine residue (19). This amphipathic pattern has been also observed in the amino acid sequences of other AdoMet radical proteins, such as ThiH, PylB, CofG, and CofH (19, 39 -41). This, in turn, confirms the prediction that these proteins belong to an AdoMet radical "subclass" characterized by a complete TIM-barrel fold and a buried active site suitable for small substrate(s) (19,39,41,42). As in BioB, in the TmHydE crystal structure the loop connecting ␤-strand 8 to ␣-helix 8 (residues 293-318 in TmHydE; sup-plemental Fig. 2) covers one end of the barrel and may act as a lid controlling access of substrate and cofactors to the active site (43). This lid loop contains the strictly conserved YXXY motif (from Tyr 303 to Tyr 306 in TmHydE; supplemental Fig. 2), which could play a role in either catalysis or structure stability.
A Patch of Detergent Molecules Covers a Region of HydE that could Interact with the Other Maturases-It has been proposed that HydE, HydF, and HydG form a ternary complex that contains at least the preassembled [FeFe] center of the FeFe-hydrogenase active site (13). This ternary complex is the only species isolated in vitro capable of producing an active hydrogenase. The x-ray structure of TmHydE has a surface region covered by several ordered detergent molecules. The superposition of the TmHydE and E. coli BioB structures indicates that this region of TmHydE matches the monomer-to-monomer contact zone in the E. coli BioB dimer. Based on amino acid sequence analyses, it is likely that HydG, the other AdoMet radical maturase, is structurally very similar to HydE, including the N-terminal extra helices. Therefore, the detergent-containing region in the crystal might be involved in a heterodimeric interaction between HydE and HydG in the in vivo maturation ternary complex (purified TmHydE is monomeric both in solution and in the crystals). This hypothesis should be tested by complementary studies in solution.
[Fe 4 S 4 ] Cluster Geometry-A comparison between the conserved [Fe 4 S 4 ] cluster in TmHydE and the high resolution structure of the classical [Fe 4 S 4 ] cluster from C. acidiurici ferredoxin (44) (Protein Data Bank code 2FDN) reveals a slight distortion of the former, corresponding to a transition of the unique iron atom coordination from tetrahedral to pseudo-octahedral (supplemental material and supplemental Table 1).
AdoMet Binding-Although the TmHydE preparations were incubated with purified AdoMet after anaerobic reconstitution of the clusters (25), we have consistently found variable proportions of the hydrolyzed derivative AdoHCys bound to the conserved [Fe 4 S 4 ] cluster. The hydrolysis of AdoMet and the presence of AdoHCys are not surprising, since the crystals took over a week to appear at pH 8, 20°C, and AdoHCys is known to be a good inhibitor of several AdoMet radical enzymes (45,46). Our high resolution structures provide a very good description of all of the interactions between AdoMet/AdoHCys and the protein ( Fig. 2A and Table 2). As in other proteins of the family (18,19,43,(47)(48)(49), the TmHydE residues that interact with AdoMet are conserved and scattered throughout the amino acid sequence. These interactions involve hydrogen bonds with protein atoms and water molecules, salt bridges, and hydrophobic  contacts through the adenosine ring. As shown in Fig. 2B, the methyl group of AdoMet and the side chain of Leu 305 are in close contact, an interaction that is similarly observed in BioB between the cofactor and an isoleucine side chain. In both proteins, the Leu/Ile residue belongs to the previously described loop lid that presumably can block access to the active site (43). In crystals where we observed equivalent occupancies for AdoMet and AdoHCys, Leu 305 was significantly disordered. Indeed, structures containing mostly either AdoMet or AdoHCys displayed clear alternative positions for this residue (Fig. 2B) (25), a feature shared with several other enzymes of this family (21)(22)(23)48). In the first TmHydE structure, obtained from crystals grown in the presence of dithiothreitol, we found an [Fe 2 S 2 ] cluster at the protein surface coordinated by Cys 311 , Cys 319 , Cys 322 , and a water molecule, 20 Å away from the conserved [Fe 4 S 4 ] center (Fig. 3). The respective positions of this [Fe 2 S 2 ] cluster and the second [Fe 4 S 4 ] cluster in MoaA are more or less equivalent (48). However, there is a major difference between the two proteins; in MoaA, this cluster, which is also coordinated by three conserved cysteine thiolates, points its iron-exchangeable coordination site to the active center cavity, potentially providing a substrate binding site (48,50). On the other hand, in TmHydE, there is a waterfilled cavity located between the second cluster and the active site, but its unique iron coordination site points toward the solvent medium, suggesting that it is not involved in substrate binding (Fig. 3C). Although a [Fe 2 S 2 ] cluster is also found in E. coli BioB, it occupies the bottom of the barrel (supplemental Fig. 2), a region that in TmHydE contains several water molecules and is closed by an additional ␤-strand (residues 337-344, depicted in green in Figs. 1B and S2).
Previous spectroscopic characterization of TmHydE has shown that its second metal binding site coordinates a [Fe 4 S 4 ] cluster in solution (25). We have been unable to reproduce this result in the crystals; instead, either our x-ray structures have an [Fe 2 S 2 ] center at this site, or the second [FeS] cluster is absent (Fig. 3B). Even when present, the [Fe 2 S 2 ] center has an occupancy that never refines above 70%. This, in turn, suggests either incomplete cluster reconstitution in the starting material or loss of iron and labile sulfide during crystallization (which would explain the discrepancy between the nature of this cluster in solution and in the crystal). Difference Fourier maps indicate that although the cluster absence provokes the partial disorder of residues Gly 308 -Cys 311 , there are no other observable effects on either the rest of the lid loop or on crystal packing and isomorphism (not shown). The high instability of the second cluster might reflect a property necessary to its function. Given the unusual structure of the [FeFe]-hydrogenase active site,  Table 2. The 2F o Ϫ F c electron density map contoured at the 1 level is depicted as a blue grid. B, Leu 305 is represented in olive green or blue with either AdoMet or AdoHCys bound, respectively. Only AdoMet is depicted, because, except for the missing C⑀, no significant differences are observed between AdoMet and AdoHCys positions.
with a small thiolate-containing active site-bridging molecule (DTN in Scheme 1B), it is tempting to assign this cluster a role as either a sulfur/iron donor or a scaffold center. For example, the homologous BioB has an additional [Fe 2 S 2 ] cluster that provides a sulfur atom to dethiobiotin, whereas LipA uses a [Fe 4 S 4 ] cluster to insert two sulfur atoms onto octanoic acid. However, there is a serious problem with assigning a functional role to the second cluster; sequence alignments distinguish two subsets of HydE proteins, one that has the three conserved cysteines of the second cluster (in addition to the common CXXXCXXC motif) and another that lacks these residues (7). Furthermore, 1) C311A, C319A, and C322A mutations in C. acetobutylicum HydE, which belongs to the same subset as TmHydE, do not prevent the production of an active [FeFe]-hydrogenase in E. coli (Fig. 4), and 2) replacement of C. acetobutylicum HydE by the Bacteroides thetaiotaomicron HydE (BtHydE), which belongs to the subset lacking the cysteine ligands for the second cluster, does not affect the maturation of the hydrogenase (Fig. 4A). Thus, despite its appeal, the notion that the second [FeS] cluster of TmHydE is involved in sulfur insertion does not seem to be correct. This cluster is also unlikely to be involved in electron transfer, because it is 20 Å away from the conserved [Fe 4 S 4 ] center and AdoMet intercalates in between. In addition, we have not found any relationship between the presence of this second cluster in HydE and either the metabolism of the organism, the cellular localization of the hydrogenase, or the existence of additional maturation proteins. When both [FeS] clusters are present, they are connected by a channel that is part of the internal cavity (Fig. 3C). A possible role for the additional [FeS] cluster could be the storage, regulation, or sensing of intracellular iron and/or sulfur availability. Finally, the second cluster could simply be a remnant from the ancestral protein that originated HydE.
The Internal Cavity-TmHydE, along with methylmalonyl coenzyme A mutase (51), is exceptional among TIM-barrel family members in that its active site-containing cavity is not found only at the top of the barrel, but it spans its entire length. This results in a significantly larger internal cavity in TmHydE with a volume of 991 Å 3 (Fig. 3, C and D), compared with 717 Å 3 in E. coli BioB. Although the active site is not  Table 1). B, structure lacking the second cluster (K11 data set in Table 1). Note that in A and B, the 2F o Ϫ F c electron density map depicted in blue is contoured at 1.0. C, view of the active site cavity. This figure was generated with the program Sybyl (52). The cavity was calculated using the K11 atomic coordinates (Table 1) with a probe of 1.2 Å radius and is colored according to the electrostatic potential value from red (Ϫ)toblue (ϩ). The electrostatic potential range is shown on the color ramp as kcal/(mol ϫ electron). Key residues lining the cavity are shown in ball-and-stick representations, and the iron sulfur clusters are depicted as van der Waals spheres. The overall charge of the cavity is positive, and two anion-binding sites, named X 1 and X 2 , are clearly visible (referred to as S1 and S3 throughout). These sites were putatively occupied by Cl Ϫ ions in the initial structure and can be replaced by Br Ϫ (see"Results"). D, view of the cavity within the TmHydE structure. The cavity was calculated using the K11 atomic coordinates (Table 1), without the [Fe 2 S 2 ] cluster and a probe of 1.0 Å radius and is colored as in C. The complete protein model is represented by its secondary structure (␣-helices are indicated in cyan, and ␤-strands are shown as red arrows). This view shows the three putative entrances to the cavity.  (Fig. 3D).
Putative Substrate/Product Binding-A close comparison between our TmHydE structure and the x-ray structures of other AdoMet radical enzymes in complex with their substrate (i.e. E. coli BioB (43) and lysine 2,3-aminomutase (49)) defines an approximately spherical volume for radical transfer to the substrate inside the barrel (brown sphere in Fig. 5A). Close to the putative active center region, an elongated electron density peak was modeled as two disordered, partially occupied, chloride ions (Fig. 5B), facing Arg 159 , Tyr 306 , and the conserved 266 PATTA motif. This electron density peak could also correspond to a diatomic molecule, such as CN Ϫ or CO, but there is no obvious source for such molecules in our crystallization solutions. Another putative anion-binding site, at the bottom of the barrel, was modeled as a fully occupied chloride ion unusually bound to the O␥ of Thr 134 and a water molecule, at 3.05 and 3.18 Å, respectively. In addition, the negative charge of the Cl Ϫ ion may be compensated by Arg 54 and Arg 155 , at 3.29 and 3.48 Å, respectively. Several conserved hydrophobic residues found at the center of the cavity separate these two sites (Met 291 , Pro 266 , Val 289 , Met 224 , Leu 157 , Val 105 , and Leu 56 ) (Fig. 3C). In order to establish unambiguously the anion-binding character of the sites, we collected diffraction data from a TmHydE crystal soaked in a NaBr-containing solution (see "Experimental Procedures"). The resulting anomalous scattering electron density map clearly displays three Br Ϫ anions bound in the internal cavity at sites called S1, S2, and S3 (Fig. 5C). S1 and S3 coincide, respectively, with the disordered and ordered Cl Ϫ sites described above (Fig. 5B). The third anion-binding site, located between the other two (S2), had been previously modeled as water. S1 and S2 have 50% Br Ϫ occupancy in the crystal, whereas S3 is fully occupied. Along the same lines, soaking a crystal grown in the presence of NaI in a NaCl-containing cryoprotecting solution indicates that S3 has the lowest exchange rate (see "Experimental Procedures") (supplemental Fig. 3). These three cavity anion-binding sites may be relevant to function, since all of the residues involved are strictly conserved in HydE enzymes. Both the apparent disorder of the ion occupying S1 and the unusual interactions of Cl Ϫ at S3 suggest that the physiological anionic ligand(s) is significantly different from halide ions.
Site-directed Mutagenesis Studies at the Top of the Barrel-Since the function of HydE is not yet known, site-directed mutagenesis effects were monitored in vivo by measuring recombinant C. acetobutylicum hydrogenase activity in E. coli (see "Experimental Procedures"). The HydE maturase from this organism, rather than the T. maritima HydE, was used to monitor the ability of selected HydE mutants to participate in [FeFe]-hydrogenase maturation. The heterologous coexpression of C. acetobutylicum HydE, in combination with HydF, HydG, and HydA from the same organism, is a well established and effective technique to obtain heterologously expressed, active [FeFe]-hydrogenases in E. coli (12,53). In contrast, there are currently no reports of the expression of an active [FeFe]hydrogenase in E. coli or any other organism with thoroughly established genetic techniques, using maturases and/or structural enzymes from thermophilic organisms. Since initial experiments indicated that maturases from T. maritima were not as effective in assembling an active [FeFe]-hydrogenase in E. coli, we used the C. acetobutylicum expression system as a reporter instead. Fig. 4A summarizes the mutagenesis results. Strictly conserved Glu 58 , Gln 107 , Leu 157 , Met 291 , Tyr 303 , and Tyr 306 ; highly FIGURE 5. Active site cavity. A, stereo view of the substrate binding region. The purple dot corresponds to the 5Ј carbon atom where the radical will be initially located upon reduction. The large semitransparent brown sphere defines the putative substrate-binding site (or at least the location where radical transfer should occur), based on comparisons with E. coli BioB (43) and lysine 2,3-aminomutase (49). The remaining spheres correspond to the location of the carboxylate moieties (red) and partial positive charges (blue) deduced from the molecular docking experiment. The two disordered and partially occupied Cl Ϫ ions are also located within the red sphere. B, view of the whole barrel cavity. The 2F o Ϫ F c electron density contoured at 1 around S1, S2, and S3 is depicted in blue. It clearly indicates an elongated feature for the species located in S1 that were modeled as two disordered and partially occupied Cl Ϫ ions. Refined occupancies are as follows: S1, 0.7/0.3; S2, 0.5; S3, 1.0. C, same view as in B after soaking the crystal in a solution containing 100 mM NaBr. The resulting anomalous Fourier map, contoured at the 4.5 level and depicted in purple, clearly shows the 3 Br Ϫ positions in the cavity at S1, S2, and S3. Refined occupancies are as follows: S1, 0.5; S2, 0.5; S3, 1.0.
conserved Leu 56 , Leu 156 , and Arg 159 ; the PATTA motif (residues 266 -270); and Leu 305 (Figs. 2B and 5A) are part of the putative substrate-binding zone. Mutation of the residue corresponding to Gln 107 in C. acetobutylicum HydE to alanine, asparagine, or glutamate decreased hydrogenase activity between 2-and 25-fold, whereas changing Leu 305 to alanine resulted in 50% loss of hydrogenase activity. On the other hand, replacing Leu 305 with valine had no appreciable effect. We speculate that alanine at position 305 renders the active site more exposed to solvent, which, in turn, may increase unproductive cleavage of the AdoMet by water instead of substrate. However, it may be that this Leu/Ile residue is mainly involved in the correct positioning of AdoMet (see "AdoMet Binding" and Fig. 2B). Arg 159 occupies a position similar to the strictly conserved Asn 222 in E. coli BioB, a residue that binds the ureido ring of dethiobiotin. Replacement of this arginine by glutamate at S1 results in inactive enzyme. Also, the double mutation T268V and T269V at the PATTA motif causes an 85% drop in hydrogenase activity. Thus, residues near S1 seem to be involved in catalysis. To test whether putative substrate/product diffusion near the active site could depend on local charges, we introduced the double E58K/K309E mutation, thus reversing a highly conserved salt bridge interaction. This mutation led to a 7-fold decrease in hydrogenase activity. However, since the single K309E mutant had no effect, Glu 58 , close to the putative substrate-binding site, is likely to be solely responsible for the drop in enzymatic activity. Although the mutation of Tyr 303 and Tyr 306 of the lid loop to phenylalanine in C. acetobutylicum HydE does not affect hydrogenase activity, replacement of these residues by alanine completely abolishes it. Maybe aromatic moieties are essential at these positions to prevent deleterious solvent access to the active site during the radical-based reaction.
Molecular Docking and Crystallographic Search for Substrates and Products-Since overexpressing hydE, hydF, and hydG along with the [FeFe]-hydrogenase gene in E. coli leads to active hydrogenase (11), it is likely that their substrates are present in this bacterium. More generally, and in an attempt to shade light on the nature of substrate(s) and product(s) of HydE, we in silico-docked over 20,000 small molecules with masses between 20 and 150 Da and net charges ranging from Ϫ2 to ϩ1 (see "Experimental Procedures"). Although there were no predicted good affinity hits according to standard criteria, some useful information could be extracted from this experiment. 1) The 50 best hits had a net charge between Ϫ2 and 0, confirming the overall anionic properties of the substrate/product. 2) Of the first 20 solutions, 16 had a carboxylate moiety with two oxygen atoms sitting at the positions occupied by the two disordered Cl Ϫ ions in S1. This observation strongly suggests that the substrate has a carboxylate group that binds at that site (red sphere in Fig. 5A). 3) When molecules had a positively charged function, it always sat between Gln 107 and Glu 58 close to the expected radical transfer region (blue sphere in Fig.  5A). This result suggests that the substrate has a partial positive charged function that binds at this site. We have considered these putative electrostatic properties to define an approximate envelope for substrate binding (Fig. 5A).
We also used crystal soakings and/or co-crystallizations to test the binding of the best hits that are likely to be E. coli metabolites (see "Experimental Procedures"). Unfortunately, under our experimental conditions, none of these molecules introduced electron density changes at the putative HydE active site pocket. On the other hand, even at 10 mM concentration and short soaks, thiocyanate fully replaced the putative chloride ion at S3, displacing three water molecules and causing the movement of several hydrophobic side chains. The most significant one concerns the side chain of the strictly conserved Leu 157 , because its rotation induces a dramatic local reduction in the width of the cavity (supplemental Fig. 4). Indeed, a calculation with a probe radius of 1.3 Å shows that upon thiocyanate binding the barrel cavity separates into two distinct pockets: a top one where the substrate-binding site is located and a bottom one that contains the triatomic ligand (supplemental Fig. 4, B and C). Unexpectedly, the nitrogen atom of thiocyanate occupies the former Cl Ϫ ion position, whereas its sulfur atom points at the top of the barrel, indicating that it binds S3 as the SϭCϭN Ϫ isomer. Another significant change corresponds to residue Thr 134 , which has switched a hydrogen bond from Cl Ϫ to the thiocyanate nitrogen atom (Figs. 6, B and C).
Site-directed Mutagenesis Studies at the Bottom of the Barrel-In order to test whether the "thiocyanate-binding" S3 is relevant to function, we mutated Thr 134 and Arg 155 . The T134V mutant provoked a 50% drop in hydrogenase activity, whereas mutation of Arg 155 to alanine or glutamine reduced hydrogenase activities by more than 90% (Figs. 4 and 6). These results indicate that this site may play a role in the reaction catalyzed by TmHydE. Surprisingly, under the same soaking conditions, the thiocyanate chemical and structural analog cyanate does not displace Cl Ϫ at S3. The same applies to cyanide. Thus, the relative high affinity of the site for thiocyanate compared with halides, cyanide, cyanate, or formate suggests that thiocyanate is likely to be very similar to the molecule (if it is not the molecule itself) that physiologically binds to HydE at the bottom of the barrel. Given the overall similarity between cyanide and thiocyanate, this result suggests that HydE may be involved in the synthesis of the latter. Because cyanide is highly poisonous to the cell, a less toxic transportable precursor may be required, as already observed for [NiFe]-hydrogenases (7). Intriguingly, cyanide can be generated by oxidative conversion of thiocyanate through a radical-based reaction (54).

DISCUSSION
One major drawback in the study of [FeFe]-hydrogenase maturases is that we currently do not know the function of any of the three gene products. As mentioned above, HydF is probably an insertase, given its GTPase activity (55). Of the other two enzymes, HydE would seem, in principle, a good candidate for synthesizing the small dithiolate-containing molecule, given its significant similarity to BioB. However, elimination of the additional [FeS] cluster by mutagenesis does not affect hydrogenase maturation. At any rate, the position of the second [FeS] cluster close to the molecular surface and its versatile composition is intriguing.
Although the high resolution structure of TmHydE displays many similarities to other AdoMet radical proteins, notably BioB, it also shows some unique features. One specific characteristic of HydE is the large size of its interior cavities. They are quite different from the ones in BioB mainly because of the relative positions of the additional [FeS] clusters. Although our crystal soaks suggest that HydE does not bind added CN Ϫ , which is a potential HydE product and hydrogenase active site component, the somewhat similar SCN Ϫ has very high affinity for S3. In vivo experiments have indicated that SCN Ϫ has no effect on [FeFe]-hydrogenase activity (not shown). However, this could be due to transport or localization problems. Conse-quently, a simpler in vitro system should be developed to study the mechanistic consequences of SCN Ϫ binding to HydE.
In an attempt to map substrate/ product binding sites, we have carried out several site-directed mutations on the related C. acetobutylicum HydE. By necessity, we have looked at hydrogenase activity as the resulting phenotype of the mutation, the only one we can measure at this time. Here, we have made the reasonable assumption that the active site is either well assembled or not assembled at all. Consequently, the drop in activity should reflect the relative amount of apohydrogenase. The possibility cannot be ruled out, however, that some mutations resulted in defective active sites retaining some residual activity. Spectroscopic characterization of the mutated C. acetobutylicum HydE will be required to distinguish between these two possibilities. We have, in general, extracted useful information from the mutants. Residues coordinating anion-binding site 1 seem to be relevant for activity, especially Arg 159 , which appears to be essential and is topologically equivalent to the arginine residue that binds dethiobiotin in BioB. The position occupied by Gln 107 is also very important, because, depending on the nature of the mutation, it introduces a significant drop in hydrogenase activity. By analogy with BioB, this residue may also be directly involved in substrate binding. The in silico screening of over 20,000 molecules has shed additional light on the characteristics of the substrate functional groups by defining carboxylate and cation potential binding sites. These, along with the Br Ϫ soaking experiments, have been useful in mapping the potential binding sites in the barrel cavity.
Taken together, our structural and functional analyses suggest that the substrate/product is a small common metabolite that contains a carboxylate moiety and a (partial) positive charge. A plausible mechanism can be postulated where HydE first catalyzes the radical-based reaction at the top of the barrel, and the resulting small anionic product hops from S1 to S2 and then to the barrel bottom, where it tightly binds to S3. This binding induces the movement of  Fig. 5, B and C. The 2F o Ϫ F c electron density contoured at 1 around S1, S2, and S3 is depicted as a blue grid. The electron density peak shape clearly indicates the presence of a thiocyanate molecule bound at S3. B and C, depiction of S3 occupied by a Cl Ϫ ion and thiocyanate, respectively. The arrows indicate the side chains of the hydrophobic residues that change positions upon thiocyanate binding. hydrophobic residues nearby, separating the bottom of the barrel from the active site-containing top cavity. Product then gets transferred to other components in the maturation ternary complex. This last step would require a significant rearrangement of several arginine residues at the bottom of the barrel.
One more general question is why the maturation of [FeFe]-hydrogenase active site requires radical chemistry, as indicated by the requirement of two AdoMet-dependent enzymes, HydE and HydG. As discussed above, DTN synthesis may involve sulfur atom insertion into unreactive carbon atoms, a difficult reaction often resorting to radical chemistry. On the other hand, CN Ϫ (and possibly CO) synthesis can be accomplished by more classic ionic reactions, as shown in the case of [NiFe]-hydrogenase maturation. Perhaps some of the radical chemistry required for [FeFe]-hydrogenase maturation reflects the enzyme's supposedly very ancient origin. Along these lines, some of us (14) have suggested that its active site or a similar precursor could have been generated abiotically and subsequently inserted within a protein matrix. Both model chemistry (56) and the fact that the active site binuclear iron cluster has only one protein ligand support this notion. We are now continuing our efforts in determining the nature of HydE substrate(s) in order to elucidate the enzyme's function.