Structural Basis for the Reaction Mechanism of S-Carbamoylation of HypE by HypF in the Maturation of [NiFe]-Hydrogenases*

Background: HypF catalyzes the transfer of a carbamoyl group to HypE in the maturation of [NiFe]-hydrogenases. Results: Crystal structures of full-length HypF alone and in complex with HypE were determined. Conclusion: HypF catalyzes three consecutive reactions without the release of intermediates. Significance: Elucidation of the reaction mechanism catalyzed by HypF is a key to understanding the maturation process of [NiFe]-hydrogenases. As a remarkable structural feature of hydrogenase active sites, [NiFe]-hydrogenases harbor one carbonyl and two cyano ligands, where HypE and HypF are involved in the biosynthesis of the nitrile group as a precursor of the cyano groups. HypF catalyzes S-carbamoylation of the C-terminal cysteine of HypE via three steps using carbamoylphosphate and ATP, producing two unstable intermediates: carbamate and carbamoyladenylate. Although the crystal structures of intact HypE homodimers and partial HypF have been reported, it remains unclear how the consecutive reactions occur without the loss of unstable intermediates during the proposed reaction scheme. Here we report the crystal structures of full-length HypF both alone and in complex with HypE at resolutions of 2.0 and 2.6 Å, respectively. Three catalytic sites of the structures of the HypF nucleotide- and phosphate-bound forms have been identified, with each site connected via channels inside the protein. This finding suggests that the first two consecutive reactions occur without the release of carbamate or carbamoyladenylate from the enzyme. The structure of HypF in complex with HypE revealed that HypF can associate with HypE without disturbing its homodimeric interaction and that the binding manner allows the C-terminal Cys-351 of HypE to access the S-carbamoylation active site in HypF, suggesting that the third step can also proceed without the release of carbamoyladenylate. A comparison of the structure of HypF with the recently reported structures of O-carbamoyltransferase revealed different reaction mechanisms for carbamoyladenylate synthesis and a similar reaction mechanism for carbamoyltransfer to produce the carbamoyl-HypE molecule.

Hydrogenases play a central role in the hydrogen metabolism of microorganisms, which includes the oxidation of H 2 to two protons for energy conservation and the reduction of protons to H 2 to eliminate excess reducing equivalents in the cell (1). To date, three types of hydrogenases: [Fe]-, [FeFe]-, and [NiFe]hydrogenases have been identified. Despite the absence of a phylogenetic relationship and differences in the composition of the metals at the active site, the three hydrogenases harbor at least one iron carbonyl [Fe(CO)] moiety in the active site as an exclusive feature of hydrogenases (2,3). In addition, [FeFe]-and [NiFe]-hydrogenases generally possess two cyano ligands (CN Ϫ ) for Fe. Each type of hydrogenase has coevolved with a unique system for the biosynthesis of Fe-diatomic ligands.
[NiFe]-hydrogenase has a NiFe(CN) 2 (CO) cluster at the active site, where two cysteine thiolates bridge the Ni and Fe atoms and two additional cysteine thiolates are terminal ligands for the Ni atom. A previous mutagenic study revealed that the availability of citruline in the cell is critical for the synthesis of functional [NiFe]-hydrogenase (4), and subsequent biochemical studies identified four enzymes involved in the synthesis of Fe(CN) 2 moiety of the active site that use carbamoylphosphate as a source of cyano ligands (5)(6)(7)(8)(9). HypF first transfers the carbamoyl group from carbamoylphosphate to the thiol of the C-terminal cysteine of HypE involving two intermediates: carbamate and carbamoyladenylate in an ATP-dependent manner (5,6) (Fig. 1A). HypF has been reported to interact with HypE, which seems to be a critical feature for the carbamoyl transfer reaction. HypE then dehydrates the carbamoyl group to yield the nitrile group by consuming another ATP. Since carbamate and carbamoyladenylate are highly unstable in solution (10), carbamoyl transfer from carbamoylphosphate to HypE catalyzed by HypF is considered to require an ingenious mechanism. The nitrile group attached to the C-terminal cysteine of HypE is reductively transferred to the Fe atom by HypC and HypD; however, the mechanism has not been fully elucidated (8,11,12). The biosynthetic routes for the CO and CN Ϫ ligands are thought to be different (13)(14)(15). The source and synthetic pathway of the CO ligand remain unknown, although the involvement of HypD has been proposed (8). A recent isotope labeling experiment has suggested that the extrinsic CO molecule may be incorporated into the bimetallic cluster but there seems to be a metabolic pathway for the synthesis of the carbonyl ligand (16).
HypF is generally composed of four domains: the acylphosphatase (ACP) 3 domain, Zn finger-like domain, YrdC-like domain, and Kae1-like domain (17) (Fig. 1B). The ACP domain includes the acylphosphate-binding site (18), where carbamoylphosphate is hydrolyzed to carbamate and phosphate. Carbamoyladenylate is then formed at the nucleotide-binding site in the YrdC-like domain using ATP and carbamate as substrates. The exact reaction that takes place at the Kae1-like domain, which contains another nucleotide-binding site and a mononuclear iron/zinc site has not been elucidated. While crystal structures of HypF from Escherichia coli have been reported for the N-terminal ACP domain (residues 1-91) (18) and for the rest of the C-terminal region (residues 92-750) (17), the intact structure of HypF is not yet available. On the other hand, the crystal structures of the full-length HypE from three different organisms have already been reported (12,19,20).
The HypF construct composed of the ACP domain alone has been reported to show no acylphosphatase activity, suggesting that the ACP domain is not functional without other domain(s) (18). In addition, the structure of the ACP domain in complex with the phosphate ion indicated that the binding manner is inconsistent with the proposed transition state for the hydrolysis reaction (21). Recently, a partial structure of HypF including the other three domains revealed that the enzyme harbors two nucleotide-binding sites (one each at the YrdC-like and Kae1-like domains) facing each other in a parallel fashion and separated by about 14 Å (17). Based on this structure, it has been proposed that carbamoyladenylate formed in the YrdClike domain is successively dislocated to the nucleotide-binding site in the Kae1-like domain without being released from the enzyme. The same mechanism has been suggested more recently for TobZ, an O-carbamoyltransferase involved in carbamoylation of antibiotics, which is composed of the YrdC-like and Kae1-like domains with the order reversed (22). However, the geometrical arrangement of the ACP domain relative to the other three domains and the binding manner of HypE to HypF cannot be established by the structural information available so far. The former is important for the elucidation of the reaction mechanism of carbamoyltransfer to adenylate after carbamate is formed in the ACP domain. The latter is a prerequisite for resolving the reaction mechanism of S-carbamoyltransfer from carbamoyladenylate to the C-terminal cysteine of HypE. In this study, we determined the crystal structures of full-length HypF alone and in complex with HypE from Caldanaerobacter subterraneus. Based on the structural information obtained in this study and for the recently reported O-carbamoyltransferase (22), a reaction mechanism of the three consecutive reactions without the release of unstable intermediates is proposed.

EXPERIMENTAL PROCEDURES
Bacterial Strain and Plasmids-C. subterraneus subsp. tengcongensis (basonym: Thermoanaerobacter tengcongensis) was obtained from NITE Biological Resource Center with the ID number of 100824. The gene for HypF was amplified from genomic DNA of C. subterraneus by PCR with the Pfu Ultra II DNA polymerase (Stratagene). The nucleotide sequence of the HypE gene was optimized for the expression in E. coli and the full-length open reading frame was synthesized (Takara Bio) due to a difficulty in the expression. The PCR products for HypE and HypF were inserted into the EheI/XhoI site of pPRO-ExHTb (Invitrogen) and BasI site of pASK IBA3 (IBA), respectively. As a result, the HypF protein used in the crystallization experiment includes an extra "Ser-Ala-Trp-Ser-His-Pro-Gln-Phe-Glu-Lys" strep-tag II peptide at its C terminus compared with the wild-type. In order to improve the quality of the crystals, the N-terminal 39 residues of HypE were not included in the expression vector. Consequently, the HypE protein used in the crystallization experiment starts from the vector-derived "Gly-Ala" sequence followed by the Leu-40 after the cleavage of the N-terminal hexahistidine tag by tobacco etch virus protease during the purification as described below. The plasmids (termed pPRO-E⌬N39 and pASK3-F) were sequenced to ensure that undesired mutations were not introduced during PCR (Sigma-Aldrich).
Protein Expression and Purification-HypE and HypF were produced in E. coli BL21(DE3) cells (Novagen) transformed with pPRO-E⌬N39 and pASK3-F, respectively. Bacteria for the production of HypE and HypF were cultured at 37°C in LB medium and Terrific Broth, respectively, and supplemented with ampicillin to a final concentration of 100 g/ml. After the culture reached an A 590 of 0.6, expressions were induced by the addition of isopropyl-␤-D-thiogalactopyranoside to 200 M for HypE and by the addition of anhydrotetracycline to 0.2 g/ml for HypF, followed by further incubation for 16 h at 24°C.
The purification of both proteins was initiated by suspension of cells with phosphate-buffered saline and cell disruption by sonication on ice. Cell debris and membrane fraction were removed by ultracentrifugation at 185,000 ϫ g for 45 min. HypF was initially purified with Strep-Tactin Superflow resin (Qiagen) according to the manufacturer's instructions except for elution of the product with a buffer containing 20 mM Na/K phosphate buffer (pH 7.4) and 2.5 mM desthiobiotin. The eluate was then loaded on a HiTrapQ anion exchange column and gradient elution of 20 -300 mM Na/K phosphate (pH 7.4) was performed. The pooled fractions were concentrated by a centrifugal concentrator (Sartorius) and further purified by a HiLoad Superdex 200 16/60 gel filtration column (GE Healthcare) equilibrated with 10 mM HEPES-NaOH (pH 7.4) and 100 mM Na citrate. HypE was first purified with a Ni-nitrilotriacetic acid resin (Qiagen) according to the manufacturer's instruction. The N-terminal hexahistidine was cleaved by tobacco etch virus protease with dialysis against a buffer containing 20 mM Tris-HCl and 5 mM ␤-mercaptoethanol at 4°C overnight. The cleaved N-terminal peptide was removed by passing the reac-tant through a Ni-nitrilotriacetic acid resin. The flowthrough was loaded on a HiTrapQ anion exchange column and eluted with the gradient elution of 100 -300 mM NaCl including 20 mM Tris-HCl. In order to obtain the HypE-HypF complex, the purified HypF was mixed with an equimolar amount of HypE and further purified by a Superdex 200 10/300 GL gel filtration column (GE Healthcare) equilibrated with 10 mM Tris-HCl (pH 7.4), 50 mM NaCl, and 0.5 mM DTT.
Crystallization and Structure Determination-All crystallization experiments were performed at 288 K by the sitting-drop vapor-diffusion method with a CrystalQuick 96-well plate (Greiner). The crystallization drop of apo HypF was prepared by mixing 1 l of the protein solution containing 15 mg/ml HypF, 10 mM HEPES-NaOH (pH 7.4), and 100 mM Na citrate with 1 l of reservoir solution containing 100 mM imidazole (pH 8.0), 1.8 M Na/K-phosphate (pH 7.7), and 200 mM NaCl. Crystals grown to a maximum size of 0.1 mm ϫ 0.1 mm ϫ 0.2 mm were soaked in a reservoir solution including 15% glycerol for cryoprotection prior to flash cooling with liquid nitrogen. Single-wavelength anomalous dispersion data of an apo HypF crystal were collected at SPring-8 beamline BL44XU, with the crystals maintained at 90 K by a gaseous nitrogen stream. All data were processed and scaled with HKL2000 (23). Eight metal sites were found by SHELXD (24) and tentatively assigned to zinc, of which the positions, occupancies, and B-factors were refined and the initial phases calculated with autoSHARP (25). After density modification using DM (26) and automatic model building by Buccaneer (27), an initial model covering ϳ50% of the HypF molecule was obtained. Subsequent model building with the density-modified experimentally phased electron density map was performed with Coot (28), and the model was refined with Refmac5 (29).
The crystals of the nucleotide-bound form of HypF were prepared by adding [[(2R,3S,4R,5R)-5-(6-aminopurin-9-yl)-3,4-dihydroxyoxolan-2-yl]methoxy-hydroxyphosphoryl]methylphosphonic acid (AMPCPP) and MgCl 2 to final concentrations of 1 mM into a crystallization drop used for the preparation of apo HypF crystals. The cryoprotection buffer also contained 1 mM AMPCPP and MgCl 2 . The crystals of the HypE-HypF binary complex were obtained by mixing the crystallization drop containing 25 mg/ml protein, 10 mM Tris-HCl (pH 7.4), 50 mM NaCl, and 0.5 mM DTT with the reservoir solution containing 100 mM PIPES (pH6.5), 0.48 M KCl, and 9% PEG3,350. The native data for the complex crystals were collected at SPring-8 beamline BL44XU. The phases were determined by the molecular replacement method with Molrep, and subsequent iterative manual model building/corrections and refinement were performed with Coot and Refmac5, respectively. The phi/psi torsion angles of protein backbone were assessed by Rampage (30). Statistics for data collection, phasing, and refinement are summarized in supplemental Table S1. The coordinates and the structure factors have been deposited in the Protein Data Bank, under the accession code of 3VTH and 3VTI. Graphical representations of the model were prepared with PyMOL (DeLano Scientific). Channels were identified by CAVER (31).

Structure Determination of HypF in Complex with
Nucleotides-Crystallization of the full-length HypF was attempted by using orthologs of several organisms, but was only achieved when C. subterraneus was used as the source. The amino acid sequence alignment suggested that HypF from C. subterraneus shares a common domain composition/order with the enzymes in other organisms (supplemental Fig. S1A). The crystal structure of the full-length HypF was determined by the single wavelength anomalous dispersion method, with endogenous metal ions (two zincs and one iron per protein molecule) used as anomalous scatters. The structure refinement was performed to a 2.0-Å resolution for the nucleotidebound form, where the protein was co-crystallized with the non-hydrolyzable ATP analog, AMPCPP 3 in the presence of Mg 2ϩ ion (supplemental Table S1). The full-length molecule consists of four domains: the ACP domain (residues 2-108), Zn finger-like domain (residues 109 -191), YrdC-like domain (residues 192-388), and Kae1-like domain (residues 389 -751) (Fig.  1B). While one of the two molecules in an asymmetric unit was well ordered throughout the entire polypeptide region except for N-terminal Val-2, the other molecule showed disorder over the whole of the ACP domain and two regions of the Zn fingerlike domain; these seemed to be in at least two major orientations relative to the other domains in the crystal. Therefore, the ACP domain and the disordered regions in the Zn finger-like domain (residues 163-167 and 186 -203) were excluded from the final model and only the ordered molecule will be described hereafter. The difference in the stability of the ACP domains between two molecules seems to be ascribable to the molecular packing in crystals; while the ordered molecule shows a close contact with a neighbor molecule via a region in the loop connecting the ACP and Zn finger-like domains (residues 101-103), there is a more space for the other molecule, which should disallow the stabilization of the ACP domain.
As anticipated from the sequence identity of 37% between the C. subterraneus and E. coli orthologs, the previously reported partial crystal structures of HypF from E. coli were superposed onto the full-length HypF with a root mean square deviation value of 1.2 Å for the ACP domain (PDB ID: 1GXT) over 82 C␣ pairs, and 1.8 Å for the other three domains (PDB ID: 3TTC) over 611 C␣ pairs.
As with HypF from E. coli (17), the nucleotide was found in each of the active sites of the YrdC-like and Kae1-like domains. We chose AMPCPP as the ATP analog because the YrdC-like domain catalyzes the hydrolysis of ATP into AMP and pyrophosphate regardless of the absence of carbamoylphosphate and HypE. As expected, the F o -F c electron density map clearly delineated the presence of one AMPCPP molecule and one Mg 2ϩ ion at the nucleotide-binding site in the YrdC-like domain. On the other hand, the F o -F c electron density map around the nucleotide-binding site in the Kae1-like domain implicated the presence of an ADP-like molecule with adenosine plus two phosphate moieties. Based on the result that no nucleotide was found at the corresponding site in the absence of AMPCPP under crystallization conditions, we assumed that AMPCPP hydrolyzed to AMPCP during crystallization or that the reagent was contaminated with AMPCP, and therefore the nucleotide in the Kea1-like domain was assigned to AMPCP. Two nucleotide-binding sites were located in the large cavity formed at the boundary of the YrdC-like and Kae1-like domains on opposite sides of the cavity (Fig. 1B). The geometric configuration between the two nucleotides was nearly identical to that observed in the previously reported crystal structure of HypF from E. coli, where the distance between the two bound nucleotides was ϳ15 Å (based on the distance between the ribose rings).
ACP Domain in the Phosphate-bound Form-As reported for ACPs from various organisms and the ACP domain of HypF from E. coli, the ACP domain of HypF from C. subterraneus is in an ␣/␤ sandwich fold with a five-stranded mixed ␤-sheet and two ␣-helices that occupy the one side of the ␤-sheet (Fig. 1B).
While peripheral loop regions show structural discrepancies, the core of the domain superposes acceptably onto the ACP domain from E. coli and other ACPs. In this study, HypF was crystallized in the presence of a high concentration of phosphate; consequently, a phosphate ion was found at the proposed substrate-binding site close to the boundary between the ACP and Zn finger-like domains (Fig. 1B). No phosphate ion was found at the corresponding position of the HypE-HypF complex, in which the crystallization condition is devoid of phosphate. The phosphate ion is located at the N-terminal edge of one of two ␣-helices (residues 25-36) and is hydrogenbonded to the five main-chain amines of Gly-22, Val-23, Gly-24, Phe-25, and Arg-26, and the guanidinium of Arg-26 ( Fig.  2A). In addition, the carboxamide of Asn-44 interacts with the phosphate ion through a water molecule. Previous mutagenic and structural studies on mammalian ACP have revealed that Arg-26 and Asn-44 are essential to the catalytic reaction (32,33), and the binding manner of the phosphate ion to the enzyme revealed in this study resembled that of the proposed catalytic transition state (21). So far, several structures of ACPs in complex that mimic the substrate-bound states are available. These include a formate, sulfate, chloride, or phosphate ion, but only the phosphate-bound forms of ACP from Pyrococcus horikoshii (PDB ID: 2W4D) (34) and Bacillus subtilis (PDB ID: 3BR8) have shown an enzyme-substrate configuration similar to that observed in this study. Although one of the crystal structures of the HypF ACP domain of E. coli has been found to accommodate a phosphate ion at nearly the same position as that observed in this study, the orientation of the phosphate ion was different (18). A detailed structural comparison suggested that the difference can be ascribed to the presence of the Zn fingerlike domain. In the full-length HypF structure, the residues from the Zn finger-like domain directly interact with Arg-26 and Asn-44 and indirectly interact with the phosphate ion through two water molecules (Fig. 2B). Furthermore, Gln-21 is fixed in place by steric hindrance caused by Tyr-151 in the Zn finger-like domain, which in turn prevents the side chain of Asn-44 from taking another conformation as observed for the APC domain alone of E. coli (corresponding to downward flipping in Fig. 2, A and B). Based on the observed arrangement, deprotonation of the water molecule (termed W1 in Fig. 2B) by a phosphate oxygen should be enhanced to produce the nucleophile hydroxide for dephosphorylation of carbamoylphosphate, as in the proposed catalytic mechanism (21,33). The absence of catalytic activity for the ACP domain of HypF without the Zn finger-like domain (18) can be explained by the observed difference in the structure of the acylphosphate-binding site. The carbamoyl group in carbamoylphosphate should be positioned around W2; this is close to the entrance of the channel connecting to the nucleotide-binding site in the YrdC-like domain, as described below.
Channel Connecting the Active Sites in the ACP and YrdClike Domains-The geometric arrangement of the ACP domain relative to the YrdC-like domain was revealed by the structure of the intact HypF molecule (Fig. 1B). The ACP domain was located near the Zn finger-like domain at the opposite side of the YrdC-like domain and also showed loose contact with the Kae1-like domain. We found a long bent channel from the acyl-FIGURE 1. Functional and structural aspects of HypF. A, reaction scheme of carbamoyltransfer from carbamoylphosphate to the C-terminal cysteine of HypE catalyzed by HypF. B, overall structure of HypF from C. subterraneus. The protein molecule is schematically represented using a different color for each domain. The phosphate ion, AMPCPP, and AMMPCP molecules are drawn as stick models, and the zinc, magnesium, and iron atoms are depicted by gray, cyan, and brown spheres, respectively. The long channel connecting the phosphate-binding site with the nucleotide-binding site in the YrdC-like domain is indicated by a light blue surface model. One of the two zincs is hidden behind the surface representation. The N and C termini of the polypeptide are labeled.
phosphate-binding site to the nucleotide-binding site in the YrdC-like domain that traversed through the Zn finger-like and YrdC-like domains (Fig. 1B). Inside the channel, a number of water molecules were recognized. Since carbamate rapidly degrades into carbon dioxide and ammonia in a physiological aqueous environment (10), the translocation of carbamate through the channel inside the protein seems to be sensible. This strategy has also been applied to carbamoylphosphate synthetase, where two long channels exist for the translocation of ammonia and carbamate (35). The direct distance from the phosphorous atom in the ACP domain to the ␣-phosphate phosphorous atom in the YrdC-like domain is ϳ33 Å whereas the moving distance is ϳ45 Å.
Carbamoyladenylation in the YrdC-like Domain-The YrdC-like domain showed a similar structure to the other members of the YrdC/Sua5 family (36 -38), which are commonly composed of a twisted ␤-sheet flanked by a few ␣-helices on both sides. At the nucleotide-binding site in the YrdC-like domain, the bound AMPCPP 3 molecule was found at the same position as the previously reported structure of HypF from E. coli, for which adenylylimidodiphosphate (AMPPNP) 3 was introduced by the soaking method. However, a significant difference was observed in the configuration of the three phosphate groups between the two structures (Fig. 3A). While the AMPCPP molecule observed in this study coordinates to Mg 2ϩ ion via three phosphate groups, the AMPPNP molecule in the other structure showed that the tri-phosphate groups were unnaturally twisted, and no Mg 2ϩ ion has been identified. We attributed the observed difference to either the presence of Mg 2ϩ ion in the crystallization condition or the property of the nucleotide analog based on the fact that the residues interacting with the nucleotides were well conserved between two orthologs (supplemental Fig. S1A).
The adenine ring of AMPCPP was accommodated in a hydrophobic pocket with only one direct hydrogen bond between the primary amine in the adenine ring and the hydroxy of Thr-303 (Fig. 3A). No specific interaction between the ribose ring and the protein was found except for the hydrogen bond between the guanidinium of Arg-382 and the 3Ј-hydroxy in the ribose ring, although two hydroxy of the ribose ring were con-nected to the hydrogen-bond network of solvent molecules. The Mg 2ϩ ion was coordinated by the ␣-, ␤-, and ␥-phosphate groups, the hydroxy of Ser-329, and one water molecule, resulting in a square pyramidal coordination. The positively charged residues Lys-245, Arg-247, Lys-250, and Arg-382 electrostatically interacted with the phosphate groups of AMPCPP.
We found a small cavity nearby the ␣-phosphate group on the opposite side of the ␤-phosphate group that was occupied by four water molecules (Fig. 3A). The cavity was sequestered by the protein and the AMPCPP molecule from the solvent region of the large cavity but was connected to the long channel by which it has been proposed that carbamate translocates from the acylphosphate-binding site. The four water molecules in the small cavity were hydrogen-bonded to the ⑀-amine of His-227, the main-chain carbonyl oxygen of Met-327, the carboxyl of Asp-367 and Asp-368, the main-chain amine of Asp-368, and the guanidinium of Arg-382. Since the space was large enough to accommodate carbamate and the position is suitable for nucleophilic attack on the phosphorous atom in the ␣-phosphate group, it is considered that some of these residues support carbamate in its proper configuration during carbamoyladenylate formation. The amino acid sequence alignment shows that the residues, whose side chains are involved in the hydrogen-bond network formation, are conserved among HypF orthologs except for Gln221 in HypF3 from Ralstonia eutropha (corresponds to His-227 in the C. subterraneus ortholog) (supplemental Fig. S1A). The residue should be able to alternate with histidine because glutamine can be a hydrogen-bond donor/acceptor.
Iron-binding Site at the Kae1-like Domain-The Kae1-like domain is composed of two ␣/␤ subdomains (Fig. 1B) and falls into the ASKHA (acetate and sugar kinases, Hsc70, actin) superfamily (39), which includes the universal protein Kae1 (40 -42) and members of the CmcH/NodU carbamoyltransferase family as closely related homologs of HypF (43). The function of Kae1 (kinase-associated endopeptidase 1) has recently been identified for the synthesis of carbamoylthreonine by N-carbamoyltransfer from carbamoylphosphate to threonine, which is an essential step for the maturation of tRNA (44). The previously reported crystal structures of Kae1 has revealed that the nucleotide-binding site contains an Fe atom coordinated by two histidines and two aspartates (40,41). In addition, the recently reported crystal structure of TobZ, a member of the CmcH/NodU carbamoyltransferase family, has demonstrated that the enzyme also includes the Fe(His) 2 (Asp) 2 center at the nucleotide-binding site (22). On the other hand, zinc has been identified in the partial HypF structure of E. coli at the corresponding site, although a substoichiometric amount of iron atoms has been detected from the purified full-length enzyme by inductively coupled plasma atomic emission spectroscopy (ICP-AES) (9). For HypF from C. subterraneus, we identified the metal as iron based on the anomalous dispersion measurement (supplemental Fig. S2) and ICP-AES analysis that showed the protein contained 1.2 Fe and 1.4 Zn per HypF molecule. UV-VIS absorption spectra of the purified protein in the absence of a reducing agent showed three small peaks at 360, 502, and 616 nm, which were significantly decreased by the addition of 10 mM dithiothreitol and disappeared when 10 mM dithionite was included (Fig. 4). On the other hand, it has been reported that Kae1 showed an absorption peak at 500 and 540 nm for nucleotide-free and nucleotide-bound forms, respectively (40,41). The inconsistency of the spectrum of the as-purified state with that of Kae1 may be explained by the partial replacement of Zn for Fe in the Zn finger-like domain, because the spectrum resembles that observed in several rubredoxin, which have a Fe(Cys) 4 site and absorption peaks at ϳ370, ϳ500, and ϳ600 nm (45) (46). If this is the case and the extinction coefficient of 9,200 L mol Ϫ1 cm Ϫ1 at 373 nm for rubredoxin is invoked (45), 0.28 Fe per HypF molecule should have been incorporated to Zn finger-like domain. This value is comparable with the extra 0.2 Fe per HypF molecule quantified by the ICP-AES analysis.
The determinant of the metal specificity for the Kae1-like domain in HypF remains unknown; the residues around the Fe-binding site are highly conserved between two orthologs of HypF, and both proteins used in the previous and current studies were produced using E. coli systems.
As reported for the ADP-bound forms of TobZ and HypF from E. coli, the metal atom showed standard octahedral coor-dination caused by interaction with the ␣and ␤-phosphate groups of the AMPCP molecule, ⑀-amines of two histidines (His-483 and His-487), and carboxyl oxygen atoms of two aspartates (Asp-507 and Asp-734) (Fig. 3B). The oxygen atoms of the ␣-phosphate group formed hydrogen bonds with the main-chain amines of Gly-702 and Asp-734. The ␤-phosphate group is hydrogen-bonded to the main-chain amines of Gly-508 and Thr-509, the side-chain hydroxy of Thr-509, and a water molecule. The main-chain carbonyl of Thr-509 shows close contact with the two oxygen atoms in the ␤-phosphate group. The structure of TobZ in complex with carbamoyladenylate has demonstrated that its carbamoyl group replaced the ␤-phosphate group of ADP without significant conformational change for the residues in contact with the substrate. In the TobZ-carbamoyladenylate structure, the carbonyl oxygen of the carbamoyl group was hydrogen-bonded to the two mainchain amines of Gly-138 and Gln-139 (corresponding to Gly-508 and Thr-509 in HypF), while the amine of the carbamoyl  group coordinated to Fe and was hydrogen-bonded to the main-chain carbonyl of Gln-139. Since the binding of the carbamoyl group only involves the main-chain groups in TobZ and the main-chain structure of HypF at the corresponding region closely resembles that in TobZ, carbamoyladenylate is likely to bind HypF in the same manner as that observed in the TobZcarbamoyladenylate structure.
Molecular Interaction in the HypE-HypF Complex-The complex formation of HypE and HypF were confirmed by size exclusion chromatography, where the complex was eluted as a dimer of heterodimers as reported for the E. coli orthologs of HypE and HypF (19). The structure of the HypE-HypF complex revealed that the two HypF molecules associated with the opposite sides of the HypE homodimer, resulting in an elongated right-handed helical shape of the complex with a length of ϳ180 Å and width of ϳ75 Å (Fig. 5A). The two HypE molecules form a homodimer via the N-terminal domain as in the previously reported crystal structures of HypE from other organisms (12,19,20). In contrast to the HypE dimers of D. vul-garis and E. coli, however, the swapping of the N-terminal segment between two HypE subunits was not found in the C. subterraneus ortholog, whereas the overall structures of the dimers are conserved among organisms regardless of the swapping (Fig. 5B). A detailed structural comparison revealed that a linker connecting the N-terminal segment with the rest of the C-terminal part of HypE of C. subterraneus is not long enough to allow the swapping, which also seems to be the case for HypE of T. kodakarensis (supplemental Fig. S1B).
The HypF molecule interacted with the C-terminal domain of HypE with no interference in the homodimerization of HypE (Fig. 5A). The interaction sites in HypF mostly resided in the Kae1-like domain, although several residues in the YrdC-like domain were also close to the HypE molecule within a distance of 4 Å. A superposition of the structure of HypF alone onto that of the HypE-HypF complex indicated that the ACP domain is significantly dislocated relative to the other domains (Fig. 6). Since HypE showed no contact with the ACP and Zn finger-like domains, the observed difference in the orientation of the ACP domain was likely to be caused by a difference in the crystallization conditions (e.g. the presence of phosphate) and/or by molecular packing effects in the crystals.
The interface between the HypE and HypF molecules is composed of the major and minor interaction sites (Fig. 5C). The former site contains hydrophobic interactions with the contribution of Leu204 E , Phe206 E , Phe276 E , Met277 E , Ile580 F , and Ile589 F (superscripts E/F denote that the residue is in the HypE/ HypF subunit). In addition, electrostatic interactions between the C-terminal end of the short ␣-helix in HypE (268 -276) and Arg553 F and two salt bridges (Asp207 E -Arg576 F and Glu275 E -Lys550 F ) were also found. Of these residues, those involved in the hydrophobic interactions showed a high degree of conservation among orthologs of HypE and HypF (supplemental Fig.  S1). The minor interaction site contained a hydrophobic core composed of L326 E , Ile338 E , Phe436 F , and Leu466 F , and a water-mediated hydrogen bond network consisting of several residues form both subunits. In contrast to the major interaction site, residues in the minor interaction site were not strictly conserved among orthologs (supplemental Fig. S1). No major conformational change in HypF was observed upon formation of the complex.
The final model of the HypE-HypF complex lacks the C-terminal 9 residues in HypE due to the loss of the electron density for residues from Ile-343 to the C terminus. The region includes the C-terminal Cys-351, which has been proposed to be the carbamoyl acceptor from carbamoyladenylate. The distance between the C␣ atom of Pro342 E and the Fe atom in the Kae1like domain is ϳ23 Å and space to accommodate the C-terminal region of HypE is available (Fig. 5C), suggesting that the C-terminal Cys351 E can reach the Fe site in the Kae1-like domain to accept the carbamoyl group without rearrangement of the subunit orientation or large conformational change in the complex.

DISCUSSION
Recently, it has been reported that TobZ, a member of the CmcH/NodU O-carbamoyltransferase family, catalyzes O-carbamoylation of secondary metabolites using carbamoylphosphate and ATP as the carbamoyl group and energy sources, respectively (22). The members of this family are two-domain enzymes composed of the YrdC-like and Kae1-like domains where the order of the domains is reversed relative to HypF (i.e. Kae1-like domain is at the N-terminal side and YrdC-like domain at the C-terminal side). On the other hand, the conserved universal proteins YrdC/Sua5 and YgjD/Kae1 exist as single-domain proteins, and have recently been shown to be involved in the modification of tRNA in a cooperative fashion with two other enzymes, where ATP, bicarbonate, and threonine are utilized as substrates (44,47). The reaction mechanism catalyzed by TobZ has been well described based on the crystal structures in complex with various substrates including carbamoylphosphate, adenine nucleotides, and tobramycin (an acceptor of the carbamoyl group) (22). In contrast to HypF, however, CmcH/NodU O-carbamoyltransferases do not harbor the ACP domain, and it has been proposed that dephosphorylation of carbamoylphosphate occurs at the active site of the YrdC-like domain, which is followed by carbamoyladenylate formation at the same site. On the other hand, a puzzling result has been reported for E. coli HypF, where the carbamoyladenylate-bound form of the partial HypF has been obtained by soaking ATP and carbamoylphosphate, even though the protein is devoid of the ACP domain and the crystallization condition contained no Mg 2ϩ ion (17). However, the functional ACP domain has been shown to be required for the hydrogenase maturation process in vivo (9), and a structural comparison between TobZ and HypF revealed that the main-chain conformation of the carbamoylphosphate-binding site identified in TobZ is considerably different from the corresponding region in HypF. In addition, the carbamoylphosphate-binding site in the ACP domain of HypF was connected to the nucleotide binding site in the YrdC-like domain through the long channel, and the proposed residues that bind carbamate at the end of the channel are exclusively conserved among orthologs of HypF (supplemental Fig. S1A). These lines of evidence strongly suggest that the ACP domain is required for the formation of carbamate, which is subsequently delivered to the YrdC-like domain for the synthesis of carbamoyladenylate. Therefore, we assume that the carbamoyladenylate-HypF complex of E. coli in the previous study (17) was obtained because of the high concentration of carbamoylphosphate in the crystallization condition, allowing spontaneously produced carbamate to react with ATP at the YrdC-like domain. The functional relevance of the flexibility of the ACP domain remains unknown. The whole-domain disorder of the ACP domain observed in one of two HypF molecules in the asymmetric unit of HypF crystals implies that the binding of phosphate to the active site per se does not stabilize the geometric arrangement of the domain. On the other hand, the structure of the phosphate-bound form of the other HypF molecule in the asymmetric unit suggests that the interaction of the zinc finger-like domain is prerequisite for the formation of the active state. Furthermore, the transfer of carbamate may also occur without major conformational change because the substrate-binding site in the ACP domain is positioned around the entrance of the channel connecting to the YrdClike domain (Fig. 1B).
One of the important issues in considering the reaction mechanisms for S-carbamoyltransfer from carbamoyladenylate to the thiol of the C-terminal cysteine of HypE is how the two groups become close to each other. The structure of the HypE-HypF complex in this study showed that the complex formation enabled the C-terminal thiol to access the S-carbamoylation active site in the Kae1-like domain. On the assumption that carbamoyladenylate is held in place as observed in the TobZ-carbamoyladenylate complex, the carbamoyl group is located on the side of the C-terminal part of HypE (Figs. 3B and 5C). The structure of the TobZ-ADP-tobramycin complex has demonstrated that the carbamoylation site in tobramycin (6Љhydroxy) resides in close proximity to the ␤-phosphate group of ADP and a proposed catalytic residue (His-14) for deprotonation of 6Љ-hydroxy in tobramycin. In contrast, no residue was found in the HypE-HypF complex around the region considered to function as a proton acceptor for the thiol of the C-terminal cysteine in HypE, whereas Arg-599 and Glu-617 were found instead as conserved residues among HypF orthologs (supplemental Fig. S1A). The members of the class of thioesterforming enzymes such as acyl-CoA synthetases and non-ribosomal peptides synthetases also do not harbor an apparent proton acceptor for the thiol in substrates, suggesting that a specific residue is not required for this step (48). Considering its close proximity to the S-carbamoylation active site, Arg-599 would be responsible for anchoring the C-terminal cysteine by electrostatic interaction with the carboxyl. On the other hand, Glu-617 may be involved in the proton transfer for the carbamoyltransfer reaction, since the calculated electron density map indicated that the side chain is highly flexible compared with those of the nearby residues.
Based on the structural information obtained in this study, we propose a catalytic mechanism of HypF (Fig. 7). Three consecutive reactions proceed in each catalytic site of the ACP, YrdC-like, and Kae1-like domains. The formations of unstable intermediates (carbamate and carbamoyladenylate) are inevitable, but they are transferred to the following catalytic site via the internal channels without being released from the protein.
A comparison of the structures of HypF and TobZ revealed that the former lacks a carbamoylphosphate-binding site in the YrdC-like domain but harbors the ACP domain with two catalytic sites connected by the channel. We therefore conclude that the ACP domain is required for the formation of carbamate as an initial step for the synthesis of the bimetal cluster of [NiFe]-hydrogenases. FIGURE 7. Schematic drawing of the reaction mechanism of the three consecutive reactions catalyzed by HypF. The first step is hydrolysis of carbamoylphosphate, where the water molecule (red) is activated for nucleophilic attack on the phosphate group to release carbamate (blue), which is transferred to the nucleotide-binding site in the YrdC-like domain via the channel identified in the crystal structure of HypF alone. The second step includes nucleophilic attack of carbamate on the ␣-phosphate group of ATP to form the carbamoyladenylate intermediate, where the protein merely provides residues for maintaining carbamate in a suitable position for the reaction. The carbamoyladenylate is then translocated to the nucleotide-binding site in the Kae1-like domain, enhanced by the change in charge properties of the molecule because of the release of the pyrophosphate group as suggested in previous studies (22,49). The last step occurs at the iron-binding site in the Kae1-like domain. The conserved Glu-617 (not shown in the figure) may play a role in proton transfer during the reaction, but the exact mechanism is still unknown. One of the hydrogen-bond acceptors is missing for all amine groups of the substrates, suggesting that a minor conformational rearrangement and/or binding of solvent molecules are required to catalyze each reaction.