The crystal structure of the hydrolase domain of 10-formyltetrahydrofolate dehydrogenase: mechanism of hydrolysis and its interplay with the dehydrogenase domain.

10-Formyltetrahydrofolate dehydrogenase (FDH) converts 10-formyltetrahydrofolate, a precursor for nucleotide biosynthesis, to tetrahydrofolate. The protein comprises two functional domains: a hydrolase domain that removes a formyl group from 10-formyltetrahydrofolate and a NADP(+)-dependent dehydrogenase domain that reduces the formyl to carbon dioxide. As a first step toward deciphering the catalytic mechanism of the enzyme, we have determined the crystal structure of the hydrolase domain of FDH from rat, solved to 2.3-A resolution. The structure comprises two domains. As expected, domain 1 shares the same Rossmann fold as the related enzymes, methionyl-tRNA-formyltransferase and glycinamide ribonucleotide formyltransferase, but, unexpectedly, the structural similarity between the amino-terminal domain of 10-formyltetrahydrofolate dehydrogenase and methionyl-tRNA-formyltransferase extends to the C terminus of both proteins. The active site contains a molecule of beta-mercaptoethanol that is positioned between His-106 and Asp-142 and that appears to mimic the formate product. We propose a catalytic mechanism for the hydrolase reaction in which Asp-142 polarizes the catalytic water molecule and His-106 orients the carbonyl group of formyl. The structure also provides clues as to how, in the native enzyme, the hydrolase domain transfers its product to the dehydrogenase domain.

10-Formyltetrahydrofolate dehydrogenase (FDH) 1 (EC 1.5.1.6) is an abundant enzyme in several tissues (1), compris-ing up to 1% of total cytosolic protein (2). In vitro, FDH converts 10-formyltetrahydrofolate (10-formyl-THF) to tetrahydrofolate in either an NADP ϩ -dependent dehydrogenase reaction or an NADP ϩ -independent hydrolase reaction (Fig. 1). The substrate for FDH, 10-formyl-THF, forms a major intracellular folate pool (3) and serves as a formyl donor in two reactions of de novo purine biosynthesis (4). The product of the FDH-catalyzed reaction, THF, can be converted to 5,10-methylene-THF, which is required for thymidine biosynthesis, and then to 5-methyl-THF, which is required for S-adenosylmethionine biosynthesis. The precise metabolic role for FDH is unclear, but one possibility is that it converts excess 10-formyl-THF from the de novo purine pathway back to the THF pool (5). Another may be a protective one against formate toxicity by removing formate, in the form of CO 2 (6).
Folate derivatives are required for the biosynthesis of the nucleotide precursors of DNA and RNA, and hence folate-metabolizing enzymes are critical for proper cell functioning. Elevated levels of these enzymes, including the key folate enzyme, dihydrofolate reductase, occur in rapidly proliferating cancer cells (1,7,8). By contrast, FDH is expressed at much lower levels in tumors compared with corresponding normal tissues (1). In fact, elevation of intracellular levels of FDH in cultured cancer cell lines inhibits proliferation and results in cytotoxicity (1). This suppressor effect of FDH is dependent upon its enzymatic activity (1). These studies imply that FDH may possess a growth-regulatory function and that cancer cells lower FDH expression to enhance proliferation.
FDH is a homotetramer of 902 amino acid residue monomers (9,10) where each monomer comprises two distinct functional domains connected by a 110-amino acid residue linker (9). The binding site for folate is located in the N-terminal region of the enzyme (termed N t -FDH, residues 1-310) (11). This domain shows sequence similarity with folate-binding domains in other proteins, including methionyl-tRNA-formyltransferase (FMT) and glycinamide ribonucleotide formyltransferase (GART) (9,11). The carboxyl-terminal domain of FDH (residues 420 -902) shows up to 50% sequence identity with aldehyde dehydrogenases (9,12,13). Interestingly, when the N-terminal domain is expressed separately, it functions as a 10-formyl-THF hydrolase, resulting in the production of formate (11), whereas the dehydrogenase reaction is catalyzed only by the full-length FDH (11,12). This suggests that the hydrolase activity of the N-terminal domain contributes to the overall dehydrogenase reaction and, therefore, some functional communication between the two domains must exist. Thus, FDH is an interesting example of a natural fusion of two unrelated proteins that creates a new enzymatic activity.
A further issue is whether the hydrolase reaction can occur in vivo independently of the dehydrogenase domain or whether * The work was supported by National Institutes of Health Grant DK54388 (to S. A. K.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The these reactions are always coupled. The significance of the hydrolase reaction alone for cellular metabolism is not clear. Since this reaction does not require NADP ϩ , one proposal is that the hydrolase reaction is used to produce THF irrespective of the redox state of the cell (14). Typically, NADPH levels in liver are 10 -30-fold higher than NADP ϩ (15), so the hydrolase activity could maintain THF levels when NADP ϩ is low. Interestingly, an enzyme called formyltetrahydrofolate hydrolase catalyzes the same conversion of 10-formyl-THF to THF and formate in Escherichia coli (16), suggesting the importance of this reaction under some circumstances. Enzymes similar to E. coli hydrolase are not apparent in eukaryotes, so this function might be performed by FDH. Contrary to this argument, if the hydrolase reaction were to occur in tandem with the synthesis of 10-formyl-THF, which is the opposite reaction, catalyzed by 10-formyl-THF synthase and coupled with ATP hydrolysis, the result will be a "futile cycle" that simply consumes energy. Moreover, in vitro, the hydrolase reaction requires millimolar concentrations of ␤-mercaptoethanol (17), and, given that ␤-mercaptoethanol is not present naturally in the cell, it is unlikely that this reaction occurs in vivo independently of the overall dehydrogenase activity of FDH (although it remains possible that other reducing agents such as glutathione could act in a similar manner).
Previously, we have proposed a sequence of steps to describe the dehydrogenase mechanism of FDH (18). In this scheme, the hydrolase domain catalyzes the removal of the formyl group from 10-formyl-THF, followed by its transfer to the aldehyde dehydrogenase domain by an unknown mechanism where it is oxidized to CO 2 . As a first step toward deciphering the catalytic mechanism of FDH, we are pursuing structural studies of this enzyme. In this paper, we report the crystal structure of N t -FDH from rat solved to 2.3-Å resolution.

MATERIALS AND METHODS
Data Collection-The expression and crystallization of N t -FDH has been described previously (19). Briefly, N t -FDH crystallizes in space group P2 1 2 1 2 with cell dimensions a ϭ 100.0 Å, b ϭ 64.6 Å, and c ϭ 64.6 Å. All crystals used for data collection were passed through a cryoprotectant solution containing 1.3 M ammonium sulfate, 0.1 M sodium acetate, pH 4.6, 27% glycerol and then flash-frozen in situ using an X-Stream cryostat (Rigaku MSC).
Diffraction data were collected on an RAXISIVϩϩ image plate detector mounted on a RU-H3R rotating anode x-ray generator fitted with Osmic Confocal optics (Rigaku MSC) and operating at 50 kV and 100 mA. Data were collected at a crystal-to-plate distance of 150 mm, with an oscillation angle of 0.5°and an exposure time of 5 min/frame. Typically 100°of data were sufficient to generate essentially complete data sets. Data were collected and processed using Crystal Clear (20). Friedel pairs were not merged in order to retain the anomalous signal. To improve the phasing (and for refinement purposes), a new native set was collected in place of one described previously (19).
Phasing-A search for derivatives was made by soaking crystals in solutions of heavy atom compounds in stabilizing buffer (1.3 M ammonium sulfate buffered in 0.1 M sodium acetate, pH 4.6). The degree of substitution by heavy atoms in each data set was evaluated using difference Patterson maps. For the first derivative so identified, heavy atom positions were determined from the Patterson maps and refined prior to phasing at 3.0 Å. Subsequent derivatives were identified by difference Fouriers using the current best phases. Where measurable, contributions from the anomalous signal of the heavy atom were included in the phasing calculations. All calculations were performed using PHASES (21).
Model Building and Refinement-Electron density maps were calculated using CCP4 programs (22) and displayed using the O program (23). Starting with the crystal structure of FMT (24) (Protein Data Bank code 1FMT) as a guide, a model was built and then refined by alternating rounds of XPLOR (25) and manual revision using O. In later rounds, water molecules were included where visible in both the 2 F o Ϫ F c and F o Ϫ F c map and with suitable hydrogen binding partners. The final rounds of refinement were performed using REFMAC (26). The stereochemistry of the final model was evaluated using PROCHECK (27). The coordinates and structure factors have been deposited with the Protein Data Bank (1S3I).
Preparation of Cysteine to Alanine Mutants-The QuikChange TM site-directed mutagenesis kit (Stratagene) was used to generate five mutants of N t -FDH, each replacing a cysteine with an alanine. Mutagenesis was performed on cDNA cloned into the pRSET vector. Each construct containing the mutation was then transformed into E. coli XL-1 Blue cells, as described previously (28).
After selection on LB-ampicillin plates, plasmids were isolated from individual colonies using the Quantum Prep kit (Bio-Rad) and the presence of the mutation was confirmed by sequencing. The entire coding region of each clone was also sequenced to ensure the absence of random mutations. Each mutant protein was expressed in E. coli and purified in the same manner as described previously for wild-type N t -FDH (28). The hydrolase activity of each mutant enzyme was assayed using 10-formyl-DDF as a substrate, as described previously (17). Each assay was performed twice, and the numbers reported are an average of these two experiments.

RESULTS
Structure Determination-The structure of N t -FDH was determined by multiple isomorphous replacement with anomalous scattering. Data from crystals soaked in mercury, platinum, and uranyl compounds were used for the phasing. The best derivative was obtained using thimerosal, and this was used to identify other weaker derivatives by difference Fouriers. The data collection and phasing statistics are shown in Tables I and II, and portions of both the experimental, calculated at 3.0 Å, and final 2F o Ϫ F c electron density maps, at 2.4 Å, are shown in Fig. 2. The refinement of this structure was rather complicated, because several regions of the molecule were difficult to fit due to weak density, and this is one contribution to the comparatively high R and R free values of the structure. In particular, the region around residues 85-88 was difficult to place correctly. In fact, the density appears to encompass at least two conformations in this region, and this suggests that it may become ordered upon binding substrate. Another contributor is the inherent quality of the crystals, which grow as thin plates and do not diffract equally well in all directions. The final structure comprises residues 1-307 and 53 water molecules and has a crystallographic R factor of 24.5% (R free ϭ 30.5%) at 2.3-Å resolution (Table III). As defined by PROCHECK (27), 88.6% of the residues lie within the most favored region of the Ramachandran plot with one outlier (Ser-87). The two C-terminal residues of the N t -FDH construct are not visible in the electron density and were omitted from the model. Interestingly, four of the five cysteines present in the structure appear to be modified by ␤-mercaptoethanol (␤ME). This was probably caused by the high concentration of ␤ME present in the crystallization solution (10 mM). Thus, four molecules of ␤ME were included in the model, modifying cysteines 86, 152, 191, and 238 (the exception is Cys-17). One of these cysteines (Cys-86) is near the active site (see below).
Site-directed Mutagenesis of Cysteine Residues-Since most of the cysteines in the crystal structure were covalently modified by ␤ME, it was important to determine the effect of these FIG. 1. The reaction scheme of FDH. A strict hydrolysis of 10formyl-THF, as occurs in N t -FDH, leads to release of formate, whereas in full-length FDH, the hydrolase reaction is coupled with the dehydrogenase reaction, such that NADP ϩ is reduced to NADPH with release of carbon dioxide. modifications on the catalytic activity of the enzyme. Modification of N t -FDH with N-ethylmaleimide, an agent that is directed toward cysteines, did not inhibit the hydrolase activity (data not shown). This implies that a cysteine residue is not involved in the hydrolase catalysis. However, a buried cysteine may be protected from chemical modification; hence, using site-directed mutagenesis, we replaced each of the cysteines with alanine. Each of the mutants, C17A, C86A, C152A, C191A, and C238A, was expressed in E. coli, purified, and characterized kinetically. All cysteine mutants catalyzed hydrolase activity at the same rate as wild type N t -FDH (Table  IV), indicating that a cysteine is not involved in the hydrolase mechanism and that the structural perturbations arising from modifications of those cysteines is unlikely to affect the functional architecture of the protein.
Structure Description-The structure comprises two domains (Fig. 3). Domain 1 contains a seven-stranded sheet of mixed parallel and antiparallel ␤-strands, comprising ␤ 1 -␤ 5 and ␤ 7 -␤ 8 , packed on both sides by ␣-helices, and is immediately recognizable as a Rossmann fold (29). ␤ 6 is separate from this sheet and forms a small two-stranded sheet with part of the loop that connects domains 1 and 2. Domain 2 is mostly all ␤ in structure and from one angle appears as a sandwich of two ␤-sheets, one five-stranded and the other four-stranded. However, two central strands, ␤ 12 and ␤ 16 , bend at approximately right angles and link the two sheets to create a single sheet. The resulting arrangement is barrel-like and reminiscent of the OB fold (30). Between the two domains lies a long stretch of intervening polypeptide chain, comprising residues 185-218, which is predominantly extended in structure. The secondary structure assignments are shown in Fig. 3c.
As expected, since it contains the folate-binding site, domain 1 of the structure shows similarity with both FMT (24) and GART (31). This domain superimposes closely in all three enzymes, with the closest overlap being the central seven-stranded sheet and the greatest differences being in the relative positions of the surrounding helices as well as the length and structure of the connections between the elements of secondary structure. For instance, the r.m.s. deviation between the main chain atoms of 145 common residues of N t -FDH and GART is 3.5 Å, whereas for the seven equivalent strands it is 0.8 Å. The same analysis for 156 common residues between N t -FDH and FMT yields an r.m.s. deviation of 3.7 Å, but for the ␤ sheet it is 1.0 Å. The divergence between N t -FDH and GART begins at residue 187 (of both enzymes), near the C terminus of GART, whereas for FMT, the structural similarity extends all the way to the C terminus of both proteins. In fact, domain 2 of N t -FDH and the C-terminal domain of FMT are topologically equivalent except for the connection between ␤ 12 and ␤ 14 , which is much longer in N t -FDH and includes an additional strand ␤ 13 , whereas in FMT this connection, between ␤ 9 and ␤ 10 , is much shorter (Fig. 4). Interestingly, although there is no topologically equivalent strand for ␤ 13 in FMT, the edge position in the sheet is occupied instead by ␤ 14 in FMT, which is at the C terminus of the protein. Given the lack of sequence similarity in the C-terminal half of these proteins, the structural overlap between N t -FDH and FMT in this region was unexpected. The functional significance of these structural similarities is discussed below.
Active Site-The location of the active site can be determined by the sequence similarity of N t -FDH to FMT and GART and, in particular, the presence of highly conserved residues that form the active sites in these enzymes. It is situated at the base of the deep cleft between domains 1 and 2. The substratebinding site is formed by the divergence of ␤ strands (␤ 4 and ␤ 5 ) in the central ␤ sheet of domain 1. As viewed in Fig. 5, the active site is formed by several loops including residues 113-117 at the base, 85-89 at the top, 104 -108 to the left (part of ␤ 5 ), 139 -144 to the right, and Phe-135 from ␤ 7 . Of note is the relatively poor density of the 85-89 region, making it difficult to fit these residues with high confidence (see above). Much of the active site is markedly hydrophobic in character due to residues such as Phe-35, Phe-135, Leu-141, Ala-114, and Ile-104. The lack of large side chains in residues 115-117 (Gly-Ala-Ser) may permit the close approach of the substrate molecule.
Two key residues in the active site are His-106 and Asp-142. These are conserved in both FMT and GART and mutation of either of these renders N t -FDH inactive as a hydrolase (18,32). Interestingly, we observe electron density bridging between these two residues and have interpreted this as a molecule of ␤ME (Fig. 5). It has been fitted such that the sulfhydryl group is within hydrogen bonding distance of His-106 and the hydroxyl group within hydrogen bonding distance of Asp-142. The involvement of ␤ME in the in vitro hydrolase mechanism of N t -FDH is discussed below.
Comparison with the Active Sites of FMT and GART-To compare the similarity of the active site with those of FMT and GART, the crystal structure of N t -FDH was superimposed onto those of FMT and GART using several conserved residues in the active site (Fig. 6). The close correspondence of many of the loops in the active sites again shows the overall similarity of the three proteins, but it also reveals some distinctive differences. Overall, there is closer similarity of N t -FDH with FMT than GART. Most notably, the loop in GART that is topologically equivalent to residues 113-117 in N t -FDH and residues 117-121 in FMT adopts a wholly different conformation and occupies a different position in that enzyme.
GART have been solved in a ternary complex with glycinamide ribonucleotide and the folate inhibitor, 5-deaza-5,6,7,8tetrahydrofolate (5dTHF) (33). 5dTHF is very similar to THF, except that a carbon is substituted for nitrogen at position 5 of the pteridine ring. The superimposition of N t -FDH with the GART-glycinamide ribonucleotide-5dTHF complex provides a view of how ligands may fit into the active site of N t -FDH (Fig.  7). Specifically, it shows that ␤ME occupies the same spatial position as the expected position of the formyl group, adjacent to N-10 of THF. This suggests that in our structure ␤ME is a mimic of formate, such that the O-C-C backbone of ␤ME is equivalent to O-C-O of formate. More pertinently, it shows that Asp-142 is best placed to orient a hydrolytic water between N-10 and the carbonyl carbon of formyl group and that His-142 may orient the carbonyl oxygen of the formyl group. This has implications for the mechanism of hydrolase, which is discussed below.
FIG. 2. Electron density maps used in the structure determination of N t -FDH. a, the experimental electron density map of N t -FDH resulting from phasing using multiple isomorphous replacement, calculated at 3.0-Å resolution. b, the 2 (F o Ϫ F c ) electron density from the final structure, calculated at 2.3 Å. Shown is a region of ␤-sheet in the core of the C-terminal domain of N t -FDH. Both maps are contoured at 1 . This figure was prepared using PyMOL (51).

DISCUSSION
Similarity of N t -FDH, GART, and FMT-There are several enzymes in the cell that use 10-formyl-THF as a substrate. Three of these enzymes, GART, AICART, and FMT, use 10formyl-THF as a formyl donor in biosynthetic reactions. Although GART and FMT show very low sequence identity with each other (19%), they share the same fold for the folatebinding domain (24). By contrast, AICART shows no sequence similarity with either GART or FMT (34), and indeed its crystal structure reveals a different fold (35). FMT is 314 residues long (36) and contains a Rossmann fold, comprising ϳ200 amino acids, and a separate carboxyl-terminal domain of around 100 residues (24). GART is a shorter protein and contains only the Rossmann fold (37,38). N t -FDH shows sequence similarity with both of these Rossmann folds but not to the C-terminal domain of FMT (11). As expected then, the fold in this region of N t -FDH is highly similar to that for FMT and GART. Unpredicted, though, is the extension of this structural similarity in FMT to encompass the C-terminal domains of FMT and N t -FDH.
Role of the C-terminal Domain of N t -FDH-One issue that arises from the structural similarity noted above is the role of the C-terminal domain in N t -FDH. In FMT, this domain is responsible for binding methionyl-tRNA (24,39) and raises the possibility that this domain of N t -FDH can also bind nucleic acids. A nucleic acid binding activity for FDH has not been reported, but several other folate-metabolizing enzymes do act as autoregulatory translational repressors by binding to mRNA (40,41). Examination of the electrostatic surface of this region of N t -FDH, however, does not show a preponderance of basic charges that would be typical of a protein that binds nucleic acids but rather an equal distribution of positive and negative charges (data not shown). Interestingly, this domain is required for function, because deletion experiments show that at least 300 residues of this domain are required for the hydrolase activity of FDH, 2 although the folate-binding site and active site for the hydrolase activity are both located within the first 200 residues of the protein. This is surprising given that E. coli 10-formyl-THF hydrolase, the catalytic domain of which is shorter than 200 amino acid residues, can function as a hydrolase (42). Moreover, GART, which has about 200 amino acid residues, also removes the formyl group from 10-formyl-THF as part of its transferase reaction (43). How the C-terminal region of N t -FDH contributes to hydrolase activity, however, remains an unanswered question.
Conservation of Active Site Architecture in N t -FDH, FMT, and GART-Three important residues, asparagine, histidine, and an aspartate, have been identified in the active sites of GART and FMT (44,45), and these residues are strictly conserved among all known GARTs and FMTs. N t -FDH contains only two of these residues, His-106 and Asp-142, and these residues reside in the hydrophobic cleft that serves as a folatebinding pocket. Both of these are strictly required for the hydrolase and dehydrogenase activities of FDH (18,32), suggesting that the hydrolase reaction is an essential part of the dehydrogenase mechanism. The superimposed structures of FMT, GART, and N t -FDH showed that positions of the aspartate and histidine overlap very closely in all three enzymes (Fig. 6), suggesting that the removal of formyl from 10-formyl-THF is likely to proceed through a similar mechanism in all three enzymes. The third residue, however, differs in FDH; in place of an asparagine, there is an isoleucine. This is an interesting difference, because in GART this residue helps to orient the formyl group (46). The absence of this asparagine suggests that Asp-142 and His-106 are the most crucial residues for the function of N t -FDH.
Role of ␤-Mercaptoethanol in the Hydrolase Mechanism-N t -FDH catalyzes the hydrolysis of 10-formyl-THF to formate and THF and the same reaction can also be assayed in full-length FDH by excluding NADP ϩ . In both cases, however, the hydrolase reaction requires millimolar concentrations of ␤ME (17). In our crystal structure, five molecules of ␤ME are observed. Four of these are modifying cysteine residues and, as discussed above, are unlikely to have a significant impact on hydrolase activity. The remaining ␤ME, however, is positioned between both of the catalytic residues, His-106 and Asp-142, such that the sulfhydryl group is hydrogen-bonded to His-106, and the hydroxyl group is hydrogen-bonded to Asp-142. This suggests that ␤ME may be involved directly in the hydrolase mechanism rather than simply serving as a reducing agent (although it cannot be ruled out that the binding of ␤ME to the active site of N t -FDH is nothing more than happenstance, as indeed happens in many protein crystal structures, and has no bearing on how ␤ME acts in hydrolase activity). Since N t -FDH is an "artificial" enzyme, the most likely explanation is that ␤ME replaces a functionality lost in the absence of the dehydrogenase domain. A number of hypotheses can be proposed to explain the requirement for ␤ME in the hydrolase mechanism. 1) Cys-707, an essential residue of the aldehyde dehydrogenase domain, has a direct role in the hydrolase reaction, and in the absence of the dehydrogenase domain in N t -FDH, this function is replaced by the sulfydryl group of ␤ME. Examination of the structures of N t -FDH and of aldehyde dehydrogenase (47,48), however, suggests that this is very unlikely, because both active sites are located at the base of relatively deep clefts and Cys-707 could not access the active site of the hydrolase domain 2 S. Krupenko, unpublished observations. without major conformational changes in both domains of FDH. 2) In the absence of the aldehyde dehydrogenase domain, hydrolysis in N t -FDH proceeds via an entirely novel mechanism that requires ␤ME, perhaps as a nucleophile to attack the carbonyl carbon of the formyl group. Again, this seems unlikely, because it is hard to envisage what property of the dehydrogenase domain is being replaced by ␤ME and, furthermore, it would mean that the hydrolase reaction in N t -FDH can operate by two different mechanisms. 3) Finally, our preferred hypothesis is that ␤ME is required for displacement of the products of the hydrolase reaction. In this mechanism, the interaction between the hydrolase and aldehyde dehydrogen-ase domains in the full-length enzyme lowers the binding affinity of the products, formate and/or THF, in the hydrolase active site. Since this domain-domain interaction is absent in N t -FDH, either formate or THF remains bound after the hydrolase reaction, but these can be displaced by ␤ME in vitro.
Given the overlap of ␤ME with the expected position for formate, we suggest that ␤ME is required to displace formate that is positioned between His-106 and Asp-142. Further support to this hypothesis arises from the apparent flexibility in the 110residue linker region between the hydrolase and dehydrogenase domains, which may act as a hinge to bring the two active sites in close proximity (49). In fact, given that formate is a FIG. 3. The structure of N t -FDH. a, stereoview showing the structure as a C␣ trace in which every 10th residue is represented as a small sphere and labeled. The N and C termini are also labeled. b, ribbon representation in which the molecule is color-ramped blue to red in the N-to C-terminal direction. ␣-Helices and ␤-strands are labeled individually. This figure was prepared using MOLSCRIPT (51) and Raster3D (52). c, the sequence of N t -FDH and the secondary structure notation used in this paper. The color scheme for the secondary structure is the same as used in b. The arrows denote ␤-strands, and boxes represent helices.
substrate for the dehydrogenase reaction, it would be logical that the hydrolase domain retains this group until the active site of the dehydrogenase domain is brought alongside. The end result may be a protected channel connecting the two active sites that prevents loss of this group to the solvent and thus enhances the overall reaction rate.
Mechanism of Hydrolase Catalysis-The superimposition of N t -FDH with GART complexed with the folate substrate ana- FIG. 4. The structural similarity of the C-terminal regions of N t -FDH (left) and FMT (right). a, ribbon representation of each domain in which spatially equivalent elements of secondary structure are colored the same. This figure was prepared using MOLSCRIPT (51) and Raster3D (52). b, topology diagram of each domain using the same color scheme as a. Note the spatial similarity of ␤ 13 in N t -FDH to ␤ 14 in FMT (both colored pink) although these strands are not topologically related.

FIG. 5. A stereoview showing the active site residues of N t -FDH.
A molecule of ␤-mercaptoethanol, situated between Asp-142 and His-106, is shown set against its 2 (F o Ϫ F c ) electron density, contoured in blue at 1 . This figure was prepared using PyMOL (W. L. DeLano; available on the World Wide Web at www.pymol.org). log, 5-deazafolate (33), revealed the likely position of the formyl in N t -FDH, and the close proximity of His-106 and Asp-142 to this group is consistent with a role for these residues in the hydrolase mechanism (18). The proposed mechanism for the hydrolase reaction is shown in Fig. 8. In this scheme, a water molecule is polarized by Asp-142, facilitating a nucleophilic attack at the carbonyl carbon of formyl-THF. This generates an intermediate comprising a quarternary amine adjacent to a tetrahedral carbon center, in which the charge of the oxyanion is stabilized by His-106, whereas Asp-142 stabilizes that of the cationic nitrogen. In the next step, the intermediate collapses to form the products, THF and formate. By polarizing the catalytic water molecule, Asp-142 also facilitates the abstraction of a proton by the tetrahydrofolate product. After the reaction is complete, formate is held in place by hydrogen bonds with His-106 and Asp-142. In the full-length enzyme, this formate is then passed to the dehydrogenase domain by an unknown mechanism (see below), but in N t -FDH it can only be displaced by ␤ME.
Hence, in our crystal structure, the view is after the hydrolase reaction is completed and ␤ME now occupies the same position as formate. A similar mechanism for the removal of formyl from the folate substrate is likely to operate in both FMT and GART, which also contain the same aspartate and histidine residues, and indeed this has been proposed for GART (50). A potential catalytic water molecule has been observed in each crystal structure of GART and FMT (24,50).
Communication between the Hydrolase and Dehydrogenase Domains-Since the product of the hydrolase reaction is probably a substrate for the dehydrogenase reaction, a key question remaining is how the hydrolase and dehydrogenase domains communicate in the full-length enzyme. As described above, the two active sites are both quite buried and lie within concave regions of their respective domains. Given this, it would appear improbable that residues of the hydrolase active site can access the active site of the dehydrogenase domain (or vice versa) to facilitate direct transfer of the formyl group. However, one mechanism of communication is suggested by the architecture of the ␤-sheet in the hydrolase domain. Interestingly, ␤ 7 and ␤ 8 form a hairpin that lies distinctly out of the plane of the sheet (Fig. 9), and Asp-142 is located at the pinnacle of the turn between these strands. Were this hairpin to fall back toward the plane of the sheet, Asp-142 would become increasingly exposed on the surface of the protein, making it possible for this residue, at least, to reach the active site of the dehydrogenase domain. How an aspartate would facilitate transfer of a formyl group is unclear, but in support of this hypothesis, the equivalent loop in GART undergoes a conformational transition in FIG. 6. A superimposition of the active sites of N t -FDH, FMT, and GART. In this stereoview, the backbone of the loops of the active sites are colored blue for N t -FDH, orange for GART, and green for FMT. The conserved residues, histidine, aspartate, and isoleucine/asparagine, are also shown clustered in the active sites of the three proteins. This figure was prepared using MOLSCRIPT (51) and Raster3D (52).
FIG. 7. ␤-Mercaptoethanol mimics the position of the formyl group. The figure was produced by superimposing the structure of N t -FDH with that of the complex of GART with the inhibitor 5dTHF. Shown is the juxtaposition of Asp-142, His-106, and ␤ME from N t -FDH (purple bonds) with 5dTHF (green bonds). Note the position of ␤ME near the formyl attachment site at position 10 on THF and the proximity of Asp-142 to the site of hydrolysis. This figure was prepared using MOLSCRIPT (51) and Raster3D (52). response to changes in pH (46). This hypothesis, of course, is highly speculative, and a better idea of how the formyl passes between domains must await a crystal structure of the fulllength enzyme.
FIG. 8. Proposed mechanism for the hydrolase reaction of FDH. In this scheme, Asp-142 polarizes a water molecule for hydrolysis of the bond between the carbonyl carbon of formyl and N-10 of THF, and His-106 helps orient the carbonyl group. After catalysis, formate remains bound in the active site until displaced by a molecule of ␤ME.

FIG. 9.
A possible mechanism of communication between the hydrolase and dehydrogenase domains of FDH. It is postulated that the loop between ␤ 7 and ␤ 8 , which carries the active site residue Asp-142, can flip back into the plane of the sheet. In so doing, Asp-142 would become exposed on the surface of the hydrolase domain and thus be accessible to the active site of the dehydrogenase domain. This figure was prepared using MOLSCRIPT (51) and Raster3D (52).