Crystal Structure of Escherichia coli PdxA, an Enzyme Involved in the Pyridoxal Phosphate Biosynthesis Pathway*

Pyridoxal 5′-phosphate is an essential cofactor for many enzymes responsible for the metabolic conversions of amino acids. Two pathways for its de novo synthesis are known. The pathway utilized by Escherichia coli consists of six enzymatic steps catalyzed by six different enzymes. The fourth step is catalyzed by 4-hydroxythreonine-4-phosphate dehydrogenase (PdxA, E.C. 1.1.1.262), which converts 4-hydroxy-l-threonine phosphate (HTP) to 3-amino-2-oxopropyl phosphate. This divalent metal ion-dependent enzyme has a strict requirement for the phosphate ester form of the substrate HTP, but can utilize either NADP+ or NAD+ as redox cofactor. We report the crystal structure of E. coli PdxA and its complex with HTP and Zn2+. The protein forms tightly bound dimers. Each monomer has an α/β/α-fold and can be divided into two subdomains. The active site is located at the dimer interface, within a cleft between the two subdomains and involves residues from both monomers. A Zn2+ ion is bound within each active site, coordinated by three conserved histidine residues from both monomers. In addition two conserved amino acids, Asp247 and Asp267, play a role in maintaining integrity of the active site. The substrate is anchored to the enzyme by the interactions of its phospho group and by coordination of the amino and hydroxyl groups by the Zn2+ ion. PdxA is structurally similar to, but limited in sequence similarity with isocitrate dehydrogenase and isopropylmalate dehydrogenase. These structural similarities and the comparison with a NADP-bound isocitrate dehydrogenase suggest that the cofactor binding mode of PdxA is very similar to that of the other two enzymes and that PdxA catalyzes a stepwise oxidative decarboxylation of the substrate HTP.

for the metabolic conversions of amino acids. Vitamin B 6 (pyridoxine) and its derivatives are also efficient singlet oxygen quenchers and potent fungal antioxidants (1). Two different pathways for de novo synthesis of pyridoxine are now recognized. One of these, found in all Archaea, eukaryotes, and in some bacteria, uses the singlet oxygen resistance (SOR1(Pdx1)) gene product, a highly conserved enzyme (2). A number of eubacteria, including Escherichia coli, utilize a specific pathway for pyridoxal phosphate synthesis that is distinct and has been characterized for some time (3). Analysis of sequences from several genomes has revealed that organisms may encode either the SOR1 or E. coli-like pyridoxine biosynthesis genes, but not both (4).
Treatment of purified PdxA with 1 mM EDTA abolishes oxidation of HTP, suggesting the presence of a tightly bound divalent metal ion. Addition of 1 mM Mn 2ϩ , Co 2ϩ , Mg 2ϩ , or Ca 2ϩ restores full activity (9), whereas 1 mM Ni 2ϩ or 2 mM Zn 2ϩ restored half of the original PdxA activity. Although the product of the reaction, 3-amino-1-hydroxyacetone 1-phosphate, undergoes facile dimerization in the absence of PdxJ and the co-substrate deoxyxylulose phosphate, we have directly detected 3-amino-1-hydroxyacetone 1-phosphate by electrospray ionization mass spectrometry of PdxA incubation mixtures. 2 None of the putative intermediate, 2-amino-3-oxo-4-hydroxybutyric acid 4-phosphate, could be detected in the reaction mixture, even as early as 30 s after initiation of the reaction, suggesting that if this compound is formed, it never leaves the PdxA active site.
In principle, PdxA might catalyze either a stepwise or a concerted oxidative decarboxylation of HTP. For example, threonine dehydrogenase catalyzes the biochemically similar, reversible NAD-dependent oxidation of L-threonine to L-2-amino-3-ketobutyrate (Scheme 2) (11). The product, L-2-amino-3ketobutyrate, which is normally converted to glycine and acetyl-CoA by 2-amino-3-ketobutyrate CoA ligase, can undergo pH-dependent decarboxylation, with a half-life ranging from 8.6 min at pH 5.9 to 140 min at pH 11.1 (12). PdxA and threonine dehydrogenase show no significant amino acid sequence similarity. By contrast, use of the PSI-BLAST sequence comparison algorithm (13) reveals ϳ30 -35% sequence identity over a short ϳ25 residue segment between E. coli PdxA and both isocitrate dehydrogenase and 3-isopropylmalate dehydrogenases (9). Interestingly, each of the latter two enzymes catalyzes a nicotinamide-and divalent metal ion-dependent oxidative decarboxylation of a ␤-hydroxy acid substrate (Scheme 2). Significantly, the isocitrate dehydrogenase reaction has been shown to proceed by a stepwise mechanism that involves the corresponding 3-keto acid, oxalosuccinate intermediate, based on multiple isotope effect ( 2 H, 13 C) studies, as well as the ability of isocitrate dehydrogenase to catalyze both the decarboxylation and the reduction of oxalosuccinate, with a ϳ10-fold preference for decarboxylation (14).
Structural characterization of the enzymes of the pyridoxine biosynthetic pathway has so far been limited to E. coli PdxJ (15)(16)(17) and pyridoxine-5Ј-phosphate oxidase from Saccharomyces cerevisiae (Protein Data Bank number 1CIO), the enzyme that oxidizes pyridoxine 5Ј-phosphate and pyridoxamine 5Ј-phosphate to pyridoxal 5Ј-phosphate. The x-ray structure of PdxA bound to Zn 2ϩ as well as the HTP presented here reveals a protein that can be divided into two subdomains, and which shares structural similarities with both isocitrate dehydrogenase and isopropylmalate dehydrogenase. These structural similarities and, in addition, sequence conservation between PdxA and the two dehydrogenases of residues involved in the cofactor binding argue for a similar manner of nicotinamide cofactor binding and similar, stepwise, biochemical mechanisms for oxidative decarboxylation. Furthermore, based on the complex with HTP in the presence of Zn 2ϩ , sequence conservation analysis among PdxA enzymes from different sources, and structural comparison with the isocitrate dehydrogenase family, the location of the PdxA active site is stipulated to be in a cleft between the two subdomains and at the interface between the two molecules of the dimer, involving a cluster of strictly conserved amino acids.

MATERIALS AND METHODS
Cloning, Expression, and Purification-The pdxA gene was cloned into a derivative of the pET-15b vector (Amersham Biosciences) con-taining a thrombin cleavage site to obtain an in-frame N-terminal fusion with a His 6 purification tag. Plasmid DNA was transformed into the E. coli methionine auxotroph DL41 for selenomethionine protein production (18). The transformed bacteria were grown at 37°C to an A 600 of ϳ0.8 in LeMaster medium supplemented with 25 mg/liter of L-selenomethionine. A 1-liter culture was induced with 100 M isopropyl-1-thio-␤-D-galactopyranoside and the culture was continued at room temperature for an additional 15 h.
Cells were harvested by centrifugation (4000 ϫ g, 4°C, 25 min) and were re-suspended in 40 ml of lysis buffer (50 mM Tris-HCl, pH 7.5, 0.4 M NaCl, 5% (w/v) glycerol, 20 mM imidazole, 10 mM ␤-mercapthoethanol) containing one dissolved tablet of Complete TM protease inhibitor mixture (Roche Diagnostics). Cells were lysed by sonication on ice for a total of five 30 s cycles with 45 s between each cycle for cooling. The lysate was then cleared by centrifugation (100,000 ϫ g, 4°C, 30 min). The protein supernatant was first loaded on a 5-ml DEAE-Sepharose (Pharmacia) column equilibrated in lysis buffer and the flow-through fraction was collected. This protein was then applied to a 5-ml nickelnitrilotriacetic acid (Qiagen) column, pre-equilibrated with lysis buffer. The column was washed extensively with buffer (50 mM Tris, pH 7.5, 50 mM imidazole, 0.4 M NaCl) and bound proteins were eluted with the same buffer containing 150 mM imidazole. Purified PdxA ran as a single band on both SDS-PAGE and native PAGE gels. Dynamic light scattering measurements were performed using a DynaPro801 molecular sizing instrument (Protein Solutions) at a protein concentration of 8.1 mg ml Ϫ1 and were carried out at 22°C.
Form II crystals were obtained from the same starting protein solution using 20% (w/v) PEG 8000, 100 mM Hepes, pH 7.5, 75 mM citrate, and 100 mM MgCl 2 as a reservoir condition. These crystals belong to the monoclinic system, space group P2 1 , with cell dimensions a ϭ 92.1, b ϭ 76.7, c ϭ 95.4 Å, ␤ ϭ 109.7 o , and four molecules in the asymmetric unit. These crystals diffracted to 2.45-Å resolution.
Crystals were briefly soaked in a cryoprotectant solution consisting of mother liquor supplemented with 20% (w/v) glycerol, picked up in a nylon loop, and flash cooled at 100 K in a N 2 (gas) cold stream (Oxford Cryosystems, Oxford, United Kingdom). Data were collected at beamline X8C, NSLS, Brookhaven National Laboratory using a Quantum 4-CCD detector (ADSC). All data sets, including Se multiwavelength anomalous diffraction data (Table I) were processed using the program HKL2000 (19).
Structure Solution and Refinement-Crystals of form I were used for structure determination. For phase determination, the resolution range from 12 to 2.4 Å was chosen. Twelve of the expected 16 Se sites within the asymmetric unit were found using the program SOLVE (20). The resulting phases gave an overall figure of merit of 0.58. Further improvement of the phases was achieved by using the program RESOLVE SCHEME 1. PdxA-catalyzed oxidative decarboxylation of HTP to 3-amino-1-hydroxyacetone 1-phosphate (AHAP) and conversion of AHAP to pyridoxol phosphate.

SCHEME 2. Nicotinamide-dependent oxidation of 3-hydroxy acids catalyzed by (a) threonine dehydrogenase, (b) isocitrate dehydrogenase, and (c) 3-isopropylmalate dehydrogenase.
(21), resulting in an increase of the overall figure of merit to 0.77. The starting electron density map was of good quality with 69% of the starting model built automatically using RESOLVE. Remaining parts of the model were built manually using the program O (22). Further cycles of model building alternating with refinement using the program CNS (23) resulted in the final model, with an R-factor of 0.217 (R free ϭ 0.267). Non-crystallographic symmetry restraints were used in the early stages of refinement, but were removed as the resolution was extended. No cutoff was used during refinement. Final refinement statistics are given in Table I. The final model comprises 328 residues for each monomer, one Zn 2ϩ , one PO 4 3Ϫ , and 267 water molecules, with the first well ordered residue in each monomer being Ala 2 . The His tag residues from the thrombin cleavage site as well as Met 1 did not have interpretable density and were not modeled.
The structure of PdxA from form II crystals was solved by the molecular replacement method using the program AmoRE (24). The starting model was the structure of the dimer of PdxA from the form I crystal. Four molecules in the asymmetric unit were clearly identified from the two top peaks of the rotation function search. The details of data collection, refinement, and quality of the model are given in Table  I. The root mean square deviation between the two crystal form structures of PdxA is 1.07 Å for all the C␣ atoms of the dimer.
Structure of the PdxA-HTP Complex-For preparation of the complex of PdxA with 4-hydroxy-L-threonine-4-phosphate, a form I crystal was soaked in mother liquor containing 7 mM HTP for 7 days. A complete data set was collected to 2.25-Å resolution using the home source. Electron density corresponding to the HTP product was observed in the difference Fourier map, and was modeled using the program O. Refinement was performed using CNS as for the apo PdxA structure.
Coordinate Deposition-Coordinates for native PdxA form I and II and the complex of form I with HTP have been deposited with the Research Co-laboratory for Structural Bioinformatics (RCSB) with Protein Data Bank codes 1PTM, 1PS7, and 1PS6, respectively.

RESULTS
Structure of the PdxA Monomer-The structure of selenomethionine-labeled PdxA from E. coli was solved by the multiwavelength anomalous diffraction method and refined to a final R-factor of 0.217 (R free ϭ 0.267) at 1.96-Å resolution for the orthorhombic crystal form of native PdxA, R-factor of 0.213 (R free ϭ 0.272) for the monoclinic crystal form of native PdxA at 2.45-Å resolution, and an R-factor of 0.206 (R free ϭ 0.258) for the complex of PdxA with HTP at 2.25-Å resolution. These models have been refined with good stereochemical parameters (Table I). Statistics for the Ramachandran plot from an analysis using PROCHECK (25) for the model gave over 90% of non-glycine residues in the most favored region.
E. coli PdxA has an ␣/␤/␣ architecture with a central 12stranded mixed ␤-sheet flanked on both sides by ␣-helices. The ␤-strand order within this sheet is 32-42-22-12-52-122 -111-62-71-101-81-91 ( Fig. 1) with the first six strands being parallel, followed by two antiparallel strands and four parallel strands running in the opposite direction to the first strands. This extended sheet is twisted along its longitudinal axis (perpendicular to the strands) by ϳ180 o , so that the first few and last few strands while antiparallel, run in the same direction (Fig. 2). There are nine ␣-helices longer than one turn, aligned nearly parallel to the ␤-strands. Five helices line one side of the sheet, whereas three helices lie along the other side of the sheet in its N-terminal part. Although there is no clear distinction of domains, the molecule could be divided into two subdomains, with residues Ala 2 -Lys 147 and Leu 288 -Ala 329 forming subdomain 1 and Lys 148 -Gly 287 forming subdomain 2. The N and C termini are within subdomain 1 and are in close proximity to each other, separated by only 9.5 Å. The substratebinding site is located near the middle of the ␤-sheet at the interface between the two subdomains, on the side with fewer ␣-helices. The overall fold of PdxA is clearly related to that found in the isocitrate/isopropylmalate dehydrogenases (26,27) as classified in the Structural Classification of Proteins data base (28).
Structure of the PdxA Dimer-Dynamic light scattering measurements indicate that the PdxA molecules form homodimers in solution. This is in contrast to an earlier study where PdxA was characterized as being monomeric based on gel filtration data (8). The presence of homodimers is consistent with the dimeric arrangement observed in the crystal structure, with each dimer having approximate dimensions of 88 ϫ 40 ϫ 32 Å. The two monomers of the dimer are related by a 2-fold non-crystallographic symmetry axis parallel to the crys-tallographic a axis. The structures of the two independent monomers in the asymmetric unit are very similar, showing root mean square (r.m.s.) deviation of 1.2 Å for all C␣ atoms, when refined without non-crystallographic symmetry restraints. Independent superposition of subdomains 1 or 2 gives an r.m.s. deviation of ϳ0.6 Å indicating that there is a small difference in the arrangement of the two subdomains relative to one another in the independent molecules.
The dimer is formed through the interactions of subdomains 2 of the PdxA monomers. These interactions are mediated by the ␣8 helices (residues 269 -275) from each monomer and additional contacts from residues in the loops following parallel ␤-strands ␤7, ␤8, ␤9, and ␤10 (Fig. 3). There are a total of 16 hydrogen bonds (Ͻ3.2 Å), including 6 salt bridges, and a network of hydrophobic interactions. Analysis of the multiple sequence alignment for enzymes showing sequence similarity to PdxA (PFAM accession number PF04166) reveals that the residues found at the dimer interface are well conserved, suggesting that the observed mode of dimerization is common to all members of this family. A pair of Zn 2ϩ ions is observed at the dimer interface, with residues from both monomers contributing to the coordination environment for each Zn 2ϩ site (Fig. 3). The surface area buried upon dimer formation calculated using a 1.4-Å probe radius is 1240 Å 2 or ϳ12% of the total surface area of each monomer.
Sequence and Structural Similarity-PdxA from E. coli and its orthologs form a conserved family of bacterial enzymes, and contain in particular the ␥-subdivision of protobacteria (7). Sequence analysis using PSI-BLAST (13) finds homologous sequences in 73 bacterial species and one similar sequence in the archaeaon Ferroplasma acidarmanus. The sequence identity between E. coli PdxA and the bacterial orthologs varies from 93% for Salmonella to 27% for Burkholderia fungorum. There is 31% sequence identity between the E. coli PdxA and that from archaeal F. acidarmanus. The conserved residues cluster in the substrate-and cofactor-binding areas as well as at the dimerization interface (Fig. 3).
Structural comparison of E. coli PdxA with other protein structures was performed using the program DALI (29). Clearly the most structurally similar proteins were found to be isocitrate dehydrogenase (Protein Data Bank code 1ISO, Z ϭ 16.5, r.m.s. of 4.0Å for 244 C␣ atoms) and isopropylmalate dehydrogenase (Protein Data Bank code 1CNZ, Z ϭ 16.3, 5.1 Å for 249 C␣ atoms). The limited amino acid sequence similarity between PdxA and the other two proteins evident from PSI-BLAST comparisons, as summarized above, has been previously noted (9). Notably, all three proteins catalyze analogous decarboxylative dehydrogenations of 3-hydroxy acids. The su- perposition of PdxA with the two above mentioned dehydrogenases shows that nine ␤-strands and seven ␣-helices are structurally and topologically equivalent in these proteins. With more stringent conditions 122 C␣ atoms of PdxA, mostly from the ␤-strands, can be superimposed on 1ISO with an r.m.s. deviation of 1.65 Å (sPDBv (30)). Furthermore, the dimerization interface of PdxA and the other two dehydrogenases involves the same regions of the proteins and is similar overall. Structural alignment shows 16 identical residues of 329 for PdxA common to all three sequences (Fig. 4).
Zn 2ϩ Binding Sites-In the electron density map of the native protein there are two strong peaks corresponding to metal ions. Based on the coordination and the type of liganding side chains we have interpreted these peaks as two Zn 2ϩ ions. As no Zn 2ϩ ions were explicitly part of the crystallization solution, these metal ions must have been acquired during expression in E. coli cells and remained associated with the enzyme throughout the purification and crystallization process. The presence of a tightly bound divalent metal cation has been previously reported by Cane et al. (9) as summarized earlier. The observed Zn 2ϩ ions are at the dimer interface and are coordinated by His 166 and His 266 from one monomer and His 211 from the second monomeric unit. The almost perfect octahedral coordination sphere of each Zn 2ϩ atom is completed by three water molecules (Fig. 5). The distances between the Zn 2ϩ and the liganding nitrogen or oxygens range from 2.1 to 2.3 Å. These three histidines are strictly conserved in all 74 homologous amino acid sequences indicating that all enzymes from this family are most likely Zn 2ϩ -dependent enzymes (PFAM family PF04166). 3 The coordination of each Zn 2ϩ ion by absolutely conserved histidine residues coming from both monomers of the dimer provides strong indication that dimerization is essential for enzymatic activity.
Substrate Binding Site-We have soaked the native crystals in a solution containing 7 mM substrate, HTP. The difference electron density map clearly showed a substrate molecule bound to one of the two molecules of the PdxA dimer (Fig. 6a). The HTP binds in a deep depression in the surface of the molecule. One side of this depression is lined by residues from the second monomer of the dimer, which converts it into a deep and rather narrow cleft (Fig. 6b). Superposition of the two independent molecules of the dimer shows that they differ somewhat in the disposition of the two subdomains. The molecule of HTP is bound to the monomer that has a less open conformation. In the native crystals of PdxA the same monomer subunit binds an inorganic phosphate ion in exactly the same place as the phosphate group of HTP. The somewhat larger opening of this cleft in the second subunit causes small displacements of atoms that would otherwise provide hydrogen bonds to the phosphate group, abolishing phosphate binding.
The incoming HTP molecule displaces the phosphate anion as well as the three water molecules that complete the Zn 2ϩ coordination sphere. The phosphate group of the HTP replaces the inorganic phosphate ion, whereas the N and OG (hydroxyl oxygen) atoms replace the water molecules, resulting in a penta-coordinated central Zn 2ϩ ion (Fig. 6c). In this HTP-Zn 2ϩ complex the N, C-2, C-3, and OG atoms are essentially constrained to a common plane, leading to the eclipsed conformation of the scissile C1-C2 and C3-H bonds. This arrangement also establishes the potential for a hydrogen bond between the N atom of the HTP and N⑀-2 of His 266 and between OG of HTP and the N⑀-2 atoms of both His 166 and His 211 , the latter protruding from the other monomeric subunit of the dimer. In addition, the carboxylate group of HTP is hydrogen bonded to the conserved residue Lys 274 and the phosphate group is hydrogen bonded to Arg 292 (conserved in ϳ90% of sequences) as well as to the backbone NH of Thr 137 , found within the highly conserved sequence (Gly-His-Thr-Glu) (Fig. 6c).
Putative NAD(P) Binding Site-The PdxA enzyme utilizes NAD ϩ or NADP ϩ as the cofactor for oxidation of HTP. Although we grew crystals of PdxA in the presence of up to 10 mM of either of these cofactors, with and without HTP, in no case could electron density corresponding to the nicotinamide cofactor be found in the map. Efforts to soak the cofactors into the crystals were also unsuccessful. This may indicate that HTP normally binds first, although this has yet to be demonstrated experimentally. Although we did not directly locate the position of the NAD(P) ϩ cofactor in PdxA, its approximate position can be inferred by a comparison with the structure of the E. coli isocitrate dehydrogenase S113E mutant complexed with NADP ϩ , isopropylmalate, and magnesium (Protein Data Bank code 1HJ6) (31). In this structure the diphosphate moiety of the NADP ϩ cofactor straddles the extended segment Thr 338 -Lys 344 , part of the loop following a ␤-strand, and the adenine and 3 www.sanger.ac.uk/cgi-bin/Pfam/getacc?. nicotinamide rings dip into depressions on either side of the loop. The adenine interacts with His 339 on one side and Tyr 345 on the other, whereas the nicotinamide packs between Gly 340 and the substrate. The 2Ј-phosphate of the adenosine fragment is on the solvent exposed edge of the NADP. The corresponding part of the structure of PdxA (strand ␤12 and the following loop) superimposes very well on that of isocitrate dehydrogenase where an equivalent segment of polypeptide (His 297 -Gly 298 ) is also present. Furthermore, these residues together with the following residue (Thr 299 ) are conserved in the sequences from the PdxA family. We superimposed the structures of PdxA complexed with HTP, and the isocitrate dehydrogenase mutant complexed with NADP (Protein Data Bank code 1HJ6). Fig. 6d shows the structure of the PdxA-HTP complex with a model of NADP ϩ from the isocitrate dehydrogenase according to the above mentioned superposition.

Substrate
Binding-The present structure shows that binding of the HTP substrate requires a precise degree of opening between the subdomains and that the flexibility of the PdxA molecule is essential for substrate binding and product release. Moreover, the structure of the PdxA-HTP complex shows that the Zn 2ϩ ion, coordinated by three conserved histidines, plays an essential role in substrate binding and that the phosphate group of HTP contributes substantially to this binding. The observed role of the phosphate group in HTP binding provides a structural explanation for the previous observation that the free alcohol 4-hydroxy-L-threonine is not a substrate for PdxA (8).
Cofactor Binding and Catalytic Mechanism-As illustrated in Fig. 6b, the 2Ј-phosphate of the adenosine fragment is on the solvent exposed edge of the NADP ϩ and would make few contacts with the protein, therefore explaining the dual specificity of PdxA for either cofactor, NAD ϩ or NADP ϩ . The nicotinamide ring of the cofactor, if positioned as in isocitrate dehydrogenase, would be in a reasonable proximity to the bound substrate HTP molecule, with the proR face of its C-4 atom positioned ϳ4 Å from the H-3 proton of HTP that is to be removed during FIG. 6. The molecule of HTP substrate bound to PdxA. a, the electron density in the omit map corresponding to the bound HTP in the active site of the PdxA molecule contoured at 2 . The residues in the radius of 3.5 Å around the HTP and Zn 2ϩ were removed from the calculation of phases. b, molecular surface of the dimer around the substrate binding site, with the HTP shown in stick mode. Each molecule is colored differently. The model of NADP ϩ was transferred from the structure of isocitrate dehydrogenase, 1HJ6. PdxA and 1HJ6 were superimposed as shown in Fig. 4a. c, the hydrogen bonding network involving HTP and the coordination of Zn 2ϩ ion. d, a model of the active site of PdxA with bound NADP ϩ . The cofactor was modeled as described in b. Dashed lines show the contacts between zinc and liganding atoms and the close contact between C-4 of NADP and C-3 of HTP. oxidation (Fig. 6d). The calculated C4 . . . H-C3 angle of ϳ150 o in the crude model is consistent with this hypothesis. By comparison, a C4 . . . H-C bond distance of 2.70 Å has been calculated for the reactive ternary nicotinamide-isocitrate complex of isocitrate dehydrogenase (32). Complexation with Zn 2ϩ should decrease the pK a of the C-3 hydroxyl (OG), similar to that in other metal ion-dependent dehydrogenases. Furthermore, because OG of HTP is within hydrogen bonding distance to the N⑀-2 of His 166 and His 266 , which are likely neutral at the pH of crystallization, the proton from this hydroxyl could be taken up by one of the histidines and transferred to the solvent, as these histidines are on the surface of the molecule. His 266 in addition makes a hydrogen bond to the conserved Asp 267 , which likely assures its proper orientation for coordination of the Zn 2ϩ ion. Conservation of Asp 247 may be explained by its interaction with Lys 274 and, through a bridging water molecule, with His 266 and Asp 267 of the second molecule, therefore contributing to the proper orientation of side chains coordinating the substrate, and at the same time to dimerization.
Comparison with Isocitrate and Isopropylmalate Dehydrogenases-PdxA, isocitrate dehydrogenase, and isopropylmalate dehydrogenase all catalyze nicotinamide-and divalent metal ion-dependent oxidative decarboxylations of 3-hydroxy acids. Both isocitrate dehydrogenase (NAD ϩ -and NADP ϩ -dependent forms) and isopropylmalate dehydrogenase have all been shown to utilize the proR face of the nicotinamide cofactor, corresponding to the same stereospecificity predicted herein for PdxA (14). In contrast to PdxA, both isocitrate dehydrogenase and isopropylmalate dehydrogenase utilize Mg 2ϩ instead of Zn 2ϩ . The comparison of the structures shows that the location of the cation is not the same in these enzymes. Indeed, the three histidines that coordinate Zn 2ϩ in PdxA, and are conserved in all PdxA enzymes, are not present in the other two dehydrogenases. In fact, previous kinetic studies of PdxA have shown that substitution of Mg 2ϩ for Zn 2ϩ results in a 2-fold increase in k cat , without establishing which metal is present in the native protein (9). More significantly, the divalent metal in both isocitrate dehydrogenase and isopropylmalate dehydrogenase is complexed between the oxygen atoms of the hydroxyl and ␣-carboxylate of the substrate. Isocitrate and isopropylmalate are held in nearly identical conformations, with a ϳ165°torsion angle between the scissile C2-H and C3-carboxylate bonds (33), in contrast to the 0°torsion angle observed for the corresponding bonds in HTP. Both the eclipsed and anti-FIG. 6-continued geometries, however, are consistent with a mechanism in which the scissile C-carboxylate bond in the initially generated 3-keto carboxylate intermediate is essentially orthogonal to the plane of the newly formed carbonyl group, as required for the subsequent decarboxylation step. It is therefore particularly significant that isocitrate dehydrogenase has been shown to catalyze a stepwise oxidative decarboxylation involving the intermediacy of oxalosuccinate (14). Finally, the substrate binding sites of isocitrate dehydrogenase and isopropylmalate dehydrogenase are highly conserved, with the labile carboxylate being bound by interactions with active site Lys and Arg residues and hydrogen bonded to the hydroxyl group of a tyrosine (33), similar to the observed interactions of the carboxylate of HTP with Lys 274 and Asn 283 as well as a bound molecule of water.