The 2.3-Å Crystal Structure of the Shikimate 5-Dehydrogenase Orthologue YdiB from Escherichia coli Suggests a Novel Catalytic Environment for an NAD-dependent Dehydrogenase*

We present here the 2.3-Å crystal structure of the Escherichia coli YdiB protein, an orthologue of shikimate 5-dehydrogenase. This enzyme catalyzes the reduction of 3-dehydroshikimate to shikimate as part of the shikimate pathway, which is absent in mammals but required for the de novo synthesis of aromatic amino acids, quinones, and folate in many other organisms. In this context, the shikimate pathway has been promoted as a target for the development of antimicro-bial agents. The crystal structure of YdiB shows that the protomer contains two (cid:1) / (cid:2) domains connected by two (cid:1) -helices, with the N-terminal domain being novel and the C-terminal domain being a Rossmann fold. The NAD (cid:3) cofactor, which co-purified with the enzyme, is bound to the Rossmann domain in an elongated fash-ion with the nicotinamide ring in the pro- R conformation. Its binding site contains several unusual features, including a cysteine residue in close apposition to the nicotinamide ring and a clamp over the ribose of the adenosine moiety formed by phenylalanine

The shikimate pathway is essential to bacteria, fungi, plants, and parasites but is not used in mammals (1,2), making the enzymes involved in this pathway attractive targets for the development of broad spectrum antibiotic drugs (3)(4)(5)(6) and herbicides (7). The pathway in Plasmodium, Toxoplasma, Cryptosporidium, and Eimeria parasites has attracted attention recently (1, 2) but also holds promise for the development of drugs to inhibit the growth of fungi and bacteria (8). The shikimate pathway in bacteria has also been considered as a route to industrial production of hydroaromatic compounds for use in chemical syntheses (9 -11).
In this article, we report the crystal structure of the YdiB protein, which is one of the two shikimate 5-dehydrogenase orthologues in E. coli (12). It is common to find multiple orthologues of individual enzymes in the shikimate pathway in microorganisms. For instance, the genome of Bacillus subtilis (13) carries two versions of the enzyme 3-dehydroquinate dehydratase (products of the aroD and aroQ genes) but just a single version of each of the next two enzymes in the pathway, shikimate 5-dehydrogenase (aroE) and shikimate kinase (aroK). In contrast, this pattern is reversed in the genome of Escherichia coli K12 (14), which carries just a single orthologue of 3-dehydroquinate dehydratase (aroD) but duplicate orthologues for both shikimate 5-dehydrogenase (products of the ydiB and aroE genes) and shikimate kinase (products of the aroK and aroL genes). The ydiB gene in E. coli K12 is cocistronic with the aroD gene, i.e. the only orthologue of 3-dehydroquinate dehydratase found in this organism. The YdiB protein is 28% identical and 50% similar to the AroE protein (12), the other shikimate 5-dehydrogenase orthologue in E. coli, but the two proteins share dramatically stronger sequence conservation in their substrate binding sites (Figs. 2C and 4C).
Interestingly, the genome of the pathogenic E. coli strain O157 (15) contains one additional orthologue of shikimate 5-dehydrogenase (i.e. three total) and at least two orthologues for every other enzyme in the shikimate pathway. Although the redundancy of these enzymes in many pathogenic organisms suggests that their function is important under physiological conditions, it also raises questions as to the feasibility of using shikimate pathway inhibitors as antibiotics. However, the results reported in this article show very tight conservation of the substrate binding site in all orthologues of shikimate 5-dehydrogenase, suggesting that the design of broad spectrum antibiotics directed against this enzyme is possible (16 -21).
Shikimate 5-dehydrogenase (EC 1.1.1.25) functions to catalyze the reduction of 3-dehydroshikimate to shikimate using the cofactor NADH (12, 14) (Fig. 1). This enzyme belongs to the superfamily of NAD(P)H-dependent oxidoreductases, which function in anabolic and catabolic enzyme pathways as well as in xenobiotic detoxification. This superfamily is usually subdivided into several families, including short chain dehydrogenases (22,23), medium chain dehydrogenases (24), aldo-keto reductases (25), and iron-activated alcohol dehydrogenases and long chain dehydrogenases (26,27). The reaction mechanism used by these enzymes involves the catalysis of hydride transfer from the NAD(P)H cofactor on the basis of stabilization of the negative charge accumulation in the hydride-accepting substrate by either an amino acid side chain acting as a general base proton donor or a prosthetic cation in the active site. The crystal structure of the YdiB shikimate dehydrogenase orthologue from E. coli that is reported in this article shows variations in the identity and spatial distribution of the active site residues compared with any other known oxidoreductase structure, suggesting that the details of its catalytic reaction mechanism are likely to be different from the other enzymes in the NAD(P)H-dependent oxidoreductase superfamily.

MATERIALS AND METHODS
Protein Expression and Purification-The full-length ydiB gene from E. coli was cloned into a pET21d (Novagen) derivative. The gene was expressed in E. coli BL21(DE3) cells (containing the rare tRNA expression plasmid pMGK), yielding a protein containing the entire native sequence plus an eight-residue affinity tag at its C terminus (LEHH-HHHH). Protein was produced in MJ9 minimal medium (28) supplemented for selenomethionine labeling (29). Expression was induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside overnight at 17°C after growth to mid-log phase at 37°C. Cells were lysed by sonication in 50 mM NaH 2 PO 4 , 300 mM NaCl, 10 mM imidazole, and 5 mM ␤-mercaptoethanol, pH 8.0. After centrifugation at 26,000 ϫ g for 45 min at 4°C, the clarified extract was loaded onto a nickel-nitrilotriacetic column (Qiagen) and eluted in the same buffer plus 250 mM imidazole. After purification on a Superdex 75-gel filtration column (Amersham Bio-sciences) equilibrated in 10 mM Tris, 100 mM NaCl, and 5 mM dithiothreitol, pH 8.0, the protein was concentrated to 10 mg/ml for crystallization.
Enzyme Assays-Activity was monitored on the basis of the increase of A 340 nm on the reduction of NAD ϩ (assuming an extinction coefficient of 6230 cm Ϫ1 M Ϫ1 for NADH) in the buffer used previously for AroE (30) (100 mm Na 2 CO 3 and 2 mM NAD ϩ , pH 10.6, at 25°C). The enzyme was used at 320 nM and shikimate or quinate (Sigma) at 4 mM. One unit is defined as a rate of 1 mol/min.
Crystallization-Crystals were grown using hanging drop vapor diffusion at 21°C over a reservoir containing 25% PEG 6000, 200 mM ammonium acetate, and 100 mM trisodium citrate dihydrate, pH 5.2. Each drop contained 2 l of protein, 1.6 l of reservoir solution, and 0.4 l of 3 M non-detergent sulfobetaine 195. Hexagonal rods grew to ϳ50 ϫ 50 ϫ 400 m overnight and were frozen in liquid propane using paratone-N as a cryoprotectant. Quinate was added to the protein solution at a 1 mM concentration before setting up the drop that produced the crystal that was solved and refined. However, this ligand was not required for crystal growth, and no evidence of it was observed in the electron density maps.
X-ray Data Collection and Structure Determination-Data were collected from a single crystal at 100 K on beamline X12C of the National Synchrotron Light Source at Brookhaven National Laboratory (Table I). Multiwavelength selenium anomalous diffraction data sets were collected on the Brandeis B4 detector in consecutive 370°sweeps at 0.9790 (peak), 0.9787 (edge), and 0.9200 Å (remote) using 1°oscillations and 5-10 s exposures. The data were processed and reduced with DENZO and SCALEPACK (31) ( Table I). The Laue symmetry of the diffraction pattern and systematic extinctions were consistent with the hexagonal space groups P6 2 or P6 4 , indicating a 56% solvent content assuming a dimer of YdiB in the asymmetric unit. SOLVE (32) identified 18 of the 22 selenium sites, yielding a map that was used for non-crystallographic symmetry averaging, solvent flattening, and automated model building in RESOLVE (33).
Model Building and Refinement-Although only 64% of the backbone and 59% of the side chains were identified during autotracing, the RESOLVE map enabled 81% of the model to be built by hand using O Crystal Structure of E. coli Shikimate 5-Dehydrogenase (34). Completion of the structure required iterative cycles of refinement in Crystallography & NMR System (CNS) (35) and manual rebuilding. Eventually, all of the residues in the protomer were included in the model except for the C-terminal affinity tag (and one adjacent residue from the native protein in one of the two subunits). An R free set containing 5% of the reflections was selected at random. Strong noncrystallographic symmetry restraints (250 kcal/Å 2 and B ϭ 1.5) were maintained throughout the model at all stages of refinement, which consisted of iterations of overall anisotropic B-factor refinement, bulksolvent correction, rigid-body refinement, positional minimization, individual isotropic B-factor refinement, torsional angle dynamics using slow cooling annealing, and automatic water addition. The final model is described in Table I.

RESULTS AND DISCUSSION
Enzymatic Activity-YdiB catalyzed the NAD ϩ -dependent reduction of both shikimate (right to left in Fig. 1) and quinate under established conditions (30). The V max of 150 -180 units/mg (data not shown) corresponds to a turnover rate of 80 -90 s Ϫ1 , which is ϳ15% of AroE.
Structure of the YdiB Protomer-The YdiB protomer is composed of two ␣/␤ domains plus a pair of C-terminal ␣-helices that bridge these two domains (Fig. 2). A deep cavity is found at the interface between these domains. This cavity is likely to form the active site of the enzyme on the basis of the fact that it is contiguous with the hydride acceptor site on the NAD ϩ cofactor, which was found bound to the C-terminal domain in the structure. The apparently flat geometry of the nicotinamide ring suggests that the cofactor is bound in the oxidized NAD ϩ state (36), although this conclusion is tentative given the resolution of our crystal structure. The cofactor was not added to the crystallization reaction, indicating that it co-purified with the enzyme through two columns and must therefore dissociate very slowly in the absence of substrate.
The N-terminal ␣/␤ domain of YdiB comprises residues 1-105. The mostly parallel six-stranded ␤-sheet at the core of this domain is flanked on each side by two ␣-helices, one of them coming from the extreme C terminus of the polypeptide (Fig. 2A). The central ␤-sheet in this domain has its strands in the order of 2-1-3-5-6-4, with only strand 5 antiparallel to the others (Fig. 2B). A systematic analysis using the DALI program identifies a number of protein domains with significant structural similarity, the closest of which is the N-terminal domain of E. coli IIB cellobiose-specific phosphotransferase (Protein Data Bank number 1iib) (37), which gives a Z-score of 5.6 for alignment of 72 residues with a root mean square deviation of 2.5 Å. However, neither this domain nor any of the other structural homologues has the same topology as the N-terminal domain in YdiB, suggesting that it is unique among proteins of known structure.
The C-terminal ␣/␤ domain of YdiB comprises residues 106 -254, which form a characteristic dinucleotide-binding Rossmann fold (38). The entirely parallel six-stranded ␤-sheet at the core of this domain is flanked by three ␣-helices on one side and two on the other (Fig. 2B). The central ␤-sheet in this domain has its strands in the order of 6-5-4-1-2-3 and only deviates from the canonical Rossmann fold by having one of its  (50)  flanking ␣-helices replaced by a long well ordered loop that contains a small 3 10 -helix (between strands ␤10 and ␤11 in Fig.  2B). The NAD ϩ cofactor binds to this domain with its nicotinamide ring in the pro-R conformation ( Fig. 2A), indicating that this enzyme is a class A NAD(P)-dependent oxidoreductase like the well studied malate, lactate, and alcohol dehydrogenases.
Structure of the YdiB Dimer-Gel filtration and static lightscattering data indicate that YdiB forms a homodimer in solution (not shown), likely corresponding to the dimer found in the asymmetric unit in the crystal structure. This species buries 2718 Å 2 of solvent-accessible surface area in its interface, which is mediated primarily by residues in strand ␤1 and helices ␣1 and ␣2 in the N-terminal ␣/␤ domain. Superimposition of the individual domains in the two independent protomers yields root mean square deviations of 0.8 Å for the N-terminal ␣/␤ domain, 0.2 Å for the C-terminal ␣/␤ domain (i.e. the Rossmann fold), and 0.4 Å for the C-terminal ␣-helices. Superimposition of just the Rossmann folds (Fig. 3A) shows that a rigid body rotation occurs between the ␣/␤ domains when the two protomers in the asymmetric unit are compared. Residue Gly-104 near the C terminus of the N-terminal ␣/␤ domain appears to act as a hinge for this rotation, which modulates the width of the active site cavity (Fig. 3A).
Structural Similarity to the Methylene Tetrahydromethanopterin Dehydrogenase, MtdA-Among proteins of known structure, the MtdA from Methylobacterium extorquens (39) shows the highest sequence homology and structural similarity to the C-terminal Rossmann domain in YdiB (Z-score of 13.7 for alignment of 135/148 residues with a root mean square deviation of 3.0 Å and 27% sequence identity). The structure similarity between the YdiB and MtdA structures extends beyond the boundary of this domain (Fig. 3B). In both enzymes, the active site is at the interface of the two ␣/␤ domains found in the protomer. Moreover, there is significant structural homology (40) in the N-terminal ␣/␤ domains (Z-score of 5.2 for alignment of 71/106 residues with a root mean square deviation of 2.5 Å and 7% sequence identity), despite the failure of iterative position-specific iterated basic alignment search tool analysis (41) to find sequence homology in this region. Although these domains share common topology in their three N-terminal ␤-strands, their overall topology is different (Figs. 2C and 3B), as observed with the other structural homologues of this domain of YdiB. Furthermore, the putative active site residues differ in MtdA and YdiB (Fig. 2C), indicating likely divergence in enzymatic mechanism.
Although the structurally conserved N-terminal region of the N-terminal ␣/␤ domain mediates formation of the oligomer interface in both YdiB and MtdA, the structural interactions at the interfaces are very different. Its position has migrated in MtdA relative to YdiB to enable the subunit-subunit interface in this trimeric enzyme to provide an extended recognition site for the much larger substrate that it oxidizes (39) (Fig. 3B). These proteins therefore demonstrate an interesting paradigm whereby structural evolution of an oligomeric interface is used to modify substrate specificity in the NAD(P)H-dependent oxidoreductase superfamily.
Binding Environment of the NAD ϩ Cofactor-The average B-factor of the bound NAD ϩ molecule is 35.6 Å 2 compared with 34.4 Å 2 for the overall polypeptide chain. This result suggests that there is full occupancy of the active site by the co-purified NAD ϩ and reinforces the conclusion that it is bound to the enzyme very tightly. The extended NAD ϩ molecule makes multiple interactions with the protein (Fig. 4, A and B). The glycine-rich loop, which anchors the pyrophosphate of the NAD(H) in the active site, has a variant sequence of GXGG in YdiB (Fig. 4, A and B), compared with the GXGXXG "fingerprint" motif found in most NAD(P)-dependent oxidoreductases (43). The backbone of this segment forms an N-terminal helixcapping structure in YdiB, which shares a conserved conformation with the canonical motif found in the other superfamily  (49) showing the environment of the crystallographically observed NAD ϩ molecule bound to YdiB. Residues within a 5-Å radius of the NAD ϩ atoms are shown. B, schematic plot of the molecular interactions of the bound NAD ϩ molecule. Distances are given in Å, and the numbers in parentheses represent the percentage of identity in all the sequences in the shikimate dehydrogenase cluster of orthologous genes (50). C, stereo pair (49) showing the proposed binding mode for a molecule of shikimate in the active site of YdiB. Distances are given in Å. All of the side chains shown here are 100% conserved in the shikimate dehydrogenase cluster of orthologous genes with the exception of Ser-67 (which is only substituted with threonine) and Tyr-234 (which is present in 37 of 43 sequences). The protein backbone is colored according to domain as in Fig. 2A. syl ribose formed by residues Lys-205 and Phe-160, respectively, which buries the NAD ϩ more than in other superfamily members. This clamp may contribute to the very slow release rate of the cofactor from YdiB.
The amide group of the nicotinamide makes extensive contacts with the N-terminal helix-capping motif on ␣9, including an H-bond to the carbonyl of Gly-255, that hold the ring in the pro-R conformation (Fig. 4, A and B). However, the six atoms of the nicotinamide ring make a total of only three or four van der Waals contacts with the protein. One of these is between the hydride-transferring atom C 4 N and the carboxylate group of the invariant residue Asp-107. Interestingly, the ring is positioned directly over two side chains containing sulfur atoms, Cys-232 and Met-258. The low pK a of the cysteine residue raises the possibility that it could be deprotonated under some circumstances in the presence of NAD ϩ to make an ionic interaction with the bound cofactor.
Shikimate Binding Site and Hypothesized Catalytic Mechanism-Analyses of the active site of YdiB combined with manual docking exercises suggest that the binding site for the shikimate substrate is located in the cleft between the two ␣/␤ domains close to the solvent-exposed Re side of the nicotinamide ring (Figs. 3B and 4C). This cavity is lined by a set of phylogenetically conserved residues (e.g. Lys-71, Asn-92, Thr-106, and Asp-107) with chemical characteristics similar to those that mediate the binding of shikimate derivatives in several other enzymes (i.e. 2,3-dehydroquinic acid to type II dehydroquinase (45) and shikimate-3-phosphate to 5-enolpyruvylshikimate-3-phosphate synthase (46)). Shikimate could be modeled into this site with minimal ambiguity on the basis of the previously established stereochemistry of ternary complexes in other NAD(P)-dependent oxidoreductases (47). Specifically, the carbon containing the hydride donor and the two flanking carbons on shikimate were aligned with the equivalent atoms in the substrate in a crystallographically observed complex, with the ring oriented so that the reactive hydride points toward the C4N acceptor on the nicotinamide ring. This exercise yields a model for the ternary complex (Fig. 4C) in which all but one of the oxygen atoms on the shikimate interact with side chain heteroatoms that are invariant in all shikimate dehydrogenase orthologues (Fig. 2C). Specifically, the C 1 carboxylate accepts a bifurcated H-bond from Ser-67, the C 4 hydroxyl donates an H-bond to Gln-262, and the C 5 hydroxyl donates a bifurcated H-bond to Asp-107. The remaining oxygen atom on shikimate (the C 3 hydroxyl) H-bonds with Tyr-234, which is conserved in AroE and 90% of the orthologous sequences (Fig. 2C). Furthermore, invariant residue Lys-71 is close to the carboxylates of both the shikimate and Asp-107. Although these groups are not within H-bonding distance when shikimate is modeled into our crystal structure of the binary complex, a rotamer change and/or the demonstrated rotational flexibility of the N-terminal ␣/␤ domain (Fig. 3A) could allow closer interactions to occur in the activated ternary complex.
This model for the ternary complex with shikimate suggests a hypothesis for the catalytic reaction mechanism of YdiB. The presence of an H-bond between Asp-107 and the C 5 hydroxyl of the shikimate suggests that the carboxylate side chain will act as the acceptor for the proton from this hydroxyl when the C5 hydride is transferred to NAD ϩ (right to left in Fig. 1). This mechanism is highly attractive, because the pK a of the aspartate makes it an excellent proton acceptor. When the reaction is run in reverse (left to right in Fig. 1), the principle of microscopic reversibility would then require the aspartate to be protonated before binding 3-dehydroshikimate, which could be promoted by the negative charge on the adjacent NADH. Comparison with all known three-dimensional structures using DALI (40) and motif searches using SPASM (48) shows that the side chain disposition in the active site YdiB is unique, suggesting that it uses a novel catalytic reaction mechanism relative to other enzymes in the NAD(P)-dependent oxidoreductase superfamily. Moreover, the key residues in YdiB that interact with the substrate and cofactor are exceedingly well conserved in all shikimate 5-dehydrogenase orthologues (12,16,30) including the AroE protein (12,30), implying that they all use a similar catalytic mechanism and that the design of broad spectrum inhibitors of this enzyme may be possible.