Structures of Shikimate Dehydrogenase AroE and Its Paralog YdiB

Shikimate dehydrogenase catalyzes the fourth step of the shikimate pathway, the essential route for the biosynthesis of aromatic compounds in plants and microorganisms. Absent in metazoans, this pathway is an attractive target for nontoxic herbicides and drugs. Escherichia coli expresses two shikimate dehydrogenase paralogs, the NADP-specific AroE and a putative enzyme YdiB. Here we characterize YdiB as a dual specificity quinate/shikimate dehydrogenase that utilizes either NAD or NADP as a cofactor. Structures of AroE and YdiB with bound cofactors were determined at 1.5 and 2.5 Å resolution, respectively. Both enzymes display a similar architecture with two / domains separated by a wide cleft. Comparison of their dinucleotide-binding domains reveals the molecular basis for cofactor specificity. Independent molecules display conformational flexibility suggesting that a switch between open and closed conformations occurs upon substrate binding. Sequence analysis and structural comparison led us to propose the catalytic machinery and a model for 3-dehydroshikimate recognition. Furthermore, we discuss the evolutionary and metabolic implications of the presence of two shikimate dehydrogenases in E. coli and other organisms.

The shikimate pathway, which links metabolism of carbohydrates to biosynthesis of aromatic compounds, is essential to plants, bacteria, and fungi (1) as well as apicomplexan parasites (2). This seven-step metabolic route leads from phosphoenolpyruvate and erythrose 4-phosphate to chorismate, the common precursor for the synthesis of folic acid, ubiquinone, vitamins E and K, and aromatic amino acids (1). This pathway is absent in metazoans, which must obtain the essential amino acids phenylalanine and tryptophan from their diet. Therefore, enzymes of this pathway are important targets for the devel-opment of nontoxic herbicides (3), as well as antimicrobial (4) and antiparasite (2) agents. The sixth step in the pathway, catalyzed by 5-enolpyruvylshikimate-3-phosphate synthase, has already been successfully targeted, with the development of glyphosate, a broad spectrum herbicide (5). However, after 20 years of extensive use, glyphosate-resistant weeds have recently emerged (6), emphasizing the importance of maintaining target diversity. In order to design new inhibitors, crystal structures of several enzymes of the shikimate pathway have been elucidated recently: 3-dehydroquinate synthase (7), type I and II dehydroquinases (8), type I and II shikimate kinases (9,10), and 5-enolpyruvylshikimate-3-phosphate synthase (11), catalyzing the second, third, fifth, and sixth steps of the pathway, respectively.
Shikimate dehydrogenase (EC 1.1.1.25) catalyzes the fourth reaction in the shikimate pathway, the NADP-dependent reduction of 3-dehydroshikimate to shikimate (Fig. 1A). Whereas dehydrogenases usually form oligomers, shikimate dehydrogenase, coded by the gene aroE in Escherichia coli, is present as a monomer in most bacteria (12,13). In higher organisms this activity is part of a multifunctional enzyme. In plants shikimate dehydrogenase is associated with type I dehydroquinase to form a bifunctional enzyme (14), whereas in fungi, such as Neurospora crassa, this enzyme forms the fifth domain of the pentafunctional AROM polypeptide, which catalyzes five of seven steps of the shikimate pathway (15). However, the molecular basis of 3-dehydroshikimate recognition and enzymatic reduction is not known.
Although in E. coli AroE is strictly specific for shikimate, some fungal shikimate dehydrogenases can also utilize quinic acid as a substrate. This compound, which differs from shikimic acid only by the addition of a hydroxyl group at C-1 (Fig.  1B), is the precursor to the ubiquitous plant secondary product chlorogenate (1). To date, two independent families of quinate/ shikimate dehydrogenases have been identified. The first consists of NAD-dependent dehydrogenases (16), and the second consists of membrane-associated dehydrogenases that utilize pyrrolo-quinoline-quinone as a cofactor (17). Both types of dehydrogenases are involved in the catabolic quinate pathway, which allows growth of microorganisms with quinate as the sole carbon source by its conversion into protocatechuate and subsequent metabolism by the ␤-ketoadipate pathway (16,17).
By using BLAST (18), ϳ130 sequences, mostly annotated as putative shikimate dehydrogenases, can be identified as homologous to AroE through the entire length of the gene, thereby defining the shikimate dehydrogenase (SDH) 1 family. It also includes the NAD-dependent quinate/shikimate dehydroge-nases, whereas the pyrrolo-quinoline-quinone-dependent enzymes compose a different protein family. This family displays no significant sequence similarity with any other NAD(P)-dependent dehydrogenases, therefore constituting a distinct dehydrogenase family. Analysis of the complete genome of E. coli K12 and pathogenic O157:H7 strains has revealed the presence of a gene of unknown function, ydiB, which shares 25% sequence identity with aroE. Thus, AroE and YdiB are paralogs, the only two proteins from the SDH family present in E. coli.
Here we report the biochemical characterization of YdiB and demonstrate that it is a quinate/shikimate dehydrogenase that can utilize either NAD or NADP as a cofactor. We have determined crystal structures of both these enzymes, AroE at 1.5 Å and YdiB at 2.5 Å resolution. These structures are the first shikimate dehydrogenase structures to be determined. Comparison of their substrate-binding sites led us to propose the catalytically important amino acid residues and to identify at the molecular level the structural differences leading to variation in cofactor specificity. Furthermore, we discuss the evolutionary and metabolic implications of the presence of two shikimate dehydrogenases in E. coli and other organisms.
Structure Solution and Refinement-Native data were collected from cryo-cooled AroE crystals to 1.5 Å resolution at station 9.6 at the Daresbury SRS using a (ADSC) Quantum-4 charge-coupled device detector. Data were collected in-house, by using a MacScience DIP2000 detector on various crystals soaked with heavy atom solutions. Crystals soaked with Hg(CN) 2 produced the only usable derivative that was isomorphous to the native crystals. This derivative was collected at a wavelength to maximize the anomalous signal on station 9.5 at the Daresbury SRS using a Mar charge-coupled device detector in a SIRAS experiment. All data were indexed and processed with the HKL suite (21), the cell dimensions and space group shown in Table I. Further processing was carried out using programs from the CCP4 package (22). From the anomalous Patterson map it was possible to identify 13 mercury sites using SHELX-90 (23), which were refined in Mlphare against the native 1.5-Å data to maximize the isomorphous signal. Phase refinement and extension was performed using the program DM with solvent flattening and histogram matching. Averaging was attempted but was unsuccessful because of the large variation in conformation in the independent molecules in the asymmetric unit. Refinement was carried out using the maximum likelihood refinement program REFMAC (24). Five percent of the data were randomly set aside as test data for calculation of R free . The structure was built automatically using the program ARP/WARP (25) and assembled into the four independent chains that were Ͼ90% complete. Manual correction of the structure and model building and addition of solvent was performed using modules within the program QUANTA (Accelrys Inc.). Nine iterations of refinement and manual rebuilding with the addition of molecules of NADP ϩ , sulfate, glycerol, and DTT with the application of individual anisotropic temperature factors in the final stages of refinement resulted in a model with the final R work of 14.7% and R free of 17.6% and good stereochemistry as assessed using the program PRO-CHECK (26). The structure was deposited with the Protein Data Bank with the code 1NYT.
YdiB crystals were soaked for ϳ30 s in a cryoprotectant solution (0.8 M KH 2 PO 4 , 0.8 M NaH 2 PO 4 , 0.1 M Hepes, pH 7.5, 2 mM NADH, 22% (v/v) glycerol), picked up in a nylon loop, transferred to the goniometer head, and kept at 100 K in a nitrogen stream. Diffraction data were collected on a Quantum-4 charge-coupled device detector (ADSC, San Diego, CA) at beamline X8C, the National Synchroton Light Source (NSLS) at Brookhaven National Laboratory in New York. Data indexing, merging, and scaling were performed using the HKL2000 package (21). Data collection and processing statistics are listed in Table I. Multiple anomalous dispersion data were collected on a Se-Met-substituted YdiB crystal to 2.5 Å resolution at inflection, peak, and hard remote wavelengths around the K absorption edge of selenium (Table I). Of the 22 expected selenium sites, 20 were found using the heavy atom search procedure of CNS (27). The phases calculated with this partial structure resulted in a figure of merit of 0.67-2.5 Å resolution. By taking advantage of the noncrystallographic symmetry (NCS), the electron density was improved by molecular averaging and solvent flipping (40% solvent) with CNS (27), yielding a final figure of merit of 0.93. The model was built manually with the program O (28) into the solvent-flipped multiple anomalous dispersion electron density map. Refinement was performed with CNS (27) with the maximum likelihood target function. The NCS restraints were applied only in the initial cycles of refinement. The experimental as well as the simulated annealing omit maps clearly showed the presence of one NAD ϩ molecule bound to each YdiB molecule. The final model for the asymmetric unit refined at 2.5 Å has an R work of 22.6% and an R free of 29.4% and consists of 4274 protein atoms, two NAD ϩ cofactors, five phosphate ions, and 156 water molecules. The relatively high R free value is likely explained by the presence of two overlapping conformations of helix ␣7 in molecule A, which render the electron density map difficult to model. In the vicinity of each molecule, several disordered electron density features were also left unassigned, because they do not respect the hydrogen bonding criteria of water molecules. The final structure was a good stereochemistry as evaluated using the PROCHECK program (26). The structure is deposited with the Protein Data Bank with the code 1O9B.
Biochemical Characterization-To evaluate the oligomeric state of YdiB, dynamic light-scattering (DLS) measurements were done on a  solution of YdiB concentrated at 1 and 11 mg/ml, in the presence or in the absence of 2 mM NADH. The measurements were performed using a DynaPro 801 instrument (Protein Solutions, Charlottesville, VA). To confirm these DLS results, gel filtration analysis was also performed using a Superdex 75 column (Amersham Biosciences) and calibrated with the reference protein mixture recommended by Amersham Biosciences. The YdiB sample (200 l, 11 mg/ml) was injected and eluted at 1 ml/min in the same buffer (20 mM Tris-HCl, pH 7.5, 200 M NaCl, 5% (w/v) glycerol, 5 mM DTT). The enzymatic activities of AroE and YdiB were assayed at 20°C by monitoring the reduction of NAD ϩ or NADP ϩ at 340 nm (⑀ ϭ 6.18 ϫ 10 Ϫ3 M Ϫ1 cm Ϫ1 ) in the presence of either shikimic acid or quinic acid. To test possible inhibition by NAD ϩ , the enzymatic activity of AroE was assayed in the following buffers: 100 mM Tris-HCl, pH 9.0, 5 mM shikimic, 200 M NADP ϩ , and 20 mM NAD ϩ . To measure the kinetic parameters for each cofactor, the assay mixture (total volume 200 l) consisted of 100 mM Tris-HCl, pH 9.0, 5 mM shikimic or quinic acid, and six different values for the cofactor NAD ϩ or NADP ϩ (4, 2, and 1 mM and 500, 250, and 125 M). Similarly to measure the kinetic parameters for both substrates, the assay mixture consisted of 100 mM Tris-HCl, pH 9.0, 5 mM NAD ϩ or NADP ϩ , and six different values for shikimic or quinic acid (4, 2, and 1 mM and 500, 250, and 125 M). To measure the activity, 10 l of enzyme ([AroE] stock ϭ 0.17 nM, [YdiB] stock ϭ 800 nM) was added to the assay mixture. These enzyme concentrations were chosen in order to follow the initial reaction rate. The absorbance at 340 nm was measured for 30 min against a blank consisting of the assay mixture without enzyme. Each measure was taken in triplicate and simultaneously using a 96-well quartz plate. The kinetic parameters were deduced by the Lineweaver-Burk method. These reactions were monitored using the Plate Reader Spectra Max (Molecular Devices, Sunnyvale, CA). All chemicals were purchased from Sigma.

RESULTS AND DISCUSSION
Characterization of YdiB as a Dual Specificity Quinate/ Shikimate Dehydrogenase-E. coli YdiB shows sequence similarity with AroE from the same organism (13) and with the quinate/shikimate dehydrogenases Qa-3 from N. crassa (29) and QutB from Emericella nidulans (30). We used this knowledge as a starting point to investigate its substrate specificity by monitoring the reduction of NAD ϩ or NADP ϩ in the presence of either shikimic acid or quinic acid. As a control, purified recombinant AroE protein was tested under the same conditions. As expected, AroE oxidized shikimic acid using NADP ϩ as cofactor but displayed no activity in the presence of NAD ϩ . The kinetic parameters are very similar for both the cofactor and the substrate (NADP K m ϭ 56 M, k cat ϭ 14,200 min Ϫ1 , shikimate K m ϭ 65 M, and k cat ϭ 14,200 min Ϫ1 ). Quinic acid, even at a high concentration of 5 mM, is not a substrate for AroE, either in the presence of NADP ϩ or NAD ϩ . To determine whether NAD ϩ acts as a competitive inhibitor with respect to NADP ϩ , the oxidation of shikimic acid was assayed with a NADP ϩ /NAD ϩ ratio of 1:100 (200 M NADP ϩ and 20 mM NAD ϩ ). Despite the large excess of NAD ϩ , the activity of AroE was not significantly altered, indicating that NAD ϩ is not a competitive inhibitor of NADP ϩ and that AroE likely does not bind NAD ϩ .
In contrast, YdiB is able to oxidize shikimic acid by using either NADP ϩ or NAD ϩ as cofactor. At saturation of shikimate, YdiB displays similar kinetic parameters for both cofactors (NADP ϩ , K m ϭ 100 M, k cat ϭ 7 min Ϫ1 ; NAD ϩ , K m ϭ 87 M, k cat ϭ 3 min Ϫ1 ). The K m values significantly differ for the shikimic acid, according to the type of cofactor used at saturation: shikimate ϩ NADP ϩ , K m ϭ 120 M, k cat ϭ 7 min Ϫ1 ; shikimate ϩ NAD ϩ K m ϭ 20 M, k cat ϭ 3 min Ϫ1 . Contrary to AroE, YdiB also displays a clear activity on quinic acid, with either NADP ϩ or NAD ϩ as a cofactor. At saturation of quinate, YdiB displays a five times lower K m for NAD ϩ (K m ϭ 116 M, k cat ϭ 3 min Ϫ1 ) than for NADP ϩ (K m ϭ 500 M, k cat ϭ 3 min Ϫ1 ). This phenomenon is accentuated for the K m of quinic acid, which is 10 times YdiB is therefore the first quinate/shikimate dehydrogenase identified in E. coli. Although this enzyme has a lower catalytic efficiency (ϳ2-4000-fold) compared with that of AroE, this is compensated by a broader substrate and cofactor specificity. The low specific activity of YdiB likely explains why it was not identified alongside AroE during the initial purification of this activity from E. coli (12). Although it is clear that YdiB is NADP/NAD-dependent dehydrogenase, we cannot exclude the possibility that its physiological substrate is neither shikimate nor quinate, considering its low catalytic efficiency.
Nevertheless, AroE and YdiB display a fairly equivalent affinity for their ligands, as shown by the similar range of their K m values. Furthermore, YdiB seems equally active on shikimic and quinic acid, because their K m values are comparable in the presence of NAD ϩ (20 and 40 M, respectively). In contrast, the behavior of YdiB is different according to which cofactor is used. YdiB has a tendency to be more "efficient" in the presence of NAD ϩ , as shown by the discrepancy between the K m values for shikimate/quinate at the saturation of either NAD ϩ (20/40 M) or NADP ϩ (120/555 M). This difference could be explained by a lower affinity for NADP ϩ , as shown by the cofactor K m in the presence of quinate or by a binding of NADP ϩ in a less productive manner (shikimate case).
Overall Structure of E. coli AroE and YdiB-The asymmetric unit of AroE crystals contains four protein molecules (Met 1 -Ser 271 ) complexed with NADP ϩ , and a total of 13 sulfate ions, 1277 water molecules, and 1 molecule of DTT bound in the active site of molecule A. The four protein molecules are related by pseudo 222 symmetry as reported previously (19). The YdiB asymmetric unit comprises two molecules (Tyr 7 -Phe 286 ), related by 2-fold noncrystallographic symmetry, each complexed with NAD ϩ , 2 phosphate ions, and 156 water molecules. The residues Met 1 -Lys 6 and Gly 287 -Ala 288 are disordered and are not included in the model. Unless specified otherwise, we will refer to residues according to AroE numbering, with those of YdiB referenced in parentheses.
Despite the relatively low sequence similarity between AroE and YdiB, the two enzymes have highly similar structures that adopt the same fold (Figs. 2 and 3). The molecules have a somewhat elongated shape (55 ϫ 40 ϫ 30 Å) and comprise two domains. The first domain is made of two discontinuous segments, Met 1 -Thr 101 (Met 7 -Thr 106 ) and Gly 237 -Ser 271 (Gly 255 -Phe 286 ), whereas the second domain encompasses Gly 119 -Asp 236 (Gly 124 -Asp 254 ). Both domains have ␣/␤ architectures and are connected by the helix ␣5 and a short linker, Asp 102 -Pro 118 (Asp 107 -Lys 123 ). The arrangement of these two domains along the connecting helices creates a deep groove in which the cofactor NADP ϩ (or NAD ϩ ) is located (Fig. 2).
The N-terminal domain consists of a mainly parallel sixstranded ␤-sheet and six ␣-helices. The strand order is 2-1-3-5-6-4, with the strand ␤5 being antiparallel to the other strands. The first three ␤-strands follow a regular ␤/␣ succession, with the helices ␣1 and ␣2 parallel to the ␤-strands, flanking opposite sides of the sheet. The next ␣/␤/␣ unit is irregular, with the helix ␣3 oriented at ϳ45°relative to the direction of the sheet, and the short, one turn helix ␣4 nearly perpendicular to the strand ␤4. The domain is completed by a C-terminal ␣-helical hairpin (␣9 and ␣10), which packs against the ␤-sheet on the same side as ␣1. According to the DALI algorithm (31), this domain shows topological and structural similarity with the C-terminal domain of glycyl-tRNA synthetase (Protein Data Bank code 1ATI), which has strand order 2-1-3-4-5 (␤4 antiparallel to the other strands). Out of 129 residues, 80 C␣ atoms can be superimposed on AroE with r.m.s.d. of 2.5 Å. In AroE/YdiB the extended loop between strands ␤3 and ␤5 contains two helices and folds back onto the ␤-sheet adding a sixth strand (␤4) at the end. The corresponding loop in glycyl-tRNA synthetase is several residues shorter and extends away from the ␤-sheet. The fold of AroE/YdiB is also similar to the N-terminal part of the molybdenum cofactor biosynthesis protein MogA (Protein Data Bank code 1DI6, r.m.s.d. of 3.1 Å over 102 residues). Although the two folds differ in the strand order (2-1-3-6-5-4 in MogA with ␤5 antiparallel to the other strands), corresponding to a switch in the relative positions of strands ␤5 and ␤6, there is additionally a good spatial overlap of several helices.
The C-terminal domain or NAD(P)-binding domain could not be recognized from its amino acid sequence; however, this domain adopts a nearly canonical Rossmann fold, i.e. a sixstranded parallel ␤-sheet, with the strand order 3-2-1-4-5-6, and ␣-helices on both sides parallel to the ␤-strands. The fourth ␣-helix present in the canonical Rossmann fold is missing in YdiB, whereas the third and fourth ␣-helices are replaced by irregular loops in AroE. As a result, the AroE/YdiB NAD(P)-binding domains are among some of the shortest reported, sharing most structural homology with S-adenosylhomocysteine hydrolase (Protein Data Bank code 1D4G, r.m.s.d. 1.64 Å over 153 C␣ atoms) and mouse class II alcohol dehydrogenase (Protein Data Bank code 1E3L, r.m.s.d. 1.87 Å over 160 C␣ atoms). The SDH family provides a new example of a protein family displaying the dinucleotide binding fold, without significant sequence homology with other Rossmann fold families; this may indicate early divergence from the ancestral fold.
Quaternary Structures of AroE and YdiB-Whereas AroE has been shown to be a monomeric protein (12,19), dynamic light scattering measurements on YdiB using different protein concentrations (1 and 11 mg/ml), both in the presence and absence of NADH, show that YdiB has a hydrodynamic radius consistent with a particle of ϳ60 kDa, indicating that this protein forms dimers. This was verified by size exclusion chromatography where the apoprotein eluted as a single species of 64 kDa. Analysis of the different protein-protein interfaces within the crystal structure of YdiB shows that the largest contact surface area is between the two molecules in the asymmetric unit. The two monomers are related by pseudo 2-fold symmetry, with the dimer interface formed by residues from strands ␤1, ␤2, and the helix ␣2 of the two N-terminal domains. This head-to-head packing of the N-terminal domains creates a highly elongated dimer with diametrically positioned active site clefts. The interface involves contacts made by 16 residues from each molecule and is predominantly hydrophobic in nature. The dimer buries 1400 Å 2 of solvent-accessible surface area (700 Å 2 from each monomer), which is at the lower end of values observed for protein-protein interfaces (32). Such an interface is not without precedent as a much smaller, solely hydrophobic, interface has been observed for the structure of Ocr from bacteriophage T7 (33). If the YdiB dimer interface has been correctly identified, then the hydrophobic residues forming this interface are mostly replaced by polar or smaller amino acids in AroE, notably YdiB (AroE) Leu 9 (Thr 3 ), Met 40 (Gly 34 ), Phe 42 (Val 36 ), Leu 59 (Ala 53 ), and Met 61 (Gly 55 ) (Fig. 3). These amino acid substitutions eliminate the hydrophobic patch on the surface of the YdiB monomer, giving a more hydrophilic character to the N-terminal domain of AroE and explaining why this protein is present as a monomer in solution.
The Cofactor Binding Site of AroE and YdiB-In all AroE and YdiB molecules the electron density for the cofactor, NAD ϩ in YdiB and NADP ϩ for AroE, is very well defined (Fig. 4). In the following description we will refer to molecule B of YdiB and molecule A of AroE as these have the lowest average B-factors. The NAD(P) ϩ cofactor is located outside the carboxyl ends of ␤-strands ␤7-␤10 at a switch point in the central ␤-sheet of the C-terminal domain. The superposition of the C-terminal domains of AroE and YdiB results in good superposition of NAD ϩ and NADP ϩ , especially of their diphosphate groups and nicotinamide rings. A somewhat larger difference, an ϳ2-Å shift, occurs in the relative position of the adenosine.
Similar Recognition of Nicotinamide and Pyrophosphate-The binding of the nicotinamide and pyrophosphate moieties is similar in AroE and YdiB. The amide group N-7 of the nicotinamide ring is hydrogen-bonded to the carbonyl group of two residues, Met 213 (Cys 232 ) and the invariant Gly 237 (Gly 255 ) (Fig.  3). The neighboring ribose forms only van der Waals contacts to the hydrophobic side chains. The pyrophosphate moiety contacts the glycine-rich loop that connects strand ␤7 and helix ␣6 (Fig. 4) and forms hydrogen bonds to the backbone N atoms of Gly 129 and Ala 130 (Gly 134 and Ala 135 ). A sequence pattern G [A,s,g] G G [A,t] [A,S,g] corresponding to the diphosphatebinding loop is conserved in the entire SDH family (Fig. 3). This fingerprint is yet another modification of the canonical pattern identified in NAD-dependent dehydrogenases: G-S2-S3-G-S5-S6-G, where S2 may be absent, S3 and S5 are variable, and S6 is always a hydrophobic residue, whose side chain is directed toward the nicotinamide moiety (34). With the missing residue S2, the main differences in the AroE fingerprint are the strict conservation of a glycine at the usually variable position S5, the presence of a less hydrophobic residue at position S6, and a small residue in place of Gly at the next position.

FIG. 2. A stereo ribbon representation of E. coli AroE (chain A)
shows the overall fold of this enzyme with ␤ strands colored in gold, ␣-helices in green, and 3 10 helices in turquoise. The cofactor NADP ϩ is shown in a ball and stick representation, with nitrogen colored blue, oxygen colored red, carbon colored gray, and phosphorus colored magenta. Figs. 2 and 6 were prepared using the program RIBBONS (47).
Cofactor Specificity Determinants within the Adenosinebinding Pocket-In contrast to the vast majority of the NAD(P)dependent dehydrogenases, which have a strong specificity for either NAD or NADP (34), members of the SDH family show a diversity of cofactor specificity. E. coli AroE, involved in biosynthesis, is strictly NADP-dependent (12), whereas N. crassa Qa-3 and E. nidulans QutB display a strong preference for NAD (29,30), and E. coli YdiB is able to use both cofactors. Therefore, the comparison of the cofactor binding sites in AroE and YdiB is of interest as it reveals the structural features necessary to discriminate between NADP and NAD in the SDH family.
The binding of the adenine moiety by both enzymes is typical for NADP-dependent dehydrogenases as it contains an arginine side chain that stacks against the adenine ring and lacks a carboxylic residue (replaced by Asn) that chelates the diol group of the ribose in NAD complexes (34). In the SDH family the loop between strand ␤8 and helix ␣7 features two strictly conserved residues, Asn 149 and Arg 150 (Asn 155 and Arg 156 , Fig.   3), which are both involved in the recognition of the adenosine moiety. There are, however, differences in their interactions with the cofactor in the two enzymes. In the AroE-NADP ϩ complex, the hydroxyl group O-3Ј of the adenosine ribose is hydrogen-bonded to Asn 149(OD1) , as well as to the main chain NH of Ala 127 , located in the glycine-rich loop (Fig. 4A). In addition, the amide of Asn 149 forms a hydrogen bond to the O-1 atom of the 2Ј-phosphate. Arg 150 forms two hydrogen bonds with the other oxygen atoms of the phosphate substituent, whereas its guanidinium group stacks against the A-face of the adenine ring. This phosphate is further stabilized by electrostatic interactions with Arg 154 from helix ␣7 and by a hydrogen bond with Thr 151(OH) . Face B of adenine contacts the side chain of Thr 188 and Ser 190 (Fig. 4A). The arginines 150 and 154 play a crucial role in adenosine phosphate binding as they form an "electrostatic clamp" that sandwiches the phosphate substituent.
In YdiB there are several substitutions affecting the interactions with NAD ϩ (Fig. 4B). Val 206 , which replaces Ser 190 of AroE, orients its aliphatic side chain perpendicularly to the B-face of the adenine ring, forming a CH--electron hydrogen bond (35). The bulging of Val 206 is accompanied by a compensating shift of Arg 156 , which maintains its stacking against the A-face of a slightly translated adenine. This arrangement provides for hydrogen bonds of O-2Ј and O-3Ј of NAD ϩ ribose to Asn 155(OD1) as well as the O-3Ј to a backbone NH of Ala 132 (Fig.  4B). The NAD ϩ binding in YdiB is favored by the substitution of Thr 151 and Arg 154 of AroE by Asp 158 and Phe 160 , respectively. Asp 158 is hydrogen-bonded to the hydroxyl group O-2Ј of the ribose and also stabilizes Arg 156 through a salt bridge. The hydrophobic residue Phe 160 creates a neutral environment, which is less discriminating than the basic binding pocket observed in the AroE structure (Fig. 5, B and C). The capacity of YdiB to also bind NADP ϩ likely involves a conformational change of Asp 158 to avoid electrostatic repulsion with the phosphate group. A low resolution structure of YdiB co-crystallized with NADP confirmed that this cofactor is located in a position similar to that of NAD ϩ . The loop ␤8-␣7, which contains Asp 158 , is displaced in this structure in order to provide a phosphate-binding site and as a result is poorly ordered.
The Active Site and Its Conformational Flexibility-The substrate-binding site is identified by the position of the nicotinamide ring of the cofactor and is delineated almost entirely by residues from the N-terminal domain. The binding site is in a pocket formed by the C-terminal ends of the ␤-strands, the N-terminal end of helix ␣1, the side of helix ␣9, the extended loop between ␤1 and ␣1, and the first residues from the connecting helix ␣5. Most of the residues absolutely conserved in the SDH family are located in this pocket, i.e. Ser 14 , Ser 16 , Lys 65 , Asn 86 , Thr 101 , Asp 102 , and Gln 244 . At position 61 (67), a serine or a threonine is also always observed in the SDH family (Fig. 3). A sulfate or phosphate ion is present in this cavity in all AroE and YdiB molecules. In molecules A and B of AroE, this anion is located at the top of the pocket and is hydrogenbonded to the hydroxyl groups of Ser 14 , Ser 16 , Thr 61 , and Tyr 215 (Fig. 6A), whereas in the remaining AroE and YdiB molecules it lies at the bottom of the cavity, hydrogen-bonded to the side chains of Lys 65 and Thr 61 (Lys 71 and Ser 67 ). In molecule A of AroE, a DTT molecule is also present in this pocket, tightly bound through numerous hydrogen bonds involving its thiol and hydroxyl groups to the conserved AroE residues: DTT SH1 -Gln 244(OE1) , DTT OH2 -Lys 69(NZ) , Asn 86(ND2) and Asp 102O(D1) , and DTT SH4 -Thr 61O(G) (Fig. 6A).
The comparison of independent molecules of AroE and YdiB shows clear differences in the relative disposition of their domains. Three different conformations are observed for AroE (molecules A/B, C, and D), whereas the two molecules of YdiB display similar conformation. Comparing individual domains of the same protein gives an r.m.s.d. in the range of 0.3-0.6 Å. Superposition of the individual domains of AroE and YdiB results in an r.m.s.d. of ϳ1.3 Å for 104 of 136 C␣ atoms and ϳ1.4 Å for 100 of 137 C␣ atoms, for the substrate-and cofactorbinding domains, respectively. However, these numbers for the entire molecules are significantly larger, 1.2-1.6 Å for the independent AroE molecules and 2.8 -3.3 Å for the comparison of AroE and YdiB molecules (Fig. 5A). Among these conformations, molecule A of AroE represent the most "closed" form, whereas molecule A of YdiB represent the most "open" form of the enzyme (Fig. 5, B and C). The transition between these two extreme conformations corresponds to a rotation of ϳ25°a round an axis passing approximately through the C␣ of Gln 26 (Lys 32 ) and Asp 102 (Asp 107 ). Consequently, the tip of the Nterminal domain traverses a distance of ϳ14 Å between the open and closed structures.
This overall conformational change is concomitant with the rearrangement of the hydrogen bonding network in the junction region between the N-and C-terminal domains. Among the five residues involved in this network, three (Asn 86 , Thr 101 , and Gln 244 (Asn 92 , Thr 106 , and Gln 262 )) are invariant in the SDH family, whereas a fourth residue, Thr 87 (Thr 93 ), is conserved in 92% of sequences. The last residue, Asn 59 , is conserved in ϳ60% of the sequences and is substituted by small residues (Gly, Ala, and Ser) in the remainder of the SDH family. In this latter group, which includes YdiB, we find a compensating replacement of Ala 248 by a glutamine (Gln 266 ), whose carboxyamide group overlaps that of Asn 59 , resulting in a spatial invariance of a polar group at this position. In the open conformation, all these residues are linked by hydrogen bonding interactions between their side chains: Asn 59(NE1) (Gln 266 )-Thr 87O(G) -Asn 86O(D1) , Asn 86(ND2) -Thr 101O(G) -Gln 244(NE2) -Asn 59O(E1) (Gln 266 ). This circular network rearranges in the closed conformation, as Gln 244 is no longer hydrogen-bonded to the side chains of Asn 59 (Gln 266 ) and Thr 101 . Instead, this glutamine side chain makes a hydrogen bond to the side chain of Asn 86 , whereas its main chain carbonyl group is hydrogenbonded to Thr 101(OH) . Because the closed conformation was found in the molecule that binds DTT, we speculate that the conformational change, which closes the central cleft, occurs upon substrate binding and is necessary for the formation of a productive active site. The cluster of conserved residues in the junction region therefore acts as a hinge, stabilizing the open conformation at the beginning of a catalytic cycle and then favoring the closing of the active site cleft when the substrate is present.
The Reaction Mechanism of Shikimate/Quinate Dehydrogenase-The presence in the closed active site of a DTT molecule and a sulfate ion, contacting invariant residues, suggests the possible interactions between shikimate dehydrogenase and its substrate (Fig. 6A). The integration of the sequence and biochemical and structural evidence led us to propose a model for the recognition of 3-dehydroshikimate (Fig. 6B). The enzyme catalyzes the stereospecific reduction of 3-dehydroshikimate to shikimate and, as such, requires precise positioning of the substrate. It was shown that hydride transfer occurs from the A-side of NADPH (36), which is consistent with the orientation of the cofactor in the active sites of the two structures. For catalysis to occur, the C-3 of 3-dehydroshikimate/3-dehydroquinate must be positioned to receive the hydrogen from C-4 of the nicotinamide ring. The location of the C-3 and C-4 of DTT in the vicinity of the C-4 of NADP ϩ is consistent with such positioning (Fig. 6A).
At the same time, we expect that the carboxylate would form specific interactions within the substrate-binding pocket. In the other enzymes in the shikimate pathway, the carboxylate of the substrate is bound by either an arginine (type I dehydroquinase (8), dehydroquinate synthase (7), 5-enolpyruvylshikimate-3-phosphate synthase (11)) or main chain amides (type II dehydroquinase (37)). Given the position of the conserved residues within the active site, the loop between ␤1 and ␣1 delineated by two strongly conserved proline residues adopts a conformation capable of binding a carboxylate in a similar manner to the type II dehydroquinase. The conserved serine residues at positions 14 and 16 most likely contribute to carboxylate bind- ing so that both carboxyl oxygens form two hydrogen bonds to the protein. The conserved tyrosine 215, located in the Cterminal domain but whose side chain points toward the substrate pocket, is also likely to establish an additional hydrogen bond with the carboxylate. The location of the sulfate ion in this region of the structure and its interactions with these three conserved residues support this hypothesis (Fig. 6A).
The substrate in this orientation will form hydrogen bonds between C-4 hydroxyl and the side chain of Lys 65 and Asp 102 , whereas the C-5 hydroxyl would be positioned down into the active site forming hydrogen bonds with Gln 244 . Such hydrogen bonds are observed between AroE and the groups OH2 and SH1 of DTT (Fig. 6A). Previous studies of Pisum sativum shikimate dehydrogenase have shown that substrate-like inhibitors of the enzyme require a C-4 hydroxyl, whereas either a C-5 hydroxyl or carboxylate group is needed for strong binding (38). By using a series of analogs of 3-dehydroshikimate that lack the C-4 and C-5 hydroxyls, Bugg and co-workers (39) demonstrated that the C-5 hydroxyl of the substrate has little effect on the specificity of E. coli AroE, whereas the C-4 hydroxyl is very significant. An estimation of the binding energy (based on k cat /K m (40)) between the C-4 hydroxyl and the enzyme suggests that this hydroxyl forms a hydrogen bond to a charged group (39). From the pH/rate profile of AroE, it has been suggested that this charged group is either a cysteine or an ␣-amino group (41). In this light, Lys 65 (Lys 71 ) seems a good candidate for the residue coordinating the C-4 hydroxyl group.
By analogy with lactate dehydrogenase (42), an acid/base catalytic group is needed to donate a proton to the carbonyl of 3-dehydroshikimate during reduction and to remove a proton during oxidation of shikimate. The invariant Lys 65 and Asp 102 are the most likely candidates to assume this role, considering their proximity to both the nicotinamide ring and the SH4 and OH3 groups of DTT (Fig. 6A). Another possibility is the involvement of Thr 61 , the 2Ј-hydroxyl of the cofactor, and His 13 in a proton relay analogous to that found in alcohol dehydrogenase (43). The pH dependence of AroE (maximum at pH 7.3) is consistent with a histidine being involved in the mechanism. Histidine-specific chemical modification of AroE by diethylpyrocarbonate at pH 7.0 has been shown to inactivate the enzyme in a time-dependent manner, which was monitored by electrospray mass spectrometry (44). Two of the histidine residues in AroE could be protected from diethylpyrocarbonate modification by the presence of shikimate; one of these was identified as His 13 (45). However, His 13 is a significant distance from the 2Ј-hydroxyl of the cofactor (ϳ11 Å), even in the closed conformation of AroE, and is not strictly conserved in the SDH family making it a less likely candidate for the catalytic acid/base. The Evolutionary and Metabolic Implications from the Presence of Two Shikimate Dehydrogenase Genes in the E. coli Genome-The presence of two shikimate dehydrogenase isoforms in E. coli raises intriguing questions concerning their specific biological roles. The existence of a second shikimate dehydrogenase also affects the design of any potential drugs, because YdiB may compensate for the inhibition of AroE. Although the substrate specificity of YdiB has been identified FIG. 6. A, stereo view of AroE active site with NADP ϩ , bound sulfate, and DTT molecules represented in thick stick (carbon atoms colored green) and protein atoms represented in thinner stick (carbon atoms colored gray). All atoms are colored according to atom type, and conserved amino acid residues are labeled. B, a molecular model of the binding of dehydroshikimate to the active site of AroE. This model does not represent the proposed ternary complex as further closing of the active site is envisaged and cannot be modeled with confidence. The model serves instead as a guide to the chemically correct orientation of the substrate which is necessary for catalysis to occur. here, it is not yet clear if YdiB participates in the shikimate pathway or has another biological function. A systematic analysis of the bacterial genomes presently listed in the TIGR data base (www.tigr.org/) revealed that 14 species possess at least two shikimate dehydrogenase isozymes located at distinct loci. These microorganisms belong to distant phyla (e.g. ␣and ␥-Proteobacteria, Deinococcus-Thermus, Actinobacteria, and Firmicute), showing that this phenomenon is not limited only to E. coli or related species. Most of these homologous proteins display a similar, relatively low sequence identity to AroE and YdiB (20 -30%), making a one-to-one assignment difficult. However, a few proteins are clearly either AroE-like (Haemophilus influenzae HI0655, Pasteurella multicida AroE, Salmonella enterica STY4396, Salmonella typhimurium STM3401, and Yersinia pestis YP00246) or YdiB-like (Listeria innocua Lin2338 and Lin0493, Listeria monocytogenes Lmo2236 and Lmo0490, S. typhimurium STM1359, and Streptococcus pyogenes SpyM18 -1592) with sequence identity varying between 45 and 90%.
The location of the aroE and ydiB genes in the E. coli genome is also informative. AroE is flanked by genes of unknown or putative functions (Yrd B, C, and D), unrelated to the shikimate pathway. In contrast, ydiB is located between the gene b1691, coding a putative amino acid transport protein, and the gene aroD, coding type I 3-dehydroquinase. According to the Regulon DB data base (46), AroE and YdiB are independently regulated, whereas the cluster b1691-ydiB-aroD is under the control of the same promoter. Such organization of ydiB and aroD in one operon is also observed in pathogenic bacteria L. innocua, L. monocytogenes (Gram ϩ ), and S. typhimurium (Gram Ϫ ). Moreover, these enzymes recognize the same substrate, 3-dehydroquinate. Therefore, YdiB may have a physiological role connected to that of aroD. More to the point, shikimate dehydrogenase and 3-dehydroquinase activities coassembled into a bifunctional protein in some plants and bacteria. Such a bifunctional enzyme could have evolved by the fusion of an ancestral bacterial ydiB-aroD gene cluster. Type I dehydroquinases like AroD are associated with biosynthesis, whereas type II dehydroquinases are known to function in synthetic and degradative pathways. The association of ydiB-aroD in one operon would therefore suggest its involvement in the shikimate pathway. In contrast, the substrate and cofactor promiscuity of YdiB would speak in favor of a different role. All presently known NAD-dependent quinate/shikimate dehydrogenases are involved in the catabolic quinate pathway. Therefore, YdiB may be essential for growth of E. coli with quinate as a sole carbon source (16), thus indicating the presence of a quinate pathway in this organism.