Crystal Structure of the NAD Biosynthetic Enzyme Quinolinate Synthase* □ S

A gene encoding a quinolinate synthase has been identified in the hyperthermophilic archaeon Pyrococcus horikoshii via genome sequencing. The gene was over-expressed in Escherichia coli , and the crystal structure of the produced enzyme was determined to 2.0 Å resolution in the presence of malate, a substrate analogue. The overall structure exhibits a unique triangular architec-ture composed of a 3-fold repeat of three-layer ( (cid:1)(cid:2)(cid:1) ) sandwich folding. Although some aspects of the fold ho-mologous to the each domain have been observed previously, the overall structure of quinolinate synthase shows no similarity to any known protein structure. The three analogous domains are related to a pseudo-3-fold symmetry. The active site is located at the interface of the three domains and is centered on the pseudo-3-fold axis. The malate molecule is tightly held near the bottom of the active site cavity. The model of the catalytic state during the first condensation step of the quinolinate synthase reaction indicates that the elimination of inorganic phosphate from dihydroxyacetone phosphate may precede the condensation reaction. NAD fundamental direct it was shown

A gene encoding a quinolinate synthase has been identified in the hyperthermophilic archaeon Pyrococcus horikoshii via genome sequencing. The gene was overexpressed in Escherichia coli, and the crystal structure of the produced enzyme was determined to 2.0 Å resolution in the presence of malate, a substrate analogue. The overall structure exhibits a unique triangular architecture composed of a 3-fold repeat of three-layer (␣␤␣) sandwich folding. Although some aspects of the fold homologous to the each domain have been observed previously, the overall structure of quinolinate synthase shows no similarity to any known protein structure. The three analogous domains are related to a pseudo-3-fold symmetry. The active site is located at the interface of the three domains and is centered on the pseudo-3-fold axis. The malate molecule is tightly held near the bottom of the active site cavity. The model of the catalytic state during the first condensation step of the quinolinate synthase reaction indicates that the elimination of inorganic phosphate from dihydroxyacetone phosphate may precede the condensation reaction.
NAD is an essential and ubiquitous coenzyme that plays a fundamental role in cellular metabolism. Besides its direct action on redox equilibrium, it was recently shown to play important roles in DNA repair, calcium-dependent signaling, and life span extension (1)(2)(3). NAD biosynthesis is accomplished through either de novo or salvage pathways that differ substantially in prokaryotes and eukaryotes (4 -6). In prokaryotes, the de novo pathway proceeds through a condensa-tion reaction between L-aspartate and dihydroxyacetone phosphate (DHAP) 1 that is catalyzed by two enzymes, L-aspartate oxidase (LAO; nadB gene product), and quinolinate synthase (QS; nadA gene product). LAO catalyzes the oxidation of Laspartate to iminoaspartate, while QS catalyzes the condensation of iminoaspartate with DHAP to produce quinolinate (QA), which is then converted to NAD via a metabolic sequence common to all organisms. Until now, efforts to understand the catalytic mechanism of QS have been hampered by the chemical instability of iminoaspartate (t1 ⁄2 ϭ 140 s at pH 8.0 and 25°C) (7). It may be that iminoaspartate is directly transferred from LAO to the QS active site (6), but there is no evidence that QS and LAO function as a multienzyme complex. Moreover, nothing is currently known about the structure of QS.
Recently, we observed the presence of LAO in Pyrococcus horikoshii, an anaerobic hyperthermophilic archaeon (8). This is the first example of the occurrence of LAO either in the Archaea or in obligate anaerobic organisms. In addition to the oxidase reaction, the enzyme catalyzed L-aspartate dehydrogenation in the presence of fumarate. The genes that encode the homologue of all other enzymes involved in the de novo NAD biosynthesis were identified in the P. horikoshii genome. Thus, we proposed that the de novo pathway functions in the organism under anaerobic conditions. To elucidate the entire aspect of the pathway in P. horikoshii, we have undertaken the structural and functional analysis of these gene products (9).
In this study, the crystal structure of QS from P. horikoshii was determined in the presence of malate, a substrate analogue of iminoaspartate. Knowledge of the structure of QS should yield information about the catalytic mechanism of the enzyme.

EXPERIMENTAL PROCEDURES
Cloning, Protein Expression, and Purification-The gene encoding QS (open reading frame ID: PH0013, the gene information is available at www. bio.nite.go.jp/dogan/Top) was amplified by PCR. The following set of oligonucleotide primers was used to amplify the QS gene fragment: one is the primer (5Ј-GGATGTTCATATGGATTTAGTTGAA-GAAATTTTGAGG-3Ј) containing a unique NdeI restriction site overlapping the 5Ј initiation codon, and the other (5Ј-CCGGATCCTCAT-TTGCTCATCTCCAGCATTCTT-3Ј) contained a unique BamHI restriction site proximal to the 3Ј-end of the termination codon. The chromosomal P. horikoshii DNA was isolated as described (10) and used as the template. The amplified 0.9-kb fragment was digested with NdeI and BamHI and ligated with the expression vector pET11a (Novagen) linearized with NdeI and BamHI to generate pEQS. The E. coli strain BL21(DE3) codon plus RIL (Stratagene) was transformed with pEQS. The transformants were cultivated at 37°C in 500 ml of a nutritionally rich medium as described previously (11) until the optical density at 600 nm reached 0.6. The induction was carried out by the addition of 0.4 mM isopropyl ␤-D-thiogalactopyranoside to the medium, and cultivation was continued for 3 h. Cells were harvested by centrifugation, suspended in buffer (10 mM potassium phosphate, pH 7.0, containing 1 mM phenylmethylsulfonyl fluoride, 10 mg/ml lysozyme from egg white, and 1 mg/ml DNase I from bovine pancreas), incubated at 37°C for 10 min, and lysed by sonication. The crude extract was heated at 90°C for 15 min in the presence of 0.2 M Na 2 SO 4 and clarified by centrifugation. The protein was subjected to gel filtration on a Superdex 200 26/60 column (Amersham Biosciences) equilibrated with 10 mM potassium phosphate, * This work was supported by the National Project on Protein Structural and Functional Analysis promoted by the Ministry of Education, Science, Sports, Culture, and Technology of Japan. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Fig. 1 and Table I pH 7.0, containing 1 mM dithiothreitol, after which the resulting protein solution was concentrated to 60 mg/ml by ultrafiltration.
Determination of Enzyme Activity-The QS activity was determined by measuring the formation of QA from iminoaspartate and DHAP with high performance liquid chromatography. The iminoaspartate was provided by LAO reaction (12). The recombinant P. horikoshii LAO was prepared as described previously (8). The reaction mixture contained 50 mM Tris/HCl buffer, pH 7.5, 50 mM L-aspartate, 2 mM DHAP, 80 units of catalase, 2.3 g of P. horikoshii LAO, and 1.5 g of purified QS in a total volume of 0.4 ml. After incubation for the appropriate time at 50°C, the reaction was stopped by the addition of 50 l of 0.4 M ice-cold perchloric acid and then neutralized with 60 l of 0.2 M K 2 CO 3 . After leaving on ice for 5 min, the precipitate was removed by centrifugation, and the supernatant solution was passed through a cellulose acetate filter (pore size 0.2 m, Advantec). An aliquot of each filtrate was subjected to a column (4.6 mm ϫ 15 cm) of TSK gel ODS-120Ts (Tosoh). K 2 HPO 4 (70 mM, pH 2.6) was used as the mobile phase at a flow rate of 0.5 ml/min. The effluent from the column was monitored by a UV detector at a wavelength of 254 nm. QS was separated on a column at a retention time of about 9.1 min.
Crystallization and Data Collection-Stura Footprint Screens (Molecular Dimensions Ltd.) were used to screen for crystallization conditions. Crystallization was performed using the hanging-drop vapor diffusion method, in which 2 l of protein solution (60 mg/ml) were mixed with an equal volume of reservoir solution. Initially, a rodclustered crystal was grown from reagent 1-1 (15% (v/v) polyethylene glycol 600, 0.2 M imidazole malate, pH 5.5), but because this crystal was difficult to reproduce, it was crushed and suspended in the reagent 1-1, and a new drop was inoculated with microseeds from the slurry. After several cycles of seeding, diffraction quality crystals were obtained after 3 days in the presence of 0.2 M imidazole, 0.1 M malate, and 18% (v/v) polyethylene glycol 600 (pH 5.3 at 20°C). It is noteworthy that no crystals were obtained when malate was removed from the reservoir solution. The crystal belongs to the hexagonal space group P6 3 , with unit cell parameters of a ϭ b ϭ 94.6 Å and c ϭ 62.6 Å and diffracts to 2.0 Å. Before cryo-cooling, crystals were stirred gently in Paratone-N (Hampton Research). We collected all diffraction data at 100 K on beamline BL5 at the Photon Factory (Tsukuba, Japan) using monochromatized radiation at ϭ 1.0 Å and an ADSC CCD detector. The data were processed using HKL2000 (13).
Phasing and Refinement-Heavy atom derivatives were prepared by soaking the crystals in reservoir solution containing 1 mM mercury acetate for 10 h or 1 mM HAuCl 4 for 11 h. Phase calculation was carried out using the multiple isomorphous replacement with an anomalous scattering (MIRAS) method using SOLVE (14). The MIRAS map at 2.1 Å was subjected to maximum likelihood density modification, followed by autotracing using RESOLVE (14). The model was built using XtalView (15) and refined at 2.0 Å resolution with Refmac5 (16) and CNS (17). A malate molecule was clearly visible in both A -weighted 2F o Ϫ F c and F o Ϫ F c density maps and was included in the latter part of the refinement. The R factor and R free values of the final model were 21.7% and 26.8%, respectively (supplemental Table I). The final structure exhibits good geometry without Ramachandran outliers. Molecular graphics figures were created using PyMOL (www.pymol.sourceforge.net/).

RESULTS AND DISCUSSION
Enzyme Activity-The predicted amino acids sequence of the P. horikoshii QS showed 35% identity to that reported for the E. coli QS (supplemental Fig. 1) (12). The formation of QA from L-aspartate, O 2 , and DHAP in the presence of the E. coli QS and LAO has been demonstrated (12). In the present study, we performed the determination of QA formed by the P. horikoshii QS and LAO and found that QA is produced in a time-dependent manner within 30 min under the used conditions. The specific activity was estimated to be 2.2 mol/min/mg. This provides direct proof that the PH0013 gene product is actually QS. Detailed characteristics of the enzyme will be described elsewhere.
Correlation among Three Domains-We also detected a 3-fold repeat of about 85 amino acid residues in the amino acid sequence of QS that corresponds to the three domains. This suggests QS evolved through a process involving two gene duplications. Fig. 2c shows the amino acid alignment of the three repeats, which are related by 20 -30% sequence identity. The folds of the three repeats are consistent with one another, except that helix ␣1 of repeat 1 extends away from the ␤ sheet in Domain 1 (Fig. 2a), and helix ␣13 presents in an arm extending from Domain 3 to helix ␣14 at the C terminus to complete the three-layer ␣␤␣ sandwich of that domain (Fig.  2b). Superposition of the three domains yields root mean square deviations of less than 1.5 Å for 47 equivalent C␣ atoms (residues 15-43/53-70 in Domain 1, 103-131/141-158 in Domain 2, and 190 -218/226 -243 in Domain 3) (Fig. 2b). At the interfaces between adjacent domains, we detected only one to four pairwise interactions based on hydrogen bonding. On the other hand, we found striking hydrophobic interactions around helices ␣13 and ␣14: Ile 262 , Leu 264 , Ile 267 , Leu 271 , and Met 274 in helix ␣13 and its vicinity were buried in a large hydrophobic cluster formed by Ile 7 (␣1), Leu 10 (␣1), Ala 16  With these interactions, the arm extending from Domain 3 to the C terminus appears to fix the three domains and could be essential for formation of the tripartite structure of QS. FIG. 1. Overall structure of QS. a, ribbon diagrams showing QS from the concave side. The rainbow drawing shows the N terminus in blue and the C terminus in red. The bound malate molecule is shown as a stick model in red. b, surface electrostatic potentials. Positive and negative surfaces are shown in blue and red, respectively. The view from the concave side (left) and a section through the active site cavity (right) are shown. The bound malate molecule is shown as a stick model in yellow. The surface was drawn using the program GRASP (29).

Crystal Structure of Quinolinate Synthase 26646
Structural Homologues-When we searched the structural databases using DALI (18), we found that the structure of the Domain 1 is most closely related to that of the N 5 -carboxyaminoimidazole ribonucleotide mutase (PurE) monomer (Protein Data Bank entry 1QCZ) (19), having a Z-score of 6.6 and an root mean square deviation of 2.2 Å over 70 C␣ positions. The central domain of PurE consists of a five-stranded parallel ␤ sheet flanked by three ␣ helices on one side and two on the other and adopts a fold akin to the dinucleotide-binding domain found in many nucleotide-binding enzymes (20). Domains 2 and 3 are most similar to sulfurtransferase (GlpE; Protein Data Bank entry 1GN0) (21), which adopts a fold typical of a single ␣/␤ rhodanese domain. Still, the overall structure of QS shows no similarity to any known protein structures. By contrast, the degree of sequence conservation among QS homologues is extremely high. Apparently, the unique overall structure of P. horikoshii QS is a common feature of the QS family.
Active Site-The active site of QS is located at the interface of the three domains on the concave side of the molecule and is centered on the pseudo-3-fold axis. The active site cavity is ϳ10 Å deep with a narrow opening and a wide inner space. The electrostatic potential surface shown in Fig. 1b Fig. 2a were used). These residues are completely conserved in all QS homologues listed in the data bases, as well as in the E. coli QS (supplemental Fig. 1). Our initial experimental electron density map showed extra density within the cavity, and after construction and refinement of the peptide chain, a malate molecule could be modeled unambiguously into that density. The map clearly defined the precise orientation of the malate (Fig. 3a): two oxygen atoms from the C-1 carboxylate of malate form hydrogen bonds with the side chains of His 21 , His 196 , and Thr 213 and the main chain amide proton of Thr 213 ; the oxygen atoms of the C-4 carboxylate are within hydrogen bonding distance of the side chains of Tyr 109 and Ser 126 , the backbone amide protons of Ser 38 and Ser 126 , and a water molecule (W1); and the oxygen atom of the C-2 hydroxyl group is hydrogen-bonded to the side chains of Tyr 109 and His 196 . With these bonds, the malate is tightly held near the bottom of the cavity (Fig. 1b).
Insight into the QS Reaction-The formation of QA from iminoaspartate and DHAP involves the removal of an inorganic phosphate and two water molecules (7). Labeling studies previously showed that the phosphate-bearing carbon of DHAP is incorporated into C-4 of QA (22) and that in the presence of both QS and LAO the aspartate nitrogen and carbon are incorporated directly into the pyridine ring of QA (23). Those results provide definitive information on the direction of approach and orientation of DHAP in the catalytic site. With that information, we were able to estimate the positions of C-1 and C-2 of the bound DHAP and the pyridine ring of QA from the orientation of malate. But we were unable to find a space to accommodate the DHAP phosphate group in our structure nor were we able to model both DHAP and malate bound to a single QS active site. For that reason, we built our model of the catalytic The malate molecule is shown as a stick model in yellow. Oxygen and nitrogen atoms are shown in red and blue, respectively. The final A -weighted 2F o Ϫ F c electron density map for malate is shown at the 1 level. b, residues in the active site pocket and proposed model for the first condensation step of the QS-catalyzed reaction. The condensation product is shown as a stick model; the scaffold from DHAP is shown in magenta and that from iminoaspartate is in yellow. The estimated QA molecule is indicated as a wire frame (cyan). Atoms are colored as described for a. state during the first condensation step of the QS reaction without phosphate (Fig. 3b).
Nasu et al. (7) proposed the following reaction mechanism for QS based on the aforementioned labeling studies (Fig. 4, Pathway 1): 1) the electron-withdrawing groups of iminoaspartate facilitate removal of a proton from C-3, and the resulting molecule carries out a nucleophilic attack on C-3 of DHAP accompanied by the elimination of inorganic phosphate; 2) the condensation product formed undergoes a keto-aldo isomerization; and 3) the loss of a proton from the resulting molecule produces an amino aldehyde that undergoes a Schiff-base formation followed by dehydration to produce QA. This mechanism differs substantially from the pathway recently proposed by Begley et al. (4) (Fig. 4, Pathway 2): 1) initially DHAP is isomerized to glyceraldehyde 3-phosphate in a reaction analogous to one catalyzed by triose phosphate isomerase; 2) subsequent imine formation precedes the elimination of phosphate; and 3) electrocyclic ring closure of the resulting molecule followed by tautomerization and dehydration yields QA. The second mechanism certainly requires formation of a condensation product with a phosphate group; thus our binding model without phosphate suggests the first mechanism is more feasible. That mechanism is also in good agreement with the observation that QS does not utilize glyceraldehyde 3-phosphate as a threecarbon precursor (24). That said, when we placed QS crystals in a reservoir solution containing DHAP, x-ray diffraction quality quickly deteriorated, which could mean that DHAP binding causes a large conformational change in the structure of the enzyme-malate complex. The structure of the tertiary complex with DHAP bound will be the focus of further investigation.
It has been suggested that QS is one of the potential targets for the development of novel antibacterial agents (25), as well as LAO (26), because the two-step conversion of aspartate to quinolinate is not present in humans, in whom de novo NAD biosynthesis occurs via degradation of tryptophan (5). Besides de novo synthesis of NAD, a salvage pathway exists that enables NAD to be recycled. This pathway usually proceeds via degradation of NAD to nicotinic acid (NA), followed by conversion of NA to nicotinic acid mononucleotide by nicotinic acid phosphoribosyltransferase (NAPRTase). Unlike most organisms, Mycobacterium tuberculosis, a tubercular pathogen, lacks NAPRTase activity and cannot recycle NA to NAD (27). Similarly, analysis of the genome of Helicobacter pylori, a major cause of gastroduodenal disease, showed it to lack NA salvage genes, including that encoding NAPRTase (25,28). This makes QS is an excellent target for the design of drugs against these organisms, and our results may provide critical information facilitating the design of such antibacterial agents.