Crystal Structures of Glutamine:Phenylpyruvate Aminotransferase from Thermus thermophilus HB8 INDUCED FIT AND SUBSTRATE RECOGNITION*

, The following three-dimensional structures of three forms of glutamine:phenylpyruvate aminotransferase from Thermus thermophilus HB8 have been determined and represent the first x-ray analysis of the enzyme: the unliganded pyridoxal 5 (cid:1) -phosphate form at 1.9 Å resolution and two complexes with 3-phenylpropionate and (cid:1) -keto- (cid:2) -methylthiobutyrate at 2.35 and 2.6 Å resolution, respectively. The enzyme shows high activity toward phenylalanine, tyrosine, tryptophan, kynurenine, methionine, and glutamine. The enzyme is a homodimer, and each subunit is divided into an N-terminal arm and small and large domains. Based on its folding, the enzyme belongs to fold type I, aminotransferase subclass Ib. a c (cid:4) one monomer Approximately 39% the the complex with 3PP obtained by cocrystallization. A droplet of 3 (cid:3) l of protein solution (6 mg/ml, 10 m M Tris-HCl buffer, 10 m M 3PP, and 1 m M EDTA, 8.0) mixed with an equal volume of reservoir solution containing 17% (w/v) PEG4000 in 200 m M disodium tartrate and equilibrated against 400 (cid:3) l of reservoir solution. The crystals of the ttGlnAT complex with KMTB were obtained under the same conditions as used for ttGlnAT (cid:2) 3PP, except for the concentration (24% (w/v)) of PEG4000. Within a week, yellow crystals of ttGlnAT (cid:2) 3PP and ttGlnAT (cid:2) KMTB appeared. Both complex crys- tals were isomorphous with the space group P 3 1 21 and had average cell dimensions of a (cid:4) b (cid:4) 123.5 and c (cid:4) 122.7 Å. There is one dimer in the asymmetric unit, and (cid:5) 55% of the crystal volume is occupied by solvent. Preliminary x-ray data collected at 293 K on the BL45PX station at SPring-8 (Hyogo, Japan).

Aminotransferase, which requires pyridoxal 5Ј-phosphate (PLP) 1 as a cofactor, reversibly catalyzes a transamination reaction that essentially consists of the same two half-transamination reactions (Scheme 1) (1). The ␣-amino group of amino acid 1 is transferred to the PLP of the aminotransferase to give pyridoxamine 5Ј-phosphate (PMP) and keto acid 1. Next, the amino group of PMP is transferred to keto acid 2 to yield amino acid 2 and regenerate PLP. For example, aspartate aminotransferase (AspAT) takes aspartate and glutamate as amino acids 1 and 2, respectively. PLP-enzyme ϩ ␣-amino acid 1 º PMP-enzyme ϩ ␣-keto acid 1 PMP-enzyme ϩ ␣-keto acid 2 º PLP-enzyme ϩ ␣-amino acid 2 SCHEME 1 In a study by the RIKEN Structural Genomics Initiative (2), a Thermus thermophilus HB8 gene homologous to the AspAT family has been identified. A homology search using FASTA with PDBSTR (3) and CLUSTAL W (4) indicated that the gene sequence was highly homologous with that of AspAT from T. thermophilus HB8 (ttAspAT; Ref. 5), human glutamine: phenylpyruvate aminotransferase (human GlnAT; Ref. 6), rat GlnAT (7), and aromatic amino acid aminotransferase from Pyrococcus horikoshii (phAroAT; Ref. 8) with sequence identities of 41.8, 36.5, 34.2, and 32.4%, respectively. We expressed the gene in Escherichia coli, purified the product protein, and physicochemically characterized it (9). The protein, which is a PLP-dependent aminotransferase, showed high activity toward tyrosine, phenylalanine, tryptophan, kynurenine, methionine, glutamine, and the corresponding keto acids with k cat /K m values of 170,000 -4,400,000 s Ϫ1 M Ϫ1 for the half-transamination reactions and low activity toward other amino acids with k cat /K m values less than 130 s Ϫ1 M Ϫ1 . Based on the primary sequence homology and the substrate specificity, we proposed that the protein is glutamine:phenylpyruvate aminotransferase (EC 2.6.1.64; ttGlnAT) in which the amino acids 1 and 2 in Scheme 1 are glutamine and an aromatic amino acid, respectively. This is the first non-eukaryotic GlnAT to be structurally characterized.
In mammals, kynurenine aminotransferase I (KynAT I) is the equivalent of prokaryotic GlnAT and catalyzes the transamination of kynurenine to produce kynurenate (7, 10, 11) (Scheme 2). Kynurenate is involved in several aspects of the central nervous system by acting as an antagonist at both the glutamate-binding site and the allosteric glycine site of the N-methyl-D-aspartate receptor and possibly by blocking the ␣7-nicotinic acetylcholine receptor (12,13). Kynurenate levels in the central nervous system are correlated to cerebral diseases such as schizophrenia and Huntington's disease. GlnAT is an interesting target enzyme for the development of a novel therapeutic compound controlling endogenous kynurenate levels because kynurenate is the product of the GlnAT-catalyzed reaction.
We have determined the structures of the native, unliganded ttGlnAT in its PLP form and its complexes ttGlnAT⅐3-phenylpropionate (3PP) and ttGlnAT⅐␣-keto-␥-methylthiobutyrate (KMTB), as the first structures of a non-eukarytotic GlnAT. In this article we describe substrate recognition by the enzyme and concomitant conformational changes at the domain level and present a computer model for the enzyme-kynurenine complex based on the crystal structure of ttGlnAT⅐3PP.

EXPERIMENTAL PROCEDURES
Crystallization and Data Collection-The expression of ttGlnAT and the purification of the expressed enzyme have been reported previously (9). The crystallization of the unliganded ttGlnAT in the PLP form was carried out at 293 K using the vapor-diffusion. The crystals were obtained by mixing 3 l of protein solution (10 mg/ml, 10 mM Tris-HCl buffer, and 1 mM EDTA, pH 8.0) with an equal volume of reservoir solution containing 1.0 M monoammonium dihydrogenphosphate in 100 mM trisodium citrate, pH 5.6, and equilibrating against 400 l of reservoir solution. Yellow crystals appeared within 2 weeks with the space group C222 1 , cell dimensions of a ϭ 77.8, b ϭ 91.9, and c ϭ 110.7 Å, and with one monomer in the asymmetric unit. Approximately 39% of the crystal volume was occupied by solvent (14).
The crystals of the ttGlnAT complex with 3PP were obtained by cocrystallization. A droplet of 3 l of protein solution (6 mg/ml, 10 mM Tris-HCl buffer, 10 mM 3PP, and 1 mM EDTA, pH 8.0) was mixed with an equal volume of reservoir solution containing 17% (w/v) PEG4000 in 200 mM disodium tartrate and equilibrated against 400 l of reservoir solution. The crystals of the ttGlnAT complex with KMTB were obtained under the same conditions as used for ttGlnAT⅐3PP, except for the concentration (24% (w/v)) of PEG4000. Within a week, yellow crystals of ttGlnAT⅐3PP and ttGlnAT⅐KMTB appeared. Both complex crystals were isomorphous with the space group P3 1 21 and had average cell dimensions of a ϭ b ϭ 123.5 and c ϭ 122.7 Å. There is one dimer in the asymmetric unit, and ϳ55% of the crystal volume is occupied by solvent. Preliminary x-ray data were collected at 293 K on the BL45PX station at SPring-8 (Hyogo, Japan).
X-ray diffraction data sets for the unliganded crystal, the ttGlnAT⅐3PP crystal, and the ttGlnAT⅐KMTB crystal were collected to 1.9-, 2.35-, and 2.6-Å resolution at 293 K on the BL6A, BL18B, and BL18B stations, respectively, at the Photon Factory, KEK (Tsukuba, Japan), using an x-ray beam with a wavelength of 1.0 Å and an ADSC Quantum 4R CCD camera. All data were processed and scaled using the program MOSFLM and SCALA (15). Data collection statistics are presented in Table I.
Structure Determination and Refinement-The structure of the unliganded ttGlnAT was determined with the program AMoRe (16), using the previously determined structure of the unliganded ttAspAT in the PLP form (5) as a search model. The modeling of the polypeptide chain was performed using the program O (17). The structure was refined by simulated annealing and energy minimization with the program CNS (18). Water molecules were picked up on the basis of the peak heights (3.0 ) and distance criteria (4.0 Å from protein and solvent) from the sigmaA weighted F o Ϫ F c map. The water molecules whose thermal factors were above the maximum thermal factor of the main chain after refinement were removed from the list. Further model building and refinement cycles resulted in an R factor of 18.1% and an R free of 21.8%, calculated for 30,807 reflections (F o Ͼ 2(F o )) ( Table I). During the last step of the refinement, unambiguous water molecules were added including those with a temperature factor higher than 50 Å 2 . The maximum temperature factor of the water molecules was 60 Å 2 .
The initial structures for ttGlnAT⅐3PP and ttGlnAT⅐KMTB were determined with AMoRe (16), using the structure of the unliganded ttGlnAT as the search model. The same refinement procedure as was used for the unliganded ttGlnAT was applied to ttGlnAT⅐3PP and ttGlnAT⅐KMTB. When the R factor value decreased below 30%, the substrate analogue 3PP or KMTB was introduced into a peak on a simulated annealing 2F o Ϫ F c map. Further model building and refinement cycles for ttGlnAT⅐3PP resulted in an R factor of 18.8% and an R free of 22.2%, calculated for 44,463 reflections (F o Ͼ 2(F o )). Further model building and refinement cycles for ttGlnAT⅐KMTB resulted in an R factor of 23.0% and an R free of 28.6%, calculated for 32,091 reflections (F o Ͼ 2(F o )) ( Table I). The maximum thermal factors of the assigned water molecules in ttGlnAT⅐3PP and ttGlnAT⅐KMTB were 66 and 64 Å 2 , respectively.
Modeling of ttGlnAT in Complex with Kynurenine-During the modeling, it was assumed that ttGlnAT⅐kynurenine as a Michaelis complex is in the closed form based on x-ray structures of ttGlnAT⅐3PP and ttGlnAT⅐KMTB. First, a model of the complex between ttGlnAT and tryptophan was constructed by replacing 3PP in ttGlnAT⅐3PP with tryptophan. The model structure was optimized by minimizing the empirical potential energy composed of stereochemical factors (bond lengths, bond angles, and torsion angles), van der Waals forces, and electrostatic interactions using the program CNS (18). The ttGlnAT complex with kynurenine was then modeled on the assumption that kynurenine is bound to the active site in a manner similar to tryptophan. The 3-anthraniloyl and alanyl portions of kynurenine were located at the corresponding indolyl and alanyl groups of tryptophan, with the o-amino group of kynurenine superimposed onto the indolyl NH group of tryptophan. The model was optimized by a slow cooling molecular dynamics simulation using a standard protocol with a starting temperature of 500 K (18).
Quality of the Structure-The final model of the unliganded ttGlnAT comprises 365 residues (1-10, 14 -368) out of 368 residues within a subunit and one PLP with 171 water molecules. The average thermal factor of the main-chain atoms is 19.4 Å 2 . The final model of ttGlnAT⅐3PP comprises all residues for subunit 1, 367 residues (1-367) for subunit 2, two PLPs, and two 3PPs with 167 water molecules. The structures of the two independent subunits are quite similar. Superimposition of subunit 1 onto subunit 2 yields an r.m.s. deviation of 0.12 Å for 367 C␣ positions. The average thermal factors of the main-chain atoms (N, C␣, C, and O) in subunits 1 and 2 are 31.5 and 36.0 Å 2 , respectively. The final model of ttGlnAT⅐KMTB comprises all residues for subunit 1, 367 residues (1-367) for subunit 2, two PLPs, and two KMTBs with 87 water molecules. The structures of the two independent subunits are quite similar. Superimposition of subunit 1 onto subunit 2 yields an r.m.s. deviation of 0.21 Å for 367 C␣ positions. The average thermal factors of the main-chain atoms in subunits 1 and 2 are 36.7 and 44.1 Å 2 , respectively. The differences in the number of residues and the average thermal factors between the subunits 1 and 2 in the final models are due to the asymmetry of the intermolecular interactions. The thermal factors for both complexes reflect the fact that crystal packing has less of an effect on subunit 2 than on subunit 1.
Analysis of the stereochemistry with PROCHECK (19) showed that SCHEME 2 all the residues except for Phe-253 (⌽ ϭ 68°, ⌿ ϭ Ϫ66°) fall within the generously allowed region of the Ramachandran plot for all structures.
On the basis of the electron density maps, it was confirmed that the conformation of Phe-253 is correct. The residues corresponding to Phe-253 are threonine in ttAspAT and serine in E. coli, pig cytosolic, and chicken mitochondrial AspATs. These residues have strained mainchain dihedral angles similar to that observed in ttGlnAT and participate in recognition of the substrate side chains (5, 20 -22). The simulated annealing 2F o Ϫ F c electron density maps for the active site around the PLP in the unliganded ttGlnAT and ttGlnAT⅐3PP are shown in Fig. 1. Fig. 2 displays the C␣ backbone tracings of ttGlnAT, ttAspAT, and E. coli AspAT in ribbon model drawings. Structure diagrams were drawn using the programs Molscript (23), Bobscript (24), and Raster3D (25).

RESULTS AND DISCUSSION
Overall Structure-ttGlnAT in the PLP form is a symmetric dimer ( Fig. 2A). The subunit structure of ttGlnAT in the PLP form, with secondary structure assignments as predicted by the program DSSP (26), is shown in Fig. 3. Each subunit is divided into an N-terminal arm comprising residues 1-10, a small domain formed by two parts of the polypeptide chain from residues 11-38 and 279 -368, respectively, and a large domain from residues 39 -278. The core of the large domain is a sharply twisted seven-stranded sheet (parallel except for strand b10), which is surrounded by helices H5, H6, and H9 from the activesite side and helices H4, H7, and H8 from the surface side (Fig.  3). Helices H3 and H10, which are located around the molecular 2-fold axis, participate in the formation of the subunit interface. ␤-Strands b5 and b6 form a hairpin ␤ structure at the domain interface. The small domain also has an antiparallel four-stranded sheet as its core, and this is flanked by four helices (H2, H11, H12, and H13) from the surface side. The molecule has two active-site cavities around a molecular 2-fold axis ( Fig. 2A). Each cavity is located at the domain interface of one subunit and at the subunit interface.
PLP-dependent enzymes have been classified into 4-fold groups (fold types I-IV) (27)(28)(29)(30). AspATs were assigned to fold type I, aminotransferase subclass I, which is further subdi-vided into subclasses Ia and Ib (31). The program DALI (32) was used to search the Protein Data Bank data base for PLPdependent enzymes possessing three-dimensional structures similar to that of the unliganded ttGlnAT. The highest Z scores (strength of structural similarity) were calculated to be 46.  (22). Due to the fact that the first three enzymes with the highest Z scores belong to the subclass Ib and the last three enzymes belong to the subclass Ia, ttGlnAT is classified as a member of subclass Ib. The subunit C␣ carbon atoms of ttGlnAT can be superimposed onto those of ttAspAT and E. coli AspAT with r.m.s. deviations of 1.6 and 2.7 Å for 355 and 337 C␣ atoms, respectively.
Open-Closed Conformation Change-One of the striking features of the AspATs in subclass Ia is the overall conformational change from the open to the closed form upon binding of substrates (20 -22, 34 -36). The small domain rotates as a rigid body to close the active site. On the other hand, in the ttAspAT of subclass Ib, only the N-terminal helix region from residues 13-30 (helix H2 in ttGlnAT) of the small domain approaches the active site to enclose the substrate inside the protein (5) (Fig. 2).
Upon binding of the substrate analogue (3PP) to ttGlnAT, a part of the small domain (the subdomain) comprising residues 14 -31 from helix H2 and ␤-strand b1 and residues 319 -332 from H12 and b13 moves to close the active site. The subunit C␣ atoms, except for the subdomain C␣ atoms in ttGlnAT⅐3PP, were superimposed onto the corresponding atoms in the unliganded ttGlnAT by least squares fitting with an r.m.s. deviation of 0.30 Å for 333 C␣ atoms. However, in this calculation, larger differences were observed in the subdomain C␣ atoms with an r.m.s. deviation of 0.71 Å for 32 C␣ atoms. This indicates that a significant conformational change occurs in the subdomain region (the assembled H2, b1, H12, and b13 regions of the small domain). This change can be approximated by a  domain of ttGlnAT that is quite similar to that observed in ttGlnAT⅐3PP. The bound 3PP or KMTB is almost shielded from the solvent region because the accessible surface areas of 315 and 306 Å 2 in 3PP and KMTB are reduced to 0 and 1 Å 2 , respectively.
Active-site Structure of ttGlnAT in the Open Form-The stereo structure and hydrogen bonding scheme of the active site are shown in Figs. 4A and 5A, respectively. PLP is located at the bottom of the active-site cavity with its si-face directed toward the protein side. PLP forms an internal aldimine bond (Schiff base linkage) with Lys-222 and interacts extensively with the residues located at or near one end of the sevenstranded sheet of the large domain in one subunit, with the exception of Tyr-57* (asterisk indicates a residue from another subunit of the dimer unit). Lys-222 (which is a catalytic residue), Asn-163, Tyr-194, and Asp-191 (which play roles in the tuning of the electronic state of the cofactor ring (37-40)), Arg-230 and Tyr-57* (which interact with the phosphate of PLP), and Phe-112 and Val-193 (which are associated with cofactor ring orientation) are conserved as subclass I aminotransferases and thus are essential for cofactor binding and/or catalytic action. Asn-163 and Tyr-194 interact weakly with the phenolic oxygen O3Ј of PLP with distances of 3.4 (O3Ј-NH 2 ) and 3.5 Å (O3Ј-OH), respectively. Asp-191 forms an ion pair with the protonated nitrogen atom of the pyridine ring of PLP. The pyridine ring of PLP interacts with the side chain of Val-193 on the si-face side and with the phenyl ring of Phe-112 on the re-face side. The phosphate group of PLP interacts with Arg-230, Tyr-57*, Ser-219, the NH groups of Ala-87 and Thr-88, and two water molecules, which act as an anchor to fix the cofactor to the active site. Ala-87 and Thr-88 belonging to the N-terminal loop of helix H5 are not conserved in the subclass I aminotransferases, but the coordination of the main-chain NH groups in the loop to the phosphate of PLP is common to subclass I aminotransferases. The negative charge of the phosphate group is compensated by the positive charge of Arg-230 and the dipole of helix H5 as is observed in the subclass I aminotransferases. Ten water molecules are located in the active-site cavity and are involved in the formation of a hydrogen-bond network with the active-site residues (Fig. 4A).
Schiff Base Linkage between PLP and Lys-222-The unliganded AspATs of subclass I whose structures have been determined so far exhibit an unusually low pK a value (about 5.1-6.9) of the Schiff base and a significant deviation of the C4ЈϭN bond from the planar pyridine ring of PLP with a dihedral angle (C3-C4 -C4Ј-N) of 52-91° (8,20,22,41). The relatively low values of dihedral angles (35 and 42°) observed in pig cytosol AspAT (34) and ttAspAT (5) are possibly due to the increase in the proportion of the protonated Schiff base by the low pH values (4.3 and 5.4, respectively) of the crystallization buffer. The deviation of the C4ЈϭN bond and the low pK a value of subclass I AspATs were proved to be caused by the stereochemical strain of the protonated Schiff base coplanar with the cofactor ring. The strain is due to the chemical linkage of the Schiff base to the protein backbone and the restricted rotation of the pyridine ring of PLP by the access of a hydrophobic side chain from the si-face of PLP (39,40).
In the unliganded ttGlnAT, the Schiff base is characterized by an unusually high pK a value of 9.3 (9) for subclass I aminotransferases and is roughly coplanar with the cofactor ring with a dihedral angle of C3-C4 -C4Ј-N ϭ 34°. Two factors are considered to be involved in the high pK a value. One factor is the weak interaction between O3Ј of PLP and Asn-163 and O3Ј of PLP and Tyr-194 when compared with those observed in other aminotransferases in subclass I. This results in the migration of the negative charge on O3Ј into the delocalized -system of the cofactor-Schiff base conjugate to increase the pK a value of the Schiff base. The other factor is the stereochemical congestion between PLP and Ala-87 (Gly in ttAspAT), and PLP and Val-193 (Ile in ttAspAT). When PLP in ttGlnAT is present in the same arrangement as that in ttAspAT, the side chain of Ala-87 makes contacts with C5Ј of PLP with a distance of 3.0 Å and that of Val-193 makes contact with C5, C6, and C5Ј of PLP with distances of 3.2, 3.0, and 3.2 Å, respectively. These short contacts push the C6 -C5-C5Ј portion of PLP from the si-face side to rotate the cofactor ring by about 22°around the C2-C3 bond toward the solvent side when compared with that of PLP in ttAspAT. This rotation decreases the deviation of C4Ј ϭ N from the cofactor ring plane to raise the pK a value. directions to interact with the hydroxy group of Ser-13, and the CϭO (Lys-11) and NH (Ser-13) group of the hydrophilic stretch, respectively. Asp-113 and Glu-249* thus play some roles in the disorder-order transition of the hydrophilic stretch and the open-closed conformational change. The flexible stretch Gly-31-Gln-32-Gly-33 changes its local conformation of the main chain (5,20,34). In the open form, only the NH group of Gly-31 interacts with Arg-347. Upon binding of 3PP, the NH group of Gly-33 approaches the carboxylate of 3PP with the hydrogen bond between Gly-31 and Arg-347 still maintained. Gln-32 changes its side-chain direction from the solvent side to the active site to form a hydrogen bond with the hydroxy group of Ser-19 on helix H2, thereby filling the space of the active-site wall.

Active-site Structures of PLP-type ttGlnAT⅐3PP and ttGlnAT⅐KMTB in the Closed
KMTB, which is the keto acid of methionine, acts as the substrate to PMP-type ttGlnAT, while KMTB interacts with PLP-type ttGlnAT as an inhibitor or a substrate analogue. The stereo structure of the active site in ttGlnAT⅐KMTB is shown in Fig. 4C. The active-site folding and the side-chain arrangement of the active-site residues are quite similar to those in ttGlnAT⅐3PP because the main-chain and side-chain atoms of ttGlnAT⅐KMTB are superimposed onto the corresponding ones of ttGlnAT⅐3PP with an r.m.s. deviation of 0.11 Å and a maximum displacement of 0.66 Å for 164 atoms. The location of the cofactor-Lys-222 conjugate is the same as that in ttGlnAT⅐3PP with a dihedral angle of C3-C4 -C4Ј-N ϭ 50°.
3PP in ttGlnAT⅐3PP may be replaced by the substrate phenylalanine to give a model of a Michaelis complex. The ␣-carboxylate and the side chain of the substrate have locations quite similar to those of 3PP. Similarly, KMTB in ttGlnAT⅐KMTB may be replaced by the substrate methionine. In the models, the ␣-amino group of the substrate is directed toward the PLP-Lys-222 Schiff-base within a distance of 3.4 Å from the C4Ј atom of PLP. This implies that the ␣-amino group is in a geometric orientation favorable to form the external aldimine with PLP.
Substrate Recognition-X-ray structures showed that the large domain of ttGlnAT has the same conformation in the open and the closed forms. In the open form, the large domain is equipped with template cavity to bind a substrate (Fig. 4). The cavity is formed by PLP and the large domain residues (Thr-88, Phe-112, Asp-113, Val-114, Asn-163, Arg-347, Tyr-57*, and Phe-253*). It is possible that a substrate approaches the built-in cavity, which is exposed to the solvent region, and this The active-site cavity is exposed to the solvent region with the hydrophilic stretch (Lys-11-Glu-12-Ser-13) at the N terminus of the small domain disordered. B, a close-up view of the active site of the ttGlnAT⅐3PP complex in the closed form. 3PP is drawn in green. The hydrophilic stretch, which is disordered in the open form, exhibits an ordered structure in the closed form to cover the bound 3PP from the solvent side. C, a close-up view of the active site of the ttGlnAT⅐KMTB complex. KMTB is drawn in green. The side chain of KMTB has a folded conformation and is located at approximately the same position as the phenyl ring of 3PP in ttGlnAT⅐3PP.
is followed by a subdomain movement toward the large domain which encapsulates the substrate within the cavity to produce a Michaelis complex.
In ttGlnAT, the ␣-carboxylate of 3PP forms a salt bridge with the guanidino group of Arg-347. The salt bridge is fixed by hydrogen bonding interactions with the NH of Gly-33 and the side chain of Asn-163 at both sides of the salt bridge (Fig. 5B). The interactions of the ␣-carboxylate of 3PP with the activesite residues observed in ttGlnAT⅐3PP are conserved in subclass I aminotransferases. The pocket for the hydrophobic side chain of 3PP is constructed by Ser-13, Phe-15, Gln-32, Thr-88, Phe-112, Asp-113, Val-114, Tyr-57*, Phe-253*, and PLP (Fig.  4B). Gln-32, Phe-253*, and Phe-112 sandwich the phenyl group of 3PP from both sides, Tyr-57* and PLP form the bottom of the hydrophobic pocket, while Phe-15 and the Ser-13-Asp-113 hydrogen-bonded pair cover the pocket from the solvent side. Thr-88 and Val-114 become directed to the top of the phenyl group of 3PP. Ser-13, Phe-15, and Gln-32 approach 3PP and become involved in the formation of a recognition site for 3PP. Therefore, the open-closed conformational change plays an important role in the recognition of the substrate.
KMTB is bound to the active site in the same manner as 3PP, with the ␣-carboxylate and the side chain of KMTB recognized by Arg-347 and the hydrophobic pocket, respectively. The side chain of KMTB exists in a folded conformation with dihedral angles of Ϫ179°around the C ␤ -C ␥ bond and Ϫ52°around the C ␥ -S ␦ bond. By taking the folded conformation, which is slightly strained compared with the extended conformation, KMTB can fit its side chain into the hydrophobic pocket of the enzyme. Phe-15, Val-114, and Phe-253*, which are involved in the formation of the hydrophobic pocket for substrates, exhibit slight differences in side-chain conformations between ttGlnAT⅐3PP and ttGlnAT⅐KMTB, which create maximum van der Waals stabilization with 3PP or KMTB.
ttGlnAT catalyzes the transamination reaction of glutamine. The substrate, glutamine, is similar to methionine in size but is different from methionine as regards the properties of its side chain. Possibly, the side chain of glutamine is included in the pocket of ttGlnAT, as is the case with methionine (refer to the KMTB structure; Fig. 4C). The inner side of the pocket is mostly hydrophobic but has four hydrophilic sites (Ser-13-Asp-113 hydrogen-bonded pair, Gln-32, Thr-88, and Tyr-57*). The most probable site to interact with the side-chain amide group of glutamine is the amide group of Gln-32 and/or the hydroxy group of Tyr-57*, although a final conclusion must await the x-ray analysis of ttGlnAT complexed with glutamine analogue. The same type of mechanism for substrate recognition has been observed in branched-chain amino acid aminotransferase from E. coli (42,43), in which the hydrophobic pocket can bind the acidic side chain of glutamate as well as the aliphatic side chain of the hydrophobic amino acids, and has four polar sites to recognize the ␥-carboxylate of glutamate.
The active-site folding of ttGlnAT is quite similar to that of ttAspAT, but the substrate specificity between them is quite different. ttAspAT is strictly specific for acidic amino acids. What differentiates the substrate specificity of ttAspAT from that of ttGlnAT? The residues that recognize the substrate ␣-carboxylate are common to both enzymes and are arranged in the same way; however, the residues that recognize the substrate side chain are extensively rearranged between the two enzymes. In ttAspAT, the ␤or ␥-carboxylate of the substrate is directly coordinated by Thr, Lys, and Trp and indirectly by Ser, Ser, and Thr through the medium of water molecules. In ttGlnAT, these residues are replaced by Phe-15, Thr-88, Phe-112, Ser-13, Val-114, and Phe-253*, respectively, which form the hydrophobic pocket with hydrophilic sites. This indicates that the change in substrate specificity from acidic amino acids to aromatic amino acids, methionine, and glutamine can be attained through the replacement of active-site residues utilizing the same main-chain conformation of the active site and a similar type of active-site closure.
Model of the Kynurenine Complex-The mammalian GlnAT is identical to KynAT I, which catalyzes the transamination of kynurenine to form kynurenate. Kynurenate acts as a competitive blocker of the N-methyl-D-aspartate receptor and as a non-competitive blocker of the ␣7-nicotinic acetylcholine receptor (12,13). To elucidate the mode of binding of kynurenine in its Michaelis complex, a model of ttGlnAT⅐kynurenine in the internal aldimine form was constructed, and the energy was minimized with CNS (18). The main-chain and side-chain atoms of the active-site residues in the model of ttGlnAT⅐ kynurenine are superimposable onto the corresponding ones in ttGlnAT⅐3PP with an r.m.s. deviation of 0.26 Å for 206 atoms (Fig. 6). Kynurenine is divided into an alanyl and a 3-anthraniloyl moiety. The hydrophobic pocket of ttGlnAT can accommodate the 3-anthraniloyl moiety by fine adjustments in sidechain conformations within the active-site residues. The modeled kynurenine fits the active-site pocket with interatomic C-C or C-N distances larger than 3.5 Å between kynurenine and the active-site residues. In the model, the planar 3-anthraniloyl moiety is perpendicular to the ␣-carboxylate of the alanyl moiety. The ␣-carboxylate forms a salt bridge with Arg-347 in the same manner as with 3PP, and the ␣-amino group is directed toward the C4Ј of PLP.
There are carbonyl and o-amino groups in the 3-anthraniloyl moiety of kynurenine. How can the hydrophobic pocket recognize these hydrophilic groups? The inner side of the pocket has hydrophilic sites. The carbonyl and o-amino group of the 3-anthraniloyl moiety in the modeled kynurenine are directed toward Tyr-57* and Asp-113, respectively, and become within the hydrogen bonding distances from them. Using the hydrophobic pocket together with the hydroxy group of Tyr-57* and the carboxylate group of Asp-113, ttGlnAT could bind kynurenine within the hydrophobic pocket without relocating the side chains of the active-site residues. It is reasonably assumed that the indolyl group of the substrate tryptophan is recognized by Asp-113, in a similar manner to the o-amino group of kynurenine.
The substrate specificity of ttGlnAT is similar to that of human KynAT I (44) and rat KynAT I (45). The primary sequences of human KynAT I and its active site are 36.5 and 74% identical to those of ttGlnAT and its active site, respectively (6). It is expected that not only the overall fold of the molecule, but also the side-chain arrangement of the active-site residues, will be similar between ttGlnAT and human KynAT I. The important residues that are involved in the formation of a cage for the 3-anthraniloyl moiety of kynurenine in the ttGlnAT⅐ kunurenine complex model are Phe-15, Gln-32, Phe-112, Asp-113, Val-114, Tyr-57*, and Phe-253* (Fig. 6). These residues are conserved in human KynAT I, with the exception of Phe-15 (Trp in human KynAT I) and Val-114 (Cys in human KynAT I). The x-ray structure of ttGlnAT⅐3PP may serve to develop novel therapeutic compounds for cerebral diseases such as schizophrenia and Huntington's disease (12,13).