Structural Determinants for Branched-chain Aminotransferase Isozyme-specific Inhibition by the Anticonvulsant Drug Gabapentin*

This study presents the first three-dimensional structures of human cytosolic branched-chain aminotransferase (hBCATc) isozyme complexed with the neuroactive drug gabapentin, the hBCATc Michaelis complex with the substrate analog, 4-methylvalerate, and the mitochondrial isozyme (hBCATm) complexed with gabapentin. The branched-chain aminotransferases (BCAT) reversibly catalyze transamination of the essential branched-chain amino acids (leucine, isoleucine, valine) to α-ketoglutarate to form the respective branched-chain α-keto acids and glutamate. The cytosolic isozyme is the predominant BCAT found in the nervous system, and only hBCATc is inhibited by gabapentin. Pre-steady state kinetics show that 1.3 mm gabapentin can completely inhibit the binding of leucine to reduced hBCATc, whereas 65.4 mm gabapentin is required to inhibit leucine binding to hBCATm. Structural analysis shows that the bulky gabapentin is enclosed in the active-site cavity by the shift of a flexible loop that enlarges the active-site cavity. The specificity of gabapentin for the cytosolic isozyme is ascribed at least in part to the location of the interdomain loop and the relative orientation between the small and large domain which is different from these relationships in the mitochondrial isozyme. Both isozymes contain a CXXC center and form a disulfide bond under oxidizing conditions. The structure of reduced hBCATc was obtained by soaking the oxidized hBCATc crystals with dithiothreitol. The close similarity in active-site structures between cytosolic enzyme complexes in the oxidized and reduced states is consistent with the small effect of oxidation on pre-steady state kinetics of the hBCATc first half-reaction. However, these kinetic data do not explain the inactivation of hBCATm by oxidation of the CXXC center. The structural data suggest that there is a larger effect of oxidation on the interdomain loop and residues surrounding the CXXC center in hBCATm than in hBCATc.

The human BCAT isozymes are 58% identical in amino acid sequence and belong along with the Escherichia coli BCAT (eBCAT) to the foldtype IV class of PLP-dependent enzymes that transfer protons on the re-face of the cofactor (17,18). Enzymes belonging to fold-type I-III classes shuttle protons on the si-face of the planar -system of the substrate-cofactor complex. The mammalian BCATs are homodimers with subunit molecular masses ranging from ϳ41,000 to 46,000 Da. The molecular mass of the hBCATc subunit is 42,800 Da; each monomer consists of 385 amino acid residues and requires one PLP as cofactor (19). The mature form of hBCATm monomer (minus its mitochondrial targeting sequence) is 365 amino acid residues (19). Structures of the fold-type IV, Bacillus sp. YM-1 D-amino acid aminotransferase (20,21), eBCAT (22)(23)(24), hBCATm (25,26), and E. coli 4-amino-4-deoxychorismate lyase (27) have been solved. The overall structures of these enzymes as homodimers (the dimeric unit of the hexamer in eBCAT) are similar.
Steady state kinetic analysis of the BCAT isozymes shows they have the same substrate specificity; nevertheless, there are subtle differences in catalytic efficiency with individual substrates (8,19). Also, the steady state k cat values for hBCATc are higher than for hBCATm (19). A more striking difference between hBCATc and hBCATm is the sensitivity of the different isozymes to inhibition by the neuroactive drug gabapentin (1-(aminomethyl)cyclohexaneacetic acid) (Scheme 1) (15). Gabapentin is used widely for seizure control (28 -30) and is now used extensively to treat neuropathic pain, migraine headache, and several other nonepileptic conditions (31). One theory of gabapentin action, the metabolic hypothesis, suggests that gabapentin interferes with neurotransmitter glutamate synthesis via inhibition of branched-chain amino acid transamination (15,32). It was shown by Hutson and coworkers (15) that gabapentin is a competitive inhibitor of hBCATc with a K i similar to the K m for leucine. The drug is not an effective inhibitor of hBCATm. Understanding the structural basis for the specificity of gabapentin for hBCATc and kinetic differences in the two isozymes has been hindered by the lack of structural information for both hBCATs.
Another feature of the BCATs is a consensus sequence, a CXXC motif, which is conserved in the mammalian proteins but not in lower eukaryotes or prokaryotes or in other fold-type IV PLP enzymes (25,33). In hBCATm, the CXXC motif is in the vicinity of the phosphate group of the cofactor PLP (25). The structures of the PLP form of hBCATm and reaction intermediates show that the cysteine residues in the CXXC center, Cys-315 and Cys-318, share a hydrogen bond (25). Biochemical studies indicate that these residues can form an intrasubunit disulfide bond under oxidizing conditions and that Cys-315 is the peroxide-reactive thiol(ate) functioning as the redox sensor in the regulation of hBCATm (34). The second cysteine, Cys-318, subsequently reacts with the nascent sulfenic acid form of Cys-315, forming a disulfide bond and permitting reversible regulation by preventing over-oxidation of hBCATm and irreversible loss of enzyme activity (34). Less is known about the role of this center in hBCATc.
The crystal structures of native hBCATm and hBCATm complexed with substrates have shed light on the "cofactor-substrate-protein" interaction, the behavior of the CXXC motif, and the reaction pathway of the catalytic reaction (25,26). Here we report the first three-dimensional structures of hBCATc complexed with gabapentin in the oxidized (hBCATc-ox⅐gabapentin) and reduced states (hBCATc⅐gabapentin) at 1.9 and 2.4 Å of resolution, respectively, the structure of the hBCATm complexed with gabapentin in the reduced state (hBCATm⅐gabapentin) at 1.8 Å resolution, and hBCATc complexed with the substrate analogue, 4-methylvalerate (4MeVA) at 2.1 Å of resolution. These are the first structures of the gabapentin complexes of the mammalian BCATs. A structural comparison of these complexes provided insight into the molecular basis for the specificity of BCATc for gabapentin, which is bulkier than 4MeVA, and the differential sensitivity of the catalytic activity of the hBCAT isozymes to the redox state of the CXXC center.

EXPERIMENTAL PROCEDURES
Crystallization and Data Collection-The expression of hBCATc and the purification of the expressed enzyme have been reported (19). All crystallization trials were performed by the hanging drop vapor diffusion method at 277 K. Drops were prepared by mixing 3 l of the protein solution (9 mg/ml, 10 mM 3-phenylpropionate (3PP)) with an equal volume of reservoir solution containing 12% (w/v) polyethylene glycol 4000 and 200 mM MgCl 2 . Two weeks after the drop was set up for crystallization at 277 K, minute crystals of hBCATc complexed with 3PP appeared and within another month grew to chunky yellow colored crystals with the space group P2 1 2 1 2 1 , cell dimensions of a ϭ 66.8 Å, b ϭ 106.4 Å, and c ϭ 109.9 Å and with one dimer in the asymmetric unit. Approximately 37% of the crystal volume was occupied by the solvent (35). Because many trials for crystallization were unsuccessful without 3PP, the addition of 3PP to the protein solution was a critical factor in producing crystals. After the initial structure of hBCATc⅐3PP was determined, two cysteine residues, Cys-335 and Cys-338, were found to form a disulfide bond, indicating that the purified enzyme was in the oxidized state of the CXXC motif (hBCATc-ox⅐3PP). Crystals of hBCATc-ox⅐gabapentin and hBCATc-ox⅐4MeVA were obtained at 293 K by soaking hBCATc-ox⅐3PP crystals in solutions containing 10 mM gabapentin or 4MeVA, respectively, for a day before data collection. Crystals of hBCATc⅐gabapentin in the reduced form were obtained at 293 K by soaking hBCATc-ox⅐3PP crystals in solutions containing 10 mM gabapentin and 2 mM DTT. Similarly, hBCATc-ox⅐3PP crystals were soaked in solution containing 4MeVA and DTT. However, 4MeVA complex crystals were not in the reduced form but in the oxidized form.
The expression of mature hBCATm in E. coli, removal of the histidine tag by thrombin cleavage, and purification of the expressed enzyme have been reported (19,36). Crystals of hBCATm were obtained using the vapor diffusion method of hanging drops at 298 K. The drop consisted of 5 l of hBCATm solution in the PLP form and 5 l of the reservoir. The drop was equilibrated against 1 ml of the reservoir. The protein solution contained 2.5 mg/ml hBCATm in a solution of 50 mM HEPES (pH 7.0), 20 mM DTT, and 50 mM EDTA. Yellow crystals grew readily in about 1-2 week in several conditions using the Hampton crystallization reagent kit. They were optimized in grids set up with 22-30% polyethylene glycol 1500, 100 mM HEPES (pH 6.9 -7.2), and 20 mM DTT. Crystals of hBCATm⅐gabapentin in the reduced form were obtained at 293 K by soaking hBCATm crystals in their crystallization solutions with 70 mM gabapentin.
The x-ray diffraction data sets for hBCATc-ox⅐3PP, hBCATc-ox⅐gabapentin, hBCATc⅐gabapentin, and hBCATc-ox⅐4MeVA were collected to 1.8, 1.9, 2.4, and 2.1 Å of resolution at 100 K on the BL6A, BL18B, NW12, and BL44B2 stations, respectively, at the Photon Factory, KEK (Tsukuba, Japan) or at the SPring-8 (Hyogo, Japan) using an x-ray beam with a wavelength of 1.0 Å and an ADSC Quantum 4R CCD camera or MarCCD165 camera. All the data were processed and scaled using the program HKL2000 (35). Data collection statistics are presented in TABLE ONE.
Diffraction data for hBCATm⅐gabapentin were collected to 1.8 Å of resolution at the synchrotron source in Cornell university on beam line 5.0.2 with a 2 ϫ 2 array CCD detector using 1.071 Å wavelength radiation monochromatized with a double crystal Si (111). DENZO (37) and SCALEPACK were used to reduce the data.
Structure Determination and Refinement-The structure of hBCATc-ox⅐3PP was determined with the program AmoRe (38) using the previously determined structure of unliganded hBCATm (PDB code 1EKF) in the PLP form (26) as a search model. The modeling of the polypeptide chain was performed using the program O (39). The structure was refined by simulated annealing and energy minimization with the program CNS (40). The initial structure for hBCATc-ox⅐gabapentin was determined with AmoRe (38) using the structure of hBCATc-ox⅐3PP as the search model. When the R factor value decreased below 30%, gabapentin was introduced into a peak on a simulated annealing 2F o Ϫ F c map. Water molecules were picked up on the basis of the peak heights (3.0 ) and distance criteria (4.0 Å from protein and solvent) SCHEME 1 NOVEMBER 4, 2005 • VOLUME 280 • NUMBER 44 from the -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 22.1% and an R free of 26.8%, calculated for 63,764 reflections (TABLE ONE). During the last step of the refinement, unambiguous water molecules were added including those with a temperature factor higher than 56 Å 2 . The maximum temperature factor of the water molecules was 62 Å 2 .

Human Cytosolic Branched-chain Aminotransferase Structures
The initial structures for hBCATc⅐gabapentin and hBCATc-ox⅐4MeVA were determined with AmoRe (38) using the structure of hBCATc-ox⅐3PP as the search model. The same refinement procedure as that used for hBCATc-ox⅐gabapentin was applied to hBCATc⅐ gabapentin and hBCATc-ox⅐4MeVA. When the R factor value decreased below 30%, gabapentin or 4MeVA (leucine analogue) was introduced into a peak on a simulated annealing 2F o Ϫ F c map. Further model building and refinement cycles for hBCATc⅐gabapentin resulted in an R factor of 21.3% and an R free of 27.2%, calculated for 28,933 reflections. Further model building and refinement cycles for hBCATc-ox⅐4MeVA resulted in an R factor of 22.8% and an R free of 29.0%, calculated for 44,715 reflections. The maximum thermal factors (B-factors) of the assigned water molecules in hBCATc⅐gabapentin and hBCATc-ox⅐4MeVA were 79 and 68 Å 2 , respectively.
The molecular replacement method was used to solve the hBCATmgabapentin structure with the earlier solved PLP form as the search probe in the program AMoRe (38), which is part of CCP4 program suite (CCP4 Collaborative Computational Project, Number 4 (1994)). The program CNS (40) was then used for all further refinements. The slow cool protocol was used to minimize model bias, and O (41, 42) was used for examination and manual adjustment of the structure during the refinement. Several cycles of positional refinement and isotropic B-factor refinement were performed after every model building cycle. The PLP was removed from the initial refinement cycles to get an unbiased view of the active site. Toward the end of the refinement (R free 26.8%), -weighted difference Fourier maps and 2F o Ϫ F c maps clearly showed the electron density for gabapentin bound to the active site. The cofactor is seen to be covalently linked to the active-site lysine, Lys-202, and gabapentin was bound as a Michaelis complex. Water molecules were automatically located using CNS and manually checked in O. A total of 341 water molecules and an acetate molecule were modeled into the -weighted (2F o Ϫ F c ) maps. Further model building and refinement cycles for hBCATm⅐gabapentin resulted in an R factor of 20.8% and an R free of 23.2%, calculated for 68,169 reflections. The refinement parameters with the average B-factors are listed in TABLE ONE.
where I is the observed intensity, and ͗I͘ is the average intensity for multiple measurements. c R factor ϭ ⌺ʈF obs ͉ Ϫ ͉F cal ʈ/͉F obs ͉. d R free was monitored with 10% of the reflection data excluded from the refinement.
Analysis of the stereochemistry with PROCHECK (43) showed that all residues except for Val-336 fall within the most favorable and additionally allowed region of the Ramachandran plot for all structures. On the basis of electron density maps, it was confirmed that Val-336 of subunit 1 in hBCATc-ox⅐gabapentin and hBCATc-ox⅐4MeVA and that of both subunits in hBCATc⅐gabapentin are in the disallowed region. Structure diagrams were drawn with the programs MOLSCRIPT (44), BOBSCRIPT (45), and PyMOL (46).
The final model of the hBCATm⅐gabapentin comprises 363 residues (3-365) for subunit 1, 363 residues (3-365) for subunit 2, two PLPs, 1 acetate, two gabapentins, and 341 water molecules. Residues 50, 105, 315, and 316 in subunit 1 have two possible conformers, each with 0.5 occupancy. Residues 105 and 315 in subunit 2 have two possible conformers each with 0.5 occupancy. The average thermal factor of the main-chain atoms is 22.6 Å 2 . Analysis of the stereochemistry done with PROCHECK (43) showed that 99.7% of the main chain atoms fall within the core and generously allowed regions of the Ramachandran plot. Gln-316 has good electron density and is the only residue seen in the disallowed region of the Ramachandran map.
Pre-steady State Kinetic Analysis of hBCAT Isozymes with Leucine and Gabapentin-The gabapentin inhibitory concentration of leucine binding was determined using stopped-flow spectroscopic analysis. The apparent rate constants (k app ) were obtained from several fixed concentrations of gabapentin and varied concentrations of leucine. The values were fit to the following equation.
K i is the inhibitory concentration of gabapentin, K d is the half-reaction dissociation constant of leucine, and k cat is the half-reaction rate constant in the reaction of hBCAT isozymes with leucine.

RESULTS AND DISCUSSION
Comparison of the Overall and Subunit Structure of the BCAT Isozymes-As shown by the amino acid sequence alignment presented in Fig. 1, the human BCAT isozymes are 58% identical in amino acid sequence. The key active-site residues involved in substrate and cofactor binding are identical in the hBCATc and hBCATm structures (see Fig. 1). Thus, observed isozyme differences in kinetic properties (8,19) and gabapentin inhibition (15) cannot be explained by mutation of active-site residues. It is more likely that the isozyme differences result from changes in the primary amino acid sequence of residues surrounding the active-site residues. Biophysical studies of the two proteins indicated that there are conformation differences in the aromatic residues in the BCAT isozyme active sites and suggested that the structure of hBCATc is more flexible than the structure of hBCATm (19).
The activity of mammalian BCAT isozymes is affected by the oxidation state of the CXXC center, with the activity of hBCATm showing greater dependence on redox state of this center than the activity of hBCATc. Oxidation of the hBCATm CXXC center to the disulfide form results in loss of enzyme activity (34). Crystals of hBCATc were formed initially only in the presence of 3PP without added DTT; hence, the crystals that formed are the oxidized form of hBCATc, with a disulfide bond between Cys-335 and Cys-338. The Michaelis complexes of hBCATc with 10 mM gabapentin and 10 mM 4MeVA (leucine analog) were obtained by soaking the oxidized hBCATc-ox⅐3PP crystals with each individual compound. By soaking the hBCATc-ox⅐3PP crystals with gabapentin in the presence of DTT, it was possible to obtain the structure of the fully reduced enzyme in a Michaelis complex with gabapentin. The resolution of the structure calculated from the hBCATc-ox⅐gabapentin crystals was higher (1.9 Å) than the resolution of the reduced hBCATc⅐gabapentin structure (2.4 Å). Using the crystallization conditions described previously (25) with 70 mM gabapentin, a high FIGURE 1. Amino acid sequence alignment of hBCATc and hBCATm. The consensus amino acids are indicated by asterisks below the amino acid residues. The secondary structure residues are indicated by lines (helix) and arrows (sheet) above the amino acid residues. The flexible interdomain loop residues are indicated by thin dotted and dashed lines above the amino acid residues. The PLP binding active-site lysine is shown by a vertical arrow. The active-site residues are indicated by boxes, and CXXC cysteines are in bold italic. The alignment was performed using ClustalW software. resolution structure (1.8 Å) of the reduced hBCATm⅐gabapentin Michaelis complex was obtained.
Comparison of the oxidized hBCATc, reduced hBCATc, and hBCATm structures with each other and the other hBCATm structures (25,26) and eBCAT dimer structures (24) shows that the overall structures of these enzymes are similar. hBCATc is a homodimer, and the C␣ atoms of the two subunits are related by a non-crystallographic 2-fold axis. The homodimer and subunit structures of hBCATc-ox⅐gabapentin are shown with secondary structure assignments by the program DSSP in Fig. 2, A and B (47). The surface area of the subunit interface was calculated to be 2,428 Å 2 , which amounts to about 16% of the subunit surface area (15,405 Å 2 ). The total surface area is comparable with 16,049 Å 2 reported for hBCATm (25,26). When the C␣ carbon atoms (except those for the interdomain loops) of the hBCATc structures are superimposed between subunits in the different complexes, the root mean square deviation (r.m.s.d.) is 0.37 Å. When the C␣ carbon atoms of the dimeric molecules are fitted between hBCATc complexes, the r.m.s.d. is 0.21 Å. The superimposition of hBCATc⅐gabapentin onto hBCATm⅐gabapentin resulted in equivalent C␣ atoms with an r.m.s.d. of 0.85 Å, indicating that the overall structure of hBCATc is quite similar to that of hBCATm. At this level the subunit and overall structures of hBCATc and hBCATm are essentially the same as those of the other fold-type IV class of PLP enzymes (20,21,27).
The hBCATc subunit consists of a small domain (N-terminal to Ser-188), an interdomain loop (Pro-189 -Pro-201), and a large domain (Val-202 to C terminus). The small and large domains are folded into an open ␣/␤ structure and a pseudo barrel structure, respectively. The large cavity is formed between the small and large domains of one subunit, and two loops from the small domain of the other subunit approach the cavity (Fig. 2). The active site is formed at the domain interface and at the subunit interface. As with other fold-type IV enzymes (20 -27) the cofactor PLP is located at the bottom of the active site with its re-face toward the protein side, forming a covalent Schiff base with the side chain of the catalytic Lys-222. Gabapentin resides on the solvent side of the PLP-Lys-222 Schiff base. The long loop connecting ␤-strands S1 and S2, which runs on the surface of the small domain, reaches the rim of the active site and overhangs the cavity. The overhang, which is not present in eBCAT, is characteristic of hBCATc and hBCATm (25,26).
Active Site of hBCATc-ox⅐gabapentin-The simulated annealing omit map showed that gabapentin was bound to the active site in the place of 3PP. The structure and the hydrogen-bonding scheme of the active site are shown in Figs. 3A and 4A, respectively. The PLP cofactor forms a Schiff-base bond with the catalytic Lys-222. The O3Ј atom and the phosphate group of PLP are on the side of Tyr-227 and on the side of the ␤-turn formed at Thr-333 and Ala-334, respectively. The ␤-turn is followed by the redox-sensitive CXXC (Cys-335-Val-336 -Val-337-Cys-338) motif. The pyridine ring of PLP is sandwiched by Leu-286 and Thr-260 from the re-and si-faces, respectively. Glu-257 forms a salt bridge with the protonated N1 atom of PLP to strengthen the electronwithdrawing effect of the pyridine ring of PLP as an electron sink, as has been observed for the enzymes of fold-types I and IV (48 -51). To finetune the electronic state of the cofactor, the OH group of Tyr-227 is hydrogen-bonded to O3Ј. The C4Ј ϭ N Schiff bond is roughly coplanar with the pyridine ring of PLP with the dihedral angle of C3-C4-C4Ј-N of Ϫ34°, indicating that the Schiff base is protonated and that an intramolecular N4 ϩ -H-O3Ј Ϫ hydrogen bond is formed. As a result, the PLP is in the ketoenamine form (19). The OH group of Tyr-90*, the guadinino group of Arg-163, the hydroxyl group Tyr-161, water molecules (W1 and W2), the main-chain CAO of Gly-97, the carboxylate of gabapentin, and the phosphate O1 atom of PLP form a hydrogen bond network and form a semicircle that encloses gabapentin on the side of the si-face of the PLP plane. The water molecule, W3, bridges Lys-99 and Gly-191 of the interdomain loop and Ala-334 of the ␤-turn by hydrogen bonds that align the side chain of Lys-99 and the interdomain loop in positions suitable for enzyme function.
Except for one residue (Val-336 in hBCATc is Gln in hBCATm), the residues forming the substrate binding pocket in hBCATm are conserved in hBCATc (25,26). The template cavity of hBCATc formed at the si-face of PLP is surrounded by Phe-49, Phe-95, Tyr-161, Tyr-227, Thr-260, Thr-333, Ala-334, Tyr-90*, Leu-173*, and Val-175* and is assumed to consist of large and small sites with Phe-49, Tyr-161, and Thr-260 located at the boundary region between the large and small sites (see Fig. 5). Gabapentin, except for the aminomethyl group, nicely fits the cavity with the cyclohexane ring and the ␣-carboxylate coordinated to the large and small sites, respectively. The large site has a round shape suitable to accommodate the cyclohexane ring of gabapentin, is enclosed in the active-site cavity with its carboxylate and aminomethyl group interacting with the ␤-turn formed at Thr-333 and Ala-334 preceding the CXXC motif and water molecule (W1) and Thr-260 on the loop between ␤-strands S10 and S11, respectively. B, close-up view of the active site of the hBCATc-ox⅐4MeVA complex. The active-site structure except for Thr-260 is quite similar to that of hBCATc-ox⅐gabapentin. 4MeVA (pink) bound to the active-site cavity is approached by Tyr-193 of the interdomain loop to be shielded in the protein inside. One of the carboxylate oxygen atoms is recognized by the ␤-turn and the W1 hydrogen bonded to Tyr-161 like the carboxylate of gabapentin, whereas the other oxygen interacts with Thr-260 like the aminomethyl group of gabapentin. The diagram was drawn with the program BOBSCRIPT (45).
which is bulkier than the side chain of the substrate leucine. The cyclohexane ring forms van der Waals contacts with Phe-49, Phe-95, Tyr-161, Tyr-227, Thr-260, Tyr-90*, Leu-173*, and Val-175*. In addition to these residues, Tyr-193 of the interdomain loop approaches the cyclohexane ring (Fig. 3A). The small site is occupied by the carboxylate group of gabapentin, which forms hydrogen bonds with the main-chain NH groups of Thr-333 and Ala-334 at the ␤-turn and with the water molecule W1. The side chain of Lys-99 approaches the main-chain carbonyl groups of the ␤-turn and interacts with them to polarize the peptide bonds leading to the enhancement of the hydrogen-bond interaction between the carboxylate of gabapentin and the main-chain NH groups of the ␤-turn (Fig. 3A). The carboxylate is, thus, recognized by the ␤-turn at Thr-333 and Ala-334 with the aid of Lys-99. Similar to Lys-99 in hBCATc, Lys-79 in the hBCATm complexes interacts with the main chain carbonyl group of the ␤-turn. The disposition of gabapentin, except for the aminomethyl group, in the active site is roughly the same as that of 4MeVA or 2-methylleucine in the eBCAT complex (22), where the cyclohexane ring and the carboxylate of gabapentin correspond to the hydrophobic side chain and the carboxylate of the substrate analog (4MeVA or 2-methylleucine), respectively. The aminomethyl group of gabapentin, which does not have its counterpart in the substrate analogue, is also located inside the template cavity at the boundary region of the large and small sites with its C-N bond projected onto the C4-C4Ј bond of the cofactor with respect to the cofactor plane. The amino group forms hydrogen bonds with the main-chain carbonyl group and the side-chain hydroxyl group of Thr-260. Thr-260 is, thus, the recognition site for the aminomethyl group. The side chain of Tyr-193 of the interdomain loop approaches the aminomethyl group from the solvent side. Gabapentin captured in the template cavity is almost shielded in the protein inside by the access of Tyr-193, because the accessible surface area (ASA) of gabapentin is 4 Å 2 , indicating that hBCATc-ox⅐gabapentin has a closed form of the active site.
hBCATc and hBCATm are characterized by a CXXC (Cys-335-Val-336 -Val-337-Cys-338 in hBCATc) consensus sequence located at the phosphate side of the cofactor PLP. In the crystals of the oxidized form of hBCATc, Cys-335 and Cys-338 form a disulfide bond with the S-S bond distance of 2.03 Å. Cys-335, Val-336, Val-337, and Cys-338 form the macrocycle at the N-terminal side of the ␤-turn, which acts as the recognition site for the carboxylate of gabapentin.
Active Site Structure of hBCATc⅐Gabapentin-An x-ray crystallographic study of hBCATm in the PLP form and of its reaction intermediates showed that the CXXC motif is in the reduced state with two cysteine residues forming a thiol-thiolate hydrogen bond, the distance of which is in the range of 3.17-3.46 Å (25,26). In hBCATc⅐gabapentin, the sulfur atom of Cys-335 is at a distance of 4.7 Å from the sulfur atom of Cys-338, indicating that the CXXC motif is in a reduced state with two thiol groups. When the active-site residues, the cofactor, and gabapentin are superimposed between subunits of reduced and oxidized gabapentin complexes, the average r.m.s.d. is 0.29 Å, indicating that they are nearly identical (Fig. 6).
On the other hand, in both the oxidized and reduced hBCATc gabapentin complexes, the interdomain loop (flexible loop) shields gabapen-tin from solvent. The interdomain loop of the reduced enzyme is closer to the central part of the active-site cavity than in the oxidized form (resulting in the closer access of Tyr-193 to gabapentin in the reduced form than in the oxidized form) (Fig. 6). Thus, in hBCATc⅐gabapentin, gabapentin is less accessible to solvent than in hBCATc-ox⅐gabapentin. ASA values of bound gabapentin in reduced and oxidized hBCATc are 2 and 4 Å 2 , respectively. Van der Waals interactions between Cys-335 and the main chain of the interdomain loop residues are not observed in the oxidized protein, but they are present in the reduced protein. In the reduced protein Tyr-193 can interact strongly with the cyclohexane ring of gabapentin, whereas this interaction is weaker in the oxidized hBCATc-ox⅐gabapentins structure. These structural differences suggest that substrate binding to the oxidized hBCATc enzyme would be weaker than binding to the reduced hBCATc.
Active Site Structure of hBCATc-ox⅐4MeVA-4MeVA is a substrate analog in which the ␣-amino group of L-leucine is replaced by a hydrogen atom. The 4MeVA is located on the si-face of PLP with one of the C␣ hydrogen atoms of 4MeVA directed toward C4Ј of the Schiff base in PLP (Figs. 3B and 4B). The active-site structure of 4MeVA complex is quite similar to that of the oxidized gabapentin complex because the active-site residues and the cofactor except for Thr-260 are superimposed onto those of the gabapentin complex with an r.m.s.d. of 0.21 and a 0.78 Å displacement of the C␣ atom of Thr-260. The dihedral angle of C3-C4-C4Ј-N in the PLP-Lys-222 conjugate is Ϫ30°, and an intramolecular N4 ϩ -H-O3Ј Ϫ hydrogen bond is formed similarly to gabapentin complexes. The water molecules (W1, W2, and W3) located at the active site of hBCATc-ox⅐gabapentin are also conserved in hBCATc-ox⅐4MeVA (Figs. 3, A and B, and 4, A and B).
The 4MeVA bound to the template cavity is isolated from the solvent region by the access of Tyr-193 of the interdomain loop with 1-Å 2 ASA for 4MeVA. The binding mode of 4MeVA to the active-site cavity is roughly the same as that of gabapentin except for the aminomethyl group. The isopropyl group and the ␣-carboxylate of the 4MeVA complex are located at the large and the small sites of the template cavity, respectively (Figs. 3B and 4B). However, the ␣-carboxylate of 4MeVA is significantly shifted toward Thr-260 compared with that of gabapentin. In the gabapentin complex, the ␤-turn preceding the CXXC motif recognizes both ␣-carboxylate oxygen atoms of gabapentin, and Thr-260 interacts with the aminomethyl group of gabapentin (Fig. 4A). On the other hand, in the 4MeVA complex, the ␤-turn interacts with one ␣-carboxylate oxygen atom of 4MeVA, and the side-chain OH group of Thr-260 forms a hydrogen bond with the other ␣-carboxylate oxygen atom (Fig. 4B). The flexible loop bearing Thr-260 plays an important role in the accommodation of not only 4MeVA (substrate leucine) but also gabapentin, which is bulkier than 4MeVA by the same template cavity. The access of gabapentin to the active site causes the short contacts between the aminomethyl group and Thr-260. Through this process, the loop connecting ␤-strands S10 and S11 moves to make room for the aminomethyl group, which is directly recognized by Thr-260 (Fig. 2).
Structure of hBCATm⅐Gabapentin-The CXXC motif is in a reduced state with two thiol groups because two sulfur atoms of the motif are at a distance from 3.9 to 4.8 Å. When the C␣ carbon atoms of hBCATm⅐gabapentin are superimposed onto those of hBCATm⅐ isoleucine and hBCATm⅐valine (25) (Fig. 7).
Unlike the situation for hBCATc, gabapentin is not an effective inhibitor of hBCATm (15). All the active-site residues of hBCATm except for Gln-316 (Val-336 in hBCATc) and the interdomain loop are conserved in hBCATc. The active-site cavity (the template cavity) has a similar shape to the template cavity in hBCATc, and C␣ atoms of the conserved residues are superimposed between these two complexes with an r.m.s.d. of 0.40 Å (Fig. 8). Gabapentin is captured in the hBCATm cavity in the same manner as observed in the hBCATc⅐gabapentin complex. The cyclohexane ring, the ␣-carboxylate, and the aminomethyl group of gabapentin are recognized by the large site of the cavity, the ␤-turn at Thr-313 (hBCATc Thr-333), Ala-314 (hBCATc Ala-334), and Thr-240 (hBCATc Thr-260), respectively. Unlike in hBCATc⅐gabapentin complex, where gabapentin is shielded from solvent, the ASA of gabapentin in the hBCATm complex is 24 -25 Å 2 , indicating gabapentin is not shielded from the solvent and hBCATm is in an open conformation. hBCATm is also found in an open form in the hBCATm complexes with its amino acid substrates isoleucine or valine (25).
Pre-steady State Kinetics of hBCAT Isozyme Inhibition by Gabapentin-Pre-steady state inhibitory constants (K i ) of gabapentin binding in the active site of both reduced and oxidized forms of hBCAT isozymes were determined and compared with the pre-steady state half-reaction dissociation constants for leucine (TABLE TWO). With reduced hBCATm, 65.4 mM gabapentin was required to inhibit the binding of leucine to the enzyme; however, only 1.3 mM gabapentin was required to inhibit binding of the same concentration of leucine to reduced hBCATc. This result is consistent with the steady state kinetic data, which showed that gabapentin is a competitive inhibitor of leucine (15). The effect of oxidation of the CXXC center on leucine and gabapentin binding was determined for both isozymes. Although our recent results show that oxidized hBCATm is inactive, the pre-steady state dissociation constants shown in TABLE TWO indicate that the first half-reaction is less efficient but not inhibited in the oxidized enzymes. Interest-ingly, oxidization of the CXXC center has less effect on the first halfreaction leucine kinetics with hBCATc than with hBCATm, 3.4-and 9-fold increase in leucine K d values, respectively. A similar pattern was observed when comparing gabapentin inhibition of the reduced and oxidized isozymes.
Implications for Gabapentin Inhibition and Redox State-dependent Activity-The question then is the following. With the close similarity of the active-site structures of hBCATm and hBCATc, why is gabapentin an effective competitive inhibitor of hBCATc but an ineffective inhibitor of hBCATm (15) (TABLE TWO)? It is likely that differences in the specificity of gabapentin for hBCATm and hBCATc result from the behavior of the isozyme interdomain loops and the relative orientation between the small and large domains (Fig. 8). The hBCATc⅐gabapentin interdomain loop residue Tyr-193 approaches gabapentin and interacts directly with it, and gabapentin is almost shielded from the solvent, with  gabapentin ASA values of ϳ2 Å 2 . On the other hand, the corresponding tyrosine in hBCATm is too far from gabapentin to interact with it, and gabapentin is solvent-accessible with ASA values of 24 -25 Å 2 . Similarly, the interaction of Tyr-193 with gabapentin significantly reduces ASA values of the Tyr-193 side chain. The side-chain ASA of Tyr-193 in hBCATc is 19 -20 Å 2 , whereas that in hBCATm is 72-77 Å 2 . The specificity of hBCATc for gabapentin could be ascribed in part to the burial of hydrophobic surfaces of the gabapentin and the phenyl ring of Tyr-193 and the van der Waals interactions between Tyr-193 and gabapentin. As described above, there are also differences in the orientation of the small domains relative to the large domains in the two isozymes. Both domains of hBCATc move to close the active-site cleft formed between the two domains, which can easily accommodate the bulky gabapentin leucine analog, resulting in a good fit for gabapentin in the active-site cavity of hBCATc. The relative orientation between the large and the small domains in hBCATm may result in weaker binding of gabapentin due to the more open conformation of the active site. In this case the interdomain loop does not shield gabapentin, and its position prevents an interaction between Tyr-173 and the hydrophobic side chain of gabapentin (Fig. 8).
In conclusion, this study presents the first detailed structural information on the hBCATc isozyme and the structural basis for the inhibition of this isozyme by the neuroactive drug gabapentin and the first structure of the oxidized form of a BCAT isozyme. Differences in alignment of the small and large domains, side chain binding pockets, and PLP binding loop residues of these two enzymes can explain the differences in gabapentin binding to the two BCAT isozymes. The effect of oxidation on the BCAT isozyme leucine and gabapentin pre-steady state half-reaction kinetic constants shows a larger effect of oxidation on hBCATm than hBCATc. A further detailed kinetic analysis is in progress to determine how structural changes in the CXXC center regulate BCAT isozyme activity.