The Molecular Structure of Hyperthermostable Aromatic Aminotransferase with Novel Substrate Specificity from Pyrococcus horikoshii *

Aromatic amino acid aminotransferase (ArAT Ph ), which has a melting temperature of 120 °C, is one of the most thermostable aminotransferases yet to be discov-ered. The crystal structure of this aminotransferase from the hyperthermophilic archaeon Pyrococcus horikoshii was determined to a resolution of 2.1 Å. ArAT Ph has a homodimer structure in which each subunit is composed of two domains, in a manner similar to other well characterized aminotransferases. By the least square fit after superposing on a mesophilic ArAT, the ArAT Ph molecule exhibits a large deviation of the main chain coordinates, three shortened a -helices, an elongated loop connecting two domains, and a long loop transformed from an a -helix, which are all factors that are likely to contribute to its hyperthermostability. The pyridine ring of the cofactor pyridoxal 5 * -phosphate co-valently binding to Lys 233 is stacked parallel to F121 on one side and interacts with the geminal dimethyl-CH/ p groups of Val 201 on the other side. This tight stacking against the pyridine ring probably contributes to the hyperthermostability of ArAT Ph . Compared with other ArATs, ArAT ( Taq DyeDeoxy Terminator Cycle Sequencing Kit, Perkin-Elmer). Overexpression and Purification of Recombinant Protein— ArAT Ph Alignment and Phylogenetic Tree— We performed a se- quence alignment of 11 aminotransferases within 1 g using on a

Aminotransferases have been widely applied in the large scale biosynthesis of unnatural amino acids, which are in increasing demand by the pharmaceutical industry for peptidomimic and other single-enantiomer drugs (1). These enzymes have been classified into four families (I-IV) (2). Family I includes the aspartate aminotransferases (AspATs), 1 aromatic amino acid aminotransferases (ArATs), alanine ATs, and histidinol phosphate aminotransferase. All members of Family I efficiently utilize ␣-ketoglutarate as an amino donor and glutamate as an amino acceptor. Eleven residues are invariant among the enzymes belonging to Family I (2). The members of Family I are further subdivided into three subfamilies according to their amino acid sequence alignment (2,3). Subfamily I␣ comprises AspATs isolated from Escherichia coli, yeast, chicken, pig, and other organisms, and ArATs from prokaryotes (E. coli and Paracoccus denitrificans). In this subfamily, an arginine residue (Arg 292 *, according to the numbering for pig cytosolic AspAT (cAspATp) (4)) 2 is conserved. The Arg 292 * residue interacts with the -carboxyl moiety of the dicarboxylic substrates (5)(6)(7). The arginine residue was not found in all members of subfamily I␥, despite the normally high degree of conservation in active site residues (2,8,9). Subfamily I␤ is specialized for histidine biosynthesis (2,4).
Recently, much research effort has been directed toward the isolation and characterization of enzymes from hyperthermophilic archaea. Interest in these enzymes has increased, because of their biotechnological potentials for novel application (10,11) and because of the need for a better understanding of their intrinsic resistance to heat and denaturing processes. The mechanisms of their stability continue to be challenging and unresolved problems in biochemistry and biotechnology (10 -13). An aspartate aminotransferase gene homolog (open reading frame identification number 1371) was identified via genome sequencing in the hyperthermophilic archaeon Pyrococcus horikoshii (14,15). The gene (ArATPh) was expressed in E. coli, the product was purified to homogeneity, and the enzyme ArATPh was determined to be an aromatic aminotransferase belonging to subfamily I␥. We present the first report of the molecular structure of hyperthermophilic ArAT, which is an essential step in the effort to comprehend its stabilizing mechanisms. We also discuss its novel substrate specificity and dual substrate binding mechanism for both acidic and aromatic amino acids on the basis of its active site structure.
* 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.
The atomic coordinates and structure factors (

MATERIALS AND METHODS
Chemicals-The pET-11a vector and ultracompetent E. coli XL2-Blue MRFЈ cell were purchased from Stratagene (La Jolla, CA). The pET-15b vector and E. coli strain BL21 (DE3) were obtained from Novagen (Madison, WI). Vent DNA polymerase was purchased from New England Biolabs (Beverly, MA). Restriction enzymes were purchased from Promega and Toyobo (Osaka, Japan) and were used according to the manufacturer's recommendations. Ultrapure dNTP solution was obtained from Amersham Pharmacia Biotech. L-Cysteinsulfinic acid, bovine DNase I, ␤-NADH, and malate dehydrogenase from porcine heart (mitochondrial) were purchased from Sigma. 2-Oxoglutaric acid monosodium salt and DTNB were purchased from Nacalai Tesque (Kyoto, Japan). Isopropyl-␤-D-thiogalactopyranoside was purchased from Takara Shuzo (Otsu, Shiga, Japan).
Cloning of Genes and Construction of Expression Vector-The complete genome sequence of P. horikoshii has been reported by Kawarabayasi et al. (14,15). Standard cloning techniques were used throughout. The aromatic aminotransferase (ArATPh) gene was amplified using polymerase chain reaction with primers having NdeI and BamHI restriction sites according to a method reported previously (16). The sequences of the primers were 5Ј-TTTTGTCGACTTACATATGGCGCTA-AGTGACAGA-3Ј (underlining indicates the upper primer containing the NdeI site) and 5Ј-TTTTGGTACCTTTGGATCCTTAACCAAGGATT-TAAACTAG-3Ј (underlining indicates the lower primer containing the BamHI site). The amplified gene was digested by NdeI and BamHI, and the digested fragment coding for ArATPh was inserted in an expression vector pET-11a cut with the same restriction enzymes. The nucleotide sequence of the inserted gene was verified by sequencing on an Applied Biosystems 373A DNA sequencer (Taq DyeDeoxy Terminator Cycle Sequencing Kit, Perkin-Elmer).
Overexpression and Purification of Recombinant Protein-The cloned gene was expressed using the pET-11a vector system in the host E. coli strain BL21 (DE3) according to the manufacturer's instructions. The host cells were transformed with the constructed pET-11a/ArATPh plasmid, after which the production of the protein was performed according to the method described previously (16). The concentration of the expressed protein was determined using a Coomassie protein assay reagent (Pierce) and utilizing bovine serum albumin as the standard protein. The crude enzyme solution was prepared from the transformant E. coli, and the enzyme was purified using chromatography in a HiTrap Q column (Amersham Pharmacia Biotech) and a HiLoad Superdex 200 column (Amersham Pharmacia Biotech) (16). The purity of the enzyme samples was analyzed using SDS-polyacrylamide gel electrophoresis (17) and isoelectric focusing using a PhastSystem (Amersham Pharmacia Biotech). Protein sequencing of recombinant ArATPh was performed by Takara Shuzo Co. Ltd. (Otsu, Shiga, Japan) using a protein sequencer PSQ-1 (Shimazu, Japan).
Pre-steady-state Kinetic Studies of Half-transamination Reactions-Aliphatic amino acid substrates with unbranched side chains were used to estimate the hydrophobic substrate specificities of the aminotransferase. We used L-form amino acids for the sC3-sC6 substrates and DL-isoforms for the sC7-sC9 substrates (18); the aminotransferase tested here cannot use D-form amino acids as substrates. All measurements were carried out at pH 8.0 and 25°C. The buffer solution contained 50 mM HEPES with 100 mM KCl and 10 mM EDTA.
The slow reaction was followed spectrophotometrically by monitoring the change in absorption of the bound coenzyme at 380 nm. When the k app value was directly proportional to the substrate concentration, the k max /K d value was calculated from the following equation (18,19).
Thus, the catalytic efficiency, k max /K d , was given by k app /[S] for these substrates. The rapid reactions were followed using stopped flow spectrophotometers from Union Giken (model RA-401) or Applied Photophysics (model SX-17MW). The reaction curves conformed to a singleexponential process. The free energy differences (⌬G T ) between the transition state and unbound enzyme plus substrate for various substrates were calculated using the following equation (19,20).
where R is the gas constant, T is the absolute temperature, k B is the Boltzmann constant, and h is the Planck constant. Temperature Dependence of Activity for Overall Transamination Reaction-The overall transamination reaction for the acidic substrate aspartate was measured spectrophotometrically at 340 nm using a coupled assay with malate dehydrogenase and NADH at pH 8.0 and 25°C (20,21), and the steady-state kinetics parameters, K m and k cat , were determined. The reaction mixtures contained 50 mM HEPES, 100 mM KCl, 0.01 mM EDTA, 0.1 mM NADH, 2.5 units/ml malate dehydrogenase, 1 M ArATPh, and various concentrations of L-aspartate or 2-oxoglutarate. The activity of the hydrophobic substrate phenylalanine was measured at pH 8.0 and 25°C. The product formation of the phenylpyruvate was monitored at 280 nm using the molar extinction coefficient difference of 450 cm Ϫ1 M Ϫ1 between phenylpyruvate and phenylalanine (22). The reaction mixture contained 50 mM HEPES, 100 mM KCl, 20 nM ArATPh, and various concentrations of L-phenylalanine or 2-oxoglutarate.
To determine temperature dependence, the activities for five sets of substrates, tryptophan-2-ketoglutarate (Trp-2OG), tryptophan-phenylpyruvate (Trp-KetoPhe), histidine-2-ketoglutarate (His-2OG), histidine-phenylpyruvate (His-KetoPhe), and glutamate-phenylpyruvate (Glu-KetoPhe) were measured at different temperatures (25-90°C) at pH 8.0. Based on the following molar extinction coefficients of 2-oxo acid derivatives, the concentration of the enzymatic products were calculated: at 310 nm, 24.5 and 3200 cm Ϫ1 M Ϫ1 for 2OG and 3-indolepyruvate, respectively, and at 280 nm, 21 and 450 cm Ϫ1 M Ϫ1 for 2OG and KetoPhe (22), respectively. The product from His was monitored at 293 nm using a coefficient difference of 3050 cm Ϫ1 M Ϫ1 obtained to subtract the spectrum for the reaction of histidine with the PLP enzyme from that of the pyridoxamine 5Ј-phosphate enzyme (23). The reaction mixture contained 50 mM HEPES, 100 mM KCl, 20 nM ArATPh, and various concentrations of amino acids or keto-acids.
Optimum Temperature and Thermostability for ArATPh Reaction-The optimum temperature for ArATPh activity was measured as described previously (24). The enzyme reaction was carried out in a solution (3.05 ml) containing ArATPh (3.8 nM), L-cysteinsulfinic acid (12.8 mM), 2-oxoglutaric acid (2.0 mM), EDTA (98 M), and DTNB (1.5 mM) in 50 mM phosphate buffer (pH 6.5) at 30 -98°C, and the rate of increase in absorbance at 412 nm because of the reduction of DTNB was monitored for 5 min. For controls, the reactions were performed under the same conditions but without the enzyme.
To determine thermostability, the enzyme solutions (0.1 mg/ml) in 20 mM phosphate buffer (pH 6.5) were incubated at 95°C for 90 and 120 min and then autoclaved in sealed Eppendorf tubes at 110°C for 5, 15, 30, and 90 min. The heated enzymes were assayed in duplicate at 90°C, as described elsewhere (24).
Spectroscopy of Coenzyme-To investigate the ionization of the internal Schiff base, the absorption spectra of the enzyme at a protein concentration of approximately 20 M in a 1-cm cell were measured at 25°C using a Hitachi spectrophotometer (model U-3000). The buffer solution was comprised of 100 mM KCl, 0.01 mM EDTA, and a buffer component of 50 mM MES, 50 mM PIPES, or 50 mM HEPES.
pH Stability-The gross conformation and pH stability of ArATPh were studied using CD spectroscopy. The CD spectra of ArATPh, at a protein concentration of approximately 0.1 mg/ml in a 1-mm cell, were measured at 25°C using a spectropolarimeter (J-720W, Jasco, Japan). The solution was comprised of 100 mM KCl and a buffer component of 50 mM acetate, 20 mM phosphate, 50 mM borate, or 20 mM carbonate.
Scanning Calorimetry-The thermal denaturation curve of ArATPh was measured using a Nano Differential Scanning Calorimeter (CSC5100, Calorimetry Science Co.). Before measurement, the enzyme solution (1 mg/ml) was dialyzed against 20 mM phosphate buffer, pH 6.5, and degassed for 15 min using an aspirator. The sample cell was filled with the degassed enzyme solution, and the reference cell was filled with the outer solution from the dialysis. The measurement was performed at a temperature range of 0 -125°C. A scan rate of 1 K/min was used throughout. The denaturation profile was analyzed using Nano differential scanning calorimetry CpCalc data analysis software (Calorimetry Science Co.).
Structure Determination, Refinement, and Model Building-Crystals were obtained using the hanging drop vapor diffusion technique. An equi-volume of 3 M 1,6-hexane-di-ol solution at pH 7.5 (100 mM HEPES buffer) containing 10 mM MgCl 2 was added to a protein solution containing 1.6% ArATPh and 20 M pyridoxal-5Ј-phosphate and a 10-l droplet of the solution was equilibrated with 1 ml of a 3 M 1,6-hexanedi-ol solution. Crystals were grown at room temperature for 1 week. X-ray diffraction experiments were carried out on an Enraf FAST differactometer equipped with a FR571 generator (40 kV, 50 mA; focal spot size, 0.2 mm), and intensity data were collected at a resolution of 2.1 Å for the native crystal and at 3.0 Å for the heavy atom derivatives.
The structure was determined using the multiple isomorphous replacement method. A structure model was built on an electron density map calculated with multiple isomorphous replacement phases with a figure of merit of 0.93. The amino acid sequence was unambiguously traced on the map and most of the side chains were identified. The structure was refined to a resolution of 2.1 Å using X-PLOR (25). All coordinates have been deposited with the RCSB Protein Data Bank as entry 1DJU. In the substrate-binding model, the coordinates of ArATPh with water molecules were fixed, but the torsion angles of the substrates were changed to find the best fitting configuration to the enzyme.

RESULTS
Sequence Alignment and Phylogenetic Tree of ArATPh-Because we were unable to construct a united alignment among aminotransferases belonging to the subfamilies 1␣, 1␤, and 1␥ from P. horikoshii, other archaea, bacteria, and eukaryotes because of a lack of similarity between each subfamily, we made the alignment using 11 candidates belonging to subfamily 1␥ to identify its conserved residues. The best alignment was obtained with the five thermophilic aminotransferases shown in Fig. 1. ArATPh showed poor identity to E. coli AspAT (AspATEc) (38), E. coli ArAT (ArATEc) (39,40), and P. denitrificans ArAT (ArATPd) (41), which are well known members of subfamily I␣. According to these results, ArATPh was nominated to the aminotransferase subfamily I␥ (2, 3). Furthermore, ArATPh was closer to the thermophilic AspATs than to the tyrosine ATs from eukaryotes in subfamily I␥.
Overexpression, Purification, and Oligomeric Structure of Recombinant ArATPh-The ArATPh gene was abundantly expressed in E. coli BL21 (DE3), and the recombinant ArATPh comprised 30% of the total protein. After heat treatment at 80°C for 15 min, which removed most of the endogenous E. coli proteins, the protein was purified to homogeneity by sequential chromatography on HiTrap Q and HiLoad Superdex 200 columns. The final preparation of the ArATPh displayed a single band (42 kDa) on SDS-polyacrylamide gel electrophoresis. Isoelectric focusing indicated a pI value of 5.2 for ArATPh. The N-terminal sequence of the recombinant ArATPh was ALS-DRLELVSASEIRKL, which was identical to that deduced from the DNA sequence without the initial methionine residue. The enzyme had an apparent molecular mass of 56 kDa as estimated by gel filtration on a calibrated TSK gel G2000SWXL column, and a subunit molecular mass of 42 kDa as estimated by SDS-polyacrylamide gel electrophoresis. This suggests that it has a dimeric structure similar to other aminotransferases (24,34,(42)(43)(44)(45).
Optimum Temperature of the Recombinant ArATPh-The optimum temperature of this enzyme was 90°C, which represents an extreme thermophilic characteristic. The k app of ArATPh increased steadily in the range of the temperature studied here. The k app for L-cysteinsulfinic acid and 2-ketoglutaric acid as substrates was 1.39 ϫ 10 2 s Ϫ1 at 90°C and pH 6.5. Like several other thermophilic enzymes (46 -48), the recombinant ArATPh shows a thermal transition in conformation as indicated in Arrhenius plots near 70°C (data not shown).
Substrate Specificity-⌬G T , the free energy difference between the unbound enzyme plus substrate (E ϩ S) and the transition state (ES ), was calculated for various substrates using Equation 1 or 2. A smaller ⌬G T value indicates higher enzyme activity. As shown in Table I, Tyr is the best substrate having a k max /K d (M Ϫ1 s Ϫ1 ) value of 1.2 ϫ 10 5 . Three aromatic amino acids (Tyr, Phe, and Trp) and Glu are good substrates for ArATPh, whereas Asp is a poor substrate having a k max /K d (M Ϫ1 s Ϫ1 ) value of 9.1. ArATPh showed moderate activity on His. The activity of ArATPh toward a series of aliphatic substrates with straight side chains was enhanced as the side chain length increased. The activity of ArATPh was maximal for an 8-carbon substrate (2-amino octanoic acid).
Comparison of the Substrate Specificity for ArATs from Different Origins-The steady-state kinetic parameters of ArATPh using an overall transamination reaction between 2OG and Asp or between 2OG and Phe were measured at 25°C (the upper section of Table II). The enzyme activity against Phe was high with a k cat /K m value of 5.2 ϫ 10 3 M Ϫ1 s Ϫ1 , but the activity against Asp was very low (2.2 M Ϫ1 s Ϫ1 ). The difference between these k cat /K m values was on the order of 10 3 , whereas the difference for ArATEc (from E. coli) and ArATPd (from P. denitrificans) was approximately 10-fold. At the optimum reaction temperature of 90°C, ArATPh has approximately a 10 2 - and 10-fold higher activity for His and Glu, respectively, than does ArATEc or ArATPd at 25°C (Table II). This indicates that Glu is the best acidic substrate for ArATPh.
Spectroscopic Properties of the Enzyme-bound Coenzymes PLP and Pyridoxamine 5Ј-Phosphate-An internal Schiff base is formed between Lys 233 (corresponding to Lys 258 in cAspATp) and the aldehyde group of the coenzyme PLP. The reaction produces a spectral change in the visible absorption region. Changes in the apparent molar absorption coefficients for the PLP form enzyme at 420 and 370 nm were plotted against pH. The Schiff base pK a value was determined to be 5.1.
Three-dimensional Structure of ArATPh-The space group of the protein crystal is P2 1 2 1 2 1 , and the cell dimensions are a ϭ 64.01, b ϭ 124.87, and c ϭ 128.78 Å. The structure was solved using the multiple isomorphous replacement method at a resolution of 3.0 Å using four heavy atom derivatives: K 2 PtCl 4 , methyl mercury chloride, p-chloromercuribenzenesulfonic acid, and mersalyl acid. These data are presented in Table III. The structure was refined at 2.1 Å resolution to the R value of 0.185 and R free of 0.254, respectively. The root mean square devia-tions of bond distances and angles from their ideal values were 0.017 Å and 3.24°, respectively. The (, ) values for all the amino acid residues except Thr 264 fell in a normal region in the Ramachandran plot (data not shown). The crystal having a V m ϭ 2.9 Å 3 /Da contains two molecules related with local 2-fold symmetry in an asymmetric unit. ArATPh has a dimer structure (Fig. 2). One dimer molecule has two active sites, and each active site binds one PLP. In both subunits, the N-terminal region of residues 2-11 form a short ␣-helix, but region 12-26 is missing in the final structure model because no significant electron density was observed in the 2F o Ϫ F c and F o Ϫ F c maps for the region (Fig. 3). The molecule consists of two domains. The large domain has a ␣/␤ structure comprised of six ␣-helices (H3-H8) and seven ␤-strands (S1-S7), as assigned by the program DSSP (49). The strands form a twisted sheet structure, on both sides of which helices are arranged. The small domain consists of three ␣-helices (H10 -H12) and a ␤-strand (S8). A long ␣-helix (H10) links to the large domain via an ␣-helix (H9).
The molecular replacement method using the structures AspATEc, ArATPd, and cAspATp as templates (6, 7, 50) was not successful for solving the ArATPh structure because of poor identity of the primary sequences and a large deviation in the main chain coordinates. By the least square fit after superposing ArATPh on ArATPd, only 295 pairs of corresponding amino acid residues were present in the C␣-C␣ distance less than 3 Å, and their root mean square deviation was 2.0 Å. Several structural differences were observed between ArATPh and ArATPd as shown in Fig. 4. The 5th, 11th, and 13th ␣-helices of ArATPd are shorten by several amino acid residues in ArATPh, corresponding to H4, H8, and H10, respectively. Interestingly, the Gross Conformation and pH Stability-The gross conformation and pH stability of ArATPh were studied using CD spectroscopy at 25°C. The CD spectrum in the region between 200 and 250 nm exhibited double negative minima at 209 and 223, which are characteristic of an ␣-helical structure (data not shown). The ␣-helical content is estimated to be approximately 40%, according to the method of Chen et al. (51). The enzyme is stable between pH 4 and 11 for 24 h at 25°C.
Heat Stability-The residual ArATPh activity remaining after heating was measured to determine the half-life of the enzyme at 95 and 110°C. The half-life of ArATPh is 30 min at 110°C, and the enzyme is stable at 95°C in 20 mM phosphate buffer (pH 6.5).   The heat capacity change of ArATPh was measured using differential scanning calorimetry from 0 to 125°C at pH 6.5. The heat capacity change was only observed during the first scan in the differential scanning calorimetry measurement, indicating that the heat denaturation profile of ArATPh is due to an irreversible denaturation process. The profile showed one major peak at 120.1°C. The molar enthalpy change, ⌬H, was calculated to be 2.4 ϫ 10 3 kJ/mol for the homodimer.

Structural Elements Providing Hyperthermostability on
ArATPh-ArATPh is one of the most thermostable aminotransferases ever to be purified (52), having a melting temperature of 120°C. ArATPh is a homodimer in which each subunit is constituted of two domains, similar to other well characterized aminotransferases, such as AspATEc, ArATPd, and cAspATp (6,7,50). However, its structure could not be solved by the molecular replacement method because of a poor sequence similarity, large deviation in the main chain coordinates, and local changes in the secondary structure, including three shortened ␣-helices and a long loop transformed from an ␣-helix (Fig. 4). Another unique characteristic is its elongated loop between H11 and S8 (Figs. 3 and 4). The loop intimately connects two domains and covers one end of the cleft formed at the interface of the two domains. Because numerous hydrophobic residues were observed inside the cleft, the elongated loop, which closely binds the two domains and acts like a lid to shield the cleft from solvents, might be one of the factors that account for the hyperthermostability of ArATPh.
On the basis of the ArATPh structures, the surface area for one amino acid residue was calculated by dividing the accessible surface area of the dimer by total residue numbers to evaluate molecular compactness. The values of ArATPh and ArATPd are 28.0 and 33.6 Å 3 , respectively. The lower surface area for one residue of ArATPh molecule might be due to tight packing of the polypeptide chain into the homodimer structure. Another prominent difference in ArATPh is the large number of charged residues (Asp, Glu, Lys, and Arg) on its molecular surface compared with the ArATPd. The occupancy of the charged residues in the accessible surface area of ArATPh and ArATPd molecules are 73.3 and 48.1%, respectively. On the contrary, the frequency of polar contacts less than 3.3 Å, including hydrogen bond and ion pair among these charged residues on the surface is decreased to 36.0% for ArATPh in comparison with the value, 51.3%, of ArATPd. The accessible surface of ArATPh has higher hydrophilicity with a lower number of ion pairs than that of ArATPd. The compact packing and the remarkably water-attractive surface of ArATPh are probably major factors contributing to its hyperthermostability.
The PLP molecule of ArATPh is fixed tightly with nine hydrogen bonds at the bottom of the active site cleft (Fig. 5A). One side of the pyridine ring of PLP interacts with the geminal dimethyl groups of Val 201 (Ala 224 in cAspATp), whereas the other side is stacked parallel with the phenyl ring of Phe 121 (Trp 140 in cAspATp). In AspATEc, the methyl group of Ala 224 interacts with the pyridine ring of PLP on one side, and on the other side, the pyridine ring stacks to Trp 140 with a 20°inclination angle. In the thermophilic enzymes of subfamily I␥, valine or isoleucine is found at the position corresponding to Val 201 of ArATPh, whereas the residue is replaced with Ala in the mesophilic enzymes of subfamily I␣ (2,3). The interaction of Ala with PLP should be weaker than those of V/I in thermophilic aminotransferases because of the lack of a geminal dimethyl-CH/ interaction (53). In subfamily I␥ of the thermophilic archaea (Fig. 1), the phenyl ring of Phe or Tyr, corresponding to Phe 121 in ArATPh, always stacks to the pyridine ring of PLP. In subfamily I␣ from the mesophilic organisms and subfamily I␥ from the thermophilic prokaryotes, these residues are replaced by tryptophan, which has a bulkier side chain with a wider surface area than does the phenyl ring (2). Consequently, a combination of the Phe 121 and Val 201 residues stacking tightly to the pyridine ring of PLP may contribute to the hyperthermophilic properties of ArATPh. Further crystallographic studies are in progress to better understand the mechanisms underlying the hyperthermostability of this enzyme.
PLP-binding Structure of ArATPh-In the PLP molecule of ArATPh, the number of hydrogen bonds fixing the phosphate  moiety is reduced from six to five (Fig. 5A), because of replacement of the Ser 255 residue, which is conserved in both AspATEc and ArATPd (6,7). The large conformational change of the phosphate moiety of PLP is induced by a shift in the side chains of Ala 96 , Ser 232 , and Arg 241 from the corresponding residues in AspATEc. The phosphate moiety moves parallel to the plane of the pyridine ring of PLP, whereas the pyridine ring is conserved at the same position as in AspATEc. The movement of the phosphate moiety in the opposite direction might be a positive adjustment of the cofactor to compensate for changes in the secondary structure that account for its hyperthermostability. Interestingly, the O position of Tyr 202 in ArATPh is almost identical to that of the corresponding Tyr 225 residue of AspATEc (Fig. 5A), whereas the coordinates of the main chain parts in both Tyr residues are shifted by more than 2 Å. The Tyr residues of both ArATPh and AspATEc are close enough to form a hydrogen bond with O 3 H of PLP. The position of the -carboxyl of Asp 199 forming a hydrogen bond with N1H of PLP is also identical to that of Asp 222 of AspATEc. These results strongly indicate that the pyridine ring must be fixed precisely at the conserved position in the active center of ArATs to make the cofactor fully active, although the phosphate moiety can be positioned according to the steric requirements. The O of the Tyr 202 residue of ArATPh can also form a hydrogen bond with the imino group of the Schiff base. The angle of the imino proton on the CϭN plane of ArATPh is sufficient to form a hydrogen bond with Tyr 202 ; however, the corresponding angle in AspATEc seems less suitable to form a hydrogen bond with Tyr 225 (Fig. 5A). The pK a of the Schiff base of ArATPh was determined to be 5.1, which is the lowest value ever reported; FIG. 3. The folding topology of the monomer of ArATPh. A, ␣-helices, ␤-strands, and loops are colored red, blue, and yellow, respectively. The green ␣-helix corresponds to the N-terminal ␣-helix followed by the disordered region. The ␣-helices and ␤-strands of the ␣/␤ structure are numbered from the N terminus. The PLP molecules are represented by a space-filling model. The figure produced using the program Turbo-Frodo. B, topology diagram of ArATPh. ␣-Helices are shown as cylinders (red), ␤-strands are arrows (green), and the numbering is the same as that in A. The sequential numbers of the first and last residues in each secondary structure element are indicated.
FIG. 5. The PLP binding structure of ArATPh. A, the stereoview for the superposition of ArATPh (red) and AspATEc (blue) by the PLP fitting. Although the overall structure of AspATEc including the PLP binding profile (6) is quite similar to that of ArATPd (7), the AspATEc was selected as a reference structure for the superposition because of its general popularity historically. The residue numbers indicate the positions in the ArATPh molecule (red). Dotted lines indicate the hydrogen bonds in ArATPh (red). B, nomenclature of atoms for PLP. the pK a values of AspATEc, T. thermophilus AspAT, and ArATEc were reported to be 6.8 (54), 6.1 (8), and 6.65 (55), respectively. This low pK a value is probably due to rotation of the CϭN plane of the Schiff base against the pyridine ring of PLP to control hydrogen bonding between the imino group and the Tyr 202 residue and may also be due to the unique environment around the PLP molecule caused by changes in the residues stacked to PLP (Fig. 5A).
Active Site Structure and Substrate Binding Models-The active site structure with the best substrate, Tyr, is shown in Fig. 6. The ␣-carboxylate of Tyr was fixed at the active site by two salt bridges with Arg 362 (corresponding to R386 in cAspATp) and three hydrogen bonds with Gly 34 , Asn 171 , and Tyr 320 . The phenyl ring of Tyr and the aromatic group of Phe 121 undergo an energetically favorable "edge-to-face" interaction (56), and the aromatic ring of Tyr 59 * is located very closely, but not in parallel, to the phenyl ring of Tyr. Thus, the best substrate can be trapped in the hydrophobic pocket formed by Phe 121 , Tyr 59 *, the pyridine ring of PLP, Met 260 * (corresponding to Arg 292 * in cAspATp), and Val 122 (Glu 141 in cAspATp). In this binding model, the OH group of Tyr is located at a distance sufficient to form hydrogen bonds with O␥1 of Thr 264 * (Ser 296 * in cAspATp) and with the phosphate moiety of PLP. The internal aldimine bond between PLP and Lys 233 (corresponding to Lys 258 in cAspATp) is located so close that a new external aldimine bond can be formed between PLP and the ␣-amino group of Tyr.
Another binding model was formed with Glu, one of the best acidic substrates. A water molecule (Xaa 126 *) is present at the center of three adjacent groups: the ␥-carboxyl of Glu, the O␥1H of Thr 264 *, and the phosphate residue of PLP. The proximity (within 3 Å) of the water molecule and the adjacent residues allows formation of a hydrogen-bond network among them. The water molecule (Xaa 126 *) may be important in binding Glu to the active center, as indicated by the reportedly complex structure of ArATPd with maleate (7). Furthermore, the ␥-carboxyl group of Glu is parallel to the phenyl ring of Tyr 59 *, suggesting a van der Waals' interaction between the two groups. This sort of weak interaction may be important for the recognition of C5 substrates (Glu and 2OG) in amino acid aminotransferases, because the Y70*S mutant of AspATEc is reportedly less active against these substrates (57). The phenyl ring at position 70 is essential for the recognition of the Glu-2OG pair as substrates. Hence, in the binding model of ArATPh, both ends of the acidic substrate, Glu, are fixed at the active center of the enzyme by three major interactions: 1) salt bridges between Arg 362 and the ␣-carboxylate of Glu, 2) two hydrogen bonds located between Gly 34 and the ␣-carboxylate, and between Gly 34 and ␣-amino groups of the substrate, and 3) the capturing of the ␥-carboxylate by the hydrogen bond network through the water molecule Xaa 126 * and by a weak interaction with Tyr 59 *. The low activity against Asp is explained in two ways: 1) the lack of an arginine residue corresponding to Arg 292 * of cAspATp, which interacts with the distal carboxylate of the acidic substrate and 2) the lack of a hydrogen bond network through the water molecule Xaa 126 * and no interaction with Tyr 59 *, because of a lack of one methylene unit at the ␥ position.
Substrate Specificity of ArATPh-As shown in Tables I and  II, ArATPh prefers the substrates in the following order in k max /K d : Tyr Ͼ Phe Ͼ Glu Ͼ Trp Ͼ His Ͼ Ͼ Met Ͼ Leu Ͼ Asp Ͼ Asn. The substrate specificity differs from those of the mesophilic ArATs, including ArATPd, with the preference in k cat /K m being Tyr Ͼ Phe ϭ Asp Ͼ Trp Ͼ Glu (41). Thermostable ArATs from Pyrococcus furiosus and Methanococcus aeolicus were also reported to have distinct substrate specificity in k cat /K m : Phe Ͼ Trp Ͼ Tyr (52,58). Consequently, ArATPh has a novel substrate specificity compared with other ArATs.
Aminotransferases are increasingly applied to the large scale synthesis of unnatural and nonproteinogenic amino acids (1). Typically exhibiting relaxed substrate specificity, rapid reaction rates, and no need for cofactor regeneration, they possess many characteristics that make them useful for biocatalysis. Because of its novel substrate specificity and high level of resistance to organic solvents (data not shown), ArATPh will continue to be a useful biocatalysis for the synthesis of unnatural compounds.