Redesigning the Substrate Specificity of an Enzyme by Cumulative Effects of the Mutations of Non-active Site Residues*

Directed evolution was used to change the substrate specificity of aspartate aminotransferase. A mutant enzyme with 17 amino acid substitutions was generated that shows a 2.1 3 10 6 -fold increase in the catalytic efficiency ( k cat / K m ) for a non-native substrate, valine. The absorption spectrum of the bound coenzyme, pyridoxal 5 * -phosphate, is also changed significantly by the mutations. Interestingly, only one of the 17 residues appears to be able to contact the substrate, and none of them interact with the coenzyme. The three-dimensional structure of the mutant enzyme complexed with a valine analog, isovalerate (determined to 2.4-Å resolution by x-ray crystallography), provides insights into how the mutations affect substrate binding. The active site is remodeled; the subunit interface is altered, and the enzyme domain that encloses the substrate is shifted by the mutations. The present results demonstrate clearly the importance of the cumulative effects of residues remote from the active site and represent a new line of approach to the redesign of enzyme activity.

Despite the impressive ability of natural enzymes to catalyze a broad array of reactions and utilize diverse substrates, attempts to make even minor modifications in enzyme activity or substrate specificity have proven to be difficult. Efforts to change the properties of existing enzymes (1)(2)(3)(4) have highlighted a limitation to enzyme design; in most cases, we can consider only amino acid residues that constitute the active site. Enzymes, however, exert their functions not only through the chemical properties of the side chains of the amino acid residues that contact substrates and cofactors. Residues distant from an active site may be important in holding the catalytic residues in their required orientations. Charge distribution throughout a whole enzyme molecule may facilitate substrate binding by electrostatically guiding the substrate into the active site. These ideas are consistent with the fact that enzymes are macromolecules composed of hundreds of amino acid residues. It has been difficult to demonstrate by protein engineering the importance of residues that are remote from the active site because the number of these residues is large and the contribution of each residue may be modest. It is also beyond our present understanding to predict a priori the effects of the mutations of remote residues on enzymatic activity through changes in the complex architecture of tertiary and/or quaternary structure. Although random mutagenesis and directed evolution have recently proven to be useful in addressing such problems in the rational redesign of enzymes (5)(6)(7)(8)(9)(10), creating enzymes with the desired activity remains challenging.
Aspartate aminotransferase (AspAT) 1 is a homodimeric enzyme and catalyzes amino group transfer between acidic amino acids, aspartate and glutamate, and their corresponding 2-oxo acids (11). Each subunit has a pyridoxal 5Ј-phosphate (PLP) molecule at the active center. The structure (12,13) and reaction mechanism (14) of AspAT have been studied extensively. AspAT from Escherichia coli shows moderate activity for aromatic amino acids. The activity for ␤-branched amino acids, valine and isoleucine, is barely detectable and even lower than that for basic amino acids (15). Several studies were previously done to increase the activity for basic or aromatic amino acids (16 -18). These results imply that the substrate specificity of AspAT like many other enzymes cannot be easily manipulated by mutating one or a few active site residues.
To alter the substrate specificity of AspAT toward ␤-branched amino acids, we established an experimental system based on directed evolution where mutant AspATs with higher activity for branched-chain substrates, especially valine and 2-oxovaline, evolve during successive rounds of selection (10). Briefly, the selection system uses an auxotrophic E. coli strain, which is deficient in the gene for branched-chain amino acid aminotransferase. The higher the activity of a plasmid-encoded mutant AspAT for 2-oxovaline is, the faster the auxotrophic E. coli carrying the plasmid grows on a selection plate. After five rounds of the selection, the catalytic efficiency (k cat /K m ) of mutant AspATs for ␤-branched substrates was increased about 10 5 -fold. One of the mutant AspATs showing the highest activity (AV5A-7) was analyzed in detail (10). The mutant had 13 amino acid substitutions, but interestingly, only one of the mutated residues seemed to interact directly with the substrate based on the three-dimensional structure of the wildtype AspAT. To elucidate the effects of these mutations on substrate binding, we set out to determine the three-dimensional structure of the mutant AspAT complexed with a ␤-branched substrate analog. The affinity of AV5A-7 for valine was, however, still too low (K m ϭ 400 mM) to yield a crystal of such an enzyme-inhibitor complex.
Here we describe the further improvement in both the catalytic efficiency and the K m value of AV5A-7 for valine. The affinity of a new mutant AspAT for valine is high enough to * This work was supported by the Ministry of Education, Science, Sports, and Culture of Japan (to T. Y.) and the Japan Society for the Promotion of Science ("Research for the Future" Program) (to H. K.). 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 allow crystallization of a complex between the mutant enzyme and a valine analog. The crystal structure shows that the mutations in residues distant from the active site cause significant changes in the higher order structure of the enzyme, which influence substrate and cofactor binding.

MATERIALS AND METHODS
Directed Evolution-An auxotrophic E. coli strain, YJ103 (⌬ilvE::kan), the selection medium, 5Ј and 3Ј primers used for the polymerase chain reaction amplification of the coding region of aspC, and the construction of the expression plasmid were reported previously (10). Briefly, the first round of selection was done as follows. The coding region of the aspC gene containing the three mutations, Ser 139 3 Gly, Asn 142 3 Thr, and Asn 297 3 Ser, was subjected to DNA shuffling (6,9,19), and the mutant genes were ligated downstream of the promoter of the tetracycline resistance gene of pBR322. After a 45-h incubation at 37°C, 116 colonies from a library of 5.6 ϫ 10 7 colonies were picked up, and a mixture of the plasmids was prepared. As for the second and third rounds of selection, the conditions were the same as those for the first round except the incubation time: the second round, a 40-h incubation, a library size of 7.9 ϫ 10 7 colonies, and 136 colonies picked; the third round, a 28-h incubation, a library size of 2.8 ϫ 10 7 colonies, and 48 colonies picked. Among the 48 clones, 9 clones exhibiting the highest activity for 2-oxovaline were chosen, and the coding regions of the aspC genes were sequenced.
Expression, Purification, and Activity Measurement of Mutant As-pATs-The coding region of the mutant AspATs was subcloned into pUC18. The mutant enzymes were expressed in E. coli TY103 (20), which is deficient in the AspAT gene, and purified as described (21). The activity for each substrate was measured at 25°C by spectrophotometrically monitoring the single turnover reaction using an Applied Photophysics stopped-flow apparatus (model SX.17MV) as described (15). The buffer system was 50 mM Hepes, pH 8.0, containing 0.1 M KCl and 10 M EDTA.
Crystallography-The crystals of ATB17 complexed with isovalerate were grown by the sitting drop vapor diffusion method. Three microliter drops containing 37 mg/ml protein were mixed with 1 l of 0.2 M sodium isovalerate and 3 l of the reservoir solution containing 1.6 M ammonium sulfate and 0.1 M Na⅐Hepes, pH 7.5. The drops were equilibrated against 0.5 ml of the reservoir solution at 20°C. An x-ray data set was collected with a Rigaku R-AXIS IIc image plate detector mounted on a Rigaku RU-200 rotating anode generator operated at 40 kV and 100 mA with monochromatized CuK ␣ radiation at room temperature. The oscillation images were processed and reduced using a data processing software, Rigaku PROCESS (22). Refinement of the structure began with the structure of the wild-type AspAT 2 (12) as an initial model using X-PLOR 3.851 (23) with parameters derived by Engh and Huber (24). After conventional positional refinement, simulated annealing using the slow cool protocol was performed. The models were improved by conventional positional refinement and the isotropic B-factor refinement, and manual rebuilding using Xfit (25) on the omit map was calculated with the coefficients ͉F o ͉ Ϫ ͉F c ͉. After R-factors were adequately lowered, water molecules were added to the models, and the structures were further refined.

Creation of a Mutant AspAT with Higher Affinity for Valine-Sequence analysis of the mutant AspATs obtained from
the previous selection showed that 5 of the 13 substitutions found in AV5A-7, Asn 34 3 Asp, Ile 37 3 Met, Ser 139 3 Gly, Asn 142 3 Thr, and Asn 297 3 Ser, were conserved among all of the mutants examined, and these substitutions were found to be functionally important (10). The fact that one substitution (Val 387 3 Leu) that significantly increased the activity for ␤-branched substrates was unique to AV5A-7 (10), however, implied that the sequence space had not yet been fully searched. Despite this finding, the initial selection system appeared to have reached its limit after five rounds of selection, because further increases in the catalytic efficiency for ␤-branched substrates did not benefit the growth of the host E.
coli cells under the most stringent selection conditions.
In the present study, we used two different strategies to find additional beneficial mutations. In one approach, we added each of the 20 unique substitutions found in four selected mutant AspATs, AV5A-1, AV5B-1, AV5B-4, and AV5B-5 (10), to AV5A-7 one mutation at a time and assayed each new mutant for 2-oxovaline activity (data not shown). Two substitutions, Ser 361 3 Phe and Ser 363 3 Gly, were chosen in this manner. In a second strategy, directed evolution was once again employed. This time, however, the experiment was started from a mutant AspAT that had three substitutions, Ser 139 3 Gly, Asn 142 3 Thr, and Asn 297 3 Ser, to facilitate the evolution. After three rounds of selection, mutant AspATs that showed similar activity for 2-oxovaline to that of AV5A-7 were obtained, and the coding regions of 9 mutant AspATs were sequenced. Each mutant AspAT had 4 -9 additional substitutions, and, again, Asn 34 3 Asp and Ile 37 3 Met were conserved in all the mutants. After adding each potential substitution one at a time to AV5A-7/Ser 361 3 Phe/Ser 363 3 Gly, 3 substitutions, Ala 11 3 Thr, Phe 24 3 Leu, and Ile 353 3 Thr, were chosen. One of the 13 substitutions of AV5A-7, Glu 7 3 Val, was mutated back to the wild-type sequence because the substitution was found to decrease the expression level of AspAT while not affecting the activity for ␤-branched substrates. The resulting mutant, ATB17, thus has 17 substitutions (Fig. 1A), 11 of which are clearly functionally important. As for the other 6 substitutions, Lys 41 3 Asn, Lys 126 3 Arg, Ala 269 3 Thr, Ala 293 3 Val, Ser 311 3 Gly, and Met 397 3 Leu, of which the contribution to the total effect was 10 -20% in AV5A-7, the importance of each substitution could not be determined (10).
Characterization of the Mutant AspAT-Compared with AV5A-7, the k cat /K m values of ATB17 for branched-chain substrates are increased, whereas those for acidic 2-oxo acid substrates are decreased (Table I). In particular, the K m value for valine is decreased 76-fold to 5.5 mM. The k cat /K m value of ATB17 for valine or 2-oxovaline is increased Ͼ2.1 ϫ 10 6 or 6.7 ϫ 10 5 -fold, respectively, and that for isoleucine or 2-oxoisoleucine is increased Ͼ6.0 ϫ 10 4 or 5.4 ϫ 10 5 -fold, respectively, compared with that of the wild-type AspAT. ATB17 retains significant activity for acidic substrates. Although this activity could have been easily eliminated by a single mutation at Arg 292 (see below and also Refs. 16 and 17), none of the mutants examined had such mutations. Probably Arg 292 was maintained simply because no selection pressure was applied to minimize the activity for acidic substrates or because acidic amino acids may serve as the amino group donors to 2-oxovaline in E. coli cells growing under the selection conditions.
During the purification of ATB17, we noticed that the color of the fractions containing the enzyme was orange, rather than yellow like the wild-type AspAT and AV5A-7. Thus, the absorption spectra of the purified AspATs were measured (Fig. 1B). AspAT has absorption bands in the region 300 -500 nm, which derive from the bound PLP molecule and are influenced significantly by amino acid residues interacting with PLP (11,26,27). The wild-type AspAT and AV5A-7 exhibit two major bands around 360 and 430 nm, whereas in ATB17 the latter band is red-shifted to 450 nm and has a broad shoulder above 500 nm. The differences in the relative intensity of the two bands between the wild-type and mutant AspATs show that the pK a of the imine nitrogen of the Schiff base formed between PLP and Lys 258 (26,27) is decreased in both mutant AspATs. These changes in the absorption spectra imply that the electronic distribution within the -electron system of the bound PLP molecule is changed significantly by the mutations, although the nature of these changes is difficult to predict.
Despite such drastic changes in the substrate specificity and absorption spectrum, only one of the mutated residues in ATB17 appears to be located at a position contacting the substrate, and none of them interact directly with PLP. The higher affinity of ATB17 for valine allowed us to crystallize ATB17 in the presence of an amino-free valine analog, isovalerate, to further investigate the effects of the mutations.

Effects of the Mutations on the Tertiary and Quaternary
Structure of AspAT-The x-ray crystal structure of ATB17 complexed with isovalerate was solved at 2.4-Å resolution (Table II) and was compared with the wild-type AspAT complexed with an aspartate analog, maleate (Fig. 2). Three features stand out in the gross structure of ATB17. First, the spatial arrangement of the two subunits of the dimer is altered. Second, the domain motion may be enhanced in ATB17. Third, two clusters of mutated residues are observed, one shown in green (Leu-24, Asp-34, Met-37, and Asn-41) and the other shown in red (Thr-353, Phe-361, Gly-363, Leu-387, and Leu-397) in Fig.  2. These findings may provide clues to elucidate how the non-active site mutations influence the substrate binding of AspAT and thus are further described below.
The rearrangement of the two subunits in ATB17 is evident. When the large domain (residues 49 -325) of one subunit of ATB17 was superimposed on that of the wild-type AspAT, the other subunits of the two enzymes overlapped poorly (deviated about 1 Å in the core region of the large domain) (Fig. 2). This difference in the subunit arrangement likely affects substrate binding because the active site is located at the subunit interface, although it is not known which mutations caused this subunit rearrangement.
The structure of AspAT changes from an "open" to a "closed" conformation when the substrate binds to the enzyme (12,13). In the ATB17-isovalerate structure, the domain closure is enhanced. The small domain of ATB17 comes closer to the active site compared with the corresponding domain of the wild-type AspAT. One of the clusters (Fig. 2, green), of which Met-37 is the only residue interacting with isovalerate, is located at the   e Parameters could not be determined experimentally because spectral changes were too small to obtain reliable parameters. Therefore, parameters of ATB17 for aspartate were estimated as follows (15): k 1 ϭ k 2 ⅐ k 3 /(k 2 ϩ k 3 ), K 1 ϭ K 2 ⅐ k 3 /(k 2 ϩ k 3 ), where k 1 and K 1 are the k cat and K m, aspartate values for the overall reaction, respectively; k 2 and k 3 are the k cat values for the single turnover reactions with aspartate and 2-oxovaline, respectively; and K 2 is the K m value for the single turnover reaction with aspartate. The parameters for the overall transamination reaction with aspartate and 2-oxovaline were determined as described (35) in the 50 mM Hepes buffer, pH 8.0, containing 0.1 M KCl and 10 M EDTA except 2-oxovaline was used instead of 2-oxoglutarate. lid of the active site which closes against the bound substrate. The other cluster (Fig. 2, red) is in the core of the small domain, except Leu-397, and may affect the domain motion given that these residues are located at the domain interface. Thus, these two clusters may influence in concert the domain motion.
Detailed analysis may help us to understand how the two clusters of mutated residues cause the observed conformational changes. Fig. 3 shows that the enzyme structure is changed especially around the two clusters. The Leu 20 -Ile 33 loop (the lid of the active site) shifts toward the active site while maintaining the overall configuration of the loop structure (Fig. 4B). The average deviation is about 1.5 Å, but the deviation is larger around the tip of the loop than around its bottom. This cluster of mutations is located at the hinge-like region and thus may cause the shift of the loop. On the other hand, the largest deviation in the whole molecule is observed at the domain interface (Fig. 3, position 363, and Fig. 4C). The backbone of Gly 363 -Gly 364 has flipped toward the solvent side. The Ser 361 3 Phe substitution would cause a steric hindrance due to its phenyl side chain. The Ser 363 3 Gly substitution yields two consecutive glycine residues of which the flexible backbone would flip to release the unfavorable strain. The effects of these two substitutions are, however, not fully cooperative because each substitution independently increased the activity for 2-oxovaline when added to AV5A-7. Other changes observed in Fig.  4C would have occurred to readjust the packing of the side chains surrounding the cluster of the mutated residues. All the above findings, including changes in the dimer interface, likely influence the substrate specificity of AspAT, but we cannot explain how these changes allow AspAT to accommodate ␤-branched substrates.
The Active Sites and Substrate-binding Modes of the Mutant and Wild-type AspATs-Trp 140 is thought to adjust the tilt of the PLP molecule during the course of catalysis by AspAT (11,14). In the ATB17-isovalerate structure, large conformational changes are observed in the Pro 138 -Thr 142 loop ( Fig. 3 and Fig.  4A). This loop contains two substitutions, Ser 139 3 Gly and Asn 142 3 Thr. Trp 140 is moved toward the "bottom" of the active site, making more space in the substrate binding pocket. The indole ring of Trp 140 is stacked against the pyridine ring of PLP by the hydroxyl group of Thr 142 , which is bulkier than the hydrogen atom at the corresponding position of the side chain of the original Asn 142 . The resulting tight interaction of the election systems of the indole ring and PLP may in part explain the spectral changes observed in ATB17. Interestingly, one of the evolved mutants we reported previously (10), AV5B-4, had an isoleucine residue at position 142, which also has a bulky  methyl group at C␤ of the side chain. Thus, the PLP molecule may be required to be pushed back by Trp 140 for AspAT to bind ␤-branched substrates.
The wild-type AspAT has two arginine residues, Arg 292 and Arg 386 , that are essential for substrate binding. Arg 292 interacts with the side chain carboxylate group of acidic substrates. The side chain of Arg 292 of ATB17 protrudes into solvent, whereas that of the wild-type AspAT flips toward the active site interacting with the carboxylate group of maleate (Fig. 4A). The same shift in the position of Arg 292 was observed previously in the crystal structure of a mutant AspAT complexed with aromatic substrate analogs (28). It is therefore possible that the difference in the position of Arg 292 between the two enzymes is caused by the difference in the bound substrate analogs, dicarboxylic maleate and monocarboxylic isovalerate, rather than by the mutations of ATB17. Arg 386 is also an important residue of which the side chain interacts with the ␣-carboxylate group of all amino acid substrates. The orientation of the Arg 386 side chain of ATB17 remains essentially unaltered despite the large conformational changes adjacent to Arg 386 (Fig. 4, A and C).
One of the side walls of the active site consists of the residues belonging to the other subunit (Fig. 4A, asterisks). Thus, the changes in the subunit interface in ATB17 have caused deviations of these residues. This portion of the active site also contains two substitutions, Ala 293 3 Val and Asn 297 3 Ser. Although the functional importance of the Ala 293 3 Val substitution is not clear, the Asn 297 3 Ser substitution increased the activity of AV5A-7 for 2-oxovaline (10). It was reported previously for the mutant AspAT with increased activity for aromatic substrates that the Asn 297 3 Ser substitution elimi-  Fig. 2. The structure of the ATB17-isovalerate complex is indicated by thick gray lines and that of the wild-type AspAT-maleate complex is indicated by thin purple lines. A, several water molecules are introduced into the active site of ATB17 (light blue spheres). One water molecule (WAT1) is located at almost the same position as a water molecule observed in the wild-type AspAT (a purple sphere). Arg 292 , Val 293 , and Ser 297 belong to the other subunit of the dimer (asterisks). B and C, viewed from the same direction as in Fig. 2.