The aglycone specificity-determining sites are different in 2, 4-dihydroxy-7-methoxy-1,4-benzoxazin-3-one (DIMBOA)-glucosidase (Maize beta -glucosidase) and dhurrinase (Sorghum beta -glucosidase).

The maize beta-glucosidase isozyme Glu1 hydrolyzes a broad spectrum of substrates in addition to its natural substrate DIMBOAGlc (2-O-beta-d-glucopyranosyl-4-hydroxy-7-methoxy-1,4-benzoxazin-3-on e), whereas the sorghum beta-glucosidase isozyme Dhr1 hydrolyzes exclusively its natural substrate dhurrin (p-hydroxy-(S)-mandelonitrile-beta-d-glucose). To study the mechanism of substrate specificity further, eight chimeric beta-glucosidases were constructed by replacing peptide sequences within the C-terminal region of Glu1 with the homologous peptide sequences of Dhr1 or vice versa, where the two enzymes differ by 4 to 22 amino acid substitutions, depending on the length of the swapped regions. Five Glu1/Dhr1 chimeras hydrolyzed substrates that are hydrolyzed by both parental enzymes, including dhurrin, which is not hydrolyzed by Glu1. In contrast, three Dhr1/Glu1 chimeras hydrolyzed only dhurrin but with lower catalytic efficiency than Dhr1. Additional domain-swapping within the C-terminal domain of Glu1 showed that replacing the peptide (466)FAGFTERY(473) of Glu1 with the homologous peptide (462)SSGYTERF(469) of Dhr1 or replacing the peptide (481)NNNCTRYMKE(490) in Glu1 with the homologous peptide (477)ENGCERTMKR(486) of Dhr1 was sufficient to confer to Glu1 the ability to hydrolyze dhurrin. Data from various reciprocal chimeras, sequence comparisons, and homology modeling suggest that the Dhr1-specific Ser-462-Ser-463 and Phe-469 play a key role in dhurrin hydrolysis. Similar data suggest that DIMBOAGlc hydrolysis determinants are not located within the extreme 47-amino acid-long C-terminal domain of Glu1.

In maize, the major function of ␤-glucosidase (DIMBOA-glucosidase) 1 is in defense against the European corn borer and other pests (3). There are two known isozymes of the enzyme, Glu1 and Glu2. The cDNAs corresponding to both isozymes were cloned and sequenced, and the deduced protein products were found to share 90% sequence identity (13). The cDNA corresponding to the sorghum ␤-glucosidase isozyme Dhr1 was also cloned and sequenced in our laboratory (14); it shares ϳ70% sequence identity with each of the two maize isozymes. The catalytically active form of both maize and sorghum ␤-glucosidases is a 120-kDa homodimer or its multimers. The primary structures of both enzymes contain the peptide motifs TFNEP and ITENG, which are shown to be highly conserved and make up the catalytic site in all family 1 ␤-glycosidases (15)(16)(17). Furthermore, the three-dimensional structures of six family 1 ␤-glucosidases (white clover linamarase, white mustard myrosinase, Lactococcus lactis 6-phospho-␤-galactosidase, Bacillus polymyxa ␤-glucosidase, Sulfolobus sulfataricus ␤-glucosidase, and Thermosphaera aggregans ␤-glucosidase) have recently been solved using crystals of the enzyme-glycosyl complexes (18 -23). In these four cases, the residues of the TFNEP and (I/V)TENG motifs or their equivalents were found to form part of a pocket or crater-shaped active site (24). The two catalytic glutamic acids in ␤-O-glucosidases (i.e. the nucleophile and the acid-base catalyst) and one glutamic acid (i.e. the nucleophile) and a glutamine (the water activator) in myrosinases were positioned within the active site at appropriate distances (2.5-3Å) on opposite sides of the glycosidic bond. There are two fundamental questions about ␤-glucosidasecatalyzed reactions. 1) How do ␤-glucosidases catalyze the hydrolysis of the ␤-glycosidic bond between two glycone residues (e.g. cellobiose and other ␤-linked oligosaccharides) or that between glucose and an aryl or alkyl aglycone (e.g. many naturally occurring substrates in plants) ? 2) What determines substrate specificity, including the site and mechanism of aglycone binding? Much progress has been made in understanding the mechanism of catalysis and defining the roles of specific amino acids involved in catalysis within the active site. However, there is virtually no information as to how ␤-glucosidases recognize their substrates and interact with them, specifically the aglycone moiety, which is the basis of tremendous diversity in natural substrates and is responsible for subtle substrate specificity differences among ␤-glucosidases. The maize ␤-glu-cosidase isozyme Glu1 and its sorghum homologue Dhr1 provide an ideal model system to address questions related to substrate specificity, because these enzymes represent extremes in substrate specificity. Although Dhr1 hydrolyzes only its natural substrate dhurrin, Glu1 hydrolyzes a broad spectrum of artificial and natural substrates in addition to its natural substrate DIMBOAGlc, but it does not hydrolyze dhurrin (Fig. 1).
The first attempt to investigate substrate specificity using chimeric ␤-glucosidases was made by Singh and Hayashi (25), who exchanged the C-terminal 58-amino acid-long domain of a prokaryotic ␤-glucosidase from Celvibrio gilvus with the Cterminal 60-amino acid-long domain of Agrobacterium tumefaciens ␤-glucosidase. They showed that the resulting chimeric enzyme exhibited the substrate specificity of A. tumefaciens ␤-glucosidase. Hoa and Hayashi 2 also show that the deletion of 70 amino acids from the C-terminal region of C. gilvus ␤-glucosidase leads to complete loss of activity. These investigators concluded that the C-terminal region of the enzyme played a major role in determining substrate specificity and catalytic activity.
The importance of specific enzyme domains in substrate specificity and catalytic efficiency was also shown in a number of other enzyme chimeras produced from two enzymes that differ with respect to substrate specificity. For example, it was shown that exchanging a small N-terminal portion between two rice ␣-amylase isozymes (Amy A and Amy 3D) resulted in a chimeric enzyme (Amy A/3D) that shows high activity on both soluble starch and oligosaccharides, whereas parental enzymes have high activity only on either soluble starch (Amy A) or oligosaccharides (Amy 3D) (26).
The purpose of the studies described in this paper was to gain insight into the mechanism of substrate (aglycone) recognition and binding in ␤-glucosidases using two plant enzymes, the maize isozyme Glu1 and sorghum isozyme Dhr1, as model systems. To this end, we have constructed eight chimeric enzymes by reciprocal domain-swapping. Target domains were selected based on amino acid sequence comparisons and analysis of modeled three-dimensional structures. We demonstrate that the maize Glu1 isozyme gains the ability to hydrolyze dhurrin when the C-terminal 47-amino acid-long region or specific subdomains within this region are replaced by corresponding Dhr1 region, whereas the reciprocal replacement has no effect on the substrate specificity of the sorghum Dhr1 isozyme, except a12-fold reduction in activity.

EXPERIMENTAL PROCEDURES
Construction of Chimeric ␤-Glucosidases-The first step toward understanding the basis of substrate specificity in maize and sorghum ␤-glucosidases was to construct cDNAs encoding chimeric enzymes by domain swapping. Since we had already cloned and expressed cDNAs encoding maize Glu1 and Glu2 isozymes as well as the sorghum Dhr1 isozyme in Escherichia coli (27), the construction and expression of chimeric cDNAs using these wild type parental templates were straightforward. The criteria for chimeric constructs was that the swapped region includes one or more amino acid substitutions within the C-terminal domain that map to or around the active center in the modeled three-dimensional structures of Glu1 and Dhr1. Chimeric cDNAs were constructed by the PCR-based recombination technique of overlap extension and the high fidelity thermostable Turbo® Pfu polymerase (Stratagene) as described (27,28). Three of the primer pairs (P100-P101, P166-P167, and P168-P169, Table I) were from conserved regions, such that they will function on both glu1 and dhr1 cDNA templates. Sequences of oligonucleotide primers used in PCR and the corresponding peptides that were swapped are shown in Table I and Fig. 2, A and B, respectively. The constructs were made so as to encode chimeric enzymes Glu1/Dhr1 or Dhr1/Glu1, where the enzyme before the slash contributed the N-terminal region, and the one after the slash contributed the C-terminal region (Fig. 2B). As a first step in defining the domain-determining substrate specificity of Glu1 and Dhr1, chimera 2 (abbreviated hereafter as chim 2) was constructed by replacing the extreme 47-amino acid-long C-terminal region (amino acids 466 -512) of Glu1 with the corresponding 53-amino acid-long region (amino acids 462-514) of Dhr1. The 5Ј portion of the chimeric cDNA was amplified on the glu1 template using the vector-specific primer T7 (sense) and gene-specific primer P101 (antisense), whereas the 3Ј portion was amplified on the dhr1 template using the primers P100 (sense) and the vector-specific T7term (antisense). Junction primers P100 (sense) and P101 (antisense) were derived from the region of cDNA encoding the peptide sequence GWFAWSL, which is invariant in Glu1 and Dhr1 ( Fig. 2A). The two PCR fragments were gel-purified, denatured, mixed, annealed (by overlapping the complementary ends that contain primers P100 and P101 sequences) and extended to obtain the full-length chimeric Glu1/Dhr1 cDNA coding sequence. The full-length chimeric cDNA was amplified by using the vector-specific primer pair of T7 (sense) and T7 ter (antisense) on the overlap extended template. The resulting PCR product was purified, digested with NheI and XhoI, and cloned into the expression plasmid pET21a. The construction of other chimeric cDNAs (chim 15,16,17,21,22,23, and 39) followed the procedure described for chim 2, except the primer pairs used for PCR (Table I). Among the eight chimeric enzymes produced, the sizes of the swapped peptides varied from 8 to 53, all coming from the C-terminal fragment 462-514 of Dhr1 and 466 -512 of Glu1 (Table I).
Expression and Purification-Wild type and chimeric ␤-glucosidases were produced in E. coli pLyS cells (F -ompT hsdSP r B -m B -gal dcm) under the control of the T7 RNA polymerase promoter in the expression plasmid pET-21a (Novagene) as described by Cicek and Esen (27). The cell lysis and protein extraction procedure was performed as described (27). For purification, ␤-glucosidase was precipitated from crude cell extracts with a 35 to 65% ammonium sulfate ((NH 4 ) 2 SO 4 ) cut. The precipitate was dissolved in 50 mM sodium acetate buffer, pH 5.0, and centrifuged at 18,000 ϫ g for 30 min. The supernatant was adjusted to a final concentration of 0.5 M (NH 4 ) 2 SO 4 and centrifuged at 18, 000 ϫ g for 30 min. Then the supernatant was applied to a ToyoPearl-butyl 650M (TosoHaas) hydrophobic interaction chromatography column (1.6 ϫ 14 cm). The column was washed to base-line absorbance with 0.5 M (NH 4 ) 2 SO 4 in buffer and eluted with approximately 5 bed volumes of 2 T. T. Hoa and K. Hayashi, unpublished results. a reverse salt gradient of 0.5 M to 0.1 M (NH 4 ) 2 SO 4 in 50 mM sodium acetate buffer, pH 5.0. The resulting fractions were assayed for ␤-glucosidase activity using the artificial substrate pNPGlc or the natural substrates DIMBOAGlc or dhurrin. The fractions with activity were pooled based on purity as judged by SDS-PAGE. The pooled fractions were adjusted to 0.5 M (NH 4 ) 2 SO 4 and were rechromatographed on a ToyoPearl-phenyl 650M (Toso Haas) column as described above. The purification protocol was the same for Glu1 and Glu1/Dhr1 chimeras except for slight changes of the (NH 4 ) 2 SO 4 concentration in the binding and elution steps of hydrophobic interaction chromatography. For purification of Dhr1, chim 17, chim 23, and chim 39, 0.05 M phosphate buffer, pH 7.0, was used in all steps because Dhr1, and Dhr1/Glu1 chimeras were not stable at pH 5.0 (acetate buffer). Again, the fractions that had ␤-glucosidase activity were checked for purity by SDS-PAGE, pooled, and concentrated approximately 10-fold using a 30,000 cut-off spin column (Gelman Sciences). The concentrated enzymes were then used for kinetic analysis with special emphasis on substrate specificity.
Enzyme Assays-For activity assays in native PAGE gels, the purified parental and chimeric enzymes were electrophoresed into 6% alkaline gels to obtain zymograms using the fluorogenic substrate 4MUGlc and the chromogenic substrate 6BNGlc as described (29,30).
Kinetic parameters, K m and k cat (V max /E tt ) for parental and chimeric ␤-glucosidases were determined by varying substrate concentration from 0.098 to 16.66 mM in citrate-phosphate buffer, pH 5.8, for the artificial substrates pNPGlc and oNPGlc. Protein content was adjusted to appropriate concentrations according to the Bradford assay (Bio-Rad) for all activity assays. K m and k cat values were determined based on the amount of p-nitrophenolate and o-nitrophenolate released from pNPGlc and oNPGlc, respectively. Each assay was performed in quadruplicate in a microtiter plate in a total volume of 140 l containing 70 l of substrate and 70 l of enzyme solution. The reaction mixture was incubated at room temperature (ϳ25°C) for 10 min, and the reaction was stopped by adding 70 l of 0.4 M Na 2 CO 3 . The absorbance of the nitrophenolate released was read in a microplate reader (Dynatech) at 410 nm. The K m and k cat values for the natural substrates DIMBOAGlc and dhurrin were determined by the peroxidase-glucose-oxidase-coupled reaction (31). Fifty-l aliquots containing 0.031 to 2.0 mM DIM-BOAGlc or dhurrin in phosphate-citrate buffer, pH 5.8, were placed in quadruplicate in wells of a microtiter plate followed by 50 l of diluted ␤-glucosidase solution, 50 l of peroxidase-glucose-oxidase enzymes, and 50 l of ABTS (2,2Ј-azinobis-3-ethylbenzthiazolinesulfonic acid). The reaction mixture was incubated at 37°C for 30 min, and the absorbance was read in the microplate reader at 410 nm.
Inhibition of Glu1 by Dhurrin and Dhr1 by DIMBOAGlc-Inhibition experiments were performed using pNPGlc as substrate for Glu1 at the substrate concentration range of 1 to 8 K m and dhurrin for Dhr1 at 1 and 10 K m . Inhibitors (dhurrin for Glu1 and DIMBOAGlc for Dhr1) were applied at four different concentrations, 0.5-2 K m . The type of inhibition and K i values for Glu1 and Dhr1 were determined and calculated by Lineweaver-Burk linearization using Enzyme Kinetics® software (Trinity Software). In addition, inhibition of Dhr1/Glu1 chimeras by DIMBOAGlc were studied by varying DIMBOAGlc concentration from 0.009 mM to 0.25 mM, and the inhibitor concentration causing 50% inhibition of dhurrin hydrolysis was determined by plotting v versus [I]. The DIMBOAGlc concentration causing 50% inhibition of dhurrin hydrolysis by Dhr1 and the dhurrin concentration, causing 50% inhibition of DIMBOAGlc hydrolysis by Glu1 were also determined by using the same procedure.
Thin Layer Chromatography-The natural substrates dhurrin and DIMBOAGlc were purified as described (27). The purified parental Glu1, Dhr1, and their chimeras (Fig. 2B) were adjusted to a final concentration of 1 g/ml in 10 mM citrate, 20 mM phosphate buffer, pH 5.8. Reactions were then incubated with 5 mM final concentrations of DIMBOAGlc and dhurrin at room temperature for 6 h. Ten l of the reaction mixture was spotted on the TLC plate (0.25-mm silica-coated Whatman PE SIL G/UV plates) and chromatographed vertically using the acetonitrile/H 2 O (85/15) mixture as the mobile phase for 45 min (32). The plate was sprayed with CH 3 OH/H 2 SO 4 (4:1; v/v) and then baked at 110°C for 10 min to visualize DIMBOAGlc, dhurrin, and the reaction product glucose resulting from hydrolysis. For each substrate, a "minus enzyme control" was included in the assay.
Molecular Modeling-Models of the three-dimensional structure of Glu1 and Dhr1 were generated by homology modeling using the Mod-eller4 program (33). The models were based on the known three-dimensional structures of linamarase, the cyanogenic ␤-glucosidase from white clover (Protein Data Bank code 1cbg), and myrosinase from white mustard (Protein Data Bank code 1myr). Five models each of Glu1 and Dhr1 were generated in Modeller4. The models were sufficiently similar that an average structure for each of the enzymes was prepared by coordinate averaging. The Leap module of AMBER4.1 (34) was used to add hydrogen atoms to the models, and bad contacts in the models were eliminated using energy minimization with the Sander module of AM-BER. For energy minimization, 100 cycles of steepest descent minimization of hydrogen atom positions was done first, after which 600 steps of steepest descent minimization of all atoms was done. Expression levels were about 10% of the total E. coli protein, and solubility was close to 30% (data not shown) when cultures were grown at 37°C and induced at room temperature. Since the expressed proteins did not contain affinity tags, they were purified by a simple two-step conventional procedure, differential solubility (35-60% (NH 4 ) 2 SO 4 cut) followed by hydrophobic  interaction chromatography. This procedure yielded essentially homogenous protein in all cases, as evident from SDS-PAGE profiles (Fig. 3A). Moreover, the native PAGE electrophoretic mobilities of two chimeras (chim 21 and 22) containing the shortest segments from Dhr1 were identical to that of Glu1 (Fig. 3, B-C, lanes 1, 6, and 7), whereas those of three other chimeras (chim 2, 15, and 16) containing longer Dhr1 segments were faster than that of Glu1 (Fig. 3, B-C, lanes 3, 4, and 5). The electrophoretic mobilities of Dhr1 and Dhr1/Glu1 chimeras (chim 17, 23, and 39) are not known because they are not active on the substrates used for zymogram development (Fig. 3, B-C, lanes 2, 8, 9, and 10).

Expression and Purification of Chimeric Enzymes-The
Substrate Specificity and Kinetics of Glu1 and Glu1/Dhr1 Chimeras-The catalytic activity of the parental enzymes and five chimeras toward natural (DIMBOAGlc and dhurrin) and artificial substrates (pNPGlc, oNPGlc, 4MUGlc, and 6BNGlc) was assayed in solution and in activity gels. The substrate specificity data showed that Glu1 had activity toward both its natural substrate DIMBOAGlc and each of the four artificial substrates tested (Tables II and III; Fig. 3, B-C). Similarly, all five Glu1/Dhr1 chimeras had activity on all of these five substrates and dhurrin. Thus, replacement of either a large (chim 2, chim 15, and chim 16) or a small (chim 21 and chim 22) portion of the C terminus of Glu1 with the homologous portion of Dhr1 altered and broadened the substrate specificity in Glu1/Dhr1 chimeras. In other words, each of the Glu1/Dhr1 chimeras hydrolyzed the physiological and artificial substrates that are hydrolyzed both by Glu1 and Dhr1 but with a different catalytic efficiency (Tables II and III). These results were also confirmed by TLC in the case of the physiological substrates dhurrin and DIMBOAGlc (Fig. 4, A and B). When substrate specificities of parental and chimeric enzymes were compared using 4MUGlc and 6BNGlc in zymogram assays after native PAGE in 6% gels, Glu1 and all five Glu1/Dhr1 chimeras hydrolyzed both of these substrates (Fig. 3, B-C, lanes 1 and 3-7).
The kinetic parameters (K m , k cat , and k cat /K m ) of both parental and chimeric ␤-glucosidases were determined, and the data are summarized in Tables II and III. As mentioned above, all five Glu1/Dhr1 chimeras exhibited substrate specificities characteristic of both Glu1 and Dhr1, hydrolyzing DIMBOAGlc (not hydrolyzed by Dhr1) and dhurrin (not hydrolyzed by Glu1), as well as the artificial substrates oNPGlc, pNPGlc, 4MUGlc, and  Tables II and III). In addition, all Glu1/Dhr1 chimeras except chim 15 showed nearly a 3-4fold increase in k cat for pNPGlc hydrolysis and a 2-3-fold increase in k cat for oNPGlc hydrolysis with little or no change in K m , when compared with Glu1. Higher specificity of these chimeras for pNPGlc and oNPGlc than Glu1 and chim 15 was clearly evident in specificity coefficients (k cat /K m ) and relative efficiencies (Table II). Moreover, Glu1 and Glu1/Dhr1 chimeras had about 4-fold higher K m values for oNPGlc than for pNPGlc. Chim 22 had the lowest K m and highest relative efficiency for oNPGlc among all five Glu1/Dhr1 chimeras. Although all Glu1/ Dhr1 chimeras hydrolyzed the natural substrates dhurrin and DIMBOAGlc, there were significant differences among them with respect to kinetic parameters (Tables II and III). The K m of Dhr1 for dhurrin is 0.051 mM, which is about one-third that determined by Hösel et al. (35). Chim 15 had the highest K m (0.51 mM) for dhurrin followed by chim 2 and 22 and the lowest specificity constants (Table III). Two chimeras (chim 16 and 21) had the lowest K m values among the five for dhurrin (ϳ0.1 mM), which were twice the value of that for the parental enzyme Dhr1. The K m and k cat values of chimeras (except chim 16) for DIMBOAGlc hydrolysis were closer to those of the parental Glu1. However, chim 16 stood out among the five with lowest specificity coefficient (k cat /K m ) and relative efficiency (32%), whereas others varied from 65 to 88% when compared with that of Glu1 (Table III). Thus, in all cases, acquiring the ability to hydrolyze dhurrin was accompanied by lowered catalytic efficiency toward DIMBOAGlc. It should be noted that chimeras 15 and 16 were obtained by splitting the 53-amino acid-long Dhr1 domain at the C terminus of chim 2 into two subdomains to further define the basis of specificity for dhurrin hydrolysis. Chim 15 contained the extreme 23-amino acid-long C-terminal region (amino acids 492-514) from Dhr1, and none of these residues mapped to the catalytic site of the modeled parental enzymes. Instead, this region resides on the surface of the tertiary structures of Glu1 and Dhr1. Kinetic data indicate that the 23-amino acid-long peptide from the extreme C-terminal region of Dhr1 still has an effect on chim 15 for dhurrin specificity, although its catalytic efficiency coefficient (k cat /K m ) for dhurrin was 57-fold and 19-fold, respectively, lower than those of Dhr1 and chim 21 (Table III). K m and k cat values for chim 15 remained similar to those of the parental enzyme Glu1 for pNPGlc, oNPGlc, and DIMBOAGlc. Chim 16 contained a 30amino acid-long internal peptide (amino acids 462-490) from Dhr1. It had a K m value similar to that of the parental enzyme Glu1 but showed nearly a 4-fold increase in k cat for pNPGlc hydrolysis and a 3-fold increase in k cat for oNPGlc hydrolysis, similar to chim 2. The kinetic values (K m and k cat ) of chim 16 for DIMBOAGlc were similar to those of chim 2, whereas its K m value for dhurrin was about one-third that for chim 2. Its k cat value for dhurrin increased nearly 1.5-fold, and the catalytic efficiency increased nearly 5-fold when compared with chim 2.
To    2 and 3), Glu1/Dhr1 chimeras (chim 2, 15, 16, 21 and 22 (lanes 3-8), and Dhr1/Glu1 chimeras (lanes 10 -12) expressed in E. coli. The plus (ϩ) denotes incubation of the substrate with parental Glu1 and Dhr1 or their chimeras, and the minus (Ϫ) denotes incubation of the substrate without any enzyme source (negative control). Note that the parental enzymes Glu1 and Dhr1 hydrolyze their natural substrate DIMBOAGlc and dhurrin, respectively, as do all five Glu1/Dhr1 chimeras, whereas all three Dhr1/Glu1 chimeras hydrolyze dhurrin only. residues (Fig. 5). Chim 21 hydrolyzed dhurrin with a catalytic efficiency one-third of that of Dhr1 but nearly 8 times that of chim 2. Its activity (e.g. efficiency coefficient k cat /K m ) toward DIMBOAGlc was similar to that of chim 2 but lower than that of Glu1 (Table III). Thus, the transfer of a total of only four amino acid substitutions (F466S, A467S, F469Y, and Y473F, numbering based on Glu1 sequence) from Dhr1 to Glu1 enabled Glu1 to hydrolyze dhurrin without substantially affecting its activity toward DIMBOAGlc and other substrates. Chim 22 also showed activity toward dhurrin. However, its catalytic efficiency was about 3-fold less than that of chim 21. In this case, a total of five substitutions (N481E, N483G, T485E, Y487T, and E490R) in the Dhr1 peptide 477 ENGCERTMKR 486 conferred to Glu1 the ability to hydrolyze dhurrin with no change in specificity for other substrates that are hydrolyzed by Glu1, including DIMBOAGlc.
The kinetic parameters (K m , k cat , and k cat /K m ) of both Dhr1 and three Dhr/Glu1 chimeras were determined, and the data are summarized in Tables II and III. Although these three chimeras hydrolyzed only dhurrin among all substrates tested and, thus, behaved like the parental enzyme Dhr1, their catalytic efficiencies differed considerably from each other and Dhr1 in that there was a negative relationship between activity and the length of the Glu1 domain replacing the C-terminal region of Dhr1. For example, chim 23 had a slightly lower K m (0.04 versus 0.05 mM) than Dhr1, and its other kinetic parameters and relative catalytic efficiency (ϳ90%) were similar to those of Dhr1. This chimera had the shortest Glu1 domain, the extreme 16-amino acid-long C-terminal (amino acids 492-508), where Dhr1 and Glu1 differ by 4 amino acid substitutions, as well as a 2-and a 4-amino acid-long addition, making the Glu1 segment 6 amino acids shorter than its Dhr1 homolog. In contrast, chim 17 and 39, although both hydrolyzed dhurrin, had higher K m values and drastically reduced catalytic efficiencies, 12-and 7-fold, respectively, when compared with the wild type Dhr1 (Table III). Of these, chim 17 (reciprocal of chim 2) had the longest Glu1 domain (47 amino acids long, amino acids 466 -512) and the poorest kinetic parameters and lowest activity toward dhurrin. Similarly, chim 39, which had a 31-amino acid-long internal C-terminal Glu1 domain (amino acids 462-492), had only slightly better kinetic parameters than chim 17.
Inhibition Studies-In view of the fact that dhurrin is not hydrolyzed by Glu1 and DIMBOAGlc is not hydrolyzed by Dhr1, the inhibitory effects of dhurrin on Glu1 and DIMBOA-Glc on Dhr1 were studied. The results showed that dhurrin is a competitive ground state inhibitor for Glu1 (or DIMBOAglucosidase) having a K i of 0.076 mM. DIMBOAGlc is also a potent competitive inhibitor for Dhr1, with a K i of 0.009 mM. The K i value for dhurrin using pNPGlc as substrate is in agreement with the data found in previous work (36). The inhibitory effect of DIMBOAGlc on Dhr1 has not been reported previously. Moreover, the three Dhr1/Glu1 chimeras, which do not hydrolyze DIMBOAGlc, were all inhibited by it, as was Dhr1. The DIMBOAGlc concentration causing 50% inhibition of dhurrin hydrolysis by these three chimeras was in the range of from 0.003 to 0.01 mM and was lower than that of Dhr1 (0.026 mM).

DISCUSSION
In this study we have designed and produced chimeric enzymes from two naturally occurring enzymes, creating a novel enzymatic function as well as improved catalytic efficiency on certain substrates. Our model system was comprised of two ␤-glucosidases, Glu1 and Dhr1, each with a strict specificity for its own physiological substrate, although they share 70% sequence identity and contain identical catalytic amino acids and glycone recognition and binding motifs (TFNEP and ITENG). Based on the modeled three-dimensional structures of Glu1 and Dhr1, we have successfully combined two different substrate specificities in a single chimeric enzyme by replacing the C-terminal domain of Glu1 with the homologous domain from Dhr1. This strategy added novel substrate specificity (e.g. dhurrin hydrolysis) to the maize Glu1 isozyme and a 2-4-fold improvement in its catalytic efficiency on other substrates (e.g. nitrophenyl ␤-glucosides, Table II).
The purpose of producing five Glu1/Dhr1 and three Dhr/Glu1 chimeric ␤-glucosidases was to delineate the regions of primary structure that contain key amino acids and sequence motifs determining substrate (or aglycone) specificity. Each of the five Glu1/Dhr1 chimeric enzymes (chim 2, 15, 16, 21, and 22) exhibited the combined substrate specificities of the parental enzymes and, on average, 2-4-fold higher catalytic efficiency on certain substrates than the parental enzyme Glu1. The basis of the broadened substrate specificity and improved catalytic efficiency is thought to reside in amino acid substitutions that occur in the 53-amino acid-long C-terminal domain of Dhr1 or its shorter fragments that were swapped with the homologous regions of Glu1. However, three Dhr1/Glu1 chimeras (chim 17, chim 23, and chim 39) hydrolyzed only dhurrin, but catalytic efficiency was drastically reduced in the case of chim 17 and 39 because the length of the exchanged Glu1 domain increased. Thus, Dhr1/Glu1 chimeras exhibited neither widened substrate specificity nor improved catalytic efficiency (Tables II and III). These results allow us to draw three conclusions. 1) The C-terminal domain of ␤-glucosidases includes residues that are necessary, but not sufficient, for aglycone recognition and binding. This is in agreement with the results of Singh and Hayashi (25); they showed a chimeric enzyme obtained by replacing the C-terminal domain of a C. gulvis ␤-glucosidase with the homologous C-terminal domain of an A. tumefaciens ␤-glucosidase had the substrate specificity of the C-terminal region donor. The N-terminal region, at least the first 41 amino acids of Glu1, does not appear to have a discernable role in substrate specificity. We replaced amino acids 1-41 of Dhr1 with the corresponding N-terminal region of Glu1. The resulting Dhr1/Glu1 chimeric enzyme hydrolyzed only dhurrin as does Dhr1 (data not shown) and exhibited kinetic properties of Dhr1, suggesting that the N terminus of Glu1 and possibly of other ␤-glucosidases is not involved in substrate (i.e. aglycone) specificity. However, the N-terminal region is required for catalysis because it contains a universally conserved amino acid (Gln-38 in Glu1 and Gln-39 in Dhr1, Fig. 2), which is in the glycone binding pocket of the active site (18 -23). 2) The dhurrin hydrolysis determinants in sorghum ␤-glucosidase (Dhr1) are among 22 amino acid substitutions in the extreme 53-amino acid-long C-terminal domain of this enzyme that distinguish it from the homologous 47-amino acid-long C-terminal domain of Glu1. More specifically, one or a combination of the amino acids Ser-462-Ser-463, Tyr-465, and Phe-469 that reside in the peptide 462 SSGYTERF 469 are essential for dhurrin hydrolysis because this peptide alone confers the capability to hydrolyze dhurrin to Glu1, as is the case in chim 21 (Table III and Fig. 4B, lane 7). 3) In contrast, the DIMBOAGlc hydrolysis determinants are not in the 47-amino acid-long Cterminal domain of Glu1, although this domain may contain residues that are involved in DIMBOAGlc or DIMBOA binding.
We postulate that although Glu1 and Dhr1 differ by 151 amino acid substitutions, 5 deletions, and 7 additions at 514 positions (ϳ30% sequence divergence), only a small number of these changes are relevant to the substrate specificity differences between them. Indeed only 4 (Ser-462-Ser-463, Tyr-465, and Phe-469) of the 22 variant amino acid sites in the extreme 47-to 53-amino acid-long C terminus map to or around the active site of the modeled enzymes (Fig. 5). Therefore, it is conceivable that more than 90% of the amino acid substitutions separating Glu1 and Dhr1 are adaptively and functionally neutral, which would be consistent with Kimura's theory of neutral evolution (37). There are well documented examples in the literature supporting this postulate. For example, eubacterial and mitochondrial isocitrate dehydrogenases differ with respect to coenzyme specificity; the former is NADP-dependent, whereas the latter is NAD-dependent. Moreover, both enzymes have essentially the same tertiary structure, although they differ by 250 amino acid substitutions at 320 positions. Only 6 of these 250 amino acid substitutions determine coenzyme specificity, as shown elegantly by shifting NADP specificity to NAD specificity or vice versa by replacing these amino acids in the coenzyme binding pocket (38). Wilks et al. (39) provide even a more dramatic example in that they changed a lactate dehydrogenase to a malate dehydrogenase by replacing a single key amino acid although the two enzymes differed by 230 amino acid substitutions. Other examples of bringing about dramatic changes in substrate specificity and catalytic properties include changing the substrate specificity and double-bond positional specificity of an acyl-carrier protein desaturase by five amino acid replacements (40), increasing catalytic efficiency and broadening substrate specificity in a chimeric protease constructed by recombining the N-terminal domain of coagulation factor X with the C-terminal domain of trypsin (41), and introducing the active site of nonheme iron superoxide dismutase into E. coli thioredoxin and changing it to a superoxide dismutase (42).
The substrate specificity data from chim 2 provided the first clue to the fact that the C-terminal 53-amino acid-long domain of Dhr1 contains key determinants for dhurrin hydrolysis. However, these determinants alone did not change the specificity of a Glu1/Dhr1 chimera entirely to that of the C-terminal region donor Dhr-1. Since there were 22 amino acid substitutions and two additions (a dipeptide and a tetrapeptide) in the 53-amino acid-long Dhr1 C-terminal domain that differentiate it from the homologous Glu1 domain (Fig. 2), it was not possible to identify specific amino acids or sequence motifs that are important for dhurrin hydrolysis. Consequently, this Dhr1 domain was split into two subdomains, which are represented in chim 15 (23-amino acid-long C-terminal subdomain) and chim 16 (30-amino acid-long N-terminal subdomain), to determine which subdomain was important for dhurrin hydrolysis. The substrate specificity and kinetic data from these two chimeras unequivocally showed that the 30-amino acid-long subdomain, which contains 10 amino acid substitutions, played a far greater role in dhurrin hydrolysis than the 23-amino acid-long subdomain. This is clearly evident in the fact that chim 16 hydrolyzes dhurrin 12 times better than chim 15 (Table III). These data are also consistent with the modeling data in that none of the 23 amino acids from the extreme C terminus of Dhr1 or Glu1 maps to and around the active site of the modeled three-dimensional structures of these enzymes. Therefore, the 8 substitutions and 4-amino acid-long addition that separate Glu1 from Dhr1 at the extreme C terminus has a rather small and probably indirect effect on substrate specificity.
The substrate specificity and kinetic data clearly suggested that one or more of the 10 amino acid substitutions in the 30-amino acid-long Dhr1 subdomain in chim 16 plays a key role in dhurrin hydrolysis. Again, to bring further clarity to the specific site(s) responsible for dhurrin hydrolysis, the 30-amino acid-long subdomain was divided into two segments after leaving out the invariant region GIVYVDR separating them. The resulting two Dhr1 peptides 462 SSGYTERF 469 and 477 ENG-CERTMKR 486 were used to replace their homologues in Glu1 by domain swapping, which yielded chim 21 and 22, respectively (Table I and Fig. 2). The substrate specificity and kinetic data obtained with these two chimeras indicated that chim 21 hydrolyzed dhurrin and had the best kinetic properties, having the lowest K m and highest k cat and efficiency coefficient among five Glu1/Dhr1 chimeras (Table III). Thus, the Dhr1 peptide 462 SSGYTERF 469 alone is sufficient to enable Glu1 to hydrolyze dhurrin when it replaces the homologous peptide 466 FAG-FTERY 473 of Glu1. Further support for the importance of pep-tide 462 SSGYTERF 469 in dhurrin hydrolysis came from three Dhr1/Glu1 chimeras. Of these, only chim 23 hydrolyzed dhurrin with kinetic constants and catalytic efficiency close to those of wild type Dhr1 (Table III; Fig. 4B, lane 11), and it was the only chimera in which the peptide 462 SSGYTERF 469 was not replaced by its Glu1 homologue. Indeed, when this peptide was replaced with its Glu1 homolog in Dhr1/Glu1 chimeras 17 and 39, their relative efficiency in dhurrin hydrolysis was 7.6to 12-fold lower than that of wild type Dhr1 (Table III). It is clear from these data that one or more of the 4-amino acid substitutions in peptide 462 SSGYTERF 469 are in the aglycone binding pocket of the active site, and they are necessary for the binding of dhurrin in correct angle and steric complementarity for hydrolysis. We postulate that Ser-462-Ser-463 and Phe-469 are the key residues because they are unique to Dhr1, and they and their homologues map to the active site in the modeled ␤-glucosidases (Fig. 5). Moreover, Ser-462-Ser-63 are ideal candidates to form hydrogen-bonding interactions through their side chain -OH group with the -OH group in the aglycone (p-hydroxy-(S)-mandelonitrile) moiety of dhurrin. Such interactions may be important in the attainment and stabilization of the transition state by the enzyme-dhurrin complex during the glycosylation step of hydrolysis. These same interactions, when coupled with those involving certain Glu1-specific residues on the Glu1 side of the primary structure, may explain the improved catalytic efficiency of the Glu1/Dhr1 chimeric enzymes toward pNPGlc and oNPGlc, since the aglycones of these artificial substrates are similar in size and shape to that of dhurrin. The role of the fourth amino acid substitution, Tyr-465 in Dhr1 and Phe-469 in Glu1, in dhurrin hydrolysis is probably indirect, if any, because this site is near but does not map to the presumptive aglycone binding pocket of the modeled enzymes. Interestingly, two of the four sites in the Glu1 and Glu2 homologs of the Dhr1 peptide 462 SSGYTERF 469 are also different, Phe-466 and Phe-469 in Glu1 and Tyr-466 and Tyr-469 in Glu2, and Glu1 and Glu2 differ in substrate specificity. The former hydrolyzes 6BNGlc and a variety of nitrophenyl ␤-glycosides, whereas the latter does not hydrolyze 6BNGlc at all and hydrolyzes nitrophenyl ␤-glycosides 5-6 times less efficiently than Glu1 (27). Moreover, the three-dimensional structure of white mustard myrosinase provided information about key residues involved in substrate binding, although the aglyconebinding site could not be directly identified. However, the docking of sinigrin into the active site suggested that the aglycone moiety is located in a hydrophobic pocket that is formed by residues Phe-331, Phe-371, Phe-473, Ile-257, and Tyr-330 (19). The Phe-473 of myrosinase is homologous with Tyr-473 of Glu1 and Phe-469 of Dhr1, which are bold-faced in peptides 466 FAG-FTERY 473 and 462 SSGYTERF 469 , respectively, supporting our hypothesis that these are one of the key residues determining aglycone recognition and, thus, substrate specificity.
The mechanism by which the Dhr1 peptide 477 ENGCERT-MKR 486 in chim 22 contributes to dhurrin hydrolysis is not apparent, although in this case the catalytic efficiency coefficient (k cat /K m ) is less than one-third that of chim 21 (Table III), which is still substantial. In other words, it is not possible to predict which of the five amino acid substitutions that are in this peptide imparts dhurrin hydrolysis capability to chim 22 because none of these five sites map to the active site of the modeled enzymes nor have their homologues been directly implicated in substrate binding and catalysis in the literature. Thus, it is conceivable that one or more of these sites have indirect global effects on the structure of the active site for the binding of dhurrin in a partially correct angle and steric complementarity for hydrolysis.
Our data from three Dhr/Glu1 chimeras (chim 17, 39, and 23) unequivocally showed that none of these three chimeras hydrolyzed DIMBOAGlc or any of the artificial substrates tested (Table II and Table III ; Figs. 3, B-C, lanes 8 -10, and 4A, lanes 10 -12), whereas their Glu1/Dhr1 reciprocals (chim 2, 16, and 15) hydrolyzed these very same substrates (Table III; Fig. 4, A-B, lanes 4 -6). Taken together, these data suggest that there are no residues in the extreme 47-amino acid-long C-terminal region of Glu1 that affect substrate hydrolysis qualitatively. However, this does not rule out the contribution of certain sites in the Glu1 C-terminal region to substrate binding and the rate of hydrolysis in combination with residues that reside elsewhere in the primary structure. The fact that the catalytic efficiency of all five Glu1/Dhr1 chimeras decreased in DIM-BOAGlc hydrolysis but increased in pNPGlc and oNPGlc hydrolysis suggests that one or more amino acid substitutions at the Dhr1 C-terminal region produce interactions favoring quantitatively nitrophenyl glucoside hydrolysis but not DIM-BOAGlc. This is not surprising because the shape and size of the nitrophenyl moiety has greater similarity to the aglycone of dhurrin than to the much bulkier DIMBOA (Fig. 1). The question of what site and in which domain in Glu1 and Glu1/Dhr1 chimeras is (or are) the residue(s) that determine the hydrolysis of DIMBOAGlc and artificial substrates (e.g. oNPGlc, pN-PGlc, MUGlc, etc) cannot be answered from the available data. However, we postulate that such residues are among the Glu1specific amino acid substitutions in the region spanning amino acids Leu-330 -Asn-450 based on preliminary data from a Dhr1/Glu1 chimera and modeling studies. In the aforementioned Dhr1/Glu1 chimera, the C-terminal half of Dhr1 (Ile-328 -Asn-514) was replaced with the homologous Glu1 C-terminal half (Leu-330 -Pro-512), and this chimera exhibited qualitatively the substrate specificity of Glu1. Our homology modeling data indicate that 12 of the 51 amino acid substitutions that are separating Glu1 from Dhr1 in this swapped C-terminal half map to the presumptive aglycone binding pocket of the active site. These substitutions are (numbering by the Dhr1 sequence position where the residue on the left occurs in Dhr1, and the one on the right in Glu1) N366K, A367P, T372M, A375P, N378Y, M407I, I410V, G413K, D414E, S462F, S463A, and F469Y (Fig. 2). The last three of these sites are postulated to be essential for dhurrin hydrolysis (see above), whereas one or more of the remaining nine are likely to be essential for the hydrolysis of DIMBOAGlc and other substrates. The precise definition of specific sites and residues required for the hydrolysis of DIMBOAGlc and other substrates will await the results of future site-directed mutagenesis studies.
One of the significant findings in this study was that the physiological substrates DIMBOAGlc and dhurrin, respectively, of Glu1 and Dhr1, were potent competitive inhibitors of their heterologous enzymes. The K i values of 0.076 mM for dhurrin and of 0.009 mM for DIMBOAGlc indicate that each substrate binds to the ground state of the heterologous enzyme with high affinity, but the hydrolysis of the ␤-glycosidic bond does not occur. In fact, the hydrolysis of dhurrin by all three Dhr1/Glu1 chimeras is also inhibited by DIMBOAGlc. It should be pointed out that the structures and substituents of the two aglycones, DIMBOA and p-hydroxy-(S)-mandelonitrile, are different (Fig. 1), the former being bulkier than the latter. The question of why DIMBOAGlc is not hydrolyzed by Dhr1 or of why dhurrin is not hydrolyzed by Glu1 in view of tight binding has only speculative answers at this time. The most plausible explanation is that the binding, although tight, does not position the glycosidic bond in the correct angle and distance (ϳ2.5 Å) with respect to the nucleophile or the acid catalyst in the first (i.e. glycosylation) step of the reaction, or it does not allow the enzyme-substrate complex to attain a transition state energetically favorable for hydrolysis. Thus, although substrate binding is the first essential step in hydrolysis, it is not sufficient for it unless the binding positions the ␤-glycosidic bond in the correct angle and distance with respect to the two catalytic glutamic acids. The above question also has bearing on the evolution and existence of two distinct ␤-glucoside biosynthesis pathways and chemical defense compounds in two closely related plant genera, Zea and Sorghum, which are thought to have diverged from a common ancestor 25-30 million years ago. For example, the dhurrin biosynthesis pathway starts with the amino acid tyrosine as the precursor and produces a cyanogenic ␤-glucoside (dhurrin) as the end product, whereas the DIMBOAGlc biosynthesis pathway starts with indole (a tryptophan analogue) as the precursor and produces a hydroxamic acid glucoside (DIMBOAGlc) as the end product. Which pathway did the common ancestor of sorghum and maize have and when, how, and why was another ␤-glucoside pathway and defense compound "invented" in one of the lineages are important questions for the evolutionary biologist to answer.
In conclusion, we were able to broaden the substrate specificity of the maize Glu1 isozyme (DIMBOA-glucosidase) to hydrolyze the sorghum natural substrate dhurrin and improve its catalytic efficiency toward the artificial substrate pNPGlc 1.5-4.4-fold and oNPGlc 1.5-3.1-fold. This was accomplished by replacing a 47-amino acid-long C-terminal domain of Glu1 and its smaller segments with the homologous Dhr1 domain and its smaller segments. The shortest Dhr1 peptide to enable Glu1 to hydrolyze dhurrin was eight amino acids long, differing by four amino acid substitutions, three of which mapped to the active site of the modeled enzymes. Although all of the five Glu1/Dhr1 chimeric enzymes hydrolyzed both dhurrin and DIMBOAGlc, none of them either equaled or exceeded their parental enzymes in terms of catalytic efficiency for these natural substrates. However, with one exception (chim 16) they were better DIMBOA-glucosidases than dhurrinase, having 65 to 88% catalytic efficiency of Glu1. In general, DIMBOAGlc hydrolysis and dhurrin hydrolysis efficiencies were negatively correlated. In contrast, three Dhr1/Glu1 chimeras (reciprocals of chim 2, 15, and 16) hydrolyzed dhurrin only, with catalytic efficiency approaching that of Dhr1 when the Dhr1 peptide 462 SS-GYTERF 469 was not replaced by its Glu1 homologue. These data, taken together, show that the C-terminal domains of ␤-glucosidases contain sites that are necessary for aglycone recognition and binding, but they alone are not sufficient to determine substrate hydrolysis. In our specific model system, the substrate specificity differences for dhurrin hydrolysis between maize and sorghum ␤-glucosidases are very likely due to four amino acid substitutions in the C-terminal Dhr1 peptide 462 SSGYTERF 469 . However, the hydrolysis of DIMBOAGlc, two NPGlcs, MUGlc, and 6BNGlc are likely determined by one or more of the 8 sites, where Glu1 and Dhr1 differ, mapping to the aglycone pocket and residing in the Glu1 C-terminal half spanning residues Leu-330 -Glu-416. Our future research will be focused on precise definition of specific residues that make up the aglycone binding pocket and, thus, the substrate specificity using three-dimensional studies on enzyme-aglycone and enzyme-competitive inhibitor complexes by x-ray crystallography as well as site-directed mutagenesis targeting candidate sites identified by domain swapping, x-ray crystallography, and modeling.