Crystal Structures of β-Primeverosidase in Complex with Disaccharide Amidine Inhibitors*

Background: β-Primeverosidase hydrolyzes disaccharide conjugates in tea leaves and releases aromatic volatiles. Results: Crystal structures of β-primeverosidase in complex with disaccharide analogues were determined. Conclusion: β-Primeverosidase has a disaccharide-binding site, and the molecular architecture for disaccharide recognition was elucidated. Significance: β-Primeverosidase adapts the active site to the 6-O-β-d-xylopyranosyl modification of β-glucoside and contributes multiple volatile emissions in tea leaves. β-Primeverosidase (PD) is a disaccharide-specific β-glycosidase in tea leaves. This enzyme is involved in aroma formation during the manufacturing process of oolong tea and black tea. PD hydrolyzes β-primeveroside (6-O-β-d-xylopyranosyl-β-d-glucopyranoside) at the β-glycosidic bond of primeverose to aglycone, and releases aromatic alcoholic volatiles of aglycones. PD only accepts primeverose as the glycone substrate, but broadly accepts various aglycones, including 2-phenylethanol, benzyl alcohol, linalool, and geraniol. We determined the crystal structure of PD complexes using highly specific disaccharide amidine inhibitors, N-β-primeverosylamidines, and revealed the architecture of the active site responsible for substrate specificity. We identified three subsites in the active site: subsite −2 specific for 6-O-β-d-xylopyranosyl, subsite −1 well conserved among β-glucosidases and specific for β-d-glucopyranosyl, and wide subsite +1 for hydrophobic aglycone. Glu-470, Ser-473, and Gln-477 act as the specific hydrogen bond donors for 6-O-β-d-xylopyranosyl in subsite −2. On the other hand, subsite +1 was a large hydrophobic cavity that accommodates various aromatic aglycones. Compared with aglycone-specific β-glucosidases of the glycoside hydrolase family 1, PD lacks the Trp crucial for aglycone recognition, and the resultant large cavity accepts aglycone and 6-O-β-d-xylopyranosyl together. PD recognizes the β-primeverosides in subsites −1 and −2 by hydrogen bonds, whereas the large subsite +1 loosely accommodates various aglycones. The glycone-specific activity of PD for broad aglycone substrates results in selective and multiple release of temporally stored alcoholic volatile aglycones of β-primeveroside.

␤-Primeverosidase (PD) is a disaccharide-specific ␤-glycosidase in tea leaves. This enzyme is involved in aroma formation during the manufacturing process of oolong tea and black tea. PD hydrolyzes ␤-primeveroside (6-O-␤-D-xylopyranosyl-␤-Dglucopyranoside) at the ␤-glycosidic bond of primeverose to aglycone, and releases aromatic alcoholic volatiles of aglycones. PD only accepts primeverose as the glycone substrate, but broadly accepts various aglycones, including 2-phenylethanol, benzyl alcohol, linalool, and geraniol. We determined the crystal structure of PD complexes using highly specific disaccharide amidine inhibitors, N-␤-primeverosylamidines, and revealed the architecture of the active site responsible for substrate specificity. We identified three subsites in the active site: subsite ؊2 specific for 6-O-␤-Dxylopyranosyl, subsite ؊1 well conserved among ␤-glucosidases and specific for ␤-D-glucopyranosyl, and wide subsite ؉1 for hydrophobic aglycone. Glu-470, Ser-473, and Gln-477 act as the specific hydrogen bond donors for 6-O-␤-D-xylopyranosyl in subsite ؊2. On the other hand, subsite ؉1 was a large hydrophobic cavity that accommodates various aromatic aglycones. Compared with aglycone-specific ␤-glucosidases of the glycoside hydrolase family 1, PD lacks the Trp crucial for aglycone recognition, and the resultant large cavity accepts aglycone and 6-O-␤-D-xylopyranosyl together. PD recognizes the ␤-primeverosides in subsites ؊1 and ؊2 by hydrogen bonds, whereas the large subsite ؉1 loosely accommodates various aglycones. The glycone-specific activity of PD for broad aglycone substrates results in selective and multiple release of temporally stored alcoholic volatile aglycones of ␤-primeveroside.
Aglycones of ␤-primeverosides are benzenoids and monoterpenoids, which are volatile organic compounds with a well known role in the defense of plants against stress and herbivore attacks (6 -8). Benzyl alcohol, 2-phenylethanol, linalool, and geraniol have antimicrobial activities against Gram-positive and Gram-negative bacteria and fungi (9 -14). Wounding and drought stress of tea leaves raise the levels of these compounds to ϳ10to 40-fold (15). Plucking, withering, and shaking procedures during the manufacture of oolong tea wounds the fresh leaves to secrete volatile compounds as sweet, flower, and honey aromatic flavors. During this process PD hydrolyzes ␤-primeverosides and releases aromatic aglycones as volatile defensive compounds against stress.
The substrate specificity of PD raises two basic questions: how does PD recognize the ␤-1,6-Xyl-␤-Glc glycone for specific hydrolysis of ␤-primeverosides, and how does PD accept four structurally varied aglycones? X-ray crystallography of PD should provide crucial information to elucidate the molecular mechanism of the substrate specificity of PD. We have established large scale expression using a baculovirus-insect cell system to obtain a sufficient amount of PD for crystallization (20). PD is a 61-kDa glycoprotein and its cDNA encodes secretion of a signal peptide of 28 residues at the N-terminal. Recombinant PD has been shown to be active and stable with post-translational modification that involves glycosylation and secretion with cleavage of the signal peptide. A synthetic affinity adsorbent was prepared for purification of PD using N-␤-primeverosylamidines as the core structure of the ligand. The affinity chromatography yielded large amounts of PD with a purity sufficiently high for crystallographic study.
N-␤-Primeverosylamidine is a derivative of ␤-glycosylamidine (21)(22)(23), which has the same sugar glycone as the natural substrate. N-␤-Primeverosylamidines are suitable substrateanalogue inhibitors for probing the substrate recognition of PD because the inhibitory activity of N-␤-primeverosylamidines exhibits glycone selectivity and broad aglycone specificity consistent with the substrate specificity of PD. N-␤-Primeverosylamidines with two aglycones, 2-phenyl-N-  (20). The monosaccharide glycoside analogue, ␤-glucosylamidine, has no inhibitory activity at 500 M, consistent with the fact that PD barely hydrolyzes ␤-D-glucopyranosides. Therefore, x-ray crystal analysis of PD in complex with N-␤-primeverosylamidine should clarify the mechanisms of the unique disaccharide glycone-aglycone recognition of PD.
This article describes studies on three-dimensional structures of PD in apo and complex forms with N-␤-primeverosylamidines. We crystallized PD after deglycosidated treatment to avoid heterogeneity, and obtained high-quality crystals. We then elucidated the structural basis of the disaccharide specific and aglycone nonspecific recognition mechanisms of PD in comparison with monosaccharide-specific ␤-glucosidase structures.

EXPERIMENTAL PROCEDURES
Chemicals-N-␤-Primeverosylamidines and affinity adsorbent were prepared by the method as described previously (20,21).
Production and Purification of PD-The full-length cDNA of PD (GenBank TM , AB088027) including the N-terminal signal sequence was cloned and expressed in a baculovirus-insect cell system as described previously (20). PD was purified by SP-Sepharose cation exchange chromatography and ConA-Sepharose. Affinity chromatography with N-␤-primeverosylamidine adsorbent was performed according to the procedure described previously (20). In brief, this included adsorption with 10 mM Tris-HCl (pH 7.0), washing with the same buffer, and elution with 10 mM acetate buffer (pH 4.0). The purified PD (10 mg/ml) was prepared as a stock solution in 10 mM sodium acetate (pH 4.0) at 4°C for use. The purified PD was treated with ␣-mannosidase before crystallization.
Crystallization-In initial screening, PD yielded only low quality needle crystals due to structural heterogeneities of posttranslational glycosylation. We performed partial digestion of glycans with ␣-mannosidase for further crystallization trials (20). Deglycosylated PD apo-crystals were grown by the sittingdrop vapor diffusion method at 20°C, using a mixture of 1.5 l of protein (10 mg/ml in 10 mM sodium acetate buffer) and 1.5 l of the reservoir solution containing 0.01% Triton X-100, 21% PEG 4000, 0.3 M ammonium acetate, and 0.1 M sodium citrate (pH 5.5). Cryoprotection was achieved by soaking in 27.5% (w/v) PEG 4000, 0.3 M ammonium acetate, 0.125 M sodium citrate (pH 6.5) for 1 h, followed by freezing in liquid nitrogen. To prepare the PD complexes, the grown crystals were soaked in 30 mM PhPA or 3 mM BsPA of the cryoprotectant solution as described above.
Data Collection and Refinement-Data were collected using an in-house x-ray generator, RIGAKU FR-E equipped with a RAXIS IV detector. Data collection was performed at the wavelength 1.5418 Å, and 0.5°oscillation images of 300-s exposure were collected for 90°(180 images) under 100 K cryo-cooling. Image processing and integrations were performed with MOSFLM (24) and scaled with SCALA, part of the CCP4 package (25). Molecular replacement was performed by MOLREP (26) using the crystal structure of the ␤-glucosidase of white clover (PDB code 1CBG) as the search model, and two monomers were observed in the asymmetric unit. ARP/wARP (27) was used for automatic model building, with subsequent refinement using the maximum likelihood method in REFMAC5 (28) and PHENIX (29), using hydrogen riding model and isotropic B-factor refinement. Visual inspection of the models was per-formed using COOT (30). The data collection and refinement statistics are summarized in Table 1. In the refinement process, N-or C-terminal residues and side chains with missing electron density were deleted. One inhibitor was found in each active site, and one PhPA molecule was incidentally found in the solvent region of the PhPA complex crystal. The Ramachandran plot was assessed using RAMPAGE (31). Superposed structures were calculated by LSQKAB of CCP4 (25). The figures were prepared using PyMOL (Schrödinger, LLC) (32).

RESULTS
The crystal structures of PD were determined in apo and in complexes with N-␤-primeverosylamidine PhPA and BsPA at resolutions of 1.9, 1.8, and 1.8 Å, respectively. The protein-fold of PD is a classical (␤/␣) 8 barrel, and PhPA/BsPA were located in the substrate-binding site at the center of the barrel (Fig. 1B). The electron densities of PhPA and BsPA were observed in both complex crystals (Fig. 2, A and B), including the ␤-1,6-Xyl-␤-Glc moiety as well as a large aglycone moiety of BsPA. The phenyl group (aglycone) of PhPA showed ambiguous F o Ϫ F c electron densities, and therefore the phenyl group was put in a position that partially fitted to the electron densities. The 2F o Ϫ F c map of the phenyl ring was ambiguous even after refinement, and the ambiguity was the same in each monomer in the asymmetric unit. There was no significant change in the (␤/␣) 8 -fold and loops among the apo and two complex structures.
Electron densities of post-translationally modified glycans were found at two NX(S/T) sites, Asn-35 and Asn-424. One GlcNAc was attached to Asn-424 by N-␤-glycosidic bonds, whereas dimeric GlcNAc with ␤-1,4-link were attached to Asn-35. The core GlcNAc on Asn-35 was further modified with a ␣-1,3-linked fucose by the N-glycosylation system of secretion proteins in insect cells (33).
Architectures of the Substrate-binding Site-The substratebinding site can be partitioned into three subsites ( Fig. 2A); binding sites for the ␤-1,6-Xyl moiety (subsite Ϫ2), ␤-Glc moiety (subsite Ϫ1), and aglycone (subsite ϩ1). The substratebinding site was shaped like a funnel and was ϳ18 Å deep, with the entrance of the pocket being 14 Å. The pocket had a large entrance at the surface of PD, and its inner space gradually narrowed toward the bottom. PhPA and BsPA were bound in the pocket by a sharp bend of the ␤-Glc moiety at the bottom of the pocket, with the ␤-1,6-Xyl moiety extending toward the entry of the pocket. The catalytic resides Glu-203 (acid/base) and Glu-416 (nucleophile) were located at the bottom of the pocket.  The structure of PhPA and BsPA showed well defined hydrogen bonding networks in subsites Ϫ2 and Ϫ1 (Fig. 2C). The heteroatoms of N-␤-primeverosylamidines, oxygen of the ␤-1,6-Xyl-␤-Glc moiety, and nitrogen of the amidino group had well defined electron densities. All the hydroxy groups of the ␤-1,6-Xyl-␤-Glc moiety formed hydrogen bonds with amino acids in subsites Ϫ2 and Ϫ1. All contacts and their distance are shown in a schematic diagram (Fig. 2C). The conformation of ␤-1,6-Xyl was in a relaxed 4 C 1 chair in subsite Ϫ2, whereas that of ␤-Glc in subsite Ϫ1 was distorted. The ␤-Dglucopyranose ring was in an 4 E-like conformation between 4 C 1 and 4 E, in which the anomeric carbon C1 was close to the plane of the glucopyranose ring. This distortion was probably induced by the 3.2-Å contact of the nucleophilic oxygen of Glu-416. Thus, this conformation is likely to reflect an early state structure of the ES complex formation, preceding the nucleophilic attack occurring in the 4 E transition state.
Disaccharide Glycone Recognition in Subsite Ϫ2 and Subsite Ϫ1-Subsite Ϫ2 held the ␤-1,6-Xyl moiety by six amino acids, Glu-470, Ser-473, and Gln-477 with hydrogen bonds and Val-386, Phe-389, and Phe-479 with hydrophobic contacts (Fig. 2C). These six residues were almost certainly responsible for the strict recognition of the ␤-1,6-Xyl moiety in the disaccharide specificity of PD. Glu-470 made hydrogen bonds with the carboxylate oxygen of the side chain and the amide carbonyl oxygen of the main chain to 2-OH and 3-OH of the ␤-1,6-Xyl moiety. Ser-473 was hydrogen bonded to 3-OH and 4-OH by a side chain hydroxy at 3.1 and 2.8 Å distance, respectively, whereas Gln-477 was hydrogen bonded with 4-OH by the side chain oxygen. Val-386, Phe-389, and Phe-479 covered the hydrophobic surface (C5) of the ␤-1,6-Xyl moiety with van der Waals interactions at 4.1, 4.2, and 3.7 Å, respectively. Other interactions, such as stacking with aromatic amino acid residues, was not observed in subsite Ϫ2.
Subsite Ϫ1, the binding site for the ␤-Glc moiety, consisted of 8 amino acid residues including the catalytic glutamates, Glu-203 and Glu-416 as the acid/base and nucleophile, respectively (Fig. 2C). The side chains of Gln-53, His-157, Asn-202, Tyr-345, Glu-416, Trp-463, Glu-470, and Trp-471 were hydrogen bonded with the hydroxy groups of the ␤-Glc moiety. Asn- 202 and Glu-416 were bonded with 2-OH at a distance of 2.7 and 3.2 Å, and Gln-53 and His-157 with 3-OH at a distance of 2.6 and 3.0 Å, respectively. Gln-53 also formed a hydrogen bond between its NH and 4-OH at 2.9 Å. Trp-463, Glu-470, and Trp-471 were bonded with 4-OH at hydrogen bonding distances. The aromatic face of Trp-463 was adjacent to the C3 and C5 of ␤-Glc. These interactions were almost conserved to that reported in GH1 ␤-glucosidases, except for Glu-470. Glu-470 was located between subsites Ϫ2 and Ϫ1, and its oxygen atom was hydrogen bonded, not only with 4-OH of ␤-Glc but also with 2-OH of ␤-1,6-Xyl in subsite Ϫ2.
Interactions between disaccharide and amino acids in subsites Ϫ2 and Ϫ1 were identical in the complex structures of PhPA and BsPA. Subsites Ϫ2 and Ϫ1 recognize the ␤-1,6-Xyl-␤-Glc moiety in the same manner, independent of the aglycone structures. The amidino group of N-␤-primeverosylamidines and Glu-203 formed electrostatic interactions (Fig. 2C), which probably enhanced the affinities of N-␤-primeverosylamidines in addition to the hydrogen bonds of the ␤-1,6-Xyl-␤-Glc moiety.
The Binding of Aglycone in Subsite ϩ1-The aglycone moiety of BsPA showed clear electron density in subsite ϩ1 (Fig. 2B). Subsite ϩ1 formed a sufficiently large space to accept the bulky, artificial aglycone of BsPA. The aglycone includes spacer (CH 2 ) 3 S and linker (N-benzyl succinimide) structures as synthetic connections used in the preparation on the affinity absorbent (21) and is therefore larger than the natural aglycones of ␤-primeverosides. In subsite ϩ1, the spacer (CH 2 ) 3 S led to the entry of the substrate binding pocket, with the N-benzyl turned into the pocket by bending at the succinimide. The cavity of subsite ϩ1 was filled with the large aglycone of BsPA.

DISCUSSION
The substrate specificity of PD has three properties (16): (i) strict selectivity for ␤-1,6-Xyl as a ␤-1,6-linked sugar moiety of ␤-primeveroside, (ii) broad aglycone specificity, and (iii) rather low activities for ␤-glucosides. In this study, we determined the crystal structure of PD in complexes with substrate analogues, and investigated the substrate-binding site to determine the amino acid residues responsible for substrate recognition. Based on the crystal structures, we formulated a structural basis for the substrate specificities of PD.
Loose Aglycone Recognition in Subsite ϩ1-The crystal structure in the complex with BsPA showed that subsite ϩ1 forms a large hydrophobic cavity to accept naturally occurring aglycones. To investigate the structural basis and differences between loose aglycone recognition of PD and specific aglycone recognition of GH1 ␤-glucosidases, we compared our crystal structure with ␤-glucosidase from Zea mays (DG) because aglycone binding of this enzyme is well studied in complexes with DIMBOA (2,4-dihydroxy-7-methoxy-1,4-benzoxazin-3-one) ␤-D-glucopyranoside (PDB code 1E56) (34). The barrel-fold of PD was very similar to that of DG with an overall root mean square deviation of 1.06 Å in the superimposed structure. The subsite ϩ1 of DG has Phe-198, Phe-205, Trp-378, and Phe-466 as crucial residues for aglycone binding. These sites are also involved in aglycone binding of dhurrinase reported with high resolution structures (19). The corresponding amino acid residues of PD were investigated in the crystal structure of subsite ϩ1.
Gly-210, Leu-217, Ala-387, and Leu-472 were identified in PD, and corresponded to the aglycone-recognizing residues of DG, Phe-198, Phe-205, Trp-378, and Phe-466 (34), respectively (Fig. 3C). The residues in PD (Gly, Leu, and Ala) were sterically smaller than the corresponding residues of DG (Phe, Trp), and these smaller residues formed a large cavity in subsite ϩ1 of PD. In particular, Ala-387 is remarkable because the corresponding amino acid in DG is Trp, a residue that is conserved in almost all GH1 ␤-glucosidases (35) and directly interacts with aglycones in crystal structures of plant GH1 ␤-glucosidases (17-19, 34, 36). These results suggest that the large cavity with the smaller residue is the basis for the broad aglycone specificity of PD, and simultaneously suggest that the large cavity allows acceptance of aglycones with bulky disaccharide-glycone, which could not be accommodated in the narrow entry cleft of DG.
Substrate Recognition of PD-A scheme for the substrate recognition mechanism indicated by the crystal structure is summarized in Fig. 4. PD recognizes the ␤-1,6-Xyl-␤-Glc (␤-primeverosyl) moiety of subsites Ϫ2 and Ϫ1 by hydrogen bonds with polar residues (Fig. 4A). Subsite Ϫ2 is attached tightly to the ␤-1,6-Xyl moiety and disfavors other ␤-1,6-linked sugars. Subsite ϩ1 is sufficiently large to accept various naturally occurring aglycones of ␤-primeverosides, and also accommodates the extremely bulky artificial aglycone of BsPA in the crystal structure.
As observed in the ambiguous electron density of phenyl group of PhPA, aglycone binding in subsite ϩ1 is loose. PD recognizes ␤-primeveroside by subsites Ϫ2 and Ϫ1, with the binding of the ␤-1,6-Xyl-␤-Glc moiety being sufficiently stable to fix the position of the substrate to the catalytic position even in the absence of specific interactions with aglycone (Fig. 4B). The ␤-1,6-Xyl-␤-Glc of PhPA and BsPA had exactly the same structure despite varied aglycone structures. On the other hand, aglycones should associate nonspecifically with hydrophobic amino acids and the hydrophobic side of ␤-1,6-Xyl in subsite ϩ1. The nonspecific association could enhance the specific polar interaction of ␤-1,6-Xyl in subsite Ϫ2, consistent  with the rather potent inhibition (K i ϭ 26 M) of BsPA filling subsite ϩ1 with the large aglycone. The fact that the hydrophobic side of ␤-1,6-Xyl faces the aglycone suggests that subsites Ϫ2 and ϩ1 could bind substrate in a concerted manner. The homology modeling of a diglycosdase cloned from Dalbergia nigrescens expected that the 6-O-linked sugar would be faced against the aglycone (37).
Monosaccharide ␤-glucosides are poor substrates for PD for two reasons: the absence of a ␤-1,6-Xyl moiety, and loose aglycone recognition in subsite ϩ1 (Fig. 4C). Although PD conserves binding residues for ␤-Glc in subsite Ϫ1, it probably requires an additional interaction to fix the position of ␤-glucosides for catalysis. The aglycone binding in subsite ϩ1 contributes to glycone binding in ␤-glucoside hydrolysis of GH1 ␤-glucosidases. Site-directed mutagenesis of GH1 ␤-glucosidases (38,39) resulted in mutations of Trp in subsite ϩ1 that markedly reduced activities and modified glycone specificity. Trp is missing in subsite ϩ1 of PD, and therefore PD barely hydrolyzes ␤-glucosides, whereas the alternative small residue (Gly, Ala, and Leu) forms a large hydrophobic cavity to accept various aglycones and ␤-1,6-Xyl moiety. In the absence of ␤-1,6-Xyl, subsite ϩ1 would lose the hydrophobic wall composed of ␤-1,6-Xyl, and the resultant large cavity would disfavor the binding of ␤-glucosides by diminishing the binding stability.
The glycone-specific and aglycone broadly specific activity of PD allows selective and multiple activation of temporally stored aglycones of ␤-primeveroside. The loose recognition of aglycone in subsite ϩ1 can activate and release all four volatile compounds. This multiple emission can be more effective than single emission in defensive roles because aglycones have different antimicrobial spectra (9 -14). Little or no activity of PD for ␤-glucosides is also important because phytohormones and toxic compounds are stored as glucose conjugates in the cell, and their activation is regulated by specific ␤-glucosidases (40 -44). PD would barely hydrolyze these glucose conjugates even in the mixture of compounds in broken cells of wounded leaves, and avoids unnecessary activation of phytohormones. The system with ␤-primeveroside and PD would be efficient for controlling defensive volatile emission because PD can activate multiple volatile compounds by one-step specific hydrolysis of the ␤-glycosidic bond of ␤-primeverosides.