Three-dimensional structure of the barley beta-D-glucan glucohydrolase in complex with a transition state mimic.

Glucophenylimidazole (PheGlcIm), a tetrahydroimidazopyridine-type inhibitor and 4H3 conformer mimic of a glucoside, binds very tightly to a barley beta-d-glucan glucohydrolase, with a Ki constant of 2 x 10(-9) m and a DeltaG of 51 kJ mol(-1). PheGlcIm binds to the barley beta-d-glucan glucohydrolase approximately 2 x 10(5) times tighter than laminarin, which is the best non-synthetic ground-state substrate found so far for this enzyme, 10(6) times tighter than 4-nitrophenyl beta-d-glucopyranoside, and 2 x 10(7) tighter than glucose. The three-dimensional structure of the beta-d-glucan glucohydrolase with bound PheGlcIm indicates that the complex resembles a hypothetical transition state during the hydrolytic cycle, that the enzyme derives substrate binding energy from the "aglycone" portion of the ligand, and that it also reveals an anti-protonation trajectory for hydrolysis. Continuous electron densities at the 1.6 sigma level form between the three active site residues Asp95, His207, and Asp285, and the C6OH, C7OH, C8OH, and C9OH groups of PheGlcIm. These electron densities correspond to the most favorable interactions in the three-dimensional structure of the beta-d-glucan glucohydrolase-PheGlcIm complex and indicate atomic distances equal to or less than 2.55 A. The crystallographic data were corroborated with ab initio molecular orbital calculations. The data indicate that the 4E conformation of the glucose part of PheGlcIm is critical for tight binding and provide the first evidence for probable substrate distortion during catalysis by this enzyme.

The enzymic transformation of a substrate into product by glucoside hydrolases proceeds through a series of intermediates and oxocarbenium (cation) ion-like transition states and is mediated via distortion/relaxation cycles of a ground-state low energy 4 C 1 chair conformation of the substrate (1)(2)(3). Evidence for the existence of oxocarbenium ion-like transition states arises from secondary kinetic isotope effects and inhibition studies with transition state analogues (3)(4)(5)(6)(7).
Barley ␤-D-glucan glucohydrolase is a family GH3 1 glycoside hydrolase that catalyzes hydrolytic removal of non-reducing glucosyl residues from a broad range of ␤-D-glucans and ␤-Dglucooligosaccharides (8). In our previous crystallographic work we used conduritol B epoxide, 2,4-dinitrophenyl 2-deoxy-2-fluoro-␤-D-glucopyranoside (2F-DNPGlc) and 4 I , 4 III , 4 V -Strithiocellohexaose to identify catalytic amino acid residues and to define three key intermediates in the catalytic sequence (9). The glucose product of the reaction does not immediately diffuse away from the enzyme surface, but instead remains bound at subsite Ϫ1. It is likely that this bound glucose is displaced from the active site by the incoming substrate during the first stage of the next hydrolytic event (9,10). The second stage of the hydrolytic pathway involves C1-O glycosidic bond cleavage, which proceeds through a double displacement. The crystallographic analysis of a S-cellobiosyl-enzyme complex, supported by quantum mechanical modeling, suggested that the S-cellobiosyl-enzyme complex mimics an oxonium intermediate, rather than the enzyme-substrate complex (9). It has also been suggested that formation of a stable oxonium intermediate inhibits distortion of a glycosyl ring (11).
The data obtained in our previous work with the ␤-D-glucan glucohydrolase indicated that no significant distortion occurred either in the enzyme-product complex or in the covalent glycosyl-enzyme intermediate (9,10). Similarly, no significant distortions could be detected in sugar conformations at subsites Ϫ1 or ϩ1 of the S-glycosyl-enzyme complexes (9,10). However, crystallographic analyses of several other endo-acting glycoside hydrolases in complex with sugars indicate distortions of the glycosyl residue that is bound at the Ϫ1 subsite. It is also assumed that distortion during oxocarbenium ion-like transition state formation is a necessary part of catalysis (3).
To further investigate intermediates and to test whether ring distortions play a role during catalysis of the barley ␤-Dglucan glucohydrolase, the three-dimensional structure of the enzyme in complex with glucophenylimidazole (PheGlcIm) has now been solved. PheGlcIm is a representative of the glucoimidazoles (GlcIm), which are excellent chemical tools for studying three-dimensional structures that mimic transition states. For example, PheGlcIm is a nagstatin-type mimic of the reactive intermediate that can be isolated from culture filtrates of Streptomyces amakusaensis (12), or prepared synthetically; nagstatin is an inhibitor of N-acetyl-␤-D-glucosaminidase (13). PheGlcIm has a trigonal anomeric center attached to an exocyclic N atom, which corresponds to the "glycosidic heteroatom," and an endocyclic N atom of the tetrahydropyridine moiety. PheGlcIm possesses a conformation similar to that of gluconojiritetrazole, which in solid state or in D 2 O adopts a 4 H 3 conformation that is similar to 4 E (14,15). Tetrahydroimidazopyridine-type inhibitors do not contain syn lone pair electrons at the "anomeric" N atom, but possess an anti-oriented, doubly occupied, non-bonding orbital at this heteroatom and are therefore suited as inhibitors of anti-protonating glycoside hydrolases (15)(16)(17).
Barley ␤-D-glucan glucohydrolase has two subsite-binding sites, designated Ϫ1 and ϩ1, in its active site (9,10), and the structure of PheGlcIm suggested that the inhibitor might be a useful mechanistic probe for the investigation of the enzymebound reactive intermediate and thus the transition state. It was expected that the tetrahydroimidazopyridine moiety and the phenyl ring would be located at the Ϫ1 and ϩ1 subsites, respectively. The primary objectives of the present work were to analyze binding interactions of PheGlcIm in the active site of the barley ␤-D-glucan glucohydrolase by x-ray crystallography, to define the structure of a possible transition state complex, and to reconcile these findings with quantum mechanical modeling.
Enzyme Isolation and Purity-Barley ␤-D-glucan glucohydrolase isoenzyme ExoI was purified from a homogenate of 8-day-old seedlings as described (8). The purity of the enzyme was assessed by SDS-PAGE, where a single protein band was detected at high protein loadings, and by NH 2 -terminal amino acid sequence analysis (8).
Inactivation of ␤-D-Glucan Glucohydrolase by Castanospermine and Glucoimidazoles-Inactivation of ␤-D-glucan glucohydrolase by castanospermine, GlcIm, PheGlcIm, and PhethGlcIm was monitored at 37°C by incubating 1-2 nmol of the purified enzyme in 100 mM sodium acetate buffer, pH 5.25, containing 0.2% (w/v) 4NPGlc, 160 g/ml BSA, and 0 -64 M castanospermine or 0 -100 nM glucoimidazoles. Each inhibitor was tested at six concentrations, in a duplicate. The residual enzyme activity was monitored spectrophotometrically at 410 nm (8). The first and second order rate inhibition constants were determined by a proportional weighted fit, using a non-linear regression analysis (20) and the GraFit program (Erithacus Software Ltd. (21)).
Dependence of k cat K m Ϫ1 and K i Ϫ1 on pH-The effects of pH on catalytic efficiency and inhibitor binding were determined by incubating 0.5-1 nmol of purified enzyme at 37°C in 50 mM citric acid/100 mM sodium dihydrophosphate (McIlvaine) buffers (pH 4.0 -7.5) containing 160 g/ml BSA and appropriate substrate (4NPGlc) concentrations. The k cat K m Ϫ1 versus pH profile was calculated from time-course curves at low 4NPGlc concentration (0.17 mM and 1/9 K m ). Division of the pseudofirst-order rate constants by the enzyme concentration produced the final k cat K m Ϫ1 values (22,23). The pH of each reaction mixture was checked during the time course to make certain that no pH changes occurred during the assays. The K i values at each pH were estimated using concentrations of inhibitor that were 0.4 -3 times the K i and concentrations of 4NPGlc substrate that were 0.5-1 times the K m constant. Dixon plots were used to determine the final K i value at each pH. Substrate hydrolysis during both types of measurements never exceeded 10% of the initial substrate concentrations. S.E. values for assays were within the range of 6 -9%. The kinetic constants were calculated using non-linear regression analyses (20) and the GraFit program (21).
After ϳ2 h at 4 Ϯ 2°C, the crystal was transferred into solution A with 11% (v/v) glycerol as a cryo-protectant. The crystal was subsequently mounted on a goniometer and flash-frozen to 100 K in a stream of nitrogen gas (Oxford Instruments, Oxford, England). Data were collected to 2.62-Å resolution using a rotating copper anode x-ray generator (MacScience Co. Ltd., Yokohama, Japan) operating at 40 kV and 50 mA, and a Rigaku Raxis IV detector (Rigaku/MSC, The Woodlands, TX) fitted with capillary optics (AXCO, Melbourne, Australia). A total of 80 data frames were collected using 1°oscillations and 1-h exposure times on a R axis II detector with crystal to film distance set at 120 mm. The diffraction data were integrated, scaled, and reduced using the HKL program (25). Autoindexing determined that the primitive tetragonal crystals belonged to the space group P4 3 2 1 2. Model refinement was performed with CNS (26). Atomic B-factor values for the protein, water molecules, and carbohydrates were reset to an overall thermal parameter of 30 Å 2 . The residues within a 10-Å radius of the active site, excluding the bound glucose, were set to zero occupancy. The initial model used to locate the position of the protein in the unit cell was constructed with a restrained rigid body refinement technique from 15 to 3.0 Å. After a final round of the rigid body refinement, the crystallographic R work was 22.29%, R free was 27.40%, and the overall B value was 31.99 Å 2 . Geometrical positional refinement followed by individual B factor refinement was subsequently calculated using the ␤-D-glucan glucohydrolase model with a bulk solvent correction applied and a maximum likelihood method implemented. The electron density map was calculated from the observed structure factors and phases using the ␤-D-glucan glucohydrolase isoenzyme ExoI structure as the starting model (PDB accession code 1IEQ (9)) and MAPMAN package (27). PheGlcIm was located in the active site pocket by examining a difference Fourier electron density map at a level greater than 4 . This showed peaks for the two rings of the PheGlcIm that were manually built into the electron density, using the graphics program "O" (28) and refined. The PROCHECK program (29) was used to assess the geometric quality of the model, and the structure was found not to deviate from the ideal geometry. The final refinement statistics are summarized in Table II. Quantum Mechanical Calculations-Ab initio molecular orbital calculations were performed with the GAMESS-US program (30). A model of the active site was constructed from the side-chain groups of residues in contact with the inhibitor. Small molecule representations of the amino acid side chains were used; acetate for Asp 95 , Asp 285 , and Glu 491 , guanidinium for Arg 158 , methylammonium for Lys 206 , imidazole for His 207 , and indole for Trp 434 . All atoms of the inhibitor were included. The systems included 66 non-hydrogen atoms, and 49 -51 hydrogen atoms, depending on the protonation state of the inhibitor and acetate representing Glu 491 . Restrained geometry optimization at the HF/3-21G level was applied, in which non-hydrogen atoms were restrained to their position found in the x-ray crystallographic analysis. Atoms were allowed to move 0.5 Å before restraints were applied; a harmonic restraint with force constant of 0.5 aJ Å Ϫ2 (mdyn Å Ϫ1 ) was used.
Several protonation states of the models were considered. The acetate representing Glu 491 was either ionized (corresponding to a charge of Ϫ1), or neutral. The protonation states of the inhibitor PheGlcIm were represented either by the ionized state, which corresponded to a charge of ϩ1 and a proton on the ring N1 atom, or by a neutral form. Four configurations were considered as follows: 1, Glu 491 and PheGlcIm ionized; 2, Glu 491 ionized and PheImGlc neutral; and 3 and 4, Glu 491 and PheGlcIm neutral, with the formal HOCO angle of 0°(see Table  III).
Improved relative energies of configurations 1, 3, and 4 were calculated at the B3LYP/6 -31G(d) level (Gaussian Inc., Pittsburgh, PA (31)), using the GAUSSIAN 98 program (Revision A.7, Gaussian Inc. (32)). The electrostatic component of the protein environment was included by incorporating partial atomic charges for all residues within 8 Å of the inhibitor, excluding those that were specified above. Partial atomic charges for amino acids were taken from the Solvent Interaction Potential (SIP) dataset (33). Ionizable residues Asp, Glu, Arg, and Lys beyond 8 Å had integer charges positioned at the C ␥ , C ␦ , C , and N z atoms, respectively.
The effects of solvent on the relative energies of 1, 3, and 4 were determined through calculations of solvation energies. The electrostatic component of the solvation energies was calculated using the DelPhi program (Delphi 97.0 Molecular Simulations Inc., San Diego, CA (34)). Hydrogen atoms were added to fill in unsatisfied valencies using the InsightII package (InsightII 97.0, Molecular Simulations Inc., 1997). The positions of all hydrogen atoms were subjected to energy minimization, while keeping protein, inhibitor, and water atoms fixed at the positions determined by x-ray crystallography, by using the Discover program and conventional valence force field parameters (Discover 97.0, Molecular Simulations Inc., 1997). Electrostatic potential-derived charges (35) were calculated for the inhibitor at the HF/6 -31ϩG(d) level and used to determine SIP atomic radii. Charges and radii for protein atoms were taken from the SIP data set of amino acids (36). All water molecules were removed for the calculation of solvation energies. The ionic strength was set to zero, and dielectric constant values of 1 and 78.54 were assumed for the solute and solvent, respectively. A grid of 197 ϫ 197 ϫ 197 points was used, yielding a grid resolution of 0.5 Å. This maintained a solvent boundary of at least 10.0 Å around the protein.
Coordinates-The coordinates of ␤-D-glucan glucohydrolase with bound PheGlcIm have been deposited with the Protein Data Bank (37), under the code 1LQ2. Table I summarizes first order rate constants of inhibition k i , dissociation constants for the enzyme-inhibitor complex K i , and secondorder rate constants of inhibition k i K i Ϫ1 , for a range of "glucose-derived" inhibitors. Although 2F-DNPGlc, conduritol B epoxide, castanospermine, and gluconolactone all inhibit within the micromolar range, the K i constants for glucoimidazoles are in the low nanomolar ranges (Table I) and in all cases show competitive inhibition (data not shown). Although variations of ϳ1 order of magnitude in K i values for various inhibitors have been reported (38), the K i constants for glucoimidazoles frequently fall within the nanomolar ranges (3, 17, 39). Thus, PheGlcIm binds to ␤-D-glucan glucohydrolase ϳ2 ϫ 10 5 times tighter than laminarin, which is the best non-synthetic ground-state substrate found so far for this enzyme (40), 10 6 times tighter than the synthetic substrate 4NPGlc (40), and 2 ϫ 10 7 tighter than glucose, which is the product of the hydrolytic reaction and a competitive inhibitor (8). The strength of the inhibition with glucoimidazoles containing phenyl or phenethyl substituents on their GlcIm core, mimicking aglycones, is ϳ30 times lower than the K i value for GlcIm (Table I). The inhibitor with the lowest K i value was PheGlcIm, that is, the inhibitor with the phenyl substituent (Table I), and this compound was therefore chosen for structural studies.

Inhibition of ␤-D-Glucan Glucohydrolase-
To investigate any correlation between binding of the inhibitor and apparent pK a values of the catalytic amino acids of ␤-D-glucan glucohydrolase, k cat K m Ϫ1 , and K i Ϫ1 values for PheGlcIm were calculated over a range of pH values. Fig. 1 shows the pH dependence of k cat K m Ϫ1 values for ␤-D-glucan glucohydrolase and indicates that the apparent pK a values for the catalytic nucleophile and the catalytic acid/base are 4.6 and 6.7, respectively. The dependence of K i Ϫ1 of PheGlcIm versus pH shows that the lowest K i value for PheGlcIm of 1.7 ϫ 10 Ϫ9 M was observed at pH 5.3. The apparent pK a constant of PheGlcIm is 4.99 (19), and this indicates that the loss of binding at pH values lower than ϳ5.0 reflects the formation of positive charge on the N1 of the imidazole moiety of PheGlcIm. The basic side of the K i Ϫ1 versus pH profile shows gradual loss of binding of the inhibitor at pH values above 5.3.
Crystallography of the ␤-D-Glucan Glucohydrolase-PheGlcIm Complex-The crystallographic data were collected to 2.62-Å resolution at 100 K from at least three different crystals using various soaking strategies to obtain the best resolution. PheGlcIm was added directly to the crystal in mother liquor in a solid form, or a solution of PheGlcIm mixed with the mother liquor to reach 2 and 10 mM concentrations. The best 2.62-Å diffraction data set was obtained in the latter case with a mosaic spread of 0.4°. The three-dimensional structure was solved to 2.62-Å resolution using a rigid body refinement technique and the three-dimensional structure of the ␤-D-glucan glucohydrolase-glucose complex as a search model. The final crystallographic R work and R free factors for ␤-D-glucan glucohydrolase-PheGlcIm complex were 20.96% and 27.29%, respectively (Table II). The structure of the ␤-D-glucan glucohydrolase consists of 602 amino acid residues, 241 bound water molecules, and 1 PheGlcIm molecule. Furthermore, we detected a heavily occupied glycosylation site at N600, which was disordered in previous crystals (1EX1 (42)), contained no detectable sugars (1IEQ, 1IEV, and 1IEX), or contained either one (1IEW (9)) or two (1J8V (10)) N-acetyl-D-glucosamine moieties at this site. The oligosaccharide attached to N600 has the following minimal structure (Structure 1).
The interactions of PheGlcIm with ␤-Dglucan glucohydrolase are schematically shown in Fig. 2. At the resolution achieved from the data at 2.62 Å, there are three continuous electron densities at the 1.6 level formed between three active site residues Asp 95 , His 207 , and Asp 285 , and the C6OH, C7OH, C8OH, and C9OH groups of PheGlcIm (Fig. 3). The C9OH group of PheGlcIm (corresponding to the C2OH in the glucosyl residue at subsite Ϫ1 (9)) interacts with the catalytic nucleophile O␦2 285 with a very short interaction of 2.48 Å, whereas the N1 atom of PheGlcIm interacts with the acid/base catalyst making hydrogen bond interactions to O⑀1 491 and O⑀2 491 of 3.07 and 2.95 Å, respectively (Fig. 2). The latter interactions are significantly longer than those observed previously in the Bacillus agaradherans endo-1,4-␤-D-glucanasecellobiose-derived imidazole (16) and Cellulomonas fimi endo-1,4-␤-D-xylanase-xylo-imidazole (17) complexes, where the distances of the inhibitor N1 atom to the catalytic acid/base are 2.58 Å and 2.49 Å, respectively. Residue Asp 95 interacts with the C4OH and C6OH groups of the glucose moiety in the ␤-D-glucan glucohydrolase-glucose complex, and His 207 , which interacts with the C3OH of the glucose in the ␤-D-glucan glucohydrolase-glucose complex, form very short interactions of 2.45 Å and 2.55 Å, respectively. Naturally, the resolution of 2.62 Å at which the structure was determined precludes placing any significance on these short distances.
There are five other residues involved in the PheGlcIm binding, namely Lys 206 , Arg 158 , Trp 286 , Trp 434 , and Glu 220 , and three water molecules that interact with amino acid residues bound to PheGlcIm (Figs. 2 and 3). One water molecule is associated with the catalytic nucleophile Asp 285 , and two water molecules are associated with the catalytic acid/base Glu 491 and Glu 220 . The two latter water molecules were similarly positioned in the ␤-D-glucan glucohydrolase-2-deoxy-2-fluoro-␣-D-glucopyranosyl complex and were coordinated through hy-  ␤-D-Glucan Glucohydrolase Transition State Complex drogen bonding interactions with Glu 491 and Glu 220 (9). These water molecules were implicated in the hydrophilic "channel" through which they are precisely directed as candidate nucleophiles for a second displacement reaction of the hydrolytic mechanism.
The sugar component of PheGlcIm is in the 4 E (envelope) or the 7 E conformation with numbering shown in Fig. 2. In this structure the C8-C9-C10-N1-C5 atoms are co-planar. The plane of the phenyl moiety of PheGlcIm is tilted by ϳ45°to the planes of the side chains of Trp 286 and Trp 434 (Fig. 3). The tilt of the phenyl substituent of PheGlcIm provides more favorable localized hydrophobic interactions with the phenyl components of indole moieties of Trp 286 and Trp 434 than if the phenyl rings were parallel to the planes of the indole moieties. These hydrophobic interactions restrict the conformational mobility of the phenyl ring of PheGlcIm, producing very clear electron density at the 4 level for the phenyl substituent of PheGlcIm (Fig. 3).
The superpositions of the ␤-D-glucan glucohydrolase-PheGlcIm complex onto the three-dimensional structures with bound glucose, cyclohexitol, and S-cellobioside moieties (9) are shown in Fig. 4. The comparison with ␤-D-glucan glucohydrolase-glucose, which represents an enzyme-product complex, indicates that O␦1 and O␦2 of Asp 285 and Asp 95 , and His 206 make pronounced movements toward the "glucose" component of PheGlcIm. On the other hand, O⑀1 491 rotates slightly away from the N1 atom of the imidazole moiety. Despite these movements, the glucose moieties in both structures are almost perfectly superimposable (Fig. 4B). The glucose and the cyclohexitol moiety (Fig. 4C) do not superpose well, because the C1 atom of the cyclohexitol moiety migrates ϳ1.2 Å toward O␦1 285 (9). The superposition of ␤-D-glucan glucohydrolase-PheGlcIm and the ␤-D-glucan glucohydrolase-S-cellobioside complexes (Fig.  4D) shows that O⑀2 491 is placed 0.2 Å closer to the S atom of the S-cellobioside moiety than O⑀2 491 is to the N1 atom in the  (Table III). In configurations 1 and 3 the N1-O ⑀1 distance is shorter than the N1-O ⑀2 distance, in disagreement with the crystallographic data, whereas in configuration 3 the N1-O ⑀1 distance is longer and the N1-O ⑀2 distance is shorter than the values found in the experiment. In configuration 2 both distances are much longer than those found experimentally, even though this configuration represents the theoretical ionization states of Glu 491 and the phenyltetrahydroimidazopyridine moiety in the crystalline environment. We have also considered configurations in which only one of either Glu 491 or PheGlcIm were charged (data not shown). These calculations did not provide a better agreement with the experimental values than those observed in configurations 1-4.
Configurations 1-4 include the three most likely, albeit different logical considerations of the ionization states of Phe-ImGlc and Glu 491 (15,16). Configurations 3 and 4 represent the same ionization states of PheImGlc and Glu 491 , but differ in the position of a proton (Table III) that is present either on O⑀1 491 or on O⑀2 491 , respectively; O⑀1 491 and O⑀2 491 have the same formal HOCO angles of 0°. At the highest level of theory studied here, B3LYP energies incorporating solvation effects, we predicted that configuration 3 lies 12.9 kJ mol Ϫ1 lower than configuration 1, whereas configuration 4 is predicted to lie 9.7 kJ mol Ϫ1 higher than 1. In summary, these calculations suggest that the energies of both ionization states are very similar. The state in which both PheGlcIm and Glu 491 are neutral is slightly lower in energy than the state in which both components are charged. The Cremer-Pople ring pucker parameters of the configurations 1-4 show that the glucosyl moieties in PheImGlc in each of the model structures are in an E-type conformation. Notably, no differences in the Cremer-Pople ring pucker parameters of the N1 protonation state of PheImGlc are found, compared with the neutral state of PheImGlc, and suggest that higher resolution x-ray data are unlikely to distinguish between them. DISCUSSION We have previously shown for the barley ␤-D-glucan glucohydrolase with bound S-cellobioside or S-laminaribioside moieties (9, 10) that pyranoside ring distortion from the 4 C 1 conformation is not necessary for occupation of the Ϫ1 subsite and have never observed distortion of intermediates or substrate analogues in the active site. Similarly, Sulzenbacher et al. (46) with an endoglucanase-cellobiose complex and Fort et al. (47) with an endoglucanase-mixed-linkage cellooligosaccharide complex showed that there is no substrate distortion at Ϫ1 subsite, although in the latter case the mixed-linkage oligosaccharide evades binding at the Ϫ1 subsite. However, the fact that the sugar is locked tightly at the Ϫ1 subsite of the ␤-D-glucan glucohydrolase through extensive hydrogen bond-ing is consistent with the theory that distortion into the transition state geometry occurs at this subsite, along the reaction coordinate (3,7).
The mechanistic aspect of substrate distortion during development of ion-like transition states has been addressed previously using transition-state-like mimics or transition-state-like analogues (3,6). Notenboom et al. (17) suggested that there are five principal factors contributing to the development of transition state characteristics, including charge distribution, a trigonal anomeric center, a half chair conformation (or close to), an appropriate relative configuration of OH groups, and an ability of the glycosidic oxygen to be directionally protonated.
In the light of these suggestions and the observed distortion of substrates in many other glycoside hydrolases, we further investigated the possibility that substrate or intermediate ring distortion occurs during hydrolysis by the barley ␤-D-glucan glucohydrolase, using transition state mimics. A comparison of the K i values for barley ␤-D-glucan glucohydrolase inhibitors revealed large differences (Table I). Castanospermine, which has previously been described as a transition state analogue (54), inhibited the barley enzyme in the micromolar range. Withers et al. (55) showed that castanospermine, being structurally unrelated to the transition state, is a "fortuitous binder" rather than a transition state analogue; no correlation was observed between pH dependences of inhibitor binding of castanospermines and catalytic efficiencies of the Agrobacterium sp. ␤-D-glucosidase. The most potent inhibitor of barley ␤-Dglucan glucohydrolase was PheImGlc, with the K i constant of 1.7 ϫ 10 Ϫ9 M. This value places PheImGlc among the strongest inhibitors found so far for glycoside hydrolases (3,19). The low K i value of PheImGlc reflects the strong binding, expected from a transition state analogue. Although the K i constant for Phe-ImGlc is in the low nanomolar ranges, the potential to design even more powerful mimics of the transition states exists. It has been suggested that these analogues should bind 10 10 -10 15 times tighter (6,14,56) or even 10 19 -10 23 times tighter (3,57) than the ground-state substrates.
To investigate whether the PheImGlc satisfies the definition of a transition state mimic (3, 22, 57), catalytic efficiency and inhibitor binding constants were monitored across a range of

␤-D-Glucan Glucohydrolase Transition State Complex
pH values (Fig. 1). Barley ␤-D-glucan glucohydrolase is a retaining enzyme (8) and has a pH optimum between pH 5.2 and 5.6 ( Fig. 1) (40). The two ionizable amino acid residues Asp 285 and Glu 491 act as the catalytic nucleophile and catalytic acid/ base, respectively (9). A correlation between binding of an inhibitor and apparent pK a constants of the catalytic machinery is one of the criteria proposed for transition state mimics (3,22,57). Fig. 1 shows a close similarity in catalytic efficiency and inhibitor binding over the pH range 4.0 -7.5 and would therefore suggest that the inhibitor is acting as a transition state mimic. In addition, the inhibitor binding versus pH curve indicates that the inhibitor is active in its neutral form (Fig. 1).
Here we report the first three-dimensional structure of family GH3 glycoside hydrolase in complex with glucoimidazoletype of inhibitor, which may be regarded as a transition state mimic or a transition state analogue. An important finding arising from the three-dimensional structure of the ␤-D-glucan glucohydrolase-PheImGlc complex is that the glucose component of PheGlcIm is in the 4 E conformation. Similarly, a xylobiose-derived imidazole in complex with an endo-1,4-␤-D-xylanase from Cellulomonas fimi (1FHD) adopts the 4 E conformation (17), whereas a cellobiose-derived imidazole in complex with an endo-1,4-␤-D-glucanase from Bacillus agaradherans (8A3H) has a 4-sofa conformation (16), which is similar to 4 E. The superposition in the active sites of all three ligands shows that their coincidence is almost perfect (Fig. 5). Values of dihedral angles of the imidazole ring of the B. agaradherans endo-1,4-␤-D-glucanase-cellobiose-derived imidazole (16), the C. fimi endo-1,4-␤-D-xylanase-xylobiose-derived imidazole complexes (17) and the ␤-D-glucan glucohydrolase-PheGlcIm complex (this work) are in conformity; the C5-N4-C10-C9 dihedral angles span the range Ϫ0.8 -0.1°, the N4-C10-C10-C9 dihedral angles are within 0.3-5.9°, and the C7-C5-N4-C10 dihedral angles span the range 24.3-30.0°. Remarkably, the relative positions of catalytic nucleophiles in the three active sites are very similar (Fig. 5). However, the relative dispositions of catalytic acid/bases in the enzyme-ligand complexes are different. Whereas in the B. agaradherans endo-1,4-␤-D-glucanase (16) the catalytic acid/base Glu 139 is positioned in the plane of the imidazole ring (Fig. 5A), the acid/base Glu 127 in the C. fimi endo-1,4-␤-D-xylanase complex (17) is slightly above the plane of the imidazole (Fig. 5B). On the contrary, the catalytic acid/base Glu 491 in the barley ␤-D-glucan glucohydrolase-PheGlcIm complex is markedly above the imidazole plane  Fig. 2) of O⑀2 228 to N1 atom and O⑀1 139 to "C2OH" group in 8A3H are 2.72 Å and 2.58 Å, respectively. The distances of O⑀2 233 to N1 atom and O⑀2 127 to "C2OH" group in 1FHD are 2.63 Å and 2.49 Å, respectively. (Fig. 5). This comparison implies that a distortion of the substrate must take place during the hydrolytic event and during this distortion the glycosidic oxygen might move "upwards," that is, closer to the position of the catalytic acid/base Glu 491 . This possibility is further illustrated by superpositions of ␤-Dglucan glucohydrolase-PheGlcIm and the ␤-D-glucan glucohydrolase-S-cellobioside complexes, where the exocyclic N1 atom in PheGlcIm is below the "glycosidic" S-atom, and thus below a predicted pseudo-axial orientation expected for the glycosidic oxygen (Fig. 6). This observation agrees with those made by Notenboom et al. (17), however, the possibility remains that 4 E represents an artifactual conformation, and the actual transition-state-like conformation during the hydrolytic cycle might be 4 H 3 .
Notenboom et al. (17) suggested that the C. fimi endo-1,4-␤-D-xylanase reaction proceeds through a classic 4 C 1 3 1 S 3 3 4 H 3 3 4 C 1 itinerary, and Sidhu et al. (53) proposed a 4 C 1 3 2 H 3 3 2 S O 3 2,5 B 3 4 C 1 itinerary for the Bacillus circulans endo-1,4-␤-D-glucanase. In our present work and in previous studies (9, 10), various stable intermediates during the hydrolytic cycle of a barley ␤-D-glucan glucohydrolase have been defined. The following itinerary can now be suggested for the barley ␤-D-glucan glucohydrolase: 4 C 1 3 (?) 3 4 E ( 4 H 3 ) 3 4 C 1 3 4 C 1 , where the individual events are a free substrate, an enzyme-substrate complex of unknown conformation, the transition state, the covalent glycosyl-enzyme intermediate, and the glucose product, respectively. At this stage we are not able to conclude whether the ␤-D-glucan glucohydrolase-PheGlcIm complex represents an early or a late transition state. According to the current understanding of the mechanism of action of retaining hydrolases (3,4,7), it is expected that an early transition state or the one that precedes the first transition state, would be closer to a distorted conformation. On the other hand, the late transition state or the one that follows the second transition state might be closer to an oxocarbenium ion-like intermediate. It then follows that these inhibitors could mimic both types of transition states, and those that mimic the distorted substrate or oxocarbenium ion-like intermediate, should bind strongly to the enzyme.
Heightman and Vasella (15) suggested that orientation of the lone pair of electrons of the glycosidic oxygen with respect to an acid/base catalyst and relative to the C1-O5 bond of the sugar moiety, defines the protonation trajectory of glycoside hydrolases, which can be either anti or syn. Whereas the anti- ␤-D-Glucan Glucohydrolase Transition State Complex protonation class of hydrolases includes retaining enzymes, which have been classified within the GH-A clan (families GH1, GH2, GH5, GH10, and GH51) and the GH-K clan (families GH18 and GH20; available at afmb.cnrs-mrs.fr/CAZY/), the syn-protonators represent both retaining and inverting enzymes, which are classified in glycoside hydrolase families GH7, GH11, GH12, GH16, GH22, GH23, and GH45 (3). The barley ␤-D-glucan glucohydrolase, as a representative of family GH3 glycoside hydrolases, is a typical retaining enzyme (e3e; nomenclature of Sinnott (4)) enzyme (8). The structure of the ␤-D-glucan glucohydrolase-PheGlcIm complex clearly reveals an anti-protonation trajectory for this enzyme (Fig. 5). The family GH5 B. agaradherans endo-1,4-␤-D-glucanase (16) and the family GH10 C. fimi endo-1,4-␤-D-xylanase (17) are also classified as anti-protonators, with a lateral mode of protonation. The barley ␤-D-glucan glucohydrolase could similarly be classified as a lateral anti-protonator, although in this case the acid/base catalyst in the active site is positioned more perpendicularly to the glycosidic oxygen (Figs. 5 and 6). However, it is important to point out that the relative positioning of the catalytic acid/base in the active site may be dictated by the overall geometry of the active site. Despite the fact that the three enzymes mentioned above are not closely related (belonging to glycoside hydrolases families GH3, GH5 and GH10; afmb.cnrs-mrs.fr/CAZY/), it is remarkable that such an exceptionally good agreement in the positioning of the enzyme catalysts in their active sites has been reached (Fig. 5).
Further to the comparison of the three-dimensional disposition of the catalytic amino acid residues in the B. agaradherans endo-1,4-␤-D-glucanase-cellobiose-derived imidazole complex (16), in the C. fimi endo-1,4-␤-D-xylanase-xylobiose-derived imidazole complex (17) and in the ␤-D-glucan glucohydrolase-PheGlcIm complex (Fig. 2), we have investigated differences in the disposition of residues that bind the C6OH group and the C5 carbon of the imidazole-sugar inhibitors at the Ϫ1 subsite. As expected for enzymes from different but related families, there is limited conservation of amino acid residues that bind the C6OH group (or the C5 carbon for xylobiose-derived imidazole) (data not shown). Nevertheless, amino acid residues that bind the C6OH groups are present in both the B. agaradherans endo-1,4-␤-D-glucanase (family GH5) and in the barley ␤-Dglucan glucohydrolase (family GH3), which indicates functional conservation of the C6OH binding. In the B. agaradherans endo-1,4-␤-D-glucanase, C␣ backbone oxygens of Ala 234 and Thr 235 bind to the C6OH group of the cellobiose-derived imidazole at Ϫ1 subsite (16), whereas Asp 95 alone is sufficient to achieve similar binding of PheGlcIm in the barley ␤-D-glucan glucohydrolase (9,10). In contrast, the C. fimi endo-1,4-␤-Dxylanase (family GH 10) has a bulky aromatic residue Trp 281 that surrounds the C5 carbon at the Ϫ1 subsite and, in addition, Gln 87 fills up the conformational space around the CЈ6OH group at the Ϫ2 subsite. Thus, Trp 281 and Gln 87 effectively hinder binding of cellobiose-derived ligands by the endo-1,4-␤-D-xylanase. However, the C. fimi endo-1,4-␤-D-xylanase does hydrolyze cellulose (58), albeit 40-times slower than xylan, and thus Trp 281 and Gln 87 must be capable of rearranging their positions in the active site to accommodate the cellulose substrate (58).
Quantum mechanical calculations were performed to predict the ionization state of the acid/base catalyst Glu 491 and the ionization state of the inhibitor PheGlcIm, so as to ascertain whether the shape of PheImGlc or its ionization state were more important for inhibition of the enzyme. These calculations suggest that PheGlcIm is neutral when bound to the enzyme. However, the difference in energy between ionized and neutral states is not large. If there is no proton transfer from the catalytic acid/base to the PheImGlc molecule, as these calculations suggest, it seems likely that the shape (glucose and PheImGlc are in 4 C 1 and 4 E conformations, respectively) must play a critical role when the inhibitor interacts with the active site of the ␤-D-glucan glucohydrolase.
␤-D-Glucan glucohydrolases in embryophytes have been implicated in wall loosening during cell elongation, in wall remodeling, in defense reactions against fungal pathogens, in the release of glucose from wall polysaccharides as an energy source in dark-grown seedlings, and in the general turnover of glucose from ␤-D-glucans and ␤-D-oligosaccharides (9,10,59). The broad specificity of the enzyme with respect to linkage position in its ␤-D-glucan and ␤-D-oligosaccharide substrates suggests that it could undertake several of these functions. In this context details of the conformational pathway of the substrates during enzymatic catalysis of the ␤-D-glucan glucohydrolase can be used as a basis for the design of novel transition state analogues, which could be used to investigate the function of the enzymes in planta. Furthermore, transition state analogues might find applications as agents controlling plant growth (60), and as probes for the investigation of catalytic mechanisms (3,6). As an adjunct to such investigations, powerful computational techniques such as quantum mechanical modeling, molecular docking, de novo design, quantitative structure-activity relationships and combinatorial library design, based on structural information, have potential to rationally predict new families of inhibitors (61)(62)(63), for application in agricultural and related technologies.