Crystal Structures of Geobacillus stearothermophilus α-Glucuronidase Complexed with Its Substrate and Products

α-Glucuronidases cleave the α-1,2-glycosidic bond between 4-O-methyl-d-glucuronic acid and short xylooligomers as part of the hemicellulose degradation system. To date, all of the α-glucuronidases are classified as family 67 glycosidases, which catalyze the hydrolysis via the investing mechanism. Here we describe several high resolution crystal structures of the α-glucuronidase (AguA) from Geobacillus stearothermophilus, in complex with its substrate and products. In the complex of AguA with the intact substrate, the 4-O-methyl-d-glucuronic acid sugar ring is distorted into a half-chair conformation, which is closer to the planar conformation required for the oxocarbenium ion-like transition state structure. In the active site, a water molecule is coordinated between two carboxylic acids, in an appropriate position to act as a nucleophile. From the structural data it is likely that two carboxylic acids, Asp364 and Glu392, activate together the nucleophilic water molecule. The loop carrying the catalytic general acid Glu285 cannot be resolved in some of the structures but could be visualized in its “open” and “closed” (catalytic) conformations in other structures. The protonated state of Glu285 is presumably stabilized by its proximity to the negative charge of the substrate, representing a new variation of substrate-assisted catalysis mechanism.

in nature. The natural degradation of these polymers, a key step in the carbon cycle on Earth, is carried out by microorganisms that can be found either free in the environment or in the digestive tract of higher animals. Efficient cellulolytic microorganisms typically secrete a battery of enzymes that can function either alone or as parts of larger enzymatic complexes such as the cellulosome (1)(2)(3). The enzymes that degrade cellulose and hemicellulose are in many cases modular and include catalytic domains of glycoside hydrolases and/or carbohydrate esterases as well as carbohydrate-binding modules that allow the soluble enzymes to adhere to insoluble substrates (4). A most informative and updated classification of these and other carbohydrate active enzymes is available on the Carbohydrate-Active Enzymes (CAZy) server (available on the World Wide Web at afmb.cnrs-mrs.fr/CAZY).
The glycosidic bond is one of the most stable bonds in nature, with a half-life of over 5 million years; glycosidases can accelerate the hydrolysis of these bonds by more than 10 17 -fold (5).
The key elements in this catalysis are the position of the catalytic residues and the distortion of the sugar ring so as to allow the stabilization of an oxocarbenium ion-like transition state (6). The enzymatic hydrolysis of glycosidic bonds results in either an overall retention or inversion of the anomeric configuration of the substrate. Retaining glycosidases follow a two-step double-displacement mechanism, usually involving two key active site carboxylic residues, one functioning as the nucleophile catalyst and the other as the acid/base catalyst. Inverting glycosidases use a single-displacement mechanism, in which usually one carboxylic acid functions as general acid and the other functions as a general base.
␣-Glucuronidases use the inverting mechanism to cleave the ␣-1,2-glycosidic bond between 4-O-methyl-D-glucuronic acid (MeGlcA) 1 and xylose units, which are part of short ␤-1,4xylooligomers, the main products of the enzymatic hydrolysis of xylan by xylanases (7,8) (Fig. 1). The different bacterial and fungal ␣-glucuronidases are classified as glycoside hydrolase family (GH) 67, and the first crystal structure of an enzyme from this family, the ␣-glucuronidase from Cellvibrio japonicus (GlcA67A), was recently described (9). The enzyme is folded into three domains, the central of which has a (␤/␣) 8 -barrel fold that accommodates the active site.
Geobacillus stearothermophilus (formerly Bacillus stearothermophilus) strain T-6 is a thermophilic bacterium that pos-sesses an extensive hemicellulolytic system (2). Many of the genes comprising this system are clustered together on the bacterial chromosome, and so far the identified components include intracellular and extracellular GH-10 xylanases; GH-39, -43, and -52 ␤-xylosidases; GH-51 ␣-L-arabinofuranosidases; acetyl-xylan esterases; xylose catabolism genes; and transport systems (10 -14). The ␣-glucuronidase T-6 (AguA) is part of a 15.5-kb-long operon involved in the utilization of D-glucuronic acid (15,16). Here we present high resolution (1.5-2.0-Å) crystal structures of AguA in complex with substrates and products. These structures provide direct evidences for several mechanistic features. We suggest that the catalytic mechanism of AguA includes a conformational change of the active-site loop carrying the general acid residue Glu 285 , a distortion of the substrate into a conformation much closer to the flat conformation required for the oxocarbenium ion-like transition state, and charge repulsion between the acidic substrate and the general acid residue so as to allow its protonated state. The structural data, combined with biochemical analysis of catalytic mutants, suggest that the two carboxylic residues Asp 364 and Glu 392 , serve as general bases, activating together the nucleophilic water molecule.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The ␣-glucuronidases from G. stearothermophilus strains T-6 and T-1 (AguA T-6 and AguA T-1; accession numbers AAC98128, and AAL32057, respectively) were cloned and overexpressed in Escherichia coli. The purification of the recombinant wild type and mutant enzymes was done as described before (16). Production of the selenomethionine derivative of AguA was carried out   2. Overall structure of AguA. a, representative electron density difference maps ("omit" map) demonstrating the reliability of the models. Top, a stereoview of the electron density section around the active site of the E285N mutant in complex with the intact aldotetraouronic acid substrate and the nucleophilic water molecule (contour level of 3.5; green, protein and water; blue, bound substrate). Bottom, a stereoview of the in the methionine auxotrophic Escherichia coli strain B834(DE3), essentially as described previously (17), and protein purification proceeded as for the wild type enzyme. ␣-Glucuronidase activity was determined by measuring the release of MeGlcA from aldotetraouronic acid using the Milner and Avigad assay for uronic acids (18), as described before (16).
Crystallization and X-ray Diffraction Data Collection-All crystals were obtained using the hanging drop crystallization method and grown at 20°C, with some modifications of the procedures described previously (19). The best crystals were obtained by mixing 5 l of the original enzyme solution (10 mg/ml) with an equal volume of reservoir solution (containing 14% (w/v) polyethylene glycol 4000, 12% (v/v) isopropyl alcohol, and 0.1 M sodium citrate, pH 5.0 -5.5) and equilibrating the drop with 1 ml of the reservoir solution. For x-ray data measurement, crystals were transferred into a cryo-solution (containing 75% reservoir solution and 25% glycerol) for about 30 s before direct flash cooling in a nitrogen gas cold stream (at 90 -100 K). The diffraction pattern of all crystals grown and flash-cooled in this procedure indicated that they are isomorphous (unit cell parameters are within Ϯ2% of the mean values) to the previously reported crystals of wild type (WT) AguA (T1 crystal form) (19), with a tetragonal space group of P4 1 2 1 2, mean unit cell dimensions of a ϭ b ϭ 74.6 Å, c ϭ 330.3 Å, and with one molecule per crystallographic asymmetric unit.
Six types of crystals were used for complete data measurement and structural analysis, as follows: 1) crystals of the selenomethionine derivative of AguA; 2) crystals of WT (uncomplexed) AguA; 3) crystals of the E285N mutant of AguA, which were co-crystallized with the substrate aldotetraouronic acid (2-O-(4-O-methyl-␣-D-glucuronosyl)-␤-D-xylotriose) (in the final model of this structure, only the MeGlcA group was resolved, and the xylotriose group had no apparent electron density); 4) crystals of the E285N mutant, which were soaked for 10 s in a solution containing ϳ5 mg/ml of aldotetraouronic acid and then flash-cooled (in the final model of this structure, an intact substrate was found to be bound to the enzyme); 5) crystals of WT AguA, which were soaked for 5 s in a solution containing ϳ4 mg/ml of aldotetraouronic acid and then flash-cooled (in the final model of this structure, the two reaction products MeGlcA and xylotriose were observed in the active site); 6) crystals of the E386Q mutant of AguA.
Two independent sets of multiple anomalous diffraction data were collected on the selenomethionine crystals, using one single crystal for each multiple anomalous diffraction experiment (20). Each of these data sets included complete diffraction data that were collected at three different wavelengths around the selenium K edge ( Table I). The two data sets were measured at 100 K, using a MAR-CCD (133-mm) detector (MAR-Research Inc.), on beamline BM14 of the European Synchrotron Radiation Facility (Grenoble, France). All other data sets were collected (95 K) at the National Synchrotron Light Source (NSLS) (Brookhaven National Laboratory). The WT AguA crystals were used for a 1.70-Å data set on NSLS/X8C, using a Quantum-4 CCD detector (ADSC Inc.). The E285N crystals were used for 1.85-and 1.75-Å data sets for the co-crystallized and soaked crystals, respectively, on NSLS/X12B using a Quantum-4 CCD detector. The substrate-soaked WT crystals were used for a 1.50-Å data set on NSLS/X25, using a Brandeis B4 CCD detector (Brandeis University). The E386Q crystals were used for a 2.0-Å data set on DESY/EMBL (Hamburg, Germany) beamline X13, using a MAR-CCD (165-mm) detector. All of the diffraction data sets were processed and reduced with the programs DENZO and SCALEPACK (21). Selected data collection parameters representing the nonmultiple anomalous diffraction data sets are shown in Table II.
Structure Determination and Refinement-The structure of the selenomethionine derivative of AguA was solved from the anomalous signal of the 14 selenomethionine residues of the modified enzyme, using the two data sets collected on selenomethionine crystals (20,22). Model building was performed with the program O (23). Due to the relatively high electron density of the selenium atoms and other landmark residues, the exact chain tracing and individual fitting of amino acid side chains was relatively straightforward. The structure of the WT AguA was solved by molecular replacement using the program AMORE (24), with the selenomethionine AguA structure as the reference model.
Further refinement was performed with the program CNS (25). The progress of refinement was monitored by following the overall values of R factor and R free (26), calculated for 10% randomly selected reflections. All other structures were solved and refined in a similar procedure (Table II).

RESULTS
The Overall Structure of AguA-The recombinant ␣-glucuronidase proteins were obtained from two related strains of G. stearothermophilus, T-1 and T-6 (AguA-T1 and AguA-T6). These two enzymes are essentially identical, differing in only 2 of the 679 amino acids of the proteins. The recombinant enzymes were overexpressed in E. coli, purified in 1-g quantities, and crystallized (16,19). Crystal structures of WT, selenomethionine derivative, catalytic mutants, and enzyme-substrate/ product complexes of AguA-T1 and AguA-T6 were obtained. All of the present structures were determined at relatively high resolutions (1.5-2.0 Å) and at good final values of R factor (13.9 -19.3%) and R free (17.2-21.8%) ( Table I). Based on these values, the average experimental error in the coordinates of these models is around Ϯ0.1 Å according to the Luzzati error estimation (27), permitting a reliable analysis of interactions and geometries in the structures presented here. Representative sections of the electron density maps of two of the complexed structures obtained are shown in Fig. 2a.
Since the structures of the WT AguA-T1 and AguA-T6 were identical within the experimental error margins (except for slight changes around residues 38 and 259, which are different between the two enzymes), we describe here only the 1.7-Å resolution structure of AguA-T1 (Fig. 2b). AguA is an ␣/␤ globular protein with overall dimensions of about 91 ϫ 56 ϫ 53 Å for a monomer. The enzyme is built of three distinct domains, which are connected by extended loops. The central domain has a (␤/␣) 8 fold (TIM barrel), and it contains about half of the protein residues (amino acids 143-471). This domain is not organized as a classical TIM barrel; there are an additional two short ␤-strands between the first ␤-strand and the first ␣-helix (␤ 7 and ␣ 4 ); the ␣-helix between strands ␤ 14 and ␤ 15 is missing; and two ␣-helices rather than one are located after both strands ␤ 11 and ␤ 15 (Fig. 2c). The N-terminal domain (residues 1-142) is made of a six-stranded ␤-sheet and three ␣-helices. The C-terminal domain, which includes the last 208 amino acids, is mostly ␣-helical and lies as an envelope covering almost half of the central domain (Fig. 2b). The final structural models consisted of almost all of the 679 amino acid residues of the protein, lacking only the first 2-4 N-terminal residues. In three of the five structural models, the loop containing residues 283-287 had disordered electron density, suggesting conformational flexibility. Interestingly, this loop is experimentally observed in two different conformations in the two other structures. The functional implications of this conformational change are discussed below.
AguA is a homodimer in solution (16), yet under the conditions employed it crystallized with one molecule per crystallographic asymmetric unit. Generation of all of the symmetryrelated molecules in the crystallographic unit cell revealed a possible dimeric form, which seems to be the most biologically relevant. The contact region between the two subunits involves the following residues from each monomer: Trp 328 and Arg 329 from the (␤/␣) 8 domain and Glu 536 , Arg 548 , Glu 654 , Asp 657 , Arg 665 , and Lys 666 , from the C-terminal domain (Fig. 3). The electron density section around the active site of the WT AguA in complex with the hydrolysis products MeGlcA and the two xyloses of the xylotriose (contour level of 4; green, protein; blue, bound products). The corresponding models are shown on the left, where carbons of protein residues are colored green, and carbons of the bound ligands are colored yellow. b, two views of the WT AguA monomer related by a 90°rotation, showing a schematic ribbon diagram of the protein. The polypeptide chain is colored from blue (N terminus) to red (C terminus). c, a schematic representation of the overall protein structure. ␣-Helices are presented as rectangles, ␤-strands as arrows, and loops as curved lines. The color coding is the same as in b. assignment of this dimerization contact is supported by mutagenesis and biochemical studies, showing that these residues are essential for dimer formation. 2 Comparison with Family 67 and Other Glycosidases-To date, the structure of GlcA67A from C. japonicus is the only other three-dimensional structure of a GH-67 glycosidase (9,28). AguA and GlcA67A share sequence identity of 42% and similarity of 59% of 665 aligned residues. The overall fold of the two enzymes is similar, and structure comparison shows high resemblance with root mean square deviations of 1.13 Å for all C-␣ atoms (structural homology was performed using the DALI server (29)). The two structures differ mainly in the C-terminal domain, where GlcA67A has an additional polypeptide chain of 33 residues, which is not present in the AguA sequence. The most striking difference between the two enzymes is in their dimeric structure. Whereas in AguA the contact region between the two monomers is relatively narrow and located at the tip of the C-terminal domain, in GlcA67A the dimer-forming residues are scattered on a much wider surface, which encloses parts from the C terminus as well as the middle of the (␤/␣) 8barrel (9). The dimeric interfaces of the two enzymes do not overlap. As a result, in GlcA67A the dimeric interface is closer to the active site than in AguA, but in both enzymes the substrate binding sites are exposed to the solvent. The difference in the dimeric forms of AguA and GlcA67A is consistent with the degree of sequence conservation of the two ␣-glucuronidases (see Supplementary Fig. 1).
Although the (␤/␣) 8 -barrel is one of the most common folds for enzymes in general and glycosidases in particular, the overall arrangement of the three domains comprising the ␣-glucuronidases seems to be unique to GH-67 glycosidases. The combination of the N-terminal domain and the central domain of the AguA structure resembles domains I and II of the Streptomyces plicatus ␤-hexosaminidase (30) and domains II and III of Serratia marcescens chitobiase (31). Superpositions reveal relatively low structural similarities, with average root mean square differences of 1.9 and 2.5 Å, between the AguA domains and the equivalent N-terminal and TIM-barrel domains, respectively. As in these GH-20 glycosidases, the function of the N-terminal domain of AguA is not yet clear. Previously, we have shown that removal of this domain does not significantly affect thermostability, but it does prevent the correct oligomerization of the protein and lowers its catalytic activity to 1% of the wild type activity (16). Interestingly, from the structure it seems that this domain is quite distant from the dimerization area, and its removal probably exposes hydrophobic residues located on the (␤/␣) 8 domain, causing the altered oligomerization observed.
Structures of AguA Complexed with Intact and Cleaved Substrates-Based on biochemical and mutational studies of AguA, it was previously suggested that the active site of the enzyme is located at the middle (␤/␣) 8 domain and that the conserved residues Glu 285 , Asp 364 , and Glu 392 have an important role in catalysis (16). From the structure of the WT AguA, it appears that Asp 364 and Glu 392 are located at the C-terminal side of the ␤-sheets of the (␤/␣) 8 domain (Fig. 4a). In an attempt to trap the Michaelis complex, we used the catalytic mutant E285N together with the natural substrate aldotetraouronic acid (2-O-(4-O-methyl-␣-D-glucuronosyl)-␤-D-xylotriose, or MeGlcAXyl 3 for short). Co-crystallization experiments have resulted in the 1.85-Å resolution binary complex of the enzyme with the reaction product MeGlcA, indicating that the residual activity of this mutant during the crystallization and the time elapsed until the data collection (about 2 months) was enough to cleave the substrate (Fig. 4b). When the E285N crystals were briefly soaked in the MeGlcAXyl 3 solution followed by flash freezing, the electron density at 1.75-Å resolution unambiguously showed the intact substrate bound to the enzyme (Figs. 2a and  4c). When a similar strategy of brief soaking and flash freezing was employed with the WT enzyme, the 1.5-Å resolution structure revealed that the substrate was cleaved, and the two reaction-products, MeGlcA and xylotriose, are trapped in the active site (Figs. 2a and 4d). In the two structures where the xylotriose moiety is present (E285N-substrate and WT-products), only the first two sugar units of the xylotriose could be resolved in the models, and the reducing end xylose unit is probably exposed to the solvent and too loose to produce clear electron density.
All of the observed ligands are found bound at the active site of the enzyme, which has a pocket topology of exo-acting glycosidases (Fig. 4e) (32). In the structure of the WT enzyme, a glycerol molecule (originating from the cryogenic freezing solution) is located in the active site, probably replacing water molecules present in these positions in aqueous solutions (Fig.  4a). Electrostatic potential analysis of the solvent-accessible surface of the protein indicates that the active site cavity is relatively polar, with a distinct positively charged region at the glucuronic acid binding site. In all of the complexed structures, the MeGlcA molecule is bound in a similar manner at the bottom of the active site pocket and held in place by stacking interactions with the conserved Trp 150 and an extensive hydrogen-bonding network with the conserved residues Glu 158 , Arg 159 , Asn 201 , Lys 281 , Arg 318 , Arg 335 , Lys 359 , Asp 364 , and Glu 392 (Fig. 4, b-d). Trp 150 is also involved, together with the conserved Val 200 , in forming the hydrophobic surrounding around the methyl group of the MeGlcA, as was observed in the GlcA67A from C. japonicus (28). In both the substrate and the product complexes, the conserved Trp 540 forms stacking interactions with the xylose at the ϩ1 subsite. Compared with the MeGlcA moiety, all of the xylose units are more exposed to the solvent and share fewer interactions with the enzyme. In glycosidases in general, the sugar at the Ϫ1 subsite (the glycon) is bound by a larger number of interactions as compared with the ϩ1 subsite (the aglycon). This is consistent with the observation that the specificity of glycosidases is governed mainly by the identity of the sugar unit in the Ϫ1 subsite, where the actual catalytic reaction involving the anomeric carbon takes place.
In the WT-products complex, the three observed sugar rings (MeGlcA and two xyloses) are in the conventional 4 C 1 (chair) conformation, indicating that the reaction products are bound in the active site in an unstrained fashion. However, in the complex of E285N with the intact substrate, the MeGlcA in subsite Ϫ1 is clearly distorted from the conventional 4 C 1 conformation and adopts instead a 2 H 3 (half-chair) conformation (Fig. 5a). In this conformation, the C-5-O-5-C-1-C-2 torsion angle is Ϫ29.8°, whereas in the 4 C 1 conformation of the cleaved MeGlcA, this angle is Ϫ62.9°. Thus, in the enzyme-substrate complex, the sugar ring at the Ϫ1 subsite is much closer to the flat conformation required for the oxocarbenium ion-like transition state.
In the structure of the E285N-substrate complex, a water molecule is found between Asp 364 , Glu 392 , and the anomeric carbon of the substrate, located 2.8 Å from the carboxylic oxygens of both Asp 364 and Glu 392 and 3.4 Å from the anomeric carbon (Figs. 2a and 5a). This geometry suggests that the water molecule may act as the nucleophilic water, attacking the anomeric carbon after being activated by the general base residue. It is not clear, however, which of the adjacent residues, Asp 364 Flexibility of the 283-287 Loop-As mentioned before, the loop containing residues 283-287 is absent in the free WT enzyme and the two E285N complex structures, but it is very well defined in the structure of the WT enzyme in complex with the reaction products. In its defined structure, the loop is in a "closed" conformation, and Glu 285 is in an ideal position to function as the general acid residue, being located 2.5 Å from the O-2 of the xylose at the ϩ1 subsite (which is the former glycosidic bond oxygen). Superposition of this structure with the E285N-substrate complex structure revealed that Glu 285 is 2.6 Å from the glycosidic oxygen of the substrate, a most suitable distance for the protonation of the leaving group (Fig. 5a). Thus, it seems that the presence of the substrate/products in the active site stabilizes this flexible loop and places Glu 285 in a position that allows catalysis.
Glu 386 is a conserved residue in GH-67, and its replacement to Gln led to a reduction of 14-fold in the catalytic activity of AguA (16). Indeed, from the present three-dimensional structure, it is evident that this residue has no direct interactions with either of the catalytic residues, and its closest contact to the substrate, between its carboxylic O ⑀2 and the carboxylic oxygen O-6B of the MeGlcA, is 4.5 Å long. Surprisingly, in the E386Q structure the 283-287 loop is well defined, but it takes a completely different conformation compared with the structure of the WT AguA. The loop is pushed away from the (␤/␣) 8barrel center, into a position that places Glu 285 carboxylate 8.5 Å away from its location in the wild type-products structure (Fig. 5b). It is tempting to speculate that this conformation is one of the possible conformations that this flexible loop takes in the free WT enzyme. What seems to stabilize this loop in its "open" form in the E386Q structure is a cascade of movements resulting from the loss of two strong hydrogen bonds between Glu 386 (that was replaced to Gln) and Arg 318 , leading eventually to the relocation of Phe 320 to the region where the 283-287 loop would be located in its closed conformation. DISCUSSION Identification of the catalytic residues of glycosidases is performed mainly by the characterization of mutant enzymes in which the conserved carboxylates are replaced by noncarboxylic residues. The generated mutants are then subjected to detailed kinetic analysis, using synthetic substrates bearing different leaving groups, activity rescue with exogenous nucleophiles such as azide ions, and pH dependence profiles of the mutants and the wild type enzymes (33). In retaining glycosidases, the identification of the nucleophilic residue can be performed by trapping the covalent glycosyl-enzyme intermediate using mechanism-based inactivators (34). AguA and most other ␣-glucuronidases are incapable in hydrolyzing synthetic substrates such as p-nitrophenyl-␣-D-glucuronide or p-nitrophenyl-␣-D-glucopyranose (16,(35)(36)(37)(38), making the identification of the catalytic residues of this family a challenging task. An attractive alternative for identifying the catalytic residues is by structural analysis of complexes with substrates, products, inhibitors, and transition state analogs. Inverting glycosidases hydrolyze the glycosidic bond using two conserved carboxylic acids, and their arrangement within the active site is derived from their role in catalysis. The general base catalyst is deprotonating a nucleophilic water molecule that attacks the anomeric carbon of the target sugar from the opposite side of the glycosidic bond. The other carboxylic residue acts as a general acid catalyst, protonating the leaving aglycone group (39,40).
The crystal structures of AguA in complex with its substrate and products show that the three conserved carboxylates Glu 285 , Asp 364 , and Glu 392 , are positioned adjacent to the scissile glycosidic bond and can act as the catalytic residues. In the enzyme-substrate complex, both Asp 364 and Glu 392 are located on the ␤-face of the MeGlcA sugar ring, in an adequate position to activate a nucleophilic water molecule and act as the general base catalytic residues. Indeed, a water molecule, positioned between Asp 364 and Glu 392 was detected in the structure of the E285N-substrate complex (Fig. 5a). Similar architecture, of two potential basic residues with a water molecule between them, was observed also in the structure of the ␣-glucuronidase GlcA67A from C. japonicus (9,28). In both enzymes, the water molecule is located in an almost equal distance from the two proposed basic residues, making it difficult to speculate, based on structural data alone, which of them is the catalytic residue. In the AguA structure, the angle between the two carboxylic oxygens of Asp 364 and Glu 392 and the oxygen of the observed water molecule is 122°, a suitable geometry for a joint activation of the water molecule by these residues. The surroundings of Asp 364 and Glu 392 are also rather similar, having both a neighboring basic residue 2.7 Å from them (His 527 near Glu 392 and Lys 359 near Asp 364 ) and being both acceptors of 2.9-Å long potential hydrogen bonds (Gln 388 near Glu 392 and Tyr 393 near Asp 364 ) (Fig. 6). These interactions are expected to lower the pK a values of Asp 364 and Glu 392 , in agreement with their suggested role as the basic catalytic residues.
Previously, we showed that the replacement of either Asp 364 or Glu 392 into noncarboxylic residues resulted in a reduction of 5 orders of magnitude in the catalytic activity (16). Taking together the structural and biochemical data, we suggest that Asp 364 and Glu 392 are activating the nucleophilic water together, and both of them could therefore be considered as the catalytic basic residues of AguA. Active site arrangement, in which two residues act together as general base catalysts, was suggested also in other inverting glycosidases, the GH-90 endorhamnosidase from the phage P22 tailspike protein (41) and the GH-28 endopolygalacturonases from Aspergillus niger and Stereum purpureum (42)(43)(44).
The assignment of Glu 285 as the catalytic general acid of AguA seems to be more straightforward, since it is located within a hydrogen bond distance from the glycosidic oxygen and can serve as the proton donor. However, the conformational flexibility of the loop carrying Glu 285 , as observed in the different AguA structures, is unique. Usually, the catalytic residues are part of an extensive hydrogen-bonding network that fixes the position and the charge for efficient catalysis (45). However, there are also reports describing mobile loops and structural changes that affect substrate binding and the position of the catalytic residues in glycosidases (46 -48). Interestingly, in the case of the GlcA67A from C. japonicus, the loop containing Glu 292 (which is equivalent to Glu 285 in AguA) is fixed in place, in both the free and complexed forms of the enzyme (9,28). In both forms, it is observed in an almost identical geometry to the "closed" conformation of the equivalent loop of AguA.
The significance. In order to properly operate as a general acid and donate a proton to the leaving aglycon, this residue needs to be protonated (high pK a ). However, in the structure of AguA in complex with its products (where the mobile loop is in its "closed" conformation), it seems that the nearby surroundings of Glu 285 would serve to lower its pK a rather than raising it. The N of Lys 281 is 2.8 Å from the O ⑀1 of Glu 285 , and a rather short hydrogen bond (2.6 Å) is donated to the O ⑀1 of Glu 285 by Tyr 322 , both expected to lower the pK a of Glu 285 by stabilizing the ionized state (Fig. 6). What can prompt the required acidic character of Glu 285 is the negatively charged glucuronic acid group of the substrate itself. The carboxylate of the bound MeGlcA is 2.7 Å from Lys 281 and could neutralize the positive charge of the nearby lysine side chain. In addition, the distance between Glu 285 and the carboxylate of the MeGlcA is only 4.0 Å, providing electrostatic repulsion between these two negative groups. In several other inverting glycosidases, other acidic residues are found in the proximity of the catalytic acidic residue and can serve to elevate its pK a (42,49,50). Interestingly, in retaining glycosidases, the catalytic residues are only about 5 Å apart, and the charge change of the nucleophilic residue (free and covalent intermediate) was shown to prompt the pK a cycle of the acid/base residue (51). It is likely that in ␣-glucuronidases, the binding of the charged substrate elevates the pK a of the general acid residue (Glu 285 in AguA) and stabilizes the protonated state, which is required for its action. This is a unique mechanism, in which the binding of the charged substrate facilitates catalysis by modulating the charge of a catalytic residue, representing a new variation of substrate-assisted catalysis mechanism (52).
As described before, in the complex of E285N with the intact substrate, the MeGlcA is found in a 2 H 3 conformation, which is closer to the planar conformation required for the oxocarbenium ion-like transition state structure. Interestingly, in the homologous structure of GlcA67A, the MeGlcA unit of the unhydrolyzed substrate is found in the relaxed 4 C 1 conformation (28). Distortion of substrates to conformations that approach the transition state is now a common observation in glycosidase structures (53,54). There are four possible sugar ring conformations, which place the C-5, O-5, C-1, and C-2 atoms of a pyranoside in one plane, as required for the oxocarbenium ion-like transition state: the 2,5 B and B 2,5 boat shapes and the 4 H 3 and 3 H 4 half-chair shapes. According to the pseudorotational itinerary of pyranosides (recently described by Davies et al. (53)), there is no direct conformational route from the observed distortion of the MeGlcA in AguA into one of these potential transition state conformations. Therefore, the trapped 2 H 3 structure is probably not the "real" Michaelis complex but rather a stable intermediate between the ground state ( 4 C 1 ) binding and the Michaelis complex, suggesting that the Michaelis complex is obtained via the 4 C 1 3 2 H 3 3 2 S o itinerary. At this point, from the 2 S o conformation of the Michaelis complex, there is a direct route for the 2,5 B shape for the transition state. The 2 S o 3 2,5 B itinerary was also suggested for two other inverting glycosidases, the GH-6 cellobiohydrolase (55) and the GH-8 endoglucanase (56), as well as for the retaining GH-11 xylanases (57,58). It should be noted, however, that since there is still no direct structural evidence for either the 2 S o or the 2,5 B shapes in ␣-glucuronidases, this proposed itinerary must be treated with caution.
The average distances between the proposed catalytic carboxylate pairs (in the "closed" catalytic conformation of Glu 285 ) are 8.2 Å for Glu 392 -Glu 285 and 6.6 Å for Asp 364 -Glu 285 . These distances are relatively short compared with most inverting glycosidases studied to date, in which the two catalytic residues are about 10 Å apart. In contrast, for retaining glycosidases, the distance between the two catalytic residues is generally 4.8 -5.5 Å (32,59). Recent findings suggest that unlike retaining glycosidases, inverting glycosidases show higher flexibility in regard to the exact geometry of the catalytic carboxylic residues relative to each other (42,43,50,60,61). Such flexibility is not surprising, since the general mechanism of inverting glycosidases involves an activated water attacking the anomeric carbon, enabling geometrical flexibility, unlike the strict requirement for a direct nucleophilic attack in retaining enzymes. In AguA (and probably other members of family 67 glycoside hydrolases), there is even more room for flexibility, since there are two, rather than one, base catalysts. The angle formed by the relevant oxygens of Asp 364 and Glu 392 , the nucleophilic water, and the anomeric carbon, is around 90°, leading to an apparently shorter distance between each of the two base catalysts and the acid catalyst Glu 285 .
In conclusion, the catalytic mechanism of AguA is believed to occur as follows (Fig. 6). First, the substrate is bound into the active site pocket and stabilizes the "closed" conformation of the 283-287 loop. The MeGlcA sugar ring in the Ϫ1 subsite is distorted into a 2 S o conformation, and the negative charge of the glucuronic acid induces the protonation of the nearby FIG. 6. A proposed catalytic mechanism for AguA based on its free and complexed structures. This mechanism presents Asp 364 and Glu 392 as the catalytic base, and Glu 285 as the catalytic acid. Active site residues that are important for the charge and position of the catalytic residues are also shown. Asp 364 and Glu 392 act together as general base, activating the nucleophilic water molecule, which then attacks the anomeric carbon of the substrate. In parallel, the general acid Glu 285 donates its proton to the glycosidic oxygen. The catalysis proceeds through an oxocarbenium ion-like transition state in which the bound MeGlcA is in a 2,5 B conformation and results in the inverted hydrolysis of the substrates.
Glu 285 . Next, Asp 364 and Glu 392 activate the nucleophilic water molecule, which attacks the anomeric carbon of the substrate, and concurrently Glu 285 donates its proton to the glycosidic oxygen. The hydrolysis proceeds through an oxocarbenium ionlike transition state, in which the MeGlcA sugar ring in the Ϫ1 subsite is likely to be distorted into a 2,5 B conformation. Last, the products are probably released from the active site in an ordered manner, in which the xylooligomer product leaves before the MeGlcA (16). The stereochemical course of the enzymatic reaction is well presented in the structures of the AguA complexes described above. The 1,2-glycosidic bond between the MeGlcA and the xylotriose in the intact substrate is clearly in an ␣ configuration, whereas in the complex of the enzyme with its products, the MeGlcA transforms into a ␤ configuration (Fig. 5a). The bound MeGlcA sugar ring is probably less stable in the ␤ configuration and slowly transforms back into the more stable ␣ configuration, as observed in the complex of E285N with MeGlcA. Alternatively, the enzyme has a higher affinity to the MeGlcA in the ␣ configuration, which diffuses into the active site at longer time periods. In any of these cases, the direct experimental observations agree well with previous assignment of GH-67 glycoside hydrolases as inverting enzymes (8).