Mannose Foraging by Bacteroides thetaiotaomicron

The human colonic bacterium Bacteroides thetaiotaomicron, which plays an important role in maintaining human health, produces an extensive array of exo-acting glycoside hydrolases (GH), including 32 family GH2 glycoside hydrolases. Although it is likely that these enzymes enable the organism to utilize dietary and host glycans as major nutrient sources, the biochemical properties of these GH2 glycoside hydrolases are currently unclear. Here we report the biochemical properties and crystal structure of the GH2 B. thetaiotaomicron enzyme BtMan2A. Kinetic analysis demonstrates that BtMan2A is a β-mannosidase in which substrate binding energy is provided principally by the glycone binding site, whereas aglycone recognition is highly plastic. The three-dimensional structure, determined to a resolution of 1.7Å, reveals a five-domain structure that is globally similar to the Escherichia coli LacZ β-galactosidase. The catalytic center is housed mainly within a (β/α)8 barrel although the N-terminal domain also contributes to the active site topology. The nature of the substrate-binding residues is quite distinct from other GH2 enzymes of known structure, instead they are similar to other clan GH-A enzymes specific for manno-configured substrates. Mutagenesis studies, informed by the crystal structure, identified a WDW motif in the N-terminal domain that makes a significant contribution to catalytic activity. The observation that this motif is invariant in GH2 mannosidases points to a generic role for these residues in this enzyme class. The identification of GH-A clan and GH2 specific residues in the active site of BtMan2A explains why this enzyme is able to harness substrate binding at the proximal glycone binding site more efficiently than mannan-hydrolyzing glycoside hydrolases in related enzyme families. The catalytic properties of BtMan2A are consistent with the flexible nutrient acquisition displayed by the colonic bacterium.

intriguing as their target substrates are found both in plant structural polysaccharides and in the N-glycan decorations of human proteins. Within the plant cell wall mannose is located in mannans, the backbone of which consists of a homopolymer of ␤1,4-linked D-mannopyranose residues, and glucomannan where the backbone comprises a heterogenous sequence of ␤1,4-linked D-glucose and D-mannose sugars (18). Both mannan and glucomannan are frequently decorated with galactosyl residues linked ␣1,6 to the backbone mannosyl moieties, which may also be acetylated at O-2 and/or O-3 (18). As with other plant cell wall polysaccharides (displayed in Fig. 1), the degradation of mannans involves an enzymatic consortium in which endo-mannanases first break down the polysaccharide into smaller oligosaccharides that are themselves further broken down to mannose through the action ␤-mannosidases (11). The situation in human N-glycans is potentially more complex, in that a repertoire of ␣ and ␤ exo-acting glycosidases are required to strip the glycan down to its terminal three sugars in which a D-mannosyl residue is linked ␤1,4 to the GlcNAc-GlcNAc base of the core.
Dissecting the mechanism by which B. thetaiotaomicron utilizes sugar polymers as major sources of nutrients is important in understanding how the bacterium is able to colonize the hind gut, where it promotes human health through the modulation of the immune system and by the synthesis of "anti-neoplaisic" molecules such as butyrate (20). Furthermore, the extensive portfolio of glycoside hydrolases produced by the bacterium could make a significant contribution to the degradation of plant biomass, which continues to be of evolving interest, not least because of the considerable importance of biofuels (21). Within a mechanistic context enzymes active on ␤-mannosidase and glycoside hydrolases that cleave ␣-linked mannosides are particularly interesting as they may use an unusual catalytic itinerary involving a boat-configured transition state (22)(23)(24)(25).
To start to unravel, both structurally and biochemically, the complex glycobiome of B. thetaiotaomicron we report here the cloning, expression, and enzymatic characterization of a GH2 ␤-mannosidase from B. thetaiotaomicron, hereafter BtMan2A. The enzyme is shown to be a classical ␤-mannosidase, with high activity on mannobiose and aryl ␤-mannosides (in contrast to exo-␤-mannanases (11)), although it is unable to accommodate the galactosyl decorations present on plant-derived mannans. The enzyme is also able to remove ␤-mannosides from Man-␤1,4-GlcNAc, found in the core pentasacharide of human N-glycans, consistent with the diversity of mannose-containing polymers presented to the bacterium in the large bowel. The three-dimensional structure of the enzyme at a resolution of 1.7 Å reveals a multidomain enzyme in which the active site pocket is housed within the TIM-barrel domain. Site-directed mutagenesis in conjunction with sequence comparisons of GH2 enzymes reveals active site residues that confer specificity for terminal mannosides.

MATERIALS AND METHODS
Cloning and Expression-Sequence comparisons indicated that open reading frame BT0458 encodes a GH2 mannosidase, defined hereafter as BtMan2A. Using primers detailed under supplementary materials the region of the gene encoding mature BtMan2A (residues 26 -839) was amplified from the B. thetaiotaomicron genomic DNA by PCR. The DNA generated was restricted with NdeI and XhoI (sites introduced in the primers employed in the PCR) and cloned into appropriately digested pET28a (Novagen) to generate pLT001. BtMan2A, encoded by pLT001, contains a C-terminal His 6 tag. To produce recombinant BtMan2A, E. coli strain BL21 (Novagen) containing pLT001 was cultured to mid-exponential phase at 37°C in Luria-Bertani media (LB). The culture was then cooled to 16°C and recombinant enzyme expression was induced by the addition of isopropyl ␤-D-thiogalactopyranoside to a final concentration of 100 M followed by incubation for a further 16 h at this lower temperature. The mannosidase was then purified by immobilized metal ion affinity chromatography, using TALON TM as the column matrix, ion exchange chromatography using a Q12 column, and size exclusion chromatography employing an S200 column (as described in, for example, in Ref. 11).
To produce seleno-L-methionine BtMan2A the E. coli methionine auxotroph B834(DE3) containing pLT001 was cultured as described following a similar protocol to that described by Charnock et al. (26) using the induction regime described above. The recombinant protein was purified as described above except all buffers were supplemented with 5 mM ␤-mercaptoethanol. All proteins were homogenous, as judged by SDS-PAGE, and matrix-assisted laser desorption ionization time-of-flight mass spectrometric analysis confirmed the incorporation of seleno-L-methionine into BtMan2A and that the enzyme had not been processed.
Construction of Site-directed Variants-Site-directed mutagenesis, employing a QuikChange kit (Stratagene), primers listed in supplementary information, and pLT001 as the template DNA, was used to construct mutants of man2A, encoding active site variants of BtMan2A. All mutants were sequenced using T7 FIGURE 1. Mannose foraging by BtMan2A. BtMan2A is able to utilize both undecorated manno-oligosaccharides (A) as a substrate, and also hydrolyzes the Man-GlcNAc disaccharide that may reflect targeting to the Man␤1,4-GlcNAc-␤1,4-GlcNAc core (B) of N-glycans. The mannosidase displays weak activity against manno-oligosaccharides that are substituted at the O-6 position of the first (R 1 ϭ ␣-Gal) or second (R 2 ϭ ␣-Gal) aglycone sugar.
forward and reverse and custom made primers to ensure that only the designed mutation had been introduced into the mannosidase gene.
Kinetic Analyses-The activity of BtMan2A was determined at 37°C in 43.5 mM sodium phosphate, 10 mM citric acid (PC) buffer, pH 5.6, containing 1 mg/ml bovine serum albumin and the appropriate amount of the substrate. Appropriate concentrations of BtMan2A were added to start the reaction, which was monitored at 400 nm during the linear initial phase (Ͻ5% of substrate hydrolyzed). The release of 2,4-dinitrophenolate (DNP) from 2,4-dinitrophenyl-␤-D-mannopyranoside (DNP-Man) was monitored continuously. When using the substrate 4-nitrophenyl-␤-D-mannopyranoside (PNP-Man) as the substrate, 4-nitrophenolate release (PNP) was measured discontinuously. At least four 200-l aliquots were removed and the reaction was stopped by the addition of 800 l of 1 M sodium carbonate that increased the pH to 12. The concentration of DNP and PNP produced was determined using a molar extinction coefficient of 15,000 and 10,000, respectively. For kinetic analysis at least six substrate concentrations that straddled the K m were employed and non-linear regression was used to estimate K m and k cat . The activity of BtMan2A against mannooligosaccharides and mannose-containing polysaccharides (supplied by Megazyme International, Bray, County Wicklow, Ireland, except Man-␤1,4-GlcNAc, which was provided by Dextra Laboratories, Reading, UK) was determined as described by Hogg et al. (14) except the reaction was carried out in PC buffer, pH 5.6. The rate of substrate hydrolysis was monitored by Dionex high performance anion-exchange chromatography (27) at a substrate concentration ϽK m (30 M) enabling k cat /K m to be determined by measuring the rate of substrate hydrolysis. The k cat for mannan hydrolysis was measured using a substrate concentration of 2 mg/ml. To determine the pH optimum of BtMan2A a discontinuous assay was used and 30 M PNP-Man as substrate and the following buffers: 50 mM sodium citrate, pH 3.3-6.5; 50 mM sodium phosphate, pH 6.0 -7.5; 50 mM Bicine, pH 6.5-9.0; 50 mM CAPSO, pH 8.0 -10.0; 50 mM CAPS, pH 9.5-11. Temperature and proteinase stability was determined as described by Andrews et al. (28). The K i for bis-Tris propane and isofagomine lactam was determined by measuring the K m for PNP-Man at three different concentrations of inhibitor. The enzyme was preincubated with the inhibitor for 1 h prior to carrying out the enzyme assays.
Crystallization and Data Collection-Pure proteins as judged by SDS-PAGE were concentrated to between 10 and 20 mg/ml and buffer exchanged into water (Sigma) using a Vivaspin 10-kDa cut-off concentrator. The protein was screened using the sitting drop vapor diffusion method together with the Hampton Crystal screen, Crystal screen 2, Hampton PEG/Ion (Hampton Research, Aliso Viejo, CA), the CSS Crystal screens I and II, and the PACT premier screen. The 100-nl droplets were formed by the mosquito (TTP LabTech Ltd, Royston, Herts, UK) liquid handling robot, mixing 50 nl of protein with 50 nl of well solution. Initial crystals were obtained in the PACT premier screen conditions F2 and PEG/Ion screen G5. These conditions were optimized further to improve crystal quality resulting in conditions of either 0.15-0.35 M NaBr and 6 -12% PEG 3350, 0.1 M bis-Tris propane, pH 7, or 0.08 -0.2 M KSCN, and 8 -15% PEG 3350,0.1 M bis-Tris propane, pH 7. A cryo-protectant solution was produced by supplementing the mother liquor with an additional 30% (v/v) glycerol. The lath-shaped crystals were harvested in rayon fiber loops then bathed in cryo-protectant solution prior to flash freezing in liquid N 2 . Selenomethionine crystals were produced in the same manner.
Data were collected at the European Synchrotron Radiation Facility (ESRF) from single crystals at 100 K with a ⌬ of 0.5°. Native data were collected at a wavelength of 0.97930 Å over an oscillation range of 200°on ID14-4 using an ADSC Q4R chargedcoupled device detector. Selenium derivative data were collected at a wavelength of 0.9795 Å over an oscillation range of 200°with ⌬ of 0.5°on ID23 using a Marmosaic 225 charged-coupled device detector. The appropriate wavelength for data collection at the Se edge, optimized for the ƒЉ component of the anomalous scattering, was determined using a fluorescence scan.
Structure Solution and Refinement-All data were indexed and integrated in MOSFLM (29). All other computing was undertaken using the CCP4 suite unless otherwise stated. Native BtMan2A crystals were found to belong to the space group P2 1 with the approximate cell dimensions of a ϭ 91.5 Å, b ϭ 116.0 Å, c ϭ 99.2 Å, and ␤ ϭ 113.4°with two molecules in the asymmetric unit. The structure was solved using singlewavelength anomalous dispersion methods. Selenium positions were determined using SHELXD (30) and phases subsequently calculated with MLPHARE from the CCP4 suite (31). Solvent flattening and phase improvement were carried out in DM (32) also from the CCP4 suite. 5% of the data were set aside for cross-validation analysis and the behavior of R free was used to monitor and guide the refinement protocols. ARP/wARP (33), in conjunction with REFMAC (34), was used to automatically build the sequence into the electron density. Refinement was undertaken in REFMAC with manual correction to the model using COOT (35). Coordinates and observed structure factor amplitudes have been deposited with the Protein Data Bank. Figures were drawn with MOLSCRIPT (36) and BOBSCRIPT (37).

Selection and Expression of a GH2
␤-Mannosidase-To identify a likely B. thetaiotaomicron GH2 ␤-mannosidase, the sequences of known, characterized ␤-mannosidases were used for BLAST searches of the UniProt sequence data base. Open reading frame BT0458 (hereafter designated man2A), one of the 32 family GH2 sequences in this organism (identified by the CAZy data base Refs. 4 and 5), displayed significant sequence similarity with both characterized and putative GH2 ␤-mannosidases, exhibiting 34 and 36% identity with the Thermotoga maritima and Thermatoga neopolitana ␤-mannosidases, respectively.
The man2A gene comprises 2592 bp encoding a 864-amino acid protein (BtMan2A) with a M r of 96,922 in which the 26 N-terminal residues are predicted to comprise a cleavable signal peptide. The man2A gene encoding mature BtMan2A was successfully expressed in E. coli, and the protein was purified by immobilized metal ion affinity chromatography to electrophoretic homogeneity (data not shown). The protein had a M r , determined by matrix-assisted laser desorption ionization time-of-flight mass spectrometric analysis, consistent with the predicted size of the processed enzyme (data not shown) indicating that the enzyme was not proteolytically degraded during its expression or subsequent purification.
Biochemical Properties of BtMan2A-BtMan2A was screened for catalytic activity using both a series of aryl-glycosides and dyed polysaccharides. The results showed that the enzyme displays no activity against xylose, arabinose, or galactose contain-ing substrates, trace activity against 4-nitrophenyl-␤-D-glucopyranoside but significant activity against PNP-Man. The enzyme displays no activity against ␣-D configured glycosides or any dyed polysaccharides. The data were entirely consistent with the designation of BtMan2A as a ␤-mannosidase displaying no endo-activity. The kinetic parameters of BtMan2A (Table 1 and Fig. 2) show that the enzyme is ϳ100,000-fold more active against manno-compared with gluco-configured substrates. Analysis of the reaction products released by BtMan2A from a range of ␤1,4-linked manno-oligosaccharides showed that mannose and a manno-oligosaccharide with a degree of polymerization of n-1, where n is the degree of polymerization of the substrate, were the initial reaction products demonstrating that the enzyme displays classical exoactivity, releasing mannose from the non-reducing end of polymeric substrates. Even at a manno-oligosaccharide concentration of 1 mM BtMan2A produced equal amounts of mannose and the n-1 product; no oligosaccharides larger than the substrate were produced. Thus, under the conditions used, BtMan2A does not display significant transglycosylating activity, in contrast to many retaining glycosidases (extensively reviewed in Ref. (38)). This may indicate that the ϩ1 subsite (39) does not bind mannose tightly and thus an activated water is more likely to be generated in the The activity of BtMan2A against the polysaccharides was evaluated using 0.5% substrate and enzyme at a concentration of 40 nM.
proximal aglycone region of the active site. The enzyme displays extremely low activity against 6 1 -␣-D-galactosyl-mannobiose and 6 1 -␣-D-galactosyl-mannotriose (ϳ40-fold lower activity than against mannobiose and mannotriose, respectively), indicating that decoration of the aglycone region of the substrate located in the ϩ2 or ϩ1 subsite can be tolerated, albeit poorly whereas no detectable hydrolysis of the doubly substituted 6 3 ,6 4 -␣-D-galactosyl-mannopentaose ( Fig. 3 and Table 1) was evident. The specificity of BtMan2A for linear unsubstituted mannans (Fig. 1) and its inability to exhibit any endo-activity explains why the enzyme is only able to produce small amounts of mannose from galactomannan or glucomannan (data not shown). Analysis of the biophysical properties of BtMan2A showed that the enzyme has a pH optimum of 5.6, typical of many GH2 enzymes, is inactivated at temperatures in excess of 58°C and is completely resistant to proteolytic attack. This latter feature is typical of extracellular glycoside hydrolases (40), and is consist-ent with the highly proteolytic environment of the large bowel where numerous anaerobic bacteria secrete a range of different proteinases.
The much higher activity (ϳ500fold) of BtMan2A for aryl-mannosides compared with disaccharides substrates reflects both the better leaving group abilities of PNP and DNP and perhaps also that the ϩ1 subsite binds sugar molecules weakly and nonspecifically. The ϩ2 subsite contributes a modest 1.5 kcal/mol to productive substrate binding. By contrast the Cellvibrio mixtus exo-mannanase, CmMan5A, is ϳ30,000-fold less active than BtMan2A against PNP-Man highlighting the pivotal role of the glycone (Ϫ1) binding site in the Bacteroides enzyme, whereas the C. mixtus glycoside hydrolase derives much of its catalytic power from substrate binding to the aglycone subsites (11). Indeed, in considerable contrast to BtMan2A the GH5 enzyme is 100-and 10,000fold more active against the ␤1,4-linked manno-oligosaccharides mannobiose and mannotriose, respectively, than PNP-Man. The importance of the Ϫ1 subsite in BtMan2A is further emphasized by its inhibition, relative to CmMan5A, by monosaccharide inhibitors. Isofagomine lactam, displays a K i of 400 M on CmMan5A (43) but binding is ϳ36 times stronger to BtMan2A with a K i value of 11 M, essentially identical for the K i of 9 M reported for the classical snail ␤-mannosidase (44). Structure of BtMan2A-The structure of BtMan2A was solved using selenomethionine single wavelength anomalous dispersion data to 2.8 Å in harness with native data to 1.7 Å (see experimental statistics in Table 2). The P2 1 crystal form contains a dimer, consistent with dynamic light scattering experiments and size exclusion chromatography (not shown) indicating that the protein is also a dimer in solution. BtMan2A consists of five clear domains (Fig. 4A) with the catalytic center the central of these. Such a modular organization is similar to that seen in some other GH2 enzymes such as the E. coli (7) and Arthrobacter sp. GH2 ␤-galactosidase (8). Domain 1 (colored blue in Fig. 4A) consisting of residues 28 -218, is a ␤ sheet domain containing a five-stranded antiparallel ␤ sheet, a fourstranded antiparallel ␤ sheet, and an ␣ helical region, reminiscent of carbohydrate-binding modules (reviewed in Ref. 45). Domains 2 (cyan) and 4 (yellow) are structurally very similar to domain 1, again displaying immunoglobulin/CBM-like folds, which, in this case, comprised residues 219 -331 and 676 -780, respectively. Both these domains consist of two antiparallel ␤ sheets one containing four ␤ strands and the other three. Domain 3 (green) contains the catalytic center and has the classical (␤/␣) 8 "TIM" barrel fold consisting of residues 332-675, discussed below, whereas domain 5 (red) comprises residues 786 -860. The first two residues of the C-terminal His tag are evident in one but not both subunits of the homodimer. The structure of BtMan2A differs from that of the E. coli LacZ ␤-galactosidase most significantly in the orientation of domain 5, which in LacZ is bent away from the active site giving the protein a more globular shape. The dimerization interface of BtMan2A is formed by interactions between domain 5 of both monomers, whereas in the E. coli ␤-galactosidase the equivalent domain also forms the basis of the interactions which lead to tetrameric oligomerization.
A DALI (46) search on the whole protein shows that BtMan2A is most similar to the H. sapiens ␤-glucuronidase and the E. coli ␤-galactosidase with an root mean square deviation of 2.9 and 3.9 Å and Z scores of 24.5 and 23.0, respectively. Taking each of the five domains separately, reveals that domain 1 (residues 2-192) has the highest similarity to the N-terminal domains of both these proteins with a root mean square deviation of 2.8 Å and a Z score of 12.4 with the E. coli ␤-galactosidase and root mean square deviation ϭ 2.9 Å and Z ϭ 11.8 for the H. sapiens ␤-glucuronidase. Domains 2 (219 -331) and 4 (676 -780) both show most similarity with the equivalent domains of the E. coli ␤-galactosidase with Z ϭ 12.7 and 10.3, respectively. Despite the general similarities between the E. coli and BtMan2A proteins the fifth domain is clearly significantly different. A DALI search carried out only on residues in this domain (758 -834) results in none of the first hits being for the GH2 enzymes. The most similar structure (Z ϭ 6.1) found during this search is to an appendage domain in mice involved in endocytosis. With almost the same Z score (5.9) an "immunoglobulin-like domain" in a sialidase from Micromonospora viri-   8 -fold that contain a conserved catalytic apparatus and hydrolyze the glycosidic bond by a double displacement or "retaining" mechanism) with DALI Z-scores of Z ϭ 23.5 to the human GH2 ␤-glucuronidase, Z ϭ 19.7 with the exo-mannanase CmMan5A, and Z ϭ 17.7 to the E. coli ␤-galactosidase. The catalytic residues, implicated as Glu-462 and Glu-555, lie on strands 4 and 7, respectively, of the (␤/␣) 8 barrel as expected, and the active site comprises a deep pocket. Comparison of the active site of BtMan2A with that of the E. coli ␤-galactosidase shows that in the case of the former domain 2 plays no role in the formation of the catalytic site, whereas in the ␤-galactosidase a loop from domain 2 forms part of the ␤ barrel. There are substantial differences between the active sites of the two proteins where strands 7, 8, 1, and 2 lie in similar orientations and the remaining strands in BtMan2A are shifted such that they occupy the same regions as occupied by the domain 2 loop in the E. coli ␤-galactosidase. Notably, in both E. coli ␤-galactosidase and BtMan2A, aromatic residues derived from domain 1 contribute to the base of the active site pocket and thus to the "exo" activity of the enzymes (discussed further below).
The active center of BtMan2A reveals unambiguous electron density for a single molecule of bis-Tris propane in the active site from the crystallization conditions (Fig. 4B). Tris itself has been observed in a number of glycosidase structures (19) and bis-Tris propane binds similarly here, with one of the Tris moieties occupying the Ϫ1 subsite such that its amino group lies approximately in the position expected for the positive charge on the transition state during catalysis and with the hydroxyl groups lying close to the positions expected for the mannoside hydroxyls of the true substrate. Bis-Tris propane indeed acts as a competitive inhibitor of the enzyme with a K i determined to be 100 M at pH 7 (see "Materials and Methods). Thus far, we have been unable to crystallize the protein in a bis-Tris propane-free form for ligand-binding studies.
Comparison of the active site of BtMan2A with the GH5 exo-mannanase CmMan5A reveals significant similarity between the Ϫ1 subsite of these two enzymes (Fig. 5) suggesting that there is some conservation in mannosyl binding machinery in these proteins. The position and orientation of the two catalytic carboxylates in the retaining mechanism, Glu-462 and Glu-555 in BtMan2A, are essentially identical to the corresponding residues, Glu-215 and Glu-330, respectively, in CmMan5A (Fig. 5). Consistent with the roles of these residues, the E462A and E555A mutants display catalytic efficiencies ϳ10 3 -and 10 5 -fold lower, respectively, than the wild type enzyme (Table 1 and Fig. 2). Indeed, the acid-base catalytic function of Glu-462 is supported by the K m of E462A for DNP-Man, which is ϳ30-fold lower than the wild type enzyme (Fig. 2 and Table 1). The decrease in this kinetic parameter indicates that the mutation has had a larger influence on k3, the deglycosylation step, than glycosylation (k2) in which protonation of the glycosidic oxygen is not critical when the leaving group has a pK a significantly lower than the pH of the reaction (pK a of DNP is 3.5). Detailed biochemical analysis of wild type and sitedirected mutants of the related C. fimi GH2 ␤-mannosidase (41,42), and other clan GH-A glycoside hydrolases, including GH10 xylanases, GH26 endo-␤1,4-mannanases, and GH53 endo-␤1,4-galactanses demonstrate that the equivalent residues to BtMan2A Glu-462 and Glu-555, in these related enzymes, fulfill the same catalytic functions.
In addition to the "catalytic" groups, the aromatic tryptophan platform "below" the Ϫ1 subsite is maintained in almost identical orientation between BtMan2A and CmMan5A (Fig. 5). In the case of BtMan2A Trp-645 provides this function, whereas the equivalent in CmMan5A is Trp-376. The importance of this residue in substrate binding, presumably via hydrophobic contacts with the mannosyl residue at the Ϫ1 subsite, is demonstrated by the W645A mutation causing a substantial increase in K m and a 10 4 -fold decrease in catalytic efficiency (Table 1). Mutagenesis studies also indicate that three other aromatic residues in the glycone binding site, Trp-198, Trp-200, and Trp-395 (the latter residue corresponds to Trp-137 in CmMan5A), contribute to substrate binding as the respective mutations W198A, W200A, and W395A significantly increase the K m and decrease the catalytic efficiency of the enzyme. By analogy to Trp-137 in CmMan5A, Trp-395 makes a major contribution to the pocket topology of the active site and hence the exo-activity of the enzyme, and is predicted to hydrogen bond to O-3 of the substrate (  into the active site (Fig. 5) and the carboxylic acid clearly makes a substantial contribution to both substrate recognition and transition state stabilization as the D199A mutation displays greatly reduced activity. Asn-461, whose orientation is stabilized through a bidentate hydrogen bond with Arg-391, is highly conserved in clan GH-A enzymes (although in GH26 it is a histidine), and by making a hydrogen bond with O-2 of the glycone sugar plays a key role in substrate binding and transition state stabilization. The low activity displayed by the mutant N461A is entirely consistent with the prominent role Asn-461 plays in catalysis contributing around 4.3 kcal/mol to catalysis, similar to that observed for the same interaction for enzymes acting on gluco-configured enzymes (6).
Comparison of BtMan2A with CmMan5A in complex with the inhibitor isofagomine lactam (43) allows identification of other residues in BtMan2A that will interact with substrate. Thus, Tyr-537 is predicted to make a hydrogen bond with substrate O-6 and, indeed, the Y537A mutation causes a 100-fold reduction in catalytic efficiency consistent with the importance of this tyrosine in catalysis. Whereas all of the functionalities in the CmMan5A active center are provided by its single (␤/␣) 8 domain this is not true of BtMan2A because a loop from domain 1 protrudes into the catalytic center and donates a number of residues that would interact with the substrate in the Ϫ1 subsite such as Asp-199 (described previously), which is spatially (but not topologically) equivalent to Glu-404 of CmMan5A (Fig. 5).

DISCUSSION
BtMan2A is a classical exo-enzyme; but, in marked contrast to the exo-mannanase CmMan5A, whose binding energy is derived predominantly from aglycone subsites (11), the Bacteroides enzyme is considerably more active on aryl-mannosides and disaccharide substrates (Table 1), and better inhibited by the monosaccharide inhibitor isofagomine lactam. BtMan2A thus displays features typical of a mannosidase. Inspection of the glycone binding site of BtMan2A reveals some similarity with CmMan5A and active side residues Trp-395, Asn-461, Glu-462, Tyr-537, Glu-555, and Trp-645 in the Bacteroides enzyme are equivalent to Trp-137, Asn-214, Glu-215, Asn-288, Glu-330, and Trp-376 in the Cellvibrio glycoside hydrolase. BtMan2A, however, must contain several additional mannosyl binding residues to account for its increased activity, relative to CmMan5A, against substrates that do not bind tightly to the aglycone subsites. Inspection of the deep pocket that comprises the glycone binding site of BtMan2A reveals several hydrophobic residues that are not present in CmMan5A, and which mutagenesis data show play an important role in substrate binding. Thus, although Trp-395 is equivalent to Trp-137 in CmMan5A, the BtMan2A tryptophan is displaced in the same plane and its indole nitrogen may interact with O-2, rather than O-3 (as in the case of Trp-137) of the substrate. Thus Trp-395 may be an important specificity determinant that distinguishes mannose, where O-2 is axial, whereas in other sugars the equivalent hydroxyl is equatorial. Trp-198 and Trp-200, which are not present in the Cellvibrio enzyme, (although Pro-401 in CmMan5A is structurally equivalent to Trp-200) are likely to make hydrogen bonds through N⑀-1 with O-3 and O-4 of the substrate, respectively, interactions that are not replicated in CmMan5A (the nitrogen in Pro-401 is too distant from the substrate to make a hydrogen bond). Indeed the three residue WDW motif, is in a domain that is not present in the Cellvibrio enzyme and it is possible that one or more of these amino acids may make a substantial contribution to the higher activity displayed by BtMan2A compared with CmMan5A for substrates that rely on tight binding to the Ϫ1 subsite. Inspection of the sequence of GH2 mannosidases (both putative and experimentally demonstrated) show that the WDW motif is invariant, providing support for the key role these residues play in the activity and possibly the specificity of these enzymes, whereas Trp-536, which is conserved in Ͼ90% of these enzymes may also contribute to catalysis in these glycoside hydrolases. The recruitment of functional residues, from domain 1, to provide exo-catalytic activity upon what might have been an endo-catalytic framework is entirely consistent with proposals made by Jain and colleagues (9) in respect of the human ␤-glucuronidase. In the absence of domain 1, BtMan2A would also appear to have a vestigial "endo" substrate-binding cleft.
In the family GH5 endo-mannanases from Hypocrea jecorina (formerly known as Trichoderma reesei), complexes are available with ligands in the aglycone sites (13). By inference the residues that are expected to bind the aglycone in the GH5 exo-mannanase CmMan5A are Trp-135 (ϩ1 susbsite), Arg-217 (ϩ2 subsite), and Trp-289 (ϩ2 subsite) (11). In contrast, BtMan2A (reported here) has no equivalent tryptophan to Trp-135 in the predicted ϩ1 subsite, instead having a cysteine residue at this position. Indeed, the topography of the cleft is quite different between the two enzymes that may contribute to the diverse specificity of BtMan2A consistent with a role in aiding glycan foraging by this glycosidase-rich organism. There is no direct equivalent to Arg-217, whose place is instead taken by Trp-444 in BtMan2A, whereas Trp-519 of the Bacteroides enzyme is roughly equivalent to Trp-289 of the Cellvibrio exomannanase. This divergence in distal aglycone binding sites is reflected in the binding energies of the subsites in that a ϩ2 mannoside contributes only 1.5 kcal/mol of binding energy of BtMan2A, as opposed to 3.5 kcal/mol in CmMan5A (11).
The plasticity in substrate recognition displayed by BtMan2A is consistent with the colonic ecosystem of the bacterium. The organism utilizes both dietary and host glycans as major sources of nutrients. The substrate specificity of BtMan2A, which releases mannose from both ␤1,4-linked manno-oligosaccharides and from the Man␤1,4GlcNAc␤1,4GlcNAc stem of the core region of N-linked human glucans, reflects the nutrient sources presented to the prokaryote. Indeed man2A is part of an operon containing a sialidase, a sialic acid-specific 9-O-acetylesterase, and three ␤-hexoaminidase genes, which degrade mammalian glycans (2), and this genetic locus is up-regulated when host mice are fed diets lacking fermentable polysaccharides, enabling the bacterium to scavenge carbohydrates from the mammals glycoproteins (2). Thus the most common natural substrate for B. thetaiotaomicron is likely to be Man-GlcNAc derived from the host glycoproteins. By contrast, in CmMan5A the aglycone binding sites play a key role in substrate recognition and thus the enzyme displays much tighter specificity for mannose-containing homopolymers. Although it is currently unclear whether the plasticity in aglycone recognition displayed by BtMan2A is a general feature of GH2 ␤-mannosidases (the activity of these enzymes are generally assessed using aryl-glycosides as substrates), it is interesting to note that the genome of B. thetaiotaomicron contains 32 GH2 open reading frames, and it is possible that tighter aglycone substrate recognition may confer more specialized roles for at least some of these enzymes. Indeed, the glycobiome of B. thetaiotaomicron, in which there has been a dramatic expansion in GH2 and GH43, which to date contain primarily exo-acting enzymes, but a paucity of enzymes in the families known to contain primarily endo-acting plant structural polysaccharidases, with no obvious candidate endo-␤1,4-mannanases, may suggest that the bacterium specializes in scavenging sugars both from other oligosaccharides produced by endo-acting glycoside hydrolases, presumably expressed by colonic microorganisms, and from the diverse human glycans in the colonic milieu.