Synergy between Cell Surface Glycosidases and Glycan-Binding Proteins Dictates the Utilization of Specific Beta(1,3)-Glucans by Human Gut Bacteroides.

Bacteroidetes are a dominant phylum of the human gut microbiota (HGM) that target otherwise indigestible dietary fiber with an arsenal of polysaccharide utilization loci (PULs), each of which is dedicated to the utilization of a specific complex carbohydrate. Here, we provide novel insight into this paradigm through functional characterization of homologous PULs from three autochthonous Bacteroides species, which target the family of dietary β(1,3)-glucans. Through detailed biochemical and protein structural analysis, we observed an unexpected diversity in the substrate specificity of PUL glycosidases and glycan-binding proteins with regard to β(1,3)-glucan linkage and branching patterns. In combination, these individual enzyme and protein specificities support taxon-specific growth on individual β(1,3)-glucans. This detailed metabolic insight, together with a comprehensive survey of individual 1,3GULs across human populations, further expands the fundamental roadmap of the HGM, with potential application to the future development of microbial intervention therapies.


H T C S T B D T S G B P -A S G B P -
. Gene products are colored analogously to panel A, and predicted N-terminal lipidation following signal peptidase II cleavage is indicated with a black squiggle.  in minimal medium containing the indicated carbon source at 0.5%, wt/vol. Points represent averages from n ϭ 6 technical replicates (microplate wells) for polysaccharides and n ϭ 3 for glucose; error bars represent the standard errors of the means. (E) RNA abundance for core PUL genes of Bacteroides spp. quantified by qRT-PCR. Bacteria were grown to mid-log phase in minimal medium containing glucose as the sole carbon source and subsequently exposed to different ␤-glucans (n ϭ 3, expression measurements from individual cultures, relative to glucose control; error bars represent the standard errors of the means).

B. uniformis possesses a distinct PUL upregulated during growth on
Biochemical basis of ␤(1,3)-glucan recognition and degradation by B. uniformis. By analogy with other PUL-encoded systems, we propose a working model of concerted ␤(1,3)-glucan saccharification and uptake by the proteins of the 1,3GUL (Fig. 1). This model involves polysaccharide capture at the cell surface by at least one SGBP, backbone hydrolysis by at least one endo-glucanase, transport of oligosaccharide fragments through the outer membrane by the TBDT, and ultimate saccharification by an exo-␤-glucosidase in the periplasm. Indeed, signal peptide analysis with SignalP, LipoP, and PSORTb predicted that BuGH16 is localized at the cell surface via N-terminal Cys lipidation (PSORTb score, 9.7), whereas BuGH3, which also harbors a type II signal peptide, is predicted to be periplasmic (PSORTb score, 9.4). Interestingly, PSORTb was unable to predict the localization of BuSGBP-A, BuSGBP-B, and BuGH158 (score, Ͻ2.5) despite the presence of a type II signal peptide, a ϩ2 serine residue in each, and expected extracellular localization of at least the SGBPs by analogy with the archetypal starch utilization system (9,10).
Subsequent Michaelis-Menten kinetics at the optimum pH and 37°C confirmed that BuGH16 is a broadly specific ␤(1,3)-glucanase with similar catalytic efficiencies toward laminarins and yBG ( Fig. 3A and Table S2 at the URL mentioned above). The k cat /K m value of BuGH16 on bMLG was 4-fold lower than those for these all-␤(1,3)-backbone glucans ( Fig. 3A and Table S2 at the URL mentioned above). The enzyme was very poorly active on the unbranched ␤(1,3)-glucan, curdlan (k cat /K m value ca. 1% of that on the laminarins) ( Fig. 3A and Table S2 at the URL mentioned above). LdLam and yBG ultimately were hydrolyzed to glucose, laminaribiose, and a trisaccharide (G3G6G or G6G3G), as determined by high-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) (Fig. 3C) and confirmed by matrixassisted laser desorption ionization mass spectrometry of the per-O-acetylated crude [Substrate] (mg/mL)   mixture (data not shown). The limit-digest products were glucose and laminaribiose from curdlan, and G4G3G and G4G4G3G from bMLG (Fig. 3C). Analyses of the hydrolysis products over time suggest that BuGH16 operates by an endo-dissociative mode of action ( Fig. S2 at the URL mentioned above). Despite extensive attempts using the full-length protein and the GH16 module only (Fig. S3 at the URL mentioned above), we were not able to obtain a crystal structure of BuGH16 to explain the observed substrate promiscuity. However, protein phylogeny unambiguously places BuGH16 in the "␤-bulge laminarinase/MLGase" clade ( Fig. S4 at https://doi.org/10.14288/1.0388792), corresponding to GH16 subfamily 3 (38), which is known to contain laminarinases with secondary MLGase activity, as well as bona fide MLGases (19). These enzymes require a ␤(1,3) linkage between the Ϫ1 and Ϫ2 subsites (44), but beyond the Ϫ2 subsite, requirements are less stringent, leading to widespread promiscuity in polysaccharide and bond cleavage specificity Notably, structural homology modeling and superposition with an exemplar laminarinase from the marine bacterium Zobellia galactanivorans DsijT (45) rationalizes the ability of BuGH16 to accommodate highly branched ␤(1,3)-glucans, i.e., laminarins and yBG ( Fig. S5 at the URL mentioned above).
In addition to the catalytic module, BuGH16 comprises a pair of PFAM 13004 domains ( Fig. S3 at the URL mentioned above). Despite initial bioinformatics predictions, carbohydrate binding has not been demonstrated for any PFAM 13004 domain to date (15). Likewise, affinity gel electrophoresis (AGE) analysis of a recombinant protein consisting of the two PFAM 13004 domains of BuGH16 revealed that they do not bind cognate polysaccharides of the 1,3GUL (Fig. S6 at the URL mentioned above). These domains likely serve a spacer function, as in B. ovatus GH5 (BoGH5) (15), and are analogous to all ␤-sheet domains in SGBPs (16,26,46).
(ii) BuGH158 is a strictly specific, retaining endo-␤(1,3)glucanase with a triose phosphate isomerase (TIM) barrel fold. BuGH158 is a member of a newly established GH family, the distantly related founding member of which, Vvad_PD1638 (sequence identity 29%) ( Fig. S7 at https://doi.org/10.14288/1.0388792), was shown to be active on the artificial proxy substrate, carboxymethyl-curdlan, in a high-throughput screen (40). Hence, we performed detailed kinetic and product analysis to more precisely delineate the specificity of BuGH158 in the context of the 1,3GUL ( Fig. 3B and Table S2 at the URL mentioned above). In contrast to BuGH16, BuGH158 is highly specific for LdLam, with a k cat /K m value ca. 2 orders of magnitude higher than those for EbLam, yBG, MLG, and curdlan ( Fig. 3B; also see and Table S2 and Fig. S1 and S2 at the URL mentioned above). The corresponding hydrolysis products were identical to those of BuGH16 (Fig. 3D).
To provide the first three-dimensional insights into substrate specificity and catalysis in GH158, we determined the tertiary structure of the enzyme to 1.8 Å by X-ray crystallography (Table S3 at the URL mentioned above). BuGH158 consists of an N-terminal (␣/␤) 8 TIM barrel domain and a C-terminal, eight-stranded immunoglobulin (Ig)-like domain in contact with helices ␣7 and ␣8 of the TIM barrel. Additional loops contribute to the extensive contact between domains, with one from the Ig-like domain extending above the TIM barrel to shape the active-site cleft (Fig. 4A). A Dali structure similarity search returned members of clan GH-A as the top 20 structural homologs (Table S4 at the URL mentioned above), confirming the bioinformatics prediction that GH158 constitutes an evolutionarily distinct family within this clan (40). Correspondingly, superposition with the top two GH2 and GH5 results show that the TIM barrel core is well conserved and that family-level differences arise from the presence or absence of accessory domains ( Fig. 4B and C).
All clan GH-A enzymes are predicted to use a canonical Koshland doubledisplacement mechanism employing a covalent glycosyl enzyme intermediate and resulting in overall retention of the anomeric stereochemistry at the site of polysaccharide backbone hydrolysis (39). This was confirmed for BuGH158 by nuclear magnetic resonance (NMR) analysis of the hydrolysis of 2-chloro-4-nitrophenyl laminaribioside (G3G-CNP), which constitutes the first stereochemical determination for the family (Fig.  S8 at the URL mentioned above). Clan GH-A members present the conserved catalytic acid/base and nucleophile residues on loops immediately following strands ␤4 and ␤7, corresponding to E137 and E220, respectively, in BuGH158 (Fig. 4D). Additionally, a conserved asparagine (N136) precedes the general acid/base (E137) and is anticipated to engage in hydrogen-bonding interactions with the substrate (Fig. 4D) (39). Indeed, site-directed mutation of N136, E137, or E220 to alanine completely abolished catalytic activity (data not shown).
(iii) Periplasmic saccharification of imported oligosaccharides is mediated by BuGH3, a specific exo-␤(1,3)-glucosidase. GH3 is known to contain members with a diversity of exo-␤-glycosidases (39), warranting detailed kinetic characterization of BuGH3 in the context of the 1,3GUL. Initial substrate screening on chromogenic pNP glycosides revealed that BuGH3 is an exo-␤-glucosidase (no other activity detected on a panel of pNP substrates; see Table S5 at https://doi.org/10.14288/1.0388792). Subsequent Michaelis-Menten kinetics on diverse ␤-gluco-oligosaccharides further established that BuGH3 is a specific ␤(1,3)-glucosidase poised to efficiently handle the hydrolysis products of the predicted cell surface GH16 and GH158 enzymes (Table S5 and Fig. S9 at the URL mentioned above). Indeed, incubation with the limit-digest products of either BuGH16 or BuGH158 confirmed that BuGH3 is capable of completely degrading all oligosaccharide products of laminarin and bMLG to glucose ( Fig. S2 and S10 at the URL mentioned above).
Despite possessing an apparently broad ability to hydrolyze ␤(1,3)-, ␤(1,4)-, and ␤(1,6)-glucosides, the particular preference of BuGH3 for ␤(1,3)-linkages is highlighted by a two-orders-of-magnitude higher k cat /K m value for laminaribiose (G3G) over cellobiose (G4G) and gentiobiose (G6G) ( Table S5 at the URL mentioned above). Concordant with this finding, catalytic efficiency toward mixed-linkage trisaccharides with a ␤(1,3)glucose at the nonreducing end (G3G4G) was 2 orders of magnitude higher than that with a ␤(1,4)-glucose at the nonreducing end (G4G3G) ( Table S5 at the URL mentioned above). Laminari-oligosaccharides of increasing degree-of-polymerization were hydrolyzed with comparable k cat values, but K m values decreased between the di-and trisaccharide, after which K m values leveled off, suggesting that the BuGH3 has two positive subsites of kinetic significance.
Divergent GH and SGBP specificities collectively dictate the range of ␤-glucans utilized by Bacteroides species. Using the B. uniformis 1,3GUL as the archetype, we identified several homologous 1,3GULs from the Bacteroidaceae, Porphyromonadaceae, and Prevotellaceae families (order Bacteroidales), of which five representatives are shown in Fig. 1. All comprise a syntenic TBDT, SGBP-A, SGBP-B, and GH3 as the core set of conserved genes. However, the predicted endo-glucanase (GH16 and GH158) content is notably variable, and the SGBPs have particularly low sequence similarity, including among related Bacteroides species ( Fig. 1 and Table S8 Table S1 at the URL mentioned above).
To elucidate the molecular basis of this species-specific ␤-glucan utilization, we characterized the transcriptional response of the B. thetaiotaomicron and B. fluxus 1,3GUL (as a measure of HTCS specificity) and the biochemistry of the predicted surface endo-glucanases and SGBPs vis-à-vis the B. uniformis system. In B. thetaiotaomicron and B. fluxus, the genes encoding the TBDT and SGBP-A were strongly upregulated by

®
LdLam and yBG, whereas they were only very weakly activated by bMLG (Ͻ10-fold) (Fig. 2E). These results are generally concordant with the observed growth phenotypes among all three Bacteroides species, although we unexpectedly observed strong upregulation of the B. fluxus 1,3GUL (Bf1,3GUL) with yBG, on which B. fluxus is not able to grow. The Bf1,3GUL still was highly activated after dialyzing the substrate to remove any potential small oligosaccharide signals (Fig. S13 at the URL mentioned above), suggesting that this system is capable of sufficiently cleaving yBG to release an oligosaccharide that activates this PUL but subsequently unable to grow appreciably on the bulk polysaccharide. These distinct regulatory profiles likely involve a combination of surface GH and transporter specificity, combined with the sensing abilities of the respective HTCS regulators. The latter signaling specificity is notable in light of high protein sequence similarity among species (80 to 85% amino acid identity and 88 to 91% similarity versus the BuHTCS; Table S8 at the URL mentioned above), and the results below show that at least B. thetaiotaomicron GH16 (BtGH16) can cleave bMLG but does generate an activating cue. Whereas B. uniformis encodes both a GH16_3 member and a GH158 member in its 1,3GUL, B. thetaiotaomicron possesses only a GH16_3 member while B. fluxus possesses only a GH158 member. Enzymology revealed that BtGH16 is a predominant laminarinase with lower, yet comparable, activity on yBG and bMLG, similar to BuGH16 (Fig. S14 and Table S2 at the URL mentioned above; the homologous Prevotella loescheii GH16 [PlGH16] also had broad activity). In contrast, BfGH158 was highly specific for LdLam, with low activity on yBG and very poor activity on bMLG, similar to BuGH158 (Table S2 at the URL mentioned above).
We  (Table S6 at the URL mentioned above).
These biochemical data show that whereas orthologous GH specificity is well predicted by the family to which it belongs, considerable diversity exists in the specificity of syntenic SGBPs encoded by 1,3GULs of B. uniformis, B. thetaiotaomicron, and B. fluxus. When combined with microbiological data, a pattern emerges whereby GH content combined with SGBP specificity predicts the range of ␤-glucan congeners a Bacteroides species can utilize. In parallel, the signaling specificity or promiscuity through the single HTCS sensor associated with each system must also be able to respond to the diversity of cues generated from cleavage of related substrates. Of the three species examined, only B. uniformis possesses the full complement of synergistic HTCS, GH, and SGBP specificities to enable growth on laminarin, yBG, and bMLG (summarized in Fig. 6).
Metagenomic survey reveals global 1,3GUL distribution in the human gut. To determine the prevalence of 1,3GULs in the human gut, we surveyed publicly available gut metagenomic data from 2,441 adults across five different continents (North American, South America, Africa, Europe, and Asia) (Fig. 7). Despite similar genetic synteny, these 1,3GULs have different nucleotide sequences (pairwise identities for B. fluxus/B. thetaiotaomicron, 44%; B. fluxus/B. uniformis, 60%; B. thetaiotaomicron/B. uniformis, 44%), allowing us to use the nucleotide sequence to distinguish the presence or absence of a specific 1,3GUL in each metagenomic sample. The B. uniformis type 1,3GUL is the most abundant across all samples (48% of the total population), perhaps reflecting the high prevalence of this species in humans from industrialized populations. This is followed by the B. thetaiotaomicron type 1,3GUL (26% of samples) and B. fluxus (0.53% of samples). The low prevalence of B. fluxus may be explained by its low abundance and prevalence in European, North American, and Asian populations, consistent with previous observations for the xyloglucan utilization locus (15). Although 1,3GULs are widely distributed throughout human populations (59%), we do not see correlation with any particular geographic region or population. This ubiquity may be explained by the prevalence of ␤(1,3)-glucans in different diets, since edible fungi and yeast fermentation products are common worldwide. Strikingly, we were unable to detect any of the three identified 1,3GULs in the indigenous Hadza and Yanomami tribes, which may be due to a high prevalence of Prevotella and not Bacteroides in these populations (51,52). The present study, in combination with our previous metagenomic survey of MLG utilization (19), reveals a broadly represented potential for specific ␤-glucan metabolism to establish niches for individual human gut bacteria.

DISCUSSION
The human large intestine is a highly competitive ecosystem, in which access to a wide range of carbohydrates confers selective advantage. Here, we outline a model in which a single PUL enables B. uniformis to utilize a range of ␤-glucan congeners (Fig. 1). In this model, LdLam, yBG, or MLG (Fig. 2) is bound by BuSGBP-B on the surface of the bacterium, whereas the SusD homolog BuSGBP-A is passive in this step. Depending on the backbone linkage composition and extent of branching, individual ␤-glucans are cleaved into oligosaccharides by one or both cell surface-anchored endo-glucanases. For each species, the binding or catalytic specificity of the proteins encoded by the endogenous 1,3GUL is indicated on the cell surface (Lam, laminarin; bMLG, barley mixed-linkage ␤-glucan; yBG, yeast ␤-glucan; NB, no binding). The capacity of each species to utilize individual ␤-glucans for growth, which arises as a combination of these individual specificities, as well as periplasmic transport/sensing, is indicated in the Venn diagram.
Comparative genomic and biochemical analysis revealed variations in GH content that, together with complementary HTCS, GH, and SGBP specificity, dictate selective ␤-glucan utilization. Among the Bacteroides species tested, B. uniformis is unique in its ability to utilize the trifecta of ␤-glucans containing contiguous or individual ␤ (1,3) linkages, viz., LdLam, yBG, and bMLG. In contrast, B. thetaiotaomicron can only access LdLam and yBG, while B. fluxus is further restricted to utilizing LdLam only (summarized in Fig. 6). None of the species tested were able to utilize curdlan, perhaps due to its poor solubility in water. Orthologous GHs from syntenic 1,3GUL share similar specificities, yet the GH complement alone is insufficient to confer growth on a particular ␤-glucan. For example, B. thetaiotaomicron produces a GH16 member that is an approximately equally competent laminarinase, yBGase, and MLGase (see Table S2 at https://doi.org/10.14288/1.0388792). However, an inability to capture bMLG at the cell surface, due to a lack of polysaccharide binding by the cognate SGBPs (Fig. 5), and an inability to transport or recognize breakdown products in the periplasm (Fig. 2) precludes growth on this abundant dietary glycan. It is interesting to ponder to what extent losses or gains of function are correlated and arise (a)synchronously through the evolution of these distinct protein families more widely among the Bacteroidetes.
In this regard, B. fluxus is the least versatile of the three Bacteroides species explored here. The inability of B. fluxus to utilize bMLG clearly is the consequence of three factors (Fig. 6): (i) the lack of a polyspecific GH16 ortholog encoded by its 1,3GUL (cf. B. uniformis and B. thetaiotaomicron), (ii) the extremely poor MLGase activity of BfGH158 (Table S2 at the URL mentioned above), and (iii) an inability to transport/sense this cereal polysaccharide in the periplasm (Fig. 2). However, we note that B. fluxus possesses an SGBP-B able to bind both laminarin and bMLG ( Fig. 5 and Table S6 at the URL mentioned above), which is correlated with MLG utilization in the absence of an MLG-binding SGBP-A in B. uniformis (Fig. 6). Further, the inability of B. fluxus to grow on yBG is consistent with weak binding by BfSGBP-A and BfSGBP-B, poor catalytic efficiency of BfGH158 toward yBG, and the lack of a compensatory GH16 ortholog. This is despite an HTCS that is equally highly responsive to yBG and LdLam (Fig. 2), as well as both SGBP types binding both of these 1,3-␤-glucans. Hence, we might anticipate that a gain-of-function mutation to introduce yBG hydrolase activity, for example, through acquisition of a versatile GH16 member or broadening the substrate scope of the extant GH158, would result in growth on yBG. Testing this hypothesis will depend on the future generation of a transformation system for B. fluxus or discovery of a corresponding wild-type strain.
The present study underscores that dietary specificity of related gut commensal strains is gleaned most precisely by systems-based approaches involving genomic, biochemical, and structural biological dissection of PULs (14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27), which otherwise could not have been predicted based on sequencing data alone. Whereas the GH complement of PULs has been shown to drive specificity in the levan/fructan system (14), to our knowledge the present study describes the first case in which the interplay of GHs and SGBPs, underpinned by HTCS specificity, collectively dictate glycan utilization among species. Indeed, in most PULs characterized to date, the specificities of the vanguard GH and SGBPs are concordant (15,17,19,23,25,26,46). On the other hand, the B. fluxus case reveals how a limited endoglycanase in the context of polyspecific SGBPs and an HTCS can restrict nutrient range. Thus, the evolution of both the gene complement and the specificity of individual components within a PUL allows bacteria to access to distinct nutrient niches in the competitive human gut environment. In the context of human nutrition and health, the detailed characterization presented here provides a validated set of molecular markers (Fig. 1) to identify ␤(1,3)-glucan utilization potential among members of the HGM (and other microbiota). This insight may prove especially transformative for our future ability to select strains and dietary formulations in tailoring microbial intervention therapies (59).
Bacterial growth experiments. Bacteroides uniformis ATCC 8492 (NCBI accession no. NZ_AAYH00000000.2, JGI accession no. 641380447), Bacteroides thetaiotaomicron NLAE-zl-H207 (NCBI accession no. NC_004663.1, JGI accession no. 2515154063), and Bacteroides fluxus YIT12057 (NCBI accession no. NZ_AFBN00000000.1, JGI accession no. 651324011) were grown in tryptone yeast extract glucose (TYG) medium at 37°C under anaerobic conditions (Coy anaerobic chamber; 85% N 2 , 10% H 2 , 5% CO 2 ). These cultures were used to inoculate minimal medium containing glucose as the sole carbon source (MM-Glc), followed by incubation at 37°C for 20 h. One-milliliter samples then were centrifuged and the bacterial pellets were gently resuspended in MM containing no carbohydrate (MM-NC). These suspensions were diluted 1:50 in MM-NC before being used to inoculate MM containing 0.5% carbohydrate. Growth experiments were performed in replicates of 6 (laminarin, yeast ␤-glucan, barley ␤-glucan, and curdlan) or 3 (glucose and H 2 O) in 96-well plates. Growth was monitored by measuring the A 600 , and data were processed using Gen5 software (BioTek). Growth was quantified in each assay by first identifying a minimum time point (A min ) at which the A 600 had increased by 10% over a baseline reading taken during the first 500 min of incubation. We then identified the time point at which the A 600 reached its maximum (A max ) immediately after exponential growth. The growth rate for each well was defined by (A max Ϫ A min )/(T max Ϫ T min ), where T max and T min are the corresponding time values for each absorbance. Cultures where the density did not increase by at least 0.1 (A 600 ) were considered to have no growth.
Quantitative reverse transcription-PCR (qRT-PCR). B. uniformis was capable of growth on all carbon sources of interest in this study and, therefore, was cultured directly in 3 ml of MM containing 0.5% (wt/vol) carbohydrate, as described above. B. thetaiotaomicron and B. fluxus were precultured on MM-Glc, pelleted, washed, resuspended twice in MM-NC, and inoculated to an A 600 of ϳ0.3 in 4 ml of MM containing 0.5% (wt/vol) carbohydrate. Bacterial cultures were harvested in triplicate (at mid-log phase [A 600 , ϳ0.6] for B. uniformis and after 5 h of incubation for B. thetaiotaomicron and B. fluxus), placed in RNA protect (Qiagen) for immediate stabilization of RNA, and then stored at -20°C. RNA was extracted and purified with the RNeasy minikit (Qiagen), and RNA purity was assessed by spectrophotometry. One g of RNA was used for reverse transcription and synthesis of the cDNA (SuperScript VILO master mix; Invitrogen). Quantitative PCRs (20 l final volume) using specific primers were performed with a SensiFast SYBR Lo-ROX kit (Bioline) on a 7500 Fast real-time PCR system (Applied Biosystems) (see Table  S9 at https://doi.org/10.14288/1.0388792). Data were normalized to 16S rRNA transcript levels, and changes in expression levels were calculated as fold change compared with levels for cultures of MM containing glucose.
Bioinformatics, gene cloning, and site-directed mutagenesis. Potential 1,3GULs were identified by homology searches of sequences available in the Joint Genome Institute Integrated Microbial Genomes and Metagenomes (JGI-IMG/M) Database (65) and other sources (41). Signal peptides and subcellular localization were predicted by protein sequence analysis (66)(67)(68). The gene fragments corresponding to BACUNI_01484 to BACUNI_01488 (encoding BuGH3, BuGH158, BuGH16, BuSGBP-B, and BuSGBP-A), H207DRAFT_02225 to H207DRAFT_02227 (encoding BtSGBP-A, BtSGBP-B, and BtGH16), HMPREF9446_00612 to HMPREF9446_00614 (encoding BfGH158, BfSGBP-B, and BfSGBP-A), and HM-PREF1991_02176 (encoding PlGH16) were PCR amplified from genomic DNA using the Q5 high-fidelity polymerase (NEB) with primers designed to exclude signal peptides and lipidation cysteines (66,69) (Table S10 at the URL mentioned above). The PCR introduced SalI/XhoI (NEB) restriction sites to the flanks of BACUNI_01484 and _01486 target genes, and the amplified DNA products were ligated into the expression vector pET28a such that the encoded recombinant proteins contain an N-terminal His 6 tag. The rest of the PCR products contained appropriate pMCSG complementary sequences for subsequent ligation-ndependent cloning into pMCSG53 or pMCSG-GST plasmids, providing an N-terminal His 6 tag or an N-terminal His 6 -glutathione S-transferase (GST) tag (70). Successful cloning was confirmed by colony PCR (GoTaq polymerase from Promega) and sequencing (Genewiz). The site-directed mutants (71) BuGH158 N136A, E137A, and E220A were generated using pMCSG53::BuGH158 as a template DNA (Table S11 at the URL mentioned above).
Recombinant protein production and purification. Recombinant proteins were produced in Escherichia coli BL21(DE3) cells cultured in TB broth containing ampicillin (50 g·ml Ϫ1 ) or kanamycin Bacteroides Beta(1,3)-Glucan Utilization Loci ® (50 g·ml Ϫ1 ) at 37°C (200 rpm). Cells were grown to mid-exponential phase (optical density at 600 nm [OD 600 ], ϳ0.4 to 0.6). Overexpression was induced by adding isopropyl ␤-D-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM, and the cultures were further grown at 16°C (200 rpm) for 18 h. The cells were harvested by centrifugation and sonicated, and His 6 -tagged recombinant proteins were purified via immobilized nickel affinity chromatography (His-Trap; GE Healthcare) utilizing a gradient elution of up to 100% elution buffer containing 20 mM sodium phosphate, pH 7.4, 500 mM NaCl, and 500 mM imidazole in a BioLogic fast protein liquid chromatography system (Bio-Rad). The purity of the recombinant proteins was determined by SDS-PAGE ( Fig. S15 and S16 at https://doi.org/10.14288/1 .0388792), and their concentrations were determined from calculated molar extinction coefficients at 280 nm using an Epoch microplate spectrophotometer (BioTek).
Selenomethionine-labeled protein was produced by inhibition of methionine biosynthesis in E. coli BL21(DE3) (72). Briefly, cells were grown in 1 liter of M9 minimal medium supplemented with 100 g ml Ϫ1 ampicillin at 37°C with shaking until the OD 600 reached 0.6. At this point, 100 mg each of Llysine, L-threonine, and L-phenylalanine and 50 mg each of L-leucine, L-isoleucine, L-valine, and L-selenomethionine were added to the medium and shaken for a further 15 min before inducing expression with 0.5 mM IPTG. The culture was transferred to 16°C and incubated for an additional 24 h. Nickel affinity chromatography was conducted as described above using HEPES buffer instead of sodium phosphate. Native and selenomethionine-labeled proteins for crystallography were further purified by size exclusion chromatography through a Superdex 75 resin (GE Healthcare Life Sciences) in an XK 16/100 column (GE Healthcare Life Sciences) run in 10 mM HEPES, pH 7.0, at 0.8 ml min Ϫ1 .
Affinity gel electrophoresis. Affinity polyacrylamide gel electrophoresis was performed for 180 min at 80 V and room temperature on nondenaturing 10% (wt/vol) polyacrylamide gels containing a polysaccharide concentration of 0.1% (wt/vol) (or water for a control), essentially as previously described (15,26,46). Five micrograms of the tested SGBP proteins, along with bovine serum albumin (BSA) used as noninteracting negative-control protein, were loaded on the gels.
ITC. Isothermal titration calorimetry (ITC) was performed using the MicroCal VP-ITC titration calorimeter equilibrated to 25°C, essentially as previously described (15,26,46). The proteins (20 to 100 M) were placed in the sample cell, and the syringe was loaded with 2.5 mg/ml polysaccharide or 0.5 to 2 mM oligosaccharide. Following an initial injection of 2 l, 25 subsequent injections of 10 l were performed with stirring at 280 rpm, and the resulting heat of the reaction was recorded. Integrated heats were fit to a single-site model using Microcal Origin v7.0 to derive n, K a , and ΔH values.
Carbohydrate analytical method. High-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) was performed on Carbopac PA200 guard and analytical columns connected in series on a Dionex ICS-5000 high-performance liquid chromatography system, operated by Chromeleon software, version 7 (Thermo Scientific), essentially as previously described (73). Solvent A was ultrapure water, solvent B was 1 M sodium hydroxide, and solvent C was 1 M sodium acetate (anhydrous bio ultra-grade; Sigma-Aldrich). The injection volume was 10 l, and the gradient was the following: 0 to 5 min, 10% B and 3.5% C; 5 to 12 min, 10% B, linear gradient from 3.5 to 30% C; 12.0 to 12.1 min, 50% B, 50% C; 12.1 to 13.0 min, exponential gradient (curve setting, 9) of B and C back to initial conditions; 13 to 17 min, initial conditions. Enzymatic assays. Polysaccharide hydrolysis was quantified using a BCA reducing-sugar assay (74). Assays were conducted in a final volume of 100 l at the optimum pH and 37°C for 10 min. Reactions were terminated by the addition of an equal volume (100 l) of BCA reagent. Color was developed by heating to 80°C for 20 min before reading A 563 . Glucose (25 to 150 M) was used to generate a standard curve for quantitation. The pH and temperature optima of each enzyme were initially determined using the same enzyme reaction assay to quantify reducing ends over 10 min of incubation with 1.0 mg/ml laminarin in different buffers at 50 mM: sodium citrate (pH 3.0 to 6.5), sodium phosphate (pH 6.5 to 8.0), and glycine (pH 9.0 to 10.5). To determine Michaelis-Menten parameters, different concentrations of polysaccharide solutions were used over the range of 0.025 to 3 mg·ml Ϫ1 with the appropriate concentration of enzyme for 10 min, and the numbers of reducing ends released were quantified as described above.
The release of glucose from oligosaccharides was quantified using the D-glucose HK assay kit from Megazyme (Bray, Ireland), modified for use as a continuous assay exactly as previously described (19).
To measure enzyme activity on chromogenic glycosides, the release of para-nitrophenyl was monitored by following the A 405 in a 1-cm-path-length quartz cuvette with a Cary 60 UV-visible spectrophotometer (Agilent Technologies). Reaction mixtures in 250 l at the optimum pH and 37°C were assayed with nine different substrate concentrations, and the rate was calculated using an extinction coefficient determined according to the buffer used. Endpoint assays were used for pH and temperature optima (same range as that described above) of BuGH3. Reactions were terminated after 10 min by the addition of 100 l of 1 M Na 2 CO 3 to raise the pH, and the A 405 was measured in 96-well plates on an Epoch microplate spectrophotometer (BioTek). An extinction coefficient of 18,100 M Ϫ1 cm Ϫ1 was used for these assays. Continuous assays were used for initial-rate saturation kinetics and reactions, initiated by adding 25 l of enzyme solution to 225 l of the remaining assay mixture in the optimum pH buffer at 37°C.
Crystallization and structure determination. Initial sitting-drop crystal screens were set up in 96-well plates using a Phoenix robot (Art Robbin Instruments) and were stored at room temperature. A hit was obtained in pHClear (Qiagen), condition E12 (0.1 M bicine, pH 9.0, 1.6 M ammonium sulfate), with purified BuGH158 at 22.7 mg ml Ϫ1 and optimized in larger hanging drops in a grid screen by varying pH and ammonium sulfate concentrations in 24-well plates. Crystals were cryoprotected in crystallization solution supplemented with 2 M lithium acetate before flash freezing with liquid nitrogen, and diffraction data were collected at the Advanced Photon Source (APS) beamline 23ID-D. Selenomethionine-labeled crystals were obtained in the same grid screen as the native crystals but required macroseeding with fine needles to obtain crystals of sufficient thickness. Crystals were cryoprotected in crystallization solution supplemented with 4 M lithium chloride before flash freezing with liquid nitrogen. Diffraction data were collected at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 9-2 at the selenium absorption peak, inflection, and high-energy remote wavelengths as determined by a fluorescence scan. Data sets were indexed and integrated with XDS (75), space groups were determined with Pointless (76), and data reduction was performed with Aimless (77). Phasing by multiple anomalous dispersion, density modification, and initial model building was performed in AutoSol (78) in the Phenix suite (79). This selenomethionine-labeled structure was used as the search model for molecular replacement with the native crystal data using Phaser (80). After initial refinement in Phenix.refine (81), iterative rounds of manual model building and refinement were conducted with Coot (82) and Refmac5 (83), respectively, in the CCP4i2 suite (84). The quality of the model was monitored throughout using Molprobity (85). A structure similarity search was conducted using the Dali server (86).
1 H-NMR determination of catalytic mechanism. BuGH158 in 50 mM sodium phosphate, pH 7.0, and laminaribiose-␤-CNP were independently lyophilized and resuspended in 99.9% D 2 O. After recording an initial 1 H-NMR spectrum of the substrate (Bruker Avance 400-MHz spectrometer), the enzyme was added to obtain final concentrations of 20 M BuGH158 and 10 mM laminaribiose-␤-CNP. Spectra were recorded at appropriate time intervals thereafter to observe the first-formed product anomer and subsequent mutarotation.
Data availability. The crystal structure data sets generated (coordinate files and structure factors) have been deposited in the Protein Data Bank (PDB) under the accession code 6PAL. We declare that the data supporting the findings of this study are available within the article and the supplemental material, the latter of which is available at https://doi.org/10.14288/1.0388792.

ACKNOWLEDGMENTS
Work at the University of British Columbia was generously supported by operating grants from the Canadian Institutes for Health Research (MOP-137134 and MOP-142472) and infrastructure support from the Canadian Foundation for Innovation (project number 30663) and the British Columbia Knowledge Development Fund. K.T. was partially supported by a 4-year doctoral fellowship from the University of British Columbia. E.C.M. acknowledges funding from the U.S. National Institutes of Health (R01 DK118024).
We thank the General Medical Sciences and Cancer Institutes Structural Biology Facility at the Advanced Photon Source (GM/CA@APS) for access to beamline 23ID-D. GM/CA@APS has been funded in whole or in part with Federal funds from the National Cancer Institute (ACB-12002) and the National Institute of General Medical Sciences (AGM-12006). This research used resources of the Advanced Photon Source, a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under contract no. DE-AC02-06CH11357.
We thank the Stanford Synchrotron Radiation Lightsource (SSRL) at the SLAC National Accelerator Laboratory (Menlo Park, CA, USA) for access to beamline 9-2, the use of which is supported by the U.S. DOE, Office of Science, Office of Basic Energy Sciences, under contract no. DE-AC02-76SF00515. The SSRL Structural Molecular Biology Program is supported by the DOE Office of Biological and Environmental Research and by the National Institutes of Health, National Institute of General Medical Sciences (including P41GM103393). The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of NIGMS or NIH.
We thank Stephanie Forget (Brumer group, UBC) for assistance with preliminary NMR analysis of GH158 reaction stereochemistry. G.D. performed bacterial growth analyses, qPCR transcript analyses, comparative bioinformatics, GH and SGBP cloning, recombinant protein production, and biochem-ical/biophysical/kinetic analyses. K.T. performed recombinant protein production and purification, all crystallography and tertiary structural analyses, and phylogenetic analysis of GH16, PFAM 13004 domain production and analysis and assisted with the characterization of B. uniformis enzymes (pH and temperature rate analyses, BuGH3 Glc-pNP kinetics). A.C. assisted G.D. with GH cloning, production, and characterization. N.J. synthesized laminaribiose-CNP and performed NMR analysis together with K.T. N.A.P. assisted G.D. with bacterial growth analyses in Michigan. G.P. performed human metagenome analysis. A.H.V. performed phylogenetic analysis of GH158. F.V.P. supervised crystallography and data analysis. H.B. and E.C.M. conceived the study and supervised research. G.D. drafted the manuscript, which was revised by K.T. and H.B.
We have no competing interests to declare.