Introduction

Complex carbohydrate utilization by the gut microbiota is linked to host metabolic health outcomes1 and represents a major nutrient source for herbivores2. The carbohydrate composition in the diet has a profound influence on shaping the microbial community composition, affecting its fitness and interactions with the host3.

Many commensal gut bacteria associated with dietary fiber breakdown, such as those from the Bacteroidota phylum, are present in both humans and herbivores, highlighting similar metabolic capabilities, to some extent, of their gut microbiota to generate energy and nutrients for the host4. Bacteroidota species are considered primary degraders of dietary fibers and are dominant in the mammalian gut5. These bacteria can metabolize a wide range of complex dietary carbohydrates within the dense and diverse gut ecosystem, which may explain their evolutionary success as symbionts6. To cope with dietary carbohydrates, they employ a broad arsenal of Carbohydrate-Active enZymes (CAZymes) arranged in gene clusters known as Polysaccharide Utilization Loci (PULs)7,8,9. PULs from Bacteroidota species found in the human gut have been elucidated, including those for type-II rhamnogalacturonan10, yeast mannans11, xyloglucans12, xylans13, arabinogalactans14, mixed-linkage β-glucans (MLGUL)15 and β(1,3)-glucans16. However, similar PULs from other ecological niches, such as from the gut of herbivores, remain underexplored, leaving a gap of knowledge regarding the impact of diet, host selection, and nutrient abundance on the functionality of such enzymatic systems and putative evolutionary trajectories for functional adaptation.

Using the gut microbiota of the capybara (Hydrochoerus hydrochaeris) as a model—a semi-aquatic herbivore considered the largest living rodent in the world and the last remnant of a long line of grass-eating rodents that evolved in South America over millions of years17,18—we investigated the composition of PULs identified in novel metagenome-assembled genomes (MAGs) belonging to the Bacteroidota phylum that are associated with carbohydrate depolymerization19. Interestingly, one of these new MAGs, denoted here as Alloprevotella sp. MAG60, showed a notable glycogenomic potential, including a PUL akin to the mixed-linkage β-glucan utilization loci (MLGUL) from the human gut Bacteroides ovatus15. Both Alloprevotella sp. and B. ovatus PULs encode a GH16_3 enzyme, two GH3 members, a cell-surface glycan-binding protein (SGBP), and the sugar transporters SusC and SusD15. Despite being similar in terms of gene composition at the family or subfamily levels, intuitively suggesting being functionally equivalent, we demonstrate that the PUL from the herbivore Bacteroidota exhibits an expansion in substrate specificity compared to the one from the human gut Bacteroidota, unveiling a gain of function towards substituted β-1,3-glucans, a polysaccharide typically found in the diversified diet ingested by herbivores like capybara.

Employing techniques such as mass spectrometry, carbohydrate enzymology, synergism studies, site-directed mutagenesis, synchrotron X-ray crystallography, and molecular dynamics simulations on the PUL-associated proteins, we uncovered the biochemical and structural basis for this gain of function, which resides (i) in the affinity of the SGBP for β(1,3)-glucans, an affinity not observed in the ortholog from human gut Bacteroidota, (ii) the broad specificity of the core endo-β-glucanase toward different types of β-1,3-glucans, and (iii) the presence of a second functional β-glucosidase specialized for β(1,6)-glucosyl linkages. Remarkably, the ortholog of this second β-glucosidase in the human gut Bacteroidota was shown to be catalytically inefficient and not essential for the MLGUL, a striking biochemical difference between these PULs.

Ultimately, the findings of this work expand the current model for non-cellulosic β-glucans utilization by commensal gut bacteria in mammals, highlighting an additional layer of functional and evolutionary complexity within the gut microbiota to cope with the diversity of dietary glycans.

Results

β-glucan enzymatic systems from herbivore and human gut Bacteroidota

The multi-omics study of the gut microbiota of the semi-aquatic herbivore capybara, also known as master of the grasses, led to the discovery of several novel lignocellulolytic MAGs19,20 from the Bacteroidota phylum including the Alloprevotella sp. MAG60 that belongs to a monophyletic clade within the order Bacteroidales along with other Alloprevotella species (Fig. 1a and Supplementary Table 1). This MAG harbors a broad repertoire of CAZymes, including a PUL encoding SusC/SusD-like proteins, a cell-surface glycan-binding protein (hereafter named CapSGBP-B), a predicted SusR-like regulator, a predicted endo-β-glucanase from the GH16_3 subfamily (CapGH16_3) and two putative β-glucosidases from the GH3 family (CapGH3a and CapGH3b) (Fig. 1b). A similar PUL composition is found in the human gut B. ovatus15, differing only in the presence of a hybrid two-component sensor/transcriptional regulator (HTCS). While PULs with similar gene composition would be considered functionally equivalent, the enzymes encoded by the Alloprevotella sp. MAG60 PUL exhibit low to moderate sequence similarity to those in the MLGUL from human B. ovatus (SusC: 35%, GH3a: 27%, GH16_3: 32% and GH3b: 55%), indicating that they could have distinguishing functional properties (Fig. 1b).

Fig. 1: Phylogeny of Alloprevotella sp. MAG60 and the gene composition of its functionally augmented PUL for plant non-cellulosic β-glucans.
figure 1

a Maximum likelihood phylogenetic tree indicating the placement of Alloprevotella sp. MAG60 alongside 181 related representative genomes recovered from the Bacteroidaceae family. The colored and labeled groups represent the taxonomy at the genus level. b Schematic representation of the PUL from Alloprevotella sp. MAG60, including a genomic comparison with other PULs with similar gene organization, derived from the literature or predicted in silico. Links between genomes represent sequence identity at the protein level in grayscale. SusC SusC-like TonB-dependent transporter, SusD SusD-like outer membrane-binding protein, SGBP cell-surface glycan-binding protein, GH16 β(1,3)(1,4)-glucanase from Glucosyl Hydrolase family 16, GH3 β-glucosidase from Glucosyl Hydrolase family 3, SusR SusR-like regulator, HTCS hybrid two-component system (transcriptional regulator).

Full size image

Supporting this, the characterization of the noncatalytic SGBP (CapSGBP-B), which is considered instrumental in recruiting polysaccharides to the cell surface, facilitating the transport of cleaved products across the outer membrane in conjunction with TonB-dependent transporters (TBDTs)16, demonstrated the ability to bind β(1,3)-glucans beyond MLGs (Supplementary Fig. 1). The corresponding protein in the B. ovatus MLGUL (BoSGBPMLG-B) is not able to interact with β(1,3)-glucans15, indicative of functional differentiation. The SGBP from the human gut B. uniformis (BuSGBP-B, 41% SEQID with CapSGBP-B) also showed the capacity to bind both MLG and β(1,3)-glucans21. This appears to compensate for the inability of BuSGBP-A (SusD-like) to bind these substrates due to loop insertions in the carbohydrate accommodation region16. However, it is worth mentioning that the B. uniformis PUL is in a different context, with a GH158 glucanase dedicated to the breakdown of β(1,3)-glucans.

Despite the SusD-like protein from the herbivore gut Bacteroidota PUL was not experimentally assessed, the structural comparison with its closest ortholog from B. thetaiotaomicron (BtSGBP-A, 61% SEQID)16 revealed conservation of the key molecular determinants for the recognition of β(1,3)-glucans (Supplementary Fig. 2). However, some residues located at the extremities of the glycan-binding interface are not conserved between CapSusD-like protein and BtSGBP-A, indicating a potential functional differentiation that may be related to the capacity to bind other β-glucans.

These analyses indicate that in the herbivore gut microbiota, Bacteroidota might have evolved a distinct molecular strategy to utilize β(1,3)-glucans as an energy and carbon source. These carbohydrates, despite their low abundance, are present in a wide variety of plant cell walls22,23 and are also associated with fungi24,25, and thus can be part of the diet of capybara and other herbivores26. Therefore, to shed light on the enzymatic mechanisms involved in the breakdown of β(1,3)-glucans in the herbivore gut Bacteroidota, each enzymatic component of this PUL was subjected to biochemical and structural characterization. This investigation revealed a distinct enzymatic cascade for the utilization of plant non-cellulosic β-glucans within the gut microbiota of mammals, including the mechanistic basis for functional divergence within CAZy families.

The herbivore gut Bacteroidota PUL encodes a broad-specificity and highly active endo-β(1,3)/β(1,4)-glucanase

The only predicted glucanase in this PUL from the herbivore gut Bacteroidota is a GH16 member belonging to the subfamily 3 (CapGH16_3). Therefore, it is expected that CapGH16_3 would be active on both MLGs and β(1,3)-glucans, consistent with the ability of CapSGBP-B to bind both polysaccharides.

Activity screening against a set of polysaccharides (Supplementary Table 2) corroborated the hypothesis, indicating that CapGH16_3 can efficiently cleave oat β-glucan, followed by laminarin from Laminaria digitata (LdLam) and laminarin from Eisenia bicyclis (EbLam) (Fig. 2a, Supplementary Fig. 3, Table 1). Compared to its ortholog in the human gut microbiota from B. ovatus (BoGH16MLG, 32% SEQID)15, CapGH16_3 exhibited an one-order-of-magnitude higher catalytic turnover rate on oat β-glucan and a catalytic turnover number of 197 s−1 on LdLam (Table 1). Substrate saturation was also attained with a highly substituted yeast β(1,3)-glucan and EbLam; however, with lower affinity and turnover rates due to the presence of many β(1,6)-linked substitutions. In contrast, the ortholog BoGH16MLG was minimally active on β(1,3)-glucans, indicating that these enzymes are functionally distinct (Table 1).

Fig. 2: The tryptophan quad is a conserved motif in GH16 members exhibiting broad substrate specificity.
figure 2

a Substrate saturation curves of CapGH16_3 over oat β-glucan, LdLam, and EbLam, highlighting the broad specificity of CapGH16_3. Results are expressed as mean ± standard deviations from three independent experiments (n = 3). The values shown represent the kcat values of CapGH16_3 for each substrate. b Cleavage pattern of β-glucan, LdLam, and EbLam after exhaustive hydrolysis by CapGH16_3, unveiling a typical dissociative endo-acting mode of action. *For clarity purposes, only linear glucooligosaccharides are represented, despite β(1,6)-substituted oligosaccharide is likely to be generated by CapGH16_3. Time-course analysis of cleavage patterns is presented in Supplementary Fig. 7. c Surface representation of the CapGH16_3 catalytic (green) and PF13004 (pink) domains, including a docked glucooligosaccharide spanning from the −3 to +2 subsites. d The conserved tryptophan quad motif among CapGH16_3 (green) and orthologs with similar broad substrate specificity, including SCLam (PDBID 6XQH28, light blue), LamR (PDBID 3ILN29, orange), ZgLamCGH16 (PDBID 4CRQ30, light pink), and LamAQ (PDBID 6JH527, yellow). G4G3G and cellobiose bound to the positive and negative subsites, respectively, were obtained from the structural superposition with SCLam (PDBIDs 6XQH, and 6XQF). e Comparative analysis of the substrate binding subsites between CapGH16_3 (green) and BoGH16MLG (PDBID 5NBP15, pink), highlighting the several changes in the active site pocket that results in the distinct substrate preference between these enzymes, including potential clashes in the positive subsites as represented in black. Cellobiose in the positive-subsite region was extracted from the structural superposition with SCLam (PDBID 6XQH), and G4G4G3G in the negative subsite region was extracted from the structural superposition with BoGH16MLG (PBDID 5NBP).

Full size image
Table 1 Kinetic parameters of CapGH16_3 and their respective orthologs
Full size table

To shed light on its mode of action, the cleavage products were analyzed by mass spectrometry using LdLam, EbLam, oat β-glucan and distinct glucooligosaccharides (β(1,3)-linked, β(1,4)-linked and mixed-β(1,3)/(1,4)-linked) as substrates (Supplementary Figs. 4–7). CapGH16_3 showed no activity on cellooligosaccharides and cellulose, indicating it strictly requires β(1,3)-linked glucosyl residues between −1 and −2 subsites to be active. The hydrolysis of oat β-glucan indicates the major products G4G3G and G4G4G3G (Supplementary Fig. 7), in accordance with the action mode of other characterized endo-β(1,3)/β(1,4)-glucanases from the GH16 family that specifically hydrolyze β(1,4)-glucosyl linkages immediately preceded by β(1,3)-linked glucosyl residue, such as BoGH16MLG15 and BuGH1621. Furthermore, CapGH16_3 can only cleave laminarioligosaccharides with minimal chain size of four glucosyl units and the digestion products of LdLam and EbLam were mainly a disaccharide followed by a trisaccharide and glucose. Time-course analysis of substrate depolymerization shows the initial production of large oligosaccharides that are progressively cleaved into short ones, compatible with a typical endo-dissociative mode of action (Fig. 2b).

CapGH16_3, as most GH16 members, behave as a monomer in solution, discarding any role of oligomerization in the stability and activity of this enzyme (Supplementary Fig. 8). Despite the full-length protein (including catalytic and PF13004 domains) was used for crystallization trials, only the catalytic domain was observed in the crystallographic structure determined at 1.90 Å resolution (Supplementary Table 3). The catalytic domain shows a canonical family fold, featuring the conserved EXDXXE consensus sequence in which Glu258 is the nucleophile, Asp260 is an electrostatic “helper” and Glu263 is the acid/base, as inferred by structural comparisons with other GH16 members15,27 (Supplementary Fig. 9). A calcium ion, involved in the structural stability of some GH16 members27, was also identified in the CapGH16_3 structure. This metal ion is localized in the opposite face of the catalytic cleft, being caged by a pentagonal-bipyramidal coordination sphere involving the carbonyl oxygen from the residues Glu144, Gly197 and Asp373, Asp373_OD2 and three solvent molecules (Supplementary Fig. 9).

The structurally characterized GH16 members most similar to CapGH16_3 are also endo-β(1,3)/β(1,4)-glucanases with similar substrate preference, being active on both MLG and LdLam such as SCLam from a soil metagenome (PDBID 6XQF, 42% SEQID)28, LamR from Rhodothermus marinus (PDBID 3ILN, 41% SEQID)29, ZgLamCGH16 from Zobellia galactanivorans (PDBID 4CRQ, 40% SEQID)30 and LamAQ from Aquimarina sp. (PDBID 6JH5, 39% SEQID)27. Despite these four enzymes were defined as laminarinases, they have a preference for MLG over LdLam being able to cleave both β(1,3)- and β(1,4)-glucosyl linkages. Structural superposition of CapGH16_3 on these enzymes showed that the subsites −2, −1 and +1 are fully conserved featuring a “tryptophan quad” (Trp253: subsite −2, Trp243: subsite −1, Trp239: subsites −1, and Trp355: subsite +1) (Fig. 2c, d). These four tryptophan residues mostly define the productive conformation of the substrate in the active site, being a conserved feature among these enzymes with similar substrate preference. Trp253, Trp243, and Trp355 establish stacking interactions with −2, −1, and +1 glucosyl residues, whereas Trp239 is perpendicular between the −1 and +1 glucosyl residues (Fig. 2d). Other subsites are not conserved between these four endo-β(1,3)/β(1,4)-glucanases (Supplementary Fig. 9), reiterating the importance of the subsites −2, −1 and +1 for substrate preference in these enzymes that feature the “tryptophan quad”.

The comparison of CapGH16_3 with BoGH16MLG in complex with G4G4G3G (PDBID 5NBP)15 showed that the −2 Trp138 residue (corresponding to Trp253 in CapGH16_3) is 2.7 Å shifted, resulting in a distinct accommodation of the –2 glucosyl residue (Fig. 2e). At the positive-subsite region, CapGH16_3 contains only one aromatic residue, (Trp355), whereas BoGH16MLG has two aromatic residues, Phe243 and Tyr181, which establish aliphatic contacts with the +1 and +2 glucosyl residues15 (Fig. 2e). These residues result in a steric hindrance for cellobiose accommodation in BoGH16MLG, imposing cellobiose relocation to fit into the positive subsites. Despite the likely distinct binding mode of the substrate in the positive-subsite region of these enzymes, no changes in the backbone linkage specificity were observed (Fig. 2e). These analyses point that such structural modifications leading to a distinct saccharide accommodation in both negative and positive-subsite regions likely play a role in substrate preference among GH16_3 family members.

These comparisons highlight that endo-β-glucanases belonging to the subfamily GH16_3 can exploit distinct molecular strategies for β-glucan selectivity and that the “tryptophan quad” is a conserved feature in CapGH16_3 and orthologs with similar activity profile such as SCLam28, LamR29, ZgLamCGH1630 and LamAQ27.

The herbivore gut Bacteroidota PUL is equipped with an exo-β-glucohydrolase active on both laminarins and β(1,3)- and β(1,4)-linked glucooligosaccharides

Considering that the core endo-β-glucanase from capybara Alloprevotella sp. MAG60 PUL is able to efficiently cleave both MLGs and β(1,3)-glucans, we further interrogated whether the GH3 enzyme (CapGH3b) with the highest sequence similarity to the functional one found in the human gut Bacteroidota MLGUL (BoGH3MLG, 55% SEQID)15 could then break down the oligosaccharides derived from both MLGs and β(1,3)-glucans. Similar to BoGH3MLG, CapGH3b can cleave both β(1,3)-glucooligosaccharides and β(1,4)-glucooligosaccharides, with a preference for longer glucooligosaccharides, specifically for those containing β(1,3) linkages such as laminaritetraose and laminarihexaose (Table 2, Supplementary Fig. 10). In addition, CapGH3b is also highly active over LdLam (Table 2) and the strict release of glucose from polysaccharides and oligosaccharides (Supplementary Figs. 11–14) indicates an exo mode of action, defining this enzyme as an exo-β-glucohydrolase.

Table 2 Kinetic parameters of CapGH3b and their respective orthologs
Full size table

In order to elucidate the molecular basis of its activity on laminarin and long glucooligosaccharides, the crystal structure of CapGH3b was determined at 1.86 Å resolution (Supplementary Table 3, Supplementary Figs. 15 and 16). Structural comparisons between CapGH3b and the AlphaFold model31 of BoGH3MLG (entry A7LY29) indicate that the capacity to recognize long oligosaccharides and laminarin might be associated with an extension in the positive-subsite region (Fig. 3a). Near to the residues Trp311 and Trp455 that are oriented to make stacking interactions with the +1 glucosyl moiety (Fig. 3a), a conserved tyrosine residue (Tyr346) along with other residues including Asp56, Met61 and Lys65 shapes a cleft-like topology. It provides extra room for the accommodation of larger substrates, which supports the fact that CapGH3b is highly active on polymeric laminarin and shows a preference for longer β(1,4)- and β(1,3)-oligosaccharides (Fig. 3b-c, and Table 2). The extension in the positive-subsite region is also found in BoGH3MLG with the conservation of the residue Tyr (Tyr345 in BoGH3MGL), corroborating its similar functional profile in comparison with CapGH3b (Supplementary Fig. 17). This finding supports that an extended positive-subsite region is a feature found in some GH3 members with preferential activity on longer β(1,4)- and β(1,3)-oligosaccharides.

Fig. 3: The extended positive-subsite region of CapGH3b enables the efficient hydrolysis of longer oligosaccharides.
figure 3

a Surface representation of the homodimeric configuration of CapGH3b, highlighting the extended positive-subsite region. The laminarihexaose was positioned in the CapGH3b active site by molecular docking calculations, showing a structural topology compatible with the higher catalytic efficiency over longer glucooligosaccharides. b Substrate saturation curves obtained with cellooligosaccharides including cellobiose (G4G, green), cellotetraose (G4G4G4G, brown), and cellohexaose (G4G4G4G4G4G, orange), including the kcat values for each substrate. c Substrate saturation curves obtained with laminarioligosaccharides including laminaribiose (G3G, gray), laminaritetraose (G3G3G3G, blue), and laminarihexaose (G3G3G3G3G3G, green), highlighting the kcat values for each substrate. Results are expressed as mean ± standard deviation from three independent experiments (n = 3). Kinetics parameters are shown in Table 2. d Stacking interaction of the Trp455 residue from the same subunit of CapGH3b with the +1 glucoside. e Instead of the respective interaction, in a GH3 recovered from metagenome (JMB19063, PDBID 3U4A33), a Phe598 residue from the subunit B is responsible for the interaction with the +1 glucoside. The β(1,4)-saccharides represented in the active site were obtained from the structural superposition with the GH3 β-glucosidase JMB19063 (PDBID 3U4A33).

Full size image

In addition, it was observed that CapGH3b behaves as a dimer in solution (Supplementary Fig. 18), in a configuration that resembles to that observed for the enzymes BoGH3B from the xyloglucan PUL of B. ovatus (PDBID 5JP0, 60% SEQID)32 and a GH3 member recovered from metagenome data (JMB19063, PDBID 3U48 and 3U4A, 32% SEQID)33 (Supplementary Fig. 18). Despite the conservation of the quaternary arrangement, in the metagenome-recovered enzyme there is a swapping of a phenylalanine residue (Phe598) between the subunits that are instrumental for substrate binding at the +1 subsite (Fig. 3d, e) and is not conserved in CapGH3b and BoGH3B (Fig. 3d, e). A similar role of the residue Phe598 at the +1 subsite is performed by a Trp residue from a loop of the same subunit in both CapGH3b (Trp455) (Fig. 3d) and BoGH3B (Trp458). These analyses between closest orthologs highlight a distinct molecular strategy for substrate binding emerging from a conserved dimeric configuration, expanding the understanding of the mechanisms underpinning the activity within the large and ubiquitous GH3 family.

A second GH3 member enables the utilization of substituted β(1,3)-glucans

Although the PUL found in the capybara gut microbiome contains a broad-specificity endo-glucanase and an exo-glucohydrolase acting on the non-reducing ends of both β(1,3)- and β(1,4)-linked glucooligosaccharides, these enzymes are unable to cope with the β(1,6)-glucosyl substitutions found in most β(1,3)-glucans. In the MLGUL from the human gut Bacteroidota, a second GH3 member is present (BACOVA_02738); however, it was considered not integral to MLGUL due to (i) its feeble activity on glucooligosaccharides and Glc-β-PNP, (ii) it is not significantly upregulated in the presence of MLG and (iii) the lack of conservation of this gene in other similar PULs15.

In the herbivore gut Bacteroidota PUL, a second GH3 member is also present, sharing only 27% of sequence identity with both CapGH3b and BACOVA_02738. On the contrary to that found in the human PUL, CapGH3a showed to be catalytically proficient on β(1,3)- and β(1,4)-linked disaccharides including mixed-linkage ones (Table 3, Supplementary Figs. 19–23). However, distinct to CapGH3b, CapGH3a showed a preference for β(1,6)-linked disaccharides (Table 3 and Supplementary Fig. 19) and is not active or present very low residual activity over polymeric substrates (LdLam, EbLam, and oat β-glucan). Furthermore, CapGH3b is at least two-order-of-magnitude more active on laminaribiose and 9-fold more efficient on cellobiose in relation to CapGH3a. This sharp difference in the substrate preference profiles highlights the complementary roles of these enzymes in processing all types of oligosaccharides that could be generated by the core endo-glucanase from MLGs and substituted β(1,3)-glucans.

Table 3 Kinetic parameters of CapGH3a and their respective orthologs
Full size table

Therefore, to test whether herbivore gut Bacteroidota PUL enzymes have complementary roles in depolymerizing plant non-cellulosic β-glucans, synergy assays were conducted using LdLam (low-density β(1,6)-substituted laminarin), EbLam (high-density β(1,6)-substituted laminarin) and oat β-glucan (mixed-linkage β-glucan) as substrates (Fig. 4a–c). Time-course analysis showed a distinct synergism degree between the enzymes depending on the degree of substitution in laminarin. In accordance with biochemical data, the combination of CapGH16_3 (core endo-glucanase) with CapGH3b (not active on β(1,6)-glucosides) was not effective in the depolymerization of EbLam, whereas CapGH16_3 and CapGH3a were able to efficiently break down this high-density β(1,6)-substituted laminarin (Fig. 4a). A further improvement in the glucose release was observed when the three enzymes were combined; however, the importance of CapGH3a is notable over high-density β(1,6)-substituted laminarin. Even over low-density β(1,6)-substituted laminarin, the activity of CapGH3a was instrumental for the efficient substrate depolymerization (Fig. 4b), highlighting the key role played by this second GH3 member in the herbivore gut Bacteroidota PUL (Supplementary Table 4). The herbivore Bacteroidota PUL enzymes also synergistically depolymerize oat β-glucan. The importance of CapGH3b for glucose release is evident, as expected, due to its similarity to the human gut Bacteroidota MLGUL (Fig. 4c), contrasting with the role of CapGH3a in the breakdown of substituted β(1,3)-glucans. It is noteworthy that the glucose release from oat β-glucan is limited compared to laminarin. This can be attributed to the lower catalytic efficiency of both β-glucosidases towards β(1,4) linkages, in contrast to β(1,3) linkages (Tables 2 and 3).

Fig. 4: Herbivore gut Bacteroidota PUL enzymes depolymerize linear and substituted β(1,3)-glucans, and mixed-linkage β-glucans.
figure 4

Synergy assays combining CapGH16_3, CapGH3b, and CapGH3a in the hydrolysis of a high-density β(1,6)-substituted laminarin EbLam, b low-density β(1,6)-substituted laminarin LdLam and c Oat β-glucan. The colored dashed lines represent the theoretical glucose concentration expected for each enzyme combination, calculated by summing the glucose released by the individual enzymes. This highlights the degree of synergism between the tested enzymes (see also Supplementary Table 4). The “+” and “−” symbols represent the presence and absence of enzymes, respectively. Results are expressed as mean ± standard deviations from three independent experiments (n = 3).

Full size image

Considering the synergism of herbivore gut Bacteroidota PUL enzymes in depolymerizing diverse types of plant non-cellulosic β-glucans and the lack of a culturable strain harboring this PUL, we explored their utilization in promoting the growth of Escherichia coli as a bacterial model, which naturally is not able to grow on β(1,3)-glucans. E. coli K12 strain was cultivated in M9 minimal media containing LdLam or EbLam as the carbon source and growth was observed only when the herbivore gut Bacteroidota PUL enzymes were added (Supplementary Fig. 24). The ability of these enzymatic system to trigger E. coli growth in the presence of complex carbohydrates was also compared to E. coli growth on M9 supplemented with glucose (Supplementary Fig. 24), indicating a similar growth profile. These results are consonant with our biochemical observations, showing that these three enzymes can release glucose from complex β-glucans, allowing bacterial growth.

Structural basis of regioselectivity in GH3 β-glucosidases

To get further insights into β(1,6)-glucosyl linkage recognition preference among GH3 enzymes, a gap of knowledge within this family, the crystal structure of CapGH3a was determined at 2.6 Å resolution (Supplementary Table 3 and Supplementary Fig. 25). In comparison to the closest structurally characterized β-glucosidases (BaBgl3 from Bifidobacterium adolescentis, 38% SEQID, PDBID 5WAB34; BglB from Acetivibrio thermocellus, 37% SEQID, PDBID 7MS235; and TnBgl3B from Thermotoga neapolitana, 33% SEQID, PDBID 2X4036), CapGH3a has an aromatic platform at the +1 subsite defined by the residue Trp88 emerging from the loop Val83-Asn95 (Fig. 5a), which is not conserved in the other cited GH3 members (Supplementary Fig. 26). Among these orthologous β-glucosidases, only BglB was enzymatically tested with gentiobiose, showing a residual activity37. This indicates that the Trp88 configuration in the +1 subsite of CapGH3a is a trait associated with the preference for β(1,6)-glucosyl linkages. Corroborating this, the structural models obtained by AlphaFold31 revealed structural conservation of a Trp residue corresponding to CapGH3a Trp88 in the biochemically characterized orthologs with preference for β(1,6)-glucosyl linkages38 such as those from B. ovatus (BACOVA_00946, 60% SEQID, Trp266) and B. thetaiotaomicron (Bt3314, 59% SEQID, Trp280) (Supplementary Fig. 26).

Fig. 5: Regioselectivity in the GH3 family is associated with the product release area.
figure 5

a Surface representation of CapGH3a highlighting the productive binding of gentiobiose, stabilized by interactions of the +1 glucoside with the aromatic residue Trp88 (Supplementary Fig. 27). The depicted structure is a representative frame of the molecular dynamics simulation. b–d Substrate saturation curves of CapGH3a WT and mutant W88A with gentiobiose, laminaribiose, and cellobiose, respectively, including the kcat values for each substrate. Results are expressed as mean ± s.d. from three independent experiments (n = 3). Kinetics parameters are shown in Table 3. e Productive binding of laminaribiose into CapGH3b. f Unproductive binding of gentiobiose into CapGH3b. The structures are representative frames of molecular dynamics simulations. The zones shadowed in blue highlight the hydrophobic interactions at the +1 subsite by the two face-to-face tryptophan residues. g Root-mean-square deviations (RMSD) of CapGH3b molecular dynamics (MD) trajectories with gentiobiose and laminaribiose.

Full size image

To get further insights into the structural determinants governing the regioselectivity to β(1,6)-glucosyl linkages, molecular docking followed by molecular dynamics simulations were carried out with CapGH3a and CapGH3b, both in complex with gentiobiose, and also CapGH3b complexed with laminaribiose. Trajectory analysis of CapGH3a revealed that the +1 glucosyl moiety is frequently stabilized by the hydrophobic interactions with Trp88 (Fig. 5a, Supplementary Fig. 27), highlighting the pivotal role of this aromatic platform at the +1 subsite for β-glucosyl residue recognition. Supporting this, the kinetics characterization of the W88A mutant showed a nearly total inactivation of the enzyme (Table 3, Fig. 5b–d, and Supplementary Fig. 28).

On the other hand, in the CapGH3b simulations, while the laminaribiose visits a productive configuration during the trajectory, the same does not occur for gentiobiose (Fig. 5e–g). In the referred system, the positioning of the glycosidic bond towards the opposite direction of the catalytic acid/base is favored, which impairs the proton transference in the first step of the catalytic reaction in retaining glycoside hydrolases. This unproductive configuration is attributed to the stereochemical restrictions imposed by the two face-to-face tryptophan residues (Trp311 and Trp456) that stack the +1 glucosyl moiety, such as observed in the crystallographic structure of the GH3 β-glucosidase from Hordeum vulgare (HvExoI) complexed with thio-gentiobioside (PDBID 3WLP)39 (Supplementary Fig. 29). Such steric barrier is not observed in the GH3 enzymes active on β(1,6)-glucosides (Supplementary Fig. 29), which contains only one aromatic platform (corresponding to Trp88 in CapGH3a), providing the necessary degrees of freedom for productive accommodation of gentiobiose.

Taken together, these analyses provide compelling evidence that the +1 subsite geometry involving two face-to-face tryptophan residues is restrictive to the proper orientation of β(1,6)-glucosyl linkages for catalysis. Additionally, the opened +1 subsite observed in GH3 members active on β(1,6)-glucosides featuring a single aromatic platform, enables the productive accommodation of gentiobiose, establishing a structural fingerprint for regioselectivity in the GH3 family.

Discussion

Dietary glycans are major nutrients for gut microbiota and studies focused on the human gut microbiota have demonstrated that some bacteria exploit diverse multi-component systems known as PULs to cope with these complex carbon sources6,7,8,9,10,11,12,13,14,15,16. However, equivalent PULs from the microbiota of other mammals or ecosystems remain uncharacterized, leaving a knowledge gap about functional conservation and/or adaptations toward the sheer complexity of dietary glycans, which may vary expressively across ecological niches.

Here we demonstrate that a PUL identified in the herbivore gut microbiota with gene composition identical to the human gut Bacteroidota MLGUL exhibit a gain of function towards polysaccharides typically found in the diet of most herbivores, the substituted β(1,3)-glucans. This finding highlights that PULs even with similar gene composition conserved at subfamily or family levels might exhibit distinct specificities and molecular mechanisms to cope with complex plant glycans, which is likely driven by the environment they belong to, and the host feeding habit.

According to the biochemical data, distinct non-cellulosic β-glucans including mixed-linkage β-glucans, linear β(1,3)-glucans and β(1,6)-substituted β(1,3)-glucans are first recognized by a surface glycan-binding protein (CapSGBP-B) and possibly by a SusD-like protein (experimentally not evaluated), which are then cleaved into oligosaccharides by a cell-surface anchored endo-β(1,3)-glucanase with broad specificity and high catalytic rates (CapGH16_3). Further oligosaccharides produced by CapGH16_3 are imported by the associated TBDT SusC-like proteins into the periplasm, where two GH3 enzymes with different regioselectivities, CapGH3a and CapGH3b, act synergistically to complete the hydrolysis (Fig. 6). CapGH3b is most active on β(1,3) followed by β(1,4)-glucosyl linkages, while CapGH3a prefers β(1,6)-glucosyl substitutions. The latter is a key activity that enables the breakdown of substituted β(1,3)-glucans. In the human gut bacterium B. ovatus PUL, the corresponding gene is considered not integral for the PUL function, a notable difference between the PULs architecture found in the herbivore and human gut bacteria.

Fig. 6: Model of plant non-cellulosic β-glucans depolymerization by the herbivore gut Bacteroidota PUL.
figure 6

General non-cellulosic β-glucans structure comprises β(1–4)-linked cellotetraose and cellotriose units connected by β(1–3) linkages, while the structure of brown algae laminarin consists of β(1,3)-linked glucosyl units with single β(1,6)-linked glucosyl substitutions. In the herbivore gut Bacteroidota, the cell-surface localized endo-glucanase CapGH16_3 cleaves polymeric non-cellulosic β-glucans into oligosaccharides that are then imported into the periplasm via a TonB-dependent transporter. Periplasmic β-glucosidases, predicted in silico using the Psortb 3.0 software64, CapGH3a, and CapGH3b, act on the non-reducing ends to release glucose. *CapSGBP-B and CapSusD-like representations are based on modeled structures.

Full size image

It is also worth noting that in the human gut microbiota, there is a PUL specialized for β(1,3)-glucans as characterized by Bacteroides uniformis that involves a GH158 strictly endo-β(1,3)-glucanase and a specific β(1,3)-glucosidase21. Therefore, while in the human gut microbiota two highly specialized PULs are involved in the breakdown of mixed-linkage glucans and β(1,3)-glucans, in the herbivore gut microbiota a single PUL with gene architecture mimicking the human gut Bacteroidota MLGUL, but functionally augmented, is able to cope with both types of β-glucans. The divergent genomic organizations of B. uniformis, B. ovatus and Alloprevotella sp. highlight the plethora of molecular strategies in the Bacteroidota phylum to cope with the diversity of plant non-cellulosic β-glucans.

Furthermore, analysis of metagenomic data from the gut microbiota of other herbivores revealed the presence of equivalent PULs also harboring a second GH3 member that conserves the molecular adaptation in the +1 subsite associated with the regioselectivity towards β(1,6)-glucosyl linkages (Supplementary Figs. 30 and 31). It indicates that this mechanism exploiting a single PUL to cope with distinct types of β-glucans including (1,3;1,4)-β-D-Glucans and glucans substituted with β(1,6)-glucosyl linkages are evolutionarily conserved in herbivores.

In conclusion, our findings reveal an unprecedented microbial mechanism to cope with complex dietary glycans, thereby expanding the current model for non-cellulosic β-glucans utilization by commensal gut bacteria. This discovery highlights an additional layer of functional and evolutionary complexity within the gut microbiota, extending beyond conventional intraspecies genomic variations to intricate biochemical adaptations.

Methods

Phylogenetic analysis

Phylogenetic analysis was performed using 181 genomes, including Bacteroidota genomes from the UBA project40, reference genomes from the Bacteroidaceae family, and genomes recovered from the capybara gut metagenome19. A set of 92 single copy genes according to the UBCG pipeline41 were concatenated and used to perform a maximum likelihood phylogeny using the amino acid sequences and FastTree software with default parameters. Whole genome comparison, average nucleotide identity (ANI) and average fraction (AF) values were calculated using fastANI software42.

Gene cloning, mutagenesis, protein expression, and purification

The genes encoding the target enzymes without the predicted signal peptide were cloned into the pET 28a (+) vector with a N-terminal 6xhis-tag. The full-length sequence of CapGH16_3 (including catalytic and PF13004 domains) was used for recombinant expression, enzymology, and structural biology. CapGH3a W88A mutant was generated by inverse PCR using a primer pair (foward: GACGAGGCGGAACAGGCGGG and reverse: CTGTTCCGCCTCGTCCCACAGC), in which the codon changes were introduced. PCR amplicons were circularized by the Gibson assembly43.

The vectors containing each gene were transformed into Escherichia coli BL21(DE3) cells and grown in 1 L of Terrific Broth (TB) medium at 37 °C until OD600nm = 0.6. Then, the induction was performed with 0.3 mM isopropyl-D-thiogalactopyranoside (IPTG) at 18 °C during overnight. The cells were harvested by centrifugation at 8000 × g (30 min, 4 °C) and lysed with 20 mL of buffer A (20 mM sodium phosphate buffer, 300 mM NaCl, 5 mM imidazole, pH 7.4), 4 mM PMSF, 0.1 mg mL−1 lysozyme and 75 µg mL−1 DNAse, incubated on ice for 40 min and disrupted by sonication. The lysed cells were clarified by centrifugation (20,000 × g, 50 min, 4 °C) and the supernatant was loaded onto a 5 mL Hi-Trap chelating HP column (GE Healthcare Biosciences, Pittsburgh, USA) coupled to an ÄKTA system (GE Healthcare Biosciences, Pittsburgh, USA) and pre-equilibrated with buffer A. The bound proteins were eluted with a nonlinear gradient of buffer B (20 mM sodium phosphate, 300 mM NaCl, 500 mM imidazole, pH 7.4). The eluted fractions were analyzed by SDS-PAGE, and those containing pure proteins were pooled, concentrated by filtration, and submitted to size-exclusion chromatography using a Superdex 200 16/60 column (GE Healthcare Biosciences, Pittsburgh, USA) previously equilibrated with 20 mM sodium phosphate, 150 mM NaCl, pH 7.4, at a flow rate of 1 mL min−1. Protein concentration was estimated measuring the absorbance at 280 nm using the specific molecular weight and the extinction coefficient of each enzyme.

Enzyme assays and kinetic parameters determination

The activity of each enzyme was screened against a set of polysaccharides (0.5% (w/v)) (Supplementary Table 2) and the synthetic substrate pNP-β-D-glucopyranoside (5 mM). For the polysaccharides, the amount of reducing sugar released was determined using the DNS method (A540 nm)44, with results expressed as glucose equivalents, based on a glucose standard solution (0–0.5 mg mL−1) for the calibration curve. For pNP-β-D-glucopyranoside, the hydrolysis was quantified by measuring the absorbance of p-nitrophenolate (calibration curve from 0 to 42 mM) at 405 nm with results expressed in mM. The optimum pH and temperature of each enzyme were determined on the substrate that enzyme showed the highest activity, ranging pH from 3.0 to 8.0 using 100 mM McIlvaine buffer and temperature from 10 to 70 °C. The substrate saturation curves were obtained using the enzyme concentrations and reaction times in the linear range.

The kinetics parameters of CapGH3a and CapGH3b against cellobiose (G4G), laminaribiose (G3G), and gentiobiose (G6G), were determined on an LTQ XL TM linear ion trap mass spectrometer (Thermo Fisher Scientific, Waltham, USA) using a direct infusion at a flow rate of 10 µL min−1, ESI (+), +3.5 kV, 275 °C and mass range m/z 150–500 Da. The kinetic parameters of CapGH3a W88A variant (against G4G, G3G and G6G) and CapGH3b (against cellotetraose (G4G4G4G), cellohexaose (G4G4G4G4G4G), laminaritetraose (G3G3G3G) and laminarihexaose (G3G3G3G3G3G)) were determined using a Synapt XS spectrometer (Waters Corporation, Milford, USA) through flow-injection analysis at a flow rate of 200 µL min−1, operated in ESI (+), +3 kV, 250 °C, and a mass range m/z 50–1200 Da. Glucose intensity was compared to the internal standard intensity (xylobiose) for analyte quantification. An internal standard calibration curve with glucose (0 to 2 mM) was used to determine the concentrations of the products in each enzymatic reaction.

The reported values for enzyme assays correspond to the mean and standard deviation calculated from three independent experiments (n = 3). Initial reaction rates (v0/[E]t) were plotted against substrate concentrations (in mM for pNP-β-D-glucopyranoside and oligosaccharides, and in mg mL−1 for polysaccharides). Kinetic parameters were determined using nonlinear regression analysis of the Michaelis–Menten plot, performed with Origin 8.1 software (Version 2022b, OriginLab Corporation, Northampton, USA).

Mass spectrometry cleavage pattern analysis

The products of enzyme assays were evaluated through mass spectrometry. The enzyme assays were performed using 4.3 mM of oligosaccharides laminaribiose (G3G), laminaritriose (G3G3G), laminaritetraose (G3G3G3G), laminaripentaose (G3G3G3G3G), laminarihexaose (G3G3G3G3G3G), cellobiose (G4G), cellotriose (G4G4G), cellotetraose (G4G4G4G), cellopentaose (G4G4G4G4G), cellohexaose (G4G4G4G4G4G), 32-β-D-Glucosyl-cellobiose (G3G4G), 31-β-D-Cellobiosyl-glucose (G4G3G), 31-β-D-Cellotriosyl-glucose (G4G4G3G) and 32-β-D-Cellobiosyl-cellobiose + 33-β-D-Glucosyl-cellotriose (G4G3G4G (~80%) + G3G4G4G (~20%)) or 2.5 mg mL−1 of polysaccharides laminarin from Eiseinia bicyclis and Laminaria digitata (EbLam and LdLam, respectively) and oat β-glucan. The enzyme amount added in the reactions were 60 ng CapGH16_3, 20 µg CapGH3a, and 4 µg CapGH3b. Reactions were performed at the optimum pH and temperature of each enzyme. Samples were collected at 0, 15, 30, 120, 360, and 1440 min, and reactions were stopped by adding methanol. Reactions were analyzed on an LTQ XL TM linear ion trap mass spectrometer (Thermo Fisher Scientific, Waltham, USA). ESI(+) ion source was employed. The spray voltage and ion source temperature was maintained at +3.5 kV and 250 °C, respectively. 5 µL of the quenched reactions were mixed with 185 µL of 12 µM xylotetraose (in water, used as the internal standard), and then, injected into the mass spectrometer in scan mode (m/z 180–1030) with direct infusion of 60 s at a flow rate of 10 µL min−1. A calibration curve was used to determine the concentrations of the products in each reaction.

Carbohydrate analytical method for oat β-glucan product identification

Products released from oat β-glucan hydrolysis by CapGH16_3 were analyzed by high-performance anion exchange chromatography coupled with pulsed amperometric detection (HPAEC-PAD) using a Dionex ICS3000 system (Thermo Fischer Scientific). Reactions containing 0.5% (w/v) of oat β-glucan, 50 mM McIlvine buffer (pH 6.0), and 100 ng of purified CapGH16_3 were incubated for 16 h at 40 °C. The samples were analyzed on the HPAEC-PAD system, equipped with a 4 × 50 mm CarboPac PA100 guard column and a 4 × 250 mm CarboPac PA100 analytical column (Thermo Fisher Scientific), maintained at 30 °C. The injection volume was 10 µL, and the analysis was performed at a flow rate of 0.6 mL min−1, using the solvents A (1 M sodium hydroxide), B (1 M sodium acetate), and C (ultrapure water). The procedure was as follows: (i) 0–40 min: linear gradient from 0 to 12% B, 22 to 10% A and 78% C; (ii) 40–45 min: 50% B, 50% A and 0% C; and (iii) 45–50 min: 0% B, 22% A and 78% C.

Synergism assays

Synergism studies with the enzymes from the herbivore gut Bacteroidota PUL were assessed using 8 mg mL−1 of the substrates EbLam, LdLam, or oat β-glucan in 50 mM McIlvaine buffer, pH 6.0. Enzymes were added to the reactions in different combinations (CapGH16_3 only, CapGH3a only, CapGH3b only, CapGH16_3 + CapGH3a, CapGH16_3 + CapGH3b, and CapGH16_3 + CapGH3a + CapGH3b) in the amount of 0.5 µg of CapGH16_3, 1 µg of CapGH3a and 1 µg of CapGH3b. Reactions were carried out at 35 °C, over 18 h, and the released glucose was determined by glucose oxidase kit (Labtest, Lagoa Santa, BRA). Data were calculated from three independent experiments (n = 3).

E. coli K12 cultivation

For growth curve analysis, E. coli K12 was cultured in LB medium at 37 °C and 200 rpm until the end-exponential phase. Then, the harvested cells were washed and transferred to reach a final OD600 nm of 0.05 to the minimal M9 medium (M9 control), minimal M9 medium with glucose 0.1% (w/v) (glucose control), or with the following substrates: 0.1% (w/v) laminarin from Eisenia bicyclis (EbLam control) or 0.1% (w/v) laminarin from Laminaria digitata (LdLam control). Additionally, the cells were transferred to the M9 medium plus substrates previously treated with herbivore gut Bacteroidota PUL enzymes (EbLam or LdLam). The treatment of substrates was performed using a mix of herbivore gut Bacteroidota PUL enzymes (1.5 µg CapGH16_3, 1.5 µg CapGH3a, and 1 µg CapGH3b) at 225 rpm, 37 °C, 16 h. The bacterial growth was monitored in a growth profiler 960 equipment (Enzyscreen, Heemstede, NLD) for 40 h at 37 °C, 225 rpm. The green values were converted to optical density at 600 nm by a calibration curve performed as described by the manufacturer. Three biological replicates were used for each condition (n = 3).

Affinity gel electrophoresis

The binding affinity of CapSGBP-B to β(1,3)-glucans was evaluated using laminarin from Laminaria digitata and oat mixed-linkage β-glucan, as previously described by Mandelli et al.45. Briefly, the polysaccharides were added in native gels at a final concentration of 0.5% (w/v), consisting of 7.5% (w/v) acrylamide in 25 mM Tris, 250 mM glycine buffer at pH 8.3. 10 µg of CapSGBP-B and 10 µg of bovine serum albumin (BSA) (used as negative control) were loaded onto the gel and subjected to non-denaturing electrophoresis at 60 mA for 2 h, at room temperature. Proteins were stained and visualized using Coomassie Brilliant Blue.

Size-exclusion chromatography coupled to multi-angle light scattering (SEC-MALS)

SEC-MALS experiments were performed using an 8-angle static light scattering detector DAWN 8 and Optilab refractive index monitor (Wyatt Technology, Santa Barbara, USA). Detectors were coupled to Agilent 1260 Infinity II Isocratic Pump and Vialsampler (Agilent Technology, Santa Clara, USA) with a WTC-030N5 analytical size-exclusion column (Wyatt Technology, Santa Barbara, USA). 30 µL of CapGH3a and CapGH3b at 2 mg mL−1 were loaded onto the column pre-equilibrated with 20 mM sodium phosphate pH 7.5 and 150 mM NaCl in a flow rate of 0.25 mL min−1. The SEC column was calibrated using BSA as a molecular weight marker, and data were processed using the ASTRA software 8.1.2 (Wyatt Technology, Santa Barbara, USA).

Protein crystallization

Crystals were grown by the hanging-drop vapor-diffusion method at 18 °C in drops containing 1.0 µL of protein solution, and 1.0 µL of the crystallization condition equilibrated against 200 µL of crystallization condition in 48-well plates. The enzymes were crystallized under the following conditions: CapGH16_3: 0.2 M sodium chloride, 0.1 M sodium phosphate citrate pH 4.2, PEG8000 (20% v/v), 0.01 M calcium chloride; CapGH3a: 0.1 M imidazole pH 8.0, PEG8000 (11% v/v); and CapGH3b: 0.1 M HEPES pH 6.5, PEG6000 (8% v/v), Glycerol (10% v/v). Crystals were cryoprotected using glycerol (30% (w/v)).

Data collection and structure determination

X-ray diffraction data were collected at the BL9-2 beamline from the Stanford Synchrotron Radiation Lightsource (Stanford, CA). Data were indexed, integrated, and scaled using the XDS package46. The crystal structure was solved by molecular replacement using the PHASER software47 with the atomic coordinates of the endo-glucanase from Rhodothermus marinus (RmLamR, PDBID 3ILN29) as a search model for CapGH16_3, the β-glucosidases from Bifidobacterium adolescentis (BaBgl3, PDBID 5WAB34) for CapGH3a and from B. ovatus (BoGH3B, PDBID 5JP015) for CapGH3b. The atomic coordinates were refined with PHENIX.REFINE48, alternating cycles of manual modeling and inspection with COOT49. The final model was validated with MolProbity server50. Data collection, processing, and refinement statistics are summarized in Supplementary Table 3.

Small-angle X-ray scattering

Small-angle X-ray scattering (SAXS) measurements were performed at the SAXS1 beamline from the LNLS (Campinas, Brazil). The wavelength was set to 1.488 Å, and X-ray scattering was recorded using a Pilatus 300 K (Dectris, Baden, CHE). The sample-to-detector distance was adjusted to a scattering-vector range of 0.01 < q < 0.5 Å−1, where q is the magnitude of the q-vector defined by q = 4πsinθ/λ (2θ is the scattering angle). 1 frame of 300 s was recorded and used for calculations, after subtracting the sample buffer scattering. The distribution curve p(r) was calculated using the GNOM package51 and was used to estimate the radius of gyration (Rg). Molecular envelopes were calculated from the experimental SAXS data using the program DAMMIF52. The theoretical scattering curve of the crystallographic structure was calculated, and the structure fitting into the SAXS molecular envelope was performed using the program SUPCOMB53.

Docking and molecular dynamics simulations

Molecular docking calculations were carried with AutoDock Vina54 and all ligands structures were retrieved from the Protein Data Bank (PDB). Box dimensions were set to 10 × 10 × 10 Å3 and 45 × 35 × 35 Å3 for docking with disaccharides and hexasaccharide, respectively. The exhaustiveness used was 30 (for disaccharides calculations) and 50 (for hexasaccharide calculation). The CapGH3b dimer was generated by structural superimposition of the coordinates from the docked complex in each of the protomers that compose the stable dimer retrieved by PDBePISA55. Missing loops of crystallographic structures were constructed with RosettaRemodel56, prior to systems preparation for molecular dynamics. Protonation states of the complexes were assigned considering the pH of 6.0 for CapGH3a and 7.0 for CapGH3b using propKa software57 and visual inspection of the chemical environment.

Molecular dynamics simulations were performed using the Amber20 package58, using the ffSB14 force field59 for proteins and GLYCAM60 for saccharides. The systems were solvated with TIP3P61 water molecules, and the total charges were neutralized by the addition of Na+ ions within the solvent. The systems were submitted to two steps of energy minimization, the first keeping both the protein and substrate fixed and the second without any restraints. After minimization, the systems were gradually heated until the temperatures of 298 K for CapGH3a and 308 K for CapGH3b, with spatial restraints applied to the protein and ligands position during the first heating interval (until 100 K). Afterwards, the systems were equilibrated in the NPT ensemble (100 ps) followed by a short MD in the NVT ensemble. Spatial restraints to the protein and ligands (distance between the acid/base and the glycosidic oxygen) were initially applied during the heating and equilibration steps, being released during the production. Production runs were extended to 100 ns with a time step of 2 fs, employing the SHAKE algorithm. Trajectory analyses were carried out using CPPTRAJ Tools62 and VMD63.