Unravelling the specificity and mechanism of sialic acid recognition by the gut symbiont Ruminococcus gnavus

ARTICLE Unravelling the specificity and mechanism of sialic acid recognition by the gut symbiont Ruminococcus gnavus C. David Owen1,7, Louise E. Tailford2, Serena Monaco3, Tanja Šuligoj2, Laura Vaux2, Romane Lallement2, Zahra Khedri4,8, Hai Yu5, Karine Lecointe2, John Walshaw2,6, Sandra Tribolo2, Marc Horrex2, Andrew Bell2, Xi Chen5, Gary L. Taylor1, Ajit Varki 4, Jesus Angulo 3& Nathalie Juge2 Ruminococcus gnavus is a human gut symbiont wherein the ability to degrade mucins is mediated by an intramolecular trans-sialidase (RgNanH). RgNanH comprises a GH33 catalytic domain and a sialic acid-binding carbohydrate-binding module (CBM40). Here we used glycan arrays, STD NMR, X-ray crystallography, mutagenesis and binding assays to determine the structure and function of RgNanH_CBM40 (RgCBM40). RgCBM40 displays the canonical CBM40 β-sandwich fold and broad specificity towards sialoglycans with millimolar binding affinity towards α2,3- or α2,6-sialyllactose. RgCBM40 binds to mucus produced by goblet cells and to purified mucins, providing direct evidence for a CBM40 as a novel bacterial mucus adhesin. Bioinformatics data show that RgCBM40 canonical type domains are widespread among Firmicutes. Furthermore, binding of R. gnavus ATCC 29149 to intestinal mucus is sialic acid mediated. Together, this study reveals novel features of CBMs which may contribute to the biogeography of symbiotic bacteria in the gut. DOI: 10.1038/s41467-017-02109-8 OPEN 1Biomolecular Sciences Building, University of St Andrews, St Andrews KY16 9ST, UK. 2The Gut Health and Food Safety Programme, Quadram Institute Bioscience, Norwich Research Park, Norwich NR4 7UA, UK. 3School of Pharmacy, University of East Anglia, Norwich Research Park, Norwich NR4 7TJ, UK. 4Glycobiology Research and Training Center (GRTC), Departments of Medicine and Cellular and Molecular Medicine, UC San Diego, La Jolla, CA 92093- 0687, USA. 5Department of Chemistry, University of California-Davis, Davis, CA 95616, USA. 6School of Computing Sciences, University of East Anglia, Norwich NR4 7TJ, UK. 7 Present address: Diamond Light Source Ltd, Diamond House, Harwell Science and Innovation Campus, Didcot OX11 0FA, UK. 8 Present address: Ajinomoto Althea Inc, 11040 Roselle Street, San Diego, CA 92121, USA. C. David Owen and Louise E. Tailford contributed equally to this work. Correspondence and requests for materials should be addressed to N.J. (email: [email protected]) NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications 1 1234567890 The human gut microbiota encompasses a complex community of bacterial species, which play a critical role in human health, through their contribution to e.g., polysaccharide digestion, immune system development and pathogen defence1. Microbiota composition varies longitudinally along the gastrointestinal (GI) tract but also transversally from the lumen to the mucosa1,2. Most gut bacteria reside in the colon, reaching 1011–1012 cells per gram, where they compete for dietary and host glycans3,4. A dysbiosis of the gut microbiota is associated with intestinal diseases, including cancers, infections and inflammatory bowel diseases5–8, underscoring the importance of understanding these host-microbe interactions in order to devise novel treatment strategies. Several factors influence the biogeography of symbiotic bacteria within the gut, including the gradient and availability of glycans within discrete physical niches2,3. The mucus layer covering the GI tract is at the interface between the gut microbiota and the host5. In the colon, the mucus layer is divided into a loose outer layer providing a habitat to commensal bacteria and an inner layer adhering to the epithelium and providing protection from bacterial invasion5. The outer mucus layer hosts a distinct intestinal microbial niche9. The intestinal mucus layers are built around large highly glycosylated gel-forming mucin MUC2 (Muc2 in mouse) secreted by goblet cells10. The glycan structures present in mucins are diverse and complex and consist of four core mucin-type O-glycans containing N-acetylgalactosamine (GalNAc), galactose (Gal) and N-acetylglucosamine (GlcNAc). Mucin O-glycosylation starts with the attachment of GalNAc residues to the hydroxyl group of Ser and Thr of the protein backbone to form the Tn antigen (GalNAcα1-Ser/Thr). This glycan is then elongated into core 1 (Galβ1-3GalNAcα1-Ser/Thr, also known as Thomsen Friedenreich-TF- or T-antigen), core 2 (Galβ1-3(GlcNAcβ1-6)GalNAcα1-Ser/Thr), core 3 (GlcNAcβ1- 3GalNAcα1-Ser/Thr) or core 4 (GlcNAcβ1-3(GlcNAcβ1-6)Gal- NAcα1-Ser/Thr)11. Core 3-derived O-glycans are important components of human colonic mucin-type O-glycans12. These core structures are further elongated by the addition of other carbohydrates (e.g., N-acetyllactosamine, LacNAc) and are most commonly terminated by fucose and sialic acid sugar residues via α1–2/3/4 and α2–3/6 linkages, respectively. These oligosaccharide chains provide binding sites and nutrients to the bacteria which have adapted to the mucosal environment13,14.Reflecting the structural diversity of mucin glycans and their prime location, commensal and pathogenic microbes have evolved a range of adhesins allowing their interaction with mucus13,15. Variation in mucosal carbohydrate availability leads to variations in the composition of the resident microbiota3,16,17 and may also impact on bacterial tropism along and across the GI tract18. Sialic acids, such as N-acetylneuraminic acid (Neu5Ac) and fucose residues in terminating positions on mucin glycan chains are prominent targets for commensal and pathogenic bacteria19,20. The ratio of sialic acid to fucose increases along the GI tract, from the ileum to the rectum in humans21 and an inverse gradient occurs in mice22. Furthermore blood group Sda/Cad related epitopes, GalNAcβ1-4(NeuAcα2-3)Gal, increase along the length of the human colon12. Over 100 complex oligosaccharides can be identified in mucins from human colonic biopsies, with most being mono-, di- or trisialylated23. Release of sialic acid by microbial sialidases allows bacteria to access free sialic acid for catabolism, decrypt host ligands for adherence, participate in biofilm formation, modulate immune function by metabolic incorporation, and expose the underlying glycans for further degradation10,14,19,20. Sialidases are often associated with additional domains including carbohydrate binding modules (CBMs), such as sialic acid-specific CBM4014,24 and broadly specific CBM3225. CBMs can enhance catalytic activity by concentrating the enzymes onto carbohydrate substrates26 or mediate adherence to host cells27. Ruminococcus gnavus is a prominent member of the gut microbiota of the healthy human gut28.R. gnavus utilisation of mucin is associated with the expression of an intramolecular trans-sialidase (IT-sialidase)29,30, which is proposed to play a key role in the adaptation of gut bacteria to the mucosal environment by providing 2,7-anhydro-sialic acid as a preferential source of nutrients31. The IT-sialidase from R. gnavus (Rg) ATCC 29149 (RgNanH) comprises a catalytic glycoside hydrolase domain, RgGH33 and a carbohydrate binding module, RgCBM40. Here, to gain insights into the role and specificity of sialic acid recognition by R. gnavus, we employed glycan microarray, X-ray crystallography, saturation transfer difference nuclear magnetic resonance spectroscopy (STD NMR), isothermal titration calorimetry (ITC), mutational analyses, and cell/tissue binding assays to identify RgCBM40 oligosaccharide binding partners. Prominent ligands were oligosaccharides with terminal sialic acid, including those which are not substrates for RgNanH activity. We propose a novel role for CBM40 in targeting gut bacteria towards sialic acid-rich regions of the GI tract. Results RgCBM40 belongs to the CBM40 subfamily.RgCBM40 crystallised as a dimer, adopting the canonical CBM40 β-sandwich fold with six antiparallel strands on the convex face and five on the concave face (Fig. 1a, for data collection and refinement statistics, see Table 1). Electron density was observed for all RgCBM40 residues present in the construct (50–237). The sialic acid binding site is on the concave face at the dimer interface (Fig. 1b), however, size exclusion chromatography with multi angle light scattering (SEC-MALS) indicated that the full-length protein, RgNanH, is monomeric in solution (Fig. 1c). The macromolecular architecture of RgCBM40 is conserved among members of the CBM40 family (Supplementary Fig. 1), with the exception of Vibrio cholerae CBM40_NanH (VcCBM40_NanH) which is proposed to be part of a separate CBM40 subfamily (Supplementary Fig. 1h)25,32. Greatest structural homology was observed to MdCBM40 NanL (RMSD: 0.3 Å) from the Macrobdella decora IT-sialidase (Supplementary Fig. 1e)33. Protein ligand complexes were achieved for both 3′SL and 6′SL (Fig. 1d, e). No significant conformational changes were observed in the binding site upon ligand binding. Definitive electron density for the Neu5Ac and galactose residues was observed in the 3′SL and 6′SL complexes. In the 6′SL complex, electron density was also observed for the glucose residues (Fig. 1e), with the lactose positioned almost perpendicular to the sialic acid (Fig. 1e). Contrastingly, for the 3′SL complex, only partial electron density was observed for the glucose residue in a single monomer (Fig. 1d), and the glucose positioning indicates that the lactose points up and away from the binding site, without further interactions with the protein. In the 3′SL complex, the lactose positioning would permit further extensions to the carbohydrate chain as would be present in more complex or anchored glycans, whereas these may be blocked in the 6′SL complex. This would provide a degree of specificity towards sialic acid linkage. Neu5Ac binds in a chair conformation (Fig. 1f, g), mimicking the solution conformation and minimising the energetic penalty paid upon binding26. Notably, the carboxylic acid group of Neu5Ac forms electrostatic interactions with an arginine dyad, Arg204 and Arg128, mimicking the coordination observed in sialidase active sites. The C4 hydroxyl group hydrogen bonds to Lys135 and Glu126, the N-acetyl group sits in a hydrophobic pocket formed by Tyr116 and Ile95. The N-acetyl group nitrogen interacts with both Glu126 and Tyr210. Glu126, Arg128 and ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 2NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications Arg204 make extensive interactions with the bound ligand and are conserved in all structurally characterised CBM40 sialic acid binding sites, discounting VcCBM40_NanH34 (Supplementary Fig. 2). The environment of the glycerol side-chain of sialic acid is generally conserved across the canonical CBM40 subfamily with the rear face (C7-H and C9-H groups) residing on a hydrophobic surface formed by Ile95 and Tyr210 in RgCBM40 (Supplementary Fig. 3a). Although VcCBM40_NanH shares the CBM40 β-sandwich fold (Supplementary Fig. 1), the location, orientation, and constitution of its sialic acid binding site is not conserved (Supplementary Fig. 2). Structure-based sequence alignment. CBM40s associated with sialidases fall into two subfamilies, the canonical subfamily exemplified by CpCBM40_NanJ25 (which also regroups RgCBM40, CpCBM40_NanI32,SpCBM40_NanA35, SpCBM40_NanB36,SpCBM40_NanC37 and MdCBM40 NanL33), and the Vibrio subfamily exemplified by VcCBM40_NanH34. Considerable sequence divergence between the Vibrio and canonical CBM40 types renders satisfactory alignments difficult to produce with standard tools, as also previously reported32. Here, by detailed manual inspection, paying particular attention to the limits of secondary structure elements and intervening loops, we produced an alignment of both types of CBM40 sequences showing well-conserved positions along its length, notwithstanding the Vibrio insertion (40 residues) near the N-terminus. The pairwise identities between the canonical representatives range from 21 to 67%, while the maximum canonical vs. Vibrio identity is 17%, reflecting that CBM40s fall into two distinct groups. This highlighted conserved residues within the canonical subfamily that may be involved in binding affinity and specificity (Fig. 2). These include (RgCBM40 numbering): an arginine dyad (Arg204 and Arg128) that interacts with the sialic acid carboxylic acid group, a glutamic acid Table 1 Data collection and refinement statistics Data set Apo 3′SL 6′SL Data collection Space group P21 P21 P21 Cell dimensions a,b,c(Å) 46.7, 72.8, 51.3 48.8, 72.4 51.5 48.7, 72.2, 51.4 β(°) 104.9 105.1 103.9 Resolution 49.56–1.73 (1.76–1.73) 39.48–1.37 (1.41–1.37) 49.91–1.30 (1.34–1.30) R merge 0.03 (0.14) 0.04 (0.34) 0.03 (0.15) I/σI47.3 (9.6) 22.9 (3.0) 32.0 (4.9) Completeness 91.8 (51.3) 74.5 (11.2) 83.9 (13.6) Redundancy 3.7 (2.4) 4.3 (2.4) 5.1 (1.4) Refinement Resolution 49.56–1.73 (1.76–1.73) 39.48–1.37 (1.41–1.37) 49.91–1.30 (1.34–1.30) No. of reflections 31,570 51,221 67,097 R work /R free 0.160/0.194 (0.82) 0.152/0.187 (0.81) 0.134/0.154 (0.87) No. of atoms 3145 3424 3704 Protein 2807 2850 3076 Ligand 0 81 123 Water 338 508 527 B-factors Protein 19.4 16.6 10.4 Ligand/ion 36.4 21.7 Water 28.1 31.7 27.3 R.m.s.d Bond lengths (Å) 0.011 0.012 0.015 Bond angle (°) 1.55 1.66 1.77 Values in parentheses refer to the highest resolution shell. For the 3′SL and 6′SL complexes the data were over 90% complete to a resolution of 1.85 Å and 1.56 Å, respectively. One crystal was used for each structure a Asp 110 Asp 131 Arg 128 Lys 135 Glu 126 Tyr 116 Ile 95 Tyr 210 Asp 110 Arg 204 Asp 131 Arg 128 Lys 135 Glu126 Tyr 116 Ile 95 Tyr 210 Arg 204 c de fg 90° b 1.0 0.8 0.6 0.4 0.2 0.0 20 40 60 80 100 100 200 150 50 0 Molar mass (kDa) Elution volume (mL) Absorbance (280 nm, normalised) Fig. 1 Crystal structure of RgCBM40 in complex with 3′SL and 6′SL. aRgCBM40 is shown in a cartoon representation with a rotation of 90° around the xaxis. bThe protein crystallised as a dimer with the ligand binding site at the dimer interface. The binding sites are shown occupied by 6′SL trisaccharides (Neu5Ac: cyan, galactose: blue, glucose: orange). cSEC-MALS performed with full length RgNanH (77 kDa). The SEC-MALS predicted molecular weight was 73 kDa, indicating that RgNanH is monomeric in solution. Bound (d)3′SL and (e)6′SL are shown with their corresponding Fo-Fc omit maps at 2σ(light cyan), 3σ(orange) and 5σ (magenta). The omit maps are carved at 1.6 Å around the bound ligand. For 3′SL, the map is carved around a dummy glucose residue to indicate the presence of partial electron density. A close-up view of RgCBM40 binding site is shown with (f)3′SL and (g)6′SL. The Neu5Ac residue is shown in cyan and the galactose residue as black lines, for clarity the glucose residue is not shown. Interacting RgCBM40 residues are shown in green with black dashed lines indicating hydrogen bonding interactions. A semi-transparent surface indicates the hydrophobic surface NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 ARTICLE NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications 3 (Glu126), which hydrogen bonds to the C4 hydroxyl; and a hydrophobic surface, which accommodates the N-acetyl moiety and the hydrophobic face of the glycerol group. Tyr116, Ile95, Tyr210 contribute to the surface of an aromatic:aliphatic:aromatic twisted platform which presents the glycerol hydroxyl groups to solvent26. Bioinformatics analyses. To gain further insights into the phylotypic distribution of the CBM40 domains within bacterial genomes, we performed a database search using pHMMs derived from our alignment as queries (canonical and Vibrio-type together, referred to as ‘combined’; canonical only; Vibrio-type only) as well as Pfam models, “Sialidase(NTD)”,“Laminin_G_3”, and “Sial-lect-inser”(see Methods section and Supplementary Methods). Our combined model successfully identified 99.9% of the CBM40 domains matched by the individual type CBM40 models (over 16,000 domain hits in the whole database of around 67,000 genomes). Further analysis of the data (Supplementary Methods) led to the identification of 51 nonredundant sequences (Supplementary Fig. 4). Of these, the canonical CBM40 domains occurred in Firmicutes with 40 sequences, representing 18 genera or pseudogenera, divided between classes Bacilli and Clostridia, as well as Erysipelotrichi and an unclassified member of the Firmicutes; and two sequences in Actinobacteria. The Vibrio-type occurred only in Gammaproteobacteria, represented by 8 sequences in five genera. The separation between the Vibriotype sequences and canonical CBM40 sequences across bacterial genomes was also apparent from a tree representation constructed using a simple distance-based model and neighbourjoining (Fig. 3). This dichotomy was fully supported by bootstrap analysis of 1000 replicates. There was no evidence for any intermediate or other CBM40 types. Only one sequence from Actinobacillus muris containing a canonical CBM40 (confirmed by pHMMs and conserved binding residues) was shown to be part of a Gammaproteobacteria clade (all other members Vibriotype) as supported by 79% of bootstraps. Further studies may indicate whether this domain is the closest to an inferred common ancestor of the canonical and Vibrio CBM40 types. The results for co-incidence of sialidase domains clearly indicated an association with CBM40s in this set of non-redundant sequences: we detected a sialidase domain in 92% of canonical-type CBM40 and in all Vibrio-type CBM40 representatives. RgCBM40 preferentially binds α2,3 linked sialosides. To further explore RgNanH ligand specificity, RgCBM40 and inactive mutant RgGH33 D282A, were tested for binding to various sialoglycans, using a slide microarray38,39. This sialoglycan microarray presents over 60 synthetically recreated naturally occurring oligosaccharide structures with diverse sialic acid forms, glycosidic linkages, and underlying glycans, representing a broad range of such targets38,39. Both recombinant proteins exclusively bound to glycans terminated with sialic acids (Fig. 4). They also showed distinct specificities. RgCBM40 bound to most but not all glycans bearing terminal Neu5Ac, Neu5Gc, Neu5,9Ac 2 and 2-keto-3 deoxynonulosonic acid (Kdn) attached with α2-3, α2-6 and α2-8 linkages, with weaker binding for Neu5Gc and kdn oligosaccharides (Fig. 4). In contrast, RgGH33 D282A interacted weakly with a narrow spectrum of sialoglycans, mainly α2-3- SpCBM40_NanA 10 20 30 40 50 60 70 150140 13012011010090 170 250 260 180 190 200 210 220 230 SpCBM40_NanB SpCBM40_NanC MdCBM40_NanL MdCBM40_NanL MdCBM40_NanL MdCBM40_NanL CpCBM40_Nanj CpCBM40_Nanl VcCBM40_NanH RgCBM40 SpCBM40_NanA SpCBM40_NanB SpCBM40_NanC CpCBM40_Nanj CpCBM40_Nanl VcCBM40_NanH RgCBM40 SpCBM40_NanA SpCBM40_NanB SpCBM40_NanC CpCBM40_Nanj CpCBM40_Nanl VcCBM40_NanH RgCBM40 SpCBM40_NanA SpCBM40_NanB SpCBM40_NanC CpCBM40_Nanj CpCBM40_Nanl VcCBM40_NanH RgCBM40 Fig. 2 CBM40 structural alignment. Structure-based alignment (α-helices and β-strands respectively in red and yellow) of CBM40 domains of RgCBM40 with C. perfringens CpCBM40_NanJ (PDB code 2V73) and CpCBM40_NanI (PDB code 5FRA), M. decora MdCBM40_NanL (PDB code 1SLI) and S. pneumoniae SpCBM40_NanA (PDB code 4C1W), SpCBM40_NanB (PDB code 2VW0) and SpCBM40_NanC (PDB code 4YZ5) and VcCBM40_NanH structure (PDB code 2W68). Amino acids identified as binding sites are highlighted in blue. RgCBM40 residues Ile95, Asp110, Tyr116, Glu126, Arg128, Arg204 and Tyr210 are at positions 104, 119, 125, 135, 137, 226 and 233 of the alignment. The alignment supplemented with other canonical and Vibrio-type CBM40 sequences, used to create the pHMM using HMMER3, is shown in Supplementary Fig. 4 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 4NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications Neu5Ac-containing glycans, primarily Neu5Acα3LacNAcβ(3′ SLN), Neu5Acα3Galβ3GlcNAcβ, Neu5Acα3Galβ3GalNAc (STF), Neu5Acα3Lacβ(GM3) and Neu5Acα3Galβ3GalNAcβ3Lac (Fig. 4). Noticeably, RgGH33 D282A recognised some of the α2-3-linked sialoglycans but not any α2-6- or α2-8-linked ones, in line with its substrate specificity30. In marked contrast, every α2-3-linked sialyl oligosaccharide present on the array could be bound by RgCBM40. RgCBM40 showed a preference for terminal Neu5Ac over Neu5Gc, and for α2-3≫α2-6 >α2-8 linkages. RgCBM40 bound generally more strongly to glycans containing LacNAc and Lac. RgCBM40 could bind Neu5Ac linked Lac with α2-3 and α2-6 linkage, albeit to a lesser degree, whereas binding to Neu5Ac linked LacNAc was α2-3-specific. Due to the glycan orientation introduced by the α2-6-sialic acid linkage the 6′SL glucose residue is close to the protein surface (Fig. 1e). Therefore, α2-6-linked LacNAc N-acetyl group may be blocked by protein residues, whereas the α2-3 linked glycan would be more solvent exposed. The highest binding was to Neu5,9Ac 2 α3GalβR1. Interestingly, RgCBM40 bound to Neu5Gcα3Galβ3GalNAcβR1 (Neu5Gc-TF) and Neu5Gc9Acα3Galβ3GalNAcβR1 (Neu5Gc9Ac-TF) although with 5–10-fold less intensity, but it could not bind to the same ligands with the αR1 linkage. RgCBM40 bound to α2-3-sialylated Lewis X (3′SLX, both Neu5Ac and Neu5Gc forms, although Neu5Ac was preferred). Sulfation of the 6 position of GlcNAc in 3′SLX (both Neu5Ac and Neu5Gc) improved binding of the protein (Fig. 4). To validate some of the glycan array data, we used STD NMR spectroscopy40,41 against a range of sialylated ligands. Since the highest STD intensities correlate with the closest ligand-protein contacts in the bound state42, STD NMR experiments provide important information on the binding epitope of the complexed ligand43. Here Neu5Ac, Neu5Gc, 2,7-anhydro-Neu5Ac, 3′SL, 6′SL, Neu5Acα3Gal (3′SGal), Neu5Acα6 Gal (6′SGal), 3′SLN, Neu5A- cα6LacNAc (6′SLN), Neu5Gcα3Lac (3′SLGc), Neu5Gcα6Lac (6′SLGc), Neu5Acα6GalαOC3H6N3 (Neu5Ac-STn), Neu5Gcα6- GalαOC3H6N3 (Neu5Gc-STn) and STFαOC3H6N3 were tested as potential ligands for RgCBM40. With the exception of the three monosaccharides, Neu5Ac, Neu5Gc and 2,7-anhydro-Neu5Ac, binding to RgCBM40 was detected for all di- and trisaccharides tested. For the latter, the binding epitope mapping was obtained and analysed as described under Methods section. Figure 5a shows the STD NMR spectra of 3′SL and 6′SL, and Fig. 5b their binding epitope mapping. The sialic acid ring was found to be the Streptococcus_pneumoniae 156 Dolosigranulum_pigrum Clostridium_perfringens_1 Clostridium_perfringens_11 Clostridium_celatum_1 Clostridium_celatum_3 Clostridium_celatum_2 Paraliobacillus_sp._PM-2_1 Paraliobacillus_sp._PM-2_3 Virgibacillus_sp._SK37 Paraliobacillus_sp._PM-2_2 Enterococcus faecalis Streptococcus_pneumoniae_117 Actinomyces neuii Actinotignum_schaalii Endozoicomonas_elysicola_2 Vibrio_sinaloensis Enterovibrio_coralii Vibrio_cholerae Pseudoalteromonas_sp._P1-26 Endozoicomonas_elysicola_1 Shewanella_pealeana Shewanella_pealeana_1 Actinobacillus Robinsoniella_sp._RHS_1 Streptococcus_iniae Erysipelothrix_rhusiopathiae Streptococcus_suis_2 Streptococcus_suis_1 Streptococcus_mitis_27 Streptococcus_pneumoniae_35 Streptococcus_oralis Streptococcus_pneumoniae_1 Subdoligranulum_sp._4_3_54A2FAA Intestinimonas_butyriciproducens Robinsoniella_sp._RHS_2 Lachnospiraceae_bacterium Blautia_hansenii Ruminococcaceae-bacterium Clostridium_sp._BAA-442_3 Clostridium_sp._BAA-442_2 Clostridium_sp._BAA-442_1 Ruminococcus_gnavus Clostridium_sp._7_2_43FAA Firmicutes-bacterium Tyzzerella_nexilis Streptococcus suis_4 Enterococcus_cecorum Lactobacillus_salivarius Staphylococcus_sp._HGB0015 Staphylococcus_pseudintermedius C B V A R L I J Fig. 3 Distance-based tree of canonical and Vibrio-type CBM40 sequences. Tree of 51 non-redundant sequences (80% identity level) calculated by neighbour-joining using evolutionary distances estimated by applying the PMB model of amino acid changes, including all sites and using a uniform rate of evolution. The representative sequences corresponding most closely (at least 97% identical) to the 7 bacterial structure-determined sequences are shown with symbols, coloured in accordance with Supplementary Fig. 1:“A”,SpCBM40_NanA; “B”,SpCBM40_NanB; “C”,SpCBM40_NanC; “I”,CpCBM40_NanI; “J”,CpCBM40_NanJ; “R”,RgCBM40; “V”,VcCBM40_NanH. Additionally, “L”denotes MdCBM40_NanL closest to the bacterial sequence of highest identity (70% identical to RgCBM40) as only bacterial sequences were searched NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 ARTICLE NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications 5 main recognition element and the binding mode was not affected by the nature of the glycosidic linkage (α2-3 or α2-6) of the sialoglycan (Supplementary Fig. 5). The same was true for the other Neu5Ac-ending ligands tested (see binding epitope mapping in Supplementary Fig. 6). The overall binding epitopes of 3′SL and 6′SL from the STD NMR in solution state are in good agreement with the crystal structures (Fig. 5), where the sialic acid is in close contact to the protein surface while the lactose moiety is solvent exposed as suggested from the very low STD intensities observed for the galactose and glucose protons. Very strong STD intensity were observed at the methyl group (Fig. 5). This is in excellent agreement with the N-acetyl group sitting in the hydrophobic pocket facing many protein protons (Hδand Hγ) from the side chains of Ile95, Tyr116, and Tyr210 (Fig. 1f, g). High intensity on H7 compared to the much lower one on the adjacent H8 agrees with H7 facing the hydrophobic side chains while H8 (Fig. 5), in trans-conformation to it, is pointing towards the solvent. Within experimental error, no stark differences were observed in the orientation of the sialic acid ring in the binding pocket of RgCBM40. RgCBM40 also showed binding to Neu5Gc- ending oligosaccharides, albeit with a lower strength. Figure 5c shows the binding epitope of 3′SLGc and 6′SLGc (STD spectra are shown in Supplementary Fig. 6). Again, sialic acid was the main recognition element of these sialoglycans, but the binding epitope mapping was slightly different, in comparison to those of 3′SL and 6′SL. For the Neu5Gc-ending ligands, stronger STD intensities on H3s and lower ones on H6 were observed, suggesting a small reorientation of the ring around C3, which would expose C6, in order to fit the bulkier hydroxyl group on the acetamide moiety. The affinity of the interaction between RgCBM40 and sialic acid ligands was further assessed by ITC. Both 3′SL and 6′SL bound with similar low affinities, with dissociation constants of 0.57 mM and 1.70 mM, respectively (Fig. 6a, b, Supplementary Table 1). This confirms that RgCBM40 is specific for the terminal residue irrespective of the glycosidic linkage but with a slight preference (~three fold) for the 2–3 linkage. Furthermore, it would suggest that the additional binding interactions observed in the crystal structure of the complex between RgCBM40 and 6′SL do not significantly promote binding, also in agreement with the STD NMR results, showing that sialic acid is the main binding epitope in solution. We confirmed that RgCBM40 binds to Neu5Gc-oligosaccharides, albeit with lower affinity, in accordance with the glycan array and STD NMR results. RgCBM40 has a Kd of ~3 mM and >10 mM towards 3′SLGc and 6′SLGc, respectively (Fig. 6c, Supplementary Table 1). Very weak (~20 mM) interaction was observed between RgCBM40 and Neu5Ac (Fig. 6d) or Neu5Gc monosaccharides (Supplementary Table 1). The STD NMR experiments were carried out with 1 mM sugar, well below the Kd, which explains why no interaction was observed using this approach. Thermodynamic analysis showed that the reaction is enthalpy-driven (Supplementary Table 2). To further assess the involvement of individual residues, we introduced point mutations specifically designed to abrogate CBM binding. Arg128, Arg204, Tyr116, Tyr210, Glu126 and Ile95 were chosen for alanine substitutions. Analysis of the secondary structure by circular dichroism (CD) suggests that the recombinant proteins were correctly folded (Supplementary Fig. 7). Binding to Neu5Ac, 3′SL and 6′SL was abolished for the double mutant R128A/R204A as well as all single mutants, with the exception of I95A as shown by ITC (Supplementary Fig. 8a, b, Supplementary Table 1). I95A binds 3′SL and 6′SL with a Kd of 1.82 and 1.37 mM, respectively, broadly similar to the binding of the wild-type enzyme (Supplementary Table 1). This suggests that Ile95 is not an essential component of the hydrophobic pocket or the aromatic:aliphatic:aromatic twisted platform, and that the Tyr Glycan structure RgCBM40 RgGH33 D282A Neu5Acα6GalNAcαR1 Neu5Acα3Galβ4GlcNAcβR1 Neu5Acα3Galβ3GlcNAcβR1 Neu5Acα3Galβ3GalNAcαR1 Neu5Acα6Galβ4GlcNAcβR1 Neu5Acα6Galβ4GlcβR1 Neu5Acα3Galβ4GlcβR1 Neu5Acα3GalβR1 Neu5Acα6GalβR1 Neu5Acα3Galβ3GalNAcβR1 Neu5Acα8Neu5Acα3Galβ4GlcβR1 Neu5Acα8Neu5Acα8Neu5Acα3Galβ4GlcβR1 Neu5Acα3Galβ4(Fucα3)GlcNAcβR1 Neu5Acα3Galβ4(Fucα3)GlcNAc6SβR1 Neu5Acα3Galβ3GlcNAcβ3Galβ4GlcβR1 Neu5Acα3Galβ4GlcNAc6SβR1 Neu5Acα6(Neu5Acα3)Galβ4GlcβR1 Neu5Acα6(Neu5Gcα3)Galβ4GlcβR1 Neu5Acα6(Kdnα3)Galβ4GlcβR1 Neu5Acα8Neu5Gcα3Galβ4GlcβR1 Neu5Acα8Neu5Gcα6Galβ4GlcβR1 Neu5Acα8Neu5Acα6Galβ4GlcβR1 Neu5,9Ac2α3Galβ4GlcNAcβR1 Neu5,9Ac2α6Galβ4GlcNAcβR1 Neu5,9Ac2α3Galβ3GlcNAcβR1 Neu5,9Ac2α3Galβ3GalNAcαR1 Neu5,9Ac2α6GalNAcαR1 Neu5,9Ac2α3GalβR1 Neu5,9Ac2α6GalβR1 Neu5,9Ac2α3Galβ3GalNAcβR1 Neu5,9Ac2α6Galβ4GlcβR1 Neu5,9Ac2α3Galβ4GlcβR1 Neu5Gcα6GalNAcαR1 Neu5Gcα3Galβ4GlcNAcβR1 Neu5Gcα3Galβ3GlcNAcβR1 Neu5Gcα3Galβ3GalNAcαR1 Neu5Gcα6Galβ4GlcNAcβR1 Neu5Gcα6Galβ4GlcβR1 Neu5Gcα3Galβ4GlcβR1 Neu5Gcα3GalβR1 Neu5Gcα6GalβR1 Neu5Gcα3Galβ3GalNAcβR1 Neu5Gcα3Galβ4(Fucα3)GlcNAcβR1 Neu5Gcα3Galβ4(Fucα3)GlcNAc6SβR1 Neu5Gcα3Galβ3GlcNAcβ3Galβ4GlcβR1 Neu5Gcα3Galβ4GlcNAc6SβR1 Neu5Gcα8Neu5Acα3Galβ4GlcβR1 Neu5Gcα8Neu5Gcα3Galβ4GlcβR1 Neu5Gc9Acα3Galβ4GlcNAcβR1 Neu5Gc9Acα6Galβ4GlcNAcβR1 Neu5Gc9Acα3Galβ3GlcNAcβR1 Neu5Gc9Acα3Galβ3GalNAcαR1 Neu5Gc9Acα6GalNAcαR1 Neu5Gc9Acα3GalβR1 Neu5Gc9Acα6GalβR1 Neu5Gc9Acα3Galβ3GalNAcβR1 Neu5Gc9Ac6Galβ4GlcβR1 Neu5Gc9Ac3Galβ4GlcβR1 Kdnα8Neu5Acα3Galβ4GlcβR1 Kdnα8Neu5Gcα3Galβ4GlcβR1 Rank 100 50 0 Fig. 4 Sialoglycan microarray analysis of binding specificities of RgCBM40 and RgGH33 D282A. Binding of the recombinant proteins RgCBM40 and RgGH33 D282A at 20 and 200 µgml −1, respectively are presented (n=4, SD). Heat map was generated using the method as previously described38,39. Binding was ranked as (glycan average RFU/ maximum glycan average RFU)×100. Red and white represent the maximum and minimum, respectively. R1 represents propyl-azide as the spacer ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 6NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications residues may compensate for the mutation of Ile95 to Ala. The binding ability of I95A to 3′SL and 6′SL was further confirmed by STD-NMR (Supplementary Fig. 9). Taken together, the STD NMR and ITC data confirmed binding of both α2-3 and α2-6 linked sugars and raise questions regarding differences in ligand specificity between the catalytic and carbohydrate binding domains constituting RgNanH. We previously showed that RgNanH is specific for α2-3-linked substrates30. To determine the influence of RgCBM40 on the sialidase activity, we compared the enzymatic activity of RgNanH and RgGH33 on a range of sialylated substrates. The reaction was monitored by HPAEC-PAD and showed no difference in catalytic activity on short oligosaccharides 3′SL, 3′SLX (Neu5Ac form) or on large polymeric MUC2 mucins (Supplementary Fig. 10), indicating that, in the conditions tested, RgCBM40 did not potentiate the enzyme activity on these substrates. RgCBM40 is a novel bacterial mucus adhesin.R. gnavus ATCC 29149 but not the E1 strain encodes the IT-sialidase required for mucin-degradation29,30. Immunogold labelling and western blotting confirmed the presence of RgNanH on R. gnavus ATCC 29149 cell-surface but not E1 (Supplementary Fig. 11a, b). Given the role of RgNanH in R. gnavus mucin glycan utilisation, the binding of RgCBM40 was tested by ELISA towards a range of mucins with different glycosylation profiles. The sialylation level H8 4.3 HO OH HO HO AcHN 8 9 76 543 OOOO OH HO OH Strong Medium Weak OH OH OH OH OOO O O HO HO HO AcHN HO HO HO OH OH COO–COO– OH OH OH O HO OH HO HO HOH2COCHN HOH 2 COCHN OOOO OH HO OH Strong Medium Weak OH OH OH OH OO O O O HO HO OH HO HO HO HO OH OH COO– COO– OHOH O 4.0 3.8 3.5 3.3 3.0 2.8 2.5 2.3 2.0 ppm 4.3 4.0 3.8 3.5 3.3 3.0 2.8 2.5 2.3 2.0 ppm H8 H9’ H9’ H5 H5 H6 H6 H7 H7 H3e H3a H3e H3a Me Me Neu5Acα2-3Lac Neu5Acα2-6Lac H4 H4 a b c Fig. 5 STD NMR analysis of RgCBM40 binding to sialoglycans. aReference (top) and difference (bottom) spectra of 3′SL and 6′SL. The strongest signals from the Neu5Ac’s protons are labelled in the difference spectra. bBinding epitope mapping from STD NMR of 3′SL and 6′SL. Legend indicates relative STD intensities normalised at H7: blue, 0–24%; yellow, 25–50%; red 51–100%; larger red dots indicate values over 100%. Sialic acid is the main recognition element. cBinding epitopes mapping from STD NMR of Neu5Gcα2-3Lac and Neu5Gcα2-6Lac. Legend as above. Sialic acid is the main recognition element. The strongest STD intensities from CH2 and the H3s, suggest a reorientation of the Neu5Gc ring in the binding pocket, in comparison to 3′SL and 6′SL 09 8 7 6 5 4 3 2 1 –0.2 –0.4 –0.6 –0.8 –0.1 –0.1 0 –0.2 –0.3 –0.4 –0.5 –0.6 –0.7 –0.8 –0.9 –1 –0.2 –0.3 –0.4 –0.5 –0.6 –0.7 –0.8 –0.9 –1 –1 DP (μcal/s) ΔH (kcal/mol) ΔH (kcal/mol) –0.1 0 –0.2 –0.3 –0.4 –0.5 –0.6 –0.7 –0.8 –0.9 –1 ΔH (kcal/mol) –0.1 0 –0.2 –0.3 –0.4 –0.5 –0.6 –0.7 –0.8 –0.9 –1 ΔH (kcal/mol) –1.2 0 –0.2 –0.4 –0.6 –0.8 –1 DP (μcal/s) DP (μcal/s) –1.2 0 –0.2 –0.4 –0.6 –0.8 –1 DP (μcal/s) Time (min) Kd 0.57 ± 0.05 Kd 1.70 ± 0.14 Kd 2.69 ± 0.86 Kd 21* Molar ratio 0 0 2 4 6 8 10 12 14 16 Molar ratio 0 2 4 6 8 10 12 14 16 Molar ratio Molar ratio 0 2 4 6 8 10 12 14 16 5 10 15 20 25 30 35 40 45 50 Time (min) Time (min) 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Time (min) 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 ab c d Fig. 6 ITC isotherms of RgCBM40 to sialoglycans. aRgCBM40 binding to 3′SL, bRgCBM40 binding to 6′SL, cRgCBM40 binding to 3′SLGc, dRgCBM40 binding to Neu5Ac. The Kd is indicated in mM. *This value is an estimate as the Kd is too high to determine with the concentration of sugar used NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 ARTICLE NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications 7 of purified commercial pig gastric mucin (pPGM), mixed and Muc2/MUC2 mucins from mice and LS174T human cell line was analysed by mass spectrometry (MS), revealing that most of the mucins tested contained >8% sialylated structures; pPGM and Muc2 from the colon of wild-type C57BL/6 mice contained <2% sialylated structures whereas the level of sialylation of LS174T MUC2 reaches 91% (Supplementary Table 3). Highest binding was observed to LS174T MUC2 whereas binding was lowest to pPGM or Muc2 from the colon of wild-type mice, which contain low levels of sialylation (Fig. 7a). The interaction was dependent on the concentration of RgCBM40 (Supplementary Fig. 12). RgCBM40 generally bound more strongly to mucins extracted from C3GnT−/−mice (mutants which lack core 3 β1-3-N-acet- ylglucosaminyltransferase, C3GnT)44 than to mucins from wildtype mice. Irrespective of the mouse model, the binding of RgCBM40 to Muc2 from the small intestine was higher than from the colon (Fig. 7a). The adhesion level correlated well with the level of sialylation between the different mucins tested (r2=0.88; Fig. 7b). RgCBM40 bound significantly less strongly to MUC2 which has been treated with trifluoroacetic acid (TFA) to remove sialic acid, or with any of the sialidases tested which included the broad-specificity sialidase from Clostridium perfringens (Cp) and the α2-3-specific sialidases from Salmonella typhimurium (St), Akkermansia muciniphila (Ak) and R. gnavus (Rg), confirming the specificity of RgCBM40 for terminal sialic acid (Fig. 7c). Consistent with the low affinity of CBM40 for Neu5Ac, this monosaccharide had no effect on adherence of RgCBM40 to mucin (Fig. 7d). However, addition of free 3′SL or 6′SL prior to binding significantly decreased adherence of RgCBM40 to MUC2 (Fig. 7d). These data indicate that RgCBM40 recognises sialylated mammalian mucins. Having shown that RgCBM40 can bind to sialylated oligosaccharides and mucins, we tested its ability to bind to mucus from mouse intestinal tissue and human cell lines cells by immunofluorescence (Fig. 8). Methacarn fixation allowed preservation of mucus in both tissue sections and cell lines. Strong binding was demonstrated to mucus produced by LS174T which correlated with staining patterns of SNA (a sialic acid-specific lectin) and MUC2 (Fig. 8a). No staining was observed in negative controls (RgCBM40 free). RgCBM40, Muc2 and lectin staining was also observed in crypts as well as on the epithelial surface of mouse colonic tissue (Fig. 8b). In addition, sialidase treatment of mouse colonic sections markedly reduced the binding of RgCBM40, as well as the SNA lectin control (Fig. 8c). SNA can outcompete RgCBM40 binding to the mucus layer in mouse colonic tissue sections, further indicating that the binding of RgCBM40 to mucus is sialic acid mediated (Fig. 8d). Similar inhibition was observed when using bacterial cells. R. gnavus ATCC 29149 was shown to bind to areas that correlated with mucus staining. This binding was blocked with the addition of ab dc * * NS Mucin LS174T MUC2 pPGM SI Colon 1.0 1.0 A450 A450 A450 1.5 0.0 Murine WT Muc2 Murine C3GnT –/– mucins Murine C3GnT –/– Muc2 Murine GF C Muc2 LS174T mucins * NSNS 200 406080100 % sialylated structures Normalised response 20 40 60 80 100 0 1.5 * 2.0 1.0 0.5 0.0 NS * Competing sugar None 6’SL Neu5Ac 3’SL * * * *** ** 1.5 2.0 1.0 0.5 Mucin treatment 0.0 None TFA Cp St Am Rg Fig. 7 ELISA of RgCBM40 binding to purified mucins. aRgCBM40 binding to a range of purified mucins; mucin 2 (MUC2) and mixed mucins (mucins) from human cell line LS174T, purified pig gastric mucin (pPGM), and murine mucins from germ free (GF), wild type (WT), and C3GnT−/−mice. bCorrelation of RgCBM40 binding with % sialylated structure for each mucin tested. The % sialylated structures was determined by MS. cRgCBM40 binding to LS174T MUC2 which has been treated chemically (TFA) or enzymatically with a sialidase from Clostridium perfringens (Cp), Salmonella typhimurium (St), Akkermansia muciniphila (Am)orRuminococcus gnavus (Rg)dRgCBM40 binding to LS174T MUC2 in competition with sugars. RgCBM40 has been preincubated with the indicated sugars. In all cases, RgCBM40 was incubated with immobilised mucins and binding detected using an anti-sialidase primary antibody and an anti-rabbit secondary antibody conjugated to horseradish peroxidase. The enzyme was incubated with TMB and the absorbance at 450 nm (A450) measured. The error bars show the standard error of the mean (SEM) of three replicates. Pvalues are indicated; NS-not significant, *p<0.05, **p<0.005, ***p<0.0005 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 8NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications SNA (Fig. 8e), confirming the importance of sialic acid recognition in R. gnavus ATCC 29149 binding to mucus. Discussion Sialic acids are often found capping mammalian glycans and are thus common binding targets of commensal or invading microbes. A wide variety of microorganisms utilise CBM- containing sialidases to process these terminal sialic acid residues. At present CBMs in family 40 are the only known examples Blank RgCBM40 SNA MUC2 Blank RgCBM40 SNA Muc2 RgCBM40 (no sialidase) RgCBM40 only RgCBM40 merged RgCBM40 and SNA RgCBM40 (sialidase) SNA (no sialidase) SNA (sialidase) RgCBM40 and SNA merged Muc2 R. gnavus and SNA R. gnavus ATCC29149 only Muc2 Primary antibody contol Primary antibody contol Secondary antibody contol Secondary antibody contol a b c d e NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02109-8 ARTICLE NATURE COMMUNICATIONS |8: 2196 |DOI: 10.1038/s41467-017-02109-8 |www.nature.com/naturecommunications 9