The human gut Firmicute Roseburia intestinalis is a primary degrader of dietary β-mannans

1The human gastrointestinal tract harbors an extremely dense and diverse microbial community, known as the gut microbiota. In a mutually beneficial relationship, the gut microbiota supplies enzymes able to depolymerize dietary carbohydrates that cannot be hydrolyzed by human enzymes,. The monosaccharides generated are further fermented into host-absorbable metabolites, including the short-chain fatty acids butyrate, acetate, and propionate. In particular, butyrate produced by commensal bacteria serves as the main energy source for colonocytes, and it exhibits anti-carcinogenic, anti-inflammatory, and barrier protective properties in the distal gut–. The relevance of this metabolic output to human health has prompted increasing interest in intentionally modulating the composition of the gut microbiota to promote wellbeing and combat disease, e.g., by the use of prebiotics. Established prebiotics have been traditionally developed based on their selective fermentation by bifidobacteria and lactobacilli generally regarded as conferring health benefits to the host. Notably, other potentially beneficial targets are the butyrate-producing Firmicutes.

2Roseburia spp., together with Faecalibacterium prausnitzii and Eubacterium rectale, constitute a group of dominant butyrate-producing Firmicutes, estimated to account for 7–24% of the total bacteria in the healthy human colon,. Interest in Roseburia spp. has increased with reports that the abundance of these bacteria is reduced in individuals affected by inflammatory diseases– and colorectal cancer,. Complementary studies have shown that Roseburia spp. play an important role in the control of gut inflammatory processes, amelioration of atherosclerosis and in the maturation of the immune system, primarily through the production of butyrate. R. intestinalis preferentially colonizes the mucin layer, and this intimate association to the host may contribute to the local level of butyrate available for the colonic epithelial cells. This species appears to be a specialist able to grow only on a few glycans, and has been recently shown to be a prominent xylan degrader in vitro and in the healthy human colon.

3β-Mannans are widespread in the human diet: they are widely used in food as thickening, stabilizing, and gelling agents (glucomannan and galactomannan, Fig. 1). They are found in the endospermic tissue of nuts (homopolymeric mannan), coffee beans, coconut palm, tomato, and legume seeds (galactomannan) (Fig. 1)–, and play vital roles in the cell wall structure and as storage polysaccharides in plants. Notably, the structure of galactoglucomannan from non-edible sources (softwood) shares striking similarities with that from edible sources (Fig. 1).

4Prevalent Gram-negative Bacteroides spp. encode β-mannan polysaccharide utilization loci (PULs) and have been recently shown to utilize mannans–. Despite members of the Firmicutes phylum being numerically dominant in the gut, insight is lacking into the metabolic strategies adopted by these Gram-positive bacteria to utilize β-mannans.

5Here, using a combination of microbiology,”omic” and enzymology approaches, we unravel the molecular mechanism evolved by R. intestinalis L1–82 to depolymerize β-mannans that are potentially available in the large intestine. Our findings show that R. intestinalis growth on β-mannan is dependent on the expression of a highly specific multi-modular cell attached endomannanase, an ATP-binding transporter and an intracellular enzyme cocktail through which linear and substituted manno-oligosaccharides are completely hydrolyzed to component monosaccharides for further metabolism. Using germ-free mice colonized with a model gut microbiota, we demonstrate that β-mannan alters the community composition, facilitating bacteria that have mannan degrading machineries. Besides extending the knowledge on the enzymatic basis of β-mannan-metabolism by members of the most numerous Firmicutes phylum, our results have implications for the design of targeted intervention strategies to manipulate the gut microbiota via supplementation of prebiotics to the diet to restore or improve health.

6R. intestinalis L1–82 grows efficiently on a variety of complex β-mannans as a sole carbon source (Fig. 2a), causing a concomitant acidification of the medium (Fig. 2b). To evaluate which fractions of β-mannan breakdown products are internalized, we analyzed the culture supernatants during R. intestinalis growth on AcGGM using high-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) (Supplementary Fig. 1a, b). Neither oligosaccharides nor monosaccharides accumulated in the stationary phase culture (Supplementary Fig. 1a, b), indicating that the bacterium possesses a highly efficient apparatus to cleave and import all the sugars derived from the breakdown of this complex glycan.

7To examine the molecular basis underlying β-mannan utilization by R. intestinalis, we performed an RNA sequencing (RNAseq) transcriptional analysis during growth on konjac glucomannan (KGM), spruce acetylated galactoglucomannan (AcGGM) and glucose (Glc). The top 20 upregulated genes in β-mannan transcriptome encode a β-mannanase belonging to the glycoside hydrolase (GH) 26 family (GH26 according to the CAZy classification), a solute binding protein (MnBP) and two permeases (MPP) of an ABC transporter, two phosphorylases (GH130), one epimerase (Mep), two β-glucosidases (GH3) and two carbohydrate esterases (CEs) (Fig. 2c and Supplementary Data 1). These genes are located in two loci, which were designated mannan-utilization locus large (MULL: ROSINTL182_05469–83) and mannan-utilization locus small (MULS: ROSINTL182_07683–85) (Fig. 2d). Among the MULL genes expression of a LacI-type transcriptional regulator, predicted glycosyl hydrolases belonging to GH113, GH36, GH1, and a phosphomutase also increased. The response was specific to β-mannan as no differential expression of these genes was observed during growth of R. intestinalis on galactose, a building block in mannan (Supplementary Table 1).

8Proteomic analysis under the same growth conditions corroborated the RNAseq results; indeed, proteins encoded by the genes located in MULL and MULS were abundant in the AcGGM samples compared to the glucose samples (Fig. 2e, Supplementary Data 2).

9We carried out a comparative genomic analysis to establish the distribution of β-mannans utilization loci equivalent to the identified MULL and MULS in other representative Roseburia spp. and Clostridium cluster XIVa members. The results showed that R. faecis and R. hominis shared an overall MULL and MULS organization with that of R. intestinalis (Supplementary Fig. 2, Supplementary Table 2), suggesting that the utilization of β-mannan is shared by these three Roseburia spp. However, the lack of the critical GH26 endomannanase, required to break down mannan (see later results for R. intestinalis β-mannanase RiGH26), is likely to render R. hominis only able to metabolize manno-oligosaccharides. Orthologous mannan utilization loci were identified in specific members of the Clostridium cluster XIVa, although a similar organization and complete conservation of all MULL and MULS components was not observed (Supplementary Fig. 2).

10RiGH26, (locus tag: ROSINTL182_07683), is a predicted extracellular modular β-mannanase comprising a carbohydrate binding module of family 27 (CBM27), a catalytic module of GH26 followed by a CBM23 (Supplementary Fig. 3a). Furthermore, two C-terminal Ig-like domains and a proline-glycine rich region likely mediate cell attachment and binding within the cell wall. The extracellular localization of RiGH26 was corroborated experimentally by immunofluorescence microscopy (Fig. 3). RiGH26 exhibited activity toward decorated mannans including KGM, carob galactomannan (CGM) and AcGGM (Fig. 4a and Supplementary Fig. 3b), generating linear and substituted manno-oligosaccharides. The enzyme was active on mannopentaose (M5) and mannotetraose (M4) but not mannobiose (M2) (Supplementary Fig. 3c). Overall, the product profiles suggest capability of endo-action and indicates that RiGH26 targets large polymers and can accommodate the galactose and acetyl decorations present in these substrates. Further analysis indicated that RiGH26 is a potent enzyme as, when used at the concentration of 10 nM, it was able to hydrolyze high concentrations of spruce AcGGM (50 mg ml−1) into oligosaccharides in 1 h at standard assay conditions (Supplementary Fig. 3d). No detectable activity was measured when RiGH26 was incubated with linear cello-oligosaccharides, birch xylan, curdlan, lichenan or barley derived β-glucan, thus confirming the specificity of RiGH26 towards β-mannan (Supplementary Fig. 3c).

11BlastP searches showed that homologs of RiGH26, including the two predicted carbohydrate binding modules CBM27 and CBM23, were exclusively found in β-mannanase encoded by Firmicutes belonging to various other members of the Clostridium cluster XIVa (Supplementary Fig. 4, Supplementary Table 3–5). To investigate the biochemical properties of the two modules, RiCBM27 and RiCBM23 were expressed in Escherichia coli and their capacities to bind to a range of different soluble cello-oligosaccharides and manno-oligosaccharides were evaluated using surface plasmon resonance (SPR). Recombinant RiCBM27 and RiCBM23 bound only manno-oligosaccharides (Table 1), but differed in their binding profiles. Similar to a previously described GH26-associated CBM27, RiCBM27 preferred mannohexaose (M6) (Kd = 165 ± 10 µM, two independent experiments, ± indicates standard deviation), (Table 1, Supplementary Fig. 5a) and its affinity dropped for ligands smaller than a tretrasaccharide (Table 1). By contrast, RiCBM23 was selective for shorter oligosaccharides with its highest affinity for M4 (Kd = 130 ± 50 µM, two independent experiments) (Table 1, Supplementary Fig. 5b), although mannotriose (M3) was also bound with good affinity (Table 1).

12Within the MULL cluster, the three genes (ROSINTL182_05477 – ROSINTL182_05479) that encode an ATP-binding cassette (ABC) importer were shown to exhibit the highest level of increased expression during growth on β-mannan (and when compared to growth on glucose). The thermodynamic binding parameters of the ABC-transporter associated solute binding protein, RiMnBP, to linear and substituted manno-oligosaccharides were determined using isothermal titration calorimetry (ITC). RiMnBP bound a range of unsubstituted manno-oligosaccharide with a preference for M5 (Kd of 2.55 µM) followed by M3 and M4 (Table 2, Supplementary Fig. 6). Acetylations had a marginal effect on the binding affinities, thus providing evidence that these fragments are efficiently captured by the transport protein. Overall, these results support the predicted role of RiMnBP in the uptake of manno-oligosaccharides generated by RiGH26, showing optimal affinity for undecorated or acetyl substituted ligands with a degree of polymerization (DP) of 4−5.

13The affinity of the solute binding protein RiMnBP to manno-oligosaccharides and the predicted intracellular location of the debranching and exo-acting enzymes is consistent with a hierarchical degradation of the internalized manno-oligosaccharides following extracellular degradation of the β-mannan polymers by RiGH26.

14The ROSINTL182_05471 (RiCEX) and ROSINTL182_05473 (RiCE2) gene products possess SGNH hydrolase-type esterase domain signatures. Comparison to previously characterized CEs revealed that RiCE2 showed 25–30% identity to a CE2 from Clostridium thermocellum and the acetyl xylan esterase Axe2C of Cellvibrio japonicus. In contrast, RiCEX did not display significant relatedness to other characterized CE catalogued in the CAZy database, which excluded RiCEX from being classified in any of the 16 CE families. RiCEX and RiCE2 showed mannan acetyl esterase activity on a mixture of oligosaccharides generated via RiGH26 hydrolysis of AcGGM (Fig. 4b). RiCE2 partially removed acetyl groups from the acetylated oligosaccharide substrate (Fig. 4b). RiCEX deacetylated the substrate mainly to free and monoacetylated oligosaccharides (Fig. 4b). These results indicate that RiCEX has a preference for oligosaccharides with a degree of acetyl substitution ≥2, but is less efficient on mono-substituted substrates. At the same time, it suggests that an acetyl group present at a specific position (O-2 or O-3) is resistant to enzymatic deacetylation by RiCEX. The combination of RiCEX and RiCE2 resulted in the almost complete deacetylation of the manno-oligosaccharides, indicating a cooperative interaction of the two esterases (Fig. 4b).

15RiGH36 released galactose from internally substituted CGM and AcGGM after the treatment with the RiGH26 β-mannanase (Fig. 4c and Supplementary Fig. 7). Interestingly, RiGH36 released galactose from CGM-endomannanase products with 100% efficiency (Fig. 4c, Supplementary Fig. 8a) as no oxidized product could be observed after treatment of these samples with galactose oxidase. The enzyme exhibited limited activity on large polymers (Supplementary Fig. 8b) consistent with the activity on internalized oligosaccharides in vivo. Similarly, α-galactosidase activity increased after de-acetylation of the oligosaccharides (Supplementary Fig. 8c, d). Beside cleaving single internal galactose residues from manno-oligosaccharides, this enzyme was capable of removing α-1,6-galactose from the reducing-end of a substituted manno-oligosaccharide (Supplementary Fig. 8e) and from an oligosaccharide containing two consecutive substitutions (Supplementary Fig. 8f). Corroborating these results, RiGH36 cleaved galactose decorations from endomannanase products of highly substituted guar gum galactomannan (Supplementary Fig. 8b).

16Sequence searches showed that the protein encoded by ROSINTL182_05483 (MULL, RiGH113) exhibited 40% identity to the only characterized enzyme from this family, the endo-β-mannanase AaManA from Alicyclobacillus acidocaldarius (Supplementary Fig. 9a). Alignment of RiGH113 and AaManA showed that the catalytic and substrate interacting residues are conserved (Supplementary Fig. 9a). When RiGH113 was assayed for activity on linear manno-oligosaccharides, it revealed an ability to cleave linear manno-oligosaccharides to yield mannose and M2 (Supplementary Fig. 9b). Strikingly, time-course analysis of RiGH113 activity revealed that this enzyme displays a different sub-specificity by hydrolyzing manno-oligosaccharides to mannose and a minor amount of M2 (Supplementary Fig. 9c). After overnight incubation with RiGH113, M2 was partially degraded to mannose (Supplementary Fig. 9d), confirming the exo-mannosidase activity as opposed to the endo-fashion cleavage reported for the AaManA. The substituted manno-oligosaccharides galactosylmannobiose (Gal1Man2) and digalactosylmannopentaose (Gal2Man5) were hydrolyzed to a lesser extent than non-substituted substrates (Supplementary Fig. 9e); no activity could be detected on Gal1Man2 while Gal2Man5 was hydrolyzed to yield mannose and digalactosylmannotetraose (Gal2Man4), which was resistant to further hydrolysis. When the reducing end of manno-oligosaccharides was blocked (Supplementary Fig. 10a–d), no RiGH113 activity could be detected demonstrating that this enzyme possesses a previously unknown reducing end mannose-releasing exo-oligomannosidase activity. Consistent with the view that RiGH113 is an intracellular enzyme, release of mannose was detected after incubation of the enzyme with RiGH26-generated CGM-oligosaccharides (Fig. 4d). The closest homologs of this enzyme are encoded by Clostridium cluster XIVa strains and a range of Firmicutes (Supplementary Fig. 10e).

17Product analysis of end point assays and a time course study revealed that both RiGH3A (ROSINTL182_07684) and RiGH3B (ROSINTL182_07685) were β-glucosidases, with redundancy in structure (Supplementary Fig. 11a, b), active on linear cello-oligosaccharides (Fig. 4e). RiGH3B completely hydrolyzed cellotetraose (G4) and cellopentaose (G5) into glucose monomers, whereas RiGH3A released glucose and a range of oligosaccharides with lower efficiency compared to that of RiGH3B (Supplementary Fig. 11c, d). Neither of these enzymes were active on manno-oligosaccharides (Supplementary Fig. 11e, f). While RiGH3B was able to hydrolyze glucosylmannose (G1M1) and, partially, mannosylglucose (M1G1) into monomers (Supplementary Fig. 11f), RiGH3A displayed activity only towards G1M1. No activity was detected on polymeric KGM (Supplementary Fig. 12a), while glucose was released after incubation of both RiGH3A and RiGH3B with RiGH26-generated KGM–hydrolysate (Supplementary Fig. 12b). Importantly, the latter results demonstrate that RiGH26 can accept a glucose moiety at the subsite +1, generating oligosaccharides with a glucose residue at the non-reducing end.

18Recombinant RiGH130_2 (MULL, ROSINTL182_05474) phosphorolyzed M4 into M3, M2 and mannose-1-phosphate (M1P) while M3 was processed to M2 and M1P (Fig. 4f). The enzyme was inactive on cello-oligosaccharides (Supplementary Fig. 13). RiGH130_2 was functional only in the presence of inorganic phosphate, confirming that RiGH130_2 is a mannosyl-phosphorylase.

19The concerted action of the MULL and MULS encoded enzymes described above on the oligosaccharides generated by RiGH26, suggest an intracellular accumulation of M2. Hydrolysis of this product into monosaccharides is accomplished through the action of two enzymes encoded by the co-transcribed genes ROSINTL182_05476 (RiMep) and ROSINTL182_05475 (RiGH130_1).

20RiMep was active on M2 and cellobiose (G2), releasing M1G1 and G1M1, respectively (Fig. 5a). These data show that RiMep is an enzyme active on the reducing end sugar and catalyzes the conversion of disaccharide substrates to the corresponding C2 epimer. This enzyme exhibited epimerization activity not only for the substrate but also for the product as, under high enzyme amount and long reaction time, it was able to convert M1G1 and G1M1 to M2 and G2, respectively (Supplementary Fig. 14a). In addition, RiMep exhibited epimerization activity towards oligosaccharides with a DP > 2 but not on monosaccharides (Supplementary Fig. 14b).

21ROSINTL182_05475 encodes a specific mannosylglucose phosphorylase belonging to the GH130 subfamily 1. RiGH130_1 was inactive on G1M1 and oligosaccharides with a DP ≥ 2 (Supplementary Fig. 15). RiGH130_1 displayed activity only towards M1G1 in the presence of inorganic phosphate, releasing glucose and M1P (Fig. 5b, c).

22RiPgm catalyzes the interconversion of M1P and mannose 6-phosphate (M6P) (Fig. 5d). In addition, the enzyme displayed activity also against D-glucose 1-phosphate (G1P) yielding D-glucose 6-phosphate (G6P) (Supplementary Fig. 16a), thus indicating that RiPgm is a phosphomannomutase (PMM)/phosphoglucomutase (PGM) which can use either glucose or mannose as substrate. Consistent with the presence of a predicted magnesium-binding loop (Supplementary Fig. 16b), the RiPgm-mediated catalytic activity was detected only when MgCl2 was present in the reaction.

23ROSINTL182_05469/70 encodes a predicted bi-functional protein consisting of an N-terminal glucosidase domain (RiGH1_D1, aa 1–246) and a C-terminal family GH1 isomerase domain (RiGH1_D2, aa 247–768). RiGH1_D1 shares 44% identity to the previously characterized β-glucosidase TmGH1 from Thermotoga maritima. The recombinant RiGH1_D1 displayed no catalytic activity against all the tested substrates, including G4, M4, M5, M6P, G6P, and fructose 6-phosphate (F6P). Thus, RiGH1_D1 functional significance is currently unclear. RiGH1_D2 is a phosphomannose isomerase catalyzing the interconversion of M6P into F6P (Fig. 5e).

24The ability of R. intestinalis to capture, breakdown β-mannan and efficiently internalize manno-oligosaccharides may increase its fitness to compete with other resident β-mannan degraders, including the glycan generalist Bacteroides. To test this hypothesis, we performed in vitro co-cultivation of R. intestinalis and the efficient β-mannan degrader Bacteroides ovatus ATCC 8483. Both bacteria showed similar growth rates in individual cultures supplemented with AcGGM (Fig. 6a). Population estimates using qPCR indicated that, in the mixed cultures, both B. ovatus and R. intestinalis grew well during the exponential growth phase, suggesting that the bacteria shared the available carbon source and maintained coexistence. (Fig. 6b). During the stationary phase, when glycan availability is limited, the mean relative abundance of R. intestinalis and B. ovatus in the culture was approximately 72.5% versus 27.5%, respectively. In contrast, R. intestinalis showed slow growth on mannose (Fig. 6c) and was outcompeted when co-cultured in this carbon source with B. ovatus (Fig. 6d).

25To elucidate whether dietary supplementation of β-mannan can result in expansion of key gut bacteria able to utilize this hemicellulose, germfree mice were colonized with a synthetic microbiota composed of 14 sequenced strains of human commensal gut bacteria. Colonized mice were fed a high-fiber diet for 14 days before being switched to a series of diet regimes with a varying amount of AcGGM (Fig. 6e). Overall, the levels of four species (R. intestinalis, Bacteroides uniformis, B. ovatus and Marvinbryantia formatexigens) gradually increased at both AcGGM doses (Fig. 6f–i) and these strains were able to suppress the bacteria foraging on the glyco-protein rich mucus layer (Akkermansia muciniphila, Bacteroides caccae, Bacteroides thetaiotamicron, Bacteroides intestinihominis) (Fig. 6e and j–m) and the amino acids degraders (E. coli, Clostridium symbiosum and Collinsella aerofaciens) (Fig. 6e and n). Upon feeding of a fiber-deficient diet, the fecal bacterial abundance of the mucin-eroding bacteria, the sulfate-reducer Desulfovibrio piger and the three amino acid degraders (Fig. 6e) rapidly increased with a corresponding decline of the fiber-degrading species.

26β-Mannans are widely present in the human diet as constituents of hemicellulose in beans, some nuts and food additives, but are recalcitrant to intestinal digestion by host enzymes. These glycans instead serve as a carbon source for mannanolytic bacteria in the distal gastrointestinal tract, primarily Firmicutes and Bacteroidetes. Recent studies have characterized a few enzymes encoded by two polysaccharide utilization loci (PULs) implicated in the metabolism of galactomannan in B. ovatus, and homopolymeric mannan in Bacteroides fragilis. To date, a full understanding of β-mannan utilization by Firmicutes, however, remains underexplored. The human gut butyrate-producing Firmicute R. intestinalis has previously been shown to utilize galactomannan and glucomannan as a carbon source and possesses predicted genes for the utilization of these substrates. However, no data are available relating the mannanolytic activity at a biochemical level. In this study, we show that two conserved loci, MULL and MULS, collectively provide R. intestinalis the capacity to depolymerize this plant polysaccharide. Detailed biochemical studies of the encoded enzymes allowed us to construct a model of sequential action for the mannan utilization system encoded by MULL-MULS (Fig. 7). The RiGH26 and the mannan ABC uptake system components RiMnBP/RiMPP1/RiMPP2 transcripts and proteins were the most upregulated in both the RNA sequencing and proteomic analyses, respectively (Fig. 2c, e). This highlights the crucial role of this endomannanase and the ABC transport system in the β-mannan metabolic pathway. RiGH26 is the only enzyme displayed on the cell surface (Fig. 3), allowing direct access to the intact β-mannan polymers through dynamic capture mediated by two appended carbohydrate binding modules (RiCBMs). The SPR data showed that RiCBM23 displays ~7- and 5-fold higher affinity for M3 and M4, respectively, than RiCBM27, suggesting that the two CBMs play different roles to mediate binding of RiGH26 to mannans. The RiCBMs’ Kd values for the preferred manno-oligosaccharides were in the 100−200 µM range (Table 1). This moderate affinity to the bound substrate constitutes an advantage as it has lower impact on the catalytic activity compared to canonical counterparts from other organisms, and suggests an evolutionary adaptation of R. intestinalis to compete with other microbial enzymes with canonical higher- affinity CBMs, but with reduced catalytic rates. Reliance on multi-modular cell-wall anchored enzymes is a common feature in Firmicutes; consistently, RiGH26 organization was primarily found in β-mannanase from other Roseburia species and members of the Clostridium cluster XIVa (Supplementary Fig. 4, Supplementary Table 3–5). Multiplicity of CBMs provides a contrast with the system for mannan metabolism in Bacteroides ovatus,, where the binding and catalytic activity are distributed between two surface located binding proteins and the single domain mannanase BoMan26B.

27Collectively, our results point to a model in which the smaller manno-oligosaccharides generated by RiGH26 are imported through a dedicated β-mannan transport system consisting of RiMnBP/RiMPP1/RiMPP2 (Fig. 7). In the cytoplasm, acetylated and galactosylated manno-oligosaccharides are systematically debranched by RiCE2, RiCEX and RiGH36, and subsequently depolymerized. Removal of glucose units from glucomannan-oligosaccharides is carried out by RiGH3A and RiGH3B. Based on the highest transcriptional and protein regulation, the main depolymerization strategy for breakdown of unsubstituted manno-oligosaccharides is mediated by the activity of two synergistic mannoside phosphorylases (RiGH130_2 and RiGH130_1) and an epimerase (RiMep), similar to the mannan catabolic pathway proposed in the ruminal bacterium Ruminococcus albus. A similar system has been reported in B. fragilis and B. ovatus, although only composed of an epimerase and a mannosylglucose phosphorylase (GH130_1) that, together, process GH26s-generated M2 units. The presence of the manno-oligosaccharide phosphorylases RiGH130_2 allows R. intestinalis to process undecorated manno-oligosaccharide of DP > 2, consistent with the internalization of large manno-oligosaccharides generated by RiGH26-hydrolysis of polymeric mannan. However, GH130_2s mainly catalyze the phosphorolysis of undecorated manno-oligosaccharides. Removal of mannose units from substituted substrates is mediated by the reducing end mannose-releasing exo-oligomannosidase RiGH113, which displays a previously undescribed specificity. The two different approaches based on the phosphorylases and the GH113 are likely to be a functional adaptation to accelerate the depolymerization process of mannan. Eventually, mannan catabolism fuels monosaccharide fermentation via glycolysis and leads to the production of butyrate, which is the primary energy source for host colonocytes,. Colonocytes oxidize butyrate to carbon dioxide, thereby keeping the epithelium hypoxic (<1% O2). This condition promotes gut homeostasis by stabilizing the hypoxia-inducible transcription HIF that coordinates barrier protection in the mucosa,. Recently, it has been shown that antibiotic-mediated depletion of butyrate-producing Clostridia increases colonocytes oxygenation and drives aerobic pathogen expansion in the gut lumen, resulting in Salmonella enterica-induced gastroenteritis. Importantly, R. intestinalis has been found to affect host histone epigenetic states, direct colonic epithelial cells metabolism away from glycolysis and towards fatty acid metabolism, reduce the levels of inflammatory markers and ameliorate atheriosclerosis in a diet-dependent fashion. The athero-protective effect was in part attributed to butyrate, as this SCFA has been shown to inhibit key inflammatory pathways involved in cardiovascular disease development.

28The absence of oligosaccharides from R. intestinalis AcGGM-spent supernatant (Supplementary Fig. 1a, b) demonstrates that the β-mannan degradation apparatus is optimized for efficient uptake of all the products released by RiGH26, maximize intracellular breakdown and avoid nutrient leakage. This will limit the access to other bacteria, such as Bacteroides spp., competing for the same resource. Using AcGGM, we have shown that R. intestinalis and B. ovatus, which possesses an equally complex β-mannan degrading system, shared the available resources and maintained coexistence (Fig. 6b). Notably, R. intestinalis outcompeted B. ovatus in the late exponential and stationary phase of growth; these results show that R. intestinalis is capable to bind and import the remaining β-mannan breakdown products (preferred by the RiMnBP transport protein) more efficiently than B. ovatus. Thus, it is likely that the β-mannan utilization apparatus provides R. intestinalis with a selective advantage during nutrient limitation when microbial competition for the available carbohydrates in the gut is intense. Understanding the mechanism by which β-mannan is degraded by key commensal members of the gut is crucial to designing intervention strategies through the use of targeted prebiotics which aim to program or reprogram the composition of the microbiota to maximize human health. Our in vivo study demonstrates that a diet supplemented with AcGGM can be used to manipulate the gut microbiota and to facilitate the growth of species equipped with a β-mannan degrading system, including R. intestinalis (Fig. 6e). This is supported by the increase in the relative abundance of R. intestinalis, B. uniformis and B. ovatus, which all possess enzymes able to degrade AcGGM (BACUNI_00371 - BACUNI_00383; BACOVA_02087–02097 and BACOVA_03386–03406 respectively). R. intestinalis was highly responsive to the AcGGM within a day, with a 10 to 30 fold increase at the 2.5% and 7.5% AcGGM diet, corroborating its ability to respond dynamically to variation in this dietary fiber. Intriguingly, R. intestinalis’ response did not last over the 7 day feeding treatment and the acetogen M. formatexigens seemed to replace it. A cluster of genes with predicted functions in β-manno-oligosaccharide utilization (BRYFOR_07194- BRYFOR_07206) was identified in the genome of M. formatexigens (Supplementary Fig.17a). The results shown in Supplementary Fig. 17b,d suggest that R. intestinalis and M. formatexigens occupy different metabolic niches in the distal gut; the former consumes complex β-mannans, whereas the acetogen feasts on mono- and oligosaccharides. When in co-culture with either R. intestinalis or B. ovatus, M. formatexigens was outcompeted in vitro (Supplementary Fig. 17e-f). A previous study with gnotobiotic mice bi-associated with the prominent saccharolytic bacterium B. thetaiotamicron and M. formatexigens showed that the presence of M. formatexigens caused a decrease in the cecal levels of B. thetaiotaomicron, compared with mono-associated controls. Transcriptional and metabolic analyses demonstrated that M. formatexigens is capable of consuming a variety of plant-derived oligosaccharides and microbial and host-derived N-glycans (such as N-acetylglucosamine), suggesting that this ability could confer a fitness advantage when competing with the glycan-consuming Bacteroides. Thus, it is likely that, when present as part of the synthetic microbial community described in this paper, M. formatexigens may be indirectly benefiting of either manno-oligosaccharides feeding/cross-feeding with other microorganisms or by its ability to grow mixotrophically, simultaneously utilizing organic carbon sources and formate or H2 for energy. Notably, M. formatexigens outcompeted B. ovatus at the 7.5% AcGGM diet, underscoring the competitiveness of this acetogen in a community setting. In the context of a complex microbial community, it is likely that M. formatexigens makes an important contribution to host nutrition improving fermentation by acting as a formate or H2 sink and by generating acetate as main metabolic product.

29Collectively, diet-induced changes involved the promotion of mannanolytic bacteria producing propionate, acetate and butyrate, metabolites that are known to regulate hepatic lipid, glucose homeostasis and health of the intestinal hepithelium. These SCFA-producers gained a competitive advantage over colonic mucin-degrading bacteria. Given that intermittent dietary fiber-deprivation results in a thinner mucus layer in mice, eventually enhancing pathogen susceptibility, our results support the concept that β-mannan-based interventions not only could contribute to preventing mucus barrier dysfunctions but also maintaining a gut environment that keeps pathogenic bacteria away. If confirmed in humans, these findings may help to prevent diseases affecting the integrity of the colonic mucus layer, such as ulcerative colitis. Indeed, the fact that the β-mannan degradation pathway is a core trait found in the majority of the human gut microbiota highlights the relevance of potential therapeutic interventions through the use of β-mannan formulations to the general population.

30Carbohydrate substrates used in this study are listed in Supplementary Table 6. All glycan stocks were prepared at 10 mg ml−1 in ddH2O and sterilized by filtration using a 0.22 µm membrane filter (Sarstedt AG & Co, Germany).

31Unless otherwise stated, R. intestinalis L1–82 was routinely grown at 37 °C without agitation in an anaerobic cabinet (Whitley A95 workstation, Don Whitley, UK) under an 85% N2/10% H2/ 5% CO2 atmosphere. Growth experiments were carried out in YCFA medium (YCFA—Yeast extract-Casein hydrolysate-Fatty Acids) supplemented with 0.5% (w/v) of the specific carbohydrate to be examined. Overnight cultures (300 µl) were used to inoculate 30 ml aliquots of YCFA plus the carbohydrate to be tested. These pre-cultures were passaged at least three times on the same media to ensure cell growth adaptation on a single carbon source prior to inoculation of the final cultures for growth experiments, RNA-sequencing and proteomic analysis. Bacterial growth was determined spectrophotochemically by monitoring changes in the optical density at 600 nm (OD600). In addition, growth on turbid substrates was assessed by measuring differences in pH compared to that of starting medium. Growth and pH curves are averages of three biological replicates, with two technical replicates each. Routine culturing of Bacteroides ovatus ATCC 8483 and M. formatexigens DSM 14469 was in anaerobic Chopped Meat Medium under static conditions at 37 °C.

32R. intestinalis was cultured in triplicate on YCFA supplemented with 0.5% (w/v) glucose, galactose, KGM or AcGGM as described above. Cells were harvested at mid-exponential phase and RNA was extracted using the RNeasy Mini Kit (Quiagen) according to the manufacturer’s instructions. RNA-seq libraries were prepared using the ScriptSeq Complete kit from Epicentre. Samples were paired-end sequenced on an Illumina Hiseq 4000 instrument at Beijing Genomics Institute (BGI). Analysis of the RNA-seq results was performed exactly as described in. Differential gene expression analysis was performed with the DeSeq2 package.

33The genes encoding mature forms of the proteins described in this study were amplified from the R. intestinalis L1–82 genomic DNA (BioProject accession number PRJNA30005 [https://www.ncbi.nlm.nih.gov/bioproject/PRJNA30005]) by PCR, using appropriate primers (Supplementary Table 7). PCR products were generated using the Q5 High-Fidelity DNA Polymerase (New England BioLabs, United Kingdom) with 50 ng DNA as template. Prior to cloning the DNA fragment encoding RiGH1_D2 (ROSINTL182_05469), sequence ambiguities at the 3’-end of ROSINTL182_05470 were corrected through sequencing the PCR product generated with the primers listed in the Supplementary Table 8. The gene ROSINTL182_07683 was synthesized without the N-terminal signal sequence predicted by SignalP v.4.1 (residues 1–27 from transcription start). The PCR amplicons were cloned into the pNIC-CH expression vector by ligation-independent cloning (LIC). The gene encoding RiMnbp (ROSINTL182_05479) was cloned in the vector pETM-11 following the method described elsewhere. Recombinant proteins generally contained a C-terminal His6-tag, although, in some cases, His-tag translation was prevented by the introduction of one or two stop codons at the end of the open-reading frame (RiMep, RiGH36, RiPgm and RiGH113). The His6-tag was excluded to prevent interaction with putative C-terminal active or catalytic residues that could be detrimental to the enzymes’ activity. Constructs were verified by sequencing (Eurofins, UK). Proteins were produced in E. coli BL21 Star (DE3) cells (Invitrogen) as previously described. Briefly, cells were cultured to mid-exponential phase in Tryptone Yeast extract (TYG) containing 50 mg ml−1 kanamycin at 25 °C. Protein overexpression was induced by adding isopropyl β-D-thiogalactopyranoside (IPTG) to a final concentration of 200 μM, followed by incubation for a further 16 h at 25 °C. Cells were harvested by centrifugation, sonicated and recombinant proteins were purified by either immobilized metal ion affinity chromatography (IMAC) or hydrophobic interaction chromatography (HIC). For IMAC purification, the clarified cell lysate was loaded onto 5 ml HisTrap HP Ni Sepharose columns (GE Healthcare) connected to an ÄKTA purifier FPLC system (GE Healthcare). Protein elution was achieved by using a linear gradient from 5 to 500 mM imidazole. RiGH113, RiGH36, RiMep and RiPgm were purified by HIC by loading the cell-free broth, adjusted to buffer A (1.5 M ammonium sulfate), onto a 5 ml HiTrap Phenyl FF (GE Healthcare) equilibrated with the same buffer. Protein was eluted by using a linear reverse gradient to 100 mM NaCl over 90 min at a flow rate of 2.5 ml min−1. After IMAC and HIC, samples were concentrated and further purified by size exclusion chromatography (SEC) using a HiLoad 16/60 Superdex G75 size exclusion column (GE Healthcare) and a running buffer consisting of 20 mM Tris-HCl pH 8.0 with 200 mM NaCl. Fractions containing the pure protein were combined, concentrated and buffer exchanged to 20 mM Tris pH 8.0, using a Vivaspin 20 (10-kDa molecular weight cutoff) centrifugal concentrators (Sartorius Stedim Biotech GmbH, Germany). Protein purity was estimated to be over 95% for all the enzymes using SDS-PAGE. Protein concentrations were determined using the Bradford assay (Bio-Rad, Germany).

34Enzyme assays, unless otherwise stated, were carried out in 10 mM sodium phosphate buffer, pH 5.8, for up to 16 h at 37 °C and 700 rpm. Reactions with RiGH130_1 and RiGH130_2 were prepared in 100 mM sodium phosphate buffer, pH 5.8. The activity of RiPgm against M1P and G1P was tested in 10 mM sodium phosphate buffer, pH 5.8, supplemented with 1 mM MgCl2. To determine the specificity of RiGH113, the recombinant protein was sequentially incubated with 0.1 mg ml−1 pre- reduced or oxidized manno-oligosaccharides at 37 °C overnight. Reduction of manno-oligosaccharides was conducted by incubating 1 mg ml−1 manno-oligosaccharides in a volume of 75 μl with sodium borodeuteride (NaBD4; 0.5 M in 100 mM NaOH) solution. The reaction was incubated overnight at room temperature then quenched by adding 25 µl of 25 mM sodium acetate. Oxidation of manno-oligosaccharides reducing-end was obtained by incubating the substrates (1 mg ml−1) with the Neurospora crassa cellobiose dehydrogenase (NcCDH) overnight at 37 °C. Both NaBD4 and NcCDH pretreated samples were diluted 10X in standard assay buffer before addition of RiGH113. Between three and five independent experiments were performed to determine the enzyme activities.

35Reaction products generated by the enzymes used in this study were analyzed by matrix-assisted laser desorption/ionization time of flight mass spectrometry (MALDI-TOF MS) as described previously. Briefly, 2 µl of a matrix, consisting of 9% 2,5-dihydroxybenzoic acid (DHB) in 30% acetonitrile, were applied to an MTP 384 ground steel target plate TF (Bruker Daltonics, Germany). Sample (1 µl) was then mixed with the matrix and dried under a stream of warm air. Samples were analyzed with an Ultraflex MALDI-ToF/ToF instrument (Bruker Daltonics, Germany), equipped with a Nitrogen 337 nm laser beam and operated in positive acquisition mode. Results were analyzed using the Bruker FlexAnalysis software (version 3.3).

36Mono- and oligosaccharides products were analyzed on a Dionex ICS-3000 HPAEC system operated by the Chromeleon software version 7 (Dionex, Thermo Scientific), as described previously. Sugars were injected onto a CarboPac PA1 2 × 250-mm analytical column (Dionex, Thermo Scientific) coupled to a CarboPac PA1 2 × 50-mm guard column kept at 30 °C. Manno-oligosaccharides and phosphorylated monosaccharides were eluted in 0.1 M NaOH at a flow rate of 0.25 ml min−1 by increasing the concentration of sodium acetate (NaOAc) exponentially from 0 to 0.3 M over 26 min (from 9 to 35 min after injection), before column reconditioning by 0.1 M NaOH for 10 min. Commercial manno-oligosaccharides with DP2−6 were used as standards. For cello-oligosaccharides, the separation was done using a multistep linear gradient going from 0.1 M NaOH to 0.1 M NaOH–0.1 M NaOAc over 10 min, 0.1 M NaOH–0.14 M NaOAc after 14 min, 0.1 M NaOH–0.3 M NaOAc at 16 min followed by a 2 min exponential gradient to 1 M NaOAc, before reconditioning with 0.1 M NaOH for 9 min. Cello–oligosaccharides with DP 2–6 were used as standards. For the analysis of disaccharides (G1M1 or M1G1) and phosphorylated monosaccharides generated from the activity of RiGH130_2, RiMep, RiGH130_1, RiPgm and RiGH1, the elution was done at 0.25 ml min−1 using a 40 min program. The program started with 0.01 M potassium hydroxide (KOH) for 15 min, reaching the concentration of 0.1 M KOH at 25 min after injection and was kept for additional 5 min at the same KOH concentration. Between each sample, the column was re-equilibrated by running initial conditions for 10 min.

37Proteins of interest were detected using anti-sera raised in rats (Eurogentec) against the corresponding recombinant RiGH26 or the previously characterized RiXyn10A.

38For immunofluorescence microscopy, R. intestinalis cells were grown in YCFA containing 0.5% AcGGM, wheat arabinoxylan (WAX) or glucose to an OD600 of 0.8, collected by centrifugation (4000 × g for 5 min) and washed twice in phosphate buffered saline (PBS). Cells were resuspended in 500 µl PBS and fixed by adding an equal volume of 2 × formalin (9% formaldehyde in PBS) on ice for 30 min. The bacterial pellet was washed twice with 1 ml PBS prior to resuspension in 1 ml of blocking buffer (1% bovine serum albumin, BSA, in PBS) and incubation at 4 °C for 16 h. After incubation the cell pellets were harvested by centrifugation and the supernatant discarded. For labelling, the bacteria were incubated with 0.5 ml of anti-sera (diluted 1:500 in blocking buffer) for 2 h at 25 °C. The cells were then pelleted, washed with 1 ml PBS and resuspended in 0.5 ml goat anti-rat IgG Alexa-Fluor 488 (Sigma-Aldrich), diluted 1:500 in blocking buffer and incubated 1 h at 25 °C. The cells were again harvested, washed with 1 ml PBS and suspended in 100 µl PBS containing one drop of ProLong Gold antifade reagent (Life Technologies). Labeled bacterial cells were mounted onto glass slides and secured with coverslips. Fluorescence microscopy was performed on a Zeiss AxioObserver equipped with the ZEN Blue software. Images were acquired using an ORCA-Flash4.0 V2 Digital CMOS camera (Hamamatsu Photonics) through a 100x phase-contrast objective. A HXP 120 Illuminator (Zeiss) was used as a fluorescence light source.

39R. intestinalis was grown in triplicate on YCFA supplemented with either 0.5% (w/v) glucose or AcGGM, respectively, as a sole carbon source. Samples (10 ml) were harvested at the mid-exponential growth phase. Cell pellet was collected by centrifugation (4500 × g, 10 min, 4 °C), resuspended in 50 mM Tris-HCl, 0.1% (v/v) Triton X-100, 200 mM NaCl, 1 mM dithiothreitol and disrupted by bead-beating using three 60 s cycles with a FastPrep24 (MP Biomedicals, CA). Proteins were precipitated with ice-cold trichloroacetic acid (TCA), final concentration of 10% (v/v), incubated on ice for 1 h, centrifuged (15,000 × g, 15 min, 4 °C) to pellet the precipitated proteins and washed with 300 μl ice-cold 0.01 M HCl in 90% acetone. Proteins were separated by SDS-PAGE with a 10% Mini-PROTEAN gel (Bio-Rad Laboratories, CA) and then stained with Coomassie brilliant blue R250. The gel was cut into five slices, after which proteins were reduced, alkylated, and in-gel digested according to a method published previously. The peptides were dried under vacuum, solubilized in 0.1% (v/v) trifluoroacetic acid (TFA) and desalted using C18 ZipTips (Merck Millipore, Germany) according to the manufacturer’s instructions.

40The peptide mixture from each fraction was analyzed using a nanoHPLC-MS/MS system as described previously, using a Q-Exactive hybrid quadrupole-orbitrap mass spectrometer (Thermo Scientific) equipped with a nano-electrospray ion source. Mass spectral data were acquired using Xcalibur (v.2.2 SP1).

41MS raw files were processed with the MaxQuant software suite (version 1.4.1.2) for identification and label-free quantification (LFQ) of proteins. Proteins were identified by searching MS and MS/MS data of peptides against the UniProtKB complete proteome of R. intestinalis L1–82 (4698 sequences) supplemented with common contaminants (e.g., keratins, trypsin, and bovine serum albumin). In addition, reversed sequences of all protein entries were concatenated to the database for estimation of false-discovery rates (FDRs). Trypsin was set as proteolytic enzyme and two missed cleavages were allowed. Protein N-terminal acetylation, oxidation of methionines, deamidation of asparagines and glutamines and formation of pyro-glutamic acid at N-terminal glutamines were defined as variable modifications while carbamidomethylation of cysteines was used as a fixed modification. The “match between runs” feature of MaxQuant, which enables identification transfer only between samples from the same carbon source based on accurate mass and retention time, was applied with default parameters. All identifications were filtered in order to achieve a protein FDR of 1%. A protein was considered “present” if it was detected in at least two of the three biological replicates in at least one glycan substrate. Missing values were imputed from a normal distribution (width of 0.3 and down shifted 1.8 standard deviations from the original distribution) in total matrix mode and differential abundance analysis was performed using an unpaired two-tailed Student’s t-test with a permutation-based FDR set to 0.05. Hierarchical clustering and heat map representations were generated using Euclidean distance measure and average linkage using Perseus (version 1.5.5.3).

42The affinity of RiCBM27 and RiCBM23 to soluble manno-oligosaccharides and cello-oligosaccharides was evaluated by SPR using a Biacore T100 (GE Healthcare). The two CBMs, diluted into 10 mM sodium acetate (pH 4.1) to 2.3 μM, were immobilized on a NTA sensor chip (GE Healthcare) to a density of 3000−4000 response units (RU). Sensograms were recorded at 25 °C in phosphate/citrate buffer (20 mM phosphate/citrate buffer; 150 mM NaCl; pH 6.5, 0.005% (v/v) P20 surfactant) at 30 μl per min with association and dissociation times of 90 s and 240 s, respectively. CBMs binding was tested towards 0.2 nM – 1 mM of carbohydrate ligands dissolved in the same buffer as above. Data were analyzed using the Biacore T100 evaluation software, and equilibrium dissociation constants (Kd) were obtained by fitting a single-site binding model to either the steady-state response data or the full sensograms.

43Binding of manno-oligosaccharides to RiMnBP was measured at 25 °C in 10 mM sodium phosphate pH 6.5 using an ITC200 microcalorimeter (MicroCal). RiMnBP in the sample cell was titrated by 19 injections of carbohydrate ligand separated by 120 s. The following concentrations were used: 900 µM of M3 in the syringe and 76.5 µM RiMnBP in the sample cell; 1365 µM of M4 or M5 in the syringe and 91 µM RiMnBP in the sample cell; 2270 µM of M6 in the syringe and 117 µM RiMnBP in the sample cell; 750 µM of diacetylated mannotetraose (M4Ac2) in the syringe and 50 µM RiMnBP in the cell; 1500 µM of diacetylated mannopentaose (M5Ac2) in the syringe and 100 µM RiMnBP in the cell. Thermodynamic binding parameters were determined using the MicroCal Origin software (version 7.0).

44R. intestinalis, B. ovatus and M. formatexigens cells were grown overnight under anaerobic conditions in YCFA containing 0.5% (w/v) AcGGM (YCFA-AcGGM) as the sole carbon source. These subcultures were used to inoculate, in approximately equal proportions (estimated by OD600), 30 ml of the same medium. A control culture of YCFA-AcGGM was also inoculated with either R. intestinalis, B. ovatus or M. formatexigens. Growth (OD600) was monitored for up to 24 h, withdrawing 1 ml samples for quantitative PCR (qPCR) analysis at selected time points. Cells were pelleted, combined with 200 μl of TE buffer (pH 7.8) and bead-beated for 2 min (FastPrep96, MP Biomedicals, CA) using ≤ 106 μm acid-washed glass beads (Sigma-Aldrich). Genomic DNA was extracted using the Mag Midi kit (LGC Group, UK) according to the manufacturer’s instructions. qPCR was performed in a LightCycler 480 II system (Roche, Germany) using specific primers for each strain (Supplementary Table 9). In addition, a high-resolution melting (HRM) analysis was conducted to evaluate the specificity of the amplification and the lack of primer dimers. The raw data were imported into the LinReg PCR program and the calculated Cq values and PCR efficiency were used to deduce the ratio of R. intestinalis, B. ovatus and M. formatexigens at each time point. Statistically significant differences were determined using the unpaired two-tailed Student’s t-test.

45All experiments involving animals complied with all relevant ethical regulations for animal testing and research and were approved by the University of Michigan, University Committee for the Use and Care of Animals. Germfree mice were colonized with a synthetic microbiota composed of 14 fully sequenced human species according to the methodology previously adopted by Desai et al. Briefly, seven 6-week-old germfree male wild-type Swiss Webster mice that had been raised on ad libitum access to a high fiber chow diet (LabDiet 5013) and autoclaved distilled water were gavaged for 3 consecutive days with 200 µl each day of a mixture of the 14 different species. Colonized mice were maintained on this high fiber diet for 14 days before being switched to a series of diet regimes with varying fibers. This feeding sequence consisted of 7 days of feeding on a gamma-irradiated fiber-free (FF) diet (TD.140343, Harlan Teklad, USA) that does not contain AcGGM or related molecules. Mice were then switched for 7 days to a custom version of the same diet that contained AcGGM at 2.5% w/w, followed by a 7-day washout period on the FF diet, and finally 7 days of feeding on a version of this diet containing AcGGM at 7.5% w/w (in both AcGGM diets an equivalent amount of glucose was removed to accommodate the prebiotic addition). Fecal samples were taken 1 day before and 1 day after each diet transition, effectively allowing us to measure changes in response to AcGGM supplementation at 1 and 7 days post exposure relative to the FF diet. The relative abundance of each microbial strain at sampled time points was measured by qPCR, using previously validated primer sets, from total DNA extracted from freshly voided fecal pellets (stored at −20 °C until extraction) exactly as described previously. Statistically significant differences were determined using the unpaired two-tailed Student’s t-test.

46Identification of similar β-mannan catabolic genes in bacteria belonging to the Clostridium XIVa cluster was performed using the Gene Ortholog Neighborhood viewer on the Integrated Microbial Genomes website (https://img.jgi.doe.gov). This was done using the genes encoding RiGH26 (ROSINTL182_07683, GenBank ABYJ02000124.1:7167–11129) and RiMEP (ROSINTL182_05476, GenBank ABYJ02000025.1:3200–4429) as the search homolog and the default threshold e-value of 1e-5. Then, a sequence comparison was conducted where each R. intestinalis L1–82 RefSeq annotated protein sequence was subjected to BLASTp searches against other Clostridium XIVa members. Sequences with coverage <60% and amino acid similarity <45% were excluded.

47Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.