Main

The mucosal surface of the human large intestine is predominantly composed of gel-forming secreted mucins, and approximately 80% by dry weight of this glycoprotein is O-glycan1,2. Mucin O-glycans include only five different monosaccharides; thus, their considerable structural heterogeneity is attributed to linkage diversity between these monosaccharides and different sulfation patterns (Fig. 1). The mucosal layer is the barrier between the host epithelial cells and the dense community of microorganisms residing in the colon3. Mucin-degrading activities are a key aspect of the complex interactions between host and microorganism, and under healthy conditions, the production and breakdown of mucin is balanced4. However, disease states of the colon, which are on the rise, are typified by the mucosal layer being disrupted and changes in the proportion of mucinolytic bacteria4,5. For Akkermansia muciniphila, multiple studies have shown that its prevalence is inversely proportional to disease and inflammation markers6,7,8,9, but for the mucinophile Ruminococcus gnavus, the opposite is observed and the population increases in disease5. It is estimated that 60% of the human gut microbiota can access mucin as a nutrient source10,11, yet this process and its links to disease has yet to be comprehensively understood.

Fig. 1: Activity of the recombinantly expressed glycoside hydrolases from AM against α-linked monosaccharides.
figure 1

a, Structural features that are expected in natural secreted mucin glycoproteins with epitopes highlighted. Only α-linkages are labelled apart from the core GalNAc monosaccharides that are also α-linkages. The structure of an O-glycan chain is generally accepted to be categorized into three sections: (1) the core, consisting of an α-linked GalNAc attached to a serine or a threonine, (2) the polyLacNAc extensions linked to the core GalNAc and (3) the terminal ‘capping’ epitopes. The polyLacNAc extensions are generally LacNAc disaccharides linked through β1,3-bonds, but variable sulfation, fucosylation and branching add to the complexity and heterogeneity along these chains. Key ganglioside and globoside structures are also included. b, Volcano plot highlighting the differential gene expression of AM when grown with PGMIII compared with glucose. Genes that did not pass the threshold of significance (FDR (false discovery rate; Benjamini–Hochberg) < 0.05) are coloured grey. Genes that passed the threshold of significance (FDR < 0.05) but had a log2(fold change (FC)) between −1.5 and 1.5 are coloured blue. Genes that were significantly differentially expressed (FDR < 0.05) and had a log2(FC) <−1.5 or >1.5 are coloured red. Enzymes from this study that were significantly differentially expressed are labelled. The total number of genes in the analysis was 2,171. c, TLC of the whole-cell assay of PGMIII-grown AM against fresh PGMIII. A smear can be seen increasing in concentration over time. The glycans in these samples were then labelled with procainamide and analysed by LC–FLD–ESI–MS. The chromatogram for the 9 h sample is shown, and the different glycan peaks are labelled. The data for all the time points are in Supplementary Fig. 6. The data allow the reconstruction of the order of monosaccharides in an oligosaccharide but do not provide information about linkages. It is also not always possible to tell where a fucose or sulfate group is along the chain. Example mass spectra and MS–MS fragment data are presented in Extended Data Figs. 2 and 3. d, A cocktail of enzymes from A. muciniphila BAA-835 can completely degrade the O-glycans from PGMIII down to the core GalNAc. The top and bottom panels are a TLC and HPAEC-PAD of the results, respectively. PGMIII was incubated with a cocktail of enzymes (in the pale-blue box), and the reaction terminated by boiling (1). The result of this reaction was a large amount of monosaccharides that can be seen on the TLC and by HPAEC-PAD. This reaction was then dialysed to remove all free monosaccharides and glycans (2) and concentrated (3). Untreated PGMIII was also included as a control (4). The application of Amuc_1008GH31 is indicated by plus signs, and in samples 3 and 4, GalNAc can be seen to be released. Standards have also been included. Enzyme assays were carried out at pH 7 and 37 °C, overnight, and with 1 μM enzymes. Neu5Ac, N-acetylneuraminic acid; Neu5Gc, N-glycolylneuraminic acid; o/n, overnight.

Source data

Full size image

A. muciniphila is from the relatively understudied Planctomycetota–Verrucomicrobiota–Chlamydiota superphylum12, is detectable in most people, colonizes early on in life and typically constitutes 1–3% of the total microbiota13,14,15. Akkermansia-like sequences have also been detected in a wide variety of vertebrates, which suggests a long evolutionary history between the mucosal surface of the gastrointestinal tract in vertebrates and Akkermansia species13. A biochemical characterization of some of the carbohydrate-active enzymes (CAZymes) from A. muciniphila ATCC BAA-835 (AM) has been carried out in detail to provide insights into how AM tackles this complicated structure16,17,18,19,20. Furthermore, characterization of glycopeptidases from AM has also illuminated how mucins are broken down21,22,23. However, a systematic approach to understanding mucin degradation has not previously been undertaken for any bacterial species and enzymes from AM remain uncharacterized. Here we provide a comprehensive picture elucidating how AM sequentially degrades mucin O-glycans, related structures and other host carbohydrates.

Results

In vivo studies of AM with mucin

AM is highly restricted in terms of the substrates it can access, with its main carbon source being mucin. These observations were reproduced here (Extended Data Fig. 1). RNA-sequencing was performed on porcine gastric mucin III (PGMIII) to investigate the enzymes that AM uses to break down this nutrient source (Fig. 1). We found that 20 CAZymes (from glycoside hydrolase (GH) families 2, 16, 20, 27, 29, 36, 89, 95, 97, 105, 109 and 123) and 4 sulfatases encoded in the AM genome were upregulated. These results were complementary to similar studies published previously (Supplementary Fig. 1). A pangenome analysis was also undertaken to examine the prevalence of different CAZymes throughout strains and species (Supplementary Fig. 2 and Table 1). Strikingly, most enzymes were highly conserved between strains and close homologues were identified in other species.

Whole-cell assays were used to assess enzyme activities on the surface of AM (Fig. 1 and Supplementary Figs. 3–5). Results were initially assessed using thin-layer chromatography (TLC), and a smear, increasing in concentration over time, could be identified for PGMIII. This corresponded to the typical migration pattern for glycan fragments, rather than monosaccharides (Fig. 1, Extended Data Figs. 2–4 and Supplementary Figs. 3 and 4). Released glycan fragments were labelled with the fluorophore procainamide, and detailed characterization was performed using liquid chromatography–fluorescence detection–electrospray–mass spectrometry (LC–FLD–ESI–MS). The data reveal that a range of mucin O-glycan fragments are produced by the enzymes on the surface of AM (Fig. 1 and Extended Data Figs. 2–4). Galactose is typically present at the reducing end, indicative of GH16 endo-O-glycanase activity, and the degree of sulfation and fucosylation is variable18. We also used PGMIII-grown cells to carry out whole-cell assays against several defined oligosaccharides (Supplementary Figs. 3 and 4). Activities were observed against TriLacNAc, human milk oligosaccharides (HMOs), Forssman antigen (FORS1) and lacto-N-biose. Whole-cell assays of PGMIII-grown cells against bovine submaxillary mucin (BSM), which has core 1 decorations24, showed only the release of sialic acids, and this was confirmed by high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD; Extended Data Fig. 5).

AM CAZymes can get down to the core GalNAc in PGMIII

Two putative GH31 enzymes are encoded in the AM genome. Amuc_1008GH31 is from subfamily 18, and other members of this subfamily, from Bacteroides caccae, Phocaeicola plebius, Enterococcus faecalis, Clostridium perfringens and Bombyx mori (domestic silk moth), have specificity for removing α-linked N-acetylgalactosamine (GalNAc) from peptide25,26,27. A phylogenetic tree of characterized bacterial GH31 enzymes shows clustering of activities, and Amuc_1008GH31 clusters with the other characterized enzymes from GH31_18 (Supplementary Fig. 6). We also found that Amuc_1008GH31 had specificity for the core α-GalNAc linked to peptide (BSM) with no activity against any other substrates (Supplementary Figs. 7–10). Quantification of GalNAc release using HPAEC-PAD showed that Amuc_1008GH31 removed the highest concentration of all the enzymes tested here (Extended Data Fig. 5). Notably, Amuc_1008GH31 could not hydrolyse the Tn antigen and therefore requires more than one amino acid for its activity. For the more complex substrate PGMIII, Amuc_1008GH31 could not liberate GalNAc when tested in isolation; however, when a cocktail of AM enzymes was used, GalNAc could then be released (Fig. 1). This observation shows that enzymes characterized from AM are capable of complete O-glycan degradation down to the polypeptide. This enzyme is predicted to be periplasmic, and there was no obvious removal of GalNAc from BSM during whole-cell assays, supporting this prediction (Supplementary Fig. 5). A comparison of the Amuc_1008GH31 model and the E. faecalis GH31_18 is provided in Supplementary Discussion and Supplementary Fig. 11.

Hydrolysis of α-linked galactose from PGMIII O-glycans

Galactose, GalNAc, N-acetylglucosamine (GlcNAc), fucose and sialic acid all cap mucin O-glycans via α-linkages at the non-reducing end, and the prevalence of these monosaccharides will vary according to mucin type and genetics, for instance. Here we determined which enzymes AM uses to tackle these different capping monosaccharides by using a panel of different substrates of varying complexity in overnight end-point assays.

There are two putative GH110 enzymes (Amuc_0480GH110 and Amuc_1463GH110) encoded by the AM genome, which share 28% identity. These enzymes showed activity towards blood group B (BGB) types I and II and α1,3-linked galactose (Gal) to either Gal or GalNAc (Extended Data Fig. 6 and Supplementary Figs. 7–9). BGB is slightly less prevalent than blood group A (BGA; 10–40%) in the human population, ranging between 0 and 30%, depending on geographical location28. These are the only enzymes that are active against BGB from AM, which predicate the removal of fucose by the fucosidases. The structure of Amuc_1463GH110 has recently been solved20.

One putative GH27 (Amuc_1187GH27) is encoded in the AM genome and was active against α-linked galactose from the non-reducing end of most of the defined substrates tested here, except BG structures. Amuc_1187GH27 can hydrolyse Galα1,3-Gal/GalNAc and globotriose, but the P1 antigen could not be completely broken down in an end-point assay. This substrate specificity is discussed in the context of a model of Amuc_1187GH27 compared with solved structures of enzymes from Homo sapiens (Supplementary Discussion and Supplementary Fig. 11).

AM also has three putative GH36 enzymes encoded in the AM genome, which have low sequence homology (Supplementary Table 2), and all three cluster in different locations on a phylogenetic tree when compared with characterized GH36 enzymes (Supplementary Fig. 12). Amuc_0855GH36 has specificity for the Galili antigen and globotriose in the defined substrate screen. Finally, AM also has one putative GH97 enzyme (Amuc_1420GH97), which showed a preference for α-linked galactose in the defined substrate screen, except in the context of BGB. A model of Amuc_1420GH97 is discussed in the context of its activity (Supplementary Discussion and Supplementary Fig. 13). All these enzymes were able to remove galactose from PGMIII, albeit in relatively small amounts, but this is likely because of the type of mucin used. The identity of the galactose was confirmed by HPEAC (Fig. 2 and Supplementary Fig. 14).

Fig. 2: Activity of the recombinantly expressed glycoside hydrolases from AM against α-linked monosaccharides capping mucin.
figure 2

a, Activity of enzymes against PGMIII (pre-treated with Amuc_1835GH33 and Amuc_1120GH95) and analysed using HPAEC-PAD. Standards were also run to identify monosaccharide products. Top panel: relevant area of chromatograms. Bottom panel: enlarged image of the area where galactose elutes, showing details (dashed blue box in the top panel). b, Activity of enzymes releasing α-GlcNAc, α-GalNAc and α-fucose from GH16-released O-glycans from PGMIII. Top panel: full chromatograms with the two types of sialic acid highlighted in pink and purple. Bottom panel: enlarged image of the smaller peaks (highlighted by the dashed orange box in the top panel). Black diamonds and black stars indicate glycan number 10 and number 20 peaks, respectively. Highlighted in yellow and blue are the number 11 and number 12 peaks. The red asterisks indicate where a glycan is absent.

Source data

Full size image

Hydrolysis of α-linked GalNAc from PGMIII O-glycans

The AM enzymes with activity against α-GalNAc capping structures are from families GH36 and GH109. All four enzymes were active against BGA structures, and they must remove this sugar before the fucosidases can act. Amuc_0216GH36 was able to act on all the substrates with α-GalNAc decorations we could test (Extended Data Fig. 6 and Supplementary Figs. 7–9). Notably, this was the only enzyme we found in AM to degrade the Tn antigen. A phylogenetic tree of currently characterized bacterial GH36 enzymes, including those from the AM genome described here, showed that Amuc_0216GH36 clustered with the other two examples of α-GalNAcases in the family (Supplementary Fig. 12). One of these previously characterized enzymes also has activity on BGA and has been shown to convert BGA whole blood to universal-type blood for transfusion applications29. Building on this application, recent work has also shown that some of the AM CAZymes can be used to generate universal blood20. Amuc_0517GH36 was much narrower in its strict specificity for BGA only. There are two putative GH109 enzymes (Amuc_0017GH109 and Amuc_0920GH109) encoded by the AM genome, which have 66% identity between them and show specificity towards α-GalNAc substrates apart from the Tn antigen. When tested against PGMIII, both GH36 enzymes could release relatively large amounts of GalNAc, with or without sialidase and fucosidase pre-treatment, and this was confirmed using HPAEC-PAD (Fig. 2 and Supplementary Fig. 14).

Hydrolysis of α-linked GlcNAc from PGMIII O-glycans

There are two putative GH89 enzymes (Amuc_0060GH89 and Amuc_1220GH89) encoded by the AM genome, and both showed activity only against GlcNAcα1,4-Gal disaccharide (stomach epitope); however, Amuc_0060GH89 could not hydrolyse all the GlcNAcα1,4-Gal overnight, which suggests that this is not the preferred substrate (Extended Data Fig. 6 and Supplementary Figs. 7–9). Amuc_1220GH89 was able to remove relatively large amounts of GlcNAc from PGMIII, and this was confirmed using HPAEC-PAD (Fig. 2 and Supplementary Fig. 14). Further discussion on models of these enzymes compared with solved structures is presented in Supplementary Discussion and Supplementary Fig. 15.

Hydrolysis of sialic acid and fucose PGMIII O-glycans

AM has three sialidases and six fucosidases that together can tackle a wide range of substrates, and their specificities have been characterized in depth previously16,17. The specificities observed for these enzymes presented in this report are comparable to previous characterizations, but we were able to identify notable further observations (Extended Data Figs. 7 and 8 and Supplementary Fig. 16). In terms of the sialidases, we found that the two GH33 enzymes from AM can act on the relatively complex GD1a and GT1b ganglioside structures and Sda antigen, which are common features of the human glycome30,31. In terms of the fucosidases, Amuc_0010GH29 shows specificity towards type 2 BG structures (not type 1) and the lacto-N-neotetraose HMO series (not the lacto-N-tetraose), whereas Amuc_1120GH95 can hydrolyse fucose from both type 1 and 2 structures. This specificity facilitates the probing of mucin glycan structures in more detail. For instance, only Amuc_1120GH95 can hydrolyse fucose from untreated PGMIII, indicating that this substrate has only type 1 structures immediately available to the fucosidases at the non-reducing ends of the O-glycans (Extended Data Fig. 8).

Combining endo-acting and exo-acting CAZymes

To further explore the specificities of the CAZymes removing α-linked monosaccharides from PGMIII, we used a series of sequential reactions, which were then labelled with procainamide and analysed by LC–FLD–ESI–MS (Fig. 2 and Supplementary Fig. 17). Complementary to the relatively high monosaccharide release seen by HPAEC-PAD, the LC–FLD–ESI–MS data show that Amuc_1220GH89, Amuc_0216GH36 and Amuc_0517GH36 could act on GH16-derived O-glycan fragments (Fig. 2). These assays therefore provide the ability to characterize the fragments produced by GH16. For example, glycan number 11, which is absent in the Amuc_1220GH89 assay, has a GlcNAcα1,4-Gal capping structure. This also confirms that GH16 can accommodate this epitope in its active site. Amuc_0216GH36 and Amuc_0517GH36 both acted on glycans number 12 and 18, confirming that they have α-GalNAc caps. Furthermore, we could also observe four of the fucosidases acting on GH16-derived number 10 O-glycan fragments. The different peaks corresponding number 10 glycan compositions will be different combinations of linkages and positioning of the fucose. For instance, the structures hydrolysed by Amuc_1120GH95 are not hydrolysed by Amuc_0010GH29, thus confirming them as type 1 α1,2-fucose structures. In addition, Amuc_0392GH29 and Amuc_0846GH29 both hydrolyse a different glycan number 10 (eluting at a different time to the glycan number 10 acted on by Amuc_1120GH95) complementary to their comparable activities seen in the defined substrate screen.

Investigating β-galactosidase activity

The combination of GH16 endo-O-glycanase activity and removal of α-capping monosaccharides from the non-reducing ends lead us to how the remaining O-glycan fragments will be broken down. This will require contribution from β-galactosidases, β-HexNAcases, fucosidases and sulfatases. There are nine putative CAZymes encoded in the AM genome that were highlighted as possible β-galactosidases from families GH2, GH35 and GH43 subfamily 24 (Supplementary Table 3). Recombinant enzymes were screened against a variety of substrates to determine their specificity (Fig. 3 and Supplementary Figs. 18–24). The results revealed a range of specificities, but between them, the β-galactosidases could break down all the defined oligosaccharides tested. There are examples of very broad-acting (Amuc_0290GH2) and highly specific β-galactosidases (Amuc_0539GH2 was found to be specific to β1,4-linked Gal with either a Gal or GalNAc in the +1 position). None of these enzymes showed activity towards 3-fucosyllactose (3-FL), Lewis A- or Lewis X-based structures, which confirms that there is a strict order of degradation, with fucosidases acting on these substrates first, followed by galactosidases.

Fig. 3: Activity of the glycoside hydrolases from families 2, 35 and 43 from AM against β-linked galactose substrates.
figure 3

a, A heat map of recombinant enzyme activities against defined oligosaccharides. The dark blue and white indicate full and no activity, respectively, and partial activities are represented by the lighter blues. Partial activity is when all the substrate has not been broken down in an end-point assay. b, The activity of the panel of β-galactosidases against PGMIII that had been sequentially degraded. Control 1, no enzymes added; control 2, Amuc_1835GH33 and Amuc_1120GH95; control 3, control 2 plus Amuc_2108GH16. The top and bottom panels are the TLC and the LC–FLD–ESI–MS results, respectively. These assays were performed in stages, so the reactions were boiled in between steps. Enzyme assays were carried out at pH 7 and 37 °C, overnight and with 1 μM enzymes. Monosaccharide standards are shown in the right lanes. c, Activity of the GH2, GH35 and GH43 enzymes against lactosylceramide. The assays were analysed using HPAEC-PAD, and the different controls confirm the release of galactose. Left panel: the different controls and samples are stacked. Right panel: the assay chromatograms are overlaid for comparison so the galactose peaks can be observed in more detail. The substrate could not be resolved using this method, so whether the reaction had gone to completion could not be determined and TLC of the samples was also inconclusive. Enzyme assays were carried out at pH 7 and 37 °C, overnight and with 1 μM enzymes. LNT, lacto-N-tetraose; LNnH, lacto-N-neohexaose; LNFP II, lacto-N-fucopentaose II; Cer, ceramide.

Source data

Full size image

We tested this panel of β-galactosidases against PGMIII that had been sequentially treated with different AM CAZymes. It should be noted that, although with the addition of only fucosidase, sialidase and a GH16 a range of sizes and compositions of O-glycan fragments were produced, the majority of these were disaccharides and trisaccharides of alternating galactose and GlcNAc, ascertained by the intensity of the fluorescence signals being relatively high for these structures (Fig. 3 and Supplementary Fig. 22). When the panel of β-galactosidases was tested, the sample was pre-treated with the three enzymes that remove α-linked GlcNAc and GalNAc (Amuc_1220GH89, Amuc_0216GH36 and Amuc_0517GH36) and a fucosidase (Amuc_0392GH29) to maximize the O-glycans with β-galactose at the non-reducing ends. A TLC of these assays clearly shows galactose being released for five of the enzymes (Fig. 3, blue-stained bands). These samples were labelled with procainamide and analysed by LC–FLD–ESI–MS (Fig. 3 and Supplementary Fig. 22). Activities for four of the β-galactosidases were observed against the trisaccharides. Amuc_0539GH2 showed no activity, and this supports the observation of only being active when a Gal or GalNAc is in the +1 subsite. Notably, doubly sulfated trisaccharide glycan number 7 was not acted upon by any β-galactosidase. The GH2 and GH35 enzymes also showed activity against lactosylceramide (ganglioside and globoside core structure), with Amuc_0290GH2, Amuc_0771GH35 and Amuc_1686GH35 releasing the most galactose (Fig. 3).

Investigating β-HexNAc activities

The GH20 family is the largest represented family in AM, with 11 putative enzymes encoded with generally low identity between sequences (Supplementary Fig. 25 and Table 4). The activity predominantly observed in this family is exo-acting β-GlcNAcases, but the activity on GalNAc has also been observed. We also included Amuc_0052GH84, Amuc_0803GH123 and Amuc_2109GH3 in this screen, with the latter having been shown to have relevant activities (pNPGlcNAc>GalNAc)32 and clusters with other β-HexNAcases on a phylogenetic tree when characterized enzymes were compared (Supplementary Fig. 26). In summary, we found a range of different β-HexNAcase specificities encoded by the AM genome and the specificities of these enzymes are generally towards the glycan structures found in mucins rather than complex N-glycans or chitin. Between them, these β-HexNAcases could access all the defined substrates provided (Fig. 4 and Supplementary Figs. 27–29). The panel of β-HexNAcases were then tested against PGMIII that had been sequentially treated (controls 1–4; Supplementary Fig. 17). The samples were then labelled with procainamide and analysed by LC–FLD–ESI–MS. The predominant glycans in ‘control 4’ were two disaccharides, one sulfated on the galactose and one not (Fig. 4). Four of the β-HexNAcases were able to break down the sulfated disaccharide, providing further insight into the different specificities of these enzymes. To further explore the capacity of β-HexNAcases against host glycan structures, we used the AM CAZymes to prepare Sda antigen from the ganglioside GD1a. Intriguingly, out of the five β-HexNAcases that could act on GA2, only Amuc_1924GH20 was able to accommodate the branching sialic acid to access the GalNAc present in the Sda antigen in an end-point reaction (Fig. 4). In addition to the sequential degradation of mucin O-glycans, the activity of the AM CAZymes against HMOs was also explored and is detailed in Supplementary Discussion and Supplementary Figs. 30 and 31.

Fig. 4: Activity of β-HexNAcases from glycoside hydrolase families 20, 84, 123 and 4 from AM.
figure 4

a, Heat map of recombinant enzyme activities against defined oligosaccharides. The dark blue and white indicate full and no activity, respectively, and partial activities are represented by the lighter blues. Asterisks represent those substrates that were generated by pre-treating with other CAZymes. b, The activity of the panel of β-HexNAcases against PGMIII that had been sequentially degraded. Control 1, no enzymes added; control 2, Amuc_1835GH33 and Amuc_1120GH95; control 3, control 2 plus Amuc_2108GH16; and control 4, control 3 plus Amuc_0771GH35. The relevant area of the chromatograms of the LC–FLD–ESI–MS data are shown, and the two glycans are highlighted. Red asterisks indicate where a glycan is not present. These assays were performed in stages, so the reactions were boiled in between steps. Enzyme assays were carried out at pH 7 and 37 °C, overnight and with 1 μM enzymes. c, TLC results of sequential assays against ganglioside GD1a. GDa1 was sequentially treated with Amuc_1835GH33 and then Amuc_0771GH35 (lane B). The sample was then boiled and a panel of β-HexNAcases added.

Source data

Full size image

Tackling sulfated substrates

We highlighted 12 potential sulfatases from the AM genome using upregulation data4,33,34,35 and the SulfAtlas database36. These enzymes could be assigned to seven different families and one unknown, which ultimately turned out not to be a sulfatase (Supplementary Table 1). The different subfamilies cluster on a phylogenetic tree, and the activities (mainly explored in Bacteroides thetaiotaomicron) within a subfamily correspond to those already described (Supplementary Fig. 32)37,38. The sequence identity between the sulfatases is low, with the highest being ~60% for pairs in the same subfamily (Supplementary Table 5). The defined sulfated substrates tested were sulfated monosaccharides and Lewis structures (Extended Data Fig. 9 and Supplementary Figs. 33 and 34). We first screened the sulfatases against the sulfated monosaccharides and found that Amuc_1655Sulf16 and Amuc_1755Sulf16 showed activity against 4S-Gal; Amuc_1755Sulf16 also showed activity against 4S-GalNAc. The subfamily 11 enzymes Amuc_1033Sulf11 and Amuc_1074Sulf11 showed activity against 6S-GlcNAc, and Amuc_0121Sulf15 showed partial activities against 6S-Gal and 6S-GalNAc. These specificities align with those previously observed for these subfamilies37.

We then used combinations of the sulfatases, α-fucosidases and β-galactosidase encoded in the AM genome to determine how the different sulfated Lewis structures are broken down. For the 3′S and 6S-Lewis A and X glycans, Amuc_0392GH29 removes the fucose to produce 3′S and 6S-N-acetyllactosamine (LacNAc) and lacto-N-biose (LNB). From here, the 3′S disaccharides are then acted on by sulfatases from subfamily 20. By contrast, the 6S disaccharides are acted on by the β-galactosidases first. In this study, we could not find a way to break down the 6′S-Lewis structures. Previous activities against these substrates have been observed for BT1624 from subfamily 15 (ref. 37), so we had predicted that Amuc_0121Sulf15 may be able to do this, but only partial activity against sulfated galactose could be detected.

AM interactions with GAGs, N-glycoproteins and starch

There are two observations from this study worthy of note, concerning the ability of AM to colonize the mucosal surface and interact with other members of the microbiota. Firstly, AM did not grow directly on any of the glycosaminoglycans (GAGs) tested (Extended Data Fig. 1). However, its genome encodes Amuc_0778PL38 and Amuc_0863GH105 and activity against all GAGs except heparin sulfate was observed for these enzymes (Extended Data Fig. 10, Supplementary Figs. 35–37 and Supplementary Tables 6 and 7). Secondly, AM grew on a high-mannose N-glycoprotein, and by analysing substrates remaining in the broth, we deciphered that AM was using the protein component and leaving the glycan (Supplementary Fig. 38).

In total, we were unable to find activities for 11 of the putative CAZymes and 3 of the sulfatases during this study (Supplementary Table 1). Some of these CAZymes are from the GH13 and 77 families, which are currently solely associated with breaking down α-glucose polymers (Supplementary Fig. 39).

Discussion

The human gut microbiota plays an intrinsic role in health and disease, with impacts now recognized to extend to many other parts of the body. One of the key processes occurring at the interface between microorganisms and host is the breakdown of mucins, and this holistic investigation of AM CAZymes described here is an important step forwards in understanding this complex relationship. Previously, we characterized the GH16 endo-O-glycanases from AM (Amuc_0724GH16, Amuc_0875GH16, Amuc_2136GH20) that hydrolyse within the polyLacNAc chains of mucin to produce fragments of O-glycans18. Here we expand on this work by taking a systematic approach to assessing the full enzymology of this organism. The capacity of the AM CAZymes to break down most of the substrates tested reveals how AM has adapted to accessing mammalian-derived glycans. The results have allowed an order of enzyme activities on different substrates to be assembled (Fig. 5), a sequential breakdown of complex substrates to be demonstrated and the determination that a cocktail of AM CAZymes could be applied to reach the core GalNAcs of a complex mucin substrate. A model of how AM degrades mucin can now be proposed (Fig. 6).

Fig. 5: Illustration of the CAZymes characterized in this report against host glycans by AM.
figure 5

The different types of substrate are grouped where possible. The enzymes listed for each reaction are the ones that will work alone, but for the Lewis B structures, ‘and’ signifies that both enzymes are required to remove the fucose. For the sialylfucosyllacto-N-tetraose (SFLNT), the possible fucosidase activities were not included, but Amuc_0392GH29 can act on this substrate also. Only α-linkages are labelled apart from the core GalNAc monosaccharides that are also α-linkages.

Full size image
Fig. 6: A model for the degradation of mucins by A. muciniphila based on current understanding.
figure 6

The GH enzymes included are only where activity has been observed, and the colour indicates the type of activity (see key). The localization of all the enzymes is based on the SignalP 6.0 prediction, apart from Amuc_2108GH16 owing to it being observed in outer membrane analysis when AM was grown on mucin65, and our whole-cell assays support a GH16 being localized to the outside of the cell. The peptidases included are both those that have been characterized and those that have been highlighted as upregulated on mucin. The signal sequences of a peptidase and sulfatases potentially localize them to the outside of the cell also. The numbers indicate the order that mucin is degraded. 1: There is some processing of mucin on the surface, with both exo- and endo-GH activities. Large sections of mucin are then imported into the cell for further breakdown. 2: Initially, there will be further exo- and endo-GH activities to produce fragments of O-glycans and mucin polypeptide with only the core glycan decoration remaining. 3: The O-glycan fragments will then be broken down to monosaccharides through the alternating action of sulfatases, fucosidases, β-galactosidases and β-HexNAcases. 4: The remaining glycopeptides are known targets for characterized A. muciniphila glycopeptidases, and Amuc_1008GH31 will remove core GalNAc from the polypeptide. The stars indicate Mul1A association.

Full size image

A limitation of this work was the use of porcine gastric mucin rather than human mucin native to AM owing to lack of availability. Future work will aim to explore human-derived mucins with CAZymes from gut microorganisms as these substrates become available. Greater availability of mucin-type substrates will also allow a more thorough exploration of mucin-targeting enzymes and organisms, such as those described here. Another limitation of this work was the approach we took to analysing AM in isolation and away from a community of microorganisms and the host environment. Our approach is useful in providing clear enzymology, but neglects to understand the role and importance of these enzymes in the context of their native environment. However, due to mucin being the sole carbon source for AM, understanding the enzymology will be critical to understanding the complicated impact of AMs on health and disease39,40.

The glycomes of different humans vary considerably, and characterization of enzymes with specificity towards different structures will now enable research into these epitopes much more broadly. For instance, blood groups are a recognizable example, which varies globally41, but Galα1,4Gal epitopes of P1 and Pk antigens are also generally present in the mucosal surface of the human large intestine. These glycans, and others like them, are receptors for a variety of pathogens and their toxins, such as Shiga toxin42. FORS1 is normally present in only 0.01% of the human population, but in that relatively small population, this antigen would be present in the gastrointestinal tract43. It can change the susceptibility to some diseases and is also expressed in some cancers, but its true prevalence and impact on different diseases are understudied44. One of the major impacts of this work will be the ability to characterize different mucins in much more detail and in a higher-throughput manner. We hope that these enzymes provide new glycotools for the community and have far-reaching applications, such as monitoring the change in mucin O-glycans between healthy and diseased tissue or the detection of disease-associated epitopes.

Methods

Substrates

A list of substrates and their sources are listed in Supplementary Data 2. PGMIII (Sigma) was prepared by dissolving it overnight in sterile deionized water. This was too turbid for bacterial growth, so the precipitate was removed by centrifugation at 1,500 × g for 5 min in sterile 15 ml Falcon tubes. Polyglucuronic acid was prepared as described45.

Recombinant protein expression and purification

Recombinant plasmids were transformed into Tuner cells (Novagen) in Luria–Bertani broth containing 50 μg ml−1 or 100 μg ml−1 kanamycin or ampicillin, respectively. The plates were used to inoculate 1 l flasks (also including relevant antibiotics), and the cells were grown at 37 °C with shaking at 180 rpm to mid-exponential growth phase; the flasks were then cooled to 16 °C, and isopropyl β-d-thiogalactopyranoside was added to a final concentration of 0.2 mM. Recombinant His-tagged protein was purified from cell-free extracts using immobilized metal affinity chromatography (using Talon resin; Takara Bio). The buffer used during the purification process was 20 mM Tris–HCl and 100 mM NaCl, pH 8. Cell-free extracts (~30 ml) were passed through a 5 ml bed volume of resin. This was then washed with 50 ml of buffer and the protein eluted with 10 mM and then 100 mM imidazole. The purification is assessed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (precast 4–20% gradient from Bio-Rad and stained with Coomassie Brilliant Blue), but proteins typically elute in the 100 mM fraction. The proteins were then concentrated using centrifugal concentrators (Vivaspin, 10 kDa molecular weight cut off), the absorbance was determined at 280 nm (NanoDrop One, Thermo Scientific) and concentrations were calculated using extinction coefficients taken from ProtParam.

Recombinant enzyme assays

The concentrations of defined oligosaccharide substrates used in the assays are presented in Supplementary Table 2. The activities of the recombinant enzymes were typically assessed in 20 mM 4-morpholinepropanesulfonic acid (pH 7) at 37 °C and a final enzyme concentration of 1 μM. The volume varies according to how much was required for subsequent analysis. For the sulfatase assays, 5 mM of CaCl was included. PNGaseL (LZ-PNGaseL-50-KIT) was used where required.

TLC

For defined oligosaccharides, 3 μl of an assay containing 1 mM substrate was spotted onto silica plates (Sigma Z740230). For assays against glycoproteins, GAGs and starch substrates, this was increased to 9 μl, 6 μl and 9 μl, respectively. The plates were resolved in running buffer containing butanol, acetic acid and water. Different ratios were used for different substrates—2:1:1 (typically resolved once) was used for most of the defined oligosaccharides, and 1:1:1 (typically resolved twice) was typically used for substrates in which large products were expected (for example, PGMIII). All TLCs were stained using a diphenylamine–aniline–phosphoric acid stain46.

Growth of AM on a variety of substrate

AM was grown on chopped meat broth (CMB) as described in ref. 18. Overnight starter cultures were inoculated from a glycerol stock into 5 ml of CMB and grown in an anaerobic cabinet (Whitley A35 Workstation, Don Whitley Scientific). Growth curves were collected in 96-well plates using Cerillo Stratus plate readers. Then, 300 μl of 2× CMB (minus the monosaccharides) was mixed with 300 μl of substrate, and 200 μl was pipetted into 3 wells to provide replicates. Growth curves were also repeated with different starter cultures and on different days.

RNA-seq of AM

AM was grown overnight in CMB at 37 °C under anaerobic conditions. The culture was diluted 1:20 into CMB minus the monosaccharides with the addition of either glucose or mucin and grown to an optical density (OD) of 0.5 at 600 nm. Cells were pelleted by centrifugation at 16,000 × g for 30 s and flash-frozen using a dry ice and ethanol bath. Sample processing, library construction and sequencing were performed by Azenta. Sequencing adapter removal and read quality trimming were performed with fastp v.0.23.2 using default parameters47. Reads were mapped to the AM (GCF_000020225.1) and enumerated using kallisto v.0.46.2 (ref. 48). Differential gene expression analyses were performed with Voom/limma v.3.40.6 (ref. 49) implemented in Degust v.4.2-dev (https://zenodo.org/records/3501067). The volcano plot was created using the package EnhancedVolcano v.1.22.0 (https://github.com/kevinblighe/EnhancedVolcano) in R v.4.4.0.

Whole-cell assays

Overnight starter cultures were inoculated from a glycerol stock into 5 ml of CMB and grown under anaerobic conditions. This was then used to inoculate a 5 ml culture of AM in CMB (minus the monosaccharides) with substrate; the cells were pooled, collected at the mid-exponential growth phase and washed twice with PBS; and 250 μl of 2× PBS was used per 5 ml culture to resuspend the cells. The assays were then mixed at a 50:50 cells-to-substrate ratio, in which substrate concentrations were the same as used in the recombinant enzyme assays. PGMIII, BSM, hyaluronic acid, chondroitin sulfate, RNaseB and Saccharomyces cerevisiae (Sc) mannan assays were 500 μl in total, and 50 μl samples were taken for each time point. Assays with defined oligosaccharides and HMOs were 200 μl, and 20 μl samples were taken for each time point. Assays were carried out at 37 °C and reactions were terminated by boiling samples.

HPAEC-PAD

To analyse monosaccharide release, sugars were separated using a CarboPac PA-1 anion exchange column with a PA-1 guard using a Dionex ICS-6000 (Thermo Fisher) and detected using PAD. Flow was 1 ml min−1 and elution conditions were 0–25 min 5 mM NaOH and then 25–40 min 5–100 mM NaOH. The software was the Chromeleon Chromatography Data System. Monosaccharide standards were 0.1 mM; all data were obtained by diluting assays 1/10 before injection, apart from the ceramide samples, which were diluted 15 μl in 100 μl.

Steady-state kinetics of AmPL38

Initial velocities of purified recombinant Amuc_0778PL38 were quantified on hyaluronic acid (HA), chondroitin sulfate A (CSA), chondroitin sulfate C (CSC) and dermatan sulfate (DS) substrates, with concentrations ranging from 0.25 to 12 g l−1 at 37 °C, 100 mM NaCl and UB4 buffer at pH 7 (ref. 50). The average initial velocities, quantified in milli-absorbance units at A235nm per second, were converted to µM of product generated by measuring the amount of Δ4,5 bonds formed per second, using the experimentally confirmed extinction coefficient for unsaturated glucuronic acid of 6,150 M−1 cm−1 (ref. 45). Kinetic parameters were determined by plotting initial velocities against substrate concentrations and fitting the Michaelis–Menten model using GraphPad Prism.

Initial rates of AmGH105

The steady-state reactions of Amuc_0778PL38 were sealed to prevent evaporation and incubated overnight at 37 °C. Subsequently, Amuc_0863GH105 was added to a final concentration of 1 µM, and the decrease in A235nm was monitored for a minimum of 2 h at 37 °C. The substrate concentration was calculated by converting the initial absorbance of the Amuc_0778PL38 steady-state reactions, minus absorbance backgrounds, to µM double bonds. The initial rates of Amuc_0863GH105 were then calculated as the loss of double bonds in µM per minute in absolute values.

LC–MS of GAG substrates

Duplicate time-course reactions for Amuc_0778PL38 were prepared under the same conditions as for the kinetics at 2 g l−1 substrate concentrations. Reactions were terminated by heating the samples at 95 °C for 5 min. Amuc_0863GH105 was added to the 20 h reaction to a final concentration of 1 µM and left to run for 2 h before heat inactivation. The final sample preparation and liquid chromatography–mass spectrometry (LC–MS) analysis were carried out using an ion trap coupled with GlycanPac chromatography as previously described51. Compounds were observed as single- or double-charge m/z, primarily as deprotonated adducts. The compounds were identified by MS and MS2 fragmentation, if possible. Extracted ion chromatograms of identified compounds were prepared, and the areas of the peaks were used to quantify the products (Supplementary Fig. 36). Fragments follow the nomenclature of ref. 52.

Glycan labelling

Released O-glycans were fluorescently labelled by reductive amination with procainamide as described previously using the LudgerTagTM Procainamide Glycan Labelling Kit (LT-KPROC-96)53. Briefly, samples in 10 μl of pure water were incubated for 60 min at 65 °C with procainamide labelling solution. Residual chemicals were removed from the procainamide-labelled samples using LudgerClean S-cartridges (LC-S-A6). The purified procainamide-labelled O-glycans were eluted with pure water (1 ml). The samples were dried by vacuum centrifugation and resuspended in pure water (50 μl) for further analysis.

LC–FLD–ESI–MS

Procainamide-labelled samples were analysed by LC–FLD–ESI–MS using an ACQUITY UPLC BEH-Glycan 1.7 µm 2.1 × 150 mm column at 40 °C on a Thermo Scientific UltiMate 3000 UPLC instrument with a fluorescence detector (λex = 310 nm, λem = 370). MS: Gradient conditions were 0–10 min, 15% A at a flow rate of 0.4 ml min−1; 10–95 min, 15–43% A at a flow rate of 0.4 ml min−1; 95–98 min, 43–90% A at 0.4–0.2 ml min−1; 98–99 min, 90% A at 0.2 ml min−1; 99–100 min, 90–15% A at 0.2 ml min−1; 100–103 min, 15% A at 0.2 ml min−1; and 103–115 min, 15% A at 0.2–0.4 ml min−1. Solvent A was 50 mM ammonium formate, pH 4.4, made from Ludger Stock Buffer (LS-N-BUFFX40); solvent B was acetonitrile (Acetonitrile 190 far UV/gradient quality; Romil number H049). Samples were injected in 15% aqueous solution and 85% acetonitrile, with an injection volume of 20 µl. The UPLC system was coupled on-line to an AmaZon Speed ETD electrospray mass spectrometer (Bruker Daltonics) with the following settings: source temperature, 180 °C; gas flow, 10 l min−1; capillary voltage, 4,500 V; ion-charge control (ICC) target, 200,000; maximum accumulation time, 50 ms; rolling average, 2; number of precursor ions selected, 3, release after 0.2 min; positive-ion mode; scan mode, enhanced resolution; and mass range scanned, 400–1,700. A glucose homopolymer ladder labelled with procainamide (Ludger; CPROC-GHP-30) was used as a system suitability standard as well as an external calibration standard for glucose units (GU) allocation. ESI–MS and MS–MS data were analysed using Bruker Compass DataAnalysis V4.1 software and GlycoWorkbench software54.

Bioinformatics

Putative signal sequences were identified using SignalP 6.0 (ref. 55). Identities between different sequences were determined using Clustal Omega using full sequences56. The CAZy database (http://www.cazy.org) was used as the main reference for CAZymes57. Phylogenetic trees were completed in SeaView58 and final trees were produced in the Interactive Tree of Life59.

Pangenome analysis

To analyse the CAZymes across different species and strains, CAZyme sequences were downloaded from UNIPROT using the get.seq function from the Bio3d package in RStudio60. Sequence selection was based on the accession numbers of the CAZymes from the most up-to-date version of the CAZy repository at the time. Sequence txt files were converted to a FASTA format using the TabulartoFasta function from the following GitHub repository: https://github.com/lrjoshi/FastaTabular. All multiple sequence alignment and phylogenetic analysis was performed using Clustal Omega56. Percentage identity matrices were downloaded directly from the programme and merged with the original file from CAZy according to shared accession numbers. The presence or absence of different enzymes and how well conserved they were across strains and species were curated by hand.

Structural comparisons

Where possible, protein models were retrieved from the AlphaFold Protein Structure Database61,62. Where the models had not been built, this was completed using AlphaFold2 Colab63. Previously published protein structures were retrieved from the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank. Different structures or models were superimposed using COOT64, and PyMOL was used to look at structures and models and generate figures. Glycan models were built using the Carbohydrate Builder on http://glycam.org. ChemDraw 18.1 was used to draw carbohydrate structures.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.