Fecal metaproteome of defined gnotobiotic model.
Using a shotgun LFQ approach, we profiled the fecal metaproteomes of germ-free mice (GF), and mice colonised with either 12 bacterial species (B - bacteria), five fungal species (Y - yeast), both bacteria and fungi (BY), and the latter group treated with an antibiotic (BY_ABX) or antifungal (BY_AFX) (Figure 1A). We enriched microbial cells from feces of four-week-old mice by differential centrifugation, digested cell lysates by trypsin, and analysed the resulting peptide mixtures using LC-MS/MS. We queried the acquired mass spectra (5.3 x 106 MS/MS) against a combined protein database of the 17 microbial species and mouse (Table S1). The proteomic search identified 70,190 unique peptide sequences (Table S3), mapped to 6,675 proteins (Table S4). Of these, 68% were bacterial, 22% fungal, and 10% were mouse proteins (Figure 1B). Up to 58% of all detected proteins were characterised by LFQ intensities using the MaxLFQ method (34). Principal component analysis based on the protein relative levels identified three main clusters (Figure 1C), suggesting similar protein profiles of Y and GF, B and BY, and BY mice groups treated with antimicrobials (BY_ABX and BY_AFX).
From the bacterial proteomes, 4.4% of detected proteins shared peptides with other protein groups (Table S5). However, most of the identified peptides belonged to a single protein group (leading protein group); that is, the sequence coverage of peptides unique for the leading protein group was considerably higher than the sequence coverage of peptides belonging also to alternative protein groups. For the fungal proteomes, a high percentage of identified peptides belonged to multiple homologous proteins from different strains of the same species (e.g., Candida albicans strains SC5314, WO-1, CD36) (Tables S5). This was due to the use of UniProtKB reference protein databases for specific fungal species (Table S1), which included several strains of the same species. The use of reference databases was necessary because the fungal strains used for the gnotobiotic mice colonisation were clinical isolates without sequenced genomes, contrary to the bacterial strains, which have their genome sequence determined (20). The shortcomings of using not perfectly matched protein databases were most apparent for two fungal strains (Pichia kudriavzevii and C. krusei), which were recently re-classified as strains of the same species (40) and did not have available individual databases. For these fungi, we used a common database based on reference P. kudriavzevii strains available in UniProtKB, resulting in decreased specificity of the MS data searches. Also, horizontal gene transfer might explain some of the homologous proteins; however, these events have been speculated to be relatively infrequent among microbial eukaryotes (41).
Using genome-specific protein databases for the bacterial stains also had the advantage of accurate assignment of proteins for the 12 different bacterial strains, for which we did not observe any species cross identifications. Nevertheless, the general UniProtKB databases did not compromise the accuracy of fungal and mouse protein identification, and we observed minimal cross-kingdom identifications (e.g., same peptides mapped to a fungal and mouse protein), with these proteins being manually filtered out.
Inter-kingdom interactions influence the proteome response of gut bacterial species to antimicrobials.
From 12 bacterial strains used for the gnotobiotic mice colonisation, Clostridium clostridioforme YL32 had the most detected proteins (Figure 1B), while Akkermansia muciniphila YL44 had the highest predicted proteome coverage (33%) (Table S5). The latter result correlated with 16S rRNA amplicon sequencing data (18), which determined A. muciniphila YL44 as the most abundant bacterial species. The number of proteins detected per condition was relatively similar for each species (Figure S5).
From 2,860 quantified bacterial proteins, we identified 1,317 proteins whose levels significantly varied between the mice groups (ANOVA followed by THSD post-hoc analysis, FDR 5%, see Table S6 for all proteins used for statistical analyses). For four bacterial species with the largest number of detected proteins (C. clostridioforme YL32, A. muciniphila YL44, Blautia coccoides YL58, Muribaculum intestinale YL27), about 50% of the quantified proteins showed differential levels (Table S5), prompting in-depth evaluation of the strains’ protein profiles.
The proteomes of these bacterial species displayed an array of responses to antimicrobial-induced ecosystem perturbances. Figure 2A shows 50 of the most significantly changed proteins for four bacterial species with the largest number of detected proteins (see list of proteins and their functional annotation in Table S7). Antibiotic or antifungal treatment appeared to have a similar effect on the proteome of A. muciniphila YL44 (comparison of BY_ABX vs. BY_AFX groups), while for M. intestinale YL27, B. coccoides YL58, and C. clostridioforme YL32, the antibiotic treatment led to more proteins with increased levels compared to the antifungal groups (Table S8). For the A. muciniphila YL44, M. intestinale YL27, and B. coccoides YL58, many proteins had significantly increased levels in response to treatment with either antifungal or antibiotic (B/BY compared to ABX/AFX). For C. clostridioforme YL32, the detected proteome showed a contrasting response to the antimicrobial treatments. For instance, compared to the fecal proteome of the B only condition, there were 2.6 and 1.3 times less proteins with elevated levels than in the BY_AFX and BY_ABX groups, respectively (Table S8). But when fungal species were present in the mouse gut (BY group), the effect of the antibiotic on C. clostridioforme YL32 appeared similar as for the other three strains (BY compared to BY_ABX), while treatment with antifungal led to a 1.3-fold increase in proteins with elevated levels in the BY group (BY compared to BY_AFX).
It should be noted that the effect of increased protein levels in the ABX/AFX groups could be further amplified by the MS detection method itself, which is less sensitive to low abundant proteins and will preferentially identify those with higher abundances (see Table S1 - Summary of MS data search). In any case, these comparisons indicate that antimicrobial perturbance targeting bacterial or fungi cause differential responses among bacterial species that depend on species identity and presence or absence of fungi.
For the rest of the bacterial strains, low numbers of quantified proteins did not allow for in-depth investigations. Nonetheless, seven of the strains showed considerable differences in the protein profiles between the mice groups (Figure S6). A general picture has emerged from this comparison, in which antimicrobial treatments targeting bacteria or fungi during the second week of the mice’s life resulted in lasting changes in the proteomes of the gut bacterial species.
Gut fungi differentially modulate bacterial proteomes.
Next, we statistically compared the bacterial protein levels between the B and BY groups to investigate the effect of fungi on the proteomes of the four strains with the highest numbers of detected proteins (Figure 2B). Similarly to what we observed for antimicrobials, fungal presence diversely influenced the proteomes of individual bacterial species. A. muciniphila YL44 and M. intestinale YL27 displayed increased amounts of proteins from various functional classes in the presence of fungi (Figure S7). We observed the opposite for C. clostridioforme YL32 and B. coccoides YL58, where the levels of 100 out of 107 and 51 out of 52 differentially produced proteins, respectively (t-test, FDR 5%), decreased when fungi were present (Table S8). The proteomic observations for A. muciniphila YL44 appeared to be inversely correlated with a decrease in A. muciniphila abundance based on 16S rRNA sequencing in the presence of fungi (18) (Figure S7). For C. clostridioforme YL32, the trend of increased protein levels in the B group compared to the BY group also appeared to inversely correlate with the 16S-based abundances; however, these trends did not reach statistical significance in the 16S sequencing data. Similarly, we could not derive a correlation with the sequencing data for M. intestinale YL27 and B. coccoides YL58, as the strains’ relative abundance appeared similar across the mice groups. Metaproteomic analyses thus provided complementary information to the sequencing data and yielded a higher resolution of the ecological interactions between the gut microbial species.
Gut bacteria and antimicrobial treatments modulate fungal proteomes.
We detected 1,492 proteins for the five fungal species colonising the gnotobiotic mice, and 39% of these proteins were assigned LFQ levels. The lower numbers of proteins detected and quantified, as compared to bacteria, were a result of a lower abundance of fungi in the gut that we previously confirmed by quantitative PCR and microbiological assays (18). Also, the LFQ approached used (34), is less sensitive to low abundant proteins, which are less likely to be quantified. Most of the identified fungal proteins were present in the Y group representing the mice colonised exclusively with fungi. We previously showed that the Y group harboured higher fungal concentration than co-colonised mice (18), confirming that a bacterial suppression of the fungal colonisation occurs in the host gut (42). Of the five fungal species, C. glabrata had the highest number of identified proteins, which also dominated the Y group (Figure 3A). However, antibiotic treatment (BY_ABX) led to higher detection of proteins derived from Pichia kudriavzevii/C. krusei. The total number of differentially abundant proteins between the mice groups and their distribution among the fungal species were rather similar for the following pairs of mice groups: Y - BY_ABX and BY - BY_AFX (Figure 3B). When we expressed the number of differentially produced fungal proteins as a percentage of each strains detected proteins, we noticed that C. glabrata proteome showed a stronger response to the antibiotic treatment (BY_ABX), as compared to Y, BY and BY_AFX.
The presence of bacteria and early-life antimicrobials had significant effects on the fungal species protein profiles. Arrangement into categories using functional annotations based on UniProtKB (35) and other databases (36, 37), showed an increased number of fungal proteins related to stress responses in the antibiotic condition (Figure 3C), suggesting that the fungal species are either negatively affected by the changes in the bacterial microbiome or by the antibiotic treatment itself. Interestingly, many of the identified fungal proteins were previously reported to be excreted via extracellular vesicles (43, 44) and to have immunogenic properties (45, 46) (Table S9). Among these proteins were enzymes of the glycolytic pathway (e.g., enolase, glyceraldehyde-3-phosphate dehydrogenase, fructose-bisphosphate aldolase, and phosphoglycerate mutase) and molecular chaperones linked to stress response (heat shock proteins). These proteins were significantly elevated in the fungal group suggesting that they are critical cytosolic proteins abundantly produced by fungal cells living in the mouse gut.
Fungal and bacterial colonisation induces distinct and persistent changes in the host fecal proteome.
We identified 674 mouse proteins as part of the mice fecal metaproteome. From proteins quantified by LFQ (405), 71% displayed differential abundance between the groups (ANOVA followed by THSD post-hoc analysis, FDR 5%, Table S10). PCA score plot based on the quantified proteins showed four clusters (Figure 4A): the first contained co-colonised mouse groups treated with antimicrobials (BY_ABX/AFX), and the second consisted of the B and BY groups. The third and fourth clusters included the Y and GF groups, and these were considerably more dissimilar from the first two clusters. Host proteins in the Y group were more clearly separated from the GF mice, in contrast to clustering when also microbial proteins were considered (Figure 1C), revealing that the proteome response to fungal colonisation may be more pronounced in the host proteome than in microbial cells. The GF group displayed the highest number of proteins associated with lipid metabolism, regulation, and molecular processing, while the BY and BY_AFX groups had the highest numbers of immune proteins (Figure 4B). In the BY_ABX group, we identified an increased number of stress response proteins. Moreover, the Y and GF groups displayed an increased number of proteins associated with energy metabolism (e.g., mitochondrial proteins).
Proteins whose levels significantly varied between the mouse groups belong to the following functional classes: processing proteins such as proteases, hydrolases, and chaperones (14%), proteins linked to cell cycle regulation, motility, and cytoskeleton, including components of the tight junctions (14%), transport and membrane proteins including epithelial cell surface receptors (13%), proteins of energy and lipid metabolism (12% and 9%, respectively), and gut immune and barrier factors (9%). To tease apart the mouse proteome response to different microbes, we statistically compared pairs of treatment groups containing either fungi or bacteria (Figure S8). The introduction of the 12-species bacterial consortium led to a striking decrease in the mouse protein levels, and this was valid both for germ-free mice (GF vs. B) and mice colonised with fungi (Y vs. BY). Although gut fungi did not have the same significant effect on the mouse proteome as bacteria, we detected a 5-fold decrease in the number of proteins whose levels increased in the Y group compared to GF (GF vs. Y). The presence of fungi and bacteria resulted in a 2-fold increase compared to bacteria alone (BY vs. B), particularly of proteins associated with the immune system, energy metabolism, and cellular processing.
Because the mouse protein levels were strongly dependent on the gut microbial consortium, we gave special attention to the individual functional classes of differentially produced proteins. Figure 4C shows 45 differentially produced mouse proteins with the largest variation between the mice groups. Levels of 27 immune proteins were the highest in GF mice, followed by those colonised only with fungi. Two immune proteins were elevated in the presence of fungi: acid mammalian chitinase (Chia), implicated in the defense response against fungi, and regenerating islet-derived protein 3-gamma (Reg3g), a bactericidal C-type lectin reported to have bactericidal activity (47, 48). Thus, our results suggest that Reg3g might also be involved in the intestinal cells’ response to fungal colonisation. However, more data will be needed to support the involvement of Reg3g in the epithelial cell response to fungi.
Bacterial presence resulted in a reduction in the mucosal Pentaxin (Mtx2) levels, a secreted protein involved in complement activation, Intelectin 1 (Itln1), a receptor that binds microbial glycans, and Dipeptidyl peptidase 4 (Dpp4), cell surface receptor and dipeptidyl protease involved in T-cell activation. In addition, levels of other immune proteins, such as Mucin 2 (Muc2), Poly-Ig receptor (Pigr), and immunoglobulin heavy chain (IgH), were altered by the shifts in microbial composition caused by antimicrobial treatments.
The absence of microbes resulted in elevated levels of mouse proteins linked to carbohydrate metabolism both at the cellular [e.g., Pyruvate Kinase (Pkr)] and host level [glycosidases such as Pancreatic Alpha-Amylase (Amy2), Lactase (Lct), and Trehalase (Treh)]. Amy2, a glycosidase that hydrolyses alpha-linked polysaccharides such as starch and glycogen, had the highest levels in the Y group, suggestive of increased Amy2 production response to either fungal polysaccharides or fungal metabolism on dietary sugars. Fungi also elicited a strong effect on host proteins associated with lipid metabolism. The highest levels were identified in the Y group for the host enzyme Colipase (Clps), a cofactor of pancreatic lipase facilitating lipids digestion, and Phospholipase A2 (Pla2g1b), a secreted protein involved in lipid degradation and innate immune mucosal response. Closely connected to the immune system are phospho- and sphingolipid metabolism enzymes, which often have regulatory properties towards immune cells. Of those, we identified Neutral Ceramidase (Asah2), a membrane protein hydrolysing sphingolipid ceramides into sphingosines and free fatty acids, whose relative levels were the lowest in BY and B groups.
Cytoskeletal and cell adhesion proteins were also significantly increased in the GF and Y mice. These included cadherin-related proteins 2 and 5 (Cdhr2 and Cdhr5), which regulate microvilli length, and Claudins 3 and 7 (Cldn3 and Cldn7), controlling tight junction-specific obliteration of the intercellular space. Notably, Keratin-33A and -81 (Krt33a and Krt81), structural proteins that form the cytoskeleton’s intermediate filaments in epithelial cells, displayed the highest levels in the Y group, followed by the B group and had the lowest levels in the BY groups treated with antimicrobials. Adhesion proteins with regulatory properties and links to the immune system included Tetraspanin-8 (Tspan 8), Epithelial Glycoprotein 314 (Epcam), and Basigin (Bsg), a cell surface receptor. All three proteins showed increased levels in the GF and Y groups.
One of the most prevalent functional classes among the differentially produced proteins were proteases, which had the highest levels in the GF and Y groups, and the lowest in the BY and B groups. Among these was for example Angiotensin-converting enzyme (Ace2), a carboxypeptidase with multiple regulatory functions, including gap junction assembly. Other proteases with regulatory properties included membrane-bound metalloproteases Meprin A (Mep1a) and B (Mep1b), implicated in the inflammatory response. Alongside proteases, protease inhibitors followed the same quantitative pattern as described above. These included Serpins, serine protease inhibitors that negatively regulate endopeptidase activity in response to cytokines (Serpinb1a, Serpina1d), innate immune response, inflammation, and cellular homeostasis (Serpinb1a, Serpinb6).
Numerous regulatory proteins were present among the proteins with increased levels in the GF and Y group, such as the Annexin family of Ca2+-regulated phospholipid-binding and membrane-binding proteins (Anxa4, Anxa11, Anxa13) and nuclear proteins such as Onzin (Plac8) suggested to regulate immune responses. Overall, the metaproteomic analyses documented an extensive impact of microbial colonisation in a controlled early life gut microbiome model, underlining the intertwined functional development of the host and its gut microbiome, and revealing new features of the host response to fungal colonisation.
Fungal colonisation drives alterations in the mouse jejunal tissue proteome.
Given the exciting findings from the mice fecal metaproteomes, we decided to investigate more closely the host proteome. We selected the small intestine as our site of interest because it contains the body’s largest immune organ (the gut-associated lymphoid tissue) and is an important site of the host-microbe interactions (49). We used a tandem mass tag labelling approach (TMT 6-plex) to gain a higher sensitivity for low abundant proteins (Figure 5A). From the TMT 6-plex experiment, we identified 10,201 peptides (Table S11) matched to 1,514 mouse proteins (Table S12), and the levels of 45 were significantly altered between the mice groups (ANOVA, THSD post-hoc analysis, FDR 5%, Table S13). PCA score plot revealed different grouping results than for the mouse fecal proteomes (Figure 5B). Functionally, the 45 significantly changed proteins were mainly associated with cellular metabolism and regulation of cellular processes (Figure 5C). Pathway analysis identified oxidation-reduction, glutathione metabolism and cell-cell adherence as significantly enriched processes (Table S14).
Compared to the fecal mouse proteome, proteome changes of the jejunal tissue related to microbial colonisation were more subtle. Figure 5D shows the relative levels of 45 differentially produced mouse jejunal proteins with significant variation between the mice groups. Among immune proteins whose levels increased with the presence of fungi were Alpha-defensin 2 (Def20) and Ig gamma-1 chain C region secreted form (Ighg1). From 10 proteins functionally connected to lipid metabolism, two Glutathione S-transferase enzymes (Gstm2 and Gsta4) were significantly decreased in the fungal group. We also detected two proteins of retinol metabolism (Retsat and Aldh1a1), which play a key role in mucosal immune responses. Fungal colonisation further appeared to impact the intestinal cells’ energy metabolism, as several mitochondrial proteins had decreased levels in the fungi group, most notably, two subunits of Cytochrome C oxidase (Cox5b and Cox7a1).
Some of the differentially produced proteins have functional links to NF-κB pathway, such as the apoptotic marker Poly (ADP-ribose) polymerase (Parp1). In some contexts, particularly in response to cellular stress, stimulation of the NF-κB pathway promotes apoptosis (50). Selenium acts as a key element that controls NF-κB activation and the half-life of its inhibitor IκBα, and we detected two selenium-binding proteins (Selenbp1 and Selenbp2). Overall, the TMT 6-plex analysis of the jejunal tissue provided further insight on host cellular pathways impacted by defined microbial colonisation and confirmed that gut fungi elicit differential effects compared to bacteria.