In vitro interaction network of a synthetic gut bacterial community

Probing directional interactions of OMM¹² strains using spent culture media

To characterize directional interactions of the OMM¹² consortium members, we chose an in vitro approach to explore how the bacterial strains alter their chemical environment by growth to late stationary phase.

Growth of the individual monocultures in a rich culture medium that supports growth of all members (AF medium, Methods, Table S1) was monitored over time (Fig. S1; SI data table 1) and growth rates (Table S2) were determined. Strains were grouped by growth rate (GR) into fast growing strains (GR > 1.5 h^–1, E. faecalis KB1, B. animalis YL2, C. innocuum I46 and B. coccoides YL58), strains with intermediate growth rate (GR > 1 h^–1, M. intestinale YL27, F. plautii YL31, E. clostridioformis YL32, B. caecimuris I48 and L.reuteri I49) and slow growing strains (GR < 1 h^–1, A. muris KB18, A. muciniphila YL44 and T. muris YL45). All strains reached late stationary phase within 20 h of growth. To probe overlap in substrate requirements and interactions between the individual OMM¹² members mediated by waste products or bacteriocins, sterile spent culture medium (SM) after growth to late stationary phase of all strains was obtained. Each OMM¹² strain was cultured in the SM of the other community members and their own SM and growth rate, the area under the growth curve (AUC) and the pH were determined (Fig. 1B; Fig. S2; SI data table 1).

A normalized inhibition factor (d_AUC) was determined by the AUC in SM relative to the AUC in fresh AF medium (( {{{{{{{{mathrm{d}}}}}}}}_{{{{{{{{mathrm{AUC}}}}}}}}} = frac{{AUC_{SM} – AUC_{AF}}}{{AUC_{AF}}}})) to quantify the influence of the different SM on the growth of the individual OMM¹² strains (Fig. 1C). Ten of the twelve SM were found to enable decreased (d_AUC < –0.5) growth of at least one other strain of the consortium. Only the SM of strains A. muris KB18 and A. muciniphila YL44 enabled reduced growth of just the strains themselves. Corresponding to decreased AUC values in SM, growth rates were found to be lower as well, resulting in linear correlation of AUC and growth rates (Fig. S3, R > 0.5, p < 0.05 for all strains). The SM of four strains, E. faecalis KB1, B. coccoides YL58, E. clostridioformis YL32 and B. caecimuris I48, were found to strongly inhibit (d_AUC < –0.5) the growth of nine other strains each (Fig. 1C). Notably, growth of E. faecalis KB1 itself was only strongly reduced in its own SM, while it was able to grow in other strains’ SM. T. muris YL45 was the only strain not showing clear growth inhibition in any of the SM while its SM strongly decreased growth (d_AUC < –0.5) of three other strains, A. muris KB18, M. intestinale YL27 and F. plautii YL31.

Individual pH profiles as indicators for niche modification

The pH of the culture medium after growth to stationary phase can be used as a measure for the extent of strain-specific environmental modification [11] and may partly explain inhibition of bacterial growth in a SM. Therefore, we determined the pH of the individual SM before and after (double spent media; DSM) growth of all OMM¹² strains (Fig. 1B; SI data table 1). From these values, we defined the ΔpH for every strain after growth in fresh medium (ΔpH_SM) and in all SM (ΔpH_DSM) by analyzing the strength (difference of pH values) and direction (more acidic or more alkaline) of the pH change (Fig. 1D). After growth in fresh AF medium with neutral pH of 7.0, the OMM¹² strains showed different degrees of ΔpH_SM. While E. faecalis KB1, B. animalis YL2, M. intestinale YL27, B. caecimuris I48 and B. coccoides YL58 distinctly acidified the medium (pH_SM < 6.2), the growth of the other strains resulted in either slightly more alkaline or nearly neutral medium. Correlating inhibition of growth in a SM (d_AUC) with the mean pH of the individual SM for each strain revealed that growth inhibition did not directly correlate with the pH. Only strains B. animalis YL2, A. muciniphila YL44 and B. caecimuris I48 showed a significant negative correlation (R < –0.5, p < 0.05) between growth inhibition and pH (Fig. S4; SI data table 1) with stronger inhibition in more acidic pH ranges. Testing monoculture growth in fresh AF medium with adjusted pH values from pH 5 to pH 7.5 revealed that these strains indeed show decreased growth rates and lower final OD values in medium with pH < 6.5 (Fig. S5; SI data table 1). pH sensitivity was further observed for M. intestinale YL27.

Most interestingly, many strains did not show the same magnitude or direction of alteration in pH when grown in SM of another strain (ΔpH_DSM) compared to growth in fresh culture medium (ΔpH_SM). This indicates an altered metabolic behavior of some strains in specific SM environments that differs from metabolic behavior in fresh AF medium (Fig. S6, Supplementary Text A).

Production of antibacterial compounds by E. faecalis KB1

Growth inhibition in SM (Fig. 1C) may further be explained by the production of antimicrobial compounds. To test for the production of antimicrobial compounds by the OMM¹² strains, we used a phenotyping approach and performed spot assays on agar plates (Fig. 1E). Inhibition zones were only seen in case of E. faecalis KB1, which produced one or several compounds active against B. animalis YL2, F. plautii YL31, E. clostridioformis YL32, C. innocuum I46 and L. reuteri I49. Genomic analysis revealed that the strain encodes genes for the production of several bacteriocins (Supplemental Text B), including enterocin L50, an enterococcal leaderless bacteriocin with broad target range among Gram-positive bacteria [33]. All other strain pairs did not show signs of growth inhibition by compound excretion under these conditions, despite the presence of genes for lanthibiotic production in the genome of B. coccoides YL58 (determined by antiSMASH) [34]. Although expression of antimicrobial molecules may be induced by specific environmental triggers, which are absent in the monoculture in vitro setting, we concluded that interference competition may only play a role in a subset of pair-wise interactions in AF medium involving E. faecalis KB1.

Substrate depletion profiles correlate with growth inhibition in SM

As pH and antimicrobial compounds only partly explained inhibition of growth in SM, we set out to gain more insights into the individual metabolic profiles in our in vitro setting. Therefore, triplicate samples of fresh AF medium and SM were analyzed by a mass spectrometry-based untargeted metabolomics approach (TripleTOF, Methods). Combining positive and negative ionization mode, 3092 metabolomic features were detected in total. From these, 2387 (77.20 %) were significantly altered (t-test, p value < 0.05) by at least one of the twelve strains (Fig. S7). Hierarchical clustering of the metabolomic feature depletion profiles (i.e., substrates used by the bacteria; Fig. 2A) reflects the phylogenetic relationship between the strains (Fig. 1A). Correlating the phylogenetic distance between the individual strains with the number of shared depleted metabolomic features in AF medium (Fig. S8) showed that phylogenetically similar strains of the consortium have a higher substrate overlap than phylogenetically distant strains (R = –0.29, p = 0.017). The total number of metabolomic features that are depleted from AF medium greatly varies for the different strains, ranging from over 600 depleted features for M. intestinale YL27 to only 42 for A. muciniphila YL44 (Fig. 2B). The number of metabolomic features overlapping with other OMM¹² strains’ features relative to the strains’ total set of depleted features was determined (Fig. 2C). Phylogenetically related strains like E. clostridioformis YL32 and B. coccoides YL58 or M. intestinale YL27 and B. caecimuris I48 share over 50% of depleted metabolic features each, suggesting a strong substrate overlap in AF medium. Visualizing the extent of overlap between substrate depletion profiles reveals that Bacteroidales, Clostridia and Bacilli strains of the consortium dominate with the highest number of commonly depleted substrates in AF medium (Fig. 2D).

Correlating the growth inhibition in SM (d_AUC) with the pairwise overlap in depletion profiles (Fig. 2C) revealed that a larger overlap is correlated with a stronger growth inhibition in the corresponding SM (R = –0.46, p = 3.1E–08, Fig. S9). This is illustrated by A. muciniphila YL44, which used only a low number of substrates from the AF medium (Fig. 2B) and the SM of which had only little effect on the growth of the other strains of the consortium (Fig. 1C). On the other hand, the strain´s growth itself was strongly reduced in the SM of most other consortium members (Fig. 1C, S2), which depleted a large spectrum of metabolomic features including those used by A. muciniphila YL44 (Fig. 2C).

Genome-informed metabolic potential of the OMM¹² consortium

To gain insights into metabolic properties of the OMM¹² strains, we reconstructed genome-scale metabolic models using gapseq [35] (SI data file). The initial metabolic models were curated by screening for metabolic pathways and transporter proteins and filling of missing reactions (gap-filling). From the genome-based metabolic models, we derived the presence and absence of metabolic pathways for central carbon metabolism (e.g., fermentation pathways, respiration), amino acid metabolism, and utilization of specific substrates, for the individual strains using MetaCyc pathways [36] (Fig. 3A, Fig. S10, SI data table 2). Further, the presence of specific substrate transporters was determined (Fig. S11, SI data table 2). Hierarchical clustering of the genome-informed metabolic potential (Fig. 3A) reflected their phylogenetic relationship in many instances. Generally, a high diversity of central and fermentation pathways was found among the consortium members. Moreover, enzymes for the degradation of amino acids (e.g., aspartate, glutamate, serine, and cysteine) are highly prevalent among consortium members. Correspondingly, systems for amino acid transport were especially prevalent among all strains of the consortium (Fig. S11).

Metabolite production and fermentation pathways of the OMM¹² strains in AF medium

To verify metabolites and fermentation products produced and consumed by the individual strains of the consortium in the given in vitro conditions, SM were analyzed using different mass spectrometry approaches (Fig. S12, S13). The combination of experimentally obtained insights and genome-based information on the presence of pathways was used to generate sketch drawings to visualize basic metabolic properties of the individual OMM¹² community members (Fig. 3B, C, S14).

To confirm if fermentation pathways identified by genomics were active under in vitro conditions, short chain fatty acid (SCFA) production and consumption were analyzed (Fig. S12A). As observed for the SM metabolic profiles (Fig. S7), hierarchical clustering revealed that closely related bacteria showed similar SCFA production and consumption profiles. Both Bacteroidales strains produced acetic acid, succinic acid as well as branched-chain fatty acids. Both Lachnospiraceae strains generated high amounts of acetic acid. Butyric acid is produced by strains F. plautii YL31 and C. innocuum I46, the latter also being the only strain of the consortium excreting valeric acid and caproic acid. Of note, F. plautii YL31 also consumed lysine, indicating the ability to produce butyric acid from lysine, which was supported by the presence of the pathway for lysine fermentation to acetate and butyrate.

Formic acid was produced by several strains and consumed by T. muris YL45 and B. coccoides YL58. B. coccoides YL58 encodes genes for a CO dehydrogenase/acetyl-CoA-synthase, the key enzyme of the Wood-Ljungdahl pathway (reductive acetyl-CoA pathway). Formic acid can be processed via this pathway to acetyl-CoA. As another prominent example of bacterial fermentation, lactate production was confirmed (Fig. S12C) for E. faecalis KB1, B. animalis YL2, F. plautii YL31, C. innocuum I46, E. clostridioformis YL32 and L. reuteri I49, all of which harbor the pathway for lactate formation.

By quantifying amino acid levels we could show that Bacteroidales and Lachnospirales strains exhibited similar amino acid depletion and production profiles. In SM of strains M. intestinale YL27 and B. caecimuris I48, elevated levels of a diverse range of amino acids including glutamic acid, histidine, methionine, proline and phenylalanine were detected. Lachnospirales strains showed increased levels of isoleucine, tryptophan and valine, while alanine was especially depleted by B. coccoides YL58. Other strains of the consortium showed specific depletion of single amino acids, e.g., F. plautii YL31 strongly depleted lysine and glutamic acid, while E. faecalis KB1 depleted serine.

Growth of OMM¹² strains in pairwise co-culture

Next, we performed a set of experiments to characterize strain-strain interactions in the dynamic community-dependent context. We first analyzed direct competition of all strains in pair-wise co-cultures over the course of 72 h, with serial dilutions every 24 h. While growth was monitored continuously by OD 600 nm, samples for pH measurements and qPCR analysis were taken every 24 h. The growth curves of most co-cultures, as well as supernatant pH differed from the corresponding strain-specific characteristics observed in monoculture (Fig. S15, Fig. S16). These differences reflect co-culture dynamics, as can be seen from change in relative abundances over time.

To identify directionality and mode of interaction between the OMM¹² strains, we analyzed the relative changes in absolute abundance (normalized 16S rRNA gene copies) as a measure of how successful a strain can grow in co-culture relative to monoculture after 72 h. The mean absolute abundance ratio was calculated for every strain in all pairwise co-cultures (( {{{{{{{{mathrm{r}}}}}}}}_{{{{{{{{mathrm{i,bm}}}}}}}}}=frac{{m_{i,co}(t72h)}}{{m_{i,mono}(t72h)}}} )) (Fig. 4A, Methods). If absolute abundance of a strain increased significantly in co-culture relative to monoculture (r_bm > 1), the interaction was categorized as positive (+), if it decreased (r_bm < 1) the interaction was categorized as negative (–) (t-test comparing the r_bm of three independent experiments, Fig. S17). If it did not significantly (p > 0.05) differ from that in monoculture (r_bm = 1), the interaction was categorized as neutral (0). By this, we created a co-culture interaction matrix (Fig. 4B): the vast majority of the interactions was classified as amensalistic (0/– and –/0, 46 of 66 of interactions). A smaller subset of interactions was either competitive (–/–, 7 of 66 of interactions) or neutral (0/0, 11 of 66 of interactions). No mutualistic interactions (+/+) were observed. However, one example for each, commensalism (0/+ and +/0) and predation (+/– and –/+), were identified.

The extent to which the individual strains altered the growth of other community members in the co-culture differed distinctly. While E. faecalis KB1 and C. innocuum I46 lead to nine negative co-culture outcomes each, A. muciniphila YL44 and A. muris KB18 only impaired growth of one and zero strains, respectively. Simultaneously, both strains are negatively influenced in most co-cultures, with a significantly decreased absolute abundance in ten and eight co-cultures, respectively. Notably, B. coccoides YL58 is involved in five of seven competitive interactions of the consortium. These observations are in line with the outcomes observed in SM experiments, as strongly negative co-culture outcome corresponds to a strong inhibition of a strain in the respective SM (Fig. S18).

Of note, the only predatory interaction was observed between C. innocuum I46 to E. faecalis KB1, where the absolute abundance of E. faecalis KB1 significantly increased in the presence of C. innocuum I46 compared to monoculture growth. This beneficial interaction might be due to a metabolic advantage that arises in co-culture. In order to identify potentially cross-fed metabolites of C. innocuum I46 to E. faecalis KB1, we mined metabolomic data of SM for features enriched in C. innocuum I46 and depleted by E. faecalis KB1. Thereby, we identified several compounds including malate, L-methionine, spermidine, and methylglyoxal (Fig. 4C). To experimentally support the idea of cross-feeding, we exemplarily tested uptake of ¹⁴C-malate and ³H-L-methionine into intact cells of E. faecalis KB1 (Fig. 4D). Both uptake of ¹⁴C-malate and uptake of ³H-L-methionine were strongly inhibited by the hydrophobic protonophores 2,4-dinitrophenol (DNP) and carbonyl cyanide m-chlorophenylhydrazone (CCCP), suggesting active transport driven by the proton motive force for both substrates. We found a very fast linear uptake of particularly ¹⁴C-malate within the first 60 s, which could explain why malate utilization confers a growth advantage to this strain. Decrease of absolute abundance of C. innocuum I46 in co-culture with E. faecalis KB1 might be due to the production of antimicrobial compounds by E. faecalis KB1 active against C. innocuum I46 (Fig. 1E).

Community structure of the OMM¹² consortium

Next, we set out to investigate if interactions found in co-cultures are transferrable to the strains’ behavior in the complete OMM¹² community. To this end, all twelve OMM strains were simultaneously co-cultured in AF medium and serially diluted 1:100 every 24 h into fresh AF medium. Relative abundance of all strains after 10 days was determined by qPCR for ten replicates each in two independent experiments from different inocula (Fig. 4E).

While each of the OMM¹² members except E. faecalis KB1 was outcompeted to a very low relative abundance in at least one pairwise culture (Fig. 4A, B), the majority (10 out of 12) of the consortium members were able to coexist in the complex community up to 10 days (Fig. 4E, Fig. S19A). Replicate communities showed reproducible community structure, even when different inocula were used. Interestingly, F. plautii YL31 dominated the community under these conditions. Furthermore, B. coccoides YL58 and E. faecalis KB1 showed a high relative abundance, which corresponds to their dominant role in SM and co-culture experiments (Figs. 1C, 4A, B). While strains B. animalis YL2 and L. reuteri I49 were not detectable at 10 days in all replicates, A. muris KB18 was found in only a few of the communities at 10 days (relative abundance < 1%).

E. faecalis KB1 strongly impacts overall in vitro community composition

Following, we investigated how the absence of E. faecalis KB1, which plays a dominant role in pair-wise interactions, would affect the overall community structure. We generated a ”dropout” community including all strains of the OMM¹² consortium except E. faecalis KB1 (OMM¹¹–E. faecalis KB1). Compositional analysis revealed increased relative abundance of C. innocuum I46 and B. animalis YL2 in the OMM¹¹–E. faecalis KB1 compared to the full OMM¹² community (Fig. 4F). In addition, the absolute abundances of strains B. animalis YL2, C. innocuum I46, B. coccoides YL58 and B. caecimuris I48 were found to increase significantly (t-test, p < 0.05) in the absence of E. faecalis KB1 (Fig. S19A (M1)). The increase in abundance of B. animalis YL2 and C. innocuum I46 may be explained by absent enterocin production or substrate competition by E. faecalis KB1. The latter may also explain increased abundance of B. caecimuris I48 and B. coccoides YL58 in the dropout community. Further, the abundance of F. plautii YL31, E. clostridioformis YL32, A. muciniphila YL44 and T. muris YL45 was found to decrease in the absence of E. faecalis KB1 (Fig. S19A (M1)). This indicates either direct positive effects of E. faecalis KB1 on these strains or indirect effects that occur through the overall shift in OMM¹¹–E. faecalis KB1 community composition compared to the OMM¹² consortium.

Influence of specific supplements on community structure

Finally, we assessed the effect of media composition on community structure under our in vitro culture conditions. Therefore, we generated a comprehensive dataset comparing the composition of the complete OMM¹² community and the E. faecalis KB1 dropout in media with different supplements that are known to promote the growth of specific gut bacteria but are missing in AF medium (mucin, C5/C6 sugars, xylan & inulin, starch; Figs. 5A, B, S19). We used a modified AF medium with reduced glucose concentration to rule out that substrate consumption may be inhibited by catabolite repression. Of note, reduction of glucose resulted in decrease of E. faecalis KB1, C. innocuum I46, and B. animalis YL2 and increase of A. muciniphila YL44 (Figs. 5B, S19). We found that the chosen supplements had characteristic effects on relative and absolute abundance of individual strains (Figs. 5A, B, S19). A. muciniphila YL44, a known mucin-degrader, was boosted by mucin. Additional supplementation with C5/C6 sugars (xylose, arabinose, lyxose, fucose and rhamnose) promoted growth of B. caecimuris I48. Further, supplementing media with xylan and inulin promoted growth of A. muris KB18 and even more enhanced levels of B. caecimuris I48. At the same time, A. muciniphila YL44 and M. intestinale YL27 were decreased. Interestingly, L. reuteri I49 was also promoted by xylan & inulin but only in the OMM¹¹–E. faecalis KB1 dropout community (Figs. 5B, S19). Finally, supplementation of AF medium with starch only promoted growth of F. plautii YL31 (Figs. 5A, B; S19). Of note, increase of C. innocuum I46 and B. animalis YL2 strains was generally observed in the E. faecalis KB1 dropout communities irrespective of the media conditions. Lack of E. faecalis KB1 had different effects on the abundance of B. caecimuris I48, M. intestinale YL27 and F. plautii YL31 depending on supplements (Fig. 5A, B, S19).

Comparison of in vitro and in vivo OMM¹² community structure

In order to evaluate, which of the in vitro conditions most closely resemble community composition in the murine gut, we analyzed community composition in the ileum, cecum, colon, and feces in 5 age-matched adult (12 weeks old) male C57Bl/6 mice. In general, intestinal communities were similar in cecum, colon, and feces, with B. caecimuris I48, B. coccoides YL58 and A. muciniphila YL44 predominating (Fig. 5C). The ileal community was distinct and dominated by B. coccoides YL58 and L. reuteri I49. The relative abundance of E. faecalis KB1 and F. plautii YL31 was comparatively low in in vivo samples. Similar as in AF media, A. muris KB18 and B. animalis YL2 were at the detection limit of the qPCR assay. Principal Component Analysis (PCA) of relative abundance data was used to compare in vivo community structures to OMM¹² and OMM¹¹–E. faecalis KB1 cultured in vitro in different media (Fig. 5D). The highest similarity was observed between OMM¹² community structure in cecum content and AF medium supplemented with mucin, C5/C6 sugars, xylan, and inulin. Conversely, OMM¹² community structure in feces and colonic content most closely resembled OMM¹² and/or OMM¹¹-KB1 in medium M1 and M10. Taken together, we identified specific supplements that can be used to shape in vitro conditions to more closely recapitulate community structure in different regions of the murine gut.

Source: Ecology - nature.com