Competition for shared resources increases dependence on initial population size during coalescence of gut microbial communities

Significance

Humans routinely encounter new microbes, but it remains unclear how the number of introduced microbes affects their ability to colonize the gut microbiome. All else being equal, species are more likely to colonize when more individuals are introduced. However, even large populations may fail to establish if they are outcompeted by resident species. To investigate how initial population size and competition affect colonization, we compared mixtures of in vitro gut microbial communities to models of resource competition. Colonization depended more strongly on initial population size in diverse communities, in which certain introduced and resident species had high niche overlap, consuming shared resources at similar rates. In these cases, communities converged slowly to equilibrium, prolonging the effect of initial population size.

Abstract

The long-term success of introduced populations depends on both their initial size and ability to compete against existing residents, but it remains unclear how these factors collectively shape colonization dynamics. Here, we investigate how initial population (propagule) size shapes the outcome of community coalescence by systematically mixing eight pairs of in vitro microbial communities at ratios that vary over six orders of magnitude, and we compare our results to neutral ecological theory. Although the composition of the resulting cocultures deviated substantially from neutral expectations, each coculture contained species whose relative abundance depended on propagule size even after ~40 generations of growth. Using a consumer–resource model, we show that this dose-dependent colonization can arise when resident and introduced species have high niche overlap and consume shared resources at similar rates. Strain isolates displayed longer-lasting dose dependence when introduced into diverse communities than in pairwise cocultures, consistent with our model’s prediction that propagule size should have larger, more persistent effects in diverse communities. Our model also successfully predicted that species with similar resource-utilization profiles, as inferred from growth in spent media and untargeted metabolomics, would show stronger dose dependence in pairwise coculture. This work demonstrates that transient, dose-dependent colonization dynamics can emerge from resource competition and exert long-term effects on the outcomes of community coalescence.

Colonization by new species plays a major role in shaping community composition and function, but the outcomes of species introductions are notoriously difficult to predict (1–7). To colonize, a species must disperse from its original population (8–10), overcome the stochastic effects of ecological drift (11, 12), and compete successfully against resident species for a niche in the community (13–17). Models of community assembly place different emphasis on the role of neutral forces—namely, dispersal and ecological drift—compared to competition (15, 18, 19), and it remains unclear how these neutral and nonneutral forces interact to shape colonization outcomes.

Ecological models differ especially in their predictions about how colonization dynamics depend on the size and frequency of introduced populations, together known as propagule pressure (5, 20). The neutral theory of biodiversity, which assumes that species are ecologically equivalent, predicts that the abundance of an introduced species will remain proportional to its initial propagule size indefinitely in the absence of subsequent perturbations (15, 18, 21). In contrast, in the simplest consumer–resource models, each introduced species eventually reaches an equilibrium abundance determined by its competitive ability, regardless of its initial abundance (19, 22). However, it remains difficult to predict how long each species will take to reach this equilibrium abundance when starting from different initial population sizes. Therefore, investigating the effect of propagule size on colonization dynamics can shed light on the structure of the interspecies interactions that impact community assembly.

So far, empirical studies have come to contradictory conclusions about the effect of propagule size on colonization. Large propagule size is often associated with successful colonization (23–27), consistent with neutral expectations. However, in many cases, large propagules fail to colonize or small propagules successfully establish (5, 20, 21, 28–34), demonstrating that the effect of propagule size can be short-lived. Many of these studies focus on the introduction of a small number of species (24–26, 28, 31, 32, 34), making it difficult to generalize their conclusions across species and community contexts. Resolving these discrepancies requires systematic, large-scale quantification of the effect of propagule size.

To address this gap, we investigated the impact of propagule size on colonization during coalescence of in vitro gut microbial communities. Understanding the effect of propagule size in the human gut is important for designing microbiome therapeutics (35, 36), but prior studies have come to inconsistent conclusions about how inoculation dose affects the outcome of probiotic introductions and fecal microbiota transplants (FMTs) (37–42). To investigate propagule size in a more tractable, laboratory setting, we used stool-derived in vitro communities, which are stable, diverse, reproducible models of the gut microbiome that can recapitulate in vivo responses to perturbation (22, 43–47). We mixed pairs of in vitro communities at ratios that varied over six orders of magnitude, allowing us to observe the outcomes of hundreds of species introductions at once. By combining these experiments with consumer–resource modeling and measurements of resource utilization, we show that resource competition can amplify the effects of initial population size in diverse communities, with lasting effects on the outcomes of community coalescence (16, 48).

Results

Diversity and Composition of Coculture Communities Vary Across Mixture Ratios but Deviate from Neutral Predictions.

We performed coalescence experiments in vitro using a set of diverse, stable communities of gut microbes. Following a previously established protocol (46), we derived these communities from stool samples of eight healthy human subjects (SI Appendix, Table S1) and passaged them 15 times with a 1:200 dilution in fresh modified Brain Heart Infusion+mucin (mBHI+mucin) medium every 48 h (Fig. 1A and Materials and Methods). To assess community composition, we performed 16S rRNA gene amplicon sequencing on multiple passages from each community and tracked the relative abundances of amplicon sequence variants (ASVs, a taxonomic proxy for species) (49). Community diversity decreased in the first few passages after initial laboratory inoculation (Fig. 1B and SI Appendix, Fig. S1 A and B), likely because some taxa in the inoculum were nonviable or were outcompeted during in vitro passaging. Nonetheless, diversity stabilized after ~3 passages, and most communities thereafter contained ASVs that comprised >50% of the relative abundance of the stool inoculum (SI Appendix, Fig. S1 C–H). Thus, these stool-derived in vitro communities are diverse and stable models of stool microbiotas that can be used to study ecological interactions among gut microbes.

Fig. 1.

Diversity and composition of community cocultures vary by mixture ratio and deviate from neutral predictions. (A) In vitro communities were inoculated in triplicate from stool samples collected from eight healthy human subjects and passaged at a 1:200 dilution every 48 h 15 times to reach stability. (B) The diversity of each community over time, quantified as the effective number of species ($e^{H^{\prime }}$) calculated from the Shannon diversity index ($H^{\prime }$), initially decreased and then plateaued by passage 3. (C) Experimental design: eight pairs of in vitro communities were mixed in triplicate at seven mixture ratios ranging from 1,000:1 to 1:1,000. The resulting cocultures were passaged alongside the parent communities (represented by the 1:0 and 0:1 mixture ratios) five times, corresponding to ~40 generations (Materials and Methods). (D) The diversity of coculture communities was similar to or lower than the diversity of parent communities across mixture ratios. For this example mixture, gray lines represent two theoretical mixtures generated under neutral ecological expectations based on the composition of two replicate parent communities after five passages (1:0 and 0:1, shown in green), and orange lines represent three replicates of experimental community cocultures after five passages. Diversity was quantified as the effective number of species ($e^{H^{\prime }}$) calculated from the Shannon diversity index ($H^{\prime }$). (E) Variation of community composition across example mixtures and mixture ratios. Each plot shows the JSD of all coculture and theoretical communities relative to both parent communities (brown and gold) for a single mixture. Gray points along the dotted line represent theoretical mixtures generated under neutral ecological expectations, and colored points represent data from experimental coculture communities after five passages. Points show averages and error bars show the full range of values across inoculation replicates.

To systematically assess the effects of initial propagule size on colonization, we mixed eight pairs of these in vitro “parent” communities (Fig. 1C) at seven mixture ratios ranging from 1,000:1 to 1:1,000 (SI Appendix, Fig. S2), which approximately recapitulate the range of doses at which microbes are introduced into the resident gut microbiota through probiotic supplementation and FMTs (SI Appendix). We cocultured the mixtures in triplicate for five passages, corresponding to ~40 generations (Materials and Methods), to allow composition to equilibrate, and we passaged two replicates of each parent community alongside for comparison. To assess the impact of propagule size on community composition and the colonization of individual species, we compared our data to the predictions of ecological neutral theory (15, 18, 21). We computationally generated a set of theoretical cocultures in which each species remains at the relative abundance at which it was introduced in the initial mixture (SI Appendix, Fig. S3 and Materials and Methods), as predicted by ecological neutral theory when the number of generations is small compared to the total population size of the community, limiting the effect of ecological drift (SI Appendix). We then quantified the similarity of our data to these theoretical communities.

Under neutral expectations, coculture diversity should be maximal at the 1:1 mixture ratio. However, we found that experimental cocultures consistently maintained levels of diversity similar to or lower than one of the parent communities (Fig. 1D and SI Appendix, Fig. S4). This finding suggests that nonneutral competition occurs during community coalescence, preventing the neutral coexistence of all species. We also compared each experimental coculture to its corresponding theoretical coculture by calculating the Jensen–Shannon divergence (JSD, SI Appendix) between them. Cocultures consistently differed from the corresponding neutral theoretical expectation at every mixture ratio (Fig. 1E and SI Appendix, Fig. S5), providing further evidence for the role of nonneutral competition in shaping coculture composition. Despite these deviations from neutral predictions, the composition of many, although not all, cocultures varied substantially across mixture ratios (Fig. 1 E, Right and SI Appendix, Fig. S5). This variation in composition highlights the continued impact of propagule size after ~40 generations of growth and suggests that cocultures had not reached the equilibrium predicted by the simplest consumer–resource models, in which communities reach the same composition determined by species’ competitive abilities regardless of mixture ratio (19). Together, these patterns of diversity and composition suggest that neither neutral theory nor consumer–resource models alone can fully predict the varied outcomes of community coalescence on the time scales of our experiments.

Dose-Dependent Colonizers Are Present in Every Community Mixture.

To investigate how the behaviors of individual species give rise to nonneutral patterns of community composition that nevertheless depend on propagule size, we examined colonization outcomes at the ASV level. We compared the relative abundance of each ASV in experimental cocultures after five passages to the neutral theoretical prediction determined solely by its propagule size in the initial mixture (SI Appendix, Fig. S6 and Materials and Methods). In contrast to neutral predictions, many ASVs that started at higher abundance in one parent community maintained a consistently high (Fig. 2 A, Left and SI Appendix, Fig. S7A) or low (Fig. 2 A, Right and SI Appendix, Fig. S7B) relative abundance across mixture ratios in experimental cocultures (n = 32, 7.8% of the 410 ASVs above the limit of detection, SI Appendix, Fig. S6). We classified these dose-independent ASVs as strong or weak colonizers depending on whether their experimental relative abundance was greater or less than the neutral prediction, respectively. For these strong and weak colonizers, deviations from neutral predictions likely reflect competitive advantages or disadvantages that ultimately outweighed the effects of propagule size.

Fig. 2.

ASVs display a wide range of dose dependence during community coalescence. (A) Strong and weak colonizers reach the same relative abundance regardless of initial mixture ratio. Each panel shows a single representative ASV from one mixture. Gray lines represent both replicates of predicted neutral relative abundances for each ASV, and orange lines represent three replicates of experimental relative abundances after five passages. (B) Dose-dependent colonizers exhibit large changes in relative abundance across initial mixture ratios. Lines are colored as in (A) for a representative ASV from one mixture. (C) Dose-dependent ASVs are present in every mixture. Bars show the number of ASVs that displayed strong, weak, or dose-dependent colonization in each set of mixtures. ASVs were counted as present if they were detected in the 1:1 mixture at the fifth passage. (D) Total relative abundance of each type of colonizer in each set of mixtures after five passages. Only the 1:1 mixture is shown. Black lines indicate distinct ASVs. Relative abundances were averaged across the three inoculation replicates. (E) Resident ASVs are present at similar relative abundances in both parent communities. Lines are colored as in (A) for a representative ASV from one mixture. (F) Noisy ASVs show large, nonmonotonic fluctuations in relative abundance between adjacent mixture ratios. Lines are colored as in (A) for a representative ASV from one mixture. (G) Colonization behavior is not associated with ASV phylogeny for dose-dependent, strong, or weak colonizers. The heatmap shows the number of times each ASV exhibited dose-dependent (DD), strong (S), or weak (W) colonization at passage 5 in any mixture.

Many of the remaining ASVs exhibited dose-dependent colonization, with relative abundances that varied systematically across mixture ratios (n = 43/410, 10.5%, Fig. 2B and SI Appendix, Figs. S6 and S7C). Dose-dependent colonizers were present in every community mixture (Fig. 2C) and accounted for up to 20% of coculture relative abundance at the 1:1 mixture ratio (Fig. 2D). Most of the remaining abundance was composed of resident ASVs, which were present at approximately the same relative abundance in both parent communities and across all mixture ratios (n = 175/410, 42.7%, Fig. 2E and SI Appendix, Figs. S6 and S8A), or by noisy ASVs whose relative abundance changed nonmonotonically across mixture ratios (n = 160/410, 39.0%, Fig. 2F and SI Appendix, Figs. S6 and S8B). The prevalence of dose dependence among these varied colonization behaviors demonstrates that propagule size can influence the colonization of a sizeable fraction of species in each community during coalescence, even when communities show substantial deviations from neutral composition (Figs. 1E and 2 C and D). Intriguingly, most ASVs present above the limit of detection exhibited distinct colonization patterns across cocultures and relative to closely related taxa (Fig. 2G, SI Appendix, Fig. S9, and Materials and Methods), suggesting that these patterns were influenced more strongly by community context than by species identity or phylogeny. Taken together, these behaviors reveal that species can exhibit varying levels of dose dependence based on community composition.

Transient Dose Dependence Arises from High Niche Overlap in a Consumer–Resource Model.

We next sought to determine how the combination of neutral and nonneutral forces in each mixture leads to dose-dependent colonization patterns. In the D1/D2 cocultures, we identified three dose-dependent Enterococcus and Lactococcus (both in the order Lactobacillales) colonizers from the D1 parent community. The relative abundances of these ASVs were approximately anticorrelated with those of another dose-dependent Lactococcus from the D2 community (Fig. 3A). This observation led us to hypothesize that dose dependence could arise from competition between phylogenetically related taxa that occupy overlapping niches.

Fig. 3.

A consumer–resource model predicts transient dose dependence. (A) ASVs from the Enterococcus and Lactococcus genera showed dose dependence during community coalescence. Enterococcus faecalis, Enterococcus casseliflavus, and Lactococcus garvieae were isolated from the D1 community, and Lactococcus lactis was isolated from the D2 community. Gray lines represent both replicates of predicted neutral relative abundances for each ASV after five passages, and colored lines represent three replicates of experimental relative abundances after five passages. (B) Schematic of a simple consumer–resource model in which two species compete neutrally (equal consumption rates $R_{1,ab}=R_{2,ab}=1$) for resource $\mathit{ab}$, and resources $a$ and $b$ are unique to species 1 and 2, respectively (Materials and Methods). (C) The dose dependence of both species in the consumer–resource model increases as the niche overlap of species 1 with species 2 (${\gamma }_{1,2}$) increases. Dose dependence also decreases from passage 3 (dotted line) to passage 5 (solid line). Colored points highlight three values of ${\gamma }_{1,2}$ for which the relative abundances over mixture ratios of each species are shown at Right. The dose dependence of each species is defined as the magnitude of the ratio of its relative abundances, after passaging, from the starting mixture ratios of 1,000:1 and 1:1,000. (D) The species that is the weaker competitor for the shared resource shows stronger dose dependence. Lines show species relative abundances in the consumer–resource model after three (dotted line) and five (solid line) passages at three ratios of the consumption rates of the shared resource $\mathit{ab}$ by the two species ($R_{2,ab}/R_{1,ab}$), with fixed ${\gamma }_{1,2}=0.975$. The dose dependence of both species decreases with increasing $R_{2,ab}/R_{1,ab}$, but the dose dependence of species 1, the weaker competitor for the shared resource, decreases more slowly than that of species 2 and therefore remains higher at all three values. Dose dependence is defined as in (C). (E) The dose dependence of both species declines over passages in the two-species consumer–resource model, even when niche overlap (${\gamma }_{1,2})$ is high. The two species consume the shared resource at equal consumption rates ($R_{1,ab}=R_{2,ab}=1$). Dose dependence is defined as in (C).

We used a simple consumer–resource model (19) to investigate the conditions under which competition for shared resources could give rise to dose-dependent colonization dynamics (Materials and Methods and SI Appendix). We first studied a system in which two species compete for a set of three common “resources” $a$, $b$, and $\mathit{ab}$ with initial concentrations $y_a$, $y_b$, and $y_{\mathit{ab}}$, respectively (Fig. 3B). In this model, each resource represents a coarse-grained grouping of many substitutable nutrients, enabling us to focus on the effects of niche overlap on colonization dynamics (SI Appendix). Resources $a$ and $b$ are consumed exclusively by species 1 and 2, respectively, and the third resource, $\mathit{ab}$, is consumed by both species. We initially focused on the regime in which species 1 and 2 consume ab at equal rates ($R_{1,ab}=R_{2,ab}$); when species 1 and 2 are identical ($y_a=y_b=0$ and $R_{1,ab}=R_{2,ab}$), this consumer–resource model matches the expectations of neutral ecological theory.

To investigate how competition between species 1 and 2 would affect the colonization behavior of these two species, we varied the niche overlap, defined as ${\gamma }_{1,2}=\frac{y_{\mathit{ab}}}{y_a+y_{\mathit{ab}}}$, from 0.025 to 1 while maintaining equal concentrations of unique resources for each species ($y_a=y_b$) and a constant total resource supply ($y_a+y_b+y_{\mathit{ab}}=1$, SI Appendix). We initialized the abundances of the two species at ratios from 1,000:1 to 1:1,000, as in our community coalescence experiments, and we performed simulations of the model to investigate how species abundances would change over five passages. To quantify the dose dependence of a species, we calculated the magnitude of the ratio of its relative abundances at the 1,000:1 vs. 1:1,000 mixture ratios.

In this system, the dose dependence of each species increased with the amount of overlap between their niches (Fig. 3 C, Left) and was equal in magnitude for both species. For example, when niche overlap was relatively low (${\gamma }_{1,2}=0.75$), the two species rapidly reached consistent relative abundances across mixture ratios (Fig. 3C). By contrast, when niche overlap was high (${\gamma }_{1,2}=0.975$), the relative abundance of each species varied by ~13-fold between the 1,000:1 and 1:1,000 mixture ratios after three passages (Fig. 3 C, Right), consistent with the magnitude of dose dependence that we often observed during community coalescence (Figs. 2B and 3A). Similar levels of dose dependence occurred even when the consumption rate of the shared resource varied by ~10% between the two species (Fig. 3 D, Left), although we observed stronger dose dependence for the species that consumed the shared resource at the lower rate (Fig. 3 D, Right and SI Appendix, Figs. S10A, S11, and S12, respectively), since its relative abundance changed by a greater order of magnitude over mixture ratios. These results show that dose-dependent colonization dynamics can arise when the growth of two species is dominated by their consumption of shared resources.

Although both species showed dose dependence when niche overlap was high, the magnitude of this dose dependence declined over time (Fig. 3 C–E). For example, when ${\gamma }_{1,2}=0.975$ and $R_{1,ab}=R_{2,ab}$, the dose dependence of species 1 declined from ~13-fold to ~8-fold between passage three and passage five (Fig. 3C). This decline was consistent with the decreases in dose dependence over passages that we observed experimentally for some ASVs during community coalescence (SI Appendix, Fig. S10B). Notably, in our model, as long as species 1 and 2 were not identical (${\gamma }_{1,2}\ne 1$), the two species always eventually reached a stable, equilibrium abundance that was consistent across mixture ratios (Fig. 3E), as expected in simple consumer–resource models (19). Nonetheless, our investigation shows that high niche overlap can slow a community’s convergence to equilibrium, creating transient, dose-dependent colonization dynamics.

Strain Isolates Show Less Dose Dependence in Pairwise Mixtures Than During Community Coalescence.

Based on the predictions of our consumer–resource model, we hypothesized that the three dose-dependent Enterococcus and Lactococcus ASVs from the D1 parent community had high levels of niche overlap with the Lactococcus ASV from the D2 community that would lead them to show dose-dependent colonization dynamics in pairwise mixtures as well as during community coalescence. To test this hypothesis, we isolated representatives of these ASVs (SI Appendix, Table S2). We identified the strains from the D1 community as E. faecalis, E. casseliflavus, and L. garvieae, and the strain from the D2 community as L. lactis (Fig. 4 A, Top). For comparison, we also isolated two additional strains from the D1 community that did not exhibit dose dependence in the D1/D2 cocultures, which we identified as Bacteroides fragilis and Parabacteroides goldsteinii (Fig. 4 A, Top, and SI Appendix, Fig. S10 C and D).

Fig. 4.

Partial and transient dose dependence in pairwise strain mixtures. (A) Growth rates of Lactobacillales and Bacteroidales strains isolated from communities D1 and D2 formed two distinct clusters that corresponded to taxonomic order. Curves (Bottom Left) represent the average of nine replicates of blank-subtracted growth curves in monoculture after smoothing over a 30-min window, with shaded regions around each line representing one standard error (SE) above and below the average. Average instantaneous growth rates (Bottom Right) were calculated from blank-subtracted growth curves (Materials and Methods). (B) Experimental design of pairwise strain mixtures: strains were mixed in triplicate at ratios ranging from 1,000:1 to 1:1,000. The resulting cocultures were passaged five times, corresponding to ~40 generations. (C) No dose dependence was observed in a pairwise coculture of strains from distinct taxonomic orders. Strain relative abundances in cocultures of B. fragilis and L. lactis. Lines are colored as in Fig. 3A. Data from passages 1, 3, and 5 are shown. Neither strain was classified as dose dependent at any passage using the methods described in SI Appendix, Fig. S6. (D) Strains from the same taxonomic order showed transient dose dependence in pairwise coculture. Strain relative abundances in cocultures of B. fragilis and P. goldsteinii. Lines are colored as in Fig. 3A. Data from passages 1, 3, and 5 are shown. B. fragilis was classified as dose dependent at passage 1, but not at passages 3 or 5, using the methods described in SI Appendix, Fig. S6, while P. goldsteinii exhibited dose dependence at all passages. (E) Strain relative abundances in cocultures of E. casseliflavus and L. lactis. Lines are colored as in Fig. 3A. Data from passages 3 and 5 are shown. Both strains were classified as dose dependent at passage 3, but not at passage 5, using the methods described in SI Appendix, Fig. S6. (F) Strain relative abundances in cocultures of L. garvieae and L. lactis. Lines are colored as in Fig. 3A. Data from passages 3 and 5 are shown. Both strains were classified as dose dependent at passage 3, but not at passage 5, using the methods described in SI Appendix, Fig. S6. (G) Strain relative abundances in co-cultures of E. faecalis and L. lactis. Lines are colored as in Fig. 3A. Data from passage 3 and 5 are shown. E. faecalis was not classified as dose dependent at either passage using the methods described in SI Appendix, Fig. S6, while L. lactis exhibited dose dependence at passage 3 but not at passage 5.

We quantified the growth rates of each strain in monoculture and identified two groupings of growth dynamics, corresponding to the Lactobacillales (Enterococcus and Lactococcus) and the Bacteroidales (Bacteroides and Parabacteroides). Despite the similarities between closely related strains, small differences between strains in the same taxonomic order suggested the potential for distinct profiles of resource utilization (Fig. 4 A, Bottom). To investigate the relationship between niche overlap and dose dependence, we performed five mixtures of strain pairs at ratios from 1,000:1 to 1:1,000 (Fig. 4B). When we mixed B. fragilis (order Bacteroidales) with L. lactis (order Lactobacillales), both strains reached a consistent relative abundance across mixture ratios after a single passage (Fig. 4C), suggesting that distantly related strains, which have dissimilar growth dynamics, show minimal dose dependence.

In contrast, when we mixed four pairs of strains from the same taxonomic order, we consistently observed dose-dependent dynamics after three passages (Fig. 4 D–G). However, after five passages, the dose dependence of nearly all strains was strongly reduced (Fig. 4 E–G), except for P. goldsteinii in the B. fragilis/P. goldsteinii cocultures (Fig. 4D). For the Lactobacillales, the transient dose dependence in pairwise mixtures contrasted with their behavior in the D1/D2 cocultures, in which dose dependence was maintained even after five passages (Fig. 3A). Although similar growth dynamics are associated with stronger dose dependence in these pairwise mixtures, this discrepancy suggests that pairwise strain mixtures rarely exhibit the same magnitude and duration of dose dependence that we observed during coalescence of more complex communities.

The Presence of a Strong Competitor Increases Dose Dependence in a Consumer–Resource Model.

To investigate the discrepancy in dose dependence between pairwise mixtures and community coalescence, we used our consumer–resource model to investigate how colonization dynamics would be affected by the addition of a third species that has niche overlap with species 1. We introduced a third species into the model that consumes a unique coarse-grained resource grouping $c$, which is supplied at the beginning of each passage at the same concentration as the resource specific to species 2 ($y_c=y_b$, Fig. 5A and SI Appendix, Figs. S13–S15). Species 3 also competes with species 1 for some of the nutrients within the coarse-grained grouping $a$, dividing this grouping into resource $a^{\prime }$, the nutrients that remain exclusive to species 1 when species 3 is present, and resource $\mathit{ac}$, nutrients shared between species 1 and 3. We define $\beta =\frac{y_{\mathit{ac}}}{y_{\mathit{ac}}+y_{a^{\prime }}}$ as the niche overlap between species 1 and 3 and $r=\frac{R_{3,ac}}{R_{1,ac}}$ as the relative rate at which species 3 consumes the shared resource $\mathit{ac}$ compared to species 1. To investigate the effects of competition between species 1 and 3 on the colonization dynamics of species 1 and 2, we varied $\beta$ while keeping $r$ constant (Fig. 5B) and vice versa (SI Appendix, Figs. S10E and S13–S15), initializing the abundances of species 1 and 2 at a range of mixture ratios from 1,000:1 to 1:1,000 as before. We set the initial abundance of species 3 equal to that of species 2 to represent them being members of the same community. We then performed simulations of the model to investigate how species abundances changed over five passages.

Fig. 5.

Dose dependence increases in mixtures with more diverse communities. (A) Modification to the consumer–resource model in Fig. 3B in which a third species competes with species 1 for a fraction of the coarse-grained nutrients within $a$, dividing this grouping into resource $a^{\prime }$, the nutrients that remain exclusive to species 1, and resource $\mathit{ac}$, nutrients shared between species 1 and 3. Species 3 consumes resource $\mathit{ac}$ with relative consumption rate $r=\frac{R_{3,ac}}{R_{1,ac}}$ compared to species 1, imposing a niche overlap $\beta =\frac{y_{\mathit{ac}}}{y_{\mathit{ac}}+y_{a^{\prime }}}$ between species 1 and 3. (B) The dose dependence of species 1 after five passages increases with increasing niche overlap between species 1 and 3 ($\beta$) at constant $r=10$, while the dose dependence of species 2 is largely unaffected. Dose dependence is defined as in Fig. 3C. (C) L. lactis shows stronger dose dependence at passage 3 when cocultured with both E. faecalis and E. casseliflavus than when cocultured with either strain alone. Strain relative abundances in cocultures of E. faecalis, E. casseliflavus, and L. lactis. Data from passages 3 and 5 are shown. L. lactis was classified as dose dependent at passage 3, but not at passage 5, using the methods described in SI Appendix, Fig. S6http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials, while neither E. faecalis nor E. casseliflavus exhibited dose dependence at either passage. (D) Dose dependence is stronger in strain-community cocultures than in pairwise cocultures. Strain relative abundances after five passages in cocultures of the D1 community and L. lactis. E. casseliflavus and L. lactis were classified as dose dependent using the methods described in SI Appendix, Fig. S6http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials. (E) Strain relative abundances after five passages in cocultures of L. garvieae and the D2 community. L. garvieae and L. lactis were classified as dose dependent using the methods described in SI Appendix, Fig. S6http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials. (F) Strain relative abundances after five passages in cocultures of E. casseliflavus and the D2 community. E. casseliflavus was classified as dose dependent using the methods described in SI Appendix, Fig. S6http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials. (G) Strain relative abundances after five passages in cocultures of E. faecalis and the D2 community. No strains were classified as dose dependent using the methods described in SI Appendix, Fig. S6http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials. Lines in (C–G) are colored as in Fig. 3A. Strain relative abundances in the cocultures shown in (D–G) after three passages are shown in SI Appendix, Fig. S16http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials, and relative abundances in the cocultures in (E–G) after eight passages are shown in SI Appendix, Fig. S17http://www.pnas.org/lookup/doi/10.1073/pnas.2322440122#supplementary-materials.

In this three-species model, species 1 showed higher dose dependence as its niche overlap with species 3 increased (Fig. 5B). When species 3 had minimal niche overlap with species 1 ($\beta =0.1$, Fig. 5B) or was a weak competitor for their shared resource ($r\ll 1$, SI Appendix, Fig. S10E), species 1 showed similar levels of dose dependence as when species 3 was absent (Fig. 3C), as expected. However, when niche overlap was high ($\beta >0.5$) and species 3 was a strong competitor for resource $\mathit{ac}$ ($r\gg 1$), the dose dependence of species 1 increased substantially, consistent with a scenario in which it is outcompeted by species 3 for the use of their shared resource $\mathit{ac}$ and relies more on the resource shared with species 2 (Fig. 5B and SI Appendix, Fig. S10E). At the same time, the dose dependence of species 2 was essentially unaffected by the degree of niche overlap between species 1 and 3, showing that the addition of a new species does not necessarily alter the colonization behavior of all community members (Fig. 5B and SI Appendix, Fig. S10E). Together, these results predict that transient, dose-dependent colonization dynamics will be longer-lasting and more pronounced in diverse communities, in which competition among multiple species limits the unique resources available for any individual species.

Dose Dependence Is Enhanced in Mixtures with More Diverse Communities.

To test the prediction that dose dependence should increase with community diversity, we mixed the L. lactis strain with a two-species mixture of equal volumes of E. faecalis and E. casseliflavus at ratios from 1,000:1 to 1:1,000. Consistent with our prediction, after three passages, the degree of dose dependence exhibited by L. lactis increased to ~200-fold (Fig. 5C) from ~60-fold in pairwise cocultures (Fig. 4 E and G), while neither E. faecalis nor E. casseliflavus exhibited dose-dependent colonization (Fig. 5C). After five passages, none of the three strains exhibited dose dependence (Fig. 5C). These results support our model’s prediction that the addition of a new species can enhance dose dependence for another species by reducing resources that are exclusive to either species, even if dose dependence still declines more rapidly than during coalescence of complex communities.

To investigate whether introducing strains into a more diverse community would increase the magnitude and duration of dose dependence for a focal species, we performed a set of strain-community mixtures in which we mixed each of the four Lactobacillales strains into either the D1 or D2 community at ratios from 1,000:1 to 1:1,000 (Fig. 5 D–G and SI Appendix, Figs. S16 and S17). We introduced each strain into the community from which it did not originate. As predicted, a subset of strains in most strain-community mixtures showed dose-dependent colonization even after five passages (Fig. 5 D–F), in contrast to the pairwise and three-strain mixtures. For example, L. lactis, L. garvieae, and E. casseliflavus each exhibited dose dependence after five passages when introduced into community D1, D2, and D2, respectively, varying by ~10-fold in relative abundance across mixture ratios (Fig. 5 D–F). In two of these three mixtures, the introduced strain showed complementary dose dependence with at least one other Lactobacillales, although the identity of this complementary strain varied with community context (Fig. 5 D and E). Even though dose dependence lasted longer in the strain-community mixtures compared to the pairwise and three-strain mixtures, it still declined between the fifth and eighth passages of the strain-community mixtures, except when E. casseliflavus was introduced into the D2 community (SI Appendix, Fig. S17). Together, these results demonstrate that the dose dependence of a focal strain can increase in mixtures with more diverse communities.

Resource-Consumption Profiles Explain Dose Dependence in Strain and Community Mixtures.

We sought to directly test our model’s prediction that stronger dose dependence arises when pairs of strains or communities have high niche overlap and similar rates of shared resource consumption. To quantify niche overlap, we compared the growth of each of the four Lactobacillales and two Bacteroidales strains in fresh medium to their growth in the spent medium of each other strain and of the D1 and D2 communities (Materials and Methods, Fig. 6 A, Left, and SI Appendix, Fig. S18A and Table S3). We calculated niche overlap for a strain in a particular spent medium as $1-\frac{\mathrm{m}\mathrm{a}\mathrm{x}({\mathrm{O}\mathrm{D}}_{\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{m}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{u}\mathrm{m}})}{\mathrm{m}\mathrm{a}\mathrm{x}({\mathrm{O}\mathrm{D}}_{\mathrm{f}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{m}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{u}\mathrm{m}})}$, where the maximum was computed over 24 h of growth.

Fig. 6.

Resource-utilization profiles predict dose dependence in strain and community mixtures. (A) Niche overlap was elevated among pairs of strains from the same taxonomic order. Left: niche overlap is quantified as one minus the ratio of the maximum OD₆₀₀ of the focal strain growing in the spent medium of the comparison strain or community after 24 h compared with fresh mBHI (Materials and Methods). Right: Colors show niche overlap between pairs of strain isolates/communities based on growth in spent media. Pairs of strains shown in Fig. 4 are highlighted. (B) Resource consumption rates predict relative dose dependence in pairwise cocultures. Colors show consumption rates of unique and shared resources for pairs of strains shown in Fig. 4. Metabolomic features were classified as unique or shared resources if they were depleted by ≥10,000-fold after 48 h in one or both strains, respectively (Materials and Methods). Coarse-grained nutrients were labeled using our model notation, in which a and b are consumed exclusively by species 1 and species 2, respectively, and ab is consumed by both strains. Consumption rates were quantified by calculating the median metabolomic feature fold change after 4 h for relevant features. (C) Strains that are weaker competitors for shared resources show stronger dose dependence in pairwise coculture. Colors show the dose dependence of strain pairs shown in Fig. 4 after three passages. For each species in a pair, dose dependence was calculated as the magnitude of the ratio of the strain’s relative abundance at the 1,000:1 vs. 1:1,000 mixture ratios, as in Fig. 3C.

Niche overlap was much higher for pairs of strains from the same taxonomic order (Fig. 6 A, Right), consistent with our finding that more closely related strains show higher levels of dose dependence in pairwise coculture (Fig. 4 C–G). Within a taxonomic order, the median niche overlap was 0.97 (n = 20, SI Appendix, Fig. S18B), while the median niche overlap was 0.60 between strains in different taxonomic orders (n = 16, Wilcoxon rank-sum one-sided test: P = 2.2 × 10⁻⁷). For example, L. lactis (Lactobacillales) had a niche overlap of 0.19 with B. fragilis (Bacteroidales), and B. fragilis had a niche overlap of 0.61 with L. lactis (Fig. 6 A, Right), in line with the observation that both strains reached a consistent relative abundance across mixture ratios after just a single passage in pairwise coculture (Fig. 4C). In contrast, L. garvieae had a niche overlap of 0.98 with L. lactis, and L. lactis had a niche overlap of 0.99 with L. garvieae, on par with the negligible amount of growth we observed when strains were grown in their own spent medium (Fig. 6 A, Right). These high niche overlaps are consistent with the dose dependence we observed in pairwise cocultures of L. garvieae and L. lactis after 3 passages (Fig. 4F). Moreover, all strains had high niche overlaps (0.93 to 0.99) with the D1 and D2 communities (Fig. 6 A, Right), in accordance with our observation that dose dependence was high in strain-community mixtures. Together, these results support our model’s prediction that dose dependence is strong in strain and community mixtures when niche overlap is high.

While our estimates of niche overlap may explain why dose dependence is generally low in pairwise cocultures of distantly related strains, we observed variation in dose dependence among pairwise mixtures of closely related strains (Fig. 4 D–G). For instance, in the L. garvieae/L. lactis coculture, L. lactis showed twice as much dose dependence as L. garvieae after three passages (Fig. 4F), even though each strain had ≥ 0.98 niche overlap with the other (Fig. 6 A, Right). This observation suggests that niche overlap alone was not sufficient to explain the levels of dose dependence we observed. Based on our model, we hypothesized that differences in resource consumption rates might account for this variation in dose dependence. To quantify consumption rates of the shared resources, we performed untargeted metabolomics on the spent medium of each strain after 4 and 48 h of growth (Materials and Methods and SI Appendix, Fig. S19). We classified metabolomic features as consumed if they were depleted ≥ 10,000-fold after 48 h compared to their peak heights in fresh medium (SI Appendix, Fig. S19 A and B), and for each pair of strains, we classified features as shared resources if they were consumed by both strains. We then quantified the consumption rate of the shared resources by calculating the median fold change of these features in the spent medium of each strain after 4 h (Fig. 6 B, Left and SI Appendix, Fig. S19E and Table S4).

Based on our two-species model, we predicted that the strain that is the weaker competitor for the shared resource would show greater dose dependence in pairwise coculture (Fig. 3D). Consistent with this prediction, we found that L. garvieae consumed its shared resources with L. lactis at ~1.6 times the rate of L. lactis (Fig. 6 B, Right), which may explain why L. lactis shows stronger dose dependence in pairwise coculture after three passages even though the two strains have similar levels of niche overlap with one another (Figs. 4F and 6 A, Right). In the B. fragilis/P. goldsteinii, E. casseliflavus/L. lactis, and E. faecalis/L. lactis strain pairs, we also observed that the strain that was the weaker competitor for shared resources showed stronger dose dependence (Fig. 6 B, Right, and C). This difference in dose dependence was especially pronounced in the E. faecalis/L. lactis coculture, in which L. lactis consumed shared resources at approximately half the rate of E. faecalis and showed ~16-fold stronger dose dependence after three passages (Fig. 6 B, Right, and C). Although we did not measure the resource consumption rates of the D1 and D2 communities, we speculate that these rates may also help explain why dose dependence is enhanced in strain-community mixtures compared to pairwise strain mixtures. Together our findings demonstrate how resource consumption rates can be combined with estimates of niche overlap to explain when dose dependence is likely to arise in strain and community mixtures.

Discussion

In this study, we quantified the effect of propagule size on hundreds of simultaneous colonization events during community coalescence of gut bacteria across a range of therapeutically relevant mixture ratios, and we compared the resulting cocultures to neutral ecological predictions. Community composition deviated systematically from neutral predictions, but initial population size had a strong effect on colonization for a subset of species in each mixture, even after ~40 generations of competition. Using a consumer–resource model, we showed that this dose-dependent colonization can arise from competition for shared resources, especially in diverse communities. We showed experimentally that dose dependence can be predicted using growth in spent media and metabolomic profiles of strain isolates and that the effect of initial propagule size increases in more diverse communities.

Our model provides an explanation for how dose-dependent colonization dynamics, which are typically associated with neutrality, can occur in a community dominated by nonneutral competition. When considering neutral theory alone (18, 50), all species in a community are expected to show lasting dose dependence, whereas simple models of resource competition alone (51, 52) dictate that all species should reach the same relative abundance regardless of initial population size. However, we observed both nonneutral (Fig. 2A) and strikingly neutral-like (Fig. 2B) colonization dynamics in community cocultures. Our model explains this co-occurrence by showing that dose-dependent behavior is neither an intrinsic property of a species, nor of the community as a whole. Instead, dose-dependent colonization dynamics are emergent phenomena (53–59) that arise from nonneutral, competitive interactions among subsets of community members, which may explain why previous studies have come to contradictory conclusions about how propagule size affects colonization success (5, 20, 26, 30, 33, 34, 37).

Another feature of our model is that dependence on propagule size is transient: communities eventually converge to a stable equilibrium, a characteristic shared with other simple consumer–resource models (19). Notably, we show that this transient dose dependence can emerge even in the absence of more complex phenomena, such as resource fluctuations (52, 60), spatial structure (61, 62), and environmental modification (7, 14), which can create alternative stable states that depend more strongly on initial conditions (63, 64). In addition, factors like phage (65, 66), self-inhibition (67), toxin production (67, 68), and quorum sensing (69) can cause growth dynamics to vary with population size, potentially modifying the degree of dose dependence. Modeling frameworks that incorporate these phenomena, along with other variables like resource essentiality (70–72) or other metabolic traits like acid production (73), may more accurately capture colonization dynamics in natural communities. However, even under our simple assumptions, both our model and experiments show that transient, dose-dependent colonization dynamics can persist for ~40 generations. The time scales of these transient states (74–77) may be long enough for introduced species to affect community composition and function (75, 78, 79), induce shifts between alternative states (60, 80), and alter the outcomes of subsequent species introductions (14, 81–85)—even if the initial species introductions are ultimately unsuccessful. Our model also sheds light on how the initial conditions of species introductions can influence the time scales over which communities approach equilibrium, an important consideration in modeling approaches that seek to infer community composition or function from these equilibrium states (86, 87).

We argue that dose-dependent colonization dynamics emerge when community members have high niche overlap and consume shared resources at similar rates, slowing the community’s convergence to equilibrium. We showed experimentally that resource-utilization profiles can explain levels of dose dependence in our pairwise strain mixtures, as our consumer–resource model leads us to expect. Notably, measurements of both niche overlap and resource consumption rates were necessary to interpret the dynamics of pairwise mixtures (Fig. 6), highlighting the value of metabolomic time courses for understanding community dynamics. However, despite the success of these approaches in explaining qualitative trends, it remains challenging to make precise, quantitative predictions about dose dependence, and further work is needed to define and measure resource-utilization profiles, particularly when strains are grown in complex communities, and to link these measurements to model predictions. For instance, we observed quantitative differences in our estimates of niche overlap from growth in spent media versus untargeted metabolomics (Materials and Methods and SI Appendix, Fig. S19D), discrepancies that likely reflect the fact that metabolomic features vary in their abundance, substitutability, and contributions to growth (88).

Despite these remaining questions, the ability of resource-utilization profiles to predict dose dependence suggests that niche structure in a community is largely determined by the metabolites available and the species that can consume them. Modifying the resource landscape by adding or subtracting nutrients would likely change patterns of niche overlap and the species pairs that show dose dependence, an effect analogous to adding or subtracting species. Future studies that combine controlled colonization experiments with targeted perturbations like nutrient manipulations and genetic knockouts will improve our ability to predict the outcomes of species introductions.

Materials and Methods

Derivation of Top–Down In Vitro Communities.

Stool samples were collected as a part of a study (89) approved by the Stanford University Institutional Review Board under Protocol 54715. We derived in vitro communities from preantibiotic samples collected by eight healthy adults (A1, A2, B1, B2, C1, C2, D1, and D2; SI Appendix, Table S1) from four household pairs. We used a Basque Engineering CXT 353 Benchtop Frozen Aliquotter to drill cores from frozen stool samples, and we generated in vitro communities following methods established in our previous studies (46, 47). First, a 50 mg fragment of each stool core was resuspended in 500 µL of filter-sterilized PBS. Next, 20 µL of the stool resuspension were added in triplicate to 180 µL of Brain Heart Infusion (BHI, BD Biosciences 237200) supplemented with 0.2 mg/mL L-tryptophan, 1 mg/mL L-arginine, 0.5 mg/mL L-cysteine, 5 µg/mL vitamin K, 1 µg/mL hemin, and 5 g/mL mucin (referred to here as mBHI+mucin) in a flat-bottom 96-well plate (Greiner Bio-One 655161). All experiments in this study that involved bacterial growth were performed in an anaerobic chamber (Coy Instruments). After inoculation, communities were grown at 37 °C and passaged 15 times, once every 48 h at a 1:200 dilution, by transferring 1 µL of saturated culture into a new 96-well plate containing 199 µL of fresh mBHI+mucin. In vitro communities grew to ~10⁸ CFUs in each 200 µL community, indicating that ~10⁶ CFUs are used to seed each new passage (SI Appendix, Table S2).

Community Coalescence Experiments.

We thawed glycerol stocks of one replicate of each community from passage 15 and reinoculated these cultures in mBHI+mucin. These communities were grown at 37 °C for 48 h. To create community mixtures, we first generated a 10-fold dilution series (1/10, 1/100, and 1/1,000) in triplicate from the single replicate of each of the unmixed parent communities. To generate each pairwise mixture, we added 1 µL of one community at the appropriate dilution to 1 µL of an undiluted parent community and 198 µL of fresh mBHI+mucin. We mixed communities in eight combinations (A1/A2, B1/B2, C1/C2, D1/D2, A1/B1, B1/C1, C1/D1, and D1/A1; SI Appendix, Table S1) at seven ratios (1,000:1, 100:1, 10:1, 1:1, 1:10, 1:100, and 1:1,000) and passaged each mixture in triplicate, resulting in 168 coculture communities (8 mixtures × 7 ratios × 3 replicates). We also inoculated 2 µL of each undiluted parent community into 198 µL of mBHI+mucin in duplicate, comprising 16 unmixed control parent communities. The resulting communities were grown at 37 °C and passaged every 48 h five times at a 1:200 dilution. This dilution allows for ~7.6 doublings per passage, corresponding to ~40 generations of bacterial growth in five passages.

16S rRNA Gene Sequencing.

We used the Qiagen DNeasy Ultraclean 96 Microbial Kit (Qiagen 10196-4) to extract DNA from 50 µL of saturated culture collected after three and five passages. We performed 16S rRNA gene amplicon sequencing targeted to the V4 region using 27 cycles of PCR with primers modified from the Earth Microbiome Project spanning 515F–806R (90, 91) at an annealing temperature of 50 °C. Illumina sequencing adapters were attached using 10 cycles of PCR with an annealing temperature of 54 °C.

Raw sequencing reads were annotated and demultiplexed using UMI-tools (92), and primer and adapter sequences were trimmed using cutadapt (93). DADA2 (49) was used to filter and truncate reads, assign amplicon sequence variant (ASV) taxonomy based on the SILVA (release 138) reference database (94), and obtain a phylogeny of all ASVs.

Comparison of Community Composition to Theoretical Predictions.

We computationally generated a set of theoretical community mixtures according to the predictions of ecological neutral theory (SI Appendix) (18). Using four summary statistics, we compared the experimental relative abundance of each ASV to theoretical predictions and categorized ASVs into six colonization patterns: low abundance, noisy, resident, dose dependent, strong colonizer, or weak colonizer (SI Appendix, Fig. S6).

Consumer–Resource Model.

We implemented a standard consumer–resource model (19):

$$\frac{d{X}_i}{\mathit{dt}}=X_i{\displaystyle \sum _{\mu =1}^M}{R}_{i,\mu }{Y}_{\mu },$$

$$\frac{d{Y}_{\mu }}{\mathit{dt}}=-Y_{\mu }{\displaystyle \sum _{i=1}^N}{R}_{i,\mu }{X}_i,$$

where $X_i$ denotes the absolute abundance of species $i$, $Y_{\mu }$ the amount of resource $\mu$, $R_{i,\mu }$ the consumption rate of resource $\mu$ by species $i$, $M$ the total number of resources, and $N$ the total number of species. Species abundances were simulated in MATLAB (Mathworks, Inc.) using the ode45 solver until all resources were depleted ($t_f=\mathrm{10,000}$ arbitrary time units). After each such passage $b$ ($t=t_f$), the final abundances were diluted 1:200, such that at the beginning ($t=0$) of passage $b+1$ the abundance of species $i$ was

$$X_i^{b+1}\left(t=0\right)=\frac{1}{200}{X}_i^b\left(t=t_f\right),$$

and resource concentrations were refreshed such that at the beginning of passage $b+1$ the amount of resource $\mu$ was

$$Y_{\mu }^{b+1}\left(t=0\right)=y_{\mu }.$$

We initialized the abundances of the two species at a range of ratios from 1,000:1 to 1:1,000, as in our community coalescence experiments, and we simulated five passages for each community across a range of conditions (SI Appendix).

Quantification of Strain Isolate Growth Dynamics.

We acquired growth curves of strain isolates using an Epoch 2 Microplate Spectrophotometer (Biotek Instruments), following two passages of outgrowth from glycerol stocks (SI Appendix). We measured growth curves in mBHI rather than mBHI+mucin because mucin interferes with measurements of optical density. We measured the optical density at 600 nm (OD₆₀₀) of nine replicates of each strain every 3 min over 48 h of growth at 37 °C with continuous orbital shaking. We normalized the OD₆₀₀ of each strain by subtracting the average OD₆₀₀ of the first three readings from each well and replacing blank-subtracted values below 0.005 with 0.005. Instantaneous growth rate was quantified from blank-subtracted OD₆₀₀ values as d ln(OD₆₀₀)/dt using the gcplyr R package (95).

CFU Counts.

Following two outgrowth passages from glycerol stocks (SI Appendix), we plated 10-fold dilutions of the saturated cultures of our six strain isolates, as well as the D1 and D2 communities, on mBHI+mucin agar plates. We incubated the plates at 37 °C and counted colonies after 48 h.

Cocultures of Strain Isolates and Strain-Community Mixtures.

We performed a set of five pairwise strain mixtures (B. fragilis/L. lactis, B. fragilis/P. goldsteinii, E. casseliflavus/L. lactis, L. garvieae/L. lactis, and E. faecalis/L. lactis), one three-strain mixture (E. faecalis/E. casseliflavus/L. lactis), and four strain-community mixtures (D1/L. lactis, L. garvieae/D2, E. casseliflavus/D2, and E. faecalis/D2). We performed mixtures at ratios from 1,000:1 to 1:1,000 in triplicate, after one 48 h passage of outgrowth in mBHI+mucin, as described above for our community coalescence experiments. For the three-strain mixture, we added 0.5 µL of E. faecalis and 0.5 µL of E. casseliflavus to 1 µL of L. lactis, at the appropriate dilutions for each mixture ratio, to inoculate cocultures. As for community coalescence, strain cocultures, strain-community mixtures, and unmixed parent strains and communities were grown at 37 °C and passaged for 48 h five times at a 1:200 dilution. We followed the procedures described above to prepare 16S rRNA gene libraries for Illumina sequencing and infer ASV abundances (SI Appendix).

Growth in Spent Media.

Following one outgrowth passage from glycerol stocks (SI Appendix), we spun down saturated cultures of our six strain isolates and the D1 and D2 communities at 4,000×g for 10 min and filtered the supernatant through 0.2-μm polyethersulfone filters (Thermo Scientific 725-2520). We inoculated 1 μL of each saturated culture into 199 μL of either spent medium or fresh mBHI, and we measured the OD₆₀₀ of two replicates of each strain in each medium over 24 h of growth at 37 °C, as described above. We measured the pH of spent media after 48 h of growth using a SympHony SB70P Digital pH Meter (VWR, SI Appendix, Table S3).

Analysis of Metabolomic Data.

Untargeted LC–MS/MS metabolomics were performed as previously described (22) on spent media from cultures of the six Lactobacillales and Bacteroidales strains isolated from the D1 and D2 communities and from the communities themselves (SI Appendix). We first normalized and filtered the raw data to remove low-quality metabolomic features. We then calculated the fold change of each feature in the spent medium of each strain or community compared to the average blank-subtracted peak height across both replicates of the medium (mBHI+mucin) blanks. For each pair of strains or communities, we classified metabolomic features as either uniquely consumed by only one strain or community or shared between both, and we used these annotations to quantify niche overlap and resource consumption rates (SI Appendix).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Sequencing data is available through the NCBI SRA at BioProject PRJNA1212511 (96). The computer code that performs the analysis is available at https://github.com/DoranG1/dose-dependent-colonization (97). All other data are included in the manuscript and/or SI Appendix.