Ecological rules for the assembly of microbiome communities

Abstract

Humans and many other hosts establish a diverse community of beneficial microbes anew each generation. The order and identity of incoming symbionts is critical for health, but what determines the success of the assembly process remains poorly understood. Here we develop ecological theory to identify factors important for microbial community assembly. Our method maps out all feasible pathways for the assembly of a given microbiome-with analogies to the mutational maps underlying fitness landscapes in evolutionary biology. Building these "assembly maps" reveals a tradeoff at the heart of the assembly process. Ecological dependencies between members of the microbiota make assembly predictable-and can provide metabolic benefits to the host-but these dependencies may also create barriers to assembly. This effect occurs because interdependent species can fail to establish when each relies on the other to colonize first. We support our predictions with published data from the assembly of the preterm infant microbiota, where we find that ecological dependence is associated with a predictable order of arrival. Our models also suggest that hosts can overcome barriers to assembly via mechanisms that either promote the uptake of multiple symbiont species in one step or feed early colonizers. This predicted importance of host feeding is supported by published data on the impacts of breast milk in the assembly of the human microbiome. We conclude that both microbe to microbe and host to microbe interactions are important for the trajectory of microbiome assembly.

Fig 1. Mathematical models to simulate and map out community assembly.

(A) One method of exploring community assembly is to draw species one by one at random from a large pool then assess whether and how they establish a diverse community. However, this approach only gives a single example of assembly per community, and due to the vast number of possibilities, we do not know whether other assembly pathways are possible. (B) In our alternative approach, we randomly select a diverse community, then determine which of its subcommunities are feasible and all possible transitions between them. In this manner, we can map out all the pathways by which a given community may have assembled (bold line in B indicates assembly pathway observed in A). (C) Examining the characteristics of a community’s assembly map enables us to determine whether it will be able to assemble from scratch and whether it will assemble in a predictable manner.

Fig 2. Strongly interdependent communities are harder to assemble.

(A) Strong interactions (moving bottom to top) reduce the probability a community will be able to assemble via the random, sequential arrival of individual species. Symmetrical interactions (−/− and +/+) are also more of a barrier to assembly than asymmetrical ones, with the most robust assembly seen in the middle of the figure (high number of +/− interactions) moving left to right. (B) A reduced ability to assemble stems from low numbers of feasible subcommunities, which lead to splits in the assembly map, such that there is no single path from uncolonized to fully colonized (Fig 1C). (C, D) Treating the assembly map as a transition matrix allows us to calculate the probability of observing the community in a given state when subject to the continuous arrival and loss of taxa. In this illustration size of dot represents the probability of observing each subcommunity at any given time (here gamma / delta = 20). (E, F) Heatmaps illustrate the most likely community size for communities that are capable of assembling when (E) the rate at which species arrive is much greater than the rate at which they are lost ($\frac{\gamma }{\delta }$ = 20) or (F) when arrival and loss rates are similar ($\frac{\gamma }{\delta }$ = 2). In each case, even when communities are capable of assembling from scratch, they will not necessarily be able to in the face of continuous arrival and loss of taxa. The more interdependent, or more strongly interacting species are, the less diverse the microbiota is likely to be at any given time. For all analyses, climax communities are drawn with S = 10, connectivity C = 0.5, from 150 independent replicates, underlying data in S1 and S2 Data.

Fig 3. Strong, positive interactions increase the predictability of assembly.

(A) Increasing interaction strength, and the number of positive interactions, decreases the number of edges within the assembly map. (B) Strong and positive interactions also increase the incidence of secondary colonizers, who cannot establish within the community until other species are present. Together, these factors increase the overall predictability of community assembly. (C, D) Increased predictability stems in part from a lower number of feasible subcommunities. (E, F) The relationship between interaction strength and sign and predictability leads to a trade-off between the robustness and predictability of assembly. For all analyses, climax communities are drawn with S = 10, connectivity C = 0.5, from 150 independent replicates, underlying data in S3 Data.

Fig 4. Published data reveal predictable assembly dynamics within the preterm infant gut, driven by interspecies dependencies.

(A) Fitting data from 13 preterm infants sampled on a near daily basis for their first 6 weeks of life to the generalized Lotka–Volterra equations reveals a rich network of microbe–microbe interactions playing a strong and consistent role in shaping microbiome dynamics. (B) The preterm gut assembles in a robust and predictable manner, with specific taxa colonizing at different points in time. (C) More helpful taxa, determined by average effect on others, tend to colonize earlier in infant life (R² = 0.32, p = 0.02). (D) We quantify the predictability of individual pairs of genera by calculating the proportion of infants in which the focal genus only colonizes with or after the partner genus. (E) Genus pairs in which the focal taxon receives a strong benefit from its partner (interaction α_ij > 0.1) display predictable assembly dynamics, with the focal taxon significantly more likely to only colonize with or after its partner than expected by chance (average predictability score 0.73 > 0.5, one-sample permutation test p = 0.002). Underlying data in S4 Data.

Fig 5. Multiple arrivals and host feeding can each increase a community’s ability to assemble.

(A) Species arriving together rather than individually enables them to bypass the interspecies dependencies that would otherwise block community assembly. (B) These multiple arrivals increase the probability a community will be able to assemble. (C) However, multiple arrivals also increase the number of edges within the assembly map, reducing the predictability of assembly. (D) We capture variable host-feeding in our framework with a multilayer network composed of the assembly map for a community with and without supplemental nutrients provided by the host. If a subcommunity is feasible in both layers of the network, we assume that at that point in assembly, the host can add/remove nutrients without disrupting the current subcommunity. This enables continuous assembly pathways that span both network layers. (E) This host feeding can increase the probability of a community assembling, provided the feeding is removed once the community is sufficiently diverse. (F) It is beneficial for the host to provide additional nutrients early in assembly but not late, as these nutrients increase the feasibility of small subcommunities but reduce the feasibility of larger subcommunities. Here, for all results, climax communities are drawn with species number S = 10, connectivity C = 0.5, interaction strength sigma = 0.05, from 100 independent replicates, underlying data in S5 Data.