Environmental covariation leads to low dimensionality, increased competition, and predictability in microbiomes

Microbiomes are complex ecosystems wherein thousands of microorganisms spanning several kingdoms of life coexist. Observed covariation in species abundances suggests that the number of degrees of freedom that govern microbiome compositions might be substantially lower than the number of organisms. To identify the effective dimensionality of microbiomes, we developed a latent variable framework derived from the consumer/resource model. In this framework, a potentially small number of effective resources and species preferences towards those resources capture the compositional variation in microbiomes. Notably, our analysis of real microbiomes showed that the dimensionality varied substantially across communities.

We hypothesized that low dimensionality may arise from two features. One, if multiple organisms prefer the same resources (i.e. high competition), the abundances of these organisms may be correlated, leading to low dimensionality. Second, if resource flows are coupled to each other, species become effectively correlated through their preferences on correlated resources, leading to low dimensionality.

Using the consumer/resource model, we observed that increased competition among the species was not sufficient to explain the low dimensionality observed in real microbiomes. In contrast, correlation amongst resources was sufficient. Having identified a potential cause of the observed low dimensionality, we further investigated its effects. By varying the strength of covariation in the resources, we were able to generate realistic microbiome compositions with varying dimensionality. This allowed us to study how different properties of the microbiome relate to its dimensionality.

Specifically, we showed that low dimensional microbiomes were predictive of their environment, an observation that was also recapitulated in real data. Moreover, while increased competition did not lead to low dimensionality, low dimensional microbiomes had an emergent increase in competition. Using flux balance analysis, we showed that this trend was also recapitulated in real microbiomes. Therefore, low dimensionality and high species-competition in microbiomes can be understood as arising from correlations in the environment.