Abstract
Microbiome dynamics influence the health and functioning of human physiology and the environment. These dynamics are driven in part by interactions between large numbers of microbial taxa, making large-scale prediction and modeling a challenge. Here, we identify states and dynamical features relevant to macroscopic processes, such as infection in the human body and geochemical cycling in the oceans, by modeling the dynamics as stochastic motion on a potential energy-like landscape. We show that gut disease processes and marine geochemical events are associated with reproducible transitions between community states, defined as topological features of the landscape. We find a reproducible two-state succession during recovery from cholera in the gut microbiomes of multiple patients. Recurrence of the late disease state prolongs disease duration. We find evidence of dynamic stability in the gut microbiome of a human subject after experiencing diarrhea during travel, in contrast to residual instability in a second human subject after clinical recovery from Salmonella infection. Finally, we find the structure of marine Prochlorococcus communities in the western Atlantic and north Pacific oceans to smoothly vary with temperature and depth. However, annual water column cycling in the Atlantic drives periodic state transitions across depths. Our approach bridges the small-scale fluctuations in microbiome composition and large-scale changes in state and phenotype, improves analyses of how changes in community composition associate with phenotype without requiring experimental characterization of underlying mechanisms, and provides a novel assessment of microbiome stability and its relation to human and environmental health.
Importance Time series of microbial communities are difficult to analyze due to the large number of interacting taxa. We developed a novel analysis based on topology to detect compositional states and state transitions in microbial time series. Our method generalizes across biological systems and can identify gut microbiome dynamics associated with recovery from disease in multiple patients on the order of weeks, and marine bacterial dynamics driven by geochemical cycling on the order of years. We furthermore propose a novel definition of ecological stability that distinguishes between complete and incomplete recovery from infection in human gut microbiomes. Our method requires minimal assumptions regarding biological mechanisms. Overall, our analysis complements current methods for identifying key ecological processes in microbial communities, and suggests further developments in modeling that may improve prediction of microbial dynamics.