Predictability of microbiome dynamics

Microbiome dynamics are both crucial indicators and drivers of human health, agricultural output, and industrial bio-applications. However, predicting microbiome dynamics is notoriously difficult because communities often show abrupt changes in species compositions. This is problematic because adverse effects, such as dysbiosis in human microbiomes cannot be anticipated. Here we show how drastic ecological community events, such as community collapse, can be quantified to forecast microbiome dynamics. We monitored 48 experimental microbiomes for 110 days (ca. 5,000 samples) and found that community-level events, including collapse and gradual compositional changes occurred depending on a defined set of environmental conditions. Specifically, major shifts in community structure among alternative transient/stable states could be forecasted based on energy landscape analyses from statistical physics14 and/or stability criteria of nonlinear dynamics. We further found that niche overlap, within metagenomic niche space, and positive feedback loops in cross-feeding networks could precede collapse from complex community structure to simpler states. Integrating these insights into early-warning signals of community dynamics will allow us to prevent catastrophic microbiome events, and to design synthetic multi-species systems with stable biological functions.