Scale-dependent signatures of microbial co-occurrence revealed via multilayer network analysis

Abstract

The composition of microbial communities underlies many ecosystem functions. Co-occurrence networks offer insights into the complexity of microbial interactions, particularly in highly diverse environments where direct observation is challenging. However, identifying the scale at which local and non-local processes structure co-occurrence networks remains challenging because it requires simultaneously analyzing network structure within and between local networks. Using the rumen microbiome of 1,012 cows in seven European farms, we built a multilayer network in which microbes (nodes) are linked within and between farms (layers), effectively creating a meta-co-occurrence network. At the local scale, we tested for a non-random transitive signature (the tendency of a microbe to close triangles) because transitivity indicates environmental similarities or interaction dependence. At the non-local scale, we tested for partner fidelity because if microbes tend to conserve their interaction partners more than expected by chance, co-occurrence is not determined locally. We further tested at which scale (farm vs. regional) the partition of microbes to modules of densely co-occurring microbes was prominent. Via comparison to shuffled networks, we discovered that, although core microbes appeared across the entire system, there were non-random signatures of local and non-local co-occurrence patterns. Microbes had transitive interactions and tended to conserved partners across farms. Microbes in four farms were clustered separately (each farm in a module), providing a signature for a strong but not complete influence of local processes. The genetic structure of cow populations partly explains modularity, indicating that environmental filtering is an important force shaping co-occurrence. Our results provide evidence for the scale at which interconnected co-occurrence networks are structured using a clear set of hypotheses. Our developed approach can also be used to explore local and non-local signatures in species interaction metanetworks or metacommunities.