Geut Galai, Dafna Arbel, Keren Klass, Ido Grinshpan, Itzhak Mizrahi, Shai Pilosof 

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. In this context, the rumen microbiome is an excellent model system because each cow contains a physically confined microbial community, which is imperative for the host's livelihood and productivity. Employing the rumen microbiome of 1,012 cows across seven European farms as our model system, we constructed and analyzed farm-level co-occurrence networks to reveal underlying microbial interaction patterns. Within each farm, microbe co-occurrence was significantly transitive with a group structure that reflected functional equivalence in co-occurrence. Knowing the group composition (obtained with stochastic block modeling) in one farm provided significantly more information on the grouping in another farm than expected. Moreover, microbes strongly conserved co-occurrence patterns across farms (also adjusted for phylogeny). We further developed a meta-co-occurrence multilayer approach, which links farm-level networks, to test scale signatures simultaneously at the farm and inter-farm levels. Consistent with the comparison between groups, the multilayer network was not partitioned into clusters. This result was consistent even when artificially disconnecting farm-level networks. Our results show a prominent signal of processes operating across farms to generate a non-random yet similar co-occurrence patterns. Comprehending the processes underlying rumen microbiome assembly can aid in developing strategies for its manipulation to mitigate greenhouse gas emissions. More broadly, our results provide new evidence for the scale at which forces shape microbe co-occurrence. Finally, the hypotheses-based approach and methods we developed provide can be adopted in other systems to detect scale signatures in species interactions.