Modular gene interactions drive modular pan-genome evolution in bacteria

Depending on the scale of observation, bacterial genomes are both organized and fluid. While individual bacterial genomes show signatures of organization (e.g., operons), pan-genomes reveal genome fluidity, both in terms of gene content and order (synteny). Here we ask how mutational forces (including recombination and horizontal gene transfer) combine with selection and gene interactions to shape genome organization and variation both within and across strains. We first build an evolutionary simulation model to assess the impact of gene interactions on pan-genome structure. A neutral evolutionary model can produce transient co-segregation of initially linked genes but is vulnerable on longer time-scales to perturbing mutational events. In contrast, incorporation of modular gene fitness interactions can produce sustainable clusters of linked and co-segregating genes, with the network of co-segregation recapitulating the defined simulation ‘ground-truth’ network of gene interactions. To test our model predictions, we exploit the increasing number of closed genomes in model species to define gene co-segregation networks in the pan-genomes of Escherichia coli and Pseudomonas aeruginosa . Using these highly curated pan-genomes, we identify modular clusters of physically linked and co-segregating genes and show that the resulting co-segregation networks map onto underlying gene-regulatory and metabolic gene interaction networks. The results imply that co-segregation networks can contribute to accessory genome annotation, and more generally that gene interactions are the primary force shaping genome structure and operon evolution.