E. coli is a commensal of the human gut but also a major opportunistic pathogen causing of the order of a million deaths worldwide each year. While it is well-known that extra-intestinal infections (urinary tract infections, bacteremia) are caused in their vast majority by a small subset of strains (ST131, ST95, ST69, etc.), the extent to which pathogenicity is determined by bacterial genetic factors had not been quantified. I will first present the result of a genome-wide association study characterizing systematically the bacterial genomic determinants of infection, where we show that pathogenicity is largely determined by E. coli genetic variation at multiple levels, and develop a predictor of pathogenicity. Secondly, we asked why pathogenicity might evolve and be selected in the bacterial population. To answer this question, we analyzed longitudinal gut microbiota datasets to investigate the epidemiology of E. coli in its natural habitat. We found that E. coli lineages face a trade-off between colonization and persistence, with niche differentiation enabling long-term coexistence. A mathematical model of colonization dynamics showed that disturbances of the human gut disproportionately impact resident strains. This suggests a general mechanism by which residence time can be selected. As resident strains tend to be the most pathogenic, this would indirectly select for pathogenicity.