Bayesian phylogenetic inference accounting for recombination in whole bacterial genomes

Abstract

Recombination is a major evolutionary force in bacteria, and ignoring it distorts phylogenies and can invalidate downstream analyses such as dating. We present ClonalFrame2, a new software for Bayesian phylogenetic inference from whole bacterial genomes that accounts for recombination. It replaces both ClonalFrame and ClonalFrameML while using the same evolutionary model, in which both mutation and recombination events occur on the branches of the clonal genealogy. The locations of recombination events are integrated out of the likelihood using a hidden Markov model, so the algorithm targets only the clonal genealogy and three recombination parameters. Ancestral sequences are recomputed whenever the tree changes and are used to propose efficient topology updates. We validated the method by analysing data-free datasets, which recovered the prior, and by comparing the approximated likelihood with the exact one on simulated data. Across 100 simulated datasets of 100 genomes each, the recombination parameters were accurately recovered, and trees inferred while accounting for recombination were many times more accurate than when ignoring it. In Staphylococcus aureus ST239, we showed that accounting for recombination gave a stronger temporal signal and a more robust dated phylogeny. In Neisseria gonorrhoeae, we identified interesting patterns of clustered recombination events happening repeatedly across the branches of the tree.