LAUSR

Despite the importance of natural selection in species’ evolutionary history, phylogenetic methods that take into account population-level processes typically ignore selection. The assumption of neutrality is often based on the idea that selection occurs at a minority of loci in the genome and is unlikely to compromise phylogenetic inferences significantly. However, genome-wide processes like GC-bias and some variation segregating at the coding regions are known to evolve in the nearly neutral range. As we are now using genome-wide data to estimate species trees, it is natural to ask whether weak but pervasive selection is likely to blur species tree inferences. We developed a polymorphism-aware phylogenetic model tailored for measuring signatures of nucleotide usage biases to test the impact of selection in the species tree. Our analyses indicate that while the inferred relationships among species are not significantly compromised, the genetic distances are systematically underestimated in a node-height dependent manner: i.e., the deeper nodes tend to be more underestimated than the shallow ones. Such biases have implications for molecular dating. We dated the evolutionary history of 30 worldwide fruit fly populations, and we found signatures of GC-bias considerably affecting the estimated divergence times (up to 23%) in the neutral model. Our findings call for the need to account for selection when quantifying divergence or dating species evolution. Significance statement Although little is known about the impact of natural selection on species tree estimation, expectations are that it occurs at a minority of loci in eukaryotic genomes and is thus unlikely to affect the divergence process. However, growing evidence suggests that a large amount of the genomic variation evolves under weak but pervasive selection (e.g., fixation biases created by GC-bias gene conversion). We tested the impact of unaccounted-for nearly neutral selection on species tree estimation and found that the estimated branch lengths are systematically biased. Our results highlight the need for selection-aware models in species tree estimation and molecular dating.