Constraint-tree methods resolve problematic branches in a recently diverged genus of New World swallow

The tree swallows (Tachycineta) is a model group of birds, and knowledge of its phylogeny is essential to interpreting a myriad of ecological, physiological, and behavioral data. However, reconstructing the phylogeny of Tachycineta has proven extremely difficult, probably because the branching order of its nine species is complicated by incomplete lineage sorting and introgression. Given earlier failures to resolve the phylogeny using Sanger-sequence data, we applied three methods of phylogenetic inference to a dataset of thousands of ultraconserved element (UCE) loci sequenced for multiple individuals of each species. Unfortunately, we recovered three well-supported discordant topologies. This result is perhaps unsurprising. The increasing incidence of alternative phylogenomic datasets yielding conflicting answers to phylogenetic questions has generated substantial debate concerning the collection and analysis of such data, and this study falls into a particularly troubling area, in which analysis of the same data by different methods leads to conflicting trees. We resolved the conflict by applying two methods, gene genealogy interrogation and analysis of per-site likelihood differences, which compare species trees and gene trees constrained to represent a set of topological hypotheses. These methods have previously been used only to resolve higher level phylogenetic problems; we demonstrate here that they also have the potential to clarify relationships in much more recently diverged groups. The conflicts we observed in the Tachycineta tree were driven by a tiny proportion of sites in the dataset; excluding less than five percent of loci from the concatenated alignment, or one percent of sites from each locus, was enough to change the results of coalescent-based phylogenetic inferences. The sensitivity of phylogenomic reconstruction to small numbers of influential sites has been observed in other studies and emphasizes both the importance of broad genomic sampling as well as the need to investigate biological sources of discord. This study makes clear that in many cases phylogenetic inference from genome-scale data will be inconclusive and will require post-inference analyses to select the most plausible tree.