Evaluating UCE data adequacy and integrating uncertainty in a comprehensive phylogeny of ants

Abstract

While some relationships in phylogenomic studies have remained stable since the era of Sanger sequencing, many challenging nodes elude resolution, even with genome-scale data. As early studies grappled with random error and insufficient information, incongruence or lack of resolution in phylogenomics is generally associated with inadequate modeling of biological phenomena combined with analytical issues leading to systematic biases. Few phylogenomic studies, however, explore the potential for random error or establish an expectation of what level of resolution should be expected from a given empirical dataset. In presenting incongruent results, phylogeneticists face a choice of providing a diverse array of results from different approaches or a single preferred tree, with few attempting to integrate uncertainties across methods. Recent phylogenetic work has uncovered many well-supported and often novel relationships, as well as more contentious findings, across the phylogeny of ants. Ants are the most species-rich lineage of social insects and among the most ecologically important terrestrial animals. As a result, they have attracted much research, including regarding systematics. To date, however, there has been no comprehensive genus-level phylogeny of the ants inferred using genomic data combined with an effort to evaluate signal and incongruence throughout. Here we provide deeper insight into and quantify uncertainty across the ant tree of life. We accomplish this with the most taxonomically comprehensive Ultraconserved Elements dataset to date, including 277 (81%) of recognized ant genera from all 16 extant subfamilies, representing over 98% of described species-level diversity. We use simulations to establish expectations for resolution, identify branches with less-than-expected concordance, and dissect the effects of data and model selection on recalcitrant nodes. We also construct a consensus tree integrating uncertainty from multiple analyses. Simulations show that hundreds of loci are needed to resolve recalcitrant nodes on our genus-level ant phylogeny, even under a best-case scenario of known model parameters and without systematic bias. This demonstrates that random error continues to play a role in phylogenomics. Our analyses provide a comprehensive picture of support and incongruence across the ant phylogeny, and our consensus topology is congruent with a recent phylogenomic study on the subfamily-level, while rendering a more realistic picture of uncertainty and significantly expanding generic sampling. We use this topology for divergence dating and find that assumptions about root age have significant impact on the dates inferred. Our results suggest that improved understanding of ant phylogeny will require both more data and better phylogenetic models. We also provide a workflow to identify under-supported nodes in concatenation analyses, outline a pragmatic way to reconcile conflicting results in phylogenomics, and introduce a user-friendly locus selection tool for divergence dating.