Multilocus Phylogenetics of a Rapid Radiation in the Genus Thomomys (Rodentia: Geomyidae)

Abstract

Species complexes undergoing rapid radiation present a challenge in molecular systematics because of the possibility that ancestral polymorphism is retained in component gene trees. Coalescent theory has demonstrated that gene trees often fail to match lineage trees when taxon divergence times are less than the ancestral effective population sizes. Suggestions to increase the number of loci and the number of individuals per taxon have been proposed; however, phylogenetic methods to adequately analyze these data in a coalescent framework are scarce. We compare two approaches to estimating lineage (species) trees using multiple individuals and multiple loci: the commonly used partitioned Bayesian analysis of concatenated sequences and a modification of a newly developed hierarchical Bayesian method (BEST) that simultaneously estimates gene trees and species trees from multilocus data. We test these approaches on a phylogeny of rapidly radiating species wherein divergence times are likely to be smaller than effective population sizes, and incomplete lineage sorting is known, in the rodent genus, Thomomys. We use seven independent noncoding nuclear sequence loci (total approximately 4300 bp) and between 1 and 12 individuals per taxon to construct a phylogenetic hypothesis for eight Thomomys species. The majority-rule consensus tree from the partitioned concatenated analysis included 14 strongly supported bipartitions, corroborating monophyletic species status of five of the eight named species. The BEST tree strongly supported only the split between the two subgenera and showed very low support for any other clade. Comparison of both lineage trees to individual gene trees revealed that the concatenation method appears to ignore conflicting signals among gene trees, whereas the BEST tree considers conflicting signals and downweights support for those nodes. Bayes factor analysis of posterior tree distributions from both analyses strongly favor the model underlying the BEST analysis. This comparison underscores the risks of overreliance on results from concatenation, and ignoring the properties of coalescence, especially in cases of recent, rapid radiations.

Resolving conflict in eutherian mammal phylogeny using phylogenomics and the multispecies coalescent model

“…Concatenation methods infer phylogenies from multilocus sequences that are combined to form a single supermatrix (1), based on the assumption that all genes have the same or similar phylogenies (1,2). However, empirical studies have shown widespread presence of gene tree heterogeneity within mammals and other clades (3,4). When a high level of gene tree heterogeneity occurs in multilocus sequence data, theory and simulations have predicted that concatenation methods can yield misleading results (5,6).…”

Multilocus phylogeography and phylogenetics using sequence-based markers

“…Like the deep coalescence approach, the Bayesian method does not rely on concatenated datasets but instead estimates gene tree distributions in a model in which gene trees are partially correlated due to their common species tree history . This method is implemented in the program Bayesian estimation of Species Trees (BEST) and has been shown to perform well in simulations (Edwards et al 2007) and can now handle multiple alleles per species (Belfiore et al 2008;Liu et al 2008).…”

How many species of mammals are there?

“…While the addition of > 300 species each of rodents and bats is unsurprising given their existing diversity, these clades may reasonably contain disproportionally high levels of cryptic diversity (e.g., Ruedi and Mayer 2001;Belfiore et al 2008), and thus the application of genetic sequence data may continue to yield greater insights. Within Eulipotyphla (most particularly in shrews), we expect that the discovery of new species will continue given their rate of recent discoveries and frequency of morphological crypsis (Esselstyn et al 2013).…”