Misleading phylogenetic inferences based on single-exemplar sampling in the turtle genus Pseudemys.

Reconstructing species trees for clades containing weakly delimited or incorrectly identified taxa is one of the most serious challenges facing systematists because building phylogenetic trees is generally predicated on correctly identifying species membership for the terminals in an analysis. A common practice, particularly in large-scale phylogenetic analyses, is to use single-exemplar sampling under the implicit assumption that the resulting phylogenetic trees will be poorly supported if the sampled taxa are not good species. We examine this fundamental assumption in the North American turtle genus Pseudemys, a group of common, widely distributed freshwater turtles whose species boundaries and phylogenetic relationships have challenged systematists for over half a century. We sequenced 10 nuclear and three mitochondrial genes from the nine currently recognized species and subspecies of Pseudemys using geographically-widespread sampling of each taxon, and analyzed the resulting 86-individual data set using population-genetic and phylogenetic methods. We found little or no evidence supporting the division of Pseudemys into its currently recognized species/subspecies. Rather, our data strongly suggest that the group has been oversplit and contains fewer species than currently recognized. Even so, when we conducted 100 replicated, single-exemplar phylogenetic analyses of these same nine taxa, most Bayesian trees were well resolved, had high posterior probabilities, and yet returned completely conflicting topologies. These analyses suggest that phylogenetic analyses based on single-exemplar sampling may recover trees that depend on the individuals that are sampled, rather than the underlying species tree that systematists assume they are estimating. Our results clearly indicate that final resolution of Pseudemys will require an integrated analysis of morphology and historical biogeographic data coupled with extensive geographic sampling and large amounts of molecular data, and we do not recommend taxonomic changes based on our analyses. If our 100-tree resampling experiments generalize to other taxa, they suggest that single-exemplar phylogenies should be interpreted with caution, particularly for groups where species are shallowly diverged or inadequately delimited.