Phylogenetic relationships among the phrynosomatid sand lizards inferred from mitochondrial DNA sequences generated by heterogeneous evolutionary processes.

Abstract

Nucleotide sequences of the mitochondrial protein coding cytochrome b (cyt b; 650 bp) and small-subunit 12S ribosomal RNA (approximately 350 bp) genes were used in analyses of phylogenetic relationships among extant phrynosomatid sand lizards, including an examination of competing hypotheses regarding the evolution of "earlessness." Sequences were obtained from all currently recognized species of sand lizards as well as representatives of the first and second outgroups and analyzed using both parsimony and likelihood methods. The cyt b data offer strong support for relationships that correspond with relatively recent divergences and moderate to low support for relationships reflecting more ancient divergences within the clade. These data support monophyly of the "earless" taxa, the placement of Uma as the sister taxon to the other sand lizards, and monophyly of all four taxa traditionally ranked as genera. All well-supported relationships in the 12S phylogeny are completely congruent with well-supported relationships in the cyt b phylogeny; however, the 12S data alone provide very little support for deeper divergences. Phylogenetic relationships within species are concordant with geography and suggest patterns of phylogeographic differentiation, including the conclusion that at least one currently recognized species (Holbrookia maculata) actually consists of more than one species. By independently optimizing likelihood model parameters for various subsets of the data, we found that nucleotide substitution processes vary widely between genes and among the structural and functional regions or classes of sites within each gene. Therefore, we compared competing phylogenetic hypotheses, using parameter estimates specific to those subsets, analyzing the subsets separately and in various combinations. The hypothesis supported by the cyt b data was favored over rival hypotheses in all but one of the five comparisons made with the entire data set, including the set of partitions that best explained the data, although we were unable to confidently reject (P < 0.05) alternative hypotheses. Our results highlight the importance of optimizing models and parameter estimates for different genes or parts thereof--a strategy that takes advantages of the strengths of both combining and partitioning data.