Different patterns of selection on the nuclear genes IRBP and DMP-1 affect the efficiency but not the outcome of phylogeny estimation for didelphid marsupials

Selection at the protein-level can influence nucleotide substitution patterns for protein-coding genes, which in turn can affect their performance as phylogenetic characters. In this study, we compare two protein-coding nuclear genes that appear to have evolved under markedly different selective constraints and evaluate how selection has shaped their phylogenetic signal. We sequenced 1,100+ bp of exon 6 of the gene encoding dentin matrix protein 1 (DMP1) from most of the currently recognized genera of New World opossums (family: Didelphidae) and compared these data to an existing matrix of sequences from the interphotoreceptor retinoid-binding protein gene (IRBP) and morphological characters. In comparison to IRBP, DMP1 has far fewer sites under strong purifying selection and exhibits a number of sites under positive directional selection. Furthermore, selection on the DMP1 protein appears to conserve short, acidic, serine-rich domains rather than primary amino acid sequence; as a result, DMP1 has significantly different nucleotide substitution patterns from IRBP. Using Bayesian methods, we determined that DMP1 evolves almost 30% faster than IRBP, has 2.5 times more variable sites, has less among-site rate heterogeneity, is skewed toward A and away from CT (IRBP has relatively even base frequencies), and has a significantly lower rate of change between adenine and any other nucleotide. Despite these different nucleotide substitution patterns, estimates of didelphid relationships based on separate phylogenetic analyses of these genes are remarkably congruent whether patterns of nucleotide substitution are explicitly modeled or not. Nonetheless, DMP1 contains more phylogenetically informative characters per unit sequence and resolves more nodes with higher support than does IRBP. Thus, for these two genes, relaxed functional constraints and positive selection appear to improve the efficiency of phylogenetic estimation without compromising its accuracy.