Phylogenetic analyses under secondary structure-specific substitution models outperform traditional approaches: case...

Many rDNA molecular phylogenetic studies result in trees that are incongruent to either alternative gene tree reconstructions and/or morphological assumptions. One reason for this outcome might be the application of suboptimal phylogenetic substitution models. While the most commonly implemented models describe the evolution of independently evolving characters fairly well, they do not account for character dependencies such as rRNA strands that form a helix in the ribosome. Such nonindependent sites require the use of models that take into account the coevolution of the complete nucleotide pair (doublet). We analyzed 28S rDNA (LSU) demosponge phylogenies using a "doublet" model for pairing sites (rRNA-helices) and compared our findings with the results of "standard" approaches using Bayes factors. We demonstrate that paired and unpaired sites of the same gene result in different reconstructions and that usage of a doublet model leads to more reliable demosponge trees. We show the influence of more sophisticated models on phylogenetic reconstructions of early-branching metazoans and the phylogenetic relationships of demosponge orders.