Returning to the roots: Resolution, reproducibility, and robusticity in the phylogenetic inference of Dissorophidae (Amphibia: Temnospondyli)

Abstract

The phylogenetic relationships of most Paleozoic tetrapod clades remain poorly resolved, which is variably attributed to a lack of study, the limitations of inference from phenotypic data, and constant revision of best practices. While refinement of phylogenetic methods continues to be important, any phylogenetic analysis is inherently constrained by the underlying dataset that it analyzes. Therefore, it becomes equally important to assess the accuracy of these datasets, especially when a select few are repeatedly propagated. While repeat analyses of these datasets may appear to constitute a working consensus, they are not in fact independent, and it becomes especially important to evaluate the accuracy of these datasets in order to assess whether a seeming consensus is robust. Here I address the phylogeny of the Dissorophidae, a speciose clade of Paleozoic temnospondyls. This group is an ideal case study among temnospondyls for exploring phylogenetic methods and datasets because it has been extensively studied (eight phylogenetic studies to date) but with most (six studies) using a single matrix that has been propagated with very little modification. In spite of the conserved nature of the matrix, dissorophid studies have produced anything but a conserved topology. Therefore, I analyzed an independently designed matrix, which recovered less resolution and some disparate nodes compared to previous studies. In order to reconcile these differences, I carefully examined previous matrices and analyses. While some differences are a matter of personal preference (e.g., analytical software), others relate to discrepancies with respect to what are currently considered as best practices. The most concerning discovery was the identification of pervasive dubious scorings that extend back to the origins of the widely propagated matrix. These include scores for skeletal features that are entirely unknown in a given taxon (e.g., postcrania in Cacops woehri) and characters for which there appear to be unstated working assumptions to scoring that are incompatible with the character definitions (e.g., scoring of taxa with incomplete skulls for characters based on skull length). Correction of these scores and other pervasive errors recovered a distinctly less resolved topology than previous studies, more in agreement with my own matrix. This suggests that previous analyses may have been compromised, and that the only real consensus of dissorophid phylogeny is the lack of one.