Polymorphic characters in the reconstruction of the phylogeny of geoemydid turtles

Several attempts to resolve the phylogeny of turtles in the clade Geoemydidae using morphology have been unsuccessful, in part because of unusually high levels of polymorphism. This has hindered the integration of the geoemydid fossil record into a phylogenetic framework. Many methods, shown to improve phylogenetic inference, allow the incorporation of different amounts of state frequency information from polymorphic characters into a phylogenetic analysis. Here, we present a new character matrix for the shell of geoemydids and assess the performance of polymorphism coding methods ('majority', 'generalized frequency coding', 'polymorphic' and 'missing') in a phylogenetic analysis by comparing the result topology of each method with a reference molecular phylogeny. The four coding methods failed to recover trees that were both well resolved and highly congruent with the reference phylogeny. Moreover, contrary to previous studies, the coding methods that made more use of character states frequencies did not perform better. However, a leave-one-out subsampling analysis suggested that despite these problems, the new matrix can still be used to place fossils in the geoemydid phylogeny with some reliability. Finally, we provide a list of characters that diagnose the major clades in our molecular reference tree.