The effects of ordered multistate morphological characters on phylogenetic analyses of eutherian mammals

Abstract

Multistate morphological characters are routinely used in phylogenetic analyses. Individual multistate characters may be treated as linearly ordered, partially ordered, or unordered. Each option implies a hypothesis of character evolution, and significant debate surrounds the appropriateness of ordering multistate characters. Several previous analyses support ordering multistate morphological characters when the character states form a morphocline. Here, we explore the effects of ordering a subset of characters in the largest morphological character matrix that is available for placental mammals. All multistate characters were assessed and were ordered only if the character states were meristic or hypothesized to form a morphocline. We then performed parsimony analyses, with and without molecular scaffolds, to examine the effects of ordering on placental mammal phylogeny. We also performed pseudoextinction analyses, which treated designated extant taxa as extinct by eliminating them from the molecular scaffold and scoring soft-tissue characters as missing, to determine if ordered or unordered characters would more accurately reconstruct the relationships of pseudoextinct placental orders. Character ordering affected the placement of a variety of taxa in non-scaffolded analyses, but the effects were less evident in scaffolded analyses. Nevertheless, one of the islands of most parsimonious trees with the ordered data set and a scaffold for extant taxa supported the inclusion of Leptictida, including the Late Cretaceous Gypsonictops, inside of crown Placentalia. Our analyses rarely supported the monophyly of Tamirtheria, the Late Cretaceous-Paleogene stem-based clade that is hypothesized to be the sister to Placentalia. Pseudoextinction analyses using ordered and unordered characters both reconstructed 12 of 19 pseudoextinct orders in positions that are incongruent with a well-supported molecular scaffold. These results suggest that the use of ordered multistate characters does not increase the proportion of well-supported molecular clades that are reconstructed with the largest available phenomic data set for placental mammals.