INCORPORATING CONTINUOUS CHARACTERS IN TOTAL EVIDENCEPHYLOGENETIC ESTIMATION OF EXTINCT LINEAGES

The need for discrete characters has been shaped by fossil preservation and subsequent interpretations used by methods for phylogenetic inference. Most fossils lack molecular markers and methods to infer phylogenetic relationships from continuous characters (e.g., length or ratios) are underdeveloped, this has left discrete characters as the dominant form of information collected for both phylogenetic and taxonomic purposes. Although phylogenetic inference focuses on discrete criteria, branch length-dependent character evolution models have been developed for use in phylogenetic comparative methods. Here, we explore the utility of modeling characters under branch length-dependent processes. We primarily explore Brownian motion (BM), which assumes a trait will change by a fixed amount (sigma) each generation, and represents evolution under drift (i.e., random walk). Current software restrictions limit our use of other models, but we intend to explore Ornstein-Uhlenbeck character evolution models, which are essentially modeling BM with a trend. Here, we explore the use of BM character evolution models alongside discrete characters and fossil ages, under a fossilized birth-death model for a total-evidence estimation of phylogenetic trees of entirely extinct lineages. We use Dicynodontia, a clade of Permian and Triassic (299 – 201.5 Ma) synapsids, that are speciose, and include a Permian radiation known by relatively complete and well-preserved specimens, and a Triassic radiation of large-bodied animals that tend to be more poorly preserved. Because of the preservation differences, the Permian radiation relies heavily on discrete characters, whereas continuous variables are more helpful in placing the Triassic taxa. All phylogenetic models are scripted into RevBayes and convergence is assessed using Tracer. With BM, we find that joint estimation outperforms continuous-only, but is not as effective as discrete-only, suggesting that continuous variables contribute information that is discordant with discrete variables, and are heavily susceptible to model-misspecification. This represents a critical step forward in developing tools for phylogenetic inference of fossil lineages that can incorporate more objective data or individuals that have lost diagnostic features to diagenesis.