Bayesian inference of paravian phylogeny with the theropod working group dataset

Abstract

Paraves, traditionally known to comprise Dromaeosauridae, Troodontidae and Avialae, is among the most well-studied dinosaur groups thanks to its importance in investigating the dinosaur-bird transition. Recent studies with different data matrices generated various maximum parsimonious topologies of paravians and thus challenged the traditionally recognized monophyly of Deinonychosauria and the affiliation of Archaeopteryx as avialans. In spite of the efforts to solve this mystery under the maximum parsimonious criterion, here we applied Bayesian techniques to reconstruct the paravian phylogeny with the latest Theropod Working Group (TWiG) coelurosaur dataset which we supplemented with new data from recently described Mesozoic paravian taxa. Despite reconstructing the evolutionary relationships, Bayesian inference would also give us information on the evolutionary rates and divergence time of relevant clades. Preliminary analyses were performed with Beast v1.84 with tip dating using the Lewis Markov model, which enabled ordering of multistate morphological characters in Bayesian analyses. Beast could infer tree topology, divergence time, and evolutionary rates simultaneously using the relaxed clock model, which is now the standard practice in phylogenetic analyses with molecular data. Our preliminary result generally agrees with the maximum parsimonious trees recovered from the same dataset. The traditional sister group status of Dromaeosauridae and Troodontidae is reaffirmed. Jurassic paravians from the northeastern China have been recovered as the most basal avialans. Unlike the maximum parsimonious result where all Archaeopteryx specimens formed a polytomy with more derived avialans, the Solnhofen Archaeopteryx has been recovered as a more derived taxon than the clade formed by other Archaeopteryx specimens in our result. Increased evolutionary rates have been detected along the stem branches of the tree, as suggested in previous likelihood-based studies with different datasets. The divergence time of major coelurosaurian clades generally matches previous speculations, but the divergence time is dependent on the constrained age at the root. An alternative analysis with MrBayes 2.3.6 was carried out with the same TWiG dataset using the Lewis Markov model without tip dating. However, a slightly different topology has been recovered compared with the result from Beast.