Time-scaled Bayesian phylogenetic analyses
To construct time-scaled trees, Bayesian molecular clock analyses were conducted using the Markov chain Monte Carlo (MCMC) method in BEAST v.1.10.4 (Suchard et al., 2018). Phylogenies were reconstructed using a) a Generalized Time-Reversible model (GTR) of nucleotide substitution (Tavaré & Miura, 1986) with a gamma distribution of site heterogeneity (also known as the Yang96 model (Yang, 1996)), b) a coalescent-based Gaussian Markov random fields (GMRF) Skyride model for estimating effective population size dynamics (Minin, Bloomquist, & Suchard, 2008), and c) an uncorrelated relaxed lognormal clock (Drummond, Ho, Phillips, & Rambaut, 2006). The BEAGLE library was used to optimize computational efficiency (Baele, Ayres, Rambaut, Suchard, & Lemey, 2019). Eight independent MCMC analyses were run for 200 million generations, sampled every 20,000 runs, and parameter convergence and effective sample size (ESS) (required to be >200) were evaluated in Tracer v.1.7.1 (Rambaut, Drummond, Xie, Baele, & Suchard, 2018). Using LogCombiner v.1.10.4, at least 10% burn-in was removed from each run and independent runs were combined to establish the maximum clade credibility (MCC) tree, from which the last 500 trees from the posterior distribution were extracted and used as the empirical tree for all subsequent phylodynamic analyses (Pagel, Meade, & Barker, 2004). Trees were visualized using Figtree v1.4.4 (Rambaut A., 2018).