Phylogenetic analyses and divergence time estimations based on mitochondrial DNA
We inferred gene trees with Bayesian and maximum parsimony (MP) methodologies using MrBayes 3.2.2 (Ronquist et al. 2012) and TNT 1.1 (Goloboff et al. 2003), respectively. The best-fit model of nucleotide substitution for each locus for the Bayesian analysis was selected using the Bayesian information criterion (BIC) implemented in jModelTest 2.1.1 (Darriba et al. 2012). HKY+I was chosen for COI and KHY for cyt b(Hasegawa et al. 1985). Both loci were placed in unlinked partitions allowing parameters to vary and to be estimated independently (except for topology and branch lengths). We conducted two independent runs of 10 million generations sampling trees every 100 generations under default priors for all parameters. We discarded the first 25% of the sampled trees as burn-in and the remaining 75,000 trees of each run were combined to generate a majority rule consensus tree. The standard deviation of split frequencies (SDSF) between runs was always < 0.01 indicating convergence. Using Tracer 1.6 (Rambaut et al. 2014), we assured that both runs reached the stationary phase and that we had a good sample of the posterior probability distributions. For the MP analysis we ran heuristic searches based on 1,000 random addition sequences (RAS) coupled with the tree bisection reconnection (TBR) branch-swapping algorithm, saving 10 trees per replication. A strict consensus tree was estimated from all the most parsimonious trees. To estimate node support we conducted a bootstrap analysis (Felsenstein 1985) that consisted of 1,000 pseudoreplicates of 100 RAS + TBR saving 10 trees per replicate.
The age of the most recent common ancestor between T. ruficapillus and T. doliatus and the time of separation of the mitochondrial lineages within T. ruficapillus were estimated by generating a time-calibrated ultrametric tree with the Bayesian approach implemented in BEAUTi/BEAST 1.8 (Drummond et al. 2012). Both mitochondrial markers were placed in separate partitions with unlinked substitution and clock models selected with jModelTest: HKY+I for COI and HKY for cyt b . The tree models were linked since the mitochondria is a single unit of inheritance. We specified a Yule speciation tree prior, assuming a constant population size and a relaxed uncorrelated lognormal clock. We used a calibration of 2.1% per million years (My) for cyt b (1.05x10-2substitutions/site/lineage/My; Weir and Schluter 2008) and 1.17x10-2 substitutions/site/lineage/My for COI (Lavinia et al. 2016). The analysis was run for 100 million generations sampling every 1,000 generations. We checked for stationarity in the estimation of the parameters and adequate ESS values using Tracer. We discarded the first 10% of the sampled trees (burn-in = 10,000) and then estimated the 95% highest posterior density (HPD) intervals of divergence dates using TreeAnnotator 1.8 (Drummond et al. 2012).