2.3 Phylogenetic analysis
Two phylogenetic trees were generated: a non-ultrametric phylogram using Maximum Likelihood (ML) (Felsenstein, 1981) and an ultrametric chronogram using Bayesian inference (BI) (Drummond et al., 2002). The ML tree was generated using Iq-Tree (Trifinopoulos et al., 2016). Node support was assessed with 1000 ultrafast bootstrap replicates (Hoang et al., 2018), using the GTR+G+I model, with the data partitioned according to codon position, as recommended by PartitionFinder version 2.1. (Lanfear et al., 2017).
The BI tree was generated using BEAST version 2.6.7 (Bouckaert et al., 2019), using default settings for all parameters. XML files were made with the BEAUti version 2.6.7 interface with the following settings: GTR+G+I substitution model, empirical base frequencies, 4 gamma categories and all codon positions partitioned with unlinked base frequencies and substitution rates. Since there is no agreement concerning the most appropriate clock and tree priors for reconstructing gene trees for species delimitation (Monaghan et al., 2009; Ratnasingham & Hebert 2013; Talavera et al., 2013; Tang et al., 2014), preliminary analyses to compare the use of two different clock (strict and relaxed lognormal) and two different tree priors (coalescent constant population and Yule) were undertaken (Rodrigues et al., 2020). The results of these exploratory analyses (data not shown) indicated the strict clock and Yule priors as the most suitable for our data set, thus these priors were used for the Bayesian inference analyses.
To account for mixing within chains and convergence among chains with reversible jump MCMC (Elworth et al., 2018), a total of 10 chains were run from different seeds for 100 million generations each. Log files from each run were combined in LogCombiner version 2.6.7 (Drummond et al., 2012) after removal of the first 10% of samples from each run as burn-in. Convergence of each run and the combined data were checked for proper mixing using effective sample size (ESS) > 200 in Tracer version 1.7.1 (Rambaut et al., 2018). Tree files from each run were resampled to retain only 10% of the total trees and combined using LogCombiner after removal of the first 10% of retained trees from each run as burn-in. A maximum clade credibility (MCC) tree was then produced using TreeAnnotator version 2.6.6 (Drummond et al., 2012) and FigTree version 1.4.4 (Rambaut, 2010) was used to visualize and edit the trees.
All phylogenetic analyses were conducted on the CIPRES Science Gateway High Performance Computing platform (Miller et al., 2011).