Phylogenetic Analysis
We estimated the phylogeny for several datasets (see Table 1 for dataset descriptions) within a maximum likelihood framework implemented in RAxML v8.2.4 (Stamatakis, 2014). We used the GTR+G model of sequence evolution. For each analysis, 1000 bootstrap replicates were calculated using the rapid bootstrapping option implemented in RAxML. The phylogenies constructed from these datasets were viewed in Figtree v1.4.1 with midpoint rooting. The phylogenies were then compared to look for congruence among clades and bootstrap values, with the final tree having the largest amount of taxon coverage and high support values used for downstream analyses (All_M20 dataset with all 157 individuals and a minimum of 20% locus occupancy; Figure 2).
Bayesian inference of the All_M20 data set was conducted in ExaBayes v1.4.1 (Aberer et al., 2014). Two independent runs of 6x107 generations with four coupled chains each were run simultaneously starting with a random starting tree. Standard deviation of split frequencies was monitored (< 0.05), and the first 25% were discarded as burn-in. An extended majority rule consensus tree was generated using the program consense within ExaBayes (Supplemental Figure S1; Aberer, Kobert, and Stamatakis 2014).
A species tree was generated from a set of gene trees using the software ASTRAL-III (Supplemental Figure S2; Zhang et al. 2018). We used the MAGNET v0.1.5 pipeline (Bagley, 2019) to estimate a gene tree for each RAD locus within the All_M20 data set using RAxML v.8.2.4, with each RAxML run implementing the GTR+G model. These gene trees were used as input for ASTRAL-III. With uncertainty in the relationships amongA. unicolor B, a well-supported clade in both maximum likelihood and Bayesian analyses but slightly differing sister relationships (see Results), additional species trees were generated using A. unicolor B -only matrices (UniB_M60 and UniB_M80; see Table 1) for aid in elucidating relationships among A. unicolor B.