QuIBL

We made use of QuIBL, a new tree-based method (Edelman et al. 2019), to differentiate between the models with ILS+introgression and with ILS only, and to obtain localized information on introgression. The method is described in detail in supplementary SMM5, Supplementary Material online. To carry out the QuIBL analysis, we used the fasta alignments we generated for our sliding window analyses and kept one sample per species that had the highest sequencing coverage (supplementary table S1, Supplementary Material online). Because S. compactum showed strong genetic structure among the populations, we used two samples from two different populations in this analysis. We used 49 scaffolds longer than 2 Mb that equal to 44% (175.6 M bases) of the total length of the reference. Since QuIBL is sensitive to recombination (Edelman et al. 2019), we extracted small 2-kb windows separated by 20 kb from each sample with Seqkit (Shen et al. 2016) to decrease the probability of sampling a window containing a recombination breakpoint (Edelman et al. 2019). We then discarded all windows that had samples with 100% of missing data and generated sliding window trees for the resulting 3,222 windows in the same manner as for our Sliding window tree analysis. We filtered the inferred ML trees based on the number of parsimony-informative sites (≥10), and used the resulted 3,195 trees as an input for QuIBL (https://github.com/michaelmiyagi/QuIBL, last accessed March 1, 2021). The QuIBL output was analyzed in the R statistical environment v3.6.3 (https://github.com/michaelmiyagi/QuIBL/tree/master/analysis, last accessed March 1, 2021), and we used the species tree topology to assign the outgroup to each triplet. We also calculated the percentage of loci supporting discordant topologies and showing significant evidence for introgression. We used the package “lattice” (Sarkar 2008), “corrplot” (Wei and Simko 2017), and “ggplot2” (Wickham 2016) to visualize the results of this analysis and the D-statistic tests.