Demographic Modeling
We optimized a three-population demographic model based on the site-frequency spectrum (SFS) using the genetic algorithm ‘GADMA’ (Noskova et al. 2020, 2023) based on the ‘moments’ engine (Jouganous et al. 2017). We first down projected the VCF file from ipyrad to a three-dimensional SFS of [27,9,21] from (28,8,20) diploid individuals from the NTA, COH, and CHR lineages (see Results) using ‘easySFS’ (https://github.com/isaacovercast/easySFS; Gutenkunst et al. 2009). Down projection was based on the allelic sample size that maximized the number of segregating sites per lineage. Mutation rate (per site, per generation) is not generally known in amphibians, so we used the mean rate for fishes of 5.97e-09; these values are relatively similar across vertebrates (e.g., mean for mammals of 7.97e-09; Bergeron et al. 2023). The generation time for Desmognathus aeneus is ~3 years (Bruce 2022). Specifying mutation rate and sequence length allow inference of θ, and generation time allows re-scaling to thousands of years. We allowed asymmetric migration, unlinked SNPs with no outgroup, and an initial and final structure with one time-period per divergence (i.e., 1,1,1). We used 120 repeated optimizations in GADMA to arrive at the final models (Fig. S8).