Selection analyses
To test if the demographic changes driven by hunting resulted in a reduction in the intensity of natural selection on FIVpco , we examined selective pressure in both time periods in each region using the RELAX hypothesis testing framework (Wertheim et al. 2015). The method builds upon random effects branch-site models (BS-REL) (Kosakovsky Pondet al. 2011) that estimates the ω ratio (the ratio of non-synonymous to synonymous mutations or dN/dS) along each branch from a discrete distribution of three ω ratio classes allowing selection pressure to vary across the phylogeny (Wertheim et al. 2015). A ω ratio of one corresponds to neutral selection with values > 1 being evidence for diversifying (positive) selection along a branch, and < 1 evidence for purifying (negative) selection along a branch. Briefly, RELAX tests for relaxation of selection pressure by dividing branches into three subsets; test branches (T), reference branches (R) and unclassified branches (U) (Wertheim et al. 2015) with ωT (resp. ωR ) being the estimated dN/dS ratio on test (resp. reference) branches. The discrete distribution of ω is calculated using BS-REL for each branch class, and then branches belonging to each subset are compared. The reference estimates of ω are raised to the power of k (an intensity parameter) so that  \(\omega_{T}=\omega_{R}^{k}\) in order to simplify model comparison. The null RELAX model is when the ω distribution and thus selective pressure is the same in R and T (whenk = 1). The null model is compared to an alternate model (using a likelihood ratio test) that allows k to vary so that whenk >1 selection pressure on the test branches was intensified or k < 1 indicating that selection pressure has been relaxed (Wertheim et al. 2015). In the relaxed scenario, k < 1 branches in R are under stronger purifying and diversifying selection compared to T branches (e.g., ω shifts from 0.1 to 0.001 or from 10 to 2). See Wertheim et al. (2015) for model details. T and R were selected from leaf branches (all other branches were Unassigned, U); individuals sampled from 2005-2011 (to the end of the lag period) were assigned to the R set and those sampled from 2012-2014 were assigned to T set. All branches not directly connecting to the tips were classified as ‘U’ as the majority had low phylogenetic support (posterior probability < 0.6). To further interrogate the sequence data to identify individual sites under selection, we performed the MEME (mixed-effects model of evolution) pipeline (Murrell et al. 2012). We performed both MEME and RELAX models using the Datamonkey web application (Weaver et al. 2018).