Population structure
We characterized population genetic structure using the SNP dataset with no missing data. We performed PCA and DAPC with the Adegenet 2.1.1 package (Jombart, 2008) in R. For the DAPC analysis, we first conductedK -means clustering and selected the number of clusters based on the lowest Bayesian Information Criterion (BIC) value. We performed cross-validation with the function xvalDapc to determine the number of PCs to retain by calculating the lowest root mean squared error. We then ran DAPC, retaining 20 PCs and 2 discriminant functions.
We used the Bayesian clustering algorithm in the program STRUCTURE v2.3.4 to infer the number of population clusters (K ) and the proportion of individual membership assigned to each cluster (qk ). We used a burn-in of 500,000 steps followed by 1,000,000 recorded steps, tracking the probability of the data givenK (LnP(D)) to ensure that we ran the program long enough for the values to stabilize. We used the admixture model, no location priors, and assumed correlated allele frequencies (Falush, Stephens, & Pritchard, 2003). We performed a simulation with K from 1 to 7 with 10 replicates each and identified meaningful K values using the ΔK method (Evanno, Regnaut, & Goutdet, 2005) implemented in STRUCTURE HARVESTER v0.6.94 (Earl & vonHoldt, 2012). We combined replicate runs using CLUMPP v1.1.2 (Jakobsson & Rosenberg, 2007).
Using the SNP dataset with 10% missing data, we performed an AMOVA and calculated pairwise F ST values between populations identified by STRUCTURE in GenAlEx v6.5 (Peakall & Smouse, 2012), with 10,000 permutations to generate the null distribution. To investigate local spatial structure, we performed a Mantel test using ade4 v1.7 (Dray & Dufour, 2007) in R on both the full and unrelated dataset. We tested for a correlation between pairwise genotypic distance and Euclidean geographic distance with 9,999 permutations to generate the null distribution. We also generated a Mantel correlogram to test for spatial autocorrelation between pairs of treeshrews at different distance classes using GenAlEx. We first calculated pairwise linear geographic and genotypic distances, and then used the ‘Spatial’ option with 9,999 permutations. We defined 7 distance classes (0.2, 1.0, 2.0, 5.0, 10.0, 15.0, and 18.0 km) based on Sturges’s Rule (Sturges, 1926), chosen to ensure sufficient comparisons within each class. Finally, we calculated the average, median, and maximum geographic distances between pairs of individuals in each kinship class corresponding to first, second, third-order, and distant relatives (Table 1). To test for significant differences between the means in each kinship class, we performed a one-way ANOVA in R with a Tukey Honest Significant Differences test and a Bonferroni correction for multiple comparisons (Combs, Puckett, Richardson, Mims, & Munshi-South, 2018).