Sampling and SNP calling
We extracted DNA from a total of 233 NIDGS and 71 SIDGS buccal swab samples for which DNA yields varied from 0.1 to 51.0 ng/μl. Of 304 samples for which we conducted RADseq, for the IDGS dataset we kept 80 individuals and 4,227 SNPs after filtering the sequence data, with an average mean read depth of coverage of 9.3 reads per individual (min. 3.6 and max. 20.6) (Figure S1A, Supporting information). For the NIDGS dataset, we kept a total of 80 individuals and 3,575 SNPs, resulting in an average mean read depth of 7.5 (min. 0.9 and max. 18.8) (Figure S1B, Supporting information). For the SIDGS dataset, we kept 41 individuals and 2,348 SNPs, resulting in an average mean read depth of 4.9 (min. 1.0 and max. 16.1) (Figure S1C, Supporting information). Heterozygosity tended to decrease with increasing missing data and decreasing mean read depth, but no single population showed a particular bias towards low read depths across all individuals (Figures S1B and S1C, Supporting information).