This program uses a Bayesian approach to generate posterior probabilities of the
assignment of individuals to each of a given number of groups (K), independently of the sampling site of origin. We ran
STRUCTURE for K ¼ 1e13, with 10 replicates each and using a model based on admixture and independent allelic frequencies,
without taking into account information regarding sampling locations. Each run had 3 million iterations with 10% additional
burn in. The ad hoc summary statistic delta K (Evanno et al., 2005) was used to select the value of K with the uppermost
hierarchical level of population structure in our data. To verify the presence of an isolation-by-distance pattern, we tested the
existence of correlations between a matrix of pair-wise log-transformed ФPT values, and log-transformed geographical distances
using a Mantel test. The latter was carried out with web service v3.1.6 (http://ibdws.sdsu.edu) (Jensen et al., 2005);
significance was tested using 1000 permutations. Geographical distances among sampling locations were measured along the
river courses using Google Earth.