FST estimates
We collected FST and FST analogs as measures of genetic differentiation (Holsinger & Weir, 2009; Meirmans & Hedrick, 2011) which we collectively refer to FSTthroughout this paper. Assuming an island model of migration-drift equilibrium, Wright (1951) developed a theoretical framework for studying the gene frequency variation among subpopulations through the fixation indices, i.e. F-statistics. In this model, FSTis the degree of gene differentiation among subpopulations for genes that have only two alleles. Nei (1973) expanded the model for polymorphic genes, and proposed GST as a measure of the gene diversity partitioned among subpopulations, relative to the total gene diversity of the population. Subsequently, Weir & Cockerham (1984) proposed a standard measure of genetic structure θ based on Wright (1951). The statistic θ is estimated per and across loci, and represents the correlation of genes, or coancestry, among individuals in a given population. Excoffier, Smouse, and Quattro (1992) proposed AMOVA (Analysis of Molecular Variance) and corresponding statistic φST; the proportion of genetic diversity partitioned among populations. Finally, Hedrick (2005) proposed a standardized measure of population differentiation, G’ST, which accounts for the level of heterozygosity of the marker used for genotyping individuals (G’ST=GSToverall/GSTmax).
The most common statistic in our dataset was θ. When θ was reported per loci, we took the mean across loci as the global FST for that species. The AMOVA derived φST was also common. Some studies reported both θ and φST, in which case we used φST as it likely better represents genetic structure among populations (Hey & Pinho, 2012). The statistics θ and φST were, however, frequently almost equivalent. Another common measure was GST; when reported for multiple pairs of populations, we used the mean across all pairs. A few studies reported G’ST. It was not possible to back-transform G’ST to GST because such studies did not report the maximum possible GST in their data (Hahn, Michalski, Fischer, & Durka, 2016). Even though G’STpotentially yields a higher value than GST (or θ and φST) based on the same data (Hedrick, 2005; Meirmans & Hedrick, 2011), we still included G’ST values, reasoning that any trend of variation in population genetic structure due to the variables here tested should still be present.