Meta-analysis: Congruence of genomic and phenotypic differentiation across diverse natural study systems

Abstract

Linking genotype to phenotype is a primary goal for understanding the genomic underpinnings of evolution. However, little work has explored whether patterns of linked genomic and phenotypic differentiation are congruent across natural study systems and traits. Here, we investigate such patterns with a meta-analysis of studies examining population-level differentiation at subsets of loci and traits putatively responding to divergent selection. We show that across the 31 studies (88 natural population-level comparisons) we examined, there was a moderate (R ² = 0.39) relationship between genomic differentiation (F _ST ) and phenotypic differentiation (P _ST ) for loci and traits putatively under selection. This quantitative relationship between P _ST and F _ST for loci under selection in diverse taxa provides broad context and cross-system predictions for genomic and phenotypic adaptation by natural selection in natural populations. This context may eventually allow for more precise ideas of what constitutes "strong" differentiation, predictions about the effect size of loci, comparisons of taxa evolving in nonparallel ways, and more. On the other hand, links between P _ST and F _ST within studies were very weak, suggesting that much work remains in linking genomic differentiation to phenotypic differentiation at specific phenotypes. We suggest that linking genotypes to specific phenotypes can be improved by correlating genomic and phenotypic differentiation across a spectrum of diverging populations within a taxon and including wide coverage of both genomes and phenomes.

Keywords: FST; GWAS; PST; candidate gene approaches; natural selection; outlier analysis.

FIGURE 1

Logit transformations are useful for quantifying differentiation. Left: F _ST and P _ST, untransformed, provide limited characterization of differentiation when differentiation is low or high (i.e., interpopulation variation is very small or large relative to intrapopulation variation). Center: logit transformations of F _ST and P _ST, however, provide a log‐linear metric of differentiation whose shape is independent of differentiation. Right: doubling differentiation (interpopulation variation) has the same effect on logit(F _ST and P _ST) regardless of starting point, whereas the effect of doubling differentiation on untransformed F _ST and P _ST depends heavily on starting point

FIGURE 2

Non‐neutral and neutral F _ST are highly correlated across studies. Gray labels show untransformed F _ST values. Each point represents average F _ST values for a unique population–population comparison, with multiple points for multiple methods (see text)

FIGURE 3

Both non‐neutral (left) and neutral (right) F _ST predict P _ST, but non‐neutral F _ST is a much stronger predictor of P _ST. Gray labels show untransformed F _ST and P _ST values. Each point represents average P _ST and F _ST values for a unique population–population comparison, with multiple points for multiple methods (see text)

FIGURE 4

Left: proportion of non‐neutral loci (the ratio of candidate, outlier, or GWAS positive loci to the total number examined) has a negative effect on P _ST. Removing this effect allows for a clearer view of the P _ST‐non‐neutral F _ST relationship (right). There is no significant interaction between proportion of non‐neutral loci and non‐neutral F _ST (Table 2). Gray labels show untransformed F _ST, P _ST, and proportion of non‐neutral loci values. Each point represents average P _ST,F _ST, and proportion of non‐neutral loci for a unique population–population comparison, with multiple points for multiple methods (see text)

FIGURE 5

Effects of common‐garden experimentation (top left), marker type (top right), broad method (bottom left), and analysis method (bottom right) on P _ST and the P _ST‐non‐neutral F _ST slope. Only marker type had a significant effect on P _ST, and none of the four variables had a significant effect on the P _ST‐non‐neutral F _ST slope (Table 2). Gray labels show untransformed F _ST and P _ST values. Each point represents average P _ST and F _ST values for a unique population–population comparison, with multiple points for multiple methods (see text). Variation due to proportion of non‐neutral loci is removed in each panel. *Only marker type (top right) had a significant effect on P _ST; R ² values and trendlines for the other three models are from the base (Figure 4) model

FIGURE 6

Neither non‐neutral F _ST (left) nor proportion of non‐neutral loci (right) predicts P _ST consistently well within studies. Each row represents a study; each symbol type represents a phenotype. Data are taken from (top to bottom): Culling et al. (2013), Hamlin & Arnold (2015), Hudson et al. (2013), Kaueffer et al. (2012), Laporte et al. (2015), Raeymaekers et al. (2007). Gray labels show untransformed F _ST, P _ST, and proportion of non‐neutral loci values. R ² values were calculated based on Equation 4, with a few negative values when study‐specific trends in non‐neutral F _ST vs. proportion of non‐neutral loci were the opposite of trends across studies