Pleiotropy facilitates local adaptation to distant optima in common ragweed (Ambrosia artemisiifolia)

Abstract

Pleiotropy, the control of multiple phenotypes by a single locus, is expected to slow the rate of adaptation by increasing the chance that beneficial alleles also have deleterious effects. However, a prediction arising from classical theory of quantitative trait evolution states that pleiotropic alleles may have a selective advantage when phenotypes are distant from their selective optima. We examine the role of pleiotropy in regulating adaptive differentiation among populations of common ragweed (Ambrosia artemisiifolia); a species that has recently expanded its North American range due to human-mediated habitat change. We employ a phenotype-free approach by using connectivity in gene networks as a proxy for pleiotropy. First, we identify loci bearing footprints of local adaptation, and then use genotype-expression mapping and co-expression networks to infer the connectivity of the genes. Our results indicate that the putatively adaptive loci are highly pleiotropic, as they are more likely than expected to affect the expression of other genes, and they reside in central positions within the gene networks. We propose that the conditionally advantageous alleles at these loci avoid the cost of pleiotropy by having large phenotypic effects that are beneficial when populations are far from their selective optima. We further use evolutionary simulations to show that these patterns are in agreement with a model where populations face novel selective pressures, as expected during a range expansion. Overall, our results suggest that highly connected genes may be targets of positive selection during environmental change, even though they likely experience strong purifying selection in stable selective environments.

Fig 1. Population structure in A. artemisiifolia.

A: Map showing locations of the populations. Map tiles by Stamen Design, under CC BY 3.0. B: Genetic variation along the first two axes of a principal components analysis (PCA). Variation explained by the PCs is shown in brackets. C: The increase of genetic distance (F_ST) as a function of geographical distance between the populations. D: Estimated admixture proportions for three different numbers of ancestral populations (K).

Fig 2. Evidence of selective sweeps at candidate genes.

Fay and Wu’s H and CLR distributions of the selection outliers, eQTLs, and eGenes are compared against the rest of the transcriptome for two datasets. The horizontal lines mark the medians of the control genes.

Fig 3. Gene network topology at the candidate genes.

A: The percentage of eQTLs and eGenes found among the LFMM and PCAdapt outliers compared against all genes. Error bars show 95% bootstrap-based CIs. B: Connectivity measures at candidate genes. The horizontal line marks the median of the control genes.

Fig 4. Visual representation of two network modules containing more than expected LFMM and PCAdapt outliers.

Circles represent genes and lines show connections between them. Distance from the core signifies decreasing connectivity. Outliers, highlighted in color, are located in more central positions than expected by chance.

Fig 5. Simulated population fitness under different levels of pleiotropy, i.e. the number of phenotypes controlled by a single QTL (horizontal axis).

Phenotypes start from an initial value 0 and selection acts to move them towards three different optima. Shown are medians and interquartile ranges (IQR) from 300 simulation. The fitness estimates were normalized in relation to the non-pleiotropic class (median = 0, IQR = 1).