Systematic detection of polygenic cis-regulatory evolution

Abstract

The idea that most morphological adaptations can be attributed to changes in the cis-regulation of gene expression levels has been gaining increasing acceptance, despite the fact that only a handful of such cases have so far been demonstrated. Moreover, because each of these cases involves only one gene, we lack any understanding of how natural selection may act on cis-regulation across entire pathways or networks. Here we apply a genome-wide test for selection on cis-regulation to two subspecies of the mouse Mus musculus. We find evidence for lineage-specific selection at over 100 genes involved in diverse processes such as growth, locomotion, and memory. These gene sets implicate candidate genes that are supported by both quantitative trait loci and a validated causality-testing framework, and they predict a number of phenotypic differences, which we confirm in all four cases tested. Our results suggest that gene expression adaptation is widespread and that these adaptations can be highly polygenic, involving cis-regulatory changes at numerous functionally related genes. These coordinated adaptations may contribute to divergence in a wide range of morphological, physiological, and behavioral phenotypes.

Figure 1. Illustration of our test for lineage-specific selection.

Four unlinked genes are shown for each of four strains (or subspecies, species, etc). The number of curved blue lines (mRNA molecules) next to each gene represents its transcript level. Red “X”s represent polymorphisms that underlie eQTLs. Comparing strains A versus B, a single trans-eQTL (shown on a fifth chromosome) up-regulates all the genes. Although all four genes show the same direction of expression change, this is not evidence for selection, since the single trans-eQTL could be neutral. Comparing strains A versus C, all four genes have independent cis-eQTLs, but the directions are split between up- and down-regulation; thus there is no evidence for selection here as well. Comparing strains A versus D, again all four genes have independent cis-eQTLs, but now they are all up-regulating. This is consistent with lineage-specific selection for altered expression of this entire gene set (although in practice, more than four genes are needed to achieve statistical significance).

Figure 2. Results of the selection test for two gene sets.

(a) Effect directions for cis-eQTLs of mitochondria-related genes in liver. A consistent bias is seen for the B6 alleles to upregulate expression. A lower-bound estimate for the number of genes with cis-regulation under lineage-specific selection is the difference in height between the two bars (numbers in green). (b) Effect directions for cis-eQTLs of adult locomotory behavior-related genes in liver. A consistent bias is seen for the CAST alleles to upregulate expression. This does not imply that the expression changes in liver are relevant for this trait, as the effect is seen in all three tissues, and thus is not tissue-specific.

Figure 3. Results of the selection test in RNA-seq data.

(a) Directions of allelic expression bias for mitochondria and locomotory-related genes in day 9.5 embryos. The significance indicated by asterisks is the same as in Figure 2. (b) Directions of allelic expression bias for calmodulin-binding and memory-related genes in day 9.5 embryos.

Figure 4. Overlap between morphological and expression QTLs.

(a) QTL scan for naso-anal length of CxB females is shown in red, where the magnitude of the log₁₀ of the trait/genotype regression p-value is plotted for each of 1,438 genetic markers (in genomic order, starting with chromosome 1). Positive values indicate the B6 allele is associated with longer mice, while negative values indicate the opposite (scale is to the left). The blue and green lines are analogous, where the traits are expression of two growth-related genes, Dcaf13 and Sp3, in CxB female brain; positive values indicate the B6 allele up-regulates expression, while negative values indicate the opposite (scale is to the right). (b) As in part (a), except for male mice, and eQTL data are shown for another growth-related gene, Ept1 (in CxB male brain).