Affiliations: 1) Johns Hopkins University; 2) University of Iceland
Keywords: Theory & Method Development
Natural selection is one of the fundamental forces shaping the evolution of living organisms, and detecting its presence is a major goal of evolutionary biology. So far, the almost singular focus of the field has been on identifying signatures of selection on DNA sequence. However, natural selection acts on phenotypes, and DNA selection signatures only reflect the extent to which the genomic sequence causally affects phenotype. Since this causal effect of DNA is often mediated via intermediate molecular features, it is of great interest to move beyond DNA and identify signatures of selection directly on such molecular features.
Here, we focus on epigenetic marks (epialleles). We formalize a notion of neutrality for epialleles, and develop a test for selection which also allows for the assessment of possible confounders. Our test captures the known biology of epialleles, including trans-regulation. Because we are leveraging the fact that each epiallele occurs multiple times throughout the genome, the test requires neither population-scale data nor data across species. It only depends on knowing genic selection coefficients against heterozygous loss-of-function alleles, which are now known for most genes in human. We apply our test to epigenome-wide data from the human male germline and find: (A) Proximal promoter methylation and gene-body H3K36me3 are under negative and positive - respectively - selection, and this is not entirely explained by their association with gene expression. (B) The size of the hypomethylated region at promoters is under positive selection, while having almost no association with expression. (C) DNA methylation at the transcriptional end site is not under selection.
Both DNA methylation and H3K36me3 affect local mutation rates in the germline; DNA methylation elevates CpG mutation rates, while H3K36me3 lowers exonic mutation rates by recruiting the mismatch repair machinery. We provide evidence suggesting that the residual selective pressure on these marks - after accounting for their relationship with gene expression - may partly be explained by their action as mutation rate modifiers. We show via simulations that such selection can overcome genetic drift because these features affect multiple genes.
Our framework is simple but general, and we anticipate its core idea to be useful for other molecular features.