Collapsed Site Coalescent Facilitates Detection of Adaptive Divergences Between Segmental Duplicates

Abstract

Segmental duplications make up a substantial fraction of many genomes and are a recurrent source of evolutionary innovation, yet they have remained largely inaccessible to population genetic inference. One reason is that, until recently, their high sequence identity precluded reliable sequencing. Long-read sequencing now resolves population haplotypes across segmental duplications. Still lacking are appropriate population genetic methods to analyze population samples. A particular difficulty arises from frequent interlocus gene conversion (IGC), the copying of sequence from one duplicate into another. Such events correspond to coalescence of lineages across distinct sites--thereby violating a core assumption of standard coalescent theory and ancestral recombination graphs (ARGs). Here, we propose an approach that can facilitate ARG inference for segmental duplicates. We propose the Collapsed Site Coalescent (CSC), a model which treats each pair of paralogous sites as a single collapsed pseudo-site with a self-contained genealogy, so that the sequence of genealogies along a collapsed locus forms an ARG. Based on the CSC, we develop a haplotype-copying inference method that estimates IGC parameters from a population sample. In simulations, estimated parameters closely track true IGC rates. We further investigate the case where an adaptive difference between duplicates is sheltered by selection, protecting it from IGC that would otherwise erase the difference, a situation we call an IGC haven. When both IGC and selection are strong, our approach allows for the detection of IGC havens. In sum, this approach brings duplicated sequence within reach of population genetics inference, opening the way for tackling longstanding questions about the evolution of gene duplicates.