EVOLUTIONARY TRIANGULATION: AN APPROACH TO IDENTIFY ANCIENT AND CONTEMPORARY SIGNALS OF ADAPTATION

Abstract

Discovering the targets of selection in a genome is fundamental for understanding how adaptation has affected the biological processes that lead to population and species diversity. However, very few variants in proteins are known to be adaptive. For example, in humans, fewer than 30 such adaptive missense variants are known. It is surprising that in the 5-6 million years since human divergence from chimpanzee so few adaptive missense variants would have accumulated, and so it is unclear whether such adaptive variants have never occurred or whether current methods cannot detect such changes. Are there such adaptive variants in the coding region of the genome? Alternatively, is everything found in non-coding regions, which is often largely thought of as “junk DNA”? These questions represent a grand challenge in the field of evolutionary biology. This work helps to resolve this paradox and shows that there, in fact, may be much more coding adaptation that was previously not detectable. We do this by developing and testing the idea that long term evolutionary patterns of genomic differences between species can be used to identify alleles that affect the fitness of individuals in a population.Many methods that aim to detect selection are often limited as to the timescale during which they are applicable, e.g., long-term or short-term only. In bridging this gap, we build on the Evolutionary Probability (EP) method, a method that uniquely integrates long term evolutionary history to provide a probability of observation under neutrality. We develop an approach that benefits from the strengths of both types of methods by leveraging long term evolutionary information from EP and short-term evolutionary information from population polymorphisms to identify alleles affected by selection. High-throughput analysis of variation for functional change is not possible given the immense sequencing data output that is increasing each day. Thus, we employ an in silico approach. We perform simulations of neutral evolution to identify thresholds of neutrality for EP. We also analyze various bona fide datasets of both neutral, adaptive, and deleterious variation found in humans comprising of hundreds of thousands of individual variants in humans alone. We find that our approach is reliably able to identify evolutionarily unexpected, non-neutral (affected by selection) alleles and requires only sequence alignments. Known adaptive variants are successfully identified, and a vast majority of disease variants are found to be evolutionarily forbidden underscoring the role of coding variation in human divergence as much more than previously thought. In fact, applying our approach across the human exome, we find that the amount of adaptation in humans is likely multiple orders of magnitude greater than previously believed. A catalog of the discovered candidate adaptive polymorphisms is available through an online personal genomics web resource. The new approach has also been shown to be generalizable to any sequence alignment, given that sufficient evolutionary time has elapsed among the species. Overall, our findings offer insight into the adaptive landscape of coding variants in humans, as well as provide a look into the role of long-term evolutionary history into generating disease in contemporary human populations, prompting a rethinking of the role of adaptation in human evolution.