Detecting Natural Selection in the Human Past Using Ancient DNA

Natural selection is a fundamental process that shapes evolution, from the emergence of antibiotic resistance to the divergence of species. Identifying genomic regions that contribute to selection processes allows us to understand the basis of adaptation and enables us to predict future responses to selective pressure. Time-series genetic datasets, sourced from ancient DNA or experimental evolution systems, are a promising tool for detecting natural selection, as they give greater insight into how allele frequencies have changed over time than samples solely from the present day. Many methods exist that use time-series genetic data to detect additive selection, whereby each copy of a beneficial allele increases an individual's fitness. However, many selective processes, such as the evolution of gene expression and host-pathogen dynamics, display non-additive dynamics. In the first part of this work, I develop a method to estimate selection coefficients from time series genetic data under a general diploid model of selection, in which the fitness of individuals with one copy of a beneficial allele and with two copies are uncoupled. Additionally, I outline a framework for classifying the dynamics of a particular dataset into one of several modes of selection. I apply this method to an empirical dataset of ancient DNA samples from Great Britain, and identify six genomic regions as targets of general diploid selection. In the second part of this work, I analyze polygenic adaptation, in which selection on a trait causes small, correlated allele frequency changes at loci that affect the trait. Polygenic adaptation has two regimes: directional selection towards an optimum trait value, and stabilizing selection to reduce variance in a population at its optimum. I propose a method that combines time-series genetic data at a number of loci with estimates of the effect each locus has on a trait to quantify the strength of selection on the trait as a whole. I apply this method to a similar ancient DNA dataset as in the first project and a set of effect sizes estimated from the UK Biobank.