Patterns of genetic diversity across the genome are affected by multiple forces of evolution, including natural selection and population demography. These effects are manifested both locally across the genome and genome-wide. While natural selection operates directly on only a small percentage of mutations within genome, it can have wide effects across neutral regions due to genetic linkage. In the context of purifying selection at linked sites, this process is called 'background selection' (BGS) and it leads to decreases in genetic diversity and skews the distribution of allele frequencies in the genome. While much theoretical and empirical investigation has gone into how BGS and demography operate independently to affect the genome, little investigation has been conducted on how these forces pattern the genome in concert. Utilizing thousands of human genomes and population genetic simulations, I have determined that the effects of BGS in humans can be magnified by population demography. I also analyzed population genetic simulations of different demographic models with BGS and found that the effects of demography and BGS are transient through time, with dips and rises in genetic diversity dependent on how far removed they are from a demographic event. Finally, in order to gain an understanding of how BGS and recent human population growth have affected patterns of the allele frequency spectrum, I analyzed genomes of varying sample size as a function of the strength of BGS. Doing so, I found that the effect of BGS on skewing the allele frequency spectrum towards rare variants in humans is dependent on sample size and leads to larger biases in demographic inference when sample size is small.

Understanding the genetic basis of complex phenotypes is a critical problem in medical and evolutionary genetics. The evolutionary forces of natural selection and demography have shaped patterns of worldwide genetic variation, which in turn have shaped the genetic architecture of human phenotypic variation. Admixed populations, including African Americans and Latinos, have recent ancestry from two or more ancestral groups and are highly underrepresented populations in human genetics research. As a result, the genetic variation that contributes to the genetic architecture of complex traits in these populations has largely been undefined. Here through a combination of data analysis, population genetic modeling and statistical genetics, we further our understanding of admixed populations and highlight the importance of studying diverse populations. First, in a study of bronchodilator drug response (BDR), we identified both population specific and shared genetic variants associated with differences in BDR in African American and Latino children with asthma. Second, in a study of Hispanics/Latinos, we show that admixture has been a dynamic process in the recent history of Mexican Americans, with ancestry proportions changing over time due to a complex mixture of small effects from several population and cultural factors. Finally, we draw attention to the biases and potential for continued health disparities that persist when utilizing genomic prediction based only on large samples of European individuals in Mexican Americans. Through these studies, we improved upon our understanding of the genetic diversity within admixed populations, its effects on human phenotypic diversity, and subsequently our ability to understand genetic contributions to complex traits and disease.

The genetic architecture of complex human traits remains largely unknown. The distribution of heritability across the minor allele frequency (MAF) spectrum for a trait will be a function of the MAF of its causal variants and their effect sizes. Assumptions about these relationships underpin the tools used to estimate heritability. We examine the performance of two widely used tools, Haseman-Elston (HE) Regression and genomic-relatedness-based restricted maximum-likelihood (GREML). Our simulations show that HE is less biased than GREML under a wide-variety of models, and that the estimated standard error for HE tends to be substantially overestimated. We then applied HE Regression to infer the heritability of 72 quantitative biomedical traits from up to 50,000 individuals with genotype and imputation data from the UK Biobank. We found that adding each individuals’ geolocation as covariates corrected for population stratification that could not be accounted for by principal components alone (particularly for rare variants). The biomedical traits we analyzed had an average heritability of 0.27, with low frequency variants (MAF≤0.05) explaining an average of 47.7% of the total heritability (and lower frequency variants with MAF≤0.02 explaining a majority of our increased heritability over previous estimates). Variants in regions of low LD accounted for 3.3-fold more heritability than the variants in regions of high LD, an effect primarily driven by low frequency variants. These findings suggest a moderate action of negative selection on the causal variants of these traits.

Over 2000 years ago, Heraclitus noted, "Everything changes and nothing stands still ." While this truth has long been evident to the wise, we have only recently developed the tools necessary to scientifically characterize sweeping patterns of change in large dynamical systems. Despite rapid progress, new methods and data sources are still sorely needed to further illuminate the intricate and dynamic nature of reality. In this dissertation, we will focus our investigations on understanding patterns of change with direct relevance to human health. In the first two chapters, we develop novel methodologies that lend insight into the evolutionary history of the human race and the genetic basis of human-specific traits and disease. Chapter 2 presents MOSAIC, a new python package for improved detection of genetically related genes between species. This inference is a foundational step towards understanding the function of proteins and the evolutionary pressures they have faced. This tool, along with a combination of other methods, facilitates our analysis in Chapter 3. In this section, we use the patterns of mutations along the human lineage to discover genes and even specific mutations that may play important roles in intelligence, obesity, mental health, as well as a variety of basic biological functions. These findings provide insight into the genetic architecture of health and disease. At the same time, they leave open questions about how genetic factors interact with the broad array of environmental and ecological variables that fundamentally shape downstream phenotypes. In Chapter 4, we introduce CauseMap, a tool I built to understand causal relationships within complex dynamical systems using time series data. It is our hope that this method will help us to interpret human health and disease as states of the bodily dynamical system embedded inextricably within an evolving social, economic, and environmental network. This perspective, we hope, will allow us to understand the characteristics of human health that emerge from an time-hewn dynamic equilibrium with the world within and around us.

Evolutionary forces such as recombination, demography, and selection can shape patterns of genetic diversity within populations and contribute to phenotypic variation. While theoretical models exist for each of these forces independently, mathematically modeling their joint impact on patterns of genetic diversity remains very challenging. Fortunately, it is possible to perform forward-in-time computer simulations of DNA sequences that incorporate all of these forces simultaneously. Here, I show that there are trade-offs between computational efficiency and accuracy for simulations of a widely investigated model of recurrent positive selection. I develop a theoretical model to explain this trade-off, and a simple algorithm that obtains the best possible computational performance for a given error tolerance. I then pivot to develop a framework for simulations of human DNA sequences and genetically complex phenotypes, incorporating recently inferred demographic models of human continental groups and selection on genes and non-coding elements. I use these simulations to investigate the power of rare variant association tests in the context of rampant selection and non-equilibrium demography. I show that the power of rare variant association tests is in some cases quite sensitive to underlying assumptions about the relationship between selection and effect sizes. This work highlights both the challenge and the promise of applying forward simulations in genetic studies that seek to infer the parameters of evolutionary models and detect statistical associations.

The role of rare variants in complex disease is hotly debated, but the design of genetic association studies to statistically associate rare variants is not well understood. Here, we simulate rare variant association studies across different case/control panel sampling strategies, sequencing methods, and genetic architecture models based on evolutionary forces to determine the statistical performance of RVATs widely in use. We find that the highest statistical power of RVATs is achieved by sampling case/control individuals from the extremes of an underlying quantitative trait distribution. We also demonstrate that the use of genotyping arrays, in conjunction with imputation from a whole genome sequenced (WGS) reference panel, recovers the vast majority (90%) of the power that could be achieved by sequencing the case/control panel using current tools. Finally, we show that the statistical performance of RVATs decreases as rare variants become more important in the trait architecture. Our work shows that RVATs are not yet well-powered enough to make generalizable conclusions about the role of rare variants in complex trait architectures.

Objectives

Methods

Results

Conclusions