University of Edinburgh
Epistatic regulation of transcription factor binding and its impact on human traits
Abstract
dc:description.abstractUnderstanding the genetic architecture of complex human traits and diseases remains a central challenge in human genetics. This thesis explores the interplay between genetic variation, molecular function, and complex phenotypic outcomes, with a particular focus on transcriptional regulation and epistatic interactions. Beginning with an overview of Mendelian and complex disease genetics, we trace key historical and conceptual developments, highlighting how genome-wide association studies (GWAS) and statistical modelling have shed light on genetic contributions to human traits. A critique of traditional statistical methods, including generalized linear models, linear mixed models, and whole-genome regression, is followed by an exploration of emerging semi-parametric and causal inference techniques that can be used to model these genetic effects on traits. One genetic contribution to complex traits is epistasis, whereby multiple variants interact and modulate the effect of each other. Epistasis is thought to be pervasive in complex human traits but has thus far been difficult to detect. We address two of the major challenges in genome-wide studies of epistasis: the enormous multiple testing burden when evaluating all possible variant-variant interactions, which is statistically and computationally burdensome, and the lack of interpretability of interaction effects. To do so, we focus on a high-confidence set of variants that influence transcription factor (TF) binding affinity. The first part of this work discusses the framework developed to identify a high-confidence set of such DNA variants. To this end, we present baal-nf, a computational pipeline for identifying high-quality allele-specific binding (ASB) events from raw Chromatin Immunoprecipitation sequencing (ChIP-Seq) data. This framework not only integrates open-source tools for data quality control, pre-processing and allelic imbalance inference, but also performs de novo motif discovery and maps inferred ASBs onto known and unknown cofactor binding sequences (termed ‘biologically relevant motifs’). We first show how the workflow was developed to identify high-quality de novo motifs and characterize these with respect to known motifs for the ChIP’d transcription factor (TF). Subsequently, this tool was applied to over 6 terabytes of input ChIP-Seq data to infer a novel ASB database of nearly 300,000 ASBs spanning 558 TFs. We also discuss a comparison of these ASBs to other methods that have inferred ASB events on a large scale. Within this set of approximately 300,000 are 1,935 high-quality ASBs that disrupt binding within biologically relevant motifs by breaking the motif, are evolutionarily conserved and frequently colocalize with molecular and trait-associated quantitative trait loci (QTLs). The next section of this thesis introduces TarGene, a state-of-the-art framework for semi-parametric estimation of genetic effects on complex traits. This methodology allows flexible, data-driven modeling of these effects, avoiding the restrictive parametric assumptions that are currently employed in linear and logistic modelling-based approaches. Novel contributions extending this software are described here, enabling its application across diverse biobank-scale cohorts beyond the UK Biobank (such as the All of Us cohort), allowing replication studies at the biobank-level and facilitating the investigation of genetic interactions involving multiple transcription factor mechanisms simultaneously. To support these analyses, scalable and reproducible workflows were developed for key pre-processing steps, including principal component analysis (PCA) to identify genetic confounders and the identification of linkage disequilibrium (LD) blocks (genomic regions containing highly correlated variants) around the DNA variants of interest. These workflows are implemented as portable Nextflow pipelines, ensuring reproducibility and scalability across both high-performance compute (HPC) and cloud-based computational environments. Lastly, we discuss a large-scale application of TarGene to investigate epistatic interactions between binding QTLs (bQTLs) of 9 nuclear hormone receptors and other trans-acting loci across 768 human traits in the UK BioBank. To achieve this aim, we leverage the ASBs presented in the first part of this thesis as ‘candidate bQTLs’, derive trans-acting loci from publicly available sources, and model genetic interactions using TarGene for semi-parametric estimation. Through this approach, previously uncharacterized epistasis was uncovered, as well as higher order interactions with genetic sex, supporting claims that epistasis is widespread, and implicating transcription factors as master regulators. Together, this work integrates functional genomics, population genetics, state-of-the-art statistical estimation techniques and large biobanks to advance our understanding of regulatory variation in genetic interactions driving complex traits. The approaches developed herein have broad applicability in both understanding the genetic underpinnings of complex disease via transcription factor mechanisms and developing genetically-informed therapeutic targets for precision medicine applications.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Roskams-Hieter, Breeshey
- Advisors dc:contributor.advisor
-
- Ponting, Chris
- Khamseh, Ava
- Beentjes, Sjoerd Viktor
Subjects
dc:subject × 6Rights
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.7488/era/7201