Population Genetics for Heart Failure: A Multi-Trait Analysis of Rare, Protein-Coding Variants

Abstract

Efforts to map the genetic architecture of heart failure (HF) have been limited by modest sample sizes and the low power of single-trait rare variant tests. This thesis applies advanced statistical methods to large-scale population data to better understand how rare genetic variants contribute to HF and related phenotypes. We leveraged exome sequencing and rich phenotypic data—including heart imaging, blood biomarkers, and clinical diagnoses—from 419,365 UK Biobank participants. We analyzed loss-of-function mutations in 9,404 genes across 13 HF-related outcomes and traits: clinical HF, dilated and hypertrophic cardiomyopathy, eight MRI-derived measures of left ventricular structure and function, and two protein biomarkers. Single-trait analyses recapitulated known associations, most notably implicating the \textit{TTN} and \textit{MYBPC3} genes in clinical HF and cardiomyopathy. To better identify true genetic signals, we proposed the novel application of a multi- trait empirical Bayes model (\texttt{mash})—originally developed for common variant studies—to rare variant analysis. This framework leverages correlations among outcomes to improve effect-size estimation and distinguish true associations from noise. Simulations demonstrated that \texttt{mash} improved both power and accuracy over single-trait testing, though performance declined when inputs were biased or skewed. In real data, the approach nominated 99 genes linked to cardiac structure or function, including novel candidates such as \textit{ENO3} and \textit{NIPSNAP3A}. Many were preferentially expressed in cardiac or skeletal muscle, and several showed that natural variation in their activity was associated with HF risk—supporting biological plausibility. These findings expand the set of genes implicated in HF and related traits. By integrating genomic, imaging, and biomarker data through a unified statistical framework, this study identifies biologically relevant, potentially actionable genes. The results support improved genetic risk prediction and highlight new directions for drug development aimed at preventing or delaying HF progression.