Mapping gene regulatory architectures driving human disease using single-cell data

Abstract

More than 90% of disease-associated genetic variants implicated in genome-wide association studies (GWAS) lie outside of protein-coding regions of the genome and are believed to mediate disease risk by regulating the expression of nearby genes. Because the regulatory regions (“enhancers”) harboring these variants often do not regulate the nearest gene, the biological mechanisms underlying these associations remains poorly understood. Genome-wide “single-cell” technologies assay various cellular modalities at the single-cell level and enable the detection of regulatory relationships at unprecedented resolution. However, existing computational methods for linking enhancers to their target genes using high-dimensional single-cell data fail to identify true regulatory relationships, implicate incorrect regulatory relationships, or do not adequately consider context specificity; indeed, methodological advancements in single-cell data analysis are needed to realize the promise of this powerful new data. Single-cell “peak-gene linking” methods link enhancers to target genes by measuring associations between enhancer activity (“peaks” of accessible chromatin) and gene expression across single cells, using single-cell “multimodal” data (which simultaneously profile gene expression and chromatin accessibility). However, existing methods exhibit low concordance and power. We developed a statistical framework that integrates existing linking strategies (including genomic distance) with expression quantitative trait loci (eQTL) data to score each candidate link. Our method outperformed existing methods in several orthogonal evaluation tasks, and restricting to a focal cell type improved power to identify cell type-specific links. We constructed enhancer-gene maps that reveal novel genes and regulatory elements underlying GWAS associations that were not implicated by other methods. Despite this substantial improvement in power, we hypothesized that correlations among peaks of accessible chromatin may induce non-causal tagging enhancer-gene links. Indeed, we demonstrate that tagging effects induced by “co-accessibility” (correlations among peaks) are pervasive, by analyzing tagging correlations in CRISPR-tested enhancer-gene links and co-accessibility stratified by functional peak categories. We show that the binding of transcription factors (TFs) (particularly pioneer TFs, which activate repressed chromatin regions) at multiple sites across the genome drives tagging effects. We also determined that statistical “fine-mapping” to distinguish causal from tagging associations improves peak-gene linking. These findings underscore the importance of accounting for tagging effects when linking enhancers to target genes. We further investigated cell state-dependent genetic effects on gene expression, which may often play an important role in disease etiology. Expression quantitative trait loci (eQTL) studies link regulatory variants to target genes by measuring associations between genotype and gene expression across multiple individuals. Using a single-cell eQTL mapping framework modeling cell state, we demonstrate that gene expression is often regulated by cell state-dependent genetic effects and that these “dynamic” regulatory effects are often independent from “main” effects (detected without accounting for cell state). We provide several examples demonstrating context-dependent regulatory architectures underlying GWAS associations. Characterizing rich and multi-layered gene regulatory architectures by linking enhancers to target genes is critical to advancing our understanding of disease etiology and elucidating potential drug targets. Here, we leverage the unprecedented power and resolution of genome-wide single-cell data to develop a new method for linking enhancers to their target genes, characterize patterns of enhancer co-accessibility inducing spurious regulatory links, and investigate cell state-dependent genetic effects on gene expression.