Comparing bulk and single-cell methodologies and models to profile gene expression, chromatin accessibility and regulatory links in endothelial cells treated with TNFα

Genome-wide association studies (GWAS) have identified thousands of non-coding variants associated with complex traits and diseases. However, it remains challenging to pinpoint the causal genes that are regulated by associated genetic variants. Connecting causal non-coding variants with genes can rely on methods that identify direct physical interactions (e.g. chromosome conformation capture) or on probabilistic models that predict regulatory links. These statistical models take advantage of gene expression and chromatin accessibility profiles generated in cells and tissues by bulk or single-cell (sc) methodologies. Here, we tested whether using bulk or sc RNAseq/ATACseq data and corresponding predictive enhancer-to-gene models impact the prioritization of causal GWAS genes. Using non-treated and TNF-treated human endothelial cells in vitro as a well-controlled experimental system, we show that bulk and sc RNAseq/ATACseq profiles are similar and highlight the same biology (e.g. biological pathways). Despite these similarities, we show using GWAS results for coronary artery disease (CAD) and diastolic blood pressure that applying enhancer-to-gene models designed for bulk or sc methodologies can yield differences in terms of captured heritability, fine-mapped variants and linked genes. For instance, at one CAD locus, the bulk-based ABC model predicts a regulatory link with BCAR1, whereas the sc-based model scE2G prioritizes a different gene (CFDP1). On the same experimental model, our results indicate that choosing between a bulk or sc approach will influence regulatory link model predictions; this should be considered when planning functional experiments to characterize GWAS discoveries.