High resolution genetic mapping of causal regulatory interactions in the human genome

Abstract

Physical interaction of distal regulatory elements in three-dimensional space poses a significant challenge for studies of common disease, because noncoding risk variants may be substantial distances from the genes they regulate. Experimental methods to capture these interactions, such as chromosome conformation capture (CCC), usually cannot assign causal direction of effect between regulatory elements, an important component of disease fine-mapping. Here, we developed a statistical model that uses Mendelian Randomisation within a Bayesian hierarchical model framework, and applied it to a novel ATAC-seq data from 100 individuals mapping over 15,000 putatively causal interactions between distal regions of open chromatin. Strikingly, the majority (>60%) of interactions we detected were over distances of <20Kb, a range where CCC-based methods perform poorly. Because we can infer the direction of causal interactions, the model also significantly improves our ability to fine-map: when we applied it to an eQTL data set we reduced the number of variants in the 90% credible set size by half. We experimentally validate one of our associations using CRISPR engineering of the BLK/FAM167A locus, which is associated with risk for a range of autoimmune diseases and show that the causal variant is likely to be a non-coding insertion within a CTCF binding motif. Our study suggests that many regulatory variants will be challenging to map to their cognate genes using CCC-based techniques, but association genetics of chromatin state can provide a powerful complement to these approaches.