Sequence sensitivity of open chromatin reveals constrained regulatory elements underlying human traits

Abstract

Open chromatin regions (OCRs) delineate cell-type-specific regulatory landscapes across the genome, yet their functional importance varies considerably. Here, we introduce CAMBUS (Chromatin Accessibility Mutation Burden Score), a machine-learning framework that quantifies the sensitivity of predicted chromatin accessibility to sequence perturbation to prioritize functionally important OCRs. Applying CAMBUS to 29 immune cell types, we identified 66,043 CAMBUS-defined OCRs (cOCRs), without directly incorporating human genetic variation or existing evolutionary constraint metrics. Nevertheless, cOCRs were strongly enriched in independently defined constrained regions (odds ratio (OR)=11.45 (95% confidence interval (CI) 9.33-14.05), P=4.7 x 10-68), and active regulatory elements, with enrichment for super-enhancers remaining after adjustment for established constraint metrics and sequence features (OR=6.18 (95% CI 3.80-10.05), P=2.1 x 10-13). cOCRs also showed cell-type-specific enrichment for the heritability of immune-mediated diseases and for putative causal variants identified by statistical fine-mapping across European and East Asian ancestries, including rare variants. Integration with statistical fine-mapping further linked 1,533 putative causal variants across 70 complex traits to cell-type-specific cOCRs, revealing plausible regulatory mechanisms underlying complex traits. Together, CAMBUS provides a region-centered framework for identifying regulatory elements whose predicted activity is sensitive to sequence disruption and that are enriched for evolutionary constraint, complementing existing regulatory and constraint annotations for interpreting non-coding genetic variation.