Genetic variants shape single-cell chromatin accessibility and disease risk

Understanding how genetic variation shapes gene regulation at single-cell resolution is crucial for elucidating complex disease mechanisms¹, but has been limited by the lack of large-scale single-cell multi-omics data^2-4. Here we present single-cell assay for transposase-accessible chromatin with sequencing of 3.5 million peripheral blood mononuclear cells from 1,042 donors with matched whole-genome sequencing. Across 28 immune cell types, we tested 440,996 chromatin accessibility peaks and identified 243,225 chromatin accessibility quantitative trait loci (caQTLs) at 100,483 peaks, 60% of them cell-type specific. Rare variant analysis identified 38,161 associated peaks, with predicted transcription factor-binding disruption implicated at individual loci. Colocalization with expression QTLs (eQTLs) from single-cell RNA sequencing data (5.4 million peripheral blood mononuclear cells)⁵ defined 28,834 candidate cis-regulatory elements. Across genome-wide association studies for 16 diseases and 44 blood traits, caQTLs colocalized 4.5-22.6% more signals than eQTLs alone. Cell-type-specific and tissue-specific context, distal effects and multi-causal-variant models raised genome-wide association study-eQTL concordance from approximately 16% to approximately 65%. Cell state further modulates genetic effects, with 27% of over 200,000 caQTL-cell-state interactions lacking a main effect. At the inflammatory bowel disease risk gene ADCY7, accessibility effects varied with genotype and G protein-mediated pathway activity in CD4⁺ T cells. A graph neural network trained on both eQTL and caQTL layers inferred peak-gene links in unpaired multiome data with up to 17.2% higher accuracy than QTL-agnostic data, uncovering 128 additional transcription factor-target pairs (22% increase). Together, these results show that genetic variants shape chromatin accessibility in a cell-type-specific and cell-state-specific manner, influencing immune function and disease risk.