Shared and unique chromatin accessibility and 3D functional interactions of substance use disorder loci implicate cell types beyond the brain

Objective: Recent genome-wide association studies (GWAS) have revealed multiple loci for substance use disorders (SUDs), including evidence that the traits overlap in genetic etiology. However, the extent of underlying SUD causal variants, effector genes, and cellular contexts, remains unclear. Recent clinical trials of glucagon-like peptide-1 (GLP-1) receptor agonists have shown promising effects on alcohol and tobacco consumption, raising the question of whether SUD genetic risk resides in metabolic as well as neural cell types. We sought to map the cell-type-specific genetic architecture of four SUDs and to test whether the observed metabolic enrichment reflects shared causal biology with type 2 diabetes (T2D). Methods: We integrated our 3D genomic datasets (high-resolution promoter-focused Capture-C/Hi-C, ATAC-seq, RNA-seq) from 59 diverse human cell types with recent GWAS summary statistics for alcohol (AUD), tobacco (TUD), opioid (OUD), and cannabis use disorder (CanUD), using Stratified LD regression (S-LDSC). We performed local genetic correlation analysis (LAVA) between SUDs and T2D, multivariate integration (DIABLO) of gene expression and chromatin accessibility at shared loci, and subsequent bidirectional two-sample Mendelian randomization (MR) to test for causal relationships between T2D and each SUD. Results: After per-trait false discovery rate (FDR) correction, TUD showed significant heritability enrichment in pancreatic {beta}-cells (4.6-fold), -cells (4.0-fold), hypothalamic neurons, and iPSC-derived neurons. CanUD showed FDR-significant enrichment concentrated in hypothalamic neural progenitor cells (7.4-fold). LAVA identified 1,550 loci with significant local genetic correlations, 114 specifically between T2D and at least one SUD. Subsequent MR revealed that genetic liability for T2D was protective against AUD (inverse-variance weighted {beta} = -0.027, P-values = 1.6x10-5, 463 instruments), consistent across four of five MR methods and free of directional pleiotropy. Cell-cell interaction modeling implicated distinct neuro-endocrine signaling axes disrupted by each SUD.