LEVERAGING CONTEXT-SPECIFIC EPIGENOMIC REGULATORY NETWORKS (EPINETS) TO DISSECT THE GENETICS OF NEUROPSYCHIATRIC DISORDERS

Abstract

Expression quantitative locus (eQTL) mapping provides deep insights into the function of disease-associated variants from Genome-wide association studies (GWAS). However, previous studies and our research reported misalignment between eQTLs and GWAS signals, likely due to bulk eQTLs mapped in non-pathology-relevant contexts and lack of cell-type resolution. Alternatively, unraveling the links between cis-regulatory elements (CREs) and genes in various cellular contexts offers an independent strategy to associate GWAS variants with their target genes beyond eQTL mapping. The state-of-the-art approaches, such as experimental assays (e.g., Promoter Capture Hi-C and macro-C) and computational models (e.g., ABC and EpiMap), provide linking resources based on different pieces of evidence, however, are each confined to limited brain cell types or cellular states.

Addressing this challenge, our study proposes a machine-learning approach to predict CRE-gene associations by combining protein-protein interactions and transcription factor (TF) binding predictions based on ATAC-seq, an assay measuring genomic accessibility. This computational approach facilitates the discovery of CRE-gene connections across different contexts (combinations of cell types and various conditions) whenever ATAC-seq data are available, enriching our understanding of the cis-regulatory networks between TF-CRE-gene.

We have amassed over 130 cell-sorted and single-cell ATAC-seq datasets encompassing a variety of brain cell types—excitatory neurons, inhibitory neurons, oligodendrocytes, oligodendrocyte progenitor cells (OPCs), astrocytes, microglia, immune cells, and brain vascular cells—under a range of conditions including chemical perturbations, genetic modifications, infections, and disease status. Utilizing this extensive data collection and our integrative pipeline, we have constructed an atlas of TF-CRE-gene linking, namely cEpiNets. We finally employ the atlas to evaluate the enrichment of GWAS signals in CRE modules under various cellular contexts and to prioritize target genes and key drivers across a spectrum of neuropsychiatric disorders.