Identifying causal genetic variants and molecular mechanisms impacting mental health

Identifying how genetic variation leads to neurodevelopmental or psychiatric disorders provides new means to study, predict, prevent and treat disease. Identifying the immediate molecular consequences of disease- associated genetic variation has necessitated the development of large-scale, multi-tissue functional genomic resources. Projects such as GTEx, Roadmap Epigenomics Project and PsychENCODE have combined molecular QTL mapping and epigenomic maps in bulk tissues to interpret various disease-associated genetic variants. However, few colocalizations between molecular QTLs and traits have been robustly identified and few causal variants mapped. As tissues like the brain constitute 100s of cell-types, we hypothesize that existing maps may mask the contributions of disease-associated variation in less-abundant cell types. One extremely powerful approach to identify cell-type specific molecular effects and their relationship to genetic diseases is through application of chromatin accessibility data – these data both allow inference of causal cell types and provide base level resolution gene regulation. Our team has considerable expertise in connecting GWAS to molecular functions and predicting causal variants through use of chromatin accessibility data. We have additionally recently collaborated to generate a comprehensive, multi-individual map single cell ATAC- seq map (scATAC-seq) of six different brain regions to detect causal cell types and predict causal variants. This work has been recently demonstrated in our fine-mapping study of Alzheimer’s and Parkinson’s disease (Corces et al, bioRxiv, 2020) but has not been systematically applied to mental health disorders. We propose to develop statistical genetics and machine learning approaches that advance the use of scATAC-seq data to connecting mental health GWAS loci to specific cell types, mechanisms and causal variants. In Aim 1, we will assemble a pipeline that leverages region and cell type-specific scATAC-seq data to identify pathological cell types for 100s of mental health and brain-related traits. We will also enhance the detection of cell-type specific molecular mechanisms by extending and applying a novel GWAS/QTL colocalization approach. Throughout these activities, variants will be validated using massively-parallel reporter assays (MPRA). In Aim 2, we will develop sophisticated machine learning models that learn regulatory grammars and score variants across the allele frequency spectrum. Predicted causal variants in GWAS loci will be further assessed with MPRAs in Aim 1 and applied in Aim 3. In Aim 3, we will demonstrate how improved detection of causal variants using our single-cell informed models aids transferability of polygenic risk scores across populations. We will provide open resources and reproducible computational methods and pipelines that integrate single cell chromatin accessibility data from multiple brain regions. This will allow detection cell-type specific genetic effects and pathological cell types in mental health GWAS, establish robust causal links between variants, genes and disease, and improve prediction of disease risk.