W14. IMPLICATING NEURAL CELL TYPES FOR PSYCHIATRIC DISORDERS AND NEUROLOGICAL DISEASES WITH NOVEL MIXER METHODS

Psychiatric disorders and neurological diseases are complex brain-related traits with shared and distinctive neurobiology. Previous genetic studies implicate various neuronal cell types for psychiatric disorders while brain-resident immune cells are implicated for several neurological diseases. Studies typically investigate each disorder separately. However, much can be gained from the cross-trait modelling of genetic signal. Therefore, we have expanded the novel gene set analysis MiXeR (GSA-MiXeR) tool to model bivariate partitioned-heritability. We apply bivariate GSA-MiXeR to pairs of psychiatric disorders and neurological diseases to identify shared and distinct genetic signal for various neural cell-types. We use genome-wide association study summary statistics for three psychiatric disorders (bipolar disorder (BD), depression, and schizophrenia (SCZ)) and three neurological diseases (Alzheimer’s disease (AD), multiple sclerosis, and Parkinson’s disease) as well as largescale single-cell RNA sequencing data from fetal and/or adult brain donors. For each single-cell dataset, we estimate an expression proportion for each gene to rank their cell-type specificity. The top 10% of genes for each cell type is used to define gene sets for all analyses. In a sample of 267 BD and 270 SCZ cases with 745 controls, we apply the recently developed MiXeR-Pred tool to compare performance of a cell-type polygenic risk score (PRS) derived from a single trait (BD or SCZ) to a PRS derived from a pair of traits (i.e., BD-with-SCZ or SCZ-with-BD). To develop and validate the bivariate GSA-MiXeR tool, we run extensive simulations. A main feature of the tool is the ability to model trait-specific and shared genetic signal which contribute to a cross-trait Akaike Information Criterion (AIC) value. This value can be used for selection of relevant gene sets for one or both modelled traits, similar to an omnibus test. Current simulations reveal that gene sets with shared genetic signal between traits have a higher cross-trait AIC than gene sets that are specific to a single trait. Consistent with previous reports, using bivariate GSA-MiXeR we find neurological diseases exhibit heritability fold-enrichment for brain-resident immune cells. Comparatively, enrichment in psychiatric disorders for the same immune cells was lower. Psychiatric disorders exhibited high enrichment for subclasses of inhibitory neurons. In general, the shared cross-trait signal across cell types was larger amongst psychiatric disorders than neurological disorders. For pairs of traits with shared genetic signal for a cell-type, PRS analyses reveal a subtle increase in performance using a bivariate cell-type PRS compared to trait-specific. For example, there was shared genetic signal for BD and SCZ for neuronal genes. The MiXeR-Pred neuronal-PRS for BD sourcing SCZ genetic signal had a larger R-squared than the neuronal-PRS derived from BD alone. Comparatively, astrocyte genes did not have shared genetic signal for BD and SCZ and the astrocyte-PRS for BD sourcing SCZ did not outperform the astrocyte-PRS for BD alone. Altogether, we illustrate neural cell associations with psychiatric disorders and neurological diseases using the novel bivariate GSA-MiXeR. We observe shared genetic signal across traits and validate these findings using MiXeR-Pred PRS performance. These tools can be further utilized to probe and understand the underlying biology of differing traits.