A case-control epigenome-wide study of schizophrenia

Schizophrenia (SZ) is severe neuropsychiatric disorder with significant costs to the individual and society. Accumulating evidence points to the role of DNA methylation (DNAm) in either pathogenesis or as a biomarker of disease risk. Studies done to date have been underpowered to detect the small methylation changes that have been observed in studies of other common diseases, and have not adequately addressed confounding and batch effect issues. In this work, we present an epigenome-wide study of SZ cases and controls using 1334 samples from three multi-site consortia: the Consortium on the Genetics of Endophenotypes in Schizophrenia [COGS], the Project among African-Americans To Explore Risks for Schizophrenia [PAARTNERS], and the Multiplex Multigenerational Family Study of Schizophrenia [MGI]. DNAm levels of 456,513 CpGs were assayed using the Infinium HumanMethylation450 array (Illumina). Findings were replicated in an independent dataset of 497 individuals obtained from the Genomic Psychiatric Cohort [GPC]. Robust linear regression adjusting for age, gender, race, smoking status, batch effects and cell-type heterogeneity revealed 923 SZ-associated differentially methylated positions (SZ-DMPs) in the discovery set, of which 172 were replicated in an independent dataset, and some are located in one of the top SZ loci identified by GWAS. To understand the functional significance of these methylation changes, we performed a cross-tissue comparison using methylation data obtained from 431 post-mortem prefrontal cortex from SZ cases and controls and tested for association with cognitive endophenotypes. The analysis revealed spatial processing as a critical domain affected in SZ, and differential methylation at 17 CpGs overlapped between blood and brain tissues, corresponding to 14 genes involved in neurodevelopment. Gene ontology analysis revealed overlap of functional pathways among the top results obtained in blood and brain discovery datasets. We also developed a statistical method that estimates and adjusts for cell-type composition by decomposing neuronal and non-neuronal differential methylation signatures. We applied this algorithm to find cell-specific differentially methylated regions (DMRs) between prefrontal cortex and hippocampus, and demonstrated the universal utility of this method by applying it to both CHARM and Infinium HumanMethylation450 array data. Taken together, the methylation results and integration with genotypic, endophenotypic data and across tissues, as well as and the statistical framework presented here, constitutes an important contribution to the field of psychiatric epigenomics