DNA methylation signatures associated with bipolar disorder in peripheral blood improve prediction models

Bipolar disorder (BD) is a major mood disorder influenced by both genetic and environmental factors. While DNA methylation from peripheral tissues can reflect both genetic and environmental influences and reveal insights into disease biology, it remains understudied in BD. DNA methylation signatures may complement polygenic scores (PGS) and hold potential as biomarkers. Here, we conducted the largest epigenome-wide association study (EWAS) of BD to date and evaluated the predictive value of polymethylation scores (PMS) in classifying case-control status.

DNA methylation from peripheral blood of 1729 cases and 1747 controls, comprising twelve cohorts, was obtained. We performed meta-analyses for the total sample, male-only, and female-only analyses. Differentially methylated regions (DMRs) were identified using the comb-p method. Polymethylation scores for BD (BD-PMS) were tested for association with BD, and in combination with PGS.

We identified 47 differentially methylated CpG positions (DMPs) in the total and four in the female-only analysis. Ninety, fourteen and six DMRs were identified in the total sample, female-only, and male-only analyses, respectively. Genes annotated to the top DMPs were enriched for immune activation and phosphorylation pathways. DMRs were annotated to genes relevant to neurotransmission, including GABBR1 and CACNA2D4. BD-PMS explained 2% of the variance in BD case-control status, and improved the variance explained from 7.9 to 8.5% when combined with PGS. For bipolar I disorder, BD-PMS explained 4.9% of the variance, and improved the variance explained by PGS from 15.9 to 18.5%. Association of BD with PMS for schizophrenia and major depression suggests pleiotropic epigenetic effects.

DNA methylation signatures of BD are detectable in blood using adequately powered data and may reveal novel BD biology that is not captured by genetic studies. PMS from large cohorts have the potential to facilitate the development of prediction tools to aid clinical decision-making.

This investigation was primarily funded by the Research Council of Norway (RCN #250299, #273446, #223273) and the University of Bergen. A complete list of funding organisations is provided in the Acknowledgements.