ABSTRACT: Major depressive disorder (MDD) is a leading cause of disability and reduced life expectancy, with a two-fold increase in prevalence in women compared to men. Over the last few years, identifying reliable molecular biomarkers of MDD has proved challenging, likely reflecting the fact that, in addition to sex-differences, a variety of environmental and genetic risk factors are implicated. Recently, epigenetic processes have been proposed as mediators of the impact of life experiences on functional regulation of the genome, with the potential to contribute to MDD biomarker development. In this context, here we characterized and integrated transcriptomic gene expression data with two upstream mechanisms for epigenomic regulation, DNA methylation and micro-RNAs. The 3 molecular layers were analyzed in peripheral blood samples from a well-characterized cohort of individuals with MDD (n=80) and healthy controls (n=89), and processed using 3 complementary bioinformatic strategies. First, we conducted case-control comparisons for each single omic layer, and contrasted sex-specific adaptations. Second, we leveraged network theory to define gene co-expression modules, followed by step-by-step annotations across omic layers. Finally, we implemented a genome-wide and multiomic integration strategy that included cross-validation and bootstrapping. The approach was used to systematically compare the accuracy of MDD prediction across 6 methods for dimensionality reduction and, importantly, for every combination of 1, 2 or 3 types of molecular data. Results showed that accuracy was higher when female and male cohorts were analyzed separately, rather than combined, and also progressively increased with the number of molecular datasets considered. While multiomic informational gain has already been illustrated in other medical fields, our results pave the way towards similar advances in molecular psychiatry, and have practical implications towards developing clinically useful biomarkers of MDD.
