Coordinated epistasis reveals symptom-specific polygenic pathway interactions in major depressive disorder

Patients diagnosed with Major Depressive Disorder (MDD) often exhibit very diverse symptom profiles, some of which have no symptoms in common. As a result, MDD is a highly heterogeneous disorder. Mounting literature suggests that MDD is likely the common outcome of many different etiological pathways. In this project, we aim to identify polygenic pathways that contribute to symptom heterogeneity in MDD. We do this through identification of interactions between their genetic effects on the disorder. We identify polygenic pathway interactions using Coordinated Epistasis (CE). CE tests for interactions between polygenic effects of etiological pathways by aggregating the effects into Polygenic Risk Scores (PRS). For a polygenic disorder like MDD, CE assumes that those genetic effects are distributed across the genome. Therefore, we approximate these pathways with random genome partitions. We characterized this polygenicity assumption through simulation. Here, we randomly partition the genome in half 100 times. This results in 100 interaction effect estimates (γ). We then apply this 100-partition test to MDD and its 14 worst-episode symptoms based on DSM-V specifiers phenotyped in the UK Biobank. To further characterize these pathway interactions, we test for the joint effect of interactions between PRS of all possible pairs of chromosomes. Lastly, we follow-up on significantly interacting chromosome-pairs through SNP-PRS interaction tests. Our simulations using 100-partition CE show that γ is a distribution over the different partitions. Additionally, the standard deviation of this distribution varies with polygenicity. Applying this to MDD and its 14 worst-episode symptoms, we find that MDD and its symptoms have varying polygenic pathway interactions. To identify the genomic regions that most effect these polygenic interactions, we test for CE between PRS of chromosome partitions. We find 2 significantly interacting chromosome pairs in MDD and 64 pairs in 13 out of 14 symptoms (at 10% FDR). Next, SNP-PRS interaction tests on these significant chromosome pairs identified 102 chromosome-wide significant loci. Strikingly, we found that the significant SNP-PRS interactions for specific worst-episode symptoms of MDD align with previously found GWAS hits for a variety of traits and diseases including: MDD disease status, neuroticism, self-injurious behavior and feelings of worry. For example, chromosomes 1 and 11 show significant PRS interactions for the worst-episode symptom hypersomnia (P = 3e-04). The locus on chromosome 11 underlying this interaction (rs4245154, SNP-PRS interaction test P = 1.2e-06) is in an intron of the gene DRD2 and has previously been associated with MDD and neuroticism. We identify, for the first time, polygenic pathway interactions in psychiatric disorder MDD and its symptoms. Our 100-partition simulations show that γ forms a distribution over partitions and simultaneously indicates trait polygenicity. We further find specific loci that contribute to pathway interactions for symptoms of MDD. These loci were previously found associated with MDD and a variety of MDD-related phenotypes. This demonstrates the relevance of these symptom interactions as part of MDD genetic architecture.