Expansion of Schizophrenia Gene Network Knowledge Using Machine Learning Selected Signals From Dorsolateral Prefrontal Cortex and Amygdala RNA-seq Data

It is widely accepted, given the complex nature of schizophrenia (SCZ) gene networks, that a few or a small number of genes are unlikely to represent the underlying functional pathways responsible for SCZ pathogenesis. Several studies from large cohorts have been performed to search for key SCZ network genes using different analytical approaches, such as differential expression tests, genome-wide association study (GWAS), copy number variations, and differential methylations, or from the analysis of mutations residing in the coding regions of the genome. However, only a small portion (<10%) of candidate genes identified in these studies were considered SCZ disease-associated genes in SCZ pathways. RNA sequencing (RNA-seq) has been a powerful method to detect functional signals. In this study, we used RNA-seq data from the dorsolateral prefrontal cortex (DLPFC) from 254 individuals and RNA-seq data from the amygdala region from 46 individuals. Analysis was performed using machine learning methods, including random forest and factor analysis, to prioritize the numbers of genes from previous SCZ studies. For genes most differentially expressed between SCZ and healthy controls, 18 were added to known SCZ-associated pathways. These include three genes (GNB2, ITPR1, and PLCB2) for the glutamatergic synapse pathway, six genes (P2RX6, EDNRB, GHR, GRID2, TSPO, and S1PR1) for neuroactive ligand-receptor interaction, eight genes (CAMK2G, MAP2K1, RAF1, PDE3A, RRAS2, VAV1, ATP1B2, and GLI3) for the cAMP signaling pathway, and four genes (GNB2, CAMK2G, ITPR1, and PLCB2) for the dopaminergic synapse pathway. Besides the previously established pathways, 103 additional gene interactions were expanded to SCZ-associated networks, which were shared among both the DLPFC and amygdala regions. The novel knowledge of molecular targets gained from this study brings opportunities for a more complete picture of the SCZ pathogenesis. A noticeable fact is that hub genes, in the expanded networks, are not necessary differentially expressed or containing hotspots from GWAS studies, indicating that individual methods, such as differential expression tests, are not enough to identify the underlying SCZ pathways and that more integrative analysis is required to unfold the pathobiology of SCZ.

Automated summary

Given the complexity of schizophrenia gene networks, it is challenging to identify key genes and pathways associated with the disease. In this study, RNA sequencing data and machine learning methods were used to prioritize genes and reveal novel gene interactions related to schizophrenia. The findings provide important insights into the pathogenesis of schizophrenia.