19409
The Discovery of Gene Modules for Autism Utilizing Co-Expression and PPI Networks
Despite extensive genetic heterogeneity underlying disorders such as autism spectrum disorders (ASD) and intellectual disability (ID), there is compelling evidence that risk genes will map to a much smaller number of biologically functional modules. However, discovering and distinguishing between these modules is still a major challenge.
Objectives:
We introduce a novel computational method (MAGI) for the discovery of disease modules enriched for mutations in probands and apply it to the recently published de novo mutations from ASD/ID samples to discover ASD biomolecular modules with specific phenotypic properties.
Methods:
MAGI simultaneously considers protein-protein interaction and RNAseq expression profiles during brain development. It is based on a combinatorial optimization algorithm that aims to find modules of genes that maximize the number of mutations seen in cases and limit the number of mutations observed in controls, while ensuring a high number of protein interactions and high coexpression among these genes. The method first finds a set of small seed pathways, enriched in de novo mutations, and then merge them into larger modules using a local search approach.
Results:
Applying the method to published exome sequencing data from 1,116 ASD and ID patients, we discovered two distinct modules (p < 0.005) that differ in their properties and associated phenotypes. The first module consists of 80 genes associated with the Wnt and Notch signaling pathways. In addition the module is significantly enriched in genes associated with chromatin remodeling and transcription regulations. Probands with truncating mutations in this module are enriched for micro and macrocephaly (KS test p = 0.013). The second module associated with synaptic function, including long-term potentiation and shows higher levels of postnatal expression. Probands with de novo mutations in these modules are found to have lower IQ. In addition, missense mutations in both modules are predicted to be more deleterious. Applying the method independently to epilepsy and schizophrenia exome sequencing cohorts, we found marked overlap among modules suggesting shared common neurodevelopmental pathways. Furthermore, analyzing the full Simon Simplex Collection (SSC) and Autism Sequencing Consortium (ASC) studies revealed an 8.1-fold enrichment of newly discovered de novo mutations in these two predicted modules (p < 10-8). Preliminary results of adding mutations found in the full Simons Simplex Collection (SSC) to the analysis (2,661 probands in total) identification more refined and biologically coherent modules (e.g. SWI/SNF complex, Wnt signaling, long-term potentiation, proteasome, and ubiquitin-mediated proteolysis pathways).
Conclusions:
Our approach provides a molecular framework for reducing the genetic heterogeneity of these diseases and a method for identifying de novo missense mutations important in ASD etiology. We believe that using MAGI to find more refined gene modules for ASD will not only improve our understanding of relevant biological pathways important for neuronal development in general and autism specifically, but also has the potential to further refine distinguished subtypes of ASD that may lead to the development of future specific disease therapy.