Abstracts

Rationale: Epileptic Encephalopathies are a group of rare devastating childhood epilepsies where the genetic basis for many patients remains unknown. Recently, large cohort massively parallel sequencing (MPS) studies have identified a considerable number of genes with single ‘hits' (variants). These candidate genes require additional evidence (e.g., a second case and/or functional support) before their role as a true Epileptic Encephalopathy gene can be confirmed. The ability to prioritize the most promising candidates can enhance these follow-up efforts. Many current candidate gene prioritization approaches have a strong reliance on text-mining and are disease agnostic (i.e., utilize non-specific resources). We reasoned a powerful, unbiased data source for the prioritization of Epileptic Encephalopathy candidates would be gene expression data specifically from the brain. By exploring gene expression networks in the developing and adult human brain we aimed to 1) determine the co-expression patterns between known Epileptic Encephalopathy genes and 2) prioritize candidate genes based on integration into Epileptic Encephalopathy expression networks. Methods: Expression data for 20,782 genes was downloaded from the Allen Human Brain Atlas for the developing and adult human brain. Known (n=29) and candidate (n=182) Epileptic Encephalopathy genes were chosen. The pairwise Pearson's correlation coefficient (r) for any two genes determined their level of co-expression. The developing and adult human brain expression resources were analyzed independently and compared to results derived using Celsius: a large, heterogeneous gene expression data warehouse. For each expression data resource (n=3), correlation values between known Epileptic Encephalopathy genes were compared to 1,000 random gene pairs (n=499,500), representing the null distribution. Significant associations were determined by an empirical 95^th percentile cut-off. A candidate gene's number of significant associations with known Epileptic Encephalopathy genes determined their connectivity (K) score. Candidates with K scores exceeding a threshold corresponding to an empirical false discovery rate (eFDR) of 0.25 were prioritized. Results: Known Epileptic Encephalopathy genes were highly correlated for each expression dataset with correlation matrices revealing interesting sub-networks; some brain-specific. Nineteen candidate genes were prioritized with a K score exceeding the eFDR threshold for at least one of the three expression resources. Two of the prioritized genes (GNAO1 and GRIN2B) have since been confirmed true Epileptic Encephalopathy genes. Conclusions: We show that known Epileptic Encephalopathy genes are significantly co-expressed and that gene expression networks can be specific to tissue-type, highlighting the advantage of incorporating disease-specific resources. With support for our prioritization approach yielding highly plausible results (i.e., GNAO1 and GRIN2B) we have identified a further 17 candidates that we believe have a stronger case for being true Epileptic Encephalopathy genes above other candidates.