Discovering genetic interactions bridging pathways in genome-wide association studies

ARTICLE https://doi.org/10.1038/s41467-019-12131-7 OPEN Discovering genetic interactions bridging pathways in genome-wide association studies 1234567890():,; Gang Fang 1,6, Wen Wang 2,6, Vanja Paunic2, Hamed Heydari 3, Michael Costanzo 3, Xiaoye Liu2, Xiaotong Liu 2, Benjamin VanderSluis 2, Benjamin Oately2, Michael Steinbach 2, Brian Van Ness4, Eric E. Schadt 1, Nathan D. Pankratz 5, Charles Boone3, Vipin Kumar2 & Chad L. Myers 2 Genetic interactions have been reported to underlie phenotypes in a variety of systems, but the extent to which they contribute to complex disease in humans remains unclear. In principle, genome-wide association studies (GWAS) provide a platform for detecting genetic interactions, but existing methods for identifying them from GWAS data tend to focus on testing individual locus pairs, which undermines statistical power. Importantly, a global genetic network mapped for a model eukaryotic organism revealed that genetic interactions often connect genes between compensatory functional modules in a highly coherent manner. Taking advantage of this expected structure, we developed a computational approach called BridGE that identifies pathways connected by genetic interactions from GWAS data. Applying BridGE broadly, we discover significant interactions in Parkinson’s disease, schizophrenia, hypertension, prostate cancer, breast cancer, and type 2 diabetes. Our novel approach provides a general framework for mapping complex genetic networks underlying human disease from genome-wide genotype data. 1 Department of Genetics and Genomic Sciences, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA. 2 Department of Computer Science and Engineering, University of Minnesota, Minneapolis, MN 55455, USA. 3 Donnelly Centre, University of Toronto, Toronto, ON M5S 3E1, Canada. 4 Department of Genetics, Cell Biology, and Development, University of Minnesota, Minneapolis, MN 55455, USA. 5 Department of Laboratory Medicine and Pathology, University of Minnesota, Minneapolis, MN 55455, USA. 6These authors contributed equally: Gang Fang, Wen Wang. Correspondence and requests for materials should be addressed to G.F. (email: ) or to V.K. (email: ) or to C.L.M. (email: ) NATURE COMMUNICATIONS | (2019)10:4274 | https://doi.org/10.1038/s41467-019-12131-7 | www.nature.com/naturecommunications 1 ARTICLE G NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12131-7 enome-wide association studies (GWAS) have been increasingly successful at identifying single-nucleotide polymorphisms (SNPs) with statistically significant association to a variety of diseases1,2 and gene sets significantly enriched for SNPs with moderate association3. However, for most diseases, there remains a substantial disparity between the disease risk explained by the discovered loci and the estimated total heritable disease risk based on familial aggregation4,5. While there are a number of possible explanations for this “missing heritability”, including many loci with small effects or rare variants4, genetic interactions between loci are one potential culprit5,6. Genetic interactions generally refer to a combination of two or more genes whose contribution to a phenotype cannot be completely explained by their independent effects5,7. One example of an extreme genetic interaction is synthetic lethality where two mutations, neither of which is lethal on its own, combine to generate a lethal double mutant phenotype. Thus, genetic interactions may explain how relatively benign variation can combine to generate more extreme phenotypes, including complex human diseases4,5,8. Several studies have reported genetic interactions between specific variants in various disease contexts7,9, and scalable computational tools have been developed for searching for interactions amongst SNPs7,10. However, systematic discovery of statistically significant genetic interactions on a genome-scale remains a major challenge. For example, a theoretical analysis estimated that ~500,000 subjects would be needed to detect significant genetic interactions under reasonable assumptions5, which remains beyond the cohort sizes available for a typical GWAS study or even the large majority of meta-GWAS studies. Genome-wide, reverse genetic screens in model organisms have produced rich insights into the prevalence and organization of genetic interactions11,12. Specifically, the mapping and analysis of the yeast genetic network revealed that genetic interactions are numerous and tend to cluster into highly organized network structures, connecting genes in two different but compensatory functional modules (e.g., pathways or protein complexes) as opposed to appearing as isolated instances11,13. For example, nonessential genes belonging to the same pathway often exhibit negative genetic interactions with the genes of a second nonessential pathway that impinges on the same essential function (Fig. 1a). Owing to their functional redundancy, the two different pathways can compensate for the loss of the other, and thus, only simultaneous perturbation of both pathways (e.g., A* and Y*) (Fig. 1a) results in an extreme loss of function phenotype, which could be associated with either increased or decreased disease risk. Importantly, the same phenotypic outcome could be achieved by several different combinations of genetic perturbations in both pathways (e.g., A-X, A-Z, B-X, B-Y, B-Z) (Fig. 1b). This model for the local topology of genetic networks, called the “between-pathway model” (BPM), has been widely observed in yeast genetic interaction networks11,14. Indeed, as many as ~70% of negative genetic interactions observed in yeast occur in BPM structures, indicating that genetic interactions are highly organized and this type of local clustering is the rule rather than the exception13. In addition to BPMs, combinations of mutations in genes within the same pathway or protein complex also tend to exhibit a high frequency of genetic interaction (Fig. 1b), a network structure referred to as a “within-pathway model” (WPM)11,14. Indeed, ~80% of essential protein complexes in yeast exhibit a significantly elevated frequency of within-pathway interactions15. In the context of human disease, a WPM may reflect an individual that inherits two variants in the same pathway, resulting in reduced flux or function of a particular pathway and an increase or decrease in disease risk. The prevalence of BPM and WPM structures observed in the yeast global genetic network has important practical implications 2 that can be exploited to explore disease-associated genetic interactions in humans based on GWAS data. Although tests to identify genetic interactions between specific SNP or gene pairs are statistically under-powered, we may be able to detect genetic interactions by leveraging the fact that pairwise interactions between genome variants are likely to cluster into larger BPM and WPM network structures similar to those observed in the yeast global (...truncated)