Network QTLs: A new methodology for multi-tissue eQTL discovery. B. Iriarte, M. Kellis, L. Ward, The GTEx Consortium MIT, Cambridge, MA.

   Single nucleotide variants (SNVs) have been associated with gene expression changes in expression quantitative trait locus (eQTL) studies. However, a variant that disrupts or creates a regulatory element can have much more complex effects on gene expression programs than those profiled in single-tissue eQTL studies. To systematically discover such network-level effects leading to more complex changes between different gene expression programs, we developed a new method we call network QTLs. We applied this method to the unique resource of genome-wide gene expression levels across 45 post-mortem tissue samples in a cohort of 175 individuals generated by the Genotype-Tissue Expression project (GTEx). We first learn common expression patterns across the 45 tissues, by clustering the expression vectors of 19,604 protein-coding genes. We find 310 distinct gene expression patterns, which include tissue-specific, tissue-restricted, and ubiquitous expression patterns. We found remarkable enrichments for common gene functions among genes belonging in the same gene expression cluster, with the vast majority of clusters showing distinct GO enrichments, even when they were enriched in similar tissues, suggesting that the subtle differences discovered by the clustering approach are biologically meaningful. We then searched for instances where SNVs between individuals are associated with changes in multi-tissue gene expression patterns, by using module membership as a quantitative trait. After imputing missing data, imputing genotypes, controlling for sex-specific expression, and projecting module-membership coefficients onto the main PCA axes of variation across individuals, we discover 55k independent netQTLs at FDR<0.05 after LD pruning the results, and these are associated with 16,145 target protein-coding genes. The discovered netQTLs include eQTLs discovered independently in individual tissues using MatrixQTL, as expected since netQTLs are a generalization of the classical eQTL discovery approach. However, we also discover a large number of novel QTLs whose effects are too subtle in any individual tissue, but strongly detectable and statistically significant when multiple tissues are combined. These are strongly enriched for regulatory region and regulatory motif annotations by ENCODE and the Roadmap Epigenomics Project, confirming they are biologically meaningful, and suggesting potential regulatory mechanisms for their action.

You may contact the first author (during and after the meeting) at