Abstract 783: Methylation-expression QTLs (meeQTLs) as part of an integrated model of the disruption of gene regulation in cancer

Abstract

Abstract This work describes a new approach aimed at untangling the complex relationship that disrupted methylation patterns share with aberrant gene expression in cancer. Typically, the association between methylation and gene expression is considered through correlation analysis. Anti-correlation of promoter DNA methylation with expression is taken to imply repression of the gene. However, this approach is limited in its ability to consider the effect of methylation on genes other than the one most proximal to a CpG whose methylation value is known. Furthermore, the lack of a statistical model means that covariates may be ignored. Additionally, unnecessary tests may be run (reducing statistical power) when all data points are treated as independent, when in reality CpG loci that are close to each other are often highly correlated. Finally, the effect of copy number variation on gene expression is often not considered. We propose an approach that addresses these issues and finds significant results with large effects even with relatively small sample sizes. The core of our method involves eQTL analysis, but not based on the typical SNP-gene association. Instead, matched tumor/normal data are used to model differentially methylated regions (DMRs) throughout the genome. Highly correlated values are aggregated, resulting in a more stable measure of methylation and fewer tests. The extant of differential methylation for an individual is then assigned as that individual's ‘genotype’ for a DMR. These genotypes are then used for an eQTL analysis, modeling the association of methylation and expression both in cis and in trans. We call this meeQTL analysis, for methylation-expression QTLs. Covariates can also be included in the model. Furthermore, the effect sizes from meeQTL analysis are adjusted, to compensate for the effect of copy number variation on the results. As a proof of concept, we ran our meeQTL analysis on breast cancer data from TCGA, and found thousands of significant meeQTLs, using just 78 matched samples. The top 10 cis-meeQTLs (by effect size) had absolute log-odds between 6 and 9 (i.e. a 1 unit increase in methylation is associated with a 6 to 9 unit change in expression), and include genes such as HIF3A, a negative regulator of hypoxia response genes, or KLF15, which regulates TP53 and NF-kappa B. The top 10 trans-meeQTLs had absolute log-odds between 16 and 19 and include multiple genes associated with a DMR in the first intron of CRADD, a death-domain containing protein. Our model also found that DMRs experience significantly fewer somatic mutations than other loci and identified regions of copy number variation associated with aberrant gene expression in trans. These results suggest our approach can provide robust delineation of methylation and expression relationships arising from the complex structure of the genome and the epigenetic mechanisms regulating its activity. Citation Format: Jeffrey A. Thompson, Carmen J. Marsit. Methylation-expression QTLs (meeQTLs) as part of an integrated model of the disruption of gene regulation in cancer. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 783.