A STATISTICAL FRAMEWORK FOR ENVIRONMENTAL EPIGENETICS

Abstract

Several lines of evidence suggest that epigenetic mechanisms such as DNA methylation can mediate exposureresponse relationships, including possible transgenerational effects like the effect of grandmaternal smoking during pregnancy on asthma in the second generation seen in the Children's Health Study. We propose a latent variable modeling framework specified in terms of 1 the risk of disease given exposure, genotype, and methylation, 2 the level of acquired methylation given exposure and inherited methylation, 3 the inherited methylation given parental methylation, and 4 the error structure ofmethylation measurements. By simulation, we show that all parameters of the model are estimable if the model is correctly specified. In a sample size of 1000 3-generation pedigrees, an RR of 3.2 (comparing 0 vs. 100% methylation) for the mediating effect of methylation on exposure-response relationships is detectable with 90% power. Coefficients of variation for other parameter estimates were 1.8% for the exposure effect and 3.2% for the methylation effect in submodel 2, and 6.7% for the transmission effect in submodel 3. We have also shown associations of PM and ozone with DNA methylation of inducible nitric oxide synthase (iNOS), modified by a promoter haplotype in NOS2A, and a 3-way interaction between PM exposure, NOS2A, and iNOS methylation on eNO measurements. This presentation will provide a unifying framework for synthesizing such observations.