Epigenome-wide contributions to individual differences in childhood phenotypes: A GREML approach

Abstract

Background: DNA methylation is an epigenetic mechanism involved in human development. Numerous epigenome-wide association studies (EWAS) have investigated the associations of DNA methylation at single CpG sites with childhood outcomes. However, the overall contribution of DNA methylation across the genome (R2Methylation) towards childhood phenotypes is unknown. An estimate of R2Methylation would provide context regarding the importance of DNA methylation explaining variance in health outcomes. Methods: We estimated the variance explained by epigenome-wide cord blood methylation (R2Methylation) for five childhood phenotypes: gestational age, birth weight, and body mass index (BMI), IQ and ADHD symptoms at school age. We adapted a genome-based restricted maximum likelihood (GREML) approach with cross-validation (CV) to DNA methylation data and applied it in two population-based birth cohorts: ALSPAC (n=775) and Generation R (n=1382). Results: Using information from >470,000 autosomal probes we estimated that DNA methylation at birth explains 45% (SDCV = 0.07) of gestational age variance and 16% (SDCV = 0.05) of birth weight variance. The R2Methylation estimates for BMI, IQ and ADHD symptoms at school age estimates were near 0% across almost all cross-validation iterations. Conclusions: The results suggest that cord blood methylation explains a moderate to large degree of variance in gestational age and birth weight, in line with the success of previous EWAS in identifying numerous CpG sites associated with these phenotypes. In contrast, we could not obtain a reliable estimate for school-age BMI, IQ and ADHD symptoms. This may reflect a null bias due to insufficient sample size to detect variance explained in more weakly associated phenotypes, although the true R2Methylation for these phenotypes is likely below that of gestational age and birth weight when using DNA methylation at birth.

Epigenetics and ADHD: Reflections on Current Knowledge, Research Priorities and Translational Potential

“…However, a central question in the field is how much variance in ADHD is jointly explained by genome-wide DNAm variation. To address this question, future studies could apply methods recently adapted from genetics, such as aggregate poly-epigenetic risk scores (comparable with PRS [ 77 ]) or methods that rely on genome-wide similarity matrices (comparable with genome-wide complex trait analysis [GCTA] [ 78 ]; for epigenetic data, see [ 79 , 80 ]), although the latter require access to very large datasets. Once this overall signal has been quantified, it will then be possible to compare it with that of other known risk factors for ADHD (e.g., PRS scores, prenatal risks, gestational age, and birth weight), and test whether it adds unique predictive power over and above these factors.…”