Using methylome data to inform exposome-health association studies : An application to the identification of environmental drivers of child body mass index

Background
The exposome is defined as encompassing all environmental exposures one undergoes from conception onwards. Challenges of the application of this concept to environmental-health association studies include a possibly high false-positive rate.

Objectives
We aimed to reduce the dimension of the exposome using information from DNA methylation as a way to more efficiently characterize the relation between exposome and child body mass index (BMI).

Methods
Among 1,173 mother–child pairs from HELIX cohort, 216 exposures (“whole exposome”) were characterized. BMI and DNA methylation from immune cells of peripheral blood were assessed in children at age 6–10 years. A priori reduction of the methylome to preselect BMI-relevant CpGs was performed using biological pathways. We then implemented a tailored Meet-in-the-Middle approach to identify from these CpGs candidate mediators in the exposome-BMI association, using univariate linear regression models corrected for multiple testing: this allowed to point out exposures most likely to be associated with BMI (“reduced exposome”). Associations of this reduced exposome with BMI were finally tested. The approach was compared to an agnostic exposome-wide association study (ExWAS) ignoring the methylome.

Results
Among the 2284 preselected CpGs (0.6% of the assessed CpGs), 62 were associated with BMI. Four factors (3 postnatal and 1 prenatal) of the exposome were associated with at least one of these CpGs, among which postnatal blood level of copper and PFOS were directly associated with BMI, with respectively positive and negative estimated effects. The agnostic ExWAS identified 18 additional postnatal exposures, including many persistent pollutants, generally unexpectedly associated with decreased BMI.

Discussion
Our approach incorporating a priori information identified fewer significant associations than an agnostic approach. We hypothesize that this smaller number corresponds to a higher specificity (and possibly lower sensitivity), compared to the agnostic approach. Indeed, the latter cannot distinguish causal relations from reverse causation, e.g. for persistent compounds stored in fat, whose circulating level is influenced by BMI.