DNA methylation proxy for exposure and outcome prediction
1 December 2016
Matthew Suderman presented at the 2016 Epigenomics of Common Diseases conference
Matthew Suderman, Gemma Sharp, Leanne Küpers, Fahimeh Falahi, Janine Felix, Harold Snieder, Ellen A. Nohr, Thorkild I. A. Sørensen, Caroline Relton, George Davey Smith
Many environmental exposures particularly in early development are known to influence later health outcomes.
Unfortunately records of exposure tend to be based on self-report long after the exposure has occurred and therefore lack precision and objectivity. A possible solution is to infer exposure from stable, accessible biomarkers. DNA methylation measured in peripheral tissues is a promising candidate because it is commonly measured in many human cohort studies and is known to retain associations with certain exposures for many years.
As a proof of principle, we created proxies for prenatal tobacco exposure (PTE) using cord blood DNA methylation levels. We then asked how well the proxies correlate with PTE and well-known outcomes of PTE. We were indeed able to derive accurate DNA methylation proxies for PTE and showed that accuracy could be improved by including certain data preprocessing steps as well as including association effect information from large meta-analyses of PTE in DNA methylation. Our proxies also related to known PTE outcomes similarly to PTE but did identify cases of collider bias introduced by certain data preprocessing steps.
In future we will further investigate approaches for improving proxy accuracy and preventing collider bias.
Please see DNA methylation proxy for exposure and outcome prediction (PDF, 718kB) for poster presentation.