Impact of confounding by leukocyte composition on associations of leukocyte DNA methylation with common risk factors.

Abstract

One concern in epigenome-wide studies investigating leukocyte DNA methylation is that observed associations may at least partly reflect differences in leukocyte composition (LC) rather than changes in methylation. We estimated the magnitude of confounding by LC for common risk factors and diseases. Variation of LC according to sex, race, age, smoking, alcohol consumption, BMI, cardiovascular fitness, hypertension, coronary heart disease and diabetes was analyzed using blood differentials from 4117 participants of NHANES. Furthermore, leukocyte DNA methylation levels of biomarkers of smoking, BMI, diabetes, age and sex were regressed on these outcomes in a sample of 989 participants of ESTHER, and regression coefficients with and without adjustment for estimated LC were compared. Aside from race and ages below 25 years, none of the investigated factors had substantial impact on LC. Adjusted and unadjusted coefficients were virtually identical. Confounding by LC might often be a minor issue.