Body Mass Index Drives Changes in DNA Methylation: A Longitudinal Study

Abstract

Rationale: Previous EWASs (Epigenome-Wide Association Studies) suggest that obesity may be the cause, not a consequence, of changes in DNA methylation (DNAm). However, longitudinal observations are lacking.

Objective: To identify 5'-cytosine-phosphate-guanine-3' in DNA (CpG) sites associated with body mass index (BMI) and examine the temporal relationship between dynamic changes in DNAm and BMI in a longitudinal cohort.

Methods and results: Race-specific EWASs were performed in 995 whites and 490 blacks from the Bogalusa Heart Study. Suggestive CpG sites were further replicated in 252 whites and 228 blacks from the Georgia Stress and Heart Study. Cross-lagged panel analysis was used to examine the temporal relationship between DNAm and BMI in 439 whites and 201 blacks who were examined twice 6.2 years apart. In discovery and replication samples, 349 CpG sites (266 novel) in whites and 36 (21 novel) in blacks were identified to be robustly associated with BMI, with 8 (1 novel) CpG sites overlapping between the 2 races. Cross-lagged panel analyses showed significant unidirectional paths (P_FDR <0.05) from baseline BMI to follow-up DNAm at 18 CpG sites in whites and 7 in blacks; no CpG sites showed significant paths from DNAm at baseline to BMI at follow-up. Baseline BMI was associated with a DNAm score (calculated from DNAm levels at the associated CpG sites) at follow-up (P<0.001 both in blacks and in whites).

Conclusions: The findings provide strong evidence that obesity is the cause, not a consequence, of changes in DNAm over time.

Keywords: DNA methylation; body mass index; causality; epigenomics; longitudinal studies; obesity; race factors.

Figure 1.

Study design

Figure 2.. Illustration of a cross-lagged path analysis model of body mass index (BMI) and DNA methylation (DNAm).

ρ₁ and ρ₂ are cross-lagged path coefficients; r₁ and r₂ are tracking correlations; r₃ is the synchronous correlation between DNAm levels and BMI at baseline. Age, sex, smoking, estimated white blood cell counts, and follow-up years were included in the analysis models for adjustment.

Figure 3.. Distribution of BMI at baseline and the DNA methylation risk score (MRS) at follow-up in Whites (a) and Blacks (b).

β, standardized coefficient. In the longitudinal analyses using generalized linear multivariable regression models, the MRS at follow-up was the dependent variable, BMI at baseline was a predictor, and covariates included MRS at baseline, changes in BMI, age at baseline, follow-up years, sex, smoking status, and six types of estimated white blood cell counts at follow-up. Solid dots and error bars were the least square means and its 95% confidence interval of MRS at follow-up by baseline BMI category.