DNA methylation near MAD1L1, KDM2B, and SOCS3 mediates the effect of socioeconomic status on elevated body mass index in African American adults

Abstract

Obesity and poverty disproportionally affect African American persons. Epigenetic mechanisms could partially explain the association between socioeconomic disadvantage and body mass index (BMI). We examined the extent to which epigenetic mechanisms mediate the effect of socioeconomic status (SES) on BMI. Using data from African American adults from the Atherosclerosis Risk in Communities (ARIC) Study (n = 2664, mean age = 57 years), education, income, and occupation were used to create a composite SES score at visit 1 (1987-1989). We conducted two methylation-wide association analyses to identify associations between SES (visit 1), BMI and cytosine-phosphate-guanine (CpG) sites measured at a subsequent visit (1990-1995). We then utilized structural equation modeling (SEM) to test whether identified sites mediated the association between earlier SES and BMI in sex-stratified models adjusted for demographic and risk factor covariates. Independent replication and meta-analyses were conducted using the Jackson Heart Study (JHS, n = 874, mean age 51 years, 2000-2004). Three CpG sites near MAD1L1, KDM2B, and SOCS3 (cg05095590, cg1370865, and cg18181703) were suggestively associated (P-value < 1.3×10-5) in ARIC and at array-wide significance (P-value < 1.3×10-7) in a combined meta-analysis of ARIC with JHS. SEM of these three sites revealed significant indirect effects in females (P-value < 5.8×10-3), each mediating 7%-20% of the total effect of SES on BMI. Nominally significant indirect effects were observed for two sites near MAD1L1 and KDM2B in males (P-value < 3.4×10-2), mediating -17 and -22% of the SES-BMI effect. These results provide further evidence that epigenetic modifications may be a potential pathway through which SES may "get under the skin" and contribute to downstream health disparities.

Keywords: epigenetics; health disparities; methylation; obesity; socioeconomic status.

Figure 1

(A and B) Path diagram for structural equation model used for testing mediating role of methylation sites in the association of socioeconomic status (SES) score and body mass index, BMI. Panel A represents the model for Atherosclerosis Risk in Communities (ARIC) females; all measures are from visit 2/3, except SES. Panel B represents the model for Jackson Heart Study (JHS) females, which only includes data from visit 1. All equations included the following covariates: Age (centered); age² (centered); smoking; alcohol use; 10 principal components; center and time between visits/10 (ARIC only). Estimated effect sizes (with standard errors in parentheses) are shown along the path lines. Error terms are associated with each CpG and BMI. ^*P-value < 5×10⁻², ^**P-value < 1×10⁻⁴.

Figure 2

(A and B) Effect sizes (standard errors) in kg/m² of the effect of SES score on BMI (direct effect) through the specified CpG site (indirect effect) in the structural equation modeling of Atherosclerosis Risk in Communities (ARIC) and Jackson Heart Study (JHS) for females (panel A) and males (panel B). Results are plotted left to right as the highest to lowest percent of the annotated total effect (i.e. direct and indirect effects) that is mediated by the indirect effect in ARIC females; percent mediated estimates were not calculated for non-significant (P-value ≥ 5×10⁻²) indirect effects in a specific study.

Figure 3

(A and B) Path diagram for structural equation model for testing the mediating role of CpG methylation, cg18181703 near SOCS3, in the association between socioeconomic status (SES, visit 1) score and body mass index (BMI) at visit 4 in Atherosclerosis Risk in Communities (ARIC) females (panel A unadjusted; panel adjusted for earlier BMI at visit 2/3). All equations included the following covariates: Age (centered); age² (centered); smoking; alcohol use; 10 principal components; center and time between visits/10. Estimated effect sizes (with standard errors in parentheses) are shown along the path lines. Error terms are associated with each CpG at visit 2/3 and BMI at visit 4. ^*P-value < 5×10⁻², ^**P-value < 1×10⁻⁴.