Table 2 shows environmental pollutants that reached the significance threshold (Cjoint of 0.01) for each lipid outcome and their estimated weights (regression coefficients) for ERS from single-pollutant models (ERS1) and a multi-pollutant model (ERS2). Figure S1 presents visual distributions of the P values for the individual environmental pollutants examined in the Stage-1 samples (Manhattan plot [41]). Out of 134 environmental pollutants, 11, 9, 5 and 23 pollutants were significantly associated with total cholesterol, HDL, LDL, and triglycerides, respectively, in single pollutant models (marginal analyses) with adjustment for the base covariates and phenotype-specific nutrients. Note that the weights in Table 2 are the regression coefficients for each log-transformed exposure in relation to the log-transformed lipid outcome, which are not directly interpretable. Generally, percent changes for a two-fold increase in exposure concentrations are presented as [exp(regression coefficient×log(2)) –1]×100%. For example, a two-fold increase in blood lead was associated with a 19% higher levels of total cholesterol ([exp(1.71×log(2)) –1]×100% = 19%). Since we used these weights to construct ERS rather than interpret the associations of individual pollutants, we presented the direct weights rather
