Gaussian graphical modeling of the serum exposome and metabolome reveals interactions between environmental chemicals and endogenous metabolites

Abstract

Given the complex exposures from both exogenous and endogenous sources that an individual experiences during life, exposome-wide association studies that interrogate levels of small molecules in biospecimens have been proposed for discovering causes of chronic diseases. We conducted a study to explore associations between environmental chemicals and endogenous molecules using Gaussian graphical models (GGMs) of non-targeted metabolomics data measured in a cohort of California women firefighters and office workers. GGMs revealed many exposure-metabolite associations, including that exposures to mono-hydroxyisononyl phthalate, ethyl paraben and 4-ethylbenzoic acid were associated with metabolites involved in steroid hormone biosynthesis, and perfluoroalkyl substances were linked to bile acids-hormones that regulate cholesterol and glucose metabolism-and inflammatory signaling molecules. Some hypotheses generated from these findings were confirmed by analysis of data from the National Health and Nutrition Examination Survey. Taken together, our findings demonstrate a novel approach to discovering associations between chemical exposures and biological processes of potential relevance for disease causation.

Figure 1

Network representation of Gaussian graphical models (GGM) of the serum exposome and metabolome measured in the whole cohort (A), in women office workers (B), and in women firefighters (C). Blue and red nodes represent endogenous metabolites and environmental chemicals, respectively. Edges connecting nodes represent significant partial correlations at FDR < 10%.

Figure 2

Linoleic acid (A,B) and bile acid metabolism (C,D) models inferred from GGMs of women FF and OW. (A) and (C) represent the sub-networks obtained from GGMs. (B) and (D) represent overlapping sub-networks of known biochemical reactions (black edges; labels correspond to regulating genes) obtained from KEGG pathway and from GGMs (red edges).

Figure 3

Exposure-metabolites subnetworks identified by the GGMs inferred from the women FF metabolomics dataset. The subnetwork A represents associations of phenols and phthalates with metabolites involved in steroid hormone biosynthesis. The subnetwork B shows associations between 4-hydroxyacetophenone, PFOS and PFHxS and metabolites involved bile acids, arachidonic and vitamin D metabolism. Blue and red nodes represent endogenous metabolites and environmental chemicals, respectively. Plain and dashed edges connecting nodes represent positive partial correlations and negative partial correlations, respectively. Edges connecting nodes represent significant partial correlations at FDR < 10%.

Figure 4

Odds ratio (95% CI) for metabolic syndrome (MetS) and individual components of MetS in women 20–79 years of age enrolled in NHANES 2003–2014 by quartile of serum PFHxS concentration. Models were adjusted for age, race/ethnicity, poverty, total caloric intake, physical activity, and smoking status. Central obesity: waist circumference ≥ 88 cm; hypertriglyceridemia: blood triglyceride ≥ 150 mg/dL; low HDL: high density lipid (HDL) cholesterol < 50 mg/dL; high blood pressure: blood pressure ≥ 130/85 mmHg or treatment for hypertension; hyperglycemia: fasting blood glucose ≥ 100 mg/dL or treatment for diabetes.