An exposome atlas of serum reveals the risk of chronic diseases in the Chinese population

Abstract

Although adverse environmental exposures are considered a major cause of chronic diseases, current studies provide limited information on real-world chemical exposures and related risks. For this study, we collected serum samples from 5696 healthy people and patients, including those with 12 chronic diseases, in China and completed serum biomonitoring including 267 chemicals via gas and liquid chromatography-tandem mass spectrometry. Seventy-four highly frequently detected exposures were used for exposure characterization and risk analysis. The results show that region is the most critical factor influencing human exposure levels, followed by age. Organochlorine pesticides and perfluoroalkyl substances are associated with multiple chronic diseases, and some of them exceed safe ranges. Multi-exposure models reveal significant risk effects of exposure on hyperlipidemia, metabolic syndrome and hyperuricemia. Overall, this study provides a comprehensive human serum exposome atlas and disease risk information, which can guide subsequent in-depth cause-and-effect studies between environmental exposures and human health.

Fig. 1. Overview of the study.

a Study design: Serum samples of 5696 participants were included in the study, 9 basic epidemiological information and 9 chronic disease-related clinical parameters were collected. Two platforms of LC-MS/MS and GC-MS/MS were used to detect 267 exposures, and 74 high-frequency exposures were determined. b Exposures characteristics: Exposures and basic epidemiological factors were associated, according to these factors, participants were stratified to analyze the distribution characteristics of exposures. c Risk discovery. First, single-exposure analysis showed the risk, stratified risk and health risk assessment of each exposure to chronic disease, and then exposure mixtures analysis showed risk effects of exposure mixtures on related chronic diseases based on 3 multi-exposure models. HbA1c glycated hemoglobin, BMI body mass index, LDL-C low density lipoprotein cholesterol, SBP systolic blood pressure, DBP diastolic blood pressure, OCP organochlorine pesticide, OPP organophosphorus pesticide, PAH polycyclic aromatic hydrocarbon, PCB polychlorinated biphenyl, PFAS perfluoroalkyl substance, WQS weighted quantile sum regression, q g-comp quantile g-computation, BKMR Bayesian kernel machine regression.

Fig. 2. Correlation analysis of the basic epidemiological factors and exposures.

a Explanation of exposure variations using epidemiological information based on variation partitioning analysis. b Correlations between each exposure and 9 basic epidemiological factors. The correlation coefficients obtained from the partial spearman correlation analysis were used to plot the heatmap. Principal component analysis score plot for 15 provinces (c) and different age ranges (d). e Correlation network of exposures and epidemiological factors. Red line represents positive correlation, blue line represents negative correlation obtained by Spearman correlation analysis. HCH hexachlorocyclohexane, DDD dichlorodiphenyldichloroethane, DDE dichlorodiphenyldichloroethylene, DDT dichlorodiphenyltrichloroethane, IBA indole-3-butyric acid, PFOA perfluorooctanoic acid, PFNA perfluorononanoic acid, PFDA perfluorodecanoic acid, PFUnDA perfluoroundecanoic acid, PFDoDA perfluorododecanoic acid, PFTrDA perfluorotridecanoic acid, PFOS perfluorooctanesulfonate, PFHxS perfluorohexanesulfonate, 6:2 Cl-PFAES 6:2 chlorinated polyfluoroalkyl ether sulfonate, 6:2 diPAP bis[2-(perfluorohexyl)ethyl] phosphate, PFPeA perfluoro-n-pentanoic acid, PFHpS perfluoroheptanesulfonic acid, MCHP monocyclohexyl phthalate, MEP monoethyl phthalate.

Fig. 3. Hierarchical analysis of exposure levels based on key epidemiological factors.

a The location of province included in this study and their total concentration of exposures. b Regional distribution of different categories of exposures depicted by stacked bar plot. Exposures that significantly increase (c, d) and decrease (e) with age. Exposures that significantly increase (f–h) with education. For figures (c–h), the concentrations scaled by the z-score method were used and error bars represent standard error of mean (n = 5696 biologically independent samples). Exposures that significantly change with gender (i) and drinking history in male (j). The geometric means of the exposures were used for figures (a, b, i, j). Gray dash lines of figures (i, j) represent fold change less than 0.8 and more than 1.3.

Fig. 4. The odds ratios of exposure to chronic diseases.

Exposures with a significant risk for hyperlipidemia (a), hyper low density lipoprotein cholesterol (b), hypercholesterolemia (c), hypertriglyceridemia (d), metabolic syndrome(e), obesity (f), diabetes (g), hyperuricemia (h), abdominal obesity (i), hypertension (j), hyper diastolic blood pressure (k) and hyper systolic blood pressure (l). The concentrations were log10 transformed so ORs represent odds ratios per one-unit increase in log-transformed exposure levels. Binary logistic regression models adjusted for age, gender, region, sampling time, education and income levels, marital status, smoking and drinking history. The position and color of diamond represent ORs and significant (two-sided, n = 5696 biologically independent samples), respectively. Error bars represent 95% confidence interval of ORs.

Fig. 5. The risk of single exposure to major chronic diseases.

a Relationship between each exposure and 9 clinical disease parameters. Included exposures were significantly associated with at least one disease outcome. Risk of exposures for hyperlipidemia and metabolic syndrome stratified by age (b) and gender (c). Scaled odds ratios were used in the heat maps of stratified risk, and * represent significant 0.01 < p < 0.05, ** represent significant 0.001 < p < 0.01, *** represent significant p < 0.001 (two-sided). Exposures with significant associations for the corresponding diseases were used for this plot based on multiple linear regression and binary logistic regression models. All of regression models adjusted for age, gender, region, sampling time, education and income levels, marital status, smoking and drinking history.

Fig. 6. The risk of exposure mixtures and dose-risk relationship between key exposures and related chronic disease.

For three chronic disease outcomes including hyperuricemia (a), hyperlipidemia (b), and metabolic syndrome (c), all exposed mixtures have a positive risk effect on them based on WQS, q g-comp, and BKMR Models. d–f Odds ratios (ORs) of the joint and each of the priority risk chemicals screened by three multi-exposure models (The chemicals given in figure were defined in at least two models). ORs of the jointed chemicals were obtained by WQS and q g-comp models, and OR of each chemical was obtained by binary logistic regression model. ORs represent odds ratios per one-unit increase in log-transformed exposure mixtures or single exposure levels. All of the three multi-exposure models adjusted for age, gender, region, smoking, and drinking history, and the binary logistic regression model adjusted for age, gender, region, sampling time, education and income levels, marital status, smoking, and drinking history. The position and color of diamond represent ORs and significant (two-sided), respectively. Gray diamonds represent no significance. Error bars represent 95% confidence interval. g Dose-risk relationship of exposures to hyperuricemia, specifically, seven key exposures screened by both single and mixed models. h Dose-risk relationship of exposures to hyperlipidemia, nine key exposures were included. i Dose-risk relationship of exposures to metabolic syndrome, three key exposures were included. The black solid line represents the OR, and gray, blue and dark green shadow represents the 95 % confidence interval of hyperuricemia (n = 927 biologically independent samples), hyperlipidemia (n = 2842) and metabolic syndrome (n = 1284) ORs, respectively.