The mixture analyses were conducted in 33,979 adults (age ≥ 20 years) with blood levels of Cd, Pb and Hg in National Health and Nutrition Examination Survey (NHANES) 2003–2018 as well as in three subsets with additional data: (1) 10,257 adults with data on four per- and polyfluoroalkyl substances (PFAS); (2) 12,847 adults with data on urine levels of eight metals; and (3) 1043 adults with data on 34 PCBs and six polychlorinated dibenzo-p-dioxins (PCDD) and dibenzofuranes (PCDF). Analyses of peripheral blood mononuclear cell telomere length and blood levels of Pb and Cd were conducted on 7360 adults with data in NHANES 1999–2002 (Fig. 1). Analysis of de-identified NHANES data is exempt from IRB review (Johnson et al., 2013).

Liver injury was defined as ALT elevation (≥ 40 IU/L, men; ≥ 31 IU/L, women) (Ma et al., 2023). Advanced-LF was defined as ALT elevation and Fibrosis-4 (FIB-4) ≥ 2.67 and/or Forns Index ≥ 6.9 and/or Aspartate aminotransferase (AST) to Platelet Ratio Index (APRI) ≥ 1.5 (Hagström et al., 2020).

The age range was 20–85 years. Sex and race/ethnicity (non-Hispanic White (NHW), NHB, Mexican American (MA), and other race (O)) were self-reported. Smoking status, alcohol use, body mass index (BMI), poverty and diabetes were defined as before (Ma et al., 2023). Smoking categories were past/current and never. Alcohol use categories were lifetime abstainers, former drinkers, non-excessive current drinkers, and excessive current drinkers. BMI categories were normal weight (<25 kg/m²), overweight (25≤ BMI <30 kg/m²), and obese (≥30 kg/m²). Diabetes was self-reported, and/or hemoglobin A1c ≥ 6.5%, and/or fasting plasma glucose ≥ 126 mg/dL. Poverty was a family poverty-income-ratio less than 1 (Ma et al., 2023). Viral hepatitis was past/current infection with hepatitis B virus (HBV), positive core antibody or surface antigen; or infection with hepatitis C virus (HCV), positive RNA or antibody (Ma et al., 2023). Blood levels of Pb, Cd, Hg, PFAS, and lipid-adjusted plasma levels of PCBs/PCCD/Fs and urine levels of eight heavy metals were analyzed as continuous (log-transformed) and binary variables (quartiles (Q)1–3 versus Q4). Telomere length was determined by quantitative polymerase chain reaction (PCR) as described elsewhere (Rattan et al., 2022). Telomere length was categorized as quartiles.

Univariable and multivariable logistic (MVL) regression and WQS regression models were used to examine associations between exposures and outcomes. MVL models were used to assess the independent association of each compound with advanced-LF or ALT elevation, adjusting for covariates. The survey estimation commands for logistic regressions in SAS OnDemand for Academics (SAS Institute Inc., Cary, NC, USA) added sample weights, allowing the results to be extrapolated to the housed and non-institutionalized U.S. population. Age standardization estimates were obtained through the direct method and were standardized to the 2000 US census population. Differences between groups were assessed using t statistics.

The “gWQS” package in R (version 4.1.0) was used to evaluate the collective impact of toxic mixtures and to identify the primary drivers (Carrico et al., 2015). The dataset was randomly divided into training and validation sets (40% vs. 60%). In the training set, toxic component weights (wi) were derived through 100X bootstrapping; weights ranged from 0 to 1 and summed to 1. The WQS index was then calculated by averaging the weights across the bootstrap sets and multiplying them by the toxic component’s decile of concentration (qi):WQS index =∑i=1cwl¯qi. Associations between the indices and liver endpoints were then evaluated in the validation set using WQS regression models. Sensitivity analysis was conducted by using Least Absolute Shrinkage and Selection Operator (Lasso) regression, a machine learning technique that encourages variables selection in model by adding a penalty term (Tibshirani et al., 1996). For dioxin-like PCBs and PCDD/Fs, a sensitivity analysis was conducted using Toxic Equivalent Quantity (TEQ) estimates. TEQs were calculated using the 2005 World Health Organization’s toxic equivalency factors (TEFs) for dioxin-like compounds (Van den Berg et al., 2006). The TEQ for each specimen was determined by summing the products of the concentration of each congener and its corresponding TEF. Missing values of covariates (<10%) were addressed using multiple imputation by chained equations (Yuan, 2010). Statistical significance was a two-sided P value <0.05.

NHANES provides cross-sectional data about the residential population of the United States. In this study we analyzed data on NHANES participants with the following data: 1) blood levels of Pb, Cd and Hg (n = 33,979); 2) blood levels of Pb, Cd, Hg and four PFAS (n = 10,257); 3) urine levels of eight metals (Pb, Cd, barium, cesium, cobalt, antimony, thallium, tungsten) and blood levels of Pb, Cd, and Hg (n = 12,847); 4) blood levels of Pb, Cd, Hg, 34 PCBs and six PCDD/Fs (n = 1043); 5) blood levels of Pb and Cd and telomere length (n = 7360) (Fig. 1). The characteristics of the NHANES participants with data about the various toxins and telomere length are presented in Appendix A Tables S1a–S1e.

Among NHB, high (Q4 vs Q1–3) blood levels of Pb were associated with advanced-LF in an MVL model that adjusted for the survey cycle, age, sex, BMI, smoking status, alcohol use, diabetes, and poverty. The odds ratio (OR) for high (Q4 vs Q1–3) Pb was 2.26 (95%CI, 1.37–3.74) (Table 1). NHB persons were both more likely to have high blood levels of Pb (Fig. 2) and they had a higher risk of advanced-LF when exposed to high Pb than NHW or any other racial/ethnic group (Table 1). In an analysis that excluded participants with viral hepatitis exposure, high Pb remained associated with advanced-LF in NHB (OR = 2.39; 95%CI, 1.18–4.86) (Appendix A Table S2). When analyzed as a continuous variable, Pb was associated with advanced-LF in NHB in a dose-dependent manner (Appendix A Table S3), but Pb was not associated with advanced-LF in any other racial/ethnic group, whether when analyzed as a categorical or continuous variable (Table 1 and Appendix A Tables S2 and S3). In contrast to high Pb, high Cd was associated with advanced-LF in the total population and in all racial/ethnic groups, with ORs ranging from 1.73–2.19. Hg was not associated with advanced-LF in any group. Diabetes was associated with advanced-LF in the total population and in NHW, MA and O, but not in NHB (OR = 0.73; 95%CI, 0.46–1.17), underscoring the distinctiveness of the factors associated with advanced-LF in NHB (Table 1).

In an investigation of factors associated with ALT elevation, high Pb (Q4 vs Q1–3) had an OR of 1.24 (95% CI, 1.01–1.53) and high Cd had an OR of 1.25 (95% CI, 1.03–1.55) in NHB. Neither high Pb nor high Cd was significantly associated with ALT elevation in any other group. Q4 blood levels of Hg were associated with ALT elevation in the total population and in MA (Appendix A Table S4).

Spearman correlations between Pb and Cd ranged up to 0.4 (Appendix A Fig. S1); thus, WQS modeling was used to identify the true “bad actors” (Carrico et al., 2015). WQS_metal indices were determined for all group 1 participants (n = 33,979) and for each racial/ethnic group. The indices were significantly associated with advanced-LF in the total population and in NHW, NHB, and MA in models that adjusted for age, sex, smoking status, alcohol use, BMI, poverty, and diabetes. ORs ranged from 1.12 (95% CI, 1.01–1.24) per decile of the WQS_metal index in MA to 1.40 (95% CI, 1.21–1.64) in NHB (Fig. 3). Pb was the most significant contributor in NHB, while Cd was the most significant contributor in the total population and in the other racial/ethnic groups (Fig. 3). Among NHB, the OR of advanced-LF for participants in the top decile of the WQS_metal index was 20.7 (95% CI, 5.6–85.8) (Fig. 3). In other analyses, the WQS_metal indices were significantly associated with ALT elevation in the total population, NHW, NHB and MA. Pb was the major contributor in NHB and MA, while Hg and Pb were the top contributors in the total population (Appendix A Table S5).

Spearman correlations between blood levels of metals and PFAS are shown in Appendix A Fig. S2. In group 2 participants (n = 10,257), the WQS_metal index was significantly associated with advanced-LF (OR = 1.17 (95% CI, 1.03–1.32)) and ALT elevation (OR = 1.03 (95% CI, 1.00–1.07)), after adjusting with age, sex, smoking status, alcohol use, poverty and diabetes, consistent with results in group 1, while the WQS_PFAS index was not associated with either advanced-LF or ALT elevation (Appendix A Figs. S3–S4). The OR of the WQS_metal-PFAS index was similar to that of the WQS_metal index for the association with advanced-LF; Cd was the major contributor (Appendix A Fig. S4). Sample size limitations precluded stratification by race/ethnicity with outcome of advanced-LF. Neither the WQS_metal-PFAS index nor the WQS_PFAS index was associated with ALT elevation in most racial/ethnic groups (Appendix A Figs. S5 and S6, respectively).

Spearman correlations among heavy metals in blood and urine are shown in Appendix A Fig. S7. In group 3 participants (n = 12,847), the WQS indices of heavy metals in a) blood, b) urine, and c) blood and urine were all significantly associated with both advanced-LF and ALT elevation. The index of the combination of blood and urine metals had the highest adjusted ORs: 1.15 (95% CI: 1.03–1.29) for advanced-LF and 1.06 (95% CI: 1.02–1.10) for ALT elevation; blood and urine Cd were the major contributors (Appendix A Figs. S8 and S9). Sample size limitations precluded stratification by race/ethnicity with outcome of advanced-LF. The indices were not associated with ALT elevation in most individual racial/ethnic groups (Appendix A Figs. S10 and S11).

Before conducting a mixture analysis, 34 PCBs, six PCDD/Fs and three heavy metals (Pb, Cd, and Hg) were analyzed individually for associations with ALT elevation in the total group (group 4, n = 1043) with adjustment for age, sex, and race/ethnicity (Appendix A Table S6). PCB209 was excluded from the subsequent mixture analysis because its concentration is inversely related to ALT elevation. Spearman correlations among the remaining pollutants are shown in Appendix A Fig. S12. The WQS_{metal-PCB-PCDD/F} index was significantly associated with ALT elevation in the total population of group 4: OR = 1.50 per decile (95% CI, 1.26–1.78), reaching an OR of 38.4 (95% CI, 8.0–179.0) in the top decile, after adjusting for age, sex, and race/ethnicity. PCB126, Cd, Pb, PCB101, PCB172, Hg and 1,2,3,4,6,7,8-heptachlororodibenzo-p-dioxin (HpCDD) were the primary contributors. PCB126 had the highest weight by far (Fig. 4). Sample size limitations precluded analysis with outcome of advanced-LF.

In a comparison of ORs for the associations between ALT elevations and WQS indices in groups 2 and 4, the WQS_{metal-PCB-PCDD/F} index had the highest value, followed by the WQS_PCB-PCDD/F index. The ORs of the indices for mixtures that included PCBs/PCDD/Fs were higher than those that included PFAS (Fig. 4), after adjusting for age, sex, and race/ethnicity. These results underscore the importance of heavy metals and selected persistent organic pollutants as potential drivers of liver injury.

An additional WQS analysis of the metal-PCB-PCDD/F mixture was carried out because many pairs of PCBs have correlation coefficients exceeding 0.7 (Appendix A Fig. S12). In this case, components with weights over 0.005 may be significant contributors (Carrico et al., 2015). Several additional compounds are above this threshold and might be considered significant contributors (Fig. 4). Lasso regression modeling was used as an additional alternative approach for identifying chemicals positively associated with ALT elevation. In general agreement with the WQS_{metal-PCB-PCDD/F} index, PCB126, PCB172, HpCDD, Cd, and Hg were selected, in addition to PCB151, PCB105, and octachlorodibenzo-p-dioxin (OCDD) (Appendix A Table S7).

In the TEQ analysis of dioxin-like PCBs and PCDD/Fs, the association between ALT elevation and the TEQ values of PCB 105, 118, 126, HpCDD, and OCDD was highly significant (OR = 1.89 (95% CI, 1.09–3.26) per 10 pg/g TEQ) when the TEQ values were treated as a continuous variable, after adjusting for age, sex, and race/ethnicity. The association was also highly significant when the TEQ values were treated as a categorical variable. Compared to participants with TEQ values in Q1, those in Q3 and Q4 had ORs of 3.43 (95% CI, 1.80–6.56) and 3.21 (95% CI, 1.90–5.41), respectively, after adjusting for age, sex, and race/ethnicity (Appendix A Table S8). Notably, PCB 105, 118, 126, HpCDD, and OCDD were significantly associated with ALT elevation in both the single-chemical analysis (Appendix A Table S6) and in the WQS mixture analysis (Fig. 4), indicating that these compounds are major contributors to liver disease risk, aligning with our primary findings. An analysis that included all of the dioxin-like PCBs and PCDD/Fs in our dataset that have reported TEQ values did not show a significant association with ALT elevation (Appendix A Table S9).

NHB had higher blood levels of liver toxicants than NHW, as indicated by their higher age-standardized weighted mean blood levels of Pb, Cd, PCB126, PCB172, HpCDD, and OCDD (Fig. 2).

NHANES 1999–2002 included 7360 adults with measurements of both blood levels of Pb and Cd and blood cell telomere length (Fig. 1 and Appendix A Table S1e). In an MVL model, telomere length had an inverse relationship with advanced-LF. Compared to Q4, the adjusted ORs for Q3 to Q1 varied from 4.31 to 5.28, indicating a higher risk of advanced-LF with decreasing telomere length (Fig. 5a). Sample size limitations precluded stratification by race/ethnicity. Telomere length was not significantly associated with ALT elevation (Appendix A Fig. S13). Heavy metals were associated with short (Q1) telomere length in WQS models in the total population, OR=1.12 (95%CI 1.08, 1.17) and in NHW and NHB; Cd was the top contributor (Fig. 5b).