Per- and Polyfluoroalkyl Substances and Blood Lipid Levels in Pre-Diabetic Adults—Longitudinal Analysis of the Diabetes Prevention Program Outcomes Study

Abstract

Exposure to per- and polyfluoroalkyl substances (PFASs) may interfere with lipid regulation. However, most previous studies were cross-sectional with the risk of reverse causation, suggesting a need for long-term prospective studies. We examined the relationship of baseline plasma PFAS concentrations with repeated measures of blood lipids. We included 888 prediabetic adults from the Diabetes Prevention Program (DPP) and DPP Outcomes Study, who had measurements of 6 plasma PFAS concentrations at baseline (1996–1999) and repeated measures of blood lipids over 15 years of follow-up, and were initially randomized to placebo or a lifestyle intervention. We used linear regression to examine cross-sectional associations of PFAS concentrations and lipid levels at baseline, and evaluated prospective risks of hypercholesterolemia and hypertriglyceridemia using Cox proportional hazard models, and tested for effect modification by study arm. Participants (65.9% female, 57.0% White, 65.9% aged 40–59 years) had comparable PFAS concentrations [e.g., median (IQR) perfluorooctanoic acid (PFOA) 4.9 ng/mL (3.2)] with the general U.S. population in 1999–2000. We observed higher total cholesterol at baseline per doubling of PFOA (β: 6.1 mg/dL, 95% CI: 3.1, 9.04), perfluorohexane sulfonic acid (PFHxS, β: 2.2 mg/dL, 95% CI: 0.2, 4.3), and perfluorononanoic acid (PFNA, β: 2.9 mg/dL, 95% CI: 0.7, 5.0). Prospectively, baseline concentrations of several PFASs, including PFOA, PFOS, PFHxS and PFNA, predicted higher risks of incident hypercholesterolemia and hypertriglyceridemia, but only in the placebo group and not the lifestyle intervention group. For example, participants in the placebo group with PFOA concentration > median (4.9 ng/mL) were almost twice as likely (HR: 1.90, 95% CI: 1.25, 2.88) to develop hypertriglyceridemia compared to those ≤ median. Findings suggest adverse effects of some PFASs on lipid profiles in prediabetic adults. However, the detrimental effect was attenuated with a lifestyle intervention.

1. Introduction

Environmental factors are important determinants of cardiovascular disease. While the influences of smoking, diet, and physical activity are well established, contributions from environmental chemicals are not as clear. Per- and polyfluoroalkyl substances (PFASs) are synthetic chemicals with waterproof and stain-resistant properties that have been widely used in household and industrial products such as food packaging, nonstick cookware, clothing, and carpets since the 1950s (Wang et al. 2014). PFASs are environmentally persistent and widely detected in the environment and human serum (>95% of the U.S. population) (Calafat et al. 2007; Kato et al. 2011b); some have low renal clearance and long elimination half-lives, 3–7 years depending on the compound (Olsen et al. 2007).

Epidemiological evidence has linked PFASs with less favorable lipid profiles in adults, including higher total cholesterol, higher low-density lipoprotein (LDL), and higher triglyceride levels (Chateau-Degat et al. 2010; Costa et al. 2009; Eriksen et al. 2013; Fisher et al. 2013; Fitz-Simon et al. 2013; Kirk et al. 2018; Nelson et al. 2010; Olsen and Zobel 2007; Sakr et al. 2007a; Skuladottir et al. 2015; Starling et al. 2014; Steenland et al. 2009). This association may be the result of PFAS-mediated changes in lipid transport genes (Fletcher et al. 2013) or androgen receptor antagonism (Monroe and Dobs 2013). However, most findings have been from cross-sectional studies, and the lack of long-term prospective evidence has made it difficult to conclude that suggestive findings did not result from reverse causality, or that results with nonsignificant effect estimates had sufficient statistical power to draw definitive inference due to samll sample size (Fu et al. 2014; Olsen et al. 2000; 2003; Olsen et al. 1999; Rotander et al. 2015). Paradoxically, some cross-sectional studies in adults (Liu et al. 2018b) and children (Mora et al. 2018) and evidence from animal studies (Butenhoff et al. 2012a; Lau et al. 2007) found higher plasma PFAS concentrations or higher estimated PFASs exposure to be associated with beneficial effects on lipid profiles.

In this study, we leveraged data and samples collected in the Diabetes Prevention Program (DPP) and Outcomes Study (DPPOS), a multi-center randomized controlled trial of a lifestyle intervention with subsequent follow-up, that recruited middle-aged adults who were at high risk of developing type 2 diabetes and hyperlipidemia (Diabetes Prevention Program Research Group 1999; Diabetes Prevention Program Research Group et al. 2009). The long-term follow-up, for a median of 15-years, allowed us to examine the prospective risk of having clinically abnormal lipid levels in relation to plasma PFAS concentrations. Further, the design of the trial, which included an intensive lifestyle intervention with coaching in diet modification, moderate physical activity, and behavioral changes (Diabetes Prevention Program Research Group 2002), created the opportunity for us to evaluate whether the lifestyle intervention modified this risk. We hypothesized that higher baseline plasma PFAS concentrations would be associated with higher total cholesterol and LDL at baseline and increased incidence of hyperlipidemia during follow-up. Moreover, we postulated that the lifestyle intervention would attenuate the adverse effects of PFASs on lipid regulation.

2. Methods

2.1. Study population

The Diabetes Prevention Program (DPP, ClinicalTrails.gov number, NCT00004992) was a randomized controlled trial that examined the effect of an intense lifestyle intervention program or oral diabetic drug (metformin) on prevention of type 2 diabetes, compared to a medication placebo group. DPP recruited participants who were overweight and had prediabetes between July 1996 and May 1999 from 27 medical centers across the United States (Diabetes Prevention Program Research Group 1999). Eligibility criteria for DPP were age at least 25 years, body mass index (BMI) of 24 kg/m² or greater (22 or higher for Asians), and glucose concentration of 95 to 125 mg/dL fasting or 140 to 199 mg/dL 2 hours after a 75-gram oral glucose load. Participants were randomized into three treatment groups: lifestyle (N=1,079), metformin (N=1,073), or placebo (N=1,082). The lifestyle intervention included intensive coaching in diet modification (reducing fat and calorie intakes), moderate physical activity (150 min per week) and behavioral changes (Diabetes Prevention Program Research Group 2002). Participants in the lifestyle arm were closely followed-up by case-mangers to achieve goals of losing 7% of initial body weight and maintaining the weight loss throughout DPP phase. After the active phase of the DPP trial (1996–2001), 88% of the surviving DPP participants were followed up in the DPP Outcomes Study (DPPOS, 2002–2014, ClinicalTrails.gov number, NCT00038727). Detailed descriptions of DPP and DPPOS have been published previously (Diabetes Prevention Program Research Group 1999; Diabetes Prevention Program Research Group et al. 2009).

The present study used de-identified data and biosamples provided by the DPP/DPPOS to the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) central repository. We included participants only from lifestyle and placebo arms but not the metformin arm for this study because we did not have a specific hypothesis about the medication’s effects on the relationship between PFASs and cardiometabolic outcomes. We excluded participants who did not have baseline assessment of variables of interests (see study flowchart in Supplementary Material, Figure S1). The median (interquartile range, IQR) total duration of follow-up was 15.0 (14.5–16.0) years and more than 95% of participants in this analysis had ≥4 years of follow-up. All DPP/DPPOS participants provided written informed consent. The present analysis was approved by the Institutional Review Board at Harvard Pilgrim Health Care. The involvement of the Centers for Disease Control and Prevention (CDC) laboratory did not constitute engagement in human subjects research.

2.2. Exposures: Plasma concentrations of per- and polyfluoroalkyl substances

This study used blood samples collected at baseline for PFAS measurements. Plasma aliquots were stored in the DPP biorepository before being transferred to the NIDDK central repository. We quantified PFAS plasma concentrations using online solid-phase extraction coupled to high-performance liquid chromatography-isotope dilution-tandem mass spectrometry as previously described in detail (Cardenas et al. 2017; Kato et al. 2011a). We included 6 highly detected PFAS in our study samples; they are perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), perfluorohexane sulfonic acid (PFHxS), N-ethyl-perfluorooctane sulfonamido acetic acid (EtFOSAA), N-methyl-perfluorooctane sulfonamido acetic acid (MeFOSAA), and perfluorononanoic acid (PFNA). Values of PFOS and PFOA were calculated by summing up linear and branched isomers (Centers for Disease Control and Prevention 2019); specifically, PFOS was the sum of linear perfluorooctanesulfonic acid PFOS (n-PFOS) and perfluoromethylheptane sulfonic acid isomers (Sm-PFOS) and PFOA was the sum of linear perfluorooctanoic acid (n-PFOA) and perfluoromethylheptanoic and perfluorodimethylhexanoic acids (Sb-PFOA). The limit of detection (LOD) for all analytes were 0.1 ng/mL. The proportion of analytes below LOD was low for all 6 PFAS, specifically 0% for PFOS and PFOA, 0.1% for PFHxS (N=1), 3.3% for EtFOSAA (N=27), 2.6% for MeFOSAA(N=25), and 6.8% for PFNA (N=56). Assay results below the LOD were replaced by the LOD/√2 (Hornung and Reed 1990).

2.3. Outcomes: Blood lipid measurements

Blood samples were collected at baseline, annual and semi-annual follow-up visits. The Central Biochemistry Laboratory (Northwest Lipid Research laboratories, University of Washington, Seattle, WA, USA), a CDC Reference Laboratory for lipids, performed all laboratory analytical measurements according to CDC Reference Methods (Diabetes Prevention Program Outcomes Study Research Group et al. 2013). Detailed descriptions of lipid measurements and calculations had been documented previously (Diabetes Prevention Program Outcomes Study Research Group et al. 2013). Briefly, fasting total plasma cholesterol and triglycerides were determined enzymatically; HDL cholesterol was analyzed using dextran sulfate-magnesium precipitating procedure, and LDL cholesterol was calculated using Friedewald equation (Friedewald et al. 1972) or directly measured if triglyceride level was ≥400 mg/dL. The analysis included concentrations of total cholesterol, triglycerides, LDL, HDL, very low-density lipoprotein (VLDL), and non-HDL cholesterol, which was calculated as the difference between total cholesterol and HDL.

We used repeated blood lipid measures to assess hyperlipidemia incidence, including hypercholesterolemia and hypertriglyceridemia, based on the National Cholesterol Education Program adult treatment panel (ATP) guideline III (Grundy et al. 1993). We defined hypercholesterolemia incidence when participants had the first record of (1) LDL ≥ 160 mg/dL, (2) LDL ≥ 130 mg/dL in those with diabetes, (3) total cholesterol ≥ 240 mg/mL or (4) initiation of lipid lowering medications; and hypertriglyceridemia incidence when participants had the first record of triglycerides ≥200 mg/dL. The use of lipid lowering therapy was assessed by research staff at the time of each blood collection using a standard questionnaire (binary, yes/no). When comparing with prescription records as a gold standard, the binary variable had 83.2% sensitivity and 98.4% specificity. We incorporated information from participants’ prescription records to evaluate the use of specific medications for hypercholesterolemia and hypertriglyceridemia outcomes respectively, see Supplementary Material, Table S1 for detail information.

2.4. Covariates

We included in our regression models a priori baseline covariates likely to be important predictors of lipid level and that may be correlated with plasma PFAS concentrations, including age, sex, race/ethnicity, educational attainment, marital status/living arrangement, smoking status, alcohol drinking, physical activity level, percent of daily calories from fat, daily fiber intake, and body size (waist circumference). Age was computed based on date of randomization and birth date and categorized into 5-year groups. Race/ethnicity categorization was based on the 1990 U.S. census questionnaire including non-Hispanic white, African American, Hispanic (of any race), and all others. Educational attainment was based on year of school completed and categorized into elementary/junior high, high school/GED, college, and graduate school. Marital status/living arrangement included married/cohabitating, single, divorced/separated, widowed. Smoking status included nonsmoker, former smoker, and current smoker. Physical activity level was assessed using the NHANES III Physical Activity Scale which captured self-reported counts of 30-minute moderate activities and exercises, including jogging, biking, swimming, aerobic exercise, dancing, yard work, weight lifting, and others (Beddhu et al. 2009). Participants who did not report engaging in any regular activities were classified as inactive and the rest of the participants were grouped into quartiles. Alcohol drinking, percent daily calories from fat, and daily fiber intake were extracted from a semi-quantitative food frequency questionnaire with 117 items that measured dietary habits over the previous year (Mayer-Davis et al. 2004). We used waist circumference (continuous) instead of BMI (which was only available as a categorical variable in the NIDDK repository database) in our analysis to better account for body size; other observational studies had also indicated waist-circumference to be a better measure of central obesity and are better predictors of risk for atherosclerotic cardiovascular disease and diabetes than weight or BMI (Lebovitz and Banerji 2005). Menopause status was extracted from self-reported data in the medical history questionnaire.

2.5. Statistical analysis

We compared participants included in the current analysis and those excluded using Student’s t-test for normally distributed variables, the Wilcoxon rank-sum test for skewed variables, and chi-square test for categorical variables. We tested for normality using the Shapiro–Wilk test; because all plasma PFAS concentrations were right-skewed, we used them as log₂ transformed or categorized them into quartiles.

We used multivariable linear regression, adjusting for potential confounders mentioned above, to examine cross-sectional associations between baseline plasma PFAS concentrations and baseline blood lipid levels, restricting to participants not taking lipid lowering medications. We used both continuous (log₂-transformed) and categorized (quartile) exposure variables for better comparison with previous findings. We presented regression parameters per doubling (log2 increase) of PFAS concentrations, adjusting for confounders, and calculated the predicated differences in lipid level for each quartile of plasma PFAS concentration, comparing to the first quartile, from regression parameters. Although triglycerides and HDL levels were not normally distributed, for easier clinical interpretation of lipid levels we did not log transform them in our models. We tested for effect modification by sex, age and baseline treatment assignment, and menopause status (among women in sex-stratified analyses), using the method of adding a multiplicative interaction term of PFAS and covariate in the models, and considered evidence for effect modification at P_interaction < 0.15. We performed sensitivity analyses replacing waist circumference with BMI (categories) and adding menopause status in the model. We identified outliers by visual inspection of scatterplots and also examined the robustness of associations by excluding outliers that had exposure and outcome values above the 95^th percentile or below the 5^th percentile, see supplementary materials Table S2 for the number of outliers removed in the sensitivity analysis. We used generalized additive models (GAM) including smoothing splines with 4 degrees of freedom on PFAS concentrations to evaluate nonlinearity and dose-response.

Given that a substantial proportion of participants initiated lipid lowering medications during study follow-up (7.1% in the placebo and 3.8% in the lifestyle arms at one year post randomization, and 32.1% in the placebo and 28.0% in the lifestyle arms at Year 5 of DPPOS), we did not model associations with blood lipid measurements as a continuous variable at follow-up visits to avoid risk of healthy participant bias (Ward 2012; Weisskopf et al. 2015). Instead, we assessed the prospective risk of hyperlipidemia, using the hypercholesterolemia and hypertriglyceridemia criteria defined in the previous section. We first used multivariable logistic regression to assess odds ratio (OR) of hyperlipidemia at baseline relative to PFAS levels. Then among participants who did not have hyperlipidemia at baseline, we used Cox-proportional hazard models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of developing hyperlipidemia relative to baseline plasma PFAS concentrations. We evaluated the proportional hazard assumption by examining Martingale-based residuals (Lin et al. 1993). For PFASs that were significantly associated with increased risk of hyperlipidemia during follow-up, we used the Kaplan-Meier method to plot the unadjusted cumulative incidence over time, stratified by baseline PFAS concentration above vs. below or equal to the median for easier interpretation of the finding. The longitudinal analyses accounted for the same set of baseline covariates as the cross-sectional analyses and did not consider time-varying variables (such as repeated waist circumference) to avoid bias from adjusting a causal intermediate. We examined effect modification by sex and treatment group assignment using the interaction term approach with significant level at P_interaction < 0.15 and performed additional analyses accounting for menopause status and use of hypertriglyceridemia medications (for the hypertriglyceridemia outcome, list of medications presented in Supplementary Material, Table S1). We also performed sensitivity analyses not conditioning on participants’ diabetic status in the definition for hyperlipidemia (i.e., using the LDL ≥ 160 mg/dL cut-off for everyone). We used SAS 9.4 for all analyses.

3. Results

3.1. Population demographics

The cross-sectional analysis on associations of plasma PFAS concentrations with blood lipids included 888 individuals initially randomized into the lifestyle (N=451) or placebo (N=437) arms of DPP who were not taking lipid lowering medications at baseline. Their demographics were comparable across the two arms and also to the original DPP cohort, with the majority of participants being female (65.9%), white (57.0%), nonsmoker (56.5%), non-alcohol drinker (54.6%), college educated (49.2%), and married or cohabitating (67.5%) (Table 1). At baseline, mean (standard deviation, SD) waist circumference was 105.0 (14.4) cm, daily total calories from fat was 34.1 (6.8) %, and daily fiber intake was 16.0 (8.0) gram (Table 1). The incidence of diabetes mellitus was 9.8% for the placebo group and 4.7% for the lifestyle group 2 years post study initiation, and 57.2% and 48.8% at Year 11 of DPPOS.

Table 1.

Baseline characteristics of participants who were not on lipid-lowering medication at baseline (N=888)

Variable	N (%) or Mean (SD)
Total	888
Sex
Male	303 (34.1)
Female	585 (65.9)
Race/ethnicity
Non-Hispanic White	506 (57.0)
African American	174 (l9.6)
Hispanic of any race	170 (l9.l)
All others	38 (4.3)
Age (years)
<40	108 (12.2)
40–44	101 (11.7)
45–49	204 (23.0)
50–54	155 (17.5)
55–59	125 (14.1)
60–64	94 (10.6)
>65	98 (11.0)
Educational attainment
< High school	43 (4.8)
High school/GED	188 (21.2)
College	437 (49.2)
Graduate school	220 (24.8)
Marital Status
Married/Cohabitating	599 (67.5)
Single	106 (11.9)
Divorced	142 (16.0)
Widowed	41 (4.6)
Alcohol (g/day)
0	485 (54.6)
<14	375 (42.2)
>14	28 (3.2)
Smoking
Nonsmoker	502 (56.5)
Current smoker	54 (6.1)
Former smoker	332 (38.4)
Number of moderate activities (per week)a
0	115 (13.0)
0 to 1	94 (10.6)
>1 to 3	254 (28.6)
>3 to 7	220 (24.8)
>7	205 (23.1)
Group assignment
Lifestyle	451 (50.8)
Placebo	437 (49.2)
Menopause status (among N=585 female participants)
Pre-menopausal	313 (53.5)
Menopausal	272 (46.5)
Percent calorie intake from fat (%)	11.7 (3.04)
Daily fiber intake (g/day)	16.0 (8.0)
Waist circumference (cm)	105.0 (14.4)

Note: PFOS: Perfluorooctanesulfonic acid (sum of linear and branched isomers); PFOA: perfluorooctanoic acid (sum of linear and branched isomers); GM: geometric mean; SD: standard deviation; IQR: interquartile range. P-values from comparing PFAS concentrations across level of covariates; PFOS and PFOA were not significantly different across levels of percent calorie intake from fat, daily fiber intake nor waist circumference (P>0.05).

^a30 minutes of moderate activities and exercises

Baseline plasma PFAS concentrations did not differ between arms and were comparable to the concentrations detected among U.S. adults in NHANES 1999–2000 (Centers for Disease Control and Prevention 2019). PFOS and PFOA were the most highly detected both in our study and in the US population. The median and IQR were 27.2 (18.0–40.4) ng/mL (PFOS), 4.9 (3.5–6.7) ng/mL (PFOA), 2.3 (1.4–3.8) ng/mL (PFHxS), 1.1 (0.6–2.1) ng/mL (EtFOSAA), 1.0 (0.6–1.7) ng/mL (MeFOSAA), and 0.6 (0.4–0.8) ng/mL (PFNA). Concentrations of most PFASs were correlated, with strongest correlation observed between PFOS and EtFOSAA (r=0.60), followed by PFOA and PFOS (r=0.59), see Supplementary Material, Figure S2 for details. There was no significant difference in baseline plasma PFAS concentration by treatment assignment, but we did observe variations in plasma PFAS concentrations by participant characteristics, which suggested differences in exposure sources and patterns as well as variability in PFAS excretion (see Supplementary Material Table S3 for concentrations stratified by participants’ characteristics). Briefly, plasma PFOS concentrations were higher among males [median (IQR): 29.0 (19.3–41.5) ng/mL] and African Americans males [median (IQR): 32.3 (19.2–47.7) ng/mL], and plasma PFOA concentrations differed significantly across age groups [highest among 55–59 years old, median (IQR): 5.5 (4.0–7.0) ng/mL] and was higher among non-Hispanic Whites [median (IQR): 5.3 (3.9–7.1) ng/mL]. We also observed differences in concentrations across educational attainment for all PFAS, where participants with high school/GED educational attainment having the highest concentrations (Supplementary Material Table S3).

Not surprisingly given that they were at high risk for diabetes, the study cohort had somewhat high levels of lipids compared to the clinical guideline (Grundy et al. 1993): mean (SD) levels in mg/dL were 204 (35.4) for total cholesterol, 166.4 (98.8) for triglycerides, 45.4 (12.1) for HDL, and 125.3 (32.0) for LDL. Lipid levels did not differ between arms at baseline, reflecting appropriate randomization. Blood lipid levels of participants in the lifestyle arm significantly improved compared to placebo one year after the randomization, but later became comparable during DPPOS follow-up (Supplementary Material, Figure S3).

3.2. Cross-sectional associations of PFAS with lipid levels

At baseline, after adjusting for confounders, each doubling of plasma PFOA concentration was associated with 6.1 mg/dL (95% CI: 3.1, 9.0) higher total cholesterol, 17.8 mg/dL (95% CI: 9.8, 25.7) higher triglycerides, 2.9 mg/dL (95% CI: 0.2, 5.6) higher LDL, and 3.7 mg/dL (95% CI: 2.2, 5.2) higher VLDL (Figure 1 and Supplementary Material Table S3). Participants in the highest quartile of plasma PFOA concentration (median: 8.4 ng/mL, IQR: 7.4, 10.3) had 13.4 mg/dL (95% CI: 6.6, 20.1) higher total cholesterol, 36.8 mg/dL (95% CI: 18.5, 55.1) higher triglycerides, 6.7 mg/dL (95% CI: 0.6, 12.9) higher LDL and 7.5 mg/dL (95% CI: 4.1, 10.9) higher VLDL compared to the lowest quartile (median: 2.6 ng/mL, IQR: 2.0, 3.0), accounting for baseline characteristics (Supplementary Material, Table S4). We found similar associations between other PFASs and adverse lipid profile at baseline. Higher PFOS plasma concentrations were associated with greater triglycerides and VLDL, and higher PFNA plasma concentrations were associated with greater total cholesterol and LDL (Figure 1 and Supplementary Material Table S3). Spline analyses using GAMs with log₂-transformed PFAS plasma concentration and untransformed lipid levels showed adding non-linear spline term did not significantly improve the fit of the models (Supplementary Material, Table S6). We did not observe evidence for effect modification by treatment group assignment, age, sex, or menopause status. Visual inspection of scatterplot did not identify influential outliers; magnitudes and directionality of associations did not change substantially after we excluded outliers with exposure and outcome values below the 5^th or above 95^th percentiles (see Supplementary Material, Table S2 for number of outliers removed in each model). We presented the scatterplots between baseline plasma PFAS and lipid level in the Supplementary Material Figure S4.

Figure 1.

Adjusted^a mean difference and 95% confidence interval in baseline total cholesterol (A), triglycerides (B), HDL (C), Non-HDL (D), LDL (E), and VLDL (F) per doubling of baseline plasma PFAS concentrations among 888 participants who were not on lipid-lowering medication in the Diabetes Prevention Program.

Note: PFAS: per- and polyfluoroalkyl substances; PFOS: Perfluorooctanesulfonic acid (sum of linear and branched isomers); PFOA: perfluorooctanoic acid (sum of linear and branched isomers); PFHxS: perfluorohexane sulfonic acid; EtFOSAA: N-ethyl-perfluorooctane sulfonamido acetic acid; MeFOSAA: N-methyl-perfluorooctane sulfonamido acetic acid; PFNA:perfluorononanoic acid; HDL: high-density lipoprotein; LDL: low-density lipoprotein; VLDL: very-low-density lipoprotein; units for effect estimates were mg/dL increase per doubling of plasma PFAS concentration (ng/mL).

^aAdjusted for age, sex, race and ethnicity, marital status, educational attainment, drinking, smoking, percent of daily calorie from fat intake, daily fiber intake, physical activity level, and waist circumference at baseline.

3.3. Risk of hyperlipidemia relative to baseline plasma PFAS concentrations

We observed increased cross-sectional and prospective risks of hypercholesterolemia and hypertriglyceridemia with some baseline plasma PFAS concentrations. At baseline, odds of hypercholesterolemia was 1.28 (95% CI: 1.05, 1.57) times higher for each doubling of baseline PFOA plasma concentration, controlling for confounders (Table 3). Plasma concentrations of other PFASs were not significantly associated with baseline prevalence of hypercholesteremia. In the prospective follow-up, excluding those with hypercholesterolemia at baseline, cumulative incidence of hypercholesteremia differed across baseline PFOA concentration and treatment group assignment (P_interaction < 0.15, see Table 3 and Figure 2). In the placebo group, participants who had PFOA levels above the median (4.9 ng/mL) had 42% higher risk of developing hypercholesterolemia (HR: 1.42, 95% CI: 1.06, 1.90) while no statistically significant elevated risk was observed in the lifestyle group over the study follow-up period (Supplementary Material, Table S7). We also found a higher incidence of hypercholesteremia with higher baseline plasma concentrations of PFNA (HR: 1.17, 95% CI: 1.04, 1.33) in the placebo but not in the lifestyle intervention group (P_interaction < 0.15, see Table 3).

Table 3.

Risks for hypercholesterolemia per doubling of baseline plasma PFAS concentrations (ng/mL)^a

	Cross-sectional association for baseline prevalent case, OR (95% CI)	Prospective risk for hypercholesterolemia incidence at follow-up, HR (95% CI)
	Total (N=940)	Total (N=730)	Placebo (N=361)	Lifestyle (N=369)	Interaction by group assignment (Pinteraction)b
PFOS	1.02 (0.85, 1.21)	1.01 (0.91, 1.12)	1.11 (0.95, 1.30)	0.97 (0.82, 1.14)	0.24
PFOA	1.29 (1.05, 1.57)*	1.06 (0.94, 1.19)	1.18 (0.98, 1.42)	1.03 (0.86, 1.23)	0.12
PFHxS	1.08 (0.94, 1.25)	1.00 (0.92, 1.09)	1.02 (0.89, 1.17)	1.02 (0.90, 1.15)	0.45
EtFOSAA	0.98 (0.87, 1.09)	1.00 (0.93, 1.07)	1.05 (0.96, 1.16)	0.98 (0.88, 1.08)	0.43
MeFOSAA	1.02 (0.89, 1.17)	1.04 (0.95, 1.12)	1.10 (0.97, 1.24)	1.01 (0.89, 1.15)	0.65
PFNA	1.11 (0.96, 1.29)	1.09 (1.00, 1.19)	1.17 (1.04, 1.33)*	1.03 (0.91, 1.18)	0.08

Note: PFAS: per- and polyfluoroalkyl substances; PFOS: Perfluorooctanesulfonic acid (sum of linear and branched isomers); PFOA: perfluorooctanoic acid (sum of linear and branched isomers); PFHxS: perfluorohexane sulfonic acid; EtFOSAA: N-ethyl-perfluorooctane sulfonamido acetic acid; MeFOSAA: N-methyl-perfluorooctane sulfonamido acetic acid; PFNA:perfluorononanoic acid; OR: odds ratio; HR: hazard ratio.

^aHypercholesterolemia defined as (1) LDL > 160 mg/dL, (2) LDL > 130 mg/dL in those with diabetes, (3) total cholesterol > 240 mg/mL or (4) initiation of lipid lowering medication; and hypertriglyceridemia defined as triglycerides >200 mg/dL. Adjusted risk calculated using logistic regression or Cox-proportional hazard model accounting for age, sex, race, education attainment, marital status, drinking level, smoking status, physical activity level, percent of daily calorie from fat intake, daily fiber intake, and waist circumference (all variables measured at baseline)

^bP-value from a fully adjusted multivariate model with a multiplicative interaction term between baseline plasma PFAS concentration and treatment assignment

^*p<0.05

Figure 2.

Kaplan-Meier curve of cumulative incidence of hypertriglyceridemia (A, B) and hypercholesterolemia (C, D) stratified by treatment assignment among those with baseline plasma perfluorooctanoic acid (PFOA) concentration >median (4.9 ng/mL) and ≤median in the Diabetes Prevention Program.

Note: PFOA: perfluorooctanoic acid (sum of linear and branched isomers); p-value calculated by log-rank test.

Hypercholesterolemia defined as (1) LDL ≥ 160 mg/dL, (2) LDL ≥ 130 mg/dL in those with diabetes, (3) total cholesterol ≥ 240 mg/mL or (4) initiation of lipid lowering medication; and hypertriglyceridemia defined as triglycerides ≥200 mg/dL.

(A) Hypertriglyceridemia in Placebo group

(B) Hypertriglyceridemia in Lifestyle group

(C) Hypercholesterolemia in Placebo group

(D) Hypercholesterolemia in Lifestyle group

Elevated risk of hypertriglyceridemia was found in relation to baseline plasma concentrations of PFOS and PFOA both at baseline and prospectively (Table 4). For example, for each doubling of baseline PFOA concentration, odds of hypertriglyceridemia at baseline was 1.47 (95% CI: 1.21, 1.80) times higher and the prospective risk was 1.23 (95% CI: 1.04, 1.45) times higher, with even higher prospective risk for those in the placebo group (HR: 1.45, 95% CI: 1.16, 1.85). Participants in the placebo group with baseline PFOA concentration above the median had approximately two-fold (HR: 1.90, 95% CI: 1.25, 2.88) greater risk of developing hypertriglyceridemia at any time point during follow-up compared to those with PFOA at or below median levels (Supplementary Material, Table S8). This difference in risk relative to baseline PFOA concentration was not observed among those who received the lifestyle intervention (Figure 2). Similar findings on elevated prospective risk were observed in relation to baseline plasma PFHxS and PFNA concentrations (Table 4). We did not find evidence of effect modification by sex or age for any associations. Additional adjustments for baseline menopause status among females and the use of triglycerides medications did not change the result (data not shown).

Table 4.

Risks for hypertriglyceridemia per doubling of baseline plasma PFAS concentrations (ng/mL)^a

	Cross-sectional association for baseline prevalent case, OR (95% CI)	Prospective risk for hypertriglyceridemia incidence at follow-up, HR (95% CI)
	Total (N=940)	Total (N=679)	Placebo (N=330)	Lifestyle (N=349)	Interaction by group assignment (Pinteraction)b
PFOS	1.23 (1.03, 1.46)*	1.09 (0.93, 1.27)	1.27 (1.03, 1.56)*	0.94 (0.73, 1.20)	0.12
PFOA	1.48 (1.21, 1.81)**	1.23 (1.04, 1.45)*	1.46 (1.15, 1.85)**	1.12 (0.85, 1.47)	0.24
PFHxS	1.03 (0.90, 1.18)	1.14 (1.00, 1.28)*	1.23 (1.03, 1.47)*	1.19 (0.98, 1.44)	0.45
EtFOSAA	1.08 (0.96, 1.20)	1.03 (0.93, 1.14)	1.08 (0.95, 1.24)	0.97 (0.83, 1.13)	0.32
MeFOSAA	1.12 (0.98, 1.28)	0.97 (0.86, 1.09)	1.04 (0.87, 1.23)	0.93 (0.78, 1.11)	0.30
PFNA	1.09 (0.94, 1.26)	1.17 (1.02, 1.33)*	1.29 (1.07, 1.56)**	1.13 (0.93, 1.37)	0.35

Note: PFAS: per- and polyfluoroalkyl substances; PFOS: Perfluorooctanesulfonic acid (sum of linear and branched isomers); PFOA: perfluorooctanoic acid (sum of linear and branched isomers); PFHxS: perfluorohexane sulfonic acid; EtFOSAA: N-ethyl-perfluorooctane sulfonamido acetic acid; MeFOSAA: N-methyl-perfluorooctane sulfonamido acetic acid; PFNA:perfluorononanoic acid; OR: odds ratio; HR: hazard ratio.

^aHypertriglyceridemia defined as triglycerides >200 mg/dL. Adjusted risk calculated using logistic regression or Cox-proportional hazard model accounting for age, sex, race, educational attainment, marital status, drinking level, smoking status, physical activity level, percent of daily calorie from fat intake, daily fiber intake, and waist circumference (all variables measured at baseline)

^bP-value from a fully adjusted multivariate model with a multiplicative interaction term between baseline plasma PFAS concentration and treatment assignment

^*p<0.05

^**p<0.01

4. Discussion

In this study of adults at high risk for diabetes with plasma PFAS concentrations comparable to the U.S. general population, higher plasma PFOA concentration was cross-sectionally associated with higher levels of total cholesterol, triglycerides, LDL and VLDL. Leveraging data over approximately 15 years of follow-up, we also observed that greater baseline plasma PFOA concentration was prospectively associated with greater incidences of hypercholesterolemia and hypertriglyceridemia in the placebo group but not in lifestyle intervention group. We also found other PFASs including PFOS, PFHxS, and PFNA to be associated with an adverse lipid profile, both cross-sectionally and prospectively, with stronger associations in the placebo group.

Our findings are consistent with the previous literature of mostly cross-sectional reports, providing additional evidence for the link between PFASs and non-favorable lipid profiles. Many studies have suggested PFAS exposure may lead to elevated total cholesterol. For example, evidence of positive associations of PFOA, PFOS, and PFHxS with total cholesterol have been reported in occupational cohorts (Costa et al. 2009; Sakr et al. 2007a) and in general populations (Fisher et al. 2013; Fu et al. 2014; Nelson et al. 2010), among healthy adults (Eriksen et al. 2013; Steenland et al. 2009; Winquist and Steenland 2014), pregnant women (Skuladottir et al. 2015; Starling et al. 2014), children and adolescents (Frisbee et al. 2010; Fu et al. 2014; Geiger et al. 2014; Nelson et al. 2010; Zeng et al. 2015), and uremic patients under hemodialysis (Liu et al. 2018c). Although in our study the absolute magnitude of elevation in total cholesterol associated with a doubling in plasma PFOA or PFOS concentration was relatively small, on a population level this effect may have important implications for shifting the mean/median as well as increasing the proportion of the population in the tail of distribution with higher total cholesterol.

The magnitude of effect we observed for plasma PFOA and total cholesterol was comparable to previous studies. A cross-sectional study conducted on a middle-age Danish population (N=753) (Eriksen et al. 2013) whose plasma PFAS concentrations (median PFOA: 5.5 ng/mL) were similar to those in our study showed 4.4 mg/dL (95% CI: 1.1, 7.8) higher total cholesterol per IQR of PFOA, while we found an adjusted mean difference in total cholesterol of 10.1 mg/dL (95% CI: 3.6, 16.7) and 13.4 mg/dL (95% CI: 6.6, 20.1) comparing the third (median: 5.6 ng/mL) and the fourth quartile (median: 8.4 ng/mL) with the first quartile (median: 2.6 ng/mL) of plasma PFOA concentration. A community survey among 46,294 residents who had high PFOA exposure from contaminated drinking water [mean (SD) serum PFOA concentration: 80.3 (236.1); median 27 ng/mL] also showed a positive cross-sectional association between PFOA and total cholesterol; a predicted increase in 10 mg/dL was observed when PFOA concentration increased from <10 ng/mL to 50 ng/mL, and the effect plateaued for PFOA concentrations above 50 ng/mL (Steenland et al. 2009). A review by Steenland and colleagues of 10 published studies (2000–2010) among mostly workers and highly exposed cohorts found a large variation in effect size; ranging from 0.001 to 2.0 mg/dL in total cholesterol per ng/mL increase in blood PFOA concentration, assuming linearity (Steenland et al. 2010).

The positive associations between PFOA and LDL were also observed in several studies among highly exposed populations. In an occupational cohort with 1,025 active workers exposed to ammonium perfluorooctanoate (APFO), serum PFOA was cross-sectionally (Sakr et al. 2007a), but not prospectively (Sakr et al. 2007b), associated with higher LDL. However, the longitudinal follow-up was among 454 workers who had longer work histories, thus the finding might be biased by healthy worker selection. In the C8 Health project, a population study of community residents exposed to PFOA-contaminated drinking water, serum PFOA concentration had cross-sectional linear association with higher LDL. Among children and adolescents in these communities (N=12,476), the adjusted mean level of LDL was 4 to10 mg/dL higher comparing the fifth (>60ngm/mL) to the first quintile (<15 ng/mL) of serum PFOA concentration, and the OR of abnormal LDL was 1.4 (95% CI: 1.2, 1.7) (Frisbee et al. 2010). Among adults (N=46,494), those with serum PFOA >50ng/mL had about 6–8 mg/dL higher LDL comparing those with serum PFOA <10ng/mL (Steenland et al. 2009). Longitudinal follow-up of a subset of 560 adults for 4.4 years showed a 50% reduction in serum PFOA to be associated with 3.6% decline in LDL (95% CI: 1.5, 5.7) (Fitz-Simon et al. 2013).

Associations of PFASs with triglycerides and other lipoproteins had generally been inconsistent, and more commonly showed null (Fisher et al. 2013; Fitz-Simon et al. 2013; Fu et al. 2014; Nelson et al. 2010; Starling et al. 2014) rather than positive results (Geiger et al. 2014; Steenland et al. 2009; Zeng et al. 2015) results, and even occasionally showed paradoxically negative (beneficial) associations (Donat-Vargas et al. 2019; Liu et al. 2018b). For example, a cross-sectional analysis among 1871 NHANES 2013–2014 adults (Liu et al. 2018b) showed higher PFOA concentration to be associated with lower likelihood of HDL < 40 mg/dL and a lower likelihood of hypertriglyceridemia. A Swedish study using repeated blood PFAS and lipid measurements from 187 healthy middle-aged subjects from a population-based nested case-control also reported inverse associations between PFAS and triglyceride levels in blood (Donat-Vargas et al. 2019). This finding could be due to the fact that PFASs activate peroxisome proliferator-activated receptor alpha (PPARα), and may work like fibrates to raise HDL cholesterol and lower triglycerides (Feingold and Grunfeld 2000). Inconsistency in results may have also resulted from differences in age, sex distribution and sample size of the study population, as well as study settings and PFAS concentrations. They may also suggest variations of effect across participants’ health status.

The mechanisms by which PFASs may interfere with blood lipids in humans are not well known. Molecular target screening (Scharmach et al. 2012) in human cells as well as metabolomics and transcriptomics analyses (Buhrke et al. 2015; Peng et al. 2013) have shown that PFOA may inhibit the nuclear receptor of hepatocyte nuclear factor 4α (HNF4α), a key regulator of lipid metabolism (Hayhurst et al. 2001) and thereby lead to a detrimental lipid profile. Notably, PFASs have structural similarity with fatty acids and are a ligand for PPARα, which would be expected to improve lipid profile, as observed in some studies of humans and animal studies (Kennedy et al. 2004). In general, PPARα-dependent effects are less pronounced in humans than in rodents (Bjork and Wallace 2009), and species-specific differences in PPARα receptor expression and activity (Palmer et al. 1998) may explain the differences in the association between PFASs and lipid profile between human and animal studies. The dramatic differences in the exposure dose between rodent studies and human studies and the potential for non-linear dose-response curves could also lead to the differences in findings.

The European Food Safety Authority (EFSA) recently issued provisional tolerable weekly intakes (TWIs) for PFOA and PFOS largely based on epidemiologic associations of these compounds with increased total cholesterol (European Food Safety Authority 2018b). During the review process, an alternative hypothesis was raised that the observed associations might be partly an artifact due a common mechanism of enterohepatic circulation (EHC) of PFAS and bile salts (European Food Safety Authority 2018a). Bile re-uptake from the gut creates a negative feedback loop, leading to reduction of de novo synthesis of bile acids from cholesterol in the liver. Several transporters required in the EHC for the uptake of bile salts into enterocytes and hepatocytes are also involved in the transport of PFASs (Zhao et al. 2015; Zhao et al. 2017). Variations in the gene expression of the transporters required for EHC (Xiang et al. 2009) can lead to variations in intestinal reabsorption of bile salts and thus total cholesterol level, and theoretically also affect PFAS concentrations in blood. However, this hypothesis has not been validated by empirical data. Longitudinal findings showing reductions in serum PFOA/PFOS correlated with reduction in total cholesterol (Fitz-Simon et al. 2013) also argued against this mechanism. The use of the lipid lowering medication, cholestyramine, which is a strong ion exchange resin that blocks re-absorption of bile acids, had been shown to increase the fecal excretion of PFAS human (Genuis et al. 2010; Genuis et al. 2013). Results from 8 adult subjects administered with cholestyramine showed greater ratio of elimination for PFHxS compared to PFOS and PFOA. In our analysis, we excluded participants on lipid lowering medications and found the strongest association between PFOA and total cholesterol, thus the abovementioned confounding by physiology was unlikely to drive our results.

We recognize the potential for reverse causation when interpreting cross-sectional findings, which lack information on the temporal order. In vitro studies have demonstrated high binding of PFOS and PFOA to albumin (99.7% to >99.9%) and β-lipoproteins (96.6% to PFOS and 39% for PFOA) (Olsen et al. 2007). Nevertheless, if positive associations were driven by this reverse causation mechanism, we should have also observed strong positive association between lipid level and PFOS, which has even stronger affinity to albumin and β-lipoproteins. Plasma density gradient fraction analysis further refuted this hypothesis by showing that only 9% of PFOS and <1% of PFOA were in the lipoprotein containing fractions, while the majority of PFASs were recovered in the lipoprotein-depleted plasma (Butenhoff et al. 2012b). The positive associations from our longitudinal analysis also argued against reverse causality.

Another important finding from our study was the effect modification by lifestyle intervention. While all DPP participants received information on healthy eating, healthy weight, active lifestyle, smoke cessation and alcohol intake (20–30 minutes overview with research staff) after randomization, those assigned to the intensive lifestyle arm had individual lifestyle coaches and went through a combination of individual and group training on diet, exercise and behavior modification skills. The DPP lifestyle protocol had a clearly defined weight loss and physical activity goals of 7% of weight reduction and at least 150 minutes of moderate intensity exercise every week (Diabetes Prevention Program Research Group 2002; Kramer et al. 2009). The initial intensive lifestyle intervention during the DPP phase attenuated the adverse effects of PFASs on lipid profile and the risk of hyperlipidemia, and the effect persisted in the 15 years of follow-up. In the DPP phase, participants who received intensive lifestyle intervention had a mean weight loss of 7 kg one year after initiation. While weight loss in the intervention group was not fully maintained, there was 58% reduction in diabetes incidence in the intervention group compared to the placebo group at the end of DPP intervention trial with about 4 kg difference between the two groups (Diabetes Prevention Program Research Group et al. 2009). We previously reported that participants in DPP/DPPOS with higher plasma PFAS concentrations, specifically PFOS, MeFOSAA, and PFNA, had greater development of adiposity over time in the placebo but not the lifestyle intervention group (Cardenas et al. 2018), so it is possible that the association between PFASs and adverse lipid profile that we observed was at least in part mediated by weight change. A diet-induced weight-loss trial (Liu et al. 2018a) recently showed that higher baseline PFAS concentration was associated with greater weight regain. Future analyses utilizing methods of statistical mediation or structural equation modeling could help elucidate this relationship.

Our study had several strengths, the strongest being the prospective study design which allowed us to include repeated measures of blood lipid levels over time and confirm the chronological sequence of exposure and outcome relationships. This is of key importance for PFAS research given the long half-lives. The cross-sectional analysis allowed for comparison with prior literature. The randomized trial design allowed us to examine the role of the lifestyle intervention. Our study population was diverse with over 40% of the study population from racial/ethnic minorities. We were able to control for important confounding factors such as fat and fiber intake, medication use, physical activity level, and waist circumference. However, we recognize the potential for residual confounding. For example, drinking water could be an important source of PFASs exposure for residents living in areas of PFASs contaminations and reduced water consumption could increase lipid concentrations (Campbell et al. 1994). Confounding by water intake could bias our results in the negative direction; nevertheless, since we did not observe many extreme high levels of plasma PFAS concentrations, the extent of this bias should be minimal. Our study also had other limitations. Plasma PFAS concentrations are considered sensitive and accurate biomarkers to internal exposure to these substances (Butenhoff et al. 2006). However, we only used a single measurement of plasma PFAS concentrations in this analysis, which may misclassify internal PFASs exposure at later time points. We found a strong correlation between PFAS concentrations measured in blood collected at baseline and at Year 2 of DPP (Cardenas et al. 2017), and with the relative long half-lives of certain PFASs, one measurement likely represents a stable body burden. We could not ascertain the exact date and clinical rationale for the initiation of lipid lowering therapy. This clinical decision may be influenced by patient’s medical history, family history of CVD, and importantly the frequency of contacts with their primary care physicians. This lack of information on the timing of initiating lipid-lowering therapy may potentially overestimate the cross-sectional association between PFAS and hyperlipidemia. More specifically, at baseline, we did not have clear information on the timing of lipid-lowering therapy initiation, so we accounted for use of lipid-lowering therapy initiated at any time point prior to baseline. The strength of association between PFAS concentrations and hyperlipidemia for participants who initiated lipid-lowering therapy at a time point farther away from the baseline would likely be weaker than what we estimated, thus our cross-sectional result could be an overestimate of the true association between PFAS and hyperlipidemia. We were careful in interpreting this cross-sectional association for baseline prevalent cases to note this limitation. We expected this limitation to minimally impact the longitudinal association as participants were followed up every six-months after study enrollment, thus, we could estimate the timing of hyperlipidemia incidence within a six-month period. The external validity of our study was limited by the study design and cohort characteristics; our findings might not be generalizable to adults with lower BMI.

5. Conclusion

Among adults with pre-diabetes and comparable plasma PFAS concentrations to the U.S. general population, we observed positive cross-sectional associations of some plasma PFAS concentrations with total cholesterol, triglycerides and LDL levels. Higher concentration of some PFASs at baseline were also prospectively associated with higher cumulative incidences of hyperlipidemia. This elevated incidence was not observed among participants who received an intensive lifestyle intervention of diet and exercise. Even though participants gradually regained weight after the first year of intervention, the protective effect of the lifestyle intervention on hyperlipidemia persisted throughout the 15 years of follow-up. Weight loss, healthy diet, and exercise may modify and attenuate the potential adverse effects of environmental chemicals, such as PFASs, on lipid metabolism.

Table 2.

Plasma PFAS concentration (ng/mL) among 888 prediabetic adults in US.

PFAS analytes	% <LOD	Median (IQR)
PFOSa		27.2 (18.0, 40.4)
n-PFOS	0%	18.8 (11.9, 29.0)
Sm-PFOS	0%	7.6 (4.9, 11.4)
PFOAb		4.9 (3.5, 6.7)
n-PFOA	0%	4.3 (3.1, 6.0)
Sb-PFOA	16.6%	0.5 (0.3, 0.9)
PFHxS	0.1%	2.3 (1.4, 3.8)
EtFOSAA	3.3%	1.1 (0.6, 2.1)
MeFOSAA	2.6%	1.0 (0.6, 1.7)
PFNA	6.8%	0.6 (0.4, 0.8)

Note: PFAS: per- and polyfluoroalkyl substances; PFOS: Perfluorooctanesulfonic acid (sum of linear and branched isomers); PFOA: perfluorooctanoic acid (sum of linear and branched isomers); PFHxS: perfluorohexane sulfonic acid; EtFOSAA: N-ethyl-perfluorooctane sulfonamido acetic acid; MeFOSAA: N-methyl-perfluorooctane sulfonamido acetic acid; PFNA:perfluorononanoic acid.

^aSummary measure calculated by adding up concentrations of linear and branched isomers (n-PFOS and Sm-PFOS).

^bSummary measure calculated by adding up concentrations of linear and branched isomers (n-PFOA and Sb-PFOA), with values <LOD imputed with LOD/V2 before summation.

Highlights.

• PFAS may disrupt lipid regulation

• We examined PFAS-lipid relationship in prediabetic adults over 15 years

• Plasma PFAS concentrations had positive cross-sectional associations with total cholesterol level

• Risk of dyslipidemia was elevated in relation to baseline PFAS levels

• Lifestyle behaviors may attenuate the adverse effect on lipid profile

Abbreviations

ATP-II

National Cholesterol Education Program adult treatment panel guideline

CDC

Centers for Disease Control and Prevention

CI

confidence interval

CVD

cardiovascular disease

DPP

Diabetes Prevention Program

DPPOS

Diabetes Prevention Program Outcome Study

EtFOSAA

2-(N-Ethyl-perfluorooctane sulfonamido) acetic acid

GAM

generalized additive models

HDL

high-density lipoprotein

HR

hazard ratios

LDL

low-density lipoprotein

LOD

limit of detection

MeFOSAA

2-(N-Methyl-perfluorooctane sulfonamido) acetic acid

NHANES

National Health and Nutrition Examination Surveys

NIDDK

National Institute of Diabetes and Digestive and Kidney Diseases

n-PFOA

linear PFOA

n-PFOS

linear PFOS

PFAS

Per- and polyfluoroalkyl substances

PFHxS

Perfluorohexane sulfonic acid

PFNA

Perfluorononanoic acid

PFOA

Perfluorooctanoic acid (sum of linear and branched isomers)

PFOS

Perfluorooctanesulfonic acid (sum of linear and branched isomers)

Sb-PFOA

sum of perfluoromethylheptanoic and perfluorodimethylhexanoic acids

SD

standard deviation

Sm2-PFOS

sum of perfluorodimethylhexane sulfonic acid isomers

Sm-PFOS

sum of perfluoromethylheptane sulfonic acid isomers

VLDL

very low-density lipoprotein