Per- and Polyfluoroalkyl Substances and Blood Pressure in Pre-Diabetic Adults—Cross-Sectional and Longitudinal Analyses of the Diabetes Prevention Program Outcomes Study

Abstract

The relationship of plasma concentration of per- and polyfluoroalkyl substances (PFAS) with blood pressure (BP) is uncertain. This study examined cross-sectional and prospective associations of PFAS with BP and hypertension. We quantified plasma PFAS concentrations from 957 participants enrolled in the lifestyle and placebo arms of the Diabetes Prevention Program (DPP), a randomized controlled trial with approximately 15 years of follow-up. We used multivariable linear and logistic regressions to test cross-sectional associations of six PFAS, including 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), with BP and hypertension prevalence, respectively, at baseline. We used generalized linear mixed models to estimate longitudinal associations between baseline PFAS and the rate of BP changes, and Cox-Proportional hazard models to estimate risk of developing hypertension relative to baseline PFAS. Models were adjusted for baseline age, sex, race/ethnicity, treatment arm, educational attainment, income, marital status, smoking habit, alcohol drinking, and diet. We tested for effect modification by the treatment arm and sex, and accounted for multiple comparisons using the False-Discovery Rate (FDR). PFAS concentrations and hypertension prevalence within the study population (65.3% female, 57.7% White, 65.3% aged 40–59 years) were comparable to the general U.S. population. Cross-sectionally, we found small but statistically significant associations of baseline plasma concentrations of PFOA with systolic BP (β per doubling: 1.49 mmHg, 95% CI: 0.29, 2.70); and MeFOSAA with hypertension (RR=1.09 per doubling, 95% CI: 1.01, 1.19). Estimates were not statistically significant after FDR adjustment. Longitudinally, we observed null associations in the placebo arm, but some inverse associations of baseline PFOS and MeFOSAA with systolic BP in the lifestyle arm, perhaps due to regression toward the mean. Baseline PFAS concentrations also were not prospectively associated with hypertension risk. Overall, there were modest and mostly null associations of plasma PFAS concentrations with BP and hypertension.

1. Introduction

Increasing blood pressure (BP) has a strong, graded association with higher risk of cardiovascular disease (CVD) (Kjeldsen 2018). While the role of many modifiable risk factors for hypertension including body mass index (BMI), cigarette smoking, physical activity, and diet, have been widely studied, knowledge about the potential contribution from environmental contaminants is limited. Per- and polyfluoroalkyl substances (PFAS) are a family of synthetic chemicals with ubiquitous human exposure and previously shown to be associated with CVD risk (Huang et al. 2018; Shankar et al. 2012) as well as several CVD risk factors including thyroid disease (Kim et al. 2018; Melzer et al. 2010; Shrestha et al. 2015), elevated lipids (Costa et al. 2009; Eriksen et al. 2013; Fisher et al. 2013; Fu et al. 2014; Lin et al. 2019; Nelson et al. 2010; Sakr et al. 2007; Steenland et al. 2010; Winquist and Steenland 2014); and higher body mass index (Cardenas et al. 2017; Andres Cardenas et al. 2018; Liu et al. 2018).

In addition, PFAS have a hypothesized relationship with BP because of their potential to increase oxidative stress and impair vasodilation (Ceriello 2008; Huang et al. 2013; Panaretakis et al. 2001; Wielsoe et al. 2015). We identified six observational studies in adults on the association between PFAS and BP with conflicting findings. Cross-sectional analyses using data from the U.S. general population sampled in the National Health and Nutrition Examination Survey (NHANES) (2003–2006, N=2200) (Min et al. 2012) and from a highly exposed population in China (China C8 Health Project, N=1612) (Bao et al. 2017) found elevated BP and increased risk of hypertension with higher serum concentrations of PFOS, perfluorooctanoic acid (PFOA), and perfluorononanoic acid (PFNA). However, these findings were not reproduced among U.S. children (NHANES 1999–2000/2003–2008; N=1655) (Geiger et al. 2014) nor in prospective studies conducted among middle-aged adults in Sweden (N=187) (Donat-Vargas et al. 2019) or in occupational and community residents from polluted areas in the USA (US C8 Health Project, N=32,254) (Winquist and Steenland 2014). Two studies with smaller sample sizes even reported weak inverse, cross-sectional associations of PFOS and PFNA with hypertension (Christensen et al. 2016; Donat-Vargas et al. 2019).

To better evaluate the relationship of PFAS with BP and hypertension, we used data from the Diabetes Prevention Program (DPP), which is a multi-center randomized controlled trial in the U.S. that collected repeated blood samples and measured BP over 15 years of follow-up. This well-characterized cohort of pre-diabetic and overweight adults allowed us to examine both cross-sectional and prospective relationships within the same population while considering important baseline modifiable risk factors for CVD and potential effect modification by sex and lifestyle intervention. We hypothesized that associations of plasma PFAS concentrations with BP and hypertension would be positive but weak given previous reports and that the initial DPP lifestyle intervention would attenuate any adverse effect of PFAS on BP outcomes.

2. Methods

2.1. Study population

The Diabetes Prevention Program (DPP, ClinicalTrails.gov number, NCT00004992) was a randomized, controlled clinical trial conducted at 27 clinical centers around the United States from 1996 to 2001; the study goal was to assess the effect of lifestyle intervention or medication (metformin) on delaying or preventing type 2 diabetes. A detailed description on the recruitment is presented in the Supplementary Materials (Figure S1) and the study design is available elsewhere (Diabetes Prevention Program Research Group 1999). Participants from the lifestyle intervention arm (N=1,079) received 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); while participants from the metformin (N=1,073) and medication placebo (N=1,082) arms took 850 mg of metformin or placebo-medication, respectively, twice a day and received standard advice about diet and physical activity. At the end of the study period of DPP (average 2.8 years of follow-up), participants in the lifestyle arm had 58% lower likelihood of developing type 2 diabetes compared to those in the placebo arm; while those in the metformin arm had 31% lower risk (Diabetes Prevention Program Research 2002). After the DPP ended, the DPP Outcomes Study (DPPOS, 2002–2014, ClinicalTrails.gov number, NCT00038727) enrolled 88% of the surviving DPP participants for ongoing follow-up.

In this analysis, we included 957 participants from the lifestyle (N=481) and medication placebo (N=476) arms who had blood samples available (≥400 μL) in the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) central repository (https://repository.niddk.nih.gov/home). We did not include participants from the metformin arm because of limited funding and unknown kinetics between PFAS and this medication. We requested blood samples from three time points: baseline (1996–1999), DPP second annual visit (1998–2001, referred to as year 2 in the rest of the paper), and DPPOS tenth annual visit (2010–2013, referred to as year 14 in the rest of the paper). The study flow chart (Supplementary Materials, Figure S1) depicts the numbers of participants at each time point of follow-up. All DPP/DPPOS participants provided written informed consent. This study used de-identified data from the February 2008 Full Scale Data Release available from NIDDK Central Repository (Diabetes Prevention Program 2008) and 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 PFAS

Baseline, year 2 and year 14 blood samples from the DPP biorepository were retrieved for quantification at the CDC (National Center for Environmental Health, Division of Laboratory Sciences). Plasma PFAS concentrations were measured using a modified version of the online solid-phase extraction coupled to high-performance liquid chromatography-isotope dilution-tandem mass spectrometry method described before (Kato et al. 2011). We quantified five linear and branched isomers of PFOS and PFOA— n-perfluorooctane sulfonate (n-PFOS), perfluoromethylheptane sulfonates (Sm-PFOS), perfluorodimethylhexane sulfonates (Sm2-PFOS), n-perfluorooctanoate (n-PFOA), and branched perfluorooctanoates (Sb-PFOA). The limit of detection (LOD) was 0.1 ng/mL for all PFAS examined. We replaced concentrations below the LOD with LOD/√2 (Hornung and Reed 1990) for subsequent statistical analyses, and calculated total PFOS and total PFOA as the sum of their linear and branched isomers using the method used in NHANES (Centers for Disease Control and Prevention 2019). This current analysis included the six most detected PFAS with >90% detection frequency at baseline: PFOS, PFOA, perfluorohexane sulfonic acid (PFHxS), N-ethyl-perfluorooctane sulfonamido acetic acid (EtFOSAA), N-methyl-perfluorooctane sulfonamido acetic acid (MeFOSAA), and PFNA. We calculated a molar sum of the 6 PFAS to generate a sum PFAS variable for easier interpretation, however, we recognized this molar sum value is an imperfect way to obtain a total for the measured PFAS compounds with high detection rates as it does not address potency differences nor account unmeasured PFAS in blood, thus we only showed the result of sum PFAS in the supplementary material as a reference.

2.3. Outcomes: BP measurements

We extracted all BP measurements available from the NIDDK Repository for each participant. Study staff measured systolic and diastolic BP semiannually according to standard protocols (Diabetes Prevention Program Outcomes Study Research Group et al. 2013). Participants were seated in a chair for 5 minutes before having BP measurements taken twice using a standard mercury manometer separated by 30 seconds, the mean of which was used for this analysis. Pulse pressure was calculated as the difference between systolic and diastolic BP. We defined hypertension at each timepoint as self-reported hypertension diagnosis, use of anti-hypertensive medications, or systolic/diastolic BP ≥140/90 mmHg (or ≥130/80 mmHg for those with diabetes).

2.4. Covariates

We evaluated a priori potential confounders likely to be correlated with plasma PFAS concentrations and BP, including baseline age, sex, race/ethnicity, treatment assignment (for post-baseline analyses), education, income, marital status, alcohol drinking, smoking, and Dietary Approaches to Stop Hypertension (DASH) diet score. We drew directed acyclic graph to evaluate the underlying causal relationship. Research staff assessed participants’ characteristics using structured questionnaire. We used self-reported dietary intake during the previous 12 months captured using a validated semiquantitative food frequency questionnaire to calculate DASH diet score (range 0 to 9 points), which is composed of a linear index of nine nutrient components, including sodium, cholesterol, saturated fat, total fat, protein, calcium, magnesium, potassium, and fiber (Lin et al. 1999; Mellen et al. 2008). Additional baseline covariates considered included BMI, menopause status, physical activity, and estimated glomerular filtration rate (eGFR), calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (Levey et al. 2009).

2.5. Statistical analysis

We used descriptive analyses to examine participants’ characteristics at baseline (Table 1) and plotted mean levels of BP and hypertension rate over time for each of the two treatment groups. We used Chi-square or Kruskal-Wallis tests to compare BP, hypertension prevalence, and plasma PFAS concentrations between treatment arms at baseline, year 2, and year 14 (Table 2). We evaluated distributions of continuous variables using the Shapiro-Wilk test. Plasma PFAS concentrations were right-skewed, thus we log-2 transformed and grouped them into quartiles for some analyses. We assessed the trend across the three repeated measures using group-based latent class trajectory modeling (Jones et al. 2001), see Supplementary Material Appendix 1 for details.

Table 1.

Baseline characteristics of study participants (N=957)

Characteristics	%, or mean ± STD
Group assignment
.Lifestyle	50.3
.Placebo	49.7
Sex
. Male	34.7
. Female	65.3
Race/ethnicity
. Non-Hispanic white	57.7
. African American	19.2
. Hispanic of any race	18.7
. All others	4.4
Age (years)
. <40	11.7
. 40–44	11.2
. 45–49	22.3
. 50–54	17.5
. 55–59	14.3
. 60–64	11.2
. >65	11.9
Educational attainment
. < High school	4.7
. High school/GED	20.9
. College	49.0
. Graduate school	25.4
Marital Status
. Married/Cohabitating	67.6
. Single	11.9
. Divorced	15.9
. Widowed	4.6
Annual Income
. $20,000	14.6
. $20,000 - <$35,000	17.5
. $35,000 - <$50,000	20.3
. $50,000 - <$75,000	19.2
. ≥ $75,000	22.4
. Refused to answer	8.1
Alcohol consumption
. Nondrinker	54.1
. <1 drink/week	16.7
. ≥1 drink/week	29.2
Current smoker	5.4
DASH Diet Score (0–9 range)	2.5 ± 1.6
Menopausal, % among female participants	54.4
Body mass index (kg/m2)	33.5 ± 5.8

Table 2.

Blood pressure, hypertension status, and plasma PFAS concentrations at three time points during DPP/DPPOS follow-up

	Baseline (1996–1999)	DPP Year 2 (1998–2001)			DPPOS Year 10 (2010–2013)
		All	Lifestyle arm	Placebo arm
Sample size, N	957	956	480	476	677
Years since enrollment, mean ± STD	0	2.0 ± 0.0			14.0 ± 0.8
Blood pressure, mean ± STD
. Systolic, mmHg	124.4 ± 14.7	122.5 ± 14.5c	120.8 ± 14.5	124.2 ± 14.4	121.6 ± 15.0
. Diastolic, mmHg	78.5 ± 9.1	76.4 ± 9.0c	75.5 ± 9.0	77.4 ± 9.0	70.1 ± 9.5
. Pulse pressure, mmHg	45.9 ± 11.7	46.0 ± 11.5c	45.3 ± 11.4	46.8 ± 11.5	51.2 ± 13.2
Change of blood pressure from baseline, mean ± STD
. Systolic, mmHg	--	−1.9 ± 13.5c			−2.4 ± 17.7
. Diastolic, mmHg	--	−2.1 ± 8.8c			−8.2 ± 11.0
. Pulse pressure, mmHg	--	0.2 ± 11.3c			5.8 ± 13.3
Hypertension status, %
. SBP ≥140mmHg or DBP ≥90mmHg	21.3	15.5c	12.3	18.7	11.1
. Hypertension diagnosisa, %	20.1	21.3			49.0
. Hypertensionb, %	31.0	30.0c	26.5	33.4	51.7
Plasma PFAS concentrations, Median (IQR)
. PFOS, ng/mL	26.7 (17.4-40.3)	27.6 (19.6-38.9)			9.8 (5.9–14.8)
. PFOA, ng/mL	4.9 (3.5–6.7)	5.7 (4.0–8.0)c	5.4 (3.8, 7.8)	5.9 (4.2, 8.2)	2.8 (2.0–3.8)
. PFHxS, ng/mL	2.3 (1.4–3.8)	2.4 (1.5–3.9)			1.7 (1.0–2.7)
. EtFOSAA, ng/mL	1.1 (0.6–2.1)	1.0 (0.5–1.8)			<LOD (<LOD)d
. MeFOSAA, ng/mL	1.0 (0.6–1.7)	1.1 (0.7–1.7)			<LOD (<LOD-0.3)d
. PFNA, ng/mL	0.6 (0.4–0.8)	0.6 (0.4–0.9)			0.9 (0.6–1.5)

Note:

^aDefined as having self-reported hypertension diagnosis or use of anti-hypertension medication

^bDefined as self-reported hypertension diagnosis, use of antihypertensive medication, systolic/diastolic blood pressure ≥140/90 mmHg or ≥130/80 mmHg for those with diabetes

^cLevel differed across group assignment, p<0.05 using Chi-square test or Kruskal-Wallis Test.

^dThe limit of detection (LOD) was 0.1 ng/mL.

We performed cross-sectional (Figure 1) and longitudinal analyses (Figure 2) of the associations between PFAS concentrations (six main PFAS and five isomers of PFOS and PFOA) and BP (systolic, diastolic and pulse pressure) and examined the prospective hazard of developing hypertension relative to plasma PFAS concentrations measured at baseline (Table 3). Each model included a total of 11 iterations (six PFAS and five isomers), and we accounted for multiple testing by adjusting the false-discovery rate (FDR) at 5% using the Benjamini and Hochberg step-up method (Benjamini and Hochberg 1995).

Figure 1.

Cross-sectional associations of plasma PFAS concentrations with blood pressure and hypertension prevalence among DPP/DPPOS participants (N=765)

Note: Estimated using multivariable linear regression among participants with no hypertension diagnosis and who were not taking anti-hypertensive medication at baseline. All models adjusted for age, sex, race, treatment assignment, education, income, marital status, alcohol drinking, smoking, and DASH diet score.

Figure 2.

Longitudinal associations between baseline plasma PFAS concentrations and BP trajectories over 15 years of follow-up in DPP/DPPOS (N=940)

Note: Effect estimates (markers) and 95% confidence interval (band) are interpreted as rate of blood pressure change (mmHg/y) per doubling of plasma PFAS concentrations measured at baseline. All models adjusted for age, sex, race, treatment assignment, education, income, marital status, alcohol drinking, smoking, and DASH diet score.

Table 3.

Prospective hazard ratio (HR) of hypertension relative to baseline plasma PFAS concentration (per doubling increase) among participants without hypertension at baseline

	Cross-sectional Risk (N=957)	Prospective Hazard (N=660)
Log-2 PFAS analytes	RR (95% CI)	HR (95% CI)	Effect modification by treatment arm (p-value)b
. PFOS	1.06 (0.96, 1.17)	0.95 (0.84, 1.06)	0.25
. PFOA	1.08 (0.96, 1.21)	0.95 (0.83, 1.08)	0.33
. PFHxS	1.04 (0.96, 1.13)	0.94 (0.86, 1.03)	0.91
. EtFOSAA	0.98 (0.92, 1.04)	0.98 (0.92, 1.06)	0.12
. MeFOSAA	1.09 (1.01, 1.19)	0.95 (0.87, 1.03)	0.93
. PFNA	1.04 (0.95, 1.14)	0.99 (0.91, 1.08)	0.52

Note:

^aHypertension defined as self-reported hypertension diagnosis, use of antihypertensive medication, systolic/diastolic blood pressure ≥140/90 mmHg or ≥130/80 mmHg for those with diabetes; All models adjusted for age, sex, race, treatment assignment, education, income, marital status, alcohol drinking, smoking, and DASH diet score.

^bEffect modification of baseline PFAS concentration by treatment arm assessed by adding a multiplicative term between plasma PFAS concentration and treatment assignment in the Cox-Proportional Hazard model.

For cross-sectional analyses of associations between PFAS and BP, we used measurements taken at baseline and restricted the sample to participants without hypertension diagnosis nor on anti-hypertensive medications. We used linear regression models to estimate the difference in mean BP levels per doubling of PFAS concentration. Covariates and confounders considered in the model included age, sex, race/ethnicity, educational attainment, income, marital status, alcohol consumption, smoking status, DASH diet score. We tested the effect of physical activity, eGFR, menopause status, and BMI on the cross-sectional associations by adding them into the model, one at a time. We used complete case analysis for all models, as missingness was <10% (See Supplementary Materials Table S1 for number of missing for each covariates). To relax the linearity assumption, we replaced continuous plasma PFAS concentrations with a categorical variable (quartiles, see Supplementary Materials Table S2 for quartile ranges), and also ran generalized additive models (GAM) including smoothing splines with 4 degrees of freedom on plasma PFAS concentrations. We also evaluated the cross-sectional relationships between PFAS and BP measurements taken at year 2 and year 14 among participants who did not previously have a hypertension diagnosis and were not taking anti-hypertensive medications, adjusting for participants’ characteristics at the respective time point. In the sensitivity analysis, we did not exclude participants with hypertension diagnosis or on anti-hypertensive medications, instead, we added a constant of 10 mmHg and 5 mmHg to participants’ systolic and diastolic BP for those who received hypertensive treatment at baseline, following the method previous reported (Tobin et al. 2005). To test for reverse causation of the above cross-sectional relationships, we regressed baseline BP, as independent variable, on later PFAS concentrations as dependent variables. For the hypertension outcome, we used Poisson regression with robust variance (Zou 2004) to estimate the cross-sectional relative risk (RR) at baseline relative to baseline plasma PFAS concentrations.

For the longitudinal analysis, we used generalized linear mixed-effect models with random intercepts and slopes to model the relationship between baseline plasma PFAS concentrations and mean BP trajectories during the approximately 15 years of follow-up. We included all participants (N=957) and did not exclude participants with hypertension diagnosis or on anti-hypertensive medications;, we added a constant of 10 mmHg and 5 mmHg to participants’ systolic and diastolic BP at timepoints when they reported hypertensive treatment. We first used fractional polynomials (FP) to obtain the best fitted average trajectory for BP and then added baseline plasma PFAS concentration into the longitudinal model to evaluate the effect of baseline PFAS on BP trajectories. Visual inspection of the best-fitted FP curve suggested stratification by treatment arms and that two knots points would sufficiently describe the BP change over time. To provide parameter estimates more relevant to the timing of the intervention, we applied the piecewise linear function in longitudinal models, segregating data into two timeframes (before year 2 and after year 2). This method produced trajectories similar to that of the best-fitted FP curves and accounted for the intensive changes in BP after the initiation of DPP intervention (Figure S2). A detailed description of the longitudinal model is presented in Appendix 2 of the Supplementary Materials. For all models, we tested model assumptions and identified the unstructured covariance model as the most optimal fit based on Akaike information criteria. The final model included baseline covariates and confounders used in the cross-sectional model and study arm, as well as a square term for time for each of the two timeframe and cross-product terms between time and log-2 transformed baseline PFAS concentrations. The regression coefficient for the cross-product term estimates average slope of BP change in mmHg/year per doubling increase in baseline PFAS concentrations. For the hypertension outcome, we used Cox-proportional hazard models to estimate hazard ratio (HRs) of developing hypertension prospectively relative to baseline plasma PFAS concentrations. We evaluated the proportional hazard assumption using Martingale-based residuals (Lin et al. 1993). We performed a sensitivity analysis restricting the time of follow-up in the Cox-proportional hazard models to 0 to 10 years, when more than 80% of the participants remained in the follow-up.

For all models, we evaluated effect modification by sex and treatment assignment by adding a multiplicative term between the effect modifier and plasma PFAS concentration and considered an evidence for effect modification if p-values of the multiplicative term was <0.10. We presented stratified effect estimates if effect modification was suggested. We used SAS 9.4 for all analyses.

3. Results

3.1. Population demographics

Table 1 summarizes participants’ characteristics at baseline. Similar to the overall DPP study population the majority of the 957 prediabetic adults included in this analysis were female (65.3%), non-Hispanic white (57.7%), college educated or higher (74.4%), married (67.6%), and non-current smoker (94.6%). About equal proportions of participants were randomized to the lifestyle (50.3%) and the placebo (49.7%) arms, with comparable demographic characteristics across the two arms (data not shown). Participants had baseline plasma PFAS concentrations similar to median serum concentrations reported among the U.S. general population sampled in NHANES 1999–2000 (Centers for Disease Control and Prevention 2019). All six PFAS were frequently detected (93~100%) with PFOS (26.7 ng/mL) and PFOA (4.9 ng/mL) having the highest concentrations. As expected, plasma PFAS concentrations remained similar at year 2, considering the relatively long half-lives of elimination in humans of several PFAS, but declined by more than 50% at year 14 for most PFAS except for PFNA (Table 2, see Supplementary Materials Figure S3 for the correlation heat map between the repeated measurements of the 6 PFAS). There was no significant difference in the median concentrations and rate of change across treatment arms at the three time points of PFAS measurements. For PFOS, PFOA, PFHxS, and PFNA, group-based latent-class modeling identified distinct trajectories that differed only by the starting baseline concentration but not the slope of PFAS concentration change (see Supplementary Materials Appendix 1), which supported the use of baseline PFAS concentrations to account for variations in PFAS concentrations in subsequent analyses. Baseline concentrations of EtFOSAA and MeFOSAA were also different across the distinct trajectories. Consistent with our previous reports, (Cardenas et al. 2017; A. Cardenas et al. 2018; Cardenas et al. 2019; Lin et al. 2019) due to variations in the exposure patterns, we observed differences in PFAS concentrations by sex, age, and race/ethnicity (Table S2).

Participants’ systolic and diastolic BP did not differ significantly between the two treatment arms at baseline, indicating successful randomization, with overall mean ± SD systolic BP being 124.4 ± 14.7 mmHg and diastolic 78.5 ± 9.1 mmHg (Table 2). BP decreased after study initiation, with participants from the lifestyle arm having experienced more reduction during the DPP phase (baseline to year 4) and returning to levels comparable to the placebo arm in the subsequent years of follow-up during DPPOS (see Supplementary Materials Figure S2 for the trend over time). The mean ± SD pulse pressure increased over time, from 45.9 ± 11.7 mmHg at baseline to 51.2 ± 13.2 mmHg at year 14. The percentage of participants with hypertension was 31.0% at baseline, 30.0% at year 2, and 51.7%, at year 14, comparable to the age-specific prevalence trend in the US (Booth et al. 2017).

3.2. Cross-sectional associations between PFAS and BP

Among 765 participants without hypertension at baseline, univariate analysis showed small but statistically significant associations between plasma concentrations of PFOA and PFNA with systolic BP [β_{per doubling of PFOA}: 1.72 mmHg (95% CI: 0.55, 2.90); β_{per doubling of PFNA}: 1.12 mmHg (95% CI: 0.34, 1.89)]; and MeFOSAA with diastolic BP (β_{per doubling of MeFOSAA}: 0.63 mmHg (95% CI: 0.15, 1.11) (Supplementary Materials, Table S3). After accounting for covariates and confounders, all effect estimates attenuated and only the association between PFOA and systolic BP [β_{per doubling of PFOA}: 1.49 mmHg (95% CI: 0.29, 2.70)] remained statistically significant at a 5% type I error rate before accounting for multiple comparisons. Figure 1 summarizes the covariate-adjusted associations between of PFAS on BP. We examined linear and branched isomers of PFOA separately and found the association to be mainly driven by the linear isomer, which was detected at higher concentration compared to the branched isomers (Supplementary Materials, Table S3). Among all covariates considered, age had the strongest positive association with BP (data not shown). Additional adjustment for physical activity, BMI, eGFR, and menopause status (among females only) in the model did not change the effect estimate substantially. A GAM with a non-linear spline term did not significantly improve the model fit. Using quartiles, individuals in the 3^rd and 4^th quartiles of plasma PFOA concentrations (3.0–4.4 ng/mL and 4.5–6.6 ng/mL, respectively) had higher systolic BP compared to the 1^st quartile (0.6 to 2.9 ng/mL), i.e. β_{Q4 vs Q1}: 2.38 mmHg (95% CI: −0.30, 5.07), β_{Q3 vs Q1}: 2.49 mmHg (95% CI: −0.11, 5.09); although effect estimates were moderate in magnitude, the confidence interval included the null (Supplementary Materials, Table S4). We did not observe evidence of effect modification by sex for any associations at baseline; the p-values of the multiplicative term between PFAS and sex were all >0.10, but we did observe stronger associations between PFOA and systolic BP among males compared to females (Supplementary Materials, Table S5). Sensitivity analysis using modified BP (adding 10 mmHg to systolic and 5 mmHg and diastolic BP) to include participants on anti-hypertensive medications at baseline yielded similar effect estimates but wider confidence intervals and none of the associations were statistically significant (data not shown). After we accounted for multiple testing using FDR, none of the covariate adjusted parameter estimates was statistically significant. When we tested for reverse causation by replacing PFAS as the dependent and BP as the independent variable in the multivariable linear regression models, baseline BP was not associated with change of PFAS concentrations from baseline to year 2 nor from baseline to year 14 (data not shown).

In separate multivariate linear regression models, we repeated the cross-sectional analyses at year 2 (N=752) and year 14 (N=346) and did not find any statistically significant associations (Table S3).

3.3. Longitudinal associations between PFAS and BP

We modelled the BP trajectories during the approximately 15 years of follow up and examined the extent to which the trajectories differed by baseline plasma PFAS concentrations among all 957 participants. Longitudinal models with interaction terms between PFAS concentrations and treatment assignment had better model fits (smaller AIC and BIC) compared to models without the interaction term. Thus, we stratified all longitudinal analysis by treatment arms. The result showed mostly null longitudinal association between baseline PFAS concentration with trajectories of BP. Figure 2 showed the rate of BP change over time relative to each doubling of baseline PFAS concentrations. We did observe higher baseline PFOS and EtFOSAA concentrations to be associated with more negative rate of BP change in the lifestyle arm [at year 2, β_PFOS: −2.13 mmHg/year (95% CI: −3.54, −0.71); β_EtFOSAA: −1.47 mmHg/year (95% CI: −2.41, −0.59) per doubling of baseline PFAS concentrations], but not in the placebo arm [at year 2, β_PFOS: −0.01 mmHg/year (95% CI: −1.51, 1.33); β_EtFOSAA: −0.54 mmHg/year (95% CI: −1.45, 0.37)]. However, we did not find evidence for effect modification by study arm or sex.

3.4. Baseline plasma PFAS concentrations and hypertension

We tested whether baseline plasma PFAS concentrations was associated with prevalence and incidence of hypertension, which was self-reported hypertension diagnosis, use of anti-hypertensive medications, or systolic/diastolic BP ≥140/90 mmHg (or ≥130/80 mmHg for those with diabetes). At baseline , we found 9% higher prevalence of hypertension per doubling of MeFOSAA [RR per doubling=1.09 (95% CI: 1.01, 1.19), Table 3]. Using concentration quartiles, the RR was 59% higher comparing the 4^th to the 1^st quartile [RR_{Q4 vs Q1}=1.59 (95% CI: 1.05, 2.41), Supplementary Materials, Table S6]. The cross-sectional association of PFOA and hypertension was not statistically significant [RR_{Q4 vs Q1}=1.24 (95% CI:0.91, 1.68), Supplementary Materials, Table S6], but was modified by sex (Supplementary Materials, Table S7); among male participants, each doubling of plasma PFOA concentration was associated with 27% higher hypertension prevalence [RR=1.27 (95% CI: 1.06, 1.54)] while the association was null among females [RR=0.99 (95% CI: 0.85, 1.15)]. Among participants who did not have hypertension at baseline (N=660), there was no significant association between hypertension incidence at the approximately 15 years of follow-up with respect to baseline plasma concentrations for any PFAS, and there was no effect modification by treatment assignment (all interaction p-value > 0.10) (Table 3). Findings remained consistent in sensitivity analyses when restricting the study follow-up time to 10 years (data not shown). We additionally examined the incidence of hypertension up to year 2 of DPP and found consistent null associationa (Table S6).

4. Discussion

In this population of overweight and prediabetic adults in the USA, whose hypertension prevalence and plasma PFAS concentrations were comparable with the general U.S. population of the same age range, we detected some cross-sectional but mostly null longitudinal associations between plasma PFAS concentrations with BP and hypertension. At baseline, we found elevated systolic BP with higher plasma PFOA concentration and increased baseline prevalence of hypertension with MeFOSAA and PFOA (among men only). However, associations were not statistically significant after accounting for multiple testing.

Literature on the relationship between PFAS and BP is relatively limited. Histological observations from animal studies showed perfluorooctane sulfonic acid (PFOS) exposure induced the production of reactive oxygen species which may lead to actin filament remodeling and endothelial permeability change in human microvascular endothelial cells (Qian et al. 2010). In vivo evidence suggests PFAS exposure causes oxidative stress in the liver and endothelial cells which are the major underlying mechanism responsible for most PFAS-induced health effects (Huang et al. 2013; Panaretakis et al. 2001; Yao and Zhong 2005). Oxidative stress has been proposed to be a key player in the pathogenesis of hypertension (Montezano and Touyz 2012; Rodrigo et al. 2011). While endothelial cells regulate arterial relaxation, and PFAS-induced oxidative stress might lead to impaired vasodilation (Ceriello 2008; Huang et al. 2013; Panaretakis et al. 2001). Previous cross-sectional findings from epidemiologic studies on the associations of PFAS with BP were inconsistent, with two reported positive associations (Bao et al. 2017; Min et al. 2012) and other studies showed null findings (Christensen et al. 2016; Donat-Vargas et al. 2019; Geiger et al. 2014; Winquist and Steenland 2014). Among the two studies that reported positive associations, the magnitude of associations was in ranges comparable to what we detected in our study. The cross-sectional analysis of 2934 adults age >20 years sampled in NHANES 2003–2004 and 2005–2006 (median PFOA=4.9 ng/mL at baseline) showed each natural log unit (approximately 2.71 times) increase in serum PFOA was associated with 2.49 mmHg (SE=0.63) higher systolic BP and 71% higher (OR=1.71, 95% CI: 1.23 to 2.36) odds of hypertension comparing the fourth to the first quartile of PFOA (Min et al. 2012). The other study among 1,612 adults in China showed each natural log unit increase in the serum concentrations of various PFAS isomers was associated with 1 to 5 mmHg increase in systolic BP [e.g. β_PFOS=4.48 (95% CI: 3.55, 6.12); β_PFOA=1.16 (95% CI: 0.25, 3.13); β_PFNA=3.01 (95% CI: 1.79, 4.23)] and 19% to 24 % increase in the odds of hypertension [e.g. OR_PFOS=1.24 (95% CI: 1.08, 1.44); β_PFNA=1.19 (95% CI: 1.04, 1.36)] (Bao et al. 2017). We should note that our study samples were pre-diabetic adults with elevated BMI who had higher risk to develop hypertension and CVD, thus, we should be careful when directly comparing our findings with those from adults sampled from the general population. Since participants in our study may experience more competing risks for adverse BP outcomes compared to healthy adults with normal BMI, any associations between PFAS and BP might be attenuated. This was a possible explanation for the weaker or null associations we observed in our study compared to the previous studies.

Our prospective analysis was the first, as far as we are aware, to examine the longitudinal change in BP relative to plasma PFAS concentration. We found that plasma PFOS and EtFOSAA concentrations at baseline were associated with more negative rate of systolic BP changes in the lifestyle but not the placebo group. We observed null effect estimates for all other longitudinal associations and did not detect effect modification by treatment arm nor sex. Previously, there were two longitudinal studies that examined the associations between PFAS and BP using population-level, rather than individual level, dataand both yielded mostly null results (Donat-Vargas et al. 2019; Winquist and Steenland 2014). The retrospective analysis on the U.S. C8 Health Project participants (N=32,254) did find some evidence of elevated hazard of hypertension with higher cumulative PFOA exposure among some subgroups (20–39 years old female and 40–59 years old male age groups), however, the trend of hazard was not monotonic across quartiles and the study used environmental monitoring data rather than individual biomarker measurements (Winquist and Steenland 2014)The inverse longitudinal associations we observed of PFOS and EtFOSAA with BP trajectories in the lifestyle group could be a result of regression toward the mean, a statistical phenomenon where a more extreme value from the population mean can experience a greater change toward the mean (Bland and Altman 1994). This finding could also suggest a more beneficial effect from lifestyle intervention for participants with higher plasma PFAS concentrations prior to study initiation.

Overall, our study showed no apparent association between plasma PFAS concentrations with BP and hypertension among prediabetic adults with elevated BMI. The fact that we observed only cross-sectional but not longitudinal association (in the placebo group) led us to use caution when interpreting the cross-sectional findings. Test of reverse causation did not suggest variations in BP could lead to variations in plasma PFAS concentration. Other potential drivers of the cross-sectional associations included unmeasured confounder such as consumption of packaged and processed food. We did adjust for DASH diet score in our analysis, which accounted for salt intake but not exposures to other chemicals or additives in processed and packaged food. It was also possible the underlying driver of the cross-sectional association was modified after the initiation of DPP intervention, diminishing the effect of PFAS on BP in the longitudinally follow-up. All participants received healthy lifestyle recommendations, which might have led the study participants to alter their diet, physical activity, and health awareness levels that could influence both PFAS exposures and BP control. Since we did not have a clear understanding of the temporal relationship of each variable nor did we have adequate measures of the time-varying confounders, we adjusted only for confounders measured at baseline in the longitudinal models. Our previous analyses of the same population showed that the DPP lifestyle intervention modified several adverse health outcomes associated with PFAS exposure, including increases in weight, body size, glycemic indicators, diabetes and total cholesterol (Cardenas et al. 2017; Andres Cardenas et al. 2018; Cardenas et al. 2019; Lin et al. 2019). For the mostly null associations we observed in this analysis between plasma PFAS concentrations and BP and hypertension, we did not expect and did not detect effect modification by lifestyle intervention, which further suggested the lack of a causal relationship of PFAS with BP and hypertension. Our study had several strengths. First, we evaluated sex-specific and isomer-specific effects systematically. Sex- and isomer-specific elimination and toxicity of certain PFAS had been reported in animal studies (Harada et al. 2005; Kudo et al. 2002; Loveless et al. 2006). We detected a sex-specific association between PFOA and hypertension prevalence at baseline, warranting further evaluation in other study populations. Potential mechanisms for having lower plasma PFAS concentration among females include transferring of PFAS to fetus during pregnancy (Berg et al. 2014), excretion via breast milk (Thomsen et al. 2010), or blood loss from menstruation (Wong et al. 2014). Another strength was the well-characterized data with repeated measures of plasma PFAS concentrations and BP of the same cohort, which was not available in previous studies. We performed careful assessment of participants’ prescription records to ensure the anti-hypertensive medication use was intended for BP control. The unique study design allowed us to test effect modification by lifestyle intervention, while the longitudinal follow-up data enabled us to examine both cross-sectional and prospective associations at the same time, reducing the bias from reverse causation.

Some limitations of the study include the potential for residual confounding and limited external validity based on participants’ characteristics. One of the residual confounders we did not account for was the inter-individual variation in PFAS absorption and elimination; though, we did control for eGFR to account for kidney function, which determines PFAS elimination efficiency. Our results reflected the associations observed among pre-diabetic adults with elevated BMI, who have higher risk for hypertension, thus, the results of our study may not be generalizable to a healthy cohort with normal weight as their lifestyle and exposure pattern to PFAS might be different. It was also possible that any true associations between PFAS and BP was masked by the other competing risks factor prevalent in study population, leading to an observed null association in our study population. However, given the current global obesity epidemic, our study provides relevant and valuable evidences on the risk of PFAS exposure on BP among an overweight/obese population with high-risk for cardiometabolic disease. Another limitation on external validity was that our study population was recruited between 1996 to 1999, and since then serum concentrations of the legacy PFAS, notably PFOS and PFOA, have declined over time among the U.S. general population (Centers for Disease Control and Prevention 2019) due to manufacture’s phasing out. Though our analysis could not reflect the current PFAS exposure level among the US population, its findings offered an opportunity to assess the potential risk at a higher exposure level.

5. Conclusions

Among an ethnically diverse population of prediabetic adults in the USA, we found mostly null associations between the PFAS examined with BP and hypertension. We detected weak positive cross-sectional associations between plasma PFOA concentration with elevated systolic BP and MeFOSAA with hypertension, with magnitudes of effects similar to those previously reported. However, the lack of longitudinal associations suggested the detected cross-sectional associations could be due to residual confounding.

Highlights.

• The relationship between PFAS and blood pressure is unclear

• The study examined this relationship among prediabetic adults in US over 15 years

• There were small associations between PFAS and BP at baseline

• Baseline PFAS were not associated with longitudinal changes in BP over time

• Baseline PFAS were not associated with incidence of hypertension over time

Abbreviations

BMI

Body mass index

BP

Blood pressure

CDC

Centers for Disease Control and Prevention

CVD

Cardiovascular disease

DASH

Dietary Approaches to Stop Hypertension

DPP

Diabetes Prevention Program

DPPOS

Diabetes Prevention Program Outcomes Study

eGFR

Estimated glomerular filtration rate

EtFOSAA

N-ethyl-perfluorooctane sulfonamido acetic acid

FDR

False-discovery rate

FP

Fractional polynomials

GAM

Generalized additive model

HR

Hazard ratio

LOD

limit of detection

MeFOSAA

N-methyl-perfluorooctane sulfonamido acetic acid

NHANES

National Health and Nutrition Examination Survey

NIDDK

National Institute of Diabetes and Digestive and Kidney Diseases

n-PFOA

n-perfluorooctanoate

n-PFOS

n-perfluorooctane sulfonate

PFAS

Per- and polyfluoroalkyl substances

PFHxS

perfluorohexane sulfonic acid

PFNA

Perfluorononanoic acid

PFOA

Perfluorooctanoic acid

PFOS

Perfluorooctane sulfonic acid

RR

Relative risk

Sb-PFOA

branched perfluorooctanoates

Sm2-PFOS

Perfluorodimethylhexane sulfonates

Sm-PFOS

Perfluoromethylheptane sulfonates