Cross-sectional associations between serum PFASs and inflammatory biomarkers in a population exposed to AFFF-contaminated drinking water

Abstract

Background:

Per- and polyfluoroalkyl substances (PFASs) are widespread and persistent environmental contaminants. Exposure to several PFASs has been associated with altered immune function in humans, including autoimmune disease and impaired response to vaccination. However, changes to the profile of inflammatory biomarkers in adults exposed to PFASs has not been extensively described.

Objective:

To estimate cross-sectional associations between serum PFASs and markers of inflammation among adults in a population exposed to aqueous film forming foam (AFFF)-contaminated drinking water.

Methods:

We quantified concentrations of 48 PFASs in non-fasting serum samples from 212 non-smoking adults. In the same serum samples, we measured concentrations of ten pro- and anti-inflammatory cytokines. We restricted analysis to seven PFASs detected in >85% of participants and the following four cytokines detected in ≥30% of participants: interleukin [IL]-1β, IL-6, IL-10, and tumor necrosis factor [TNF]-α. We fit multiple linear regression or logistic regression models, adjusted for potential confounders, to estimate associations between concentrations of each PFAS and either continuous or categorical (above vs below limit of detection) concentrations of each cytokine. We additionally applied Bayesian kernel machine regression to describe the combined effect of the PFAS mixture on each cytokine outcome.

Results:

Certain PFAS concentrations in this sample were elevated compared to a US nationally representative sample; median levels of PFHxS, ΣPFOS and ΣPFOA in this sample were 13.8, 2.1 and 1.7 times higher, respectively, than medians observed in the U.S. sample. Higher concentrations of multiple PFASs were significantly associated with lower odds of detectable IL-1β. Weaker associations were observed with other cytokines. In general, perfluoroalkyl carboxylic acids had inverse associations with TNF-α, whereas the perfluoroalkyl sulfonic acids showed positive associations.

Conclusions:

We observed preliminary evidence of altered inflammatory profiles among adults with elevated serum concentrations of PFASs due to contaminated drinking water. Modifications to inflammatory pathways may be one mechanism by which PFAS exposures produce adverse health effects in humans, but this finding requires verification in longitudinal studies as well as phenotypic anchoring to immune function outcomes.

1. Introduction

Per- and polyfluoroalkyl substances (PFASs) are a class of synthetic chemicals widely used in various commercial products and industrial processes since the 1950s because of their oil-, grease-, water-, stain- and heat-resistant properties (CDC 2017). PFASs are highly persistent in the environment and many can accumulate in the human body, with estimated elimination half-lives in humans as long as two to eight years for some of the longer chain (e.g., ≥6 carbons) PFASs (Bartell et al., 2010; Hu et al., 2016; Li et al., 2018; Olsen et al., 2007; Xu et al., 2020). Because of these properties and their widespread use, many different PFASs have been detected in the blood of most people (>99%) in the United States (U.S.) (CDC ATSDR, 2020).

Experimental studies provide evidence in support of immunotoxic effects of PFAS exposure. In several animal models, perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS) have been reported to alter adaptive and innate immune responses, including inflammation and cytokine production (DeWitt et al., 2012). Exposure to PFOA and PFOS in adult mice suppresses T-cell-dependent immunoglobulin M (IgM) production, one of the most sensitive predictive measures of immune function (DeWitt et al., 2012). The strength of these findings was confirmed by a systematic review of immunotoxicological evidence for PFOA and PFOS by the National Toxicology Program (NTP) and with support from epidemiological findings, the NTP determined that PFOA and PFOS were presumed immune hazards to humans (NTP 2016). The mechanism of action for immunotoxicity remains to be elucidated and results vary by species, strain, sex, and route of exposure. Changes in concentrations of circulating inflammatory cytokines may be one indicator of underlying immunotoxic effects of PFASs in humans. Additionally, disturbances to cytokine signaling may be involved in other adverse health effects of PFAS exposure, such as hepatotoxicity (Bassler et al., 2019).

In humans, there is evidence that PFAS exposure is associated with increased cholesterol levels, changes in liver enzymes, increased risk for kidney and testicular cancer, increased risk of high blood pressure and pre-eclampsia in pregnant women, and lower infant birth weight (ATSDR 2020). Of particular interest, these compounds also have been reported to impair immune system function in humans (ATSDR 2021). Epidemiological findings concerning PFAS immunotoxicity demonstrate suppressed vaccine responses across different populations and mixed results for findings of inappropriate immune stimulation such as allergy, asthma, and autoimmune disease across populations (ATSDR 2017, 2021; Costa et al., 2009; Steenland et al., 2010). However, multiple studies have indicated that exposure to elevated levels of PFASs is associated with a reduced humoral response to childhood immunizations with some evidence for reduced response to the influenza vaccine among adults (Grandjean et al. 2012, 2017; Granum et al., 2013; Kielsen et al., 2016; Looker et al., 2014; Mogensen et al., 2015; Stein et al. 2016a, 2016b). Additionally, in one large community-based study, exposure to PFOA was associated with increased odds of ulcerative colitis, an autoimmune disease (Steenland et al., 2013). In an occupationally exposed cohort, incidence of ulcerative colitis and rheumatoid arthritis was highest among individuals with the highest cumulative occupational and residential exposure to PFOA (Steenland et al., 2015).

We therefore aimed to study the association between PFAS concentrations in serum and circulating inflammatory cytokines in a human population with elevated PFAS exposure from contaminated drinking water. From 2013 to 2016, concentrations of PFOA and PFOS above the U.S. Environmental Protection Agency (EPA) health advisory level (70 ng/L) were detected in municipal water systems downstream of the Peterson Space Force Base, impacting approximately 80,000 people in Fountain Valley, Colorado (Barton et al., 2020). The source of the contamination was believed to be aqueous film-forming foams (AFFF) released from the nearby Space Force base after use in training exercises and in extinguishing fuel fires, with use dating back to the 1970s. Use of AFFF is common in the U.S., with over 533 civilian airports and 290 military bases reporting their use as of 2015 (Hu et al., 2016). Some AFFFs contain high concentrations of perfluorohexane sulfonate (PFHxS) and its precursors, which are less studied in humans compared to PFOA and PFOS. A recent animal study demonstrated that mixtures of AFFF-derived PFASs impair immune function in mice and that mixtures of PFASs may have more adverse effects than exposure to individual PFASs (McDonough et al., 2020b).

Each of the inflammatory biomarkers analyzed in this study have distinct roles in regulating the inflammatory process. While TNF-α, IL-6, and IL-1β are typically pro-inflammatory and IL-10 is generally anti-inflammatory, IL-6 can be either pro- or anti-inflammatory(Chatzantoni and Mouzaki 2006; Iyer and Cheng 2012; Kaneko et al., 2019; Scheller et al., 2011). TNF-α is often elevated during infection and higher circulating concentrations have been linked with insulin resistance and several autoimmune diseases(Chatzantoni and Mouzaki 2006; Nieto-Vazquez et al., 2008). Similarly, elevations in IL-6 and IL-1β are associated with autoimmune diseases, diabetes, and cancer (Kaneko et al., 2019; Robertson and Coveney 2021). IL-10 plays a role in controlling allergies and asthma, and IL-10 deficiency has been associated with inflammatory bowel disease and autoimmune disease (Iyer and Cheng 2012). IL-1β is a key pro-inflammatory mediator and as it is produced by many cell types, it may be a better marker of generalized immunosuppression than other cytokines. These biomarkers are useful in understanding immune system function and linked with autoimmune diseases. The goal of this study is to describe the cross-sectional associations between individual PFASs, and an AFFF-related mixture of PFASs, with circulating pro- and anti-inflammatory biomarkers.

2. Methods

2.1. Study population

The PFAS Assessment of Water and Resident Exposure (PFAS- AWARE) Study population, design and procedures are described in detail elsewhere (Barton et al., 2020). In brief, the PFAS-AWARE study included 220 non-smoking adults who lived in one of three PFAS-impacted water districts (Fountain, Security or Widefield) or whose water was provided by a PFAS-impacted private well in one of these areas (n = 16). In June 2018, all participants provided a 20-mL non-fasting blood sample and completed an exposure and health history questionnaire. For this analysis, seven participants were excluded because they did not live in the PFAS-impacted area for at least two years during the time of known contamination (2012–2015; prior to remediation efforts that began in fall of 2015 in Fountain), and one participant was excluded because they were undergoing dialysis treatment at the time of the blood draw, resulting in a final sample size of 212. For all analyses with TNF-α as the outcome, two individuals were excluded due to use of TNF-α altering medications as reported on the questionnaire. Medications considered to be TNF-α altering included: Remicade (infliximab), Enbrel (etanercept), Humira (adalimumab), Cimzia (certolizumab pegol), and Simponi (golimumab).

The study protocol was approved by the Colorado Multiple Institutional Review Board and all participants provided informed consent prior to study procedures.

2.2. Serum preparation

Venous blood was collected into BD vacutainer red top tubes (Becton Dickinson, New Jersey, USA) and within 30 min of collection, was centrifuged in a 5804R centrifuge (Eppendorf, Hamburg, Germany) for 15 min at 1300×g to separate serum. 500 μl serum aliquots were obtained and placed on dry ice (maximum time on dry ice was 54 h) until they were transported to the laboratory where they were stored at −80 ^◦C until analysis.

2.3. Cytokine analysis

A panel of pro- and anti-inflammatory cytokines were selected based on toxicologic relevance and previously published literature (Bassler et al., 2019; Mitro et al., 2020; Stein et al., 2016a; Zota et al., 2018). Interleukins (IL-1β, IL-2, IL-4, IL-5, IL-6, IL-8, IL-10), interferon gamma (IFNγ), granulocyte-macrophage colony-stimulating factor (GM-CSF) and tumor necrosis factor alpha (TNF-α) were initially measured in duplicate assays using the Cytokine Human Magnetic 10-plex panel for Luminex platform (Life Technologies Corporation, Maryland, USA). Due to low sensitivity of the Life Technologies assay, interleukins (IL-1β, IL-2, IL-6, and IL-10), IFNγ and TNF-α were measured in duplicate assays using human high sensitivity cytokine A Luminex performance assays (R&D, Minnesota, USA) to determine absolute cytokine concentrations.

2.4. Analysis of serum PFASs

Serum samples were analyzed at the Colorado School of Mines; detailed methods and details on quantitation are provided elsewhere (McDonough et al., 2021). In brief, samples were analyzed with online solid-phase extraction (SPE) using a SCIEX Exion high performance liquid chromatography (HPLC) system coupled to an X500R quadrupole-time-of-flight mass spectrometer (QTOF-MS). PFASs were quantified via negative electrospray ionization (ESI-) with SWATH® Data-Independent Acquisition using a suspect screening workflow (McDonough et al., 2020a). Quantitation was done for 48 targeted PFASs by isotope dilution. For PFASs with both branched and linear isomers detected (PFOS and PFHpS), total concentration of all isomers was measured and reported. Novel PFASs identified via suspect screening were measured semi-quantitatively by assigning response factors from structurally-related target compounds. Personnel conducting the laboratory analysis were not provided with any information about characteristics of the participant providing the specimen. Multiple analytical runs were used to analyze the sample sets resulting in variations in detection limits between runs.

2.5. Statistical analysis

All study data were input into and managed using REDCap (Research Electronic Data Capture) tools hosted at the University of Colorado Denver (Harris et al., 2009).

Summary statistics (median, range, percentiles) were computed for seven PFASs (perfluorooctanoic acid [PFOA], perfluorononanoic acid [PFNA], perfluorodecanoic acid [PFDA], perfluorohexanesulfonic acid [PFHxS], perfluoroheptanesulfonic acid [ΣPFHpS], perfluorooctanesulfonic acid [ΣPFOS], unsaturated perfluorooctanesulfonic acid [U-PFOS]) detected in >85% of participants. Distributions and outliers were also visually assessed via histograms and scatterplots, and pairwise correlations between PFASs were described using Spearman’s rank correlation coefficient. All statistical analyses were conducted after substituting ½ the limit of detection (LOD) for PFAS values below the LOD (up to 15% of values) (U.S. Environmental Protection Agency, 2006). The LOD for each sample and analyte was specific to the sample batch.

Of the 10 cytokines measured, only those detected in ≥30% of participants (IL-1β, IL-6, IL-10, TNF-α) were included in this analysis. Cytokines with >15% of values below the LOD were not analyzed as continuous outcomes, but instead classified into two categories: at or above vs below the LOD. Substitution or imputation of greater than 30% of values has been shown to produce biased results with inaccurate measures of variance (Lubin et al., 2004). Pairwise correlations between cytokines were described using Spearman’s rank correlation coefficient.

Associations between each of the seven PFASs and each of the four cytokines were analyzed separately using 1) linear regression (TNF-α only) and 2) logistic regression (IL-1β, IL-6, IL-10) where the outcome was binary (detected vs. not detected). In both linear and logistic regression models, PFASs were natural log-transformed to reduce the influence of outliers and modeled as continuous variables. All models were adjusted for age (continuous), race/ethnicity (binary: non-Hispanic white vs all others), smoking history (binary: former smokers vs never smokers), sex (binary), and continuous BMI (kg/m²) a priori based on previous literature (Mitro et al., 2020; Olden and White 2005; Stein et al., 2016a; Timmermann et al., 2020; Williams et al., 2016; Zota et al., 2018). Further, as an exploratory analysis, we evaluated the potential for effect modification by the following variables by including product interaction terms in adjusted models: sex, race/ethnicity, and BMI (categorized into <25 kg/m² or ≥ 25 kg/m²). The selected variables were evaluated as effector modifiers because of the following previous literature. Zota et al. observed different effects for several endocrine disrupting chemicals (including PFASs) and cytokines among participants with and without obesity (Zota et al., 2018). Endocrine disrupting chemicals have been observed to effect men and women differently in multiple settings including response to immunizations (Timmermann et al., 2020). Race/ethnicity was evaluated as an effect modifier due to current empirical literature suggesting that the lived experience of systemic racism may cause greater susceptibility of the effects of environmental exposures (Olden and White 2005; Williams et al., 2016). Stratified models are presented in the supplement for those outcomes where at least one PFAS-cytokine association had a statistically significant interaction term (p <0.05).

To assess the appropriateness of linear regression, we explored the potential for non-linear associations between PFASs and cytokine outcomes by fitting generalized additive models (GAMs). GAMs were adjusted for the same covariates as the linear and logistic regression models described above (age, race/ethnicity, smoking history and sex). GAM results were then plotted for visual assessment of non-linearity. If the Akaike information criterion (AIC) for the GAM with a spline- transformed exposure term was less than the AIC for the corresponding untransformed model (indicating a better fit), and if visual inspection of plotted results confirmed a non-linear association (i.e., not driven by outliers), then multiple regression was performed with the PFAS exposure divided into quartiles and treated as a categorical predictor.

In addition to single pollutant models detailed above, Bayesian kernel machine regression (BKMR) was performed to assess the combined effect of the seven PFASs on each cytokine outcome (Bobb et al., 2015). Using BKMR allows for visualization of the exposure-response function while accounting for interactions between pollutants (Bobb et al., 2015). All PFAS concentrations were scaled and mean-centered for BKMR analyses and categorical covariates were included as binary variables (Bobb et al., 2015). As previously described, TNF-α was treated as a continuous outcome while IL-1β, IL-6, IL-10 were treated as binary outcomes. BKMR models for all four cytokines were adjusted for age, race/ethnicity, smoking history, BMI, and sex.

All statistical analyses were conducted using the statistical software R Studio Version 1.3.959 (RStudio Team 2020). The significance level for all statistical analyses was set at p <0.05.

3. Results

3.1. Study population demographics and serum PFAS concentrations

The study population included in this analysis consisted of 212 adults that lived in the affected Fountain Valley water districts for at least two years between 2012 and 2015. As described in Table 1, the population was largely female (62%) with a median age of 61 years, 74% non-Hispanic white, largely overweight (44%) or obese (34%), and predominantly never smoking (61%). Slightly more participants were recruited from Security (45%), the water district nearest to the presumed source of contamination.

Table 1.

Select characteristics of study population.

Characteristic	Study Population (N = 212)
Sex
Male	80 (37.7%)
Female	132 (62.3%)
Age (years)
Mean (SD)	59.3 (14.6)
Median [Min, Max]	61 [22, 93]
Race/Ethnicity a
Non-Hispanic White	157 (74.1%)
Other	55 (25.9%)
BMI b
<25 kg/m2	48 (22.6%)
25 kg/m2 ≤ BMI <30 kg/m2	93 (43.9%)
≥30 kg/m2	71 (33.5%)
Smoking Status
Never Smoker	130 (61.3%)
Former Smoker	82 (38.7%)
Water District (2012–2015) c
Fountain	52 (24.5%)
Security	95 (44.8%)
Widefield	65 (30.7%)

Abbreviations: N, number of participants; SD, standard deviation; Min, minimum; Max, maximum; BMI, body mass index.

^aBMI calculated from self-reported height and weight.

^bThe “Other” category includes individuals who identified as the following: Asian, Black or African, American Indian or Alaska Native, Other, and/or Hispanic. A binary grouping was made to avoid small cell sizes in these categories.

^cIndividuals on private wells were assigned a water district based on well location.

Summary statistics for the seven PFASs included in this analysis are presented in Table 2. PFOA, PFNA, PFHxS, ΣPFHpS, and ΣPFOS were detected in ≥98% of samples while U-PFOS and PFDA were detected in 91% and 86% of samples, respectively. All frequently detected PFASs were targeted perfluoroalkyl acids (PFAAs) except for U-PFOS, a novel unsaturated analog of PFOS which was tentatively identified by high-resolution mass spectrometry. Concentrations of U-PFOS may represent total concentrations of multiple positional isomers. Identification and semi-quantitation of this novel PFAS are described in detail in McDonough et al., (2021).

Table 2.

Concentrations (ng/mL) of PFASs included in these analyses, among 212 participants in the PFAS-AWARE study.

PFAS	% Detected	DLa	Min	25th %	Median	75th %	Max	Median U.S.b
PFOA	100	0.01–0.1	0.18	2.2	3.3	5.6	17.2	1.57
PFNA	98	0.01–0.2	<DL	0.3	0.4	0.6	6.6	0.60
PFDA	86	0.01–0.2	<DL	0.07	0.1	0.2	1.9	0.10
PFHxS	100	0.11–1.0	0.9	7.7	16.6	33.7	164.4	1.2
ΣPFHpS	99	0.01–0.04	<DL	0.4	0.9	1.6	10.6	N/A
ΣPFOS	100	0.1–2.0	1.2	4.8	8.2	14.0	50.2	4.8
U-PFOS c	91	0.01–0.2	<DL	0.2	0.3	0.4	1.9	N/A

Abbreviations: PFASs, per and polyfluoroalkyl substances; N, number of participants; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexansulfonic acid; PFHpS, perfluoroheptanesulfonic acid; PFOS, perfluorooctanesulfonic acid; U-PFOS, unsaturated perfluorooctanesulfonic acid; Min, minimum; Max, maximum; DL, detection limit; ng/mL, nanograms per milliliter.

^aMultiple analytical runs were used to analyze sample sets, causing some run-to-run variation in detection limits. The range of detection limits for each compound is provided.

^bData from the 2015–2016 cycle of the US National Health and Nutrition Examination Survey (NHANES), https://www.cdc.gov/exposurereport/

^cValue is semi-quantitative.

Because of variability in the detection limits between multiple analytical runs, ranges of detection limits are presented. This PFHxS, ΣPFHpS and ΣPFOS dominated PFAS mixture is characteristic of exposure to AFFF contamination (Barton et al., 2020). Median concentrations (Table 2) observed in this study population were two times median levels measured through the U.S. National Health and Nutritional Examination Survey (NHANES) for PFOA and PFOS, and nearly 14 times the U.S. national median for PFHxS (U.S. Department of Health and Human Services Centers for Disease Control and Prevention, 2021).

Spearman’s rank correlation analysis (r_s, Supplemental Table 1) shows that all PFASs were moderately to strongly correlated with one another (range r_s 0.22–0.93). The most strongly correlated PFASs were ΣPFHpS with U-PFOS (r_s =0.93), ΣPFHpS with PFHxS (r_s = 0.88), PFHxS with PFOA (r_s = 0.85), and PFHxS with U-PFOS (r_s = 0.85). While still statistically significant, the weakest correlations observed were PFNA with PFHxS (r_s = 0.26), and PFDA with PFHxS (r_s = 0.22).

3.2. Inflammatory biomarkers

IL-1β, IL-2, IL-6, IL-10, IFNγ, and TNF-α were analyzed in all study participants, but only TNF-α, IL-1β, IL-6, and IL-10 were detected in ≥30% of participants. To assess the quality of results in each batch we used the coefficient of variation (CV), a measure of variance between replicate data points within an assay, and found CVs to be generally low (i.e., <10%) (Supplemental Table 2). Summary statistics are presented in Table 3, including the ranges of detection limits which varied across multiple runs. TNF-α was the only highly detected and normally distributed cytokine in this study population. Due to relatively low frequencies of detection (30–85%), we categorized concentrations of IL- 1β, IL-6, and IL-10 as above versus below the limit of detection. Spearman’s rank correlation analysis (r_s, Supplemental Table 3) shows that all cytokines were weakly correlated with one another (range r_s −0.15 –0.43). The most strongly correlated cytokines were IL-10 with IL-6 (r_s = 0.43) while the weakest correlation was between IL-1β with IL-6 (r_s = 0.02).

Table 3.

Concentrations (pg/mL) of cytokines included in these analyses, among 212 participants in the PFAS-AWARE study.

Cytokine	% Detected	DLa pg/mL	Min pg/mL	5th % pg/mL	25th % pg/mL	Median pg/mL	75th % pg/mL	95th % pg/mL	Max pg/mL
TNF-α	100	0.3–1.1	2.0	4.0	6.2	7.9	10.0	13.4	23.4
IL-1β	31	0.1–0.4	<DL	<DL	<DL	<DL	0.2	0.4	0.8
IL-6	40	0.5–1.3	<DL	<DL	<DL	<DL	1.3	3.6	32.2
IL-10	35	0.1–0.5	<DL	<DL	<DL	<DL	0.2	0.6	1.6

Abbreviations: N, number of participants; Min, minimum; Max, maximum; DL, detection limit; TNF, tumor necrosis factor; IL, interleukin; IFN, interferon; pg/mL, picograms per milliliter.

^aMultiple analytical runs were used to analyze sample sets, causing some run-to-run variation in detection limits. The range of detection limits for each compound is provided.

3.2.1. TNF-α

As indicated in Table 4 and Fig. 1, while not statistically significant, all of the analyzed perfluoroalkyl carboxylic acids (PFCAs), including PFOA, PFNA, and PFDA, were inversely associated with TNF-α, whereas the perfluoroalkyl sulfonic acids (PFSAs), including PFHxS, ΣPFHpS, ΣPFOS, and U-PFOS, were positively associated with TNF-α. As a mixture, higher concentrations of all seven PFASs combined were associated with slightly higher TNF-α compared to the scenario where all PFASs were at their median value, although prediction intervals were wide and included the null value (Fig. 2). In other words, with increasing quantile of exposure, we observe that the effect of the seven PFAS mixture on TNF-α concentration is positive. This suggests that the effect of the mixture is heavily influenced by the sulfonates which also have a positive direction of association with TNF-α in the single pollutant models (Fig. 1).

Table 4.

Associations between individual PFASs and TNF-α serum concentrations, estimates from single pollutant multiple linear regression models^a.

PFAS	N = 210b
	Change in TNF-a per ln-unit increase in PFAS	95% CI	P-value
PFOA	−0.14	−0.73, 0.44	0.63
PFNA	−0.14	−0.81, 0.52	0.68
PFDA	−0.12	−0.47, 0.23	0.50
PFHxS	0.22	−0.21, 0.65	0.32
ΣPFHpS	0.08	−0.32, 0.49	0.68
ΣPFOS	0.15	−0.51, 0.81	0.65
U-PFOSc	0.25	−0.28, 0.78	0.35

Abbreviations: PFASs, per and polyfluoroalkyl substances; TNF-α, tumor necrosis factor alpha; N, number of participants; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexansulfonic acid; PFHpS, perfluoroheptanesulfonic acid; PFOS, perfluorooctanesulfonic acid; U-PFOS, unsaturated perfluorooctanesulfonic acid; CI, confidence interval; BMI, body mass index.

^aModels adjusted for age, BMI, sex, smoking history and race/ethnicity.

^bExcludes two individuals who reported taking TNF-α altering medication.

^cValue is semi-quantitative.

Fig. 1.

Associations between individual PFASs and TNF-α serum concentrations, estimates from single pollutant multiple linear regression models. Models adjusted for age, sex, BMI, smoking history and race/ethnicity. Diamonds represent beta coefficients and bars represent 95% confidence intervals.

Fig. 2.

Combined effect of seven PFASs on TNF-α estimated using Bayesian kernel machine regression, comparing the value of the exposure-response function when all PFASs are at a given quantile as compared to when all PFASs are at their median value. Models are adjusted for age, sex, BMI, smoking history and race/ethnicity.

In an exploratory analysis, we evaluated the potential for effect modification by sex, race/ethnicity, and BMI categories, by including a product interaction term with each PFAS and each potential modifier in fully adjusted models. Due to a small sample size, all results of interaction models should be interpreted with caution. We observed some significant interaction terms for PFAS-by-BMI category interaction terms, therefore we present results stratified by BMI category (<25 kg/m² or ≥ 25 kg/m²) in Supplemental Table 4. PFHxS was significantly positively associated with TNF-α among participants with BMI <25 kg/m² but not among participants with BMI ≥25 kg/m². A similar, but non-significant, trend was observed for PFHpS.

3.2.2. IL-1β

As indicated in panel 1 of Table 5 and Fig. 3, odds of detectable IL-1β were significantly reduced with a 1-ln-unit increase in PFOA (OR = 0.64 [95%CI: 0.41, 0.98]), PFDA (OR = 0.78 [95%CI: 0.91, 1.00]), and ΣPFHpS (OR = 0.70 [95%CI: 0.52, 0.94]). Consistent direction of association was noted in the models for PFNA, PFHxS, ΣPFOS and U-PFOS, though confidence intervals included the null. Assessing all seven PFASs as a mixture, there was a clear inverse linear pattern of association between the PFAS mixture and the odds of detectable IL-1β (Fig. 4).

Table 5.

Associations between individual PFAS serum concentrations (ln-ng/mL) and the odds of detectable concentrations of inflammatory cytokines. Estimates from single pollutant multiple logistic regression models^a, among 212 participants in the PFAS-AWARE study.

PFAS	IL-1β			IL-6			IL-10
	Odds Ratio	95% CI	P-Value	Odds Ratio	95% CI	P-Value	Odds Ratio	95% CI	P-Value
PFOA	0.64	0.41, 0.98	0.05	1.09	0.72, 1.69	0.68	0.81	0.53, 1.22	0.31
PFNA	0.69	0.41, 1.13	0.14	1.36	0.84, 2.27	0.22	0.78	0.48, 1.25	0.31
PFDA	0.78	0.61, 1.00	0.05	1.01	0.79, 1.31	0.93	0.97	0.76, 1.25	0.81
PFHxS	0.78	0.56, 1.07	0.12	0.99	0.73, 1.36	0.96	0.93	0.68, 1.27	0.65
ΣPFHpS	0.70	0.51, 0.94	0.02	1.19	0.88, 1.64	0.27	0.87	0.65, 1.15	0.33
ΣPFOS	0.78	0.48, 1.25	0.29	1.33	0.82, 2.16	0.25	0.79	0.49, 1.26	0.32
U–PFOSb	0.73	0.49, 1.07	0.11	1.07	0.73, 1.62	0.73	0.99	0.68, 1.47	0.97

Abbreviations: PFASs, per and polyfluoroalkyl substances; IL, interleukin; N, number of participants; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexansulfonic acid; PFHpS, perfluoroheptanesulfonic acid; PFOS, perfluorooctanesulfonic acid; U-PFOS, unsaturated perfluorooctanesulfonic acid; CI, confidence interval; BMI, body mass index.

^aModels adjusted for age, BMI, sex, smoking history and race/ethnicity.

^bValue is semi-quantitative.

Fig. 3.

Associations between individual PFAS serum concentrations (ln-ng/mL) and the odds of detectable concentrations of inflammatory cytokines. Estimates from single pollutant multiple linear regression models, among 212 participants in the PFAS-AWARE study. ^aModels adjusted for age, sex, BMI, smoking history and race/ethnicity.

Fig. 4.

Bayesian kernel machine regression results: combined effect of all seven PFASs on odds of cytokine detection, comparing the value of the exposure-response function when all PFASs are at a given quantile as compared to when all PFASs are at their median value. Models are adjusted for age, sex, BMI, smoking history and race/ethnicity, N =212.

In an exploratory analysis, there was some evidence for potential effect modification by race/ethnicity for the associations between PFOA and U-PFOS and the odds of detectable IL-1β (Supplemental Table 5). Odds of detecting IL-1β was inversely associated with PFOA and U-PFOS among those of race/ethnicity other than non-Hispanic white. Additionally, there was some evidence of potential effect modification by sex (Supplemental Table 6). Odds of detecting IL-1β were inversely associated with concentrations of PFOA, PFHxS, ΣPFHpS, and U-PFOS among male participants only.

3.2.3. IL-6

As indicated in panel 2 of Table 5 and Fig. 3, odds of detecting IL-6 were positively associated with certain PFAS concentrations but none of the findings were statistically significant. Assessing all seven PFASs as a mixture, odds of detecting IL-6 were positively associated with increasing exposure to the PFAS mixture (Fig. 4). In exploratory results, stratified by BMI category, positive associations between PFASs and odds of detecting IL-6 were observed among participants with BMI <25 kg/m² but not among participants with BMI ≥25 kg/m² (Supplemental Table 7).

3.2.4. IL-10

As indicated in panel 3 of Table 5 and Fig. 3, odds of detecting IL-10 were weakly inversely associated with measured PFAS concentrations but confidence intervals were wide and included the null. When all seven PFASs were assessed as a mixture, there appeared to be a weakly inverse association with detectable IL-10 (Fig. 4). In exploratory results, stratified by BMI category, PFASs were positively associated with detectable IL-10 among healthy weight participants but inversely associated with detected PFASs among overweight participants, though confidence intervals were imprecise and included the null (Supplemental Table 8).

Because potentially non-linear associations between PFHxS and ΣPFOS and detection of IL-10 were observed in the GAMs, we conducted sensitivity analyses with these PFASs modeled as quartiles (Supplemental Table 9). Odds of detected IL-10 were highest in the third quartile of measured PFHxS concentrations compared to the first, with OR = 3.02 (95% CI: 1.26, 7.51). Odds of detecting IL-10 were highest in the second quartiles of measured ΣPFOS concentrations compared to the first but none of the exposure groups were significantly different from the lowest exposure group.

4. Discussion

In this study of adults highly exposed to PFASs via AFFF-contaminated drinking water, we found associations between PFASs and multiple circulating biomarkers of inflammation. Higher concentrations of PFOA, PFDA, and ΣPFHpS were significantly associated with lower odds of detectable IL-1β. We observed different patterns of association for each cytokine analyzed. Although confidence intervals were wide and included the null, PFAS concentrations were associated with lower odds of detectable IL-10, and with higher odds of detectable IL-6. In general, perfluoroalkyl carboxylic acids (PFOA, PFNA, and PFDA) had weakly inverse associations with TNF-α, while the perfluoroalkyl sulfonic acids (PFHxS, ΣPFHpS, ΣPFOS, and U-PFOS) showed positive associations.

While some of the existing human and animal data are conflicting regarding the directionality of effects of PFAS exposure on a range of immune endpoints, the weight of evidence is that exposure to well-studied perfluoroalkyl acids (PFAAs) (e.g., PFOA and PFOS) is immunosuppressive; both adaptive and innate immune responses are reduced (DeWitt et al., 2012). For example, several studies have demonstrated that immune responses to immunizations are reduced in populations that are highly exposed to PFASs (Grandjean et al. 2012, 2017; Granum et al., 2013; Kielsen et al., 2016; Mogensen et al., 2015; Stein et al., 2016b). Suppression of specific cytokines (in this case, IL-1β) could reflect an overall suppressive effect of PFAS exposure.

Autoimmune diseases have been previously linked with PFAS exposure in human studies. The largest cohort of PFOA-exposed individuals studied to date, the C8 Study, reported a probable link between PFOA and diagnosed high cholesterol, chronic kidney disease, ulcerative colitis, thyroid disease, testicular and kidney cancer, and pregnancy-induced hypertension and preeclampsia (C8 Science Panel, 2012). In the same population, other autoimmune outcomes including rheumatoid arthritis, Crohn’s disease, type I diabetes, lupus, and multiple sclerosis were also evaluated, but were not found to be significantly more likely in more highly exposed participants (Steenland et al., 2013). In the subset of the cohort with occupational exposure to PFOA, both incident ulcerative colitis and rheumatoid arthritis were associated with PFOA exposure (Steenland et al., 2015). While the present study was too small to evaluate autoimmune disease outcomes, altered patterns of cytokine production may be involved in the pathogenesis of some of these diseases.

Few previous studies of PFAS exposure have evaluated circulating inflammatory biomarkers as outcomes. In a random sample of 200 participants from the C8 Study, several pro-inflammatory cytokines (including TNF-α, IL-6, IL-8, and IFN-γ) were measured along with biomarkers of liver disease, under the hypothesis that cytokine dysregulation was linked to liver injury (Bassler et al., 2019). The results supported this hypothesis, demonstrating that hepatocyte apoptosis was increased in individuals with higher PFAS serum concentrations, while TNF-α and IL-8 were downregulated. The authors concluded that hepatocyte apoptosis was mechanistically linked to decreased TNF-α (Bassler et al., 2019). Similarly, we observed weakly inverse associations between PFCAs and TNF-α, although confidence intervals were imprecise and included the null. IL-8 was not measured in our study and IFN-γ was only detected in two participants. A recent study in an elderly population identified decreased proteomic inflammatory markers associated with exposure to PFASs, another novel approach to measuring inflammation (Salihovic et al., 2020).

While this cohort did not include pregnant women, recent studies have demonstrated associations between PFASs and inflammation in pregnancy. In a cohort of women with overweight or obesity from the San Francisco Bay area, a doubling of ΣPFAS concentrations was associated with a 20.87% (3.46, 41.22) increase in IL-6 during pregnancy, with inconsistent and non-significant results for IL-10 and TNF-α (Zota et al., 2018). Similarly, in the Project Viva cohort study, pregnancy plasma concentrations of 2-(N-ethyl-perfluorooctane sulfonamido) acetic acid (EtFOSAA) and 2-(N-methyl-perfluorooctane sulfonamido) acetic acid (MeFOSAA) were associated with greater 3-year postpartum IL-6 (Mitro et al., 2020). In this study, IL-6 was 10.8% higher [95%CI: 3.3, 18.9] and 14.5% higher [95%CI: 5.7, 24.1] per doubling in EtFOSAA and MeFOSAA, respectively. The authors hypothesized that IL-6 may be upregulated in response to higher oxidative stress after PFAS exposure (Mitro et al., 2020).

Evidence from animal studies and cell culture models may be helpful in understanding the mechanism behind this observed association in a more controlled setting. For example, exposure to PFOA increased production of TNF- α, IL-1β, IL-6, and IL-8 in IgE-stimulated mast cells (RBL-2h3 cells) in culture (Lee et al., 2017). These results support the assumption that PFOA exacerbated mast cell-derived allergic inflammation via activation of NF-kB. In stimulated mast cells, PFDA and PFUnA exposure significantly increased IL-1β whereas none of the PFCAs (PFHpA, PFNA, PFDA, PFUnA) significantly increased IL-1β expression in unstimulated cells (Lee and Kim 2018). In vitro, in human bronchial epithelial cells, both PFOA and PFOS stimulated the release of IL-1β in virus stimulated cells (no effect observed in non-stimulated cells) at human relevant concentrations. Other PFASs did not induce cytokine release at the concentrations tested (Sorli et al., 2020). These findings reinforce that cytokine function and dysregulation is highly context dependent, and our results provide only a snapshot of a complex process of regulating inflammatory signaling.

Results from animal studies to date have been inconclusive regarding the influence of PFAS exposure on IL-1β. In mice exposed to PFOA via drinking water, IL-1β was significantly increased in the spleen but decreased with dose in the thymus (Son et al., 2009). TNF-α and IL-6 were significantly increased in the spleen at the highest administered dose but did not differ across exposure groups in the thymus (Son et al., 2009). These findings highlight that tissue-specific levels of cytokines can vary, making the interpretation of levels of cytokines in serum more challenging to interpret. By contrast, in zebrafish exposed to lower levels of PFOA, IL-1β expression in the spleen was significantly reduced at concentrations≥0.01 mg/L. This study indicates that zebrafish may be more vulnerable than mice to the effects of PFOA exposure on cytokine production (Zhang et al., 2014). In another study of zebrafish, production of IL-1β after PFOA exposure was both time and dose dependent (Zhang et al., 2021). It is possible that the cytokine response to PFAS exposure in humans also depends on the dose and the duration of exposure, and on the specific mixture of PFASs received.

Our study has several limitations. This was a pilot study to establish blood concentrations of AFFF-related PFASs in a population with contaminated drinking water and may not be powered to detect associations of small magnitude or be generalizable to other, non-AFFF-exposed populations. As with any cross-sectional study, there is a potential for reverse causation. Additionally, we measured cytokines at a single time point, although concentrations in blood may vary substantially from day to day, introducing some degree of outcome measurement error. Both cytokines and serum PFAS concentrations can vary based on dietary intake (Aziz 2015; Roth et al., 2020; Seshasayee et al., 2021; van Bussel et al., 2011). The blood samples were non-fasting and cytokine concentrations may have been influenced by recent food intake (Aziz 2015; van Bussel et al., 2011), however we did not collect these data. While certain foods may be sources of PFAS exposure, we do not expect the fasting status of participants impacted serum PFAS concentrations due to their persistent nature (Sunderland et al., 2019). Further, while covariates were selected based on published literature, measures of socioeconomic status and typical food consumption were not collected and could result in residual confounding. We performed numerous statistical tests and therefore it is possible that some results reached statistical significance by chance.

Despite these limitations, measuring cytokines as a biomarker of inflammation is a minimally invasive and cost effective first step in evaluating the potential immunotoxic effects of exposure to AFFF-associated PFAS mixtures in humans. Strengths of our study include quantification of a large panel of PFASs and consideration of the combined effects of multiple PFASs in the exposure mixture using advanced statistical methods. Previous studies of cytokines have not been conducted in populations with AFFF exposure, so this study provides novel data on the potential immune-modifying effects of this exposure profile, with PFHxS as the predominant substance measured.

5. Conclusion

Modifications of inflammatory pathways may be one mechanism by which PFAS exposures produce adverse health effects in humans. We observed preliminary evidence of altered inflammatory profiles among adults with elevated serum concentrations of several AFFF-associated PFASs due to contaminated drinking water. Further studies are needed to examine whether altered cytokine profiles associated with this exposure mixture may be linked to increased risk of autoimmune disease or other evidence of immunotoxicity.