The Association between Multiple Per- and Polyfluoroalkyl Substances’ Serum Levels and Allostatic Load

Background/Objective: This study aimed to explore the association between allostatic load (AL), an index of chronic stress, with nine per- and polyfluoroalkyl substances (PFASs), a group of organic compounds used in commercial and industrial applications. The PFASs explored were perfluorohexane sulfonic acid (PFHS), perfluorodecanoic acid (PFDE), perfluorobutane sulfonic acid (PFBS), perfluoroheptanoic acid (PFHP), perflurododecanoic acid (PFDO), perfluorononanoic acid (PFNA), perfluoroundecanoic acid (PFUA), perfluorooctanoic acid (PFOA), and perfluorooctane sulfonic acid (PFOS). This study was performed to better understand the association between PFASs and AL, which may be a mediator of several diseases. Methods: This study was performed on adults aged 20 and older, using the National Health and Nutrition Examination Survey (NHANES) 2007–2014 data. AL was calculated as a cumulative index of ten biomarkers from the cardiovascular, inflammatory, and metabolic system, which was dichotomized into high risk (assigned a value of 1) or low risk (assigned a value 0) depending on if the index value was ≥3 (chronic physiological stress) or <3 (less stressed). In this study, PFASs and covariates such as age, gender, ethnicity, alcohol consumption, smoking, and physical activity were explored using descriptive statistics and logistic regression modeling. Results: The results indicated that in adults, AL was more elevated in men as compared to women, in those aged ≥60 years, and varied by ethnicity. For instance, non-Hispanic Blacks had higher AL levels (mean of 3.92) compared to other ethnicities. A significant number of the participants tested for PFBS, PFHP, PFDO were below the LOD and thus these PFASs were excluded from the analysis. Our analysis demonstrated multicollinearities between variables such as PFNA, PFOS, and PFOA with variance inflation factor (VIF) values of 6.197, 6.212, and 5.139, respectively. Thus, PFASs were analyzed individually and adjusted for age, gender, ethnicity, physical activity, smoking, and alcohol consumption. The results indicated a statistically significant positive association between AL and most of the PFASs, except PFUA which was not statistically significant with a p value of 0.531. Conclusions: The findings of this study suggest that exposure to PFDE, PFNA, PFOS, PFOA, and PFHS are associated with AL when adjusted for age, gender, ethnicity, alcohol consumption, smoking, and physical activity. Future studies looking to model the effects of these factors together must consider their relationship with each other and choose different analytical approaches.


Introduction
Per-and polyfluoroalkyl substances (PFASs) are persistent organic pollutants that have been widely used in consumer products for more than seven decades. Because of PFASs unique thermal stability and surface activity properties, such as hydro-and lipophobicity, they have been used in commercial and industrial applications [1]. For example, PFASs have been utilized in food packaging, firefighting foams, and coatings

Description of Cohort
This study analyzed de-identified demographic and biomarker data collected for National Health and Nutrition Examination Surveys (NHANES), which is administe by the Centers for Disease Control and Prevention (CDC). NHANES uses a complex, m tistage probability design that samples the civilian noninstitutionalized population re ing in all 50 states within the United States and the District of Columbia (DC). The tar population for this study was adults aged 20 and older.

Blood Serum Measurements
NHANES collected biological specimens for laboratory analysis to provide deta information about participants' health and nutritional status at the mobile examina center (MEC). Sera were stored in polypropylene and polyethylene containers. The bl sample collection depended on the age of the participant and collected a minimum of mL of serum, preferably. Blood was processed, then refrigerated or frozen before be shipped to laboratories throughout the United States. The controlled environment of MEC permitted laboratory measurements to be completed under similar condition each survey location.
Participants in this study appointed to a morning session were asked to fast for After the primary blood draw, they were asked to consume 75 g of dextrose (10 o glucose solution) within ten minutes. A second blood sample was collected two ho later [16,17].

PFASs Extraction and Quantitation
The extraction of PFASs was performed. After dilution with formic acid, one aliq of 50 μL of serum was injected into a commercial column switching system allowing concentration of the analytes (PFASs) on a solid-phase extraction column. The NHAN followed the recommended phlebotomy practices for the collection of blood and sep tion of blood serum. The PFAS analysis was completed on participants' blood sera w completed and passed the questionnaires to screen for conditions such as hemoph

Description of Cohort
This study analyzed de-identified demographic and biomarker data collected for the National Health and Nutrition Examination Surveys (NHANES), which is administered by the Centers for Disease Control and Prevention (CDC). NHANES uses a complex, multistage probability design that samples the civilian noninstitutionalized population residing in all 50 states within the United States and the District of Columbia (DC). The target population for this study was adults aged 20 and older.

Blood Serum Measurements
NHANES collected biological specimens for laboratory analysis to provide detailed information about participants' health and nutritional status at the mobile examination center (MEC). Sera were stored in polypropylene and polyethylene containers. The blood sample collection depended on the age of the participant and collected a minimum of 0.5 mL of serum, preferably. Blood was processed, then refrigerated or frozen before being shipped to laboratories throughout the United States. The controlled environment of the MEC permitted laboratory measurements to be completed under similar conditions at each survey location.
Participants in this study appointed to a morning session were asked to fast for 9 h. After the primary blood draw, they were asked to consume 75 g of dextrose (10 oz of glucose solution) within ten minutes. A second blood sample was collected two hours later [16,17].

PFASs Extraction and Quantitation
The extraction of PFASs was performed. After dilution with formic acid, one aliquot of 50 µL of serum was injected into a commercial column switching system allowing for concentration of the analytes (PFASs) on a solid-phase extraction column. The NHANES followed the recommended phlebotomy practices for the collection of blood and separation of blood serum. The PFAS analysis was completed on participants' blood sera who completed and passed the questionnaires to screen for conditions such as hemophilia, having been administered chemotherapy in the previous four weeks, and other reasons that would preclude using the participant's arms for a blood draw [18,19].
Partitioning of the analytes from each other and other serum components was performed with high-performance liquid chromatography. Detection and quantification were done using a negative-ion Turbo Ion Spray (TIS) ionization source, which is a variant of the electrospray ionization source that is used to convert liquid phase ions into gas-phase ions incorporated into tandem mass spectrometry. This technique allows for rapid detection of these PFASs in human serum with limits of detection (LOD) in the low parts per billion range [16,20,21].

PFASs Detection Limits
The detection limits were constant for all the PFASs analytes in the data set (0.10 ng/mL). In the NHANES, two variables' names were provided for each of these analytes. The value "0" meant that the result was at or above the limit of detection, "1" indicated that the result was below the limit of detection. For analytes with analytic results below the lower limit of detection, an imputed fill value was placed in the analyte results field. This value is the lower limit of detection divided by the square root of 2 (LLOD/sqrt(2)), which is 0.10/ √ 2 = 0.07. As such, the LOD for each PFAS was either 0.10 or 0.07 [16,20,21]. A significant number of the participants tested for PFBS, PFHP, PFDO in our sample were below the LOD and thus these PFASs were excluded from the analysis.

Operationalizing Allostatic Load
Informed by prior studies [14], AL was operationalized by developing a cumulative index of physiologic dysfunction of the cardiovascular (systolic blood pressure-SBP, diastolic blood pressure-DBP, triglycerides, high-density lipoprotein (HDL) cholesterol, and total cholesterol), inflammatory (C-reactive protein-CRP), and the metabolic systems (body mass index-BMI, hemoglobin A1C, albumin, and creatinine clearance). Based on the distribution of AL markers in the dataset, the markers of interest were split into quartiles with the top quarter of their distribution considered to be high-risk for all markers apart from albumin, creatinine clearance, and HDL cholesterol, for which the bottom 25% of the distribution were considered to have the highest risk as determined by the literature [22][23][24][25][26][27][28]. Within the data, each participant in the study was assigned a value of 1 for those considered to be in the high-risk category or a value of 0 if for those in the low-risk category for all markers to add up to a total AL value out of 10. We dichotomized allostatic load scores as high if AL was ≥3 and low if AL was <3 [29,30].

Statistical Analysis
The dependent variable for this study was AL as a binary outcome (1 for high risk and 0 for low risk). The independent variables were the following PFASs: PFHS, PFDE, PFBS, PFHP, PFDO, PFNA, PFUA, PFOA, and PFOS. We performed Pearson's correlation coefficient analysis and the variance inflation factor (VIF) test to check for multicollinearity. Due to the presence of multicollinearity, PFDE, PFNA, PFOS, PFUA, PFOA, PFHS were examined individually for their relationship with AL.
Descriptive statistics summarized the data within this study. In addition, logistic regression models were used to examine the associations between AL and each PFAS. Sociodemographic variables such as gender, age, and ethnicity were explored to see their distribution by PFASs and AL, how they affect the association between PFASs and AL, and subsequent adjustment for them in logistic regression modelling. Furthermore, behavioral variables including alcohol consumption, cigarette or tobacco use, and physical activity were also investigated to see their distribution in terms of PFASs and AL, how they affect the association between PFASs and AL, and subsequent adjustment for them in logistic regression modelling. These variables were adjusted for due to the literature indicating their relationship with AL and or PFASs [9,[31][32][33].
To examine the associations between the response variable and the predictor variables, the Wald chi-squared t-test was used for categorical variables and the Wilcoxon t-test for numerical variables. In our analysis, to check the assumptions that PFASs variables were normally distributed, we performed a Kolmogorov-Smirnov Test (KS), which found that the PFASs were not normally distributed. Thus, the PFASs concentrations were logtransformed to satisfy normality assumptions. Considering the nonlinear exposure pattern of PFASs, the logistic regression analysis techniques used in this study offer better insight into the relationship between AL and PFASs as compared to studies that used linear analytic techniques.
All analyses were conducted using R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria) with the RStudio platform version 9.1.372.

Results
The total number of participants for this analysis was 23,482, with 51.88% being female. The mean age was 49.62 years. The ethnic composition of the study was as follows: 14.8 percent Hispanic Mexican American (n = 3480), 10.2 percent other Hispanic (n = 2384), 43.7 percent non-Hispanic White (n = 10,250), 21.2 percent non-Hispanic Black (n = 4981), and approximately 10.2 percent for other non-Hispanic races, including Asian and multiracial) (n = 2387). PFOA was differentially distributed by age, gender, and race/ethnicity ( Figure 2). Prior studies using NHANES data have indicated the PFASs in this study are differentially distributed by race/ethnicity and gender [34,35]. To examine the associations between the response variable and the predictor variables, the Wald chi-squared t-test was used for categorical variables and the Wilcoxon ttest for numerical variables. In our analysis, to check the assumptions that PFASs variables were normally distributed, we performed a Kolmogorov-Smirnov Test (KS), which found that the PFASs were not normally distributed. Thus, the PFASs concentrations were logtransformed to satisfy normality assumptions. Considering the nonlinear exposure pattern of PFASs, the logistic regression analysis techniques used in this study offer better insight into the relationship between AL and PFASs as compared to studies that used linear analytic techniques.
All analyses were conducted using R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria) with the RStudio platform version 9.1.372.

Results
The total number of participants for this analysis was 23,482, with 51.88% being female. The mean age was 49.62 years. The ethnic composition of the study was as follows: 14.8 percent Hispanic Mexican American (n = 3480), 10.2 percent other Hispanic (n = 2384), 43.7 percent non-Hispanic White (n = 10250), 21.2 percent non-Hispanic Black (n = 4981), and approximately 10.2 percent for other non-Hispanic races, including Asian and multiracial) (n = 2387). PFOA was differentially distributed by age, gender, and race/ethnicity ( Figure 2). Prior studies using NHANES data have indicated the PFASs in this study are differentially distributed by race/ethnicity and gender [34,35].  Table 1 summarizes the ethnic composition of the study participants by gender. The composition was: Mexican American (49.3 percent males), other Hispanic (44.5 percent males), non-Hispanic White (49.3 percent were males), non-Hispanic Black (48.6 percent males), and non-Hispanic including multiracial and Asian (48.9 percent males). Table 2 summarize the variables by mean levels and standard deviation (SD) of the selected PFAS of interest and behavioral health variables (smoking, alcohol consumption, and physical activity).   , and non-Hispanic including multiracial and Asian (48.9 percent males). Table 2 summarize the variables by mean levels and standard deviation (SD) of the selected PFAS of interest and behavioral health variables (smoking, alcohol consumption, and physical activity).   Table 3 show the mean AL score by age category and ethnicity of participants in the study. AL was elevated for non-Hispanic Blacks in every age category. These variables include AL high ≥ 3 vs. low < 3 AL. AL levels increased with age. Older people had higher AL levels than other age groups as follows, non-Hispanic Black with a mean of 3.92, followed by other Hispanic ethnicity with a mean of 3.64, Mexican (mean 3.58), and other Hispanic with a mean of 3.47. Table 3. AL's score (≥, high and <3, low) and percentage by ethnicity and age group.

Age Group
Mean Score Allostatic Load   Figure 2 presents the levels of PFOA for participants of different ethnicities and by gender and demonstrates that males have higher PFOA concentrations than females. Figure 3 reveals that PFOA is more elevated in those aged sixty and older for both females and males.   We ran separate models for each PFAS (unadjusted models) and used three different methods to check for multicollinearity: Table 4 presents the Pearson correlation coefficients of the relationship between the selected PFAS with each other. PFAS were generally correlated with each other. A correlation of 0.4 and up presents concerns of multicollinearity. The analysis revealed that the PFAS have a correlation from moderate to strong which indicates the potential for multicollinearity. We ran separate models for each PFAS (unadjusted models) and used three different methods to check for multicollinearity:  Table 4 presents the Pearson correlation coefficients of the relationship between the selected PFAS with each other. PFAS were generally correlated with each other. A correlation of 0.4 and up presents concerns of multicollinearity. The analysis revealed that the PFAS have a correlation from moderate to strong which indicates the potential for multicollinearity.     All three methods demonstrated multicollinearities between variables such as PFNA, PFOS, and PFOA with VIF values of 6.197, 6.212, and 5.139, respectively (Table 5). A VIF greater than five indicates a high/strong or potentially severe correlation, and thus the existence of multicollinearity.

(b) Variance Inflation Factor (VIF) test (Table 5)
As such, AL and each individual PFAS, adjusted for age, gender, ethnicity, physical activity, smoking, and alcohol consumption was used in the analysis. Table 6 shows the association between PFASs and AL in the adjusted models. Each PFAS was entered into its own model and adjusted for the covariates of interest: age, gender, ethnicity, physical activity, smoking, and alcohol consumption. Logistic regressions modeling was used to investigate the association between AL each of the selected PFAS with covariates of interest included in the model. The results indicated a statistically significant positive association between AL and most of the PFASs, except PFUA, which was not statistically significant with a p value of 0.531.
Logistic regression modeling also found that AL was positively and significantly associated with age (p-value < 0.001, physical activity (p-value < 0.001) ethnicity (p-value < 0.001), alcohol consumption (p-value < 0.001), and smoking (p-value < 0.001). The summary statistics of the behavioral health covariates for participants (Table 7) gives context to these findings by gender.

Discussion
This study explored the association between AL and PFASs. The major PFASs are environmentally and biologically persistent, with many having long half-lives in humans [36]. Allostatic load, an index of the chronic stress response, represents the physiological wear and tear on the body from extended exposure to stressful events. Markers of AL potentially shield the body in the short term and promote adaptation, but in the longer term, AL causes alterations in the body that bring forth disease. Indeed, AL has been shown to predict cardiovascular events and all-cause mortality, as it represents biological mediators in adaptation and maladaptation of the individual circumstances of life [37]. Our methods for calculating AL, which mirrored that of other studies [12,14,38,39] helped give insight into this understudied area.
Understanding the association between the cumulative physiological burden of stress (AL) and PFASs in the context of life's demands, behavioral factors, and other factors [40] helps to close the gap in the literature on the role of environmental toxicants on stress and health. This is especially the case, since people are exposed to stressors at all points in life with the cumulative burden altering health in the absence of resilience [41,42].
In our study, non-Hispanic Blacks had higher allostatic load scores across all age groups. This may be attributed to factors such as social and environmental exposures promoting the expression of genes more likely to put individuals in allostatic load. It may also be due to perceived racial discrimination or internalized racism [30].
This study found positive significant associations between AL and all the PFASs, apart from PFUA, in adjusted models evaluating AL and individual PFAS serum levels. In unadjusted models, AL was only significantly associated with PFOA p-value = 0.006. This indicates that the perceived association between PFASs and AL is likely due to confounding by behavioral factors, where people who smoke or consume alcohol are likely to have elevated PFASs levels as well as a higher AL index.
Other studies' [11,40] findings were similar to this study. They found an association between AL and health-damaging habits, such as smoking, low physical activity, and extensive alcohol consumption.
Males with higher serum PFOA had higher AL than females (Figures 2 and 3). These differences may be explained by the divergent toxicokinetic of PFOA between the genders [43], the excretion pathway of females, such as menstruation and lactation [44], or its interaction with estrogen receptors [45]. That said, further analysis is needed by gender.
More research is needed to understand how PFASs contributes to AL. In addition, multicollinearity made analyzing all PFASs together impossible. Ultimately, behaviors such as physical exercise and limiting smoking and alcohol consumption can help to reduce AL. These behavioral changes, in addition to increasing resilience, can help promote longevity and decrease the likelihood of chronic diseases. If AL is unmanaged, the long-term effects of chronic, unmitigated, multi-year stress may be disease and subsequent death.
Limitations: This cross-sectional study depicts only a snapshot in time. A longitudinal study of the group may have yielded distinct results as people's unique circumstances (finances, education, family, etc.) change resulting in people moving to different geographic areas, creating different social networks, gaining access to vital knowledge about PFASs exposure, and hence potentially decreasing exposure risk. In addition, due to the study design, it is unclear if PFASs serum levels increase AL or if AL makes once more likely to expose themselves to PFASs. In addition, multicollinearity made analyzing all PFASs together impossible: machine-learned models, Bayesian kernel machine regression, or weighted quantile sum (WQS) analysis may offer more insight into the interaction of multiple PFASs in promoting AL. Finally, a detailed examination of the associations by among different races/ethnicities and gender may have yielded different results.

Conclusions
AL levels in US adults are associated with PFASs serum levels. A reduction in alcohol consumption and smoking and increased physical activity may alter health outcomes. Future works should explore these findings in detail by gender and ethnicity and investigate the degree to which PFASs and other environmental and social exposures contribute to AL since AL is likely a mediator for several chronic diseases such as cardiovascular disease and cancer. Funding: This research was funded in part by NHLBI grant R25 HL105400.

Institutional Review Board Statement:
This study did not require IRB approval because de-identified secondary data were used. In the collection of the data by the Centers for Disease Control and Prevention, the study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of the Centers for Disease Control and Prevention.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The NHANES dataset is publicly available online, accessible at cdc. gov/nchs/nhanes/index.htm (accessed on 12 February 2022).