Per- and Polyfluoroalkyl Substances (PFAS), Perceived Stress, and Depressive Symptoms in a Prospective Cohort Study of Black Women

Abstract

Background:

Per- and polyfluoroalkyl substances (PFAS) are endocrine-disrupting chemicals with neurotoxic properties. PFAS have been associated with depressive symptoms in women in some studies, but little research has evaluated the effects of PFAS mixtures. Further, no study has investigated interactions of PFAS-depression associations by perceived stress, which has been shown to modify PFAS effects on other health outcomes.

Objective:

In a prospective cohort study of reproductive-aged Black women, we investigated associations between PFAS and depressive symptoms and the extent to which perceived stress modified these associations.

Methods:

We analyzed data from 1,499 participants (23–35 years) from the Study of Environment, Lifestyle, and Fibroids. We quantified concentrations of nine PFAS in baseline plasma samples using online solid-phase extraction-liquid chromatography-isotope dilution tandem mass spectrometry. Participants reported perceived stress via the Perceived Stress Scale (PSS-4; range=0–16) at baseline and depressive symptoms via the Center for Epidemiologic Studies Depression Scale (CES-D; range=0–44) at the 20-month follow-up visit. We used Bayesian Kernel Machine Regression to estimate associations between PFAS concentrations, individually and as a mixture, and depressive symptoms, and to assess effect modification by PSS-4 scores, adjusting for confounders.

Results:

Baseline perfluorodecanoic acid concentrations were associated with greater depressive symptoms at the 20-month follow-up, but associations for other PFAS were null. The PFAS were not associated with depressive symptoms when evaluated as a mixture. The association between the 90^th percentile (vs. 50^th percentile) of the PFAS mixture with CES-D scores was null at the 10^th (β=0.03; 95% CI=0.20, 0.25), 50^th (β=0.02; 95% CI=−0.16, 0.19), and 90^th (β=0.01; 95% CI=0.18, 0.20) percentiles of PSS-4 scores, suggesting perceived stress did not modify PFAS mixture.

Conclusion:

In this prospective cohort study, PFAS concentrations—assessed individually or as a mixture—were not appreciably associated with depressive symptoms, and there was no evidence of effect modification by perceived stress.

Graphical Abstract

1. INTRODUCTION

Per- and polyfluoroalkyl substances (PFAS) are a class of environmentally persistent endocrine-disrupting chemicals (EDCs) known to dysregulate hormonal systems (Agency for Toxic Substances and Disease Registry, 2021). PFAS are used extensively in industrial applications (e.g., firefighting foams) and in consumer products (e.g., clothing, cookware, food packaging, cosmetics) (Agency for Toxic Substances and Disease Registry, 2021); human exposure to PFAS is therefore widespread (U.S. CDC, 2023). Data from the National Health and Nutrition Examination Survey (NHANES) show that 98% of Americans have detectable concentrations of PFAS in their blood (U.S. CDC, 2023). Given that PFAS have been associated with numerous adverse health outcomes, such as metabolic and neurological disorders (Brown-Leung and Cannon, 2022; Fenton et al., 2021), the ubiquity of exposure to PFAS is a public health concern.

Depression, a mood disorder that affects nearly 20% of U.S. adults, is characterized by feelings of sadness, irritability, disrupted sleep, changes to appetite, impaired cognition, and loss in interests (Chand and Arif, 2023; Lee et al., 2023; Otte et al., 2016). Though depression affects a large proportion of adults, inequities in the prevalence of depressive symptoms are well documented: women, for example, are twice as likely to report depressive symptoms as men (Albert, 2015). Moreover, Black women, relative to white women, are more likely to have chronic and severe depressive symptoms but are far less likely to seek care (Nelson et al., 2020; Williams et al., 2007; Woods-Giscombe et al., 2016). Depressive symptoms among Black women may relate to experiences of structural racism and discrimination (Martin et al., 2023; Patel et al., 2023; Quist et al., 2022a), and have in turn been linked with elevated body mass index (BMI), hypertension, and risk of cardiovascular disease (Gaffey et al., 2022; Harris et al., 2021).

In addition to psychosocial stressors, toxicological data support the role of environmental chemicals like PFAS in the etiology of depressive symptoms. The neurotoxic properties of PFAS have been well characterized: PFAS alter the functional connectivity of the brain, including regions that regulate emotion, motivation, and mood (e.g., prefrontal cortex, hippocampus) (Bremner et al., 2000; Mayberg et al., 1999; Pandya et al., 2012; Zhang et al., 2018). Additionally, PFAS disrupt neurotransmitter metabolism (i.e., dopamine, glutamate, gamma-aminobutyric acid [GABA]) and calcium ion signaling, particularly in the hypothalamus, cortex, amygdala, and hippocampus, which have been implicated in the development of depressive symptoms (Belujon and Grace, 2017; Brown-Leung and Cannon, 2022; Fogaça and Duman, 2019; Onaolapo and Onaolapo, 2021; Pandya et al., 2012). PFAS further dysregulate the hypothalamic-pituitary-adrenal (HPA) and hypothalamic-pituitary-thyroid (HPT) axes and have been associated with altered concentrations of hormones (e.g., cortisol) that play a role in the stress response and anxiety symptoms (Keller et al., 2017; Varghese and Brown, 2001). These toxicological mechanisms are supported by animal data illustrating increased post-partum anxiety-like behavior among dams treated with a low-dose of perfluorooctanoic acid (PFOA) during pregnancy (Merrill et al., 2022).

Only a handful of epidemiologic studies have investigated PFAS exposure and depressive symptoms, with mixed results. Four studies measured prenatal PFAS biomarker concentrations and assessed maternal depressive symptoms; two studies found that maternal PFAS concentrations measured during pregnancy (first, second, or third trimester) were associated with higher depression scores during pregnancy (measured with the Center for Epidemiologic Studies Depression Scale) and 6 months post-partum (measured with the Edinburg Postnatal Depression Scale [EPD]) (Aung et al., 2023; Hu et al., 2024). However, two studies reported that maternal PFAS (serum or plasma) were not associated with greater depressive symptoms 42 days (measured with the EPD) or 4 weeks through 8 years (measured with the Beck Depression Inventory II) post-pregnancy (Vuong et al., 2020; Wang et al., 2023). An analysis of data from NHANES (2005–2018) among participants ≥20 years further reported non-linear associations of serum PFAS concentrations with depression scores (measured with the Patient Health Questionnaire-9 [PHQ-9]), such that only concentrations above a threshold (i.e., 39.7 ng/mL for total PFAS) were associated with depressive symptoms (Yi et al., 2023). Alternatively, a separate analysis of NHANES data (2005–2018) reported inverse associations between select PFAS (perfluorohexanesulfonic acid [PFHxS], PFOA, perfluoroundecanoic acid [PDUnDA]) and depression scores among participants 18 years and older (Sun et al., 2024).

Previous studies included small numbers of Black participants (Aung et al., 2023; Vuong et al., 2020; Yi et al., 2023). Because Black women report greater severity of chronic depressive symptoms (Nelson et al., 2020; Williams et al., 2007; Woods-Giscombe et al., 2016), understanding the potential role of PFAS in the etiology of depression among Black women is an important research objective. Further, psychosocial stress contributes to depressive symptoms (Quist et al., 2022b), likely through similar toxicological mechanisms as PFAS (i.e., HPA/HPT disruption) (Brown-Leung and Cannon, 2022; Busse et al., 2017), suggesting that psychosocial stress may modify the effects of PFAS, leading to greater depressive symptoms. Although perceived stress has been shown to modify associations of PFAS with other health endpoints (e.g., birth outcomes, corticotropin-releasing hormone) (Eick et al., 2022, 2023, 2021), we are not aware of any previous studies that have examined psychosocial stress as a joint exposure or effect measure modifier of PFAS in relation to depression. Finally, few previous studies of depression considered PFAS mixtures (Aung et al., 2023; Hu et al., 2024; Wang et al., 2023); because PFAS tend to target similar regions of the brain and have overlapping mechanisms of toxicity, exposure to multiple PFAS may have joint or interactive effects (Agency for Toxic Substances and Disease Registry, 2021). Consistent with animal studies of PFAS mixtures (Foguth et al., 2020; Grønnestad et al., 2021; Yu et al., 2016), two studies reported a joint adverse association of a PFAS mixture with depression scores that was stronger than when the PFAS were considered individually (Aung et al., 2023; Hu et al., 2024). However, these studies did not examine potential interactions among the PFAS.

In a prospective cohort study of reproductive-aged Black women, we investigated the association between plasma concentrations of a mixture of PFAS with depressive symptoms and assessed perceived stress as a modifier of the PFAS mixture.

2. METHODS

2.1. Study Population

We analyzed data from the Study of Environment, Lifestyle, and Fibroids (SELF), a prospective cohort of 1,693 reproductive-aged Black women from the Detroit metropolitan area (Baird et al., 2015; Wise, 2015). The study was designed to investigate the development and growth of uterine leiomyomata (UL). Participants were recruited between 2010 and 2012 and were eligible for the study if they: 1) were 23–35 years at enrollment, 2) self-identified as Black/African American, 3) had an intact uterus, 4) had no prior diagnosis of UL, and 5) had no previous diagnosis of cancer or an autoimmune disease that required regular medication use. We subsequently followed participants to ascertain UL incidence and collected data at multiple (20-month, 40-month, and 60-month) follow-up visits. Institutional Review Boards at Henry Ford Health (HFH), the National Institute of Environmental Health Sciences (NIEHS), and Boston University Medical Campus approved all study protocols. The involvement of the Centers for Disease Control and Prevention (CDC) laboratory did not constitute engagement in human subjects research.

2.2. Plasma Collection and Quantification of PFAS

The majority of participants (n=1,499) provided a non-fasting whole blood sample at baseline with sufficient volume to measure PFAS. HFH processed biospecimens and separated plasma from whole blood by centrifuging the whole blood for 10 minutes at 4°C and 1300g. Plasma samples were shipped to the NIEHS repository, aliquoted and stored in straws (Cryo Bio System) at −80°C. Plasma in the straws was aliquoted into polypropylene cryovials and shipped to the CDC overnight on dry ice in two batches (n=781 and 718, respectively). CDC staff used online solid-phase extraction-liquid chromatography-isotope dilution tandem mass spectrometry (Kato et al., 2018) to quantify plasma concentrations of nine PFAS: perfluorohexanesulfonic acid (PFHxS), linear perfluorooctanesulfonic acid (n-PFOS), branched perfluorooctanesulfonic acid (Sm-PFOS), linear perfluorooctanoic acid (n-PFOA), branched perfluorooctanoic acid (Sb-PFOA), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), perfluoroundecanoic acid (PFUnDA), and 2-(N-methyl-perfluorooctane sulfonamido) acetic acid (MeFOSAA). We calculated total PFOS (PFOS) and total PFOA (PFOA) by summing concentrations of linear and branched PFOS and PFOA, respectively. Full details on laboratory methods and quality assurance/quality control have been described previously (Wise et al., 2021). The CDC used previously validated laboratory methods utilized for NHANES to measure PFAS concentrations in SELF, including reagent blanks, analytic standards, and concentration quality control samples (Botelho and Pirkle, 2018; Kato et al., 2018).

The limit of detection (LOD) for all PFAS was 0.1 ng/mL; we included PFAS with a detection frequency >60% in the final analysis (Table S1): PFHxS, PFOS, PFOA, PFNA, PFDA, and MeFOSAA. We imputed PFAS concentrations <LOD in the final analysis as the LOD/√2 (Hornung and Reed, 1990).

2.3. Assessment of Depressive Symptoms

We measured depressive symptoms using the 11-item Iowa short-form of the CES-D (Table S2), which ascertains participants’ well-being and emotional state in the previous week and was constructed to identify individuals at high-risk of depression (Radloff, 1977). Participant responses included never, hardly ever, sometimes, often, or always (scored as 0, 1, 2, 3 or 4, respectively). Two questions (“I was happy” and “I enjoyed life”) reflected lower levels of depressive symptoms; we therefore reverse-coded responses to these questions. We summed scores across the 11 questions (range: 0—44), where higher scores suggested greater depressive symptoms. The CES-D, including shorter versions of the scale, has good reliability and internal consistency (Carpenter et al., 1998), including in Black and African American populations (Atkins, 2014; Canady et al., 2009; Cosco et al., 2017; Kohout et al., 1993; Makambi et al., 2009a; Torres, 2012).

We invited participants to complete the CES-D at the first follow-up visit that they attended, i.e., only once during follow-up. The majority (88.3%) of participants completed the CES-D at the first (20-month) follow-up visit; an additional 5.4% and 1.8% of participants completed the CES-D at the second (40-month) and third (60-month) follow-up visits, respectively. The remaining participants (4.5%) did not attend any follow-up visits and therefore did not have any CES-D data.

2.4. Measurement of Covariates

We collected baseline covariate data using interviews, in-person clinic visits, and self-administered questionnaires. Socioeconomic and demographic variables included age, educational attainment, annual household income, employment status, and marital status. Anthropometric and behavioral variables included BMI (calculated using height and weight measured at the baseline clinic visit), smoking status, and alcohol consumption. Reproductive variables included parity and breastfeeding duration summed across all prior births.

We measured baseline perceived stress, which we considered as a modifier of the PFAS mixture, using the Perceived Stress Scale (PSS-4) (Cohen et al., 1983). The PSS-4 asks participants to rate their perceived stress in the previous 30 days using the following four questions: have you felt: 1) that you were unable to control the important things in life, 2) confident about your ability to handle your personal problems, 3) that things were going your way, and 4) difficulties were piling up so high that you could not overcome them. Participant response options were never, almost never, sometimes, fairly often, or very often (scored as 0, 1, 2, 3, and 4, respectively). We reverse coded responses to questions 2) and 3) and then summed scores for all four questions to reflect total perceived stress levels at baseline (range: 0–16, where higher scores reflected greater stress levels). The PSS-4 has high short-term (2-month) reliability in Black populations (Cohen et al., 1983; Sharp et al., 2007).

2.5. Statistical Analysis

2.5.1. Multiple Imputation

The analytic sample for this analysis included 1,499 participants with sufficient plasma volume to measure baseline PFAS concentrations. We did not restrict our analytic sample to participants without a history of depression at baseline because we hypothesized that PFAS may influence symptom severity among participants with prior diagnoses of depression. To reduce potential for selection bias, we multiply imputed missing data on covariates (<1%; see Table S3) and CES-D scores (12% of participants who did not attend the 20-month follow-up visit; Figure 1). A subset of participants (~7%) who did not have CES-D scores measured at the 20-month follow-up had CES-D scores measured at a subsequent follow-up visit (Figure 1); we used the multiply imputed 20-month follow-up CES-D scores for these participants, rather than CES-D scores measured at a subsequent (40-month or 60-month) follow-up visit, in statistical analyses to reduce the potential for exposure misclassification resulting from varying time periods between plasma sampling (at baseline) and CES-D measurement. Importantly, multiple imputation of missing data is a valid method that can reduce the influence of bias (Cole et al., 2023; Madley-Dowd et al., 2019).

Figure 1.

Schematic illustrating the timing of CES-D measurement for participants in the SELF cohort (n=1,499). We imputed scores for participants that completed the CES-D at the 40-month (5.4%) or 60-month visit (1.8%). We also imputed scores for participants that did not have the CES-D measured at any follow-up visit (4.5%)

We generated five imputed datasets with Monte Carlo Markov Chain methods utilizing all relevant variables (PFAS, confounders, baseline PSS-4 scores, 20-month CES-D scores) under the assumption that the data were missing at random (Zhou et al., 2001). Plasma PFAS concentrations were similar among participants with CES-D scores measured at the 20-month follow-up compared with participants who completed the CES-D at a subsequent follow-up visit, and to participants who did not complete the CES-D at any follow-up visit (Table S4). Median imputed 20-month CES-D scores were similar, though lower, compared to CES-D scores measured at 40 months and 60 months among participants who completed the CES-D at the 40-month (n=81) and 60-month (n=27) follow-up visits, respectively (Figure 1, Table S5).

2.5.2. Summary Statistics and Data Transformations

We calculated descriptive statistics for plasma PFAS concentrations, potential confounders, PSS-4 scores, and CES-D scores (Table 1). PFAS plasma concentrations and CES-D scores were right-skewed; we therefore naturally log (ln) transformed these variables prior to modeling to satisfy modelling assumptions and reduce the influence of extreme values. We calculated Spearman correlation coefficients to estimate correlations between plasma concentrations of each PFAS.

Table 1.

Characteristics of 1,499 SELF participants in the analytic sample, 2010–2012.

Variable	N (percent) or median (25th, 75th percentiles)
Sociodemographic characteristics at baseline
Age at enrollment (years)	29 (26, 32)
Educational attainment
High School degree/GED	334 (22.3%)
Some college/Associate’s/Technical school	746 (49.8%)
Bachelor’s/Master’s/Doctorate	419 (27.9%)
Annual household income
<$20,000	696 (46.4%)
$20,000–50,000	544 (36.3%)
>$50,000	259 (17.3%)
Employment status
Not employed	581 (38.8%)
Employed <30 hours/week	179 (11.9%)
Employed ≥30 hours/week	739 (49.3%)
Marital status
Never married	885 (59.0%)
Currently married	401 (26.8%)
Previously married	213 (14.2%)
BMI (kg/m2)	32.4 (26.4, 39.6)
Smoking status
Never	1104 (73.6%)
Former smoker	114 (7.6%)
Current smoker <10 cigarettes/day	208 (13.9%)
Current smoker ≥10 cigarettes/day	73 (4.9%)
Alcohol consumption
Low	445 (29.7%)
Moderate	757 (50.5%)
Heavy	297 (19.8%)
Parity	1 (0, 2)
Perceived stress score (PSS-4) 1	7 (4, 8)
Depression score (CES-D) 2	13 (8, 19)
PFAS plasma concentrations (ng/mL)
PFHxS	0.6 (0.4, 1.0)
PFOS3	4.3 (2.8, 6.3)
PFOA3	1.3 (0.9, 1.8)
PFNA	0.5 (0.4, 0.7)
PFDA	0.2 (0.1, 0.3)
MeFOSAA	0.1 (0.1, 0.2)

¹PSS-4= Perceived Stress Scale; range= 0–16.

²CES-D= 11-item Center for Epidemiologic Studies Depression Scale; range= 0–44.

³PFOS and PFOA are the sum of their respective linear and branched isomers.

2.5.3. Selection of Confounders

We used a directed acyclic graph (DAG) informed by the literature to guide the selection of confounders for statistical models (Figure S1) (Aung et al., 2023; Chen et al., 2020; Gaum et al., 2019, 2017; Koh et al., 2017; Mutic et al., 2021; Peltier et al., 2021; Schildroth et al., 2021; Vuong et al., 2020). We adjusted all models for the following baseline characteristics: age (continuous, years), educational attainment (≤high school degree or GED, some college/Associate’s degree/Technical degree, or ≥Bachelor’s degree), annual household income (<$20,000, $20,000-$50,000, or >$50,000), employment status (not employed, employed <30 hours/week, or employed ≥30 hours/week), marital status (never married, currently married, or previously married), BMI (continuous, kg/m²), smoking status (never, former, current <10 cigarettes/day, or current ≥10 cigarettes/day), alcohol consumption (low [<10 drinks/year], moderate [> low, but lower than heavy], or heavy [≥6 drinks per day when drinking alcohol or ≥4 drinks at a single sitting twice per month]), and parity (continuous). Because perceived stress is a known predictor of depressive symptoms in women (Leonard et al., 2020; Li et al., 2021), we included PSS-4 scores as a covariate in all statistical models. We also considered PSS-4 scores as a potential effect measure modifier of associations between PFAS and depressive symptom (CES-D) scores in both Bayesian Kernel Machine Regression (BKMR) and multivariable linear regression models (described below).

2.5.4. Bayesian Kernel Machine Regression

We used BKMR as our primary model to examine associations between the plasma PFAS mixture and depressive symptoms. BKMR utilizes a flexible kernel function to model the mixture, allowing for nonlinearity of the mixture components, pairwise interactions, higher-order interactions between multiple mixture components, and estimation of joint effects (Bobb et al., 2018, 2015). We included PSS-4 scores as a component in the kernel function to assess effect measure modification of the PFAS mixture by perceived stress (Eick et al., 2021). We z-standardized all ln-transformed plasma PFAS concentrations and PSS-4 scores prior to BKMR modeling to keep components included in the kernel function on the same scale. The BKMR model took the following form:

$$CES-D\mathit{\text{\hspace{0.17em}score}}{s}_i=h(PFO{S}_i,PFO{A}_i,\mathit{\text{\hspace{0.17em}PFHx}}{S}_i,PFD{A}_i,PFN{A}_i,\mathit{\text{\hspace{0.17em}MeFOSA}}{A}_i,PSS-4\mathit{\text{\hspace{0.17em}score}}{s}_i)+{\beta }_1*Ag{e}_i+{\beta }_2*\mathit{\text{Educatio}}{n}_i+{\beta }_3*\mathit{\text{Incom}}{e}_i+{\beta }_4*\mathit{\text{Employment\hspace{0.17em}statu}}{s}_i+{\beta }_5*\mathit{\text{Marital\hspace{0.17em}statu}}{s}_i+{\beta }_6*BM{I}_i+{\beta }_7*\mathit{\text{Smokin}}{g}_i+{\beta }_8*\mathit{\text{Alcohol\hspace{0.17em}intak}}{e}_i+{\beta }_9*\mathit{\text{Parit}}{y}_i+e_i$$ [1]

where h represents the kernel function, CES – D scores reflect depressive symptoms, PSS – 4 scores reflect perceived stress, and e_i represents the error term.

We fit BKMR models for all five imputed datasets with the non-informative prior specifications, 50,000 iterations, and a 50% burn-in. We selected every 5^th iteration for analysis to reduce autocorrelation. We employed the component-wise variable selection option and estimated posterior inclusion probabilities (PIPs), reflecting the relative importance of each exposure variable included in the kernel function, for all PFAS and PSS-4 scores. We pooled findings from each BKMR fit across the five imputed datasets using Rubin’s rule with code developed previously (Devick, 2019; Rubin, 2004).

We estimated summary measures and plotted the following to describe the association of the mixture with depressive symptoms: 1) exposure response profiles for each mixture component, when all other mixture components were held at their medians; 2) joint association at varying percentiles (10^th – 90^th) of the mixture, compared to the 50^th percentile; 3) associations of each mixture component, for an increase from the 25^th to 75^th percentiles, at the 25^th, 50^th, and 75^th percentiles of all other mixture components; and 4) exposure response profiles for each mixture component, at the 25^th, 50^th, and 75^th percentiles of a second mixture component, holding all other mixture components at their medians. To investigate the modifying role of perceived stress on the overall PFAS mixture, we further estimated the joint association of the PFAS mixture at varying percentiles (10^th – 90^th), compared with the 50^th percentile, fixing PSS-4 scores at their 10^th, 50^th, and 90^th percentiles.

We estimated beta (β) coefficients and 95% credible intervals (CrIs) for all summary measures. To aid in the interpretation of our findings, we also presented β coefficients as the percent difference in CES-D scores, calculated as:

$$\%\mathit{\text{\hspace{0.17em}difference\hspace{0.17em}in\hspace{0.17em}CES}}-D\mathit{\text{\hspace{0.17em}scores}}=(e^{\beta }-1)*100$$ [2]

2.5.5. Multivariable Linear Regression

We used multivariable linear regression models to complement the findings from our BKMR model. There was limited evidence of nonlinearity in associations of plasma PFAS concentrations or PSS-4 scores with CES-D scores in the BKMR model and Generalized Additive Models (GAMs), which we fit to assess the linearity of associations between plasma PFAS concentrations and CES-D scores. We therefore modeled plasma PFAS concentrations and PSS-4 scores continuously in subsequent linear regression models. We adjusted for the same set of confounders as in the BKMR models, and mutually adjusted for all plasma PFAS concentrations and PSS-4 scores. We investigated the potential for pairwise interactions between PFAS and perceived stress by including cross-product terms between each PFAS and PSS-4 scores in regression models. All cross-product terms had p-values >0.20, and we therefore did not retain these terms in the final models.

We fit linear regression models for all five imputed datasets and estimated pooled β coefficients and 95% confidence intervals (CIs) using Rubin’s rule (Rubin, 2004). To improve interpretability, we back-transformed the β coefficients and 95% CIs using Equation [2] for PSS-4 scores and the following equations for the PFAS:

$$\%\mathit{\text{\hspace{0.17em}difference\hspace{0.17em}in\hspace{0.17em}CES}}-D\mathit{\text{\hspace{0.17em}scores}}=(e^{\text{ln(2)}*\beta }-1)*100$$ [3]

$$95\%CI=(e^{\text{ln(2)}*(\beta \pm 1.96*SE)})-1*100$$ [4]

where SE represents the standard error.

We interpreted the β coefficients as the percent difference in CES-D scores per doubling in PFAS concentrations or point increase in PSS-4 scores.

2.5.6. Sensitivity Analyses

We performed a series of sensitivity analyses to examine the robustness of our findings. We first fit BKMR and multivariable linear regression models: 1) only among participants with CES-D scores measured at the 20-month follow-up (n=1,323); 2) utilizing the 40-month and 60-month CES-D scores, rather than imputed 20-month CES-D scores, for participants who complete the CES-D at the 40- and 60-month follow-up visits, respectively; and 3) additionally adjusting for duration of breastfeeding (continuous, in months), given breastfeeding is a known route of elimination for PFAS and has been associated with depressive symptoms (Hamdan and Tamim, 2012; Pope and Mazmanian, 2016; Schildroth et al., 2021; Wise et al., 2021). Next, we fit BKMR models changing the: 1) smoothness of the kernel function from the default (b=100) to lower and higher degrees of smoothness (b=50 and b=1000, respectively), and 2) default uniform prior distribution to a gamma distribution (Bauer et al., 2020; Valeri et al., 2017). Finally, we fit linear regression models that were not co-adjusted for all PFAS.

3. RESULTS

3.1. Study Population Characteristics

The median age at enrollment of SELF participants in the analytic sample was 29 years (25^th-75^th percentiles: 26–32 years), and the median BMI was 32.4 kg/m² (25^th-75^th percentiles: 26.4–39.6 kg/m²; Table 1). At baseline, 27.9% of participants had a Bachelor’s degree or higher, 46.4% had annual household incomes <$20,000/year, 38.8% were not employed, and 59.0% had never been married. Seventy-four percent of participants were never smokers and 50.5% were moderate consumers of alcohol. Median (25^th-75^th percentiles) scores for perceived stress (PSS-4) and depressive symptoms (CES-D) were 7 (4–8) and 13 (8–19), respectively. Thirty-seven percent of participants had CES-D scores (≥16) indicative of clinically relevant depression (Park, 2021).

Median (25^th-75^th percentiles) plasma PFAS concentrations were highest for PFOS (4.3 ng/mL; 2.8–6.3 ng/mL) and PFOA (1.3 ng/mL; 0.9–1.8 ng/mL), and lowest for MeFOSAA (0.1 ng/mL; 0.1–0.2 ng/mL; Table 1 and Table S6). PFAS concentrations were weakly to moderately correlated, where Spearman correlation coefficients ranged from 0.11 (PFNA-MeFOSAA) to 0.68 (PFOA-PFNA; Table S7).

3.2. Associations of PFAS with Depressive Symptoms

Considered individually, most PFAS concentrations were not materially associated with CES-D scores in multivariable linear regression or BKMR models (Figure 2). However, a doubling in plasma PFDA concentrations was associated with an 11.5% (95% CI=-0.4%, 24.9%) increase in CES-D scores in multivariable linear regression models (Figure 2A). Likewise, when we evaluated PFAS using the BKMR model, PFDA was associated with an 10.5% increase in CES-D scores when the remaining mixture components were held at their 50^th percentiles (for an increase in PDFA concentrations from the 25^th to 75^th percentiles: β=0.10, 95% CrI=-0.06, 0.27; Figure 2B), but this association was less precise.

Figure 2.

Associations of plasma PFAS concentrations and PSS-4 scores with CES-D scores in the (A) multivariable linear regression model and (B) BKMR model (for an increase from the 25^th to 75^th percentiles, holding all other mixture components at their 25^th, 50^th, or 75^th percentiles). Both models were adjusted for age, educational attainment, annual household income, employment status, marital status, BMI, smoking status, alcohol consumption, and parity (n=1,499). The multivariable linear regression model was also co-adjusted for all plasma PFAS concentrations and PSS-4 scores. PFAS were ln-transformed for BKMR and multivariable linear regression models; ln-transformed PFAS and PSS-4 scores were z-standardized for the BKMR model. Beta coefficients from the multivariable linear regression model were back-transformed to percent (%) differences in CES-D scores per doubling in plasma PFAS concentrations or point increase in PSS-4 scores.

Higher perceived stress (PSS-4) scores were associated with greater CES-D scores in the multivariable linear regression model (β=8.5%, 95% CI=6.0%, 11.2%) and BKMR model (for an increase from the 25^th to 75^th percentiles, holding all other mixture components at their 50^th percentiles: β=0.34, 95% CrI=0.23, 0.45; Figure 2).

Increasing percentiles of the mixture, including all PFAS and PSS-4 scores, were jointly associated with higher CES-D scores in BKMR models (Figure S2); this association was driven exclusively by PSS-4 scores, which also had the highest PIP (1.0) relative to the PFAS (range: 0.06–0.12)

There was no evidence of pairwise or higher-order interactions between any of the PFAS in relation to CES-D scores in BKMR models (Figure S3 and Figure 2B), nor was there evidence that perceived stress modified associations of individual PFAS with CES-D scores in either the BKMR model (Figures S3) or the multivariable linear regression model.

We further examined potential effect modification by perceived stress of the association between the PFAS mixture and CES-D scores by estimating joint associations between the PFAS mixture (PFHxS, PFOS, PFOA, PFNA, PFDA, MeFOSAA) held at varying percentiles (compared to the 50^th percentile) when PSS-4 scores were held at their 10^th, 50^th, and 90^th percentiles (Figure 3). There was no evidence that the PFAS mixture was appreciably associated with CES-D scores, or that perceived stress modified the association between the PFAS mixture and CES-D scores. For example, the association of the 90^th percentile of the PFAS mixture, compared with the 50^th percentile, with CES-D scores was null when PSS-4 scores were held at their 10^th (β=0.03; 95% CI= −0.20, 0.25), 50^th (β=0.02; 95% CI= −0.16, 0.19), and 90^th (β=0.01; 95% CI= −0.18, 0.20) percentiles (Figure 3).

Figure 3.

Associations of varying percentiles (10^th—90^th) of the PFAS mixture (PFHxS, PFOS, PFOA, PFNA, PFDA, MeFOSAA) with CES-D scores, compared to the 50^th percentile of the PFAS mixture, when PSS-4 scores were held at their 10^th, 50^th, and 90^th percentiles, respectively. Models were adjusted for age, educational attainment, annual household income, employment status, marital status, BMI, smoking status, alcohol consumption, and parity (n=1,499). Plasma PFAS concentrations scores were ln-transformed and z-standardized; PSS-4 scores were z-standardized.

3.3. Sensitivity analyses

Our main findings did not differ materially from sensitivity analyses where we: 1) restricted the analytic sample to participants with CES-D scores measured at the 20-month follow-up (i.e., a complete case analysis, n=1,323; Figures S4 and S5); 2) utilized 40-month and 60-month CES-D scores, rather than the imputed 20-month CES-D scores, for participants who completed the CES-D at the 40- and 60-month follow-up visits (Figures S6 and S7); 3) additionally adjusted for prior breastfeeding duration (Figures S8 and S9); 4) changed the default prior BKMR specifications (Figure S10–S12); or 5) did not co-adjust linear regression models for all PFAS (Table S8). However, the association for PFDA observed in primary linear regression models was attenuated in the linear regression model that was not co-adjusted for all other PFAS (Table S8). As with our main findings, there was no evidence that plasma PFAS concentrations (individually or as a mixture) were appreciably associated with CES-D scores.

DISCUSSION

We estimated associations of a mixture that included six PFAS and perceived stress with depressive symptoms in a cohort of reproductive-aged Black women. Plasma PFDA concentrations were associated with increased depressive symptoms in linear regression and BKMR models, but other PFAS were not materially associated with depressive symptoms. There was also no evidence of joint or interactive effects of the PFAS mixture. Although self-reported levels of perceived stress were associated with increased depressive symptoms, perceived stress did not modify associations between the PFAS mixture and depression scores.

Environmental chemicals, including PFAS, have been hypothesized to play a role in the etiology of depressive symptoms given their ability to interfere with hormonal systems (Agency for Toxic Substances and Disease Registry, 2021). Disruption of the HPA/HPT axes has been implicated in the development of mood disorders because these axes regulate the stress response (Chu et al., 2022; Nuguru et al., 2022; Russell and Lightman, 2019; Varghese and Brown, 2001). Notably, individuals with depression have elevated levels of cortisol (Keller et al., 2017; Varghese and Brown, 2001), thyroid stimulating hormone (TSH) (Nuguru et al., 2022), and oxidative stress biomarkers (Bajpai et al., 2014; Correia et al., 2023), but reductions in triiodothyronine concentrations (Morin, 2015). Alternatively, studies have reported both increased and reduced levels of thyroxine (T4) among individuals with depressive symptoms (Hage and Azar, 2012; Odawara et al., 2023). PFAS exposure has been associated with higher levels of reactive oxygen species (ROS) and inflammatory biomarkers (Jiao et al., 2021; Ojo et al., 2021; Omoike et al., 2021; Taibl et al., 2022; Tan et al., 2023), compromised oxidative stress defenses (i.e., glutathione) (Jiao et al., 2021; Ojo et al., 2021), increased corticotropin-releasing hormone (CRH) (Eick et al., 2021), higher TSH levels, and lower T4 levels (Coperchini et al., 2020; Derakhshan et al., 2022; Inoue et al., 2019; Lebeaux et al., 2020; Preston et al., 2020, 2018; Reardon et al., 2019; Wang et al., 2013; Xiao et al., 2020; Zhang et al., 2023), supporting the potential induction of depressive symptoms by PFAS through HPA/HPT disruption.

In addition to HPA/HPT dysregulation, the etiology of depression is related to changes in functional connectivity in the brain, particularly the prefrontal cortex and limbic system (Bremner et al., 2000; Mayberg et al., 1999; Pandya et al., 2012; Zhang et al., 2018), activation of the cortico-amygdalastriatal circuits (Pandya et al., 2012; Price and Drevets, 2010; Zhang et al., 2018), and changes in neurotransmitter levels (e.g., dopamine, glutamate, GABA) (Belujon and Grace, 2017; Fogaça and Duman, 2019; Onaolapo and Onaolapo, 2021; Pandya et al., 2012). Importantly, regions of the brain implicated in depression are targets for PFAS neurotoxicity. PFAS induce ROS and disrupt endothelial tight junctions of the blood-brain-barrier, increasing permeability and allowing toxicants, including PFAS, to enter the brain (Brown-Leung and Cannon, 2022; Qian et al., 2010; Reistad et al., 2013; Wang et al., 2011; Yu et al., 2020). PFAS neurotoxicity is characterized by disruptions to neurotransmission through dysregulation of calcium ion signaling and neurotransmitter (e.g., dopamine, glutamate, GABA) synthesis, metabolism, and receptor binding in the cortex, hippocampus, amygdala, and hypothalamus (Brown-Leung and Cannon, 2022; Foguth et al., 2020, 2019; Grønnestad et al., 2021; Li et al., 2017; Long et al., 2013; López-Doval et al., 2015; Mshaty et al., 2020; Salgado et al., 2016, 2015; Tukker et al., 2020). These data support the hypothesis that PFAS exposure can impact depressive symptoms through multiple neurotoxic mechanisms affecting regions of the brain implicated in emotional regulation, mood, and motivation.

Despite extensive toxicological data implicating PFAS in the etiology of depression, epidemiologic studies investigating associations between PFAS and self-reported depressive symptoms are inconsistent. Four studies examined serum or plasma PFAS concentrations among pregnant women (in the first, second, or third trimester) in relation to depression scores measured during pregnancy or the post-partum period. While PFAS (PFOA, PFOS, PFHxS, PFNA, PFDA, PFUnDA, perfluorododecanoic acid, perfluorobutanesulfonic acid, and perfluoroheptanoic acid) were not associated with post-partum depressive symptoms measured at multiple timepoints following birth (42 days or 4 weeks through 8 years post-pregnancy) (Vuong et al., 2020; Wang et al., 2023), PFOA, PFOS, PFHxS, PFNA, PFDA, and MeFOSAA were associated with greater CES-D scores during the second trimester of pregnancy or greater EPD scores 1 to 6 months post-partum (Aung et al., 2023; Hu et al., 2024). These findings are consistent with an animal study that similarly reported higher post-partum anxiety-like behavior among low-dose PFAS-exposed dams (Merrill et al., 2022). Interestingly, the median PFOS concentration (2.0 ng/mL for participants with CES-D scores <16, 1.6 ng/mL for participants with CES-D scores ≥16) in Aung et al. was lower than that in the current study (4.3 ng/mL) (Aung et al., 2023), suggesting that the PFAS concentrations do not necessarily explain discrepant findings between our studies. While we measured PFAS concentrations among non-pregnant participants, Aung et al. measured PFAS during pregnancy (Aung et al., 2023). Pregnancy is a dynamic period characterized by anatomical and functional changes in the brain and may be a critical period of exposure to neurotoxicants (Barba-Müller et al., 2019). The timing of exposure assessment may therefore explain adverse PFAS-depression associations that Aung et al. reported. Alternatively, median PFAS concentrations were consistently higher in Hu et al., which, in addition to differences in exposure timing, may explain discrepant findings between our studies (Hu et al., 2024).

A more recent study of adults aged ≥20 years from NHANES (2005–2018 cycles) reported non-linear associations between PFAS and self-reported depressive symptoms measured using the PHQ-9 (Yi et al., 2023). Using GAMs, the authors identified inflection points in the dose-response curves for PFOS (25.4 ng/mL), PFOA (5.8 ng/mL), and PFNA (0.9 ng/mL), where serum concentrations below the inflection point were associated with lower depression scores and serum concentrations above the inflection point were associated with either increases (PFOS, PFNA) or no change (PFOA) in depression scores (Yi et al., 2023). The inflection point concentrations reported in Yi et al. at which adverse associations were observed were higher than median concentrations in our study (PFOS: 25.4 ng/mL vs. 4.3 ng/mL; PFNA: 0.9 ng/mL vs. 0.5 ng/mL), possibly explaining the inconsistency in results between their study and the current analysis. Moreover, the adverse associations Yi et al. observed were strongly attenuated among females in sex-stratified models, suggesting potential sex-specific effects; therefore, the lack of association in SELF may be explained by weaker effects of PFAS on depression among females.

We found little evidence of either interactions or joint effects of the PFAS mixture, likely because the individual PFAS were not strongly associated with depressive symptoms in either BKMR or linear regression models. These findings are contrary to prior studies in animal models that demonstrate exposure to PFAS mixtures had a greater effect on altered neurotransmitter concentrations (i.e., reduced glutamate, dopamine, serotonin levels) than PFAS considered individually (Foguth et al., 2020; Grønnestad et al., 2021; Yu et al., 2016). To our knowledge, only a handful of prior epidemiologic studies considered a PFAS mixture in relation to depressive symptoms. Using quantile-based g-computation, Aung et al. and Hu et al. both reported that serum or plasma concentrations of PFAS mixtures that included PFHxS, PFOS, PFOA, PFNA, PFDA, PFUnDA, perfluoroheptanesulfonic acid (PFHpS), 6:2 chlorinated polyfluorinated ether sulfonic acid (6:2 Cl-PFESA), or MeFOSAA were jointly associated with greater depression scores (primarily driven by PFOS and MeFOSAA in Aung et al. and PFHxS, PFOA, PFNA, PFHpS, and 6:2 Cl-PFESA in Hu et al.) (Aung et al., 2023; Hu et al., 2024). In Aung et al., this association was markedly stronger among immigrant women (driven by PFHxS, PFOS, PFNA, PFDA, and MeFOSAA), who also had lower serum PFAS concentrations than U.S.-born women for whom the joint association of the mixture was null. The authors concluded that the stronger association of the PFAS mixture with depression scores among immigrant women may reflect modification of the toxic effects of the PFAS mixture by social stressors (e.g., discrimination, hardship, social isolation) (Aung et al., 2023), highlighting the importance of considering both chemical and non-chemical stressors in mixtures research.

Although PFAS (considered individually or as a mixture) were not associated with depressive symptoms in SELF, we did observe strong associations between self-reported perceived stress and depressive symptoms, which has been reported previously (Catabay et al., 2019; Qu et al., 2012). Several types of stressors (i.e., violence, discrimination, financial hardship) have been associated with greater depression scores (Catabay et al., 2019; Gillespie et al., 2021; Holden et al., 2013; Quist et al., 2022b), as well as altered cortisol levels (Adam et al., 2015; Johnson et al., 2022; Lee et al., 2018; Nam et al., 2022; Suglia et al., 2010). However, there was no evidence that perceived stress modified the PFAS mixture in this study, and the association of the overall mixture (PFHxS, PFOS, PFOA, PFNA, PFDA, MeFOSAA, and PSS-4 scores) with depressive symptoms was driven exclusively by PSS-4 scores. The lack of effect modification by perceived stress on the association between PFAS and depression in SELF could relate to the absence of meaningful associations between most PFAS and depression scores. However, several prior studies reported joint and interactive associations of PFAS and perceived stress in relation to pregnancy outcomes (e.g., birthweight) (Eick et al., 2022, 2023) and circulating hormone levels (i.e., CRH) (Eick et al., 2021). Future studies should consider perceived stress in mixtures research for outcomes where the HPA/HPT axes are implicated.

This study represents one of only a handful of epidemiologic studies examining associations between PFAS concentrations and depressive symptoms. We investigated this research question in a large, established cohort with comprehensive chemical biomarker data measured using the same established methods and laboratory used to analyze NHANES samples, and prospective measurements of depressive symptoms. Our study also focused on reproductive-aged Black women, who are inequitably impacted by depression and remain underrepresented in environmental health research (Nelson et al., 2020; Williams et al., 2007; Woods-Giscombe et al., 2016). Further, we measured depressive symptoms using the CES-D, which has demonstrated good internal consistency and reliability (Atkins, 2014; Canady et al., 2009; Cosco et al., 2017; Kohout et al., 1993; Makambi et al., 2009a; Torres, 2012). Utilizing continuous CES-D scores allowed us to investigate potential subclinical effects and may have improved our outcome sensitivity compared to using depression diagnoses given that Black women are less likely to seek out and receive clinical diagnoses of depression (Nelson et al., 2020; Williams et al., 2007; Woods-Giscombe et al., 2016). Notably, CES-D scores in SELF tended to be elevated (i.e., 37% above the clinical cutoff for the CES-D-11) compared to other study populations of Black women (i.e., the Black Women’s Health Study, where 21%-26% of participants scored above the clinical cutoff for the CES-D-20) (Makambi et al., 2009b). Our focus on PFAS mixtures is also noteworthy because, to date, only one other study has investigated the potential for joint associations between a PFAS mixture and depressive symptoms in women. We utilized a state-of-the-art statistical approach to model the mixture, which allowed for the investigation of pairwise interactions, higher-order interactions, and joint effects. This is also one of the first studies to incorporate both chemical and non-chemical stressors as components in a complex mixture. Finally, we were able to adjust for a suite of potential confounders given the collection of extensive of socioeconomic, demographic, and behavioral data in SELF.

Our study had several limitations. Although we were able to adjust for several important confounders, we were unable to adjust for other stressors (e.g., structural racism, discrimination) that could be related to both PFAS exposure and depressive symptoms (Catabay et al., 2019; Gillespie et al., 2021; Holden et al., 2013; Quist et al., 2022b). We also did not adjust for dietary factors or other environmental exposures that may be associated with PFAS exposure and depression (Schildroth et al., 2021; Selvaraj et al., 2022; Wise et al., 2021). Though there were little missing data in SELF, ~12% of participants were missing CES-D scores at the 20-month follow-up. We multiply imputed CES-D scores for these participants to reduce the potential for selection bias. It is possible that utilizing the imputed scores led to some degree of non-differential misclassification of the outcome; however, we would generally expect this misclassification to create a bias towards the null (Yland et al., 2022). Importantly, PFAS biomarker concentrations were similar for participants who completed the CES-D (at the 20-month, 40-month, and 60-month follow-up visits) and for participants who did not complete the CES-D at any follow-up visit (Table S4), and our findings did not materially differ from sensitivity analyses in which we: 1) restricted to participants with complete 20-month outcome data, and 2) used the CES-D scores from subsequent follow-ups (40-month and 60-month), suggesting that our findings were not strongly affected by how we handled missing outcome data.

In this prospective study of reproductive-aged Black women, we found limited evidence that plasma concentrations of PFAS, considered individually or as a mixture, were associated with depressive symptoms. Although perceived stress has been shown to modify the effects of PFAS on other health outcomes in previous studies (Eick et al., 2021), and perceived stress was strongly related to depressive symptoms in our cohort, there was no indication that perceived stress modified associations of the PFAS mixture with depressive symptoms.

HIGHLIGHTS.

• PFAS have been associated with depressive symptoms, but little is known about PFAS mixtures.

• We evaluated associations of a PFAS mixture with depressive symptoms in Black women.

• The PFAS mixture was not appreciably associated with depressive symptoms.