Perfluoroalkyl and polyfluoroalkyl substances (PFASs) are a class of chemicals composed of fluorinated alkyl chains and a polar head group^[1]. They are widely used in industry and consumer products because of their excellent thermal stability^[2]. Humans can be exposed to them in a variety of ways, such as air, dust, diet, and drinking water^[3-5]. PFASs, as typical endocrine disrupting chemicals (EDCs), can have various negative effects on human health, including diabetes, cardiovascular diseases and metabolic dysregulation^[6-9]. Different PFASs may exert diverse toxic effects on human health, probably caused by differences in chemical structure, carbon atom length, and functional groups between different PFASs^[10,11]. Therefore, it is necessary to consider the structural characteristics of various PFASs to accurately assess their effects on human and environmental health^[12].

An estimated 603.7 million adults are obese, according to the Global Burden of Disease Obesity Collaboration. Additionally, almost 40% of Americans are obese, according to a nationally representative survey^[13,14]. Obesity poses a major burden on health and is associated with a greater incidence of chronic disorders^[15,16]. Susceptibility to obesity-related diseases depends largely on specific differences in body fat composition^[17]. Unlike body mass index (BMI), which fails to identify individuals with excess visceral fat/heterotopic fat, fat distribution has become an important predictor of the risk of obesity-related diseases^[17]. Specific regional fat accumulation, especially in the central abdomen, increases the risk of metabolic and cardiovascular diseases and even death^[18]. Therefore, as a potent risk factor for the development of obesity and cardiovascular diseases, abdominal fat distribution is more worthy of study as an intermediate variable in the risk of metabolic diseases caused by PFASs.

A number of studies have examined the associations between PFASs and body fat distribution. Children with relatively high levels of perfluorooctane sulfonate (PFOS) and perfluorohexane sulfonate (PFHxS) presented relatively small increases in total fat and trunk fat mass, especially in subcutaneous fat mass^[19]. A prospective cohort study of a population at high risk for diabetes revealed a positive association between various plasma PFAS concentrations and a greater sum of skinfold thickness, subcutaneous fat and visceral fat^[20]. However, these two studies focused only on the effects of individual PFASs on adiposity. Humans are exposed to multiple PFASs from different sources in the natural environment, and PFAS exposure varies significantly among populations in different regions. In this study, we focused on the dose-response relationships between PFAS congeners of different structures and fat distributions and explored the possible toxic effects of single PFAS exposure, the trait spectrum exposure pattern of PFASs, and different structural-based PFAS exposures on regional fat accumulation.

The National Health and Nutrition Examination Survey (NHANES) is a survey conducted by the National Center for Health Statistics (NCHS) to evaluate the health and nutritional status of the civilian noninstitutionalized U.S. population. The survey consists of interviews with household members, standardized physical examinations, and the collection of medical histories and biological samples. All participants under the age of 18 years provided written approval from their parents or guardians. The current study used continuous NHANES data from four cycles from 2011 to 2018. The study sample was limited to adults (20–59 years) with dual-emission X-ray absorptiometry (DXA) measurement information (n = 4,362).

We detected 16 PFASs in the serum samples from the participants: perfluorooctanoic acid (PFOA), PFOS, 2-(N-ethyl-perfluorooctane sulfonamido) acetic acid (Et-PFOSA-AcOH), perfluorooctane sulfonamide (PFOSA), PFHxS, 2-(N-methylperfluoroctanesulfonamido) acetic acid (Me-PFOSA-AcOH), perfluorodecanoic acid (PFDeA), perfluorononanoic acid (PFNA), perfluoroundecanoic acid (PFUA), linear perfluorooctanoate (n-PFOA), branch perfluorooctanoic acid isomers (Sb-PFOA), linear perfluorooctane sulfonate (n-PFOS), monomethyl branched isomers of PFOS (Sm-PFOS), perfluorobutane sulfonic acid (PFBuS), perfluoroheptanoic acid (PFHpA), and perfluorododecanoic acid (PFDoA). The concentrations of PFAS were measured in blood samples collected during the medical examination. PFASs were quantified in serum via solid-phase extraction-high-performance liquid chromatography-turbo-ion spray ionization-tandem mass spectrometry (SPE-HPLC-TCI-MS/MS)^[21]. Eight PFASs (PFOA, PFOS, PFHxS, PFDeA, PFNA, n-PFOA, n-PFOS, and Sm-PFOS) were detected in at least 79% of the samples and were retained in subsequent analyses. The chemical information, structures, and distributions of these PFASs are shown in Supplementary Tables S1–S2 (available in www.besjournal.com). Samples with PFAS concentrations below the limit of detection (LOD) were assigned a value equal to the LOD divided by the square root of 2.

From 2011–2018, whole-body and android and gynoid (A/G) region DXA scans were acquired with Hologic Discovery A densitometers (Hologic, Inc.), reviewed and analyzed via Hologic APEX version software with the NHANES Body Composition Analysis option. This study included six major fat distribution variables: trunk fat mass (TR), android fat mass (AN), gynoid fat mass (GY), subcutaneous fat mass (SA), total abdominal fat mass (TA) and visceral adipose tissue mass (VFA). The visualization of fat composition is shown in Supplementary Figure S1, available at www.besjournal.com. Moreover, given that the android to gynoid fat ratio (A/G ratio) is a good predictor of several diseases, we also included it in our analysis^[22].

Figure S1.  Study design overview.

The NHANES collected data on potential confounding variables that might be connected to changes in PFAS concentrations through questionnaires. The variables included race/ethnicity (Mexican American, other Hispanic, non-Hispanic white, non-Hispanic black, other races), household income-to-poverty ratio (< 1.30, 1.30–3.49, > 3.50), physical activity (1–2 times/week, ≥ 3 times/week), smoking status (never smoker, former smoker, current smoker), drinking status (drinker, nondrinker), BMI (< 25.0, 25.0–29.9, > 30.0 kg/m²), waist circumference and high-density lipoprotein (HDL).

For the analyses, we applied the primary sampling unit, strata, and weights calculated using the four sampling cycles (2011–2018 cycles) according to the NHANES analytic and reporting guidelines. We stratified all analyses by age and sex group, 20–39 years/39–59 years and female/male subgroups, since previous studies have reported that age-specific and sex-specific differences in fat distribution exist^[23,24]. The demographic characteristics are presented as the means ± standard deviations (SDs) or n (%). We compared the demographic characteristics of individuals in the age-specific and sex-specific subgroups by using the chi-square test to compare categorical variables and analysis of variance to compare continuous variables.

Relationships of log-transformed PFAS concentrations with fat distribution and composition were examined by multiple linear regression models to calculate beta values (β) as model regression coefficients and 95% confidence intervals (95% CI). Model 1 was crude (unadjusted), and Model 2 included race, PIR, physical activity, smoking status, and drinking status. Model 3 was further adjusted for BMI, waist circumference and HDL level. Missing covariate values were substituted with median values^[25]. The potential effect of the PFAS concentration on the A/G ratio was also evaluated. We created a fat depot score by summing the concentrations of PFASs in 6 regional fat depots, weighted by the multivariable-adjusted risk estimates (β coefficients) caused by PFASs in the present analysis to capture the joint effect of regional fat depots on PFASs. All of the models were fully adjusted for confounders. Moreover, we explored the associations between tertiles of different distributions of body fat and PFASs in subgroups whose BMIs were < 25.0 kg/m².

Moreover, we classified PFAS congeners on the basis of the structural characteristics of specific PFASs to examine the structure-specific associations between PFASs and fat distribution: a) branched and linear isomers; b) different carbon chain lengths; and c) different functional head groups. We constructed the following summary measures: short-chain perfluoroalkyl carboxylates (PFCAs)/perfluoroalkyl sulfonates (PFSAs), long-chain PFCAs/PFSAs, and total PFCAs/PFSAs. The associations between structurally grouped PFASs and fat distribution were also explored. Considering the complexity of mixed exposure to chemicals in the human body, we explored the potential toxicity of different combinations of PFAS exposures on fat distribution, including 2 PFAS combinations, 3 PFAS combinations, 4 PFAS combinations, 5 PFAS combinations and 6 PFAS combinations. An interaction analysis of PFAS and stratification variables (physical activity, BMI, PIR, smoking status and drinking status) was conducted on the fat distribution. The dose‒response relationship between PFAS exposure and fat distribution was fitted by a cubic spline, and 4 knots were set in the models.

In addition, we estimated the ability of PFAS exposure to predict the risk of obesity (BMI > 30.0 kg/m²). To build and evaluate the prediction model, we randomly assigned the participants to the training set and the test set at a ratio of 7:3. The factors were selected to establish the conventional model according to the results of single-factor regression, including marital status, PIR, smoking status, drinking status, waist circumference, diastolic blood pressure (DBP), HDL, direct HDL-C, total cholesterol, glycohemoglobin (%), insulin, and the insulin resistance index (HOMA-IR). Insulin resistance was estimated by HOMA-IR via the following equation: HOMA-IR = fasting insulin (μIU/mL) × fasting glucose (mmol/L)/22.5^[26].

We attempted different combinations (conventional model plus single PFAS exposure) to determine which combination could improve the discriminative ability of the model. The evaluation results in the hold-out testing sets, including areas under the receiver operating characteristic curve (AUCs) of the receiver operating characteristic (ROC) curve, were calculated. The discriminative abilities of the models were assessed using the AUC, and improvements in subject risk.

Additionally, we performed sensitivity analyses to test whether the correlation between PFASs and fat distribution was robust. Concentrations of PFASs more than three times the 75th percentile were removed^[12]. We also conducted sensitivity analyses by adjusting for low-density lipoprotein (LDL). All analyses were performed with R (version 4.0.4). To be considered statistically significant, we required a two-tailed P value of less than 0.05.

The study was carried out in compliance with the Declaration of Helsinki. Ethical review and approval were not required for this study, as it involved the use of publicly available data and did not qualify as human subject research in the United States.

The demographic characteristics of the study population are shown in Supplementary Table S3 (available in www.besjournal.com). Our cohort included 4,362 individuals, including 2,276 (52.2%) females and 2,086 (47.8%) males. We divided the population into four groups by age and sex (males aged 20–39 years and 39–59 years, females aged 20-39 years and 39–59 years). The majority of the participants were non-Hispanic white (33.7%). We observed significant differences in BMI, education, marital status, family, the income‒to‒poverty ratio, physical activity, smoking status and drinking status among the four groups. Moreover, we also analyzed the age-specific and sex-specific distributions of the concentrations of serum PFASs in the study population (Supplementary Table S4, available in www.besjournal.com). PFAS exposure was significantly greater in males than in females, whereas PFAS levels were greater in the 39–59-year-old age group than in the 20–39-year-old age group since PFAS may accumulate.

We found significant associations between single-PFAS exposure and changes in TR, AN, SA, TA, and VFA in Model 3 (Table 1). The results of Model 1 and Model 2 are shown in Supplementary Table S5 and Table S6 (available in www.besjournal.com). For each doubling of PFOS, the change in TR in females aged 39–59 years was 1.52 (95% CI: 0.14289). The participants aged 39–59 with higher concentrations of PFHxS and PFNA had greater accrual of AN (β = 0.40, 95% CI: 0.03, 0.78; β = 0.44, 95% CI: 0.11, 0.77). Moreover, n-PFOA, n-PFOS and Sm-PFOS were associated with fewer SA depots (n-PFOA in females aged 20-39 years, β = −0.64, 95% CI: −1.16, −0.13; n-PFOS in males aged 39–59 years, β = −0.55, 95% CI: −0.99, −0.11); Sm-PFOS in males aged 20–39 years, β = −0.38, 95% CI: −0.72, −0.05; Sm-PFOS in males aged 39–59 years, β = −0.63, 95% CI: –1.03, −0.23). The concentrations of PFOS, PFHxS, and PFNA were shown to be linked with an increase in TA, whereas n-PFOA, n-PFOS, and Sm-PFOS were found to be related to a decrease in TA. Females aged 39–59 years with higher concentrations of PFOS, PFHxS, PFDeA, PFNA and n-PFOA had greater accumulation of VFAs (PFOS, β = 0.68, 95% CI: 0.16, 1.21; PFHxS, β = 0.22, 95% CI: 0.02, 0.42; PFDeA, β = 0.25, 95% CI: 0.08, 0.42; PFNA, β = 0.31, 95% CI: 0.13, 0.49; n-PFOA, β = 0.25, 95% CI: 0.03, 0.47). Moreover, we explored the changes in the A/G ratio. We found that doubling the concentrations of PFOA, PFDeA and PFNA was associated with a significant increase in the A/G ratio in men aged 20-39 years (Supplementary Table S7, available at www.besjournal.com) (PFOA, β = 0.97, 95% CI: 0.04, 1.89; PFDeA, β = 0.59, 95% CI: 0.13, 1.04; PFNA, β = 0.53, 95% CI: 0.04, 1.03). The associations between PFAS exposure and joint assessments of body fat are shown in Supplementary Table S8, available at www.besjournal.com. Among women aged 39–59 years, PFOS, PFHxS, PFDeA, PFNA, n-PFOA, n-PFOS and Sm-PFOS were significantly associated with increased body fat [PFOS, β = 1.68, 95% CI: 0.43, 2.93; PFHxS, β = 0.24, 95% CI: 0.07, 0.42; PFDeA, β = 0.18, 95% CI: 0.07, 0.29; PFNA, β = 0.41, 95% CI: 0.17, 0.65; n-PFOA, β = 0.46, 95% CI: 0.17, 0.76; n-PFOS, β = 0.22, 95% CI: 0.05, 0.38; Sm-PFOS, β = 0.29, 95% CI: 0.07, 0.51). We also observed that PFDeA, n-PFOA and Sm-PFOS were significantly associated with an increase in joint body fat in the subpopulation of men aged 20–39 years [PFDeA, β = 0.14, 95% CI: 0.02, 0.26; n-PFOA, β = 0.14, 95% CI: 0.01, 0.27; Sm-PFOS, β = 0.48, 95% CI: 0.01, 0.96). n-PFOS and Sm-PFOS exposure was associated with a greater risk of fat distribution [n-PFOS, β = 0.62, 95% CI: 0.11, 1.12; Sm-PFOS, β = 0.77, 95% CI: 0.21, 1.34). We performed additional analyses of the relationships between abnormal body fat distribution and PFAS exposure when BMI was normal (Supplementary Tables S9–S11, available in www.besjournal.com). In the BMI < 25.0 subgroup, we found that PFOA exposure may be positively associated with tertile 2 of the VFA in males aged 39–59 years [β = 0.77, 95% CI: 0.26, 1.29), whereas PFHxS and PFNA were positively associated with tertile 2 of the VFA in females aged 39–59 years [PFHxS, β = 0.40, 95% CI: 0.00, 0.80; PFNA, β = 0.65, 95% CI: 0.28, 1.02).

In the sensitivity analysis, by removing concentrations of PFAS more than three times the 75th percentile (Supplementary Table S12, available in www.besjournal.com), we found that participants with higher concentrations of PFOS, PFHxS and PFNA had less accrual of AN, TA, VFA and TR. Moreover, in the sensitivity analysis adjusted for LDL (Supplementary Table S13, available in www.besjournal.com), most of the results remained the same as those in the primary analyses, which means that our results are relatively robust. The structure-specific associations with outcomes were examined by grouping PFAS congeners according to their structural characteristics, namely, total PFSAs, short-chain PFSAs (PFHxS), long-chain PFSAs (Sm-PFOS, n-PFOS, PFOS) and long-chain PFCAs (n-PFOA, PFOA, PFNA, PFDeA), and the relationships between each of the PFAS and abdominal fat distribution were compared from the perspective of chemical structure, as shown in Figure 1. In all subgroups, the effect of a single PFAS in PFSA congeners on body fat decreased with an increasing number of carbon atoms, reaching the highest level among the effects caused by PFOS exposure. No significant regularity was observed among the PFCAs. The associations of structurally grouped PFAS concentrations with regional fat distribution are shown in Supplementary Table S14, available in www.besjournal.com. Exposure to short-chain PFSAs was significantly associated with increased AN (0.39), TA (0.49), and VFA (0.22) in females aged 39–59. The accumulation of VFA was affected by long-chain PFCAs [β = 0.31, 95% CI: (0.10, 0.52)].

Figure 1.  Effect estimates and 95% confidence intervals for associations of single PFAS with regional fat grouped by structural characteristics. PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonate; PFHxS, perfluorohexane sulfonate; PFDeA, perfluorodecanoate; PFNA, perfluorononanoate; n-PFOA, n-perfluorooctanoate. n-PFOS, linear perfluorooctane sulfonate; Sm-PFOS, monomethyl branched isomers of PFOS; PFAS, Per and polyfluoroalkyl substance.

Considering the differences in the distribution of PFASs in different regions that may lead to the diversity of PFAS exposure in the human body, we comprehensively evaluated the effects of different PFAS combination exposures on regional fat depots (Figure 2). The results revealed that combinations containing PFOS and/or PFOA as exposure traits were more likely to be positively associated with abdominal fat, and this association was more pronounced in women aged 39–59 years. Positive associations were focused on the endpoint of visceral fat.

Figure 2.  Association of trait spectrum exposure pattern of per and polyfluoroalkyl substance (PFAS) with regional fat depots.

Next, we used restricted cubic splines to visualize the relationship between PFAS exposure and body fat in various regions (Supplementary Figure S2, available in www.besjournal.com). We found that both PFOA and PFOS have nonlinear associations with TR in men aged 20-39 years, with TR relatively flat at first and then increasing rapidly until PFOA exposure of approximately 9.25 µg/dL (P values for nonlinearity were 0.0182 and 0.0203, respectively). In women aged 20-39 years, PFHxS, n-PFOA and Sm-PFOS were nonlinearly associated with VFA (P values for nonlinearity were 0.0041, 0.0069 and 0.0303, respectively). We explored the interaction effects of several indicator variables (smoking status, drinking status, BMI, physical activity and PIR) that may affect the effect of PFAS exposure on each component of body fat (Supplementary Tables S16–S19, available in www.besjournal.com). There was a marked difference in the interaction between PFAS exposure and abdominal fat distribution in subgroups with different smoking habits, physical activity levels and PIR. For males aged 20–39 years, smoking status (P for interaction: AN = 0.022; GY = 0.013; TA = 0.030; TR = 0.018) and physical activity (P for interaction: AN = 0.006; GY = 0.003; TA = 0.007; TR = 0.014) interacted with Sm-PFOS in AN, GY, TA and TR. Moreover, Sm-PFOS interacted with physical activity in SA (P for interaction: 0.005) and smoking status, and the PIR level in VFA (P for interaction: 0.016; 0.008), n-PFOS interacted with smoking status in VFA (P for interaction: 0.015), n-PFOA interacted with physical activity in GY (P for interaction: 0.039), and PFHxS interacted with physical activity in SA (P for interaction: 0.019). In males aged 39–59 years, the effect of PFHxS was modified by smoking status in AN and TA (P for interaction: 0.040; 0.036). n-PFOS interacted with smoking status and the PIR level in VFA (P for interaction: 0.002; 0.016). Sm-PFOS was modified by smoking status in VFA and TR (P for interaction: 0.016; 0.018), physical activity in TR (P for interaction: 0.014), and PIR level in VFA (P for interaction: 0.008). In addition, PFHxS interacted with BMI in VFA (P for interaction: 0.012), and n-PFOS interacted with BMI in SA and TR in females aged 20–39 years (P for interaction: 0.011; 0.044). In females aged 39–59 years, PFOS was modified by physical activity in AN (P for interaction: 0.038), whereas PFHxS was modified by smoking status in AN, GY, and SA (P for interaction: 0.006, 0.003, 0.009), and BMI in VFA (P for interaction: 0.012). Moreover, PFDeA was modified by smoking status in GY (P for interaction: 0.020), and n-PFOS interacted with BMI in TR (P for interaction: 0.044).

Figure S2.  Dose-response association of PFAS level with regional fat depots.

ROC analyses were used to explore the predictive power of PFAS exposure and conventional risk factors for obesity in the age-specific and sex-specific population (Supplementary Table S15, Figure 3, available in www.besjournal.com). The AUCs of the conventional models ranged from 0.9557 to 0.9886. Among the single PFAS exposures, the addition of PFDeA and PFNA increased the AUC to 0.9770 in the male participants (20–39; 39–59 years). Moreover, only PFDeA increased the ability to predict obesity in the younger female population. In women aged 39 to 59 years, the AUC of the prediction model increased to a certain degree after being combined with three single chemicals (PFHxS, PFDeA, and PFNA).

Figure 3.  ROC curves comparing the prediction of different models of obesity.