Associations between serum per- and polyfluoroalkyl substance concentrations and β-cell function and insulin resistance in adult females

Abstract

Context

Epidemiological evidence of exposure to precursor and alternative per- and polyfluoroalkyl substances (PFASs) and metabolic health outcomes is lacking.

Objective

To quantify associations between concentrations of 31 PFAS and metabolic biomarkers of glucose homeostasis and β-cell function.

Methods

We used data from a 2018-2021 follow-up of the Maternal-Infant Research on Environmental Chemicals (MIREC) study, which included measurements of serum concentrations of PFAS and metabolic biomarkers in samples provided by 274 adult female participants. Our primary outcomes were composite measures of pancreatic β-cell function (proinsulin:insulin [PI:INS] and proinsulin:C-peptide [PI:CP] ratios) and insulin resistance (homeostatic model assessment for insulin resistance [HOMA-IR] and triglyceride-glucose [TyG] index). We used multivariable linear regression models to quantify the percent difference in outcome measures. Per- and polyfluoroalkyl substances with >50% detection (n = 17) were log₂-transformed; PFAS with 10-50% detection (n = 14) were dichotomized at the limit of detection. We used quantile g-computation (qgcomp) and weighted quantile sum (WQS) regression to evaluate PFAS mixtures. We also modeled arithmetic sums of 17 PFAS detected in >50% of participants (Σ17PFAS) and 7 PFAS specified in the National Academies of Sciences, Engineering and Medicine report (Σ7PFAS).

Results

Each doubling of Σ7PFAS, PFOS, and PFHxS was associated with a 5-9% increase in PI:INS ratio. Σ7PFAS, but not the Σ17PFAS, was also positively associated with the PI:INS ratio in qgcomp models. We observed inverse associations between Σ7PFAS and HOMA-IR and fasting insulin. Many results were of small magnitude or imprecise.

Conclusion

In this cross-sectional analysis, exposure to certain legacy, alternative, and precursor PFAS were associated with β-cell dysfunction.

Per- and polyfluoroalkyl substances (PFASs) are a class of environmental chemicals known for their amphipathic properties (1). Their unique grease- and water-repellent characteristics render them ideal for use in a variety of consumer and industrial products (2). Due to their chemical stability, PFAS are highly resistant to degradation (3), contributing to their environmental persistence and widespread human exposure (4-7).

PFAS exposure has been associated with numerous adverse health effects, including immunotoxicity, elevated risk of cardiovascular disease, and increased risk of certain types of cancers (8-11). Epidemiological studies have evaluated associations between PFAS exposure and insulin resistance or β-cell dysfunction, key factors that increase risk of type 2 diabetes (T2D) development in adults (12-16); however, findings remain equivocal. For example, some studies have reported positive associations between certain legacy PFAS, including perfluorooctanoic acid (PFOA) and perfluorooctane sulfonate (PFOS), and insulin resistance and/or measures of β-cell function (12, 13, 15), whereas others reported null (14, 15), or inverse associations (16), depending on the specific PFAS examined. Previous studies evaluating the relationship between PFAS concentrations and markers of glucose homeostasis or β-cell function have primarily focused on legacy PFAS such as PFOA, PFOS, and PFHxS. Due to global regulations imposed on the manufacturing and use of legacy PFAS, alternative and short-chain PFAS are increasingly used as substitutes (17, 18). The potential health effects associated with these replacement PFAS remain poorly characterized.

Our objective was to quantify cross-sectional associations between serum concentrations of 31 legacy, alternative, and precursor PFAS and markers of β-cell function and glucose homeostasis in adult female participants from a pan-Canadian cohort study. Our primary outcomes were composite markers of insulin resistance and β-cell function. Specifically, we assessed β-cell function using proinsulin:C-peptide (PI:CP) and proinsulin:insulin (PI:INS) ratios and insulin resistance using the homeostatic model assessment for insulin resistance (HOMA-IR) and triglyceride-glucose (TyG) index. Our secondary outcomes were individual biomarkers of β-cell function, insulin resistance, and glycemic control.

Materials and Methods

Study participants

The Maternal-Infant Research on Environmental Chemicals (MIREC) study recruited individuals during their first trimester of pregnancy from obstetric and prenatal clinics in 10 Canadian cities (2008-2011). Participants were eligible if they were at least 18 years of age, able to communicate in English or French, less than 14 weeks of gestation and planning to deliver at a local hospital (19). The MIREC cohort was invited to participate in a follow-up study from 2018 to 2021 (MIREC-ENDO). At the MIREC-ENDO follow-up clinic visit (n = 308), participants provided a fasting blood sample (n = 291), physical measures, including anthropometry and blood pressure, and completed a questionnaire to provide relevant sociodemographic and health history information. The present investigation included participants with both PFAS and metabolic biomarkers from the 7-9-year follow-up visit, and excluded those who self-reported taking hypoglycemic agents or insulin (n = 9), and those who were pregnant at the follow-up visit (n = 4) for a total analytical sample size of 274 participants (Fig. S1) (20).

All participants provided informed consent prior to participating. The MIREC and MIREC-ENDO studies were approved by the Research Ethics Boards at Health Canada/Public Health Agency of Canada, the Sainte-Justine's Hospital (Montreal, QC, Canada), and the institutions of all participating recruitment sites. The present investigation was approved by Health Canada's Research Ethics Board (REB 2021-003H), as well as Carleton University's Research Ethics Board (REB #115741).

Serum PFAS measurement

Blood samples were collected using 10 mL sterile vacutainer tubes. Within 2 hours of the blood draw, samples were centrifuged to separate serum and aliquoted into smaller cryovials and stored at −80 °C. Analysis of 40 legacy, alternative and precursor PFAS was completed in 2022 by AXYS Analytical Services Ltd. (Sidney, BC, Canada) using isotope dilution ultra-performance liquid chromatography-tandem mass spectrometry (AXYS method MLA-110). The laboratory is ISO 17025 accredited by the Canadian Association for Laboratory Accreditation. Further details on compound names and laboratory analyses, including batch-specific limits of detection (LODs) and quality control measures, have been published (21). Nine out of 40 PFAS had <10% detection rates and were excluded from the present analysis. For the remaining PFAS, we substituted concentrations that were <LOD using batch-specific LOD/√2.

Biomarkers of β-cell function and insulin resistance

As reported in our previous publication (22), we assessed β-cell function using PI:INS and PI:CP ratios (23, 24) and insulin resistance using HOMA-IR (25) and the TyG index (26, 27). Increased PI:INS and PI:CP ratios reflect an inefficient conversion of proinsulin to insulin or C-peptide within the β-cell, resulting in increased circulating proinsulin and a higher ratio (23, 24). Higher HOMA-IR and TyG index measurements reflect increased insulin resistance (25-27). Our secondary outcomes were the individual concentrations of fasting intact proinsulin, mature insulin, C-peptide, hemoglobin A1c (HbA1c), and glucose. Laboratory methods and quality control measures are detailed in the supplemental methods (20). Briefly, fasting serum insulin, C-peptide and intact proinsulin were measured using an enzyme-linked immunosorbent assay (C-peptide RRID: AB_3665034; Insulin RRID: AB_2894946; Proinsulin RRID: AB_3665033).

All outcome biomarkers were detected in 100% of participants.

Covariates

We identified covariates using a priori knowledge of determinants of plasma PFAS concentrations and β-cell function and insulin resistance, as shown in a directed acyclic graph (Fig. S2) (20). In our final model, we included the following potential confounders recorded at the time of the MIREC-ENDO clinic visit: age, body mass index (BMI), race and ethnicity, parity, education, cigarette smoking status, lifetime breastfeeding history, and estimated glomerular filtration rate (eGFR). BMI (kg/m²) was calculated using measured weight and height. Breastfeeding is a known excretion route of PFAS (28, 29) and is associated with increased insulin resistance postpartum (30) and long-term metabolic health (31). Lifetime breastfeeding history was derived using information from a detailed obstetrical history questionnaire. Age, BMI, lifetime breastfeeding history, and eGFR were modeled as continuous variables, while other variables were categorized (Table 1). eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (32).

Table 1.

Sociodemographic characteristics of adult female participants (n = 274) in the MIREC-ENDO study (2018-2021)

	Mean ± SD or n (%)
Age (years)	42.5 ± 4.8
BMIa (kg/m2)	26.3 ± 5.6
Lifetime breastfeeding historyb (months)	30.6 ± 24.7
eGFR (mL/min/1.73 m²)	92.2 ± 13.2
Education
College diploma or less	66 (24)
Undergraduate classes or degree	126 (46)
Graduate degree	82 (30)
Smoking status
Never	253 (92)
Occasional	9 (3.3)
Daily	12 (4.4)
Race and ethnicity
White	244 (89)
Other	25 (9.1)
Missing	5 (1.8)
Parity
1	28 (10)
2	141 (52)
≥ 3	90 (33)
Missing	15 (5.5)
Menopausal status
Premenopause	164 (60)
Perimenopause	43 (16)
Postmenopause	10 (3.6)
Unknown due to contraceptive use that prevents menstruation	44 (16)
Missing	13 (4.7)

^a Missingness for BMI (n = 25).

^b Missingness for Lifetime breastfeeding history (n = 37).

Abbreviations: BMI, body mass index; eGFR, estimated glomerular filtration rate.

Statistical analysis

Descriptive statistics

We assessed Spearman correlations among PFAS and metabolic biomarkers. We calculated the geometric mean for PFAS and metabolic outcome measures detected in >50% of participants. We visualized associations between individual PFAS and metabolic biomarkers using locally estimated scatterplot smoothing plots to assess linearity. We also assessed regression diagnostic plots to ensure the assumptions of linear regression were met.

Individual regression models

We quantified associations between PFAS and metabolic outcome biomarkers using multivariable linear regression models, adjusted for covariates. To satisfy the regression model assumption of normality of residuals, all outcome biomarkers were log₂-transformed to normalize distributions. In linear regression models, PFAS that were detected in 50% or more of the population were log₂-transformed and evaluated as continuous measures. β coefficients were back-transformed using the formula $(2^{\mathrm{\beta }}-1)\times 100$ to obtain a percent difference in outcome biomarker concentrations per doubling of PFAS concentrations.

For PFAS that were detected in between 10% and 49.9% of individuals, PFAS were dichotomized as either detected (≥LOD) or not detected (<LOD). β coefficients were back-transformed using the formula $(2^{\mathrm{\beta }}-1)\times 100$ and interpreted as the percent difference in the outcome upon detection of the specified PFAS.

We imputed missing covariate data for BMI, parity, race and ethnicity, and lifetime breastfeeding history, using multiple imputation by chained equations (number of imputations = 5) (33). Missingness varied between 0% and 13.5% (Table 1).

Since long-term metabolic health may be influenced by preexisting dysglycemia, we conducted 2 sensitivity analyses to exclude participants who had reported being diagnosed with gestational diabetes (GDM) (n = 16) or prior type 1 or type 2 diabetes (n = 2). Additionally, recognizing that menopausal status can influence metabolic health (34), we conducted a third sensitivity analysis restricted to premenopausal participants only (n = 164).

In accordance with guidance from the American Statistical Association and epidemiologists, we focus our interpretation of results on magnitude and direction of effect rather than null hypothesis testing (35-38). Our interpretation of results also focuses on consistency in results across models. Furthermore, given the limited literature on alternative and replacement PFAS in relation to biomarkers of β-cell function and glucose homeostasis, we prioritized minimizing risk of type 2 over type 1 errors. We, therefore, did not adjust for multiple comparisons (39).

Joint effects

We calculated arithmetic sums of the 17 PFAS (∑17PFAS) detected in >50% of participants as well as 7 PFAS (PFOA, PFOS, perfluorohexanesulfonate [PFHxS], perfluorononanoic acid [PFNA], perfluorodecanoic acid [PFDA], perfluoroundecanoic acid [PFUnDA], and N-Methylperfluorooctanesulfonamidoacetic acid [N-MeFOSAA]) (∑7PFAS) identified in the 2022 National Academies of Science, Engineering and Medicine (NASEM) Guidance Report on PFAS Exposure, Testing and Clinical Follow-Up (40). Upon summing raw concentrations of the specified PFAS, we log₂-transformed both arithmetic sums and modeled them in relation to each outcome biomarker.

We also applied quantile g-computation (qgcomp) regression to assess the effect of a one-quartile increase in either the 17 PFAS identified in >50% of participants or the 7 PFAS identified in the NASEM report on our outcome measures (41, 42). qgcomp is a generalized linear model-based implementation of g-computation that estimates the parameters of a marginal structural model. We prioritized qgcomp for the mixture model analysis due to the lack of directional homogeneity assumption, which enables the detection of both positive and negative associations between the specified mixture and outcome. In the qgcomp models, weights sum to 0 (negative weights sum to −1 and positive weights sum to +1).

As a secondary analysis, we evaluated PFAS as a mixture using weighted quantile sum (WQS) regression (43, 44). For WQS models, we applied 1000 bootstraps with a 40/60 validation split. We used separate WQS models to estimate the effects of a one-quartile increase in the specified PFAS mixture in either the negative or positive direction. The weights represent the pooled contribution of all PFAS within the specified mixture to either the overall positive or negative association and sum to 1 for each outcome biomarker.

We applied both qgcomp and WQS to 5 imputed datasets and calculated the pooled estimates using Rubin's Rules to appropriately account for the inter-dataset variance (45). For the qgcomp analysis, the reported weights are of the first imputed dataset only. The coefficients were back-transformed to allow for interpretation of a percent difference in the outcome per one-quartile increase in the PFAS mixture.

All statistical analyses were performed using R Statistical Software (Version 4.3.2) (46). Data were visualized using GraphPad Prism (Version 10.1.0).

Results

Study population characteristics

Participants in our analytical sample (n = 274) were predominantly White (89%), had more than a college diploma (76%), never smoked (92%), and self-reported as premenopausal (60%). They also had a mean age of 42.5 years, a mean BMI of 26.3 kg/m², and a mean lifetime duration of breastfeeding of 30.6 months (Table 1). Among PFAS detected in 50% or more of participants, geometric mean and 95th percentile concentrations (µg/L) ranged between 0.001 and 0.004 (PFMBA) to 1.70 and 4.63 (PFOS), respectively (Table 2). Among these 17 PFAS, correlations ranged between −0.24 (Perfluoropentanoic acid [PFPeA] and 6:2 fluorotelomersulfonate [FTS]) and 0.62 (PFOS and PFHxS) (Fig. S3) (20). Generally, stronger correlations were observed among the 7 PFAS identified in the NASEM report, with the exception of N-MeFOSAA (Fig. S3) (20). Among metabolic outcome biomarkers, stronger correlations were generally observed among measures of β-cell function compared to measures of insulin resistance (Fig. S4) (20). In general, concentrations of metabolic biomarkers were within reference ranges and consistent with our previous publication (Table 3) (22). Within our analytical sample, approximately 86% of participants had a ∑7PFAS concentration between 2 and 20 µg/L, which, according to authors of the NASEM report, corresponds to a potential risk of adverse health effects such as dyslipidemia, hypertensive disorders of pregnancy, and breast cancer. No participants exceeded the 20 µg/L concentration threshold, which was identified as being linked to an increased risk of health effects (40).

Table 2.

Descriptive statistics of serum concentrations (µg/L) of 31 PFAS measured in adult female participants in the MIREC-ENDO study (2018-2021)

	n	% >LOD	25th percentile	Median	75th percentile	95th percentile	GM (95% CI)
∑17PFASa	—	—	2.972	4.141	5.147	9.472	4.141 (3.875, 4.426)
∑7PFASb	—	—	2.505	3.620	4.656	7.610	3.510 (3.307, 3.725)
PFOSc	274	100	1.159	1.670	2.350	4.626	1.702 (1.586, 1.826)
PFHxSc	274	100	0.199	0.342	0.526	0.982	0.322 (0.295, 0.351)
N-EtFOSE	274	99.6	0.001	0.001	0.010	0.052	0.003 (0.002, 0.003)
PFNAc	274	99.3	0.277	0.383	0.522	0.819	0.367 (0.342, 0.395)
PFOSA	274	98.9	0.001	0.001	0.002	0.010	0.002 (0.002, 0.002)
PFOAc	274	97.5	0.445	0.711	0.962	1.533	0.508 (0.438, 0.589)
7:3 FTCA	274	97.1	0.004	0.004	0.155	1.427	0.024 (0.019, 0.031)
6:2 FTS	254	86.5	0.002	0.011	0.060	0.260	0.014 (0.011, 0.019)
PFPeA	274	79.6	0.001	0.020	0.038	0.077	0.009 (0.007, 0.011)
PFDAc	274	75.9	0.080	0.146	0.222	0.430	0.141 (0.130, 0.153)
PFBA	274	71.9	—	0.093	0.218	0.334	0.090 (0.078, 0.104)
N-MeFOSE	274	67.2	—	0.009	0.015	0.040	0.006 (0.005, 0.006)
PFHpS	274	66.4	—	0.019	0.042	0.078	0.012 (0.010, 0.014)
PFUnDAc	274	60.6	—	0.100	0.180	0.309	0.086 (0.076, 0.097)
PFMBA	274	56.6	—	0.001	0.002	0.004	0.001 (0.001, 0.001)
4:2 FTS	274	54.0	—	0.004	0.008	0.026	0.004 (0.003, 0.004)
N-MeFOSAAc	274	51.1	—	0.008	0.036	0.124	0.013 (0.011, 0.015)
8:2 FTS	274	49.6	—	—	0.046	0.332	—
HFPO-DA	274	46.0	—	—	0.014	0.035	—
PFHpA	274	39.4	—	—	0.111	0.184	—
N-EtFOSA	274	38.0	—	—	0.008	0.020	—
N-MeFOSA	274	37.6	—	—	0.010	0.027	—
PFHxA	274	35.8	—	—	0.097	0.235	—
PFBS	274	32.5	—	—	0.024	0.170	—
PFPeS	274	21.2	—	—	—	0.231	—
PFMPA	274	21.2	—	—	—	0.025	—
ADONA	274	16.4	—	—	—	0.034	—
9Cl-PF3ONS	274	16.8	—	—	—	0.121	—
N-EtFOSAA	274	12.0	—	—	—	0.075	—
PFDS	274	11.3	—	—	—	0.016	—
PFTeDA	274	10.6	—	—	—	0.296	—

Abbreviations: GM, geometric mean; LOD, limit of detection.

^a ∑17PFAS arithmetic sum of 17 PFAS detected at >50%.

^b ∑7PFAS is the arithmetic sum of 7 PFAS (PFOA, PFOS, PFHxS, PFNA, PFDA, PFUnDA, and N-MeFOSAA) outlined in the NASEM report.

^c PFAS included in the ∑7PFAS measure as indicated by the NASEM report.

Table 3.

Individual biomarkers and composite measures of β-cell function and insulin resistance among 274 adult female participants in the MIREC-ENDO study (2018-2021)

Metabolic outcomea	5th percentile	25th percentile	Median	75th percentile	95th percentile	GM (95% CI)
Proinsulin (pM)	0.72	1.02	1.33	1.96	3.20	1.43 (1.34, 1.54)
Insulin (pM)	14.7	25.1	34.4	54.3	106.8	37.5 (34.9, 40.4)
C-peptide (pM)	186	259	348	489	817	361 (341, 382)
HbA1cb	4.8	5.0	5.2	5.4	5.8	5.2 (5.2, 5.3)
Glucose (mg/dL)	77.0	83.3	89.0	96.0	107.0	89.6 (88.4, 90.8)
PI:INS ratio	0.018	0.027	0.038	0.051	0.093	0.038 (0.036, 0.041)
PI:CP ratio	0.0019	0.0028	0.0039	0.0052	0.0087	0.0040 (0.0037, 0.0042)
TyG index	7.4	7.9	8.1	8.5	9.1	8.2 (8.1, 8.2)
HOMA-IR	0.49	0.90	1.26	2.10	4.29	1.38 (1.28, 1.50)

Abbreviations: GM, geometric mean; HbA1c, hemoglobin A1c; HOMA-IR, homeostatic model of insulin resistance; LOD, limit of detection; PI:CP, proinsulin:C-peptide; PI:INS, proinsulin:insulin; TyG, triglyceride-glucose.

^a All participants were above LOD for all metabolic outcome biomarkers.

^b HbA1c was available among 272 participants.

Multivariable regression models: PFAS detected >50%

In multiple linear regression models with PFAS detected in >50% of samples, most associations were either null or of small magnitude (Figs. 1 and 2; Fig. S1) (20). However, we did observe associations with concentrations of Σ7PFAS and some alternative PFAS (Fig. 1; Table S1) (20).

Figure 1.

Percent difference (95% CI) in maternal metabolic composite measurements of β-cell function and insulin resistance, including (A) PI:INS ratio, (B) PI:CP ratio, (C) HOMA-IR, and (D) TyG index, per doubling of serum PFASs that were detected in >50% of adult female participants in MIREC-ENDO (2018-2021). Models were adjusted for maternal age, BMI, race and ethnicity, education, parity, smoking status, eGFR, and lifetime breastfeeding history.

Figure 2.

Percent difference (95% CI) in maternal metabolic biomarkers, including (A) Proinsulin, (B) Insulin, (C) C-peptide, (D) HbA1c, and (E) glucose per doubling of serum PFASs that were detected in >50% of adult female participants in MIREC-ENDO (2018-2021). Models were adjusted for maternal age, BMI, race and ethnicity, education, parity, smoking status, eGFR, and lifetime breastfeeding history.

A doubling of the Σ7PFAS concentrations was associated with 9.2% (95% CI: −0.3, 19.6) higher PI:INS ratio, and 7.5% (95% CI: −16.4, 2.3) lower fasting insulin concentrations (Figs. 1 and 2). Among the PFAS used in the Σ7PFAS, 2-fold increases in serum concentrations of PFHxS and PFOS were associated with a 7.1% (95% CI: 0.7, 14.0) and 5.5% (95% CI: −2.3, 14.0) increase in the PI:INS ratio, respectively (Fig. 1).The association between Σ7PFAS and the PI:CP ratio was of weaker magnitude (% change: 6.2 [95% CI: −2.3, 15.5]); for this outcome, we observed a positive association with PFHxS (% change: 4.8 [95% CI: −1.0, 10.9]) (Fig. 1). In addition, a doubling in Σ7PFAS concentrations corresponded to an 8.0% (95% CI: −17.7, 2.7) decrease in HOMA-IR (Fig. 1). These associations were preserved in sensitivity analyses excluding individuals with prior diabetes and GDM but were attenuated when the sample was restricted to premenopausal participants (Table S1) (20).

6:2 FTS and 7:3 fluorotelomer carboxylic acid (FTCA) were both inversely associated with some measures of β-cell function (Figs. 1 and 2); however, these results were of small magnitude (<2.5%). Similarly, 7:3 FTCA was associated with a 3% increase in both HOMA-IR and fasting insulin (Figs. 1 and 2).

Multivariable regression models: PFAS detected between 10% and 50%

Among PFAS detected between 10% and 50% of samples, the majority of associations were null or of small magnitude (Figs. 3 and 4; Table S2) (20). Detectable concentrations of N-EtFOSA and N-EtFOSAA were associated with a 12.6% (95% CI: −23.2, −0.7) and 19.0% (95% CI: −33.0, −2.1) decrease in the PI:INS ratio, respectively (Fig. 3). Detectable concentrations of PFHpA were associated with a 13.7% (95% CI: 1.1, 27.9) increase in the PI:CP ratio (Fig. 3). In addition, detectable concentrations of perfluorobutanesulfonate (PFBS) were associated with an 11.8% (95% CI: −20.8, −1.8) decrease in circulating C-peptide concentrations (Fig. 4).

Figure 3.

Percent difference (95% CI) in maternal metabolic composite measurements of β-cell function and insulin resistance, including (A) PI:INS ratio, (B) PI:CP ratio, (C) HOMA-IR, and (D) TyG index, among those with detectable vs nondetectable concentrations of PFAS detected in 10-50% of adult female participants in MIREC-ENDO (2018-2021). Models were adjusted for maternal age, BMI, race and ethnicity, education, parity, smoking status, eGFR, and lifetime breastfeeding history.

Figure 4.

Percent difference (95% CI) in maternal metabolic biomarkers, including (A) Proinsulin, (B) Insulin, (C) C-peptide (D) HbA1c, and (E) glucose among those with detectable vs nondetectable concentrations of PFAS detected in 10-50% of adult female participants in MIREC-ENDO (2018-2021). Models were adjusted for maternal age, BMI, race and ethnicity, education, parity, smoking status, eGFR, and lifetime breastfeeding history.

Detectable concentrations of 8:2 FTS were associated with a 12.0% (95% CI: −2.2, 28.3) and 3.8% (95%: 1.3, 6.4) increase in HOMA-IR and fasting glucose, respectively (Figs. 3 and 4). We did not observe any notable associations between any PFAS detected between 10% and 50% of participants and fasting insulin, HbA1c or the TyG index.

Mixtures models

In qgcomp models, a one-quartile increase in the 7 PFAS mixture was associated with a 7.2% (95% CI: −2.7, 18.9) increase in PI:INS ratio (Table 4), with PFOA and PFOS contributing the most to this association (Table S3) (20). Additionally, a one-quartile increase in the 7 PFAS mixture was inversely associated with HOMA-IR (−12.3% [95% CI: −21.5, −1.4]) (Table 4). A one-quartile increase in the same mixture was also associated with 11.1% (95% CI: −19.9, −1.4) and 10.5% (95% CI: −17.1, −2.7) decrease in the concentrations of insulin and C-peptide, respectively (Table 4). In general, results for the 7 PFAS mixture were broadly inconsistent between WQS (positive and negative) and qgcomp models (Table 4; Tables S3-S5) (20).

Table 4.

Joint associations between PFAS mixtures and individual biomarkers and composite measures of β-cell function and insulin resistance using quantile g-computation

	17 PFAS		7 PFAS
	n	Percent difference (95% CI)	n	Percent difference (95% CI)
Proinsulin	254	−20.5 (−40.5, 7.2)	274	−4.7 (−13.5, 5.7)
Insulin	254	−12.3 (−34.9, 18.1)	274	−11.1 (−19.9, −1.4)
C-peptide	254	0.0 (−20.4, 26.6)	274	−10.5 (−17.1, −2.7)
PI:INS ratio	254	−9.2 (−31.7, 21.4)	274	7.2 (−2.7, 18.9)
PI:CP ratio	254	−20.5 (−39.3, 4.3)	274	6.4 (−3.4, 16.5)
HOMA-IR	254	−14.1 (−38.0, 18.9)	274	−12.3 (−21.5, −1.4)
Glucose	254	0.7 (−2.7, 3.5)	274	0.0 (−0.7, 1.4)
HbA1c	253	−2.7 (−6.0, 0.7)	272	0.0 (−0.7, 1.4)
TyG Index	254	−2.1 (−8.0, 4.2)	274	−0.7 (−2.7, 1.4)

Values represent the percent difference per one-quartile increase in PFAS mixture. Bold estimates are statistically significant (P < .05). Models were adjusted for maternal age, BMI, race and ethnicity, education, parity, smoking status, eGFR, and lifetime breastfeeding history.

Abbreviations: HbA1c, hemoglobin A1c; HOMA-IR, homeostatic model of insulin resistance; PI:CP, proinsulin:C-peptide; PI:INS, proinsulin:insulin; TyG, triglyceride-glucose.

Among the 17 PFAS mixture, we noted several consistencies among mixture model results between qgcomp and negative-WQS, but not with positive-WQS models (Table 4; Tables S3, S4, and S6) (20). Specifically, in both mixture models, the mixture of 17 PFAS was negatively associated with proinsulin and the PI:CP ratio (Table 4; Table S4) (20), with 7:3 FTCA and N-EtFOSE contributing the most to these associations (Tables S3 and S6) (20).

Discussion

In this cross-sectional analysis of Canadian adult females enrolled in the MIREC-ENDO study, we show that the 7 PFAS included in the NASEM clinical testing guidelines were positively associated with the PI:INS ratio when modeled as an arithmetic sum and as a mixture in qgcomp models. We also observed that this sum was inversely associated with HOMA-IR in individual and qgcomp mixture models. Other associations in our analyses were either null, of small magnitude, or imprecise.

The positive association between the ∑7PFAS and PI:INS ratio is likely driven by fasting insulin; we observed an inverse relationship between ∑7PFAS and insulin, and null findings between ∑7PFAS and proinsulin. We also observed a positive association between the ∑7PFAS and the PI:CP ratio, although estimates were of lesser magnitude and greater imprecision than the PI:INS ratio. Given the inverse association between ∑7PFAS and HOMA-IR, we hypothesize that any PFAS-related deleterious effects on metabolic health may be linked to an inefficient conversion of proinsulin to insulin in pancreatic β cells rather than increased insulin resistance. While our results may seem counterintuitive, they are consistent with experimental evidence demonstrating that PFAS may directly impact insulin secretion independent of peripheral insulin resistance. Specifically, studies conducted in rodent and human β-cell lines demonstrate that PFOS can dysregulate glucose-stimulated insulin secretion (47-51). Furthermore, congruent with our findings, Yan et al observed that mice exposed to PFOA had decreased fasted serum insulin and an increase in insulin sensitivity (52). This hypothesis, however, does require further exploration and confirmation in mechanistic research using other model systems.

In contrast to the associations between ∑7PFAS and impaired β-cell function, the association between the ∑17 PFAS and indicators of β-cell function were null or inverse. These findings provide preliminary evidence that the joint effect of the ∑7PFAS, as indicated by the NASEM report, is potentially relevant to adult female metabolic health. The inclusion of 10 additional PFAS to derive the ∑17PFAS does not appear to be equated with additional risk or more precise estimates. The low detection rate and concentrations of many of the alternative and precursor PFAS likely contribute noise to the joint exposure effect estimates of this larger mixture.

The observed associations between PFAS and the PI:INS ratio are generally consistent with the limited existing literature. Authors of the Swedish Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) cohort study (2001-2004) reported positive associations between serum PFOA, but not PFOS or PFHxS, concentrations and the PI:INS ratio (15). Serum PFAS concentrations in PIVUS participants were ∼4- to 11-fold higher than MIREC concentrations (15), consistent with the older age of participants and earlier sampling period in the PIVUS study (2001-2004) compared to MIREC (2018-2021). Our findings may also differ due to the older age of PIVUS participants (PIVUS: 70 years of age vs MIREC: 42.5 years) (20, 52). The MIREC-ENDO population, therefore, represents a younger cohort that may be less impacted by age-related metabolic outcomes compared to the PIVUS cohort (19, 53).

Several studies have observed associations between PFAS and other calculated measures of β-cell function (eg, HOMA-β, HOMA2-β%), insulin resistance (eg, HOMA-IR, HOMA2-IR) and fasting insulin; however, findings are heterogenous (12-14, 16, 53). Congruent with our HOMA-IR findings, authors of a Swedish nested case–control study and the US-based Project Viva cohort both reported inverse associations between certain PFAS and measures of insulin resistance (15, 55). In contrast, authors of other studies reported positive or null associations between certain PFAS and HOMA-IR, fasting insulin, and improved β-cell function (12, 13, 54). The apparent discrepancy in findings may be due to differences in timing of data collection, serum legacy PFAS concentrations and study populations. Furthermore, our findings of proinsulin cannot be directly compared to HOMA-based assessments, which do not consider proinsulin concentrations (22, 23, 56).

Our research is the first identified epidemiological investigation of multiple precursors and alternative PFAS and measures of β-cell function and glucose homeostasis. Therefore, it is challenging to contextualize our observed associations with PFOS precursors (N-EtFOSA and N-EtFOSAA) and alternatives (PFBS and PFHpA). It is also a challenge to interpret our findings due to reliance on dichotomous measures of exposure for these PFAS. As such, these results should be considered exploratory and interpreted with caution. Furthermore, observed concentrations in those with values above the LOD were low (eg, 75th percentiles ranged between 0.008 and 0.111 µg/L).

Our analysis is strengthened by the availability of a large suite of PFAS biomonitoring data that have yet to be explored in the context of β-cell function and glucose homeostasis. Second, the prospective nature of the broader MIREC cohort allowed us to adjust for variables relating to pregnancy, including parity and lifetime breastfeeding history, which may influence PFAS burden and metabolic function. Additionally, we were also able to adjust for eGFR as a measure of kidney function in all our analyses (57, 58). As part of a sensitivity analysis, we also excluded participants based on other prior metabolic health conditions or menopausal status to assess the robustness of our findings. Given the timing of sample collection in MIREC-ENDO, our study uniquely evaluates associations between PFAS concentrations and markers of metabolic dysfunction at an age when some participants would be in the perimenopausal transition. Finally, this analysis was conducted using PFAS serum concentrations collected more recently than previous studies of PFAS exposure and β-cell function/insulin resistance, allowing us to examine associations with PFAS at more contemporary levels of exposure following decades of regulation and industry phase-out activities.

Interpretation of our results is hindered by the cross-sectional study design and corresponding lack of temporality between our exposure and outcome. Although there is potential for reverse causality in cross-sectional studies, evidence from longitudinal studies supporting an association between PFAS and T2D risk (13, 16, 59, 60) suggests that reverse causality is unlikely to be the sole explanation for our findings. Furthermore, due to the long half-lives and persistence of PFAS (61, 62), the measurements during the MIREC-ENDO study are likely reflective of prior exposures. Our analysis is subject to 4 additional limitations. First, due to the lack of available data, we were not able to control for diet quality or physical activity. Second, the narrow distributions of several metabolic outcome measures (eg, HbA1c, TyG index, and glucose) made it challenging to assess meaningful contrasts in relation to PFAS concentrations. Third, given the few postmenopausal participants in our analytical sample, we were unable to comprehensively assess differences in associations by menopausal status. Fourth, due to the lack of oral glucose tolerance test data, we were unable to directly assess peripheral insulin sensitivity and the use of a ratio makes it challenging to draw conclusions about the direction of associations with the individual biomarkers.

As with all observational epidemiology studies, it is possible that some of our observed associations are due to chance. The potential for type 1 error in our analysis is heightened by the number of modeled associations; however, as previously noted, we prioritized minimizing type 2 error due to the novelty of our research question. The generalizability of our findings is limited by the relatively homogenous sociodemographic and health profile of the MIREC study population, who are predominantly White with above average levels of education and income, and largely nonsmoking (19). The relative uniformity within the MIREC population does however minimize potential confounding by sociodemographic status. In addition, it is also worth noting that concentrations of PFAS in this population tended to be low. For example, we did not have any participants that exceeded the 20 μg/L threshold outlined by NASEM (39). It is possible that effects would be stronger in a population with higher levels of exposure but our results are not generalizable to this exposure scenario. Similarly, we also observed null associations for many alternative and precursor PFAS. Given that exposure levels for these PFAS were also quite low in this study, future studies investigating potential adverse effects in populations experiencing higher levels of exposure are warranted. Finally, a larger sample size may increase the precision of our results.

In conclusion, we provide evidence in this population that the ∑7PFAS, both as an arithmetic sum and as a mixture, is positively associated with β-cell dysfunction, and inversely associated with measures of insulin resistance. We speculate that this relationship is likely driven by lower insulin concentrations. Finally, we provide some evidence that other alternative and precursor PFAS may be associated with measures of β-cell dysfunction and insulin resistance, although further prospective studies are required to confirm these findings.

Funding

The MIREC study is supported by the Government of Canada Chemicals Management Plan, the Ontario Ministry of the Environment, and the Canadian Institutes for Health Research (CIHR) (grant MOP-81285). This study was supported by a CIHR Project Grant (PJT-186282). J.E.B. is supported by an Ontario Early Researcher Award. J.P. is supported by the Guiding interdisciplinary Research on Women’s and girls’ health and Wellbeing scholarship and Ontario Graduate Scholarship. C.P. is supported by an Investigator Grant from BC Children's Hospital Research Institute and BC Children's Hospital Foundation.

Disclosures

The authors have nothing to disclose.

Data availability

Restrictions apply to the availability of some or all data generated or analyzed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will, on request, detail the restrictions and any conditions under which access to some data may be provided.