Prenatal PFAS exposures and cardiometabolic health in middle childhood in the Maternal-Infant Research on Environmental Chemicals-ENDO (MIREC-ENDO) cohort

Abstract

Studies on prenatal exposure to per- and polyfluoroalkyl substances (PFAS) and cardiometabolic health in childhood have produced inconsistent results. In this study, we evaluated associations between prenatal PFAS exposures, individually and as a mixture, and cardiometabolic outcomes including insulin resistance, beta cell function, blood lipids, blood pressure and central adiposity during middle childhood (7–9 years of age) in a Canadian maternal-child cohort (n=281). We also explored effect measure modification based on child sex and physical activity. We quantified maternal second trimester plasma concentrations of six PFAS and measured 11 offspring cardiometabolic outcomes at a 7–9-year follow-up. In single-exposure models, ten-fold higher prenatal PFDA (β: −0.82, 95% CI: −1.36, −0.28), PFNA (β: −0.8, 95% CI: −1.41, −0.19), and PFOA (β: −0.69, 95% CI: −1.18, −0.19) concentrations were associated with lower diastolic blood pressure z-scores. This association did not persist when considering PFAS exposures as a mixture using quantile g-computation. Associations between PFAS exposures, individually or as a mixture, and other cardiometabolic outcomes were null. We observed no effect measure modification by child sex or physical activity (p-values for interaction ≥0.2). Our results contradict existing studies that suggest prenatal PFAS exposures are associated with adverse childhood cardiometabolic outcomes. Future studies should consider alternative markers of cardiometabolic health, trajectories in cardiometabolic health throughout childhood, and further explore potentially protective health behaviors.

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) are a group of thousands of anthropogenic chemicals known for their oil, water, and heat resistant properties.(Gluge et al., 2020) These characteristics, while valuable for commercial and industrial applications, contribute to their environmental persistence and certain PFAS can bioaccumulate in humans, with known biological half-lives ranging from 2 to 35 years. (Agency for Toxic Substances and Disease Registry, 2021; Brase et al., 2021) Widespread use and subsequent environmental contamination have led to nearly universal exposure to PFAS among the general US and Canadian populations.(Brunn et al., 2023; Centers for Disease Control and Prevention, 2023; Cousins et al., 2022; Dasu et al., 2022; Health Canada, 2010, 2023; Herrick et al., 2017; Lewis et al., 2015) Infants and children generally exhibit a higher PFAS body burden compared to adults as a result of placental and lactational transfer, higher body size to surface area ratios, and closer proximity to ground level dust.(Rappazzo et al., 2017) Understanding the human health effects of exposure to these environmentally persistent and ubiquitous chemicals is a pressing public health issue, particularly in early life.

PFAS exposures have been associated with numerous adverse health effects, including a cluster of cardiometabolic conditions sometimes referred to as Metabolic Syndrome or MetS. MetS includes excess central adiposity, elevated blood pressure, insulin resistance, and dyslipidemia.(Hall & Braun, 2023; Meigs, 2003) Importantly, components of MetS in childhood are strongly predictive of cardiovascular disease (CVD) outcomes and type II diabetes (T2D) in adulthood.(Jacobs et al., 2022; Pool et al., 2021; Raitakari et al., 2022; Serbis et al., 2023) In 2019, CVD and diabetes accounted for 18 and 2 million global deaths respectively with most of these deaths thought to be preventable.(World Health Organization, 2013, 2021, 2023) Therefore, identifying modifiable environmental determinants of poor cardiometabolic health, especially in children, is of utmost importance as it can inform environmental PFAS regulations and development of clinical recommendations and targeted interventions to reduce the overall disease burden of CVD and T2D and their cardiometabolic risk factors.(Ghassabian et al., 2022)

In experimental studies of rodents, prenatal exposure to PFOA and PFOS have been associated with developmental delays, growth deficits, and neonatal morbidity/mortality.(Fenton et al., 2021) Of the few animal studies that investigated the effect of prenatal PFAS exposure on offspring metabolic health, some found that prenatal PFOA of PFOS exposure is associated with alterations in growth and liver damage.(Hines et al., 2009; Marques et al., 2021; van Esterik et al., 2016) Some of these studies show that sex and maternal diet may modify these associations; this latter finding suggests that lifestyle factors could modify the effect of PFAS exposure on health.(Tompach et al., 2024) In humans, several studies have investigated associations between prenatal PFAS exposures and components of MetS in children, including excess adiposity, insulin resistance, and blood pressure.(Hall & Braun, 2023; Rappazzo et al., 2017) However, results from these studies have been mixed, likely owing to differences in statistical analyses, imprecision due to limited sample sizes, parametrization of specific MetS component conditions, and study sample characteristics.(Hall & Braun, 2023; Rappazzo et al., 2017) Additionally, one study reported that greater adolescent physical activity mitigated the adverse relationship between prenatal PFAS and MetS,(Braun et al., 2022) a finding consistent with studies of adult PFAS exposure and CVD-related traits.(Borghese et al., 2022; Cardenas et al., 2019)

The goal of this study was to evaluate associations between prenatal PFAS exposures, both as individual chemicals and as a mixture, with cardiometabolic outcomes in a Canadian pregnancy and birth cohort. As a secondary analysis, we explored effect measure modification of these relationships by physical activity to provide insight into potential areas of targeted interventions during middle childhood (7–9 years of age). Finally, well studied PFAS are recognized endocrine disruptors with potential sexually dimorphic effects.(Braun, 2017) However, research on sex-specific impacts and associated cardiometabolic outcomes remains inconsistent, with some studies reporting effects predominantly in males, others in females, and some finding no significant impact so we also explored effect measure modification by child sex. (Gaston et al., 2020; Hall & Braun, 2023)

2. Methods

2.1. Cohort and sample

We used data from the Canadian Maternal–Infant Research on Environmental Chemicals (MIREC) Research Platform. Recruitment, inclusion, and participation in the MIREC study has been previously published.(Arbuckle et al., 2013) Briefly, the MIREC study is a pregnancy and birth cohort that enrolled 2,001 pregnant persons during their first trimester of pregnancy. Recruitment for MIREC occurred at 11 study sites (10 cities) across six Canadian provinces between 2008 and 2011. The MIREC Research Platform consists of the original MIREC study, associated Biobank, and several follow-up studies, including the MIREC-ENDO study.(Arbuckle et al., 2013; Fisher et al., 2023)

To investigate cardiometabolic health in middle childhood, we used data from phase one of the MIREC-ENDO study. Phase one of MIREC-ENDO conducted follow-up of 7–9-year MIREC participants, and was designed to investigate environmental chemical determinants of puberty onset and adolescent growth. Inclusion criteria for MIREC-ENDO required participants to be in the original MIREC study, have agreed to be contacted for future research, and have consented to participation in the MIREC biobank and present study. For phase one of MIREC-ENDO, conducted 2018–2021, 584 children participated, with 311 children participating in an on-site clinical visit (in-person assessments were disrupted due to the COVID-19 pandemic). (Fisher et al., 2023) In this study, we included 281 singletons who had data on at least one cardiometabolic outcome captured between 7–11 years of age, and whose mothers had at least one second trimester plasma PFAS measurement.

2.2. Exposures

Pregnant MIREC participants provided venous blood samples during their second trimester (mean: 20 weeks; range: 15–37 weeks). Prior to analysis, plasma from these samples was stored at −20°C in polypropylene tubes. The Laboratoire de Toxicologie Institut National de Santé Publique du Quebec measured plasma concentration of six PFAS: perfluorodecanoic acid (PFDA, CAS RN: 335-76-2), perfluorohexanesulfonic acid (PFHxS, CAS RN: 355-46-4), perfluorononanoic acid (PFNA, CAS RN: 375-95-1), perfluorooctanoic acid (PFOA, CAS RN: 335-67-1), perfluorooctane sulfonic acid (PFOS, CAS RN: 1763-23-1), and perfluoroundecanoic acid (PFUdA, CAS RN: 2058-94-8). The laboratory used solid phase extraction with an anionic exchange medium to separate PFAS from plasma. Extracts were then measured using UPLC-MS-MS (Waters Acquity UPLC; tandem mass detector Waters Xevo TQ-S with MassLynx software) with an electrospray source in MRM negative mode. These maternal plasma concentrations are a valid proxy for fetal PFAS exposures as PFAS can cross the placenta and previous studies have found high correlation between maternal and child PFAS measurements.(Agency for Toxic Substances and Disease Registry, 2021; Fisher et al., 2016)

2.3. Outcomes

For this study, we examined 10 childhood cardiometabolic outcomes: waist circumference to height ratio, blood pressure (systolic and diastolic), triglycerides, cholesterol (total, HDL, and LDL), insulin resistance (HOMA-IR), proinsulin to insulin ratio, and proinsulin to c-peptide ratio.(Jacobs et al., 2022; Pool et al., 2021; Raitakari et al., 2022; Serbis et al., 2023) During MIREC-ENDO clinic visits, study staff collected biometrics including standing height (m), waist circumference (cm), and systolic and diastolic blood pressure (mmHg). We calculated waist to height ratio as $\frac{\mathit{\text{waist}}\mathit{\text{circumference}}(cm)}{\mathit{\text{standing}}\mathit{\text{height}}(m)}$. For systolic and diastolic blood pressure, we converted raw mmHg values to a percentile, conditional on the child’s sex, age, and height using a Pediatric Task Force database of normal weight children as a reference population.(National Heart, 2004) All of the above biometric measures were collected in duplicate and, if a measurement exceeded pre-determined thresholds, in triplicate. We used the average of repeat measurements in our analyses to increase precision.

In addition to direct measures, MIREC-ENDO participants also provided venous blood at clinic visits; almost all (99%) of which were fasting samples. As with maternal blood, plasma was extracted from these samples and stored at −20°C prior to analyses. The Advanced Research and Diagnostic Laboratory at the University of Minnesota measured plasma concentrations of triglycerides, HDL cholesterol, total cholesterol, and glucose. To measure triglycerides, they used Triglyceride L reagent on the Roche Cobas c502 chemistry analyzer (Roche Diagnostics, Indianapolis, IN 46250).(Wahlefeld, 1974) For total cholesterol, they used a cholesterol oxidase method (Roche Diagnostics, Indianapolis, IN 46250) on a Roche Cobas 6000 Chemistry Analyzer (Roche Diagnostics Corporation). For HDL cholesterol, they used the Roche HDL-Cholesterol 3rd generation direct method (Roche Diagnostics, Indianapolis, IN 46250) on a Roche Cobas 6000 Chemistry Analyzer (Roche Diagnostics Corporation). Using total and HDL cholesterol measurements they calculated LDL cholesterol as $LDL=TC-HDL-\frac{TG}{5.0(mg/dL)}$. (Friedewald et al., 1972) For glucose, they used the Roche hexokinase method (Roche Diagnostics, Indianapolis, IN 46250) on a Roche Cobas 6000 Chemistry Analyzer (Roche Diagnostics Corporation). Laboratory inter-assay coefficients of variation for triglyceride, total cholesterol, HDL cholesterol, and glucose QA/QC samples were all ≤3%. The Bruin Lab at Carleton University (Ottawa, Canada) measured plasma insulin, proinsulin, and c-peptide using Human Insulin Chemiluminescence ELISA kits (#80-INSHU-CH01, ALPCO Diagnostics, RRID: AB_2894946), Human Intact Proinsulin ELISA kits (#TE1012, TECO Medical Group, Sissach, Switzerland), and C-peptide Chemiluminescence ELISA kits (#80-CPTHU-CH01, ALPCO Diagnostics, Salem, NH) respectively. The intra- and inter-assay coefficients of variation for insulin, proinsulin, and c-peptide QC/QA samples ranged from 4 to 12%. Eight proinsulin measurements were below the limit of detection (0.15 pmol/L) so we imputed these values as $\frac{LOD}{\sqrt{2}}$. (Hornung & Reed, 1990) As a measure of insulin resistance, we used homeostasis model assessment (HOMA-IR) calculated as $\frac{\mathit{\text{glucose}}(\frac{mg}{dL})+\mathit{\text{insulin}}(\frac{mU}{L})}{405}$. Higher values on the HOMA-IR scale indicate more insulin resistance.(Matthews et al., 1985) As measures of pancreatic beta cell function, we calculated proinsulin to insulin ratio as $\frac{\mathit{\text{Plasma}}\mathit{\text{proinsulin}}(\frac{\mathit{\text{pmol}}}{L})}{\mathit{\text{Plasma}}\mathit{\text{insulin}}(\frac{\mathit{\text{pmol}}}{L})}$ and proinsulin to c-peptide as $\frac{\mathit{\text{Plasma}}\mathit{\text{proinsulin}}(\frac{\mathit{\text{pmol}}}{L})}{\mathit{\text{Plasma}}c-\mathit{\text{peptide}}(\frac{\mathit{\text{pmol}}}{L})}$. Higher values of these ratios are considered a risk factor for T2D and indicate less conversion of proinsulin to mature insulin because of lower pancreatic beta cell function.(Kahn et al., 1995; MykkÄnen et al., 1995; Pradhan et al., 2003; Serbis et al., 2023)

2.4. Covariates

We chose our covariates as confounders or precision variables based on an a priori directed acyclic graph (supplemental Figure 1). These included study site; baseline maternal sociodemographic characteristics: age (continuous years), country of birth (Canada/elsewhere), race (White/other race), educational attainment (high school or less/college+), income (CA$), and parity (nulliparous/primiparous or multiparous); maternal pre-pregnancy BMI (continuous kg/m²); maternal first trimester plasma cotinine (<5.2 ng/mL/≥5.2 ng/mL); self-reported maternal daily supplemental folic acid intake during pregnancy (continuous mcg); baseline maternal diet quality score (continuous); and maternal first trimester fish consumption (times per month). Information on all these covariates was collected via questionnaire during the original MIREC study, (2008–2012) except for first trimester plasma cotinine. First trimester cotinine was measured in plasma samples via two LC-MS/MS methods with differing sensitivities depending on self-reported smoking status where non-smoker samples were analyzed using the more sensitive method. We refer the reader to Arbuckle et al., 2018 for a thorough explanation of these methods.(Arbuckle et al., 2018) Diet quality score was estimated from a baseline abbreviated food frequency questionnaire designed to assess intake of iron, folate and vitamin D. (Morisset et al., 2016) The calculation of diet quality score was based on the Healthy Eating Index, which calculates diet quality benchmarked against the United States Department of Agriculture’s Dietary Guidelines for Americans. Higher scores indicate adequate consumption of total fruits, whole fruits, total vegetables, greens and beans, whole grains, dairy, total protein foods, seafood and plant proteins, fatty acids and moderated consumption of refined grains, sodium and empty calories.(National Cancer Institute, 2024) We did not adjust for the week of blood sample collection, but would not expect any resulting confounding of our effect estimates. While some studies have shown PFAS levels vary throughout pregnancy,(Luo et al., 2024) variability in the week in which mothers had blood samples drawn is unlikely to be related to offspring cardiometabolic health and thus, cannot be a confounder of the causal association.(VanderWeele, 2019)

2.5. Effect measure modifiers

We examined effect measure modification by physical activity level and child sex at birth. For physical activity, we used average time (minutes per day) of moderate or vigorous physical activity. Moderate or vigorous physical activity was considered as time spent at an energy expenditure > 3 metabolic equivalents of tasks (METs) as captured by an ActiGraph GT3X+ accelerometer worn at the right mid-axillary line on a belt around their waist for at least 7 consecutive days for 24-hours per day.(Borghese et al., 2017) ActiGraph GT3X+ data were initialized at midnight, collected at a sampling rate of 80 Hz, downloaded in 1-s epochs, and aggregated to 15-s epochs. Time spent sleeping, as well as non-wear time defined as 20 consecutive minutes of zero counts, ending with any non-zero minute, was removed prior to deriving estimates of movement behaviors.(Trost, 2000) A valid measurement day was defined as having ≥10 hours of accelerometer data. Participants had to have at least 4 valid measurement days to be included.(Trost, 2000) Cutoffs for activity intensity based on accelerometer data were vigorous: ≥1003 counts/15 s, moderate: between <1003 counts and ≥574 counts /15 s, light: between <574 counts and >25 counts /15 s, and sedentary: <25 counts/15 s.(Evenson et al., 2008; Wong et al., 2011) To explore effect measure modification we dichotomized physical activity according to Canadian Society for Exercise Physiology guidelines for this age group (< 60 minutes per day on average vs. ≥ 60 minutes per day on average).(Tremblay et al., 2016)

2.6. Statistical analysis

2.6.1. Descriptive analysis

To investigate potential selection bias due to study attrition or missing data, we first examined the distribution of PFAS concentrations and participant characteristics for 1,537 live-born, singleton MIREC participants from ENDO eligible study sites for whom we had data on at least child sex, stratified by whether they were included in this analysis. We generated descriptive statistics for all exposures and outcomes in both the MIREC study and Phase 1 of the MIREC-ENDO study. We also calculated the sum of PFDA, PFHxS, PFNA, PFOA, PFOS, and PFUdA and reported the median and IQR for this measure, stratified by categorical covariates and continuous covariates dichotomized at the median.

2.6.2. Missing Data Imputation

To address missing data, we used multiple imputation by chained equations (MICE) using R’s mice package.(van Buuren & Groothuis-Oudshoorn, 2011) We imputed values for all variables with missingness using a classification and regression algorithm, with a complexity parameter of 0.05 and a minimal terminal node size of seven. This approach is similar to the frequently used predictive mean matching method except predictive means are estimated using a CART algorithm instead of a traditional regression model.(van Buuren, 2018) We used all exposures, outcomes, covariates, and effect modifiers as predictors in our classification and regression trees, except for gestational glucose tolerance, daily supplemental folic acid intake, and child daily moderate to vigorous physical activity due to large amounts of missingness (31, 14, and 44% respectively). We pooled estimates from 20 imputed datasets over five iterations using Rubin’s rule(van Buuren & Groothuis-Oudshoorn, 2011) and used data visualizations for quality control.

2.6.3. Regression analysis

To standardize outcomes across age and sex, we regressed each cardiometabolic outcome on child age (years) and sex at birth (female/male), extracted the residuals, and transformed them into z-scores. We natural log-transformed triglycerides and HOMA-IR to approximate the normality assumption of our models. We then used these age- and sex-specific z-scores as the dependent variables in all subsequent regression models.

To estimate the association between individual PFAS and cardiometabolic outcomes, we ran linear regression models for each log₁₀ transformed plasma PFAS independently. For each cardiometabolic outcome-PFAS pair, we fit three models: a crude model; a minimally adjusted model including study site and maternal sociodemographic characteristics; and a fully adjusted model with adjustment for all covariates. Coefficients from these regression models represent the mean difference in age- and sex-specific z-scores for a 10-fold increase in plasma PFAS concentration.

We used quantile-based g-computation (qgcomp) to estimate associations between our PFAS mixture and cardiometabolic outcomes. Quantile-based g-computation is a method for estimating the health effects of a mixture of correlated exposures. The details of this method have been previously discussed.(Keil et al., 2020) In brief, continuous exposures (in this study, 2^nd trimester plasma PFAS concentrations) are quantized and the outcome of interest is regressed onto an optimally weighted sum of quantized exposures and covariates. A marginal structural model is then employed to estimate ψ, representing the effect of a one unit increases in all quantized exposures simultaneously.(Keil et al., 2020) In our case, a quartile split was employed for quantizing all PFAS exposures of interest. For our crude, minimally adjusted, and fully adjusted models, we fit qgcomp with the same adjustment sets as our individual exposure models. Lastly, we conducted an exploratory analysis to assess potential PFAS exposure interactions. To accomplish this, we used a backwards model selection approach based on Akaike Information Criteria (AIC). For each cardiometabolic outcome, we started the backwards selection procedure with a saturated model including all pairwise interaction terms between PFAS exposures. We forced PFAS exposure main effects and covariates into the final selected model. For all best fit models where at least one PFAS interaction term was retained, we used qgcomp to estimate ψ and ψ² terms. We used these models to generate plots of predicted cardiometabolic outcome z-scores. For these plots, we used non-parametric bootstrapped 95% confidence intervals from 1,00 iterations across four joint exposure quantiles. All statistical analyses were performed using R version 4.4.1 and a number of user created packages.(Keil, 2024; Long, 2022; R Core Team, 2024; van Buuren & Groothuis-Oudshoorn, 2011; Venables & Ripley, 2002; Wei & Simko, 2024)

3. Results

Of 1,537 live born singleton children originally enrolled at a follow-up eligible MIREC study site with baseline data on at least child sex, 281 (18%) met our inclusion criteria. Compared to those not included, these participants were more likely to have mothers who were White (90 vs. 83%), born in Canadian (87 vs. 84%), college educated (92 vs. 82%), and female sex at birth (55 vs. 46%). (Table 1)

Table 1:

Participant characteristics stratified by inclusion in analytic sample.^a

	Included (n= 281)	Excluded (n=1,256)	Total* (n=1,537)	p-value†
Maternal age at baseline (years),
mean (sd)	32.6 (4.7)	31.8 (5.2)	31.9 (5.1)	0.013
Missing	0	45	45
Maternal race
White	254 (90.4)	1,043 (83.0)	1,297 (84.4)
Other	27 (9.6)	213 (17.0)	240 (15.6)	0.003
Maternal country of origin
Canada	244 (86.8)	1,020 (84.2)	1,264 (84.7)
Elsewhere	37 (13.2)	191 (15.8)	228 (15.3)	0.317
Missing	0	45	45
Maternal education
High school or less	22 (7.8)	224 (18.5)	246 (16.5)
College +	259 (92.2)	985 (81.5)	1,244 (83.5)	<0.001
Missing	0	47	47
Income (CA$), median [iqr]	90,000 [65,000, 110,000]	90,000 [55,000, 110,000]	90,000 [55,000, 110,000]	<0.001
Missing	7	104	111
Parity
Nulliparous	128 (45.6)	533 (44.0)	661 (44.3)
Primiparous/Multiparous	153 (54.4)	678 (56.0)	831 (55.7)	0.688
Missing	0	45	45
Child sex
Male	127 (45.2)	681 (54.2)	808 (52.6)
Female	154 (54.8)	575 (45.8)	729 (47.4)	0.008
Gestational age (weeks)
>=39 weeks	199 (71.1)	834 (68.9)	1,033 (69.3)
<39 weeks	81 (28.9)	376 (31.1)	457 (30.7)	0.529
Missing	1	46	47
Mode of delivery
Spontaneous	174 (62.6)	788 (65.2)	962 (64.7)
Instrumental vaginal	25 (9.0)	99 (8.2)	124 (8.3)
Cesarean section	79 (28.4)	321 (26.6)	400 (26.9)	0.704
Missing	3	48	51
Second trimester PFDA (μg/L),
median [iqr]	0.2 [0.1, 0.3]	0.2 [0.1, 0.3]	0.2 [0.1, 0.3]	0.549
Missing	0	197	197
Second trimester PFHxS (μg/L),
median [iqr]	0.8 [0.4, 1.2]	0.8 [0.5, 1.2]	0.8 [0.5, 1.2]	0.536
Missing	0	192	192
Second trimester PFNA (μg/L),
median [iqr]	0.5 [0.4, 0.7]	0.6 [0.4, 0.8]	0.6 [0.4, 0.8]	0.191
Missing	0	192	192
Second trimester PFOA (μg/L),
median [iqr]	1.5 [0.9, 2.0]	1.4 [1, 2]	1.4 [1, 2]	0.883
Missing	0	192	192
Second trimester PFOS (μg/L),
median [iqr]	4 [2.9, 5.9]	4.3 [2.9, 6.2]	4.3 [2.9, 6.2]	0.222
Missing	0	192	192
Second trimester PFUDA (μg/L),
median [iqr]	0.1 [0.1, 0.2]	0.1 [0.1, 0.2]	0.1 [0.1, 0.2]	0.179
Missing	0	192	192

^*All live-born, singleton MIREC participants from ENDO eligible study sites for whom we had data on at least child sex.

^†P-values from Chi-square test, two sample t-test, and Wilcoxon Rank Sum test where appropriate.

^a-Missing is N.

At the MIREC-ENDO follow-up, children were 8 years old on average (range: 7–11 years). According to guidelines set by the National Heart, Lung and Blood Institute’s Expert Panel on Integrated Guidelines for Cardiovascular Health and Risk Reduction in Children and Adolescents, median blood concentrations of total cholesterol, LDL, HDL, and triglycerides in this cohort are in normal, typical rages.(Khoury et al., 2022) Children in this study also had notably higher diastolic blood pressure compared to the Pediatric Task Force population used to calculate sex, age, and height standardized percentiles. (Table 2)

Table 2:

Descriptive statistics for cardiometabolic outcomes stratified by child sex.^a

	Male (n=127)	Female (n=154)	Total (n=281)
HOMA-IR, median [iqr]	1.2 [0.9, 1.8]	1.2 [0.7, 1.8]	1.2 [0.8, 1.8]
Missing	42	42	84
Proinsulin (pmol/L)	1 [0.5, 1.5]	1 [0.6, 1.6]	1 [0.6, 1.6]
Missing	48	49	97
Proinsulin to insulin ratio*	2.5 [1.6, 4.5]	2.8 [1.7, 4.5]	2.7 [1.6, 4.5]
Missing	48	49	97
Proinsulin to c-peptide ratio*	0.5 [0.3, 0.8]	0.4 [0.3, 0.8]	0.5 [0.3, 0.8]
Missing	48	49	97
Total cholesterol (mg/dL), median [iqr]	151 [136, 172]	151 [134, 170]	151 [134, 171]
Missing	34	36	70
LDL cholesterol (mg/dL), median [iqr]	78 [65, 95]	79.5 [60.0, 92.5]	79 [61.5, 94.5]
Missing	34	36	70
HDL cholesterol (mg/dL), median [iqr]	61 [54, 69]	59 [53, 67]	59 [53.0, 67.5]
Missing	34	36	70
Triglycerides (mg/dL), median [iqr]	54 [40, 70]	58.5 [47, 76]	57 [44.0, 73.5]
Missing	34	36	70
Systolic blood pressure (percentile†), median [iqr]	50.5 [24.0, 69.8]	56 [36.0, 74.5]	54 [29, 73]
Missing	13	11	24
Diastolic blood pressure (percentile†), median [iqr]	65 [47, 80]	73 [57, 86]	69 [53, 84]
Missing	13	11	24
Waist to height ratio, median [iqr]	0.5 [0.4, 0.5]	0.4 [0.4, 0.5]	0.4 [0.4, 0.5]
Missing	10	10	20

^*Ratios multiplied by 100 for readability

^†Percentile conditional on the child’s sex, age, and height using the Pediatric Task Force database as a reference population

^a-Missing is n.

All six PFAS were detectable in 100% of samples. Median second trimester plasma concentrations were highest for PFOS (median: 4 μg/L, IQR: 2.9, 5.9), followed by PFOA (median: 1.5 μg/L, IQR: 0.9, 2.0), and lowest for PFUdA (median: 0.1 μg/L, IQR: 0.1, 0.2) and PFDA (median: 0.2 μg/L IQR: 0.1, 0.3). (Table 1) Median (IQR) PFAS concentrations were similar in participants included and excluded from our analysis. Few participant characteristics were substantive predictors of PFAS exposure. (Supplemental Table 1) Most PFAS concentrations were highly correlated, with Spearman correlations coefficients ranging from 0.03 (PFUdA and PFHxS) to 0.82 (PFNA and PFDA). (Supplemental Figure 2) The cardiometabolic outcomes we considered were mostly weakly correlated with the strongest correlations between measures of cholesterol, blood pressure, and beta cell function. (Supplemental Figure 3)

In fully adjusted single exposure models, each ten-fold higher concentration of PFDA (β: −0.82, 95% CI: −1.36, −0.28), PFNA (β: −0.8, 95% CI: −1.41, −0.19), and PFOA (β: −0.69, 95% CI: 1.18, −0.19) was associated with lower diastolic blood pressure. (Figure 1 and supplement Table 2) For both single exposure and qgcomp models, all other PFAS-cardiometabolic outcome association point estimates were close to the null and all confidence intervals included the null. (Figures 1 and 2) Results from crude and minimally adjusted linear and qgcomp models were comparable with fully adjusted models. (supplement Tables 2 and 3)

Figure 1:

Beta coefficients from fully adjusted models regressing conditional cardiometabolic outcome z-scores at ages 7–11 years on log₁₀ transformed 2nd trimester maternal plasma PFAS. Betas represent difference in means outcome z-score per 10-fold increase in PFAS concentration. Models adjusted for study site; maternal sociodemographic characteristics: age at baseline (continuous years), country of birth (Canada/elsewhere), race (White/other race), educational attainment (high school or less/college+), income (CA$), and parity (nulliparous/primiparous or multiparous); maternal pre-pregnancy BMI (continuous kg/m²); maternal first trimester plasma cotinine (<5.2 ng/mL/≥5.2 ng/mL);(Arbuckle et al., 2018) maternal daily supplemental folic acid intake during pregnancy (continuous mcg); maternal diet quality score (continuous); and maternal first trimester fish consumption (times per month).

Figure 2:

Psi coefficients from fully adjusted quantile g-computation models regressing conditional cardiometabolic outcome z-scores at ages 7–11 years on a mixture of 6, untransformed, 2nd trimester maternal plasma PFAS. Psi coefficient represents mean change in outcome simultaneous increase in all PFAS by one quartile. Models adjusted for study site; maternal sociodemographic characteristics: age at baseline (continuous years), country of birth (Canada/elsewhere), race (White/other race), educational attainment (high school or less/college+), income (CA$), and parity (nulliparous/primiparous or multiparous); maternal pre-pregnancy BMI (continuous kg/m²); maternal first trimester plasma cotinine (<5.2 ng/mL/≥5.2 ng/mL);(Arbuckle et al., 2018) maternal daily supplemental folic acid intake during pregnancy (continuous mcg); maternal diet quality score (continuous); and maternal first trimester fish consumption (times per month).

Effect measure modification by child sex or physical activity was not statistically significance (all p-values for interaction terms ≥0.2). (Figure 3 and supplement Table 4) In our analysis exploring potential interactions between individual PFAS comprising the mixture, we found that, in some cases, models with interaction terms were an improvement over models with no interaction terms with respect to AIC (supplement Table 5). However, visualizations from these models indicate the deviations from linearity was negligible. (Supplement Figure 4)

Figure 3:

Psi coefficients from fully adjusted quantile g-computation models regressing conditional cardiometabolic outcome z-scores at ages 7–11 years on a mixture of 6, untransformed, 2nd trimester maternal plasma PFAS stratified by child sex at birth and physical activity. Psi coefficient represents mean change in outcome simultaneous increase in all PFAS by one quartile. Models adjusted for study site; maternal sociodemographic characteristics: age at baseline (continuous years), country of birth (Canada/elsewhere), race (White/other race), educational attainment (high school or less/college+), income (CA$), and parity (nulliparous/primiparous or multiparous); maternal pre-pregnancy BMI (continuous kg/m²); maternal first trimester plasma cotinine (<5.2 ng/mL/≥5.2 ng/mL);(Arbuckle et al., 2018) maternal daily supplemental folic acid intake during pregnancy (continuous mcg); maternal diet quality score (continuous); and maternal first trimester fish consumption (times per month).

4. Discussion

In this Canadian pregnancy and birth cohort, we found higher prenatal exposure to PFDA, PFNA, and PFOA was associated with lower diastolic blood pressure in middle childhood. When considering PFAS exposures as mixture, we found no convincing PFAS-cardiometabolic outcome associations. Lastly, we observed little evidence of effect measure modification by child sex or physical activity, or of substantive interactions between individual components of the PFAS mixture although our analysis of PFAS interactions was exploratory given limited sample size.

Our findings are somewhat inconsistent with previous studies that found higher prenatal PFAS exposures were associated with higher adiposity measures in childhood and adolescence.(Braun et al., 2016; Liu et al., 2020; Liu et al., 2023; Starling et al., 2024) Furthermore, some studies in the U.S. and China have found sexually dimorphic associations between prenatal PFAS exposures and childhood adiposity, a finding consistent with some prior animal studies,(Fenton et al., 2021) whereas our results were consistently null in girls and boys. (Chen et al., 2019; Mora et al., 2017; Zhang et al., 2022)

Fewer studies have examined the association of prenatal PFAS exposures with cardiometabolic outcomes other than adiposity. In this study, we observed no association between prenatal PFAS exposures and insulin resistance as measured by HOMA-IR, contradicting several existing studies.(Li et al., 2021; Valvi et al., 2023; Valvi et al., 2021) For example, a study by Li et al. found that prenatal exposure to PFOA was associated with higher HOMA-IR in 12 year-olds in Cincinnati, Ohio.(Li et al., 2021) When considering PFAS individually, we unexpectedly found that, several PFAS were associated with lower diastolic blood pressure in middle childhood. Notably, a study by Zhang et al. in the Massachusetts-based maternal-child cohort, Project Viva, found no such association. However, they did find that higher prenatal exposure to EtFOSAA was associated with lower diastolic blood pressure in infancy.(Zhang et al., 2023) In contrast, Du et al. observed a positive association between prenatal PFAS mixture exposure and diastolic blood pressure at age 4 in a birth cohort from Shanghai, China.(Du et al., 2024)

Given the large body of research linking prenatal and early-life PFAS exposures to health outcomes, it is critical not only to prevent or reduce exposure, but to identify protective interventions, especially in highly exposed populations with historical exposure. A 2022 study by Braun et al. reported that physical activity modified the associations between PFAS exposures and childhood cardiometabolic outcomes but only when physical activity was measured using the self-reported Physical Activity Questionnaire for Children (PAQ-C).(Braun et al., 2022) However, both Braun et al. and our study found no evidence of effect measure modification by physical activity measured with the accelerometer-based data. Notably, both our study and Braun et al had relatively small samples sizes to evaluate effect measure modification. As such, larger future studies should investigate effect measure modification of these associations by physical activity as well as other modifiable behaviors like diet.

How PFAS might influence cardiometabolic health is unclear. However, it is well documented that, in animal studies, PFAS interact with humanized peroxisome proliferator-activated receptor (PPAR)-α and PPAR-γ.(Evans et al., 2022) PPARs are a family of transcription factors involved in many physiological processes in humans including lipid and glucose metabolism whose dysregulation could be a contributing factor in CVD.(Lin et al., 2022) Other studies suggest that prenatal PFAS have persistent effects on leukocyte DNA methylation of genes related to cancers, cognitive health, cardiovascular disease, and kidney function.(Liu et al., 2022) However, the inherent complexity of biological systems makes understanding potentially biological pathways of PFAS action difficult.

This is the first study to examine the association between low-level prenatal PFAS exposures and middle childhood cardiometabolic health in a Canadian birth cohort. We also qgcomp to estimate main and interaction effects of prenatal PFAS exposure mixtures, something few previous studies have done.(Du et al., 2024; Li et al., 2021; Liu et al., 2023; Starling et al., 2024; Zhang et al., 2023; Zhang et al., 2022) This is important as PFAS exposure does not typically occur in isolation, thus, evaluation of PFAS as a mixture provides better insight into how these chemicals can affect health outcomes. Other strengths of our study include the use of objectively measured child physical activity to explore effect measure modification and impressive longitudinal follow-up when children were 7–11 years of age paired with exposure assessment in utero. The latter provides insight into potentially protracted effects of prenatal PFAS exposures.

This study has some limitations. First, while our results are not subject to collider stratification bias (aka selection bias) given that inclusion in the analytic sample was not related to prenatal PFAS exposures,(Hernan et al., 2004) the MIREC cohort is also homogenous with respect to sociodemographic characteristics which may limit the generalizability of our findings.(Hernan, 2017) We used an objective accelerometer-based measure of physical activity, but we had substantial missing data (n=124). Lastly, follow-up for the MIREC-ENDO study was interrupted by the COVID-19 pandemic. As such, some of the data collection for MIREC-ENDO was transitioned to remote study visits during which objective measures of the cardiometabolic outcomes were not obtained. This led to a modest sample size of 281 participants, increasing random error in our findings, and hindering definitive conclusions.

5. Conclusions

When viewed broadly, studies of prenatal PFAS exposures and childhood cardiometabolic health provide mixed conclusions. Our study adds to the weight of evidence indicating prenatal PFAS exposures, at the concentrations observed in the MIREC cohort, do not influence cardiometabolic outcomes in middle childhood, contradicting some existing studies. Future studies should consider alternative markers of cardiometabolic health including body mass index, body fat percentage, and markers of inflammation; trajectories of cardiometabolic health throughout childhood; and further explore potentially protective health behaviors.

Highlights.

• In a Canadian maternal-child cohort, we examined exposure to six PFAS in utero in relation to 11 cardiometabolic outcomes in middle childhood

• Prenatal exposure to PFAS individually and as a mixture was not convincingly associated with adverse cardiometabolic health.

• We also investigated potential effect modification by child sex and physical activity, finding neither.

• Our results contradict some existing studies, particularly of prenatal PFAS exposures and childhood adiposity.