Prospective associations of mid-childhood plasma per- and polyfluoroalkyl substances and pubertal timing

Abstract

Background

Exposure to per- and polyfluoroalkyl substances (PFAS) may disrupt pubertal timing. Higher PFAS plasma concentrations have been associated with later pubertal timing in girls, but cross-sectional findings may be explained by reverse causation.

Objectives

To assess prospective associations between PFAS plasma concentrations in mid-childhood and markers of pubertal timing in male and female adolescents.

Methods

We studied 640 children in Project Viva, a Boston-area prospective cohort. We examined associations of plasma concentrations of 6 PFAS measured at mean 7.9 (SD 0.8) years (2007–2010) with markers of pubertal timing. Parents reported a 5-item pubertal development score at early adolescence (mean 13.1 (SD 0.8) years) and reported age at menarche annually. We calculated age at peak height velocity using research and clinical measures of height. We used sex-specific linear and Cox proportional hazards regression to estimate associations of single PFAS with outcomes, and we used Bayesian Kernel Machine Regression (BKMR) to estimate associations of the PFAS mixture with outcomes.

Results

Plasma concentrations were highest for perfluorooctane sulfonate (PFOS) [median (IQR) 6.4(5.6) ng/mL], followed by perfluorooctanoate (PFOA) [4.4(3.0) ng/mL]. In early adolescence, girls were further along in puberty than boys [pubertal development score mean (SD) 2.9 (0.7) for girls and 2.2(0.7) for boys; age at peak height velocity mean (SD) 11.2y (1.0) for girls and 13.1y (1.0) for boys]. PFAS was associated with later markers of pubertal timing in girls only. For example, each doubling of PFOA was associated with lower pubertal development score (−0.18 units; 95% CI: −0.30, −0.06) and older age at peak height velocity (0.23 years; 95% CI: 0.06, 0.40)]. We observed similar associations for PFOS, perfluorodecanoate (PFDA), and the PFAS mixture. PFAS plasma concentrations were not associated with age at menarche or markers of pubertal timing in boys.

Discussion

Higher PFAS plasma concentrations in mid-childhood were associated with later onset of puberty in girls.

Introduction

Per- and polyfluoroalkyl substances (PFAS) are a group of synthetic chemicals with water, oil, and dirt resistant properties used in the manufacture of everyday consumer products, including food packaging, nonstick cookware, and clothing (Lindstrom et al. 2011). As a result of both pervasive exposure to and the biological persistence of perfluorooctane sulfonate (PFOS), perfluorooctanoate (PFOA), and other long-alkyl chain PFAS (Olsen et al. 2007), nearly all U.S. children and adolescents have detectable concentrations of these PFAS in their blood (Calafat et al. 2019; Ye et al. 2018).

PFAS exposure has been linked to endocrine health outcomes, including markers of pubertal timing (Rappazzo et al. 2017; White et al. 2011). PFAS may delay puberty in girls by lowering estradiol concentrations (Knox et al. 2011; Shi et al. 2009), and in boys by lowering testosterone concentrations (Lopez-Espinosa et al. 2016) and antagonizing the androgen receptor (Kjeldsen and Bonefeld-Jorgensen 2013). Consistent with this physiology, some, but not all, rodent studies (Butenhoff et al. 2004; Lau et al. 2006; Yang et al. 2009) have shown prenatal PFOA exposure to be associated with delayed vaginal opening in female offspring (Lau et al. 2006; Yang et al. 2009) and delayed preputial separation in male (Butenhoff et al. 2004) offspring. Epidemiologic studies similarly support an association between maternal PFAS plasma concentrations during pregnancy and delayed pubertal development (e.g., menarche) in offspring, particularly girls (Christensen et al. 2011; Kristensen et al. 2013). Only one previous study has investigated the association between childhood PFAS plasma concentrations and markers of pubertal timing. In a cross-sectional analysis of the C8 Health Project, girls with higher PFAS concentrations were less likely to have achieved menarche compared to their peers (Lopez-Espinosa et al. 2011); however, cross-sectional associations may be explained by reverse causation related to potential decreases in PFAS plasma concentrations as body volume increases during puberty (Wu et al. 2015). Prospective studies are needed to better understand associations between childhood PFAS concentrations and pubertal timing. Here, we investigate prospective associations of PFAS plasma concentrations measured in mid-childhood with markers of pubertal timing in adolescents in the Project Viva cohort, a Boston-area prospective cohort. In addition to incorporating several markers of pubertal development in boys and girls, our study is the first to relate complex PFAS mixtures to markers of pubertal timing. We hypothesized that higher PFAS plasma concentrations would be associated with later puberty in both boys and girls.

Materials and Methods

Study population

Participants were enrolled in Project Viva, a Boston-area prospective pre-birth cohort. Women were recruited between 1999 and 2002 at their first prenatal visit at one of eight participating Atrius Harvard Vanguard Medical Associates clinics (Oken et al. 2015). Women were considered eligible to participate in Project Viva if they had a singleton pregnancy, were <22 weeks gestation at their initial visit, could participate in English, and planned to stay in the area following delivery. Of 2,128 mother-infant dyads, 1,122 (52.7%) had follow-up in mid-childhood. Among the 653 children who had a blood draw in mid-childhood [mean 7.9 (SD 0.8) years (2007–2010)], 640 (98%) reported at least one pubertal marker and were therefore eligible for this analysis. Participants included in the analysis (N=640) were more likely to have an annual household income below $70,000 and had a slightly different race/ethnicity distribution compared to excluded participants (N=482), but otherwise appeared similar (Table S1).

The institutional review boards of participating institutions approved this study. Mothers provided written informed consent, and children provided verbal assent at the mid-childhood visit. The involvement of the Centers for Disease Control and Prevention (CDC) did not constitute engagement in human subjects research.

Quantification of PFAS plasma concentrations

We measured PFAS concentrations in plasma collected in mid-childhood in 2007–2010, as described previously (Harris et al. 2017). Samples were shipped overnight to the laboratory at the CDC to measure PFAS using online solid-phase extraction coupled with isotope dilution high-performance liquid chromatography-tandem mass spectrometry, the same method used to measure PFAS concentrations in 2013–2014 National Health and Nutrition Examination Survey samples (CDC 2013). Low-concentration quality control materials (QCs) and high-concentration QCs, prepared from a calf serum pool, were analyzed with the study samples, analytical standards, and with reagent and serum blanks to ensure the accuracy and reliability of the data (Kato et al. 2011). We replaced concentrations below the limit of detection (LOD) of 0.1 ng/mL (Table S2) with the $\text{LOD/}\sqrt{2}$ (Hornung and Reed 1990). We calculated total PFOA as the sum of its isomers [i.e., n-PFOA and the sum of perfluoromethylheptanoates and perfluorodimethylhexanoates (Sb-PFOA)]. Similarly, we calculated total PFOS as the sum of its isomers [i.e., n-PFOS, sum of perfluoromethylheptane sulfonates (Sm-PFOS), and sum of perfluorodimethylhexane sulfonates (Sm2-PFOS)]. We a priori decided to restrict analyses to PFAS detected in >60% of samples, which included n-PFOA, n-PFOS, Sm-PFOS, PFOA, PFOS, perfluorodecanoate (PFDA), perfluorohexane sulfonate (PFHxS), 2-(N-methyl-perfluorooctane sulfonamide) acetate (MeFOSAA), and perfluorononanoate (PFNA).

Age at peak height velocity

Linear growth accelerates during puberty due to activation of the hypothalamic-pituitary axis, and age at peak height velocity is a common marker of pubertal timing. We estimated age at peak height velocity as previously described (Aris et al. 2019). Briefly, we used longitudinal height data from multiple research visits and medical records [mean (range): 13.3 (3 to 34) height measurements per child; age range: 2.5–18.3 years] and fitted subject-specific height growth curves across childhood and adolescence using the SITAR (SuperImposition by Translation and Rotation) growth model (Cole et al. 2010). Subsequently, we differentiated the individually predicted height curves and located the maximum inflection point during adolescence for each individually differentiated curve to obtain the age at peak height velocity.

Pubertal development score

We calculated a pubertal development score using a 5-item written pubertal development scale (Petersen et al. 1988) completed by a parent or guardian at the early adolescent visit [mean 13.1 (SD 0.8) years]. Ninety-nine percent of reporters were mothers. Items included on the boys’ pubertal development scale were voice deepening, body hair growth, facial hair growth, acne, and growth spurt. Items included on the girls’ pubertal development scale were breast development, body hair growth, acne, growth spurt, and menarche. Response options and scores for each pubertal development scale item except menarche were: 1 point for “not yet started”, 2 points for “barely started”, 3 points for “definitely started”, and 4 points for “seems complete”. We coded menarche as 4 points if menarche was present and 1 point if not present. To assign a summary pubertal development score to each participant, we calculated the average score across all items. Previous studies have reported moderate to strong correlations between pubertal development score and other markers of pubertal development, including physician breast Tanner staging in girls (Brooks-Gunn et al. 1987; Koopman-Verhoeff et al. 2020) and growth velocity (Petersen et al. 1988).

Age at menarche

We queried mothers on their daughters’ menarcheal status and the month and year of menarche via annual questionnaires from ages 9–12 years and at age 14 years and used the first reported age at menarche to minimize misclassification (Dorn et al. 2013).

Covariates

Mothers reported their age at enrollment in the cohort, educational attainment (categorized as with or without a college degree), annual household income (categorized as <$40,000, $40,000–$70,000, or >$70,000), pregnancy smoking status (smoked during pregnancy, former smoker, never smoked), and age at menarche. Mothers also reported their child’s race/ethnicity, which we categorized as white, Black, Asian, Hispanic, or other, and their child’s breastfeeding duration (months). We obtained child date of birth and sex from medical records. Mothers reported their child’s sugar-sweetened beverage (i.e., non-diet soda) intake using the PrimeScreen administered at mid-childhood (Rifas-Shiman et al. 2001).

Statistical analyses

We examined distributions of PFAS plasma concentrations and calculated Spearman correlations between individual PFAS. As our primary approach, we used sex-specific linear regression to evaluate the associations between plasma concentrations of each individual PFAS as continuous variables with age at peak height velocity and pubertal development score. After confirming associations were generally linear via generalized additive models, we used single PFAS linear regression models as our primary approach because they provide a quantifiable measure of association that can be easily compared to other studies. For all analyses, PFAS plasma concentrations were log-2 transformed to satisfy model assumptions and aid interpretation.

We first ran unadjusted models (except for models of pubertal development score, which we adjusted for age at outcome to improve precision and align with the other two outcomes, which are age-based). Next, we ran covariate-adjusted models. We selected the following covariates a priori based on previously described associations with PFAS and/or pubertal timing: mother’s education, household income, mother married/cohabiting, maternal age at enrollment, maternal smoking history, maternal age at menarche, child race/ethnicity, and breastfeeding duration (Ellis 2004; Herman-Giddens et al. 1997; Mogensen et al. 2015). To account for time trends in PFAS concentrations, we also adjusted for year of mid-childhood blood draw. Because PFAS have been associated with body composition in children (Domazet et al. 2016; Lin et al. 2009; Nelson et al. 2010), and adiposity is an important predictor of pubertal timing (Biro et al. 2006b; Dunger et al. 2005), we considered child BMI to be a potential mediator and did not include it in multivariable models.

For the primary analysis, we used chained equation multiple imputation to impute values for missing covariate data (N=188, 29%). Using an imputation model that included all PFAS concentrations, outcomes, and covariates for all 2,128 mother-child pairs in Project Viva, we generated 50 imputed data sets using PROC MI (SAS) (Harel et al. 2018).

Single PFAS models do not account for the complex mixture of PFAS that constitutes human exposure. Therefore, we additionally used Bayesian Kernel Machine Regression (BKMR) to simultaneously account for all mid-childhood PFAS plasma concentrations, while also allowing for potential non-linear and non-additive associations between PFAS and markers of pubertal timing (Bobb et al. 2018; Bobb et al. 2015). We used the variable selection option, which calculates the relative importance of each PFAS, the Gaussian kernel, and 10,000 Markov chain Monte Carlo iterations to model associations and 95% credible intervals. We used BKMR to examine the strength and linearity of the associations between a) the overall PFAS mixture and each pubertal outcome, comparing the effect of setting all PFAS at a given quantile versus the median and b) each PFAS and each pubertal outcome, setting other PFAS to the median. Next, we examined whether a given PFAS acted synergistically with the other PFAS. To do this, we examined the change in each pubertal outcome when a given PFAS was at the 75^th vs 25^th percentile a) when all other PFAS were set to the 75^th percentile and b) when all other PFAS were set to the 25^th percentile, and we compared these two values. If we detected interaction, we additionally investigated relevant pairwise (single PFAS-single PFAS) interactions. For BKMR models, we standardized PFAS plasma concentrations by dividing by the SD and modeled categorical covariates using indicator variables. We collapsed the values of several covariates (household income >$70,000/≤$70,000; smoked/ did not smoke during pregnancy; white/ nonwhite) due to small cell counts. We singly-imputed missing covariate data using stochastic imputation.

We additionally calculated the hazard ratio (HR) of time to menarche (in years and months) per doubling of plasma PFAS using singly-imputed Cox proportional hazards models with a time scale of age in years. We censored girls who never reported menarche at their last menarcheal status report. We evaluated the proportional hazards assumption using Schoenfeld residuals. Several covariates (child race, income, and maternal smoking) violated the proportional hazards assumption; however, after stratifying these variables by age at follow-up (<11 years, 11–12.9 years, ≥13 years) the proportional hazards assumption were met for all analyses. Due to sparse cell counts, we collapsed race (white/nonwhite). We used the Kaplan Meier method to calculate the median unadjusted age at menarche. We present only single-PFAS regression models for PFAS plasma concentrations and age at menarche, as BKMR models have not yet been adapted for use with survival models.

We performed several sensitivity analyses to test the robustness of our findings. First, we investigated PFAS-pubertal outcome associations for PFAS isomers, which may have different associations with health outcomes than other isomers (Cardenas et al. 2017; Lin et al. 2020). Second, because emerging literature has linked prenatal PFAS exposure to delayed puberty (Christensen et al. 2011; Kristensen et al. 2013), we tested the degree to which our findings were driven by PFAS concentrations measured in childhood (rather than maternal prenatal PFAS concentrations) by additionally adjusting models for log-2 transformed individual PFAS plasma concentrations measured in mothers in early pregnancy (median=10 weeks). Correlations between maternal prenatal PFAS and mid-childhood PFAS were generally weak, ranging from 0.02 for PFNA to 0.36 for PFHxS. Third, we also considered sugar-sweetened beverage intake, and maternal and paternal height; however, including these variables did not substantially change our results. We limited our analysis of dietary factors to sugar-sweetened beverages because previous studies by our group and others have related sugar-sweetened beverages, but not other dietary factors (Carwile et al. 2015b; Gaskins et al. 2017), to both plasma PFAS (Seshasayee et al. 2021) and age at menarche (Carwile et al. 2015a; Carwile et al. 2015b; Gaskins et al. 2017; Mueller et al. 2015). Finally, we conducted a complete case analysis in which analyses were restricted to participants without missing covariate data (for pubertal development score, N=221 for girls and N=264 for boys; for age at peak height velocity, N=219 for girls and N=256 for boys; for menarche, N=221).

We used R (Core Team. 2018. R: A language and environment for statistical computing. Vienna Austria:R Foundation for Statistical Computing) for BKMR and menarche survival models; other analyses were performed in SAS EG (version 7.1, SAS Institute Inc.).

Results

Overall, our study population was 47% female and 60% white, with 70% reporting a household income >$70,000 (Table 1). PFAS were measured at (mean ± SD) 7.9 ± 0.8 years and pubertal development score was reported at 13.1 ± 0.8 years. The mean pubertal development score in early adolescence (mean ± SD) was 2.9 ± 0.7 for girls and 2.2 ± 0.7 for boys. As expected, mean pubertal development score increased with age (data not shown). Peak height velocity (mean ± SD) was 11.2 ± 1.0 years for girls and 13.1 ± 1.0 years for boys. The majority (71%) of girls reported menarche during follow-up, and the remaining 29% were censored due to non-response; none reported being premenarcheal at the end of follow-up at age 14 years. The median age at menarche was 12.6 years. Mothers of children who had higher PFAS plasma concentrations were slightly older, more likely to be white, and have a higher household income. Plasma concentrations were highest [median (interquartile range (IQR))] for PFOS 6.4 (5.6) ng/mL, followed by PFOA [4.4 (3.0) ng/mL] and PFHxS [1.9 (2.3) ng/mL]. Individual PFAS were correlated, with the strongest correlation observed between PFOA and PFOS (Spearman correlation coefficient 0.72). PFOA and PFOS were also moderately correlated with PFDA, PFHxS, and MeFOSAA (e.g., for PFOS and MeFOSAA, Spearman correlation coefficient 0.59) (Table S2).

Table 1.

Participant characteristics overall (N=640) and by quartiles of perfluorooctanoate (PFOA) in mid-childhood, N (%).

		Quartiles of PFOAa
	Overall N=640	Q1 N=173	Q2 N=152	Q3 N=153	Q4 N=162
Parental/neighborhood characteristics
Maternal age at enrollment (y), mean ± SD	31.9 ± 5.6	30.1 ± 6.4	32.0 ± 5.7	32.4 ± 5.0	33.1 ± 4.5
Maternal age at menarche (y), mean ± SD	12.6 ± 1.5	12.5 ± 1.8	12.5 ± 1.4	12.7 ± 1.4	12.8 ± 1.4
Maternal pregnancy smoking statusb
Smoked during pregnancy	68 (11)	23 (13)	13 (9)	15 (10)	17 (11)
Former	123 (19)	26 (15)	31 (21)	37 (24)	29 (18)
Never	448 (70)	124 (72)	107 (71)	101 (66)	116 (72)
Married/cohabiting	568 (89)	132 (77)	136 (91)	143 (94)	157 (97)
Mother college graduate	414 (65)	75 (44)	100 (66)	110 (72)	129 (80)
Annual household incomeb
<$40,000	86 (14)	41 (26)	17 (12)	16 (11)	12 (8)
$40,000–70,000	96 (16)	33 (21)	20 (14)	23 (16)	20 (13)
>$70,000	418 (70)	81 (52)	104 (74)	108 (73)	125 (80)
Child characteristics
Female	303 (47)	83 (48)	76 (50)	66 (43)	78 (48)
Race/ethnicityb
White	381 (60)	55 (32)	90 (59)	104 (68)	132 (81)
Black	134 (21)	72 (42)	29 (19)	22 (14)	11 (7)
Asian	17 (3)	8 (5)	2 (1)	3 (2)	4 (3)
Hispanic	32 (5)	13 (8)	11 (7)	6 (4)	2 (1)
Other	74 (12)	24 (14)	20 (13)	17 (11)	13 (8)
Breastfeeding duration (months)	6.4 ± 4.5	5.0 ± 4.2	6.4 ± 4.6	6.8 ± 4.3	7.4 ± 4.5
Year of mid-childhood blood draw
2007	68 (11)	8 (5)	17 (11)	16 (10)	27 (17)
2008	220 (34)	25 (14)	41 (27)	62 (41)	92 (57)
2009	205 (32)	64 (37)	58 (38)	50 (33)	33 (20)
2010	147 (23)	76 (44)	36 (24)	25 (16)	10 (6)
Age (y) at mid-childhood blood draw, mean ± SD	7.9 ± 0.8	8.3 ± 1.0	8.0 ± 0.8	7.8 ± 0.7	7.7 ± 0.6
Age (y) at early adolescent visit, mean ± SDc	13.1 ± 0.8	13.1 ± 0.8	13.0 ± 0.8	13.1 ± 0.9	13.1 ± 0.8

^aPFOA quartile minimum and maximum values (in ng/mL): Q1, < 0.1–3.1; Q2 >3.1–4.4; Q3, >4.4–6.0; Q4, >6.0–14.3.

^bNumbers do not sum to 100 due to rounding.

^cAge at early adolescent questionnaire

Missing data for participants overall: maternal age at menarche, N=93; maternal pregnancy smoking status, N=1; married/cohabiting, N=5; mother college graduate, N=4; annual household income, N=40; child race/ethnicity, N=2; breastfeeding duration, N=52

Abbreviations: PFOA, perfluorooctanoate; SD, standard deviation

PFAS plasma concentrations and puberty in girls

Girls with higher plasma concentrations of several PFAS had later pubertal timing in single PFAS models. In covariate-adjusted models, each doubling of PFOA was associated with a 0.18 unit lower [95% confidence interval (CI): −0.30, −0.06] pubertal development score and a 0.23 year (95% CI: 0.06, 0.40) older age at peak height velocity [i.e., 2.76 months (95% CI: 0.72, 4.8)] (Figures 1 & 2, Table S3). We observed similar though weaker associations for PFOS and PFDA, and null associations for PFHxS and MeFOSAA. Point estimates for PFNA indicated associations with lower pubertal development score and older age at peak height velocity, but 95% confidence intervals crossed the null. A doubling of PFOA plasma concentrations was associated with a 12% lower hazard of menarche [HR: 0.88 (95% CI: 0.68, 1.13)] (Figure 3, Table S4)]. We observed similar associations for PFOS [HR: 0.93 (95% CI: 0.78, 1.13)] and PFDA [HR: 0.91 (95% CI: 0.77, 1.06)].

Figure 1.

Associations of mid-childhood per- and polyfluoroalkyl substance (PFAS) plasma concentrations with pubertal development score in girls (N=258) and boys (N=297). Adjusted for age at outcome, mother’s education, household income, married/cohabiting, maternal age at enrollment, maternal smoking history, maternal age at menarche, child race/ethnicity, breastfeeding duration, and year of blood draw. Missing covariate data was multiply imputed. For effect estimates, see Table S3.

Abbreviations: CI, confidence interval, PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFDA, perfluorodecanoate; PFHxS, perfluorohexane sulfonate; MeFOSAA, 2-(N-methyl-perfluorooctane sulfonamide) acetate; PFNA, perfluorononanoate

Figure 2.

Associations of mid-childhood per- and polyfluoroalkyl substances (PFAS) plasma concentrations with age at peak height velocity in girls (N=297) and boys (n=329). Adjusted for mother’s education, household income, married/cohabiting, maternal age at enrollment, maternal smoking history, maternal age at menarche, child race/ethnicity, breastfeeding duration, and year of blood draw. Missing covariate data was multiply imputed. For effect estimates, see Table S3.

Abbreviations: CI, confidence interval, PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFDA, perfluorodecanoate; PFHxS, perfluorohexane sulfonate; MeFOSAA, 2-(N-methyl-perfluorooctane sulfonamide) acetate; PFNA, perfluorononanoate

Figure 3.

Adjusted¹ Cox proportional hazards survival curves (log scale) for time until menarche (years) (N=258).

¹ Adjusted for mother’s education, household income, married/cohabiting, maternal age at enrollment, maternal smoking history, maternal age at menarche, child race/ethnicity, breastfeeding duration, and year of blood draw. Missing covariate data was multiply imputed. For effect estimates, see Table S4.

Abbreviations: CI, confidence interval, PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFDA, perfluorodecanoate; PFHxS, perfluorohexane sulfonate; MeFOSAA, 2-(N-methyl-perfluorooctane sulfonamide) acetate; PFNA, perfluorononanoate

Results from mixture models were generally consistent with those from the single PFAS models. Among girls, higher concentrations of the overall PFAS mixture were linearly associated with lower pubertal development score (i.e., less pubertal development in early adolescence) (Figure S1, panel A, Table S3) and an older age at peak height velocity (Figure S2, panel A, Table S3), although confidence intervals for mixture-outcome associations crossed the null. When we used the mixture model framework to examine individual PFAS-outcome associations, we observed that higher plasma concentrations of PFOA, PFOS, and PFNA were associated with a lower pubertal development score, and higher plasma concentrations of PFOA and PFDA were associated with older age at peak height velocity. In contrast, higher concentrations of PFHxS and MeFOSAA were associated with a higher pubertal development score (Figure S1, panel B). PFNA and PFHxS contributed most to the overall fit for the pubertal development score model, and PFOA and PFDA contributed most to the fit of the age at peak height velocity model (Table S5). Individual PFAS-outcome associations were not modified by other PFAS for either pubertal outcome (Figure S1 and S2, panel C).

PFAS plasma concentrations and puberty in boys

PFAS plasma concentrations were not associated with pubertal development score (Figure 1) or age at peak height velocity in boys (Figure 2) in single PFAS models. In mixture models, we similarly observed no effect of the overall PFAS mixture on pubertal development score (Figure S1, panel D), although the overall PFAS mixture was associated with an increase in age at peak height velocity (Figure S2, panel D). In mixture models, associations between individual PFAS and pubertal development scores were null with wide 95% CIs (Figure S1, panel E). Higher PFOA plasma concentrations were non-linearly associated with older age at peak height velocity, and higher PFNA plasma concentrations were linearly associated with younger age at peak height velocity (Figure S2, panel E). Individual PFAS were otherwise not associated with either pubertal outcome, and the individual PFAS-outcome associations were not modified by other PFAS (Figure S1 & S2, panel F).

Sensitivity analyses

The magnitude of the associations between PFOA or PFOS isomers and pubertal outcomes were similar to that of the associations between the sum of their respective isomers and pubertal outcomes. For example, in girls, each doubling of n-PFOA plasma concentrations was associated with a 0.20 unit (95% CI: −0.33, −0.08) lower pubertal development score compared to a 0.18 lower (95% CI: −0.30, −0.06) pubertal development score for PFOA (other isomer data not shown). Our findings were robust to additional adjustment for individual prenatal PFAS plasma concentrations (Tables S3 and S4, Model 3). Effect estimates from complete case analyses were similar to models in which missing covariates were imputed (data not shown).

Discussion

In this large prospective study of US children, we found that mid-childhood plasma concentrations of select PFAS were associated with later puberty in girls, but not boys.

We found the strongest associations between plasma concentrations of PFOA, PFOS, and PFDA in mid-childhood and later puberty in girls, consistent with a previous epidemiological study of childhood PFAS plasma concentrations and pubertal outcomes. In a cross-sectional analysis of girls aged 8–18 years in the C8 Health Project, a large sample of Mid-Ohio Valley residents exposed to PFOA via water contamination, girls in the highest PFOA and PFOS quartiles were less likely to have reached menarche than those in the lowest PFAS quartiles (Lopez-Espinosa et al. 2011). Similar findings have also been reported in studies examining prenatal PFAS plasma concentrations. Kristensen et al. found that prenatal PFOA concentrations were associated with later age at menarche in a population-based birth cohort in Denmark (Kristensen et al. 2013), and prenatal PFOS, but not other PFAS, was associated with a non-statistically significant reduced odds of reaching menarche before age 11.5 years in a case-control study of UK girls (Christensen et al. 2011). In the Danish National Birth Cohort, prenatal PFDA was non-monotonically associated with later menarche; however, there was no clear pattern of an association between prenatal PFAS and markers of pubertal timing (e.g., age at menarche and Tanner stages 2–5) in girls (Ernst et al. 2019). Our findings suggest that some, but not all, PFAS are associated with delayed pubertal timing in girls, although this conclusion requires replication in additional cohorts.

The association between higher PFAS plasma concentrations and later puberty in girls may be due to a lowering of estradiol levels or reduction in estrogen receptor expression by PFAS. Pre-pubertal female rats treated with 3 mg/kg-d PFDA had lower estradiol concentrations and reduced expression of estrogen receptors α and β relative to untreated mice (Shi et al. 2009). Similarly, in a cross-sectional analysis of peri-menopausal and menopausal women in the C8 Health Project, PFOS was inversely associated with estradiol (Knox et al. 2011), although PFOS concentrations were much higher than those observed in Project Viva (median 18–33 ng/mL vs. 6.4 ng/mL). Consistent with a reduction in estradiol, treatment of mouse dams with PFOA delayed vaginal opening, a sign of puberty, in female offspring (Lau et al. 2006; Yang et al. 2009).

Our findings also align with previous epidemiological studies that have reported mixed findings for PFAS concentrations and puberty in boys. In boys in the Danish National Birth Cohort, higher PFHxS concentrations were associated with younger age at self-reported markers of pubertal development (e.g., voice break and acne), but other PFAS-pubertal outcome associations were generally non-monotonic with confidence intervals that crossed the null (Ernst et al. 2019). In contrast, in the cross-sectional analysis of the C8 Health Project, boys aged 8–18 years with higher PFOS, but not PFOA, had lower odds of having reached puberty, defined as a total testosterone concentration >50 ng/dL (Lopez-Espinosa et al. 2011). Due to exposure to industrially-contaminated drinking water, PFAS plasma concentrations in C8 Health Project participants were around three-times higher than in those of Project Viva participants (e.g, median 19.4 vs. 6.4 ng/mL).

The present study is the first to our knowledge to examine the role of PFAS mixtures in childhood on pubertal timing. Mixture models may be less biased than single PFAS models because they account for correlations and interactions between chemicals. We present single PFAS models as our primary results because they are more interpretable and facilitate comparison to other studies; however, the observation that results from BKMR mixture models were generally consistent with results from single PFAS linear regression models adds credibility to our findings. In both models, we found higher plasma concentrations of PFOA and PFOS in mid-childhood to be associated with less pubertal development (i.e., lower pubertal development score) in early adolescence in girls. Contrary to our findings from single PFAS models, in BKMR models, girls with higher concentrations of MeFOSAA and PFHxS appeared to have greater pubertal development (i.e., higher pubertal development score) in early adolescence. One possible explanation is that BKMR models account for correlations between chemicals whereas single PFAS models do not, and concentrations of MeFOSAA, and to a lesser extent, PFHxS, were correlated with PFOS and PFOA. PFOS and PFOA were associated with less pubertal development in early adolescence, which may explain null findings for MeFOSAA and PFHxS in single PFAS models. Correlations between PFNA and other PFAS (PFOS, PFOA, and PFDA) could also, in part, explain our observation that associations between the overall PFAS mixture and pubertal development score were most strongly driven by PFNA, even though this PFAS did not have the strongest association with pubertal development score in single PFAS models. Another possible reason that results from BKMR and single PFAS models may differ is because BKMR takes into account interactions among PFAS (Bobb et al. 2015); however, we did not find evidence for strong interactions between any PFAS and the rest of the PFAS mixture.

Our finding of an association between higher plasma concentrations of select PFAS and later pubertal timing in girls has implications for bone health. The majority of bone mass accumulates during puberty, and children with later puberty have an abbreviated window for bone mass to accrue, resulting in reduced peak bone mass (Chevalley et al. 2008). Consequently, women who experienced later menarche are at increased risk of fractures across the life course (Chevalley et al. 2012; Johnell et al. 1995). We previously reported a cross-sectional association between PFAS concentrations and lower bone mineral density in Project Viva in 8-year old children. This association was evident even when we restricted to prepubertal children, suggesting that PFAS may impact bone mineral density through mechanisms other than through later pubertal timing (Cluett et al. 2019). Future studies should explore whether the observed association between PFAS exposure in childhood and lower bone mineral density persists through adolescence and whether this may be partly driven by later puberty. The extent to which a 2–3 month older age at puberty impacts bone outcomes remains unclear, and its public health importance should be considered in the context of the many known adverse health effects associated with earlier puberty in girls, including early sexual debut, depression, and increased risk of breast cancer (Golub et al. 2008; Graber 2013; Hsieh et al. 1990).

Our study has several potential limitations. First, the pubertal development score, which is a subjective report of pubertal status, is subject to misclassification. We considered differential misclassification of the pubertal development score unlikely because participants were unaware of their PFAS plasma concentrations; however, it is possible that misclassification could vary based on other factors (e.g., socioeconomic status) associated with PFAS plasma concentrations. Also, the pubertal development score was typically reported by the participants’ mothers, and maternal report of pubertal status is less accurate than clinical assessment, and possibly, assessments performed by the children themselves (Terry et al. 2016). Mothers may also report pubertal development scores more accurately for their daughters compared to their sons (Carskadon and Acebo 1993). Our null findings in boys could be explained by greater misclassification of the pubertal development score in boys, but this finding was present across pubertal outcomes, and misclassification is unlikely for age at peak height velocity, which was based on objective height measurements and calculated using a SITAR model, considered to be the least biased method (Simpkin et al. 2017). Mean pubertal development score and median age at menarche in our study was similar to those of other US studies (Carwile et al. 2015a; Chumlea et al. 2003; Duncan Cance and Ennett 2012). However, the mean age at peak height velocity in our study population was slightly younger than those from other US and UK cohorts (Abbassi 1998; Cole 2020), which could be due to differences in participant characteristics or methods of calculating age at peak height velocity, or temporal changes in pubertal timing. Additionally, while Cox proportional hazards models are limited in their ability to account for changing hazards and potential for selection bias (Hernan 2010), they allow for comparison to other studies.

Our study was limited to select markers of pubertal timing (i.e., pubertal development score in early adolescence, age at peak height velocity, and age at menarche), and we did not directly test whether PFAS are also associated with earlier or later markers of pubertal timing, including age at pubertal onset. However, while puberty trajectory can be nonlinear, onset of puberty is moderately correlated with age at menarche and age at peak height velocity (Biro et al. 2006a; German et al. 2018), and associations of PFAS with delayed vaginal opening and preputial separation in rodents further support an association between PFAS and onset of puberty (Butenhoff et al. 2004; Lau et al. 2006; Yang et al. 2009). Direct characterization of the association of PFAS with earlier and later markers of pubertal timing, as well as pubertal tempo (Pantsiotou et al. 2008), require further investigation.

Our study also has several important strengths. We examined multiple validated markers of puberty and found that PFAS-pubertal timing associations were generally consistent across all pubertal outcomes. In addition to self- and parent-reported outcomes, we calculated age at peak height velocity from repeated height measures obtained at research and clinical visits, providing an objective measure of pubertal timing. Another strength of our study is that we were able to account for PFAS concentrations during gestation, supporting an independent association between PFAS concentrations during childhood and pubertal timing. Third, the prospective design of our study allowed us to avoid reverse causation, which may be introduced in cross-sectional analyses if PFAS plasma concentrations decrease following the initiation of menses (i.e., PFAS loss via excretion) or when body volume increases during the growth spurt (i.e., PFAS dilution) (Wu et al. 2015). Although many Project Viva participants are white and college education, the generalizability of our findings is supported by the similarity of the median age at menarche (Adgent et al. 2012; Herman-Giddens et al. 1997) and PFAS plasma concentrations (Ye et al. 2018) to other populations.

In this prospective study of US children, we found higher plasma concentrations of select PFAS at mid-childhood to be associated with later puberty in girls, but not boys. Our findings add to a growing body of literature supporting the ability of certain PFAS, particularly PFOA and PFOS, to disrupt normal endocrine function at concentrations detected in the general US population.

1. We examined associations of mid-childhood PFAS and markers of pubertal timing.

2. In girls, PFOS, PFOA, and PFDA were associated with later markers of pubertal timing.

3. PFAS were not associated with markers of pubertal timing in boys.

4. Results from single PFAS models were generally consistent with BKMR models.