Serum PFAS and Urinary Phthalate Biomarker Concentrations and Bone Mineral Density in 12-19 Year Olds: 2011-2016 NHANES

Abstract

Context

Per- and polyfluoroalkyl substances (PFAS) and phthalates are 2 families of environmental endocrine disruptors that may be associated with areal lower bone mineral density (aBMD).

Objective

To examine associations between serum PFAS and urinary phthalate biomarker concentrations and their mixtures with aBMD Z-scores in adolescents.

Design, Patients, and Measures

We examined serial cross-sectional data from male (n = 453) and female (n = 395) 12- to 19-year-old participants in the 2011 through 2016 National Health and Nutrition Examination Survey with measures of serum PFAS, urinary phthalate metabolites, and dual-energy X-ray absorptiometry aBMD Z-scores (total body less head). In sex-specific models, we used linear regression to examine associations of individual PFAS and phthalate biomarkers with aBMD Z-scores, and Bayesian kernel machine regression to examine the association of the overall PFAS/phthalate biomarker mixture with aBMD Z-scores. We replicated the analysis, stratifying by race/ethnicity.

Results

Participants were (mean ± SD) 15 ± 2.1 years of age. In males, each doubling of serum perfluorooctanoate (PFOA), perfluorooctane sulfonate, urinary mono-isobutyl phthalate (MiBP), mono-n-butyl phthalate, and the overall PFAS/phthalate mixture was associated with a lower aBMD Z-score (eg, for PFOA: -0.24; 95% CI, -0.41 to -0.06). Serum PFOA and urinary MiBP were associated with higher aBMD Z-scores in females (eg, for PFOA: 0.09; 95% CI, -0.07 to 0.25). Findings did not differ by race/ethnicity.

Conclusions

Certain PFAS and phthalates may be associated with reduced bone mineral density in adolescent males. Bone mineral density tracks across the life course, so if replicated in longitudinal cohorts, this finding may have implications for lifelong skeletal health.

Bone accrual primarily occurs during childhood and adolescence (1, 2); therefore, it is critically important to identify modifiable factors that negatively affect bone during this period to better promote bone health across the lifespan. The role of the synthetic chemical classes per- and polyfluoroalkyl substances (PFAS) and phthalates on bone health is of particular interest because of their high exposure prevalence in US children and adolescents (3) and ability to disrupt hormone-signaling pathways. PFAS impart water-, oil-, and dirt-resistant properties to diverse items, including nonstick cookware, clothing, and food packaging (4). Low-molecular-weight phthalates are primarily used in personal care products, whereas high-molecular-weight phthalates provide flexibility to plastics in building materials, food processing, and children’s toys (5). PFAS and phthalates may affect bone homeostasis through shared biological mechanisms. Both groups of chemicals activate peroxisome proliferator-activated receptor gamma, suppressing osteoblast formation (6, 7), and are androgen receptor antagonists that may inhibit androgen-mediated osteoblastogenesis (8, 9).

Previous epidemiological studies support associations of PFAS and phthalate biomarkers with lower areal bone mineral density (aBMD), but information is lacking for adolescents and on whether data are generalizable across race/ethnicities. Higher PFAS concentrations have been associated with lower aBMD Z-scores in 6- to 10-year-old children in a Boston-based cohort, in a small pilot study of 8- to 12-year-old obese children, and in adults in the National Health and Nutrition Examination Survey (NHANES) (10-12). Also, several studies have shown that postmenopausal females with higher urinary phthalate metabolite concentrations have lower aBMD (13-15). However, it is important to determine if prior findings on PFAS, phthalate metabolites, and bone health are generalizable to more racially and socioeconomically diverse populations because PFAS and phthalate exposure profiles differ by race, ethnicity, and socioeconomic status (16-18). There is also a gap in the literature regarding exposure vulnerability in adolescence, when bone accrual is at its peak.

Here, we address these gaps by describing associations between serum PFAS and urinary phthalate biomarker concentrations, and their complex mixtures, with aBMD Z-score in 12 to 19 year olds in the 2011 through 2016 NHANES. We hypothesize that adolescents with higher concentrations of serum PFAS and urinary phthalate biomarkers will have a lower aBMD Z-score.

Materials and Methods

Study Population

We combined data from 2011 and 2012, 2013 and 2014, and 2015 and 2016 NHANES, a serial cross-sectional survey of the noninstitutionalized US population. We selected these cycles because both PFAS and phthalate biomarkers have strong temporal trends (19, 20), making data from the most recent cycles most relevant. Also, previous cycles did not measure total body less head (TBLH) aBMD, a preferred skeletal site in children and adolescents because the skull is not responsive to environmental factors (21, 22). Participants completed in-person interviews, provided urine and blood samples, and underwent a physical examinations including dual-energy x-ray absorptiometry to measure aBMD. We restricted the study population to adolescents 12 to 19 years of age because serum PFAS was not measured in younger children and because adolescent (vs child) aBMD tracks more strongly with adult aBMD (23). Of 4027 participants between 12 and 19 years of age, 1015 had complete data on serum PFAS and urinary phthalate biomarker concentrations, and, of these, 896 also had data on aBMD. After excluding participants missing data on diet (n = 26) or other covariates (n = 22), 848 participants were included in this analysis. The Maine Medical Center institutional review board determined that this study was not human subjects research.

Serum PFAS and Urinary Phthalate Metabolite Measurements

During the examinations, NHANES technicians collected blood and urine samples. Serum and urine samples were stored at -20°C until shipped to the National Center for Environmental Health at the Centers for Disease Control and Prevention (CDC). CDC laboratory staff measured serum PFAS and urinary phthalate metabolite concentrations in a one-third random subsample of participants. CDC staff measured serum PFAS concentrations using online solid-phase extraction coupled to high-performance liquid chromatography-turbo ion spray ionization-tandem mass spectrometry, and quantified urinary phthalate metabolites using solid-phase extraction coupled with online high-performance liquid chromatography and tandem mass spectrometry (24, 25). NHANES analysts imputed values below the limit of detection (LOD) with $\mathrm{L}\mathrm{O}\mathrm{D}/\surd 2$ (26).

We a priori limited our analysis to PFAS and phthalate metabolites for which at least 60% of samples were > LOD, thus we included 5 serum PFAS—perfluorooctanoate (PFOA), perfluorooctane sulfonate (PFOS), perfluorodecanoate (PFDA), perfluorohexane sulfonate, and perfluorononanoate—and 7 urinary phthalate metabolites—monocarboxyoctyl phthalate (MCOP), monocarboxy-isononyl phthalate (MCNP), mono(3-carboxypropyl) phthalate (MCPP), monobenzyl phthalate (MBzP), monoethyl phthalate (MEP), monobutyl phthalate (MBP), and mono-isobutyl phthalate (MiBP). We also included ∑ di-2-ethylhexyl phthalate (∑DEHP), which we estimated as the molar sum of the DEHP metabolites, mono (2-ethylhexyl) phthalate, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, and mono(2-ethyl-5-carboxypentyl phthalate (µmol/L) (19). We categorized ∑ DEHP, MCOP, MCNP, MCPP, and MBzP as high-molecular-weight urinary phthalate biomarkers, and MEP, MBP, and MiBP as low-molecular-weight urinary phthalate biomarkers. In 2011 and 2012, NHANES analysts calculated total serum PFOA and PFOS. For data from 2013 through 2016, we calculated PFOA as the sum of branch perfluorooctanoic acid isomers and n-perfluorooctanesulfonic acid, and calculated PFOS as the sum of n-perfluorooctanesulfonic acid and perfluoromethylheptanesulfonic acid isomers. These calculated sums are considered comparable to the total PFOS and PFOA as reported in 2011 and 2012 because the summed isomers of each chemical represent more than 95% of what was previously reported (25).

Bone Mineral Density

All eligible participants were invited to complete dual-energy x-ray absorptiometry scans, which were performed by a radiology technologist using a Hologic Discovery model A densitometer (27). The accuracy of the scans was ensured by regular quality control procedures and expert review of all scans. We used TBLH aBMD in our analysis because the International Society for Clinical Densitometry considers this skeletal site to be the most accurate and reliable for children and adolescents (21). We calculated aBMD Z-scores using published age-, sex-, height-, and race-specific reference curves (28).

Covariates

NHANES staff measured participants’ height with stadiometers using standardized protocols. Participants (or, if younger than 16 years, their proxy) reported their age, sex, race/ethnicity, and family size (29). NHANES calculates income to poverty ratio as the ratio of household income to the Department of Health and Human Services poverty threshold specific to their family size. Participants separately reported their moderate (eg, brisk walking, bicycling) and vigorous (eg, running, basketball) physical activity. We calculated hours per week of moderate or vigorous physical activity by multiplying the days per week spent engaged in an activity by daily minutes of activity, dividing by 60, and summing across moderate and vigorous activities. Participants were asked whether any household members smoked inside their home. CDC laboratory technicians measured cotinine in serum samples.

Participants also completed a 24-hour dietary recall, in which they reported the types and amounts of all foods consumed in the previous 24 hours (30). Reported foods were coded into food groups and nutrients using the US Department of Agriculture’s Food and Nutrient Database for Dietary Studies (31). We defined sugar-sweetened beverage intake using What We Eat in America food categories, a classification scheme of more than 150 unique foods and beverages (32), and categorized intake in cups (245 g). We categorized total dairy intake (cup equivalents) using the Food Patterns Equivalents Database, which converts dietary intake into the relatively broader US Department of Agriculture Food Patterns components (33).

Statistical Analysis

We examined distributions of serum PFAS and urinary phthalate metabolite concentrations and calculated Spearman correlations between individual biomarkers.

We first assessed linearity of associations using generalized additive models. Next, we used linear regression to examine single-chemical associations of serum PFAS and urinary phthalate biomarker concentrations with TBLH aBMD Z-scores. Because endocrine disruptor-health outcome associations, including for PFAS and aBMD (12, 34), often vary by sex, we present sex-specific findings for PFAS and phthalate biomarkers. We log-2 transformed serum PFAS and urinary phthalate biomarker concentrations to meet model assumptions and aid interpretation. We first ran models that were unadjusted for PFAS and models adjusted only for urinary creatinine for phthalates. We adjusted for urinary creatinine in phthalate models to improve precision (35). Next, we ran models additionally adjusted for covariates that we selected a priori based on previously described associations with PFAS and phthalate biomarkers and aBMD (19, 36-39): sociodemographics (age, race/ethnicity), physical activity, indoor smoking by a household member, total dairy intake, sugar-sweetened beverage intake, and temporal trend (ie, NHANES cycle [2011-2012, 2013-2014, 2015-2016]). Because effect estimates were similar between models with and without additional adjustment for income to poverty ratio and serum cotinine (data not shown), we excluded these variables from final models. Although our group and others have related additional food items to PFAS (eg, fish, meat/poultry) and phthalate biomarkers (eg, fish, baked goods, starchy vegetables) (40-43), we did not include them in our model because these foods are not strongly associated with aBMD (1). We did not adjust for body size and pubertal status because these variables may mediate the association between PFAS/phthalate biomarkers and aBMD (44-47). We adjusted for NHANES survey weights to account for bias that may have been introduced by oversampling (or undersampling) specific populations (eg, race/ethnicity) as part of the NHANES survey methodology (48). To facilitate comparison with Bayesian kernel machine regression (BKMR) mixture models that cannot incorporate survey analysis procedures, we also did not use survey analysis procedures to account for the complex sample design in single chemical models. Thus, like other NHANES studies using novel mixture methods (49), the findings from our study leverage the large sample size and racial/ethnic diversity of NHANES, but cannot be considered generalizable to the US population. Because serum PFAS and urinary phthalate biomarkers vary by race/ethnicity (17, 18), we stratified models by the most frequently sampled categories of race/ethnicity (non-Hispanic White, non-Hispanic Black, and Hispanic [Mexican American and other Hispanic]).

Because exposure to multiple PFAS and phthalates occurs simultaneously and both classes of chemicals may act through similar pathways to lower aBMD, we used BKMR to examine the association between the overall chemical mixture and aBMD Z-score. BKMR takes into account correlations and interactions between chemical biomarkers as well as potential nonlinearity of chemical biomarker–aBMD Z-score associations (50, 51). The BKMR model characterizes the association between individual chemical biomarkers and aBMD Z-score, holding other chemicals constant and also provides an overall mixture effect. We used a hierarchical variable selection method and posterior inclusion probabilities in the BKMR model to quantify the relative importance of each class of chemical (ie, PFAS, low-molecular-weight phthalate biomarkers, and high-molecular-weight phthalate biomarkers) and individual chemicals within each class to the overall mixture effect. We considered low-molecular-weight and high-molecular-weight phthalate biomarkers separately because their associations with health outcomes often vary (52). Within the BKMR model, we also assessed for chemical–chemical interactions by calculating the effect estimate for a single chemical (75th vs 25th percentile) on aBMD Z-scores under 2 scenarios, when all other chemicals were set to the 75th percentile and when all other chemicals were set to the 25th percentile, and comparing these 2 values. We modeled associations and 95% credible intervals using a Gaussian kernel function and Markov chain Monte Carlo algorithm with 10 000 iterations. Because we modeled serum PFAS and urinary phthalate biomarkers simultaneously, we used a common set of covariates, including age, race/ethnicity, physical activity, indoor smoking by a household member, total dairy intake, sugar-sweetened beverage intake, urinary creatinine, NHANES cycle, and NHANES weights.

We conducted statistical analyses using SAS EG, version 7.1 (SAS Institute, Inc.) and R (Core Team. 2018. R: A language and environment for statistical computing. Vienna Austria:R Foundation for Statistical Computing).

Results

Participants were (mean ± SD) 15 ± 2.1 years old and 47% female (Table 1). Twenty-seven percent of participants were non-Hispanic White, 23% were Mexican American, and 23% were non-Hispanic Black. Participants who were male, non-Hispanic Black, or reported more physical activity tended to have higher aBMD Z-scores. PFOS had the highest median concentration (3.2 ng/mL) of the serum PFAS, and MEP had the highest median concentration (34.5 ng/mL) of the urinary phthalate biomarkers (Table S1) (53). We observed a wide range of positive correlations between individual serum PFAS (eg, for PFDA and PFOS, correlation = 0.56; for PFDA and perfluorohexane sulfonate, correlation = 0.09) and individual urinary phthalate biomarkers (eg, for ∑ DEHP and MCNP, correlation = 0.84; for ∑ DEHP and MEP, correlation = 0.09); correlations between serum PFAS and urinary phthalate biomarkers were low (Figure S1) (53).

Table 1.

Unweighted^a characteristics of 12- to 19-year-old participants (n = 848) in the 2011-2016 NHANES, overall and by total body less head aBMD Z-score quartiles

		aBMD Z-score quartilesb
	Total (N = 848)	Q1 (N = 212)	Q2 (N = 212)	Q3 (N = 212)	Q4 (N = 212)
	Mean ± SD or N (%)
Age, y	15 ± 2.1	15 ± 2.1	15 ± 2.2	15 ± 2.1	15 ± 2.0
Female	395 (47%)	97 (46%)	101 (48%)	101 (48%)	96 (45%)
Race/ethnicity
Mexican American	199 (23%)	42 (20%)	49 (23%)	51 (24%)	57 (27%)
Other Hispanic	94 (11%)	20 (9%)	22 (10%)	27 (13%)	25 (12%)
Non-Hispanic White	225 (27%)	67 (32%)	58 (27%)	54 (25%)	46 (22%)
Non-Hispanic Black	198 (23%)	47 (22%)	51 (24%)	52 (25%)	48 (23%)
Non-Hispanic Asian	89 (10%)	27 (13%)	21 (10%)	21 (10%)	20 (9%)
Other/multi-racial	43 (5%)	9 (4%)	11 (5%)	7 (3%)	16 (8%)
Physical activity, hours per week	5.9 ± 7.3	4.1 ± 5.5	5.4 ± 7.4	6.6 ± 7.6	7.3 ± 8.1
Household member smokes indoors	109 (13%)	26 (12%)	25 (12%)	28 (13%)	30 (14%)
Total dairy, cups	1.8 ± 1.6	1.6 ± 1.3	1.9 ± 1.6	1.9 ± 1.6	1.9 ± 1.7
Sugar-sweetened beverages, cups	0.87 ± 1.5	0.85 (± 1.4)	0.76 (± 1.2)	1.0 (± 2.1)	0.83 (± 1.3)
NHANES cycle
2011-2012	258 (30%)	63 (30%)	81 (38%)	55 (26%)	59 (28%)
2013-2014	319 (38%)	74 (35%)	69 (33%)	92 (43%)	84 (40%)
2015-2016	271 (32%)	75 (35%)	62 (29%)	65 (31%)	69 (33%)

Abbreviations: aBMD, area bone mineral density; NHANES, National Health and Nutrition Examination Survey; Q, quartile.

^aModels do not account for the complex sample design, and therefore results cannot be considered generalizable to the US population.

^baBMD Z-score quartile minimum and maximum values: Q1: -4.01 to -1.12; Q2: 1.11 to -0.30; Q3: -0.30 to 0.39; Q4: 0.39 to 2.72.

Chemical Concentrations and aBMD Z-Scores in Males

In multivariable single-chemical models in males, each biomarker doubling was associated with a lower aBMD Z-score for serum PFOA (difference in aBMD Z-score, -0.24; 95% CI, -0.41 to -0.06) and urinary MBP (-0.09; 95% CI, -0.17 to 0.00) (Fig. 1A). Serum PFOS and urinary MiBP were also associated with lower aBMD, but 95% CIs crossed the null. Other serum PFAS and urinary phthalate biomarkers were not associated with aBMD Z-scores. Findings from unadjusted models were generally similar (Table S2)(53).

Figure 1.

Single-chemical associations from unweighted^a multivariable^b linear regression models of individual serum PFAS and urinary phthalate biomarkers with total body less head aBMD Z-scores in 12- to 19-year-old males (A) and females (B), 2011-2016 National Health and Nutrition Examination Survey (males, N = 453; females, N = 395). ^aModels do not account for the complex sample design, and therefore results cannot be considered generalizable to the US population. ^bAdjusted for age, race/ethnicity, moderate/vigorous physical activity, indoor smoking by a household member, recent sugar-sweetened beverage consumption, recent total dairy consumption, NHANES weights, and NHANES cycle. Models for phthalates were additionally adjusted for urinary creatinine. Abbreviations: aBMD, areal bone mineral density; DEHP, di(2-ethylhexyl)phthalate; MBP, mono-n-butyl phthalate; MBzP, mono-benzyl phthalate; MCNP, mono(carboxynonyl) phthalate; MCOP, mono(carboxyoctyl) phthalate; MCPP, mono-(3-carboxypropyl) phthalate; MEP, mono-ethyl phthalate; MIBP, mono-isobutyl phthalate; PFDA, perfluorodecanoate; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate.

In line with the single-chemical models, in the BKMR mixture models for males, higher serum PFOA and PFOS and urinary MBP and MiBP were linearly associated with lower aBMD Z-scores, with the strongest associations for PFOA and MBP (Figure S2A) (53). ∑DEHP and MBzP were nonlinearly associated with aBMD Z-score, with males with the lowest and highest concentrations having the lowest aBMD Z-scores. The overall serum PFAS/urinary phthalate biomarker mixture was also associated with a lower aBMD Z-score, although 95% credible intervals crossed the null (Fig. 2A). Associations between individual serum PFAS or urinary phthalate biomarkers with aBMD Z-scores were not modified by other chemical biomarkers (data not shown). Posterior inclusion probabilities indicated that of the 3 chemical classes, high- and low-molecular-weight urinary phthalate biomarkers contributed the most to the overall fit of the model. PFOA had the highest contribution among the PFAS, ∑DEHP among the high-molecular-weight phthalate biomarkers, and MiBP among the low-molecular-weight phthalate biomarkers (Table 2), indicating that these chemicals are the primary drivers of the association with lower aBMD Z-score.

Figure 2.

Associations (95% credible intervals) from unweighted^a Bayesian kernel machine regression models of the overall urinary phthalate metabolite and serum PFAS mixture with total body less head aBMD Z-score^b in 12-19 year old males (A) and females (B), 2011-2016 National Health and Nutrition Examination Survey (males, n = 453; females, n = 395). ^aModels do not account for the complex sample design, and therefore results cannot be considered generalizable to the US population. ^bFrom BKMR models of the association between the overall urinary phthalate metabolite/serum PFAS mixture when all chemical biomarkers are set at a given quantile versus the median. Models are adjusted for age, race/ethnicity, moderate/vigorous physical activity, indoor smoking by a household member, recent sugar-sweetened beverage consumption, recent total dairy consumption, urinary creatinine concentrations, NHANES weights, and NHANES cycle. Abbreviation: aBMD, areal bone mineral density; NHANES, National Health and Nutrition Examination Survey; PFAS, per- and polyfluoroalkyl substances.

Table 2.

PIPs from Bayesian kernel machine regression models^a of chemical mixtures and total body less head aBMD Z-scores in 12- to 19-year-old males (n = 453) and females (n = 395), 2011-2016 NHANES

	Biomarker groups	Group PIP	Individual biomarkers	Individual PIP
Males	Serum PFAS	0.65	PFOA	0.76
			PFOS	0.12
			PFDA	0.03
			PFHxS	0.06
			PFNA	0.03
	High-molecular-weight urinary phthalate biomarkers	0.79	∑DEHPb	0.65
			MBzP	0.22
			MCOP	0.02
			MCNP	0.06
			MCPP	0.05
	Low-molecular-weight urinary phthalate biomarkers	0.77	MBP	0.39
			MiBP	0.58
			MEP	0.03
Females	Serum PFAS	0.33	PFOA	0.26
			PFOS	0.10
			PFDA	0.17
			PFHxS	0.23
			PFNA	0.24
	High-molecular-weight urinary phthalate biomarkers	0.26	∑DEHPb	0.19
			MCOP	0.23
			MCPP	0.23
			MBzP	0.19
			MCNP	0.16
	Low-molecular-weight urinary phthalate biomarkers	0.36	MBP	0.28
			MiBP	0.51
			MEP	0.21

Abbreviations: aBMD, areal bone mineral density; BKMR, Bayesian kernel machine regression; DEHP, di(2-ethylhexyl)phthalate; MBzP, mono-benzyl phthalate; MCNP, mono(carboxynonyl) phthalate; MCOP, mono(carboxyoctyl) phthalate; MCPP, mono-(3-carboxypropyl) phthalate; MEP, mono-ethyl phthalate; MIBP, mono-isobutyl phthalate; NHANES, National Health and Nutrition Examination Survey; PFDA, perfluorodecanoate; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; MBP, mono-n-butyl phthalate; PIP, posterior inclusion probability.

^aPIPs indicate the relative contribution of individual or group biomarkers to the overall fit of the BKMR model. Models are adjusted for age, race/ethnicity, physical activity, indoor smoking by a household member, urinary creatinine concentrations, NHANES weights, and NHANES cycle.

^bMolar sum of mono (2-ethylhexyl) phthalate, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, and mono(2-ethyl-5-carboxypentyl) phthalate.

Chemical Concentrations and aBMD Z-Score in Females

In females, serum PFOA and urinary MiBP were associated with higher aBMD Z-scores, but 95% CIs crossed the null (Fig. 1B). Other serum PFAS and urinary phthalate metabolites were not associated with aBMD Z-scores in single-chemical models. Mixture models were generally consistent with single-chemical models. Serum PFOA and urinary MiBP were positively associated with aBMD Z-score, and urinary MBP was associated with lower aBMD Z-score. There was no association between other individual serum PFAS, urinary phthalate metabolites (Figure S2B) (53), or the overall serum PFAS/urinary phthalate biomarker mixture (Fig. 2B) with aBMD Z-score in BKMR models. Individual chemical-aBMD Z-score associations were not modified by other serum PFAS or urinary phthalate biomarkers (data not shown).

Sensitivity Analyses

Effect estimates for were generally similar across race/ethnicity categories for both males and females in multivariable single-chemical models (Figure S3) (53) (eg, for PFOA, Hispanic males, -0.31 [95% CI, -0.69 to 0.07]; non-Hispanic White males, -0.31 [95% CI, -0.65 to 0.03]; non-Hispanic Black males, -0.26 [95% CI, -0.64 to 0.12]).

Discussion

In 2011 through 2016 NHANES, higher concentrations of several serum PFAS and urinary phthalate biomarkers, as well as the complex mixture of PFAS and phthalate biomarkers, were associated with lower aBMD Z-scores in adolescent males. In females, select serum PFAS and urinary phthalate biomarkers were associated with higher aBMD Z-scores, although 95% CIs crossed the null. Effect estimates were similar across different race/ethnicities.

Our findings align with previous studies that have shown prenatal (54, 55) and childhood (10, 11) PFAS concentrations are associated with lower aBMD in childhood. Our finding of an association of serum PFOS and PFOA with lower aBMD Z-scores in males is consistent with a recent study in the Health Outcomes and Measures of the Environment Study that reported an association between higher concentrations of prenatal PFOA with lower aBMD Z-scores at some skeletal sites in 12-year-old males, but not females (54). Our finding of a positive association between serum PFOA and aBMD in female adolescents, however, is in contrast to the Avon Longitudinal Study of Parents and Children, which found higher prenatal concentrations of several PFAS (including PFOS and PFOA) to be associated with lower aBMD in 17-year-old females (55). In the Boston-area Project Viva cohort, cross-sectional associations of higher PFAS with lower aBMD Z-scores in 8-year-old children did not differ by sex (10). Notably, the associations we observed were modestly stronger than the associations in males in Project Viva (change in TBLH aBMD Z-scores per doubling of PFOA: -0.26 [95% CI, -0.45 to -0.08] in NHANES vs -0.11 [95% CI, -0.23 to 0.00] in Project Viva) (10). The differences by sex and in the magnitude of the PFAS-aBMD association in NHANES vs Project Viva may reflect the timing of the aBMD measurement because peak bone accrual occurs somewhat earlier in adolescent females (~age 12 years) vs males (~14 years), and children in the Project Viva study had not yet initiated peak bone accrual. The field would benefit from longitudinal studies relating PFAS and phthalate metabolite concentrations to aBMD across childhood and adolescence to elucidate whether chemical-aBMD associations vary by age at aBMD measurement.

We found males with higher urinary MBP and MiBP to have lower aBMD Z-scores and females with higher urinary MiBP to have higher aBMD Z-scores. Positive associations between maternal phthalate biomarker concentrations and aBMD, particularly in females, were also reported in the Health Outcomes and Measures of the Environment Study (56). In the Generation R cohort, there was no association between maternal phthalate biomarker concentrations and bone mineral content in childhood (57). In contrast, previous cross-sectional studies consistently indicate that phthalate biomarkers, and particularly low-molecular-weight phthalate biomarkers, are associated with lower aBMD in postmenopausal females (13-15). In a cross-sectional analysis of the Women’s Health Initiative, those not using hormone replacement therapy with higher urinary concentrations of ∑ dibutyl phthalate (parent diester of MBP), MCPP, and ∑ diisobutyl phthalate (parent diester of MiBP) had lower total hip aBMD (14). In postmenopausal females, higher urinary MBP, MCPP, and MBzP were associated with lower total hip and femoral neck aBMD in the 2005 through 2008 NHANES (13), and higher concentrations of urinary MEP and ∑ low-molecular-weight phthalate metabolites were associated with lower total spine aBMD in the 2005 through 2010 NHANES (15). Associations between phthalate biomarkers and aBMD may be more evident later in life, when there is more variability in aBMD and low bone mass becomes prevalent.

Serum PFAS and urinary phthalate biomarkers may disrupt common pathways to affect aBMD. Both chemicals activate peroxisome proliferator-activated receptor-γ (58, 59), which impairs osteoblastogenesis (60). In addition, both chemicals induce hormonal changes that may disrupt bone homeostasis. PFAS and phthalates increase inflammatory cytokines (61, 62), lower IGF-1 (63, 64), antagonize androgen receptor pathways (8, 9), and lower testosterone levels in children and adults (64-66). Our finding of associations between higher levels of select chemicals and lower aBMD Z-score in males but not females may be related to differences in biological impact across chemicals. For example, in a large cross-sectional analysis of children in the C8 Health Project, males with higher biomarker concentrations of PFOA and PFOS, but not other PFAS, had lower testosterone (64). Likewise, MBP was weakly associated with lower testosterone in an analysis of 12- to 20-year-old males in the 2011 and 2012 NHANES. Notably, although urinary ∑ DEHP metabolites and MCOP have also been associated with lower testosterone in males (66), they were not associated with aBMD Z-score in males in our study. Because testosterone is a more important factor in bone formation in males compared with females (67), decreased testosterone could also explain the observed sex-specific effects on aBMD. Future studies should examine whether associations between ∑ DEHP metabolites and hormone concentrations follow a nonlinear dose-response such as we observed for the ∑ DEHP metabolites-aBMD Z-score association.

An important strength of our analysis was our use of sophisticated environmental mixture models to account for chemical coexposures. Consistent with single-chemical models, when we used BKMR to account for correlations and interactions between all of the measured chemicals, PFOA and MBP, and to a lesser extent, PFOS and MiBP, were associated with low aBMD Z-scores in males. Although ∑ DEHP and MBzP also had weak linear associations with aBMD Z-scores in males in single-chemical generalized additive and linear regression models, BKMR models revealed nonlinear associations, suggesting the importance of accounting for chemical coexposures. In females, we observed positive associations of serum PFOA and urinary MiBP with aBMD Z-score across both single-chemical and BKMR models, reinforcing our findings.

In the present study, we found that the chemical-aBMD associations were similar across non-Hispanic White, non-Hispanic Black, and Hispanic individuals. This is another important addition to the literature because chemical exposure profiles and bone health may vary by race/ethnicity. Prior studies have shown that non-Hispanic Black children have lower concentrations of several plasma PFAS (18) and higher urinary concentrations of low-molecular-weight phthalate metabolites (17). Also, non-Hispanic Black children have been shown to have higher aBMD (28). Future studies in larger populations would benefit by examining associations across a wider range of race/ethnicities.

Our study has several limitations. First, the cross-sectional design limits inference about the temporality of the association. Also, although a single measure of urinary phthalate metabolites has been shown to be moderately predictive of longer term (6-month) exposure (68-70), urinary phthalate metabolites have a short half-life (< 24 hours). Thus, whereas aBMD is a marker of lifetime bone accrual, a single measure of phthalate metabolites reflects exposure within the recent past, limiting our ability to see an association. The representativeness of a single biomarker is less of a concern for serum PFAS, which have half-lives of years (71), thus better reflecting longer term exposure in relation to aBMD. Second, prospective studies are needed to examine the extent to which pubertal status and body size mediate associations of PFAS and phthalate biomarkers with aBMD Z-scores. Finally, although we accounted for potential confounders including diet, physical activity, and environmental tobacco smoke, error in these measurements may have resulted in some residual confounding.

In conclusion, in a racially diverse population of 12 to 19 year olds, males with higher concentrations of PFOA, PFOS, MBP, and MiBP had lower aBMD Z-scores, whereas females with higher concentrations of PFOA and MiBP had higher aBMD Z-scores. Our findings have public health implications because reducing exposure to PFAS and phthalates may improve peak bone accrual in adolescent males and set the stage for improved bone health across the lifespan. Exposure to PFAS and phthalates can be reduced through avoidance of affected consumer products or, more equitably, through public policies requiring substitution of these chemicals with safer alternatives.

Glossary

Abbreviations

aBMD

areal bone mineral density

BKMR

Bayesian kernel machine regression

CDC

Centers for Disease Control and Prevention

DEHP

di(2-ethylhexyl)phthalate

LOD

limit of detection

MBP

mono-n-butyl phthalate

MBzP

mono-benzyl phthalate

MCNP

mono(carboxynonyl) phthalate

MCOP

mono(carboxyoctyl) phthalate

MCPP

mono-(3-carboxypropyl) phthalate

MEP

mono-ethyl phthalate

MIBP

mono-isobutyl phthalate

NHANES

National Health and Nutrition Examination Survey

PFAS

per- and polyfluoroalkyl substances

PFDA

perfluorodecanoate

PFOA

perfluorooctanoate

PFOS

perfluorooctane sulfonate

TBLH

total body less head

Funding

This work was supported by the National Institutes of Health [R01ES030101].

Disclosures

The authors have nothing to disclose.

Data Availability

All datasets generated during and/or analyzed during the current study are publicly available and free of charge.