Prenatal and childhood exposure to per- and polyfluoroalkyl substances (PFASs) and child cognition

Abstract

Background

Per- and polyfluoroalkyl substances (PFASs) are suspected developmental toxicants, but epidemiological evidence on neurodevelopmental effects of PFAS exposure is inconsistent. We examined associations of prenatal and childhood PFAS exposure with performance on assessments of cognition in children.

Methods

We included mother-child pairs from Project Viva, a longitudinal Boston-area birth cohort enrolled during 1999–2002. We quantified concentrations of eight PFASs, including perfluorooctane sulfonate (PFOS), perfluorooctanoate (PFOA), and perfluorohexane sulfonate (PFHxS), in plasma collected from women during pregnancy (median 9.7 weeks gestation) and from children at a visit in mid-childhood (median age 7.7 years). In early childhood (median age 3.2 years) we administered standardized assessments of visual motor skills and vocabulary comprehension, and in mid-childhood we assessed visual motor skills, visual memory, and verbal and non-verbal intelligence. Using multivariable regression, we estimated associations of prenatal and childhood PFAS plasma concentrations with children’s cognitive assessment scores, adjusted for relevant covariates including breastfeeding, maternal intelligence, parental education, and household income. Samples sizes ranged from 631–971, depending on analysis.

Results

Prenatal PFAS concentrations were associated with both better and worse cognitive performance; children with top quartile prenatal concentrations of some PFASs had better visual motor abilities in early childhood and non-verbal IQ and visual memory in mid-childhood, while children with upper quartile prenatal PFOA and PFOS had lower mid-childhood visual-motor scores. In cross-sectional analyses of mid-childhood PFAS concentrations and cognitive assessments, visual-motor scores on the Wide Range Assessment of Visual Motor Abilities (WRAVMA) (standardized mean=100, standard deviation=15) were lower among children with higher PFHxS (fourth quartile (Q4) vs. Q1: −5.0, 95% confidence interval (CI): −9.1, −0.8). Upper quartiles of childhood PFOA and PFOS were also associated with somewhat lower childhood WRAVMA scores, but childhood PFASs were not associated with verbal or non-verbal IQ or visual memory.

Conclusions

We present evidence suggesting associations of prenatal and childhood PFAS exposure with lower childhood visual motor abilities. Other results were inconsistent, with higher prenatal PFASs associated in some cases with better cognitive outcomes.

1. INTRODUCTION

Per- and polyfluoroalkyl substances (PFASs) are synthetic chemicals used in consumer and industrial products including oil-resistant coatings for food packaging and stain-resistant fabric treatments.^1,2 Several PFASs, including perfluorooctane sulfonate (PFOS) and perfluorooctanoate (PFOA), have been identified as persistent and bioaccumulative chemicals that may have neurotoxic properties.² PFASs are frequently detected in human biosamples, including those of pregnant women^1–5 and children,^6–8 and cross the placenta.^9–11

Rodent studies suggest that PFASs may disrupt neurodevelopment,^12–14 but epidemiological evidence is inconsistent and relatively limited. In prospective cohort studies, prenatal PFAS concentrations were associated with worse motor skills at age two years,¹⁵ behavior problems at age 5–9 years,¹⁶ and poorer executive function at age 5–8 years.¹⁷ Other cohort studies of prenatal or perinatal PFAS exposures and neurobehavioral outcomes in children have reported null^18–24 or protective^60,64,65,66 associations. Cross-sectional studies in U.S. children and adolescents have linked serum concentrations of PFASs with attention deficit hyperactivity disorder (ADHD) diagnosis^6,25 and greater impulsivity.²⁶

Effects of PFASs on neurodevelopment may vary depending on of timing of exposure, but few prior studies have examined PFAS exposure at multiple time points in relation to childhood cognition or behavior.^24,27 A cohort study from the Faroe Islands (n=539) reported associations of PFASs measured at age 5 with behavior problems assessed at age 7, but no associations with prenatal PFASs,²⁴ while in a small U.S. cohort (n=167), higher prenatal and childhood PFASs were associated with better child reading skills.²⁷ In a prospective cohort of U.S. mothers and children, we examined associations of prenatal (maternal plasma from pregnancy) and childhood plasma PFAS concentrations with children’s performance on assessments of visual motor skills and vocabulary comprehension in early childhood and visual motor skills, visual memory, and verbal and non-verbal intelligence in mid-childhood. We hypothesized that higher PFAS exposure would be associated with poorer cognitive outcomes, and sought to identify potentially sensitive exposure periods for effects of PFAS exposure on cognitive development.

2. METHODS

2.1 Study population

Participants were drawn from Project Viva, a prospective pre-birth cohort enrolled 1999–2002 at mothers’ first prenatal visit (median 9.9 weeks of gestation) at eight locations of Atrius Harvard Vanguard Medical Associates, a multispecialty group practice in urban and suburban Eastern Massachusetts.²⁸ Eligibility criteria included gestational age <22 weeks at enrollment, ability to answer questions in English, and no plans to move away from the study area before delivery. A total of 2,128 mothers with live, singleton births was initially enrolled, and a subset (n=1,668; 78%) agreed to contribute blood samples at an in-person visit in pregnancy (median 9.7 weeks gestation, range 4.8–21.4). Mother-child pairs were followed prospectively; mothers provided demographic and health-related information through annual questionnaires and children attended periodic in-person visits assessing health and development, including a visit in in early childhood (2.8–6.3 years) attended by 1,294 pairs (61%) and a visit in mid-childhood (6.6–10.9 years), attended by 1,116 pairs (52%). A subset of children (n=702; 63% of those attending visit, 34% of those enrolled at birth) agreed to contribute blood samples at the mid-childhood visit in 2007–10. Additional description of the Project Viva cohort and participation flow is available in a published cohort profile.²⁸

The Institutional Review Boards of participating institutions approved all study protocols. 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) laboratory did not constitute engagement in human subjects research. All study forms are available at https://www.hms.harvard.edu/viva/.

2.2 Cognitive assessments

A pediatric neuropsychologist trained Project Viva staff to administer standardized assessments of cognitive development to child participants during in-person study visits in early childhood (median age 3.2 years) and mid-childhood (median age 7.7 years) (Table 1).

Table 1.

Cognitive assessments administered at Project Viva in-person study visits in early childhood (median age 3.2 years; range 2.8–6.3) and mid-childhood (median age 7.7 years; range 6.6–10.9).

Assessment	Subtest	Primary cognitive domains assessed	Administered at early childhood study visit	Administered at mid- childhood study visit
Peabody Picture Vocabulary Test (PPVT-III)	–	Word knowledge	×
Wide Range Assessment of Visual Motor Abilities (WRAVMA)	Drawing	Visual-motor	×	×
	Pegboard	Fine motor	×
	Matching	Visual-spatial perception	×
Kaufman Brief Intelligence Test (KBIT-2)	Verbal IQ	Word knowledge, crystallized intelligence		×
	Non-verbal IQ	Visual-spatial perception, fluid intelligence		×
Visual Memory Index of the Wide Range Assessment of Memory and Learning (WRAML2)	Design Memory	Visual memory, visual-spatial perception		×
	Picture Memory	Visual memory		×

In early childhood, staff administered the Peabody Picture Vocabulary Test, 3rd edition (PPVT-III),²⁹ which assesses vocabulary comprehension, and the Wide Range Assessment of Visual Motor Abilities (WRAVMA),³⁰ which assesses visual-motor (drawing subtest), fine motor (pegboard subtest) and visuospatial (matching subtest) skills. The PPVT-III is valid for use in children and adults from 2.5 to 90+ years and was standardized based on a sample representative of the general U.S. population; PPVT-III scores in children correlate well (r ≥ 0.90) with verbal and full-scale intelligence quotient (IQ) scores on the full length Wechsler Intelligence Scale for Children-III (WISC-III).²⁹ The WRAVMA is designed for children aged 3 to 17 and was normed using a representative sample of U.S. children; WRAVMA composite scores are moderately correlated (r=0.62) with WISC-III full-scale IQ and moderately to highly correlated with other standardized tests of visual-motor ability.³⁰

In mid-childhood, staff assessed verbal and non-verbal intelligence using the Kaufman Brief Intelligence Test, Second Edition (KBIT-2),³¹ visual-motor skills using the WRAVMA drawing subtest, and visual memory (design memory and picture memory) using the Visual Memory Index of the Wide Range Assessment of Memory and Learning, Second Edition (WRAML2).³² The KBIT-2 is a valid and reliable measure for children and adults age 4 to 90, standardized using a representative U.S. sample; KBIT-2 scores are moderately to highly correlated with relevant subscores on the WISC-III (r=0.76 for KBIT-2 IQ composite and WISC-III full-scale IQ).³¹ The WRAML2 is standardized for ages 5 to 90 based on a representative U.S sample; among children aged 6–16, the WRAML2 Visual Memory Index showed moderate correlation with relevant indices from the Children’s Memory Scale (r=0.48 for General Memory, 0.52 for Attention/Concentration).³²

Assessors rated their confidence in the accuracy of each assessment on a scale of 1–5 (based on testing conditions, child’s mood, etc.); assessments with low confidence (≤2) were excluded from primary analyses. Project Viva staff double-scored all cognitive assessments using published scoring guidelines; in cases where two scorers disagreed, group adjudication was used. WRAVMA visual-motor scoring also employed supplementary guidelines developed by a pediatric neuropsychologist to ensure consistency among scorers.³³ Scaled scores were standardized to mean=100, standard deviation (SD)=15 for the PPVT-III, KBIT-2 and WRAVMA, and mean=10, SD=3 for WRAML2 design memory and picture memory subscores, using published reference data.^29–32 Staff administering and scoring assessments had no knowledge of participants’ PFAS plasma concentrations, nor did participants themselves.

2.3 PFAS measurements

We quantified PFAS concentrations in maternal plasma samples from pregnancy (4.8–21.4 weeks gestation) and child plasma samples from mid-childhood (6.6–10.9 years). Following collection, plasma specimens were stored in PFAS-free cryovials in liquid nitrogen freezers. Samples were subsequently thawed, aliquoted, shipped to the Division of Laboratory Sciences at CDC and analyzed for concentrations of eight PFASs: PFOS, PFOA, perfluorohexane sulfonate (PFHxS), perfluorononanoate (PFNA), 2-(N-ethyl-perfluorooctane sulfonamido) acetate (EtFOSAA; also known as Et-PFOSA-AcOH), 2-(N-methyl-perfluorooctane sulfonamido) acetate (MeFOSAA; also known as Me-PFOSA-AcOH), perfluorodecanoate (PFDeA), and perfluorooctane sulfonamide (FOSA; also known as PFOSA) using on-line solid-phase extraction coupled with isotope dilution high-performance liquid chromatography-tandem mass spectrometry. Analytical methods for the maternal and child samples were the same as those used to analyze PFAS concentrations in the 2011–2012³⁴ and 2013–2014³⁵ National Health and Nutrition Examination Survey (NHANES) cycles, respectively. To ensure accuracy and reliability, the laboratory analyzed low and high-concentration quality control materials, analytical standards, and reagent and serum blanks along with the study samples; the laboratory successfully participated in external Quality Assessment schemes.³⁶ PFOS and PFOA measures represented total PFOS and PFOA (sum of linear and branched isomers). Prenatal samples were shipped and analyzed in 2014 and child plasma samples were shipped and analyzed in 2015; 1,645 prenatal plasma samples (of 1,668; 99%) and 653 child plasma samples (of 702; 93%) had sufficient volume for PFAS quantification. Limits of detection (LOD) were 0.1 ng/mL, except for prenatal PFOS measures, for which the LOD was 0.2 ng/mL; values below LOD were estimated as LOD/√2.³⁷

2.4 Covariates

Study staff collected data on participant demographics and health-related behaviors via study questionnaires and interviews and assessed mothers’ cognition using the PPVT-III (at the early childhood visit) and the KBIT-2 (mid-childhood visit). At the mid-childhood visit, mothers completed the Home Observation for Measurement of the Environment - Short Form (HOME-SF), which measures support for cognitive development in the child’s home. HOME-SF scores range from 0 to 22, with higher scores representing better support.³⁸ Plasma creatinine and albumin were measured in the same maternal pregnancy plasma samples analyzed for PFASs. We used plasma creatinine measurements to estimate maternal glomerular filtration rate (GFR) at the time of pregnancy blood draw according to the Cockcroft-Gault equation.³⁹

2.5 Statistical analyses

Missing covariates were imputed using a chained equation multiple imputation model (PROC MI in SAS) including all exposure and outcome variables, all study covariates, and auxiliary variables potentially predictive of exposures, outcomes, covariates or missingness, including maternal and child dietary factors and child anthropometric and clinical measurements. We generated 50 imputed data sets including all Project Viva participants (n=2,128)⁴⁰ and pooled beta estimates from imputed data sets according to Rubin’s rules.⁴¹ The use of multiply imputed data allows for unbiased effect estimates if data are missing at random, meaning that missingness depends only on factors included in imputation model, a less restrictive assumption than is necessary for complete case analyses.⁴⁰ Primary models included participants with imputed covariate data, but excluded participants lacking PFAS measures or cognitive outcome scores. In sensitivity analyses, we also ran complete case analyses and models including imputed exposures and outcomes for all participants with at least one measure of PFASs (prenatal and/or childhood) and at least one cognitive assessment (early childhood and/or mid-childhood) (n=1,226).

Correlations among PFASs plasma concentrations, outcomes, and covariates were determined using Spearman rank correlation coefficients. We examined associations of maternal plasma PFASs with early and mid-childhood cognitive assessments, as well as cross-sectional associations of child plasma PFAS with mid-childhood cognitive assessments. To assess the shape of exposure-outcome associations, we fit generalized additive models with cubic regression splines (three degrees of freedom) for continuous exposures.

All models were adjusted for child sex and age at cognitive testing, along with covariates hypothesized to be potential confounders of the studied associations based on prior knowledge, according to Directed Acyclic Graph (DAG) theory (Figures S1 and S2).^42,43 Covariates included for all models were: year of blood collection (for prenatal or mid-childhood blood samples, depending on model), maternal race/ethnicity (black, white, Hispanic, Asian, other), age (<25, 25–34, ≥35 years), maternal and paternal education (<college/college/graduate degree), and maternal intelligence scores (PPVT-III for early childhood cognition models, KBIT-2 for mid-childhood cognition models). Models for early childhood cognition also included annual household income at enrollment (<$40K, $40-≤70K, >$70K), while models for mid-childhood cognition included mid-childhood annual household income (<$40K, $40-≤70K, $70-≤150K, >$150K) and HOME-SF score.

As changes in plasma volume expansion and renal function in pregnancy may affect measured plasma concentrations of PFASs and may also relate to pregnancy health,⁴⁴ models for maternal plasma PFASs were additionally adjusted for estimated GFR and the gestational week in which plasma was collected, along with pre-pregnancy body mass index (kg/m²). Prenatal PFAS models were also adjusted for maternal smoking status in pregnancy, a predictor of maternal PFAS plasma concentrations in this population.⁴⁵ Finally, because PFASs can partition into breastmilk, prior breastfeeding may influence maternal PFAS levels and breastfeeding may represent a source of PFAS exposure for children.^46,47 Information on prior breastfeeding was not available in Project Viva, so we created a proxy indicator of prior breastfeeding derived from parity and report of breastfeeding (≥1 month) of the studied child. Prenatal PFAS models were adjusted for this indicator of prior breastfeeding incorporating maternal parity (nulliparous, parous with no breastfeeding, parous with breastfeeding), while childhood PFAS models were adjusted for breastfeeding duration (in months) of the studied child and maternal parity (0, ≥1).

We assessed collinearity among covariates for each model by calculating variance inflation factors (VIF); VIFs were <2 for all variables, suggesting that covariate collinearity did not substantially reduce precision of effect estimates. As covariate-adjusted spline models suggested non-linearity in a number of the studied PFAS-cognitive assessment associations, we categorized PFAS concentrations into quartiles for primary analyses. Quartiles were determined separately for prenatal PFASs and childhood PFASs based on concentrations among all children with the relevant PFAS measurement and at least one measured cognitive outcome (n=1095 for prenatal PFASs and n=653 for childhood PFASs).

As primary analyses, we ran separate multivariable linear regression models assessing associations of prenatal plasma concentrations of each PFAS with each early childhood and mid-childhood cognitive assessment score, and cross-sectional associations of child plasma concentrations of each PFAS with each mid-childhood cognitive assessment score. As secondary analyses, to examine whether cross-sectional associations might be confounded by prenatal PFAS exposure, we ran cross-sectional models for each mid-childhood PFAS with additional adjustment for maternal pregnancy plasma concentrations of the same PFAS, modeling maternal PFAS concentrations as cubic regression splines (3 degrees of freedom) to allow for non-linear dose-response relationships; these analyses included only the subset of the study population with both prenatal and childhood PFAS measures (n=511). To assess whether sex might be an effect measure modifier, we re-ran models stratified by child sex, and conducted log-likelihood ratio tests to assess whether model fit improved with inclusion of multiplicative interaction terms for PFAS quartile × sex.

As sensitivity analyses, we re-ran prenatal PFAS models with additional adjustment for plasma albumin, as albumin is a primary binding site for PFASs and plasma albumin may also reflect hemodynamic variability in pregnancy. To evaluate the influence of adjusting for breastfeeding, we ran models excluding prior breastfeeding and breastfeeding duration covariates. To evaluate potential bias related to inter-rater variability, we ran models adjusted for assessor and excluding assessments from assessors who conducted <20 assessments. Finally, we ran models excluding assessments with assessor confidence scores of 3/5 (primary models excluded assessments with confidence scores ≤2).

We prepared data sets and completed multiple imputation in SAS Version 9.3 (SAS Institute Inc., Cary, NC); all other analyses were conducted in R Version 3.1.3 (R Foundation for Statistical Computing, Vienna, Austria). We based our conclusions about potential relationships between exposures and outcomes on the magnitude and precision of estimates and patterns observed across results.

3. RESULTS

3.1 Participant characteristics and PFAS concentrations

Our study included participants with at least one measure of PFAS plasma concentrations (prenatal and/or mid-childhood) and at least one completed cognitive assessment (in early childhood and/or mid-childhood). Sample sizes ranged from 631 to 971 (representing 30 to 46% of the 2,128 children enrolled at birth); Figure 1 outlines inclusion criteria and sample sizes for each analysis. Table 2 outlines characteristics of study participants, stratified by analysis. Mothers were predominantly white (62–73%, depending on analysis), tended to have high educational attainment (65–70% with college or graduate degrees), and the majority had household income >$70,000/year (60–69%). Distributions of covariates were similar when imputed covariate data were excluded (data not shown); proportions of imputed data were low (≤10%) (Table 2).

Figure 1.

Flow diagram for inclusion in study population (with % of original cohort included at each stage)

Table 2.

Characteristics of study participants, stratified by analysis

Characteristic	N (%) or mean ± standard deviation
	Included in analyses of prenatal PFASs and early childhood cognition (n=986)	Included in analyses of prenatal PFASs and mid-childhood cognition (n=865)	Included in analyses of childhood PFASs and mid-childhood cognition (n=642)
Early childhood cognitive assessments
Peabody Picture Vocabulary Test (PPVT-III)a (n=948)	104.4 ± 14.1	-	-
Wide Range Assessment of Visual Motor Abilities (WRAVMA) Composite Scorea (n=919)	102.1 ± 11.5	-	-
WRAVMA Visual-Motor (a (n=971)	99.0 ± 11.3	-	-
WRAVMA Fine Motora (n=968)	98.0 ± 10.8	-	-
WRAVMA Visual-Spatiala (n=937)	108.1 ± 13.8	-	-
Mid-childhood cognitive assessments
Kaufman Brief Intelligence Test (KBIT-2) Verbal IQa (n=851 with prenatal PFASs, 631 with child PFASs)	-	112.1 ± 14.9	111.1 ± 15.4
KBIT-2 Non-Verbal IQa (n=862 with prenatal PFASs, 640 with child PFASs)	-	106.5 ± 16.9	105.8 ± 16.9
WRAVMA Visual-Motor a (n=853 with prenatal PFASs, 635 with child PFASs)	-	92.1 ± 16.7	93.1 ± 17.2
Visual Memory Index of the Wide Range Assessment of Memory and Learning (WRAML2) Design Memory (WRAML2)b (n=853 with prenatal PFASs, 636 with child PFASs)	-	8.0 ± 2.8	8.1 ± 2.7
WRAML 2 Picture Memoryb (n=855 with prenatal PFASs, 635 with child PFASs)	-	8.9 ± 3.0	9.0 ± 3.0
Maternal characteristics
Year of enrollment
1999	245 (25)	212 (25)	-
2000	362 (37)	329 (38)	-
2001	343 (35)	289 (33)	-
2002	36 (4)	35 (4)	-
Age at enrollment (years)	32.4 ± 5.1	32.1 ± 5.3	31.8 ± 5.6
Parity at enrollment
Nulliparous	468 (47)	415 (48)	271 (42)
Parous	518 (53)	450 (52)	371 (58)
Proxy for breastfeeding of a prior childc
Nulliparous (assume no prior breastfeeding)	468 (47)	415 (48)	271 (42)
Parous, did not breastfeed Project Viva child	100 (10)	76 (9)	57 (9)
Parous, breastfed Project Viva child	391 (40)	343 (40)	283 (44)
Missing data	27 (3)	31 (4)	31 (5)
Education (at enrollment)
Less than college degree	286 (29)	275 (32)	225 (35)
College degree	378 (38)	300 (35)	230 (36)
Graduate degree	320 (32)	286 (33)	183 (29)
Missing data	2 (<1)	4 (1)	4 (1)
Race/ethnicity
White	722 (73)	594 (69)	397 (62)
Black	115 (12)	129 (15)	133 (21)
Asian	40 (4)	39 (5)	29 (5)
Hispanic	60 (6)	59 (7)	46 (7)
Other	47 (5)	40 (5)	33 (5)
Missing data	2 (<1)	4 (<1)	4 (1)
Smoking in pregnancy
Smoked in pregnancy	110 (11)	166 (19)	73 (11)
Former smoker	206 (21)	88 (10)	122 (19)
Never smoker	666 (68)	608 (71)	446 (70)
Missing data	4 (<1)	3 (<1)	1 (<1)
Pre-pregnancy body mass index (kg/m2)	24.8 ± 5.3	24.8 ± 5.3	25.0 ± 5.3
Missing data	2 (<1)	6 (1)	4 (1)
Gestational duration at time of pregnancy blood draw (weeks)	9.6 ± 2.2	10.1 ± 2.3	-
Estimated glomerular filtration rate at time of pregnancy blood draw (mL/min per 1.73 m2)	110.5 ± 49.5	109.4 ± 38.5	-
Missing data	7 (<1)	13 (2)	-
Plasma albumin at time of pregnancy blood draw (g/dL)	8.3 ± 2.8	8.2 ± 2.0	-
Missing data	46 (5)	44 (5)	-
PPVT-III score at early childhood cognitive assessment	106.1 ± 14.6	-	-
Missing data	23 (2)	-	-
IQ (KBIT-2 composite) at mid-childhood cognitive assessment	-	106.9 ± 15.2	105.1 ± 15.6
Missing data	-	15 (2)	13 (2)
Paternal characteristics
Education (at enrollment)
Less than college degree	287 (29)	251 (29)	199 (31)
College degree	332 (34)	284 (33)	199 (31)
Graduate degree	281 (28)	241 (28)	168 (26)
Missing data	86 (9)	89 (10)	76 (12)
Child characteristics
Sex
Female	471 (48)	419 (48)	305 (48)
Male	515 (52)	446 (52)	337 (52)
Duration of breastfeeding (months up to 12)	6.3 ± 4.5	6.5 ± 4.6	6.4 ± 4.5
Missing data	51 (5)	58 (7)	52 (8)
Age at early childhood assessment (years)	3.3 ± 0.4	-	-
Age at mid-childhood assessment (years)	-	8.0 ± 0.9	7.9 ± 0.8
Year of mid-childhood blood draw
2007	-	-	67 (10)
2008	-	-	220 (34)
2009	-	-	206 (32)
2010	-	-	149 (23)
Household characteristics
HOME-SF scored at mid-childhood assessment	-	18.4 ± 2.2	18.2 ± 2.3
Missing data	-	26 (3)	13 (2)
Household annual income at enrollment
≤$40 K	115 (12)	-	-
>$40–≤$70 K	217 (22)	-	-
>$70 K	591 (60)	-	-
Missing data	63 (6)	-	-
Household annual income at mid-childhood
≤$40 K	-	99 (11)	90 (14)
>$40–≤$70 K	-	110 (13)	96 (15)
>$70–≤$150 K	-	367 (42)	272 (42)
>$150 K	-	234 (27)	148 (23)
Missing data	-	55 (6)	36 (6)

^aPPVT-III, WRAVMA and KBIT-2 scores standardized to mean=100, SD=15.

^bWRAML2 scores standardized to mean=10, SD=3.

^cEstimated using maternal parity and breastfeeding data for the Project Viva child.

^dHome Observation for Measurement of the Environment (Short Form); scale: 0–22, with higher scores representing better support for cognitive development in home.

Participants excluded due to missing exposure or outcome data differed somewhat from included participants. For example, participants excluded from analyses of mid-childhood PFAS and mid-childhood cognitive outcomes had somewhat lower mean WRAVMA (90.7 vs. 92.2) and WRAML2 Picture Memory scores (7.9 vs. 8.9), slightly higher KBIT-2 (verbal: 112.9vs. 111.7; non-verbal: 107.1 vs. 106.2) and WRAML2 Design Memory scores (8.8 vs. 8.0) and slightly higher median prenatal PFOA (6.0 ng/mL vs. 5.4) and PFOS (26.2 ng/mL vs. 24.3) concentrations than included participants (Table S1). Excluded children also had shorter duration of breastfeeding, higher annual household income at mid-childhood, lower maternal parity, higher maternal IQ, and a higher proportion of white and lower proportion of black mothers than included children (Table S1).

Plasma concentrations of PFASs are presented in Table 3. We detected PFOS, PFOA, PFHxS, and PFNA in >98% of prenatal and childhood samples. EtFOSAA and MeFOSAA were detected in >99% of prenatal samples, but less often in childhood samples (5.1% and 65.2%, respectively), while PFDeA was detected more often in childhood samples (88.3%) versus prenatal samples (45.1%). FOSA was detected in 10.9% of prenatal samples and 0.1% of childhood samples. Given low detection frequencies, PFDeA and FOSA were excluded from analyses of prenatal PFASs and EtFOSAA, MeFOSAA, and FOSA were excluded from analyses of childhood PFASs. Correlations among PFASs quantified in the same sample were moderate to high (Spearman correlation coefficients 0.19–0.72 for prenatal samples, 0.14–0.78 for childhood), with PFOA and PFOS most highly correlated (Tables S2–S3). Correlations between prenatal and childhood concentrations of the same PFAS ranged from 0.09 for PFNA to 0.40 for PFHxS (Table S4).

Table 3.

Concentrations (in ng/mL) of per- and polyfluoroalkyl substances (PFASs) in maternal pregnancy plasma and child plasma (median age 7.7) for study participants^a

PFAS	% Detect (≥LOD)	Median (25th %ile–75th %ile)	Minimum	Maximum
Prenatal concentrations (sampled 1999–2002, n=1095)
PFOA	100	5.6 (4.1–7.7)	0.8	49.3
PFOS	99.9	24.9 (18.4–34.4)	<LOD	185.0
PFHxS	99.3	2.4 (1.6–3.7)	<LOD	46.4
PFNA	98.9	0.6 (0.5–0.9)	<LOD	6.0
EtFOSAA	99.6	1.2 (0.7–1.9)	<LOD	33.6
MeFOSAA	100	1.9 (1.2–3.0)	0.1	29.7
PFDeA	45.1	<LOD (<LOD–0.3)	<LOD	3.0
FOSA	10.9	All <LOD	<LOD	1.8
Childhood concentrations (sampled 2007–2010, n=642)
PFOA	99.5	4.4 (3.1–6.0)	<LOD	14.3
PFOS	99.5	6.2 (4.2–9.7)	<LOD	51.4
PFHxS	99.5	1.9 (1.2–3.4)	<LOD	56.8
PFNA	99.5	1.5 (1.1–2.3)	<LOD	25.7
EtFOSAA	5.1	All <LOD	<LOD	1.4
MeFOSAA	65.2	0.3 (<LOD–0.6)	<LOD	6.7
PFDeA	88.3	0.3 (0.2–0.5)	<LOD	1.9
FOSA	0.1	All <LOD	<LOD	0.5

^aLimit of detection (LOD) was 0.1 ng/mL except for prenatal PFOS measures (LOD=0.2 ng/mL)

3.2 PFAS exposures and cognitive outcomes

In covariate-adjusted models assessing quartiles of PFAS plasma concentrations (described below), associations varied across PFASs and cognitive outcomes, and differed depending on timing of PFAS measurement (prenatal or mid-childhood) and outcome assessment (early childhood or mid-childhood).

3.2.1 Prenatal PFASs and early childhood WRAVMA and PPVT-III (Figure 2)

Figure 2.

Associations of maternal pregnancy plasma PFAS concentrations in quartiles with mean differences (+95% confidence intervals) in children’s early childhood (median age 3.2 years) Wide Range Assessment of Visual Motor Abilities (WRAVMA) Composite score (n=919) [a] and Peabody Picture Vocabulary Test (PPVT-III) (n=948) [b]. Standard mean (SD) for WRAVMA and PPVT-III=100 (15); higher scores indicate better performance. Models adjusted for: year of blood collection, gestational age at blood collection, estimated glomerular filtration rate, maternal race/ethnicity, age, education, IQ, pre-pregnancy body mass index, and smoking status, paternal education, household income, child’s sex and age at assessment, and proxy for breastfeeding of a prior child (incorporating parity).

Early childhood WRAVMA total scores were higher (representing better performance) among children born to mothers in the highest quartile of prenatal PFOA plasma concentrations (2.3 points, 95% confidence interval (CI): 0.1, 4.5) compared to those in the lowest quartile (Figure 2; Table S5). We also observed trends suggesting higher early childhood WRAVMA scores among children born to mothers with the highest quartile prenatal concentrations of PFOS, PFHxS, PFNA, and MeFOSAA. Early childhood visual motor and receptive vocabulary (PPVT-III) scores were higher among those born to mothers in the second (though not third or fourth) versus first quartile of MeFOSAA concentrations; other prenatal PFASs did not appear associated with PPVT-III scores.

3.2.2 Prenatal and mid-childhood PFASs and mid-childhood WRAVMA (Figure 3)

Figure 3.

Associations of prenatal and mid-childhood plasma PFAS concentrations (in quartiles) with mean differences (+95% confidence intervals) in mid-childhood (median age 7.7 years) Wide Range Assessment of Visual Motor Abilities (WRAVMA) Visual-Motor scores, adjusted for relevant covariates (n=853 for prenatal PFASs, 635 for childhood PFASs). Standard mean (SD) for WRAVMA=100 (15); higher scores indicate better performance. All models adjusted for year of blood collection, maternal race/ethnicity, age, education, and IQ, paternal education, household income, home environment, and child’s sex and age at assessment. Prenatal PFAS models additionally adjusted for: gestational age at blood collection, estimated glomerular filtration rate, maternal pre-pregnancy body mass index and smoking in pregnancy, and proxy for breastfeeding of a prior child (incorporating parity). Childhood PFAS models additionally adjusted for: breastfeeding duration (months up to 12) and maternal parity.

Mid-childhood visual-motor scores were somewhat lower among those with upper quartile prenatal exposure to PFOA and PFOS, and among children in the second and third, but not fourth, quartile of prenatal PFHxS concentration (vs. first quartile) (Figure 3; Table S6). Conversely, mid-childhood visual-motor scores were higher among those born to mothers in the second (though not third or fourth) quartile of EtFOSAA concentrations. Prenatal concentrations of other PFASs were not associated with mid-childhood WRAVMA scores. Mid-childhood visual-motor scores (WRAVMA Drawing subtest) were lower among children in the highest three quartiles of mid-childhood PFHxS concentration (Q2: −5.1, 95% CI: −8.9, −1.3; Q3: −5.0, 95% CI: −9.0., −0.9; Q4: −5.0, 95% CI: −9.1, −0.8), in the fourth quartile of mid-childhood PFOA concentration (−6.1, 95% CI: −10.5, −1.6) and in upper quartiles of mid-childhood PFOS concentration (Q3: −4.6, 95% CI: −8.7, −0.5; Q4: −2.0, 95% CI: −6.3, 2.2); mid-childhood PFNA and PFDeA were not associated with WRAVMA scores (Figure 3; Table S7).

3.2.3 Prenatal and mid-childhood PFASs and mid-childhood KBIT-2 (Figure 4)

Figure 4.

Associations of prenatal and mid-childhood plasma PFAS concentrations (in quartiles) with mean differences (+95% confidence intervals) in mid-childhood (median age 7.7 years) Kaufman Brief Intelligence Test (KBIT-2) Verbal IQ and Non-Verbal IQ scores, adjusted for relevant covariates (n=851 for prenatal PFASs/Verbal IQ, 862 for prenatal PFASs/Non-Verbal IQ, 631 for childhood PFASs/Verbal IQ, 640 for childhood PFASs/Non-Verbal IQ). Standard mean (SD) for KBIT-2=100 (15); higher scores indicate better performance. All models adjusted for year of blood collection, maternal race/ethnicity, age, education, and IQ, paternal education, household income, home environment, and child’s sex and age at assessment. Prenatal PFAS models additionally adjusted for: gestational age at blood collection, estimated glomerular filtration rate, maternal pre-pregnancy body mass index and smoking in pregnancy, and proxy for breastfeeding of a prior child (incorporating parity). Childhood PFAS models additionally adjusted for: breastfeeding duration (months up to 12) and maternal parity.

KBIT-2 verbal IQ scores were lower among children with higher prenatal concentrations of PFOA, although the dose-response pattern appeared non-linear, with weaker associations observed for the third and fourth quartiles (Figure 4). Similarly non-linear patterns were observed for prenatal PFOA and KBIT-2 non-verbal IQ, as well as prenatal PFOS and verbal and non-verbal IQ; children born to women with top quartile PFOS concentrations had higher non-verbal IQ (Figure 4; Table S6). We observed potential trends of lower mid-childhood KBIT-2 IQ scores with higher mid-childhood concentrations of PFOA (verbal IQ and non-verbal IQ) and PFHxS (non-verbal IQ); other mid-childhood PFASs were not associated with KBIT scores (Figure 4; Table S7).

3.2.4 Prenatal and mid-childhood PFASs and mid-childhood WRAML2 (Figure 5)

Figure 5.

Associations of prenatal and mid-childhood plasma PFAS concentrations (in quartiles) with mean differences (+95% confidence intervals) in mid-childhood (median age 7.7 years) Visual Memory Index of the Wide Range Assessment of Memory and Learning (WRAML2) Design Memory and Picture Memory scores associated with prenatal and mid-childhood plasma PFAS concentrations in quartiles, adjusted for relevant covariates (n=853 for prenatal PFASs/Design Memory, 855 for prenatal PFASs/Picture Memory, 636 for childhood PFASs/Design Memory, 635 for childhood PFASs/Picture Memory). Standard mean (SD) for WRAML2 scores=10 (3); higher scores indicate better performance. All models adjusted for year of blood collection, maternal race/ethnicity, age, education, and IQ, paternal education, household income, home environment, and child’s sex and age at assessment. Prenatal PFAS models additionally adjusted for: gestational age at blood collection, estimated glomerular filtration rate, maternal pre-pregnancy body mass index and smoking in pregnancy, and proxy for breastfeeding of a prior child (incorporating parity). Childhood PFAS models additionally adjusted for: breastfeeding duration (months up to 12) and maternal parity.

Mid-childhood WRAML2 design memory scores were higher among children born to women in the top quartiles of PFOA, PFOS, PFHxS, and PFNA and WRAML2 picture memory scores were higher among children born to women with higher PFNA concentrations (Figure 5; Table S6). Mid-childhood PFASs did not appear associated with mid-childhood WRAML2 design memory or picture memory scores (Figure 5; Table S7).

3.3 Sensitivity analyses

Results for mid-childhood PFASs and mid-childhood cognitive assessment scores did not change appreciably in models additionally adjusted for maternal plasma PFAS concentrations (Table S8). Results of complete case analyses were similar to those of primary models (data not shown).

Results of models including imputed exposures and outcomes differed from primary models for some analyses (Tables S9–S11); most notably, associations of childhood PFHxS and mid-childhood WRAVMA visual motor scores were attenuated in models including imputed exposures and outcomes (e.g. Q4 vs. Q1: −2.8, 95% CI: −6.2, 0.9 compared to −5.0, 95% CI: −9.1, −0.8 in primary models).

Sensitivity analyses adjusting for assessor and excluding assessments from assessors who conducted <20 assessments (n=54 for early childhood, 28 for mid-childhood), those excluding individual cognitive assessment scores in which assessors had lower confidence (2–10% of scores, depending on assessment), and those with no adjustment for prior breastfeeding (prenatal PFAS models) or breastfeeding duration (childhood PFAS models) also yielded similar results (data not shown). Log-likelihood ratio tests and sex-stratified models did not suggest consistent patterns of effect measure modification by sex (data not shown). Results of models adjusted for plasma albumin were similar to those of primary models (data not shown).

4. DISCUSSION

In a prospective pre-birth cohort of Boston-area mothers and children, we observed lower visual motor abilities in mid-childhood (median age 7.7) among children with higher plasma PFHxS measured cross-sectionally; mean WRAVMA visual-motor scores were approximately five points, or a third of a standard deviation, lower among children in each of the top three quartiles versus the first quartile. Results suggested that higher plasma PFOA and PFOS in mid-childhood and PFOA, PFOS and PFHxS in maternal plasma from pregnancy might also be related to lower visual motor abilities in mid-childhood, though dose-response relationships appeared non-linear. We also observed trends suggesting lower mid-childhood intelligence test scores among those with higher mid-childhood PFOA (verbal IQ and non-verbal IQ) and PFHxS (non-verbal IQ).

Conversely, and contrary to our hypotheses, children born to women with the highest concentrations of PFOA, PFOS, PFHxS, and PFNA performed better on selected mid-childhood cognitive assessments (design memory for PFOA, non-verbal IQ and design memory for PFOS, design memory for PFHxS, and design memory and picture memory for PFNA). Children born to mothers with the highest concentrations of PFOA also scored higher on early childhood (median age 3.2) assessments of visual motor abilities, and we observed higher visual motor and receptive vocabulary scores among those in the second quartile of MeFOSAA. Prenatal PFAS concentrations were otherwise not associated with assessments of cognition conducted in early childhood. Adjustment for prenatal plasma PFASs had little effect on associations of mid-childhood plasma PFASs and cognitive outcomes, suggesting that gestational exposure to PFASs did not explain the observed relationships of mid-childhood PFASs and cognitive performance.

PFAS concentrations of mothers in our study (sampled 1999–2002) were similar to those of women in the 1999–2000 cycle of the nationally-representative U.S. NHANES (e.g. geometric mean (GM) PFOA=5.7 ng/mL in Project Viva and 4.8 ng/mL in NHANES women),^45,48 and concentrations among children (sampled 2007–2010) were similar to those of adolescent participants in the 2007–2008 and 2009–2010 NHANES (e.g. GM PFOA=4.2 ng/mL in Project Viva and 3.9 ng/L in 2007–8 NHANES adolescents).^49,50 Concentrations of the studied PFASs mostly decreased in NHANES over the 1999–2012 period following phase-outs by major U.S. manufactures of PFOS and PFOA, along with precursor chemicals that break down to PFOS and PFOA.^50–53 Our analyses were adjusted for year of blood sampling to reduce potential confounding by temporal trends related to these changes. Prenatal and child PFAS concentrations in Project Viva were also broadly comparable to those measured in European birth cohorts from similar time periods (e.g. GM PFOA=3.2 ng/mL for mothers, 4.5 ng/mL in 7-year-old children in a Faroe Islands cohort enrolled 1996–2000;²⁴ median maternal PFOA=5.4 ng/mL in Danish National Birth Cohort enrolled 1998–2002¹⁹).

There are multiple mechanisms through which PFAS exposure may adversely affect neurodevelopment. In vitro models suggest that PFASs directly influence neuronal differentiation.⁵⁴ In mice, neonatal exposure to PFOA, PFOS, or PFHxS was associated with alterations in levels of proteins related to synaptogenesis in the hippocampus and cerebral cortex in adulthood;^12,55 exposed mice also had irregular nicotinic responses, suggesting alterations to the cholinergic system, and changes in adult behaviors potentially related to cognitive function.^13,56 PFASs are known to activate the peroxisome proliferator-activated receptor alpha (PPAR-α), a nuclear receptor involved in regulation of metabolism and cell growth^1,2,57 that may play a role in regulating the cholinergic activation of dopaminergic neurons.⁵⁸ Finally, epidemiologic evidence suggests that, in pregnant women, PFASs may affect thyroid hormone levels, which are important for fetal brain development.^11,59

Prior investigators have also noted mechanisms through which PFAS exposure could exert neuroprotective effects.^60,61 PFASs appear to be partial agonists of peroxisome proliferator-activated receptor gamma (PPAR-γ),⁶² and PPAR-γ-agonists have been observed to induce anti-inflammatory effects in the central nervous system that may be neuroprotective.⁶³

We observed evidence that PFAS exposure is associated with both better and worse cognitive performance in children, with findings varying across PFAS, cognitive endpoint, dose, and timing of exposure (prenatal versus postnatal). Prior cohort studies assessing PFAS exposure at multiple time points have also reported inconsistent associations with brain development across PFASs and exposure periods. In a Faroe Islands birth cohort (n=539), childhood (age 5) PFOA, PFNA, and PFDeA concentrations were associated with greater behavior problems at age 7, but no associations were observed with age 5 PFOS or PFHxS or prenatal PFAS measures (in that study, PFAS measures at age 7 were associated with greater behavior problems in girls, but with fewer behavioral problems in boys).²⁴ In a Cincinnati, Ohio-based birth cohort (n=167), prenatal, age 3, and age 8 childhood PFNA, PFOA and PFOS (but not PFHxS) were associated with better reading skills at ages 5 and 8.²⁷ Other studies in prospective cohorts have reported negative,^15–17 null,^18–23 and positive^60,64–66 associations of prenatal or perinatal PFAS exposure with childhood neurodevelopmental outcomes. Childhood and adolescent exposures to certain PFASs have been linked to ADHD diagnoses^6,25 and greater impulsivity²⁶ in cross-sectional analyses, but these studies do not allow for comparison of prenatal and postnatal windows of exposure. Our results require confirmation in additional longitudinal studies examining cognitive development in relation to PFAS exposure at multiple time points.

The varied and inconsistent relationships between prenatal PFAS exposure and cognitive outcomes in our study mirror the inconsistency in the existing literature, and could reflect that biological mechanisms of action vary across PFASs or differ depending on plasma PFAS concentration. For instance, a potential mechanism of PFAS neurotoxicty is competition with thyroxine (T4) in binding with human thyroid hormone transport protein transthyretin (TTR), resulting in reduced concentrations of free T4, which could alter fetal neurodevelopment. The studied PFASs have been shown to vary in TTR binding potency, with PFHxS more potent than PFOS and PFOA, and PFNA less potent, while EtFOSAA and MeFOSAA were conversely associated with a slight increase in T4-TTR binding at high concentrations.⁶⁷ Additional research into the potentially complex mechanisms through which PFASs may influence the developing brain is needed.

Limitations of our study should be noted. Cross-sectional associations of mid-childhood PFAS concentrations and cognitive outcomes could be impacted by reverse causation if children’s cognitive abilities were related to behaviors that might influence PFAS exposure. Only a subset of originally enrolled Project Viva participants with available exposure and outcome data were included in each of our analyses (30–46% of the original cohort, depending on analysis), and excluded participants differed somewhat from included participants. In order to minimize potential selection bias due to differential loss to follow-up, we adjusted our models for covariates observed to predict drop-out, but bias is still possible. The observed attenuation in some effect estimates, including associations of mid-childhood PFHxS and mid-childhood WRAVMA visual motor scores, in models including imputed exposures and outcomes may indicate that selection bias influenced these results, though this attenuation could also stem from misclassification of imputed exposure and outcome values.

We relied on a proxy measure of prior breastfeeding, meaning that residual confounding resulting from incomplete adjustment for prior breastfeeding could have biased observed associations with prenatal PFASs. Sensitivity analyses excluding this proxy variable yielded very similar results to those of primary models, however, indicating that confounding related to prior breastfeeding was likely minimal. There were moderate to high correlations between studied PFASs, limiting our capacity to fully differentiate the effects of individual chemicals; reported associations with individual PFASs could therefore reflect combined effects of concurrent exposure to multiple PFASs.

We based estimates of prenatal PFAS exposure on a single measurement of maternal plasma PFAS concentration in the first or second trimester of pregnancy (5.6–20.9 weeks of gestation), rather than a more direct measure of fetal exposure, such as cord blood. Maternal gestational PFAS concentrations, however, likely represent a good estimate of PFAS fetal exposure throughout gestation because half-lives of PFASs in humans are relatively long (e.g. 2.3 years for PFOA, 5.4 years for PFOS, 8.5 years for PFHxS)¹ and concentrations of PFASs in maternal peripheral blood correlate highly with levels in cord blood.^9–11,68 Similarly, concentrations of PFASs in mid-childhood plasma likely reflect PFAS exposure over multiple preceding years.

Cognitive outcomes were measured using well-validated assessments, and assessor training and double-scoring protocols were employed to ensure consistent scoring across assessors, but outcome misclassification related to challenges inherent in assessing cognitive function in young children is possible. Results of sensitivity analyses adjusting for assessor were similar to those of primary models, suggesting that inter-rater variability is unlikely to have introduced substantial bias.

The study also has a number of strengths. PFAS concentrations in the study cohort were similar to concentrations observed in NHANES during the same time periods, and therefore representative of exposures experienced by the general U.S. population. Our study was one of the first to examine multiple windows of PFAS exposure (prenatal and childhood) in relation to cognitive development. The study was conducted in a relatively large, prospective cohort with rich data on maternal and child health and behaviors, which enabled adjustment for important potential confounders including physiologic measures in pregnancy, maternal IQ, support for cognitive development in the home, breastfeeding, and sociodemographic factors.

5. CONCLUSIONS

Our study suggests that exposure to PFASs (including PFOA, PFOS, and PFHxS) during gestation and childhood may be associated with lower childhood visual motor abilities. Other results were inconsistent, with higher prenatal PFASs associated in some cases with better cognitive outcomes. These findings indicate that PFASs may influence children’s cognitive development, though magnitude and direction of effects may vary depending on PFAS, dose, cognitive endpoint, and timing of exposure (prenatal versus postnatal). As PFAS exposures are ubiquitous among pregnant women and children, even modest adverse impacts of PFASs on cognitive development could cause a substantial neurodevelopmental burden at the population level.⁶⁹ These findings highlight the need for further research on mechanisms through which PFASs may act on the developing brain.

Highlights.

• Per- and polyfluoroalkyl substances (PFASs) are suspected developmental toxicants.

• We examined prenatal and childhood PFAS exposure in relation to child cognition.

• Prenatal PFASs were associated with both better and worse cognitive scores.

• Childhood PFASs were associated cross-sectionally with lower visual motor abilities.

Abbreviations

ADHD

attention deficit hyperactivity disorder

CDC

Centers for Disease Control and Prevention

DAG

Directed Acyclic Graph

EtFOSAA

2-(N-ethyl-perfluorooctane sulfonamido) acetate

FOSA

perfluorooctane sulfonamide

GFR

glomerular filtration rate

GM

geometric mean

HOME-SF

Home Observation for Measurement of the Environment - Short Form

IQ

intelligence quotient

KBIT-2

Kaufman Brief Intelligence Test, Second Edition

LOD

limit of detection

MeFOSAA

2-(N-methyl-perfluorooctane sulfonamido) acetate

NHANES

National Health and Nutrition Examination Survey

PFAS

per- and polyfluoroalkyl substance

PFDeA

perfluorodecanoate

PFHxS

perfluorohexane sulfonate

PFNA

perfluorononanoate

PFOA

perfluorooctanoate

PFOS

perfluorooctane sulfonate

PPAR-α

peroxisome proliferator-activated receptor alpha

PPAR-γ

peroxisome proliferator-activated receptor gamma

PPVT-III

Peabody Picture Vocabulary Test, 3rd edition

SD

standard deviation

T4

thyroxine

TTR

transthyretin

VIF

variance inflation factor

WISC-III

Wechsler Intelligence Scale for Children-III

WRAML2

Wide Range Assessment of Memory and Learning, Second Edition

WRAVMA

Wide Range Assessment of Visual Motor Abilities