Childhood urinary organophosphate esters and cognitive abilities in a longitudinal cohort study

Abstract

The use of organophosphate esters (OPEs) as flame retardants, which has increased over the past two decades, raises concerns that OPEs may be harmful to humans, especially children. Animal studies and some human studies have reported that OPEs may adversely impact brain development, but few human studies evaluated OPE exposure during early childhood and neurodevelopmental outcomes. We aimed to fill this knowledge gap with the present study on urinary OPE metabolite concentrations at ages 1–5 years and cognitive abilities at 8 years. We used data of 223 children from the Health Outcomes and Measures of the Environment (HOME) Study, a prospective pregnancy and birth cohort in Cincinnati, Ohio. The point estimates for bis-2-chloroethyl-phosphate (BCEP) and bis(1,3-dichloro-2-propyl)-phosphate (BDCIPP) in association with IQ tended to be small and positive, while the point estimates for diphenyl-phosphate (DPHP) were small and negative, with 95% CIs including the null. However, we did find that socioeconomic status (SES) variables modified associations between OPEs and child IQ, with adverse OPE-IQ associations being stronger in socioeconomically disadvantaged children than in others. We identified an additional 1- to 2-point decrease in Full Scale IQ for every log-unit increase in BDCIPP, BCEP, and DPHP among those with lower maternal education, non-white race, lower income, or living in more deprived neighborhoods. We observed similar results for the Perceptual Reasoning, Verbal Comprehension, and Working Memory Index Scores. We suspect that there is residual confounding related to socioeconomic disadvantage, which was not captured with the available SES variables typically used in epidemiologic studies.

1. Introduction

Organophosphate esters (OPEs) are chemicals added to consumer products to function as flame retardants, plasticizers, and lubricants, among other functions (van der Veen and de Boer, 2012). Since the mid-2000s, the use of OPEs has dramatically increased in the United States and other countries largely because of the toxicity of polybrominated diphenyl ether (PBDE) flame retardants (Lam et al., 2017a; De Wit, 2002). OPEs are applied in flame retardant chemical mixtures (van der Veen and de Boer, 2012) and then separate from the consumer products and settle into house dust, where humans are exposed through accidental ingestion, inhalation, and dermal sorption (Kim et al., 2019).

Experiments in animals suggest that OPEs may affect brain development (Fu et al., 2013; McGee et al., 2012; Noyes et al., 2015), neurobehavior (Patisaul et al., 2013), and hormone regulation (Farhat et al., 2013; Springer et al., 2012), but few epidemiological studies of the health effects of OPE exposure have been completed. There has been an effort to examine gestational exposure to OPEs and neurobehavioral outcomes in children(Percy et al., 2021a; Castorina et al., 2017; Doherty et al., 2019), but childhood exposure to OPEs and its consequences are still understudied. Children have smaller body sizes than adults, increasing their total body burden for toxicants, and their hand-to-mouth behaviors and closer proximity to floors and other high-exposure surfaces may further increase the risk of exposure to chemicals present in house dust (Bearer, 1995). It is critical to understand how OPE exposure during childhood may contribute to future health effects.

The purpose of this study was to examine associations between urinary metabolites of three common OPEs measured up to four times during childhood with cognitive abilities evaluated at 8 years of age in a well-established pregnancy and birth cohort.

2. Methods

2.1. Study participants

We recruited 468 pregnant women in the Cincinnati, Ohio area to participate in the HOME Study, a prospective cohort study to examine gestational and childhood exposures to environmental toxicants and health effects in children from March 2003 to February 2006 (Braun et al., 2017, 2020). Women were eligible for the original cohort if they were 1) at least 18 years of age, 2) at 16 ± 3 weeks’ gestation, and 3) currently living in a home built prior to 1978. We excluded women who were diagnosed with bipolar disorder, schizophrenia, diabetes, or cancer that required radiation or chemotherapy; HIV positive; taking medication for seizures or thyroid disorders; not fluent in English; or who were planning to move outside of the Greater Cincinnati Area. To be eligible for this analysis, singleton child participants had at least one urinary OPE metabolite measurement at ages 1, 2, 3, or 5 years of age and cognitive abilities assessed at age 8 (n = 221) or 12 (n = 12) years. All women provided informed consent for themselves and their children, and the study protocol was approved by the Institutional Review Board (IRB) at Cincinnati Children’s Hospital Medical Center (CCHMC). The Centers for Disease Control and Prevention (CDC) deferred to the CCHMC IRB as the IRB of record.

2.2. Environmental chemical measurements

Children provided spot urine samples at approximately 1, 2, 3, and 5 years of age and blood samples at 5 years of age. To collect the urine, we used surgical diaper inserts for children who were not yet toilet-trained, training potty liners for children in the process of being toilet-trained, and polypropylene specimen cups for toilet-trained children. We refrigerated diaper inserts within 24 h of collection and then expressed urine from the insert with a syringe into a specimen collection cup in the laboratory. All urine samples were aliquoted and frozen at −20 °C. Sera were isolated from blood samples and frozen at −80 °C until they were shipped overnight to the CDC’s National Center for Environmental Health (Watkins et al., 2014). Pregnant women provided spot urine samples at approximately 16 weeks of pregnancy. Samples were collected in polypropylene specimen cups and then aliquoted and frozen at −20 °C until analysis at the CDC’s National Center for Environmental Health Laboratory.

At the CDC, we analyzed urine for three OPE metabolites: bis(1,3-dichloro-2-propyl) phosphate (BDCIPP), the metabolite of tris(1,3-dichloro-2-propyl) phosphate (TDCIPP); bis-2-chloroethyl phosphate (BCEP), the metabolite of tris(2-chloroethyl) phosphate (TCEP); and diphenyl phosphate (DPHP), the metabolite of triphenyl phosphate (TPHP) and 2-ethylhexyl diphenyl phosphate (EHDPP). The urine measurements for these OPE metabolites were used as an index of OPE exposure. The OPE metabolites were conjugated and then they underwent enzymatic hydrolysis in 200 μL of urine followed by automated off-line solid-phase extraction. Next, they were separated via reversed-phase high-performance liquid chromatography, and detected by isotope dilution-electrospray ionization tandem mass spectrometry (Jayatilaka et al., 2017, 2019). The limit of detection (LOD) was 0.1 μg/L for all three OPE metabolites and coefficients of variation were 2.1–9.8% (Jayatilaka et al., 2019). Trained staff at the Cincinnati Children’s Hospital Laboratory measured urinary specific gravity using the ATAGO PAL-10 S pocket refractometer (ATAGO CO., Tokyo, Japan). For more analysis details, including quality control procedures, refer else-where (Jayatilaka et al., 2019; Percy et al., 2020).

Child sera, collected at the 5-year study visit, were analyzed for 10 PBDE congeners: BDE-17, −28, −47, −66, −85, −99, −100, −153, −154, and −183 with gas chromatography/isotope dilution high-resolution mass spectrometry (Sjödin et al., 2004). We derived serum lipid concentrations using the Phillips formula(Phillips et al., 1989), which sums total triglycerides and cholesterol. Final PBDE serum concentrations were expressed in ng/g lipids.

2.3. Child and maternal cognitive abilities measurements

Three blinded and trained research assistants assessed children’s cognitive abilities at age 8 years using the Wechsler Intelligence Scale for Children-IV (WISC-IV) (Wechsler, 2003). Research assistants were evaluated without notice every six months to confirm the accuracy of their test administration and scoring in comparison to the gold standard examiner (Yolton). For children who were not assessed at age 8 years, we used the WISC-IV at age 12 years (n = 12) (Watkins and Smith, 2013). We will henceforth refer to all the child cognitive abilities assessments as age 8 years regardless of which timepoint the measure was collected. Trained research assistants assessed mothers’ cognitive abilities using the Wechsler Abbreviated Scale of Intelligence (WASI) (Wechsler, 1999). All assessments were conducted in a study clinic for a standardized environment free of distractions. The WISC and the WASI scores are standardized to population norms with a mean of 100 and a standard deviation of 15, where a higher score indicates better performance. Each test yields a Full-Scale IQ (FSIQ) score, indicating global performance, and the WISC provides Index Scores for the specific intelligence domains of Verbal Comprehension, Perceptual Reasoning, Working Memory, and Processing Speed.

2.4. Statistical analysis

We standardized urinary OPE metabolite concentrations by urine specific gravity to account for urinary dilution with the following formula(MacPherson et al., 2018):

$$OP{E}_{SG\_Adj}=OP{E}_i\frac{\left(S{G}_m-1\right)}{\left(S{G}_i-1\right)}$$

where OPE_i is the observed OPE metabolite concentration, SG_m is the median specific gravity for the cohort at the appropriate time point, and SG_i is the specific gravity of the urine sample.

After specific gravity correction, all metabolite concentrations were natural log-transformed to reduce the influence of outliers prior to regression modeling. We calculated intraclass correlation (ICC) across the four early childhood time points for urinary OPE metabolite concentrations using specific gravity-corrected and natural log-transformed values from linear mixed effect models. To assess reproducibility, we applied cut-offs described by Rosner: ≤0.4 is poor, 0.4–0.75 is fair to good, and ≥0.75 is excellent (Rosner, 2011).

Detection frequency was above 90% for the three metabolites; fewer than 0.56% of samples had nondetectable concentrations of DPHP (Table 1), suggesting no need for multiple imputation methods (Lubin et al., 2004). For concentrations below the LOD (2.5% of measurements), we imputed using the LOD/√2 (Hornung and Reed, 1990). We estimated Pearson correlations on the three urinary OPE metabolites and serum PBDE congener concentrations (BDEs-17, −28, −47, −66, −85, −99, −100, −153, −154), which were natural log-transformed and summed (∑PBDEs) prior to analysis.

Table 1.

Child urinary OPE metabolite concentrations (μg/L) corrected for urinary specific gravity.

OPE Metabolite	N	% >LOD	Quantile				GM	GSD
			25	50	75	95
BCEP
1 Year	176	95.45	0.42	0.99	2.15	10.98	1.02	3.77
2 Years	161	90.06	0.41	0.98	2.22	12.10	0.99	4.35
3 Years	177	94.92	0.32	0.93	2.48	11.32	1.00	4.60
5 Years	176	93.75	0.34	0.64	1.61	7.41	0.74	3.64
BDCIPP
1 Year	178	98.88	0.59	1.33	3.52	13.21	1.52	3.72
2 Years	165	98.79	0.89	1.84	4.76	18.48	1.98	3.73
3 Years	177	98.87	1.07	2.93	7.35	21.06	2.85	3.78
5 Years	173	99.42	1.56	2.88	6.19	26.64	3.19	3.21
DPHP
1 Year	177	100	1.20	2.03	3.82	9.38	2.15	2.37
2 Years	164	100	1.31	2.59	4.78	16.16	2.69	2.67
3 Years	177	99.44	1.25	2.42	4.75	13.98	2.47	2.79
5 Years	177	100	1.34	2.49	5.70	20.22	2.88	2.93

The LOD was 0.1 μg/L for all metabolites. Abbreviations: LOD, limit of detection; GM, geometric mean; GSD, geometric standard deviation; BCEP, bis-2-chloroethyl phosphate; BDCIPP, bis(1,3-dichloro-2-propyl) phosphate; DPHP, diphenyl phosphate.

We created a directed acyclic graph to determine the appropriate covariates for regression modeling (Supplemental Fig. 1). Covariates used in the final models were maternal race/ethnicity (white vs. non-white), maternal IQ, breastfed status (yes vs. no), Home Observation for Measurement of the Environment (HOME) score (which captures the quality and quantity of caregiving in the child’s home environment) (Caldwell and Bradley, 1984), and household income. We also used maternal education, maternal race/ethnicity, household income, and child sex in secondary analyses to determine whether any effect modification by individual-level socioeconomic status (SES) or sex was present. To test for neighborhood-level SES effect modification, we used a deprivation index consisting of variables from the 2010 5-year American Community Survey (Brokamp et al., 2019). The six census tract variables used to create the deprivation index are fraction of the population in poverty, median household income, fraction of the population with a high school education, fraction of the population without health insurance, fraction of household receiving public assistance, and fraction of vacant housing.

We tested the associations between OPEs and child cognitive abilities for linearity using generalized additive models with smoothing terms and did not find enough evidence to reject linearity. Then, we used generalized linear models with Generalized Estimating Equations (GEE) to account for the repeated measurements of OPEs during childhood. We also tested for chemical-by-timepoint interactions and did not find any significant interaction terms (defined as p_interaction<0.1). Final GEE models did not retain this interaction term; therefore, they estimate the overall association between childhood urinary OPE metabolite concentrations at 1–5 years of age and cognitive abilities at 8 years of age. Finally, we performed a sensitivity analysis in which the respective maternal urinary OPE metabolite from 16 weeks of pregnancy was added to our main models as an additional covariate to ensure that the estimates in our main models were not influenced by prenatal OPE exposures. All analyses were performed using R version 4.0.2 (R Core Team. R, 2019).

3. Results

Of the 223 children included in the present study, 60.9% were non-Hispanic white, 58.3% came from families with a mean household income of at least $40,000, and 45.5% were male (Table 2). Children who were excluded from the present study due to lack of available IQ data were more likely to be non-Hispanic white, have mothers with graduate-level education, and come from households with an annual income of >$80,000.

Table 2.

Select demographic characteristics of HOME Study children and their mothers for all children with urinary OPE metabolite concentrations, for children included in the present study with both OPE and IQ data available, and for children who were excluded from the present study due to missing IQ data.

	Full Cohort N = 330	OPEs and IQ N = 223	IQ Unavailable N = 107
	N (%) or Mean (SD)†	N (%) or Mean (SD)†	N (%) or Mean (SD)†
Maternal race
Non-Hispanic white	217 (65.8)	142 (60.9)	79 (73.8)
Non-Hispanic black and others	113 (34.2)	91 (39.1)	28 (26.2)
Maternal education
High school or less	75 (22.7)	61 (26.2)	18 (16.8)
Any college	183 (55.5)	131 (56.2)	56 (52.3)
Graduate or professional	72 (21.8)	40 (17.2)	33 (30.8)
Household income (USD)
<$40,000	121 (36.7)	97 (41.6)	30 (28.0)
$40,000 – $79,999	92 (27.9)	77 (33.0)	35 (32.7)
>$80,000	117 (35.5)	59 (25.3)	42 (39.3)
Child sex
Male	149 (45.2)	106 (45.5)	48 (44.9)
Female	181 (54.8)	127 (54.5)	59 (55.1)
Child IQ
Full-Scale IQ†		101.8 (16.3)
Verbal Comprehension index†		99.9 (15.8)
Perceptual Reasoning Index†		105.7 (16.4)
Working Memory Index†		101.2 (15.3)
Processing Speed Index†		96.4 (14.9)
Maternal IQ†		105.4 (15.2)
Deprivation Index†	0.36 (0.16)	0.36 (0.16)	0.31 (0.14)
Maternal age at delivery (years)†	29.7 (5.6)	29.1 (5.8)	30.6 (5.3)

Urinary BCEP was detected at the lowest concentrations in the cohort (geometric mean [GM] = 0.74–1.02 μg/L), DPHP was detected at the highest concentrations at ages 1 and 2 years (GM = 2.15 and 2.69 μg/L, respectively), and BDCIPP was detected at the highest concentrations at ages 3 and 5 years (GM = 2.85 and 3.19 μg/L, respectively; Table 1). Concentrations of urinary BCEP declined as children aged, while concentrations of BDCIPP and DPHP rose with age. ICCs of urinary OPE metabolites between the four time points was <0.30 for all three metabolites, indicating poor reproducibility in early childhood.

We also examined the correlations between urinary OPE metabolites and the older generation of flame retardants, PBDEs, at age 5 years due to the potential for confounding. Although the sample size was moderate for children with both OPEs and PBDEs data at age 5 years (n = 140), cross-sectional Pearson correlations between serum ∑PBDEs and the three urinary OPE metabolites were poor (Supplemental Fig. 2). BDCIPP was not significantly correlated with ∑PBDEs, and BCEP and DPHP had slight positive correlations with ∑PBDEs (r = 0.180 and 0.178, respectively). These results suggest that PBDEs are unlikely to be a major confounding variable in the associations between urinary OPE metabolites and child cognitive abilities.

We tested the associations of urinary OPE metabolites at ages 1–5 years with child cognitive abilities at 8 or 12 years, as measured by the WISC-IV, using covariate-adjusted GEE models. In our main models, we found that the associations of urinary OPE metabolites with child FSIQ and the four IQ Index Scores had 95% confidence intervals which all included the null (Table 3).

Table 3.

Generalized estimating equations models of specific gravity corrected and natural log-transformed child urinary OPE metabolites at ages 1–5 years and cognitive abilities at age 8 years. Models are adjusted for maternal race, maternal IQ, breastfeeding, HOME score, and household income.

	BCEP		BDCIPP		DPHP
	β	95% CI	β	95% CI	β	95% CI
FSIQ	0.48	−0.27, 1.24	0.42	−0.40, 1.23	−0.50	−1.14, 0.14
Perceptual Reasoning	0.37	−0.36, 1.11	0.63	−0.20, 1.45	−0.33	−1.07, 0.41
Verbal Comprehension	0.15	−0.66, 0.95	−0.18	−1.06, 0.70	−0.60	−1.42, 0.22
Working Memory	0.65	−0.19, 1.49	0.61	−0.28, 1.50	−0.34	−1.12, 0.45
Processing Speed	0.47	−0.37, 1.30	0.44	−0.57, 1.45	−0.22	−1.02, 0.59

Abbreviations: FSIQ, Full-scale intelligence quotient; CI, confidence interval; BCEP, bis-2-chloroethyl phosphate; BDCIPP, bis(1,3-dichloro-2-propyl) phosphate; DPHP, diphenyl phosphate.

In secondary analyses, we tested for any effect modification by socioeconomic status (SES) factors, including maternal education, race/ethnicity, household income, and neighborhood deprivation. We found a clear pattern of children in lower SES circumstances experiencing decreases in FSIQ for every natural log-unit increase in OPE metabolite concentrations across multiple measures of SES (Fig. 1). We found a similar patterns across the IQ Index Scores of Perceptual Reasoning, Verbal Comprehension, and Working Memory (Table 4 and Supplemental Tables 1–3). In contrast, we found no effect modification in Processing Speed Index Score by SES.

Fig. 1.

Effect modification of the associations between urinary OPE metabolites at ages 1–5 years and child cognitive abilities at age 8 years by socioeconomic status variables. Models are adjusted for maternal race, maternal IQ, breastfeeding, HOME score, and household income.

Abbreviations: SES, socioeconomic status; IQ, Intelligence Quotient; H.S., high school; IQR, interquartile range; BCEP, bis-2-chloroethyl phosphate; BDCIPP, bis(1,3-dichloro-2-propyl) phosphate; DPHP, diphenyl phosphate.

Table 4.

Effect modification of the associations between urinary OPE metabolites at ages 1–5 years and child cognitive abilities at age 8 years by maternal education. Models are adjusted for maternal race, maternal IQ, breastfeeding, HOME score, and household income.

	BCEP			BDCIPP			DPHP
	β	95% CI	p interaction	β	95% CI	p interaction	β	95% CI	p interaction
FSIQ			0.01			0.02			0.02
H.S. or less	−0.93	−1.77, −0.08		−0.79	−1.97, 0.40		−1.75	−2.86, −0.64
Any college	0.79	−0.17, 1.75		0.85	−0.17, 1.87		−0.10	−0.85, 0.64
Graduate school	1.71	−0.62, 4.04		1.01	−1.56, 3.58		1.59	−1.77, 4.96
Perceptual Reasoning			0.05			0.26			0.13
H.S. or less	−0.69	−1.54, 0.16		−0.10	−1.43, 1.23		−1.21	−2.27, −0.16
Any college	0.86	−0.18, 1.90		0.82	−0.25, 1.88		−0.06	−1.11, 0.99
Graduate school	0.54	−1.48, 2.55		1.71	−0.49, 3.91		1.69	−1.33, 4.72
Verbal Comprehension			0.001			0.003			0.01
H.S. or less	−1.43	−2.35, −0.52		−1.48	−2.69, −0.28		−1.89	−3.15, −0.64
Any college	0.47	−0.43, 1.36		0.53	−0.43, 1.48		−0.15	−0.99, 0.70
Graduate school	2.01	−0.69, 4.72		−0.69	−4.49, 3.12		1.71	−2.64, 6.06
Working Memory			0.04			0.08			0.04
H.S. or less	−0.64	−1.68, 0.40		−0.50	−1.83, 0.82		−1.68	−2.79, −0.57
Any college	0.81	−0.27, 1.90		0.91	−0.21, 2.02		0.14	−0.87, 1.15
Graduate school	1.96	−0.24, 4.15		1.50	−1.11, 4.11		1.19	−1.76, 4.14
Processing Speed			0.86			0.90			0.95
H.S. or less	0.37	−0.84, 1.57		0.03	−1.53, 1.60		−0.57	−2.17, 1.04
Any college	0.31	−0.82, 1.45		0.42	−0.98, 1.81		−0.16	−1.16, 0.85
Graduate school	0.59	−1.75, 2.93		1.16	−1.10, 3.43		0.00	−3.38, 3.38

Abbreviations: FSIQ, Full-Scale IQ; H.S., high school; CI, confidence interval; BCEP, bis-2-chloroethyl phosphate; BDCIPP, bis(1,3-dichloro-2-propyl) phosphate; DPHP, diphenyl phosphate.

The relationship between child urinary OPE metabolites and cognitive abilities was modified by maternal education (Table 4). In all models except for those for Processing Speed, children of mothers with a high school education or less had negative effect estimates for the association between urinary OPE metabolites and cognitive abilities, indicating poorer outcomes. These findings were especially apparent for DPHP and FSIQ; children of mothers with the lowest level of education had 1.75 lower IQ score for each log-unit increase in urinary DPHP (95% CI: −2.86, −0.64). Urinary BCEP was also negatively associated with FSIQ in children of mothers with less education (β: −0.93; 95% CI: −1.77, −0.08). The Verbal Comprehension sub-scale was also significantly associated with all three urinary OPE metabolites in children of mothers with the lowest level of education (BCEP = β: −1.43; 95% CI: −2.35, −0.52; BDCIPP = β: −1.48; 95% CI: −2.69, −0.28; DPHP = β: −1.89; 95% CI: −3.15, −0.64).

We also observed effect modification by race/ethnicity for the associations between urinary OPE metabolites and FSIQ, Verbal Comprehension, Perceptual Reasoning, and Working Memory (Supplemental Table 1). Urinary OPE metabolite concentration ranges for the various strata of maternal education and race are reported in Supplemental Table 2. Supplemental Table 3 shows that an increase of $100,000 of household income was associated with a further increased effect estimate for the association between DPHP and both FSIQ and Perceptual Reasoning (β: 1.84; 95% CI: 0.04, 3.63 and β: 1.83; 95% CI: 0.27, 3.40, respectively). In Supplemental Table 4, we observe that an IQR increase in neighborhood deprivation is associated with the greatest effect modification of the relationships between BCEP and BDCIPP and Perceptual Reasoning (β: −1.01; 95% CI: −1.76, −0.25 and β: −1.19; 95% CI: −2.09, −0.30, respectively).

We also tested for effect modification by child sex, but none of the associations were significantly modified by sex (Supplemental Table 5). To test whether the associations we observed in our main models were influenced by prenatal exposures to OPEs, we performed a sensitivity analysis adjusting for maternal urinary OPE metabolite concentrations at 16 weeks of pregnancy for each respective child OPE metabolite model. Supplemental Table 6 demonstrates that the models including prenatal OPEs as covariates were not substantially different from our main models.

4. Discussion

In this study of early-life urinary OPE metabolites and cognitive abilities at age 8 years, we found that OPEs were not significantly associated with IQ in our main models. However, we did consistently observe stronger adverse effects of OPEs on cognitive abilities for children from more disadvantaged SES backgrounds, using both individual-and neighborhood-level variables. Child sex did not modify the associations between urinary OPE metabolites and cognitive abilities.

We are not aware of any other studies of postnatal OPE exposure and cognitive abilities in children, but some studies of prenatal exposure and child cognitive abilities have reported mixed findings. Two studies reported evidence of decreased child cognitive abilities in association with higher concentrations of some urinary OPE metabolites, both from US-based cohorts (Castorina et al., 2017; Doherty et al., 2019). Additionally, we previously published our findings of null associations between OPEs measured at three timepoints during the prenatal period and IQ measured at age 8 years in the HOME Study, which are similar to our main results reported here (Percy et al., 2021a). We are aware of two other studies of postnatal exposure to OPEs and child neurobehavior (e. g. memory, conduct, mood, social skills, etc.) although neither measured internal biomarkers of OPE exposure. Sugeng et al. measured OPEs in household dust and hand wipe samples from children aged 8–16 months and found that increased levels of TCEP in floor dust were associated with worse externalizing symptoms at age 18 months (Sugeng et al., 2021). Another study examined cross-sectional associations between OPEs measured with silicone wristbands and social skills in children aged 3–5 years. They found that higher OPEs were associated with less responsible behavior and greater externalizing problems, as rated by their preschool teachers (Lipscomb et al., 2017). Neither study reported impact on cognitive skills or effect modification by SES variables.

One of the commercial uses of OPEs is to add flame retardancy to consumer goods to replace the industry demand for neurotoxic PBDEs. While the literature is not yet clear on whether OPEs represent a significant risk of harm to human health, there is a consensus that PBDEs are harmful (Lam et al., 2017b). Vuong et al. explored postnatal serum PBDE levels and child neurobehavior in the HOME Study and found that increased PBDEs were associated with decreases in FSIQ and increases in hyperactivity and aggressive behaviors (Vuong et al., 2017).

Our findings of effect modification across multiple SES variables suggests that the relationship between OPE exposure and cognitive abilities is modified by related factors with more direct biological consequences, such as the experience of stress due to hardship. For instance, while it is not biologically plausible that race itself modifies how OPE exposures may affect brain development, the stress of experiencing racism may impact this pathway. Likewise, other hardships relating to having a lower SES or living in a more deprived neighborhood could similarly activate stress pathways and modify the OPE-IQ relationship. The concept of stress modifying associations between chemicals and brain development has already been explored for other environmental toxicants. For example, stress can modify the associations between lead exposure and various human health effects, including cognition, via the hypothalamic-pituitary-adrenal axis (Cory-Slechta et al., 2008). Stress has also been shown to alter associations between air pollution and central nervous system changes through the same mechanisms (Thomson, 2019). Further, one study found that racism amplified the adverse effects of air pollution on adolescent conduct problems (Karamanos et al., 2021). We are not aware, however, of any experimental or human studies exploring how stress may modify the impacts of OPE exposure on brain development.

Additionally, there are other factors besides stress that could potentially modify how OPEs interact with the developing brain in certain population subsets. For example, children who belong to the lower-SES strata may experience greater exposures to other environmental toxicants that could work additively or synergistically with OPEs to result in the observed IQ deficits. Although we did not find any evidence of this phenomenon in the case of PBDEs, future studies should be designed to interrogate how chemical mixtures may differentially impact various groups.

The observed results concerning SES effect modification were relatively consistent for FSIQ, Verbal Comprehension, Perceptual Reasoning, and Working Memory. The associations between OPEs and the Processing Speed Index Score were not similar to those for the other cognition domains, suggesting that the mechanism by which OPEs may impact some areas of neurodevelopment is not related to the same neurobiological pathways that impact Processing Speed. Magistro et al. reported that Processing Speed in young adults is associated with global white matter volume (Magistro et al., 2015), whereas others have found that FSIQ and the other IQ Indices are related to total grey matter volume and grey matter volume in specific brain regions known to be involved in memory, response selection, and executive function (Frangou et al., 2004; Gläscher et al., 2009). Further, early-life disturbances in thyroid hormones, which OPEs are suspected of causing (Ren et al., 2015; Percy et al., 2021b; Tao et al., 2021), have been associated with changes in child grey matter volume, but not white matter volume (Korevaar et al., 2016). We speculate that the observed differences in effects between the various IQ sub-scales may relate to the mechanisms by which OPE affect grey and white matter in the brain.

This study had some limitations. We lacked complete data about PBDE exposure during childhood. While we did explore correlations between urinary OPE metabolites and serum PBDEs in a subset of the cohort at age 5 years, data in the entire cohort at multiple time points would have allowed us to better explore the possibility of confounding by other flame retardant chemical exposures. In addition, the lack of correlations between urinary OPE metabolites and serum PBDEs may relate to differences in the chemicals’ half-lives. PBDEs have an average half-life of 1–3 years(Trudel et al., 2011), while OPEs have half-lives ranging from 9 to 54 days (Wang et al., 2020). Thus, serum PBDEs concentrations represent past and current exposures, but urinary OPE metabolites represent only current exposures. Therefore, we cannot entirely rule out confounding by PBDEs in the present study for OPE exposures earlier than those we assessed or other unmeasured confounding. Additionally, we cannot specify which underlying factors related to socioeconomic status are truly responsible for the effect modification that we observed. Future studies could explore this by including data on stress, hardship, or built environment. Other cohorts with larger sample size of non-white and lower education participants may also be better posed to investigate the observed underlying relationships between OPEs, cognitive development, and SES.

However, there are also many strengths of this study. We have multiple measurements of urinary OPE metabolites in early childhood, which gives us a clearer picture of exposure during this critical window. The ICC values that we calculated indicate high exposure variation between measurement time points, which highlights the value of frequent OPE measurement during early childhood. We also performed extensive secondary analyses to explore potential effect modification by SES factors, which is in line with current efforts in the field of environmental epidemiology to study the joint and individual effects of both chemical and non-chemical stressors. While SES variables are often added to statistical models as covariates, this is intended to control for confounding, and it does not assess for differences in the main effect based on levels of the potential modifying variable. We urge other researchers to consider performing similar effect modification analyses when examining environmental chemical exposures and human health effects if sample size allows and effect modification is suspected.

5. Conclusions

OPEs are of growing concern due to their ubiquity in homes and their potential for neurotoxicity. We found that SES variables modified the associations of early childhood urinary OPE metabolites on cognitive abilities in children from more disadvantaged backgrounds. Additional studies would be useful to validate our findings and examine if SES modifies associations between other chemicals and health effects.

Data availability

The data that has been used is confidential.