Background

Neurobehavioral development is a lifelong, dynamic process which encompasses a host of psychosocial and biological processes that influence behavior, emotion, and learning [1, 2]. Environmental chemical exposures are increasingly recognized as major risk factors for adverse neurobehavioral outcomes, ranging in effects from subclinical deficits in neurobehavioral functioning to increased risks of neurobehavioral disorders [2,3,4]. The prenatal period is a particularly susceptible window for neurobehavioral development given the rapid cascade of tightly controlled and sequenced biological processes that occur in utero, resulting in heightened susceptibility to environmental exposures [2]. Even minor, incremental disruptions to prenatal biological processes from low-level chronic exposures to environmental chemicals have the potential to result in lifelong health effects [3, 5].

Flame retardants are anthropogenic chemical additives incorporated into materials to prevent or delay fires and to meet flammability regulations in the United States, particularly in California [6, 7]. For many decades, legacy flame retardants, such as polybrominated diphenyl ethers (PBDEs), were the most frequently used [8, 9]. However, due to their bioaccumulation in the environment, persistence, and neurotoxicity to children, PBDEs have been phased out of the US market and banned from production in the European Union [10]. As a result, organophosphate esters (OPEs) have dramatically increased in use as replacement flame retardants in recent years [11,12,13]. However, emerging literature suggests that OPEs may be a regrettable substitution for PBDEs and may also adversely impact neurobehavioral and neurodevelopmental outcomes [14].

OPEs are commonly used as plasticizers and lubricants, contributing to their environmental ubiquity [7]. OPEs are also applied as additives to various consumer, industrial, and electronic products, such as polyurethane foam, textiles, and building materials [7, 15]. Due to their physical incorporation within a product matrix and their semivolatile nature, OPEs easily volatize and leach into surrounding environments, commonly settling into dust particles in homes and environmental media such as soil, surface water, sediment, and agricultural products and facilitating human exposure to OPEs [

Methods

Study design

The MADRES study is an ongoing prospective pregnancy cohort of predominately low-income Hispanic/Latino mother-child pairs living in urban Los Angeles, CA. A detailed description of the MADRES study population and protocol have been previously described [41]. In brief, participants were recruited into the study prior to 30 weeks’ gestation at three partner community health clinics, one private obstetrics and gynecology practice in Los Angeles, and through self-referrals from community meetings and local advertisements. Eligible participants at time of recruitment were: (1) less than 30 weeks’ gestation, (2) over 18 years of age, and (3) fluent in English or Spanish. Exclusion criteria included: (1) multiple gestation, (2) having a physical, mental, or cognitive disability that prevented participation or ability to provide consent, (3) current incarceration, and (4) HIV positive status. Written informed consent was obtained at study entry for each participant and the study was approved by the University of Southern California’s Institutional Review Board.

Nine urinary OPE metabolite concentrations were measured in 426 participants’ urine samples provided during the third trimester study visit (mean GA at sample collection ± SD = 31.4 ± 1.8 weeks) from 2017 to 2019. Child neurobehavior was assessed using the Child Behavioral Checklist 1.5–5 years (CBCL 1.5–5) composite scales, including the internalizing problems, externalizing problems, and total problems scales, administered at the 36-month timepoint. As shown in the consort diagram (Fig. 1), mother-child participants with complete information on the exposure, outcome, and key covariates of interest were included in the final analytic sample. A total of 204 mother-child dyads with available data on OPE metabolite concentrations, the CBCL administered at 36 months, and key covariates were included in this study.

Fig. 1
figure 1

Consort diagram of included mother-infant dyads

Full size image

OPE metabolites

Single spot urine samples were collected in 90 mL sterile specimen containers during a third trimester study visit. Urine specimens were aliquoted into 1.5 mL aliquot cryovials and specific gravity was measured in room temperature urine samples using a digital handheld refractometer (ATAGO PAL-10s pocket refractometer). Samples were stored at -80 ºC prior to shipment and sent to the Wadsworth Center’s Human Health Exposure Analysis Resource (HHEAR) lab hub for the analysis of the following nine OPE metabolites: diphenyl phosphate (DPHP), composite of di-n-butyl phosphate and di-isobutyl phosphate (DNBP + DIBP), bis(1,3,-dichloro-2-propyl) phosphate (BDCIPP), bis(2-chloroethyl) phosphate (BCEP), bis(butoxethyl) phosphate (BBOEP), bis(1-chloro-2-propyl) phosphate (BCIPP), bis(2-ethylhexyl) phosphate (BEHP), bis(2-methylphenyl) phosphate (BMPP), and dipropyl phosphate (DPRP). Additional information on each metabolite, the corresponding parent compound, and common uses are described in Table S1.

Urinary OPE metabolites were quantified following methods similar to those previously described, with some slight modifications [42]. In brief, urine samples (0.5 mL) were aliquoted into pre-baked glass tubes and spiked with 1 ng of deuterated internal standard (IS) mixtures of OPEs and 1 mL of 10 mM ammonium acetate buffer (pH 5). The samples were passed through solid phase extraction (SPE) cartridges (STRATA-X-AW: 60 mg, 3 cc, Phenomenex, Torrance, CA, USA) which were conditioned by successive passage with 2 mL of 5% (v/v) ammonia/methanol, 2 mL of methanol, and 2 mL of water. The samples were loaded with the valves partially opened. The SPE cartridges were then dried under vacuum for 3 min after washing with 1.0 mL of water. Analytes were eluted with 2 times 0.5 mL of 5% (v/v) ammonia/methanol, concentrated under a gentle stream of nitrogen at 37 °C to near dryness, and reconstituted with 0.1 mL of acetonitrile.

High-performance liquid chromatography (HPLC, ExionLC™ system; SCIEX, Redwood City, CA, USA), coupled with an AB SCIEX QTRAP 5500+ triple quadrupole mass spectrometer (TQMS, Applied Biosystems, Foster City, CA, USA), was used in the identification and quantification of target compounds. Nine OPE diester metabolites and corresponding 9 internal standards were separated by a Kinetex HILIC column (100 mm × 2.1 mm, 2.6 μm particle size; Phenomenex) serially connected to a Betasil C18 guard column (20 mm × 2.1 mm, 5 μm particle size; Thermo Scientific). The analytes were quantified by isotopic dilution method and an 11-point calibration curve (at concentrations ranging from 0.02 to 50 ng/mL) with the regression coefficient ≥ 0.998. Matrix spikes (synthetic and urine pool spiked with 1 ng of native standards and 1 ng of internal standards) were analyzed with real samples as quality control (QC) samples. For each batch of samples, replicates of reagent blanks, matrix blanks, and matrix spiked samples were processed. Replicates of HHEAR Urine Quality Control (QC) Pools Standard Reference Materials (SRM3672 and SRM3673, NIST, Gaithersburg, MD, USA) were analyzed with every batch of samples. Trace levels of all OPE diester metabolites were found in procedural blanks. OPE diester metabolite concentrations measured in blanks were subtracted from sample values. Matrix spiked samples had average recoveries of 70.4–133% (CV: ± 9–19%). Repeated analysis of HHEAR Urine QC Pools A and B among batches showed coefficients of variation of ± 12–31% and ± 12–30% respectively. SRM3672 and SRM3673 had coefficients of variation of ± 12–40% and ± 12–27% respectively. Target analytes limit of detection (LOD) ranged from 0.012 to 0.044 ng/mL. Due to poor chromatographic separation and co-elution of peaks accompanying a similar mass transition for DNBP and DIBP, these two isomers were reported as sum concentration of di-n-butyl phosphate and di-isobutyl phosphate (DNBP + DIBP).

OPE metabolites with concentrations below the LOD were imputed using the LOD/\(\surd 2\) [43]. Metabolites were then specific gravity (SG) adjusted using the following formula: Pc = P[(SGm-1)/(SG-1)], where Pc is the specific gravity corrected toxicant concentration (ng/mL), P is the observed toxicant concentration (ng/mL), SGm is the median SG value among the study population (median = 1.016), and SG = the SG value of the sample.

Health outcome assessment

The Child Behavior Checklist for ages 1½ through 5 years old (CBCL 1.5–5) is a 99-item questionnaire which has been validated and widely used to assess a broad range of emotional and behavioral problems in children [44]. The questionnaire was orally administered to maternal participants during the 36 month study visit who indicated the frequency of behaviors in their child within the prior 2 months on a 3-point Likert scale (not true = 0, sometimes true = 1, or very often true = 2), with each raw scale created by summing together relevant items and t-scores and corresponding borderline (t-scores: 60–63) and clinical symptom categories (t-scores: ≥ 64) calculated based on previously described criteria to quantify areas that may warrant evaluation by a professional [45]. Higher scores across all CBCL scales indicate increasing problems. The CBCL consists of seven scored syndrome scales (emotionally reactive (9 items), anxious/depressed (8 items), somatic complaints (11 items), withdrawn (8 items), sleep problems (7 items), attention problems (5 items), aggressive behavior (19 items), and other problems (33 items)). These syndrome scales can be summed to create two composite scales, internalizing problems (emotionally reactive, anxious/depressed, somatic complaints, and withdrawn) and externalizing problems (attention problems and aggressive behavior). The CBCL additionally includes a total problems score which is the summed total of all 99 questionnaire items, plus the highest score on any additional problems listed under an open-ended item, question 100 (score range = 0–200). For the purposes of this analysis, the raw internalizing problems, externalizing problems, and total problems scores were each analyzed to encapsulate the breadth of potential behavioral and emotional developmental problems experienced by participants and to facilitate comparisons to prior studies similarly examining impacts of OPEs on raw CBCL scores [40]. However, sensitivity analyses examining associations between OPEs and CBCL t-scores were also evaluated to assess the robustness of our results after standardizing raw scores to a normative US sample of children.

Covariates

Covariates assessed in this analysis were study design or sample collection variables or were identified based on previous literature which examined impacts of neurotoxic chemicals on early neurobehavioral development [3, 31, 39, 40]. Relationships between prenatal OPE metabolites and neurobehavioral development were visualized using a Directed Acyclic Graph (DAG) created using DAGitty (Fig. S1) [46]. All models were adjusted for variables identified in the DAG’s minimal sufficient adjustment set (maternal age, parity, pre-pregnancy BMI, race/ethnicity, income, and education) and study design or sample collection variables whose inclusion in models changed the effect estimate of our exposure of interest by 10% or more (recruitment site, specimen collection season, GA at sample collection, and child adjusted age at CBCL administration). The only exception to these criteria was adjustment for maternal-reported smoking during pregnancy. Prenatal smoking was identified in the minimal sufficient adjustment set, but, given the small frequency of maternal smoking (n = 5, 2.5%), we instead evaluated its impact in sensitivity analyses by removing participants who reported smoking during pregnancy. Additionally, child sex was adjusted for in all models since it is an important predictor of neurobehavioral outcomes and was also evaluated as an effect modifier in adjusted models.

Maternal age (years), household annual income during pregnancy (< $50,000, ≥ $50,000, do not know), education (≤ 12th grade, > 12th grade), race/ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, Multiracial non-Hispanic/Other non-Hispanic), maternal smoking during pregnancy (yes, no), and parity (first born, ≥ second born, missing) were collected via interviewer administered questionnaires in the participant’s preferred language (English or Spanish). Pre-pregnancy BMI was calculated using participant-reported pre-pregnancy weight and standing height measured by study staff at the first study visit using a commercial stadiometer (Perspectives Enterprises model P-AIM-101). Child sex assigned at birth was primarily abstracted from electronic medical records (n = 200, 98.0%), followed by maternal-reported child sex (n = 4, 2.0%) for cases in which abstracted sex could not be obtained. Child adjusted age at time of questionnaire administration was calculated in weeks using date of birth and date of questionnaire administration, corrected for premature birth (< 37 weeks).

Statistical analysis

We examined participant demographic characteristics using means and frequencies. OPE metabolite distributions were explored using histograms, geometric means, percentile distributions, and metabolite detect frequencies. Given the generally right skewed distribution of OPE metabolites, Kruskal Wallis tests were conducted to evaluate bivariate associations between categorical covariates and OPE concentrations and Spearman correlations were performed to evaluate associations between OPE metabolites.

The distribution of CBCL raw scores was right skewed with 7.4% and 2.5% of scores with a 0 on the internalizing and externalizing problems scales, respectively; therefore, CBCL scores were offset by 0.1 and natural log transformed prior to linear regression modeling. Locally Weighted Scatterplot Smoothing (LOWESS) plots between prenatal OPEs and CBCL composite scales were then evaluated, and due to non-linear associations that persisted after natural log transformation, OPE metabolites were categorized into exposure tertiles prior to linear regression modeling. For OPE biomarkers detected in > 80% of participants (DPHP, DNBP + DIBP, BDCIPP), OPE metabolites were categorized into tertiles of specific gravity adjusted exposure concentrations. For OPE metabolites detected in 50–80% of participants (BCEP, BBOEP, BCIPP), a three-level categorical variable was created, with the lowest category defined as concentrations < LOD, and the remaining detected values categorized as < median or ≥ median. For OPE biomarkers detected in < 50% of participants (BMPP, BEHP, DPRP), we modeled OPE biomarkers as binary variables that were detected (> LOD) or not detected (≤ LOD). Modeling assumptions for all linear regressions were evaluated and met. A statistical interaction between each OPE metabolite and child sex was also tested in linear regression models. Data were managed and linear regression models were analyzed using SAS v9.4 (SAS Institute, Inc., Cary, NC, USA).

Generalized Additive Models (GAMs) with a smoothing term for natural log transformed OPE metabolites were also performed to evaluate possible non-linear associations between OPE metabolites and neurobehavioral outcomes using the R package “mgcv”. Consistent with prior literature, only metabolites with a detect frequency > 60% (DPHP, DNBP + DIBP, BDCIPP, BCEP, BBOEP) were evaluated using GAMs [47,48,37, 38].

This study has several important strengths. Its prospective design provided us with the opportunity to collect urine samples during potentially sensitive periods (i.e., pregnancy) to measure OPEs prior to our outcome of interest. An additional strength of this study was the use of prenatal urinary metabolites as a measure of in utero exposure to OPEs, given that maternal urinary OPE metabolites are considered reliable indicators of potential fetal OPE exposures [15]. We also measured various previously understudied OPE metabolites, including DNBP + DIBP, BCIPP, BCEP, BBOEP, DRPR, BMPP, and BEHP, which advances opportunities for risk assessment and subsequent interventions. Furthermore, the population evaluated in this study was largely comprised of pregnant individuals of Latin American origin, who are historically underrepresented in U.S. biomedical and population health research and disproportionally burdened by environmental exposures [75], providing us with the opportunity to inform environmental justice solutions. An additional strength of this study is the use of a flexible environmental mixture modeling approach to assess the association between mixtures of OPE metabolites and neurobehavioral outcomes at 36 months.

However, our study also has some limitations. Since single spot urine samples collected during the third trimester were used to assess OPE exposures throughout pregnancy, there may have been some exposure misclassification. However, previous studies indicate moderate to good reproducibility for DPHP and BDCIPP levels throughout pregnancy [76, 77]. Additionally, although many key covariates identified in the literature were adjusted for, residual confounding could still be present, especially for postnatal OPE exposures, which could impact neurobehavioral outcomes. The relatively modest analytical sample analyzed in this study is another limitation since we may have been underpowered to detect associations between OPE mixtures and neurobehavioral outcomes. Furthermore, although our use of a flexible environmental mixture modeling approach was used to assess joint OPE exposures, we were unable to explore the impacts of joint OPE exposures among metabolites with low detect frequencies, such as BMPP, which we found to adversely impact neurobehavioral development.

Conclusion

In this prospective pregnancy cohort of predominately low-income and Hispanic pregnant individuals living in Los Angeles, we found adverse associations between prenatal exposures to multiple previously understudied OPEs and children’s neurobehavioral outcomes at 36 months. There was also suggestive evidence of interactions between metabolites, highlighting the importance of evaluating OPEs beyond the effects of a single metabolite, along with non-linear and sex-specific associations between OPEs and children’s neurobehavioral development. Given the scarcity of studies evaluating associations between prenatal OPE metabolites and early neurobehavioral outcomes, additional studies exploring these associations, for exposures during both the prenatal and postnatal periods, are warranted.