Category Archives: Health

Topics such as diseases, healthcare, Medicare, Medicaid, healthcare, and medical research.

Prenatal Exposure to Perfluoroalkyl Substances and Cardiometabolic Risk in Children from the Spanish INMA Birth Cohort Study

Author Affiliations open

1ISGlobal, Centre for Research in Environmental Epidemiology (CREAL), Barcelona, Spain

2Spanish Consortium for Research on Epidemiology and Public Health (CIBERESP), Madrid, Spain

3Universitat Pompeu Fabra, Barcelona, Spain

4Epidemiology and Environmental Health Joint Research Unit, FISABIO–Universitat Jaume I–Universitat de València, Valencia, Spain

5Health Research Institute of Palma (IdISPa), University Hospital Son Espases, Palma de Mallorca, Spain

6Spanish Consortium for Research on Obesity and Nutrition (CIBEROBN), Madrid, Spain

7Public Health Department of Gipuzkoa, San Sebastián, Spain

8Health Research Institute BIODONOSTIA, San Sebastián, Spain

9Institute for Occupational Medicine, RWTH Aachen University, Aachen, Germany

10Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA

11Miguel Hernandez University, San Juan de Alicante, Spain

PDF icon PDF Version (408 KB)

  • Background:
    Perfluoroalkyl substances (PFAS) may affect body mass index (BMI) and other components of cardiometabolic (CM) risk during childhood, but evidence is scarce and inconsistent.
    Objectives:
    We estimated associations between prenatal PFAS exposures and outcomes relevant to cardiometabolic risk, including a composite CM-risk score.
    Methods:
    We measured perfluorohexanesulfonic acid (PFHxS), perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorononanoic acid (PFNA) in maternal plasma (first trimester). We assessed weight gain from birth until 6 mo. At 4 and 7 y, we calculated the age- and sex-specific z-scores for BMI, waist circumference (WC), and blood pressure (BP) (n≈1,000). At age 4, we calculated the age-, sex-, and region-specific z-scores for cholesterol, triglycerides (TGs), high-density (HDL-C), and low-density lipoprotein cholesterol (LDL-C) (n=627). At age 4, we calculated a CM-risk score (n=386) as the sum of the individual age-, sex-, and region-specific z-scores for WC, BP, HDL-C, and TGs. We used the average between the negative of HDL-C z-score and TGs z-score to give similar weight to lipids and the other components in the score. A higher score indicates a higher cardiometabolic risk at age 4.
    Results:
    PFOS and PFOA were the most abundant PFAS (geometric mean: 5.80 and 2.32 ng/mL, respectively). In general, prenatal PFAS concentrations were not associated with individual outcomes or the combined CM-risk score. Exceptions were positive associations between prenatal PFHxS and TGs z-score [for a doubling of exposure, β=0.11; 95% confidence interval (CI): 0.01, 0.21], and between PFNA and the CM-risk score (β=0.60; 95% CI: 0.04, 1.16). There was not clear or consistent evidence of modification by sex.
    Conclusions:
    We observed little or no evidence of associations between low prenatal PFAS exposures and outcomes related to cardiometabolic risk in a cohort of Spanish children followed from birth until 7 y. https://doi.org/10.1289/EHP1330
  • Received: 07 November 2016
    Revised: 13 July 2017
    Accepted: 21 July 2017
    Published: 20 September 2017

    Address correspondence to C. B. Manzano-Salgado, ISGlobal–Centre for Research in Environmental Epidemiology (CREAL), Doctor Aiguader, 8808003 Barcelona, Catalonia, Spain. Phone: +34 932 147 314. Email: cyntia.manzano@isglobal.org

    Supplemental Material is available online (https://doi.org/10.1289/EHP1330).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (244 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (185 KB)

Introduction

Childhood obesity has been steadily increasing during the past four decades (de Onis et al. 2010). Overweight children are more likely to present obesity, hypertension, dyslipidemia, and cardiovascular disease in adulthood than normal-weight children (Deshmukh-Taskar et al. 2006; Janssen et al. 2005). Obesity is known to be caused by a mismatch between caloric intake and energy expenditure, but early-life exposure to obesogens, including perfluoroalkyl substances (PFAS), may play a role (Holtcamp 2012; La Merrill and Birnbaum 2012). PFAS are synthetic chemicals widely used in industrial and commercial applications, including nonstick cookware, consumer products, textiles, and food packaging (Buck et al. 2011; Casals-Casas and Desvergne 2011; Stahl et al. 2011). PFAS have been detected in cord blood samples, suggesting that PFAS exposure starts prenatally (Inoue et al. 2004; Manzano-Salgado et al. 2015).

Rodents prenatally exposed to perfluorooctanoic acid (PFOA) showed higher weight gain, body fat accumulation, and cardiovascular disease (Hines et al. 2009; Lv et al. 2013; Tan et al. 2013). Epidemiological studies of the potential effects of PFAS on cardiometabolic outcomes in children have focused primarily on the two most common PFAS, perfluorooctanesulfonic acid (PFOS) and PFOA (reviewed by Vrijheid et al. 2016). Prenatal exposure to PFOS and PFOA was associated with lower weight at 5–12 mo of life (Andersen et al. 2010) among >1,000 children from the Danish National Birth Cohort (DNBC) study, but not with body mass index (BMI) and waist circumference (WC) in a later follow-up of >800 children from the same cohort at age 7 (Andersen et al. 2013). Prenatal PFOS and/or PFOA exposure also was associated with higher weight at 20 mo in a study of 320 girls from the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort (Maisonet et al. 2012), higher risk of waist-to-height ratio (WHtR) >0.5 at 5–9 y in 1,022 children from Greenland and Ukraine (Høyer et al. 2015), higher adiposity at 8 y in 204 American children (Braun et al. 2016), and higher BMI among 345 Danish women, but not among 320 Danish men, at age 20 (Halldorsson et al. 2012). Prenatal PFAS exposure was associated with higher BMI, skinfold thickness, and total fat mass, measured using dual-energy X-ray absorptiometry (DXA) in 7-y-old girls (n=466), but not boys (n=522) from a U.S. birth cohort (Mora et al. 2016). However, estimated early-life PFOA exposure was not associated with self-reported BMI at 20–40 y among 8,764 adults who resided near a PFOA manufacturing facility in the United States (Barry et al. 2014).

In addition to overweight and adiposity, cardiometabolic risk factors include elevated blood pressure (BP), lipid abnormalities, and abnormal glucose homeostasis, all of which are considered components of metabolic syndrome (Kassi et al. 2011). A cross-sectional study of adolescent participants (12–19 y of age) in the U.S. National Health and Nutrition Examination Survey (NHANES) study reported inverse associations between perfluorononanoic acid (PFNA) and metabolic syndrome, but positive associations of PFOS with markers of abnormal glucose homeostasis (Lin et al. 2009), while a second NHANES study reported that PFOA and PFOS were inversely associated with hypertension in adolescents (Geiger et al. 2014b). Two additional cross-sectional studies of PFOA and PFOS include a study of Danish children at 8–10 y of age (n=342–499) that reported no associations with cardiometabolic risk factors in the population as a whole (Timmermann et al. 2014), and a cross-sectional study of >12,000 U.S. children and adolescents (1–18 y) with potential exposure via contaminated drinking water, which reported positive associations with serum lipid levels (Frisbee et al. 2010). Two longitudinal studies of PFAS and cardiometabolic risk factors, other than overweight and adiposity, include a study of girls in the ALSPAC cohort, which reported that increasing prenatal PFOA concentrations within the lowest tertile of the distribution (but not in the second or third tertiles) was positively associated with total cholesterol (TC) and low-density lipoprotein cholesterol (LDL-C) at 7 and 15 y (111 and 88 girls, respectively) (Maisonet et al. 2015). A U.S. study reported no association between prenatal PFAS and insulin resistance at approximately 8 y of age (n=441), though concurrent serum PFOA and PFOS were inversely associated with insulin resistance in a cross-sectional analysis (n=541) (Fleisch et al. 2016).

In the present study, we evaluated prenatal exposures to four PFAS in association with individual cardiometabolic risk factors (anthropometric measurements, blood pressure, and serum lipids) in early and midchildhood among participants in a Spanish birth cohort. In addition, we estimated longitudinal associations with a combined CM-risk score as an alternative predictor of overall cardiometabolic risk (Eisenmann 2008; Pandit et al. 2011).

Methods

Study Population

We used data from the INMA (INfancia y Medio Ambiente, Environment and Childhood) birth cohort study including three Spanish regions: Gipuzkoa, Sabadell, and Valencia. A total of 2,150 pregnant women were recruited in the first trimester of pregnancy [gestational age: median=12.71 wk; interquartile range (IQR)=5.85] during the years 2003–2008. The inclusion criteria were: 16 y old or older, intention of giving birth at the reference hospital, singleton pregnancy, no language barrier, and no assisted pregnancy (Guxens et al. 2012). Child anthropometric data at 6 mo was abstracted from medical records. Mother–child pairs were then assessed when children were 4 and 7 y old. Maternal PFAS concentrations were measured in serum samples from 1,243 women (58% of the initial cohort) who participated in follow-up when their children were 4 y of age (Manzano-Salgado et al. 2016). The present analysis was limited to mother–child pairs with prenatal PFAS concentrations and weight gain data for children at 6 mo (n=1,154), and with BMI at 4 y (n=1,230) and 7 y (n=1,086) (Figure 1). In addition, WC and BP measurements were available at 4 y of age for children from the Valencia and Sabadell subcohorts (n=839, 68%); blood lipid levels were available for a subset of children from all three of the subcohorts (n=627, 51%); and our derived CM-risk score, which was based on WC, BP, high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG) (as described in detail below), was available for 386 children (31%) (Figure 1). All participating women signed written informed consent. This study was approved by the regional ethical committees of each cohort (Guxens et al. 2012).

Flow chart
Figure 1. Flow chart of sample populations in our study. Note: BP, blood pressure; CM, cardiometabolic; Gip, Gipuzkoa; PFAS, perfluoroalkyl substances; Sab, Sabadell; Val, Valencia; WC, waist circumference.

Perfluoroalkyl Substances Determination

We collected maternal blood samples during the first trimester of pregnancy (median=12.71 wk; IQR=5.85). Plasma was aliquoted in 1.8 mL cryotube vials (Sigma-V7884, Nunc® CryoTubes®) and stored at −80°C until their analysis at the Institute for Occupational Medicine, RWTH Aachen University (Aachen, Germany), as previously described (Manzano-Salgado et al. 2015). Briefly, plasma concentrations of perfluorohexanesulfonic acid (PFHxS), PFOS, PFOA, and PFNA were determined by column-switching liquid chromatography (Agilent 1,100 Series high-performance liquid chromatography apparatus; Agilent Technologies) coupled with tandem mass spectrometry (Sciex API 3,000 liquid chromatography coupled with tandem mass spectrometry system in electrospray ionization-negative mode; Applied Biosystems), according to a modified protocol described by Kato et al. (2011). The limits of quantification (LOQ) and detection (determined as a signal-to-noise ratio of 6 in the vicinity of the analytes) were 0.20 ng/mL and 0.10 ng/mL, respectively, for PFHxS, PFOS, and PFOA, and 0.10 ng/mL and 0.05 ng/mL, respectively, for PFNA (Manzano-Salgado et al. 2015).

Anthropometric Measurements

Assessment of anthropometric outcomes in INMA has been previously described (Valvi et al. 2013, 2015). Briefly, we abstracted height and weight measurements at birth and 6 mo from medical registries, and calculated age- and sex-specific z-scores for weight gain from birth to 6 mo using World Health Organization (WHO) reference values (de Onis et al. 2009, 2007). Early rapid growth during the first months of life has been associated with later obesity (Monteiro and Victora 2005); therefore, we also evaluated early rapid growth, defined as a z-score >0.67 standard deviation (SD) for weight gain from birth until 6 mo.

At 4 and 7 y, we measured weight (nearest 0.10 g) and height (nearest 0.10 cm) using a standard protocol (with no shoes and in light clothing) (Valvi et al. 2013). We calculated BMI (weight in kg/height in cm2) and age- and sex-specific BMI z-scores using the WHO reference (de Onis et al. 2009, 2007), and defined overweight as BMI z-score ≥85th percentile (de Onis et al. 2009, 2007). WC was measured using an inelastic tape (SECA model 201; SECA) at the midpoint between the right lower rib and the iliac crest, and after a slight breath out. We derived age-, sex-, and region-specific WC z-scores as the standardized residuals from a regression model of WC as the dependent variable, and age, sex, and region as the predictors, following the method of Eisenmann (2008), with standardization by region used to account for differences among the three regional subcohorts (Table S1) (Manzano-Salgado et al. 2016). We used WC (cm) and height (cm) to calculate the WHtR, and defined high abdominal adiposity as WHtR> 0.5 based on previous studies (Graves et al. 2014; Martin-Calvo et al. 2016; Mokha et al. 2010).

Blood Pressure

We used a digital automatic monitor (OMRON CPII) to measure systolic and diastolic BP (SBP and DBP). At age 4 y, we did a single measurement after five min of rest only, while at age 7 y, we did two measurements (with an additional 5-min rest period in between) and averaged the paired values for SBP and DBP, respectively. We derived average BP at each age as the mean of the SBP and DBP values. Similar to WC, we used regression models to derive BP z-scores standardized by age, sex, height, and region (Sabadell and Valencia at 4 y; Gipuzkoa, Sabadell, and Valencia at 7 y).

Lipids

In INMA, lipids were measured in all the children that agreed to provide blood samples in the follow-up at 4 y (n=740). For the purpose of the present study, we selected 627 children because they also had matched maternal PFAS concentrations. Lipids were measured using nonfasting blood samples (samples were fasting for Valencia, but not for Sabadell or Gipuzkoa) collected by venipuncture. We measured total TC, HDL-C, and TG levels using standard analytical techniques (ABX-Pentra 400; Horiba Medical). LDL-C was calculated based on TC, HDL-C, and TG concentrations using the Friedewald formula (Fukuyama et al. 2008). As for WC, we derived age-, sex-, and region-specific z-scores for TC and each individual lipid.

Cardiometabolic Risk Score

We derived a continuous CM-risk score at 4 y of age that is similar to the pediatric metabolic syndrome (MetS) score derived for the IDEFICS study by Ahrens et al. (2014) for >16,000 children at 2–9 y of age from eight European countries. Specifically, our cardiometabolic (CM)-risk score was derived as the sum of the standardized z-scores for WC, BP, and the mean of the HDL-C and TG z-scores, with HDL-C multiplied by −1 because it is inversely associated with cardiometabolic risk:

Our CM-risk score differs from the Identification and prevention of dietary- and lifestyle-induced health effects in children and infants (IDEFICS) MetS score in that it does not include a measure of glucose homeostasis; blood lipid levels were not always measured in fasting blood samples; z-scores for individual components were standardized by INMA study region (and by height for BP), as well as by age and sex; and we used the average of SBP and DBP, rather than mean arterial pressure, to represent the BP component. A higher CM-risk score suggests higher cardiometabolic risk.

Covariates

At enrollment (during the first trimester of pregnancy), mothers provided blood samples and completed self-reported questionnaires on sociodemographic and dietary factors, including the maternal country of birth (Spain or other), region of residence (Gipuzkoa, Sabadell, and Valencia), parity (0, 1, and ≥2), age (in years), and weekly intakes of fish and seafood consumption during the previous 3 mo of pregnancy (based on a food frequency questionnaire) (Manzano-Salgado et al. 2016). In INMA, mothers self-reported the duration of previous breastfeeding as the total number of weeks for any previous pregnancy. Then we combined all the durations into a single variable and classified their previous breastfeeding as none, <4 mo, 4–6 mo, or >6 mo. However, in the present study, we used the continuous variable in the models, that is, the total number of weeks of any previous breastfeeding. Regarding prepregnancy BMI, in INMA, height was measured, and prepregnancy weight was self-reported. Self-reported prepregnancy weight and measured weight at 12 wk of pregnancy were highly correlated; r=0.96; p<0.0001, (Casas et al. 2013). We then used the reported prepregnancy weight and measured height to calculate BMI (kg/m2), and classified mothers as underweight, normal weight, overweight, and obese. Further, the association between PFAS and fetal growth may be confounded by maternal glomerular filtration rate (GFR) during pregnancy (Verner et al. 2015). Therefore, we also measured maternal plasma creatinine (n=800) and calculated GFR using the Cockcroft-Gault formula [GFR=(140-maternal age)×weight (kg)×1.04/serum creatinine (μmol/L)]. Because plasma albumin is the binding site of PFAS (D’eon et al. 2010), we measured maternal albumin levels using the same maternal plasma sample that was used for measuring PFAS concentrations (n=800). Finally, in the follow-ups at the ages of 14 mo, 4 y, and 7 y, mothers completed questionnaires with information on postnatal characteristics of the index child, such as the duration of breastfeeding (total number of weeks) and the level of physical activity (Guxens et al. 2012).

Statistical Analysis

Since maternal PFAS concentrations were skewed to the right, we transformed PFAS concentrations to a 2-base logarithm. For PFAS with values under the LOQ (<4% for all compounds), we used multiple imputations to assign a value between 0 and the LOQ (0.20 ng/mL, and 0.10 ng/mL for PFNA) using the Stata 14.0 command ice. A detailed description of this procedure is provided in Table S2. We used generalized additive models (GAMs) of each log2–transformed PFAS and each outcome (adjusted for maternal region of residence, country of birth, parity, prepregnancy BMI, previous breastfeeding, and by the age at follow-up and sex of the child) to assess potential departures from linearity (p-value< 0.05). Based on this criterion, we found evidence of nonlinearity for exposure–outcome relations at age 7 (data not shown), and therefore modeled associations with PFAS concentrations categorized into quartiles, as well as associations with log2 PFAS as continuous variables; otherwise, we modeled continuous log2 PFAS concentrations only. We estimated associations between individual PFAS and each outcome using multivariable linear regression models for continuous outcomes (CM-risk score and z-scores for weight gain, BMI, WC, BP, TC, HDL-C, LDL-C, and TG) and Poisson regression models for binary outcomes (rapid growth, overweight, and WHtR> 0.5). Linear regression coefficients represent the unit difference in each outcome (where 1 unit is equivalent to a 1-SD difference in z-scores, or a 1-unit difference in the CM-risk score) with a doubling of prenatal PFAS concentration. Covariates included in the model were selected based on our previous study of determinants of PFAS concentrations in our cohort (Manzano-Salgado et al. 2016) and whether these determinants were associated with at least one of the outcomes (p-value< 0.10). We further adjusted all of our models by the age and sex of the child. Final models were adjusted for maternal country of birth (Spain or other), parity (number of pregnancies), previous breastfeeding (number of weeks), age (years), prepregnancy BMI (kg/m2), and by the age and sex of the child. We used multiple imputations to impute missing covariate data (<5%) (Donders et al. 2006; Sterne et al. 2009), under the assumption that covariate data were missing at random. We used Stata 14.0 command ice to create 20 different datasets for each age point (i.e., 6 mo, 4 y, and 7 y) and region of residence (i.e. Gipuzkoa, Sabadell, and Valencia), and imputed missing values as maternal and child characteristics (age, country of birth, fish intake during pregnancy, education, etc.). A detailed description of the imputation procedure is provided in Table S2. Distributions of covariates in all the imputed datasets were similar to those observed (see Table S3 for 4-y-old children from Sabadell; otherwise, data not shown).

We performed various sensitivity analyses. Because PFAS effect may differ by sex (Andersen et al. 2010; Halldorsson et al. 2012), we evaluated if the sex of the child modified our results by including the interaction term in the models (i.e., considered significant if p-value for PFAS*sex-interaction< 0.05) and by stratifying our analysis. Moreover, children with low birth weight (LBW) (i.e., <2,500 g) or preterm (i.e., before 37 wk of gestation) might have a different growth pattern than the rest of children, so we repeated our analysis excluding children with LBW (n=56) and born preterm (n=48). Further, we excluded children born by cesarean section (n=210) because a higher risk of obesity later in life has been suggested (Blustein et al. 2013). We also introduced maternal GFR as a confounder to evaluate if it changed our associations. Because albumin levels decrease during pregnancy (Glynn et al. 2012; van den Akker et al. 2008), we introduced albumin levels during pregnancy as a potential confounder in our models. Further, we stratified our models by breastfeeding duration of the index child as a proxy of postnatal PFAS exposure [never, short-term (<4 mo), long-term (4–6 mo), and very long-term (>6 mo)]. Finally, as PFAS are correlated in our cohort (Manzano-Salgado et al. 2015, 2016), we assessed whether introducing all PFAS into a single model changed the estimates for associations with CM-risk scores only.

We used the STATA 14.1 statistical software (Stata Corporation) for our analysis. We considered a p-value< 0.05 to be statistically significant.

Results

Mothers included in our study were more likely to be older, nulliparous, and with university studies compared to mothers without PFAS measured (this comparison does not include children who were excluded from analyses of lipids, BP, or CM-risk score) (Table S4). Geometric mean concentrations of PFAS ranged between 0.61 ng/mL for PFHxS and 5.80 ng/mL for PFOS (Table 1). PFAS were moderately correlated, with PFOA and PFNA (Spearman rho: 0.68 p-value< 0.001) being the most correlated, and PFHxS and PFNA being the least correlated (Spearman rho: 0.43 p-value< 0.001). Twenty-four percent of our children were classified as rapid growers at 6 mo, and 27% and 34% were classified as overweight at 4 and 7 y, respectively (Table 2). At 4 and 7 y, 26% and 19% of children had high abdominal adiposity (i.e., WHtR> 0.5), respectively (Table 2). Overall, the associations between PFAS and rapid growth, overweight, and WHtR> 0.5 were close to the null and not significant at any age (see Table S5). In this study, we describe the results from the adjusted models because we consider that they are closer to the true association of PFAS on cardiometabolic risk during childhood (for unadjusted models, see Table S6). Because no major differences in the magnitude or direction of the effect estimates were observed between the analyses using imputed data vs. complete cases only (n≤1,100) (Table S7), we only present the results based on imputed datasets. Differences in the CM-risk scores were more pronounced, but we do not consider this will affect interpretation (Table S7).

Table 1. Perfluoroalkyl substance (PFAS) concentrations (ng/mL) in maternal plasma during pregnancy.
Compound n(%)>LOD p5 p25 GM p75 p95
PFHxS 1,185 (96.3) 0.28 0.43 0.61 0.84 1.39
PFOS 1,230 (100) 2.54 4.52 5.80 7.84 11.40
PFOA 1,230 (100) 0.97 1.63 2.32 3.31 5.24
PFNA 1,222 (99.3) 0.30 0.50 0.66 0.90 1.49
Note: Concentrations are from the study population at 4 y (n=1,230). GM, geometric mean; LOD, limit of detection; p5, 25, 75, 95, percentiles 5, 25, 75, and 95; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.
Table 2. Summary of child characteristics in our study.
Child characteristic 6 mo n=1,154 4 y n=1,230 7 y n=1,086
Age (months or years) (mean±SD) 6.1±0.1 4.4±0.2 7.4±0.5
Anthropometric measurements
 Weight (kg) (mean±SD) 7.7±0.9 18.1±2.6 27.3±5.7
 Height (cm) (mean±SD) 67.0±2.4 105.7±4.5 125.2±6.0
 Weight gain z-score (mean±SD) 0.09±1.0
 Rapid growth (yes) n (%)a 270 (24)
 BMI (kg/m2) (mean±SD) 17.1±1.4 16.2±1.6 17.2±2.6
 BMI z-score (mean±SD) −0.9±0.9 0.6±1.0 0.8±1.2
 Overweight (yes) n (%) 336 (27) 367 (34)
 Waist circumference (cm) (mean±SD) (n=839 at 4 y) 51.2±4.2b 58.2±6.6
 Waist circumference z-score (n=839 at 4 y) (mean±SD) −0.01±0.95b −0.01±1.0
 Waist-to-height ratio>0.5 (yes) n (%) (n=839 at 4 y) 219 (26)b 204 (19)b
BP, mmHg (mean±SD) (n=839 at 4 y)
 Systolic BP 102.5±15.6b 107.3±9.8
 Diastolic BP 66.0±15.4b 63.5±8.5
 BP z-score 0.01±1.0b 0.00±1.0
Lipids, mg/dL (mean±SD) (n=627)
 Total cholesterol 167.7±26.6
 HDL-C 51.5±12.1
 LDL-C 100.8±22.0
 Triglycerides 76.6±36.9
CM-risk score (mean±SD) (n=386)b,c −0.84±4.0
Note: —, no data; BMI, body mass index; BP, blood pressure; CM, cardiometabolic; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SD, standard deviation.

aRapid growth was defined as a weight gain z-score from birth until 6 mo >0.67 SD.

bOnly Sabadell and Valencia subcohorts have available data for this outcome.

cCM-risk score is the z-scores for WC, BP, and the mean of the HDL-C and TG z-scores, with HDL-C multiplied by −1.

We observed few associations between prenatal PFAS concentrations, anthropometric measurements, and BP at any age (Table 3). Maternal PFOA concentrations were positively associated with weight gain z-scores at 6 mo in boys [for a doubling of PFOA, β=0.13; 95% confidence interval (CI): 0.01, 0.26] but not girls (β=−0.03; 95% CI: −0.14, 0.08) (p-value for sex-interaction=0.28) (Table 3). PFHxS was negatively associated with weight gain from birth to 6 mo in the overall population (β=−0.06; 95% CI: −0.15, 0.03) and in boys and girls, though the association was not statistically significant (Table 3). There were no other significant associations for the other PFAS at this age (Table 3). At 4 and 7 y, we did not observe any statistically significant association between PFAS and anthropometric measurements and BP (Table 3). However, we observed a pattern of inverse associations between PFHxS, BMI, and WC z-scores in the overall population and in boys (Table 3), whereas for girls, positive associations were observed (p-value for sex-interaction at 4 y≥0.12 and at 7 y≥0.16). At these ages, PFOS, PFOA, and PFNA showed patterns of positives associations with BMI z-scores in the overall population and in boys (p-value for sex-interaction> 0.18). Further, at 7 y, we observed nonsignificant associations between PFOA and BP z-scores that were positive in boys but negative in girls (p-value for sex-interaction=0.11). Finally, at age 7, we repeated the analysis using quartiles of PFAS exposure (instead of the continued variable) because GAM models showed nonlinearity for exposure–outcome relations at this age. Using quartiles of PFAS exposure yielded patterns of positive associations, although not statically significant, between PFOS, WC, and BP z-scores, and PFNA, BMI, and WC z-scores at 7 y (Figure S1). On the contrary, a pattern of inverse associations was observed between PFOA and BP z-scores at 7 y (Figure S1).

Table 3. Adjusted associations between maternal perfluoroalkyl substance (PFAS) concentrations (log2-transformed, in ng/mL) and cardiometabolic components during childhood.
Cardiometabolic components n PFHxS PFOS PFOA PFNA
β (95% CI) p-Value sex interaction β (95% CI) p-Value sex interaction β (95% CI) p-Value sex interaction β (95% CI) p-Value sex interaction
From birth until 6 mo
 Weight gain z-score, overall 1,154 −0.06 (−0.15, 0.02) −0.02 (−0.11, 0.07) 0.04 (−0.04, 0.12) 0.01 (−0.07, 0.09)
  Girls 568 −0.09 (−0.20, 0.02) 0.93 −0.09 (−0.21, 0.04) 0.54 −0.03 (−0.14, 0.08) 0.28 0.00 (−0.11, 0.11) 0.86
  Boys 586 −0.03 (−0.15, 0.10) 0.05 (−0.08, 0.19) 0.13 (0.01, 0.26) 0.04 (−0.09, 0.17)
At 4 y
 BMI z-score, overall 1,230 −0.02 (−0.10, 0.07) 0.04 (−0.05, 0.13) 0.04 (−0.04, 0.13) 0.05 (−0.03, 0.13)
  Girls 600 −0.02 (−0.13, 0.08) 0.70 0.02 (−0.10, 0.14) 0.99 0.00 (−0.11, 0.10) 0.31 0.02 (−0.08, 0.12) 0.26
  Boys 630 −0.02 (−0.15, 0.11) 0.05 (−0.08, 0.18) 0.09 (−0.03, 0.22) 0.08 (−0.04, 0.19)
 WC z-scoreb, overall 839 −0.04 (−0.14, 0.05) −0.03 (−0.13, 0.07) 0.00 (−0.09, 0.10) 0.02 (−0.07, 0.10)
  Girls 412 0.03 (−0.10, 0.16) 0.12 −0.04 (−0.18, 0.10) 0.73 −0.03 (−0.16, 0.10) 0.78 0.02 (−0.10, 0.14) 0.97
  Boys 427 −0.11 (−0.25, 0.03) −0.02 (−0.17, 0.13) 0.04 (−0.10, 0.18) 0.02 (−0.11, 0.14)
 BP z-scoreb, overall 839 −0.01 (−0.10, 0.09) −0.05 (−0.15, 0.06) −0.06 (−0.16, 0.04) −0.01 (−0.10, 0.08)
  Girls 412 0.10 (−0.09, 0.18) 0.60 −0.06 (−0.22, 0.09) 0.74 −0.04 (−0.18, 0.10) 0.99 0.05 (−0.08, 0.18) 0.39
  Boys 427 −0.06 (−0.20, 0.08) −0.02 (−0.18, 0.14) −0.08 (−0.23, 0.07) −0.07 (−0.20, 0.06)
Lipids
 TC z-score, overall 627 0.02 (−0.09, 0.12) 0.02 (−0.10, 0.15) 0.02 (−0.10, 0.15) 0.00 (−0.11, 0.12)
  Girls 318 0.04 (−0.12, 0.20) 0.96 0.05 (−0.13, 0.23) 0.74 0.09 (−0.08, 0.26) 0.53 0.05 (−0.11, 0.21) 0.85
  Boys 309 −0.02 (−0.17, 0.13) 0.00 (−0.18, 0.17) −0.05 (−0.22, 0.13) −0.05 (−0.22, 0.12)
 HDL-C z-score, overall 627 −0.01 (−0.11, 0.10) −0.03 (−0.14, 0.09) −0.04 (−0.15, 0.08) −0.03 (−0.14, 0.08)
  Girls 318 0.04 (−0.10, 0.19) 0.97 −0.03 (−0.18, 0.13) 0.71 0.12 (−0.02, 0.27) 0.10 0.04 (−0.10, 0.18) 0.34
  Boys 309 −0.06 (−0.22, 0.10) −0.02 (−0.20, 0.16) −0.20 (−0.01, −0.03) −0.10 (−0.27, 0.07)
 LDL-C z-score, overall 627 −0.01 (−0.12, 0.09) 0.02 (−0.10, 015) 0.03 (−0.08, 0.15) 0.01 (−0.10, 0.12)
  Girls 318 0.00 (−0.15, 0.15) 0.94 0.07 (−0.11, 0.25) 0.51 0.04 (−0.12, 0.21) 1.00 0.02 (−0.13, 0.18) 0.71
  Boys 309 −0.04 (−0.18, 0.10) −0.02 (−0.20, 0.15) 0.02 (−0.15, 0.19) −0.01 (−0.17, 0.16)
 Triglycerides z-score, overall 627 0.11 (0.01, 0.21) 0.05 (−0.06, 0.17) 0.04 (−0.07, 0.15) 0.03 (−0.07, 0.14)
  Girls 318 0.07 (−0.08, 0.22) 0.85 0.01 (−0.17, 0.19) 0.80 −0.01 (−0.17, 0.16) 0.79 0.05 (−0.11, 0.20) 0.68
  Boys 309 0.16 (0.03, 0.30) 0.09 (−0.06, 0.24) 0.08 (−0.06, 0.22) 0.02 (−0.12, 0.16)
 CM-risk scoreb,c overall 386 −0.09 (−0.64, 0.45) 0.28 (−0.33, 0.89) 0.27 (−0.35, 0.89) 0.60 (0.04, 1.16)
  Girls 197 −0.14 (−0.93, 0.64) 0.77 0.10 (−0.73, 0.93) 0.73 −0.22 (−1.10, 0.66) 0.45 0.50 (−0.27, 1.27) 0.86
  Boys 189 −0.09 (−0.86, 0.68) 0.47 (−0.44, 1.37) 0.72 (−0.17, 1.62) 0.70 (−0.13, 1.54)
At 7 y
 BMI z-score, overall 1,086 −0.04 (−0.14, 0.06) 0.03 (−0.08, 0.14) 0.03 (−0.08, 0.13) 0.06 (−0.04, 0.16)
  Girls 535 0.02 (−0.10, 0.14) 0.44 0.05 (−0.09, 0.20) 0.60 −0.01 (−0.13, 0.12) 0.48 0.00 (−0.12, 0.13) 0.18
  Boys 551 −0.10 (−0.25, 0.05) 0.02 (−0.15, 0.19) 0.07 (−0.10, 0.23) 0.12 (−0.04, 0.28)
 WC z-score, overall 1,086 −0.04 (−0.12, 0.04) 0.00 (−0.09, 0.09) −0.02 (−0.11, 0.06) 0.02 (−0.07, 0.10)
  Girls 535 0.04 (−0.07, 0.15) 0.16 0.04 (−0.09, 0.17) 0.42 −0.05 (−0.16, 0.07) 0.49 −0.03 (−0.14, 0.08) 0.26
  Boys 551 −0.11 (−0.23, 0.01) −0.02 (−0.16, 0.11) 0.01 (−0.12, 0.14) 0.07 (−0.05, 0.19)
 BP z-score, overall 1,086 0.04 (−0.04, 0.13) 0.06 (−0.04, 0.15) −0.02 (−0.11, 0.07) 0.00 (−0.08, 0.09)
  Girls 535 0.07 (−0.06, 0.19) 0.99 0.06 (−0.09, 0.20) 0.92 −0.08 (−0.21, 0.04) 0.11 0.00 (−0.12, 0.13) 0.70
  Boys 551 0.00 (−0.11, 0.11) 0.04 (−0.08, 0.17) 0.04 (−0.08, 0.16) −0.01 (−0.12, 0.11)
Note: Coefficients represent the average difference in the outcome with a doubling of the exposure. BMI, body mass index; BP, blood pressure; CI, confidence interval; CM, cardiometabolic; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid; TC, total cholesterol; WC, waist circumference.

aModel adjusted by: maternal characteristics (i.e., region of residence, country of birth, previous breastfeeding, age, and prepregnancy BMI), and the age and sex of the child. Note that sex of the child was not included in the models stratified by sex.

bOnly Sabadell and Valencia subcohorts have available data for this outcome.

cCM-risk score is the z-scores for WC, BP, and the mean of the HDL-C and TG z-scores, with HDL-C multiplied by −1.

As for the lipids measured at 4 y, we did not observe any statistically significant association between PFAS and z-scores of TC, HDL-C, or LDL-C at 4 y. In general, associations between prenatal PFAS and blood lipids at 4 y were close to the null (overall and when stratified by sex) (Table 3). One exception was the association between PFHxS and TGs at age 4 y, which was positive overall (β=0.11; 95% CI: 0.01, 0.21) and in boys (β=0.16; 95% CI: 0.03, 0.30) and girls (β=0.07; 95% CI: −0.08, 0.22) (p-value for sex-interaction=0.85). In addition, while the association between PFOA and HDL-C was close to the null in the combined population (β=−0.04; 95% CI: −0.15, 0.08), it was positive for girls (β=0.12; 95% CI: −0.02, 0.27) and negative for boys (β=−0.20; 95% CI: −0.37, −0.03), with p-value for sex-interaction=0.10 (Table 3).

Children with available CM-risk scores had lower BMI (mean BMI: 16.05 vs. 16.26; p-value=0.03) and TGs (mean TGs: 71.08 vs. 85.34 mg/dL; p-value<0.001), but higher HDL-C levels (mean HDL-C: 54.96 vs. 46.05 mg/dL; p-value< 0.001) than the full sample at 4 y (data not shown). Prenatal exposures to PFOS, PFOA, and PFNA were positively associated with CM-risk scores at 4 y in the overall population (Figure 2, Table 3), with a significant association for PFNA (β=0.60; 95% CI: 0.04, 1.16) that was similar in magnitude for boys and girls (p-value for sex-interaction=0.86). There were no significant differences in associations with CM-risk scores between boys and girls, though associations with PFOA were on opposite sides of the null (p-value for sex-interaction=0.45).

Plots with confidence intervals showing difference in CM risk score per log 2 unit increase in each PFAS (95 percent confidence intervals) (y-axis) across concentrations of PFAS, namely, PFHxS, PFOS, PFOA, and PFNA (x-axis), for girls (n equals 197) and boys (n equals 189) and overall (n equals 386).
Figure 2. Adjusted associations between maternal PFAS concentrations (log2–transformed, in ng/mL) and cardiometabolic risk score at 4 y. Abbreviations: CI, confidence interval; CM, cardiometabolic; PFHxS, perfluorohexanesulfonic acid; PFOS, perfluorooctanesulfonic acid; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid. Model adjusted by: maternal characteristics (i.e., region of residence, country of birth, previous breastfeeding, age, and prepregnancy BMI), and the age and sex of the child. Only Sabadell and Valencia subcohorts have available data for this outcome. CM-risk score is the z-scores for waist circumference (WC), blood pressure (BP), and the mean of the high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG) z-scores, with HDL-C multiplied by −1. Note: Coefficients represent the average difference in the outcome with a doubling of the exposure. Upper values represent the p-values of the sex interaction term.

Regarding the sensitivity analyses, the associations between PFOA and BMI z-scores at 4 and 7 y increased with breastfeeding duration of the index child, and these associations were statistically significant in very long-term breastfeeding duration, i.e., at 4 y: β=0.14 (0.03, 0.26; p-value for breastfeeding-interaction=0.09) and at 7 y: β=0.17 (0.02, 0.32; p-value for breastfeeding-interaction=0.01) (Figure S2). Associations with the CM-risk score differed when estimated using a model that included all four PFAS (compared with estimates from single PFAS models), with coefficients suggesting a stronger positive association with PFNA (β=0.85; 95% CI: 0.01, 1.69), a weaker positive association with PFOS (β=0.11; 95% CI: −0.73, 0.95), and a stronger negative association with PFHxS (β=−0.36; 95% CI: −1.05, 0.33), while the association with PFOA changed from positive to negative (β=−0.26; 95% CI: −1.20, 0.68). However, the mutually adjusted estimates were imprecise, and, with the exception of the association with PFNA, none were clearly different from the null. After including GFR in our models, all estimated associations remained close to null, but the association between PFHxS and BMI z-scores at 4 y changed from negative to positive, and the CIs were reduced for the associations between PFOS and PFOA and BMI z-scores at 4 y (Table S8). Similar results were observed after including maternal albumin levels in our models (Table S9). Finally, estimates remained similar after excluding cesarean section (Table S10), LBW, or preterm infants (data not shown because few observations were excluded).

Discussion

In this study with low prenatal PFAS concentrations, we observed little or no association with cardiometabolic risk components from birth until 7 y. None of the PFAS were significantly associated with anthropometric measurements or BP at 4 or 7 y. A doubling of prenatal PFHxS was associated with significantly higher TGs z-scores at 4 y of age, and prenatal PFNA was associated with a significantly higher CM-risk score. We did not observe any other statistically significant associations between PFAS concentrations and lipid z-scores or the CM-risk score in the overall population, or when stratified by sex.

In our study, low exposure levels of PFAS were not significantly associated with weight gain, BMI, overweight, or WC at any age. Other studies with similar exposure levels found that prenatal PFOS and PFOA exposure was associated with WHtR> 0.5 at 7 y (n=1,022) (Høyer et al. 2015), and PFAS were also associated with higher BMI, WC, skinfold thickness, and DXA total fat mass in girls aged 7 (n=1,006) (Mora et al. 2016). In the Health Outcomes and Measures of the Environment Study (n=204) with PFOA exposure levels above the U.S. average (median: 5.3 ng/mL vs. 2.3 ng/mL), prenatal PFOA exposure was associated with higher BMI, WC, and adiposity at 8 y (Braun et al. 2016). On the contrary, the DNBC, where PFOS exposure levels (median: 33.18 ng/mL) are considerably higher than ours, prenatal exposure to PFOS and PFOA was nonsignificantly associated with self-reported lower BMI (n=811), overweight, and WC (n=804) at age 7 (Andersen et al. 2013). Differences in the outcome measure might explain the conflicting results in prospective studies, though there are many other causal and noncausal factors that also might contribute to variation among studies.

We did not find clear or consistent evidence of differences in associations between boys and girls. Prenatal PFOA exposure was associated with higher weight gain from birth until 6 mo in boys, but not girls, though the difference was not significant (p-value for sex-interaction=0.28). The opposite pattern was seen in DNBC, where prenatal PFOA exposure (median: 5.25 ng/mL) was associated with lower weight and BMI at 5 and 12 mo in boys, but not in girls (Andersen et al. 2010). Effects might differ between the present study and the DNBC because exposures were higher in the DNBC, though other causal and noncausal explanations are also possible.

Early-life BP is predictive of cardiovascular health in adult life (Bao et al. 1995). In our study, we did not observe any statistically significant association between PFAS concentrations and BP at any age. In line with our results, cross-sectional data from the United States (n=1,655) showed no association between PFAS and hypertension in children 12–18 y old (Geiger et al. 2014b). In the present study, BP was measured twice at 7 y of age and averaged as recommended (Pickering et al. 2005), but was measured only once at 4 y of age, and only in children from two of the three study regions. We used the average of SBP and DBP as our BP outcome, consistent with the study of Ahrens et al. (2014), though other studies have evaluated SBP and DBP as separate outcomes, or have used mean arterial pressure (Eisenmann 2008; Geiger et al. 2014b; Sardinha et al. 2016; Shafiee et al. 2013).

Elevated TG levels during childhood have been associated with metabolic syndrome later in life (Berenson et al. 1998; Miller et al. 2011). TGs tend to accumulate in the liver, and PFAS are known to activate peroxisome proliferator–activated alpha receptor, which is a nuclear receptor that regulates lipid homeostasis in the liver (Lau et al. 2007; Rosen et al. 2008). In our study, higher prenatal PFHxS concentrations were associated with higher TG levels at 4 y, with higher point estimate for boys, but given the low precision of the estimates and p-value for sex-interaction=0.85, there is not clear evidence of a stronger association in boys than in girls. Few studies have assessed PFAS effect on lipids during childhood and adolescence, and even though they are of cross-sectional design, they suggest that PFAS, especially PFOS and PFOA, alter the lipid profile in children (Frisbee et al. 2010; Geiger et al. 2014a; Lin et al. 2009; Zeng et al. 2015). The study of Frisbee et al. (2010) from the C8 project (n>12,000 children aged 1–18 y old) only measured PFOS and PFOA, and observed positive associations of both with TC and LDL-C, with PFOS also positively associated with HDL-C. The study of Geiger et al. (2014a) from the NHANES (n=814 children aged 12–18 y old) only evaluated PFOA and PFOS, and reported that PFOA was positively associated with LDL-C>110 mg/dL and with HDL-C<40 mg/dL (these were used as parameters of dyslipidemia), and PFOS was positively associated with LDL-C>110 mg/dL only. The study from Lin et al. (2009), also from NHANES (n=474 children aged 12–18 y old), assessed PFHxS, PFOS, PFOA, and PFNA, and observed that PFNA negatively associated with HDL-C. The study of Zeng et al. (2015) in 225 Taiwanese children (aged 12–15 y), measured eight PFAS, and observed that PFOS and PFNA associated with TC, LDL-C, and TGs. Also, in a prospective study, prenatal PFOA in the lowest tertile was positively associated with LDL-C, but not with TC, HDL-C, or TG in girls at 7 and 15 y old (Maisonet et al. 2015).

Our CM-risk score included three of four individual components (e.g., anthropometric measurements, BP, and lipids) that are typically used to define metabolic syndrome (Eisenmann 2008). In our study, we observed higher CM-risk scores with higher PFOS, PFOA, and PFNA, but the association was only significant for PFNA. In contrast, Lin et al. (2009) reported that serum PFHxS, PFOA, PFOS, and PFNA concentrations in NHANES participants 12–19 y of age were inversely associated with the prevalence of metabolic syndrome (based on ≥three of the following conditions: high WC, high serum TG, low serum HDL, elevated SBP or DBP, medication for hypertension, or elevated fasting blood glucose or medication to reduce blood glucose), with a significant negative association for PFNA. However, direct comparisons between our study and Lin et al. (2009) are not possible given differences in the study design, population age, and outcome. In a Danish birth cohort study, prenatal exposure to PFOA was associated with overweight at 20 y in women, but not men (Halldorsson et al. 2012). Future research should include a prospective assessment of prenatal PFAS exposure and follow-up beyond early and midchildhood, with evaluation of sex-specific associations in larger populations.

Maternal excretion rates during pregnancy may influence the associations between prenatal exposure to PFAS and weight of the child (Verner et al. 2015). In our study, we adjusted our models by maternal GFR, showing that excretion rates are unlikely to have confounded the association between maternal PFAS concentration in plasma and childhood cardiometabolic outcomes. In the Project Viva cohort (United States), Fleisch et al. (2016) reported that after adjusment for GFR, their exposure–outcome estimates did not change by more than 10%. In the same cohort, Mora et al. (2016) reported that adjusting for GFR marginally strenghtened the associations between prenatal PFAS exposure and adiposity in midchildhood, suggesting some confounding by GFR. Both in the Project Viva and in our cohort, maternal PFAS concentrations were measured early in pregnancy when changes in GFR might not have a big impact on PFAS concentrations (Verner et al. 2015).

The main strengths of our study are its prospective design and large sample size for analyses of BMI and weight gain, and the ability to estimate associations between prenatal PFAS exposures and outcomes that may contribute to future cardiometabolic risk, including weight gain from birth to 6 mo; BMI, WC, and BP at 4 and 7 y of age; and blood lipids and a composite CM-risk score (based on WC, BP, and lipids) at age 4. Nevertheless, some methodological limitations should be considered. First, lipid levels were measured using fasting samples in children from the Valencia region, but nonfasting samples for children from Sabadell and Gipuzkoa, which may influence lipid levels, especially TGs. In addition, lipid concentrations were measured in only a subset of children (n=627) at 4 y of age. Second, our CM-risk score does not include a marker of glucose homeostasis, which is one of the components that is normally used to define metabolic syndrome (Ahrens et al. 2014; Eisenmann 2008). Therefore, our CM-risk score might not fully characterize the potential impact of PFAS on the prevalence of metabolic syndrome at age 4 or the future risk of cardiometabolic disease. Future follow-ups with available information on glucose homeostasis or insulin resistance at later ages are recommended. Third, we could only calculate the CM-risk score in 386 children that were generally healthier than the rest, thus limiting the extrapolation of our result to the full sample at 4 y (n=1,230). Fourth, we observed a pattern of positive associations between PFOA and BMI z-scores at ages 4 and 7 y with longer duration of breastfeeding for the index child. This finding suggests that postnatal PFAS exposure may play a role on childhood cardiometabolic risk, as similarly seen in other studies (e.g., Domazet et al. 2016; Zeng et al. 2015); however, we lack a direct measurement of postnatal PFAS exposure in our cohort. Fifth, women included in this study were more likely to be older, nulliparous, and with higher education than those excluded from the analysis. Given that older and nulliparous women tend to have higher PFAS levels (Manzano-Salgado et al. 2016), we probably included women and children with exposures that were higher than exposures in the cohort as a whole. Sixth, we cannot rule out the possibility of chance findings due to multiple comparisons in our study, or the possibility of uncontrolled confounding or bias due to measurement errors or missing data. Finally, in this study, small sample sizes for some of our analyses resulted in unstable or imprecise estimates of association, particularly for the CM-risk score, lipids, WC, and BP at 4 y, and for estimates stratified by sex.

Conclusions

In this study population exposed to low levels of PFAS, we found little evidence of an association between prenatal PFAS exposure and cardiometabolic risk during childhood. Although concerns have been raised about the potential for bias due to changes in maternal GFR during pregnancy, we measured PFAS early in pregnancy (before such changes are likely to be pronounced), and adjusting for GFR in a sensitivity analysis had little effect on our findings. Moreover, we evaluated multiple outcomes that may contribute to cardiometabolic risk, including anthropometric measurements, BP, lipid levels, as well as a combined risk score. Our findings are consistent with two previous prospective studies that reported little or no evidence of associations between prenatal PFAS exposure and obesity during early and midchildhood (Andersen et al. 2013; Mora et al. 2016). Future studies with follow-ups beyond midchildhood and with measures of glucose homeostasis are needed to further elucidate the effect of prenatal PFAS exposure on cardiometabolic risk.

Acknowledgments

We would particularly like to thank all the participants for their generous collaboration. A full roster of the INMA Project Investigators can be found at: http://www.proyectoinma.org/presentacion-inma/listado-investigadores/en_listado-investigadores.html.

This study was funded by grants from the European Union (FP7-ENV-2011 cod 282957 and HEALTH.2010.2.4.5-1), and from Spain: Instituto de Salud Carlos III and Ministry of Health (Red INMA G03/176; CB06/02/0041; PI041436, PI081151, PI06/0867, PS09/00090, PI13/02187; FIS-FEDER: PI03/1615, PI04/1509, PI04/1112, PI04/1931, PI05/1079, PI05/1052, PI06/1213, PI07/0314, PI09/02647, PI11/01007, PI11/02591, PI11/02038, PI12/01890, PI13/1944, PI13/2032, PI14/00891, and PI14/1687; predoctoral grant PFIS – FI14/00099; and Miguel Servet-FEDER: CP11/0178 and CPII16/00051), CIBERESP; the Conselleria de Sanitat, Generalitat Valenciana; Department of Health of the Basque Government (2005111093 and 2009111069); the Provincial Government of Gipuzkoa (DFG06/004 and DFG08/001); and the Generalitat de Catalunya-CIRIT (1999SGR 00241); and from the United Stated of America: National Institute of Environmental Health Sciences of the National Institutes of Health (NIH) (grant number ES021477). ISGlobal is a member of the Centres de Recerca de Catalunya (CERCA) Programme, Generalitat de Catalunya. This study has been reviewed and approved by the accredited committees of the following institutions: the Municipal Institute of Sanitary Assistance of Barcelona, La Fe University Hospital of Valencia, and Donostia Hospital de Zumarraga.

References

Ahrens W, Moreno LA, Mårild S, Molnár D, Siani A, De Henauw S, et al. 2014. Metabolic syndrome in young children: definitions and results of the IDEFICS study. Int J Obes 38(suppl 2):S4–S14, PMID: 25376220, 10.1038/ijo.2014.130.

Andersen CS, Fei C, Gamborg M, Nohr EA, Sørensen TIA, Olsen J. 2010. Prenatal exposures to perfluorinated chemicals and anthropometric measures in infancy. Am J Epidemiol 172(11):1230–1237, PMID: 20940176,10.1093/aje/kwq289.

Andersen CS, Fei C, Gamborg M, Nohr EA, Sørensen TIA, Olsen J. 2013. Prenatal exposures to perfluorinated chemicals and anthropometry at 7 years of age. Am J Epidemiol 178(6):921–927, PMID: 23825166, 10.1093/aje/kwt057.

Bao W, Threefoot SA, Srinivasan SR, Berenson GS. 1995. Essential hypertension predicted by tracking of elevated blood pressure from childhood to adulthood: the Bogalusa heart study. Am J Hypertens 8(7):657–665, PMID: 7546488, 10.1016/0895-7061(95)00116-7.

Barry V, Darrow LA, Klein M, Winquist A, Steenland K. 2014. Early life perfluorooctanoic acid (PFOA) exposure and overweight and obesity risk in adulthood in a community with elevated exposure. Environ Res 132:62–69, PMID: 24742729, 10.1016/j.envres.2014.03.025.

Berenson GS, Srinivasan SR, Bao W, Newman WP, Tracy RE, Wattigney WA. 1998. Association between multiple cardiovascular risk factors and atherosclerosis in children and young adults. The Bogalusa Heart Study. N Engl J Med 338(23):1650–1656, PMID: 9614255,10.1056/NEJM199806043382302.

Blustein J, Attina T, Liu M, Ryan AM, Cox LM, Blaser MJ, et al. 2013. Association of caesarean delivery with child adiposity from age 6 weeks to 15 years. Int J Obes (Lond) 37(7):900–906, PMID: 23670220, 10.1038/ijo.2013.49.

Braun JM, Chen A, Romano ME, Calafat AM, Webster GM, Yolton K, et al. 2016. Prenatal perfluoroalkyl substance exposure and child adiposity at 8 years of age: The HOME Study. Obesity (Silver Spring) 24(1):231–237, PMID: 26554535, 10.1002/oby.21258.

Buck RC, Franklin J, Berger U, Conder JM, Cousins IT, de Voogt P, et al. 2011. Perfluoroalkyl and polyfluoroalkyl substances in the environment: terminology, classification, and origins. Integr Environ Assess Manag 7(4):513–541, PMID: 21793199, 10.1002/ieam.258.

Casals-Casas C, Desvergne B. 2011. Endocrine disruptors: from endocrine to metabolic disruption. Annu Rev Physiol 73:135–162, PMID: 21054169, 10.1146/annurev-physiol-012110-142200.

Casas M, Chatzi L, Carsin AE, Amiano P, Guxens M, Kogevinas M, et al. 2013. Maternal pre-pregnancy overweight and obesity, and child neuropsychological development: two Southern European birth cohort studies. Int J Epidemiol 42(2):506–517, PMID: 23569191, 10.1093/ije/dyt002.

D’eon JC, Simpson AJ, Kumar R, Baer AJ, Mabury SA. 2010. Determining the molecular interactions of perfluorinated carboxylic acids with human sera and isolated human serum albumin using nuclear magnetic resonance spectroscopy. Environ Toxicol Chem 29(8):1678–1688, PMID: 20821619,10.1002/etc.204.

de Onis M, Blössner M, Borghi E. 2010. Global prevalence and trends of overweight and obesity among preschool children. Am J Clin Nutr 92(5):1257–1264, PMID: 20861173, 10.3945/ajcn.2010.29786.

de Onis M, Garza C, Onyango AW, Rolland-Cachera MF, le Comité de nutrition de la Société française de pédiatrie
. 2009. [WHO growth standards for infants and young children]. Arch Pediatr 16(1):47–53, PMID: 19036567, 10.1016/j.arcped.2008.10.010.

de Onis M, Onyango AW, Borghi E, Siyam A, Nishida C, Siekmann J. 2007. Development of a WHO growth reference for school-aged children and adolescents. Bull World Health Organ 85(9):660–667, PMID: 18026621, PMID: 10.2471/blt.07.043497.

Deshmukh-Taskar P, Nicklas TA, Morales M, Yang SJ, Zakeri I, Berenson GS. 2006. Tracking of overweight status from childhood to young adulthood: the Bogalusa Heart Study. Eur J Clin Nutr 60(1):48–57, PMID: 16132057, 10.1038/sj.ejcn.1602266.

Domazet SL, Grøntved A, Timmermann AG, Nielsen F, Jensen TK. 2016. Longitudinal associations of exposure to perfluoroalkylated substances in childhood and adolescence and indicators of adiposity and glucose metabolism 6 and 12 years later: The European Youth Heart Study. Diabetes Care 39(10):1745–1751, PMID: 27489335, 10.2337/dc16-0269.

Donders AR, van der Heijden GJ, Stijnen T, Moons KG. 2006. Review: A gentle introduction to imputation of missing values. J Clin Epidemiol 59(10):1087–1091, PMID: 16980149, 10.1016/j.jclinepi.2006.01.014.

Eisenmann JC. 2008. On the use of a continuous metabolic syndrome score in pediatric research. Cardiovasc Diabetol 7:17, PMID: 18534019, 10.1186/1475-2840-7-17.

Fleisch AF, Rifas-Shiman SL, Mora AM, Calafat AM, Ye X, Luttmann-Gibson H, et al. 2016. Early Life Exposure to Perfluoroalkyl Substances and Childhood Metabolic Function. Environ. Health Perspect 125(3):481–487, PMID: 27586368,10.1289/EHP303.

Frisbee SJ, Shankar A, Knox SS, Steenland K, Savitz DA, Fletcher T, et al. 2010. Perfluorooctanoic acid, perfluorooctanesulfonate, and serum lipids in children and adolescents: results from the C8 Health Project. Arch Pediatr Adolesc Med 164(9):860–869, PMID: 20819969, 10.1001/archpediatrics.2010.163.

Fukuyama N, Homma K, Wakana N, Kudo K, Suyama A, Ohazama H, et al. 2008. Validation of the Friedewald equation for evaluation of plasma LDL-cholesterol. J Clin Biochem Nutr 43(1):1–5, PMID: 18648653, 10.3164/jcbn.2008036.

Geiger SD, Xiao J, Ducatman A, Frisbee S, Innes K, Shankar A. 2014a. The association between PFOA, PFOS and serum lipid levels in adolescents. Chemosphere 98:78–83, PMID: 24238303, 10.1016/j.chemosphere.2013.10.005.

Geiger SD, Xiao J, Shankar A. 2014b. No association between perfluoroalkyl chemicals and hypertension in children. Integr Blood Press Control 7:1–7, PMID: 24520202, 10.2147/IBPC.S47660.

Glynn A, Berger U, Bignert A, Ullah S, Aune M, Lignell S, et al. 2012. Perfluorinated Alkyl acids in blood serum from primiparous women in Sweden: serial sampling during pregnancy and nursing, and temporal trends 1996-2010. Environ Sci Technol 46(16):9071–9079, PMID: 22770559, 10.1021/es301168c.

Graves L, Garnett SP, Cowell CT, Baur LA, Ness A, Sattar N, et al. 2014. Waist-to-height ratio and cardiometabolic risk factors in adolescence: findings from a prospective birth cohort. Pediatr Obes 9(5):327–338, PMID: 23894119, 10.1111/j.2047-6310.2013.00192.x.

Guxens M, Ballester F, Espada M, Fernández MF, Grimalt JO, Ibarluzea J, et al. 2012. Cohort Profile: the INMA–INfancia y Medio Ambiente–(Environment and Childhood) Project. Int J Epidemiol 41(4):930–940, PMID: 21471022, 10.1093/ije/dyr054.

Halldorsson TI, Rytter D, Haug LS, Bech BH, Danielsen I, Becher G, et al. 2012. Prenatal exposure to perfluorooctanoate and risk of overweight at 20 years of age: a prospective cohort study. Environ Health Perspect 120(5):668–673, PMID: 22306490, 10.1289/ehp.1104034.

Hines EP, White SS, Stanko JP, Gibbs-Flournoy EA, Lau C, Fenton SE. 2009. Phenotypic dichotomy following developmental exposure to perfluorooctanoic acid (PFOA) in female CD-1 mice: Low doses induce elevated serum leptin and insulin, and overweight in mid-life. Mol Cell Endocrinol 304(1-2):97–105, PMID: 19433254, 10.1016/j.mce.2009.02.021.

Holtcamp W. 2012. Obesogens: an environmental link to obesity. Environ Health Perspect 120(2):a62–a68, PMID: 22296745, 10.1289/ehp.120-a62.

Høyer BB, Ramlau-Hansen CH, Vrijheid M, Valvi D, Pedersen HS, Zviezdai V, et al. 2015. Anthropometry in 5- to 9-Year-Old Greenlandic and Ukrainian children in relation to prenatal exposure to perfluorinated alkyl substances. Environ Health Perspect 123(8):841–846, PMID: 25809098, 10.1289/ehp.1408881.

Inoue K, Okada F, Ito R, Kato S, Sasaki S, Nakajima S, et al. 2004. Perfluorooctane sulfonate (PFOS) and related perfluorinated compounds in human maternal and cord blood samples: assessment of PFOS exposure in a susceptible population during pregnancy. Environ Health Perspect 112(11):1204–1207, PMID: 15289168, 10.1289/ehp.6864.

Janssen I, Katzmarzyk PT, Srinivasan SR, Chen W, Malina RM, Bouchard C, et al. 2005. Utility of childhood BMI in the prediction of adulthood disease: comparison of national and international references. Obes Res 13(6):1106–1115, PMID: 15976154, 10.1038/oby.2005.129.

Kassi E, Pervanidou P, Kaltsas G, Chrousos G. 2011. Metabolic syndrome: definitions and controversies. BMC Med 9:48, PMID: 21542944, 10.1186/1741-7015-9-48.

Kato K, Basden BJ, Needham LL, Calafat AM. 2011. Improved selectivity for the analysis of maternal serum and cord serum for polyfluoroalkyl chemicals. J Chromatogr A 1218(15):2133–2137, PMID: 21084089, 10.1016/j.chroma.2010.10.051.

La Merrill M, Birnbaum LS. 2012. Childhood obesity and environmental chemicals. Mt Sinai J Med 78(1):22–48, PMID: 21259261, 10.1002/msj.20229.

Lau C, Anitole K, Hodes C, Lai D, Pfahles-Hutchens A, Seed J. 2007. Perfluoroalkyl acids: a review of monitoring and toxicological findings. Toxicol Sci 99(2):366–394, PMID: 17519394, 10.1093/toxsci/kfm128.

Lin CY, Chen PC, Lin YC, Lin LY. 2009. Association among serum perfluoroalkyl chemicals, glucose homeostasis, and metabolic syndrome in adolescents and adults. Diabetes Care 32(4):702–707, PMID: 19114613, 10.2337/dc08-1816.

Lv Z, Li G, Li Y, Ying C, Chen J, Chen T, et al. 2013. Glucose and lipid homeostasis in adult rat is impaired by early-life exposure to perfluorooctane sulfonate. Environ Toxicol 28(9):532–542, PMID: 23983163, 10.1002/tox.20747.

Maisonet M, Näyhä S, Lawlor DA, Marcus M. 2015. Prenatal exposures to perfluoroalkyl acids and serum lipids at ages 7 and 15 in females. Environ Int 82:49–60, PMID: 26010307, 10.1016/j.envint.2015.05.001.

Maisonet M, Terrell ML, McGeehin MA, Christensen KY, Holmes A, Calafat AM, et al. 2012. Maternal concentrations of polyfluoroalkyl compounds during pregnancy and fetal and postnatal growth in British girls. Environ Health Perspect 120(10):1432–1437, PMID: 22935244, 10.1289/ehp.1003096.

Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Basterrechea M, Grimalt JO, et al. 2015. Transfer of perfluoroalkyl substances from mother to fetus in a Spanish birth cohort. Environ Res 142:471–478, PMID: 26257032,10.1016/j.envres.2015.07.020.

Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Martinez D, Ibarluzea J, et al. 2016. Variability of perfluoroalkyl substance concentrations in pregnant women by socio-demographic and dietary factors in a Spanish birth cohort. Environ Int 92-93:357–365, PMID: 27132161,10.1016/j.envres.2016.04.004.

Martin-Calvo N, Moreno-Galarraga L, Martinez-Gonzalez M. 2016. Association between body mass index, waist-to-height ratio and adiposity in children: a systematic review and meta-analysis. Nutrients 8(8):512, PMID: 27556485,10.3390/nu8080512.

Miller M, Stone NJ, Ballantyne C, Bittner V, Criqui MH, Ginsberg HN, et al. 2011. Triglycerides and cardiovascular disease: a scientific statement from the American Heart Association. Circulation 123(20):2292–2333, PMID: 21502576, 10.1161/CIR.0b013e3182160726.

Mokha JS, Srinivasan SR, Dasmahapatra P, Fernandez C, Chen W, Xu J, et al. 2010. Utility of waist-to-height ratio in assessing the status of central obesity and related cardiometabolic risk profile among normal weight and overweight/obese children: the Bogalusa Heart Study. BMC Pediatr 10:73, PMID: 20937123, 10.1186/1471-2431-10-73.

Monteiro PO, Victora CG. 2005. Rapid growth in infancy and childhood and obesity in later life – a systematic review. Obes Rev 6(2):143–154, PMID: 15836465, 10.1111/j.1467-789X.2005.00183.x.

Mora AM, Oken E, Rifas-Shiman SL, Webster TF, Gillman MW, Calafat AM, et al. 2016. Prenatal Exposure to Perfluoroalkyl Substances and Adiposity in Early and Mid-Childhood. Environ Health Perspect 125(3):467–473, PMID: 27352404,10.1289/EHP246.

Pandit D, Chiplonkar S, Khadilkar A, Kinare A, Khadilkar V. 2011. Efficacy of a continuous metabolic syndrome score in Indian children for detecting subclinical atherosclerotic risk. Int J Obes Relat Metab Disord 35(10):1318–1324, PMID: 21772245,10.1038/ijo.2011.138.

Pickering TG, Hall JE, Appel LJ, Falkner BE, Graves J, Hill MN, et al. 2005. Recommendations for blood pressure measurement in humans and experimental animals: part 1: blood pressure measurement in humans: a statement for professionals from the Subcommittee of Professional and Public Education of the American Heart Association Council on High Blood Pressure Research. Circulation 111(5):697–716, PMID: 15699287,10.1161/01.CIR.0000154900.76284.F6.

Rosen MB, Abbott BD, Wolf DC, Corton JC, Wood CR, Schmid JE, et al. 2008. Gene profiling in the livers of wild-type and PPARα-null mice exposed to perfluorooctanoic acid. Toxicol Pathol 36(4):592–607, PMID: 18467677, 10.1177/0192623308318208.

Sardinha LB, Santos DA, Silva AM, Grøntved A, Andersen LB, Ekelund U. 2016. A comparison between BMI, waist circumference, and waist-to-height ratio for identifying cardio-metabolic risk in children and adolescents. PLoS One 11(2):e0149351, 10.1371/journal.pone.0149351.

Shafiee G, Kelishadi R, Heshmat R, Qorbani M, Motlagh ME, Aminaee T, et al. 2013. First report on the validity of a continuous Metabolic syndrome score as an indicator for Metabolic syndrome in a national sample of paediatric population – the CASPIAN-III study. Endokrynol Pol 64(4):278–284, PMID: 24002955,10.5603/EP.2013.0006.

Stahl T, Mattern D, Brunn H. 2011. Toxicology of perfluorinated compounds. Environ Sci Eur 23(1):38, 10.1186/2190-4715-23-38.

Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, et al. 2009. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 338:b2393, PMID: 19564179, 10.1136/bmj.b2393.

Tan X, Xie G, Sun X, Li Q, Zhong W, Qiao P, et al. 2013. High fat diet feeding exaggerates perfluorooctanoic acid-induced liver injury in mice via modulating multiple metabolic pathways. PLoS One 8(4):e61409, PMID: 23626681, 10.1371/journal.pone.0061409.

Timmermann CA, Rossing LI, Grøntved A, Ried-Larsen M, Dalgård C, Andersen LB, et al. 2014. Adiposity and glycemic control in children exposed to perfluorinated compounds. J Clin Endocrinol Metab 99(4):E608–E614, PMID: 24606078, 10.1210/jc.2013-3460.

Valvi D, Casas M, Mendez MA, Ballesteros-Gómez A, Luque N, Rubio S, et al. 2013. Prenatal bisphenol a urine concentrations and early rapid growth and overweight risk in the offspring. Epidemiology 24(6):791–799, PMID: 24036610, 10.1097/EDE.0b013e3182a67822.

Valvi D, Casas M, Romaguera D, Monfort N, Ventura R, Martinez D, et al. 2015. Prenatal phthalate exposure and childhood growth and blood pressure: evidence from the Spanish INMA-Sabadell Birth Cohort Study. Environ Health Perspect 123(10):1022–1029, PMID: 25850106, 10.1289/ehp.1408887.

van den Akker CH, Schierbeek H, Rietveld T, Vermes A, Duvekot JJ, Steegers EA, et al. 2008. Human fetal albumin synthesis rates during different periods of gestation. Am J Clin Nutr 88(4):997–1003, PMID: 18842786.

Verner MA, Loccisano AE, Morken NH, Yoon M, Wu H, McDougall R, et al. 2015. Associations of Perfluoroalkyl Substances (PFAS) with Lower Birth Weight: An Evaluation of Potential Confounding by Glomerular Filtration Rate Using a Physiologically Based Pharmacokinetic Model (PBPK). Environ Health Perspect 123(12):1317–1324, PMID: 26008903,10.1289/ehp.1408837.

Vrijheid M, Casas M, Gascon M, Valvi D, Nieuwenhuijsen M. 2016. Environmental pollutants and child health-A review of recent concerns. Int J Hyg Environ Health 219(4-5):331–342, PMID: 27216159, 10.1016/j.ijheh.2016.05.001.

Zeng XW, Qian Z, Emo B, Vaughn M, Bao J, Qin XD, et al. 2015. Association of polyfluoroalkyl chemical exposure with serum lipids in children. Sci Total Environ 512–513:364–370, PMID: 25638651, 10.1016/j.scitotenv.2015.01.042.

Integrating Health into Local Climate Response: Lessons from the U.S. CDC Climate-Ready States and Cities Initiative

Author Affiliations open

1Johns Hopkins Bloomberg School of Public Health, Department of Health Policy and Management, Baltimore, Maryland, USA

PDF icon PDF Version (180 KB)

  • Summary:
    Public health has potential to serve as a frame to convey the urgency of behavior change needed to adapt to a changing climate and reduce greenhouse gas emissions. Local governments form the backbone of climate-related public health preparedness. Yet local health agencies are often inadequately prepared and poorly integrated into climate change assessments and plans. We reviewed the climate health profiles of 16 states and two cities participating in the U.S. Centers for Disease Control and Prevention (CDC)’s Climate-Ready States and Cities Initiative (CRSCI) that aims to build local capacity to assess and respond to the health impacts of climate change. Following recommendations from a recent expert panel strategic review, we present illustrations of emerging promising practice and future directions. We found that CRSCI has strengthened climate preparedness and response in local public health agencies by identifying critical climate-health impacts and vulnerable populations, and has helped integrate health more fully into broader climate planning. Promising practice was found in all three recommendation areas identified by the expert panel (leveraging partnerships, refining assessment methodologies and enhancing communications), particularly with regard to health impacts of extreme heat. Vast needs remain, however, suggesting the need to disseminate CRSCI experience to non-grantees. In conclusion, the CRSCI program approach and selected activities illustrate a way forward toward robust, targeted local preparedness and response that may serve as a useful example for public health departments in the United States and internationally, particularly at a time of uncertain commitment to climate change agreements at the national level. https://doi.org/10.1289/EHP1838
  • Received: 01 March 2017
    Revised: 11 April 2017
    Accepted: 12 April 2017
    Published: 20 September 2017

    Address correspondence to M. Sheehan, Johns Hopkins Bloomberg School of Public Health, Dept. of Health Policy and Management, 615 N. Wolfe St., Baltimore, MD 21205 USA. Telephone: 703-663-0536. Email: msheeh10@jhu.edu

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Introduction

The risks to human health of a warmer, more extreme, and variable “new climate normal” (World Bank 2014) are diverse and increasingly apparent. They include illnesses, injuries, and deaths as the direct result of excessive heat and more violent storms, as well as harm caused indirectly through deteriorating air quality; a wider range and greater frequency of insect-, food-, and water-borne diseases; growing risks to food and water supplies; and the enhanced mental stress these risks bring (McMichael 2014; Patz et al. 2014; Smith et al. 2014). Population vulnerability, due to varying exposure and sensitivity, plays a role in who bears the highest risks, while adaptive capacity helps determine who is best able to manage these risks (Smith et al. 2014; Crimmins et al. 2016). The World Health Organization has identified climate change as “the defining issue” for public health in the 21st century, and urged that human health be placed at the center of climate change efforts (WHO 2016, 2017). Public health is a useful lens for conveying the urgency of behavior change needed to reduce greenhouse gas emissions and develop resilience to a changing climate (Maibach et al. 2010; Humphreys 2014). The recent Climate and Health Conference hosted by a group of stakeholders, including former U.S. Vice President Gore, along with the American Public Health Association designation of 2017 as the “Year of Climate Change and Health,” are steps toward framing climate change as a public health challenge.

Yet the growing number of health risks associated with a changing climate—whether well-known, such as heat-related illness (HRI), or less widely recognized, such as the additional burden of injuries due to storm-related motor vehicle accidents (Liu et al. 2017)—are often inadequately integrated into broader climate change adaptation efforts. Such risks require modeling to guide design of health adaptive policies, such as vulnerability-targeted early warning systems (Ebi and Rocklov 2014) that are incorporated into broader national and local climate adaptation plans (Araos et al. 2016). Similarly, more frequent inclusion of co-benefits to health, such as from reduced chronic cardiovascular and respiratory disease burden due to lower greenhouse gas emissions (Thompson et al. 2016; Watts et al. 2016), could strengthen the effectiveness of national and local climate strategies and plans. While recent progress has been made, climate-related health research still lags behind climate research in other fields, such as transport and energy (Jessup et al. 2013; Verner et al. 2016), and evaluation evidence regarding best practice interventions to manage climate-induced health risks is sparse (Hosking and Campbell-Lendrum 2012; Bouzid et al. 2013).

The health effects of climate change occur at the individual, family, and community levels. Local (referred to as subnational, whether city, state, or other nonnational jurisdiction) governments are therefore the first line of defense; they form the backbone of public health preparedness, surveillance, and response (Frumkin et al. 2008; Maibach et al. 2008). With intimate knowledge of population needs, local governments can often be more agile, innovative, and proactive in addressing these needs than national governments (Barata and Ligeti 2011). However, evidence suggests many local health departments remain at early stages of climate-related adaptation (Roser-Renouf et al. 2016; Araos et al. 2016). The U.S. Centers for Disease Control and Prevention (CDC) Climate and Health Program is one of few programs that builds capacity at the subnational level to assess and respond to the health impacts of climate change, supporting innovative approaches to adapt and protect health in U.S. communities through the Climate-Ready States and Cities Initiative (CRSCI) (APHA 2015). Since 2010, CRSCI has awarded annual grants ranging from $100,000 to $250,000 (CDC 2017) on a competitive basis to public health departments in 16 states and 2 cities, with some states passing these funds through to cities, counties, and other localities. The state CRSCI grantees are: Arizona, California, Florida, Illinois, Maine, Maryland, Massachusetts, Michigan, Montana, New Hampshire, New York, North Carolina, Oregon, Rhode Island, Vermont, and Wisconsin; the city grantees are New York City and San Francisco (https://www.cdc.gov/climateandhealth/crsci_grantees.htm).

CRSCI is guided by a stepwise risk assessment and management framework called Building Resilience Against Climate Effects (BRACE), which is grounded in the principles of adaptive management, a learning-focused iterative approach developed for interventions in complex systems (Hess et al. 2012). BRACE’s five-step framework is oriented toward testing adaptation solutions in the context of locally relevant risks and vulnerabilities, and is designed to help grantees build public health adaptive capacity while managing and minimizing population health impacts (Marinucci et al. 2014; CDC 2014). BRACE’s five steps are to a) forecast climate health impacts and vulnerabilities; b) project climate-related disease burden; c) assess relevant public health interventions; d) develop climate and health adaptation plans; and e) evaluate implementation. At the current stage in CRSCI implementation, grantees have undertaken Climate Health Profile Reports (CHPRs) or similar analytical work that report on locally relevant climate hazards, health impacts of concern, and vulnerable populations (step 1 in the BRACE framework), and are moving toward subsequent steps, including preparation of climate health adaptation plans.

To contribute to dissemination of lessons on building adaptive capacity for local public health departments in the United States and elsewhere, in this brief communication, we present results of a survey of CRSCI grantee CHPRs. The survey was conducted via an online search and review of the 18 grantee health department websites, augmented with other publicly available grantee and CRSCI publications. We reviewed CHPRs and related analyses published between 2014 and February 2017 with a view of synthesizing self-reported hazards, health risks, and vulnerabilities; identifying promising practice; and highlighting future challenges. This survey was done in conjunction with an expert panel review of the CRSCI program convened in early 2016 by Johns Hopkins University investigators under contract to CDC’s National Center for Environmental Health. With the goal of providing strategic guidance to the CRSCI program after 5 y of implementation, the expert panel had identified core recommendations for enhancing the CRSCI program in three broad areas:

  • Expand and leverage climate and health partnerships, including sharing technical knowledge and building local public health workforce capacity, reducing fragmentation through greater cross-disciplinary integration, and building links across existing programs and toward new partners, for example, reaching out toward cities.
  • Refine climate and health assessment methodologies, including simplifying and developing how-to guides for quantitative analyses; recognizing the value of qualitative information, particularly for less-resourced jurisdictions; and placing priority on estimation of health co-benefits from climate change mitigation efforts.
  • Enhance climate and health communications, including by employing risks to human health as a useful frame for climate change, proactively evaluating and disseminating promising practice, and communicating in ways that resonate with target populations, including storytelling and linking climate change to other well-known public health messages.

We first summarize features of CHPRs, then highlight examples of promising practice from these reports and grantee websites, followed by implications for gaps, challenges, and future directions. We use the expert panel recommendations as a framework for these illustrations.

Discussion

Our review identified CHPRs or other climate and health profile information in varying formats for all 18 CRSCI grantees. Most reports were found on local health department websites, which, in many cases, contained additional climate and health-related resources, including locally conducted epidemiology, vulnerability indexes, risk maps, and risk communication tools. Most CHPRs were prepared and issued by the grantee public health department (alone or with a technical partner), and virtually all CHPRs involved partnership with a university or another specialized agency (Table 1). The following sections are based on self-reported information provided in grantee publications, as referenced.

Table 1. Features of CRSCI grantee climate health profile reports.
CHPR features Number (share) grantees
Prepared and issued by
 - State or city health department 12 (67%)
 - Health department with technical partner(s) 3 (17%)
 - Technical partner 3 (17%)
Referred to technical partnerships 18 (100%)
Climate impact geographic scope
 - State (or city)-wide 14 (78%)
 - Focus regions 4 (22%)
Climate impact selectivity
 - Inclusive impacts 14 (78%)
 - Specific ranked impacts 4 (22%)
Reported downscaling climate models 14 (78%)
Vulnerability assessment
 - Developed vulnerability indexes 10 (56%)

Note: CHPR, Climate Health Profile Reports; CRSCI, Climate-Ready States and Cities Initiative.

In addition to addressing the first step of the BRACE assessment process, CHPRs aimed at one or more of the following goals: to serve as a public outreach tool; to guide new, or fine-tune existing, adaptation strategies and interventions; to identify climate change and health data availability and gaps; and to contribute to identifying and developing good practice (e.g., OHA 2014; MIAEH 2016). Reports also aimed to provide baseline data, gaps, and analytical and institutional underpinning for subsequent BRACE framework steps, including projecting climate-related disease burdens and identifying interventions. Several grantees reported carrying out analytical work, including developing toolkits, outreach, and educational material; health department capacity gap assessments; and inventories of potential interventions. CHPRs took different approaches, falling broadly into three categories: a) geographically comprehensive assessments, often focused on selected high-priority health impacts, e.g., Rhode Island (RIDH 2015) and Vermont (VDH 2016); b) assessments piloted in several counties covering multiple health impacts, often with an emphasis on capacity building, e.g., Oregon (OHA 2014) and Maryland (MIAEH 2016); and c) assessments aimed at capacity and data gap identification, focused on informing local health departments, e.g., New York State (NYSDH 2015) or as public outreach, e.g., Wisconsin (WDHS 2017). (Table 1).

Near-term, direct impacts of more frequent and extreme weather, including heat, storms (often defined as winter and/or summer), and flooding, were concerns common to all CHPRs. The most frequently reported specific climate-related health risks across grantees were increased HRI and heat-related mortality and increased storm- and flood-related risks (e.g., injuries, motor vehicle accidents, carbon monoxide poisoning), critical service interruptions (affecting hospitals and pharmacies), and mental health impacts. Grantees identified a range of barriers and constraints within these two impact categories. For example, regarding extreme heat, a standard national definition does not exist for HRI, and few states are legally obligated to report this group of health outcomes (FDH 2015); several states proposed specific definitions and surveillance approaches. In the case of extreme storms, public health preparedness involves collaboration across departments (e.g., emergency services, public works, transportation, building, and zoning) that can be challenging to implement; many CHPRs identified concrete ways to enhance partnerships across disciplines and services (e.g., Cameron et al. 2015; MDPH 2014). On the other hand, indirect and longer-term climate change drivers and health risks varied; among commonly reported concerns were worsening air pollution and aeroallergens, water quality and contamination (particularly with combined sewage and drainage infrastructure), and ecosystem changes that modify disease vector patterns. In these categories, additional health risks frequently cited by grantees included pollution-related respiratory disease (including asthma), vector-borne disease (most commonly Lyme disease and West Nile virus), and illness due to contamination of food and water (harmful algal blooms and vibrio). The mix of these concerns differed across geographic, population, urbanization, and climate contexts; for example, health risks in Arizona focused on HRI and air pollution; those in Illinois centered on HRI, vector-borne disease, and allergies; and those in New York City involved HRI, storm-related injuries, and risks from critical infrastructure outages.

Grantees took differing approaches to evaluating population vulnerability. Over half reported developing social vulnerability or other indexes, mapping, and other quantitatively derived tools to assess vulnerability, often with the assistance of a university technical partner (Table 1). Population characteristics frequently cited as associated with vulnerability included: being elderly, being very young, having a preexisting health condition, having lower income, being a minority, working outdoors, living in vulnerable geographic locations, and lacking protective infrastructure, such as air-conditioning. Several states with larger rural populations identified tribal groups, agricultural workers, and those with private well water as at greater risk. Most CHPR vulnerability assessments provided practical insights into near-term ways to enhance the effectiveness of ongoing interventions. This was most evident in the case of extreme heat, where several grantees reported creating vulnerability indexes based on sensitivity and exposure factors tracked by census block and mapping the results. For example, San Francisco reported using a heat vulnerability index and mapping to improve the effectiveness of heat wave early warning and response plans (SFDPH 2014). Other illustrations of improved heat response targeting undertaken or planned based on CHPRs include piloting use of risk communication via social media (Chuang et al. 2015), definition of an HRI indicator to improve disease identification and monitoring (FDH 2015), enhancing surveillance and voluntary reporting of HRI (FDH 2015; Fernandez et al. 2015; NYSDH 2015; RIDH 2015), educating outdoor workers on heat safety (Chuang et al. 2015), identifying education and communication outreach venues (Cameron et al. 2015), and identifying factors leading to unhealthy heat exposure in city apartment buildings (Kinney et al. 2015).

Promising Practices

Expanding and Leveraging Climate and Health Partnerships

Networked learning partnerships.

As reported in CHPRs, grantees developed networked learning partnerships for multiple purposes. For example, several states collaborate in a northeast regional group to examine the impacts of heat on social and geographic vulnerabilities, with the goal of developing consistent methodologies and sharing knowledge across states (RIDH 2015). Similarly, a western state collaborative is working to develop common social vulnerability metrics that can be compared across states (OHA 2014). Several northeast state grantees have initiated a community of practice to assess regional climate impacts on Lyme disease (RIDH 2015). Many state CHPRs highlighted training of county public health agency staff as an explicit goal, either through programs directly with these localities or through broader statewide training initiatives.

Partnership across existing programs.

Nearly all CHPRs identified local and national partnerships. For example, several grantees had ongoing partnerships with the National Oceanic and Atmospheric Administration, state climate agencies, or locally convened science panels. Other types of partnerships included participation in the National Association of County and City Health Officials (NACCHO) Climate Change Workgroup. A majority of grantees grounded their CHPR in collaboration with local Environmental Public Health Tracking (EPHT) programs, whether to develop baseline assessments of climate relevant health outcomes or to enhance monitoring of climate-related health indicators and outcomes, such as HRI or Lyme disease (e.g., MDPH 2014; NYSDH 2015; OHA 2014).

Greater cross-disciplinary integration.

Several grantee programs also illustrated ways of building integration across state agencies within multidisciplinary adaptation planning. For example, New York State’s CHPR outlines a structure with four cross-cutting climate impact teams setting the goal of shifting climate change from an environmental to a public health issue (NYSDH 2015). In New York City, the health department participates in multiagency planning around the priority of ensuring that urban built infrastructure is more resilient and protects public health (Kinney et al. 2015). Massachusetts developed interactive mapping of risk factors based on EPHT and other state data sources integrated across agencies in the service of localities and residents (MDPH 2014). Several states chose to decentralize their program to pilot localities, which developed integrated assessments, adaptation plans, and partnerships across local agencies, for example, the New Hampshire Lake District’s focus on Lyme disease (NH LRPPH 2016).

Refining Climate and Health Assessment Methodologies

Simplifying and developing how-to guides for quantitative analyses.

The most data-intensive quantitative analyses of the BRACE assessment framework steps are guided by a CDC Climate and Health Technical Report Series (https://www.cdc.gov/climateandhealth/publications.htm) designed to assist grantees in tasks including downscaling climate models to derive finer-scale resolution local climate projections (e.g., Hess et al. 2014; Schramm et al. 2014) and development of vulnerability assessments (e.g., Manangan et al. 2014). Through implementation of the CRSCI program, however, collaboration between the health department, universities, and other partners has resulted in new, practice-informed guidance tools. For example, recognizing the need to build familiarity with climate modeling and implications for disease burden projection among public health professionals, the Florida CRSCI program documented cross-disciplinary collaboration between regional climate scientists and public health staff as a guide for other public health departments describing methodologies and providing case studies that address drought, heat, and tropical cyclones (Conlon et al. 2016).

Focusing on mitigation.

A few grantees reported having begun to focus on mitigation by estimating health co-benefits from efforts to reduce greenhouse gas emissions. For example, in Oregon, two health impact assessments were conducted based on regional transport plans that would reduce reliance on light-duty vehicles; quantified health benefits were derived both from increased physical activity as well as improved air quality (OHA 2014). New Hampshire’s assessment suggested considering primary, secondary, and tertiary prevention as a lens to plan interventions connecting public health to other sectors (e.g., transportation, energy, water management) and estimating co-benefits associated with primary and secondary prevention that would reduce chronic cardiovascular and respiratory disease as well as minimize vulnerability (UNH 2015).

Recognizing the value of qualitative methods.

Qualitative methods, such as surveys and community-based participatory approaches, can provide valuable information. The CHPR prepared by New York State was among those that used qualitative methods to identify population vulnerability and partnerships. Needs assessment surveys of health department staff knowledge regarding climate and health population vulnerability factors provided insights based on direct experience and highlighted differences in perceptions across stakeholders regarding partnerships; for example, health department staff prioritized collaboration with emergency management services, while external stakeholders considered partnerships with schools and agencies to be more critical (NYSDH 2015).

Enhancing Climate and Health Communications

Health as a frame for climate change.

Our review of CHPRs suggests that CRSCI has helped grantees integrate health more effectively into broader climate change efforts. This occurred in different ways, depending on the context. For example, in California, CRSCI appears to have contributed to enhancing an ongoing climate and health program that was already part of a broader climate change initiative implemented by an interdisciplinary team (Maizlish et al. 2017; CalBRACE 2017). In Maryland, the program appears to have helped elevate the local health department’s role within a wider multisectoral climate change effort in which health had previously not featured prominently (MCCCAR 2008; MIAEH 2016), while in Michigan, CRSCI seems to have spurred the launch of a new climate and health program with potential to provide leadership on broader climate change efforts (Cameron et al. 2015). In the majority of grantees, the CRSCI-supported program appears to have fallen into one or both the two first categories by raising the visibility of climate-driven health risks and their importance to vulnerable local populations, and better integrating public health within an existing broader climate change effort.

Evaluating and disseminating promising practice.

Several CHPRs recognized the need for evaluation of interventions and noted a lack of evidence-based knowledge of good practice. However, grantees demonstrated ways to share promising practice, including in Oregon, where the five local jurisdictions chosen as pilots developed climate health adaptation plans and formed an online toolbox to share effective practice for natural hazard preparedness with other local jurisdictions. These included modules on flooding and wildfires that are based on lessons and communication guidance from past experiences with these hazards (OHA 2014). In a similar way, the CRSCI programs in Florida, California, and Minnesota report creating extreme heat toolkits for local jurisdictions (FDH 2015; Maizlish et al. 2017; MDH 2015). Rhode Island has developed a Lyme disease prevention toolkit based on vulnerable populations and high-risk occupations (RIDH 2015).

Communicating in ways that resonate.

Several state grantees reported having begun to communicate in simpler and more direct ways with an emphasis on storytelling. For example, the Florida program developed touchstone event summaries for historic extreme weather events, including the 1990 flood and 2006–2008 drought; using photos, data records, and personal stories, touchstone summaries maintain knowledge of these events in community memory (FDH 2014). Illinois’s CHPR highlighted several personal stories of confronting weather extremes, including the 2013 extreme flood, and developed a video on preparation for extreme weather (UICSPH and IDPH 2016). California has developed case stories of successful activities to reduce climate-related risks in communities (CalBRACE 2017). North Carolina created an educational campaign on HRI (NCDHHS 2015), and Maine developed eighth-grade teaching modules on climate change and health, Spanish-language guidance on how to detect vibrio in locally caught seafood, and cartoon versions of guidance for extreme weather strategies (MCDC 2017).

Future Directions

The CHPRs reviewed suggest that grantee states and cities have developed climate impact and vulnerability assessments that serve several relevant goals, including identifying key climate hazards, associated health risks, and population vulnerability factors. They have also addressed health department training and capacity building and public outreach, and have informed next steps toward development of climate health adaptation plans. This suggests the utility of the BRACE framework and the potential applicability of the CHPR approach for other local governments. CHPRs have also served in a hands-on way to inform refinement and targeting of existing health interventions. This was particularly true for extreme heat early warning systems, among the few climate health interventions that have been evaluated and considered likely to be cost-effective (Bouzid et al. 2013; Toloo et al. 2013). Research suggests that vulnerability targeting, such as reported by grantees, is likely key to enhance early warning effectiveness (Lowe et al. 2011; Hess and Ebi 2016). A systematized review of CRSCI grantee extreme heat-related activities, including use of heat vulnerability indexes and mapping approaches, could help identify areas of promising practice, opportunities for intervention evaluation, and scope for scaling up of activities. Some lessons learned in heat early warning may be applicable to other health risks, such as air and water pollution alerts and contagious disease outbreak early warning.

We found numerous illustrative examples in CHPRs and associated CRSCI activities that were consistent with strategic directions suggested by the recent expert review panel, in particular, the development of knowledge networks, progress toward practice-informed guidance on quantitative methods, and integrating public health more centrally within broader climate changes efforts, a key program achievement. In some of these areas, it may be useful to review grantee experience with a view to developing additional toolkits, guidance documents, and identifying scope for case studies. However, we found fewer examples to report in some of the other strategic areas recommended by this expert panel. For examples, we found less evidence of the use of qualitative methods, such as surveys and community-based participatory research; few CHPRs reported a focus on the health co-benefits associated with climate mitigation efforts. And while we identified networking partnerships among grantees and between grantees, nongrantees, and academic, government, and nongovernment partners, the CRSCI program has potential to expand its networks even further toward nongrantees through existing and new networks (e.g., through NACCHO or other organizations). Development of further practice-informed guidance documents (e.g., for co-benefit estimates or for use of qualitative methods for vulnerability assessment) would also be a useful addition and would support dissemination of lessons to other local governments. New communication strategies are also being tested by grantees; evaluation of their effectiveness and sharing lessons of success will be helpful to others. Tools such as these could be shared through an online dissemination channel with an international reach; particularly in the context of scarce resources, this would help to scale up lessons and make them more widely available to local health departments, both in the United States and worldwide. Also of use would be extending the program reach toward additional cities, increasingly important actors in climate change adaptation and mitigation. Program enhancements would also be useful in areas such as estimating the health co-benefits of climate mitigation efforts, support to climate and health institutional capacity building and lesson dissemination, and systematic evaluation of climate and health interventions.

Conclusion

Through relatively small amounts of grant support over the last 6 y, CRSCI has helped local public health agencies in sixteen states and two cities—whose combined population reaches half of the U.S. total—identify critical climate impacts and vulnerable populations. In the process, the program has helped to integrate health more fully into local climate change efforts. As a result of CRSCI support, these local public health agencies—the backbone of public health climate response capacity—have tools to enhance real-life adaptive capacity and increase the effectiveness of existing interventions, such as heat response plans. They are also better prepared to take the next steps toward developing climate and health adaptation plans. However, the CRSCI program is only a start; vast needs remain in the United States as well as globally. At a time of uncertain commitment to climate change agreements at the national level, the challenge of building adaptive capacity in public health rests in large part at the local level. Subnational governments worldwide have a role to play in adapting to the health risks of a changing climate and enhancing the urgency of needed mitigation policies. As demonstrated with the examples here, the CRSCI program approach illustrates a way forward toward robust, targeted, and resilient local preparedness and response that may serve as a useful model for public health departments in the United States and internationally as the climate continues to change.

Acknowledgments

The authors thank the expert panel participants for offering their time and advice at the strategic planning workshop. A. Briskin-Limehouse contributed to an early draft of the paper. This work was supported by the Centers for Disease Control and Prevention, contract number 200I2015IMI87947.

References

APHA (American Public Health Association). 2015. Adaptation in Action: Grantee Success Stories from CDC’s Climate and Health Program. Washington, DC: American Public Health Association. https://www.apha.org/∼/media/files/pdf/topics/environment/adapt_in_action.ashx [accessed 10 February 2017].

Araos M, Austin SE, Berrang-Ford L, Ford JD. 2016. Public health adaptation to climate change in large cities: A global baseline. Int J Health Serv 46(1):53–78, PMID: 26705309, 10.1177/002073141562145.

Barata M, Ligeti E, De Simone G, Dickinson T, Jack D, Penney J, et al. 2011. Climate change and human health in cities. In: Climate Change and Cities: First Assessment Report of the Urban Climate Change Research Network. Rosenzweig C, et al, eds. Cambridge, UK:Cambridge University Press. http://uccrn.org/files/2014/02/ARC3-Chapter-7.pdf [accessed 10 February 2017].

Bouzid M, Hooper L, Hunter PR. 2013. The effectiveness of public health interventions to reduce the health impact of climate change: a systematic review of systematic reviews. PloS One 8(4):e62041, PMID: 23634220, 10.1371/journal.pone.0062041.

CalBRACE. 2017. Preparing for Climate Change in California – A Public Health Approach. CalBRACE: California Building Resilience against Climate Effects. Sacramento, CA:California Department of Public Health. https://www.cdph.ca.gov/Programs/OHE/Pages/ClimateHealthProfileReports.aspx [accessed 27 June 2017].

Cameron L, Ferguson A, Walker R, Brown D, Briley L. 2015. “Michigan Climate and Health Profile Report 2015: Building Resilience Against Climate Effects on Michigan’s Health.” Lansing, MI:Michigan Department of Health and Human Services. http://www.michigan.gov/documents/mdhhs/MI_Climate_and_Health_Profile_517517_7.pdf [accessed 10 February 2017].

CDC (U.S. Centers for Disease Control and Prevention). 2014. CDC’s Climate-Ready States and Cities Initiative. Atlanta, GA:Centers for Disease Control and Prevention. https://www.cdc.gov/climateandhealth/climate_ready.htm [accessed 10 February 2017].

CDC. 2017. CDC Grant Funding Profiles – Summary View. Atlanta, GA:Centers for Disease Control and Prevention. https://wwwn.cdc.gov/FundingProfilesApp [accessed 10 February 2017].

Chuang W-C, Karner A, Selover N, Hondula D, Chhetri N, Middel A, et al. 2015. “Arizona Extreme Weather, Climate and Health Profile Report. A report prepared for Arizona Department of Health Services and the United States Centers for Disease Control and Prevention ClimateIReady States and Cities Initiative.” Arizona Department of Health Services, Phoenix. http://www.azdhs.gov/documents/preparedness/epidemiology-disease-control/extreme-weather/pubs/arizona-extreme-weather-climate-and-health-profile-report-2015-executive-summary.pdf [accessed 10 February 2017].

Conlon KC, Kintziger KW, Jagger M, Stefanova L, Ueijo CK, Konrad C. 2016. Working with climate projections to estimate disease burden: perspectives from public health. Int J Environ Res Public Health.
13(8). pii: E804, PMID: 27517942, 10.3390/ijerph13080804.

Crimmins A, Balbus J, Gamble JL, Beard CB, Bell JE, Dodgen D, et al. 2016. “The Impacts of Climate Change on Human Health in the United States: A Scientific Assessment,” executive summary. U.S. Global Change Research Program, Washington, DC, 10.7930/J0R49NQX [accessed 17 January 2017].

Ebi KL, Rocklöv J. 2014. Climate change and health modeling: horses for courses. Glob Health Action 7:24154, PMID: 24861341.

FDH (Florida Department of Health). 2015. “Health Effects of Summer Heat in Florida.” Tallahassee, FL:Florida Department of Health. http://www.floridahealth.gov/environmental-health/climate-and-health/_documents/heat-profile.pdf [accessed 10 February 2017].

Fernandez IJ, Schmitt CV, Birkel SD, Stanicoff E, Pershing AJ, Kelly JT, et al. 2015. “Maine’s Climate Future, 2015 Update.” Orono, ME:Climate Change Institute, University of Maine. http://climatechange.umaine.edu/research/publications/climate-future [accessed 10 February 2017].

Frumkin H, Hess JJ, Luber G, Maililay J, McGeehin M. 2008. Climate change: the public health response. Am J Public Health 98(3):435–445, 10.2105/AJPH.2007.119362.

Hess JJ, Ebi KL. 2016. Iterative management of heat early warning systems in a changing climate. Ann NY Acad Sci 1382(1):12–30.

Hess JJ, McDowell JZ, Luber G. 2012. Integrating climate change adaptation into public health practice: using adaptive management to increase adaptive capacity and build resilience. Environ Health Perspect 120(2):171–179, PMID: 21997387, 10.1289/ehp.1103515.

Hess JJ, Saha S, Schramm PJ, Conlon KC, Uejio CK, Luber G. 2014. “Projecting Climate-Related Disease Burden: A Guide for Health Departments.” CDC Climate and Health Technical Report Series. Atlanta, GA:National Center for Environmental Health, Centers for Disease Control and Prevention. https://www.cdc.gov/climateandhealth/pubs/projectingclimaterelateddiseaseburden1_508.pdf [accessed 5 April 2017].

Hosking J, Campbell-Lendrum D. 2012. How well does climate change and human health research match the demands of policymakers? A scoping review. Environ Health Perspect 120(8):1076–1082, PMID: 22504669, 10.1289/ehp.1104093.

Humphreys G. 2014. Reframing climate change as a health issue. Bull World Health Organ 92(8):551–552, PMID: 25177068, 10.2471/BLT.14.020814.

Jessup CM, Balbus JM, Christian C, Haque E, Howe SE. et al. 2013. Climate change, human health, and biomedical research: Analysis of the National Institutes of Health research portfolio. Environ Health Perspect 121(4):399–404, PMID: 23552460, 10.1289/ehp.1104518.

Kinney PL, Matte T, Knowlton K, Madrigano J, Petkova E, Weinberger K, et al. 2015. New York City Panel on Climate Change 2015 Report. Chapter 5: Public health impacts and resiliency. Ann NY Acad Sci, 1336:67–88, PMID: 25688947, 10.1111/nyas.12588.

Liu A, Soneja SI, Jiang C, Huang C, Kerns T, Beck K, et al. 2017. Frequency of extreme weather events and increased risk of motor vehicle collision in Maryland. Sci Total Environ 580:550–555, PMID: 27988189, 10.1016/j.scitotenv.2016.11.211.

Lowe D, Ebi KL, Forsberg B. 2011. Heatwave early warning systems and adaptation advice to reduce human health consequences of heatwaves. Int J Environ Res Public Health 8(12):4623–4648, PMID: 22408593, 10.3390/ijerph8124623.

Maibach EW, Chadwick A, McBride D, Chuk M, Ebi KL, Balbus J. 2008. Climate change and local public health in the United States: preparedness, programs and perceptions of local public health department directors. PloS One 3(7):e2838, PMID: 18665266, 10.1371/journal.pone.0002838.

Maibach EW, Nisbet M, Baldwin P, Akerlof K, Diao G. 2010. Reframing climate change as a public health issue: an exploratory study of public reactions. BMC Public Health 10:299, PMID: 20515503, 10.1186/1471-2458-10-299.

Maizlish N, English D, Chan J, Dervin K, English P. 2017. “Climate Change and Health Profile Report: Fresno County.” Sacramento, CA: Office of Health Equity, California Department of Public Health. https://archive.cdph.ca.gov/programs/Documents/CalBRACE%202015CHPR/CHPR019Fresno_County2-23-17.pdf [accessed 5 April 2017].

Manangan AP, Uejio C, Saha S, Schramm PJ, Marinucci GD, Brown CL, et al. 2014. “Assessing Health Vulnerability to Climate Change: A Guide for Health Departments.” CDC Climate and Health Technical Report Series. Atlanta, GA:National Center for Environmental Health, Centers for Disease Control and Prevention. https://www.cdc.gov/climateandhealth/pubs/assessinghealthvulnerabilitytoclimatechange.pdf [accessed 5 April 2017].

Marinucci GD, Luber G, Uejio CK, Saha S, Hess JJ. 2014. Building resilience against climate effects: a novel framework to facilitate climate readiness in public health agencies. IJERPH 11(6):6433–6458, 10.3390/ijerph11060643.

MCCCAR (Maryland Commission on Climate Change Adaptation and Response). 2008. Comprehensive Strategy for Reducing Maryland’s Vulnerability to Climate Change. Baltimore, MD:Maryland Department of the Environment. http://dnr2.maryland.gov/ccs/Publication/Comprehensive_Strategy.pdf [accessed 10 February 2017].

MCDC (Maine Center for Disease Control and Prevention) 2017. Environmental Public Health, Maine Tracking Network. Augusta, ME:Maine Center for Disease Control and Prevention. https://data.mainepublichealth.gov/tracking/home [accessed 10 February 2017].

McMichael AJ. 2014. Earth as humans’ habitat: Global climate change and the health of populations. Int J Health Policy Manag 2(1):9–12, 10.15171/ijhpm.2014.03.

MDH (Minnesota Department of Health). 2015. “Minnesota Climate and Health Report.” St. Paul, MN:Minnesota Department of Health. http://www.health.state.mn.us/divs/climatechange/docs/mnprofile2015.pdf [accessed 10 February 2017].

MDPH (Massachusetts Department of Public Health). 2014. Environmental Public Health Tracking, Climate and Health Profile. Boston, MA:Massachusetts Department of Public Health, Bureau of Environmental Health. https://matracking.ehs.state.ma.us/Climate-Change/climate_and_health_profile.html [accessed 10 February 2017].

MIAEH (Maryland Institute for Applied Environmental Health). 2016. “Maryland Climate and Health Profile Report.” College Park, MD:University of Maryland School of Public Health. http://phpa.dhmh.maryland.gov/OEHFP/EH/Shared%20Documents/Climate%20Change/Reports/MD_climate_and_health_FullReport_04182016%20Final.pdf [accessed 10 February 2017].

NCDHHS (North Carolina Department of Health and Human Services). 2015. “North Carolina Climate and Health Profile.” Raleigh, NC:North Carolina Department of Health and Human Services. http://epi.publichealth.nc.gov/oee/climate/ClimateAndHealthProfile.pdf [accessed 10 February 2017].

NH LRPPH (New Hampshire Lakes Region Partnership for Public Health). 2016. Health and Climate Change in New Hampshire’s Lakes Region: An Action Plan. Laconia, NH:New Hampshire Lakes Region Partnership for Public Health. https://www.dhhs.nh.gov/dphs/climate/documents/climate-plan-lakes.pdf [accessed 10 February 2017].

NYSDH (New York State Department of Health). 2015. “Climate and Health Profile.” Albany, NY:New York State Department of Health. https://www.health.ny.gov/environmental/weather/docs/climatehealthprofile6-2015.pdf [accessed 10 February 2017].

OHA (Oregon Health Authority). 2014. “Oregon Climate and Health Profile Report.” Portland, OR:Oregon Health Authority. https://public.health.oregon.gov/HealthyEnvironments/climatechange/Documents/oregon-climate-and-health-profile-report.pdf [accessed 10 February 2017].

Patz JA, Frumkin H, Holloway T, Vimont DJ, Haines A. 2014. Climate change: challenges and opportunities for global health. JAMA 312(15):1565–1580, 10.1001/jama.2014.13186.

RIDH (Rhode Island Department of Health). 2015. “Climate Change and Health Resiliency Report.” Providence, RI:Rhode Island Department of Health. http://health.ri.gov/publications/reports/ClimateChangeAndHealthResiliency.pdf [accessed 10 February 2017].

Roser-Renouf C, Maibach EW, Li J. 2016. Adapting to the changing climate: an assessment of local health department preparations for climate change-related health threats, 2008-2012. PLoS ONE 11(3):e0151558, PMID: 26991658, 10.1371/journal.pone.0151558.

Schramm PJ, Uejio CK, Hess JJ, Marinucci GD, Luber G, 2014. “Climate Models and the Use of Climate Projections: A Brief Overview for Health Departments.” CDC Climate and Health Technical Report Series. Atlanta, GA:National Center for Environmental Health, Centers for Disease Control and Prevention. https://www.cdc.gov/climateandhealth/pubs/climate_models_and_use_of_climate_projections.pdf [accessed 5 April 2017].

SFDPH (San Francisco Department of Public Health). 2014. “San Francisco Climate Health Profile.” San Francisco, CA: San Francisco Department of Public Health. http://www.sfclimatehealth.org/wp-content/uploads/2015/01/SFDPH_ClimateHealthProfile_FinalDraft.pdf [accessed 10 February 2017].

Smith KR, Woodward A, Campbell-Lendrum D, Chadee DD, Honda Y, Liu Q, et al. 2014. Chapter 11: Human health: impacts, adaptation and co-benefits. In: Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Field CB, Barros VR, Dokken DJ, Mach KJ, Mastrandrea MD, Bilir TE, et al, eds. Cambridge, UK and New York, NY:Cambridge University Press. https://www.ipcc.ch/pdf/assessment-report/ar5/wg2/WGIIAR5-Chap11_FINAL.pdf [accessed 17 January 2017].

Thompson TM, Rausch S, Saari RK, Selin NE. 2016. Air quality co-benefits of subnational carbon policies. J Air Waste Manag Assoc 66(10):988–1002, PMID: 27216236, 10.1080/10962247.2016.1192071.

Toloo GS, Fitzgerald G, Aitken P, Verrall K, Tong S. 2013. Are heat warning systems effective? Environ Health 12:27, PMID: 23561265, 10.1186/1476-069X-12-27.

UICSPH and IDPH (University of Illinois at Chicago School of Public Health and Illiniois Department of Public Health). 2016. “Climate and Health in Illinois.” Chicago, IL:Illinois Department of Public Health. http://www.dph.illinois.gov/sites/default/files/publications/publicationsoprclimatehealthreport.pdf [accessed 10 February 2017].

UNH (University of New Hampshire), 2015. Climate Change and Human Health in New Hampshire: An Impact Assessment. Durham, NH: University of New Hampshire, New Hampshire Department of Health and Human Services. http://sustainableunh.unh.edu/climate-solutions-new-england/sites/climatesolutionsne.org/files/candhreport4.30.pdf [accessed 10 February 2017].

VDH (Vermont Department of Health). 2016. “Vermont Climate and Health Profile Report.” Montpelier, VT:Vermont Department of Public Health. http://healthvermont.gov/sites/default/files/documents/2017/01/CHPR_Sept7_2016.pdf [accessed 10 February 2017].

Verner G, Schütte S, Knop J, Sankoh O, Sauerborn R. 2016. Health in climate change research from 1990 to 2014: positive trend, but still underperforming. Glob Health Action 9:30723, PMID: 27339855, 10.3402/gha.v9.30723.

Watts N, Adger WN, Ayeb-Karlsson S, Bai Y, Byass P, Campbell-Lendrum D, et al. 2016. The Lancet countdown: Tracking progress on health and climate change. Lancet 389(10074):1151–1164, 10.1016/S0140-6736(16)32124-9.

WDHS (Wisconsin Department of Health Services). 2017. “Understanding the Link between Climate and Health.” Madison, WI:Bureau of Environmental and Occupational Health, Division of Public Health. https://www.dhs.wisconsin.gov/publications/p0/p00709.pdf [accessed 10 February 2017].

WHO (World Health Organization). 2016. WHO Director-General Addresses Human Rights Council on Climate Change. http://www.who.int/dg/speeches/2016/human-rights-council/en [accessed 10 February 2017].

WHO. 2017. World Health Day Message from WHO Director-General. http://www.who.int/world-health-day/dg_message/en [accessed 10 February 2017].

World Bank. 2014. “Turn Down the Heat: Confronting the New Climate Normal. Report No. 3.” Washington, DC: World Bank. http://www.worldbank.org/en/topic/climatechange/publication/turn-down-the-heat [accessed 10 January 2017].

Prenatal Fluoride Exposure and Cognitive Outcomes in Children at 4 and 6–12 Years of Age in Mexico

Author Affiliations open

1Dalla Lana School of Public Health, University of Toronto, Toronto, Ontario, Canada

2University of Michigan School of Public Health, Ann Arbor, Michigan, USA

3Indiana University School of Dentistry, Indiana University-Purdue University Indianapolis, Indianapolis, Indiana, USA

4Faculty of Agricultural and Environmental Sciences, McGill University, Montreal, Quebec, Canada

5Center for Human Growth and Development, University of Michigan, Ann Arbor, Michigan, USA

6Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA

7Icahn School of Medicine at Mount Sinai, New York, New York, USA

8Instituto Nacional de Perinatología, Mexico City, Mexico

9Instituto Nacional de Salud Pública, Cuernavaca, Morelos, Mexico

PDF icon PDF Version (815 KB)

  • Background:
    Some evidence suggests that fluoride may be neurotoxic to children. Few of the epidemiologic studies have been longitudinal, had individual measures of fluoride exposure, addressed the impact of prenatal exposures or involved more than 100 participants.
    Objective:
    Our aim was to estimate the association of prenatal exposure to fluoride with offspring neurocognitive development.
    Methods:
    We studied participants from the Early Life Exposures in Mexico to Environmental Toxicants (ELEMENT) project. An ion-selective electrode technique was used to measure fluoride in archived urine samples taken from mothers during pregnancy and from their children when 6–12 y old, adjusted for urinary creatinine and specific gravity, respectively. Child intelligence was measured by the General Cognitive Index (GCI) of the McCarthy Scales of Children’s Abilities at age 4 and full scale intelligence quotient (IQ) from the Wechsler Abbreviated Scale of Intelligence (WASI) at age 6–12.
    Results:
    We had complete data on 299 mother–child pairs, of whom 287 and 211 had data for the GCI and IQ analyses, respectively. Mean (SD) values for urinary fluoride in all of the mothers (n=299) and children with available urine samples (n=211) were 0.90 (0.35) mg/L and 0.82 (0.38) mg/L, respectively. In multivariate models we found that an increase in maternal urine fluoride of 0.5mg/L (approximately the IQR) predicted 3.15 (95% CI: −5.42, −0.87) and 2.50 (95% CI −4.12, −0.59) lower offspring GCI and IQ scores, respectively.
    Conclusions:
    In this study, higher prenatal fluoride exposure, in the general range of exposures reported for other general population samples of pregnant women and nonpregnant adults, was associated with lower scores on tests of cognitive function in the offspring at age 4 and 6–12 y. https://doi.org/10.1289/EHP655
  • Received: 14 June 2016
    Revised: 08 May 2017
    Accepted: 09 May 2017
    Published: 19 September 2017

    Please send correspondence to M. Bashash, Dalla Lana School of Public Health, 6th floor, 155 College St., Toronto, Ontario M5R3M7 Canada. Telephone: +1-416-978-6512. Email: m.bashash@utoronto.ca

    Supplemental Material is available online (https://doi.org/10.1289/EHP655).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (36 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (26 KB)

Introduction

Community water, salt, milk, and dental products have been fluoridated in varying degrees for more than 60 y to prevent dental caries, while fluoride supplementation has been recommended to prevent bone fractures (Jones et al. 2005). In addition, people may be exposed to fluoride through the consumption of naturally contaminated drinking water, dietary sources, dental products, and other sources (Doull et al. 2006). Whereas fluoride is added to drinking water [in the United States at levels of 0.7–1.2 mg/L (Doull et al. 2006)] to promote health, populations with exceptionally high exposures, often from naturally contaminated drinking water, are at risk of adverse health effects, including fluorosis.

In the United States, the U.S. Environmental Protection Agency (EPA) is responsible for establishing maximum permissible concentrations of contaminants, including fluoride, in public drinking-water systems. These standards are guidelines for restricting the amount of fluoride contamination in drinking water, not standards for intentional drinking-water fluoridation. In 2006 the U.S. EPA asked the U.S. National Research Council (NRC) to reevaluate the existing U.S. EPA standards for fluoride contamination, including the maximum contaminant level goal (MCLG, a concentration at which no adverse health effects are expected) of 4 mg/L, to determine if the standards were adequate to protect public health (Doull et al. 2006). The committee concluded that the MCLG of 4 mg/L should be lowered because it puts children at risk of developing severe enamel fluorosis, and may be too high to prevent bone fractures caused by fluorosis (Doull et al. 2006). The Committee also noted some experimental and epidemiologic evidence suggesting that fluoride may be neurotoxic (Doull et al. 2006).

The National Toxicology Program (NTP) recently reviewed animal studies on the effects of fluoride on neurobehavioral outcomes and concluded that there was a moderate level of evidence for adverse effects of exposures during adulthood, a low level of evidence for effects of developmental exposures on learning and memory, and a need for additional research, particularly on the developmental effects of exposures consistent with those resulting from water fluoridation in the United States (Doull et al. 2006; NTP 2016). Human studies have shown a direct relationship between the serum fluoride concentrations of maternal venous blood and cord blood, indicating that the placenta is not a barrier to the passage of fluoride to the fetus (Shen and Taves, 1974). Fluoride was shown to accumulate in rat brain tissues after chronic exposures to high levels, and investigators have speculated that accumulation in the hippocampus might explain effects on learning and memory (Mullenix et al. 1995). An experimental study on mice has shown that fluoride exposure may have adverse effects on neurodevelopment, manifesting as both cognitive and behavioral abnormalities later in life (Liu et al. 2014).

Most epidemiologic studies demonstrating associations between fluoride exposure and lower neuropsychological indicators have been conducted in populations living in regions with endemic fluorosis that are exposed to high levels of fluoride in contaminated drinking water. The epidemiologic evidence is limited, however, with most studies using an ecologic design to estimate childhood exposures based on neighborhood measurements of fluoride (e.g., drinking water levels) rather than personal exposure measures. Moreover, almost all existing studies of childhood outcomes are cross-sectional in nature, rendering them weak contributors towards causal inference.

The main objective of this study was to assess the potential impact of prenatal exposures to fluoride on cognitive function and test hypotheses related to impacts on overall cognitive function. We hypothesized that fluoride concentrations in maternal urine samples collected during pregnancy, a proxy measure of prenatal fluoride exposure, would be inversely associated with cognitive performance in the offspring children. Overall, to our knowledge, this is one of the first and largest longitudinal epidemiologic studies to exist that either address the association of early life exposure to fluoride to childhood intelligence or study the association of fluoride and cognition using individual biomarker of fluoride exposure.

Methods

This is a longitudinal birth cohort study of measurements of fluoride in the urine of pregnant mothers and their offspring (as indicators of individual prenatal and postnatal exposures to fluoride, respectively) and their association with measures of offspring cognitive performance at 4 and 6–12 y old. The institutional review boards of the National Institute of Public Health of Mexico, University of Toronto, University of Michigan, Indiana University, and Harvard T.H. Chan School of Public Health and participating clinics approved the study procedures. Participants were informed of study procedures prior to signing an informed consent required for participation in the study.

Participants

Mother–child pairs in this study were participants from the successively enrolled longitudinal birth cohort studies in Mexico City that comprise the Early Life Exposures in Mexico to Environmental Toxicants (ELEMENT) project. Of the four ELEMENT cohorts [that have been described elsewhere (Afeiche et al. 2011)], Cohort 1 and Cohort 2B recruited participants at birth and did not have archived maternal-pregnancy urine samples required for this analysis; they were thus excluded. Mothers for Cohort 2A (n=327) and 3 (n=670) were all recruited from the same three hospitals in Mexico City that serve low-to-moderate income populations. Cohort 2A was an observational study of prenatal lead exposure and neurodevelopmental outcomes in children (Hu et al. 2006). Women who were planning to become pregnant or were pregnant were recruited during May 1997–July 1999 and were considered eligible if they consented to participate; were ≤14 wk of gestation at the time of recruitment; planned to stay in the Mexico City study area for at least 5 y; did not report a history of psychiatric disorders, high-risk pregnancies, gestational diabetes; did not report current use of daily alcohol, illegal drugs, and continuous prescription drugs; and were not diagnosed with preeclampsia, renal disease, circulatory diseases, hypertension, and seizures during the index pregnancy.

Cohort 3 mothers were pregnant women (≤14 wk of gestation) recruited from 2001 to 2003 for a randomized trial of the effect of calcium supplementation during pregnancy on maternal blood lead levels (Ettinger et al. 2009). Eligibility criteria were the same as for Cohort 2A, and 670 agreed to participate.

Exposure Assessment

By virtue of living in Mexico, individuals participating in the study have been exposed to fluoridated salt (at 250 ppm) (Secretaría-de-Salud 1995, 1996) and to varying degrees of naturally occurring fluoride in drinking water. Previous reports, based on samples taken from different urban and rural areas, indicate that natural water fluoride levels in Mexico City may range from 0.15 to 1.38 mg/L (Juárez-López et al. 2007; Martínez-Mier et al. 2005). Mean fluoride content for Mexico City’s water supply is not available because fluoride is not reported as part of water quality control programs in Mexico.

Mother–child pairs with at least one archived urine sample from pregnancy and measures of neurocognitive function in the offspring were included in this study. In terms of when the archived samples were collected, the pregnant mothers were invited for assessments with the collection of samples during trimester 1 (13.6±2.1 wk for Cohort 3 and 13.7±3.5 wk for Cohort 2A), trimester 2 (25.1±2.3 wk for Cohort 3 and 24.4±2.9 wk for Cohort 2A), and trimester 3 (33.9±2.2 wk for Cohort 3 and 35.0±1.8 wk for Cohort 2A).

A spot (second morning void) urine sample was targeted for collection during each trimester of pregnancy of ELEMENT mothers as well as the offspring children at the time of their measurements of intelligence at 6–12 y old. The samples were collected into fluoride-free containers and immediately frozen at the field site and shipped and stored at −20°C at the Harvard T.H. Chan School of Public Health (HSPH), and then at −80°C at the University of Michigan School of Public Health (UMSPH).

A procedure for urine analysis of fluoride described elsewhere (Martínez-Mier et al. 2011) was adapted and modified for this study. The fluoride content of the urine samples was measured using ion-selective electrode-based assays. First, 3 M sulfuric acid saturated with hexamethyldisiloxane (HMDS) was added to the sample to allow fluoride to diffuse from the urine for 20–24 hr. The diffused fluoride was allowed to collect in 0.05 M of sodium hydroxide on the interior of the petri dish cover. Once the diffusion was complete, 0.25 M of acetic acid was added to the sodium hydroxide to neutralize the solution and then analyzed directly using a fluoride ion-selective electrode (Thermo Scientific Orion, Cat#13-642-265) and pH/ISE meter (Thermo Scientific Orion, Cat#21-15-001). All electrode readings (in millivolts) were calculated from a standard curve. Analyses were performed in a Class 100/1,000 clean room. Quality control measures included daily instrument calibration, procedural blanks, replicate runs, and the use of certified reference materials (Institut National de Santé Publique du Québec, Cat #s 0910 and 1007; NIST3183, Fluoride Anion Standard). Urinary fluoride concentrations were measured at the UMSPH and the Indiana University Oral Health Research Institute (OHRI) as previously described (Thomas et al. 2016). A validation study comparing measures taken by the two labs in the same samples revealed a between-lab correlation of 0.92 (Thomas et al. 2016).

There were a total of 1,484 prenatal samples measured at the UMSPH lab. All of these samples were measured in duplicate. Of these, 305 (20%) of them did not meet the quality control criteria for ion-selective electrode-based methods (i.e., RSD<20% for samples with F level<0.2 ppm or RSD<10% when F level>0.2 ppm) (Martinez-Mier et al. 2011). Of these 305, 108 had a second aliquot available and were successfully measured at the OHRI lab in Indiana (sufficient urine volume was not available for the remaining 197 samples). The OHRI lab in Indiana also measured an additional 289 samples. Of the 397 total samples measured at the OHRI lab in Indiana, 139 (35%) were measured in duplicate, for which >95% complied with the quality control criteria above; thus, all 139 values were retained. The remaining 258 (65%) were not measured in duplicate because of limitations in available urine volume, but were included in the study given the excellent quality control at the OHRI lab. In total, we ended up with 1,576 prenatal urine samples with acceptable measures of fluoride.

Of these 1,576 urine samples, 887 also had data on urinary creatinine and were associated with mother–offspring pairs who had data on the covariates of interest and GCI or IQ in the offspring. The urinary creatinine data were used to correct for variations in urine dilution at the time of measurement (Baez et al. 2014). Creatinine-adjusted urinary fluoride concentrations were obtained for each maternally derived sample by dividing the fluoride concentration (MUF) in the sample by the sample’s creatinine concentration (MUC), and multiplying by the average creatinine concentration of samples available at each trimester (MUCaverage) using the formula: (MUF/MUC)×MUCaverage. The values of average creatinine concentration used for the MUCaverage at each trimester were derived from the larger pool of trimester-1, -2, and -3 samples from Cohorts 2A and 3 examined in our previous report on maternal fluoride biomarker levels (Thomas et al. 2016): 100.81, 81.60, and 72.41 (mg/L), respectively. For each woman, an average of all her available creatinine-adjusted urinary fluoride concentrations during pregnancy (maximum three samples and minimum one sample) was computed and used as the exposure measure (MUFcr). For children, as creatinine measurements were not available, urinary fluoride values (CUF) were corrected for specific gravity (SG) using the formula CUFsg=CUF(1.02−1)/(SG−1) (Usuda et al. 2007).

After calculating MUFcr for the 887 urine samples noted above, 10 values of MUFcr were identified as extreme outliers (>3.5 SDs) and were dropped, leaving 877 measures of MUFcr. These 877 measures of MUFcr stemmed from 512 unique mothers. Of these 512, 71 participants had measurements from each of the three trimesters; 224 had measurements from two of the three trimesters (74, T1 and T2; 131, T1 and T3; and 19, T2 and T3); and 217 had measurements from only one of the trimesters (159, T1; 34, T2; and 24, T3).

Measurement of Outcomes

At age 4 y, neurocognitive outcomes were measured using a standardized version of McCarthy Scales of Children’s Abilities (MSCA) translated into Spanish (McCarthy 1991). MSCA evaluates verbal, perceptual-performance, quantitative, memory, and motor abilities of preschool-aged children, and it has previously been successfully used in translated versions (Braun et al. 2012; Julvez et al. 2007; Kordas et al. 2011; Puertas et al. 2010). For this analysis, we focused on the General Cognitive Index (GCI), which is the standardized composite score produced by the MSCA (McCarthy 1991). For children 6–12 y old a Spanish-version of the Wechsler Abbreviated Scale of Intelligence (WASI) (Wechsler 1999) was administered. WASI includes four subtests (Vocabulary, Similarities, Block Design, and Matrix Reasoning), which provide estimates of Verbal, Performance, and Full-Scale IQ (Wechsler 1999). Both tests were administered by a team of three psychologists who were trained and supervised by an experienced developmental psychologist (L.S.). This team of three psychologists applied all of the McCarthy tests as well as the WASI-FSIQ tests. At the time of follow-up visits (age 4 and 6–12 y), each child was evaluated by one of the psychologists who was blind to the children’s fluoride exposure. The inter-examiner reliability of the psychologists was evaluated by having all three psychologists participate in assessments on a set of 30 individuals. For these 30, the inter-examiner reliability of the psychologists was evaluated by calculating the correlation in GCI scores by two of the psychologists with the scores of a third psychologist whom they observed applying the test in all three possible combinations with 10 participants for each observers–examiner pair (i.e., psychologist A (applicant) was observed by psychologist B and psychologist C; psychologist B (applicant) was observed by psychologist A and psychologist C; and psychologist C (applicant) was observed by psychologist A and psychologist B). The mean observer–examiner correlation was 0.99. All raw scores were standardized for age and sex (McCarthy 1991). Inter-examiner reliability was not examined on the WASI test.

Measurement of Covariates

Data were collected from each subject by questionnaire on maternal age (and date of birth), education, and marital status at the first pregnancy visit; on birth order, birth weight, and gestational age at delivery; and on maternal smoking at every prenatal and postnatal visit. Gestational age was estimated by registered nurses. Maternal IQ was estimated using selected subtests of the Wechsler Adult Intelligence Scale (WAIS)-Spanish (Information, Comprehension, Similarities, and Block Design), which was standardized for Mexican adults (Renteria et al. 2008; Wechsler et al. 1981). Maternal IQ was measured at the study visit 6 mo after birth or at the 12-mo visit if the earlier visit was not completed.

The quality of the children’s individual home environments was assessed using an age-appropriate version of the HOME score. However, the measure was not available for all observations because it was only added to on-going cohort evaluation protocols beginning in April 2003, when a version of the HOME score instrument that is age-appropriate for children 0–5 y old was adopted, following which a version of the HOME score instrument that is age-appropriate for children ≥6 y old was adopted in September 2009 (Caldwell and Bradley 2003). Thus, we adjusted for HOME score using the measures for 0- to 5-y-old children in the subset of children who had this data in our analyses of GCI, and we adjusted for HOME score using the measures for >6-y-old children in the subset of children who had this data in our analyses of IQ.

Statistical Analyses

Univariate distributions and descriptive statistics were obtained for all exposure variables, outcome variables, and model covariates. For each variable, observations were classified as outliers if they were outside the bounds of the mean±3.5 SDs. Primary analyses were conducted with exposure and outcome outliers excluded. Statistical tests of bivariate associations were conducted using chi-square tests for categorical variables and analysis of variance (ANOVA) to compare the means of the outcomes or exposure within groups defined according to the distribution of each covariate. Spearman correlation coefficients were used to measure the correlation between MUFcr and CUFsg. Regression models were used to assess the adjusted associations between prenatal fluoride and each neurocognitive outcome separately. Generalized additive models (GAMs) were used to visualize the adjusted association between fluoride exposure and measures of intelligence [SAS statistical software (version 9.4; SAS Institute Inc.)]. Because the pattern appeared curvilinear, and because GAMs do not yield exact p-values for deviations from linearity, we used a Wald p-value of a quadratic term of fluoride exposure to test the null hypothesis that a quadratic model fit the data better than the model assuming a linear relationship, and thus obtained a p-value for deviation from linearity of the fluoride–outcome associations. Residual diagnostics were used to examine other model assumptions and identify any additional potentially influential observations. Visual inspection of default studentized residual versus leverage plot from SAS PROC REG did not identify potential influential observations. Visual inspection of the histogram of the residuals did not indicate lack of normality; however, a fanning pattern in the residual versus predicted value plot indicated lack of constant variance (data not shown). Hence, robust standard errors were obtained using the “empirical” option in SAS PROC GENMOD.

Our overall strategy for selecting covariates for adjustment was to identify those that are well known to have potential associations with either fluoride exposure or cognitive outcomes and/or are typically adjusted for as potential confounders in analyses of environmental toxicants and cognition. All models were adjusted for gestational age at birth (in weeks), birthweight (kilograms), birth order (first born yes vs. no), sex, and child’s age at the time of the neurocognitive test (in years). All models were also adjusted for maternal characteristics including marital status (married vs. others), smoking history (ever-smoker vs. never-smoker), age at delivery, IQ, and education (itself also a proxy for socioeconomic status). Finally, all models adjusted for potential cohort effects by including indicator variables denoting from which cohort (Cohort 2A, Cohort 3+Ca supplement, and Cohort 3 -placebo) the participants came. We used 0.5 mg/L, which was close to the interquartile range of MUFcr for the analyses of both GCI (IQR=0.45) and IQ (IQR=0.48), as a standard measure of incremental exposure. SAS statistical software (version 9.4; SAS Institute Inc.) was used for all data analyses described.

Sensitivity Analyses

Models were further adjusted for variables that relate to relatively well-known potential confounders (but for which we were missing a significant amount of data) and variables that were less-well known but possible confounders. The HOME scores were subject to sensitivity analyses because, as noted in the “Methods” section, they were not added to the subject evaluation protocols until 2003, resulting in a significantly smaller subsample of participants with this data. Models of the association between prenatal fluoride exposure (MUFcr) and IQ at 6–12 y old were also adjusted for the child’s urine fluoride concentration at 6–12 y of age (CUFsg), a measure that was collected in a significantly smaller subset of individuals, to evaluate the potential role of contemporaneous exposure. Associations between prenatal fluoride exposure (MUFcr) and GCI at 4 y old could not be adjusted for contemporaneous fluoride exposure because urine samples were not collected from children when the MSCA (from which the GCI is derived) was administered. Maternal bone lead measured by a 109-Cd K-X-ray fluorescence (KXRF) instrument at 1 mo postpartum, a proxy for lead exposure from mobilized maternal bone lead stores during pregnancy (Hu et al. 2006), was included in the model to test for the possible confounding effect of lead exposure during pregnancy. We focused on the subset of women who had patella bone lead values because these were found to be most influential on our previous prospective study of offspring cognition (Gomaa et al. 2002). Average maternal mercury level during pregnancy was also tested for being a potential confounder (Grandjean and Herz 2011). Mercury was measured as total mercury content in the subsample of women who had samples of archived whole blood samples taken during pregnancy with sufficient volume to be analyzed using a Direct Mercury Analyzer 80 (DMA-80, Milestone Inc., Shelton, CT, USA) as previously described (Basu et al. 2014).

To address the potential confounding effect of socioeconomic status (SES) we conducted sensitivity analyses that adjusted our model for SES (family possession score). The socioeconomic questionnaire asked about the availability of certain items and assets in the home. Point values were assigned to each item, and SES was calculated based on the sum of the points across all items (Huang et al. 2016). Given that the calcium intervention theoretically could have modified the impact of fluoride, in examining our results, we repeated the analyses with and without the Cohort 3 participants who were randomized to the calcium intervention to omit any potential confounding effect of this intervention. Another sensitivity test was performed to examine the potential effect of the psychologist who performed the WASI test by including tester in the regression model. The information about psychologists who performed the WASI was available for 75% of participants, as recording this data was added later to the study protocol. We also re-ran models with exposure outliers included as a sensitivity step. Finally, we ran models that focused on the cross-sectional relationship between children’s exposure to fluoride (reflected by CUFsg) and IQ score, unadjusted; adjusting for the main covariates of interest; and adjusting for prenatal exposure (MUFcr) as well as the covariates of interest.

Results

Flow of Participants

Of the 997 total mothers from two cohorts evaluated, 971 were eligible after removing mothers <18 y old. Of these 971, 825 had enough urine sample volume to measure fluoride in at least one trimester urine sample, and of these 825 participants, 515 participants had urine samples with previously measured creatinine values, enabling calculation of creatinine-adjusted urinary fluoride (MUFcr) concentrations. Of these 515, 3 participants were excluded based on the 10 extreme outlier values identified for MUFcr (see the “Methods” section, “Exposure Assessment” subsection) and not having any other MUFcr values to remain in the analysis. Thus, we had a total of 512 participants (mothers) with at least one value of MUFcr for our analyses (Figure 1).

Flowchart.
Figure 1. Flowchart describing source of mother–offspring subject pairs, fluoride and cognition study. Cohort 2A was designed as an observational birth cohort of lead toxicodynamics during pregnancy, with mothers recruited early during pregnancy from 1997 to 2001. Cohort 3 was designed as a randomized double-blind placebo-controlled trial of calcium supplements, with mothers recruited early during pregnancy from 2001 to 2006. “Ca” denotes subjects who were randomized to the calcium supplement; “placebo” denotes subjects who were randomized to the placebo. GCI is the McCarthy Scales General Cognitive Index (administered at age 4 y). IQ is the Wechsler Abbreviated Intelligence Scales Intelligence Quotient (administered at age 6–12 y and age-adjusted).

Of these 512 mothers, 312 had offspring with outcome data at age 4 (i.e., GCI), and 234 had offspring with outcome data at age 6–12 (i.e., IQ). Of these, complete data on all the covariates of main interest (as specified in the “Methods” section) were available on 287 mother–child pairs for the GCI analysis and 211 mother–child pairs for the IQ analysis. A total of 299 mother–child pairs had data on either GCI or IQ, and 199 mother–child pairs had data on both GCI and IQ (Figure 1).

Number of Exposure Measures per Subject

In terms of repeated measures of MUFcr across trimesters, of the 287 participants with data on GCI outcomes; 25 participants had MUFcr data for all three trimesters (11 from Cohort 2A and 14 from Cohort 3), 121 participants had MUFcr data from two trimesters (48 from Cohort 2A and 73 from Cohort 3), and 141 participants had MUFcr data from one trimester (51 from Cohort 2A and 90 from Cohort 3). Of the 211 participants with data on IQ outcomes, 10 participants had MUFcr data for all three trimesters (6 from Cohort 2A and 4 from Cohort 3), 82 participants had data from two trimesters (32 from Cohort 2A and 50 from Cohort 3), and 119 participants had data from one trimester (40 from Cohort 2A and 79 from Cohort 3).

Comparisons across the Cohorts

In terms of the mother–child pairs who had data on all covariates as well as data on either GCI or IQ (n=299), the mean (SD) values of creatinine–corrected urinary fluoride for the mothers was 0.90 (0.36) mg/L. The distributions of the urinary fluoride, outcomes (GCI and IQ), and additional exposure variables examined in our sensitivity analyses (maternal bone lead, maternal blood mercury, and children’s contemporaneous urinary fluoride) across the three cohort strata (Cohort 3-Calcium, Cohort 3-placebo, and Cohort 2A) and all strata combined are shown in Table 1 for the mother–child pairs who had data for the GCI outcome (n=287) and the IQ outcome (n=211). The distributions showed little variation across the cohort strata except for bone lead and possibly blood mercury, for which, in comparison with Cohort 3, Cohort 2A clearly had higher mean bone lead levels (p<0.001) and possibly higher blood mercury levels (p=0.067). The mean (SD) values of specific gravity–corrected urinary fluoride for the children who had these measures (only available for those children who had IQ; n=189) were 0.82 (0.38) mg/L.

Table 1. Comparisons across cohorts with respect to the distributions of biomarkers of exposure to prenatal fluoride (MUFcr), prenatal lead (maternal bone Pb), prenatal mercury (maternal blood Hg), and contemporaneous childhood fluoride (CUFsg); and cognitive outcomes (GCI and IQ).
Analysis Measurement Cohort N Mean SD Min Percentiles Max p-valuea
25 50 75
GCI Analysis GCI Cohort 3-Ca 84 96.88 14.07 50 88 96 107 124 0.997
Cohort 3-placebo 93 96.80 13.14 50 89 96 105 125
Cohort 2A 110 96.95 15.46 56 88 98 110 125
Totalb 287 96.88 14.28 50 88 96 107 125
MUFcr (mg/L) Cohort 3-Ca 84 0.92 0.41 0.28 0.60 0.84 1.14 2.36 0.57
Cohort 3-placebo 93 0.87 0.34 0.23 0.62 0.82 1.10 2.01
Cohort 2A 110 0.92 0.33 0.23 0.68 0.86 1.11 2.14
Totalb 287 0.90 0.36 0.23 0.65 0.84 1.11 2.36
Maternal bone Pb (μg/g) Cohort 3-Ca 62 7.30 7.37 0.05 0.75 4.40 12.93 26.22 <0.01
Cohort 3-placebo 43 9.21 7.31 0.11 1.50 8.60 13.97 27.37
Cohort 2A 62 13.60 11.36 0.15 5.35 10.52 19.46 47.07
Totalc 167 10.13 9.41 0.05 2.37 8.22 15.37 47.07
Maternal blood Hg (μg/L) Cohort 3-Ca 38 3.32 1.40 0.73 2.40 3.00 4.15 7.06 0.12
Cohort 3-placebo 28 2.80 1.33 1.27 1.89 2.53 3.40 7.22
Cohort 2A 75 4.53 5.61 0.77 2.30 3.24 4.37 35.91
Totalc 141 3.86 4.25 0.73 2.20 3.08 4.15 35.91
IQ Analysis IQ Cohort 3-Ca 58 94.91 9.86 76 87 96 100 120 0.69
Cohort 3-placebo 75 96.29 9.63 75 89 97 102 124
Cohort 2A 78 96.47 13.20 67 87 96 107 131
Totald 211 95.98 11.11 67 88 96 107 131
MUFcr (mg/L) Cohort 3-Ca 58 0.89 0.38 0.29 0.57 0.84 1.10 1.85 0.86
Cohort 3-placebo 75 0.87 0.35 0.23 0.61 0.82 1.11 2.01
Cohort 2A 78 0.90 0.34 0.23 0.67 0.85 1.09 2.14
Totald 211 0.89 0.36 0.23 0.64 0.82 1.07 2.14
Maternal bone Pb (μg/g) Cohort 3-Ca 67 6.97 7.20 0.05 0.76 4.36 11.73 26.22 <0.01
Cohort 3-placebo 48 9.07 7.42 0.11 1.00 8.49 14.41 27.37
Cohort 2A 62 13.60 11.36 0.15 5.35 10.52 19.46 47.07
Totale 177 9.86 9.33 0.05 2.29 7.95 15.22 47.07
Maternal blood Hg (μg/L) Cohort 3-Ca 43 3.25 1.41 0.51 2.43 2.87 4.02 7.06 0.067
Cohort 3-placebo 31 2.66 1.36 0.78 1.81 2.40 3.26 7.22
Cohort 2A 75 4.53 5.61 0.77 2.30 3.24 4.37 35.91
Totale 149 3.77 4.16 0.51 2.19 2.90 4.11 35.91
CUFsg (mg/L) Cohort 3-Ca 71 0.84 0.4 0.31 0.53 0.78 1.12 2.8 0.29
Cohort 3-placebo 53 0.85 0.38 0.35 0.57 0.75 1.14 1.85
Cohort 2A 65 0.76 0.34 0.18 0.51 0.7 0.89 1.76
Totale 189 0.82 0.38 0.18 0.54 0.73 1.01 2.8
All available measurements GCI Cohort 3-Ca 133 97.32 13.67 50 88 96 107 124 0.57
Cohort 3-placebo 149 95.99 13.07 50 88 96 106 125
Cohort 2A 150 97.57 14.63 56 88 99 109 131
Totalf 432 96.95 13.80 50 88 96 107 131
IQ Cohort 3-Ca 91 95.92 10.15 76 88 95 103 120 0.92
Cohort 3-placebo 114 96.56 9.84 75 89 96 102 124
Cohort 2A 111 96.25 12.67 67 87 95 105 131
Totalf 316 96.27 10.97 67 88 96 103 131
MUFcr (mg/L) Cohort 3-Ca 181 0.89 0.36 0.28 0.64 0.83 1.09 2.36 0.11
Cohort 3-placebo 183 0.84 0.31 0.02 0.61 0.81 1.02 2.01
Cohort 2A 148 0.91 0.35 0.23 0.67 0.86 1.10 2.15
Totalf 512 0.88 0.34 0.02 0.64 0.82 1.07 2.36
Maternal bone Pb (μg/g) Cohort 3-Ca 97 7.07 7.26 0.01 0.83 4.36 11.78 26.22 <0.01
Cohort 3-placebo 74 9.15 8.38 0.11 0.85 8.62 13.41 40.8
Cohort 2A 86 13.77 11.30 0.15 5.49 10.52 20.58 47.07
Totalf 257 9.91 9.51 0.01 2.01 7.64 15.31 47.07
Maternal blood Hg (μg/L) Cohort 3-Ca 55 3.03 1.41 0.51 2.12 2.77 3.62 7.06 0.09
Cohort 3-placebo 48 2.87 2.09 0.34 1.82 2.37 3.34 13.47
Cohort 2A 104 4.06 4.88 0.77 2.14 3.10 4.16 35.91
Totalf 207 3.51 3.70 0.34 2.07 2.80 3.79 35.91
CUFsg (mg/L) Cohort 3-Ca 104 0.84 0.39 0.31 0.56 0.75 1.07 2.80 0.227
Cohort 3-placebo 84 0.90 0.46 0.35 0.58 0.75 1.09 2.89
Cohort 2A 96 0.79 0.34 0.18 0.53 0.73 0.92 2.11
Totalf 284 0.84 0.40 0.18 0.57 0.74 1.00 2.89

aAnalysis of variance across cohorts.

bTotal number of subjects included in GCI main analysis.

cTotal number of subjects included in GCI sensitivity analysis.

dTotal number of subjects included in IQ main analysis.

eTotal number of subjects included in IQ sensitivity analysis.

fTotal number of subjects with available measurements, combining Cohort 2A and Cohort 3.

In terms of the comparability of the participants across Cohort 2A and Cohort 3 with respect to our covariates, the distribution of the covariates was very similar with the exception of age of the offspring when IQ was measured, for which the mean ages were 7.6 and 10.0 y, respectively; and birth weight in the GCI analysis, for which Cohort 3 participants were slightly heavier than Cohort 2 participants (see Table S1).

GCI versus IQ Scores

There was a significant correlation between GCI at 4 y and IQ at 6–12 y old (Spearman r=0.55; p<0.01). There was no significant correlation between prenatal MUFcr and offspring CUFsg (Spearman r=0.54, p=0.44).

Comparisons of Participants in Relation to Missing Data

In comparing the participants who were included for the GCI and IQ analyses with the participants who were not included (based on data missing on GCI, IQ or other covariates), the distribution of covariates were similar except for sex, for which the proportion of females was somewhat higher in the included versus excluded group for both the GCI and IQ analyses (Table 2).

Table 2. Analysis comparing subjects with and without data of interest [n (%) or mean±SD] with respect to characteristics of mothers and children and sensitivity analysis covariates.
Characteristic GCI analysis IQ analysis
Included Excluded Included Excluded
Total numbera 287 710 211 786
Sex
 Female 160 (56%) 244 (47%) 116 (55%) 288 (48%)
 Male 127 (44%) 275 (53%) 95 (45%) 307 (52%)
Birth order
 First child 96 (33%) 184 (35%) 93 (32%) 279 (36%)
 ≥2nd child 191 (67%) 335 (65%) 118 (68%) 507 (65%)
 Birth weight (kg) 3.11±0.45 3.11±0.44 3.11±0.46 3.11±0.43
 Gestational age (wk) 38.66±1.84 38.58±1.68 38.56±1.80 38.63±1.72
 Age at outcome assessment (y) 4.04±0.05 4.05±0.05 8.50±1.31 8.83±1.64
 Maternal age at delivery (y) 26.78±5.53 26.49±5.37 27.16±5.61 26.41±5.36
 Maternal education (y)b 10.63±2.76 10.75±3.08 10.80±2.85 10.69±3.03
 Maternal IQc 88.63±12.17 89.27±14.6 89.01±12.45 88.27±13.00
 Marital statusd 3.11±0.45 3.11±0.44 3.11±0.46 3.11±0.43
 Married 201 (70%) 493 (70%) 149 (71%) 544 (69%)
 Other 86 (30%) 216 (30%) 62 (29%) 240 (31%)
Maternal smokinge
 Ever 141 (49%) 335 (51%) 102 (48%) 374 (51%)
 Never 146 (51%) 325 (49%) 109 (52%) 362 (49%)
Cohort
 Cohort 3-Ca 93 (32%) 241 (34%) 76 (36%) 259 (33%)
 Cohort 3-placebo 84 (29%) 252 (36%) 59 (28%) 278 (35%)
 Cohort 2A 110 (38%) 217 (31%) 78 (37%) 249 (32%)
Sensitivity Analyses
 HOME score f N=138 N=87 N=124 N=55
35.24±6.31 33.23±6.55 35.54±7.46 35.8±7.44
 SESg N=188 N=110 N=199 N=98
6.35±2.43 6.94±2.72 6.36±2.41 6.98±2.79
 Maternal Bone Pb (μg/g)h N=167 N=91 N=177 N=80
9.26±10.55 8.97±10.32 9.02±10.43 9.48±10.55
 Maternal Blood Hg (μg/L)i N=141 N=67 N=149 N=58
3.86±4.25 2.76±1.95 3.77±4.16 2.83±2.01
 CUFsgj (mg/L) N=124 N=55
35.54±7.46 35.8±7.44

aThe total number of subjects (n=997) are all mother–offspring pairs who participated in the original Cohort 2A and Cohort 3 studies.

bMaternal education at the time of the child’s birth.

cMaternal IQ measured at 6 mo after child’s birth.

dMother’s marital status at the time of the child’s birth.

eHistory of any maternal smoking.

fHOME score measured using the separate age-appropriate instruments pertaining to children of ≤5 y old; and children >5 y old.

gFamily socioeconomic status (SES) measured by questionnaire of family possessions at follow-up.

hMaternal patella bone lead measured by KXRF after birth.

iMaternal average blood mercury during pregnancy.

jChildren’s specific gravity–corrected urinary fluoride measured at the time of each child’s IQ test (6–12 y old).

N† Number of subjects with measurements of MUFcr, cognitive outcome, main covariates, and sensitivity covariates (they are included in the sensitivity model).

N‡ Number of subjects with measurements of sensitivity covariates, but missing data on exposure, outcomes, or main covariates (they are excluded from the sensitivity model).

In terms of the sensitivity analyses, for each sensitivity variable of interest, we compared participants who had data on our exposures, outcomes, covariates, and the sensitivity variable of interest (and were thus included in the sensitivity analysis) versus participants who had data on the sensitivity variable of interest but were missing data on the exposure, outcomes, and/or covariates of interest (and were thus excluded from the sensitivity analysis; Table 2). It can be seen that for each sensitivity analysis, most of the participants with data on the sensitivity variable of interest also had data on the exposures, outcomes, and covariates and were therefore included in the sensitivity analysis. In addition, the distributions appeared to be similar comparing those included with those excluded in each sensitivity analysis (means were within 10% of each other), with the exception of maternal blood Hg, for which the mean levels for those included were 28.5% and 24.9% higher than the mean levels for those excluded in the GCI and IQ analyses, respectively.

Comparisons of GCI and IQ across Covariates

Table 3 shows mean and SD values for MUFcr and offspring cognitive scores across categories of the covariates. In the participants with GCI data, the offspring cognitive scores were higher among mothers with higher levels of education, measured IQ, and HOME scores for both analyses; and scores were higher among first children and girls. In the IQ analysis a statistically significant difference was observed in MUFcr as a function of child sex. No significant differences in MUFcr values across levels of other covariates were observed. A modest difference (not statistically significant), was observed in MUFcr as a function of maternal IQ (p=0.09), and MUFcr as a function of child sex (p=0.09). Among other co-variates there were significant differences in age (p<0.01) in both analyses.

Table 3. Distributions of maternal creatinine-adjusted urinary fluoride (MUFcr) and offspring cognitive scores across categories of main covariates.
Covariate GCI Analysis IQ Analysis
n MUFcra p-value GCI (Age 4) p-value n MUFcra p-value IQ (Age 6–12) p-value
Mothers
Age
 ≥25 y 123 0.88±0.36 0.45 96.22±14.12 0.50 88 0.89±0.37 0.98 95.75±11.64 0.80
 <25 y 164 0.92±0.36 97.37±14.43 123 0.89±0.35 96.15±10.76
Education
 <12 y 153 0.91±0.4 0.92 94.22±14.23 0.001 111 0.87±0.37 0.53 93.09±10.54 <0.001
 12 y 97 0.89±0.34 98.56±14.46 70 0.93±0.35 98.29±10.72
 >12 y 37 0.89±0.42 103.49±11.21 30 0.85±0.31 101.3±11.16
Marital status
 Married 201 0.90±0.37 0.81 96.40±14.46 0.39 62 0.90±0.35 0.79 96.55±11.06 0.63
 Other 86 0.91±0.33 98.00±13.88 149 0.88±0.36 95.74±11.16
Smoking
 Ever smoker 141 0.90±0.36 0.80 97.77±13.9 0.30 102 0.90±0.36 0.56 97.21±10.7 0.12
 Nonsmoker 146 0.91±0.35 96.01±14.63 109 0.87±0.35 94.83±11.41
HOME score b
 Mid-low≤30 49 0.88±0.37 0.47 90.73±13.36 <0.001 32 0.87±0.36 0.85 89.88±8.45 0.011
 High>30 137 0.92±0.38 99.29±14.61 92 0.88±0.38 99.05±11.65
Maternal IQ
 Mid-low≤85 116 0.95±0.35 0.09 93.16±15.04 <0.001 86 0.92±0.36 0.23 91.26±9.72 <0.001
 High>85 171 0.87±0.36 99.4±13.21 125 0.86±0.35 99.23±10.87
Children
Sex
 Boy 127 0.94±0.36 0.09 93.93±13.98 0.002 95 0.96±0.38 0.008 96.82±12.02 0.32
 Girl 160 0.87±0.36 99.22±14.12 116 0.83±0.32 95.29±10.31
Birthweight
 ≥3.5 kg 241 0.91±0.36 0.57 96.52±14.36 0.33 201 0.89±0.36 0.88 95.66±11.29 0.58
 <3.5 kg 46 0.87±0.35 98.76±13.88 10 0.88±0.34 97.38±9.42
Gestational age
 ≤39 wk 192 0.90±0.35 0.90 96.66±14.23 716 146 0.89±0.36 0.712 95.71±11.62 0.65
 >39 wk 95 0.90±0.37 97.32±14.46 65 0.88±0.34 96.58±9.91
First child
 Yes 96 0.91±0.38 0.75 99.97±12.87 0.009 68 0.88±0.36 0.91 97.00±11.00 0.36
 No 191 0.90±0.35 95.32±14.73 143 0.89±0.36 95.50±11.17
CUFsgc
 ≥0.80 mg/L 112 0.86±0.32 0.49 96.80±11.16 0.37
 <0.80 mg/L 77 0.90±0.38 95.37±10.31

aMaternal creatinine-adjusted urinary fluoride (mg/L).

bHome Observation for the Measurement of the Environment (HOME) score, measured using the separate age-appropriate instruments pertaining to children of ≤5 y old; and children >5 y old.

cChild contemporaneous specific gravity–adjusted urinary fluoride (available at the time of each child’s IQ test).

Regression Models of GCI

Before adjustment, a 0.5 mg/L increase in MUFcr was negatively associated with GCI at 4 y old [mean score −3.76; 95% confidence interval (CI): −6.32, −1.19] (Table 4). The association was somewhat attenuated after adjusting for the main covariates (model A, −3.15; 95% CI: −5.42, −0.87). The smooth plot of the association between GCI and maternal prenatal urinary fluoride from an adjusted GAM model suggested a linear relation over the exposure distribution (Figure 2).

Scatter plot with a regression line indicating McCarthy GCI at 48 months (y-axis) across concentration of prenatal M U F subscript cr in milligrams per liter (x-axis).
Figure 2. Adjusted association of maternal creatinine-adjusted urinary fluoride (MUFcr) and General Cognitive Index (GCI) scores in children at age 4 y. Adjusted for gestational age, weight at birth, sex, parity (being the first child), age at outcome measurement, and maternal characteristics including smoking history (ever smoked vs. nonsmoker), marital status (married vs. others), age at delivery, IQ, education, and cohort (Cohort 3-Ca, Cohort 3-placebo and Cohort 2A). Shaded area is 95% confidence interval. Short vertical bars on the x-axis reflect the density of the urinary fluoride measures. Individual data points are individual observations, n=287.
Table 4. Multivariate regression models: unadjusted and adjusted differences in GCI and IQ per 0.5 mg/L higher maternal creatinine-adjusted urinary fluoride (MUFcr).
Estimate GCI IQ
n β (95%CI) p-value n β±S.E (95%CI) p-value
Unadjusted 287 −3.76 (−6.32, −1.19) <0.01 211 −2.37 (−4.45, −0.29) 0.03
model Aa 287 −3.15 (−5.42, −0.87) 0.01 211 −2.50 (−4.12, −0.59) 0.01
Model A–HOME 138 −3.63 (−6.48, −0.78) <0.01 124 −2.36 (−4.48, −0.24) 0.03
Model A+HOME 138 −3.76 (−7.08, −0.45) 0.03 124 −2.49 (−4.65, −0.33) 0.02
Model A−CUFsg 189 −1.79 (−3.80, 0.22) 0.08
Model A+CUFsg 189 −1.73 (−3.75, 0.29) 0.09
Model A−SES 188 −4.55 (−7.23, −1.88) 0.01 199 −2.10 (−4.02, −0.18) 0.03
Model A+SES 188 −4.45 (−7.08, −1.81) 0.01 199 −2.10 (−4.06, −0.15) 0.04
Model A–Pb 167 −5.57 (−8.48, −2.66) <0.01 177 −3.21 (−5.17, −1.24) <0.01
Model A+Pb 167 −5.63 (−8.53, −2.72) <0.01 177 −3.22 (−5.18, −1.25) <0.01
Model A−Hg 141 −7.13 (−10.26, −4.01) <0.01 149 −4.59 (−7.00, −2.17) <0.01
Model A+Hg 141 −7.03 (−10.19, −3.88) <0.01 149 −4.58 (−6.99, −2.16) <0.01
Model A−Ca 194 −3.67 (−6.57, −0.77) 0.01 136 −3.23 (−5.88, −0.57) 0.02

aCoefficients from linear regression models adjusted for gestational age, weight at birth, sex, parity (being the first child), age at outcome measurement, and maternal characteristics including smoking history (ever smoked during the pregnancy vs. nonsmoker), marital status (married vs. others), age at delivery, IQ, education, and cohort (Cohort 3-Ca, Cohort 3-placebo and Cohort 2A). Model A–HOME, model A for subset of cases who have data on Home Observation for the Measurement of the Environment (HOME) scores (but the model did not include HOME score). Model A+HOME, model A for subset of cases with HOME score, adjusted for HOME score. Model A−CUFsg, model A for subset of cases who have data on child contemporaneous specific gravity–adjusted urinary fluoride CUFsg (but the model did not include CUFsg). Model A+CUFsg, model A for subset of cases with CUFsg, adjusted for CUFsg. Model A−SES, model A for subset of cases who have data on socioeconomic status (family possession measured by questionnaire of family possessions) (but the model did not include SES). Model A+SES, model A for subset of cases with SES data, adjusted for SES. Model A–Pb, model A for subset of cases who have data on maternal bone lead (but the model did not include maternal bone lead). Model A+Pb, model A for subset of cases with data on maternal bone lead, adjusted for maternal bone lead. Model A−Hg, model A for subset of cases who have data on maternal blood mercury (but the model did not include maternal blood mercury). Model A+Hg, model A for subset of cases who have data on maternal blood mercury, adjusted for maternal blood mercury. Model A−Ca, model A for subset of cases who did not receive the Ca supplement (they received the placebo).

Regression Models of IQ

A 0.5 mg/L increase in prenatal fluoride was also negatively associated with IQ at age 6–12 y based on both unadjusted (−2.37; 95% CI: −4.45, −0.29) and adjusted models (−2.50; 95% CI: −4.12, −0.59) (Table 4). However, estimates from the adjusted GAM model suggest a nonlinear relation, with no clear association between IQ scores and values below approximately 0.8 mg/L, and a negative association above this value (Figure 3A). There was a nonsignificant improvement in the fit of the model when a quadratic term was added to the linear model (p=0.10).

Scatter plot with a regression line indicating WASI (IQ) at 6 to 12 years old (y-axis) across concentration of prenatal M U F subscript cr in milligrams per liter (x-axis). Scatter plot with a regression line indicating WASI (IQ) at 6 to 12 years old (y-axis) across concentration of prenatal M U F subscript cr in milligrams per liter (x-axis).
Figure 3. (A) Adjusted association of maternal creatinine-adjusted urinary fluoride (MUFcr) and children’s IQ at age 6–12 y. Adjusted for gestational age, weight at birth, sex, parity (being the first child), age at outcome measurement, and maternal characteristics including smoking history (ever smoked vs. nonsmoker), marital status (married vs. others), age at delivery, IQ, education, and cohort (Cohort 3-Ca, Cohort 3-placebo and Cohort 2A). Short vertical bars on the x-axis reflect the density of the urinary fluoride measures. Individual data points are individual observation, n=211. (B) Association of maternal creatinine-adjusted urinary fluoride (MUFUcr) and children’s IQ at age 6–12 y, adjusted for specific gravity–adjusted child urinary fluoride (CUFsg). Adjusted for gestational age, weight at birth, sex, parity (being the first child), age and CUFsg at outcome measurement, and maternal characteristics including smoking history (ever smoked vs. nonsmoker), marital status (married vs. others), age at delivery, IQ, education. and cohort (Cohort 3-Ca, Cohort 3-placebo and Cohort 2A). Shaded area is 95% confidence interval. Short vertical bars on the x-axis reflect the density of the urinary fluoride measures. Individual data points are individual observation, n=189.

Sensitivity Analyses

In sensitivity analyses, adjustment for HOME score increased the magnitude of the association between MUFcr and GCI, though the difference was less pronounced when associations with and without adjustment for HOME score were both estimated after restricting the model to the subset of 138 children with HOME score data (Table 4). The association of IQ scores with MUFcr did not substantially change after adding HOME score to the model (Table 4).

The association between MUFcr and IQ was attenuated slightly after adjusting for contemporaneous children’s urinary fluoride (CUFsg) and comparing estimates with [−1.73 (95% CI: −3.75, 0.29)] and without [−1.94 (95% CI: −4.15, 0.26)] adjustment for CUFsg among the 189 children with this data (Table 4). In addition, the evidence of nonlinearity was more pronounced, with no clear evidence of an association for MUFcr <1.0 mg/L based on the GAM model (Figure 3B), and a significant improvement in model fit when a quadratic term was added to the linear regression model (p=0.01).

When we restricted models to subsets of children with available data for each additional covariate, there was little difference between adjusted and unadjusted associations between MUFcr and GCI or IQ when socioeconomic status (family possession), maternal bone lead, and blood mercury, were added to models (Table 4). However, the effect estimates associated with MUFcr for these analyses appear to be higher in the subsets with available data for these variables.

Adding tester (psychologist who performed WASI) in the model did not substantially change the results (data not shown). In the sensitivity analyses in which we excluded Cohort 3 participants who received the calcium supplement, we continued to observe a negative association between MUFcr and GCI [0.5 mg/L increase in MUFcr associated with 3.67 lower GCI (95% CI: −6.57, −0.77), n=194]; and between MUFcr and IQ [0.5 mg/L increase in MUFcr associated with 3.23-lower IQ (95% CI: −5.88, −0.57), n=136].

In sensitivity analyses in which we re-ran models that included the 10 outliers with respect to fluoride exposure (for each of seven participants already in our models, an additional value of MUFcr [from a different trimester]; for three participants, a value of MUFcr that then allowed the participants to be added to our models), the results did not change in any meaningful way (data not shown). There were no outliers with respect to cognitive outcomes.

Independent Influence of Child Fluoride Exposure

Finally, in models that focused on the cross-sectional relationship between children’s exposure to fluoride (reflected by their specific gravity–adjusted urinary fluoride levels) and IQ score and that contained the main covariates of interest, there was not a clear, statistically significant association between contemporaneous children’s urinary fluoride (CUFsg) and IQ either unadjusted or adjusting for MUFcr. A 0.5 mg/L increase in CUFsg was associated with a 0.89 lower IQ (95% CI: −2.63, 0.85) when not adjusting for MUFcr; and 0.77-lower IQ (95% CI: −2.53, 0.99), adjusting for MUFcr (n=189).

Discussion

In our study population of Mexican women and children, which accounted for two of the three cohorts included in the ELEMENT study, higher prenatal exposure to fluoride (as indicated by average creatinine-adjusted maternal urinary fluoride concentrations during pregnancy) was associated with lower GCI scores in children at approximately 4 y old, and with lower Full-Scale IQ scores at 6–12 y old. Estimates from adjusted linear regression models suggest that mean GCI and IQ scores were about 3 and 2.5 points lower in association with a 0.5 mg/L increase in prenatal exposure, respectively. The associations with GCI appeared to be linear across the range of prenatal exposures, but there was some evidence that associations with IQ may have been limited to exposures above 0.8 mg/L. In general, the negative associations persisted in sensitivity analyses with further adjustment for other potential confounders, though the results of sensitivity analyses were based on subsets of the population with available data.

Overall, our results are somewhat consistent with the ecological studies suggesting children who live in areas with high fluoride exposure (ranging from 0.88 to 11.0 mg/L fluoride in water, when reported) have lower IQ scores than those who live in low-exposure or control areas (ranging from 0.20 to 1.0 mg/L fluoride in water) (Choi et al. 2012) and with results of a pilot study of 51 children (mean age 7 y) from southern Sichuan, China, that reported that children with moderate or severe dental fluorosis (60% of the study population) had lower WISC-IV digit span scores than other children (Choi et al. 2015). A distinction is that our study, which was longitudinal with repeated measures of exposure beginning in the prenatal period, found associations with respect to prenatal fluoride exposures.

To our knowledge, the only other study that is similar to ours was only recently published. Valdez Jiménez et al. (2017) studied the association of prenatal maternal urinary fluoride levels (not corrected for dilution) and scores on the Bayley Scales of Infant Development II among 65 children evaluated at age 3–15 mo (average of 8 mo). The mothers in their study had urinary fluoride levels of which the means at each of the three trimesters of pregnancy (1.9, 2.0, 2.7 mg/L) were higher than the mean MUFcr in our participants (0.88 mg/L) (Valdez Jiménez et al. 2017). These levels of exposure were found to be associated with statistically significantly lower scores on the Bayley Scales’ Mental Development Index (MDI) score after adjusting for gestational age, age of child, a marginality index, and type of drinking water (Valdez Jiménez et al. 2017). By comparison, our study had much longer periods of follow-up and larger sample sizes, controlled for a much larger set of covariates and sensitivity variables, and used creatinine–corrected urinary fluoride measures (which, by adjusting for urinary dilution effects, provides a more reliable measure of internal fluoride exposure).

With respect to understanding the generalizability of our findings to other populations, there are very few studies that measured prenatal fluoride levels among women derived from population-based samples. Gedalia et al. (1959) measured urinary fluoride in multiple samples collected from each of 117 healthy pregnant women living in Jerusalem, where fluoride in the water was approximate 0.50 mg/L, and reported mean levels per person that ranged from 0.29 to 0.53 mg/L. However, these analysis were not conducted utilizing modern analytical techniques. In a study of 31 pregnant women living in Poland, Opydo-Szymaczek and Borysewicz-Lewicka (2005) measured urinary fluoride in healthy pregnant women patients of a maternity hospital in Poland, where fluoride in the water ranged from 0.4 to 0.8 mg/L, and found a mean level of 0.65 mg/L for women in their 28th week of pregnancy, 0.84 mg/L in their 33rd week, and 1.30 mg/L in healthy non-pregnant women of similar age. This would suggest that the mothers in our study, who had a mean MUFcr value of 0.90 mg/L, had fluoride exposures slightly higher than prior-mentioned populations.

In terms of comparing our findings with other studies of fluoride (using urinary fluoride as a biomarkers of exposure) and intelligence (i.e., those not involving prenatal exposures), of the 27 epidemiologic studies on fluoride and IQ reviewed by Choi et al. in their 2012 meta-analysis, only 2 had measures of urinary fluoride. Both were of urinary fluoride measures in children (not pregnant mothers), and neither corrected for dilution (either by correcting for urinary creatinine or specific gravity). Of these two, in comparison with the urinary fluoride levels of both our mothers (0.88 mg/L) and our children (0.82 mg/L), the mean levels of children’s urinary fluoride were higher in the non-fluorosis (1.02 mg/L) and high-fluorosis (2.69 mg/L) groups found by Li et al. (1995) as well as the control (1.5 mg/L) and high-fluorosis (5.1 mg/L) groups described by Wang et al. (2007).

Among the limitations of our study are that we measured fluoride in spot (second morning void) urine samples instead of 24-hr urine collections. However, others have noted a close relationship between the fluoride concentrations of early morning samples and 24-hr specimens (Watanabe et al. 1994; Zohouri et al. 2006). Another limitation relates to the potential differences in the distribution of covariates over our study cohorts, raising the issue of potential bias. In the analyses we conducted across cohorts, we saw that, in comparison with Cohort 3, Cohort 2A clearly had higher mean bone lead levels (p<0.001) and possibly higher blood mercury levels (p=0.067). However, we saw no other differences and the differences in these measures have a clear likely explanation: Cohort 2A had bone lead levels measured in 1997–2001 and Cohort 3 had bone lead levels measured in 2001–2005. Given that environmental lead and mercury exposures were steadily decreasing during this time interval (due to the phase-out of lead from gasoline), this difference likely relates to an exposure–time–cohort effect. We do not anticipate that this phenomenon would have introduced a bias in our analyses of fluoride and cognition controlling for bone lead.

Another limitation relates to the missing data that pertain to our covariate and sensitivity variables. In the comparisons of participants in relation to missing data (Table 2A,B), the proportion of females was somewhat higher in the included versus excluded group for both the GCI and IQ analyses, and the mean levels of maternal blood Hg for those included were 28.5% and 24.9% higher than the mean levels for those excluded in the GCI and IQ analyses, respectively. We also note that the coefficients for the associations between fluoride on cognition varied substantially in some of the sensitivity analyses, particularly with respect to the subgroups of participants who have data on SES, lead exposure, and mercury exposure (of which, for the latter, the effect estimates almost doubled). We do not have a ready explanation for this phenomenon, given that there is no obvious way that each of the selection factors governing which mothers had these measurements (discussed above) could have influenced the fluoride–cognition relationship. Nevertheless, it is not possible to entirely rule out residual confounding or in the population as a whole (that might have been detected had we had full data on larger sample sizes) or bias (should the subpopulations that had the data for analysis have a different fluoride–cognition relationship than those participants who were excluded from the analyses).

Other limitations include the lack of information about iodine in salt, which could modify associations between fluoride and cognition; the lack of data on fluoride content in water given that determination of fluoride content is not reported as part of the water quality monitoring programs in Mexico; and the lack of information on other environmental neurotoxicants such as arsenic. We are not aware of evidence suggesting our populations are exposed to significant levels of arsenic or other known neurotoxicants; nevertheless, we cannot rule out the potential for uncontrolled confounding due to other factors, including diet, that may affect urinary fluoride excretion and that may be related to cognition.

Another potential limitation is that we adjusted maternal urinary fluoride levels based on urinary creatinine, whereas we adjusted children’s urinary fluoride levels based on urinary specific gravity; however, these two methods are almost equivalent in their ability to account for urinary dilution. We also had no data to assess the inter-examiner reliability of the testers administering the WASI test; however, the excellent reliability of these same testers in administering the McCarthy tests provides some reassurance that the WASI tests were conducted in a consistent manner.

Finally, our ability to extrapolate our results to how exposures may impact on the general population is limited given the lack of data on fluoride pharmacokinetics during pregnancy. There are no reference values for urinary fluoride in pregnant women in the United States. The Centers for Disease Control and Prevention has not included fluoride as one of the population exposures measured in urine or blood samples in its nationally representative sampling. The WHO suggests a reference value of 1 mg/L for healthy adults when monitoring renal fluoride excretion in community preventive programs (Marthaler 1999). As part of the NRC’s review of the fluoride drinking-water standard, it was noted that healthy adults exposed to optimally fluoridated water had urinary fluoride concentrations ranging from 0.62 to 1.5 mg/L.

Conclusion

In this study, higher levels of maternal urinary fluoride during pregnancy (a proxy for prenatal fluoride exposure) that are in the range of levels of exposure in other general population samples of pregnant women as well as nonpregnant adults were associated with lower scores on tests of cognitive function in the offspring at 4 and 6–12 y old.

Community water and salt fluoridation, and fluoride toothpaste use, substantially reduces the prevalence and incidence of dental caries (Jones et al. 2005) and is acknowledged as a public health success story (Easley 1995). Our findings must be confirmed in other study populations, and additional research is needed to determine how the urine fluoride concentrations measured in our study population are related to fluoride exposures resulting from both intentional supplementation and environmental contamination. However, our findings, combined with evidence from existing animal and human studies, reinforce the need for additional research on potential adverse effects of fluoride, particularly in pregnant women and children, and to ensure that the benefits of population-level fluoride supplementation outweigh any potential risks.

Acknowledgments

This study was supported by the U.S. National Institutes of Health (NIH; grants R01ES021446 and R01-ES007821); the National Institute of Environmental Health Sciences/the U.S. Environmental Protection Agency (NIEHS/EPA; grant P01ES022844), the NIEHS (grant P42-ES05947 and NIEHS Center Grant P30ES017885), and by the National Institute of Public Health/Ministry of Health of Mexico. The American British Cowdray Hospital provided facilities used for this research. The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NIEHS, NIH, or the U.S. EPA. David Bellinger collaborated on the design and execution of this study’s cognitive testing.

References

Afeiche M, Peterson KE, Sánchez BN, Cantonwine D, Lamadrid-Figueroa H, Schnaas L, et al. 2011. Prenatal lead exposure and weight of 0- to 5-year-old children in Mexico City. Environ Health Perspect 119(10):1436–1441, PMID: 21715242, 10.1289/ehp.1003184.

Baez R, Petersen PE, Marthaler T. 2014. Basic Methods for Assessment of Renal Fluoride Excretion in Community Prevention Programmes for Oral Health. Geneva, Switzerland:World Health Organization.

Basu N, Tutino R, Zhang Z, Cantonwine DE, Goodrich JM, Somers EC, et al. 2014. Mercury levels in pregnant women, children, and seafood from Mexico City. Environ Res 135:63–69, PMID: 25262076, 10.1016/j.envres.2014.08.029.

Braun JM, Hoffman E, Schwartz J, Sanchez B, Schnaas L, Mercado-Garcia A, et al. 2012. Assessing windows of susceptibility to lead-induced cognitive deficits in Mexican children. Neurotoxicology 33(5):1040–1047, PMID: 22579785, 10.1016/j.neuro.2012.04.022.

Caldwell BM, Bradley RH. 2003. Administration Manual: HOME Observation for Measurement of the Environment. Little Rock, AK:University of Arkansas at Little Rock.

Choi AL, Sun G, Zhang Y, Grandjean P. 2012. Developmental fluoride neurotoxicity: a systematic review and meta-analysis. Environ Health Perspect 120(10):1362–1368, PMID: 22820538, 10.1289/ehp.1104912.

Choi AL, Zhang Y, Sun G, Bellinger DC, Wang K, Yang XJ, et al. 2015. Association of lifetime exposure to fluoride and cognitive functions in Chinese children: a pilot study. Neurotoxicol Teratol 47:96–101, PMID: 25446012, 10.1016/j.ntt.2014.11.001.

Doull J, Boekelheide K, Farishian B, Isaacson R, Klotz J, Kumar J, et al. 2006. Fluoride in Drinking Water: A Scientific Review of EPA’s Standards. Committee on Fluoride in Drinking Water, Board on Environmental Studies and Toxicology, Division on Earth and Life Studies, National Research Council of the National Academies. Washington, DC:National Academies Press.

Easley MW. 1995. Celebrating 50 years of fluoridation: a public health success story. Br Dent J 178(2):72–75, PMID: 7848761, 10.1038/sj.bdj.4808658.

Ettinger AS, Lamadrid-Figueroa H, Téllez-Rojo MM, Mercado-García A, Peterson KE, Schwartz J, et al. 2009. Effect of calcium supplementation on blood lead levels in pregnancy: a randomized placebo-controlled trial. Environ Health Perspect 117(1):26–31, PMID: 19165383, 10.1289/ehp.11868.

Gedalia I, Brzezinski A, Bercovici B. 1959. Urinary fluorine levels in women during pregnancy and after delivery. J Dent Res 38(3):548–551, PMID: 13654605, 10.1177/00220345590380031701.

Gomaa A, Hu H, Bellinger D, Schwartz J, Tsaih SW, Gonzalez-Cossio T, et al. 2002. Maternal bone lead as an independent risk factor for fetal neurotoxicity: a prospective study. Pediatrics 110(1):110–118, PMID: 12093955.

Grandjean P, Herz KT. 2011. Methylmercury and brain development: imprecision and underestimation of developmental neurotoxicity in humans. Mt Sinai J Med 78(1):107–118, PMID: 21259267, 10.1002/msj.20228.

Hu H, Téllez-Rojo MM, Bellinger D, Smith D, Ettinger AS, Lamadrid-Figueroa H, et al. 2006. Fetal lead exposure at each stage of pregnancy as a predictor of infant mental development. Environ Health Perspect 114(11):1730–1735, PMID: 17107860, 10.1289/ehp.9067.

Huang S, Hu H, Sánchez BN, Peterson KE, Ettinger AS, Lamadrid-Figueroa H, et al. 2016. Childhood blood lead levels and symptoms of attention deficit hyperactivity disorder (ADHD): a cross-sectional study of Mexican children. Environ Health Perspect 124(6):868–874, PMID: 26645203, 10.1289/ehp.1510067.

Jones S, Burt BA, Petersen PE, Lennon MA. 2005. The effective use of fluorides in public health. Bull World Health Organ 83:670–676.

Juárez-López M, Hernández-Guerrero JC, Jiménez-Farfán D, Molina-Frechero N, Murrieta-Pruneda F, Lopez-Jimenez G. 2007. Fluoride Urinary Excretion in Mexico City’s Preschool Children [in Spanish]. Revista de investigacion clinica; organo del Hospital de Enfermedades de la Nutricion 60:241–247.

Julvez J, Ribas-Fito N, Torrent M, Forns M, Garcia-Esteban R, Sunyer J. 2007. Maternal smoking habits and cognitive development of children at age 4 years in a population-based birth cohort. Int J Epidemiol 36(4):825–832, PMID: 17550944, 10.1093/ije/dym107.

Kordas K, Ettinger AS, Bellinger DC, Schnaas L, Téllez Rojo MM, Hernández-Avila M, et al. 2011. A dopamine receptor (DRD2) but not dopamine transporter (DAT1) gene polymorphism is associated with neurocognitive development of Mexican preschool children with lead exposure. J Pediatr 159(4):638–643, PMID: 21592505, 10.1016/j.jpeds.2011.03.043.

Li XS, Zhi JL, Gao RO. 1995. Effect of fluoride exposure on intelligence in children. Fluoride 28(4):189–192.

Liu F, Ma J, Zhang H, Liu P, Liu YP, Xing B, et al. 2014. Fluoride exposure during development affects both cognition and emotion in mice. Physiol Behav 124:1–7, PMID: 24184405, 10.1016/j.physbeh.2013.10.027.

Marthaler T. 1999. Monitoring of Renal Fluoride Excretion in Community Preventive Programmes on Oral Health. Geneva, Switzerland:World Health Organization.

Martínez-Mier EA, Cury JA, Heilman JR, Katz BP, Levy SM, Li Y, et al. 2011. Development of gold standard ion-selective electrode-based methods for fluoride analysis. Caries Res 45(1):3–12, PMID: 21160184, 10.1159/000321657.

Martínez-Mier EA, Soto-Rojas AE, Buckley CM, Zero DT, Margineda J. 2005. Fluoride concentration of bottled water, tap water, and fluoridated salt from two communities in Mexico. Int Dent J 55(2):93–99, PMID: 15880964.

McCarthy D. 1991. Manual for the McCarthy Scales of Children’s Abilities. Spanish, User’s Guide [in Spanish]. Madrid, Spain:TEA Ediciones.

Mullenix PJ, Denbesten PK, Schunior A, Kernan WJ. 1995. Neurotoxicity of sodium fluoride in rats. Neurotoxicol Teratol 17(2):169–177, PMID: 7760776.

NTP (National Toxicology Program). 2016. Systematic Literature Review on the Effects of Fluoride on Learning and Memory in Animal Studies. NTP Research Report 1. Research Triangle Park, NC:NTP.

Opydo-Szymaczek J, Borysewicz-Lewicka M. 2005. Urinary fluoride levels for assessment of fluoride exposure of pregnant women in Poznan, Poland. Fluoride 38:312–317.

Puertas R, Lopez-Espinosa MJ, Cruz F, Ramos R, Freire C, Pérez-García M, et al. 2010. Prenatal exposure to mirex impairs neurodevelopment at age of 4 years. Neurotoxicology 31(1):154–160, PMID: 19818364, 10.1016/j.neuro.2009.09.009.

Renteria L, Li ST, Pliskin NH. 2008. Reliability and validity of the Spanish language Wechsler Adult Intelligence Scale (3rd Edition) in a sample of American, urban, Spanish-speaking Hispanics. Clin Neuropsychol 22(3):455–470, PMID: 17853132, 10.1080/13854040701336428.

Secretaría-de-Salud. 1995. Norma oficial Mexicana nom-040-ssa-1-1993. Sal yodatada y sal fluorada [in Spanish]. México:Diario Oficial de la Federación, 12–27.

Secretaría-de-Salud. 1996. Norma oficial Mexicana nom-127-ssa1-1994. Salud ambiental. Agua para uso y consumo humano. Límites permisibles de calidad y tratamientos a que debe someterse el agua para su potabilización [in Spanish]. México:Diario Oficial de la Federación, 41–46.

Shen YW, Taves DR. 1974. Fluoride concentrations in the human placenta and maternal and cord blood. Am J Obstet Gynecol 119(2):205–207, PMID: 4823388.

Thomas DB, Basu N, Martinez-Mier EA, Sánchez BN, Zhang Z, Liu Y, et al. 2016. Urinary and plasma fluoride levels in pregnant women from Mexico City. Environ Res 150:489–495, PMID: 27423051, 10.1016/j.envres.2016.06.046.

Usuda K, Kono K, Shimbo Y, Fujihara M, Fujimoto K, Kawano A, et al. 2007. Urinary fluoride reference values determined by a fluoride ion selective electrode. Biol Trace Elem Res 119(1):27–34, PMID: 17914216, 10.1007/s12011-007-0044-6.

Valdez Jiménez L, López Guzmán OD, Cervantes Flores M, Costilla-Salazar R, Calderón Hernández J, Alcaraz Contreras Y, et al. 2017. In utero exposure to fluoride and cognitive development delay in infants. Neurotoxicology 59:65–70, PMID: 28077305, 10.1016/j.neuro.2016.12.011.

Wang SX, Wang ZH, Cheng XT, Li J, Sang ZP, Zhang XD, et al. 2007. Arsenic and fluoride exposure in drinking water: children’s IQ and growth in Shanyin county, Shanxi province, China. Environ Health Perspect 115(4):643–647, PMID: 17450237, PubMed Central PMCID: PMC1852689. 10.1289/ehp.9270.

Watanabe M, Kono K, Orita Y, Ydote T, Usuda K, Takahashi Y, et al. 1994. Influence of dietary fluoride intake on urinary fluoride concentration and evaluation of corrected levels in spot urine. In: Proceedings of the XXth Conference of the International Society for Fluoride Research, Beijing, China. Beijing, China:Ministry of Public Health of People’s Republic of China, 246–247.

Wechsler D. 1999. Wechsler Abbreviated Scale of Intelligence. San Antonio, TX:Psychological Corporation.

Wechsler D, Jorge M, Velaco A. 1981. WAIS-Español: Escala de Inteligencia para Adultos: El Manual Moderno [in Spanish]. México, DF:El Manual Moderno, S.A.

Zohouri F, Swinbank C, Maguire A, Moynihan P. 2006. Is the fluoride/creatinine ratio of a spot urine sample indicative of 24‐h urinary fluoride? Community Dent Oral Epidemiol 34(2):130–138, PMID: 16515677, 10.1111/j.1600-0528.2006.00269.x.

In Memoriam: Herbert L. Needleman

PDF icon PDF Version (254 KB)

  • Published: 19 September 2017

    Environ Health Perspect 125(8): 091601 (2017). https://doi.org/10.1289/EHP2636

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • Published: 19 September 2017

    Environ Health Perspect 125(8): 091601 (2017). https://doi.org/10.1289/EHP2636

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.


  • Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Herbert L. Needleman, MD, pediatrician, child psychiatrist, and hero for children’s environmental health, died in Pittsburgh, Pennsylvania, on 18 July 2017.

[PhotographHerbert L. Needleman, 1927–2017
(Steve McCaw/Image Associates)

Needleman was one of the world’s foremost researchers on lead poisoning. He conducted seminal studies that illuminated with great clarity the enduring impacts of lead on children’s health. He was a deeply moral man with a strong sense of social justice, a courageous and highly effective advocate who successfully translated his scientific findings into robust, evidence-based interventions that safeguarded the health of millions. He conducted much of this work in the face of powerful opposition.

Needleman’s scientific work began when he was a young pediatrician at the Children’s Hospital of Philadelphia in the late 1960s and early 1970s. He noted that children who had recovered from acute lead poisoning appeared frequently to have chronic residual neuropsychological impairment. He came to realize that lead poisoning was not an all-or-none phenomenon from which a child either died or recovered completely, as had previously been taught. Instead, he hypothesized that lower levels of exposure to lead that produce no clinically evident symptoms might nonetheless be associated with permanent neuropsychological deficits, albeit of lesser magnitude, a phenomenon now termed “subclinical toxicity.”

Through his subsequent studies, Needleman documented that the loss of intelligence, the shortening of attention span, and the disruption of behavior that results from subclinical lead poisoning is permanent, untreatable, and irreversible. He therefore argued that the only rational approach to treatment of lead poisoning is to prevent exposure to lead.

In his first major epidemiologic study, Needleman conducted a cross-sectional evaluation of two groups of asymptomatic elementary school children in Boston. Needleman found a consistent mean decrement in intelligence of 6–7 IQ points in highly exposed children in comparison with their peers without elevated blood lead. The most severe deficits were seen in children with the greatest lead burdens (Needleman et al. 1979). Needleman also noticed that the highly exposed children were far less likely to have IQ scores>125 and much more likely to have IQ scores< 70. They were also more likely to have behavioral problems, including an inability to focus, a shortened attention span, and impulsive hyper-aggressive behavior.

In a follow-up examination of the same two groups 11 y later, Needleman found that the association between lead exposure and impaired neuropsychological development had not gone away and now included a significantly higher frequency of school dropout, more learning disabilities, and poorer eye–hand coordination than their peers without elevated lead levels showed (Needleman et al. 1990). Even more troubling, the children who had been exposed to lead had a higher reported frequency of encounters with the police.

In his next major study, Needleman explored the impact on children’s health of prenatal lead exposure. He and his team found that deficits in children’s development were associated with blood lead levels at birth as low as 10–20 μg/dL (Bellinger et al. 1986). Previously, toxicity had not been thought to occur at blood lead levels below 25–40 μg/dL.

In his last major studies, Needleman examined the association between lead exposure and criminality. He compared the body lead burdens of incarcerated young men and nonincarcerated youth from the same communities. He found that the incarcerated young men had significantly higher lead burdens than their peers (Needleman et al. 1996). This finding strengthened Needleman’s conclusion that strong links exist between lead exposure in early life and later increased risk for criminal behavior and added to the growing literature on the association between lead exposure and “violent crime” (Nevin 2000; Reyes 2007).

Needleman’s work has had major consequences for public health and the economy. His findings provided the intellectual basis for decisions made by federal agencies to remove lead from gasoline and interior paint and to remediate lead contamination in many thousands of housing units across the United States. These interventions produced an estimated 94% reduction in blood lead levels in American children and also resulted in sharp decreases in incidence of lead poisoning (Annest et al. 1983). It has also been suggested that the nationwide reduction in lead exposure is at least partly responsible for a 5-point increase in mean IQ scores for children born in the United States since 1980, in comparison with those of children born in earlier decades (Grosse et al. 2002). Furthermore, the overall economic benefit of reducing lead exposures has been estimated at US$100–300 billion per year (Grosse et al. 2002).

Internationally, Needleman’s findings provided the catalyst for nations around the world to remove lead from gasoline. Several countries, including Finland, Greece, India, and Thailand, have removed lead from gasoline, and children’s blood lead levels in all those countries have fallen sharply as a result. The aggregate intelligence and economic productivity of entire societies have been enhanced as the direct result of Needleman’s work (UNEP 2017).

In recognition of his scientific leadership and contribution to public health, Needleman was honored by the presentation of multiple awards, among them the Heinz Award for the Environment, the Charles A. Dana Award for Pioneering Achievement in Public Health, the Prince Mahidol Award for outstanding achievements in medicine and public health presented by the King of Thailand, the New York Academy of Sciences Sarah L. Poiley Memorial Award, the National Wildlife Federation Conservation Achievement Award in Science, the University of Pittsburgh Chancellor’s Distinguished Public Service Award, the Physicians Forum’s Edward K. Barsky Award, the Society for Occupational and Environmental Health Vernon Houk Award, Muhlenberg College’s Dr. John V. Shankweiler Prize, the Toxicology Landmarks Program Award from the Society of Toxicology, and the University of Pennsylvania’s Distinguished Graduate Award.

Needleman’s work, and especially his elaboration of the concept of subclinical toxicity, revolutionized scientific understanding of the neurodevelopmental impact not only of lead, but of a wide range of toxic chemicals. Needleman’s work thus created a new paradigm for developmental neurotoxicology that took the field far beyond its earlier focus on acute, high-dose toxicity. Needleman demonstrated that harmful effects could occur at every level of exposure to a neurotoxicant, an observation that has now been replicated for many chemicals, among them methyl mercury, polychlorinated biphenyls, organophosphate pesticides, phthalates, and brominated flame retardants. Additionally, Needleman’s work showed that exposure in early life to lead and other neurodevelopmental toxicants can produce lasting injury to the brain that, when widespread in a society, can have profound consequences for health, economic productivity, and political stability.

Herb Needleman was a warm, caring, and deeply ethical man. He was a generous and beloved mentor who nurtured and led a generation of younger researchers. He inspired us not only through the extraordinarily high quality of his science, but also through his deep thirst for social justice, his care for the oppressed, and his heroism in the face of adversity. Needleman’s passing is a loss for America and for the world. He will be deeply missed.

References

Annest JL, Pirkle JL, Makuc D, Neese JW, Bayse DD, Kovar MG. 1983. Chronological trend in blood lead levels between 1976 and 1980. N Engl J Med 308(23):1373–1377, PMID: 6188954, 10.1056/NEJM198306093082301.

Bellinger D, Leviton A, Needleman HL, Waternaux C, Rabinowitz M. 1986. Low-level lead exposure and infant development in the first year. Neurobehav Toxicol Teratol 8(2):151–161, PMID: 2423895, 10.1289/ehp.6941.

Grosse SD, Matte TD, Schwartz J, Jackson RJ. 2002. Economic gains resulting from the reduction in children’s exposure to lead in the United States. Environ Health Perspect 110(6):563–569, PMID: 12055046, 10.1289/ehp.02110563.

Needleman HL, Gunnoe C, Leviton A, Reed R, Peresie H, Maher C, et al. 1979. Deficits in psychologic and classroom performance of children with elevated dentine lead levels. N Engl J Med 300(13):689–695, PMID: 763299, 10.1056/NEJM197903293001301.

Needleman HL, Riess JA, Tobin MJ, Biesecker GE, Greenhouse JB. 1996. Bone lead levels and delinquent behavior. JAMA 275(5):363–369, PMID: 8569015, 10.1016/S0892-0362(02)00269-6.

Needleman HL, Schell A, Bellinger D, Leviton A, Allred EN. 1990. The long-term effects of exposure to low doses of lead in childhood. An 11-year follow-up report. N Engl J Med 322(2):83–88, PMID: 2294437, 10.1056/NEJM199001113220203.

Nevin R. 2000. How lead exposure relates to temporal changes in IQ, violent crime, and unwed pregnancy. Environ Res 83(1):1–22, PMID: 10845777, 10.1006/enrs.1999.4045.

Reyes JW. 2007. Environmental policy as social policy? The impact of childhood lead exposure on crime. BE J Econ Anal Policy 7(1):Article 51, .

UNEP (United Nations Environment Programme). 2017. Partnership for Clean Fuels and Vehicles. http://www.unep.org/transport/pcfv/ [accessed 3 August 2017].

Association between Exposure to p,p′-DDT and Its Metabolite p,p′-DDE with Obesity: Integrated Systematic Review and Meta-Analysis

Author Affiliations open

1Department of Environmental Toxicology, University of California, Davis, Davis, California, USA

2Office of Environmental Health Hazard Assessment, California Environmental Protection Agency, Oakland, California, USA

PDF icon PDF Version (3.2 MB)

  • Background:
    The prevalence of obesity is increasing in all countries, becoming a substantial public health concern worldwide. Increasing evidence has associated obesity with persistent pollutants such as the pesticide DDT and its metabolite p,p′-DDE.
    Objectives:
    Our objective was to systematically review the literature on the association between exposure to the pesticide DDT and its metabolites and obesity to develop hazard identification conclusions.
    Methods:
    We applied a systematic review-based strategy to identify and integrate evidence from epidemiological, in vivo, and in vitro studies. The evidence from prospective epidemiological studies was quantitatively synthesized by meta-analysis. We rated the body of evidence and integrated the streams of evidence to systematically develop hazard identification conclusions.
    Results:
    We identified seven epidemiological studies reporting prospective associations between exposure to p,p′-DDE and adiposity assessed by body mass index (BMI) z-score. The results from the meta-analysis revealed positive associations between exposure to p,p′-DDE and BMI z-score (β=0.13 BMI z-score (95% CI: 0.01, 0.25) per log increase of p,p′-DDE). Two studies constituted the primary in vivo evidence. Both studies reported positive associations between exposure to p,p′-DDT and increased adiposity in rodents. We identified 19 in vivo studies and 7 in vitro studies that supported the biological plausibility of the obesogenic effects of p,p′-DDT and p,p′-DDE.
    Conclusions:
    We classified p,p′-DDT and p,p′-DDE as “presumed” to be obesogenic for humans, based on a moderate level of primary human evidence, a moderate level of primary in vivo evidence, and a moderate level of supporting evidence from in vivo and in vitro studies. https://doi.org/10.1289/EHP527
  • Received: 17 May 2016
    Revised: 04 May 2017
    Accepted: 09 May 2017
    Published: 18 September 2017

    Please address correspondence to M.A. La Merrill, Dept. of Environmental Toxicology, University of California, Davis, 1 Shields Ave., 4245 Meyer Hall, Davis, CA 95616-5270 USA. Telephone: (530) 754-7254. Email: mlamerrill@ucdavis.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP527).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (1.7 MB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (392 KB)

Introduction

The Obesity Society defines obesity as a disease characterized by an excess of body fat, either total body fat or a particular depot of body fat, which increases the likelihood of comorbidities such as diabetes, hypertension, coronary heart disease, stroke, some cancers, obstructive sleep apnea, or osteoarthritis (Allison et al. 2008; Arnold et al. 2015; Jokinen 2015). Obesity has been increasing in all countries, with prevalence doubling during the past three decades to become a substantial public health concern worldwide (Ogden et al. 2014; WHO 2014). Among children and adolescents, the prevalence of obesity follows similar time trends and those akin comorbidities are also diagnosed at early ages (I’Allemand et al. 2008).

Excess caloric consumption and sedentary behavior are some of the risk factors traditionally identified as the main promoters of obesity and overweight. These risks alone do not explain the increased body weight and odds of obesity have also been observed among primates and rodents serving as experimental controls, feral rodents, and domestic dogs and cats across recent decades in the United States (Klimentidis et al. 2011). Instead the complex etiology of this condition involves multiple interrelated causes, such as genetic, social, and environmental factors (Speakman and O’Rahilly 2012; WHO 2014). Some environmental pollutants, including lipophilic persistent organic pollutants, have been associated with an increased risk of overweight and obesity in epidemiological and experimental studies (Lee et al. 2014; Taylor et al. 2013; Thayer et al. 2012). This evidence supports the “obesogen” hypothesis, which predicts that some xenobiotic chemicals “inappropriately regulate lipid metabolism and adipogenesis to promote obesity” (Grün and Blumberg 2006). Extensive data in support of this hypothesis illustrates that the developmental period is a vulnerable window during which transient environmental exposures may inappropriately regulate energy balance or adiposity over the long term (Grün and Blumberg 2006; La Merrill and Birnbaum 2011).

The body of evidence for obesogenic effects of the pesticide dichlorodiphenyltrichloroethane (DDT) and its metabolite dichlorodiphenyldichloroethylene (DDE) has increased notably in the last decade, with a particular focus on exposure during prenatal development. Technical DDT is a persistent organic pesticide mixture of three isoforms, p,p′-DDT, o,p′-DDT, and p,p′-DDD. In the present paper we use the term DDTs to identify the molecular family including these DDT isoforms and their metabolites (e.g., p,p′-DDE). The commercial formulation was widely used for the control of disease (e.g., malaria, typhus) vectors in most countries from the mid-1940s to the late 20th century. DDT is still manufactured in India for control of malaria primarily in India and Africa, where the quantity used for vector control (71% of total) has not changed substantially since the Stockholm Convention restricted its use (ATSDR 2002; Rogan and Chen 2005; UNEP 2010). Moreover, due to the extremely high persistence and lipophilicity of DDTs, internal exposure to DDTs remains ubiquitous in many countries decades after the ban was enforced (Rogan and Chen 2005; Smith 1999).

DDTs are listed by the California Environmental Protection Agency (CalEPA) as causing developmental and reproductive toxicity and have recently been classified by the International Agency for Research on Cancer (IARC) as probably carcinogenic to humans (Cal/EPA-OEHHA 2016; Loomis et al. 2015), yet no study of its obesogenic effects has attempted to deliver hazard identification conclusions by means of a systematic approach to integrate all the evidence. The advantages of applying a Grading of Recommendations Assessment, Development, and Evaluation (GRADE) working group approach has been extensively demonstrated in fields such as clinical medicine and public health, and some recent studies have depicted the benefits of its implementation in environmental health assessment to increase the transparency, rigor, and reproducibility on the decision-making process (Lam et al. 2014; Morgan et al. 2016; Sheehan and Lam 2015; Woodruff and Sutton 2014). Thus, the main objective of this study was to systematically review and integrate the available literature on the association between exposure to the pesticide DDT and obesity to deliver hazard identification conclusions.

Materials and Methods

We applied a systematic review–based strategy to evaluate and integrate evidence from epidemiological, in vivo, and in vitro studies. The methodological approach is based on the National Toxicology Program Office of Health Assessment and Translation’s (NTP/OHAT) Handbook for Conducting a Literature-Based Health Assessment with support of the Navigation Guide, both of which provide a standardized methodology to implement the GRADE approach to environmental health assessments (OHAT 2015a; Rooney et al. 2014). We followed a pre-specified protocol (see “1. SYSTEMATIC REVIEW PROTOCOL” in the Supplemental Material) that was slightly modified throughout an iterative process to refine the integrated systematic review seeking to answer the study question.

In terms of logistics, the NTP/OHAT Handbook considers two valid approaches depending on the size and complexity of the project: the main review maybe either a) independently conducted by two members of the review team; or b) conducted by one member of the review team, with a second member of the team confirming the exclusion determination of the first reviewer. Accordingly, we implemented the second approach, where the screening, data extraction, and data synthesis process were performed by one reviewer (G.C.-S.) after checking the reproducibility, reliability and validity of outcomes by means of a full-duplicated pilot trial where two reviewers (G.C.-S. and M.A.L.) performed the entire process in a subsample of studies and compared the outcomes. Results from the pilot trial demonstrated that no improvement of accuracy and reliability nor reduction of errors were observed when we compared the results from both reviewers. Discrepancies were discussed with a third reviewer (A.G.S.) and external expert advisors. The confidence on the body of evidence rating was performed through a panel discussion with the presence and final agreement of the reviewers.

Study Question and Eligibility Criteria

We formulated the search question: “Does exposure to DDT increase obesity in humans?” Accordingly, we defined the eligibility criteria for the key elements (population, exposure, comparators, and outcomes; PECO) summarized in the PECO statement (Table 1).

Table 1. PECO statement.
Study type Population Exposure Comparators Outcomes
Epidemiological studies Humans studied prospectively without restrictions on country, race, religion, sex. Exposure to DDT and derivatives or isoforms based on administered dose or concentrations, environmental measures or indirect measures. The exposure must be measured individually using direct validated biomonitoring methods. We excluded studies to assess the therapeutic use of o,p′-DDD isoform, commercially known as mitotane or lysodren. Reference groups of population exposed at lower levels of DDTs than the rest of population groups. Primary outcome: body mass index (BMI) and z-score, overweight, and obesity.
All ages and/or life-stage at exposure or outcome assessment were included with exception of newborn (birth outcomes were excluded).
In vivo studies Any animal model, sex, age, lifestage at exposure or outcome assessment. Exposure to all types of DDT and derivatives or isoforms and their mixtures, including all ranges of concentrations, duration, and routes of exposure. Experimental animals receiving vehicle-only treatment. Primary outcome: adiposity (e.g., relative or absolute weight, DXA, EchoMRI).
We excluded studies including DDT in mixtures with other pollutants. Secondary outcomes: dyslipidemia, abnormal lipids, other markers of metabolic homeostasis, energy balance.
In vitro studies Any cell lines and/or in vitro procedures. Exposure to all types of DDT and derivatives or isoforms and mixtures, including all ranges of concentrations, duration, and routes of exposure. Cells receiving vehicle-only treatment. Adipogenic differentiation, gene expression of metabolic regulators, adipokines.
We excluded studies including DDT in mixtures with other pollutants.

We initially included human prospective studies reporting associations between DDTs and health outcomes related to increased adiposity, overweight, and obesity, considering continuous body mass index (BMI) and its z-score (BMI-z) as a primary outcome. The preferred choice of BMI-z and BMI among clinicians and their extensive use in epidemiology prompted us to choose these outcomes as primary. Cross-sectional studies were excluded to avoid potential reverse causality that can result from the effect of adiposity on circulating lipophilic chemical levels (La Merrill et al. 2013). The metabolite p,p′-DDE was considered the major biomarker of exposure to DDT given its high occurrence, but we also explored the associations with other isoforms.

We retained in vivo studies reporting associations between DDTs (excluding mixtures with other pollutants) and adiposity as the main stream of evidence. The use of crude body weight has limited applicability to characterize obesity in animal models if adiposity or other related outcomes are not measured, and risk of misclassification has already been demonstrated (Nascimento et al. 2008; Woods et al. 2003).

Based on a preliminary literature search, we anticipated a limited number of studies addressing our primary research question with animals, and expanded the evidence with health outcomes directly related to adiposity. The supporting body of evidence included in vivo studies detailing associations between DDTs and energy imbalance; on the basis that an imbalance between energy intake and energy expenditure is considered the primary etiology for excess fat accumulation (Drenowatz 2015; Martinez 2000), measurements of thermogenesis and energy expenditure were considered directly applicable. We considered abnormal lipids (circulating and hepatic) as additional supporting outcomes because dyslipidemia is a principal metabolic comorbidity associated with obesity (Bays et al. 2013). Adipokines were also considered secondary to adiposity because of the association of adiponectin with adipocyte differentiation and the proportional relationship between circulating levels of leptin and fat mass (Stern et al. 2016).

As depicted in the NTP/OHAT framework, we also considered supporting evidence from in vitro studies that addressed mechanisms underlying the causes of obesity. Among in vitro studies, we considered enhanced adipogenic differentiation of cells, including lipid, protein, and RNA changes associated with this process. Additionally, we considered the adipokines as reliable markers of in vitro adipocyte expansion for their high association with differentiation and regulatory role on lipid homeostasis (Fu et al. 2005; Stern et al. 2016).

Concerning the publication type, we only considered reports that contained original data and were peer-reviewed, thus excluding reviews. All publication dates were considered and articles not written in English were excluded. Conference papers were excluded.

Search Strategy

The search string (see “1.3.2. SEARCH” in the Supplemental Material) was applied to three electronic literature databases [MEDLINE (http://www.ncbi.nlm.nih.gov/pubmed), EMBASE (https://www.embase.com), and Scopus (http://www.scopus.com)] on 23 March 2015, and a follow-up search was performed on 8 January 2016. The search strategy was developed to identify human, in vivo, and in vitro studies reporting original data on the associations of DDTs with obesity given that some outcomes of interest here may be indexed under co-morbidities of obesity such as diabetes, dyslipidemia, and metabolic syndrome. We also searched for measures of energy imbalance, and adipogenic differentiation, as well as protein and RNA measures associated with these processes, as indicators of mechanisms underlying the potential causes of obesity. The search was run without filters and without limitation on publication date. The records were pooled in Endnote X7 and screened manually to eliminate duplicates. The resulting library was uploaded to DistillerSR online software (Evidence Partners) to carry out the selection of studies.

Selection of Studies

The selection was performed in DistillerSR (Evidence Partners) software in a two-step process: During the first step, the studies were screened based on the title and abstract. The included and doubtful studies were screened in a second step, using the full-text to conclude if the studies meet the inclusion criteria.

Data Extraction

The data was extracted using data forms specifically designed for human, in vivo, and in vitro studies (see “5. DATA FORMS” in the Supplemental Material) in DistillerSR and exported to Excel. The data was extracted by a main reviewer (G.C.-S.) and checked by an additional external reviewer (M.A.L.) to ensure accuracy. Discrepancies and controversial issues were discussed by the reviewer team, and external advice was requested when it was required. We contacted the authors to request additional data when it was required.

Data Synthesis and Meta-Analysis

We synthesized the data from human epidemiological studies by means of meta-analysis of effect estimates. Data from in vivo and in vitro studies were synthesized and displayed to summarize the direction of the effect while comparing of doses among studies using forest plots adapted from Thayer et al. (2012). The effect estimates initially considered for pooling the data were beta regression coefficients (β) for continuous outcomes, and risk ratios (RR) and/or odds ratios (OR) for dichotomous outcomes. However, the different methodological approaches, metrics, and outcomes used in the different studies only allowed pooling β estimates for continuous models with BMI-z as the dependent variable and p,p′-DDE as the independent variable with the corresponding covariables. The effect sizes were summarized using the inverse variance method for random-effects meta-analysis (DerSimonian and Laird 1986). Studies also provided the measures of variance of the effect size, such as confidence intervals (CI). Between-study variance in a random-effects meta-analysis was represented by tau squared (τ2). Heterogeneity was assessed with the I2 statistic, which quantifies the heterogeneity and degree of inconsistency among studies. The results were interpreted according Cochrane’s criteria: I2 between 0% and 40%: heterogeneity might not be important; between 30% and 60%: may represent moderate heterogeneity; between 50% and 90%: may represent substantial heterogeneity; and between 75% and 100%: considerable heterogeneity (Higgins and Green 2011). Potential small-study bias was evaluated by funnel plots and Egger’s test (Harbord et al. 2006). The influence of each individual study in a meta-analysis was investigated by omitting each study in turn and reestimating the summary estimate.

Rating and Integrating the Evidence for Hazard Identification

We applied the NTP/OHAT framework (OHAT 2015a), based on the GRADE approach (Guyatt et al. 2011a), to rate the confidence in the body of evidence, translate to a level of evidence, and integrate the different streams of evidence to deliver the hazard identification conclusions. The overall work-flow process is illustrated in Figure 1, considering two primary bodies of evidence from human studies (increased BMI-z) and animal studies (increased adiposity). Two additional bodies of evidence (secondary outcomes from in vivo and in vitro studies) were included as complementary information to support the associations and its biological plausibility. The confidence and level of evidence was evaluated independently for each body of evidence (e.g., human primary outcomes, in vivo primary outcomes, in vivo secondary outcomes, and in vitro secondary outcomes), establishing an initial confidence rating based on key study design features. The body of evidence from in vivo and in vitro studies were initially rated with high confidence because they control the exposure levels, which in turn are prior to the outcome, the outcome measure is collected at the individual level, and a comparison group equal in all conditions save the exposure is always used. In contrast, the body of evidence composed of human prospective studies was initially classified as moderate confidence because observational studies fail to control the exposure levels or provide a comparison group known to be absolutely free of all sources of confounding (compared with randomized controlled trials). Despite the limitations related to the uncontrolled exposure, the prospective observational studies are considered reliable approaches to establish causative associations between pollutant exposures and disease. Moreover, given the ethical limitations on carrying out controlled trials with pollutants in humans, this epidemiological design is considered the most feasible and reliable approach (Johnson et al. 2014).

Work flow.
Figure 1. Flow chart for rating the quality and integration of evidence from human and animal evidence, and the judgments of primary and supporting evidence for hazard identification conclusions.

Subsequently, these initial ratings were subjected to a sequential process considering those factors that may affect (upgrading or downgrading) the confidence, including the risk of bias, imprecision, publication bias, indirectness, magnitude, dose response, plausible confounding, and consistency across populations and models (Figure 1). The risk of bias was evaluated by means of risk of bias tools specifically designed for human epidemiological studies and animal studies and slightly adapted for DDTs and obesity outcomes (Koustas et al. 2014; OHAT 2015b). The rationale for risk of bias rating and results may be found in the Supplemental Material for humans (see “6. INSTRUCTIONS TO ASSESS THE RISK OF BIAS OF HUMAN EPIDEMIOLOGICAL STUDIES” and Tables S10–S17) and animal studies (see “7. INSTRUCTIONS TO ASSESS THE RISK OF BIAS OF IN VIVO STUDIES” and Tables S24–S32). The extended rationale for rating the confidence and integrating the evidence is reported in the protocol (see “1.4. RATING THE BODY OF EVIDENCE” and “1.5. INTEGRATION OF EVIDENCE AND HAZARD IDENTIFICATION CONCLUSIONS” in the Supplemental Material). We did not assess the risk of bias of in vitro studies because of the lack of risk of bias tools or guidance to assess the internal quality; however, we considered the remaining confidence factors (Figure 1) to rate the confidence in the in vitro body of evidence (Rooney et al. 2016). In brief, the NTP/OHAT’s risk of bias tiered approach considers key elements or risk of bias domains to establish the risk of bias classification for each individual study (Tier 1 to 3). Individual studies are classified in the Tier 1 when the key elements are considered as having “definitely low” or “probably low” risk of bias, and classified in the Tier 3 when the key elements are considered as having “definitely high” or “probably high” risk of bias (see “1.4. RATING THE BODY OF EVIDENCE” in the Supplemental Material). In the second level of risk of bias evaluation, the rating of the overall risk of bias in the body of evidence is classified as “not likely,” “serious” or “very serious,” depending on whether most information is gathered from studies classified as Tier 1, 2 or 3, respectively (see Table S2). The confidence rating process was completed considering the upgrading and downgrading factors and balanced together to deliver a final rating for each of the four bodies of evidence (Figure 1). This final confidence rating for each body of evidence (human, in vivo primary, in vivo supporting, and in vitro supporting) was translated to a level of evidence (low, moderate, or high) for each of the primary evidence streams and for the supporting evidence stream, considering additionally the nature and direction of the effect (“health effect” and “no health effect”).

The two primary bodies of evidence were integrated using the hazard identification scheme to provide a preliminary classification of the obesogen hazard identified for DDTs (“known,” “presumed,” “suspected,” or “not classifiable” hazard for humans; Figure 1). The final level of evidence (low, moderate, or high) from the supporting body of evidence (in vivo secondary outcomes and in vitro secondary outcomes) was considered for its indication that evidence exploring biological plausibility warranted an upgrade or downgrade of the preliminary hazard classification.

Software

The libraries were created in Endnote X7 (Thomson Reuters) and Excel 2010 (Microsoft Windows). The selection and data extraction was managed by the on-line software DistillerSR (Evidence Partners, Ottawa, CAD) and exported to Excel. Statistical meta-analysis was performed with Stata version 14 (StataCorp, College Station, TX, USA).

Results

Study Acquisition

Initially 5,024 articles were identified from PubMed (n=662 articles), Scopus (n=2,198 articles), and Embase (n=2,164 articles), which were reduced to 3,585 after manual removal of 1,439 duplicates (Figure 2). After screening the titles and abstracts, we retained 330 records for full-text screening, resulting in 39 full-text peer-reviewed articles retained for data extraction, which comprised 13 human studies, 19 in vivo studies, and 7 in vitro studies. Among the 13 human studies that met the eligibility criteria only 7 studies were retained for quantitative synthesis in the meta-analysis. We were unable to include 6 prospective studies (see Table S18) in the meta-analysis due to heterogeneity introduced by the reported outcomes, for example, beta estimates of BMI, and risk and trend estimates for overweight and obesity. In addition, there was no formal guidance for the qualitative synthesis of those studies in the NTP/OHAT handbook. For space considerations, the study characteristics were placed in the Supplemental Material data for each stream of evidence quantitated here, for example, human studies (see Table S8) and nonhuman studies (see Tables S19–S21).

Flow chart.
Figure 2. Flow chart of the systematic review process.

Evidence from Human Epidemiological Studies

Study characteristics.

The effect estimates initially considered for pooling the data were β regression coefficients for continuous outcomes, and risk ratios and/or odds ratios for dichotomous outcomes. However, the paucity of many possible exposure metrics and outcomes used in the studies led us to conduct the meta-analysis on the most prevalently reported exposure and outcome combination: β estimates for continuous models with BMI-z as a dependent variable and p,p′-DDE as an independent variable.

The population size varied among the seven included prospective human studies, ranging from 114 (Delvaux et al. 2014) to 788 participants (Cupul-Uicab et al. 2010). Most studies reported the results pooled for males and females, and only three studies provided stratified or independent results for each gender (Cupul-Uicab et al. 2010; Tang-Péronard et al. 2015; Warner et al. 2014). The rates of participation ranged from 36% (Høyer et al. 2014) to 91% (Cupul-Uicab et al. 2010). The cohorts were from the United States (Warner et al. 2014), Spain (Agay-Shay et al. 2015), Belgium (Delvaux et al. 2014), Greece (Vafeiadi et al. 2015), Greenland (Høyer et al. 2014), Poland (Høyer et al. 2014), Ukraine (Høyer et al. 2014), Mexico (Cupul-Uicab et al. 2010), and Denmark (Tang-Péronard et al. 2015).

Health outcome assessment.

The anthropometric measurements were performed by clinicians, primarily on children between 4 and 9 y of age (Agay-Shay et al. 2015; Delvaux et al. 2014; Høyer et al. 2014; Warner et al. 2014), though two studies focused on the early, up to 2.5 y (Cupul-Uicab et al. 2010), or later, 20 y (Tang-Péronard et al. 2015), stages of life. Given the early age of participants in most studies, obesity and/or overweight were often ascertained by means of standardized anthropometric measurements, such as the body mass index (BMI) z-scores. Due to the lack of unified criteria among clinicians, different reference charts and guidelines were used to calculate BMI-z scores, including standards such as the 2000 Centers for Disease Control and Prevention (CDC) Growth Charts (Cupul-Uicab et al. 2010; Warner et al. 2014), the International Obesity Task Force Growth Charts (Vafeiadi et al. 2015), or the British growth reference data (Delvaux et al. 2014).

Exposure assessment.

Most studies evaluated the association of obesity with prenatal exposure to p,p′-DDE in mothers’ sera, yet one study explored the associations with postnatal (at age 8–10 y) exposure to p,p′-DDE in the index children’s serum (Tang-Péronard et al. 2015). Most studies were performed in the framework of national biomonitoring programs, providing external references on the validity and analytical performance of methodological procedures. These methodologies commonly utilized gas chromatography–high resolution mass spectrometry, which is able to detect nearly 100% of p,p′-DDE. The exposure levels varied largely among studies (see Table S9). Among those studies reporting the exposure levels of p,p′-DDE standardized by lipid content, the median concentrations ranged between 1.1 ng/g (Warner et al. 2014) and 2,700 ng/g (Cupul-Uicab et al. 2010). For those studies reporting exposures by their wet-based values, the median exposure levels of p,p′-DDE presented narrower estimates ranging from 0.24 μg/L (Delvaux et al. 2014) to 1.9 μg/L (Vafeiadi et al. 2015). Beyond the absolute differences in the exposure levels between cohorts, especially large differences were noticed between the arbitrary boundaries of exposure and reference groups when comparing the different studies (see Table S9).

Lipid adjustment.

The complex relationships between the levels of DDTs and other lipophilic pollutants, serum lipids, and obesity are not fully understood (La Merrill et al. 2013), and researchers commonly infer assumptions about these relationships to formulate their causal models. The three most common approaches are to a) model the exposure levels in lipid basis (e.g., ratio of pollutant levels by the triglyceride and cholesterol content), b) include the blood lipid content as a covariate in the regression model, or c) use the unadjusted wet-weight values (Li et al. 2013). Some simulation studies revealed that the first approach (ratio chemical exposure by lipids) may bias the estimates compared with the other approaches (Gaskins and Schisterman 2009; Schisterman et al. 2005). However, there is no consensus on which is the best approach to apply in complex scenarios such as the obesogenic effect of DDTs, where a lipophilic compound is causally related with an obesity outcome and circulating lipid levels (Patel et al. 2012). The two main approaches were present among the studies included in the present meta-analysis, with five studies modeling the exposure concentration levels standardized by lipids (lw) as a ratio of p,p′-DDE/serum lipid levels (Agay-Shay et al. 2015; Cupul-Uicab et al. 2010; Høyer et al. 2014; Tang-Péronard et al. 2015; Warner et al. 2014), and two studies modeling the wet-weight (ww) levels of p,p′-DDE (Delvaux et al. 2014; Vafeiadi et al. 2015), while including the lipid content as a covariate (see Table S8).

Confounding bias.

The studies retained for meta-analysis addressed potential confounding bias by adjusting for known confounders in multivariate regression models (see Table S8). Most studies adjusted for maternal BMI, or occasionally for maternal weight and/or height. Most analyses also included adjustment for maternal age, education, parity, and breastfeeding and an indicator of socioeconomic status (race, education, income, social class, and/or socioeconomic index). Birth weight was also included in the model of two studies (Agay-Shay et al. 2015; Vafeiadi et al. 2015). Physical activity and/or diet were adjusted in models of three studies (Agay-Shay et al. 2015; Høyer et al. 2014; Tang-Péronard et al. 2015). Maternal smoking was modeled as a confounder in the majority of studies retained here (Cupul-Uicab et al. 2010; Delvaux et al. 2014; Høyer et al. 2014; Vafeiadi et al. 2015) with the exception of one study in which maternal smoking did not modify the effect estimate (Warner et al. 2014). One study concluded that risk of obesity associated with DDTs would be exacerbated by maternal smoking (Cupul-Uicab et al. 2010). Maternal alcohol consumption was included as a confounder in the regression models of one study (Høyer et al. 2014).

Sex.

The estrogenic effect of o,p′-DDT and the anti-androgenic effects of p,p-DDT and p,p′-DDE emphasize the potential effect modification of sex and most studies anticipated this by adjusting the model by sex (Agay-Shay et al. 2015; Delvaux et al. 2014; Høyer et al. 2014; Vafeiadi et al. 2015) and/or modeling the data stratified by sex after testing the interaction of sex with p,p′-DDE (Tang-Péronard et al. 2015; Warner et al. 2014; Delvaux et al. 2014). However, the interaction results of individual studies demonstrated no consistent sex-specific trends. Whereas two studies indicated males were more at risk of an association between DDTs and obesity measures (Tang-Péronard et al. 2015; Warner et al. 2014), another study indicated females were more at risk of an association between DDTs and obesity measures (Delvaux et al. 2014).

Results from the meta-analysis.

Seven studies reporting associations between blood p,p′-DDE and continuous BMI-z by means of adjusted β coefficients were included in the meta-analysis. The associations of BMI-z with the exposure to the other DDTs were by far less evaluated, and the meta-analysis of such subsamples was not feasible. Similarly, other health outcomes were reported (e.g., BMI, waist circumference, overweight, and obesity) by few studies with heterogeneous methodologies, making the meta-analysis underpowered and inaccurate (see Table S18). The pooled β coefficients for males and females were selected for all studies with the exception of Tang-Péronard et al. (2015), where a gender interaction was reported and thus was plotted individually. In the case of Delvaux et al. (2014), we detected a typo in the manuscript and contacted the authors who provided the correct estimate (β=0.22 instead of 0.95 BMI-z, 95% CI: −0.06, 0.51). When the studies provided different β values for different percentiles or tertiles of p,p′-DDE instead of continuous trends, we selected the highest estimate (worst-case scenario).

The overall association between exposure to p,p′-DDE and BMI-z was significantly positive with a β of 0.13 BMI-z per log increase of p,p′-DDE (95% CI: 0.01, 0.25 BMI-z; n=7 studies; Figure 3). The stratified analysis of units (lipid weight vs. wet weight) indicated the associations were on a similar scale across these units and suitable for pooling, supported by the global heterogeneity (I2 of 39.5%). A sensitivity analysis excluding one study each time showed the confidence intervals overlapped the null in five of the eight possible combinations (see Figure S5). Despite the low number of studies, the funnel plots did not show marked asymmetry and Egger’s test did not reveal statistically significant small-study effects (see Figure S6).

Forest plot.
Figure 3. Forest plot of the association between exposure to p,p′-DDE with BMI-z from human prospective studies, stratified by exposure units (lipid weight and wet weight). The effect size estimate is the adjusted coefficient regression (β) with 95% confidence intervals (units in BMI z-score per log increase of p,p′-DDE) for combined gender (males and female) unless the strata is specifically reported in the cohort label: (F) females or (M) males. Cohorts: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; EYHS, Danish part of the European Youth Heart Study; FLEHS I, first Flemish Environment and Health Study; RMCC, Rhea Mother–Child Cohort; INMA-Sabadell, Infancia y Medio-Ambiente Child and Environment birth cohort. Countries: GL, Greenland; SK, Warsaw Poland; UA, Ukraine; US, United States. Age: age at outcome assessment.
Confidence in the body of evidence and level of evidence.

The full rationale and results for rating the confidence in the body of evidence are provided in “4.2. RATING THE CONFIDENCE IN THE OF BODY OF EVIDENCE FROM HUMAN STUDIES AND LEVEL OF EVIDENCE FOR HEALTH EFFECT” in the Supplemental Material, and the results of the confidence rating process are summarized in Figure 1 for each stream of evidence considering those upgrading and downgrading factors, as well as the initial and final confidence rating decisions and integration of evidence. We provided a preliminary rating for the confidence with the body of evidence of “moderate” for human studies based on the intrinsic characteristics of observational prospective studies (OHAT 2015a).

We considered risk of bias, unexplained inconsistency, indirectness, imprecision, and publication bias among the downgrading factors (Figure 1). The rationale and results from risk of bias assessment from each individual study are described in Tables S10–S17). The risk of bias domains we most critically considered were confounding bias, performance bias, and detection bias.

We initially considered that confounding bias could be an issue because relevant confounding variables such as physical activity or diet were only included in three of eight models. However, we have not seen any evidence in the experimental literature indicating DDTs cause hyperphagia or sedentary activity (La Merrill 2014; Howell et al. 2015). Indeed, based on a preliminary search of the literature, we generated a directed acyclic graph (DAG) (see Figure S5) to select potential key confounders the maternal BMI, maternal smoking, and sex, which were controlled by all studies. Hence, we did not feel confounding bias was more substantial than allowed for in the penalization of the initial confidence rating to moderate.

We considered that another potential risk of bias was performance bias due to the extended use of single-pollutant models where simultaneous exposure to complex mixtures of xenogenous chemicals was reported by the authors or highly suspected. Only one study addressed this potential performance bias; principal component analysis identified an association of the DDE-containing component with both increased BMI z-score and risk of overweight, but no other components (Agay-Shay et al. 2015). Following the OHAT risk of bias rating tool (2015b), we only penalized studies that did not control for other exposures if the sample population had high exposure; however, only one study population was occupational or acutely contaminated (Warner et al. 2014). From this, we did not find that performance bias was a concerning bias domain.

Detection bias was also discussed because the studies estimated exposure from a single measurement; thus some risk of exposure misclassification could be suspected. Yet given the narrow windows of exposure, we judged the risk of bias to be low due to the high correlation of exposure estimates in biological samples collected serially across the prenatal and neonatal periods (Longnecker et al. 1999). Further, all studies used gas chromatography with mass spectrometry, the gold standard method to assess DDE levels. Overall, we classified most human studies in the Tier 1 of risk bias because we did not suspect bias among the other domains, and the key domains of confounding and detection bias were judged as having definitively or probably low risk of bias, being that the overall risk of bias was considered “not likely.”

The between-studies low heterogeneity (I2 39.5%) and variance (τ2<0.013) were not considered concerning enough to downgrade the confidence for unexplained inconsistency. Some inconsistencies may be explained by the sex-stratified results of Tang-Péronard et al. (2015), the only study that reported a statistically significant interaction of exposure effects with sex effects. We did not penalize the confidence rating concerning indirectness because the human studies prospectively assessed obesity and adiposity outcomes associated with exposures to DDTs. Despite large differences between the exposure groups and reference levels across studies, the narrow confidence intervals of the meta-estimates indicated no evidence for a lack of precision on the meta-estimates; hence we concluded unexplained imprecision was not serious enough to downgrade. The funnel plots did not show asymmetry; however, considering the absence of private funding or conflict of interest, as well as, the lack of potentially unpublished studies (e.g., conference abstracts, gray literature), we determined publication bias was not serious.

Another source of concern was the potential selection bias associated with the exclusion of six studies (see Table S18) from the meta-analysis solely because their outcome metrics differed from the seven included studies. Among the six different studies reporting results from five different cohorts, results from four cohorts indicated some positive associations between p,p′-DDE and measures of adiposity in both children and adults (Michigan fisheaters, Faroe Islands, PIVUS, and AMICS-INMA-Menorca), whereas null associations were reported in one cohort (CPP). Overall we had no reason to suspect that those results could threaten the confidence in the body of evidence included in the present study or reveal new insights on the direction and magnitude of our estimates.

We also considered those factors that may upgrade the confidence, such as the magnitude, dose response, residual confounding, and consistency across populations (Figure 1). We concluded that the magnitude of the effect was too modest to justify upgrading that confidence rating (see Figure S5). The presence and shape of a dose–response trend was inconsistent across studies (see Table S9). This may reflect a nonmonotonic trend and/or wide variability in the levels of DDTs used to define boundaries of reference and exposure groups across studies. We were concerned about potential residual confounding caused by lipid adjustment of exposure levels given that lipid adjusting the levels of contaminants has been demonstrated to provide more biased results than those models using wet values and including the lipid concentration as a covariate in the model (Gaskins and Schisterman 2009; Schisterman et al. 2005). However, if we consider the in vivo results further expanded upon below (Figure 4), we see inconclusive support of the hypothesis that abnormal blood lipids are in the causal pathway between DDTs and obesity (see Figure S7). For instance, only half of the experimental evidence available demonstrated positive relationships between DDTs and blood triglycerides and cholesterol (Figure 4A) in spite of the consistent lipid disruption in liver (Figure 4B); for these reasons, we did not upgrade the residual confounding or consistency.

red upward pointing triangle blue downward pointing triangle black circle Forest plot.
Figure 4. Forest plot of the associations between exposure to p,p′-DDT and p,p′-DDE and (A) energy balance, (B) circulating adipokines, (C) abnormal liver lipids, and (D) abnormal blood lipids from in vivo studies. Symbols: upward-pointing triangle, increase effect; downward-pointing triangle, decrease effect; circle, no statistical effect. Upward-pointing triangle means adversity of the health effect. Abbreviations: CBA, conjugated bile acids; CHO, cholesterol; DPC, day post-coitus; F, females; FAs, fatty acids; GD, gestational day; HFD, high-fat diet; LE, lipogenic enzymes; LW, liver weight; M, males; NS, no specified; PG, parental generation; PHO, phospholipids; PND, postnatal day; SD, Sprague-Dawley; T, temperature; TAG, triacylglycerol; TL, total lipids; TnG, transgenerational. Doses were approximated to the daily body weight basis using the conversion factors specified in Table S22.

After balancing the upgrading and downgrading factors, the final rating of the confidence with the body of evidence was finally appraised to be “moderate.” The results supported the direction of the association towards the “health effect,” thus we translated the confidence into a “moderate” level of evidence for the associations between exposure to p,p′-DDE and increased adiposity in humans (Figure 1).

Evidence from Primary in Vivo Studies

We retained two studies evaluating the associations between DDTs and adiposity (La Merrill et al. 2014; Skinner et al. 2013) as a primary body of evidence from in vivo studies (Table S19). Due to the low number of studies, we synthetized the results qualitatively instead of using meta-analysis.

One study evaluated adiposity longitudinally by EchoMRI™ after perinatal exposure to p,p′-DDT and o,p′-DDT (1.7 mg/kg bw, from gestational day 11.5 to postnatal day 5). Perinatal DDTs caused a transient increase in body and fat mass for several months in young adult female but not male mice, and no differences in female or male body composition when later fed high-fat diet. The effects of perinatal DDTs on adiposity were further explained by reductions in thermogenesis and energy expenditure. The metabolic disruption by perinatal DDTs was accompanied by dyslipidemia, glucose intolerance, hyperinsulinemia, and altered bile acid metabolism (La Merrill et al. 2014).

A transgenerational study performed with adult rats exposed to p,p′-DDT (25 and 50 mg/kg body weight) followed up obesity status in the subsequent three generations. The classification of obesity was established by an increase of body weight and abdominal adiposity. Among DDT-exposed lineages, only male and female offspring from the F3 generation and male offspring from the F4 generation had an increased prevalence of obesity. The authors concluded that the etiology of obesity in DDT-lineage rats may be in part due to environmentally induced transgenerational inheritance of differential DNA methylation regions in sperm (Skinner et al. 2013).

Confidence in the body of evidence and level of evidence.

The initial rating of the confidence in the body of primary evidence of experimental animal data was considered to be “high” (Figure 1; see also “4.3. RATING THE CONFIDENCE IN THE BODY OF PRIMARY EVIDENCE FROM IN VIVO STUDIES AND LEVEL OF EVIDENCE FOR HEALTH EFFECT” in the Supplemental Material), comparable to human randomized controlled trials where the exposure levels were controlled at individual level prior to the health outcomes, and using suitable control groups. As with the human evidence stream, we evaluated those factors that could modify this preliminary classification. Both studies were rated at low or probably low risk of bias for most of bias domains, which included their proper considerations of litter effects. The exemption was that one study was classified as probably high risk of bias in the sequence generation domain due to the lack of randomization of treatments (see Table S24). We judged the overall risk of bias to be of the “serious” risk of bias rating given that this rating has the criteria that most information was from Tier 1 and 2 studies, although plausible biases raise some doubt about the results (Figure 1; see also Table S2). The results from both experimental studies had some relevant inconsistencies. For instance, Skinner et al. (2013) observed obesity only in the third and fourth generations, whereas La Merrill et al. (2014) reported increased adiposity in the first generation. Inconsistencies in the methodological approaches (e.g., timing, dose, and route of exposure; rodent model) may explain these disparities; however, because there are only two studies, we concluded consistency is unknown in accordance with NTP/OHAT guidance (OHAT 2015a) and thus we did not downgrade due to inconsistency. According GRADE guidelines, downgrading by indirectness may be only justified when there is some compelling reason to suspect the different biology could modify the magnitude of the effect (Guyatt et al. 2011b). Both studies used rodent models (C57BL/6J mice and Hsd:Sprague-Dawley rats), which are considered directly applicable to human obesity, thus we rated indirectness as “not serious.” An acceptable number of animals per treatment and controls were used in both studies (n=15, La Merrill et al. 2014; n=30, Skinner et al. 2013), providing accurate estimates with narrow error bars; accordingly, we decided imprecision was not serious. Despite the limited number of studies, we judged publication bias was not serious enough to downgrade the confidence.

We further considered factors dictating an upgrade the initial rating of the confidence, but we concluded that the limited magnitude of the effect, the absence of dose-response analysis, and the absence of plausible residual confounding would not justify a decision to upgrade the confidence (see “4.3. RATING THE CONFIDENCE IN THE BODY OF PRIMARY EVIDENCE FROM IN VIVO STUDIES AND LEVEL OF EVIDENCE FOR HEALTH EFFECT” in the Supplemental Material). Despite the consistent direction of results in two mammalian species, and in turn with the human epidemiological results, we did not upgrade for consistency because of the limited number of available studies to conclude such a relationship.

Overall, we considered the main body of evidence from animal studies, assembled by only two studies, as having “serious” risk of bias. Thus, we downgraded the initial confidence, finally rating the confidence in the body of evidence as “moderate.” The nature or direction of the effect was to a health effect, thus the confidence was translated to a “moderate” level of evidence of obesogenic effects of DDTs in in vivo studies (Figure 1).

Evaluating the Support for Biological Plausibility: in Vivo Studies

We considered measures of energy imbalance such us body temperature and cold intolerance, as well as associated protein and RNA measures, as indicators of mechanisms underlying the potential causes of obesity. We considered tissue lipid levels as secondary outcomes in the supporting body of evidence as they are merely correlated with an obese state. Six studies were retained because of their reported associations with end points closely related to metabolic homeostasis such as energy imbalance and adipokines (Figure 4A,B). We retained 15 in vivo studies reporting associations between DDTs and abnormal lipids (Figure 4C,D), which is one of the main metabolic comorbidities associated with excessive body fat (Bays et al. 2013; Klop et al. 2013). Some studies simultaneously reported evidence of different outcomes giving a total count of 19 studies (Tables S20 and S21).

Markers of metabolic disruption.

We considered impaired energy expenditure and changes in circulating adipokines as markers of metabolic disruption associated with obesity (Figure 4A,B). Two of three studies found that exposure to DDTs decreased rectal temperature, a surrogate marker of thermogenesis that contributes 60–90% to total energy expenditure (Landsberg 2012), in two rodent species. Perinatal exposure to DDTs (1.7 mg/kg bw) decreased body temperature, energy expenditure and cold tolerance of female mice (La Merrill et al. 2014), consistent with risk of obesity. Those findings where mechanistically supported by the reduction of brown adipose mRNA expression of Ppargc1a, master regulator of thermogenesis, and Dio2, an upstream mediator of thermogenesis. In agreement with these results, obese Sprague-Dawley rats exposed to DDT (5.60 μg DDTs/kg body weight/d) exhibited lower core body temperature compared with the control during HFD feeding and subsequent 60% caloric restriction periods (Ishikawa et al. 2015). However in the third study, mice with acutely toxic DDT exposure (535–1,821 mg/kg) had no change in rectal temperature (Ahdaya et al. 1976). The exposure of C57BL/6H mice to p,p′-DDE at 0.4 and 2 mg/kg for 5 or 13 wk had no effect on the serum levels of adipokines closely related with energy balance such as leptin, adiponectin, or resistin (Howell et al. 2014; Howell et al. 2015).

Abnormal lipids.

We defined abnormal lipids as elevated lipids (cholesterol, triglycerides, or fatty acids) in blood or liver, increased liver weight (as surrogate measurement of hepatic steatosis), and increased activity of liver lipogenic enzymes. An overall positive association was seen between a wide range of DDTs and abnormal lipids consistently in the livers of rats and mice, where DDTs increased hepatic lipids, total liver weights, and lipogenic enzymes (Figure 4C). The majority of conflicting findings were clustered in blood (9 null of 16 data points in blood; Figure 4D), suggesting weaker evidence associating DDTs with dyslipidemia, particularly blood fatty acids. However, fatty acid composition and distribution in adipose tissue from Wistar rats was disrupted after oral exposure to p,p′-DDE (100 μg/kg/day) for 12 wk (Rodríguez-Alcalá et al. 2015). Lipid homeostasis was also disrupted by experimental treatment of DDTs in two nonmammalian systems, sailfin mollies and Japanese quail. For instance, the whole-body levels of total lipids and triglycerides in sailfin mollies were reduced at the highest exposure levels of 50 μg/L o,p′-DDT (Benton et al. 1994). The DDT isoform 1,1-di-p-chlorophenyl-2 chloroethylene (DDMU) increased the liver weights and hepatic triglycerides of Japanese quail (Westlake et al. 1979).

Confidence in the body of evidence and level of evidence.

We established a preliminary rating of “high” confidence with the body of evidence based on the features of animal study design (Figure 1; see also “4.4. RATING THE CONFIDENCE IN THE BODY OF SUPPORTING EVIDENCE FROM IN VIVO STUDIES”). Considering most studies have “probably high” risk of bias for randomization, concealment, and blinding when these methods are not reported, we downgraded the confidence based on risk of bias (Figure 1). The central role of energy imbalance in causing obesity and the close relationship of lipid abnormalities with adiposity were the main rationale to judge the directness sufficient. We did not consider imprecision serious and had no reason to suspect publication bias.

The available evidence for the increased hepatic lipids and impaired thermogenesis by DDTs was consistent across two mammalian species (Figure 4) and with the expectations from the positive meta-estimates of the associations of p,p′-DDE with adiposity in human studies (Figure 3). However, because of the lack of consistency of blood lipid disruption and absence of effects on adipokine levels, we did not upgrade due to consistency. There was no justification for upgrading confidence based on the modest magnitude of effects in most studies, the infrequent assessment of dose response, and unlikelihood of residual confounding. After assessing the different factors that may affect the confidence, we modified the initial confidence rating of “high” to “moderate” as the final rating of confidence and level of evidence for in vivo supporting evidence considering the direction of the effect to the presence of “health effects” (Figure 1).

Evaluating the Support for Biological Plausibility: in Vitro Studies

Typical phenotypic changes during pathological fat expansion may include increased adipogenesis (number of differentiated cells and/or quantity of fat accumulation), disruption of lipid metabolism (lipolytic and lipogenic processes) and disruption of adipokines involved in energy balance (e.g., leptin and adiponectin) (Bays et al. 2013). The most reported mechanism of action for obesogen compounds involve the disruption of PPAR-γ, which is considered a master regulator of adipogenesis and lipid homeostasis (Gore et al. 2015). Reflecting on these prior observations, we considered measures of enhanced adipogenic differentiation—as well as lipid, protein, and RNA measures associated with this process—as those mechanisms evidencing a potential cause of obesity.

Adipocyte differentiation and lipogenesis.

We retained seven references reporting associations between exposure to DDTs and outcomes related to adiposity using in vitro models (see Table S23). Exposure to p,p′-DDT consistently increased the adipogenic differentiation (Figure 5A) of 3T3-L1 preadipocytes (Moreno-Aliaga and Matsumura 2002) and mesenchymal stem cells (MSCs) (Strong et al. 2015). The presence of p,p′-DDT also initially accelerated the differentiation process of 3T3-F442A cells; however, their differentiation was not complete (Moreno-Aliaga and Matsumura 2002). The adipogenic effect of p,p′-DDT may relate to its estrogen receptor agonism (Nelson 1974), given that adipogenic differentiation of MSCs was strongly inhibited by the estrogen receptor inhibitor fulvestrant (Strong et al. 2015).

red upward pointing triangle</alt-text><alt-text>blue downward pointing triangle</alt-text><alt-text>black circle</alt-text><alt-text>Forest plot.
Figure 5. Forest plot of the association between exposure to p,p′-DDT and p,p′-DDE and (A) adipogenic differentiation, (B) expression of metabolic regulators, and (C) adipokines from in vitro studies. Symbols: upward-pointing triangle, increase effect; downward-pointing triangle, decrease effect; circle, no statistical effect. Upward-pointing triangle means adversity of the health effect. Abbreviations: ATGL, adipose triglyceride lipase; CEBP enhancer-binding protein; Fabp4, Fatty acid binding protein 4; Fasn, fatty acid synthase; Insig1, Insulin-induced gene-1; LpL, lipoprotein lipase; PPAR, peroxisome proliferator-activated receptor; Srebf1, sterol regulatory element-binding protein 1c.

Unlike p,p′-DDT, exposure to p,p′-DDE showed inconsistent effects on adipogenesis (Figure 5A). Although one study found insignificant effects on adipogenic differentiation up to 100 μM p,p′-DDE (Taxvig et al. 2012), two studies reported significant effects at low (0.01–2 μM) and high (10–100 μM) concentrations (Ibrahim et al. 2011; Mangum et al. 2015). Mangum et al. (2015) argued that some methodological limitations could justify the null results in their previous study (Howell and Mangum 2011). Furthermore, p,p′-DDE increased the fatty acid uptake in NIH3T3-L1 and increased the proliferation of human preadipocytes (Chapados et al. 2011; Howell and Mangum 2011).

Expression of metabolic regulators.

Overall, positive associations between exposure to either p,p′-DDT or p,p′-DDE and both adipokines and master regulators of adipogenesis were reported in mice preadipocytes and human MSCs (Figure 5B,C).

Consistent with increased adipogenesis, the master regulator of adipocyte differentiation PPAR-γ, was more highly expressed in p,p′-DDT treated cells than the controls in differentiated 3T3-L1 cells (Moreno-Aliaga and Matsumura 2002) and mesenchymal stem cells (Strong et al. 2015). The effect of p,p′-DDE was not consistent; whereas one study did not show statistically significant effects (Mangum et al. 2015), the other showed decreased activation at the highest doses (Taxvig et al. 2012). Srebf1 RNA, encoding the downstream target of PPAR-γ and mediator of adipogenic differentiation SREBP1C, was also overexpressed in cells treated with p,p′-DDE (Mangum et al. 2015). Protein C/EBP-α, considered with PPAR-γ the key transcription regulation factors in adipogenesis and lipogenesis, was also increased after incubation with p,p′-DDT, but the expression of C/EBP-β was unaffected (Moreno-Aliaga and Matsumura 2002). Above these doses of DDTs, the activation of PPAR-γ was reduced in NIH-3T3 cells (Taxvig et al. 2012).

The majority of studies found increased adipokine parameters in pre- and differentiated adipocytes by DDTs (Howell and Mangum 2011; Mangum et al. 2015; Taxvig et al. 2012), with the exception of one study whose authors found a decrease of resistin at the lowest concentration tested (5 μM) (Taxvig et al. 2012).

Confidence in the body of evidence and level of evidence.

Similar to the in vivo studies, we classified the body of evidence from in vitro studies with an initial high level of confidence based on the features of routine in vitro study designs (Figure 1; see also “4.5. RATING THE CONFIDENCE IN THE BODY OF SUPPORTING EVIDENCE FROM IN VITRO STUDIES” in the Supplemental Material). Among downgrading and upgrading factors, we noted a lack of consistency among the results of adipogenic differentiation caused by p,p′-DDE. Only half of the results showed statistically significant increases, and the positive results were not consistent acoss overlapping dosing concentrations (Ibrahim et al. 2011; Mangum et al. 2015; Taxvig et al. 2012). Similarly, lack of consistency extended to the effects of p,p′-DDE on mRNA expression of the main master regulator of adipogenic differentiation PPAR-γ (Figure 5B). We contrasted these p,p′-DDE inconsistencies with the generally consistent increased differentiation and Pparg expression with p,p′-DDT exposure that supported the main in vivo evidence. We decided to downgrade due to the inconsistency observed in p,p′-DDE (Figure 1) given that its relevance to the human stream of evidence and that risk of bias could not be assessed here but was deemed serious in all other experimental streams of evidence evaluated. The remaining downgrading and upgrading factors were not considered compelling enough to modify the overall evaluation (see “4.5. RATING THE CONFIDENCE IN THE BODY OF SUPPORTING EVIDENCE FROM IN VITRO STUDIES” in the Supplemental Material), including the applicability of the tested doses (see Figure S9). For example, although adipokine parameters were consistent across all in vitro studies, their lack of consistency with in vivo secondary outcomes precluded upgrade on this basis. The balance led to “moderate” as the final rating of confidence and level of evidence, accounting for the presence of “health effects” as the nature of the associations (Figure 1).

Integration of the Body Evidence and Hazard Identification

We first integrated the two streams of primary evidence—moderate level of human evidence and moderate level of in vivo evidence—and thus we classified p,p′-DDT and p,p′-DDE as “presumed” obesogens to humans (Figure 1). We applied a systematic approach to integrate the supporting evidence with the preliminary classification of the human and in vivo primary evidence. According our conceptual framework, a high or low level of supporting evidence of biological plausibility from in vitro and/or in vivo studies may justify upgrading or downgrading the preliminary classification, respectively. In this regard, the moderate supporting evidence from in vivo and in vitro studies did not justify any modification of the preliminary hazard classification (Figure 1). Thus, the final hazard identification conclusion was that p,p′-DDT and p,p′-DDE are “presumed” to be obesogenic in humans, based on a moderate level of primary human evidence, a moderate level of primary in vivo evidence, and a moderate level of evidence from in vivo and in vitro studies that supported the biological plausibility of the association.

Conclusions

Obesity is characterized by the expansion of adipose tissue mass, which is often accompanied by metabolic dysfunctions. Results from this meta-analysis, limited to prospective epidemiological studies, demonstrated a significant positive association between exposure to p,p′-DDE and adiposity. These epidemiological observations were integrated with experimental evidence of increased rodent adiposity, impaired energy expenditure, fatty liver, and adipogenic expansion that were estimated to fall within the range of the human exposures from prospective studies (see Figures S8 and S9).

The risk of obesity was observed among human populations exposed to p,p′-DDE, mainly during the prenatal period although one study also provided estimates from postnatal exposure. The increased risk of human obesity due to prenatal exposure to p,p′-DDE was also in agreement with in vivo primary evidence, demonstrating prenatal exposure to DDTs increases the adiposity of subsequent generations of rodents. In accordance with the primary in vivo evidence, the limited in vitro studies available reported higher adipogenic differentiation among preadipocytes exposed to p,p′-DDT but inconsistently so when exposed to p,p′-DDE.

Developmental exposure is one of the main pillars of the obesogenic hypothesis because the vulnerability of developing tissues to adult metabolic disease (Barker 1990; La Merrill and Birnbaum 2011). The latency of developmental exposure effects on obesity is postulated to result from impaired adipocyte differentiation (La Merrill et al. 2013) and/or epigenetic changes (Heindel et al. 2015), both evidenced here. For example, in the rats whose obesity was associated with the exposure of their ancestors to DDT, DNA methylation of sperm differed according to that trans-generational DDT exposure (Skinner et al. 2013). The potential human relevance of that provocative finding is suggested by the recent demonstration that both obesity and surgical weight loss also cause dramatic changes in the DNA methylation of sperm collected from men (Donkin et al. 2016).

Obesity is ultimately the result of energy imbalance, and energy expenditure via thermogenesis was impaired in two mammalian species exposed to DDTs. Although the mechanism(s) underlying this physiological phenomenon were sparsely studied, one in vivo study further supported biological plausibility by demonstrating a decrease in brown adipose RNA responsible for regulating thermogenesis (La Merrill et al. 2014).

Epidemiological Research Needs

The research on DDTs has focused on the associations between obesity outcomes and the major metabolite p,p′-DDE however comparatively little is known about the role of the primary commercially important parent compound p,p′-DDT. Given that p,p′-DDT exposure is primarily due to its manufacture or use, whereas p,p′-DDE exposure can be attributed to contamination of the environment and food supply, distinguishing the causal obesogen among them would have substantial implications for public health policies. The meta-analysis of dose–response profiles across populations remains analytically prohibitive with respect to the variable background exposure levels and increments between exposure categories. Improvements in both the quantitation and statistical analysis tools for the exposome can address these needs.

Several potential residual confounding factors should be considered in future models because of their relevant role in obesity etiology. For example, few epidemiological studies controlled for poor diet or sedentary lifestyle, two substantial obesity risk factors. However, we found no evidence that DDTs reduce exercise or cause hyperphagia (Howell et al. 2015; La Merrill et al. 2014). Future epidemiological studies should evaluate measures of energy intake and expenditure as potential confounders of the association between DDTs and obesity given the current paucity of this investigation.

Experimental Research Needs

There is a substantial need for further in vivo and in vitro mechanistic studies to demonstrate biological plausibility of the association between DDTs and obesity. For instance, the causal role of the endocrine system (e.g., insulin, thyroid, estrogen, and androgen axes (Chen et al. 1997; Kelce et al. 1995; La Merrill et al. 2014; Nelson 1974) on obesity associated with exposure to DDTs is uncharacterized despite evidence that both exposure and outcome are associated with these endocrine systems. Indeed mechanistic questions involving DDTs and obesity related outcomes should be conducted at numerous doses relevant to the human condition in multiple species. For example, in vitro studies of human cells are urged to demonstrate that mechanistic findings exhibited in mouse cells can operate in humans. Such efforts will allow for more rigorous dose–response analyses, and for stronger evidence of consistency.

Future experimental obesogen studies with animals must measure adiposity directly. The use of crude body weight has very limited applicability to characterize obesity in animal models if adiposity or other related outcomes are not measured and greatly limited the number of primary in vivo studies analyzed here (Nascimento et al. 2008; Woods et al. 2003). The body of evidence from animal studies addressing primary outcomes also needs to be strengthened by corroborating existing results, especially considering dose relevance and transgenerational outcomes. Better mechanistic characterization of the effects of DDTs on diseases often comorbid with obesity, for example, type 2 diabetes, dyslipidemia, and hepatic steatosis, will also be critical in informing causal models underlying epidemiological analyses.

Primary Limitations and Strengths of This Study

Limitations
  • This study relied on a search strategy designed to address multiple outcomes from multiple streams of evidence, focusing on high sensitivity rather specificity. More specific search strategies may better characterize mechanistic pathways; nonetheless, we did not suspect we missed relevant publications.
  • The paucity of prospective epidemiological data on DDTs and adiposity outcomes did not allow us to parse out robust stratified meta-analysis. For instance, comparing effect modification between DDTs, lipid adjustment, and exposure and outcome windows.
  • The variable and narrow range of exposure to DDTs across studied human populations is likely a poor reflection of the entire dose-response relationship. The impact this has on variability in the defined range of reference groups, the absence of p,p′-DDT data in humans, and the possibility of non-monotonic dose–response among populations could underestimate the meta-estimate effect size.
  • Judgmental inference was required to rate the confidence and integrate evidence. Potential subjective influence was minimized by critical review of multiple coauthors iteratively until consensus was reached.
  • Risk of bias tools and guidelines were unavailable for in vitro studies. They are needed for applying the evidence-based framework to in vitro data here and to the growing body of evidence from high-throughput screening programs.
Strengths
  • The meta-analysis of human evidence, limited to prospective studies determined with quantitative biomonitoring techniques, exhibited moderate heterogeneity and statistically significant positive associations.
  • Experimental evidence from in vivo studies, limited to adiposity as primary outcome, was consistent with impaired thermogenesis, a secondary outcome relevant to obesity etiology.
  • In this study, we applied a systematic and structured approach to data collection, data analysis, evidence rating, and integration using the GRADE-based NTP/OHAT protocol to draw hazard identification conclusions on the obesogenic effects of DDTs; that increases the rigor, transparency, and reproducibility.

To the best of our knowledge, this is the first study to systematically integrate evidence about the obesogenic effects of the pesticide DDT and its metabolite DDE. We integrated different streams of evidence from human, primary in vivo, and secondary in vivo and in vitro studies, and determined that each provides a moderate level of evidence supporting our conclusion that DDT and DDE exposures during the developmental period can be classified as “presumed” human obesogens. This is essential to inform decisions in the ongoing cost–benefit debate of the continued use of DDT as an insecticide (Conis 2010). Further, this study also highlights metabolic disruption triggered by environmental pollutants as a novel end point to be considered in risk assessment frameworks. Finally, it has been estimated that the annual economic impact of obesity as a consequence of exposure to DDT and DDE is 62 million USD in the European Union and United States (Attina et al. 2016). Thus, policy makers should also consider the preventive strategies reducing the exposure to obesogen compounds in overall disease and budget management.

Acknowledgments

We are grateful for the technical support of K. Thayer on the implementation of the National Toxicology Program/Ongoing Methods Development Activities (NTP/OHAT) framework and for the feedback of the anonymous reviewers.

This work is supported by the U.S. Department of Agriculture’s National Institute of Food and Agriculture (Hatch project 1002182), the California Office of Environmental Health Hazard Assessment (award 13-E0014-1), and the National Institutes of Health (grant ES019919).

References

Agay-Shay K, Martinez D, Valvi D, Garcia-Esteban R, Basagaña X, Robinson O, et al. 2015. Exposure to endocrine-disrupting chemicals during pregnancy and weight at 7 years of age: a multi-pollutant approach. Environ Health Perspect 123:1030–1037, PMID: 25956007, 10.1289/ehp.1409049.

Ahdaya SM, Shah PV, Guthrie FE. 1976. Thermoregulation in mice treated with parathion, carbaryl, or DDT. Toxicol Appl Pharmacol 35:575–580, PMID: 817420.

Allison DB, Downey M, Atkinson RL, Billington CJ, Bray GA, Eckel RH, et al. 2008. Obesity as a disease: a white paper on evidence and arguments commissioned by the Council of the Obesity Society. Obesity (Silver Spring) 16:1161–1177, 10.1038/oby.2008.231.

Appleton BS, Shriver CN, Arnrich L, Hathcock JN. 1981. Effects of 3-methylcholanthrene and DDT on cholesterol 7 alpha-hydroxylase in rats. Durg-Nutr Interact 1(1):15–21, PMID: 6926812.

Arnold M, Pandeya N, Byrnes G, Renehan AG, Stevens GA, Ezzati M, et al. 2015. Global burden of cancer attributable to high body-mass index in 2012: a population-based study. Lancet Oncol 16:36–46, PMID: 25467404, 10.1016/S1470-2045(14)71123-4.

ATSDR (Agency for Toxic Substances and Disease Registry). 2002. Toxicological profile for DDT, DDE, and DDD. Atlanta, GA:ATSDR.

Attina TM, Hauser R, Sathyanarayana S, Hunt PA, Bourguignon JP, Myers JP, et al. 2016. Exposure to endocrine-disrupting chemicals in the USA: a population-based disease burden and cost analysis. Lancet Diabetes Endocrinol 4(12):996–1003, PMID: 27765541, 10.1016/S2213-8587(16)30275-3.

Barker DJ. 1990. The fetal and infant origins of adult disease. BMJ 301:1111, PMID: 2252919, 10.1136/bmj.301.6761.1111.

Bays HE, Toth PP, Kris-Etherton PM, Abate N, Aronne LJ, Brown WV, et al. 2013. Obesity, adiposity, and dyslipidemia: a consensus statement from the National Lipid Association. J Clin Lipidol 7:304–383, PMID: 23890517, 10.1016/j.jacl.2013.04.001.

Benton MJ, Nimrod AC, Benson WH. 1994. Evaluation of growth and energy storage as biological markers of DDT exposure in sailfin mollies. Ecotoxicol Environ Saf 29:1–12, PMID: 7529156.

Cal/EPA-OEHHA (California Environmental Protection Agency–Office of Environmental Health Hazard Assessment, Safe Drinking Water and Toxic Enforcement Act of 1986). 2016. Chemicals Known to the State to Cause Cancer or Reproductive Toxicity.

Chapados NA, Casimiro C, Batal M, Blais JM, Haman F, Robidoux MA, et al. 2011. Increased proliferative effect of organochlorine compounds on human preadipocytes [Abstract]. Can J Diabetes 35:213, 10.1016/S1499-2671(11)52271-3.

Chen CW, Hurd C, Vorojeikina DP, Arnold SF, Notides AC. 1997. Transcriptional activation of the human estrogen receptor by DDT isomers and metabolites in yeast and MCF-7 cells. Biochem Pharmacol 53:1161–1172, PMID: 9175721.

Conis E. 2010. Debating the health effects of DDT: Thomas Jukes, Charles Wurster, and the fate of an environmental pollutant. Public Health Rep 125(2):337–342, PMID: 20297762, 10.1177/003335491012500224.

Cupul-Uicab LA, Hernández-Avila M, Terrazas-Medina EA, Pennell ML, Longnecker MP. 2010. Prenatal exposure to the major DDT metabolite 1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene (DDE) and growth in boys from Mexico. Environ Res 110:595–603, PMID: 20566194, 10.1016/j.envres.2010.06.001.

Darsie J, Gosha SK, Holman RT. 1976. Induction of abnormal fatty acid metabolism and essential fatty acid deficiency in rats by dietary DDT. Arch Biochem Biophys 175(1):262–269, PMID: 952519.

Delvaux I, Van Cauwenberghe J, Den Hond E, Schoeters G, Govarts E, Nelen V, et al. 2014. Prenatal exposure to environmental contaminants and body composition at age 7–9 years. Environ Res 132:24–32, PMID: 24742724, 10.1016/j.envres.2014.03.019.

DerSimonian R, Laird N. 1986. Meta-analysis in clinical trials. Control Clin Trials 7:177–188, PMID: 3802833.

Donkin I, Versteyhe S, Ingerslev LR, Qian K, Mechta M, Nordkap L, et al. 2016. Obesity and bariatric surgery drive epigenetic variation of spermatozoa in humans. Cell Metab 23:369–378, PMID: 26669700, 10.1016/j.cmet.2015.11.004.

Drenowatz C. 2015. Reciprocal compensation to changes in dietary intake and energy expenditure within the concept of energy balance. Adv Nutr 6:592–599, 10.3945/an.115.008615.

Fu Y, Luo N, Klein RL, Garvey WT. 2005. Adiponectin promotes adipocyte differentiation, insulin sensitivity, and lipid accumulation. J Lipid Res 46(7):1369–1379, PMID: 15834118, 10.1194/jlr.M400373-JLR200.

Gaskins AJ, Schisterman EF. 2009. The effect of lipid adjustment on the analysis of environmental contaminants and the outcome of human health risks. Methods Mol Biol 580:371–381, PMID: 19784610, 10.1007/978-1-60761-325-1_20.

Gore AC, Chappell VA, Fenton SE, Flaws JA, Nadal A, Prins GS, et al. 2015. EDC-2: the Endocrine Society’s Second Scientific Statement on Endocrine-Disrupting Chemicals. Endocr Rev 36:E1–E150, PMID: 26544531, 10.1210/er.2015-1010.

Grün F, Blumberg B. 2006. Environmental obesogens: organotins and endocrine disruption via nuclear receptor signaling. Endocrinology 147(6 suppl):S50–S55, PMID: 16690801, 10.1210/en.2005-1129.

Guyatt GH, Oxman AD, Schünemann HJ, Tugwell P, Knottnerus A. 2011a. GRADE guidelines: a new series of articles in the Journal of Clinical Epidemiology. J Clin Epidemiol 64:380–382, PMID: 21185693, 10.1016/j.jclinepi.2010.09.011.

Guyatt GH, Oxman AD, Sultan S, Glasziou P, Akl EA, Alonso-Coello P, et al. 2011b. GRADE guidelines: 9. Rating up the quality of evidence. J Clin Epidemiol 64:1311–1316, PMID: 21802902, 10.1016/j.jclinepi.2011.06.004.

Harbord RM, Egger M, Sterne JA. 2006. A modified test for small-study effects in meta-analyses of controlled trials with binary endpoints. Stat Med 25:3443–3457, PMID: 16345038, 10.1002/sim.2380.

Heindel JJ, vom Saal FS, Blumberg B, Bovolin P, Calamandrei G, Ceresini G, et al. 2015. Parma consensus statement on metabolic disruptors. Environ Health 14:54, 10.1186/s12940-015-0042-7.

Higgins JPT, Green S. 2011. Cochrane Handbook for Systematic Reviews of Interventions, Version 5.1.0. http://handbook.cochrane.org/ [accessed 15 August 2015].

Howell G III, Mangum L. 2011. Exposure to bioaccumulative organochlorine compounds alters adipogenesis, fatty acid uptake, and adipokine production in NIH3T3-L1 cells. Toxicol In Vitro 25:394–402, 10.1016/j.tiv.2010.10.015.

Howell GE III, Meek E, Kilic J, Mohns M, Mulligan C, Chambers JE. 2014. Exposure to p,p′-dichlorodiphenyldichloroethylene (DDE) induces fasting hyperglycemia without insulin resistance in male C57BL/6H mice. Toxicology 320:6–14, 10.1016/j.tox.2014.02.004.

Howell GE III, Mulligan C, Meek E, Chambers JE. 2015. Effect of chronic p,p′-dichlorodiphenyldichloroethylene (DDE) exposure on high fat diet-induced alterations in glucose and lipid metabolism in male C57BL/6H mice. Toxicology 328:112–122, PMID: 25541407, 10.1016/j.tox.2014.12.017.

Høyer BB , Ramlau-Hansen CH, Henriksen TB, Pedersen HS, Góralczyk K, Zviezdai V, et al. 2014. Body mass index in young school-age children in relation to organochlorine compounds in early life: a prospective study. Int J Obes (Lond) 38:919–925, PMID: 24718355, 10.1038/ijo.2014.58.

I’Allemand D, Wiegand S, Reinehr T, Müller J, Wabitsch M, Widhalm K, et al. 2008. Cardiovascular risk in 26,008 European overweight children as established by a multicenter database. Obesity (Silver Spring) 16(7):1672–1679, PMID: 18451769, 10.1038/oby.2008.259.

Ibrahim MM, Fjære E, Lock EJ, Naville D, Amlund H, Meugnier E, et al. 2011. Chronic consumption of farmed salmon containing persistent organic pollutants causes insulin resistance and obesity in mice. PloS One 6:e25170, PMID: 21966444, 10.1371/journal.pone.0025170.

Ishikawa T, Graham JL, Stanhope KL, Havel PJ, La Merrill MA. 2015. Effect of DDT exposure on lipids and energy balance in obese Sprague-Dawley rats before and after weight loss. Toxicol Rep 2:990–995, 10.1016/j.toxrep.2015.07.005.

Johnson PI, Sutton P, Atchley DS, Koustas E, Lam J, Sen S, et al. 2014. The Navigation Guide—evidence-based medicine meets environmental health: systematic review of human evidence for PFOA effects on fetal growth. Environ Health Perspect 122:1028–1039, PMID: 24968388, 10.1289/ehp.1307893.

Jokinen E. 2015. Obesity and cardiovascular disease. Minerva Pediatr 67:25–32, PMID: 25387321.

Kelce WR, Stone CR, Laws SC, Gray LE, Kemppainen JA, Wilson EM. 1995. Persistent DDT metabolite p,p′-DDE is a potent androgen receptor antagonist. Nature 375:581–585, PMID: 7791873, 10.1038/375581a0.

Klimentidis YC, Beasley TM, Lin HY, Murati G, Glass GE, Guyton M, et al. 2011. Canaries in the coal mine: a cross-species analysis of the plurality of obesity epidemics. Proc Biol Sci 278:1626–1632, 10.1098/rspb.2010.1890.

Klop B, Elte JW, Cabezas MC. 2013. Dyslipidemia in obesity: mechanisms and potential targets. Nutrients 5:1218–1240, PMID: 23584084, 10.3390/nu5041218.

Koustas E, Lam J, Sutton P, Johnson PI, Atchley DS, Sen S, et al. 2014. The Navigation Guide—evidence-based medicine meets environmental health: systematic review of nonhuman evidence for PFOA effects on fetal growth. Environ Health Perspect 122:1015–1027, PMID: 24968374, 10.1289/ehp.1307177.

La Merrill M, Birnbaum LS. 2011. Childhood obesity and environmental chemicals. Mt Sinai J Med 78:22–48, PMID: 21259261, 10.1002/msj.20229.

La Merrill M, Emond C, Kim MJ, Antignac JP, Le Bizec B, Clément K, et al. 2013. Toxicological function of adipose tissue: focus on persistent organic pollutants. Environ Health Perspect 121:162–169, PMID: 23221922, 10.1289/ehp.1205485.

La Merrill M, Karey E, Moshier E, Lindtner C, La Frano MR, Newman JW, et al. 2014. Perinatal exposure of mice to the pesticide DDT impairs energy expenditure and metabolism in adult female offspring. PloS One 9:e103337, PMID: 25076055, 10.1371/journal.pone.0103337.

Lam J, Koustas E, Sutton P, Johnson PI, Atchley DS, Sen S, et al. 2014. The Navigation Guide—evidence-based medicine meets environmental health: integration of animal and human evidence for PFOA effects on fetal growth. Environ Health Perspect 122:1040–1051, PMID: 24968389, 10.1289/ehp.1307923.

Landsberg L. 2012. Core temperature: a forgotten variable in energy expenditure and obesity? Obes Rev 13 (suppl 2):97–104, 10.1111/j.1467-789X.2012.01040.x.

Lee DH, Porta M, Jacobs DR Jr, Vandenberg LN. 2014. Chlorinated persistent organic pollutants, obesity, and type 2 diabetes. Endocr Rev 35:557–601, 10.1210/er.2013-1084.

Li D, Longnecker MP, Dunson DB. 2013. Lipid adjustment for chemical exposures: accounting for concomitant variables. Epidemiology 24:921–928, PMID: 24051893, 10.1097/EDE.0b013e3182a671e4.

Longnecker MP, Klebanoff MA, Gladen BC, Berendes HW. 1999. Serial levels of serum organochlorines during pregnancy and postpartum. Arch Environ Health 54:110–114, PMID: 10094288, 10.1080/00039899909602244.

Loomis D, Guyton K, Grosse Y, El Ghissasi F, Bouvard V, Benbrahim-Tallaa L, et al. 2015. Carcinogenicity of lindane, DDT, and 2,4-dichlorophenoxyacetic acid. Lancet Oncol 16:891–892, PMID: 26111929, 10.1016/S1470-2045(15)00081-9.

Mangum LH, Howell GE III, Chambers JE. 2015. Exposure to p,p′-dde enhances differentiation of 3T3-L1 preadipocytes in a model of sub-optimal differentiation. Toxicol Lett 238:65–71, 10.1016/j.toxlet.2015.07.009.

Martinez JA. 2000. Body-weight regulation: causes of obesity. Proc Nutr Soc 59:337–345, PMID: 10997649.

Moreno-Aliaga MJ, Matsumura F. 2002. Effects of 1,1,1-trichloro-2,2-bis(p-chlorophenyl)-ethane (p,p′-DDT) on 3T3-L1 and 3T3-F442A adipocyte differentiation. Biochem Pharmacol 63:997–1007, PMID: 11911853.

Morgan RL, Thayer KA, Bero L, Bruce N, Falck-Ytter Y, Ghersi D, et al. 2016. GRADE: assessing the quality of evidence in environmental and occupational health. Environ Int 92-93:611–616, PMID: 26827182, 10.1016/j.envint.2016.01.004.

Narayan S, Dani HM, Misra UK. 1990. Changes in lipid profiles of liver microsomes of rats following intratracheal administration of DDT or endosulfan. J Environ Sci Health Part B Pestic Food Contam Agric Wastes 25(2):243–257, PMID: 2380487, 10.1080/10934529009375554.

Nascimento AF, Sugizaki MM, Leopoldo AS, Lima-Leopoldo AP, Nogueira CR, Novelli EL, et al. 2008. Misclassification probability as obese or lean in hypercaloric and normocaloric diet. Biol Res 41:253–259, PMID: 19399338, 10.4067//S0716-97602008000300002.

Nelson JA. 1974. Effects of dichlorodiphenyltrichloroethane (DDT) analogs and polychlorinated biphenyl (PCB) mixtures on 17β-(3H)estradiol binding to rat uterine receptor. Biochem Pharmacol 23:447–451, PMID: 4360348.

Ogden CL, Carroll MD, Kit BK, Flegal KM. 2014. Prevalence of childhood and adult obesity in the United States, 2011–2012. JAMA 311:806–814, PMID: 24570244, 10.1001/jama.2014.732.

OHAT (Office of Health Assessment and Translation). 2015a. Handbook for Conducting a Literature-Based Health Assessment Using OHAT Approach for Systematic Review and Evidence Integration. Washington, DC:National Toxicology Program.

OHAT. 2015b. OHAT Risk of Bias Rating Tool for Human and Animal Studies. Washington, DC:National Toxicology Program.

Okazaki Y, Katayama T. 2003. Effects of dietary carbohydrate and myo-inositol on metabolic changes in rats fed 1,1,1-trichloro-2,2-bis (p-chlorophenyl) ethane (DDT). J Nutr Biochem 14(2):81–89, PMID: 12667599.

Okazaki Y, Katayama T. 2008. Dietary inositol hexakisphosphate, but not myo-inositol, clearly improves hypercholesterolemia in rats fed casein-type amino acid mixtures and 1,1,1-trichloro-2,2-bis (p-chlorophenyl) ethane. Nutr Res 28(10):714–721, PMID: 19083479, 10.1016/j.nutres.2008.07.003.

Patel CJ, Cullen MR, Ioannidis JP, Butte AJ. 2012. Systematic evaluation of environmental factors: persistent pollutants and nutrients correlated with serum lipid levels. Int J Epidemiol 41:828–843, PMID: 22421054, 10.1093/ije/dys003.

Platt DS, Cockrill BL. 1967. Liver enlargement and hepatoxicity: an investigation into the effects of several agents on rat liver enzyme activities. Biochem Pharmacol 16(12):2257–2270, PMID: 4294662.

Platt DS, Cockrill BL. 1969. Biochemical changes in rat liver in response to treatment with drugs and other agents. II. Effects of halothane, DDT, other chlorinated hydrocarbons, thioacetamide, dimethylnitrosamine and ethionine. Biochem Pharmacol 18(2):445–457, PMID: 4305095.

Rao TN, Sanyal S, Agarwal N, Subrahmanyam D. 1981. Effect of a single oral dose of DDT on lipid metabolism in rat. Toxicol Lett 7(3):263–266, PMID: 7222101.

Rodríguez-Alcalá LM, Sá C, Pimentel LL, Pestana D, Teixeira D, Faria A, et al. 2015. Endocrine disruptor DDE associated with a high-fat diet enhances the impairment of liver fatty acid composition in rats. J Agric Food Chem 63:9341–9348, PMID: 26449595, 10.1021/acs.jafc.5b03274.

Rogan WJ, Chen A. 2005. Health risks and benefits of bis(4-chlorophenyl)-1,1,1-trichloroethane (DDT). Lancet 366:763–773, PMID: 16125595, 10.1016/S0140-6736(05)67182-6.

Rooney AA, Boyles AL, Wolfe MS, Bucher JR, Thayer KA. 2014. Systematic review and evidence integration for literature-based environmental health science assessments. Environ Health Perspect 122:711–718, PMID: 24755067, 10.1289/ehp.1307972.

Rooney AA, Cooper GS, Jahnke GD, Lam J, Morgan RL, Boyles AL, et al. 2016. How credible are the study results? Evaluating and applying internal validity tools to literature-based assessments of environmental health hazards. Environ Int 92–93:617–629, PMID: 26857180, 10.1016/j.envint.2016.01.005.

Schisterman EF, Whitcomb BW, Buck Louis GM, Louis TA. 2005. Lipid adjustment in the analysis of environmental contaminants and human health risks. Environ Health Perspect 113:853–857, PMID: 16002372, 10.1289/ehp.7640.

Sheehan MC, Lam J. 2015. Use of systematic review and meta-analysis in environmental health epidemiology: a systematic review and comparison with guidelines. Curr Environ Health Rep 2:272–283, PMID: 26231504, 10.1007/s40572-015-0062-z.

Skinner MK, Manikkam M, Tracey R, Guerrero-Bosagna C, Haque M, Nilsson EE. 2013. Ancestral dichlorodiphenyltrichloroethane (DDT) exposure promotes epigenetic transgenerational inheritance of obesity. BMC Med 11:228, 10.1186/1741-7015-11-228.

Smith D. 1999. Worldwide trends in DDT levels in human breast milk. Int J Epidemiol 28:179–188, PMID: 10342677.

Speakman JR, O’Rahilly S. 2012. Fat: an evolving issue. Dis Model Mech 5:569–573, PMID: 22915015, 10.1242/dmm.010553.

Stern JH, Rutkowski JM, Scherer PE. 2016. Adiponectin, leptin, and fatty acids in the maintenance of metabolic homeostasis through adipose tissue crosstalk. Cell Metab 23:770–784, PMID: 27166942, 10.1016/j.cmet.2016.04.011.

Strong AL, Shi Z, Strong MJ, Miller DF, Rusch DB, Buechlein AM, et al. 2015. Effects of the endocrine-disrupting chemical DDT on self-renewal and differentiation of human mesenchymal stem cells. Environ Health Perspect 123:42–48, PMID: 25014179, 10.1289/ehp.1408188.

Tang-Péronard JL, Jensen TK, Andersen HR, Ried-Larsen M, Grøntved A, Andersen LB, et al. 2015. Associations between exposure to persistent organic pollutants in childhood and overweight up to 12 years later in a low exposed Danish population. Obes Facts 8:282–292, 10.1159/000438834.

Taxvig C, Dreisig K, Boberg J, Nellemann C, Schelde AB, Pedersen D, et al. 2012. Differential effects of environmental chemicals and food contaminants on adipogenesis, biomarker release and PPARγ activation. Mol Cell Endocrinol 361:106–115, PMID: 22526026, 10.1016/j.mce.2012.03.021.

Taylor KW, Novak RF, Anderson HA, Birnbaum LS, Blystone C, Devito M, et al. 2013. Evaluation of the association between persistent organic pollutants (POPs) and diabetes in epidemiological studies: a National Toxicology Program workshop review. Environ Health Perspect 121:774–783, 10.1289/ehp.1205502.

Thayer KA, Heindel JJ, Bucher JR, Gallo MA. 2012. Role of environmental chemicals in diabetes and obesity: a National Toxicology Program workshop review. Environ Health Perspect 120:779–789, PMID: 22296744, 10.1289/ehp.1104597.

UNEP (United Nations Environment Programme). 2010. Report of the Expert Group on the Assessment of the Production and Use of DDT and Its Alternatives for Disease Vector Control. Third Meeting. UNEP/POPS/COP.3/24. Geneva, Switzerland:
United Nations.

Vafeiadi M, Georgiou V, Chalkiadaki G, Rantakokko P, Kiviranta H, Karachaliou M, et al. 2015. Association of prenatal exposure to persistent organic pollutants with obesity and cardiometabolic traits in early childhood: the Rhea mother–child cohort (Crete, Greece). Environ Health Perspect 123:1015–1021, 10.1289/ehp.1409062.

Warner M, Wesselink A, Harley KG, Bradman A, Kogut K, Eskenazi B. 2014. Prenatal exposure to dichlorodiphenyltrichloroethane and obesity at 9 years of age in the CHAMACOS study cohort. Am J Epidemiol 179:1312–1322, PMID: 24722999, 10.1093/aje/kwu046.

Westlake GE, Bunyan PJ, Stanley PI, Walker CH. 1979. The effects of 1,1-di(p-chlorophenyl)-2-chloroethylene on plasma enzymes and blood constituents in the Japanese quail. Chem Biol Interact 25:197–210, PMID: 466732.

WHO (World Health Organization). 2014. Global Status Report on Noncommunicable Diseases 2014. Geneva, Switzerland:WHO.

Woodruff TJ, Sutton P. 2014. The Navigation Guide systematic review methodology: a rigorous and transparent method for translating environmental health science into better health outcomes. Environ Health Perspect 122:1007–1014, PMID: 24968373, 10.1289/ehp.1307175.

Woods SC, Seeley RJ, Rushing PA, D’Alessio D, Tso P. 2003. A controlled high-fat diet induces an obese syndrome in rats. J Nutr 133:1081–1087, PMID: 12672923.

Lifelong Residential Exposure to Green Space and Attention: A Population-based Prospective Study

Author Affiliations open

1ISGlobal, CREAL (Centre for Research in Environmental Epidemiology), Barcelona, Spain

2Universitat Pompeu Fabra, Barcelona, Spain

3CIBERESP (Centro de Investigación Biomédica en Red Epidemiología y Salud Pública), Madrid, Spain

4Epidemiology and Environmental Health Joint Research Unit, FISABIO (La Fundación para el Fomento de la Investigación Sanitaria y Biomédica de la Comunitat Valenciana)−Universitat Jaume I−Universitat de València, Valencia, Spain

5Agency for Health Quality and Assessment of Catalonia (AQuAS), Catalonia Ministry of Health, Spain

6Department of Medicine, University of Oviedo, Oviedo, Spain

7Faculty of Psychology, University of the Basque Country UPV/EHU, San Sebastián, Basque Country, Spain

8Biodonostia Health Research Institute, San Sebastian, Basque Country, Spain

9Servicio de Pediatría, Hospital San Agustín, Avilés, Spain

10Department of Child and Adolescent Psychiatry/Psychology, Erasmus University Medical Centre–Sophia Children’s Hospital, Rotterdam, The Netherlands

11Cancer Epidemiology Unit-CeRMS (Il Centro di Ricerca in Medicina Sperimentale), Department of Medical Sciences, University of Turin and CPO-Piemonte, Turin, Italy

12Sub-Directorate for Public Health of Guipúzcoa, Department of Health, Government of the Basque Country, San Sebastián, Basque Country, Spain

13IMIM-Parc Salut Mar, Barcelona, Spain

PDF icon PDF Version (345 KB)

  • Background:
    Natural environments, including green spaces, may have beneficial impacts on brain development. However, longitudinal evidence of an association between long-term exposure to green spaces and cognitive development (including attention) in children is limited.
    Objectives:
    We evaluated the association between lifelong residential exposure to green space and attention during preschool and early primary school years.
    Methods:
    This longitudinal study was based on data from two well-established population-based birth cohorts in Spain. We assessed lifelong exposure to residential surrounding greenness and tree cover as the average of satellite-based normalized difference vegetation index and vegetation continuous fields, respectively, surrounding the child’s residential addresses at birth, 4–5 y, and 7 y. Attention was characterized using two computer-based tests: Conners’ Kiddie Continuous Performance Test (K-CPT) at 4–5 y (n=888) and Attentional Network Task (ANT) at 7 y (n=987). We used adjusted mixed effects models with cohort random effects to estimate associations between exposure to greenness and attention at ages 4–5 and 7 y.
    Results:
    Higher lifelong residential surrounding greenness was associated with fewer K-CPT omission errors and lower K-CPT hit reaction time-standard error (HRT-SE) at 4–5 y and lower ANT HRT-SE at 7 y, consistent with better attention. This exposure was not associated with K-CPT commission errors or with ANT omission or commission errors. Associations with residential surrounding tree cover also were close to the null, or were negative (for ANT HRT-SE) but not statistically significant.
    Conclusion:
    Exposure to residential surrounding greenness was associated with better scores on tests of attention at 4–5 y and 7 y of age in our longitudinal cohort. https://doi.org/10.1289/EHP694
  • Received: 23 June 2016
    Revised: 20 July 2017
    Accepted: 21 July 2017
    Published: 18 September 2017

    Address correspondence to P. Dadvand, ISGlobal, Doctor Aiguader, 88, 08003, Barcelona, Spain. Telephone: 34 93 214 7317. Email: payam.dadvand@isglobal.org

    Supplemental Material is available online (https://doi.org/10.1289/EHP694).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (1.69 MB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (54 KB)

Introduction

It has been proposed that exposure to natural environments, which include green spaces, is important for normal neurodevelopment (Kahn and Kellert 2002; Kellert 2005). Natural en-vironments provide children with unique opportunities for engagement, discovery, risk-taking, creativity, mastery, and control, and for strengthening the child’s sense of self; in addition, they also may inspire basic emotional states (including a sense of wonder) and enhance psychological restoration, all of which may positively influence cognitive development and attention (Kahn and Kellert 2002; Kellert 2005; Bowler et al. 2010). Proximity to green spaces also may benefit cognitive development through indirect mechanisms involving increased physical activity (James et al. 2015), reduced exposure to air pollution and noise (Gidlöf-Gunnarsson and Öhrström 2007; Dadvand et al. 2012, 2015b), and exposure to an enriched microbial environment (Rook 2013), each of which may contribute to improved cognitive function in general and attention in particular (Fedewa and Ahn 2011; Klatte et al. 2013; Rook 2013; Sunyer et al. 2015). However, population-based evidence of the association between long-term exposure to green spaces and cognitive development and attention in children is limited (Dadvand et al. 2015a).

In a previous study of 2,593 children attending primary school in Barcelona (Dadvand et al. 2015a), exposure to surrounding greenness at enrollment was associated with greater progress in working memory and attention over a 12-mo period. However, children were evaluated at 7–10 years of age, after substantial cognitive development had already occurred, and we were not able to account for exposures during prenatal and early postnatal periods of rapid brain development that may be especially vulnerable to effects of environmental exposures (Grandjean and Landrigan 2014). Therefore, the aim of the present study was to evaluate longitudinal associations between lifelong residential exposure to greenness, including exposure during prenatal and early postnatal periods, and measures of attention during preschool and at 7 y of age.

Materials and Methods

Study Population

Our study was based on data from two well-established population-based birth cohorts that are part of the INMA (INfancia y Medio Ambiente; Childhood and Environment) network of birth cohorts in Spain. The overall goals of INMA are to identify biological, social, and environmental determinants of normal and abnormal growth, development, and health, from fetal life to adulthood (Guxens et al. 2012). The Sabadell and Valencia INMA cohorts are located in northeastern and eastern Spain, respectively. Both locations have a Mediterranean climate characterized by hot and dry summers, mild and rainy winters, and maximum vegetation between autumn and spring.

The data was collected prospectively during 2003–2013 for these two cohorts using the INMA common protocol (Guxens et al. 2012). Briefly, pregnant women who fulfilled the inclusion criteria [age ≥16 y, singleton pregnancy, no use of assisted reproductive techniques, intention to deliver at the reference hospital, and ability to speak and understand Spanish or a local language (e.g., Catalan)] were recruited during the first trimester of pregnancy at primary healthcare centers or public hospitals. A baseline survey was performed at enrollment (approximately 12 wk of pregnancy), and follow-up surveys were performed at 20 and 32 wk of pregnancy, at birth, and when children were 6 mo, 1 y, 2 y, 4 y, or 5 y (in Sabadell and Valencia, respectively), and 7 y of age. Additional information on the cohorts and data collection has been published elsewhere (Guxens et al. 2012). All participants gave written informed consent before enrollment in the cohorts. Each cohort obtained ethical approval from the ethical committee in its corresponding region.

Residential Surrounding Greenness

The assessment of residential surrounding greenness was based on two satellite-based indices of greenness: a) Normalized Difference Vegetation Index (NDVI) an indicator of greenness including all types of vegetation and b) Vegetation Continuous Fields (VCF), an indicator of tree canopy cover.

NDVI is based on land surface reflectance of visible (red) and near-infrared parts of spectrum (Weier 2011). Its values range between −1 and 1, with higher numbers indicating more greenness and negative values indicating water bodies, snow, and barren areas of rock and sand. VCF indicates the percentage of land (in each image pixel) covered by the woody vegetation with a height greater than five meters (Sexton et al. 2013). To develop NDVI and VCF maps for our study regions, we used Landsat data at 30 m×30 m resolution as detailed in Supplemental Materials (see Table S1 and Figures S1 and S2).

For each participant, we derived estimates of residential surrounding greenness (NDVI) and residential surrounding tree cover (VCF) within 100 m, 300 m, and 500 m buffer areas(representing immediate, intermediate, and neighborhood areas, respectively) surrounding the residential address at birth, at the 4–5 y follow-up, and at the 7-y follow-up, resulting in 18 estimates (3 time points×3 buffers×2 indices) for (Dadvand et al. 2012, 2014, 2015a, 2016). For each greenness index and buffer area, we derived lifelong exposure estimates at 4–5 y [the mean value of the index at birth, and at 4- or 5-y (for the Sabadell and Valencia cohorts, respectively)] and at 7 y (the mean value of the index at birth, 4–5 y, and 7-y).

Assessment of Attention

We used two computer-based tests to assess attention in INMA children: Conners’ Kiddie Continuous Performance Test (K-CPT) at 4 y of age for the Sabadell cohort and at 5 y of age for the Valencia cohort and Attentional Network Task (ANT) at 7 y of age for both cohorts.

The K-CPT.

The K-CPT (K–CPT™ v.5) is designed to characterize attention in children aged 4 to 7 y (Conners 2000). The K-CPT has been demonstrated to be a valid tool to characterize attention in comparison with clinical (Epstein et al. 2003; Homack and Riccio 2006) and parental evaluations (Barnard et al. 2015). To conduct this task, children were instructed to press the space bar when they saw any image on the computer screen except a ball. Three main outcomes of the K-CPT were used in our analyses: a) omission errors (e.g., the child failed to respond when she or he should); b) commission errors (e.g., the child responded when she or he should not); and c) hit reaction time–standard error (HRT-SE) (SE of RT for correct responses), a measure of response speed consistency throughout the test (Conners and Staff 2000). A higher HRT-SE indicates highly variable reactions related to inattentiveness.

The ANT.

The ANT is a task developed to assess attention in subjects older than 6 y (Rueda et al. 2004). To perform this test, children were asked to press the left or right key on the computer mouse, depending on whether the centrally located fish in a horizontal row of five yellow fish was pointing to the left or right. As for the K-CPT, we derived counts of omission errors and commission errors, and the HRT-SE, for each participant. We have previously shown that in a sample of ∼2,900 primary schoolchildren in Barcelona, the ANT indicators have statistical dependency with age, school performance, attention deficit/hyperactivity disorder (ADHD) clinical criteria, behavioral problems, and maternal education (Forns et al. 2014).

Statistical Analysis

Because of the multilevel nature of the data (i.e., children within cohorts), we used mixed effects models with attentional parameters as outcomes (one parameter for each test at a time), measures of exposure to green spaces (one at a time) as a fixed effect predictor, and the cohort as the random effect. Random intercepts were used to account for clustering of subjects into cohorts (Chu et al. 2011). For commission and omission errors (count data), we developed negative binomial mixed effects models and for HRT-SE (continuous data), we developed linear mixed effects models. The regression coefficients of negative binomial models were exponentiated to obtain mean ratios. Separate sets of models were developed for K-CPT and ANT with 4- or 5-year exposure measures being used for K-CPT analyses and 7-y exposure for ANT analyses. All models were further adjusted for the following covariates identified a priori: age (at the time of 4- or 5-y follow-up for the K-CPT analyses and at the time of 7-y follow-up for the ANT analyses), sex, preterm birth (<37 weeks of gestation, yes/no), maternal cognitive performance [assessed at 4- or 5-y follow-up, using the Wechsler Adult Intelligence Scale (WAIS-IV) Similarities subscale, one of four subscales used to measure verbal comprehension], maternal smoking during pregnancy (yes/no), and exposure to environmental tobacco smoke (smoking by any resident of the child’s home at 4-y follow-up for the K-CPT analyses and at 4- or 5-y as well as 7-y follow-ups for the ANT analyses, yes/no). In addition, we adjusted for maternal educational attainment at enrollment (none or primary school only, secondary school only, or university) as an indicator of individual-level socioeconomic status (SES), and for the Urban Vulnerability Index (Spanish Ministry of Public Works 2012), a measure of neighborhood SES, at each census tract (using the address at the time of outcome assessment), as an indicator of area-level SES. We estimated the difference in average outcome scores associated with one interquartile range (IQR) increase (based on all study participants) in average lifetime NDVI or VCF at 4–5 y or 7 y.

Sensitivity Analyses

We conducted a number of sensitivity analyses to evaluate the robustness of our findings. First, we performed models with additional adjustments for parity (continuous), for whether the child had been breastfed (yes/no), for child’s birth weight (continuous), and for the following factors at the time of the outcome assessment: sleep duration (average hours per day), time spent watching TV (average hours per week), time spent on sedentary activities (average hours per week), parental marital status (single parent: yes/no), and social class [Clasificación Nacional de Ocupaciones (CNO-94) (three categories)]. In addition, we estimated associations using simple negative binomial and linear regression models with cohort as a categorical predictor, as an alternative to using mixed effects models with a random intercept for the study cohort. We used NDVI maps from two different satellite sensors to assess green-space exposure in each cohort. To explore whether differences between the sensors influenced our findings, we derived standardized NDVI estimates for each map and buffer size (100 m, 300 m, and 500 m) as follows:
where NDVIij is the value of NDVI for subject i in center j (e.g., Sabadell or Valencia), NDVIj is the average NDVI in center j, and sd_NDVIj is the standard deviation of NDVI in center j. We averaged the standardized NDVI estimates for each buffer and time point (birth, 4 or 5 y, and 7 y, as appropriate) to construct alternative measures of lifelong residential exposure to green space, and we repeated the analyses. Furthermore, because the 16 May 2007 NDVI map used to estimate NDVI at birth for Valencia participants was not cloud-free, we conducted a sensitivity analysis excluding participants for whom >10% of the NDVI pixels in each buffer were affected by clouds (n=6, 8, and 9 for the 100-m, 300-m, and 500-m buffers, respectively).

Results

Study Population

Of 1,527 children with data available at birth (740 from Sabadell, 787 from Valencia), 1,199 (77.6%) and 1,044 (68.5%) participated in the 4- or 5-y and 7-y follow-ups, respectively (Figure 1 and Table S1). There were no statistically significant differences (α=0.05) in neighborhood SES, maternal verbal comprehension, child’s sex, or preterm birth between participants with available data at birth and those with measures of attention at 4–5 y (n=888, 364 from Sabadell, 524 from Valencia) or 7 y (n=978, 530 from Sabadell, 448 from Valencia) in the combined cohorts (Table 1) or individual cohorts (see Table S2). However, the children included in the analyses for 4- or 5-y and 7-y follow-ups tended to have mothers with higher educational attainment in comparison with the children participating at birth. Furthermore, the mothers of children included in the 7-y analyses were less likely to have smoked during pregnancy than all mothers with data available at birth.

Flow chart
Figure 1. The number of participants in each follow-up and those with available data on attention tests in Valencia and Sabadell, Spain.
Table 1. Description of characteristics of the study participants at birth and in 4- and 7-year follow-ups.
Characteristics Birth 4–5-y follow-up p-Valuec p-Valuea 7-y follow-up p-Valuec p-Valued p-Valuee
Includea (n=888) Excludedb (n=311) Includea (n=978) Excludedb (n=66)
Sex (Female) 48.0% 48.8% 46.1% 0.39 0.73 48.8% 40.9% 0.21 0.72 0.99
Preterm birth (yes) 4.4% 3.7% 3.9% 0.89 0.42 3.4% 7.7% 0.08 0.21 0.70
Maternal educational attainment 0.04 0.01 0.21 <0.01 0.81
 No or primary school 31.0% 26.1% 29.9% 24.7% 30.8%
 Secondary school 43.0% 42.7% 46.4% 43.6% 47.7%
 University 26.1% 31.2% 23.7% 31.7% 21.5%
Maternal IQf 10.5 (3.7) 10.5 (4.4) 10.5 (4.4) 0.53 0.84 10.5 (4.4) 10.5 (4.7) 0.66 0.30 0.40
Maternal smoking during pregnancy (yes) 35.6% 33.6% 33.1% 0.88 0.32 31.7% 33.3% 0.78 0.04 0.38
Exposure to environmental tobacco smoke at 4-year follow-up (yes) *** 50.2% 47.7% 0.44 *** 51.2% 48.4% 0.67 *** 0.69
Exposure to environmental tobacco smoke at 7-year follow-up (yes) *** *** *** *** *** 30.3% 15.2% 0.01 *** ***
Neighborhood socioeconomic statusg 0.6 (0.2) 0.6 (0.2) 0.7 (0.3) 0.33 0.22 0.6 (0.2) 0.6 (0.2) 0.12 0.09 0.65
Note: p-Values are reported for chi-squared test for categorical variables and Mann–Whitney U test for continuous variables. For continuous variables, median (IQR) and for categorical variables count (percentage) of each category has been reported.

ap-Value for the difference between children with available data at birth and those included at the 4-year follow-up.

bChildren participated in the 4-y follow-up but did not have data on K-CPT.

cp-Value for the difference between children included and excluded in each follow-up.

dp-Value for the difference between children with available data at birth and those included at the 7-year follow-up.

ep-Value for the difference between children with available data at the 4-year follow-up and those included at 7-y follow-up.

fWechsler Adult Intelligence Scale (WAIS-IV), Similarities subscale.

gUrban vulnerability index.

Greenness Exposure

Of 888 participants with available data on K-CPT, 194 (21.8%) had changed their address of residence between birth and the 4- or 5-y follow-up. Of 978 participants with available data on ANT, 252 (25.8%) had moved home between birth and the 7-y follow-up. The description of exposure measures in each follow-up separately for participants at each center has been presented in Table 2. Participants in Sabadell generally had higher levels of residential surrounding greenness and canopy cover in comparison with participants in Valencia. As presented in Table S3, NDVI and VCF values at 500 m were moderately to strongly correlated with values of the same exposure metric at different time points (Spearman’s correlations 0.68–0.83 for NDVI, 0.83–0.94 for VCI), with stronger correlations when limited to children who had not changed addresses between the follow-ups (Table S2). Correlations between NDVI and VCF at the same time points also were moderate to strong (0.46–0.79). There was no statistically significant difference in greenness exposure between those included and excluded in each follow-up or between those included in each follow-up and participants with available data at birth (data not shown).

Table 2. Median (25th and 75th percentiles) of measures of attention [Conners’ Kiddie Continuous Performance Test (K-CPT) at 4–5 y and attentional network task (ANT) at 7 y] and exposure [averages of normalized difference vegetation index (NDVI) and vegetation continuous fields (VCF, % tree cover)].
Variables 4–5-y follow-up 7-y follow-up
Sabadell (n=364) Valencia (n=524) p-Valuea Sabadell (n=530) Valencia (n=448) p-Valuea
Attentionb
  Omissions (counts) 24 (13, 37) 10 (5, 20) <0.01 3 (1, 7) 3 (1, 7) 0.1
  Commissions (counts) 22 (15, 30) 22 (15, 29) 0.63 5 (2, 8) 4 (2, 7) <0.01
  Hit Reaction Time-Standard Error (ms) 332.8 (252.8, 441.6) 229.7 (171.9, 307.0) <0.01 332.0 (273.8, 381.1) 326.6 (268.6, 385.3) 0.55
Residential surrounding greenness (NDVI)b
  100-m buffer 0.20 (0.16, 0.24) 0.19 (0.17, 0.24) 0.89 0.20 (0.17, 0.25) 0.19 (0.16, 0.24) 0.01
  300-m buffer 0.24 (0.18, 0.29) 0.22 (0.19, 0.27) 0.06 0.24 (0.19, 0.30) 0.22 (0.19, 0.27) <0.01
  500-m buffer 0.26 (0.21, 0.32) 0.24 (0.21, 0.29) <0.01 0.26 (0.21, 0.32) 0.23 (0.20, 0.28) <0.01
Residential surrounding tree cover (VCF)b
  100-m buffer 1.6 (1.2, 2.3) 0.3 (0.1, 0.7) <0.01 1.7 (1.3, 2.4) 0.4 (0.2, 0.8) <0.01
  300-m buffer 1.9 (1.4, 2.7) 0.6 (0.3, 1.0) <0.01 2.0 (1.6, 2.8) 0.7 (0.4, 1.0) <0.01
  500-m buffer 2.2 (1.5, 3.0) 0.7 (0.5, 1.0) <0.01 2.2 (1.7, 3.1) 0.8 (0.6, 1.1) <0.01

ap-value for Mann–Whitney U test.

bFor the follow-up at 4 or 5 y, the NDVI was averaged for addresses at birth and 4–5 y and for the follow-up at 7 y, the NDVI was averaged for addresses at birth, 4 or 5 years, and 7-years.

Attention

The description of the performance of study participants in attention tests is presented in Table 2. Median K-CPT scores for omissions and HRT-SE were lower for children from Valencia than for children from Sabadell, which may at least partly reflect the difference in the age at which children in each cohort were tested (5 y vs. 4 y, respectively) (Table 2). The ANT measures were comparable with those of our other study (Forns et al. 2014) conducted in Barcelona reporting median (IQR) of 2 (4), 5 (5), and 310.3 (122.4), respectively, for omission and commission errors and HRT-SE among a population-based sample of 7-y-old children.

Greenness and Attention

K-CPT.

Increases in residential surrounding greenness (NDVI) in all buffer areas during the first 4–5 y of life were associated with lower K-CPT omission errors and HRT-SE (Table 3). In contrast, estimated associations between NDVI and K-CPT commission errors were essentially null. Although increases in residential surrounding tree cover (VCF) also were associated with lower average K-CPT omission errors and average HRT-SE, all estimated differences were very close to the null (Table 3).

Table 3. Adjusted mean ratios (95% confidence interval) in omission and commission errors and regression coefficient (95% CI) for hit reaction time-standard error (HRT-SE) associated with an IQR increase in the average of normalized difference vegetation index (NDVI) and vegetation continuous fields (VCF, % tree cover) surrounding participants’ residences.
Greenness exposure Median (IQR) greenness (0 to 4–5 years) K-CPTa (n=888) Median (IQR) greenness (0–7 years) ANTb (n=978)
Omission error Commission error HRT-SE (ms) Omission error Commission error HRT-SE (ms)
NDVI
100-m buffer 0.193 (0.074) 0.90 (0.85, 0.96) 1.01 (0.97, 1.04) −1.0 (−2.0, −0.1) 0.194 (0.087) 1.00 (0.90, 1.11) 1.03 (0.95, 1.11) −4.1 (−10.6, 2.5)
300-m buffer 0.229 (0.089) 0.88 (0.82, 0.94) 1.00 (0.96, 1.05) −1.3 (−2.5, −0.2) 0.232 (0.097) 0.97 (0.87, 1.07) 1.02 (0.95, 1.11) −6.6 (−13.6, 0.3)
500-m buffer 0.245 (0.091) 0.88 (0.81, 0.95) 1.01 (0.97, 1.06) −1.3 (−2.5, −0.1) 0.247 (0.102) 0.95 (0.85, 1.06) 1.01 (0.93, 1.10) −7.9 (−15.1, −0.8)
VCF
100-m buffer 0.700 (1.315) 0.98 (0.94, 1.02) 1.01 (0.99, 1.04) −0.2 (−0.9, 0.6) 1.162 (1.514) 1.01 (0.96, 1.06) 1.02 (0.94, 1.10) −2.3 (−7.1, 2.4)
300-m buffer 0.964 (1.295) 0.99 (0.94, 1.03) 1.01 (0.98, 1.03) −0.1 (−0.8, 0.6) 1.388 (1.469) 1.00 (0.95, 1.04) 1.02 (0.97, 1.09) −2.6 (−6.8, 1.6)
500-m buffer 1.088 (1.372) 0.99 (0.95, 1.03) 1.00 (0.99, 1.04) −0.2 (−0.9, 0.5) 1.495 (1.573) 0.99 (0.94, 1.03) 1.02 (0.96, 1.08) −3.0 (−7.1, 1.2)
Note: Mixed effects models with random intercepts for cohort (binomial for omission and commission errors and linear for HRT-SE) adjusted for age at the time of attention test, sex, history of preterm birth, maternal educational attainment, maternal IQ, maternal smoking during pregnancy, exposure to environmental tobacco smoke, and neighborhood socioeconomic status.

aConners’ Kiddie Continuous Performance Test.

bAttentional Network Task.

The ANT.

More residential surrounding greenness during the first seven years of life was inversely associated with ANT HRT-SE at the age of 7 y, with associations being statistically significant for the 500-m buffer (Table 3). There was little or no evidence of associations for NDVI or VCF with ANT omission or commission errors. VCF was associated with lower HRT-SE at 7 y, but associations were not significant (Table 3).

The cohort-specific associations between measures of greenness exposure and K-CPT and ANT indicators were generally consistent with the pooled analyses (see Table S3). Although there was some variation in corresponding estimates between the cohorts, estimates were imprecise, and clear differences between the regions were not evident (see Table S4).

Sensitivity Analyses

Our findings after further adjustment of analyses for the parity, breastfeeding, birth weight, sleep duration, time spent watching TV, time spent on sedentary activities, parental marital status, and social class were generally consistent with those of the main analyses in terms of direction and statistical significance (data not shown). Similarly, the results of simple negative binomial and linear regression models with cohort as a categorical predictor in the models were in line with those of the main analyses (see Table S5). Moreover, the direction and statistical significance of the associations based on standardized NDVI values were consistent with those of the main analyses (data not shown). Similarly, the associations were consistent with those of main analyses after excluding participants with more than 10% of the NDVI pixels in each buffer around their homes affected by clouds (data not shown).

Discussion

To our knowledge, this prospective study is the first to estimate associations of lifelong residential exposure to greenness with measures of attention in children, and the first to report separate estimates for associations with tree cover (VCF) and greenness (NDVI). We made use of data from two well-established population-based cohorts, utilized computerized tests (K-CPT and ANT) to assess attention, and used remote-sensing indices (NDVI and VCF) to estimate exposure to greenness. Higher lifelong residential surrounding greenness was associated with fewer K-CPT omission errors and lower K-CPT HRT-SE at 4–5 y of age, and with lower ANT HRT-SE at 7 y of age, consistent with better attention. Point estimates were close to the null for greenness (NDVI) and K-CPT commission errors and ANT omission and commission errors; and for tree cover (VCF) and all K-CPT outcomes and ANT omission and commission errors (Table 3). Tree cover was inversely associated with ANT HRT-SE, though estimates were not significant.

Interpretation of Results

Exposure to residential greenness was inversely associated with omission errors and HRT-SE, but was not associated with commission errors. K-CPT omission errors and HRT-SE may be measures of “focused attention,” whereas K-CPT commission errors may be more relevant to “hyperactivity-impulsivity” (Egeland and Kovalik-Gran 2010). Thus, our findings suggest that exposure to greenness might influence focused attention, rather than hyperactivity-impulsivity, consistent with attention restoration theory, as described below.

The inverse association between exposure to residential surrounding greenness and omission errors at 4–5 y of age was not present at 7 years. One explanation might be the use of different tools (K-CPT vs. ANT) to characterize attention at each time point. Furthermore, nonresidential exposure to greenness, at school or at the homes of other children, may increase with age, leading to greater exposure misclassification when exposure is estimated based only on residential address. Assessing greenness exposures at additional locations may be beneficial for future studies.

The associations for residential surrounding tree cover were close to null (with the exception of ANT HRT-SE, which showed indications of inverse associations) and none attained statistical significance. These findings might suggest that tall trees could exert fewer benefits on attention, in comparison with other types of vegetation, such as grasses and shrubs. However, the contrast in exposure to residential surrounding tree cover among our study participants was small, which could have underpowered our analyses to detect an association between this exposure and attention. Such a low contrast in exposure was not unexpected in our study regions, given their Mediterranean climate and high density of built-up areas, but it also could have resulted, at least in part, from our use of VCF to assess this exposure, which does not take account of trees shorter than five meters. The possibility that different types of greenness might have different impacts on neurodevelopment remains an open question for future studies.

Available Evidence

Surrounding greenness were associated with lower prevalence rates for depression in Dutch children <12 years of age in an ecological study based on medical records data and land use information (Maas et al. 2009). Another ecological study of 905 public schools in Massachusetts, United States, reported that higher levels of greenness surrounding the schools (measured as the average of NDVI) was associated with better student performance at schools (Wu et al. 2014). Experimental studies have suggested that walking in nature or watching photos of nature might improve directed-attention abilities in adults (Berman et al. 2008) and reduce ADHD symptoms in children (Taylor et al. 2001; Kuo and Taylor 2004; Taylor and Kuo 2009; van den Berg and van den Berg 2011). A study by Wells (2000) reported that relocation to residences with higher “naturalness” (a combination of visual access to greenness and presence of vegetation in houses’ yards) improved attention in a sample of 17 children (Wells 2000). In a previous cross-sectional analysis, we found a protective association between residential surrounding greenness and behavioral problems, including hyperactivity and inattention in primary schoolchildren in Barcelona (Amoly et al. 2014), an observation that was replicated by another cross-sectional study in Germany (Markevych et al. 2014). In another study (Dadvand et al. 2015a) based on a sample of 2,593 primary schoolchildren (aged 7–10 y) residing in Barcelona (2012–2013), we observed higher total surrounding greenness index (defined as the average of NDVI around a home, within a school, and surrounding commuting routes) between home and school was associated with reduced inattentiveness as characterized by the 12-mo trajectory of HRT-SE from four repeated ANTs measures (three months apart).

Potential Underlying Mechanisms

The Biophilia hypothesis suggests that humans have important evolutionary bonds to nature (Wilson 1984; Kellert and Wilson 1993). Accordingly, it has been proposed that contact with nature is important for brain development in children (Kahn 1997; Kahn and Kellert 2002). In addition, the theory of attention restoration proposes that contact with nature may enhance attention (Kaplan and Kaplan 1989; Kaplan 1995; Berman et al. 2008). Our findings extend the prospect of attention-restoration theory by evaluating the long-term association between lifelong exposure to green spaces and attention in children.

Greenness surrounding children’s schools and residences has been associated with lower exposure to air pollution (Dadvand et al. 2012, 2015b), and school air pollution exposure was positively associated with ANT HRT-SE, indicating greater inattentiveness, in a study of Barcelona school children (Sunyer et al. 2015). Perceived access to green spaces was associated with less noise annoyance in a study of urban adults (Gidlöf-Gunnarsson and Öhrström 2007; Dadvand et al. 2012, 2015b), and performance on tests of attention is reduced when children are exposed to noise during the tests (Klatte et al. 2013; Sunyer et al. 2015). Moreover, proximity to green spaces, particularly parks, has been suggested to increase physical activity (James et al. 2015), and higher levels of physical activity are related to improved cognitive development (Fedewa and Ahn 2011). However, the body of evidence on the effect of proximity to green spaces on physical activity is not consistent, and a notable heterogeneity exists in the reported direction and strength of associations (Timperio et al. 2008; Lachowycz and Jones 2011; Lovasi et al. 2011). Furthermore, psychological stress and depression in parents have been associated with adverse impacts on cognitive development in their children (Ramchandani and Psychogiou 2009), and residential surrounding greenness has been associated with evidence of stress-reduction effects and reduced depression in adults (Dadvand et al. 2016; McEachan et al. 2016). A growing body of evidence also suggests that a failure of the immunoregulatory pathways due to a reduced exposure to macro- and microorganisms in Westernized populations might play a role in impairment of brain development (Rook 2013; Rook et al. 2013), with childhood as a particular window of vulnerability (Rook et al. 2014). Therefore, the potential ability of surrounding greenness to enhance immunoregulation-inducing microbial input from the environment (Rook 2013) could have been another mechanism underlying our observed association between surrounding greenness and attention.

Strengths and Limitations

This prospective study was based on computerized tests to objectively characterize attention for each study participant. Our use of objective measures of attention could be considered a step forward as these computerized cognitive tests are less prone to outcome misclassification due to subjectivity in comparison with questionnaire-based methods used in previous studies (Forns et al. 2014). Questionnaire-based methods, however, have the advantage of assessing cognition in a realistic setting. Therefore, for future studies, an ideal assessment of attention could include both computerized cognitive tests and questionnaire-based methods.

We accounted for changes in residential addresses at each follow-up interview when estimating lifelong exposures, but we could not account for the timing of each address change or additional moves between study visits. Therefore we could not estimate associations with time-varying cumulative exposures, or associations with exposures during specific time windows. Also, children included in the analyses, in comparison with those children who were excluded (due to loss to follow-up or unavailability of data for attention tests) were more likely to have mothers with higher educational attainment, which could have resulted in attrition bias. Furthermore, we did not have information on the family context and support for cognitive development available to the child in the home environment, which could have been a potential source of residual confounding. Additionally, our study could not disentangle the short- and long-term associations between greenness exposure and attention.

Conclusions

We studied the association of lifelong exposure to residential surrounding greenness and tree cover with attention in children. Higher exposure to residential surrounding greenness areas was associated with fewer K-CPT omission errors and lower K-CPT HRT-SE at age 4–5 y and lower ANT HRT-SE at age 7 y, consistent with a beneficial association between this exposure and attention. The associations for residential surrounding tree cover were not conclusive; however, our analyses might have been underpowered to detect such associations. These findings warrant further replications in other settings with different climates and environments.

Acknowledgment

We thank all the funding agencies for supporting our research. We are particularly grateful to all the participants for their generous collaboration. A full roster of the INMA Project Investigators can be found at: http://www.proyectoinma.org/presentacion-inma/listado-investigadores/en_listado-investigadores.html. We are grateful to J. Julvez for his help in the implementation of the cognitive tests and interpretation of their results.

C.T. is a recipient of a European Respiratory Society Fellowship (RESPIRE2–2015–7251) P.D. is funded by a Ramón y Cajal fellowship (RYC-2012-10995) awarded by the Spanish Ministry of Economy and Competitiveness. S.L. is funded by a Miguel Servet-FEDER fellowship (MS15/0025) awarded by the Spanish Ministry of Economy and Competitiveness. M.G. is funded by a Miguel Servet-FEDER fellowship (MS13/00054) awarded by the Spanish Ministry of Economy and Competitiveness.

References

Amoly E, Dadvand P, Forns J, López-Vicente M, Basagaña X, Julvez J. 2014. Green and blue spaces and behavioral development in Barcelona schoolchildren: The BREATHE Project. Environ Health Perspect 122(12):1351–1358. PMID: 25204008, 10.1289/ehp.1408215.

Barnard H, Rao R, Xu Y, Froehlich T, Epstein J, Lanphear BP, et al. 2015. Association of the Conners’ Kiddie Continuous Performance Test (K-CPT) performance and parent-report measures of behavior and executive functioning. J Atten Disord pii:1087054715578271, PMID: 25846228 10.1177/1087054715578271.

Berman MG, Jonides J, Kaplan S. 2008. The cognitive benefits of interacting with nature. Psychol Sci 19(12):1207–1212, PMID: 19121124, 10.1111/j.1467-9280.2008.02225.x.

Bowler D, Buyung-Ali L, Knight T, Pullin A. 2010. A systematic review of evidence for the added benefits to health of exposure to natural environments. BMC Public Health 10:456, PMID: 20684754, 10.1186/1471-2458-10-456.

Chu R, Thabane L, Ma J, Holbrook A, Pullenayegum E, Devereaux P. 2011. Comparing methods to estimate treatment effects on a continuous outcome in multicentre randomized controlled trials: a simulation study. BMC Med Res Methodol 11(1):21, 10.1186/1471-2288-11-21.

Conners CK. 2000. Conners’ Continuous Performance Test II: Computer Program for Windows Technical Guide and Software Manual. Staff MHS, eds. North Tonawanda, NY:Multi-Health Systems.

Dadvand P, Bartoll X, Basagaña X, Dalmau-Bueno A, Martinez D, Ambros A, et al. 2016. Green spaces and general health: roles of mental health status, social support, and physical activity. Environ Int 91:161–167, PMID: 26949869, 10.1016/j.envint.2016.02.029.

Dadvand P, de Nazelle A, Triguero-Mas M, Schembari A, Cirach M, Amoly E, et al. 2012. Surrounding greenness and exposure to air pollution during pregnancy: an analysis of personal monitoring data. Environ Health Perspect 120(9):1286–1290, PMID: 22647671, 10.1289/ehp.1104609.

Dadvand P, Nieuwenhuijsen MJ, Esnaola M, Forns J, Basagaña X, Alvarez-Pedrerol M, et al. 2015a. Green spaces and cognitive development in primary schoolchildren. Proc Natl Acad Sci USA U S A 112(26):7937–7942, PMID: 26080420, 10.1073/pnas.1503402112.

Dadvand P, Rivas I, Basagaña X, Alvarez-Pedrerol M, Su J, De Castro Pascual M, et al. 2015b. The association between greenness and traffic-related air pollution at schools. Sci Total Environ 523:59–63, PMID: 25862991, 10.1016/j.scitotenv.2015.03.103.

Dadvand P, Sunyer J, Basagaña X, Ballester F, Lertxundi A, Fernández-Somoano A, et al. 2012. Surrounding greenness and pregnancy outcomes in four Spanish birth cohorts. Environ Health Perspect 120(10):1481–1487, PMID: 22899599, 10.1289/ehp.1205244.

Dadvand P, Wright J, Martinez D, Basagaña X, McEachan RRC, Cirach M, et al. 2014. Inequality, green spaces, and pregnant women: roles of ethnicity and individual and neighbourhood socioeconomic status. Environ Int 71:101–108, PMID: 24997306, 10.1016/j.envint.2014.06.010.

Egeland J, Kovalik-Gran I. 2010. Measuring several aspects of attention in one test: the factor structure of Conners’s Continuous Performance Test. J Atten Disord 13(4):339–346, PMID: 18728238, 10.1177/1087054708323019.

Epstein JN, Erkanli A, Conners CK, Klaric J, Costello JE, Angold A. 2003. Relations between continuous performance test performance measures and ADHD behaviors. J Abnorm Child Psychol 31(5):543–554, PMID: 14561061.

Fedewa AL, Ahn S. 2011. The effects of physical activity and physical fitness on children’s achievement and cognitive outcomes. Res Q Exerc Sport 82(3):521–535, PMID: 21957711, 10.1080/02701367.2011.10599785.

Forns J, Esnaola M, López-Vicente M, Suades-González E, Alvararez-Pedrerol M, Julvez J, et al. 2014. The n-back test and the attentional network task as measures of child neuropsychological development in epidemiological studies. Neuropsychology 28(4):519–529, PMID: 24819069, 10.1037/neu0000085.

Gidlöf-Gunnarsson A, Öhrström E. 2007. Noise and well-being in urban residential environments: the potential role of perceived availability to nearby green areas. Landscape and Urban Plan 83(2):115–126, 10.1016/j.landurbplan.2007.03.003.

Grandjean P, Landrigan PJ. 2014. Neurobehavioural effects of developmental toxicity. Lancet Neurol 13(3):330–338, PMID: 24556010, 10.1016/S1474-4422(13)70278-3.

Guxens M, Ballester F, Espada M, Fernández MF, Grimalt JO, Ibarluzea J, et al. 2012. Cohort profile: The INMA—INfancia y Medio Ambiente—(Environment and Childhood) Project. Int J Epidemiol 41(4):930–940, PMID: 21471022, 10.1093/ije/dyr054.

Homack S, Riccio CA. 2006. Conners’ Continuous Performance Test (2nd ed.; CCPT-II). J Atten Disord 9(3):556–558, PMID: 16481673, 10.1177/1087054705283578.

James P, Banay RF, Hart JE, Laden F. 2015. A review of the health benefits of greenness. Curr Epidemiol Rep 2(2):131–142, PMID: 26185745, 10.1007/s40471-015-0043-7.

Kahn PH. 1997. Developmental psychology and the biophilia hypothesis: children’s affiliation with nature. Dev Rev 17(1):1–61, 10.1006/drev.1996.0430.

Kahn PH, Kellert SR. 2002. Children and nature: psychological, sociocultural, and evolutionary investigations. Cambridge, MA: Massachusetts Institute of Technology.

Kaplan R, Kaplan S. 1989. The experience of nature: a psychological perspective. New York: Cambridge University Press.

Kaplan S. 1995. The restorative benefits of nature: toward an integrative framework. J Environ Psychol 15(3):169–182, 10.1016/0272-4944(95)90001-2.

Kellert SR. 2005. Building for Life: Designing and Understanding the Human-Nature Connection. Washington, DC: Island Press.

Kellert SR, Wilson EO. 1993. The biophilia hypothesis. Washington, DC: Island Press.

Klatte M, Bergström K, Lachmann T. 2013. Does noise affect learning? a short review on noise effects on cognitive performance in children. Front Psychol 4:578, PMID: 24009598, 10.3389/fpsyg.2013.00578.

Kuo FE, Taylor AF. 2004. A potential natural treatment for attention-deficit/hyperactivity disorder: evidence from a national study. Am J Public Health 94(9):1580–1586, PMID: 15333318.

Lachowycz K, Jones AP. 2011. Greenspace and obesity: a systematic review of the evidence. Obes Rev 12(5):e183–e189, PMID: 21348919, 10.1111/j.1467-789X.2010.00827.x.

Lovasi G, Jacobson J, Quinn J, Neckerman K, Ashby-Thompson M, Rundle A. 2011. Is the environment near home and school associated with physical activity and adiposity of urban preschool children? J Urban Health 88(6):1143–1157, PMID: 21826583, 10.1007/s11524-011-9604-3.

Maas J, Verheij RA, de Vries S, Spreeuwenberg P, Schellevis FG, Groenewegen PP. 2009. Morbidity is related to a green living environment. J Epidemiol Community Health 63(12):967–973, PMID: 19833605, 10.1136/jech.2008.079038.

Markevych I, Tiesler CMT, Fuertes E, Romanos M, Dadvand P, Nieuwenhuijsen MJ, et al. 2014. Access to urban green spaces and behavioural problems in children: results from the GINIplus and LISAplus studies. Environ Int 71:29–35, PMID: 24953038, 10.1016/j.envint.2014.06.002.

McEachan RRC, Prady SL, Smith G, Fairley L, Cabieses B, Gidlow C, et al. 2016. The association between green space and depressive symptoms in pregnant women: moderating roles of socioeconomic status and physical activity. J Epidemiol Community Health 70(3):253–259, 10.1136/jech-2015-205954.

Ramchandani P, Psychogiou L. 2009. Paternal psychiatric disorders and children’s psychosocial development. Lancet 374(9690):646–653, PMID: 19411102, 10.1016/S0140-6736(09)60238-5.

Rook GAW. 2013. Regulation of the immune system by biodiversity from the natural environment: an ecosystem service essential to health. Proc Natl Acad Sci USA 110(46):18360–18367, PMID: 24154724, 10.1073/pnas.1313731110.

Rook GAW, Lowry CA, Raison CL. 2013. Microbial ‘Old Friends’, immunoregulation and stress resilience. Evol Med Public Health 2013(1):46–64, PMID: 24481186, 10.1093/emph/eot004.

Rook GAW, Lowry CA, Raison CL. 2014. Hygiene and other early childhood influences on the subsequent function of the immune system. Brain Res 1617:47–62, PMID: 24732404, 10.1016/j.brainres.2014.04.004.

Rueda MR, Fan J, McCandliss BD, Halparin JD, Gruber DB, Lercari LP, et al. 2004. Development of attentional networks in childhood. Neuropsychologia 42(8):1029–1040, PMID: 15093142, 10.1016/j.neuropsychologia.2003.12.012.

Sexton JO, Song X-P, Feng M, Noojipady P, Anand A, Huang C, et al. 2013. Global, 30-m resolution continuous fields of tree cover: Landsat-based rescaling of MODIS vegetation continuous fields with lidar-based estimates of error. Int J Digit Earth 6(5):427–448, 10.1080/17538947.2013.786146.

Spanish Ministry of Public Works. 2012. Atlas of Urban Vulnerability in Spain. Methodology and Contents. http://www.fomento.gob.es/NR/rdonlyres/40668D5E-26B6-4720-867F-286BD55E1C6B/135960/20160201METODOLOGIAATLASVULNERABILIDAD2001Y2011.pdf [accessed 27 May 2016].

Sunyer J, Esnaola M, Alvarez-Pedrerol M, Forns J, Rivas I, López-Vicente M, et al. 2015. Association between traffic-related air pollution in schools and cognitive development in primary school children: a prospective cohort study. PLoS Med 12(3):e1001792, PMID: 25734425, 10.1371/journal.pmed.1001792.

Taylor AF, Kuo FE. 2009. Children with attention deficits concentrate better after walk in the park. J Atten Disord 12(5):402–409, PMID: 18725656, 10.1177/1087054708323000.

Taylor AF, Kuo FE, Sullivan WC. 2001. Coping with ADD. The surprising connection to green play settings. Environ Behav 33(1):54–77, 10.1177/00139160121972864.

Timperio A, Giles-Corti B, Crawford D, Andrianopoulos N, Ball K, Salmon J, et al. 2008. Features of public open spaces and physical activity among children: findings from the CLAN study. Prev Med 47(5):514–518, PMID: 18718847, 10.1016/j.ypmed.2008.07.015.

van den Berg AE, van den Berg CG. 2011. A comparison of children with ADHD in a natural and built setting. Child Care Health Dev 37(3):430–439, PMID: 21143265, 10.1111/j.1365-2214.2010.01172.x.

Wells NM. 2000. At home with nature effects of greenness on children’s cognitive functioning. Environ Behav 32(6):775–795, 10.1177/00139160021972793.

Wilson EO. 1984. Biophilia. Cambridge, MA:Harvard University Press.

Wu C-D, McNeely E, Cedeño-Laurent JG, Pan W-C, Adamkiewicz G, Dominici F, et al. 2014. Linking student performance in Massachusetts elementary schools with the “greenness” of school surroundings using remote sensing. PloS one 9(10):e108548, PMID: 25310542, 10.1371/journal.pone.0108548.

Respiratory, Dermal, and Eye Irritation Symptoms Associated with Corexit™ EC9527A/EC9500A following the Deepwater Horizon Oil Spill: Findings from the GuLF STUDY

Author Affiliations open

1Epidemiology Branch, National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health (NIH), Department of Health and Human Services (DHHS), Research Triangle Park, North Carolina, USA

2Department of Epidemiology, UNC Gillings School of Global Public Health, Chapel Hill, North Carolina, USA

3Exposure Assessment Applications, LLC, Arlington, Virginia, USA

4Stewart Exposure Assessments, LLC, Arlington, Virginia, USA

PDF icon PDF Version (167 KB)

  • Background:
    The large quantities of chemical oil dispersants used in the oil spill response and cleanup (OSRC) work following the Deepwater Horizon disaster provide an opportunity to study associations between dispersant exposure (Corexit™ EC9500A or EC9527A) and human health.
    Objectives:
    Our objectives were to examine associations between potential exposure to the dispersants and adverse respiratory, dermal, and eye irritation symptoms.
    Methods:
    Using data from detailed Gulf Long-term Follow-up ( GuLF) Study enrollment interviews, we determined potential exposure to either dispersant from participant-reported tasks during the OSRC work. Between 27,659 and 29,468 participants provided information on respiratory, dermal, and eye irritation health. We estimated prevalence ratios (PRs) to measure associations with symptoms reported during the OSRC work and at study enrollment, adjusting for potential confounders including airborne total hydrocarbons exposure, use of cleaning chemicals, and participant demographics.
    Results:
    Potential exposure to either of the dispersants was significantly associated with all health outcomes at the time of the OSRC, with the strongest association for burning in the nose, throat, or lungs [adjusted PR (aPR)=1.61 (95% CI: 1.42, 1.82)], tightness in chest [aPR=1.58 (95% CI: 1.37, 1.81)], and burning eyes [aPR=1.48 (95% CI: 1.35, 1.64). Weaker, but still significant, associations were found between dispersant exposure and symptoms present at enrollment.
    Conclusions:
    Potential exposure to Corexit™ EC9527A or EC9500A was associated with a range of health symptoms at the time of the OSRC, as well as at the time of study enrollment, 1–3 y after the spill. https://doi.org/10.1289/EHP1677
  • Received: 25 January 2017
    Revised: 20 June 2017
    Accepted: 26 June 2017
    Published: 15 September 2017

    Address correspondence to D.P. Sandler, Epidemiology Branch, National Institute of Environmental Health Sciences, PO Box 12233, MD A3-05, 111 TW Alexander Dr., Research Triangle Park, NC 27709-2233 USA. Telephone: 919-541-4668. Email: Dale.Sandler@nih.gov

    Supplemental Material is available online (https://doi.org/10.1289/EHP1677).

    M.R.S. provided a deposition for the Celanese Chemical Company, which has no oil or gas holdings, in August 2016. During this deposition, M.R.S. served as a corporate representative and not an expert witness on Celanese’s asbestos-related practices in the 1950s, 1960s, and 1970s. M.R.S. worked for Celanese Chemical Group from 1972 to 1992 as the Manager of Industrial Hygiene. All other authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF 193 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (114 KB)

Introduction

Background

Over 4.9 million barrels of crude oil was released into the Gulf of Mexico between the explosion of the Deepwater Horizon rig on 20 April 2010 and the top-capping of the wellhead on 15 July 2010 (United States Coast Guard 2011). As part of the oil spill response and cleanup (OSRC), approximately 1.8 million gallons (6.8 million liters) of oil dispersant was applied both to the sea surface [1.07 million gallons (4.05 million liters)] and directly into the stream of oil leaving the wellhead 5,000 feet (1.5 km) underwater [0.77 million gallons (2.9 million liters)] (United States Coast Guard 2011). Dispersants are typically used to reduce the interfacial tension between crude oil and water and facilitate the breakup of oil slicks into small droplets that are thought to be more easily dispersed by natural processes such as wind and wave action (Chapman et al. 2007). Two dispersants were used in the Deepwater Horizon spill response: Corexit™ EC9500A (9500A), which was applied at both the water surface and the subsurface wellhead, and Corexit™ EC9527A (9527A), which was applied only at the water surface (Kujawinski et al. 2011). Both dispersants are composed of propylene glycol and organic salts including dioctyl sodium sulfosuccinate (DOSS). Additionally, 9500A contains petroleum distillates, whereas 9527A does not contain petroleum distillates but does contain 2-butoxyethanol (Wise and Wise 2011).

Dispersants were applied to the surface either through aerial spraying or by vessels within 3 nautical miles (5.5 km) of the wellhead area. Aerial application consisted of both 9527A and 9500A from 22 April until 22 May, after which 9500A was used exclusively. Vessels in the wellhead area applied 9500A exclusively (BP Gulf Science Data 2016a). Subsurface application of 9500A was accomplished through a remotely operated underwater vehicle injecting dispersant directly into the stream of oil leaving the wellhead (BP Gulf Science Data 2016b). Based on these uses, the most likely avenues for human exposure among responders are from dermal exposure and from inhalation of dispersant aerosol droplets.

Previous epidemiologic studies have found adverse health effects associated with oil spill cleanup work (Aguilera et al. 2010; Laffon et al. 2016). Effects have included increased lumbar pain, migraine, dermatitis, eye and throat irritation, and respiratory symptoms. Most epidemiologic studies have focused on the acute effects of crude oil exposure during spill cleanup, although Zock et al. (2012) found an association between participating in cleanup work and self-reported respiratory symptoms 5 y after the spill response in workers who responded to the Prestige oil spill in 2002. Although dispersants were used in some of these previous OSRC operations, no studies looked at distinguishing effects of exposure to the dispersants.

In contrast, much of the research on dispersants has focused on their efficacy in dispersing oil (Prince et al. 2013) and on potential adverse impacts on the environment (Kleindienst et al. 2015). Wise and Wise (2011) published a review of studies examining the toxicity of various dispersants, including 9500A and 9527A, in certain model species, finding that both dispersants exhibited similar acute toxicity to crustaceans and mollusks and that oral exposure to 9527A adversely affected intestinal function in rat models.

In response to public concerns regarding the use of dispersants during the OSRC, the U.S. Environmental Protection Agency (EPA) commissioned a series of trials testing the acute toxicity of eight oil dispersants in representative Gulf species, classifying 9500A as either “slightly toxic” or “practically nontoxic” depending on the species (Hemmer et al. 2010a). Similarly, dispersant–oil mixtures were reported to be no more toxic to those Gulf species than crude oil alone (Hemmer et al. 2010b). However, the U.S. EPA did not investigate the effects of exposure to 9527A. Since the spill, additional research has shown that dermal exposure to 9500A resulted in dermal irritation and sensitization in mice (Anderson et al. 2011), although a 5-h inhalation exposure to 9500A did not appear to result in significant adverse pulmonary symptoms in rats (Roberts et al. 2011). However, these studies of mice and rats did not include a coexposure of crude oil or other petroleum by-products that would be expected to be present in the Deepwater Horizon oil spill environment.

Using in vitro cultures of human bronchial cells, Shi et al. (2013) showed that exposure to either 9500A or 9527A resulted in a significant loss of cell viability and that the loss of viability was dose-dependent. Similarly, Major et al. (2016) found that exposing human bronchial cells to individual mixtures of 9500A and 9527A with crude oil induced both cytotoxic and genotoxic effects. Although data on toxicity of dispersants are limited, there is some evidence of effects caused by the component ingredients of each dispersant. Human exposure studies have shown that propylene glycol is a mild irritant when applied dermally, and animal studies of respiratory effects due to inhalation exposure are inconclusive (ATSDR 1997). In contrast, 2-butoxyethanol is considerably more toxic, with acute respiratory effects and eye irritation observed in both human and animal studies, although minimal dermal effects have been observed in human studies (ATSDR 1998). The Material Safety Data Sheet (MSDS) for DOSS lists the chemical as irritating to the skin and eyes and as a possible respiratory irritant (Acros Organics 2013). We were unable to find any studies of direct effects of either 9500A or 9527A on human health.

Given the lack of epidemiologic research into the effects of dispersant exposure on human health, our study used data from a large cohort of workers participating in the Deepwater Horizon OSRC to assess two related objectives. The first objective was to quantify associations of potential exposure to dispersants with adverse respiratory, eye irritation, and dermal effects at the time of the OSRC; the second objective was to quantify associations of potential exposure to dispersants with adverse respiratory, eye irritation, and dermal effects in the 30 days before study enrollment, 1–3 years after the OSRC.

Methods

Study Design

Data were from the Gulf Long-term Follow-up Study (GuLF STUDY), a prospective cohort study of persons involved in the OSRC following the 2010 Deepwater Horizon oil spill (Kwok et al. 2017). Briefly, a total of 32,608 participants completed a telephone interview to enroll in the study between March 2011 and March 2013. This detailed interview collected information on particulars of the participant’s OSRC work, if any, in addition to demographic factors, lifestyle information, and medical history/symptoms both at the time of the OSRC and at the time of the interview. The interview for Vietnamese-speaking participants was abbreviated and did not collect information that could be used to assess potential dispersant exposure; therefore, those participants (n=999) were excluded from this analysis. Additionally, we excluded any remaining participants with missing data on the exposure of interest, on the outcomes of interest, or on covariates, leaving study populations of 28,636 for analyses of respiratory outcomes, 27,659 for dermal outcomes, and 29,468 for eye irritation outcomes. The study was approved by the Institutional Review Board of the National Institute of Environmental Health Sciences/National Institutes of Health. After receiving information about the study in the mail, participants provided verbal consent for the enrollment telephone interview.

Exposure Assessment

Participants were categorized as workers if they worked at least one day engaged in OSRC activities. Nonworkers received safety training but never worked on the response. For respiratory and eye irritation analyses, dispersant exposure for workers was classified as “ever/never” based on a positive response to any interview question asking about direct work with dispersants or work on a ship from which dispersants were applied (see Table S1). Additionally, participants were classified as exposed if they responded positively to working on any task that involved dispersant-related equipment, such as pumps, for more than half of the time. For dermal analyses, dermal dispersant exposure for workers was classified as “ever/never” based on self-reported skin or clothing contact with dispersants during the OSRC for breaking up the oil on or below the surface of the water. Although the questionnaire did not specifically refer to either Corexit™ 9527A or Corexit™ 9500A by name, these were the only oil dispersants used during the OSRC, and it is therefore reasonable to consider reported exposure to dispersants as reported exposure to either Corexit™ 9527A or Corexit™ 9500A. Office workers, workers who said no to all dispersant-related questions, and those who received safety training but did not work on the OSRC were assumed to be unexposed and were categorized as such for all analyses.

Using publicly available data on dates and locations of use of specific dispersants in the spill response (BP Gulf Science Data 2016a, 2016b), we categorized exposed participants as potentially exposed to 9500A only or as potentially additionally exposed to 9527A. Because 9527A was used in aerial applications only prior to 22 May 2010, we categorized participants who reported working on the relevant tasks during that period as potentially exposed to 9527A. Those who only worked on spraying dispersant from vessels or pumping dispersant to the wellhead and those who only reported working with dispersants after 22 May were classified as potentially exposed to 9500A only.

Additionally, among those classified as exposed in the respiratory and eye irritation analyses, participants were classified as directly working with dispersants if they had a positive response to any questionnaire item related to personally working with dispersants (see Table S1). Any participant who reported a positive response to a question about dispersant exposures but did not report this direct exposure was categorized as indirectly working with dispersants.

Outcome Assessment

Outcomes were based on participant responses to questions on the enrollment interview about the frequency of symptoms at the time of the OSRC or at the time of the enrollment interview. Participants reported the frequency of symptoms on an ordinal scale: “never,” “rarely,” “sometimes,” “most of the time,” or “all of the time,” and a symptom was considered present (yes vs. no) if the response was “most of the time” or “all of the time.” Five distinct respiratory symptoms (cough, wheeze, tightness in chest, shortness of breath, burning in nose/throat/lungs), one dermal symptom (≥2 d of eczema, dermatitis, other skin rashes, sores, or blisters), and two eye irritation symptoms (itchy eyes, burning eyes) were assessed.

Potential Confounders

We considered a variety of potential confounders. Demographic data including age, race, gender, and education level, as well as smoking status, employment status, financial worry, preexisting lung conditions, potential exposure to equipment decontamination chemicals and skin/clothing contact with oil or decontamination chemicals were assessed from responses on the enrollment interview. Residential proximity to the spill site was categorized based on the proximity of the county of residence to the Gulf [adjacent to the Gulf Coast, counties one county inland from the Gulf coast, other Gulf state (AL, FL, LA, MS, TX) counties, or non–Gulf state counties]. Approximate maximum daily time-weighted average airborne level of total hydrocarbons (THC) exposure over all OSRC tasks was estimated using an ordinal job exposure matrix (Kwok et al. 2017; Stewart et al. in press). Perceived stress was assessed using Cohen’s Perceived Stress Scale (Cohen et al. 1983).

Statistical Analyses

Owing to the cross-sectional nature of the data and the moderately high prevalence of the outcomes of interest among the study population (Table 1), we calculated adjusted prevalence ratios (aPRs) as the measure of effect rather than odds ratios because of ease of interpretation (Thompson et al. 1998). We fit multivariable log-binomial regression models to directly estimate the PR for each outcome in the exposed group compared with the unexposed group. Models were adjusted for a variety of a priori potential confounders depending on the outcome of interest. All models were adjusted for age at enrollment (<30, 30–45, >45), race (white, black, or other), gender, and education level (>high school degree or not). Models of dermal symptoms were also adjusted for skin or clothing contact with oil or tar (yes/no) and skin or clothing contact with equipment decontamination chemicals (yes/no). Models of both eye irritation and respiratory symptoms were also adjusted for residential proximity to the spill site, smoking status (never smoker, former smoker, light current smoker, or heavy current smoker), potential exposure to equipment decontamination chemicals (yes/no), and the maximum estimated airborne level of THC exposure (<0.3 ppm, 0.3–0.99 ppm, 1.00–2.99 ppm, ≥3.00 ppm) across all OSRC work. Models for respiratory symptoms were further adjusted for the presence of self-reported preexisting lung conditions (yes/no). All models of symptoms at the time of enrollment were further adjusted for employment status (employed, unemployed, disabled, retired), financial worry (yes/no), and Cohen’s Perceived Stress Scale (0–5, 6–10, ≥11) at enrollment.

Table 1. Enrollment characteristics of each analysis population.
Characteristic Respiratory analysis groupa(n=28,636) Eye irritation analysis groupa(n=29,468) Dermal analysis groupa (n=27,659)
Dispersant exposure(n=2,178) No dispersant exposure(n=26,458) Dispersant exposure(n=2,238) No dispersant exposure(n=27,230) Dispersant exposure(n=1,039) No dispersant exposure(n=26,620)
Age at enrollment, years
 <30 532 (24%) 5,182 (20%) 543 (24%) 5,285 (19%) 187 (18%) 5,241 (20%)
 30–45 years 911 (42%) 9,861 (37%) 932 (42%) 10,130 (37%) 458 (44%) 9,882 (37%)
 >45 735 (34%) 11,415 (43%) 763 (34%) 11,815 (43%) 394 (38%) 11,497 (43%)
Gender
 Male 1,961 (90%) 21,137 (80%) 2,014 (90%) 21,735 (80%) 962 (93%) 21,153 (79%)
 Female 217 (10%) 5,321 (20%) 224 (10%) 5,495 (20%) 77 (7%) 5,467 (21%)
Race
 White 952 (44%) 18,105 (68%) 976 (44%) 18,501 (68%) 481 (46%) 17,861 (67%)
 Black 986 (45%) 5,701 (22%) 1,011 (45%) 5,918 (22%) 411 (40%) 5,983 (22%)
 Other 240 (11%) 2,652 (10%) 251 (11%) 2,811 (10%) 147 (14%) 2,776 (10%)
Education
 High school education or less 1,318 (61%) 11,741 (44%) 1,355 (61%) 12,195 (45%) 625 (60%) 11,946 (45%)
 Greater than high school education 860 (39%) 14,717 (56%) 883 (39%) 15,035 (55%) 414 (40%) 14,674 (55%)
Maximum total hydrocarbon exposure, ppm
 <0.3 28 (1%) 5,040 (19%) 28 (1%) 5,136 (19%)
 0.3–0.99 369 (17%) 6,857 (26%) 373 (17%) 7,050 (26%)
 1.0–2.99 635 (29%) 5,966 (23%) 658 (29%) 6,147 (23%)
 ≥3.0 1,146 (53%) 1,949 (7%) 1,179 (53%) 2,002 (7%)
Exposed to decontamination chemicals
 Yes 1,612 (74%) 5,068 (19%) 1,656 (74%) 5,254 (19%)
 No 566 (26%) 21,390 (81%) 582 (26%) 21,976 (81%)
Residential proximity to spill
 Directly affected county 1,301 (60%) 14,208 (54%) 1,338 (60%) 14,705 (54%)
 Indirectly affected county 163 (7%) 1,717 (6%) 164 (7%) 1,778 (7%)
 Other Gulf state county 465 (21%) 5,434 (21%) 483 (22%) 5,582 (20%)
 Non–Gulf state residence 249 (11%) 5,099 (19%) 253 (11%) 5,165 (19%)
Smoking status
 ≥1 pack per day 229 (11%) 2,687 (10%) 233 (10%) 2,768 (10%)
 <1 pack per day 629 (29%) 4,888 (18%) 640 (29%) 5,057 (19%)
 Former smoker 357 (16%) 5,865 (22%) 369 (16%) 6,022 (22%)
 Never smoker 963 (44%) 13,018 (49%) 996 (45%) 13,383 (49%)
Preexisting lung condition
 Yes 263 (12%) 3,186 (12%)
 No 1,915 (88%) 23,272 (88%)
Any relevant PPE useb
 Yes 777 (48%) 2,149 (21%) 1,012 (97%) 13,284 (89%)
 No 839 (52%) 8,196 (79%) 27 (3%) 1,684 (11%)
Skin or clothing contact with oil/tar
 Yes 1,006 (97%) 8,306 (31%)
 No 33 (3%) 18,314 (69%)
Skin or clothing contact with decontamination chemicals
 Yes 624 (60%) 1,434 (5%)
 No 415 (40%) 25,186 (95%)

Note: PPE, personal protective equipment.

aDashes (–) indicate variables that were not examined as covariates in that analysis population.

bn=11,961 for respiratory analysis population, n=16,007 for dermal analysis population.

Because nonconvergence can be an issue with log-binomial regression owing to the model’s restricted parameter space, we used the weighted COPY method outlined by Deddens and Petersen (2008) to approximate the maximum likelihood estimates and related standard errors for any models that did not initially converge, using 1,000,000 virtual copies of the data set. All analyses were performed using SAS version 9.3 (SAS Institute Inc.). We used a significance level of p<0.05 for all analyses.

Sensitivity/Secondary Analyses

Because these data are self-reported, there is a potential for recall and reporting bias to influence the participant’s reporting of symptoms. We used self-reported excessive hair loss at the time of the OSRC or in the 30 d before enrollment to identify participants who may have over-reported their symptoms because there is no known biological mechanism that would relate dispersant exposure to excessive hair loss. All analyses were repeated after restricting the study populations to those without self-reported excessive hair loss at the time point of interest (i.e., symptoms during the spill or within 30 d of enrollment) to help assess any potential impact of reporting bias on the results.

Some nonworkers who completed safety training but were not hired to work on the OSRC may have had preexisting health conditions or other factors (such as obesity) that either prevented them from working or made them less attractive to contractors charged with staffing the cleanup effort. As such, having them as a part of the comparison group in the analyses could result in biased estimates in comparison with an analysis comparing exposed workers with unexposed workers. To help assess any impact of such a potential healthy-worker selection effect, we repeated all analyses with nonworkers excluded from the analysis.

Because some participants who worked on land reported working directly with dispersants, there was concern that participants may have confused dispersants with chemicals used to clean and decontaminate equipment because these chemicals were used to “disperse” the oil from the equipment. Therefore, we conducted a sensitivity analysis excluding any respondents whose only exposure was reported handling of dispersants on land who also reported active participation in equipment decontamination activities. We also conducted sensitivity analyses excluding exposed workers who reported being exposed outside the known dates of dispersant use; furthermore, we performed sensitivity analyses assessing potential confounding resulting from relevant personal protective equipment (PPE) use during the OSRC (i.e., respirators/facemasks for respiratory symptoms, rubber gloves or coveralls for dermal symptoms). We also assessed potential effect measure modification by PPE (i.e., respirator or face mask) use among respiratory outcomes. More than 97% of those reporting dermal exposure also reported the use of PPE (i.e., gloves, protective clothing); therefore, there was insufficient heterogeneity to investigate potential effect modification of dispersant contact by this variable (Table 1). In the questionnaire, we did not ascertain whether PPE was worn during specific tasks, so here, we used overall PPE use as a proxy.

Among OSRC workers, models of respiratory and eye irritation were stratified based on three mutually exclusive worker locations—ever worked on the water in the area of the wellhead, ever worked on the water but not near the wellhead, or worked only on land—to capture potential environmental differences between the locations. For example, workers near the wellhead area would likely have been exposed to increased particulate matter from the flaring of oil/gas by two vessels in the wellhead area. For models of symptoms within 30 d of enrollment, participants were also stratified by the reported presence/absence of the symptom of interest at the time of the spill in order to investigate persistence across time points.

Given that those exposed to dispersants largely worked in areas where they might have also had exposure to THC, we investigated potential effect modification of the dispersant by the estimated maximum airborne level of THC (<0.3 ppm, 0.3–0.99 ppm, 1.0–2.99 ppm, and ≥3.0 ppm) over all OSRC jobs. We also investigated potential effect modification by exposure to decontamination chemicals. Because 97% of those reporting skin/clothing contact with dispersants also reported skin/clothing contact with oil/tar (Table 1), there was insufficient heterogeneity to investigate whether oil/tar contact modified the effect of dispersant contact.

Results

Table 1 shows the baseline characteristics of participants in each analytic study population. Approximately 7.6% of participants included in the respiratory symptom and eye irritation analyses were considered exposed to dispersants. Among those included in the analysis of dermal symptoms, 3.8% had direct skin or clothing contact with dispersants.

The study population was overwhelmingly male; approximately 80% of the unexposed were men as were ∼90% of the exposed. Those exposed to dispersants were more likely to be nonwhite, less likely to have any education beyond high school, and more likely to be <45 y old compared with those who were unexposed. For the respiratory and eye irritation analyses, those exposed were more likely to be current smokers (39% vs. 29% for the unexposed) and live in Gulf Coast counties. Given where dispersants were used during the OSRC, it is unsurprising that those exposed to dispersants were more likely to have also been exposed to levels of airborne THC >3.0 ppm (ppm) (53% vs. 7%) and to have worked with equipment decontamination chemicals (74% vs. 19%). Among those included in the respiratory analyses, there was no difference in the prevalence of preexisting lung conditions between the exposed and unexposed groups. For the dermal analyses, those exposed to dispersants were substantially more likely to have also come into contact with oil or tar (97% vs. 31% for the unexposed group) as well as more likely to have come into contact with equipment decontamination chemicals (60% vs. 5%).

Table 2 presents the prevalence of each symptom reported at the time of the OSRC along with aPRs. The prevalence of each respiratory symptom reported at the time of the OSRC (cough, wheeze, shortness of breath, tightness in the chest, and burning in the nose, throat, or lungs) was significantly higher in the exposed group than in the unexposed group, with aPRs ranging from 1.36 [95% confidence interval (CI): 1.23, 1.52] for wheeze to 1.61 (95% CI: 1.42, 1.83) for burning in the nose, throat, or lungs. Similarly, the adjusted prevalences at the time of the OSRC for burning in the eyes and for itching in the eyes, as well as the adjusted prevalence for ≥2 d of itching or dermatitis, were significantly higher in the exposed group than in the unexposed group. Direct exposure to dispersants was more strongly associated with each respiratory and eye irritation outcome at the time of the OSRC than was indirect exposure, with nonoverlapping confidence intervals for shortness of breath (Table 3).

Table 2. Symptoms at spill response associated with dispersant exposure (GuLF STUDY, 2011–2013).
Symptom Exposed [n (%)] Unexposed [n (%)] aPR (95% CI)a
Coughb 534 (25%) 2,642 (10%) 1.41 (1.28, 1.55)
Wheezeb 426 (20%) 2,050 (8%) 1.36 (1.23, 1.52)
Tightness in chestb 305 (14%) 1,248 (5%) 1.58 (1.38, 1.81)
Shortness of breathb 387 (18%) 1,827 (7%) 1.41 (1.26, 1.58)
Burning in nose, throat, lungsb 367 (17%) 1,325 (5%) 1.61 (1.42, 1.83)
Burning eyesc 512 (23%) 2,261 (8%) 1.49 (1.35, 1.64)
Itching eyesc 659 (29%) 3,362 (12%) 1.35 (1.24, 1.46)
Skin irritationd 548 (53%) 4,345 (16%) 1.34 (1.25, 1.43)

Note: aPR, adjusted prevalence ratio; CI, confidence interval.

aAll models adjusted for gender, age, race, education. Skin irritation models further adjusted for contact with oil/tar, contact with cleaning chemicals, and dispersant/oil interaction. Respiratory and eye irritation models at spill further adjusted for smoking, residential proximity to oil spill, level of oil exposure (total hydrocarbons, THC), use of decontamination chemicals, and preexisting lung disease (respiratory models).

bn=28,636 (2,178 exposed, 26,458 unexposed).

cn=29,468 (2,238 exposed, 27,230 unexposed).

dn=27,659 (1,039 exposed, 26,620 unexposed).

Table 3. Respiratory and eye irritation symptoms at the time of spill response associated with dispersant exposure, differentiating exposure by direct or indirect exposure (GuLF STUDY 2011–2013).
Outcome Direct work with dispersants Indirect work with dispersants
PRa (95% CI) PRa (95% CI)
Cough 1.47 (1.31, 1.63) 1.29 (1.13, 1.47)
Wheeze 1.45 (1.28, 1.63) 1.17 (1.01, 1.36)
Tightness in chest 1.74 (1.48, 2.04) 1.30 (1.07, 1.58)
Shortness of breath 1.63 (1.43, 1.85) 1.07 (0.90, 1.27)
Burning in nose/throat/lungs 1.75 (1.52, 2.02) 1.30 (1.09, 1.55)
Burning eyes 1.58 (1.41, 1.76) 1.28 (1.12, 1.47)
Itchy eyes 1.39 (1.27, 1.52) 1.17 (1.05, 1.31)

Note: CI, confidence interval; PR, prevalence ratio.

aAdjusted for gender, age, race, education, smoking, residential proximity to oil spill, level of oil exposure (total hydrocarbons, THC), use of decontamination chemicals, and preexisting lung disease (respiratory models).

For most symptoms at the time of the OSRC, aPRs were higher for possible exposure to 9527A than for exposure to only 9500A, although the aPRs were not markedly different except for tightness in the chest [aPR=1.79 (95% CI: 1.45, 2.21) vs. aPR=1.33 (95% CI: 1.10, 1.63), respectively] and burning in the nose, throat, or lungs [aPR=1.82 (95% CI: 1.52, 2.19) vs. aPR=1.22 (95% CI: 1.01, 1.47) respectively], and only the latter symptom had nonoverlapping confidence intervals (see Table S2).

There was little difference in the associations between symptoms and dispersant use in analyses stratified by work location, and the confidence intervals overlapped, although aPRs tended to be higher among those who worked on the water away from the wellhead than among those who did not work on the water and those who worked near the wellhead (see Table S3). Exclusion of participants reporting hair loss (n=578, n=612, and n=617 for respiratory, eye, and dermal symptoms, respectively) did not materially change any of the PR estimates or confidence intervals, although there was a nonsignificant positive association between self-reported hair loss and dispersant exposure [aPR=1.24 (95% CI: 0.95, 1.61)]. Similarly, excluding participants whose only dispersant exposure was on land and who also worked with cleaning chemicals did not affect the results, indicating that any potential misclassification resulting from confusing cleaning chemicals with dispersants was minor. Exclusion of nonworkers and exclusion of participants with invalid self-reported dates of dispersant use also had negligible effects on the results.

Stratification by the estimated maximum level of THC exposure over all OSRC jobs did not reveal any meaningful differences in the associations between dispersant exposure and any of the respiratory and eye irritation symptoms at the time of the spill (see Table S4). Reported PPE use at any time during the OSRC did not confound the association between dispersant exposure and any of the respiratory or dermal symptoms at the time of the spill, and there was no significant evidence of effect measure modification among any of the respiratory symptoms.

For each symptom, the aPR was lower for symptoms present at the time of study enrollment than for symptoms present at the time of the OSRC (Table 4). However, dispersant exposure remained significantly associated with the prevalence of symptoms at the time of study enrollment, with the exception of cough and skin irritation.

Table 4. Symptoms within 30 days of study enrollment associated with dispersant exposure (GuLF STUDY, 2011–2013).
Symptom Exposed [n (%)] Unexposed [n (%)] aPR (95% CI)a
Cougha 594 (27%) 3,896 (15%) 1.03 (0.96, 1.11)
Wheezea 479 (22%) 2,400 (9%) 1.16 (1.06, 1.26)
Tightness in chesta 337 (16%) 1,455 (6%) 1.30 (1.16, 1.46)
Shortness of breatha 434 (20%) 2,129 (8%) 1.15 (1.06, 1.26)
Burning in nose, throat, lungsa 246 (11%) 1,005 (4%) 1.55 (1.34, 1.80)
Burning eyesc 363 (17%) 1,699 (6%) 1.44 (1.28, 1.61)
Itching eyesc 482 (22%) 2,758 (11%) 1.24 (1.12, 1.36)
Skin irritationd 207 (21%) 8,891 (35%) 0.84 (0.74, 0.95)

Note: aPR, adjusted prevalence ratio; CI, confidence interval.

aAll models adjusted for gender, age, race, education, unemployment, disability, financial and perceived stress. Skin irritation models further adjusted for contact with oil/tar, contact with cleaning chemicals, and dispersant/oil interaction. Respiratory and eye irritation models at spill further adjusted for smoking, residential proximity to oil spill, level of oil exposure (total hydrocarbons, THC), use of decontamination chemicals, and preexisting lung disease (respiratory models).

bn=28,183 (2,163 exposed, 26,020 unexposed).

cn=28,363 (2,181 exposed, 26,182 unexposed).

dn=26,249 (1,001 exposed, 25,248 unexposed).

Among participants who reported the presence of a given symptom at the time of the OSRC, dispersant use remained significantly associated only with burning eyes at the time of study enrollment. However, among those participants who did not report a given symptom at the time of the OSRC, dispersant use was significantly associated with all outcomes except cough and itching eyes at the time of study enrollment (see Table S5). Exposure to dispersants was associated with a decreased likelihood of reported skin irritation within 30 d of enrollment among those who did not report skin irritation at the time of the spill.

Discussion

This study is the first to evaluate associations between potential exposure to dispersants, specifically Corexit™ EC9527A or EC9500A, and respiratory, eye irritation, and dermal symptoms both during the OSRC and at study enrollment 1–3 y later. OSRC workers with potential exposure to either Corexit™ product were more likely to have reported adverse symptoms at the time of the spill. Previous studies have shown an association between exposure to crude oil and adverse effects among spill responders (Aguilera et al. 2010; Laffon et al. 2016), but no previous spill involved this level of dispersant use or resulted in an investigation of potential associations between dispersant use and adverse health effects. In vitro results suggested that 9527A and 9500A may have adverse effects on human lung tissue (Major et al. 2016; Shi et al. 2013). Our results provide epidemiological evidence to suggest that exposure to 9527A or 9500A may be associated with adverse health effects, even after taking into account exposure to the crude oil.

Based on the known irritant properties of chemicals in the dispersants, we hypothesized that there might be acute effects at the time of the cleanup, but we did not expect there to be longer-term effects at the time of enrollment. Although we observed associations between dispersant use and symptoms at both time points, among participants with a given symptom at the time of the OSRC, only increased prevalence of burning eyes at enrollment remained significantly associated with dispersant exposure, consistent with a lack of persistent effects of the dispersants. The significant associations between exposure and symptoms at the time of enrollment among those who did not have symptoms at the time of the OSRC were unexpected and are difficult to explain. Although it is possible that these associations are measuring some latent effect of exposure to the dispersants, another possibility is that some of these symptoms may have been present at the time of the OSRC but were not intense enough for the study participants to recall 1–3 y later. The inverse association between skin/clothing contact with dispersant during the OSRC and skin irritation reported at the time of enrollment was also unexpected and is difficult to explain. Many media reports at the time of the study linked skin lesions with work or recreational activities involving contact with water from the Gulf of Mexico (Marsa 2016; Landau 2010). We were unable to account for current recreational contact with the water in this analysis.

As would be expected, direct work with dispersants was more strongly associated with the respiratory and eye irritation outcomes than indirect exposure through working in an area where dispersants were used. Even so, for most symptoms, indirect exposure was significantly associated with the symptom, indicating that these likely lower exposures may also be important. Stratification by airborne level of THC exposure showed no evidence for effect modification by THC on the associations between dispersant exposure and either respiratory or eye irritation symptoms.

The exposure measures used in this analysis were based on self-reported responses to questions about work locations and dispersant-related tasks and do not allow exploration of exposure–response relationships. A quantitative job exposure matrix for Corexit™ exposure that takes into account the chemical and physical properties of the chemicals and external information on patterns of use may allow evaluation of exposure–response relationships in the future.

The GuLF STUDY is the largest prospective study of OSRC workers to date, and it provides an excellent opportunity to investigate less-common spill-related exposures and health outcomes (Kwok et al. 2017). The detailed questionnaire provided the opportunity to assess previously understudied health effects associated with dispersants while also taking into account a wide variety of potentially confounding factors.

However, our approach relied almost entirely on self-reported data, which provides several opportunities for bias. When possible, these potential sources of bias were investigated using a variety of sensitivity analyses. One potential concern is the over-reporting of symptoms. We addressed this concern by investigating the relationship between each exposure metric and self-reported unusual amount of hair loss at the time of the spill, which does not have any known biological relationship to exposure to either 9527A or 9500A. The positive association between dispersant exposure and self-reported excessive hair loss, although not statistically significant, suggests the possibility of bias due to over-reporting. However, excluding participants who reported excessive hair loss did not meaningfully change the results, suggesting that over-reporting does not explain our findings. Similarly, exclusion of nonworkers did not result in a meaningful difference in any of the results, indicating no appreciable effect on the overall associations resulting from nonworkers potentially having preexisting worse health than workers (i.e., a healthy worker selection effect). Associations between dispersant exposure and each outcome were somewhat stronger among workers who spent time on the water away from the wellhead than among workers who worked only on land or those workers who worked near the wellhead. However, associations at all work locations remained significant, providing evidence that the overall associations were not being driven by unmeasured characteristics of a particular work location.

Misclassification of exposure is another potential problem because of the reliance on self-reported information about the work performed. For example, there was some evidence in open-ended responses in the questionnaire that some participants were confused by the term “dispersant” when responding to questions about decontamination tasks. We attempted to address this issue by excluding participants who reported only a land-based exposure and also reported working on equipment decontamination. The results of that analysis are qualitatively similar to the overall results, indicating that this potential misclassification was unlikely to have played an appreciable role in our results.

It would be expected that the proper use of PPE would help reduce received exposure and any potential adverse effects of this exposure. The MSDSs for both dispersants recommend the use of gloves and standard protective clothing, as well as the use of respirators when concentrations exceed recommended limits (NALCO 2012a, 2012b). Measurements taken by BP and by the National Institute for Occupational Safety and Health (NIOSH) during the OSRC indicate that it is unlikely that airborne concentrations of either 2-butoxyethanol or propylene glycol exceeded recommended limits (BP Gulf Science Data 2016c; NIOSH 2010). No measurements were available for airborne concentrations of DOSS, nor were any dermal exposure measurements during the OSRC available. Although we had no way to ascertain if PPE was used specifically during dispersant-related tasks, a sensitivity analysis among participants who reported PPE use during the OSRC indicated no confounding of the main association by reported PPE use, nor any effect measure modification among respiratory outcomes.

Although our results suggest an association between exposure to 9527A, 9500A, or both and adverse acute symptoms, we were not able to completely distinguish these exposures. Participants who were potentially exposed to 9527A, as identified by date and method of use, reported slightly higher prevalence of most symptoms than those who would have been exposed to only 9500A. Although this outcome could have been caused by the presence of more acutely toxic agents in 9527A, a substantially larger quantity of dispersants was applied during the early period of the OSRC, when both dispersants were being used, than in the later stages of the OSRC, when only 9500A was used (BP Gulf Science Data 2016a, 2016b).

Conclusion

Our findings suggest associations between exposure to dispersants, specifically Corexit™ EC9527A or Corexit™ EC9500A, and adverse acute health effects at the time of the OSRC as well as with symptoms that were present at the time of study enrollment 1–3 y later.

Acknowledgements

The authors thank A. Hodges, J. McGrath, and the rest of the staff at Social and Scientific Systems for data collection and management. We also thank the GuLF STUDY cohort members.

This study was funded by the National Institutes of Health (NIH) Common Fund and the Intramural Program of the National Institute of Environmental Health Sciences/NIH (ZO1 ES 102945).

References

Acros Organics. 2013. Dioctyl sulfosuccinate, sodium salt safety data sheet. http://194.7.155.215/DirectWebViewer/private/document.aspx?prd=ACR11710∼∼PDF∼∼MTR∼∼CLP1∼∼EN∼∼2013-12-11%2012:08:23∼∼Dioctyl%20sulfosuccinate%20%20sodium%20salt∼∼ [accessed 15 July 2015].

Aguilera F, Mendez J, Pasaro E, Laffon B. 2010. Review on the effects of exposure to spilled oils on human health. J Appl Toxicol 30(4):291–301, PMID: 20499335, 10.1002/jat.1521.

Anderson SE, Franko J, Lukomska E, Meade BJ. 2011. Potential immunotoxicological health effects following exposure to Corexit 9500a during cleanup of the Deepwater Horizon oil spill. J Toxicol Environ Health Part A 74(21):1419–1430, PMID: 21916747, 10.1080/15287394.2011.606797.

ATSDR (Agency for Toxic Substances and Disease Registry). 1997. Toxicological Profile for Propylene Glycol. Atlanta, GA:U.S. Department of Health and Human Services, Public Health Service, ATSDR.

ATSDR. 1998. Toxicological Profile for 2-Butoxyethanol and 2-Butoxyethanol Acetate. Atlanta, GA:U.S. Department of Health and Human Services, Public Health Service, ATSDR.

BP Gulf Science Data. 2016a. Application of dispersants to surface oil slicks by aircraft and by boat in approved areas of the Gulf of Mexico from April 22, 2010 to final application on July 19, 2010. Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC). https://data.gulfresearchinitiative.org/data/BP.x750.000:0017 [accessed 3 February 2016].

BP Gulf Science Data. 2016b. Subsea dispersant application records collected during the Deepwater Horizon (DWH) accident near the Mississippi Canyon block 252 wellhead from April 30 to July 22, 2010. Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC). https://data.gulfresearchinitiative.org/data/BP.x750.000:0018 [accessed 16 January 2017].

BP Gulf Science Data. 2016c. Monitoring of the personal breathing zone of response workers for chemicals and oil mists from April 2010 to January 2012. Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC). https://data.gulfresearchinitiative.org/data/BP.x750.000:0007 [accessed 16 January 2017]

Chapman H, Purnell K, Law RJ, Kirby MF. 2007. The use of chemical dispersants to combat oil spills at sea: A review of practice and research needs in Europe. Mar Pollut Bull 54(7):827–838, PMID: 17499814, 10.1016/j.marpolbul.2007.03.012.

Cohen S, Kamarck T, Mermelstein R. 1983. A global measure of perceived stress. J Health Soc Behav 24(4):385–396, PMID: 6668417.

Deddens JA, Petersen MR. 2008. Approaches for estimating prevalence ratios. Occup Environ Med 65(7):501–506, PMID: 18562687, 10.1136/oem.2007.034777.

Hemmer MJ, Barron MG, Greene RM. 2010a. Comparative toxicity of eight oil dispersant products on two Gulf of Mexico aquatic test species. EPA/600/R-11/113. Washington, DC:U.S. Environmental Protection Agency. https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=226426 [accessed 19 May 2015].

Hemmer MJ, Barron MG, Greene RM. 2010b. Comparative toxicity of Louisiana Sweet Crude oil (LSC) and chemically dispersed LSC to two Gulf of Mexico aquatic test species. EPA/600/R-11/112. Washington, DC:
U.S. Environmental Protection Agency. https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=228203 [accessed 19 May 2015].

Kleindienst S, Seidel M, Ziervogel K, Grim S, Loftis K, Harrison S, et al. 2015. Chemical dispersants can suppress the activity of natural oil-degrading microorganisms. Proc Natl Acad Sci USA 112(48):14900–14905, PMID: 26553985, 10.1073/pnas.1507380112.

Kujawinski EB, Kido Soule MC, Valentine DL, Boysen AK, Longnecker K, Redmond MC. 2011. Fate of dispersants associated with the Deepwater Horizon oil spill. Environ Sci Technol 45(4):1298–1306, PMID: 21265576, 10.1021/es103838p.

Kwok RK, Engel LS, Miller AK, Blair A, Curry MD, Jackson WB, et al. 2017. The GuLF STUDY: A prospective study of persons involved in the Deepwater Horizon oil spill response and clean-up. Environ Health Perspect 125(4):570–578, 10.1289/EHP715.

Laffon B, Pásaro E, Valdiglesias V. 2016. Effects of exposure to oil spills on human health: Updated review. J Toxicol Environ Health B Crit Rev 19(3–4):105–128, PMID: 27221976, 10.1080/10937404.2016.1168730.

Landau E. 2010. Gulf states tracking oil-related illnesses. CNN. http://www.cnn.com/2010/HEALTH/06/09/oil.spill.health.impact/index.html [accessed 27 July 2017].

Major D, Derbes RS, Wang H, Roy-Engel AM. 2016. Effects of Corexit oil dispersants and the WAF of dispersed oil on DNA damage and repair in cultured human bronchial airway cells, BEAS-2B. Gene Rep 3:22–30, PMID: 27563691, 10.1016/j.genrep.2015.12.002.

Marsa L. 2016. Deepwater Horizon oil disaster extends its toxic reach. Newsweek. http://www.newsweek.com/2016/10/21/deepwater-horizon-bp-oil-spill-sickened-gulf-residents-508362.html [accessed 31 July 2017].

NALCO Environmental Solutions LLC. 2012a. Corexit™ EC9527A safety data sheet. http://www.nalcoenvironmentalsolutionsllc.com/wp-content/uploads/COREXIT%E2%84%A2-EC9527A-GHS-SDS-USA.pdf [accessed Day Month Year].

NALCO Environmental Solutions LLC. 2012b. Corexit™ EC9500A safety data sheet. http://www.nalcoenvironmentalsolutionsllc.com/wp-content/uploads/COREXIT%E2%84%A2-EC9500A-GHS-SDS-USA.pdf [accessed Day Month Year].

NIOSH (National Institute for Occupational Safety and Health). 2010. Health hazard evaluation of Deepwater Horizon response workers. http://www.cdc.gov/niosh/topics/oilspillresponse/xls/niosh_sampling_data_updated_11.17.2010.xls [accessed 24 Jan 2017]

Prince RC, McFarlin KM, Butler JD, Febbo EJ, Wang FCY, Nedwed TJ. 2013. The primary biodegradation of dispersed crude oil in the sea. Chemosphere 90(2):521–526, PMID: 22967931, 10.1016/j.chemosphere.2012.08.020.

Roberts JR, Reynolds JS, Thompson JA, Zaccone EJ, Shimko MJ, Goldsmith WT, et al. 2011. Pulmonary effects after acute inhalation of oil dispersant (Corexit EC9500A) in rats. J Toxicol Environ Health Part A 74(21):1381–1396, PMID: 21916744, 10.1080/15287394.2011.606794.

Shi Y, Roy-Engel AM, Wang H. 2013. Effects of Corexit dispersants on cytotoxicity parameters in a cultured human bronchial airway cells, beas-2b. J Toxicol Environ Health A 76(13):827–835, PMID: 24028667, 10.1080/15287394.2013.821396.

Stewart T, Stenzel M, Kwok RK, Engel L, Ramachandran G, Banerjee S, et al. In press. Development of a job-exposure matrix for workers in the GuLF STUDY responding to the Deepwater Horizon disaster. J Expo Sci Environ Epidemiol.

Thompson ML, Myers JE, Kriebel D. 1998. Prevalence odds ratio or prevalence ratio in the analysis of cross sectional data: what is to be done? Occup Environ Med 55(4):272–277, PMID: 9624282, 10.1136/oem.55.4.272.

United States Coast Guard. National Response Team (U.S.). 2011. On Scene Coordinator Report: Deepwater Horizon Oil Spill. Washington, DC:U.S. Dept. of Homeland Security, U.S. Coast Guard.

Wise J, Wise JP. 2011. A review of the toxicity of chemical dispersants. Rev Environ Health 26(4):281–300, PMID: 22435326, 10.1515/REVEH.2011.035.

Zock JP, Rodríguez-Trigo G, Rodríguez-Rodríguez E, Espinosa A, Pozo-Rodríguez F, Gómez F, et al. 2012. Persistent respiratory symptoms in clean-up workers 5 years after the Prestige oil spill. Occup Environ Med 69(7):508–513, PMID: 22539655, 10.1136/oemed-2011-100614.

Prenatal Exposure to Nonpersistent Endocrine Disruptors and Behavior in Boys at 3 and 5 Years

Author Affiliations open

1Institute for Advanced Biosciences, INSERM U1209, CNRS UMR 5309, University Grenoble Alpes, Grenoble, France

2Centers for Disease Control and Prevention, Atlanta, Georgia, USA

3U1153 Epidemiology and Biostatistics Sorbonne Paris Cité Research Centre (CRESS), Early Origin of the Child’s Health and Development (ORCHAD) Team, Inserm, Villejuif, France

4Université Paris Descartes, Villejuif, France

5Faculty of Pharmacy, Université Paris-Sud, Université Paris-Saclay, Châtenay-Malabry, France

PDF icon PDF Version (197 KB)

  • Background:
    Sex-specific associations have been reported between phthalates, bisphenol A (BPA), and child behavior. No data on large study populations are available for other phenols with possible endocrine-disrupting properties.
    Objectives:
    We aimed to study associations between prenatal exposure to phthalates and several phenols on behavior among male infants.
    Methods:
    We quantified 11 phthalate metabolites and nine phenols (four parabens, benzophenone-3, BPA, two dichlorophenols, triclosan) in spot urine samples collected during pregnancy among EDEN cohort mothers who delivered a boy. Mothers completed the Strength and Difficulties Questionnaire (SDQ) when their children were 3.1 (n=529) and 5.6 (n=464) y old.
    Results:
    BPA was positively associated with the relationship problems subscale at 3 y [incidence rate ratio (IRR): 1.11; 95% confidence interval (CI): 1.03, 1.20] and the hyperactivity–inattention subscale scores at 5 y (IRR: 1.08; 95% CI: 1.01, 1.14). Mono-n-butyl phthalate (MnBP) was positively associated with internalizing behavior, relationship problem, and emotional symptom scores at 3 y. Monobenzyl phthalate (MBzP) was positively associated with internalizing behavior and relationship problems scores at 3 y. After dichotomizing SDQ scores, triclosan tended to be positively associated with emotional symptom subscales at both 3 and 5 y.
    Conclusions:
    The observed associations between BPA, MnBP, and behavior in boys are consistent with previous findings. Further health impact assessment studies based on dose–response functions corrected for exposure misclassification are required to quantify the public health burden possibly entailed by such associations. https://doi.org/10.1289/EHP1314
  • Received: 02 November 2016
    Revised: 23 May 2017
    Accepted: 19 June 2017
    Published: 15 September 2017

    Address correspondence to C. Philippat, Institut for Advanced Biosciences, Site Santé – Allée des Alpes, 38700 La Tronche, France. Phone: +33 4 76 54 94 66, Email: claire.philippat@inserm.fr

    Supplemental Material is available online (https://doi.org/10.1289/EHP1314).

    The EDEN Mother–Child Cohort Study Group includes: I. Annesi-Maesano, J. Bernard, J. Botton, M.A. Charles, P. Dargent-Molina, B. de Lauzon-Guillain, P. Ducimetière, M. de Agostini, B. Foliguet, A. Forhan, X. Fritel, A. Germa, V. Goua, R. Hankard, B. Heude, M. Kaminski, B. Larroque, N. Lelong, J. Lepeule, F Pierre, L. Marchand, C. Nabet, R. Slama, M.J. Saurel-Cubizolles, M. Schweitzer, O. Thiebaugeorge.

    The authors declare they have no actual or potential competing financial interests.

    The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (258 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (173 KB)

Introduction

Neurodevelopment is a complex process that starts in humans as early as the second gestational week with neurulation and continues through adolescence (Rice and Barone 2000). Neurogenesis, neuron proliferation, migration, and differentiation of cells of the nervous system as well as synaptogenesis happen during fetal and early postnatal life. Disruption of these processes might be deleterious for the nervous system and may lead to neurodevelopmental disorders later in life. Neurodevelopmental disorders, such as intellectual disability, attention deficit/hyperactivity disorders, learning disabilities, or autism spectrum disorders affect about 12% of children worldwide (National Institute of Health and Medical Research 2002). Most disorders are likely to be caused by a combination of genetic and environmental factors (Sandin et al. 2014; van Loo and Martens 2007).

Several environmental chemicals, such as lead, mercury, and polychlorinated biphenyls, have been identified as neurotoxic, while a wide range of less extensively studied chemicals, including phenols and phthalates, are also suspected to affect child neurodevelopment (Grandjean and Landrigan 2006). Phenols include bisphenol A (BPA), a component of polycarbonate plastics and epoxy resins used in many consumer products (e.g., digital media, construction glazing, some toys, medical devices, food packaging); triclosan, an antibacterial agent used in personal care products such as antibacterial soaps or toothpastes; benzophenone-3, an ultraviolet (UV) filter; dichlorophenols, metabolites of intermediates used in the production of several herbicides and insecticides; and parabens used as preservatives in cosmetics and food. Phthalates are also used in a wide range of products, including personal care products (e.g., cosmetics, fragrances, shampoos), food packaging, and indoor residential environments (e.g., polyvinyl chloride flooring and wall covering, vinyl tiles, and shower curtains) (Hauser and Calafat 2005; Philippat et al. 2015).

Phenols and phthalates can interact with pathways that are crucial for the development of the nervous system. Some of these chemicals can indeed disrupt hormonal pathways and calcium signaling [reviewed by Miodovnik et al. (2014) for phthalates and Mustieles et al. (2015) for BPA]. Most epidemiological studies that evaluated potential associations between early life exposure to phenols and behavioral outcomes in children focused on BPA and reported effects that were sex-specific (Braun et al. 2011; Casas et al. 2015; Evans et al. 2014; Harley et al. 2013; F Perera et al. 2012; Perera et al. 2016; Roen et al. 2015). In vitro studies suggested endocrine-disrupting properties for other phenols, such as parabens and triclosan. Estrogenic effects have been indeed reported for parabens (Golden et al. 2005), while triclosan has been suspected to affect thyroid hormone homeostasis (Wu et al. 2016). However, to our knowledge, no study has investigated the associations between these other phenols and child behavior. Regarding phthalates, most epidemiological studies reported increased behavioral problems with prenatal exposure to at least one phthalate (Engel et al. 2010; Kobrosly et al. 2014; Lien et al. 2015; Whyatt et al. 2012), with the exception of one study conducted in Spain (Gascon et al. 2015). However, the type of behavior (e.g., hyperactivity, conduct problems) and the phthalate implicated often varied across studies.

We aimed at studying the associations between prenatal exposure to several phenols and phthalates and behavior among boys at 3 and 5 y.

Population and Methods

Study Population

The study population consisted of a subsample of mother–son pairs from the EDEN (study of pre- and early postnatal determinants of the child’s development and health) cohort that recruited pregnant women in the obstetric departments of Nancy and Poitiers University hospitals (France) between February 2003 and January 2006. Participating in the study was proposed to all women visiting the prenatal clinics of Nancy and Poitiers University hospitals before their 24th week of amenorrhea. Exclusion criteria included multiple pregnancies, known diabetes before pregnancy, French illiteracy, or planning to move out of the region within the next 3 y. More details on the cohort are available in previous paper (Heude et al. 2016).

Phenol and phthalate biomarker concentrations were assessed for the mother of 604 boys of the EDEN cohort in the framework of previous projects that aimed at studying associations with growth (Botton et al. 2016; Philippat et al. 2012, 2014) and male genital anomalies (Chevrier et al. 2012). Selection criteria for these previous projects were: being a boy, the mother having one urine sample available during pregnancy for biomarker assessment, and having data on growth during the pre- and postnatal period (up to 3 y). Because of limited funding, we focused on a single sex rather than on two equal-size groups of girls and boys. Among the 604 children included in these previous projects, 546 had data on behavior at 3 and/or 5 y and were included in the present study.

The EDEN cohort received approval from the ethics committee of Kremlin-Bicêtre University Hospital. The participants gave informed written consent for themselves and for their children to be included in the cohort.

Assessment of Child Behavior

Behavior was assessed at 3.1 [standard deviation (SD): 0.1] and 5.6 (SD: 0.2) y using the Strength and Difficulties Questionnaire (SDQ) completed by the mother (Goodman 1997). This questionnaire includes 25 items scored on a 3-point scale (0: not true; 1: somewhat true; 2: certainly true) that were combined into 4 difficulties subscores: conduct problems, hyperactivity–inattention, peer relationship problems, and emotional symptoms subscores, and one strength subscore: prosocial behavior. The score for each subscale ranged from 0 to 10. We then computed an externalizing (sum of the conduct problems and hyperactivity–inattention scores) and an internalizing behavior score (sum of the peer relationship problems and emotional symptoms scores). Score of the strength subscale was reversed prior to analysis so that for all SDQ subscales higher scores meant increased difficulties. Construct of each subscore is detailed in Table S1.

Exposure Assessment

Women were asked to collect first morning urine void at home just before the prenatal study visit. If forgotten, urine collection was done at the hospital during the prenatal visit. We quantified 9 phenols and 11 phthalate metabolites in maternal spot urine samples collected between 22 and 29 gestational wk using online solid phase extraction–high-performance liquid chromatography–isotope dilution tandem mass spectrometry at the Centers for Disease Control and Prevention (CDC) laboratory (Silva et al. 2007; Ye et al. 2005). Creatinine, a proxy of urine dilution, was also measured. The analysis of blinded urine specimens at CDC was determined not to constitute engagement in human subjects research. We used instrumental reading values for biomarker concentrations below the limits of detection. Instrumental reading values equal to 0 (i.e., indicative of no signal) were replaced by the lowest instrumental reading value divided by the square root of 2. We computed the total parabens and dichlorophenol concentrations by summing molar concentrations of the four parabens (∑parabens) and the two dichlorophenols (∑dichlorophenols), respectively. For phthalates, we computed the total di(2-ethylhexyl) phthalate (DEHP) metabolite concentration (∑DEHP) by summing molar concentrations of mono(2-ethyl-5-carboxypentyl) phthalate (MECPP), mono(2-ethyl-5-hydroxy-hexyl) phthalate (MEHHP), mono(2-ethylhexyl) phthalate (MEHP), and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP).

Statistical Analysis

Biomarker concentrations were standardized for creatinine and sampling conditions (e.g., hour of sampling, gestational age at sampling) prior to analysis using a two-step standardization approach developed by our team (Mortamais et al. 2012; Philippat et al. 2014). This method consists of a) studying the associations between each biomarker concentration and sampling conditions and creatinine concentrations through adjusted linear regression modeling; and b) using the measured biomarker concentrations and the effects of sampling conditions and of creatinine estimated during step 1 to predict standardized concentrations that would have been observed if all samples had been collected under the same conditions. These standardized concentrations were used in all analyses.

We used adjusted negative binomial regressions to study the associations between each standardized biomarker concentration and each SDQ subscore. Biomarker concentrations were log–transformed (base 2) prior to analysis so that the incidence rate ratios (IRRs) corresponded to the multiplicative change in the probability of the SDQ scores increasing by one unit for a doubling in biomarker concentration.

Adjustment factors were selected based on a priori knowledge and included factors possibly related to both exposures and child behavior, or to child behavior only. These factors included child age at assessment, recruitment center (Poitiers vs. Nancy), maternal age (continuous), parity (0, 1, 2 or more children), maternal body mass index (continuous), parental education (average length of the mother and father’s education), breastfeeding duration (never, ≤3 mo, >3 mo), household income [≤€1,500; €1,500 to €3,000, ≥€3,000 (euros) per mo], smoking during pregnancy (yes/no), and maternal psychological difficulties during pregnancy (yes/no). The maternal psychological difficulties score was constructed by combining scores of the Center for Epidemiologic Studies Depression Scale Revised (CESD), a questionnaire designed to assess depression and scores of the State-Trait Anxiety Inventory (STAI) that evaluates anxiety. The CESD and the STAI were completed by the mother during pregnancy.

Sensitivity Analyses

Several sensitivity analyses were conducted. To explore to which extent associations could be confounded by exposure to other phenols or phthalates, we ran analyses simultaneously adjusted for all phenols (BPA, benzophenone-3, triclosan, ∑dichlorophenols, ∑parabens), and phthalate metabolites [monoethyl phthalate (MEP), mono-n-butyl phthalate (MnBP), mono-isobutyl phthalate (MiBP), monobenzyl phthalate (MBzP), monocarboxyoctyl phthalate (MCOP), mono(3-carboxypropyl) phthalate (MCPP), monocarboxynonyl phthalate (MCNP), and ∑DEHP] concentrations.

Benzophenone-3 is used as a UV filter in sunscreens. Use of sunscreens is likely to vary across seasons. Because season of birth has been associated with increased risks of neurodevelopmental disorders such as autism spectrum disorders and schizophrenia (Torrey et al. 1997), we performed an additional analysis in which models that looked at the associations between benzophenone-3 and SDQ scores were additionally adjusted for season of urine collection.

We also dichotomized the SDQ scores at the 85th percentile (Melchior et al. 2015) and used adjusted logistic regression models to study the associations of these binary outcomes with biomarker concentrations.

Finally, as an attempt to take into account the effect of measurement error resulting from the use of a spot urine sample to assess exposure, we reported in the supplemental material the effect estimates corrected for exposure misclassification using the following formula (Rappaport et al. 1995):
where β̂jobs  is the estimated association from our main analysis between a biomarker j and an SDQ score, β̂jcorr our corrected estimate, and ICCj the intraclass correlation coefficient for the biomarker j considered. A simulation study has shown that this a posteriori disattenuation approach led to unbiased effect estimates in the presence of exposure measurement error of classical type (Perrier et al. 2016). We had only one urine sample per participant in our cohort and were not able to compute ICCs from our data. For this reason, we used ICCs from previous studies among pregnant women. ICCs were 0.1 for MCNP; 0.2 for BPA, ∑DEHP, and MCPP; 0.3 for MCOP; 0.4 for MnBP, MBzP, MEP, and MiBP; and 0.6 for ∑dichlorophenols, ∑parabens, benzophenone-3, and triclosan (Cantonwine et al. 2014; Philippat et al. 2013).

Given the relatively limited evidence regarding possible effects of phenols and phthalates on child neurodevelopment and the breadth of data available in our cohort, we chose to adopt an exploratory approach and reported in the results section all associations (p-values<0.05 and p-values between 0.05 and 0.1). In a sensitivity analysis, we accounted for multiple comparisons by applying a false discovery rate (FDR) correction (Benjamini and Hochberg 1995).

Analyses were performed using Stata/SE 14.1 (StataCorp LLC).

Results

Characteristics of the study population are presented in Table 1. Parents tended to be highly educated: 57% of the mothers and 47% of the fathers had 2 or more y of education after high school; 28% of the mothers reported having psychological difficulties during pregnancy, and 27% reported having never breastfed their child. Spearman correlation coefficients between the SDQ scores at 3 y and at 5 y were below 0.5 for the emotional, peer relationship problem, and prosocial scores, and ranged from 0.51 to 0.66 for the three other scores (Table S2). Phenols were detected in 73% (triclosan) to 100% (BPA; 2,5-dichlorophenols; methyl-paraben) of maternal urine samples, while all phthalate metabolites were detected in more than 97% of the samples (Table S3). We observed moderate to high Spearman correlation coefficients (rho) between the two dichlorophenols (rho=0.69), the four parabens (rho≥0.48), the four DEHP metabolites (rho≥0.78), and between MNBP and MCPP (rho=0.62). The remaining correlation coefficients were below 0.43 (Table S4).

Table 1. Characteristics of our study population (n=546a mother–son pairs of the French EDEN mother–child cohort).
Characteristics n (%)
Recruitment center
 Poitiers 323 (59)
 Nancy 223 (41)
Parity
 Nulliparous 255 (47)
≥1 290 (53)
 Missing 1 (0)
Maternal education
≤2 y after high school 226 (41)
 High school +2 y 127 (23)
 High school +3 y 186 (34)
 Missing 7 (1)
Paternal education
 ≤2 y after high school 257 (47)
 High school +2 y 118 (22)
 High school +3 y 137 (25)
 Missing 34 (6)
Household monthly income (euros)
 ≤1,500 70 (13)
 1,500–3,000 328 (60)
 ≥3,000 146 (27)
 Missing 2 (0)
Smoking during pregnancy
 No 419 (76)
 Yes 125 (23)
 Missing 2 (0)
Maternal psychological difficultiesb
 No 394 (72)
 Yes 152 (28)
Breastfeeding duration
 Never 145 (27)
 ≤3 mo 221 (40)
 3 mo 180 (33)
Mean (SD)
Maternal age at pregnancy 29.7 (4.7)
Body mass index (kg/m2) 23.2 (4.5)
Note: SD, standard deviation; SDQ, Strength and Difficulties Questionnaire.

aIncludes mother–son pairs that have biomarker assessments and SDQ scores at 3 or 5 years.

bMaternal depression and/or maternal anxiety during pregnancy.

Phenols and Child Behavior

Maternal BPA was associated with the internalizing behavior score [IRR for a doubling in concentration: 1.06; 95% confidence interval (CI): 1.00, 1.12] and the peer relations problem (IRR: 1.11; 95% CI: 1.03; 1.20) score at 3 y. BPA was also associated with the externalizing behavior score (IRR: 1.05; 95% CI: 1.00, 1.11) and the hyperactivity–inattention score (IRR: 1.08; 95% CI, 1.01; 1.14) at 5 y (Tables 2 and 3).

Table 2. Adjusted associations between phenols, phthalate metabolites and behavior at 3 y among boys of the EDEN mother–child cohort (n=518 to 520 mother–son pairs, depending of the subscale).
Emotional symptoms Conduct problems Peer relationship problems Hyperactivity–inattention problems Prosocial behavior Externalizing behavior Internalizing behavior
IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI
Phenols
∑Dichlorophenols 1.02 (0.98, 1.06) 1.01 (0.99, 1.04) 0.99 (0.95, 1.03) 1.00 (0.98, 1.03) 1.00 (0.98, 1.03) 1.01 (0.99, 1.03) 1.00 (0.97, 1.03)
 Bisphenol A 1.01 (0.94, 1.09) 1.03 (0.98, 1.08) 1.11 (1.03, 1.20)** 1.04 (0.99, 1.08) 1.05 (0.99, 1.10) 1.03 (0.99, 1.07) 1.06 (1.00, 1.12)*
 Benzophenone-3 1.00 (0.96, 1.03) 0.99 (0.97, 1.01) 1.01 (0.98, 1.05) 1.01 (0.99, 1.03) 0.99 (0.96, 1.01) 1.00 (0.98, 1.02) 1.00 (0.98, 1.03)
 Triclosan 1.02 (1.00, 1.04)* 1.01 (1.00, 1.03)* 1.00 (0.98, 1.03) 1.01 (1.00, 1.03) 1.01 (0.99, 1.03) 1.01 (1.00, 1.03)* 1.01 (0.99, 1.03)
 ∑Parabens 1.00 (0.97, 1.04) 0.99 (0.97, 1.02) 0.99 (0.95, 1.02) 0.98 (0.96, 1.00) 0.97 (0.94, 1.00)** 0.99 (0.97, 1.01) 1.00 (0.97, 1.03)
Phthalates
 MEP 1.03 (0.97, 1.08) 1.00 (0.97, 1.04) 0.97 (0.92, 1.03) 1.00 (0.97, 1.03) 0.99 (0.95, 1.03) 1.00 (0.97, 1.03) 1.00 (0.96, 1.04)
 MnBP 1.06 (1.00, 1.12)** 1.01 (0.97, 1.04) 1.06 (1.00, 1.12)** 0.99 (0.95, 1.02) 0.99 (0.95, 1.03) 1.00 (0.97, 1.03) 1.06 (1.01, 1.11)**
 MiBP 0.97 (0.90, 1.05) 1.00 (0.96, 1.05) 1.05 (0.98, 1.13) 0.95 (0.91, 1.00)** 0.99 (0.94, 1.05) 0.97 (0.94, 1.01) 1.01 (0.95, 1.07)
 MCPP 1.03 (0.96, 1.11) 0.98 (0.94, 1.03) 1.04 (0.97, 1.12) 0.96 (0.92, 1.00)* 0.95 (0.90, 1.00)* 0.97 (0.93, 1.01) 1.04 (0.98, 1.10)
 MBzP 1.02 (0.96, 1.09) 1.00 (0.97, 1.04) 1.07 (1.01, 1.13)** 0.98 (0.94, 1.01) 0.97 (0.93, 1.02) 0.99 (0.96, 1.02) 1.04 (1.00, 1.09)*
 MCOP 1.02 (0.95, 1.10) 1.01 (0.97, 1.06) 0.99 (0.92, 1.06) 0.98 (0.94, 1.03) 1.00 (0.95, 1.05) 1.00 (0.96, 1.04) 1.01 (0.95, 1.06)
 MCNP 1.01 (0.95, 1.07) 0.99 (0.95, 1.03) 1.00 (0.94, 1.06) 0.97 (0.93, 1.01) 0.98 (0.94, 1.03) 0.98 (0.95, 1.01) 1.00 (0.95, 1.05)
 ∑DEHP 1.04 (0.97, 1.11) 0.99 (0.95, 1.04) 1.05 (0.98, 1.12) 0.97 (0.93, 1.01) 0.99 (0.94, 1.04) 0.98 (0.94, 1.02) 1.04 (0.99, 1.10)
Note: CI, confidence interval; ∑DEHP, molecular sum of di(2-ethylhexyl) phthalate metabolites; ∑dichlorophenols, molecular sum of 2,4 and 2,5-dichlorophenols; IRR, incidence rate ratio; MBzP, monobenzyl phthalate; MCNP, monocarboxynonyl phthalate; MCOP, monocarboxyoctyl phthalate; MCPP, mono(3-carboxypropyl) phthalate; MEP, monoethyl phthalate; MiBP. mono-isobutyl phthalate, MnBP, mono-n-butyl phthalate; ∑parabens, molecular sum of methyl, ethyl, propyl, and butyl parabens. Associations adjusted for recruitment center, maternal age, parity, parental education, breastfeeding duration, household income, smoking during pregnancy, maternal psychological difficulties during pregnancy, and child age at assessment. IRR are reported for a doubling in biomarkers concentrations.

*p≤ 0.10;

**p≤ 0.05.

Table 3. Adjusted associations between phenols, phthalate metabolites and behavior at 5 y among boys of the EDEN mother–child cohort (n=457 or 458 mother–son pairs, depending of the subscale).
Emotional symptoms Conduct problems Peer relationship problems Hyperactivity–inattention problems Prosocial behavior Externalizing behavior Internalizing behavior
IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI
Phenols
 ∑Dichlorophenols 1.00 (0.97, 1.04) 1.03 (1.00, 1.07)* 0.99 (0.94, 1.04) 1.01 (0.98, 1.04) 1.00 (0.96, 1.04) 1.02 (0.99, 1.05) 1.00 (0.97, 1.03)
 Bisphenol A 1.03 (0.95, 1.11) 1.02 (0.95, 1.10) 1.02 (0.93, 1.13) 1.08 (1.01, 1.14)** 1.05 (0.96, 1.14) 1.05 (1.00, 1.11)** 1.03 (0.96, 1.10)
 Benzophenone-3 0.96 (0.93, 1.00)** 1.00 (0.97, 1.04) 1.01 (0.97, 1.06) 1.02 (0.99, 1.04) 1.02 (0.98, 1.06) 1.01 (0.99, 1.04) 0.98 (0.95, 1.01)
 Triclosan 1.01 (0.99, 1.03) 1.00 (0.98, 1.02) 1.01 (0.98, 1.04) 1.00 (0.98, 1.01) 0.99 (0.97, 1.02) 1.00 (0.98, 1.02) 1.01 (0.99, 1.03)
 ∑Parabens 0.99 (0.95, 1.02) 0.99 (0.96, 1.03) 0.99 (0.94, 1.04) 0.98 (0.95, 1.00) 0.99 (0.95, 1.03) 0.98 (0.96, 1.01) 0.99 (0.96, 1.02)
Phthalates
 MEP 1.01 (0.95, 1.06) 1.03 (0.98, 1.08) 0.99 (0.92, 1.05) 0.99 (0.95, 1.03) 0.96 (0.91, 1.02) 1.01 (0.97, 1.05) 1.00 (0.95, 1.04)
 MnBP 1.00 (0.94, 1.05) 1.02 (0.97, 1.07) 1.05 (0.99, 1.13) 1.01 (0.97, 1.05) 1.01 (0.95, 1.07) 1.01 (0.97, 1.06) 1.02 (0.97, 1.07)
 MiBP 0.97 (0.90, 1.05) 1.01 (0.94, 1.08) 1.00 (0.91, 1.09) 0.97 (0.91, 1.02) 1.04 (0.96, 1.13) 0.98 (0.93, 1.04) 0.98 (0.92, 1.05)
 MCPP 0.98 (0.91, 1.05) 1.04 (0.97, 1.11) 1.06 (0.97, 1.16) 1.01 (0.95, 1.07) 0.99 (0.91, 1.08) 1.02 (0.97, 1.08) 1.01 (0.95, 1.08)
 MBzP 1.03 (0.97, 1.09) 0.98 (0.93, 1.04) 1.01 (0.94, 1.09) 1.00 (0.95, 1.04) 0.98 (0.92, 1.04) 0.99 (0.95, 1.03) 1.02 (0.97, 1.08)
 MCOP 1.04 (0.97, 1.12) 1.00 (0.94, 1.06) 0.99 (0.91, 1.08) 1.01 (0.96, 1.07) 0.94 (0.87, 1.02) 1.01 (0.96, 1.06) 1.02 (0.96, 1.09)
 MCNP 1.04 (0.98, 1.10) 0.99 (0.93, 1.04) 1.03 (0.96, 1.11) 0.99 (0.94, 1.04) 0.98 (0.92, 1.05) 0.99 (0.94, 1.03) 1.04 (0.98, 1.09)
 ∑DEHP 1.01 (0.94, 1.08) 0.99 (0.93, 1.06) 1.03 (0.94, 1.12) 1.01 (0.96, 1.07) 0.97 (0.90, 1.05) 1.00 (0.95, 1.06) 1.02 (0.95, 1.08)
Note: CI, confidence interval; ∑DEHP, molecular sum of di(2-ethylhexyl) phthalate metabolites; ∑dichlorophenols, molecular sum of 2,4 and 2,5-dichlorophenols; IRR, incidence rate ratio; MBzP, monobenzyl phthalate; MCNP, monocarboxynonyl phthalate; MCOP, monocarboxyoctyl phthalate; MCPP, mono(3-carboxypropyl) phthalate; MEP, monoethyl phthalate; MiBP, mono-isobutyl phthalate; MnBP, mono-n-butyl phthalate; ∑parabens, molecular sum of methyl, ethyl, propyl, and butyl parabens. Associations adjusted for recruitment center, maternal age, parity, parental education, breastfeeding duration, household income, smoking status, maternal psychological difficulties during pregnancy, and child age at assessment. IRRs are reported for a doubling in biomarkers concentrations.

*p≤0.10;

**p≤0.05.

Triclosan tended to be positively associated with several SDQ scores at 3 y with IRRs closer to 1 than those observed for BPA: IRR of 1.01 (95% CI: 1.00, 1.03), 1.02 (95% CI: 1.00, 1.04), and 1.01 (95% CI: 1.00; 1.03) for the externalizing, emotional symptoms, and conduct problem scores at 3 y, respectively (Table 2).

No other phenol was associated with the internalizing and externalizing scores at 3 or 5 y (p-values>0.14, Tables 2, 3). However, when we looked at each SDQ subscore separately, scores of the conduct problems subscale at 5 y tended to increase in association with ∑dichlorophenols (IRR: 1.03; 95% CI: 1.00, 1.07). Benzophenone-3 was associated with a decreased emotional symptom score at 5 y (IRR, 0.96; 95% CI: 0.93, 1.00), while ∑parabens were associated with increased prosocial behavior score at 3 y (IRR, 0.97; 95% CI: 0.94, 1.00) (Tables 2, 3).

Phthalates and Child Behavior

Several phthalate metabolites were associated with SDQ scores at 3 y (Table 2). MnBP was associated with internalizing (IRR: 1.06; 95% CI: 1.01, 1.11) as well as with the relationship problems (IRR: 1.06; 95% CI: 1.00, 1.12) and emotional problems (IRR: 1.06; 95% CI: 1.00; 1.12) scores at 3 y. MBzP was associated with increased internalizing (IRR: 1.04; 95% CI: 1.00, 1.09) and relationship problems scores (IRR: 1.07; 95% CI: 1.01, 1.13). Two phthalates were associated with lower SDQ scores: MiBP and MCPP concentrations were indeed associated with decreased hyperactivity–inattention scores [IRRs were 0.95 (95% CI: 0.91, 1.00) and 0.96 (95% CI: 0.92, 1.00) for MiBP and MCPP, respectively]. MCPP was also associated with decreased prosocial behavior scores (IRR: 0.95; 95% CI: 0.9; 1.0). None of the associations we observed between phthalate metabolites and SDQ scores at 3 y persisted at 5 y (Table 3).

Sensitivity Analyses

Studying SDQ scores as dichotomized variables overall led to similar results as studying them as count outcomes for BPA, MBzP, MnBP, and MCPP (Tables 4, 5). For triclosan, the positive association seen at 3 y with the emotional symptoms score persisted at 5 y when SDQ scores were dichotomized (Tables 4, 5). The inverse associations we observed between benzophenone-3 and emotional symptoms at 5 y, MiBP and hyperactivity–inattention at 3 y, and ∑parabens and prosocial behavior at 3 y were strongly attenuated when SDQ scores were dichotomized and did not remain significant. Odds ratios (ORs) were 0.93 (95% CI: 0.82, 1.05), 0.91 (95% CI: 0.72, 1.16), and 0.92 (95% CI: 0.80, 1.06) for benzophenone-3, MiBP, and ∑parabens, respectively (Tables 4, 5). Similarly, the association between ∑dichlorophenols and conduct problems at 5 y did not remain when this score was dichotomized (OR: 0.90; 95% CI: 0.80, 1.10, Table 5). Finally, a few associations emerged: ∑DEHP was associated with increased risk of emotional symptoms (OR: 1.27, 95% CI: 1.01, 1.60) and internalizing behavior (OR: 1.41; 95% CI: 1.13, 1.76) at 3 y, while MCNP was associated with decreased risk of hyperactivity–inattention at 3 y (OR: 0.72; 95% CI: 0.55, 0.93).

Table 4. Adjusted associations between phenols, phthalate metabolites and dichotomized Strength and Difficulties Questionnaire (SDQ) scores at 3 y (n=518 to 520 mother–son pairs, depending of the subscale).
Emotional symptoms Conduct problems Peer relationship problems Hyperactivity–inattention problems Prosocial behavior Externalizing behavior Internalizing behavior
OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Phenols
 ∑Dichlorophenols 1.06 (0.94, 1.20) 0.97 (0.83, 1.13) 0.99 (0.85, 1.14) 1.04 (0.92, 1.18) 1.02 (0.90, 1.17) 1.05 (0.93, 1.18) 1.04 (0.93, 1.18)
 Bisphenol A 0.92 (0.70, 1.19) 1.08 (0.81, 1.43) 1.40 (1.07, 1.84)** 1.07 (0.83, 1.38) 1.07 (0.82, 1.39) 1.01 (0.79, 1.29) 1.22 (0.96, 1.55)*
 Benzophenone-3 1.00 (0.89, 1.13) 1.02 (0.89, 1.16) 1.03 (0.91, 1.17) 1.01 (0.89, 1.13) 0.98 (0.86, 1.11) 0.98 (0.87, 1.10) 1.02 (0.91, 1.14)
 Triclosan 1.07 (0.99, 1.17)* 1.10 (1.00, 1.21)** 1.06 (0.97, 1.16) 1.02 (0.94, 1.10) 1.07 (0.99, 1.17)* 1.05 (0.97, 1.13) 1.05 (0.97, 1.14)
 Parabens 1.01 (0.89, 1.15) 1.00 (0.87, 1.16) 0.97 (0.84, 1.12) 1.01 (0.89, 1.15) 0.92 (0.80, 1.06) 0.94 (0.83, 1.07) 1.03 (0.92, 1.17)
Phthalates
 MEP 1.14 (0.95, 1.36) 0.95 (0.77, 1.17) 1.04 (0.86, 1.26) 1.07 (0.89, 1.29) 0.98 (0.81, 1.19) 1.00 (0.84, 1.20) 0.98 (0.83, 1.17)
 MnBP 1.10 (0.91, 1.31) 0.99 (0.80, 1.22) 1.16 (0.95, 1.41) 0.95 (0.79, 1.16) 0.92 (0.75, 1.13) 0.93 (0.78, 1.12) 1.22 (1.03, 1.44)**
 MiBP 0.87 (0.68, 1.12) 1.13 (0.86, 1.49) 1.16 (0.89, 1.53) 0.91 (0.72, 1.16) 1.10 (0.85, 1.43) 0.93 (0.74, 1.17) 1.09 (0.87, 1.38)
 MCPP 1.02 (0.80, 1.31) 0.79 (0.59, 1.07) 1.13 (0.86, 1.47) 0.79 (0.61, 1.04)* 0.80 (0.61, 1.06) 0.82 (0.64, 1.05) 1.20 (0.95, 1.51)
 MBzP 0.99 (0.81, 1.22) 0.99 (0.78, 1.25) 1.25 (1.00, 1.56)** 0.94 (0.76, 1.15) 0.94 (0.76, 1.17) 0.94 (0.78, 1.15) 1.19 (0.98, 1.44)
 MCOP 1.02 (0.81, 1.30) 1.14 (0.89, 1.47) 1.04 (0.79, 1.35) 0.85 (0.66, 1.10) 0.98 (0.77, 1.25) 1.07 (0.86, 1.34) 1.08 (0.87, 1.35)
 MCNP 0.90 (0.72, 1.14) 1.00 (0.78, 1.27) 1.09 (0.87, 1.36) 0.72 (0.55, 0.93)** 0.84 (0.65, 1.08) 0.85 (0.68, 1.07) 1.00 (0.81, 1.23)
 ∑DEHP 1.27 (1.01, 1.60)** 1.10 (0.85, 1.43) 1.09 (0.84, 1.41) 1.03 (0.81, 1.30) 1.18 (0.94, 1.50) 1.07 (0.85, 1.34) 1.41 (1.13, 1.76)*
Note: CI, confidence interval; ∑DEHP, molecular sum of di(2-ethylhexyl) phthalate metabolites; ∑dichlorophenols, molecular sum of 2,4 and 2,5-dichlorophenols; MBzP, monobenzyl phthalate; MCNP, monocarboxynonyl phthalate; MCOP, monocarboxyoctyl phthalate; MCPP, mono(3-carboxypropyl) phthalate; MEP, monoethyl phthalate; MiBP, mono-isobutyl phthalate; MnBP, mono-n-butyl phthalate; OR, odds ratio; ∑parabens, molecular sum of methyl, ethyl, propyl, and butyl parabens. Associations adjusted for recruitment center, maternal age, parity, parental education, breastfeeding duration, household income, smoking status, maternal psychological difficulties during pregnancy, and child age at assessment. IRRs are reported for a doubling in biomarkers concentrations.

*p≤ 0.10;

**p≤ 0.05.

Table 5. Adjusted associations between phenols, phthalate metabolites and dichotomized Strength and Difficulties Questionnaire (SDQ) scores at 5 y (n=457 or 458 mother–son pairs, depending of the subscale).
Emotional symptoms Conduct problems Peer relationship problems Hyperactivity–inattention problems Prosocial behavior Externalizing behavior Internalizing behavior
OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Phenols
 ∑Dichlorophenols 0.99 (0.87, 1.12) 0.94 (0.80, 1.10) 0.94 (0.79, 1.11) 1.10 (0.97, 1.24) 1.01 (0.87, 1.18) 1.05 (0.93, 1.18) 0.94 (0.83, 1.07)
 Bisphenol A 1.04 (0.80, 1.36) 1.04 (0.76, 1.41) 1.10 (0.80, 1.50) 1.40 (1.08, 1.82)** 1.09 (0.80, 1.47) 1.20 (0.93, 1.54) 1.00 (0.96, 1.10)
 Benzophenone-3 0.93 (0.82, 1.05) 1.00 (0.87, 1.16) 0.99 (0.86, 1.14) 1.05 (0.93, 1.19) 1.05 (0.91, 1.20) 1.06 (0.95, 1.18) 0.97 (0.95, 1.01)
 Triclosan 1.08 (1.00, 1.18)* 1.06 (0.96, 1.17) 1.10 (0.99, 1.22)* 1.01 (0.93, 1.09) 1.00 (0.91, 1.09) 1.00 (0.93, 1.08) 1.07 (0.99, 1.03)
 ∑Parabens 0.95 (0.84, 1.08) 0.96 (0.82, 1.11) 1.01 (0.87, 1.18) 0.98 (0.86, 1.12) 0.97 (0.84, 1.13) 1.01 (0.89, 1.14) 0.96 (0.96, 1.02)
Phthalates
 MEP 1.16 (0.97, 1.39)* 1.13 (0.92, 1.38) 1.02 (0.82, 1.27) 1.05 (0.87, 1.27) 0.79 (0.63, 1.00)** 1.08 (0.91, 1.29) 0.93 (0.95, 1.04)
 MnBP 1.02 (0.85, 1.23) 1.06 (0.85, 1.31) 1.24 (1.01, 1.52)** 1.03 (0.85, 1.25) 0.87 (0.68, 1.10) 1.04 (0.86, 1.25) 1.14 (0.97, 1.07)
 MiBP 1.01 (0.79, 1.29) 1.02 (0.77, 1.35) 1.05 (0.79, 1.41) 0.95 (0.74, 1.23) 1.02 (0.75, 1.38) 1.00 (0.79, 1.27) 0.97 (0.92, 1.05)
 MCPP 1.01 (0.79, 1.29) 1.07 (0.81, 1.43) 1.20 (0.91, 1.58) 0.98 (0.75, 1.27) 0.86 (0.63, 1.17) 1.04 (0.81, 1.33) 1.16 (0.95, 1.08)
 MBzP 1.06 (0.87, 1.29) 1.02 (0.81, 1.29) 0.96 (0.75, 1.22) 1.02 (0.83, 1.25) 0.79 (0.61, 1.03)* 0.91 (0.75, 1.12) 1.04 (0.97, 1.08)
 MCOP 1.15 (0.91, 1.45) 0.93 (0.70, 1.23) 1.03 (0.78, 1.36) 1.23 (0.97, 1.56) 0.88 (0.66, 1.17) 1.03 (0.82, 1.30) 1.09 (0.96, 1.09)
 MCNP 1.01 (0.81, 1.25) 0.90 (0.68, 1.18) 0.99 (0.76, 1.28) 1.02 (0.82, 1.28) 1.06 (0.84, 1.35) 1.00 (0.80, 1.24) 1.04 (0.98, 1.09)
 ∑DEHP 1.08 (0.85, 1.36) 0.99 (0.74, 1.31) 1.18 (0.90, 1.55) 1.07 (0.84, 1.36) 0.79 (0.59, 1.07) 1.05 (0.83, 1.32) 1.15 (0.95, 1.08)
Note: CI, confidence interval; ∑DEHP, molecular sum of di(2-ethylhexyl) phthalate metabolites; ∑dichlorophenols, molecular sum of 2,4 and 2,5-dichlorophenols; MBzP, monobenzyl phthalate; MCNP, monocarboxynonyl phthalate; MCOP, monocarboxyoctyl phthalate; MCPP, mono(3-carboxypropyl) phthalate; MEP, monoethyl phthalate; MiBP, mono-isobutyl phthalate; MnBP, mono-n-butyl phthalate; OR, odds ratio; ∑parabens, molecular sum of methyl, ethyl, propyl, and butyl parabens. Associations adjusted for recruitment center, maternal age, parity, parental education, breastfeeding duration, household income, smoking status, maternal psychological difficulties during pregnancy, and child age at assessment. IRRs are reported for a doubling in biomarkers concentrations.

*p≤ 0.10;

**p≤ 0.05.

Adjustment for season did not change the association observed between benzophenone-3 and the emotional subscale. Running models simultaneously adjusted for all of the biomarkers measured in our study did not strongly affect our findings, and the effect estimates were close to those of the main analyses that consisted of studying each biomarker separately (Tables S5, S6).

As expected, IRRs and ORs corrected for exposure measurement error using the formula based on ICCs were larger (absolute values) than the uncorrected ones (Tables S7 to S10). As an example, after correction, OR was as high as 5.4 (95% CI: 1.45, 20.0) for the association between BPA, one of the chemicals with the highest variability, and the hyperactivity and inattention scores at 5 y (Table S10). OR without correction was 1.40 (95% CI: 1.0, 1.82, Table 5).

When we applied a correction for multiple comparisons using an FDR method, none of the associations reported in the results section remained significant, the lowest corrected p-value being 0.42 for the association between BPA and the peer relations problem score at 3 y.

Discussion

In our study of 546 women who delivered a boy, three phenols (BPA, triclosan, and ∑dichlorophenols) and two phthalates (MnBP, MBzP) biomarkers were positively associated with scores in one or more subscales of the SDQ, while inverse association was observed for other biomarkers (benzophenone-3, MiBP, MCPP). Most of the observed IRRs were close to 1. It should be kept in mind that a detailed assessment of the public health impact of these exposures would require a thorough health impact assessment, taking into account the distribution of exposure levels and dose–response functions, if possible, corrected for exposure misclassification. Our study is among the first to simultaneously consider a rather large number of biomarkers from the phthalates and phenols families in relation to several behavioral scales. Because we tested many associations, some of our results might be chance finding, as suggested by the fact that none of the observed p-values remained significant after FDR correction. For this reason, in the following discussion, we focused on the associations for which some consistency was observed with the previous human literature focusing on prenatal exposure, or with animal literature when no human study was available.

In our study population, BPA was associated with increased scores on the internalizing behavior and peer relationship problem subscales at 3 y and on the externalizing behavior and hyperactivity–inattention subscales at 5 y. Associations between prenatal BPA concentrations and internalizing and externalizing behavior scores in boys have been reported in most of the previous studies evaluating such behaviors (Evans et al. 2014; Harley et al. 2013; FP Perera et al. 2012; Roen et al. 2015) (Table S11). Regarding specific subscales, an increased score on the hyperactivity–inattention subscale in association with prenatal exposure to BPA has been previously reported among boys at 4, but not at 7 y (Casas et al. 2015). Other studies have reported opposite (Braun et al. 2011) or null (Evans et al. 2014; Harley et al. 2013) associations with this subscale. Interpretation of the human literature on BPA should be done keeping in mind that BPA is one of the biomarkers with the highest within-subject variability [with intraclass coefficients of correlation of 0.1–0.2 being typically reported during pregnancy (Braun et al. 2012; Philippat et al. 2013)], limiting the power of studies relying on one spot urine sample to assess exposure, as was the case of most previous studies. Despite this limitation, most of the published studies (including ours) reported an association with one or several behavioral scores among boys (Table S11), suggesting that BPA may affect some aspects of behavior among boys. This is supported by experimental studies on laboratory animals that have also reported associations between perinatal exposure to BPA and behavior (Anderson et al. 2013; Ishido et al. 2011; Komada et al. 2014; Nakagami et al. 2009) and by in vitro studies that suggested that BPA can interact with some pathways that are crucial for brain development. BPA can indeed bind with estrogen and androgen receptors as well as interact with thyroid hormone pathways [reviewed by Mustieles et al. (2015)].

In our population, triclosan, a biocide used in some toothpastes, antibacterial soaps, and detergents, tended to be positively associated with the externalizing behavior score as well as with the emotional symptom and conduct problems scores at 3 y. When SDQ scores were dichotomized, associations with the emotional symptom score was also observed at 5 y. To our knowledge, this is the first study investigating triclosan potential effect on child neurodevelopment. Both in vitro studies and studies in rodents suggested that triclosan can affect pathways involved in the development of the nervous system, such as the thyroid hormone pathway (Paul et al. 2010, 2012). In humans, a cross-sectional study reported increased total triiodothyronine concentrations in association with increased triclosan concentrations among adolescents (Koeppe et al. 2013). Other mechanisms whereby triclosan might affect child neurodevelopment include disruption of the sex hormone homeostasis because triclosan can bind, albeit with low affinity, to both the androgen and estrogen receptors [reviewed by Witorsh (Witorsch 2014)].

Maternal urinary concentrations of ∑dichlorophenols tended to be associated with increased risk of conduct problems at 5 y. However, this association did not remain when SDQ scores were dichotomized. ∑Dichlorophenols included 2,5-dichlorophenol, a metabolite of 1,4-dichlorobenzene used in mothballs; indoor deodorizers; toilet bowl disinfectants; and 2,4-dichlorophenol, a metabolite of 2,4-dichlorophenoxyacetic acid used as intermediate in the production of some herbicides. Our study is the first to focus on these compounds in humans for behavior-specific factors. A study in which rats were exposed during pre- and postnatal life to 2,4-dichlorophenoxyacetic acid reported deleterious effect on behaviors (Bortolozzi et al. 1999) that included hyperactivity, stereotypic behavior (excessive grooming), and serotonin syndrome behaviors (forepaw tapping, sprawling of limbs, and mobility) (Bortolozzi et al. 1999). Translation to epidemiological research is difficult, given the fact that behavioral outcomes and routes of exposure differ between rats and humans.

We observed increased scores on the internalizing, relationship problems, and emotional symptoms subscales at 3 y in association with urinary concentrations of MnBP. These associations did not remain at 5 y. In a previous study among mother–child pairs from New York City, 3-y-old boys experienced more internalizing and withdrawn behaviors as well as more emotionally reactive problems in association with maternal urinary concentrations of this phthalate metabolite (Whyatt et al. 2012). Other studies among older boys (4.5 to 10 y) did not report such associations (Engel et al. 2010; Kobrosly et al. 2014; Lien et al. 2015). Experimental studies among rodents also suggested behavioral effects for this phthalate (Farzanehfar et al. 2016; Hoshi and Ohtsuka 2009; Yan et al. 2016).

In our study population, MBzP was associated with increased scores on the internalizing behaviors and relationship problems subscales at 3, but not at 5 y. Associations between MBzP and behavioral scores have been reported previously among boys, but with other subscales, including oppositional behavior and conduct problems (Kobrosly et al. 2014).

When SDQ scores were dichotomized, ∑DEHP was associated with increased risk for emotional symptoms and internalizing behavior. A few studies reported associations between DEHP biomarkers (as a sum of several metabolites or for single DEHP metabolites) and behavior in boys. These associations were seen with other subscales than those reported in our study, including increased social, delinquent, somatic, and externalizing behavior subscores (Kobrosly et al. 2014; Lien et al. 2015). However, these two previous studies examined behavior at older age (6 to 10 y old) and had relatively low sample size (n<80 for boys). Among studies of bigger sample size, one reported no association between prenatal exposure to DEHP and behavior at 3 y (Whyatt et al. 2012), while the other reported better social competence and lower hyperactivity–inattention scores at 7 y in association with DEHP biomarkers (Gascon et al. 2015). Phthalates can disrupt the thyroid and sex hormone homeostasis and calcium signaling, and can alter brain’s lipid profile [reviewed by Miodovnik et al. (2014)]. They are also weak agonists of the aryl hydrocarbon (Krüger et al. 2008) and peroxisome proliferator-activated receptors (Miodovnik et al. 2014) involved in numerous developmental pathways.

Four biomarkers (benzophenone-3, ∑parabens, MiBP, MCPP) were associated with lower SDQ scores, suggesting improved behaviors. When SDQ scores were dichotomized, only the associations between MCPP and the prosocial behavior score at 5 y and the hyperactivity–inattention scores at 3 y remained. Previous studies that assessed this phthalate in maternal urine did not report any association with hyperactivity–inattention among boys (Engel et al. 2010; Gascon et al. 2015; Lien et al. 2015). Regarding parabens and benzophenone-3, this is, to our knowledge, the first epidemiological study that looked at their human behavioral effects, which makes it difficult to discuss the plausibility of our associations with lower SDQ scores. However, previous studies in rats exposed to parabens pre- and postnatally (via lactation) reported increased anxiety-like behavior (Kawaguchi et al. 2009) and autism-like symptoms (Ali and Elgoly 2013), as well as less social interactions and learning memory (Ali and Elgoly 2013) in exposed animals compared to controls, which is not in line with our findings. This difference and the fact that the association we observed with ∑parabens did not remain when SDQ scores were dichotomized should lead to cautious interpretation of the association we observed that suggested that higher concentration of ∑parabens was associated with more prosocial behavior.

Although most studies (including ours) that examined the associations between phthalate biomarkers and child behavior in humans reported deleterious associations, the biomarkers implicated and the affected behaviors often varied across studies. This may be explained by the large heterogeneity in age, ranging from 1 to 10 y; the different instruments used to evaluate behavior; the large number of comparisons performed in each study, which usually did not correct for multiple testing; and issues related to measurement error in the exposure assessment of biomarkers with short half-lives. Exposure was assessed using only one, rarely two, urine samples per participant collected at different time points during pregnancy, which is insufficient to characterize exposure for biomarkers with high intraindividual variabilities, as reported for some phthalates and phenols (Adibi et al. 2008; Philippat et al. 2013), and is likely to lead to attenuation bias (Perrier et al. 2016). As in most previous studies, we assessed exposure during pregnancy and were missing exposure occurring during the first years of life, also a potentially crucial time point for brain development. Socioeconomic status is likely to be a strong confounder of the associations between exposure to environmental chemicals and child neurodevelopment (Bellinger 2004). We adjusted our analyses for many potential confounders, but we cannot exclude that residual confounding remained. Among 728 Spanish women for whom exposure to 81 chemicals was assessed, BPA was not highly correlated with other environmental biomarkers, while phthalate metabolites were correlated among each other and were negatively correlated with some metals, such as cadmium and copper [rho ranged between −0.3 and −0.5, (Robinson et al. 2015)]. Although correlations between exposures possibly differ across countries, this finding suggests that the associations we observed with BPA and phthalate biomarkers are unlikely to be confounded by exposure to other chemicals. In line with this finding, associations varied little after adjusting our analyses for the other phenols and phthalates biomarkers measured in the study. We relied on the SDQ, a validated and widely used questionnaire in Europe, to assess child behavior at two different time points.

Strengths of our study include the sample size, larger than those of previous studies that included 122 to 460 boys and girls and among which analyses stratified for child sex have been often performed. The fact that we focused on boys limits generalizability of our findings, but is not a source of bias, especially in the context of endocrine disruptors for which sex-specific effects have been reported previously (Casas et al. 2015; Perera et al. 2012; Roen et al. 2015). In an attempt of correcting our findings for exposure measurement error, we present in the supplemental material the effect estimates corrected using an a posteriori disattenuation method (Perrier et al. 2016). As expected, the corrected effect estimates were bigger (absolute values) than the corresponding uncorrected IRRs and ORs. We believe that these corrected effect estimates are relevant for the purpose of future meta-analyses that will combine studies relying on a different number of urine samples per participant to assess exposure.

Conclusion

Several phenol and phthalate biomarkers were associated with increased scores on the SDQ subscales at 3 and/or 5 y. The associations observed for BPA and MnBP were consistent with those reported previously among boys in the same age range (Evans et al. 2014; Harley et al. 2013; Perera et al. 2012; Whyatt et al. 2012) and are further supported by the animal literature. Associations observed with MBzP, triclosan, and ∑dichlorophenols need cautious interpretation, since they have not been observed in previous studies (MBzP), or this study is the first to explore their potential effects on child behavior (triclosan, ∑dichlorophenols). Harmonizing protocols across studies (including tools used to assess behavior and age at assessment) would enhance results comparability across studies. In addition, further studies that aim at looking at the potential effects of phenols and phthalates on child neurodevelopment would benefit from incorporating new approaches to improve exposure assessment, one of the main limitations of the current epidemiological literature for these biomarkers.

Acknowledgments

We acknowledge L. Giorgis-Allemand and A. Forhan for data management, the midwife research assistants (L. Douhaud, S. Bedel, B. Lortholary, S. Gabriel, M. Rogeon, and M. Malinbaum) for data collection, the psychologists (M.-C. Cona and M. Paquinet), the data entry operators (P. Lavoine, J. Sahuquillo and G. Debotte), and X. Ye, A. Bishop, X. Zhou, M. Silva, E. Samandar, J. Preau, and L. Jia for technical assistance in measuring the urinary biomarkers. We are grateful to the EDEN participating families.

This research was supported by grants from the French Agency for Food, Environmental and Occupational Health and Safety [ANSES (grant EST-2010/2/126)] and Fondation de France (grant 2015-00059545). R.S. is supported by a European Research Council consolidator grant [grant N°311765–E-DOHaD (Environmentally-induced Developmental Origins of Health and Disease)]. The Eden cohort is supported by grants from Fondation pour la Recherche Médicale (FRM), Inserm, de Recherche en Santé Publique (IReSP), Nestlé, French Ministry of Health, Agence National de la Recherche (ANR), Univ. Paris-Sud, Santé Publique France, ANSES, and Mutuelle Générale de l’Education Nationale (MGEN).

References

Adibi JJ, Whyatt RM, Williams PL, Calafat AM, Camann D, Herrick R, et al. 2008. Characterization of phthalate exposure among pregnant women assessed by repeat air and urine samples. Environ Health Perspect 116(4):467–473, PMID: 18414628, 10.1289/ehp.10749.

Ali EH, Elgoly AH. 2013. Combined prenatal and postnatal butyl paraben exposure produces autism-like symptoms in offspring: comparison with valproic acid autistic model. Pharmacol Biochem Behav 111:102–110, PMID: 24036472, 10.1016/j.pbb.2013.08.016.

Anderson OS, Peterson KE, Sanchez BN, Zhang Z, Mancuso P, Dolinoy DC. 2013. Perinatal bisphenol A exposure promotes hyperactivity, lean body composition, and hormonal responses across the murine life course. FASEB J 27(4):1784–1792, PMID: 23345456, 10.1096/fj.12-223545.

Bellinger DC. 2004. Assessing environmental neurotoxicant exposures and child neurobehavior: confounded by confounding? Epidemiology 15(4):383–384, PMID: 15232396, 10.1097/01.ede.0000129525.15064.a4.

Benjamini Y, Hochberg Y. 1995. Controlling the false discovery rate – a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol 57(1):289–300.

Bortolozzi AA, Duffard RO, Evangelista de Duffard AM. 1999. Behavioral alterations induced in rats by a pre- and postnatal exposure to 2,4-dichlorophenoxyacetic acid. Neurotoxicol Teratol 21(4):451–465, PMID: 10440489.

Botton J, Philippat C, Calafat AM, Carles S, Charles MA, Slama R, et al. 2016. Phthalate pregnancy exposure and male offspring growth from the intra-uterine period to five years of age. Environ Res 151:601–609, PMID: 27596487, 10.1016/j.envres.2016.08.033.

Braun JM, Kalkbrenner AE, Calafat AM, Yolton K, Ye X, Dietrich KN, et al. 2011. Impact of early-life bisphenol A exposure on behavior and executive function in children. Pediatrics 128(5):873–882, PMID: 22025598, 10.1542/peds.2011-1335.

Braun JM, Smith KW, Williams PL, Calafat AM, Berry K, Ehrlich S, et al. 2012. Variability of urinary phthalate metabolite and bisphenol A concentrations before and during pregnancy. Environ Health Perspect 120(5):739–745, PMID: 22262702, 10.1289/ehp.1104139.

Cantonwine DE, Cordero JF, Rivera-González LO, Anzalota Del Toro LV, Ferguson KK, Mukherjee B, et al. 2014. Urinary phthalate metabolite concentrations among pregnant women in Northern Puerto Rico: distribution, temporal variability, and predictors. Environ Int 62:1–11, PMID: 24161445, 10.1016/j.envint.2013.09.014.

Casas M, Forns J, Martínez D, Avella-García C, Valvi D, Ballesteros-Gómez A, et al. 2015. Exposure to bisphenol A during pregnancy and child neuropsychological development in the INMA-Sabadell cohort. Environ Res 142:671–679, PMID: 26343751, 10.1016/j.envres.2015.07.024.

Chevrier C, Petit C, Philippat C, Mortamais M, Slama R, Rouget F, et al. 2012. Maternal urinary phthalates and phenols and male genital anomalies. Epidemiology 23(2):353–356, PMID: 22317818, 10.1097/EDE.0b013e318246073e.

Engel SM, Miodovnik A, Canfield RL, Zhu C, Silva MJ, Calafat AM, et al. 2010. Prenatal phthalate exposure is associated with childhood behavior and executive functioning. Environ Health Perspect 118(4):565–571, PMID: 20106747, 10.1289/ehp.0901470.

Evans SF, Kobrosly RW, Barrett ES, Thurston SW, Calafat AM, Weiss B, et al. 2014. Prenatal bisphenol A exposure and maternally reported behavior in boys and girls. Neurotoxicology 45:91–99, PMID: 25307304, 10.1016/j.neuro.2014.10.003.

Farzanehfar V, Naderi N, Kobarfard F, Faizi M. 2016. Determination of dibutyl phthalate neurobehavioral toxicity in mice. Food Chem Toxicol 94:221–226, PMID: 27311797, 10.1016/j.fct.2016.05.006.

Gascon M, Valvi D, Forns J, Casas M, Martínez D, Júlvez J, et al. 2015. Prenatal exposure to phthalates and neuropsychological development during childhood. Int J Hyg Environ Health 218(6):550–558, PMID: 26095249, 10.1016/j.ijheh.2015.05.006.

Golden R, Gandy J, Vollmer G. 2005. A review of the endocrine activity of parabens and implications for potential risks to human health. Crit Rev Toxicol 35(5):435–458, PMID: 16097138.

Goodman R. 1997. The Strengths and Difficulties Questionnaire: a research note. J Child Psychol Psychiatry 38(5):581–586, PMID: 9255702.

Grandjean P, Landrigan PJ. 2006. Developmental neurotoxicity of industrial chemicals. Lancet 368(9553):2167–2178, PMID: 17174709, 10.1016/S0140-6736(06)69665-7.

Harley KG, Gunier RB, Kogut K, Johnson C, Bradman A, Calafat AM, et al. 2013. Prenatal and early childhood bisphenol A concentrations and behavior in school-aged children. Environ Res 126:43–50, PMID: 23870093, 10.1016/j.envres.2013.06.004.

Hauser R, Calafat AM. 2005. Phthalates and human health. Occup Environ Med 62(11):806–818, PMID: 16234408, 10.1136/oem.2004.017590.

Heude B, Forhan A, Slama R, Douhaud L, Bedel S, Saurel-Cubizolles MJ, et al. 2016. Cohort Profile: The EDEN mother-child cohort on the prenatal and early postnatal determinants of child health and development. Int J Epidemiol 45(2):353–363, PMID: 26283636, 10.1093/ije/dyv151.

Hoshi H, Ohtsuka T. 2009. Adult rats exposed to low-doses of di-n-butyl phthalate during gestation exhibit decreased grooming behavior. Bull Environ Contam Toxicol 83(1):62–66, PMID: 19399352, 10.1007/s00128-009-9729-1.

Ishido M, Masuo Y, Terasaki M, Morita M. 2011. Rat hyperactivity by bisphenol A, but not by its derivatives, 3-hydroxybisphenol A or bisphenol A 3,4-quinone. Toxicol Lett 206(3):300–305, PMID: 21884766, 10.1016/j.toxlet.2011.08.011.

Kawaguchi M, Irie K, Morohoshi K, Watanabe G, Taya K, Morita M, et al. 2009. Maternal isobutyl-paraben exposure alters anxiety and passive avoidance test performance in adult male rats. Neurosci Res 65(2):136–140, PMID: 19560493, 10.1016/j.neures.2009.06.011.

Kobrosly RW, Evans S, Miodovnik A, Barrett ES, Thurston SW, Calafat AM, et al. 2014. Prenatal phthalate exposures and neurobehavioral development scores in boys and girls at 6-10 years of age. Environ Health Perspect 122(5):521–528, PMID: 24577876, 10.1289/ehp.1307063.

Koeppe ES, Ferguson KK, Colacino JA, Meeker JD. 2013. Relationship between urinary triclosan and paraben concentrations and serum thyroid measures in NHANES 2007-2008. Sci Total Environ 445-446:299–305, PMID: 23340023, 10.1016/j.scitotenv.2012.12.052.

Komada M, Itoh S, Kawachi K, Kagawa N, Ikeda Y, Nagao T. 2014. Newborn mice exposed prenatally to bisphenol A show hyperactivity and defective neocortical development. Toxicology 323:51–60, PMID: 24953688, 10.1016/j.tox.2014.06.009.

Krüger T, Long M, Bonefeld-Jørgensen EC. 2008. Plastic components affect the activation of the aryl hydrocarbon and the androgen receptor. Toxicology 246(2-3):112–123, PMID: 18294747, 10.1016/j.tox.2007.12.028.

Lien YJ, Ku HY, Su PH, Chen SJ, Chen HY, Liao PC, et al. 2015. Prenatal exposure to phthalate esters and behavioral syndromes in children at 8 years of age: Taiwan Maternal and Infant Cohort Study. Environ Health Perspect 123(1):95–100, PMID: 25280125, 10.1289/ehp.1307154.

Melchior M, Hersi R, van der Waerden J, Larroque B, Saurel-Cubizolles MJ, Chollet A, et al. 2015. Maternal tobacco smoking in pregnancy and children’s socio-emotional development at age 5: The EDEN mother-child birth cohort study. Eur Psychiatry 30(5):562–568, PMID: 25843027, 10.1016/j.eurpsy.2015.03.005.

Miodovnik A, Edwards A, Bellinger DC, Hauser R. 2014. Developmental neurotoxicity of ortho-phthalate diesters: review of human and experimental evidence. Neurotoxicology 41:112–122, PMID: 24486776, 10.1016/j.neuro.2014.01.007.

Mortamais M, Chevrier C, Philippat C, Petit C, Calafat AM, Ye X, et al. 2012. Correcting for the influence of sampling conditions on biomarkers of exposure to phenols and phthalates: a 2-step standardization method based on regression residuals. Environ Health 11(1):29, PMID: 22537080, 10.1186/1476-069X-11-29.

Mustieles V, Pérez-Lobato R, Olea N, Fernández MF. 2015. Bisphenol A: Human exposure and neurobehavior. Neurotoxicology 49:174–184, PMID: 26121921, 10.1016/j.neuro.2015.06.002.

Nakagami A, Negishi T, Kawasaki K, Imai N, Nishida Y, Ihara T, et al. 2009. Alterations in male infant behaviors towards its mother by prenatal exposure to bisphenol A in cynomolgus monkeys (Macaca fascicularis) during early suckling period. Psychoneuroendocrinology 34(8):1189–1197, PMID: 19345509, 10.1016/j.psyneuen.2009.03.005.

National Institute of Health and Medical Research. 2002. “Troubles mentaux: dépistage et prévention chez l’enfant et l’adolescent.” Paris, France:INSERM.

Paul KB, Hedge JM, Bansal R, Zoeller RT, Peter R, DeVito MJ, et al. 2012. Developmental triclosan exposure decreases maternal, fetal, and early neonatal thyroxine: a dynamic and kinetic evaluation of a putative mode-of-action. Toxicology 300(1-2):31–45, PMID: 22659317, 10.1016/j.tox.2012.05.023.

Paul KB, Hedge JM, Devito MJ, Crofton KM. 2010. Developmental triclosan exposure decreases maternal and neonatal thyroxine in rats. Environ Toxicol Chem 29(12):2840–2844, PMID: 20954233, 10.1002/etc.339.

Perera F, Nolte EL, Wang Y, Margolis AE, Calafat AM, Wang S, et al. 2016. Bisphenol A exposure and symptoms of anxiety and depression among inner city children at 10-12 years of age. Environ Res 151:195–202, PMID: 27497082, 10.1016/j.envres.2016.07.028.

Perera F, Vishnevetsky J, Herbstman JB, Calafat AM, Xiong W, Rauh V, et al. 2012. Prenatal bisphenol a exposure and child behavior in an inner-city cohort. Environ Health Perspect 120(8):1190–1194, PMID: 22543054, 10.1289/ehp.1104492.

Perera FP, Tang D, Wang S, Vishnevetsky J, Zhang B, Diaz D, et al. 2012. Prenatal polycyclic aromatic hydrocarbon (PAH) exposure and child behavior at age 6-7 years. Environ Health Perspect 120(6):921–926, PMID: 22440811, 10.1289/ehp.1104315.

Perrier F, Giorgis-Allemand L, Slama R, Philippat C. 2016. Within-subject pooling of biological samples to reduce exposure misclassification in biomarker-based studies. Epidemiology 27(3):378–388, PMID: 27035688, 10.1097/EDE.0000000000000460.

Philippat C, Bennett D, Calafat AM, Picciotto IH. 2015. Exposure to select phthalates and phenols through use of personal care products among Californian adults and their children. Environ Res 140:369–376, PMID: 25929801, 10.1016/j.envres.2015.04.009.

Philippat C, Botton J, Calafat AM, Ye X, Charles MA, Slama R. 2014. Prenatal exposure to phenols and growth in boys. Epidemiology 25(5):625–635, PMID: 25061923, 10.1097/EDE.0000000000000132.

Philippat C, Wolff MS, Calafat AM, Ye X, Bausell R, Meadows M, et al. 2013. Prenatal exposure to environmental phenols: concentrations in amniotic fluid and variability in urinary concentrations during pregnancy. Environ Health Perspect 121(10):1225–1231, PMID: 23942273, 10.1289/ehp.1206335.

Philippat C, Wolff MS, Calafat AM, Ye X, Bausell R, Meadows M, et al. 2012. Exposure to phthalates and phenols during pregnancy and offspring size at birth. Environ Health Perspect 120(3):464–470, PMID: 21900077, 10.1289/ehp.1103634.

Rappaport SM, Symanski E, Yager JW, Kupper LL. 1995. The relationship between environmental monitoring and biological markers in exposure assessment. Environ Health Perspect 103(suppl 3):49–53, PMID: 7635112.

Rice D, Barone S Jr. 2000. Critical periods of vulnerability for the developing nervous system: evidence from humans and animal models. Environ Health Perspect 108(suppl 3):511–533, PMID: 10852851, 10.2307/3454543.

Robinson O, Basagaña X, Agier L, de Castro M, Hernandez-Ferrer C, Gonzalez JR, et al. 2015. The pregnancy exposome: multiple environmental exposures in the INMA-Sabadell Birth Cohort. Environ Sci Technol 49(17):10632–10641, PMID: 26168307, 10.1021/acs.est.5b01782.

Roen EL, Wang Y, Calafat AM, Wang S, Margolis A, Herbstman J, et al. 2015. Bisphenol A exposure and behavioral problems among inner city children at 7-9 years of age. Environ Res 142:739–745, PMID: 25724466, 10.1016/j.envres.2015.01.014.

Sandin S, Lichtenstein P, Kuja-Halkola R, Larsson H, Hultman CM, Reichenberg A. 2014. The familial risk of autism. JAMA 311(17):1770–1777, PMID: 24794370, 10.1001/jama.2014.4144.

Silva MJ, Samandar E, Preau JL Jr, Reidy JA, Needham LL, Calafat AM. 2007. Quantification of 22 phthalate metabolites in human urine. J Chromatogr B Analyt Technol Biomed Life Sci 860(1):106–112, PMID: 17997365, 10.1016/j.jchromb.2007.10.023.

Torrey EF, Miller J, Rawlings R, Yolken RH. 1997. Seasonality of births in schizophrenia and bipolar disorder: a review of the literature. Schizophr Res 28(1):1–38, PMID: 9428062.

van Loo KM, Martens GJ. 2007. Genetic and environmental factors in complex neurodevelopmental disorders. Curr Genomics 8(7):429–444, PMID: 19412416, 10.2174/138920207783591717.

Whyatt RM, Liu X, Rauh VA, Calafat AM, Just AC, Hoepner L, et al. 2012. Maternal prenatal urinary phthalate metabolite concentrations and child mental, psychomotor, and behavioral development at 3 years of age. Environ Health Perspect 120(2):290–295, PMID: 21893441, 10.1289/ehp.1103705.

Witorsch RJ. 2014. Critical analysis of endocrine disruptive activity of triclosan and its relevance to human exposure through the use of personal care products. Crit Rev Toxicol 44(6):535–555, PMID: 24897554, 10.3109/10408444.2014.910754.

Wu Y, Beland FA, Fang JL. 2016. Effect of triclosan, triclocarban, 2,2′,4,4′-tetrabromodiphenyl ether, and bisphenol A on the iodide uptake, thyroid peroxidase activity, and expression of genes involved in thyroid hormone synthesis. Toxicol In Vitro 32:310–319, PMID: 26827900, 10.1016/j.tiv.2016.01.014.

Yan B, Guo J, Liu X, Li J, Yang X, Ma P, et al. 2016. Oxidative stress mediates dibutyl phthalateinduced anxiety-like behavior in Kunming mice. Environ Toxicol Pharmacol 45:45–51, PMID: 27262985, 10.1016/j.etap.2016.05.013.

Ye X, Kuklenyik Z, Needham LL, Calafat AM. 2005. Quantification of urinary conjugates of bisphenol A, 2,5-dichlorophenol, and 2-hydroxy-4-methoxybenzophenone in humans by online solid phase extraction-high performance liquid chromatography-tandem mass spectrometry. Anal Bioanal Chem 383(4):638–644, PMID: 16132150, 10.1007/s00216-005-0019-4.

Adipogenic Effects and Gene Expression Profiling of Firemaster® 550 Components in Human Primary Preadipocytes

Author Affiliations open

Environmental Health Science and Research Bureau, Health Canada, Ottawa, Ontario, Canada

PDF icon PDF Version (1.3 MB)

  • Background:
    Exposure to flame retardants has been associated with negative health outcomes including metabolic effects. As polybrominated diphenyl ether flame retardants were pulled from commerce, human exposure to new flame retardants such as Firemaster® 550 (FM550) has increased. Although previous studies in murine systems have shown that FM550 and its main components increase adipogenesis, the effects of FM550 in human models have not been elucidated.
    Objectives:
    The objectives of this study were to determine if FM550 and its components are active in human preadipocytes, and to further investigate their mode of action.
    Methods:
    Human primary preadipocytes were differentiated in the presence of FM550 and its components. Differentiation was assessed by lipid accumulation and expression of peroxisome proliferator-activated receptor γ (PPARG), fatty acid binding protein (FABP) 4 and lipoprotein lipase (LPL). mRNA was collected for Poly (A) RNA sequencing and was used to identify differentially expressed genes (DEGs). Functional analysis of DEGs was undertaken in Ingenuity Pathway Analysis.
    Results:
    FM550 triphenyl phosphate (TPP) and isopropylated triphenyl phosphates (IPTP), increased adipogenesis in human primary preadipocytes as assessed by lipid accumulation and mRNA expression of regulators of adipogenesis such as PPARγ, CCAAT enhancer binding protein (C/EBP) α and sterol regulatory element binding protein (SREBP) 1 as well as the adipogenic markers FABP4 LPL and perilipin. Poly (A) RNA sequencing analysis revealed potential modes of action including liver X receptor/retinoid X receptor (LXR/RXR) activation, thyroid receptor (TR)/RXR, protein kinase A, and nuclear receptor subfamily 1 group H members activation.
    Conclusions:
    We found that FM550, and two of its components, induced adipogenesis in human primary preadipocytes. Further, using global gene expression analysis we showed that both TPP and IPTP likely exert their effects through PPARG to induce adipogenesis. In addition, IPTP perturbed signaling pathways that were not affected by TPP. https://doi.org/10.1289/EHP1318
  • Received: 02 November 2016
    Revised: 18 May 2017
    Accepted: 23 May 2017
    Published: 14 September 2017

    Address correspondence to E. Atlas, Environmental Health Science and Research Bureau, Health Canada, 50 Colombine Driveway, Ottawa, Ontario, K1A 0K9, Canada. Telephone: 613-668-6151. Email: ella.atlas@canada.ca

    Supplemental Material is available online (https://doi.org/10.1289/EHP1318).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (179 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (22 KB)

Introduction

Stringent flammability standards set in the state of California resulted in the widespread use of chemical flame retardants in commercial products (Dodson et al. 2012). Of these, the polybrominated diphenyl ethers (PBDEs) were among the most abundantly used; however, due their toxicity and bioaccumulative properties, they were phased out of commerce. As such, industry was required to find alternatives such as the proprietary mixture Firemaster® 550 (FM550), which is used in commercial products including furniture, textiles, and electronics (Belcher et al. 2014; Stapleton et al. 2008).

FM550 is composed of four different compounds: bis(2-ethylhexyl)-2,3,4,5-tetrabromophthalate (TBPH) (8%), 2-ethylhexyl-2,3,4,5-tetrabromobenzoate (TBB) (30%), triphenyl phosphate (TPP) (17%), and isopropylated triphenyl phosphates (IPTPs) (45%) (Stapleton et al. 2008). The IPTPs consist of a mixture of mono-isopropylphenyl, diphenyl-phosphate, di-isopropylphenyl, phenyl-phosphate, and tris-isopropylphenyl-phosphate in various proportions (Phillips et al. 2016). Measurement of the FM550 metabolites diphenyl-phosphate (DPHP) and isopropyl-diphenyl-phosphate (ip-DPHP) in urine confirmed that exposure to the chemical mixture FM550 is ubiquitous (Hoffman et al. 2014). In addition, recent studies show that some of the FM550 components concentrations can reach 15,030 ng/g in house dust (Stapleton 2008) and that the metabolite ip-DPHP is ubiquitous in the urine of children at concentrations up to 24 ng/mL (Hoffman et al. 2014).

A few studies suggest that FM550 and its components have metabolic effects and act as environmental obesogens. For example, in a rodent model, perinatal and lactational exposure to FM550 induced behavioral and endocrine effects, increased adipose mass, and induced insulin resistance in the offspring (Patisaul et al. 2013). In addition, a study in murine stem cells showed that the FM550 components TPP and IPTP divert osteogenesis to the adipogenesis pathway through activation of peroxisome proliferator-activated receptor γ (PPARγ) (Pillai et al. 2014).

Although aforementioned studies (Patisaul et al. 2013; Pillai et al. 2014) suggest that FM550 is an endocrine disruptor and an environmental obesogen in murine cell cultures and animal models, little is known regarding its effects on human health and obesity. Primary human preadipocytes are a relevant tool to test the ability of chemicals to induce adipogenesis in human specimens and hence can identify a potential role for these chemicals to cause metabolic effects in humans (Boucher et al. 2014a; Boucher et al. 2014b). Previous work showed that the transcriptional cascade differs in human and murine differentiating preadipocytes (Tomlinson et al. 2006, 2010). This suggests that chemicals have potentially different specific targets in human cells compared with mouse cells. Further, human preadipocytes have different requirements for optimal differentiation compared with the mouse models (Tomlinson et al. 2006). One major difference is the requirement of clonal expansion for the murine cell model (3T3-L1) but not for the human primary preadipocytes (Janderová et al. 2003; Yeh et al. 1995). In addition, human primary preadipocytes require both dexamethasone (glucocorticoid agonist) and troglitazone (PPARG agonist) to induce differentiation (Janderová et al. 2003). As such, this model presents us with the opportunity to investigate whether the chemicals of interest are acting through PPARG activation or through the glucocorticoid pathway. Conversely, murine 3T3-L1 preadipocytes differentiate with either dexamethasone or a PPARG agonist (Ahmed and Atlas 2016). Finally, human preadipocytes are primary cells, and therefore provide an in vitro model that is more relevant to the human condition and relevant for obesogen screening.

The purpose of this study was to determine the effects of FM550 and its individual components on adipogenesis in human primary preadipocytes, and to explore modes of action through analysis of global transcriptomic response to these chemicals. We show that FM550 and its two major components, TPP and IPTP, induce adipogenesis in human primary cells. Furthermore, global gene expression analysis revealed that pathways other than PPARG may also be involved in the adipogenic changes induced in differentiating human cells in response to FM550 components.

Materials and Methods

Reagents

Chemicals were purchased from the following manufacturers: human insulin (Roche Diagnostics, Indianapolis, IN, USA); 3-isobutyl-1-methylxanthine (IBMX), dexamethasone (DEX), troglitazone, triphenyl phosphate (TPP), and dimethyl sulfoxide (DMSO) (Sigma-Aldrich, Oakville, Ontario, Canada); 2-ethylhexyl-2,3,4,5-tetrabromobenzoate (TBB) and bis(2-ethylhexyl)-2,3,4,5-tetrabromophthalate (Toronto Research Chemicals, Toronto, Ontario, Canada). Isopropylated triphenyl phosphate (IPTP) was a generous gift from W. Casey (National Institute of Environmental Health Sciences, National Institutes of Health in the United States). Analysis of the IPTP mixture was hampered by the lack of availability of pure standards for the various congeners suspected to be present. By comparing mass estimates of the whole molecules (from positive chemical ionization analyses) with those of molecular fragments (from electron impact mode analyses) we estimate that the majority of the material is composed of three congeners. These were triphenyl phosphate (TPP ∼18%); monoisopropylphenyl, diphenylphosphate (mITP ∼55%); and di(isopropylphenyl), phenylphosphate (dITP ∼25%). In addition, there was a small amount of tris(isopropylphenyl)phosphate present in the test material, but at a substantially lower concentration (∼1.5%). Firemaster® 550 (FM550) was a generous gift from B. Chittam (Wellington Laboratories, Guelph, Ontario, Canada).

Culture and Differentiation of Human Primary Subcutaneous Preadipocytes

Primary human subcutaneous preadipocytes (ZenBio, Inc., Research Triangle Park, NC, USA) from female donors ages 25, 38, 40, 39, and 34 y with body mass indices of 18.8, 21.6, 23.3, 22.6, 21.5 (kg/m2) and who were of several ethnicities (one Caucasian, two Hispanic, one Asian, and one unknown) were taken from thigh, back, abdomen, and flank depots and differentiated as previously described (Boucher et al. 2016) with modifications. Briefly, human primary preadipocytes were seeded in six-well dishes in subcutaneous preadipocyte media (PM-1, Zenbio Inc.), containing 10% fetal calf serum (Wisent, Montreal, Quebec, Canada). When cells reached confluence (day 0), they were treated with 100 nM insulin (I) and 500 μM IBMX (M) until day 4, with a media change on day 2. From day 4 onward, cells received insulin with media changes on days 4 and 8. For the positive control (MIDT), where the cells require both dexamethasone and troglitazone, dexamethasone (D) (1 μM) was added from days 0 to 14 and troglitazone (T) (5 μM) from days 2 to 14 in addition to MI. Treatments with the chemicals of interest were performed as follows: a) when the chemicals were tested for their ability to replace troglitazone, the cells were treated with dexamethasone from day 0 and the test chemical (0–200 μM FM550, 0–200 μM IPTP, 0–20 μM TPP, 0–20 μM TBPH, 0–20 μM TBB) from day 2, with media replacements on days 4 and 8 (MID condition); b) when the chemicals were to replace dexamethasone, the test chemical was added from day 0, and troglitazone with the test chemical were added from day 2 onward and replaced with media changes on days 2, 4, and 8 (MIT condition). IBMX and insulin were added in all treatments as described above for the MIDT condition.

Nile Red Staining of Lipids

Primary human preadipocytes were differentiated as described above for 14 d with the indicated treatments and controls in black collagen-coated 96-well plates (Fisher Scientific, Canada). The level of differentiation was assessed using a fluorescence plate reader as follows. At day 14, cells were fixed with 4% paraformaldehyde (VWR, Canada) for 30 min followed by PBS washes. Background fluorescence was read in PBS at 485/528 for Nile red and 360/460 for DAPI. Cells were then stained with Nile red (1 μg/mL) to stain for lipid droplets and DAPI (1 μg/mL) to stain nuclei as previously described (Greenspan et al. 1985). Nile red fluorescence was read at 485/528 nm (excitation/emission) in a Synergy 2 fluorescence plate reader (BioTek Instruments, Inc., Winooski, VT, USA) and DAPI staining, nuclei staining, was measured at 360/460 nm . To calculate Nile red to DAPI ratios the background fluorescence was first subtracted from the readings of Nile red and DAPI at the respective wavelengths, and Nile red/DAPI ratios were calculated for each of the wells.

Western Blotting

Human primary preadipocytes were seeded in six-well dishes and treated according to the differentiation protocol described above. On day 14, cells were lysed in RIPA buffer (20 mM Tris pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% sodium deoxycholate, 2% NP-40, 0.4% SDS, 10% glycine) containing protease inhibitors (Roche Diagnostics, Laval, Quebec, Canada). Western blots were performed by probing with primary fatty acid binding protein (FABP) 4 and β-actin ACTB antibodies (Cell Signaling Technology, Danvers, MA, USA) followed by appropriate HRP-linked secondary antibodies, and developed using Clarity Western ECL Substrate (BioRad, Hercules, CA, USA). Relative optical densities were quantified using Image Lab software (BioRad), and values of terminal differentiation markers were normalized to ACTB levels.

mRNA Extraction and Real-Time Quantitative PCR (RT-qPCR)

Total RNA was extracted from differentiating human primary preadipocytes, treated as previously described. Samples were taken on days 4, 6, 9, and 12 using the RNeasy Mini kit and genomic DNA was eliminated using the RNase-Free DNase Kit (Qiagen, Mississauga, Ontario, Canada). RNA was reverse transcribed using iScript cDNA Synthesis Kit (BioRad). cDNA expression levels were analyzed by the CFX96-PCR Detection System using the iQSYBR SsoFast EvaGreen Supermix (BioRad). Primer sequences for each gene are summarized in Table S1. Primer efficiencies were ≥90% and specificity was confirmed by sequence blast and melting curve analysis. All target gene transcripts were normalized to ACTB expression, which was not affected by treatment. Fold inductions were calculated using time-matched, control solvent-treated samples, and the comparative CT (ΔΔCT) method was used for data analysis.

For RNA-seq analysis, total RNA was extracted from differentiating cells treated as described above using the RNeasy Kit and genomic DNA was eliminated using the RNase-Free DNase Kit (both from Qiagen, Mississauga, Ontario, Canada). RNA was quantified and RNA quality was determined using a BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA). RNA samples with A260/A280 ratios >1.8 and RNA integrity numbers (RIN) >8.0 were used.

Ion ProtonTM Sequencing (RNA-Seq)

Total RNA from preadipocytes from five donors treated with MID, MITD, TPP, or IPTP was collected on day 6 of differentiation of preadipocytes, from five donors were used for RNA-Seq. Poly (A) RNA enrichment (DynaBeads® mRNA DIRECT Micro Kit, Thermo Fisher, Waltham, MA, USA) was performed for each sample on 1 μg of total RNA. The Ion Total RNA-Seq Kit version 2 (Thermo Fisher, Waltham, MA, USA) was used to fragment and prepare the sample libraries from poly (A)–enriched samples. The 3′-end barcode adapters provided in the Ion XpressTM RNA-Seq Barcode Kit were ligated to the ends of the fragmented libraries (each PCR product receiving its own unique barcode). Libraries were then amplified using the Platinum® PCR SuperMix High Fidelity (Thermo Fisher, Waltham, MA, USA). Each amplified library was quantified/qualified using the High Sensitivity D1000 ScreenTape Kit and the Agilent® 2,200 TapeStation Instrument. Aliquots of each library were pooled together for a total final concentration of 50 pM. Emulsion PCR and chip loading was performed on the Ion Chef with the Ion P1 HI-Q Chef kit (Thermo Fisher, Waltham, MA, USA). The chips (P1 version 3) were run on the Ion Proton using the HI-Q chemistry. The Proton™ Torrent Server version 4.3 was used to interpret the sequencing data and generate unaligned binary version of a sequence aignment map (BAM) files for each barcoded sample. Reads were trimmed to remove low quality read prefixes, and then aligned to the reference genome (GRCh38v77) using Star (Dobin et al. 2013) and Bowtie (Langmead and Salzberg 2012). Following alignment, gene counting was performed with HT-Seq count (http://www-huber.embl.de/users/anders/HTSeq/doc/count.html) with the m parameter set to “intersection-nonempty” using the Ensembl GTF annotation (GRCh38v77). The table of counts was then imported into R (version 3.1.0; R Development Core Team) where genes that did not obtain a minimum total count of at least 0.5 reads per million within at least one dose group were removed from further analyses. The EdgeR (Robinson et al. 2010) package was used for the analysis. The data was normalized with TMM (Robinson and Oshlack 2010). The calculation of differentially expressed genes was performed using a paired design and using the GLM function. The data are publically available from Sequence Read Archive (http://www.ncbi.nlm.nih.gov/Traces/sra/sra.cgi?view=studies; BioProject ID SRP100037).

Biological and Pathway Analyses

All genes with a false discovery rate (FDR) p<0.05 and fold change>±1.5-fold compared with matched controls were considered DEGs and selected for further analyses. Biological functions, canonical pathways, and upstream regulatory molecules/networks were analyzed using Ingenuity Pathway Analysis (IPA) (Ingenuity Systems, Redwood City, CA, USA). DEGs were uploaded into IPA for functional analysis. Within IPA, a standard analytical workflow was used to identify the enriched canonical pathways, diseases, and biological functions, using adipose tissue as the target. The significance of the association between the gene expression data set and the pathways and functions in IPA was measured using IPA’s built in Fischer’s exact test (deemed significant if p≤0.05). (For further details see https://www.ingenuity.com/wp-content/themes/ingenuity-qiagen/pdf/ipa/IPA-netgen-algorithm-whitepaper.pdf).

Statistical Analyses

One-way ANOVA, followed by a Dunnett’s or Tukey’s post hoc test were used when comparing multiple means within an experiment. Student’s t-test was used when comparing two means. Significance was defined as p≤0.05. Statistical analyses were performed using SigmaPlot software version 12.5 (San Jose, CA, USA) or GraphPad Prism version 7 (La Jolla, CA, USA).

Results

Effects of FM550, IPTP, and TPP on Adipogenesis in Human Primary Preadipocytes

Previous studies in murine cell models have shown that FM550, and its component TPP, increased adipogenesis through direct activation of PPARγ (Pillai et al. 2014). We investigated whether this is also true for human primary preadipocytes by assessing lipid accumulation and expression levels of adipogenic markers. Two conditions were used to establish whether FM550 and its components induce differentiation in human primary preadipocytes and if they act only through PPARG in this model system. One condition was a differentiation protocol where the chemicals of interest were added in the presence of IBMX, insulin, and dexamethasone (MID). The second condition was when the chemicals were added to human preadipocytes exposed to IBMX, insulin, and troglitazone (MIT). In positive controls, IBMX, insulin, dexamethasone, and troglitazone were added (MIDT).

Lipid accumulation was measured on day 14 by Nile red fluorescence values normalized to DAPI fluorescence. Human primary preadipocytes treated with the FM550 mixture exhibited a dose-dependent increase in lipid accumulation, with statistical significance achieved in the MID condition at 200 μM FM550, where a 4-fold increase as compared with MID alone was achieved (Figure 1A). TPP also induced a 4-fold increase in lipid accumulation relative to MID controls at 20 μM (Figure 1B). An increasing trend was also seen with TPP treatment in the MIT condition, but this was not statistically significant. IPTP treatment induced a significant ∼4-fold increase in lipid accumulation by Nile red staining at both 10 and 20 μM, as compared with the MID condition (Figure 1C). An increasing trend was also observed in the MIT condition; however, this increase did not reach statistical significance (Figure 1C). TBB and TBPH did not increase lipid accumulation, although an increasing trend was observed in the TBB dose–response curve (Figure 1D,E). As expected, the positive control MIDT increased lipid accumulation (see Figure S1).

Figures 1A, 1B, 1C, 1D, and 1E are bar graphs with standard errors plotting changes in lipid accumulation (y-axis) for MID and MIT groups in response to treatment with varying concentrations of Firemaster 550, TPP, IPTP, TBB, and TBPH, respectively (all components in micromolar) (x-axis).
Figure 1. Effects of Firemaster 550®, and its components TPP and IPTP, on lipid accumulation in differentiating human preadipocytes. Human primary preadipocytes were induced to differentiate for 14 d in the presence of MI and 1 μM dexamethasone (MID) or MI and 5 μM troglitazone (MIT) supplemented with either FM550 (0–200 μM) or its components (TPP, IPTP, TBB, and TBPH; 0–20 μM). At day 14 of differentiation, lipid accumulation was quantified by Nile red staining normalized to DAPI. Data from all treatments, normalized to their respective control condition (MID or MIT), are graphically presented as mean±SEM of four separate donor samples. *p<0.05 compared with respective controls, as assessed by one-way ANOVA with Dunnett’s post hoc tests.

Effects of FM550, IPTP, and TPP on FABP4 Protein Expression in the Presence of Dexamethasone

To further assess the extent of adipocyte differentiation and to delineate whether in human preadipocytes FM550 and its components up-regulate the expression of known PPARG targets, protein levels of the mature adipocyte marker, FABP4, were measured by Western blotting. In the presence of MID, 200 μM FM550 significantly increased FABP4 expression by ∼4-fold compared with MID controls; however, this effect was not observed when the mixture was added in the presence of MIT (Figure 2A). This may be partially due to the higher background in the presence of troglitazone given that FABP4 is a known target of PPARG. Similarly, 20 μM TPP induced a 6-fold increase in FABP4 expression relative to MID (Figure 2B), although increasing trends were observed at lower concentrations but did not reach statistical significance. Again, this effect was not observed in the presence of MIT (Figure 2B). Remarkably, IPTP treatment induced an increase not only at 20 μM (10-fold), but also at 10 μM (5-fold) compared with MID controls (Figure 2C). Note that, at equivalent concentrations (10 and 20 μM), IPTP induced a higher expression of FABP4 relative to TPP treatment (Figure 2B,C). TBPH and TBB did not affect FABP4 protein expression in the presence of either MID or MIT (data not shown).

Figure 2A shows an immunoblot of FABP4 and ACTB expressions in the presence of MID and MIT supplemented with Firemaster 550, and a bar graph with standard error plotting changes in FABP4 protein levels (y-axis) in response to treatment with varying concentrations of Firemaster 550 (micromolar) (x-axis). Figure 2B shows an immunoblot of FABP4 and ACTB expressions in the presence of MID and MIT in response to treatment with varying concentrations of TPP, and a bar graph with standard error plotting FABP4 protein levels (y-axis) in response to treatment with varying concentrations of TPP (micromolar) (x-axis). Figure 2C comprises an immunoblot of FABP4 and ACTB expressions in the presence of MID and MIT in response to treatment with varying concentrations of IPTP, and a bar graph with standard error plotting FABP4 protein levels (y-axis) in response to treatment with varying concentrations of (micromolar) (x-axis).
Figure 2. Effects of Firemaster 550®, and its components TPP and IPTP, on FABP4 protein expression in the presence of dexamethasone or troglitazone. Human primary preadipocytes were induced to differentiate for 14 d in the presence of MI and 1 μM dexamethasone (MID) or MI and 5 μM troglitazone (MIT) supplemented with either FM550 (0–200 μM) (A), or its components (TPP and IPTP; 0–20 μM) (B, C). Equal amounts of solubilized cellular proteins were separated by SDS-PAGE and immunoblotted with antibodies against FABP4 and ACTB as a loading control. Densitometric data from four separate donor samples, normalized to loading control, are graphically presented as means±SEM. *p<0.05 compared with respective controls, as assessed by one-way ANOVA with Dunnett’s post hoc tests.

Effects of TPP and IPTP on the mRNA Expression of Transcriptional Regulators of Adipogenesis and Adipogenic Markers in Human Primary Preadipocytes

Our initial findings indicated that the MID condition was optimal for FM550-induced human adipocyte differentiation and that the FM550 components TPP and IPTP-containing 18% TPP were adipogenic. Therefore, the temporal changes in the mRNA levels of transcription factors and differentiation markers of adipogenesis were assessed in response to 20 μM TPP and 20 μM IPTP relative to MID controls.

First we assessed the mRNA levels of master regulators of adipogenesis, PPARG and CCAT enhancing binding protein α (CEBPA). PPARG expression was increased through the course of differentiation in TPP- and IPTP-treated cells relative to MID controls (Figure 3A). CEBPA levels were also increased following treatment with both TPP and IPTP during adipogenesis compared with MID controls (Figure 3B). In the TPP-treated cells, this increase was observed earlier in differentiation; however, similar levels were attained by both chemicals by the end of differentiation (Figure 3E). For the mature adipocyte markers, both FABP4 and lipoprotein lipase (LPL) mRNA expression levels were increased with adipogenesis in response to both TPP and IPTP relative to MID control. Again, although the increase with IPTP was delayed relative to TPP, both components led to similar levels by the end of differentiation (Figure 3C, D). Perilipin (PLIN) mRNA expression, indicative of lipid droplet formation, was also up-regulated by TPP and IPTP during adipogenesis as compared with MID control. TPP increased PLIN levels as early as day 6, and the levels were maintained throughout the differentiation process. However, the increase by IPTP was to a lower extent than in TPP-treated cells and reached statistical significance only at day 9 (Figure 2E). We also measured the mRNA levels of the lipogenic transcription factor, sterol regulatory element binding transcription factor 1 (SREBF1), and found that both TPP and IPTP increased its expression during differentiation as compared with MID (Figure 3F). As expected, the positive control (i.e., MIDT) increased the expression of PPARG, CEBPA, FABP4, LPL, PLIN, and SREBF1 throughout the differentiation process as illustrated in Figure S2.

Figures 3A, 3B, 3C, 3D, 3E, and 3F are line graphs with standard error plotting changes in PPAR gamma mRNA levels, CEBP alpha mRNA levels, FABP4 mRNA levels, LPL mRNA levels, PLIN mRNA levels, and SREBF1 mRNA levels, respectively (y-axis), with increase in time in days (x-axis) for treatment with TPP and IPTP.
Figure 3. Effects of TPP and IPTP on the mRNA expression levels of transcriptional regulators of adipogenesis and adipogenic markers in differentiating human preadipocytes. Human primary preadipocytes were induced to differentiate for 4, 6, 9, and 12 d in the presence of 1 μM dexamethasone (MID) supplemented with either TPP (20 μM) or IPTP (20 μM). RNA was isolated and the mRNA levels of adipogenic markers, PPARG (A), CEBPA (B), FABP4 (C), LPL (D), PLIN1 (E), and SREBF1 (F) were quantified by real-time qPCR. Levels were normalized to endogenous ACTB mRNA, and expressed as a fold over the control condition (MID) for each time point. Results from five separate donor samples are graphically presented as means±SEM. #p<0.05 for TPP-treated cells, and *p<0.05 for IPTP-treated cells compared with MID controls, as assessed by one-way ANOVA with Dunnett’s post hoc tests.

Transcriptional Profiling of Human Primary Preadipocytes Treated with TPP, IPTP, and Troglitazone

To further investigate the effects of TPP and IPTP on the adipogenic pathway, human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP, or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq. Overall, TPP and IPTP treatments resulted in 380 and 713 DEGs, respectively, whereas troglitazone treatment resulted in 3,277 (Figure 4). Of the DEGs affected by TPP, only 5.8% were distinct; in contrast, 29% of DEGs affected by IPTP were unique. Most of the genes up-regulated by TPP (84%) were in common with troglitazone; whereas only 65% of the IPTP affected genes were also affected by troglitazone. Hierarchical cluster analysis of the fold changes for all DEGs from IPTP, troglitazone, and TPP shows that TPP- and IPTP-induced expression profiles cluster together and apart from troglitazone, indicating that they more closely resemble each other than the troglitazone-treatment group (Figure 4B). In addition, pairwise analysis using fold changes of the common genes shows that TPP-induced expression profiles are more strongly correlated with troglitazone and IPTP (correlation=0.87 and 0.86, respectively), than IPTP is correlated with troglitazone (correlation=0.75). All correlations are significant with a p-value of <0.00001. Overall, this analysis demonstrates that the DEGs in common across the treatments are highly correlated, and that IPTP is less similar to troglitazone than TPP. The top 10 up- and down-regulated DEGs affected by TPP, IPTP, and troglitazone are listed in Table 1. Table 1 shows that 45% of the top DEGs were common between TPP and IPTP, 20% of the top DEGs in TPP-treated cells were in common with troglitazone, and only 5% of the top DEGs in IPTP-treated cultures were in common with troglitazone.

Figure 4A is a Venn diagram with three circles labeled TPP, IPTP, and troglitazone. The three circles intersect each other. Figure 4B is a graphical output of hierarchical cluster analysis.
Figure 4. Differentially expressed genes (DEGs) affected by TPP, IPTP, and troglitazone. Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP, or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq analysis. The overlap of significant (−1.5≤fold change≥1.5, FDR p<0.05) DEGs between all treatments are depicted in a Venn diagram (A). Venn diagrams were produced using the on-line tool Venny (http://bioinfogp.cnb.csic.es/tools/venny/). (B) Hierarchical cluster analysis of the fold change for all DEGs from IPTP, troglitazone, TPP having an FDR of p<0.05 and fold change >1.5. The distance is metric 1− the Spearman correlation, values greater than 5 were truncated strictly for the color scale.
Table 1. Top 10 differentially expressed genes in differentiating human primary preadipocytes.
Gene symbol Gene name GO biological process Fold change FDR (p-value)
TPP
ITIH1 Inter-alpha-trypsin inhibitor heavy chain 1 Hyaluronan metabolic process 7.9 <0.0001
PCK1 Phosphoenolpyruvate carboxykinase 1 Glucose metabolic process 7.1 <0.0001
FABP4 Fatty acid binding protein 4 Cholesterol homeostasis 6.6 <0.0001
GPBAR1 G-protein–coupled bile acid receptor 1 Cell surface bile acid receptor signaling pathway 5.7 <0.0001
PPP1R1A Protein phosphatase 1 regulatory inhibitor subunit 1A Glycogen metabolic process 5.6 <0.0001
SLCO4C1 Solute carrier organic anion transporter family, member 4C1 Cell differentiation 5.3 <0.0001
FNDC5 Fibronectin type III domain-containing protein 5 Response to muscle activity 4.6 <0.0001
ADIPOQ Adiponectin, C1Q, and collagen domain containing Adiponectin-activated signaling pathway 4.5 <0.0001
CDHR1 Cadherin-related family member 1 Homophilic cell adhesion via plasma membrane adhesion molecules 4.3 <0.0001
ITIH5 Inter-alpha (globulin) inhibitor H5 Negative regulation of peptidase activity 4.3 <0.0001
BPIFB4 BPI fold containing family B member 4 Lipid binding −3.0 0.003
GPR78 G-protein–coupled receptor 78 Adenylate cyclase-activating G-protein–coupled receptor signaling pathway −2.9 0.009
MYH11 Myosin heavy chain 11 Elastic fiber assembly −2.6 0.0006
AMZ1 Archaelysin family metallopeptidase 1 Proteolysis −2.5 <0.0001
FBXL22 F-box and leucine-rich repeat protein 22 Proteasome-mediated ubiquitin-dependent protein catabolic process −2.4 <0.0001
RGS5 Regulator of G-protein signaling 5 Positive regulation of GTPase activity −2.3 <0.0001
D4S234E Neuron-specific gene family member 1 Positive regulation of receptor recycling −2.3 0.0008
CORIN Corin, serine peptidase Peptide hormone processing −2.2 <0.0001
KRT18 Keratin 18 Anatomical structure morphogenesis −2.2 <0.0001
CNN1 Calponin 1 Actomyosin structure organization −2.1 <0.0001
IPTP
SLCO4C1 Solute carrier organic anion transporter family, member 4C1 Cell differentiation 7.9 <0.0001
TRIM63 Tripartite motif containing 63 Protein ubiquitination 6 <0.0001
ABCG1 ATP-binding cassette subfamily G member 1 Cholesterol metabolic process 5.9 <0.0001
ITIH1 Inter-alpha-trypsin inhibitor heavy chain 1 Hyaluronan metabolic process 5.7 <0.0001
FABP3 Fatty acid binding protein 3 Fatty acid metabolic process 4.8 <0.0001
FNDC5 Fibronectin type III domain containing 5 Response to muscle activity 4.9 <0.0001
PLEKHG6 Pleckstrin homology and RhoGEF domain containing G6 Positive regulation of GTPase activity 4.7 <0.0001
RNF157 Ring finger protein 157 Zinc ion binding 4.6 <0.0001
EFNA1 Ephrin A1 Activation of MAPK activity 4.6 <0.0001
PPP1R1A Protein phosphatase 1 regulatory inhibitor subunit 1A Glycogen metabolic process 4.2 <0.0001
MYH11 Myosin heavy chain 11 Elastic fiber assembly −4.8 <0.0001
LBP Lipopolysaccharide binding protein Lipopolysaccharide-mediated signaling pathway −4.5 <0.0001
RGS5 Regulator of G-protein signaling 5 Positive regulation of GTPase activity −4.5 <0.0001
OXTR Oxytocin receptor G-protein–coupled receptor signaling pathway −4.1 <0.0001
KRT18 Keratin 18 Anatomical structure morphogenesis −3.7 <0.0001
DKK2 Dickkopf WNT signaling pathway inhibitor 2 Wnt signaling pathway −3.6 <0.0001
FBXL22 F-box and leucine-rich repeat protein 22 Proteasome-mediated ubiquitin-dependent protein catabolic process −3.5 <0.0001
CXCL5 C-X-C motif chemokine ligand 5 G-protein–coupled receptor signaling pathway −3.4 <0.0001
CORIN Corin, serine peptidase Peptide hormone processing −3.4 <0.0001
TLL1 Tolloid like 1 Cell differentiation −3.2 <0.0001
Troglitazone
SCN4A Sodium voltage-gated channel alpha subunit 4 Regulation of ion transmembrane transport 261.5 <0.0001
CD96 CD96 molecule Cell adhesion 147.7 <0.0001
PCK1 Phosphoenolpyruvate Carboxykinase 1 Glucose metabolic process 121.3 <0.0001
ADAMTS18 ADAM metallopeptidase with thrombospondin type 1 motif 18 Proteolysis 113.0 <0.0001
MOGAT1 Monoacylglycerol O-acyltransferase 1 Diacylglycerol biosynthetic process 107.2 <0.0001
ADRA1A Adrenoceptor alpha 1A G-protein–coupled receptor signaling pathway 98.3 <0.0001
SCGN Secretagogin, EF-hand calcium binding protein Regulation of cytosolic calcium ion concentration 93.3 <0.0001
PPP1R1A Protein phosphatase 1 regulatory inhibitor subunit 1A Glycogen metabolic process 67.8 <0.0001
CDH22 Cadherin 22 Calcium-dependent cell-cell adhesion via plasma membrane cell adhesion molecules 61.6 <0.0001
ADIPOQ Adiponectin, C1Q, and collagen domain containing Adiponectin-activated signaling pathway 56.6 <0.0001
CCL13 C-C motif chemokine ligand 13 G-protein–coupled receptor signaling pathway −9.8 <0.0001
STEAP4 STEAP4 metalloreductase Iron ion homeostasis −6.7 <0.0001
PPP2R2C Protein phosphatase 2 regulatory subunit Bgamma Regulation of protein phosphatase type 2A activity −6.1 <0.0001
PMAIP1 Phorbol-12-myristate-13-acetate-induced protein 1 Intrinsic apoptotic signaling pathway −6.1 <0.0001
BPIFB4 BPI fold containing family B member 4 Lipid binding −6.1 <0.0001
CSF3 Colony stimulating factor 3 Cellular response to cytokine stimulus −5.9 <0.0001
RAB27B RAB27B, member RAS oncogene family Small GTPase-mediated signal transduction −5.9 <0.0001
WISP1 WNT1 inducible signaling pathway protein 1 Wnt signaling pathway −5.8 <0.0001
FRMPD4 FERM and PDZ domain containing 4 Positive regulation of synapse structural plasticity −5.4 <0.0001
RGCC Regulator of cell cycle Fibroblast activation −5.2 <0.0001

Note: Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP, or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq analysis. The top 10 up- and down-regulated genes are shown. FDR, false discovery rate; GO, gene ontology.

Validation of Select Differentially Expressed Genes by RT-qPCR

We validated some of the DEGs found in the RNA-seq analysis by RT-qPCR. Of the up-regulated genes, FABP5, PLIN4, and phosphoenolpyruvate carboxykinase 1 (PCK1) were increased in both TPP- and IPTP-treated cells. However, only TPP significantly increased the mRNA levels of the above genes (Figure 5A–C). By contrast, other genes including ATP-binding cassette subfamily G member 1 (ABCG1), solute carrier organic anion transporter family, member 4C1 (SLCO4C1), and FABP3 were significantly increased by IPTP and not TPP (Figure 5D–F). It is possible that by adding more repeats we would have reached statistical significance for FABP5, PCK1, PLIN4, ABCG1, and SLCO4C1 for both treatments. Some genes were equally increased by both treatments (Figure 5G–I). All of these genes were up-regulated by troglitazone control (see Figure S3). Interestingly, lipopolysaccharide binding protein (LBP) levels were down-regulated in IPTP-treated cells, not affected by TPP exposure, and up-regulated in troglitazone positive controls (Figure 5J; see also Figure S3). Further, keratin 18 (KRT18) levels were decreased by both TPP and IPTP; however, the expression of this gene was not changed in response to troglitazone (Figure 5K; see also Figure S3).

Figures 5A, 5B, 5C, 5D, 5E, 5F, 5G, 5H, 5I, 5J, and 5K are bar graphs with standard error plotting FABP5 mRNA levels, PLIN4 mRNA levels, PCK1 mRNA levels, ABCG1 mRNA levels, SLCO4C1 mRNA levels, FABP3 mRNA levels, LDLR mRNA levels, CIDEC mRNA levels, FASN mRNA levels, LBP mRNA levels, and KRT18 mRNA levels (y-axis) across control-treated, 20 micromolar TPP-treated, and 20 micromolar IPTP-treated samples (x-axis).
Figure 5. Validation of selected differentially expressed genes (DEGs) by RT-qPCR. Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP or 20 μM IPTP. At day 6 of differentiation, RNA was collected for RNA-seq analysis. The mRNA levels of select DEGs from RNA-seq analysis were quantified by RT-qPCR. Levels were normalized to endogenous ACTB mRNA, and expressed as a fold over the control condition (MID) for each treatment. Results from five separate donor samples are graphically presented as means±SEM. *p<0.05, **p<0.01, and ***p<0.001 for TPP- and IPTP- treated samples compared with control; #p<0.05, ##p<0.01, and ###p<0.001 for TPP-treated compared with IPTP-treated samples, as assessed by one-way ANOVA with Tukey’s post hoc tests.

Canonical Pathways and Upstream Regulators Identified in IPA’s Knowledge Base in Human Primary Preadipocytes Treated with TPP and IPTP

The DEGs obtained from the RNA-seq analysis of TPP, IPTP, and troglitazone treatments were analyzed using IPA to identify enrichment of canonical pathways and upstream regulators. The top canonical pathways for each treatment identified in IPA are summarized in Table 2. As expected, the adipogenesis pathway was significantly enriched in TPP-, IPTP-, and troglitazone-treatment groups. In addition, there were several common pathways between TPP and IPTP treatments, such as liver X receptor (LXR)/retinoid X receptor (RXR) activation, cholesterol biosynthesis I, and lipopolyscharide (LPS)/interleukin-1 (IL-1)–mediated inhibition of RXR function (Table 2A,B). Of these pathways, only LPS/IL-1–mediated inhibition of RXR function pathway was also in the top 20 pathways affected by troglitazone (Table 2C). Only 65% of the canonical pathways were common to both TPP and IPTP treatments. Some of the distinct pathways identified in TPP-treated cells were protein kinase A signaling, AMP-activated protein kinase (AMPK) signaling, and type II diabetes mellitus signaling. In the IPTP treatment, unique pathways included caveolar-mediated endocytosis signaling, clathrin-mediated endocytosis signaling, and acute phase response signaling. Of the top 20 pathways affected by IPTP and TPP treatment, several pathways overlapped and included superpathway of cholesterol biosynthesis, cholesterol biosynthesis (I, II, III), LXR/RXR activation (Table 2A,B). Troglitazone treatment was also associated with effects on pathways such as cAMP-mediated signaling and PXR/RXR activation, in addition to mitochondrial dysfunction and oxidative phosphorylation pathways (Table 2C).

Table 2. Top 20 significant canonical pathways identified by IPA.
IPA canonical pathway p-Value Ratio
TPP
 LXR/RXR activation <0.0001 0.227
 LPS/IL-1–mediated inhibition of RXR function <0.0001 0.109
 Superpathway of cholesterol biosynthesis <0.0001 0.688
 TR/RXR activation <0.0001 0.177
 Protein kinase A signaling 0.009 0.0526
 Tight junction signaling 0.0002 0.0909
 AMPK signaling 0.001 0.0758
 Adipogenesis pathway 0.0003 0.0947
 Hepatic fibrosis/hepatic stellate cell activation 0.001 0.0796
 ILK signaling 0.002 0.0769
 Cholesterol biosynthesis I <0.0001 0.889
 Cholesterol biosynthesis II (via 24,25-dihydrolanosterol) <0.0001 0.889
 Cholesterol biosynthesis III (via Desmosterol) <0.0001 0.889
 Cellular effects of sildenafil (Viagra) <0.0001 0.133
 Type II diabetes mellitus signaling 0.0008 0.093
 G-protein–coupled receptor signaling 0.019 0.0552
 Cardiac β-adrenergic signaling 0.001 0.0972
 cAMP-mediated signaling 0.01 0.0631
 Stearate biosynthesis I (animals) <0.0001 0.231
 FXR/RXR activation 0.003 0.0968
IPTP
 LXR/RXR activation <0.0001 0.288
 Hepatic fibrosis/hepatic stellate cell activation <0.0001 0.159
 LPS/IL-1–mediated inhibition of RXR function <0.0001 0.124
 Clathrin-mediated endocytosis signaling 0.003 0.0917
 Superpathway of cholesterol biosynthesis <0.0001 0.625
 Acute phase response signaling 0.003 0.0962
 Tight junction signaling 0.005 0.0909
 ILK signaling 0.008 0.0855
 FXR/RXR activation 0.0003 0.145
 TR/RXR activation 0.0003 0.145
 Agranulocyte adhesion and diapedesis 0.005 0.0978
 Cellular effects of sildenafil (Viagra) 0.001 0.133
 Cholesterol biosynthesis I <0.0001 0.778
 Cholesterol biosynthesis II (via 24,25-dihydrolanosterol) <0.0001 0.778
 Cholesterol biosynthesis III (via Desmosterol) <0.0001 0.778
 Atherosclerosis signaling 0.01 0.101
 Hepatic cholestasis 0.04 0.0761
 Adipogenesis pathway 0.05 0.0737
 Caveolar-mediated endocytosis signaling 0.008 0.118
 Paxillin signaling 0.03 0.0896
Troglitazone
 Role of macrophages, fibroblasts, and endothelial cells in rheumatoid arthritis 0.017 0.236
 Mitochondrial dysfunction <0.0001 0.425
 Noradrenaline and adrenaline degradation 0.001 0.5
 Mitochondrial L-carnitine shuttle pathway 0.005 0.545
 LPS/IL-1–mediated inhibition of RXR function <0.0001 0.38
 cAMP-mediated signaling 0.003 0.279
 Fatty acid β-oxidation  I <0.0001 0.625
 Cardiac β-adrenergic signaling 0.001 0.319
 PXR/RXR activation 0.028 0.295
 Agranulocyte adhesion and diapedesis 0.019 0.261
 Melanocyte development and pigmentation signaling 0.005 0.316
 Glutaryl-CoA degradation <0.0001 0.9
 PPARα/RXRα activation <0.0001 0.316
 p38 MAPK signaling 0.011 0.295
 Sperm motility 0.011 0.295
 Oleate biosynthesis II (animals) 0.019 0.571
 Endothelin-1 signaling 0.028 0.25
 Myc-mediated apoptosis signaling 0.039 0.289
 Regulation of the epithelial–mesenchymal transition pathway 0.02 0.252
 Glutathione-mediated detoxification 0.017 0.412

Note: Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP, or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq. Genes that had ≥+1.5 or ≤−1.5 a fold change were uploaded in to IPA for analysis using the adipose tissue as the target organ. The pathways were sorted by number of molecules involved and the top 20 are shown.

Using IPA we also identified potential upstream regulators of the DEGs. As expected, PPARG was identified as a common top upstream regulator in an activated state for all treatments. In addition, SREBF1, another adipogenic transcriptional regulator, was found among the top upstream regulators. Accordingly, the insulin-induced gene (INSIG) 1 and 2, proteins known to be involved in the negative regulation of SREBF1 function, were predicted to be inhibited in all treatments. The membrane-bound transcription factor site-1 protease (MBTPS1) (also known as serine protease 1: S1P), a serine protease involved in SREBF1 activation, was predicted to be activated in all treatments as expected although not always in the top 20 (Inoue et al. 2001). By contrast, the transcription factor Krüppel-like factor (KLF) 15, known to be involved in adipocyte differentiation (Mori et al. 2005), was predicted to be active only in TPP and troglitazone treatments but not IPTP (Table 3). Furthermore, CEBPA, another important transcription factor in adipogenesis (Lane et al. 1996) was found as an upstream regulator in TPP and troglitazone treatments (not in the top 20), but not IPTP. Finally, lamin B1 (LMNB1), a matrix protein involved in cytoskeletal organization and adipogenesis (Verstraeten et al. 2011), was predicted to be inhibited by IPTP but not TPP or troglitazone treatment.

Table 3. Top 20 upstream regulators identified by IPA.
Upstream regulator Molecule type Predicted activation state Activation z-score p-Value
TPP
 PPARG Ligand-dependent nuclear receptor Activated 5.895 <0.0001
 SREBF1 Transcription regulator Activated 5.424 <0.0001
 SCAP Other Activated 4.618 <0.0001
 CEBPA Transcription regulator Activated 3.652 <0.0001
 NR1H3 Ligand-dependent nuclear receptor Activated 3.546 <0.0001
 PPARGC1A Transcription regulator Activated 3.465 <0.0001
 ATP7B Transporter Activated 3.464 <0.0001
 PPARGC1B Transcription regulator Activated 3.248 <0.0001
 KLF15 Transcription regulator Activated 2.905 <0.0001
 NR1H2 Ligand-dependent nuclear receptor Activated 2.865 <0.0001
 INSIG1 Other Inhibited −4.967 <0.0001
 POR Enzyme Inhibited −3.069 <0.0001
 ELOVL5 Enzyme Inhibited −2.976 <0.0001
 EPAS1 Transcription regulator Inhibited −2.969 <0.0001
 INSIG2 Other Inhibited −2.934 <0.0001
 ASXL1 Transcription regulator Inhibited −2.646 <0.0001
 TNF Cytokine Inhibited −2.620 <0.0001
 MKL1 Transcription regulator Inhibited −2.591 <0.0001
 LEP Growth factor Inhibited −2.480 <0.0001
 PML Transcription regulator Inhibited −2.462 <0.0001
IPTP
 SREBF1 Transcription regulator Activated 4.43 <0.0001
 PPARG Ligand-dependent nuclear receptor Activated 4.20 <0.0001
 SCAP Other Activated 4.05 <0.0001
 NR1H3 Ligand-dependent nuclear receptor Activated 3.70 <0.0001
 ATP7B Transporter Activated 3.32 <0.0001
 PPARGC1B Transcription regulator Activated 2.93 <0.0001
 MBTPS1 Peptidase Activated 2.43 <0.0001
 PPARD Ligand-dependent nuclear receptor Activated 2.38 <0.0001
 FAS Transmembrane receptor Activated 2.35 <0.0001
 NR1H2 Ligand-dependent nuclear receptor Activated 2.23 <0.0001
 INSIG1 Other Inhibited −4.64 <0.0001
 EPAS1 Transcription regulator Inhibited −3.08 <0.0001
 ELOVL5 Enzyme Inhibited −2.98 <0.0001
 MKL1 Transcription regulator Inhibited −2.94 <0.0001
 IKBKB Kinase Inhibited −2.84 <0.0001
 TGFB1 Growth factor Inhibited −2.83 <0.0001
 INSIG2 Other Inhibited −2.76 <0.0001
 POR Enzyme Inhibited −2.75 <0.0001
 LEP Growth factor Inhibited −2.48 <0.0001
 LMNB1 Other Inhibited −2.43 <0.0001
Troglitazone
 PPARG Ligand-dependent nuclear receptor Activated 6.706 <0.0001
 PPARA Ligand-dependent nuclear receptor Activated 6.302 <0.0001
 PPARGC1A Transcription regulator Activated 6.283 <0.0001
 SREBF1 Transcription regulator Activated 5.048 <0.0001
 SCAP Other Activated 4.942 <0.0001
 RB1 Transcription regulator Activated 4.767 <0.0001
 KLF15 Transcription regulator Activated 4.712 <0.0001
 INSR Kinase Activated 4.667 <0.0001
 PPARGC1B Transcription regulator Activated 4.223 <0.0001
 PNPLA2 Enzyme Activated 3.763 <0.0001
 RICTOR Other Inhibited −5.182 <0.0001
 INSIG1 Other Inhibited −5.099 <0.0001
 TNF Cytokine Inhibited −4.972 <0.0001
 KDM5A Transcription regulator Inhibited −4.494 <0.0001
 TWIST1 Transcription regulator Inhibited −3.656 <0.0001
 F2R G-protein–coupled receptor Inhibited −3.359 <0.0001
 TGFB1 Growth factor Inhibited −3.189 <0.0001
 INSIG2 Other Inhibited −3.088 <0.0001
 Aldosterone Chemical–endogenous mammalian Inhibited −3.084 <0.0001
 HSD17B4 Enzyme Inhibited −2.985 <0.0001

Note: Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP, or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq. Genes that had ≥+1.5 or ≤−1.5 a fold change were uploaded into IPA for analysis using the adipose tissue as the target organ. The upstream regulators were sorted by z-score and the top 20 are shown.

Using the common and unique DEGs to the three treatments identified by the Venn diagram (Figure 4) upstream regulators were identified in IPA. Table 4 shows the top upstream regulators using the common DEGs and Table 5 lists the upstream regulators based on the unique DEGs for TPP, IPTP, and troglitazone. As expected, PPARG was the top common upstream regulator for all treatments. Interestingly, based on the unique DEGs for TPP there were only few upstream regulators identified (Table 5A). IPTP had many more upstream regulators identified in IPA from which the top ones based on the number of molecules involved are being listed (Table 5B). As expected, troglitazone had the most upstream regulators identified in IPA based on a much larger number of unique DEGs, and the top 10 are listed in Table 5C.

Table 4. Top upstream regulators identified by IPA for TPP, IPTP, and troglitazone overlapping DEGs.
Upstream regulator Molecule type p-Value
TPP, IPTP, MIDT
 PPARG Ligand-dependent nuclear receptor <0.001
 NR1H2 Ligand-dependent nuclear receptor <0.001
 GHRL Growth factor <0.001
 Dihydrotestosterone Chemical–endogenous mammalian <0.001
 FGF21 Growth factor <0.001
 LEP Growth factor <0.001
 NR4A1 Ligand-dependent nuclear receptor <0.001
 TNF Cytokine <0.001

Note: Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq. Genes that had ≥+1.5 or ≤−1.5 a fold change were uploaded into Venny version 2.0 for analysis and overlapping DEGs were uploaded into IPA. Only pathways containing more than five molecules are shown.

Table 5. Top upstream regulators identified by IPA using unique DEGs to TPP, IPTP, and troglitazone.
Upstream regulator Molecule type p-Value
TPP
 DICER1 Enzyme <0.0001
 miR-7155-5p (miRNAs w/seed CUGGGGU) Mature microRNA <0.0001
 miR-4667-5p (and other miRNAs w/seed CUGGGGA) Mature microRNA <0.0001
IPTP
 beta-Estradiol Chemical–endogenous mammalian <0.002
 Lipopolysaccharide Chemical drug <0.002
 Dexamethasone Chemical drug <0.002
 TNF Cytokine <0.002
 TGFB1 Growth factor <0.002
 IFNG Cytokine <0.002
 ESR1 Ligand-dependent nuclear receptor <0.002
 TP53 Transcription regulator <0.002
 IL1B cytokine <0.002
Troglitazone
 TP53 Transcription regulator <0.0001
 ERBB2 Kinase <0.0001
 CCND1 Transcription regulator <0.0001
 TGFB1 Growth factor <0.0001
 RB1 Transcription regulator <0.0001
 E2F4 Transcription regulator <0.0001
 CDKN1A Kinase <0.0001
 NUPR1 Transcription regulator <0.0001
 calcitriol Chemical drug <0.0001
 Vegf Group <0.0001

Note: Human primary preadipocytes were differentiated in the presence of MID supplemented with 20 μM TPP, 20 μM IPTP or 5 μM troglitazone. At day 6 of differentiation, RNA was collected from five donors and used for RNA-seq. Genes that had ≥+1.5 or ≤−1.5 a fold change were uploaded into Venny version 2.0 for analysis and unique genes for TPP, IPTP, and troglitazone were uploaded into IPA. Data were sorted by number of molecules per pathway and only pathways containing more than three molecules for TPP and five molecules for IPTP and troglitazone are shown.

Discussion

We found that FM550, and its components TPP and IPTP, induce adipogenesis in human primary preadipocytes as demonstrated by lipid accumulation and expression of adipogenic markers. Further, we found that IPTP was able to increase lipid accumulation and FABP4 protein expression at lower concentrations than TPP. Our findings are in agreement with a previous study showing that treatment of murine pluripotent cells with the FM550 components TPP and IPTP diverted the cells to an adipogenic fate, as assessed by lipid accumulation and perilipin levels (Pillai et al. 2014). However, (Pillai et al. 2014) used an IPTP mixture containing 40% TPP and concluded that the main adipogenic component in FM550 was likely TPP. Here we show in human primary preadipocytes that an IPTP mixture containing only 18% TPP was more potent than pure TPP at inducing lipid accumulation and FABP4 expression.

The human preadipocyte differentiation model system allowed us to explore the mode of action of the chemicals of interest leading to adipogenesis and, potentially, obesity whether through PPARG activation or glucocorticoid pathways. Chemicals are tested for their ability to activate PPARG when replacing troglitazone in the differentiation medium and via the glucocorticoid pathway when chemicals are replacing dexamethasone. FM550 and its components, TPP and IPTP, are able to increase lipid accumulation in the presence of dexamethasone in human preadipocytes, although a positive trend, which did not reach statistical significance, was also observed in the presence of troglitazone for this end point in our study. It is very likely that statistical significance would have been achieved if more donors were used for the experiments. Using a murine model, others have shown that TPP and IPTP increased lipid accumulation; however, it is unclear which nuclear receptor these chemicals are targeting because murine cell models require either dexamethasone or troglitazone for differentiation (Gimble et al. 1990). Our lipid accumulation data suggest that FM550, TPP, and IPTP may have other effects in addition to PPARG activation. Further, we have shown that in the murine 3T3-L1 model, the lipogenesis (lipid accumulation) mediated by dexamethasone TPP and IPTP was not inhibited by the PPARG antagonist GW9662, whereas troglitazone-mediated lipogenesis was (Tung et al. 2017). Moreover, in our study the test chemicals were added in the presence of 5 μM troglitazone, a concentration by which PPARG is likely to be saturated (Nagai et al. 2011) and unlikely to be further activated, supporting that mechanisms beyond PPARG activation are at work for lipogenesis. Therefore one may hypothesize that the glucocorticoid receptor may be a target. However, TPP and IPTP were not able to activate the glucocorticoid receptor in luciferase reporter assays (data not shown). Of further note, Pillai et al. (2014) showed a modest increase in PPAR transactivation by TPP and IPTP relative to solvent control, and also showed that TPP was more potent than IPTP in mediating this effect. Therefore, these authors concluded that TPP was the adipogenic component in FM550 (Pillai et al. 2014). Interestingly, we observed a higher PPARG transactivation in IPTP treatments compared with TPP treatments in luciferase assays (data not shown). Of importance, our mixture of isopropylated compounds (IPTP) contained only 18% TPP by analysis, whereas Pillai et al. (2014) used a mixture containing 40%. Therefore, one may not conclude that TPP is the main PPARG activator in the FM550 mixture or that this is the only mechanism of action of the chemicals.

In addition to using a human model to assess the effects of FM550 components in inducing adipogenesis, we also performed global gene expression analysis to evaluate the mode of action of these chemicals. To the best of our knowledge, this study is the first to compare global transcriptomic changes in response to FM550 components in differentiating human preadipocyte or any other system. Interestingly, we found that IPTP treatment resulted in twice the number of affected DEGs as TPP. About half of the DEGs affected by IPTP were distinct (i.e., were not shared with either TPP or troglitazone). In contrast, most of the TPP-affected DEGs were in common with troglitazone treatment. This suggests that TPP acts mainly through PPARG activation, whereas IPTP perturbed a variety of regulatory pathways beyond just PPARG. Indeed, ∼84% of the DEGs induced by TPP treatment were in common with DEGs induced by the PPARG agonist troglitazone, as opposed to 64% DEGs in common between IPTP and troglitazone.

When we assessed the transcript levels of select DEGs by RT-qPCR, we found striking differences between the three treatments. The fold change in the expression of some DEGs was larger in the TPP treatment versus the IPTP treatment, and vice versa. For example, PCK1, PLIN4, and FABP5 were all increased to a greater extent by TPP. All these genes are involved in lipogenesis, an integral part of adipocyte differentiation (Ducharme and Bickel 2008). Interestingly, IPTP treatment induced higher expression levels of transporters, such as ABCG1 and SLCO4C1 than TPP. ABCG1 expression is positively correlated with triglyceride accumulation during adipogenesis by increasing fatty acid influx (Frisdal et al. 2015). Fatty acids are known to act as endogenous PPARG agonists (Krey et al. 1997), and therefore this may be a plausible mechanism by which these chemicals indirectly activate PPARG. SLCO4C1 is a member of the transporter superfamily mediating the transport of thyroid hormones, T3 and T4 (van der Deure et al. 2010). It has been shown that when preadipocytes are transduced with an inactive thyroid hormone receptor mutant a decrease in the expression of adipogenic markers is observed (Liu et al. 2015). This suggests that increased intracellular thyroid hormone levels, which may be mediated by SLCO4C1, enhance human adipogenesis.

The distinct effects observed in the TPP and IPTP treatments are also apparent in the canonical pathways enriched in IPA. For TPP treatment, many of the pathways were in common with the troglitazone treatment, as expected, although some were unique. Of those, protein kinase A signaling is relevant, as it is elevated early in adipogenesis (Klemm et al. 1998). During a standard differentiation protocol, IBMX is used to increase intracellular cAMP concentrations, which in turn activates PKA (Reusch et al. 2000). Because all treatments contained IBMX, a phosphodiesterase inhibitor, it is remarkable that only TPP further increased this process. Another relevant pathway that was enriched in TPP treatment was AMPK Signaling, which is involved in energy metabolism and fatty acid oxidation (Hardie et al. 2006). Some of the IPTP perturbed pathways included caveolar-mediated endocytosis signaling and clathrin-mediated endocytosis signaling, both of which are involved in insulin receptor and solute carrier family 2 (facilitated glucose transporter), member 4 (GLUT4) endocytosis, respectively (Fagerholm et al. 2009; Leto and Saltiel 2012). In addition, IPTP also affected the paxillin signaling pathway, which is involved in cytoskeletal remodeling, an important process in adipocyte differentiation (Kawaguchi et al. 2003; Parsons et al. 2012). This further indicates that although both chemicals may modulate adipogenesis, their modes of action may be different.

The top upstream regulators identified by IPA for all treatments are known regulators of the adipogenic process. With respect to these identified regulators, the mode of action of TPP appears to resemble the mode of action of troglitazone. In addition, the few upstream regulators that were identified in IPA using the unique DEGs to TPP did not appear to have any known function in adipogenesis. However, for IPTP dexamethasone was identified as one of the top upstream regulators using unique DEGs. Of note, CEBPA, an important transcription factor in adipogenesis was identified as an upstream regulator in TPP and troglitazone-treated cells (not in the top 10) but not in IPTP. In addition, CEBPA was not a DEG identified in the IPTP treatment; however, this may be due to the delayed increase in mRNA expression of CEBPA by IPTP treatment, which reached significance only at day 9 (Figure 3). This is not surprising considering that the mRNA used in the RNA-seq analysis was obtained at an earlier time point (day 6). This is one limitation of this study, where we chose one most potent concentration and one time point to generate the data due to prohibiting costs. Further, the concentration of FM550 and its components used in this study were in the μM range, whereas human exposure is estimated at the nM range. However, it is difficult to extrapolate in vitro doses to human exposure. First, human exposure is chronic and some of the chemicals may have lipophilic properties and may be accumulative in tissues, including the fat tissue. In fact, accumulation of FM550 components was detected in the blubber of some marine species, showing that these compounds may accumulate in adipose tissue (Lam et al. 2009). In addition, chemicals may adhere to the plastic or media components in vitro, and therefore the intracellular concentrations may be different than the ones applied. More knowledge is needed on the pharmacokinetics of FM550 and its components in humans and in cell culture in order to be able to model how the doses used in vitro correlate to the in vivo exposure. Future studies will focus on DEGs of interest to establish the mode of action of TPP and IPTP in more depth.

Conclusion

In conclusion, we demonstrated that FM550 and its components, TPP and IPTP, induced adipogenesis in human primary preadipocytes in vitro. Although others have concluded that TPP is the main adipogenic component in FM550 and an activator of PPARG in a murine model system, using the human model we show that IPTP is also adipogenic. Global gene expression profiles revealed that both components had a gene signature that supported an adipogenic end point. However, of the two, TPP seemed to have a mode of action more similar to that of a known PPARG agonist. Interestingly, IPTP exhibited a more distinct gene profile from the PPARG agonist. Taken together, we show that although both components of FM550 increase human adipocyte differentiation, they may also exert other effects via different mechanisms. Overall, the results of this study suggest that human exposure to FM550 may promote adverse metabolic effects and as such, further investigation into the mechanisms of action of these chemicals is needed.

Acknowledgments

The authors thank R. Farmahin and N. Chepelev for reviewing this manuscript. Grant funding was provided by the Health Canada Chemical Management Plan.

References

Ahmed S, Atlas E. 2016. Bisphenol S- and bisphenol A-induced adipogenesis of murine preadipocytes occurs through direct peroxisome proliferator-activated receptor gamma activation. Int J Obes (Lond) 40(10):1566–1573, PMID: 27273607, 10.1038/ijo.2016.95.

Belcher SM, Cookman CJ, Patisaul HB, Stapleton HM. 2014. In vitro assessment of human nuclear hormone receptor activity and cytotoxicity of the flame retardant mixture FM 550 and its triarylphosphate and brominated components. Toxicol Lett 228(2):93–102, PMID: 24786373, 10.1016/j.toxlet.2014.04.017.

Boucher JG, Ahmed S, Atlas E. 2016. Bisphenol S induces adipogenesis in primary human preadipocytes from female donors. Endocrinology 157(4):1397–1407, PMID: 27003841, 10.1210/en.2015-1872.

Boucher JG, Boudreau A, Atlas E. 2014a. Bisphenol A induces differentiation of human preadipocytes in the absence of glucocorticoid and is inhibited by an estrogen-receptor antagonist. Nutr Diabetes 4:e102, PMID: 24418828, 10.1038/nutd.2013.43.

Boucher JG, Husain M, Rowan-Carroll A, Williams A, Yauk CL, Atlas E. 2014b. Identification of mechanisms of action of bisphenol A-induced human preadipocyte differentiation by transcriptional profiling. Obesity (Silver Spring) 22(11):2333–2343, PMID: 25047013, 10.1002/oby.20848.

Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. 2013. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29(1):15–21, PMID: 23104886, 10.1093/bioinformatics/bts635.

Dodson RE, Perovich LJ, Covaci A, Van den Eede N, Ionas AC, Dirtu AC, et al. 2012. After the PBDE phase-out: a broad suite of flame retardants in repeat house dust samples from California. Environ Sci Technol 46(24):13056–13066, PMID: 23185960, 10.1021/es303879n.

Ducharme NA, Bickel PE. 2008. Lipid droplets in lipogenesis and lipolysis. Endocrinology 149(3):942–949, PMID: 18202123, 10.1210/en.2007-1713.

Fagerholm S, Ortegren U, Karlsson M, Ruishalme I, Strålfors P. 2009. Rapid insulin-dependent endocytosis of the insulin receptor by caveolae in primary adipocytes. PLoS One 4(6):e5985, PMID: 19543529, 10.1371/journal.pone.0005985.

Frisdal E, Le Lay S, Hooton H, Poupel L, Olivier M, Alili R, et al. 2015. Adipocyte ATP-binding cassette G1 promotes triglyceride storage, fat mass growth, and human obesity. Diabetes 64(3):840–855, PMID: 25249572, 10.2337/db14-0245.

Gimble JM, Dorheim MA, Cheng Q, Medina K, Wang CS, Jones R, et al. 1990. Adipogenesis in a murine bone marrow stromal cell line capable of supporting B lineage lymphocyte growth and proliferation: biochemical and molecular characterization. Eur J Immunol 20(2):379–387, PMID: 2178944, 10.1002/eji.1830200222.

Greenspan P, Mayer EP, Fowler SD. 1985. Nile red: a selective fluorescent stain for intracellular lipid droplets. J Cell Biol 100(3):965–973, PMID: 3972906.

Hardie DG, Hawley SA, Scott JW. 2006. AMP-activated protein kinase – development of the energy sensor concept. J Physiol 574(pt 1):7–15, PMID: 16644800, 10.1113/jphysiol.2006.108944.

Hoffman K, Fang M, Horman B, Patisaul HB, Garantziotis S, Birnbaum LS, et al. 2014. Urinary tetrabromobenzoic acid (TBBA) as a biomarker of exposure to the flame retardant mixture Firemaster® 550. Environ Health Perspect 122(9):963–969, PMID: 24823833, 10.1289/ehp.1308028.

Inoue J, Kumagai H, Terada T, Maeda M, Shimizu M, Sato R. 2001. Proteolytic activation of SREBPs during adipocyte differentiation. Biochem Biophys Res Commun 283(5):1157–1161, PMID: 11355894, 10.1006/bbrc.2001.4915.

Janderová L, McNeil M, Murrell AN, Mynatt RL, Smith SR. 2003. Human mesenchymal stem cells as an in vitro model for human adipogenesis. Obes Res 11(1):65–74, PMID: 12529487, 10.1038/oby.2003.11.

Kawaguchi N, Sundberg C, Kveiborg M, Moghadaszadeh B, Asmar M, Dietrich N, et al. 2003. ADAM12 induces actin cytoskeleton and extracellular matrix reorganization during early adipocyte differentiation by regulating β1 integrin function. J Cell Sci 116(pt 19):3893–3904, PMID: 12915587, 10.1242/jcs.00699.

Klemm DJ, Roesler WJ, Boras T, Colton LA, Felder K, Reusch JE. 1998. Insulin stimulates cAMP-response element binding protein activity in HepG2 and 3T3-L1 cell lines. J Biol Chem 273(2):917–923, 10.1074/jbc.273.2.917.

Krey G, Braissant O, L’Horset F, Kalkhoven E, Perroud M, Parker MG, et al. 1997. Fatty acids, eicosanoids, and hypolipidemic agents identified as ligands of peroxisome proliferator-activated receptors by coactivator-dependent receptor ligand assay. Mol Endocrinol 11(6):779–791, PMID: 9171241, 10.1210/mend.11.6.0007.

Lam JC, Lau RK, Murphy MB, Lam PK. 2009. Temporal trends of hexabromocyclododecanes (HBCDs) and polybrominated diphenyl ethers (PBDEs) and detection of two novel flame retardants in marine mammals from Hong Kong, South China. Environ Sci Technol 43(18):6944–6949, PMID: 19806725.

Lane MD, Lin FT, MacDougald OA, Vasseur-Cognet M. 1996. Control of adipocyte differentiation by CCAAT/enhancer binding protein alpha (C/EBP alpha). Int J Obes Relat Metab Disord 20 (suppl 3):S91–S96, PMID: 8680485.

Langmead B, Salzberg SL. 2012. Fast gapped-read alignment with Bowtie 2. Nat Methods 9(4):357–359, PMID: 22388286, 10.1038/nmeth.1923.

Leto D, Saltiel AR. 2012. Regulation of glucose transport by insulin: traffic control of GLUT4. Nat Rev Mol Cell Biol 13(6):383–396, PMID: 22617471, 10.1038/nrm3351.

Liu YY, Ayers S, Milanesi A, Teng X, Rabi S, Akiba Y, et al. 2015. Thyroid hormone receptor sumoylation is required for preadipocyte differentiation and proliferation. J Biol Chem 290(12):7402–7415, PMID: 25572392, 10.1074/jbc.M114.600312.

Mori T, Sakaue H, Iguchi H, Gomi H, Okada Y, Takashima Y, et al. 2005. Role of Krüppel-like factor 15 (KLF15) in transcriptional regulation of adipogenesis. J Biol Chem 280(13):12867–12875, PMID: 15664998, 10.1074/jbc.M410515200.

Nagai H, Ebisu S, Abe R, Goto T, Takahashi N, Hosaka T, et al. 2011. Development of a novel PPARγ ligand screening system using pinpoint fluorescence-probed protein. Biosci Biotechnol Biochem 75(2):337–341, PMID: 21307572, 10.1271/bbb.100810.

Parsons SA, Sharma R, Roccamatisi DL, Zhang H, Petri B, Kubes P, et al. 2012. Endothelial paxillin and focal adhesion kinase (FAK) play a critical role in neutrophil transmigration. Eur J Immunol 42(2):436–446, PMID: 22095445, 10.1002/eji.201041303.

Patisaul HB, Roberts SC, Mabrey N, McCaffrey KA, Gear RB, Braun J, et al. 2013. Accumulation and endocrine disrupting effects of the flame retardant mixture Firemaster® 550 in rats: an exploratory assessment. J Biochem Mol Toxicol 27(2):124–136, PMID: 23139171, 10.1002/jbt.21439.

Phillips AL, Chen A, Rock KD, Horman B, Patisaul HB, Stapleton HM. 2016. Editor’s highlight: transplacental and lactational transfer of Firemaster® 550 components in dosed Wistar rats. Toxicol Sci 153(2):246–257, PMID: 27370412, 10.1093/toxsci/kfw122.

Pillai HK, Fang M, Beglov D, Kozakov D, Vajda S, Stapleton HM, et al. 2014. Ligand binding and activation of PPARγ by Firemaster® 550: effects on adipogenesis and osteogenesis in vitro. Environ Health Perspect 122(11):1225–1232, PMID: 25062436, 10.1289/ehp.1408111.

Reusch JE, Colton LA, Klemm DJ. 2000. CREB activation induces adipogenesis in 3T3-L1 cells. Mol Cell Biol 20(3):1008–1020, PMID: 10629058.

Robinson MD, McCarthy DJ, Smyth GK. 2010. edgeR: a bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26(1):139–140, PMID: 19910308, 10.1093/bioinformatics/btp616.

Robinson MD, Oshlack A. 2010. A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biol 11(3):R25, PMID: 20196867, 10.1186/gb-2010-11-3-r25.

Stapleton HM, Allen JG, Kelly SM, Konstantinov A, Klosterhaus S, Watkins D, et al. 2008. Alternate and new brominated flame retardants detected in U.S. house dust. Environ Sci Technol 42(18):6910–6916, PMID: 18853808.

Tomlinson JJ, Boudreau A, Wu D, Abdou Salem H, Carrigan A, Gagnon A, et al. 2010. Insulin sensitization of human preadipocytes through glucocorticoid hormone induction of forkhead transcription factors. Mol Endocrinol 24(1):104–113, PMID: 19887648, 10.1210/me.2009-0091.

Tomlinson JJ, Boudreau A, Wu D, Atlas E, Haché RJ. 2006. Modulation of early human preadipocyte differentiation by glucocorticoids. Endocrinology 147(11):5284–5293, PMID: 16873539, 10.1210/en.2006-0267.

Tung EWY, Ahmed S, Peshdary V, Atlas E. 2017. Firemaster® 550 and its components isopropylated triphenyl phosphate and triphenyl phosphate enhance adipogenesis and transcriptional activity of peroxisome proliferator activated receptor (Pparγ) on the adipocyte protein 2 (aP2) promoter. PLoS One 12(4):e0175855, PMID: 28437481, 10.1371/journal.pone.0175855.

van der Deure WM, Peeters RP, Visser TJ. 2010. Molecular aspects of thyroid hormone transporters, including MCT8, MCT10, and OATPs, and the effects of genetic variation in these transporters. J Mol Endocrinol 44(1):1–11, PMID: 19541799, 10.1677/JME-09-0042.

Verstraeten VL, Renes J, Ramaekers FC, Kamps M, Kuijpers HJ, Verheyen F, et al. 2011. Reorganization of the nuclear lamina and cytoskeleton in adipogenesis. Histochem Cell Biol 135(3):251–261, PMID: 21350821, 10.1007/s00418-011-0792-4.

Yeh WC, Bierer BE, McKnight SL. 1995. Rapamycin inhibits clonal expansion and adipogenic differentiation of 3T3-L1 cells. Proc Natl Acad Sci U S A 92(24):11086–11090, PMID: 7479942.

Changes in Transportation-Related Air Pollution Exposures by Race-Ethnicity and Socioeconomic Status: Outdoor Nitrogen Dioxide in the United States in 2000 and 2010

Author Affiliations open

1Department of Civil, Environmental, and Geo-Engineering, University of Minnesota, Minneapolis, Minnesota, USA

2Department of Civil and Environmental Engineering, University of Washington, Seattle, Washington, USA

3Department of Soil, Water, and Climate, University of Minnesota, St. Paul, Minnesota, USA

PDF icon PDF Version (1.5 MB)

  • Background:
    Disparities in exposure to air pollution by race-ethnicity and by socioeconomic status have been documented in the United States, but the impacts of declining transportation-related air pollutant emissions on disparities in exposure have not been studied in detail.
    Objective:
    This study was designed to estimate changes over time (2000 to 2010) in disparities in exposure to outdoor concentrations of a transportation-related air pollutant, nitrogen dioxide (NO2), in the United States.
    Methods:
    We combined annual average NO2 concentration estimates from a temporal land use regression model with Census demographic data to estimate outdoor exposures by race-ethnicity, socioeconomic characteristics (income, age, education), and by location (region, state, county, urban area) for the contiguous United States in 2000 and 2010.
    Results:
    Estimated annual average NO2 concentrations decreased from 2000 to 2010 for all of the race-ethnicity and socioeconomic status groups, including a decrease from 17.6 ppb to 10.7 ppb (−6.9 ppb) in nonwhite [non-(white alone, non-Hispanic)] populations, and 12.6 ppb to 7.8 ppb (−4.7 ppb) in white (white alone, non-Hispanic) populations. In 2000 and 2010, disparities in NO2 concentrations were larger by race-ethnicity than by income. Although the national nonwhite–white mean NO2 concentration disparity decreased from a difference of 5.0 ppb in 2000 to 2.9 ppb in 2010, estimated mean NO2 concentrations remained 37% higher for nonwhites than whites in 2010 (40% higher in 2000), and nonwhites were 2.5 times more likely than whites to live in a block group with an average NO2 concentration above the WHO annual guideline in 2010 (3.0 times more likely in 2000).
    Conclusions:
    Findings suggest that absolute NO2 exposure disparities by race-ethnicity decreased from 2000 to 2010, but relative NO2 exposure disparities persisted, with higher NO2 concentrations for nonwhites than whites in 2010. https://doi.org/10.1289/EHP959
  • Received: 15 August 2016
    Revised: 07 June 2017
    Accepted: 09 June 2017
    Published: 14 September 2017

    Please address correspondence to J.D. Marshall, Dept. of Civil, Environmental, and Geo-Engineering, University of Washington, 201 More Hall, Seattle, WA 98195 USA. Telephone: (206) 685-2591. Email: jdmarsh@uw.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP959).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (1.2 MB)

    Zip icon Supplemental Code and Data Zip File (944 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (142 KB)

Introduction

Environmental injustice describes conditions in which more vulnerable communities experience disproportionate burdens of environmental health risks, such as exposure to air pollution. Environmental injustice in air pollution has been widely documented in the United States: many (>140) studies, covering a range of pollutants and U.S. locations, found higher air pollution exposures for lower-income groups and/or for race-ethnicity minority groups (Marshall et al. 2014). A key knowledge gap is whether environmental injustice has changed over time in the United States (Mohai and Saha 2015; Hajat et al. 2015). Longitudinal studies are needed to evaluate impacts of environmental policies on equity (Bento et al. 2015; Post et al. 2011), to explore the underlying causes of environmental injustice (Pastor et al. 2001), to enable tracking of environmental justice outcomes over time (Payne-Sturges and Gee 2006), and to test relationships between health disparities and exposure disparities over time (Mohai et al. 2009).

The goal of the present study was to estimate changes over time in environmental injustice in exposure to outdoor concentrations of a transportation-related air pollutant (TRAP) for the contiguous United States. Previous studies explored environmental injustice aspects of distributions of benefits (e.g., accessibility) and costs (e.g., noise) of transportation (Schweitzer and Valenzuela 2004). We focused on exposure to air pollution as a cost of transportation emissions that often differs by race-ethnicity and/or socioeconomic status in the United States. Racial minorities and low-income households are disproportionately likely to live near a major road [e.g., 27% of racial minorities vs. 19% of the total population lived near high traffic volume roads in the United States in 2010 (based on an analysis of national census and traffic data; Rowangould 2013)], where TRAP concentrations are typically highest (e.g., nitrogen dioxide concentrations were on average 2.9 times higher near major roads than urban background levels [based on a synthesis of monitoring studies in multiple cities; Karner et al. 2010]).

Previous U.S.-based longitudinal air pollution environmental justice studies have focused on exposure to industrial air pollution or proximity to polluting industrial facilities. Ard (2015) studied annual average concentrations of industrial air pollution nationwide during 1995–2004 and found that exposures decreased for all race-ethnicity groups over time, but African Americans remained more exposed than whites and Hispanics (by a factor of ∼50%). Longitudinal case studies on residential proximity to polluting industrial facilities [e.g., Seattle, 1990–2007 (Abel and White 2011); southern California, 1990–2000 (Hipp and Lakon 2010); in a national cohort, 1990–2007 (Pais et al. 2014)] found that race-ethnicity minority groups and/or lower socioeconomic status groups experienced closer average proximity to industrial facilities compared with other groups, and this pattern persisted over time. Few U.S.-based studies have explored temporal trends in environmental injustice for ambient air pollution or for transportation-related air pollution. Brajer and Hall (2005), studying ozone and coarse particulate matter in southern California during 1990–1999, found that on average, as air pollution decreased over time, Asians and Hispanics experienced larger reductions in ozone concentrations but smaller reductions in coarse particulate matter concentrations, compared with other groups. Kravitz-Wirtz et al. (2016), studying nitrogen dioxide and particulate matter exposures in the United States for a cohort of ∼9,000 families during 1990–2009, found that as exposures decreased over time, exposures remained higher for blacks and Hispanics than for whites.

We focused on nitrogen dioxide (NO2) as a TRAP. Transportation sources accounted for an estimated 60% of anthropogenic NOx emissions in the United States in 2010 (U.S. EPA 2016), and NO2 is an indicator of local transportation-related emissions (Brook et al. 2007; Burnett et al. 2004; Levy et al. 2014) with high within-urban spatial variability (Hewitt 1991; Apte et al. 2017). The U.S. EPA regulates outdoor annual NO2 as one of six criteria pollutants, in part because exposure to NO2 (together with other co-emitted TRAPs) is associated with health impacts, including low birth weight (Brauer et al. 2008), asthma in children (Takenoue et al. 2012), and cardiovascular mortality (Jerrett et al. 2013).

Air quality improved substantially in the United States after the 1990 Clean Air Act Amendments (Clean Air Act Amendments of 1990). From 2000 to 2010, estimated annual anthropogenic NOx emissions in the United States decreased by ∼50% (U.S. EPA 2016). It is unknown to what extent these estimated emission-reductions impacted NO2 exposure disparities by race-ethnicity and by socioeconomic status. To investigate, we combined NO2 air pollution data from a spatially precise (Census block scale) temporal land use regression model (Bechle et al. 2015) with Census demographic data (MPC 2011) and then estimated changes in TRAP environmental injustice over a decade (2000 to 2010) for the contiguous United States.

Methods

Study Area and Time Points

Analyses covered the contiguous United States (48 states plus the District of Columbia; selected based on availability of air pollution data) for two time points (selected based on availability of decennial Census demographic data): year 2000 (population: 280 million) and year 2010 (population: 306 million).

Datasets

Air pollution data.

Air pollution estimates were annual average NO2 concentrations for 2000 and 2010. These values were from a monthly land use regression (LUR) model incorporating satellite-based and ground-based observations (Bechle et al. 2015) for Census blocks [in 2010, n=8.2 million; mean area=0.97 km2 (total), 0.048 km2 (urban), 1.8 km2 (rural)].

Demographic data.

Demographic data were population estimates from the Census by race-ethnicity, socioeconomic status, language, and age. Demographic data included race (seven categories: white alone, black or African American alone, Asian alone, Native Hawaiian or other Pacific Islander alone, American Indian or Alaska Native alone, other race alone, two or more races), ethnicity (two categories: Hispanic, non-Hispanic), per capita income (continuous variable), household income [five categories (approximate annual household income quintiles): <$20,000, $20,000–$35,000, $35,000–$50,000, $50,000–$75,000, >$75,000], poverty (two categories: below poverty level, at or above poverty level), highest level of education for population >25 y (five categories: less than high school degree, high school degree, some college, college degree, graduate degree), employment for population >16 y (two categories: employed, unemployed), household language (five categories: English only, Spanish, other Indo-European language, Asian language, other language), household linguistic isolation [two categories: linguistically isolated (no one >14 y speaks English “well” or “very well”), not linguistically isolated], and age [four categories: younger children (<5 y), older children (5–18 y), younger adults (18–65 y), older adults (>65 y)]. Demographic data for 2000 were from the Decennial Census (2000 estimated populations for all demographic characteristics) and for 2010 were from the decennial Census (2010 estimated populations by race-ethnicity and by age) and the American Community Survey (2008–2012 five-year estimated populations for all other demographic characteristics not reported in the 2010 decennial Census) at the Census block group level [in 2010, n=210,000 (total); area mean (interquartile range)=36 km2 (0.49 km2–9.1 km2) (total); 1.1 km2 (0.34 km2–1.3 km2) (urban); 200 km2 (32 km2–150 km2) (rural)], the finest spatial scale for which detailed Census data are publicly available.

Spatial and Temporal Matching of Air Pollution and Demographic Data

To match the air pollution data (block level) with demographic data (block group level), we calculated population-weighted mean annual NO2 concentrations for all block centroids within each block group boundary, for 2000 and 2010. Boundaries for Census urban areas (defined based on population, population density, land cover, and other criteria; U.S. Census Bureau 2011) and boundaries for smaller Census geographies (blocks and block groups) changed during 2000 to 2010. For analyses comparing consistent block group boundaries over time, we applied the National Historic Geographic Information System time-series data: estimates of 2000 population counts and race-ethnicity within 2010 block group boundaries (MPC 2011). To match urban area data over time, we applied the 2010 urban area definitions to both 2000 and 2010 block groups, including block groups for which all blocks were inside the urban area boundary.

Urban and Rural Block Group Definitions

For urban versus rural comparisons, we applied the following definitions based on 2010 Census urban definitions: urban block groups contain only urban blocks (65%; n=140,000 in 2010), rural block groups contain only rural blocks (13%; n=28,000 in 2010), and mixed block groups contain both urban and rural blocks (22%; n=47,000 in 2010).

Exposure Assessment

Exposure assessment was based on residential block group LUR estimates of outdoor annual average NO2 concentrations.

Analyses Estimating Changes in NO2 Environmental Injustice over Time

We applied three related approaches to estimating changes in NO2 environmental injustice over time: a) we estimated and compared NO2 concentrations for populations defined by demographic characteristics (e.g., race-ethnicity groups); b) we estimated and compared NO2 concentrations for block groups (as proxies for “neighborhoods” or “local areas”) by demographic characteristics (e.g., per capita income); and c) we estimated and compared NO2 environmental injustice metrics on a national basis and for regions, states, counties, and urban areas.

Estimated changes in NO2 concentrations by demographic groups.

Our analyses by demographic groups focused on categories of race-ethnicity (14 groups), age (4 groups), household income (5 groups), and educational attainment (5 groups). We also performed analyses with race-ethnicity dichotomized as “white” or “nonwhite,” where the white population was defined as the race-ethnicity majority group (i.e., the “white alone, non-Hispanic” population; 69% of population in 2000, 64% in 2010), and the nonwhite population included all other race-ethnicity minority groups combined (i.e., the non-“white alone, non-Hispanic” population). In addition, we performed supplemental analyses of populations by household primary language and linguistic isolation (combined, 13 groups), employment status (unemployed, employed), and poverty (below or above poverty level). For all analyses by demographic groups, we conducted analyses for the total population, and separately for the urban and rural populations.

We estimated the population-weighted mean annual NO2 concentration C for each demographic group j in each year t (2000 or 2010) as
where ci is the annual mean NO2 concentration for block group i, pji is the population of demographic group j in block group i, and n is the total number of block groups.

To compare population-weighted mean NO2 concentrations between demographic groups j1 and j2 in year t (cross-sectional comparisons), we estimated absolute differences as Cj1Cj2, and relative percent differences as {100(Cj1Cj2)/[(Cj1+Cj2)/2]}. To compare population-weighted mean NO2 concentrations between 2000 (t1) and 2010 (t2) for group j (temporal comparisons), we estimated the absolute change as Ct2Ct1, and the relative percent change as {100(Ct2>−Ct1)/[(Ct2+Ct1)/2]}. Changes were calculated such that negative values indicate a decrease in NO2 concentration over time.

Estimated changes in NO2 concentrations by block group demographic characteristics.

To quantify differences by local (i.e., block group) demographic characteristics, we compared estimated mean NO2 concentrations in each year between block groups with proportions of nonwhite residents in the highest and lowest 5% of the distribution for all block groups in the United States in each year. We analyzed data for all block groups combined and separately for urban and rural block groups.

To explore block group differences in NO2 concentrations by race-ethnicity, income, and size of urban area, we categorized urban block groups by percent nonwhite in each year (quintiles), average per capita income in 2010 (20 equal groups), and total urban area population in 2010 (tertiles; large: 4.2 million to 18 million residents, n=8, total population=61 million; medium: 830,000 to 3.8 million residents, n=35, total population=63 million; and small: 14,000 to 800,000 residents, n=438, total population=61 million). We then compared estimated mean NO2 concentrations according to average per capita income (by approximate interquartile range in 2010 per capita income: $18,000 to $33,000) between urban block groups with percent nonwhite populations in the highest and lowest quintile of the national distribution for each year, after stratifying by small, medium, or large urban area population size.

Estimated changes in NO2 environmental injustice metrics.

To quantify how environmental injustice has changed over time on a national basis and for different U.S. geographies, we calculated and compared environmental injustice metrics in 2000 and 2010 on a national basis and by region, state, county, and urban area. Our core environmental injustice metric is the difference in estimated population-weighted mean NO2 concentration (Equation 1) for nonwhites versus whites [i.e., (population-weighted mean NO2 concentration for nonwhites) − (population-weighted mean NO2 concentration for whites)], hereafter referred to as the “nonwhite–white NO2 disparity.” As supplements to the nonwhite–white NO2 disparity, we calculated alternate environmental injustice metrics by race-ethnicity {for the three largest minority race-ethnicity groups: black–white NO2 disparity [difference in estimated population-weighted mean NO2 concentration for non-Hispanic blacks and non-Hispanic whites], Hispanic–white NO2 disparity [difference in estimated population-weighted mean NO2 concentration for Hispanics of any race(s) and non-Hispanic whites], and Asian–white NO2 disparity [difference in estimated population-weighted mean NO2 concentration for non-Hispanic Asians and non-Hispanic whites]} and by income (difference in estimated population-weighted mean NO2 concentration for the population with income below the poverty level and the population with income two times the poverty level). We calculated correlations (Pearson’s correlation coefficient; Spearman’s rank coefficient) among the changes in the alternate environmental injustice metrics for states, counties, and urban areas.

Potential Influence of Changes in NO2 Emissions and Changes in Demographic Patterns to Changes in Environmental Injustice over Time

As a preliminary step in understanding underlying mechanisms for changes over time in TRAP environmental injustice, we explored potential contributions of two factors: emission-reductions and residential demographic patterns. To estimate the potential extent to which each factor separately contributed to changes in NO2 environmental injustice, we considered two counterfactual scenarios with the following assumptions: a) NO2 concentrations changed as observed (from 2000 to 2010), but residential demographic patterns remained constant (at year-2000 values); and b) residential demographic patterns changed as observed (from 2000 to 2010), but NO2 concentrations remained constant (at year-2000 values). We then calculated the core national environmental injustice metric (nonwhite–white NO2 disparity) for each scenario. To estimate the contribution of changes in NO2 concentrations alone, we divided the predicted change in the national nonwhite–white NO2 disparity calculated under counterfactual scenario a by the observed change in the national nonwhite–white NO2 disparity. To estimate the contribution of changes in residential demographic patterns alone, we divided the change in national nonwhite–white NO2 disparity calculated under counterfactual scenario b by the observed change in the nonwhite–white NO2 disparity.

Potential Relevance of Changes in Environmental Injustice for Public Health

As a preliminary step to explore the potential health relevance of the observed gaps in NO2 exposures, we a) compared estimated exposures to health-based air quality guidelines and b) conducted an illustrative (“back-of-the-envelope”) health impact calculation. We compared the proportion of nonwhites versus whites living in block groups with NO2 concentrations above the WHO annual guideline [>40 μg/m3 (corresponds approximately to >21 ppb) NO2; WHO 2005] and below 50% of the WHO guideline (<11 ppb). [All block groups were below the U.S. EPA annual standard for NO2 (53 ppb) in 2000 and 2010.] We estimated potential health impacts for one outcome [ischemic heart disease (IHD) mortality, the most common cause of death in the United States (CDC 2015)] attributable to the difference in national mean NO2 concentration for nonwhites and whites in 2000 and 2010. We assumed the relative risk (RR) of IHD mortality associated with outdoor annual average NO2 concentration was 1.066 [95% confidence interval (CI): 1.015, 1.119] per 4.1 ppb NO2 (based on a cohort of 74,000 adults in California during 1982–2000; Jerrett et al. 2013). Relative risks (RR) for NO2 concentrations experienced by nonwhites and whites were calculated using: RR=exp (βC), where C is the population-weighted mean NO2 concentration (Equation 1), and β=ln(1.066)/(4.1 ppb)=0.0156 ppb−1. To obtain a simplified estimate that reflects only the estimated potential impact of changes in NO2 exposure over time experienced on average by nonwhites and whites (all else equal), our health risk calculations assumed that the underlying IHD mortality rate was constant over time [using the year-2011 estimate: 109 deaths per 100,000 (CDC 2012), although IHD mortality rates decreased during this time period in the United States (Finegold et al. 2013; WHO 2016)], and that the underlying mortality rate was the same for nonwhites and whites and the same by U.S. location [although IHD mortality rates differed by race-ethnicity and by U.S. location during this time period (CDC 2016)].

Sensitivity Analyses on Uncertainty in NO2 LUR Model Estimates

To assess the potential impact of exposure misclassification on our findings, we tested whether NO2 LUR model residuals showed systematic bias with respect to demographic characteristics. We compared annual average NO2 concentrations based on measurements from 366 U.S. EPA monitors in 2006 (the base year for the temporal LUR model; Bechle et al. 2015) with the LUR-based estimates for each block group in which a monitor was located. We then compared the distributions of the LUR model residuals (i.e., the measured – predicted values) among block groups categorized by tertiles of percent nonwhite residents and tertiles of average per capita income in 2010. In addition, we compared the nonwhite–white NO2 disparity (core environmental injustice metric) based on U.S. EPA monitor data versus LUR model estimates for the 366 block groups with U.S. EPA monitors.

Results

Estimated Changes in NO2 Concentrations by Demographic Groups

Consistent with national trends, outdoor annual average NO2 concentrations decreased substantially across all race-ethnicity, income, education, and age groups during 2000 to 2010. Overall, on a national basis, the estimated population-weighted mean NO2 concentration decreased from 14.1 ppb in 2000 to 8.9 ppb in 2010, an absolute change of −5.2 ppb and a relative change of −37% (Table 1). Estimated changes among groups defined by race-ethnicity, income, age, and education ranged from −3.5 ppb to −8.6 ppb (−33% to −42%).

Table 1. Estimated NO2 population-weighted mean concentration (ppb) for year 2000, year 2010, and estimated change over time (year 2010–year 2000), by race-ethnicity, household income quintile, educational attainment, and age.
Demographic characteristic Population (%) Mean NO2 concentration (ppb) Change in mean NO2 concentration:
Absolute (ppb) Relative (%)
2000 2010 2000 2010 2010–2000 2010–2000
Total 100 100 14.1 8.9 −5.2 −37
Race-ethnicity
Non-Hispanic 87 84 13.4 8.4 −5.0 −37
 White alone 69 64 12.6 7.8 −4.7 −38
 Black or African American alone 12 12 16.2 10.0 −6.1 −38
 American Indian or Native American alone 0.7 0.7 10.1 6.6 −3.5 −35
 Asian alone 3.4 4.5 20.2 12.1 −8.1 −40
 Native Hawaiian or other Pacific Islander alone 0.1 0.1 17.7 10.6 −7.1 −40
 Other race alone 0.2 0.2 17.9 10.8 −7.1 −40
 Two or more races 1.6 1.8 16.1 9.3 −6.8 −42
Hispanic 13 16 18.9 11.2 −7.7 −41
 White alone 6.0 8.7 17.6 10.6 −7.0 −40
 Black or African American alone 0.3 0.4 20.8 12.2 −8.6 −41
 American Indian or Native American alone 0.1 0.2 18.8 11.2 −7.6 −41
 Asian alone 0.04 0.1 19.3 11.8 −7.5 −39
 Native Hawaiian or other Pacific Islander alone 0.01 0.02 18.4 10.8 −7.6 −41
 Other race alone 5.3 6.0 20.2 12.0 −8.2 −41
 Two or more races 0.8 1.0 19.3 11.3 −8.0 −41
Household income quintilea
 <$20,000 8.3 6.7 14.2 9.0 −5.2 −36
 $20,000–$35,000 7.3 5.9 13.7 8.7 −5.0 −37
 $35,000–$50,000 6.2 5.1 13.7 8.6 −5.0 −37
 $50,000–$75,000 7.3 6.8 13.8 8.6 −5.2 −38
 >$75,000 8.4 13 14.6 9.0 −5.7 −39
Educational attainmentb
 <High school degree 13 19 14.9 9.3 −5.6 −37
 High school degree 19 10 13.2 8.8 −4.4 −33
 Some college 18 12 13.7 8.9 −4.9 −35
 College degree 10 5.5 14.6 9.3 −5.3 −36
 Graduate degree 5.7 6.2 14.9 9.3 −5.6 −38
Age (y)
 <5 6.8 6.5 14.4 9.0 −5.4 −38
 5–17 19 17 14.0 8.8 −5.2 −37
 18–65 62 63 14.2 9.0 −5.2 −37
 >65 12 13 13.7 8.4 −5.3 −38

aHousehold income quintiles are based on year-2000 population and income data. Income is reported for householders (38% of the total population in year 2000).

bEducational attainment data is reported for population >25 y (65% of the total population in year 2000).

In general, the groups with the highest estimated NO2 exposures in 2000 experienced the largest reductions in NO2 concentrations from year 2000 to year 2010 (see Figures S1 and S2). As an example, the Hispanic black group, the group with the highest estimated mean NO2 exposure in 2000 [20.8 ppb; 6.6 ppb (38%) higher than the national mean] experienced the largest estimated reduction in NO2 exposure from 2000 to 2010 [−8.6 ppb, a 3.3 ppb (48%) greater concentration reduction than the national mean reduction].

In 2000 and 2010, disparities in estimated mean NO2 concentrations were larger by race-ethnicity group than by income, education, or age group (Table 1). For example, in 2000, mean NO2 concentrations for race-ethnicity groups ranged from 10.1 ppb (non-Hispanic American Indian group) to 20.8 ppb (black Hispanic group), a maximum difference of 10.7 ppb, compared with maximum differences of 1.7 ppb, 0.9 ppb, and 0.7 ppb between the education, income, and age groups with the highest and lowest mean exposures, respectively. In 2010, mean NO2 concentrations for race-ethnicity groups ranged from 6.6 ppb to 12.2 ppb (a maximum difference of 6.5 ppb), whereas mean values for all individual education, income, and age subgroups were within 1.0 ppb of the national average.

On a national basis, rankings (most to least exposed groups) remained fairly consistent over time (Figure 1). For the six largest race-ethnicity groups, rank-order by estimated population-weighted mean NO2 concentration remained constant with time: the non-Hispanic Asian group was most exposed and the non-Hispanic American Indian group was least exposed over time. Differences by age, income, and education were small compared with differences by race-ethnicity in both time periods.

Box-and-whisker plots for race-ethnicity indicate nitrogen dioxide concentration for Asian, non-Hispanic; Hispanic, any race; black, non-Hispanic; two or more races, non-Hispanic; white, non-Hispanic; and American Indian, non-Hispanic groups. The plots for household income quintile indicate nitrogen dioxide concentration for less than 20,000 dollars, 20,000 to 35,000 dollars, 35,000 to 50,000 dollars, 50,000 to 75,000 dollars, and greater than 75,000 dollars groups. The plots for educational attainment indicate nitrogen dioxide concentration for less than high school degree, high school degree, some college, college degree, and graduate degree groups. The plots for age indicate nitrogen dioxide concentration for less than 5-y, 5- to 17-y, 18- to 65-y, and greater than 65-y groups.
Figure 1. Estimated NO2 concentration (ppb) by race-ethnicity, household income quintile, educational attainment, and age group, for year 2000 and year 2010. Box-and-whiskers indicate the 90th, 75th, 50th, 25th, and 10th percentile concentrations, and circles indicate population-weighted mean concentration. Race-ethnicity groups shown above are the six largest groups (Table 1 includes remaining race-ethnicity groups). Income groups are quintiles on a national basis for year-2000 households (38% of total population in year 2000). Educational attainment is reported for population over 25 y (65% of total population in year 2000).

After controlling for urban versus rural location (see Figures S3 and S4, Table S1), disparities in NO2 concentrations by race-ethnicity persisted (with higher concentrations and higher disparities in urban than in rural locations), with some differences in exposure patterns for demographic groups by urban versus rural location in each year. For example, estimated population-weighted mean NO2 concentrations were lower for non-Hispanic American Indians than non-Hispanic whites in rural locations (−1.3 ppb in 2000; −0.5 in 2010) but higher in urban locations (+0.2 ppb 2000; +0.1 ppb in 2010).

Results for supplemental measures of socioeconomic status (poverty, employment) and language (see Table S2) were generally consistent with the core demographic characteristics (race-ethnicity, income, education, and age). NO2 concentrations were higher for people below the poverty level than above the poverty level, for households with a language other than English than households with only English, and for linguistically isolated than nonlinguistically isolated households. NO2 concentrations were higher for employed than for unemployed populations.

Estimated Changes in NO2 Concentrations by Block Group Demographic Characteristics

Consistent with population-based results, block groups with a higher proportion of race-ethnicity minority residents tended to have higher concentrations of NO2, and this pattern was consistent over time (Figure 2). In 2000, the 5% of block groups with the highest proportion of nonwhite residents had 2.5 times higher [+13.2 ppb (22.1 ppb vs. 8.9 ppb)] estimated mean NO2 concentrations than the 5% of block groups with the lowest proportion of nonwhite residents; in 2010, the 2.5-fold gap had increased slightly, to 2.7-fold [+8.9 ppb (14.1 ppb vs. 5.2 ppb)]. Considering urban versus rural block groups separately (see Figure S5), urban results were consistent with national results [the 5% of urban block groups with the highest versus lowest proportion of nonwhite residents had 1.8 times higher [+10.3 ppb (23.6 ppb vs. 13.3 ppb)] mean NO2 concentration in 2000 and 1.8 times higher [+6.9 ppb (15.0 ppb vs. 8.1 ppb)] mean NO2 concentration in 2010), whereas rural results had the reverse pattern to a minor extent: NO2 concentrations were lower in block groups with a higher proportion of nonwhite residents (the 5% of rural block groups with the highest vs. lowest proportion of nonwhite residents had 0.7 times lower [−1.9 ppb (5.4 ppb vs. 7.3 ppb)] mean NO2 concentrations 2000] and 0.8 times lower [−0.9 (3.8 ppb vs. 4.6 ppb)] mean NO2 concentration in 2010).

Scatter plot showing estimated mean nitrogen dioxide concentration in parts per billion (y-axis) across population of nonwhite in percentage (x-axis) for the years 2000 and 2010.
Figure 2. Estimated mean NO2 concentration versus percent nonwhite population for block groups in year 2000 and year 2010. Each point represents the mean NO2 concentration for 1% of the 210,000 block groups in the United States, binned by percent nonwhite residents. (The first point represents the 1% of block groups with the lowest percent nonwhite population, and the last point represents the 1% of block groups with the highest percent nonwhite population.)

In urban areas, disparities in block group estimated mean NO2 concentrations by race-ethnicity (for nonwhites vs. whites) persisted over time, regardless of average block group per capita income or the size of the urban area (large, medium, or small), and were generally larger than disparities by income (see Figure S6). For example, in large urban areas in 2010, estimated mean NO2 concentrations were 3.0 ppb higher (16.8 ppb vs. 13.8 ppb) for block groups with the highest versus lowest quintile percent nonwhite residents at the 25th percentile income ($18,000) and 4.2 ppb higher (16.4 ppb vs. 12.2 ppb) for block groups with the highest versus lowest quintile nonwhite residents at the 75th percentile income ($33,000). Estimated mean NO2 concentrations were 1.6 ppb higher (13.8 ppb vs. 12.2 ppb) for the block groups at the 25th percentile income than at the 75th percentile income among lowest quintile percent nonwhite block groups, and 0.4 ppb (16.8 ppb vs. 16.4 ppb) higher among the highest quintile percent nonwhite block groups. In large urban areas, in 2000, the estimated mean NO2 concentration was 2.9 ppb higher for highest income category block groups with the highest quintile nonwhite residents (mean per capita income: $74,000; mean percent nonwhite residents: 88%; mean NO2: 25.4 ppb; population: 56,000) than the lowest income block groups with the lowest quintile nonwhite residents (mean per capita income: $6,400; mean percent nonwhite residents: 2.9%; mean NO2: 22.4 ppb; population: 14,000), and in 2010, 1.2 ppb higher (16.7 ppb vs. 15.5 ppb).

Estimated Changes in NO2 Environmental Injustice Metrics

Nationally, on an absolute basis, environmental injustice declined from 2000 to 2010. The nonwhite–white NO2 disparity decreased from 5.0 ppb in 2000 to 2.9 ppb in 2010 (−2.1 ppb [−42%]; Table 2). However, nationally, on a relative basis, environmental injustice persisted. Nonwhites remained more exposed to outdoor NO2 air pollution than whites on average in 2010, and there was little change in the relative NO2 difference between nonwhites and whites between 2000 and 2010: The nonwhite–white NO2 difference was 33% in 2000 (nonwhites were 40% more exposed than whites) and 31% in 2010 (nonwhites were 37% more exposed than whites).

Table 2. Estimated population-weighted mean NO2 concentrations (ppb) for nonwhites and whites: year 2000, year 2010, and change over time (year 2010–year 2000).
Race-ethnicity 2000 2010 Change: 2010–2000
Nonwhitesa 17.6 10.7 −6.9 (−39%)
Whitesb 12.6 7.8 −4.7 (−38%)
Difference: nonwhites–whites 5.0 (33%) 2.9 (31%) −2.1 (−42%)

aNonwhites includes all race-ethnicity minority groups (i.e., people who reported any race-ethnicity other than white alone, non-Hispanic).

bWhites includes people who reported white alone, non-Hispanic race-ethnicity.

Environmental injustice declined in most, but not all, locations. In all regions and in most (>75%) states, counties, and urban areas, the nonwhite–white NO2 disparity decreased over time (Figure 3). The nonwhite–white NO2 disparity decreased by >1 ppb in 16 urban areas (accounting for 32% of the urban area population; 49 million in year 2000), including Detroit (Michigan), Los Angeles (California), New Orleans (Louisiana), and Chicago (Illinois). The nonwhite–white NO2 disparity increased by >1 ppb in two urban areas (accounting for <1% of the urban population): Watertown (New York) and Delano (California): both are urban areas for which mean NO2 concentrations were higher for whites than nonwhites in 2000, and for which concentrations decreased to a greater extent for whites than for nonwhites during 2000 to 2010. Similar patterns hold among counties: the nonwhite–white NO2 disparity decreased by >1 ppb in 75 counties (accounting for 16% of the population in 2000), and increased by >1 ppb in 6 counties (accounting for <0.1% of the population in 2000), for all of which NO2 concentrations were higher for whites than for nonwhites in 2000.

Maps of USA stratifying the regions, states, counties, and urban areas according to the difference in mean nitrogen dioxide concentration in parts per billion as follows: less than negative 3.0, negative 3.0 to negative 2.0, negative 2.0 to negative 1.0, negative 1.0 to negative 0.5, negative 0.5 to negative 0.1, negative 0.1 to 0.1, 0.1 to 0.5, 0.5 to 1.0, 1.0 to 2.0, 2.0 to 3.0, and greater than 3.0.
Figure 3. Estimated environmental injustice metric (absolute difference in population-weighted mean NO2 concentration (ppb) between nonwhites and whites) (a) in year 2000, (b) in year 2010, and, (c) change over time (year 2010–year 2000) for United States (1) regions (n=9), (2) states (n=49 [including District of Columbia]), (3) counties (n=3,109), and (4) urban areas (n=481). For maps in columns (a) and (b), red indicates that annual mean NO2 concentrations are higher for nonwhites than whites, blue indicates that annual mean NO2 concentrations are higher for whites than nonwhites, and white indicates that annual mean NO2 concentrations are equal for nonwhites and whites. For maps in column (c), red indicates that the absolute difference in annual mean NO2 concentration between nonwhites and whites increased over time, blue indicates that the absolute difference decreased over time, and white indicates no change in the absolute difference over time. For maps in row (4), circle icons are located at the centroid of the urban area. For all plots, the box-and-whiskers indicate 90th, 75th, 50th, 25th, and 10th percentiles, and circles indicate maximum and minimum. Map boundary data are from the National Historical Geographic Information System (MPC 2011).

The alternate environmental injustice metrics considered (see Figures S7–S10) were moderately correlated (see Tables S3–S5). For example, for urban areas, changes in alternate environmental injustice metrics were moderately correlated (Pearson’s correlation coefficient, r, range: 0.3–0.8; Spearman’s rank coefficient, s, range: 0.2–0.9). New York and California had large reductions (high decile reductions) in all five environmental injustice metrics, and North Dakota had increases (low decile reductions) in all five environmental injustice metrics. Similar to the patterns for the nonwhite–white NO2 disparity, the black–white, Hispanic–white, and Asian–white NO2 disparity decreased in most (>75%) regions, states, counties, and urban areas from 2000 to 2010. In contrast, the poverty-based NO2 disparity increased in nearly half of states and counties, although in general, the poverty-based NO2 disparities were smaller than the race-based NO2 disparity metrics (e.g., among states the mean change in the poverty-based NO2 disparity was −0.2 ppb vs. −1.0 ppb for the Asian–white NO2 disparity). Estimated population-weighted mean NO2 concentrations and environmental injustice metrics for each region, state, county, and urban area included in our analyses are available in Supplemental Material (Excel Tables A-D).

Potential Influence of Changes in NO2 Emissions and Changes in Demographic Patterns to Changes in Environmental Injustice over Time

When we estimated what population-weighted mean NO2 concentrations in 2010 would have been if residential demographic patterns changed as observed but NO2 concentrations were fixed as in 2000, we predicted a decrease in mean NO2 exposure for nonwhites from 17.6 ppb to 16.6 ppb (−1.0 ppb) and for whites from 12.6 ppb to 12.1 ppb in whites (−0.5 ppb), for a change of −0.6 ppb in the nonwhite–white NO2 disparity over time (5.0 ppb in 2000, 4.5 ppb in 2010), in contrast with the estimated change of −2.1 ppb in the nonwhite–white NO2 disparity (Table 2). When we estimated what population-weighted mean NO2 concentrations in 2010 would have been if residential demographic patterns were fixed as in 2000 but NO2 concentrations decreased as observed, we predicted a decrease in mean NO2 exposure for nonwhites to 11.4 ppb (−6.3 ppb) and for whites to 8.1 ppb (−4.5 ppb), for a change of −1.8 ppb in the nonwhite-white NO2 disparity over time (5.0 ppb in 2000, 3.3 ppb in 2010). This analysis of counterfactual scenarios suggests that both changes in NO2 and changes in residential demographic patterns contributed to the observed reductions in the national nonwhite-white NO2 disparity, with changes in NO2 contributing to a larger extent (83%, i.e., −1.8 ppb of the −2.1 ppb observed change in environmental injustice metric) than changes in residential demographic patterns (26%, i.e., −0.6 ppb of the −2.1 ppb observed change in environmental injustice metric). [The individual contributions of these two factors sum to greater than 100%, indicating interaction effects (9%) due to air pollution and population changing together.]

Potential Relevance of Changes in Environmental Injustice for Public Health

In 2000 and in 2010, nonwhites were more likely than whites to live in block groups with NO2 concentrations above international health-based guidelines. In 2000, 30% of nonwhites and 10% of whites lived in block groups with NO2 concentrations above the WHO annual guideline (>21 ppb), compared with 5% of nonwhites and 2% of whites in 2010 (see Figures S11–S12, Table S6). Thus, nonwhites were three times as likely as whites to live in a block group above the WHO guideline in 2000, and 2.5 times as likely in 2010. Conversely, 23% of nonwhites and 44% of whites lived in block groups with NO2 concentrations below 50% of the WHO guideline (<11 ppb) in 2000, compared with 56% of nonwhites and 80% of whites in 2010. Thus, nonwhites were 0.5 and 0.7 times as likely as whites to live in a block group with population-weighted mean NO2 concentrations <50% of the WHO guideline in 2000 and 2010, respectively. For the urban population in 2000 and 2010, nonwhites were 2.1 times and 3.5 times as likely, respectively, to live in a block group with mean NO2 concentrations above the WHO guideline. Most of the rural population (95% of whites and 97% of nonwhites in 2000; 99% of whites and nonwhites in 2010) lived in blocks groups with NO2 concentrations below 50% of the WHO guidelines.

Based on the simplified health impact calculation, the estimated mean NO2 concentration burden for nonwhites relative to whites (5.0 ppb in 2000, 2.9 ppb in 2010) was associated with an estimated ∼7,000 (95% CI: 2000, 10,000) additional premature IHD deaths for nonwhites in the United States in 2000 and an estimated ∼5,000 (95% CI: 1,000, 9,000) in 2010 (calculations presented in Table S7). Thus, the reduction in the mean nonwhite–white NO2 disparity (−2.1 ppb between 2000 and 2010) was associated with preventing an estimated ∼2,000 (95% CI: 400, 3,000) premature IHD deaths per year among nonwhites. The purpose of this simplified (back-of-the-envelope) calculation was to provide background and context for concentration disparities reported here. This health impact calculation was limited by several important simplifying assumptions and considerations [i.e., this calculation assumed that the U.S. population breathed the national mean NO2 concentration, considered only one health impact (IHD mortality), assumed that the IHD mortality rate is constant over time and by race-ethnicity and U.S. location, and did not adjust for differences in age by race-ethnicity or over time]. This simplified health impact calculation suggests that the estimated nonwhite–white NO2 disparity may have been associated with potentially large health impacts (i.e., thousands of IHD deaths per year in the United States); more detailed analyses are needed to fully investigate the implications of NO2 disparities for public health.

Sensitivity Analyses on Uncertainty in NO2 LUR Model Estimates

When we compared LUR model-based NO2 estimates for the 366 block groups with U.S. EPA monitors to the monitor-based NO2 observations, median model-based NO2 concentrations were lower for block groups in the middle and highest tertiles of percent nonwhite residents, and higher for block groups in the lowest tertile of percent nonwhite residents (see Figure S13). Median model-based estimates were also higher than monitor-based estimates for block groups in the highest tertile of average per capita income. When we estimated the nonwhite–white NO2 disparity for these block groups in 2006 (the year for which monitor data were available; 670,000 people, 48% nonwhite) the disparity was larger when based on monitor data (3.3 ppb; 13.4 ppb vs. 10.1 ppb for nonwhites and whites, respectively) than LUR model predictions (2.3 ppb; 12.2 ppb vs. 9.8 ppb for nonwhites and whites, respectively). These findings suggest that our model-based results may under-estimate disparities in exposures.

Discussion

Estimated average NO2 concentrations decreased for almost all U.S. populations and locations from 2000 to 2010. Disparities in average NO2 concentrations by race-ethnicity decreased on an absolute basis (e.g., the nonwhite–white difference decreased from 5.0 ppb in 2000 to 2.9 ppb in 2010). However, despite these improvements, estimated average annual concentrations continued to be higher for nonwhite populations than for white populations in 2010 (nonwhite–white difference: 31% in 2010, 33% in 2000). In 2010, the estimated average concentration in the 5% of block groups with the highest proportion of nonwhite residents was 2.7 times higher than in the 5% of block groups with the lowest proportion of nonwhite residents (2.5 times higher in 2000). Therefore, our findings suggest that over time, NO2 concentrations decreased; disparities by race-ethnicity decreased on an absolute basis but on a relative basis have persisted.

Our finding that, on a relative basis, NO2 air pollution disparities by race-ethnicity persisted in the United States over time is consistent with a recent U.S. cohort study that reported that estimated NO2 concentrations during 1990 to 2009 were ∼10% higher for blacks and Hispanics than whites, even after controlling for individual socioeconomic characteristics (income, employment, home ownership) and metropolitan area characteristics (residential segregation, industry) (Kravitz-Wirtz et al. 2016). Our findings are also consistent with a national study of industry-related air pollution that reported that, although estimated exposures to industrial hazardous air pollutants (HAPs) decreased in the United States during 1994–2005, HAPs exposures remained ∼1.5 times higher for African Americans than whites (Ard 2015); in our study, NO2 exposures remained ∼1.3 times higher for African Americans than whites.

Our findings suggested that most of the reduction in nonwhite–white NO2 disparities between 2000 and 2010 was attributable to overall reductions in outdoor NO2 concentrations. Emissions-reductions were achieved in part via emission-control technology in motor vehicles (particularly in gasoline vehicles during this time period; McDonald et al. 2012) and stationary sources (e.g., power plants) (U.S. EPA 2016). In addition, U.S. metropolitan regions became more suburban, and suburban areas became more racially diverse during 2000 to 2010 (Howell and Timberlake 2014). Shifts in demographic residential patterns leading to larger proportions of race-ethnicity minorities in suburban locations (where TRAP concentrations are typically lower compared with central cities or downtown locations) also may have contributed to reductions in NO2 disparities by race-ethnicity during this time period.

Our evidence of larger NO2 disparities by race-ethnicity than by income is consistent with previous studies of environmental injustice in TRAP (e.g., Clark et al. 2014) and with persistent patterns of residential segregation in U.S. metropolitan regions, which remain more segregated by race than by income (Reardon et al. 2015). Additional work is needed to further investigate potential underlying causes (e.g., changes in patterns of residential segregation) of changes in environmental injustice in exposure to TRAP over time.

Although absolute NO2 exposure disparities reduced substantially during this period, there remain potentially large public health benefits from eliminating these disparities: nonwhites remained 2.5 times more likely than whites to live in block groups above WHO guidelines for NO2 in 2010, and based on the back-of-the-envelope calculation described above, the estimated nonwhite–white NO2 disparity may have been associated with thousands of premature IHD deaths among nonwhites in 2010.

Our analyses have several important limitations. Due to limitations in the spatial resolution of the Census data, we were unable to explore spatial patterns in air pollution and demographics at spatial scales finer than Census block groups. We focused on outdoor air pollution exposures, and we were unable to explore the potential influence of time-activity patterns for which air pollution exposure gradients by race-ethnicity and socioeconomic status may exist, including exposures during commuting, at work, or indoors (O’Neill et al. 2003). In addition, we evaluated only one pollutant at only two time points. Spatial patterns may differ for other TRAPs or for cumulative exposures to multiple pollutants. We also did not account for joint effects (interactions) of race-ethnicity and socioeconomic characteristics. Finally, our estimates were limited by uncertainties in the NO2 LUR model estimates and Census data. The impact of uncertainties in the Census data, particularly for national race-ethnicity data that represent an almost complete sample of ∼300 million people, is likely to be small relative to the potential impact of uncertainties in NO2 LUR model estimates. Findings from a sensitivity analysis comparing results when NO2 exposure estimates were based on U.S. EPA monitor data instead of our LUR model suggested that exposure misclassification may have varied in a way that would have caused us to underestimate true disparities by race-ethnicity in outdoor NO2 concentrations in the United States. However, we were unable to directly test the potential consequences of exposure misclassification on our national-scale estimates of environmental injustice.

Conclusion

During 2000 to 2010, estimated annual average exposures to outdoor NO2 air pollution declined across all race-ethnicity and socioeconomic groups [range of mean change: −33% to −42% (−3.5 ppb to −8.6 ppb)]. The most exposed groups in 2000 experienced, on average, the largest reductions in NO2 during 2000 to 2010. Disparities in NO2 exposure were larger by race-ethnicity than by other demographic characteristics (income, education, age) in 2000 and 2010, with higher exposures for race-ethnicity minorities. The estimated national mean nonwhite–white NO2 disparity decreased from 5.0 ppb in 2000 to 2.9 ppb in 2010. Most of this reduction in the national mean nonwhite–white NO2 disparity over time is attributable to reductions in outdoor NO2 concentrations, suggesting that existing efforts to reduce TRAP are also reducing TRAP exposure disparities by race-ethnicity over time. Despite these improvements in absolute exposures, relative exposure disparities persisted, with nonwhites remaining exposed to 37% more NO2 than whites on average in 2010, and 2.5 times more likely than whites to live in a block group with NO2 concentration above WHO guidelines in 2010. Overall, these findings suggest that continued improvements to air quality may further reduce TRAP exposure disparities by race-ethnicity. However, eliminating disparities may require additional policies and interventions that target the underlying causes of environmental injustice.

Acknowledgments

The authors thank M. Bechle for providing the calculated Census block descriptive statistics and for his contributions to method design and data interpretation.

The authors are grateful for support from the National Science Foundation (NSF; Sustainability Research Network award 1444745 and grant 0853467) and the U.S. Environmental Protection Agency (EPA; Assistance Agreement RD83587301). This article has not been formally reviewed by the NSF or the U.S. EPA; views expressed herein are solely those of authors and do not necessarily reflect those of either agency.

References

Abel TD, White J. 2011. Skewed riskscapes and gentrified inequities: environmental exposure disparities in Seattle, Washington. Am J Public Health 101(suppl 1):S246–S254, PMID: 21836115, 10.2105/AJPH.2011.300174.

Apte JS, Messier KS, Gani S, Brauer M, Kirchsetter TW, Lunden MM, et al. 2017. High-resolution air pollution mapping with Google street view cars. Environ Sci Technol 51(12):6999–7008, PMID: 28578585, 10.1021/acs.est.7b00891.

Ard K. 2015. Trends in exposure to industrial air toxins for different racial and socioeconomic groups: a spatial and temporal examination of environmental inequality in the U.S. from 1995 to 2004. Soc Sci Res 53:375–390, PMID: 26188461, 10.1016/j.ssresearch.2015.06.019.

Bechle MJ, Millet DB, Marshall JD. 2015. National spatiotemporal exposure surface for NO2: monthly scaling of a satellite-derived land-use regression, 2000–2010. Environ Sci Technol 49(20):12297–12305, PMID: 26397123, 10.1021/acs.est.5b02882.

Bento A, Freedman M, Lang C. 2015. Who benefits from environmental regulation? Evidence from the Clean Air Act Amendments. Rev Econ Stat 97(3):610–622, 10.1162/REST_a_00493.

Brajer V, Hall JV. 2005. Changes in the distribution of air pollution exposure in the Los Angeles basin from 1990 to 1999. Contemp Econ Policy 23(1):50–58, 10.1093/cep/byi005.

Brauer M, Lencar C, Tamburic L, Koehoorn M, Demers P, Karr C. 2008. A cohort study of traffic-related air pollution impacts on birth outcomes. Environ Health Perspect 116(5):680–686, PMID: 18470315, 10.1289/ehp.10952.

Brook JR, Burnett RT, Dann TF, Cakmak S, Goldberg MS, Fan X, et al. 2007. Further interpretation of the acute effect of nitrogen dioxide observed in Canadian time-series studies. J Expo Sci Environ Epidemiol 17(suppl 2):S36–S44, PMID: 18079763, 10.1038/sj.jes.7500626.

Burnett RT, Stieb D, Brook JR, Cakmak S, Dales R, Raizenne M, et al. 2004. Associations between short-term changes in nitrogen dioxide and mortality in Canadian cities. Arch Environ Health 59(5):228–236, PMID: 16201668, 10.3200/AEOH.59.5.228-236.

CDC (Centers for Disease Control). 2012. National Vital Statistics Reports: Deaths, Preliminary Data for 2011. http://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_06.pdf [accessed 1 December 2013].

CDC. 2015. Heart Disease Fact Sheet. https://www.cdc.gov/heartdisease/facts.htm [accessed 17 April 2017].

CDC. 2016. National Center for Health Statistics. Underlying Cause of Death 1999–2015 on CDC WONDER Online Database, released December 2016. http://wonder.cdc.gov/ucd-icd10.html [accessed 30 May 2017].

Clark LP, Millet DB, Marshall JD. 2014. National patterns in environmental injustice and inequality: outdoor NO2 air pollution in the United States. PLoS One 9(4):e94431, PMID: 24736569, 10.1371/journal.pone.0094431.

Clean Air Act Amendments of 1990. 1990. U.S. Public Law 101-549. https://www.congress.gov/bill/101st-congress/senate-bill/1630 [accessed 19 July 2017].

Finegold JA, Asaria P, Francis DP. 2013. Mortality from ischaemic heart disease by country, region, and age: statistics from World Health Organisation and United Nations. Int J Cardiol 168(2):934–945, PMID: 23218570, 10.1016/j.ijcard.2012.10.046.

Hajat A, Hsia C, O’Neill MS. 2015. Socioeconomic disparities and air pollution exposure: a global review. Curr Environ Health Rep 2(4):440–450, PMID: 26381684, 10.1007/s40572-015-0069-5.

Hewitt CN. 1991. Spatial variations in nitrogen dioxide concentrations in an urban area. Atmospheric Environment Part B Urban Atmosphere 25(3):429–434, 10.1016/0957-1272(91)90014-6.

Hipp JR, Lakon CM. 2010. Social disparities in health: disproportionate toxicity proximity in minority communities over a decade. Health Place 16(4):674–683, PMID: 20227324, 10.1016/j.healthplace.2010.02.005.

Howell AJ, Timberlake JM. 2014. Racial and ethnic trends in suburbanization of poverty in U.S. metropolitan areas, 1980–2010. J Urban Aff 36(1):79–98, 10.1111/juaf.12030.

Jerrett M, Burnett RT, Beckerman BS, Turner MC, Krewski D, Thurston G, et al. 2013. Spatial analysis of air pollution and mortality in California. Am J Respir Crit Care Med 188(5):593–599, PMID: 23805824, 10.1164/rccm.201303-0609OC.

Karner AA, Eisinger DS, Niemeier DA. 2010. Near-roadway air quality: synthesizing the findings from real-world data. Environ Sci Technol 44(14):5334–5344, PMID: 20560612, 10.1021/es100008x.

Kravitz-Wirtz N, Crowder K, Hajat A, Sass V. 2016. The long-term dynamics of racial/ethnic inequality in neighborhood air pollution exposure, 1990–2009. Du Bois Review 13(2):237–259, 10.1017/S1742058X16000205.

Levy I, Mihele C, Lu G, Narayan J, Brook JR. 2014. Evaluating multipollutant exposure and urban air quality: pollutant interrelationships, neighborhood variability, and nitrogen dioxide as a proxy pollutant. Environ Health Perspect 122(1):65–72, PMID: 24225648, 10.1289/ehp.1306518.

Marshall JD, Swor KR, Nguyen NP. 2014. Prioritizing environmental justice and equality: diesel emissions in Southern California. Environ Sci Technol 48(7):4063–4068, PMID: 24559220, 10.1021/es405167f.

McDonald BC, Dallman TR, Martin EW, Harley RA. 2012. Long-term trends in nitrogen oxide emissions from motor vehicles at national, scale, and air basin scales. J Geophys Res 117:D00V18, 10.1029/2012JD018304.

Mohai P, Lantz PM, Morenoff J, House JS, Mero RP. 2009. Racial and socioeconomic disparities in residential proximity to industrial facilities: evidence from the Americans’ Changing Lives Study. Am J Public Health 99(suppl 3):S649–S656, PMID: 19890171, 10.2105/AJPH.2007.131383.

Mohai P, Saha R. 2015. Which came first, people or pollution? A review of theory and evidence from longitudinal environmental justice studies. Environ Res Lett 10(12):125011, 10.1088/1748-9326/10/12/125011.

MPC (Minnesota Population Center). 2011. National Historical Geographic Information System: Version 2.0. http://www.nhgis.org [accessed 1 December 2015].

O’Neill MS, Jerrett M, Kawachi I, Levy JI, Cohen AJ, Gouveia N, et al. 2003. Health, wealth, and air pollution: advancing theory and methods. Environ Health Perspect 111(16):1861–1870, PMID: 14644658, 10.1289/ehp.6334.

Pais J, Crowder K, Downey L. 2014. Unequal trajectories: racial and class differences in residential exposure to industrial hazard. Soc Forces 92(3):1189–1215, PMID: 25540466, 10.1093/sf/sot0</em99.

Pastor M, Sadd J, Hipp J. 2001. Which came first? Toxic facilities, minority move-in, and environmental justice. J Urban Aff 23(1):1–21, 10.1111/0735-2166.00072.

Payne-Sturges D, Gee GC. 2006. National environmental health measures for minority and low-income populations: tracking social disparities in environmental health. Environ Res 102(2):154–171, PMID: 16875687, 10.1016/j.envres.2006.05.014.

Post ES, Belova A, Huang J. 2011. Distributional benefit analysis of a national air quality rule. Int J Environ Res Public Health 8(6):1872–1892, PMID: 21776207, 10.3390/ijerph8061872.

Reardon SF, Fox L, Townsend J. 2015. Neighborhood income composition by household race and income, 1990–2009. Ann Am Acad Pol Soc Sci 660(1):78–97, 10.1177/0002716215576104.

Rowangould GM. 2013. A census of the US near-roadway population: public health and environmental justice considerations. Transp Res D Transp Environ 25:59–67, 10.1016/j.trd.2013.08.003.

Schweitzer L, Valenzuela A. 2004. Environmental injustice and transportation: the claims and the evidence. J Plan Lit 18(4):383–398, 10.1177/0885412204262958.

Takenoue Y, Kaneko T, Miyamae T, Mori M, Yokota S. 2012. Influence of outdoor NO2 exposure on asthma in childhood: meta-analysis. Pediatr Int 54(6):762–769, PMID: 22640481, 10.1111/j.1442-200X.2012.03674.x.

U.S. Census Bureau 2011. Urban Area Criteria for the 2010 Census; Notice. Federal Register 76(164):53030–53043.

U.S. EPA (U.S. Environmental Protection Agency). 2016. Air Pollutant Emissions Trends Data. https://www.epa.gov/air-emissions-inventories/air-pollutant-emissions-trends-data [accessed 21 March 21 2016].

WHO (World Health Organization). 2005. Air Quality Guidelines: Global Update 2005. http://www.who.int/phe/health_topics/outdoorair/outdoorair_aqg/en/ [accessed 20 June 2016].

WHO. 2016. WHO Mortality Database. http://apps.who.int/healthinfo/statistics/mortality/whodpms/ [accessed 9 January 2017].

Maternal Exposure of BALB/c Mice to Indoor NO2 and Allergic Asthma Syndrome in Offspring at Adulthood with Evaluation of DNA Methylation Associated Th2 Polarization

Author Affiliations open

1College of Environment and Resource, Research Center of Environment and Health, Shanxi University, Taiyuan, Shanxi, People’s Republic of China

2State Key Laboratory of Environmental Chemistry and Ecotoxicology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, People’s Republic of China

PDF icon PDF Version (176 KB)

  • Background:
    Fetal stress has been proposed to be associated with diseases in both children and adults. Epidemiological studies suggest that maternal exposure to nitrogen dioxide (NO2) contributes to increased morbidity and mortality of offspring with allergic asthma later in life.
    Objectives:
    We aimed to test whether maternal NO2 exposure causes allergic asthma-related consequences in offspring absent any subsequent lung provocation and whether this exposure enhances the likelihood of developing allergic asthma or the intensity of developed allergic airway disease following postnatal allergic sensitization and challenge. In addition, if such consequences and enhancements occurred, we sought to determine the mechanism(s) of these responses.
    Methods:
    Pregnant BALB/c mice were exposed to either NO2 (2.5 ppm, 5 h/day) or air daily throughout the gestation period. Offspring were sacrificed on postnatal days (PNDs) 1, 7, 14, 21, and 42, and remaining offspring were sensitized by ovalbumin (OVA) injection followed by OVA aerosol challenge during postnatal wk 7–9. We analyzed the lung histopathology, inflammatory cell infiltration, airway hyper-responsiveness (AHR), immune responses, and gene methylation under different treatment conditions.
    Results:
    Maternal exposure to NO2 caused a striking increase in inflammatory cell infiltration and the release of type 2 cytokines in the lungs of offspring at PNDs 1 and 7; however, these alterations were reversed during postnatal development. Following OVA sensitization and challenge, the exposure enhanced the levels of allergic asthma-characterized OVA-immunoglobulin (Ig) E, AHR, and airway inflammation in adult offspring. Importantly, differentiation of T-helper (Th) 2 cells and demethylation of the interleukin-4 (IL4) gene occurred during the process.
    Conclusions:
    Maternal exposure to indoor environmental NO2 causes allergic asthma-related consequences in offspring absent any subsequent lung provocation and potentiates the symptoms of allergic asthma in adult offspring following postnatal allergic sensitization and challenge; this response is associated with the Th2-based immune response and DNA methylation of the IL4 gene. https://doi.org/10.1289/EHP685
  • Received: 19 June 2016
    Revised: 07 June 2017
    Accepted: 19 June 2017
    Published: 13 September 2017

    Address correspondence to N. Sang, College of Environment and Resource, Shanxi University, Taiyuan, Shanxi 030006, PR China. Telephone: 86-351-7011932. Email: sangnan@sxu.edu.cn

    Supplemental Material is available online (https://doi.org/10.1289/EHP685).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (215 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (107 KB)

Introduction

Nitrogen dioxide (NO2) is a gaseous pollutant found in both outdoor and indoor environments. It is emitted via automobile exhaust, fossil fuel combustion, and power plant operations, all of which compose the majority sources of outdoor exposures, whereas indoor exposures are commonly due to emissions from cooking gas, heating systems, and sites with increased occupational exposures such as garages and ferries (Ezratty et al. 2014; Gaffin et al. 2014). Importantly, indoor air NO2 pollution is likely more serious than outdoor pollution (Ezratty et al. 2014) with maximum concentrations indoors reaching peaks of 2–4 ppm (WHO 2010). NO2 exposure has been associated with asthma symptoms and the decline of lung function, and recent evidence has shown that early life exposure to NO2 is a potential stimulus for the early genetic encoding of asthma and allergic responses (Clark et al. 2010; Deng et al. 2016b; Khreis et al. 2017). In addition, such exposure has been considered as an important determinant in the later development of asthma in childhood and adulthood (ATSDR 2014; Deng et al. 2015, 2016a). A study conducted by Mortimer et al. (2008) in California indicated that prenatal NO2 exposure led to a decrease in the forced vital capacity (FVC) and forced expiratory volume in one second (FEV1) among children. Another study reported that nitrogen oxide (NOx, primarily NO2) exposure during the second trimester induced impaired lung development and worsened asthma symptoms in the offspring (Padula et al. 2010). Recently, it has been reported that maternal exposure to the air pollutant NO2 during pregnancy, especially during certain trimesters, was associated with an increased risk of the development of asthma in children (Deng et al. 2016b).

Allergic asthma is a chronic inflammatory disorder of the lung characterized by airway hyper-responsiveness (AHR), the recruitment of eosinophils and increased serum immunoglobulin E (IgE) levels. The dysregulation of specific T lymphocytes and their associated cytokines plays an important role in the etiology of allergic asthma (Lambrecht and Hammad 2015; Holgate 2012; Lloyd and Hessel 2010). The onset of allergic asthma is characterized by the increased infiltration of naïve CD4+ T lymphocytes into the bronchial mucosa. When these CD4+ T cells are sensitized by an allergen, they become activated and differentiate into proallergic T-helper (Th) 2 cells instead of the counterregulatory Th1 cells (Hwang et al. 2017; Muehling et al. 2017). The incidence of allergic asthma has risen steadily over the past 20 years and currently represents a significant health and financial burden. The reasons for this increase remain unclear, although a link between air pollution and allergic sensitization may be correlated with the prevalence of allergic asthma. Experimental studies have provided evidence of a biological basis for gases such as NO2 and particulate air pollutants as risk factors for allergic sensitization as indicated by enhanced IgE production (Weir et al. 2013). NO2 inhalation can augment the degree of allergic airway inflammation and prolong allergen-induced airway hyper-responsiveness (AHR) in rodent models of asthma (Poynter et al. 2006). Similarly, epidemiological studies of two birth cohorts in Germany and Sweden have shown a positive association between air pollution and allergic sensitization (Morgenstern et al. 2008; Nordling et al. 2008). Increased levels of NO2 have been consistently associated with an increased prevalence of allergic sensitization (Weir et al. 2013); however, information regarding the effects of NO2 exposure during early life, especially fetal development stages, and in combination with postnatal exposure to allergens, on the manifestation of allergic asthma later in life has rarely been assessed, and the lack of detailed experimental evidence warrants further investigation.

In our study, we treated pregnant BALB/c mice with NO2 and subjected postnatal offspring to selective allergic sensitizations and subsequent challenges to address the following: a) whether allergic asthma-related consequences manifest in juvenile offspring in the absence of subsequent lung provocation; b) what responses, if any, develop in adult offspring to postnatal allergic sensitization and challenge; and c) how maternal NO2 exposure causes postnatal respiratory and immune abnormalities following allergen sensitization and challenge.

Methods

Animal Protocol

BALB/c mice (7–8 wk of age) were purchased from the Beijing Vital River Laboratory Animal Technology Co., Ltd. and maintained on a 12-h light/dark cycle with unrestricted access to ovalbumin (OVA)-free food and water, except when kept in the dynamic exposure chamber. All animal care and experiments were approved by the Institutional Animal Care and Use Committee of Shanxi University. The animals were treated humanely and with regard to the alleviation of pain and suffering. At 9–10 wk of age, 156 female mice were mated (one male and two females per cage), and 101 plug-positive mice were considered pregnant. The day on which a vaginal plug was observed was defined as gestation day 0 (GD0).

NO2 Exposure and OVA Sensitization/Challenge

Plug-positive mice were randomly divided into an NO2 inhalation group containing 51 animals and a control group of 50 animals. For NO2 exposure, plug-positive animals, in individually housed wire cages, were exposed to NO2 (2.5 ppm, 5 h/day) daily in a dynamic exposure chamber starting at GD0, and the treatment lasted throughout the gestation period. The NO2 gas was diluted with air at the intake port of the exposure chamber to achieve the desired concentration, and the mixed gas was uniformly distributed throughout the entire chamber through two perforated gas-dispersion plates. One plate was placed at the intake port, and the other was placed on the gas outlet connected to an aspirator pump. The NO2 concentration in the exposure chamber was measured using a real-time NO2 monitor (FIX550-NO2-A, WANDI Technology Co.), and the exhaust gas was absorbed by an alkali absorption device. Correspondingly, the control plug-positive animals were continually exposed to air in the other exposure chamber using the same protocol. During the exposure period, the animals and their cages were placed in the exposure chamber, and the mice were free to move around in the cages but were denied access to water and food. Pregnant mice were housed in separate cages until the offspring could be weaned, and none of the offspring were exposed to NO2 after birth. At postnatal day (PND) 21, the offspring from the different litters were housed together (no more than 10 animals/cage), and males and females were separated. The offspring from the NO2 inhalation group and from the vehicle control group were sacrificed on PND1, 7, 14, 21, and 42, and the lungs were dissected for histological examination (two females and two males per time point); flow cytometry analysis (fluorescence-activated cell sorting (FACS), five females and five males per time point); enzyme-linked immunosorbent assay (ELISA) detection (10 females and 10 males on PND1, eight females and eight males on PND7, and five females and five males on PND14, 21, and 42); and methylation status analysis (three mice per time point).

The remaining offspring from the NO2 inhalation and vehicle control groups were administered either OVA [80 μg OVA and 1.3 mg colloid Al(OH)3; Sigma Chemical Co.], colloid Al(OH)3 (1.3 mg; Sigma Chemical Co.) or sterilized saline via a single intraperitoneal (i.p.) injection at 7 and 8 wk of age. Then, the OVA or saline challenge was followed by inhalational exposure to either 1% OVA or saline at 9 wk of age (30 min/day for 7 d) using an ultrasonic nebulizer until the offspring were sacrificed at 10 wk. Following the Al(OH)3 or OVA sensitization and OVA challenge, the offspring from mothers exposed to air or NO2 were designated the AAO [Air+Al(OH)3+OVA] (n=40), AOO (Air+OVA+OVA) (n=41), NAO [NO2+Al(OH)3+OVA] (n=40), and NOO (NO2+OVA+OVA) (n=43) groups, and the corresponding control groups subjected to saline treatment were classified as the ASS (Air+saline+saline) (n=40) and NSS (NO2+saline+saline) (n=42) groups. After allergen sensitization and challenge, the offspring from the various groups were sacrificed, and the lungs were dissected for FACS analysis (five females and five males), histological examination (two females and two males), the measurement of lung function (five females and five males), ELISA (five females and five males), and the assessment of methylation status (four mice). The pre- and postnatal treatment protocol is shown in Figure 1.

Timeline indicating air or nitrogen dioxide exposure during pregnancy from GD 0 to PND 1; sacrifice on PND 1, 7, 14, 21, and 42; intraperitoneal saline or aluminum hydroxide or OVA exposure at 7 and 8 weeks of age; aerosol saline or OVA exposure at 9 and 10 weeks of age; and ASS, AAO, AOO, NSS, NAO, and NOO sacrifice at 10 weeks of age.
Figure 1. Experimental timeline for prenatal and postnatal treatments and sacrifice indicating the times of maternal NO2 exposure and postnatal OVA sensitization and OVA challenges. Times are expressed relative to the birth date of the offspring. Note: GD, gestation day; PND, postnatal day; w, week; i.p., intraperitoneal.

Measurement of AHR

At 24 h after the final treatment, the AHR was assayed as previously described using an AniRes2005 Lung Function System (version 3.5; Bestlab Technology Co.) according to the manufacturer’s instructions (You et al. 2014). Briefly, mice were anesthetized with 50 mg/kg pentobarbital sodium (Beijing Solarbio Technology Co., Ltd.) and connected to a computer-controlled ventilator via the tracheal cannula. The time of expiration/inspiration and the respiratory rate were preset at 1.5:1 and 90/min, respectively. The resistance of the lung (RL), resistance of expiration (Re), and respiratory dynamic compliance (Cdyn) were recorded to evaluate the reaction of mice to a methacholine chloride [Acetyl-β-methylcholine chloride (MCH) (Sigma)] gradient (0.025, 0.05, 0.1, and 0.2 mg/kg body weight); this compound was injected into the jugular vein at 5-min intervals using a fine needle.

FACS Analysis

Mouse lungs were carefully perfused, excised, diced, and incubated with collagenase type IA (0.5 mg/mL; Sigma-Aldrich Corp.) and type IV bovine pancreatic DNase (20 μg/mL; Sigma-Aldrich Corp.) for 45 min at 37°C in Hank’s Balanced Salt Solution (HBSS) containing 5% fetal bovine serum (FBS). The digested lung tissue was passed through a 40-μM cell strainer to form a single-cell suspension, and TruStain fcX (antimouse CD16/32) was added to the suspension to block the Fc receptors on the cells. To determine the percentage of macrophages, eosinophils, neutrophils, and lymphocytes in the lung tissue, the cells were stained for extracellular antigens using specific antibodies according to standard instructions (Table 1) (Barletta et al. 2012; Ford et al. 2012; Sharma et al. 2016). The percentage of CD4+ interleukin (IL)-4+ (Th2) cells was measured after the suspensions were washed with PBS before surface staining. To determine the intracellular expression of IL-4, cells were first treated with a cell stimulation and protein transport inhibition cocktail containing PMA, Ionomycin, Brefeldin A, and Monensin (500x, eBioscience, Thermo Fisher Scientific (China) Co., Ltd.) for 4 h followed by surface staining (PerCP-conjugated antimouse CD45, APC-conjugated antimouse CD3, and FITC-conjugated antimouse CD4 [BioLegend Inc.,]). The cells were fixed with 2% PFA, permeabilized with 0.5% Triton X-100 and stained with a PE-conjugated antimouse IL-4 antibody (eBioscience, Thermo Fisher Scientific (China) Co., Ltd.) (Jeon et al. 2014; Wang et al. 2013). All the experiments set isotype control. The FACS analysis was performed using a BD LSRII flow cytometer (BD Biosciences) according to standard protocols as previously described. The details of the gating strategy are shown in Figure S1.

Table 1. The antibodies used for staining antigens by FACS analysis.
Number Antigen Fluorochrome Company
1 CD3 APC Biolegend
2 CD4 FITC Biolegend
3 CD45 Percp Biolegend
4 CD11b PE/cy7 Biolegend
5 Gr-1 PE Biolegend
6 I-A/I-E FITC Biolegend
7 F4/80 PE Biolegend
8 Siglect-f eflour 660 eBioscience
9 IL-4 PE eBioscience

Determination of Cytokines

After the mice were sacrificed, the lungs were harvested, and 60 mg of tissue per animal was homogenized in 300 μL PBS containing 1% (wt/vol) Triton X-100 (Beijing Solarbio Technology Co., Ltd.) and pepstatin A, leupeptin, and aprotinin (all at 20 ng/ml, pH 7.4; Beijing Solarbio Technology Co., Ltd.) and incubated on ice for 30 min as previously described (Wieland et al. 2006). The homogenate was then centrifuged (4°C, 1,500 g, 15 min), and the supernatant was assayed for IL-4, IL-13, and interferon (IFN)-γ using the appropriate ELISA reagent kits (R&D Systems, Inc.) according to the manufacturer’s instructions.

Histological Examination

After a lung lavage and the removal of the right lung lobes, the left lung lobe was intratracheally perfused with 10% neutral buffered formalin at a constant pressure of 30 cm of fixative. After 1 h, the trachea was ligated, and the inflated left lung lobe was immersed in a large volume of the same fixative for at least 24 h. The following day, the tissue was embedded in paraffin, and segments (5- to 6-μm-thickness) were prepared for staining with hematoxylin and eosin (H&E), Alcian blue (pH 2.5), and Periodic acid Schiff’s (AB/PAS) reagent or Masson’s trichrome (MT); five sections were used per stain (Li et al. 2009).

Total IgE and OVA-Specific IgE Detection

At 24 h after the last OVA aerosol exposure, blood specimens were collected using evacuated tubes (Taizhou Tianyi Medical Devices Co., Ltd.) and incubated at room temperature (RT, 20–25°C) for 30 min. The blood samples were then centrifuged at 1,000 g for 10 min at RT, and the resulting supernatant (serum) was stored at −80°C until further analysis. The levels of total IgE and OVA-specific IgE in the serum were measured using ELISA reagent kits (Mouse IgE ELISA MAX Deluxe Sets and LEGEND MAX and Mouse OVA-Specific IgE ELISA Kit with Pre-coated Plates; BioLegend Inc.) according to the manufacturer’s instructions.

Analysis of the Methylation Status of the IL4 and IL13 Promoters

The methylation status of the IL4 and IL13 promoters was measured as previously described (Zhou et al. 2016). Genomic DNA was extracted using a Tissue DNA Kit (Omega Bio-tek, Inc.) and treated with bisulfite according to the EZ DNA Methylation-Gold™ Kit instruction manual (Zymo Research). The IL4 and IL13 promoter sequences were amplified from the isolated DNA using touchdown PCR, extracted from an agarose gel and incorporated into the Puc18-T vector [Sangon Biotech (Shanghai) Co., Ltd.] for TA cloning and sequencing by Sangon Biotech [Sangon Biotech (Shanghai) Co., Ltd.]. The PCR conditions were as follows: 98°C for 4 min; 20 cycles of 45 s at 94°C, 45 s at 66°C (with a 0.5°C decrement per cycle), and 1 min at 72°C; 20 cycles of 45 s at 94°C, 45 s at 56°C and 1 min at 72°C; and a final extension at 72°C for 8 min. The prediction of the CpG islands and the designation of the BSP primers for the IL4 core promoter sequence (−566 to −311 and −41 to +244) and the IL13 sequence (−1915 to −1554 and −355 to −81) were performed by Sangon Biotech. The BSP primer sequences were as follows: CpG region 1 of IL4, forward 5′-TTGTAAGATTAGTTGGTTTAGGATG-3′, and reverse 5′-TTTCAACATAAAAAATTACACCATA-3′; CpG region 2 of IL4, forward 5′-GTTAGTATTGTATTGTTAGTATTTTTTGAT-3′, and reverse 5′-ATCTCTTAAACTTTATCCCTAATCCTA-3′; CpG island 1 of IL13, forward 5′-TTTATTGTAGYGGGGYGGT-3′, and reverse 5′-ACCTTAAACRCTACATAAATAAATCA-3′; and CpG region 2 of IL13, forward 5′-GGTTAGTATTGGGTTGGTTGTTTAG-3′, and reverse 5′-CCTAAACTACTAACTTATAACCTTAACCTA-3′. In detail, five individual clones from one mouse, a total of 15 clones of three mice from PND1 and PND42, and five individual clones from one mouse and a total of 20 clones of four mice from each treatment were sequenced. In a given sample, the number of the methylated CpG sites was divided by the total number of detected CpG sites among the five clones to evaluate the methylation percentage of each CpG region for one animal; furthermore, the methylation status of each region within the IL4 and IL13 promoters from each treatment group were calculated by averaging the methylation rate of each CpG region from three or four animals (Tang et al. 2012; Zhang et al. 2014). Visualization and analysis of the methylation status of the CpG regions surrounding the IL4 and IL13 promoters were conducted using a BiQ Analyzer (http://biq-analyzer.bioinf.mpi-inf.mpg.de/tools/MethylationDiagrams/index.php).

Statistical Analysis

The data are expressed as the mean±standard error. A one-way analysis of variance (ANOVA) followed by Fisher’s least significant difference (LSD) test was used to determine significant differences between all the treatment groups and the control group. A two-tailed Student’s t test was used to analyze the experimental results between pairs of groups. Significance was defined as p<0.05, and the data analysis and figure generation were conducted using Origin® 8.0 (OrigenLab Corp.).

Results

There were 267 offspring from 39 litters in the control group comprising 139 female and 128 male offspring, and there were 286 offspring from 42 litters in the NO2 inhalation group comprising 160 female and 126 male offspring. There were no significant differences in either the delivery rate or litter size between the NO2– and air-exposed dams; additionally, none of the offspring died due to exposure.

Maternal NO2 Exposure, Airway Inflammation, and Th2 Polarization in Juvenile Offspring

We first investigated whether allergic asthma-related consequences arose in the offspring subjected to maternal NO2 exposure from GD0 to birth in the absence of subsequent lung provocation. As shown in Figure 2, maternal NO2 inhalation caused inflammatory cell infiltration in the peribronchial and perivascular areas (H&E staining) and peribronchial collagen deposition (MT staining) in PND1 and seven mice, although no obvious mucous cell metaplasia (AB-PAS staining) was observed at any time point (data not shown). Interestingly, these pathological abnormalities were attenuated during postnatal development and were nearly eliminated at PND21 and 42. Consistent with this, we observed that the total number of cells in the lungs of the offspring at PND1 increased slightly by 1.52-fold in comparison with that in the corresponding control pups, suggesting that maternal NO2 exposure mildly promotes the recruitment of cells. To quantify the potential inflammatory response, we further analyzed the leukocyte subtypes by differentiating the recruited cell populations using FACS. Figure 3 indicates that maternal NO2 exposure resulted in increased recruitment of macrophages, eosinophils, and lymphocytes, with eosinophils presenting the most obvious increase in the lungs of offspring at PND1, 7 and 14. However, no significant increase in the number of neutrophils was observed at any time point. As expected, the total number of cells began to decline at PND7, and the recruitment of macrophages and lymphocytes was gradually eliminated by PND14; however, eosinophils were still detected.

Photomicrographs

Figure 2. Representative photomicrographs of histology pictures of airway structures in the lungs of offspring at PND1, 7, 14, 21, and 42. H&E staining shows the inflammatory cell infiltration; MT staining shows subepithelial collagen deposition.

Note: H&E, hematoxylin and eosin stain; MT, Masson’s trichrome stain; al, airway lumen; ap, alveolar parenchyma; bv, blood vessel; e, airway surface epithelium; arrows, sites of fibrosis. Bar=100 μm.

Bar graphs with standard errors plotting inflammatory cell counts in lungs per 10 super 4 lung cells (left y-axis) and total lung cells (right y-axis) during air and nitrogen dioxide exposures as measured at PND 1, 7, 14, 21 and 42, where n equals 10, for Mac, Eos, Neu, Lym, and total groups (x-axis).
Figure 3. Effects of maternal NO2 exposure on the differentiation of inflammatory cells in the lungs of offspring at PND1, 7, 14, 21, and 42. The values are expressed as the mean±SE. *p<0.05 compared with the air exposure group.Note: Mac, macrophages; Eos, eosinophils; Neu, neutrophils; Lym, lymphocytes.

Cellular inflammation of the airways with eosinophils is a characteristic feature of allergic asthma (Zissler et al. 2016; Loutsios et al. 2014). Coupled with airway inflammation, T cells in the airways in human and animal models of allergic asthma present cytokine profiles characteristic of Th2 cells (Zissler et al. 2016; Singh et al. 2011). Based on the above findings, we hypothesized that maternal NO2 exposure may differentially induce naïve CD4+ T-cell polarization toward Th2 cells instead of Th1 cells. To test this hypothesis, we analyzed the expression of the type 2 cytokines IL-4 and IL-13 and the type 1 cytokine IFN-γ. As shown in Figure 4, maternal NO2 exposure elevated the IL-4 levels and suppressed IFN-γ expression at PND1 and PND7; moreover, these effects tended to subside at PND14 and even returned to normal levels at PND21 and 42. Additionally, the IL-13 content was statistically increased at PND1 but restored at PND7.

Figures 4A, 4B, and 4C are bar graphs with standard errors plotting lung IL-4, lung IL-3, lung IFN-gamma in picograms per milliliter, respectively, (y-axis) during air and nitrogen dioxide exposures as measured at across PND1 (n equals 20), PND7 (n equals 16), PND14 (n equals 10), PND21 (n equals 10), and PND42 (n equals 10) (x-axis).
Figure 4. Effects of maternal NO2 exposure on the levels of the (A) type 2 cytokines IL-4 and (B) IL-13 and (C) the type 1 cytokine IFN-γ in the lungs of offspring at PND1, 7, 14, 21, and 42. The values are expressed as the mean±SE. *p<0.05 compared with the air exposure group.

Importantly, we controlled for sex in the statistical analyses and provide these data in the Supplemental Materials (Figure S2). The results indicated that the basic values in male offspring were slightly different from those in female offspring, including the number of eosinophils and the IL-4 level, but the fold changes were equal in the two sexes, suggesting no sex-related significant difference in the inflammatory responses and the expression of the type 2 cytokines IL-4 and IL-13 and the type 1 cytokine. These findings suggest that maternal NO2 exposure caused allergic asthma-related consequences in the offspring, including airway inflammation and Th2 polarization; however, these effects subsided during postnatal development in the absence of subsequent lung provocation.

Maternal NO2 Exposure, OVA-Specific IgE Release, AHR Alteration, and Lung Cell Differentiation in Offspring in Response to OVA Sensitization and Challenge

Next, we sought to clarify the effects (if any) of maternal NO2 exposure on offspring subjected to allergic sensitization and challenge. To address this issue, we combined maternal NO2 exposure with postnatal exposure to OVA to elicit an allergic response via sensitization and subsequent challenge with OVA. The OVA sensitization and subsequent challenge apparently increased the levels of total IgE regardless of the type of maternal inhalational exposure, and the levels in the AOO and NOO groups were 3.86- and 3.94-fold higher than the values in the ASS and NSS groups and 1.44- and 1.53-fold higher than the values in the AAO and NAO groups, respectively. No significant difference was observed between the NOO and AOO groups. OVA-specific IgE was undetectable in the ASS and NSS groups but apparently increased in the AOO and NOO groups, suggesting that allergic asthma was successfully established in this study. Importantly, the levels of OVA-specific IgE in the NOO group significantly increased to 1.96- and 2.58-fold of those in the AOO and NAO groups, respectively (Figures 5A and B).

Figures 5A and 5B are bar graphs with standard errors plotting serum total IgE and serum OVA-IgE, respectively, in nanograms per milliliter (y-axis) across control groups ASS, AAO, AOO, NSS, NAO, and NOO (x-axis), where n equals 10. Figure 5C comprises three line graphs with standard errors plotting changes in Re (centimeter of H sub 2 O per seconds per milliliter), R sub L (centimeter of H sub 2 O per seconds per milliliter), and Cdyn (milliliter per centimeter of H sub 2 O), respectively, (y-axis) with increases in MCH dose in milligrams per kilogram in control groups ASS, AAO, AOO, NSS, NAO, and NOO, where n equals 10.
Figure 5. Effects of maternal NO2 exposure on the symptoms of allergic asthma in offspring subjected to OVA sensitization and challenge. (A) Total IgE in the serum of offspring from the different treatment groups. (B) OVA-specific IgE in the serum of offspring from the different treatment groups. (C) AHR measurements of offspring from the different treatment groups. The values are recorded as the mean±SE. *p<0.05, **p<0.01, ***p<0.001 compared with the saline groups. #p<0.05, ##p<0.01, ###p<0.001 compared with the Al(OH)3 groups. †p<0.05, ††p<0.01 compared with the AOO group.

As predicted, OVA sensitization and challenge significantly increased AHR in the offspring regardless of maternal exposure to air or NO2 (Figure 5C). For all the groups, the Re values increased with increasing doses of MCH, and at 0.2 mg/kg MCH, the Re values in the AOO and NOO groups were 1.36- and 1.96-fold higher than those in the ASS and NSS groups and 1.38- and 1.86-fold higher than those in the AAO and NAO groups, respectively. RL showed the same trend as Re, whereas Cdyn showed an opposing trend to that of Re. Furthermore, the AHR in the NOO group was significantly higher than that in the AOO and NAO groups. However, maternal NO2 exposure without OVA sensitization and challenge did not induce apparent AHR. AHR directly reflects changes in the airway wall structure. Common structural changes in the airway include thickening of the subbasement membrane, excessive mucus secretion, subepithelial fibrosis, inflammatory cell infiltration, and extracellular matrix deposition in the subepithelial layer (You et al. 2014). Consistent with this, we observed that OVA-sensitized and subsequently challenged mice presented inflammatory cell infiltration in the peribronchial and perivascular areas, mucous cell metaplasia, and peribronchial collagen deposition. Importantly, these alterations in the offspring from the NOO group were more profound than those in the AOO and NAO groups (Figure 6).

Photomicrographs.
Figure 6. Representative photomicrographs of histology images of airway structures in the lungs of offspring subjected to OVA sensitization and challenge. H&E staining shows the inflammatory cell infiltration; MT staining shows subepithelial collagen deposition; AB/PAS staining shows mucous cells. Note: H&E, hematoxylin and eosin stain; AB/PAS, Alcian blue and Periodic acid Schiff double stain; MT, Masson’s trichrome stain; al, airway lumen; ap, alveolar parenchyma; bv, blood vessel; e, airway surface epithelium; arrows, sites of fibrosis and mucous cell metaplasia. Bar=100 μm.

Next, we quantified the inflammatory cell differentiation. Figure 7 shows that the OVA sensitization and subsequent challenge significantly recruited more total cells, macrophages, eosinophils, and Th2 cells into the lung in the absence or presence of maternal NO2 exposure, and the number of eosinophils in the AOO and NOO groups was 2.34- and 4.00-fold greater than that in the ASS and NSS groups and 2.11- and 2.90-fold greater than that in the AAO and NAO groups, respectively. Additionally, the number of macrophages in the AOO and NOO groups was 1.73- and 2.84-fold higher than that in the ASS and NSS groups and 1.66- and 2.52-fold higher than that in the AAO and NAO groups, respectively. Regarding Th2 cells, the number in the AOO and NOO groups was 2.01- and 3.04-fold higher than that in the ASS and NSS groups and 1.78- and 3.20-fold higher than that in the AAO and NAO groups, respectively. Moreover, the numbers of eosinophils, macrophages, and Th2 cells in the NOO group were 1.75-, 1.64- and 1.78- fold higher than those in the AOO group, respectively. Interestingly, no significant increases in the neutrophil count were observed in any of the groups. These results suggested that maternal NO2 exposure enhanced the allergic asthma responses in offspring when combined with postnatal exposure to antigen.

Bar graph with standard errors plotting inflammatory cell counts in lungs per 10 super 4 lung cells (left y-axis) and total lung cells (right y-axis) during nitrogen dioxide exposures across Mac, Eos, Neu, Th2, and total groups (x-axis) observed in control groups ASS, AAO, AOO, NSS, NAO, and NOO, where n equals 10.
Figure 7. Effects of maternal NO2 exposure on inflammatory cell differentiation in the lungs of offspring subjected to OVA sensitization and challenge. The values are expressed as the mean±SE. *p<0.05, **p<0.01, ***p<0.001 in comparison with the saline groups. #p<0.05, ##p<0.01, ###p<0.001 in comparison with the Al(OH)3 groups. †p<0.05, †††p<0.001 in comparison with the AOO group. Note: Mac, macrophages; Eos, eosinophils; Neu, neutrophils.

Maternal NO2 Exposure and Th2 Polarization in Offspring following OVA Sensitization and Challenge

To elucidate the primary immune cell response underlying how maternal NO2 exposure enhanced symptoms of allergic asthma to OVA sensitization and challenge, we measured the levels of the type 2 and type 1 cytokines. As shown in Figures 8, the levels of IL-4 and IL-13 in OVA-sensitized and subsequently challenged offspring were significantly elevated in the lungs of the offspring, regardless of maternal exposure to air or NO2. The IL-4 levels in the AOO and the NOO groups were 1.85- and 2.42-fold higher than those in the ASS and NSS groups and 1.63- and 2.16-fold higher than levels in the AAO and NAO groups, respectively. The trend for IL-13 expression was similar to that for IL-4 expression. Importantly, the levels of IL-4 and IL-13 in the NOO group were 1.34- and 1.61-fold higher than those in the AOO group, respectively. In contrast, the IFN-γ levels were significantly decreased in OVA-sensitized and subsequently challenged offspring regardless of maternal exposure to air or NO2, and the values in the AOO and the NOO groups were 0.72- and 0.64-fold of those in the ASS and NSS groups and 0.80- and 0.67-fold of those in the AAO and NAO groups, respectively. In particular, the content of IFN-γ in the NOO group was 0.82-fold of that in the AOO group. These findings imply that the unbalanced differentiation of naïve CD4+ T cells into Th2 cells instead of Th1 cells plays an important role in the maternal NO2 exposure-mediated enhancement of allergic asthma in offspring when combined with postnatal exposure to OVA sensitization and challenge.

Figures 8A, 8B, and 8C are bar graphs with standard errors plotting lung IL-4, lung IL-3, lung IFN-gamma in picograms per milliliter, respectively, (y-axis) during nitrogen dioxide exposures observed in control groups ASS, AAO, AOO, NSS, NAO, and NOO (x-axis), where n equals 10.
Figure 8. Effects of maternal NO2 exposure on (A) the type 2 cytokines IL-4 and (B) IL-13 and (C) the type 1 cytokine IFN-γ in the lungs of offspring subjected to OVA sensitization and challenge. The values are expressed as the mean±SE. *p<0.05, **p<0.01, ***p<0.001 in comparison with the saline groups. #p<0.05, ##p<0.01, ###p<0.001 in comparison with the Al(OH)3 groups. †p<0.05 in comparison with the AOO group.

We also controlled for sex differences in the FACS analysis (five females and five males), lung function measurements (five females and five males), and ELISA detection (five females and five males). As shown in Figure S3, the basic values in the male offspring were slightly different from those of the female offspring, including the AHR, number of eosinophils, OVA-IgE, IL-4, and so on, but the fold changes between the two sexes were equal. These findings suggest no significantly sex-related influence on enhancing the intensity of allergic asthma.

IL4 Promoter Demethylation in Offspring Subjected to Maternal NO2 Inhalation and Postnatal OVA Sensitization and Challenge

Increasing evidence shows that epigenetic mechanisms are involved in regulating T-cell differentiation, cytokine expression, allergic sensitization, and allergic asthma development (Harb and Renz 2015; Lovinsky-Desir and Miller 2012). Importantly, the methylation status of several CpG sites at the IL4 and IL13 genes during Th1/Th2 differentiation and T-cell stimulation has been studied by other researchers (Lee et al. 2015). Therefore, the methylation status of the IL4 and IL13 promoters in the lungs of the offspring subjected to different treatment conditions was evaluated. We first investigated whether alterations in gene expression were accompanied by changes in the methylation status of region 1 or region 2 of the IL4 flanking sequence. Prior to OVA sensitization and subsequent challenge, the methylation status of region 2 in the IL4 promoter was similar to that of the control group, whereas the methylation status of region 1 was decreased by 20% (Figure 9A) on PND1 but mildly reduced on PND42 (Figure S4A), which is consistent with the observed changes in gene expression. Further, increases in IL4 expression mediated by OVA sensitization and challenge were associated with reduced promoter methylation of region 1, but the methylation status of region 2 was unchanged (Figure 9B). Interestingly, maternal NO2 exposure followed by postnatal OVA sensitization and subsequent challenge significantly decreased the methylation rate of region 1 by 9% in comparison with normal air exposure followed by OVA sensitization and subsequent challenge, which is consistent with the change of IL4 expression. Furthermore, we assessed the methylation status of the IL13 gene on both CGI (region 1) and nonCGI CpG dinucleotides (region 2). Prior to OVA sensitization and subsequent challenge, the methylation status of region 1 in the IL13 promoter was similar to that of the control group, but the levels of methylation at region 2 were decreased by 10% (Figure 10A) on PND1 and completely absent on PND42 (Figure S4B), which was consistent with the results of the gene expression analysis. However, OVA sensitization and challenge did not alter the promoter methylation of regions 1 and 2 regardless of maternal NO2 exposure (Figure 10B), which does not correspond with the changes of IL13 gene expression. These findings suggest that IL4 promoter demethylation was associated with Th2 polarization in the offspring in response to maternal NO2 inhalation and postnatal OVA sensitization and subsequent challenge.

Figure 9A shows methylation status of the IL4 promoter in the lungs at PND1, when exposed to air (n equals 3) and nitrogen dioxide (n equals 3). Figure 9B shows methylation status of the IL4 promoter in the lungs from the different treatment groups ASS, AOO, NSS, and NOO, where n equals 4.
Figure 9. Effects of maternal NO2 exposure on the methylation status of the IL4 promoter. (A) The methylation status of the IL4 promoter in the lungs of offspring at PND1. (B) The methylation status of the IL4 promoter in the lungs of offspring from the different treatment groups. Five individual clones from three mice and a total of 15 clones at PND1 were sequenced, and five individual clones from four mice, and a total of 20 clones from the different treatment groups were sequenced. Each row represents an individual clone of the promoter, and a total of five CpG sites on region 1 and 10 CpG sites on region 2 for the IL4 promoter were analyzed. Each circle represents a CpG site within the promoter; white circles represent unmethylated CpGs, and black circles represent methylated CpGs. Met %, average percent of total CpG methylation. The data are expressed as the mean±standard error. *p<0.05, **p<0.01 in comparison with air or saline groups. †p<0.05 in comparison with the AOO group. S1, S2, S3, and S4 refer to sample 1, sample 2, sample 3, and sample 4, respectively.
Figure 10A shows methylation status of the IL13 promoter in the lungs at PND1, when exposed to air (n equals 3) and nitrogen dioxide (n equals 3). Figure 10B shows methylation status of the IL13 promoter in the lungs from the different treatment groups ASS, AOO, NSS, and NOO, where n equals 4.
Figure 10. Effect of maternal NO2 exposure on the methylation status of the IL13 promoter. (A) The methylation status of the IL13 promoter in the lungs of offspring at PND1. (B) The methylation status of the IL13 promoter in the lungs of offspring from the different treatment groups. Five individual clones from three mice and a total of 15 clones at PND1 were sequenced, and five individual clones from four mice and a total of 20 clones from the different treatment groups were sequenced. Each row represents an individual clone of the promoter, and a total of 36 CpG sites on region 1 and six CpG sites on region 2 for the IL13 promoter were analyzed. Each circle represents a CpG site within the promoter; white circles represent unmethylated CpGs, and black circles represent methylated CpGs. Met % and Met ‰, average percent of total CpG methylation. Note: The data are expressed as the mean±standard error. S1, S2, S3, and S4 refer to sample 1, sample 2, sample 3, and sample 4, respectively. *p<0.05 in comparison with air or saline groups.

Discussion

Although epidemiological studies have suggested that early-life NO2 exposure may be associated with the increased incidence and severity of asthma, the results were controversial (Clark et al. 2010; Deng et al. 2015; Morales et al. 2015; Ranzi et al. 2014). The results of our current study revealed two important points. First, maternal NO2 exposure caused eosinophilic airway inflammation and an associated release of type 2 cytokines in the lung shortly after birth; second, the effects were almost totally restored to normal physiological levels during postnatal development in the absence of subsequent lung provocation. Similarly, exposure of pregnant dams to lipopolysaccharide (LPS) caused greater pulmonary inflammation in pups at PND 0, 2, 6, and 14, although these differences disappeared by PND 21 (Cao et al. 2009). In utero environmental tobacco smoke (ETS) exposure without further lung challenge was observed to have no measurable effect on either pulmonary function or histology in offspring at PND 42 (Penn et al. 2007). Emerging evidence suggests that full maturation of the alveolus occurs during the alveolarization stage (PND 0–14) (Herriges and Morrisey 2014). Following this, we determined the mean linear intercept, alveolar surface area per unit of lung volume, the ratio of lung to body weight, and body weight during postnatal development, and we did not observe any apparent delayed lung development. Thus, the delayed recovery of eosinophilic inflammation in the lung of juvenile offspring might be related to inflammatory cell infiltration, increased collagen deposits in the airways, and increased thickness of the subepithelial basement membrane zone. Although the recruitment of inflammatory cells and associated factors declines during postnatal development, these early abnormalities resemble a maternal NO2 exposure-induced hyper-responsive airway that becomes more pronounced after additional lung stress or challenge.

To address this implication, we established a mouse model of allergic asthma using OVA sensitization and subsequent challenge and then sequentially evaluated the effects of maternal exposure to NO2 on allergic manifestations in the offspring. IgE is an important biomarker of asthma that is significantly increased in the serum of individuals with asthma and in animal models of asthma (Froidure et al. 2016). Increased levels of IgE may cause different immune effector cells to release a variety of mediators that promote AHR (Rabe et al. 2011), which is characterized as uncontrolled bronchoconstriction in response to various stimuli. Importantly, exposure of pregnant females to air pollution can lead to elevated IgE levels, long-term deficits in lung function, and even asthma in the offspring (Zacharasiewicz 2016; Peters et al. 2013). AHR in asthma is related to eosinophilic airway inflammation. The eosinophil is a central effector cell that, when localized to an inflamed asthmatic airway, can cause bronchial epithelial damage and airflow obstruction (Kim et al. 2010). Here, we observed that OVA-specific IgE, AHR, and eosinophilic inflammation were all significantly increased in offspring that were sensitized and challenged with OVA and maternal exposed to NO2 in comparison with offspring that were not exposed to either NO2 or OVA-sensitization and challenge. It has been reported that a Th2>Th1 immune disequilibrium contributes to the pathogenesis of allergic diseases (Romagnani 1994). Importantly, prenatal stress exposure increased the susceptibility to allergic AHR and inflammation, which were accompanied by a Th2-dominated immune response (Lee et al. 2015; Penn et al. 2007). Consistent with this observation, the results of the present study show that offspring subjected OVA sensitization and challenge and born to dams exposed to NO2 during pregnancy exhibited significantly increased IL-4 and IL-13 expression and decreased IFN-γ expression. Th2 cells are activated by various allergens and subsequently initiate allergic immune responses in mild-to-moderate eosinophilic asthma (Vroman et al. 2015). Excessively produced type 2 cytokines (IL-4 and IL-13) cause the infiltration of inflammatory cells into the airway, such as eosinophil influx (Lambrecht and Hammad 2015). In addition, these inflammatory cells produce more cytokines, thus exacerbating airway inflammation and further facilitating allergic responses (Kim et al. 2010). Similarly, prenatal diesel exhaust particle (DEP) exposure enhanced the expression of type 2 cytokines (IL-4, IL-5 and IL-13) and suppressed the expression of type 1 cytokines (IFN-γ) in offspring subjected to OVA immunization and challenge, which resulted in AHR and eosinophilic inflammation (Manners et al. 2014). These findings suggest that maternal NO2 exposure likely enhances the risk of developing allergic airway diseases or increases the intensity of developed allergic airway disease in offspring subjected to postnatal allergic sensitization and challenge and that the action may be associated with disrupting the balance of Th1/Th2 cell differentiation and the subsequent activation of Th2 immune responses. The underlying mechanisms of NO2-mediated promotion of allergic asthma in mouse pups remain unclear; however, increasing evidence of the resulting immune responses have been described. NO2 exposure impaired local bronchial immunity (Guarnieri and Balmes 2014), induced inflammatory cytokine release, and disrupted the balance of Th1/Th2 differentiation (Ji et al. 2015).

Early-life exposure to air pollution can induce epigenetic alterations in gene expression and affect the disease risk for asthma and airway allergies later in life (Martino and Prescott 2011). These alterations frequently involve fluctuations of an aberrant DNA methylation pattern on promoters, accompanied by changes in gene expression; these pattern changes include global hypomethylation and gene-specific hypermethylation or hypomethylation (Lee et al. 2015). Specifically, DNA methylation has been classified as an important modulatory process for the establishment and maintenance of Th2 bias in asthmatic and allergic symptoms (Yang et al. 2015). DNA methylation at the promoter regions of the IL4 and IL13 genes are likely associated with an allergic phenotype with Th2 activation (Bégin and Nadeau 2014). Experimental evidence has shown that in utero ETS exposure increases the risk of pulmonary inflammation and AHR associated with altered DNA methylation (Lee et al. 2015). Moreover, when female mice (F0) were exposed to A. fumigatus during early gestation, the F2 generation developed increased airway eosinophilia and reduced levels of methylation of IL4 CpG−408 and CpG−393 (Niedzwiecki et al. 2012). Furthermore, NO2 exposure during pregnancy was associated with differential DNA methylation in mitochondria- and antioxidant-related genes in the offspring (Gruzieva et al. 2017). In this study, we observed a significant difference in the DNA methylation status at the IL4 promoter in the lungs of PND1 offspring from the air- and NO2-exposed dams, but this difference was absent by PND42. Importantly, the DNA hypomethylation status at the IL4 promoter in the lungs of offspring was statistically increased following postnatal OVA sensitization and subsequent challenge regardless of maternal NO2 exposure; however, offspring from dams exposed to NO2 and subjected to the postnatal OVA sensitization and challenge exhibited an enhanced DNA demethylation rate. Additionally, the identified demethylation alterations of the IL4 gene were consistent with the changes in cytokine expression. In contrast, although the DNA methylation status at the IL13 promoter in the lung at PND1 showed slight alterations, few methylation sites in the IL13 promoter were shared among the different treatment groups, which suggests that changes in IL-13 expression were not associated with DNA methylation under these described treatment conditions, and future studies should be conducted to determine how IL13 is involved in the response to maternal NO2 exposure. Allergic asthma is defined as bronchial constriction and Th2-dominated airway inflammation following allergen sensitization and subsequent challenge. During this process, IL-4 is an important cytokine predominantly released by Th2 that triggers a humoral immune response toward IgE up-regulation and eosinophil accumulation in the airways (Oeser et al. 2015). Consistent with this, Kwon et al. (2008) reported the demethylation of CpG (−80) on the IL4 promoter of human CD4+ T lymphocytes isolated from patients with allergic asthma following in vitro exposure to dust-mite allergens. In response to allergens, the demethylation of type 2 cytokine genes induces a change in the chromatin structure, allowing the DNA to unfold and recruit transcription factors for the immediate expression of type 2 cytokines (Bégin and Nadeau 2014). Our findings provide initial experimental evidence that maternal NO2 inhalation-induced IL4 hypomethylation was associated with Th2 polarization in the offspring, which predisposes the offspring to allergic asthma. Considering that maternal NO2 exposure increases the number of Th2 cells in the lungs of offspring, regardless of OVA sensitization and challenge, and that IL-4 is primarily produced by Th2, we cannot exclude the notions that IL4 demethylation is due to the abundance of Th2 cells and that the hypomethylation of IL4 may facilitate the differentiation of these cells. Furthermore, considering that there are many kinds of cells in the lungs of mice and that the experimental conditions in this reported study did not allow for the specific identification of a cellular subpopulation, the question of whether the altered methylation status is an artifact of differential cellular composition should be addressed in future studies.

Conclusions

Collectively, our findings show that maternal NO2 exposure resulted in eosinophil-associated airway inflammation and Th2 polarization in the offspring; additionally, these effects were reversed in the absence of subsequent allergen provocation. However, the symptoms of allergy asthma following OVA sensitization and subsequent challenge were enhanced in offspring from dams exposed to NO2. These effects were associated with a Th2-biased response and DNA demethylation of the IL4 gene promoter in the offspring (Figure 11).

Conceptual diagram illustrating changes in pregnant mice when expose to nitrogen dioxide.
Figure 11. An illustration summarizing the relationship among maternal NO2 exposure, enhancement of the allergic asthma syndrome in BALB/c offspring, and the associated Th2 polarization and DNA methylation at the IL4 promoter.

Acknowledgments

This study was supported by National Science Foundation of China (No. 21477070, 21377076, 21222701, 91543203, 21425731, and 21637004), the Research Project for Young Sanjin Scholarship of Shanxi, the Program for the Outstanding Innovative Teams of Higher Learning Institutions of Shanxi, the Research Project Supported by Shanxi Scholarship Council (No. 2015-006), the National “973” Program (No. 2014CB932000), and the Strategic Priority Research Program of the Chinese Academy of Sciences (No. XDB14000000).

References

ATSDR (Agency for Toxic Substances and Disease Registry). 2014. Environmental Triggers of Asthma. https://www.atsdr.cdc.gov/csem/csem.asp?csem=32&po=6 [accessed: 9 August 2016].

Barletta KE, Cagnina RE, Wallace KL, Ramos SI, Mehrad B, Linden J. 2012. Leukocyte compartments in the mouse lung: distinguishing between marginated, interstitial, and alveolar cells in response to injury. J Immunol Methods 375(1-2):100–110, PMID: 21996427, 10.1016/j.jim.2011.09.013.

Bégin P, Nadeau KC. 2014. Epigenetic regulation of asthma and allergic disease. Allergy Asthma Clin Immunol 10(1):27, PMID: 24932182, 10.1186/1710-1492-10-27.

Cao L, Wang J, Tseu I, Luo D, Post M. 2009. Maternal exposure to endotoxin delays alveolarization during postnatal rat lung development. Am J Physiol Lung Cell Mol Physiol 296(5):L726–L737, PMID: 19218354, 10.1152/ajplung.90405.2008.

Clark NA, Demers PA, Karr CJ, Koehoorn M, Lencar C, Tamburic L, et al. 2010. Effect of early life exposure to air pollution on development of childhood asthma. Environ Health Perspect 118(2):284–290, PMID: 20123607, 10.1289/ehp.0900916.

Deng Q, Lu C, Li Y, Sundell J, Norbäck D. 2016b. Exposure to outdoor air pollution during trimesters of pregnancy and childhood asthma, allergic rhinitis, and eczema. Environ Res 150:119–127, PMID: 27281689, 10.1016/j.envres.2016.05.050.

Deng Q, Lu C, Norbäck D, Bornehag CG, Zhang Y, Liu W, et al. 2015. Early life exposure to ambient air pollution and childhood asthma in China. Environ Res 143(Pt A):83–92, PMID: 26453943, 10.1016/j.envres.2015.09.032.

Deng Q, Lu C, Ou C, Chen L, Yuan H. 2016a. Preconceptional, prenatal and postnatal exposure to outdoor and indoor environmental factors on allergic diseases/symptoms in preschool children. Chemosphere 152:459–467, PMID: 27003368, 10.1016/j.chemosphere.2016.03.032.

Ezratty V, Guillossou G, Neukirch C, Dehoux M, Koscielny S, Bonay M, et al. 2014. Repeated nitrogen dioxide exposures and eosinophilic airway inflammation in asthmatics: a randomized crossover study. Environ Health Perspect 122(8):850–855, PMID: 24747297, 10.1289/ehp.1307240.

Ford AQ, Dasgupta P, Mikhailenko I, Smith EM, Noben-Trauth N, Keegan AD. 2012. Adoptive transfer of IL–4Rα+ macrophages is sufficient to enhance eosinophilic inflammation in a mouse model of allergic lung inflammation. BMC Immunol 13:6, PMID: 22292924, 10.1186/1471-2172-13-6.

Froidure A, Mouthuy J, Durham SR, Chanez P, Sibille Y, Pilette C. 2016. Asthma phenotypes and IgE responses. Eur Respir J 47(1):304–319, PMID: 26677936, 10.1183/13993003.01824-2014.

Gaffin JM, Kanchongkittiphon W, Phipatanakul W. 2014. Reprint of: Perinatal and early childhood environmental factors influencing allergic asthma immunopathogenesis. Int Immunopharmacol 23(1):337–346, PMID: 25308874, 10.1016/j.intimp.2014.09.028.

Gruzieva O, Xu CJ, Breton CV, Annesi-Maesano I, Antó JM, Auffray C, et al. 2017. Epigenome-wide meta-analysis of methylation in children related to prenatal NO2 air pollution exposure. Environ Health Perspect 125(1):104–110, PMID: 27448387, 10.1289/EHP36.

Guarnieri M, Balmes JR. 2014. Outdoor air pollution and asthma. Lancet 383(9928):1581–1592, PMID: 24792855, 10.1016/S0140-6736(14)60617-6.

Harb H, Renz H. 2015. Update on epigenetics in allergic disease. J Allergy Clin Immunol 135(1):15–24, PMID: 25567039, 10.1016/j.jaci.2014.11.009.

Herriges M, Morrisey EE. 2014. Lung development: orchestrating the generation and regeneration of a complex organ. Development 141(3):502–513, PMID: 24449833, 10.1242/dev.098186.

Holgate ST. 2012. Innate and adaptive immune responses in asthma. Nat Med 18(5):673–683, PMID: 22561831, 10.1038/nm.2731.

Hwang YH, Paik MJ, Yee ST. 2017. Diisononyl phthalate induces asthma via modulation of Th1/Th2 equilibrium. Toxicol Lett 272:49–59, PMID: 28300662, 10.1016/j.toxlet.2017.03.014.

Ji X, Han M, Yun Y, Li G, Sang N. 2015. Acute nitrogen dioxide (NO2) exposure enhances airway inflammation via modulating Th1/Th2 differentiation and activating JAK-STAT pathway. Chemosphere 120:722–728, PMID: 25462318, 10.1016/j.chemosphere.2014.10.039.

Jeon H, Vigdorovich V, Garrett-Thomson SC, Janakiram M, Ramagopal UA, Abadi YM, et al. 2014. Structure and cancer immunotherapy of the B7 family member B7x. Cell Rep 9(3):1089–1098, PMID: 25437562, 10.1016/j.celrep.2014.09.053.

Khreis H, Kelly C, Tate J, Parslow R, Lucas K, Nieuwenhuijsen M. 2017. Exposure to traffic-related air pollution and risk of development of childhood asthma: a systematic review and meta-analysis. Environ Int 100:1–31, PMID: 27881237, 10.1016/j.envint.2016.11.012.

Kim HY, DeKruyff RH, Umetsu DT. 2010. The many paths to asthma: phenotype shaped by innate and adaptive immunity. Nat Immunol 11(7):577–584, PMID: 20562844, 10.1038/ni.1892.

Kwon NH, Kim JS, Lee JY, Oh MJ, Choi DC. 2008. DNA methylation and the expression of IL-4 and IFN–γ promoter genes in patients with bronchial asthma. J Clin Immunol 28(2):139–146, PMID: 18004650, 10.1007/s10875-007-9148-1.

Lambrecht BN, Hammad H. 2015. The immunology of asthma. Nat Immunol 16(1):45–56, PMID: 25521684, 10.1038/ni.3049.

Lee JW, Jaffar Z, Pinkerton KE, Porter V, Postma B, Ferrini M, et al. 2015. Alterations in DNA methylation and airway hyperreactivity in response to in utero exposure to environmental tobacco smoke. Inhal Toxicol 27(13):724–730, PMID: 26525176, 10.3109/08958378.2015.1104402.

Li N, Wang M, Bramble LA, Schmitz DA, Schauer JJ, Sioutas C, et al. 2009. The adjuvant effect of ambient particulate matter is closely reflected by the particulate oxidant potential. Environ Health Perspect 117(7):1116–1123, PMID: 19654922, 10.1289/ehp.0800319.

Lloyd CM, Hessel EM. 2010. Functions of T cells in asthma: more than just T(H)2 cells. Nat Rev Immunol 10(12):838–848, PMID: 21060320, 10.1038/nri2870.

Loutsios C, Farahi N, Porter L, Lok LS, Peters AM, Condliffe AM, et al. 2014. Biomarkers of eosinophilic inflammation in asthma. Expert Rev Respir Med 8(2):143–150, PMID: 24460178, 10.1586/17476348.2014.880052.

Lovinsky-Desir S, Miller RL. 2012. Epigenetics, asthma, and allergic diseases: a review of the latest advancements. Curr Allergy Asthma Rep 12(3):211–220, PMID: 22451193, 10.1007/s11882-012-0257-4.

Manners S, Alam R, Schwartz DA, Gorska MM. 2014. A mouse model links asthma susceptibility to prenatal exposure to diesel exhaust. J Allergy Clin Immunol 134(1):63–72, PMID: 24365139, 10.1016/j.jaci.2013.10.047.

Martino D, Prescott S. 2011. Epigenetics and prenatal influences on asthma and allergic airways disease. Chest 139(3):640–647, PMID: 21362650, 10.1378/chest.10-1800.

Morales E, Garcia-Esteban R, de la Cruz OA, Basterrechea M, Lertxundi A, de Dicastillo MD, et al. 2015. Intrauterine and early postnatal exposure to outdoor air pollution and lung function at preschool age. Thorax 70(1):64–73, PMID: 25331281, 10.1136/thoraxjnl-2014-205413.

Morgenstern V, Zutavern A, Cyrys J, Brockow I, Koletzko S, Krämer U, et al. 2008. Atopic diseases, allergic sensitization, and exposure to traffic-related air pollution in children. Am J Respir Crit Care Med 177(12):1331–1337, PMID: 18337595, 10.1164/rccm.200701-036OC.

Mortimer K, Neugebauer R, Lurmann F, Alcorn S, Balmes J, Tager I. 2008. Air pollution and pulmonary function in asthmatic children: effects of prenatal and lifetime exposures. Epidemiology 19(4):550–557, PMID: 18520616, 10.1097/EDE.0b013e31816a9dcb.

Muehling LM, Lawrence MG, Woodfolk JA. 2017. Pathogenic CD4+ T cells in patients with asthma. J Allergy Clin Immun, PMID: 28442213, 10.1016/j.jaci.2017.02.025.

Niedzwiecki M, Zhu H, Corson L, Grunig G, Factor PH, Chu S, et al. 2012. Prenatal exposure to allergen, DNA methylation, and allergy in grandoffspring mice. Allergy 67(7):904–910, PMID: 22583153, 10.1111/j.1398-9995.2012.02841.x.

Nordling E, Berglind N, Melén E, Emenius G, Hallberg J, Nyberg F, et al. 2008. Traffic-related air pollution and childhood respiratory symptoms, function and allergies. Epidemiology 19(3):401–408, PMID: 18379426, 10.1097/EDE.0b013e31816a1ce3.

Oeser K, Maxeiner J, Symowski C, Stassen M, Voehringer D. 2015. T cells are the critical source of IL-4/IL-13 in a mouse model of allergic asthma. Allergy 70(11):1440–1449, PMID: 26214396, 10.1111/all.12705.

Padula AM, Mortimer KM, Tager I. 2010. The causal effects of prenatal air pollution on pulmonary function in children with asthma: an example of causal inference methods in the FACES LiTE cohort [Abstract]. Am Thorac Soc 181:A3311.

Penn AL, Rouse RL, Horohov DW, Kearney MT, Paulsen DB, Lomax L. 2007. In utero exposure to environmental tobacco smoke potentiates adult responses to allergen in BALB/c mice. Environ Health Perspect 115(4):548–555, PMID: 17450223, 10.1289/ehp.9780.

Peters JL, Boynton-Jarrett R, Sandel M. 2013. Prenatal environmental factors influencing IgE levels, atopy and early asthma. Curr Opin Allergy Clin Immunol 13(2):187–192, PMID: 23385288, 10.1097/ACI.0b013e32835e82d3.

Poynter ME, Persinger RL, Irvin CG, Butnor KJ, van Hirtum H, Blay W, et al. 2006. Nitrogen dioxide enhances allergic airway inflammation and hyperresponsiveness in the mouse. Am J Physiol Lung Cell Mol Physiol 290(1):L144–L152, PMID: 16085673, 10.1152/ajplung.00131.2005.

Rabe KF, Calhoun WJ, Smith N, Jimenez P. 2011. Can anti-IgE therapy prevent airway remodeling in allergic asthma? Allergy 66(9):1142–1151, PMID: 21645010, 10.1111/j.1398-9995.2011.02617.x.

Ranzi A, Porta D, Badaloni C, Cesaroni G, Lauriola P, Davoli M, et al. 2014. Exposure to air pollution and respiratory symptoms during the first 7 years of life in an Italian birth cohort. Occup Environ Med 71(6):430–436, PMID: 24659182, 10.1136/oemed-2013-101867.

Romagnani S. 1994. Regulation of the development of type 2 T-helper cells in allergy. Curr Opin Immunol 6(6):838–846, PMID: 7710707.

Sharma P, Sharma A, Vishwakarma AL, Agnihotri PK, Sharma S, Srivastava M. 2016. Host lung immunity is severely compromised during tropical pulmonary eosinophilia: role of lung eosinophils and macrophages. J Leukoc Biol 99(4):619–628, PMID: 26489428, 10.1189/jlb.4A0715-309RR.

Singh SP, Gundavarapu S, Peña-Philippides JC, Rir-Sima-ah J, Mishra NC, Wilder JA, et al. 2011. Prenatal secondhand cigarette smoke promotes Th2 polarization and impairs goblet cell differentiation and airway mucus formation. J Immunol 187(9):4542–4552, PMID: 21930963, 10.4049/jimmunol.1101567.

Tang WY, Levin L, Talaska G, Cheung YY, Herbstman J, Tang D, et al. 2012. Maternal exposure to polycyclic aromatic hydrocarbons and 5′-CpG methylation of interferon – γ in cord white blood cells. Environ Health Perspect 120(8):1195–1200, PMID: 22562770, 10.1289/ehp.1103744.

Vroman H, van den Blink B, Kool M. 2015. Mode of dendritic cell activation: the decisive hand in Th2/Th17 cell differentiation. Implications in asthma severity? Immunobiology 220(2):254–261, PMID: 25245013, 10.1016/j.imbio.2014.09.016.

Wang P, You D, Saravia J, Shen H, Cormier SA. 2013. Maternal exposure to combustion generated PM inhibits pulmonary Th1 maturation and concomitantly enhances postnatal asthma development in offspring. Part Fibre Toxicol 10:29, PMID: 23856009, 10.1186/1743-8977-10-29.

Weir CH, Yeatts KB, Sarnat JA, Vizuete W, Salo PM, Jaramillo R, et al. 2013. Nitrogen dioxide and allergic sensitization in the 2005-2006 national health and nutrition examination survey. Respir Med 107(11):1763–1772, PMID: 24045117, 10.1016/j.rmed.2013.08.010.

WHO (World Health Organization). 2010. WHO Guidelines for Indoor Air Quality: Selected Pollutants. Geneva: WHO.

Wieland CW, Kerver ME, Florquin S, Nolte MA, Borst J, van Lier R, et al. 2006. CD27 contributes to the early systemic immune response to mycobacterium tuberculosis infection but does not affect outcome. Int Immunol 18(11):1531–1539, PMID: 16966496, 10.1093/intimm/dxl086.

Yang IV, Pedersen BS, Liu A, O’Connor GT, Teach SJ, Kattan M, et al. 2015. DNA methylation and childhood asthma in the inner city. J Allergy Clin Immunol 136(1):69–80, PMID: 25769910, 10.1016/j.jaci.2015.01.025.

You H, Chen S, Mao L, Li B, Yuan Y, Li R, et al. 2014. The adjuvant effect induced by di-(2-ethylhexyl) phthalate (DEHP) is mediated through oxidative stress in a mouse model of asthma. Food Chem Toxicol 71:272–281, PMID: 24953552, 10.1016/j.fct.2014.06.012.

Zacharasiewicz A. 2016. Maternal smoking in pregnancy and its influence on childhood asthma. ERJ Open Res 2(3), PMID: 27730206, 10.1183/23120541.00042-2016.

Zhang J, Ma X, Wang H, Ma D, Huang G. 2014. Elevated methylation of the RXRA promoter region may be responsible for its downregulated expression in the myocardium of patients with TOF. Pediatr Res 75(5):588–594, PMID: 24513686, 10.1038/pr.2014.17.

Zhou W, Tian D, He J, Wang Y, Zhang L, Cui L, et al. 2016. Repeated PM2.5 exposure inhibits BEAS-2B cell P53 expression through ROS-Akt-DNMT3B pathway-mediated promoter hypermethylation. Oncotarget 7(15):20691–20703, PMID: 26942697, 10.18632/oncotarget.7842.

Zissler UM, Esser-von Bieren J, Jakwerth CA, Chaker AM, Schmidt-Weber CB. 2016. Current and future biomarkers in allergic asthma. Allergy 71(4):475–494, PMID: 26706728, 10.1111/all.12828.

Randomized Trial of Interventions to Improve Childhood Asthma in Homes with Wood-burning Stoves

Author Affiliations open

1School of Public and Community Health Sciences, University of Montana, Missoula, Montana, USA

2Community Medical Center, Missoula, Montana, USA

3Department of Statistics, University of Kentucky, Lexington, Kentucky, USA

4Department of Mathematical Sciences, University of Montana, Missoula, Montana, USA

PDF icon PDF Version (547 KB)

  • Background:
    Household air pollution due to biomass combustion for residential heating adversely affects vulnerable populations. Randomized controlled trials to improve indoor air quality in homes of children with asthma are limited, and no such studies have been conducted in homes using wood for heating.
    Objectives:
    Our aims were to test the hypothesis that household-level interventions, specifically improved-technology wood-burning appliances or air-filtration devices, would improve health measures, in particular Pediatric Asthma Quality of Life Questionnaire (PAQLQ) scores, relative to placebo, among children living with asthma in homes with wood-burning stoves.
    Methods:
    A three-arm placebo-controlled randomized trial was conducted in homes with wood-burning stoves among children with asthma. Multiple preintervention and postintervention data included PAQLQ (primary outcome), peak expiratory flow (PEF) monitoring, diurnal peak flow variability (dPFV, an indicator of airway hyperreactivity) and indoor particulate matter (PM) PM2.5.
    Results:
    Relative to placebo, neither the air filter nor the woodstove intervention showed improvement in quality-of-life measures. Among the secondary outcomes, dPFV showed a 4.1 percentage point decrease in variability [95% confidence interval (CI)=−7.8 to −0.4] for air-filtration use in comparison with placebo. The air-filter intervention showed a 67% (95% CI: 50% to 77%) reduction in indoor PM2.5, but no change was observed with the improved-technology woodstove intervention.
    Conclusions:
    Among children with asthma and chronic exposure to woodsmoke, an air-filter intervention that improved indoor air quality did not affect quality-of-life measures. Intent-to-treat analysis did show an improvement in the secondary measure of dPFV.
    Trial registration:
    ClincialTrials.gov NCT00807183. https://doi.org/10.1289/EHP849
  • Received: 21 July 2016
    Revised: 13 June 2017
    Accepted: 16 June 2017
    Published: 13 September 2017

    Address correspondence to C.W. Noonan, School of Public and Community Health Sciences, 32 Campus Dr., University of Montana, Missoula, MT 59812 USA. Telephone: (406) 243-4957; Fax: (406) 243-2807. Email: curtis.noonan@umontana.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP849).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (128 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (150 KB)

Introduction

Several studies of children with asthma in urban environments have targeted the household environment to improve health outcomes (Bryant-Stephens 2009; Butz et al. 2011; Eggleston et al. 2005; Krieger et al. 2005; Levy et al. 2006; Morgan et al. 2004; Sweet et al. 2014). These intervention studies have targeted asthma triggers by employing education strategies and/or directly mitigating exposure to allergens through pest management and/or use of allergen-impermeable mattress and pillow encasings. Three of these studies incorporated air-filter devices in the intervention strategy to reduce children’s exposure to airborne allergens, dust, or environmental tobacco smoke (Butz et al. 2011; Eggleston et al. 2005; Morgan et al. 2004). The intervention strategies consistently demonstrated declines in asthma-related symptom measures. However, all of these studies were conducted in urban environments, and the nature of these exposures likely differ from that of exposures in rural environments.

Attention to home heating sources in asthma intervention trials has been limited. One randomized trial showed lower frequency of asthma symptoms and health care visits compared with controls when improved heating-source interventions were employed to increase indoor temperature and decrease NO2 concentrations (Howden-Chapman et al. 2008). To our knowledge, no randomized controlled trials among children with asthma have been conducted that target indoor particulate matter (PM) exposures in rural-area homes heated by residential woodstoves.

Biomass combustion is known to be an important contributor to household air pollution in developing countries with resultant adverse health effects (Smith et al. 2014). The exposure scenarios and health consequences related to biomass combustion in developing-country settings are distinct from those in high-income countries. Nevertheless, burning wood for residential heating is an important PM exposure source in many developed countries, particularly in rural settings. The U.S. Department of Energy estimates that there are 11.6 million homes that use wood as a primary or secondary heating fuel (U.S. Department of Energy 2009), translating to more than three million children living in woodstove-heated homes (Noonan et al. 2015). Estimates of the contribution of wood-burning to ambient air quality can vary widely (Naeher et al. 2007), but woodsmoke accounts for 80– 90% of the ambient PM concentrations in communities with a high proportion of wood-burning households (Johnston et al. 2013; McGowan et al. 2002; Ward and Lange 2010; Ward et al. 2006). In the European Union, it is estimated that domestic woodstoves will be the dominant source of ambient fine PM (PM2.5), accounting for 38% of all emissions by 2020 (Sigsgaard et al. 2015).

Indoor PM2.5 concentrations in homes that heat with wood often exceed the World Health Organization daily ambient air-quality standard of 25 μg/m3 (Noonan et al. 2012; Semmens et al. 2015; Ward et al. 2011), likely the result of indoor PM sources and infiltration of ambient PM from local sources (Allen et al. 2003; Barn et al. 2008; Hystad et al. 2009; Semmens et al. 2015). Few exposure-reduction interventions have been tested in such wood-burning households (Allen et al. 2011; Noonan et al. 2012; Ward et al. 2011), and none has been evaluated in the context of impact on children’s health.

We report here results from a randomized controlled trial to evaluate interventions targeting biomass smoke PM from older-model residential woodstoves in homes of children with asthma. Older-model stoves are models defined as not being certified by the U.S. Environmental Protection Agency (EPA), or those that have emissions of >4.1 grams/hour of PM for a catalytic stove or 7.5 grams/hour for a noncatalytic stove. The Asthma Randomized Trial of Indoor Wood Smoke (ARTIS) took place in rural areas of Montana, Idaho, and Alaska, where residential wood combustion is a major source of ambient and indoor PM2.5 and the primary source of home heating during coldtemperature periods. Our aims were to test the hypothesis that (a) improved-technology wood-burning stoves or (b) air-filtration units would result in improvements, relative to placebo, in asthma measures among children in participating homes.

Methods

Study Overview

The ARTIS study design has been previously described (Noonan and Ward 2012). Briefly, ARTIS was a three-arm randomized placebo-controlled intervention trial with two intervention strategies for reducing in-home woodsmoke PM (Figure 1). Eligible participants were identified by parent-response screening questionnaire (Magzamen et al. 2005) and included children with asthma, age 6–18 y, residing in a non-tobacco-smoking household that used an older-model woodstove as their primary source of heating. If a household had more than one eligible child with asthma, the child with more severe asthma based on screening questions was designated as the primary participant in the household, but all eligible, consented children with asthma were included in the analyses. Overall, ARTIS was conducted over five years. Each household participated in two consecutive winter periods with household interventions occurring between the two winter periods (Figure 2). Homes were randomly assigned to one of three treatments: the woodstove-intervention group receiving improved-technology wood-burning appliances (i.e., EPA-certified woodstoves), the air-filter group receiving functioning air-filtration devices, and the placebo group receiving sham air-filtration devices. Detailed specifications of the intervention treatments can be found in Supplemental Material. This study was approved by the University of Montana Institutional Review Board. Participating children completed documented assent procedures, and a parent or guardian signed parental permission forms.

Flow chart

Figure 1. Trial profile. Note: The woodstove changeout arm (W) was discontinued prior to recruiting the final cohort of homes.

Graphical representation showing study locations according to years of participation (Winter No. 1, 2008-2009; Winter No. 2, 2009-2010; Winter No. 3, 2010-2011; Winter No. 4, 2011-2012; and Winter No. 5, 2012-2013), and the corresponding sample sizes for the three arms. The study locations are as follows: Hamilton, MT [F (n equals 4), W (n equals 5), and P (n equals 6)]; Missoula, MT; Nez Perce, ID [F (n equals 9), W (n equals 10), and P (n equals 10)]; Butte, MT; Fairbanks, AK [F (n equals 5), W (n equals 4), and P (n equals 5)]; and Western MT [F (n equals 25) and P (n equals 24)].

Figure 2. Study locations according to years of participation. Note: The woodstove changeout arm (W) was discontinued prior to recruiting the final cohort of homes. F, air filter arm; W, woodstove changeout arm; P, placebo arm.

Health Measures

The primary health outcome was change in the score on the Pediatric Asthma Quality of Life Questionnaire (PAQLQ) administered directly to children at each visit. The PAQLQ is a 23-item asthma-specific battery that provides an overall score as well as domain scores for symptoms (10 items), activity limitation (5 items), and emotional function (8 items) (Juniper et al. 1996). The instrument requires the participating children to reflect on the experiences over the prior week; thus, it is a one-week retrospective self-assessment (Figure 3). This instrument has strong correlation, both longitudinal and cross-sectional, with other measures of asthma severity, such as peak flow monitoring, medication use, and symptom diaries, among children with asthma ages 7 through 17 y (Juniper et al. 1996). The score for each domain is the average of the items in it, with scores ranging from 1 to 7 (with 7 as the optimal score), and the overall PAQLQ score was the mean score across the three domains. The primary hypothesis was that the preintervention to postintervention change in PAQLQ scores would be greater in the woodstove changeout or air filter arms relative to placebo. Our a priori power estimates indicated that when comparing two groups, a sample size of 30 subjects per group would provide 80% power to demonstrate a difference in PAQLQ scores of 1.0 unit or greater at the 95% confidence level (CI).

Timeline plotting PAQLQ assessment on Day 1, reflecting on past week; the indoor sampling for Day 1, Day 2, and Day 3; and twice daily PEF and FEV sub 1 monitoring for two weeks, from Day 1 to Day 14.
Figure 3. Schematic of sampling visit. Each visit occurred multiple times during each of two winter periods, a preintervention winter and a postintervention winter. For a given visit, study personnel administered the PAQLQ to participants. The PAQLQ is a self-assessment questionnaire reflecting on the child’s past week of experiences. At the sampling visit, children are trained in the use of the expiratory flow monitor and begin recording twice-daily measures of PEF and FEV1 for two weeks. Indoor air monitoring was initiated on Day 0 and continued for 48 h. Note: FEV1, forced expiratory volume in the 1st second; PAQLQ, pediatric asthma quality of life questionnaire; PEF, peak expiratory flow.

Secondary outcomes included home-based peak flow monitoring. Parents and children were trained to collect twice-daily measures (in the morning and in the evening) of peak expiratory flow (PEF) and forced expiratory volume in the first second (FEV1) using the PiKo-1 Peak Flow Meter (Ferraris). These measures were collected and logged during a two-week period beginning with the first day of an exposure sampling visit. For each measure, the child performed a peak flow maneuver three times, and the device stored the best result for PEF and FEV1. Outcomes from these two-week recordings included average percent predicted morning PEF and FEV1, average percent predicted evening PEF and FEV1, and average diurnal PEF variability (dPFV). Daily dPFV was the amplitude as a percentage of the mean of consecutive evening to morning readings, or (previous night PEF–morning PEF)/(mean of the two measures) (Lebowitz et al. 1997). Whereas lower percent predicted PEF and FEV1 are considered worse outcomes, the opposite is true for dPFV, where higher dPFV is considered a worse outcome for children with asthma.

Additional child data were collected to characterize baseline health. During each two-week sampling period, parents of participating children were also asked to record daily health-related events for their child, including daytime cough or wheezing, nighttime cough or wheezing, activity limitations due to asthma symptoms, and use of asthma medications. These data together with peak-flow monitoring data collected during the baseline (preintervention) winter were used to classify asthma severity based on the 2007 National Heart Lung and Blood Institute (NHLBI) guidelines (National Heart Lung and Blood Institute 2007). For methods used to translate these data to the 2007 NHLBI asthma severity classifications and for overall severity classification of participants and severity classification according to specific components, see Supplemental Material and Table S1.

Exposure Assessment

The exposure and health assessments took place on two, 48-h observation periods during each of two winter periods, before and after intervention. A DustTrak® Aerosol Monitor 8520/8530 (TSI) was used to continuously measure PM2.5 concentrations that were corrected for woodsmoke (McNamara et al. 2011). A Lighthouse 3016-IAQ particle counter (Lighthouse Worldwide Solutions) was used to continuously measure particle number concentrations of varying size fractions (0.30–0.49, 0.50–0.99, 1.00–2.49, 2.5–5.0, and 5.0–10.0, 10.0+μm). Particle number concentrations (PNCs), reported as the number of particles per cubic centimeter, were summed across the size fractions 2.5–10.0 μm and were used as a surrogate for concentrations of coarse PM (PMc). Samplers had 60-s recording intervals, and were zero calibrated prior to each sampling event. We included only those 48-h averages that were generated from data that were at least 80% complete to ensure that the averages were representative of concentrations experienced during the entire sampling events. Instrument malfunctions (e.g., flow errors) or power failures were the primary reasons for sampling events with less than 100% air-sampling data capture.

Prior to sampling, participants completed demographics and home characteristics surveys. These surveys documented information such as household income, education, type/age/size of home, age of woodstove and specific activities that occurred in the home during the 48-h sampling periods.

Statistical Analysis

We used analysis of variance or chi-square tests, as appropriate, to compare differences between treatment groups in descriptive characteristics such as age, sex, race/ethnicity, and baseline health measures. We followed the principle of intent-to-treat and included all participants who were randomized to a treatment arm and had at least one baseline measure. The objective of our statistical analysis was to evaluate whether intervention assignment modified the preintervention-to-postintervention winter change in PAQLQ or peak flow measure. As such, the linear mixed effects model included fixed effects for a) winter (i.e., preintervention or postintervention winter), b) intervention assignment (e.g., air filter), and c) a multiplicative interaction term between these two variables to assess effect modification by intervention assignment. These models also included fixed effects for age and gender. Initially, random effects were included for both the intercept and the slope (i.e., winter), which in effect allowed each participant to have his or her own intercept and slope describing change in asthma measure from the preintervention to postintervention winter. However, there were problems with estimation of the model when we included a random effect for the slope (i.e., winter). As a result, our final model included a random effect only for the intercept. Winters were nested within participants, and participants were nested within homes. The linear mixed-model analysis amounts to adopting likelihood-based available-case analysis for handling missing data. In post-hoc sensitivity analyses, we examined the impact on results when we adjusted for additional covariates, including baseline asthma measures, use of asthma medications, and ambient temperature because these factors could influence both the use of the woodstove and asthma measures. We also performed separate analyses for each level of baseline asthma severity. All intent-to-treat analyses were conducted using SAS (version 9.3; SAS Institute Inc.).

The interventions utilized in this study were designed to improve various measures of childhood asthma by improving indoor air quality. Specifically, the objective was to lower concentrations of PM2.5 in homes. To assess the extent to which PM reductions contributed to intervention efficacy, we added both natural log-transformed PM2.5 and coarse PM as time-dependent variables to our primary models and examined potential attenuation of efficacy of the interventions. In addition, to examine further whether impacts on health were consistent with changes in PM, we used linear mixed models to assess the effect of PM2.5 concentrations on PAQLQ scores and peak flow measures. These post-hoc analyses were no longer based on a randomized design and were adjusted for treatment assignment, gender, and age. PM2.5 concentrations were skewed and were natural log-transformed. Thus, results are presented as the change in PAQLQ or peak flow measure associated with a doubling of PM2.5.

Results

Almost 7,000 recruitment surveys were administered and follow-up contact with the 408 potentially eligible households resulted in 115 recruited children with asthma living in 98 eligible households (Figure 1). Insufficient baseline data were collected for one participant to classify according to asthma severity, and they were not included in further analysis. Among the 114 remaining participants, health and exposure data were captured at 422 overall visits (226 during the preintervention winter and 196 during the postintervention winter). Prior to enrollment of the final cohort of homes, the woodstove intervention arm was discontinued as interim analyses indicated that the woodstove change-out was not efficacious in reducing indoor PM2.5 (Ward et al. 2017). Thus, the sample size in the air filter and placebo arms were approximately twice that of the woodstove arm.

Descriptive and baseline health characteristics by assigned treatment arm are shown in Table 1. The overall mean [standard deviation (SD)] age was 12.4 y (3.0), and 48.3% of participants were female. Age, gender, and body mass index were similar across treatment arms. Most participants were white (82%) and non-Hispanic (95%), but race and ethnicity were not reported by parents for seven children. Among those with available race data, 4 of the 41 participants assigned to the air-filter arm were American Indian/Alaskan Native (AI/AN), and no participants reported being of AI/AN race in the other two arms. In addition, 101 of the 114 participants (89%) were classified as having moderate to severe persistent asthma (Table 1 and Table S1). A majority of caregivers (57%) reported household incomes greater than $40,000 per year, and 72% reported having some postsecondary education. The majority of homes (87%) reported having a dog or cat. These factors were similar by treatment arm assignment. During the preintervention winter, overall median PM2.5 concentrations were 17.5 μg/m3, and were higher, relative to placebo, in homes assigned to the woodstove changeout arm of the trial. Median PMc was 0.32 particles per cm3 and similar in all treatment arms.

Table 1. Baseline participant and household characteristics.
Participant characteristics Woodstove changeout (n=22) Air filter (n=46) Placebo (n=46)
Age, mean years (SD) 12.3 (3.1) 12.7 (3.3) 12.2 (2.5)
Body mass index percentile, mean (SD) 67.6 (28.4) 69.7 (27.5) 62.4 (30.1)
Female sex, n (%) 10 (45.5) 25 (54.5) 20 (43.5)
Race, n (%)a
 American Indian/Alaskan Native 0 (0.0) 4 (9.8) 0 (0.0)
 White 16 (76.2) 32 (78.1) 40 (88.9)
 Other 5 (23.8) 5 (12.2) 5 (11.1)
Non-Hispanic ethnicity, n (%)a 20 (95.2) 39 (95.1) 43 (95.6)
Household income ≥$40,000, n (%)a 12 (57) 20 (50) 28 (62)
Household post-secondary education, n (%)a 15 (75) 29 (74) 31 (69)
Dogs or cats in home, n (%) 17 (77) 39 (85) 41 (89)
Baseline (preintervention) Health Measures
 Asthma Severity, n (%)
  Intermittent or Mild 2 (9.1) 5 (10.9) 6 (13.0)
  Moderate or Severe 20 (90.9) 41 (89.1) 40 (87.0)
 PAQLQ, mean (SD)
  Overall 5.1 (1.0) 5.4 (1.0) 5.4 (1.0)
  Symptoms 4.9 (1.2) 5.3 (1.0) 5.3 (1.1)
  Limitation of Activity 4.6 (1.3) 5.0 (1.1) 5.0 (1.3)
  Emotion 5.7 (0.8) 5.6 (1.1) 5.7 (1.0)
 Two-week spirometry monitoring, mean (SD)
  Evening FEV1 percent predicted 75.3 (19.0) 87.2 (18.1) 88.3 (24.9)
  Morning FEV1 percent predicted 73.1 (19.4) 86.2 (18.2) 87.5 (24.4)
  Evening PEF percent predicted 72.4 (25.8) 84.9 (24.7) 85.1 (21.9)
  Morning PEF percent predicted 70.8 (25.6) 82.3 (22.8) 82.9 (23.0)
  Diurnal PEF variability (dPFV), % 18.5 (8.5) 17.0 (10.4) 14.4 (6.3)
 Two-week reporting of symptoms and medical usage
 Daytime cough/wheezing, days
  Mild 5.7 (3.8) 4.3 (3.2) 4.2 (3.1)
  Moderate/Severe 1.9 (2.9) 1.0 (1.5) 0.9 (1.6)
 Nighttime cough/wheezing, days
  Mild 3.2 (3.0) 2.6 (2.7) 3.3 (3.0)
  Moderate/Severe 2.9 (4.6) 0.8 (1.4) 0.9 (1.6)
 Altered activity/behavior, days
  Mild 3.8 (2.9) 2.0 (2.3) 2.1 (2.4)
  Moderate/Severe 1.3 (1.9) 0.8 (1.4) 1.0 (2.4)
Medication usage, doses
  Long-term control medications 4.4 (6.2) 7.5 (10.6) 4.0 (7.4)
  Quick relief medications 4.9 (7.1) 4.1 (7.9) 3.8 (6.4)

Note: FEV1, forced expiratory volume in the first second; PAQLQ, pediatric asthma quality of life; PEF, peak expiratory flow.

aRace/ethnicity, household income, and household education not reported for 7, 8, and 10 participant parents, respectively.

Among the baseline health measures, averaged over preintervention winter sampling visits, asthma severity and PAQLQ did not differ among the treatment arms (Table 1). Children in the woodstove changeout arm had a slightly higher proportion of participants experiencing moderate to severe asthma and lower mean PAQLQ scores. FEV1 and PEF percent predicted were also lower among participants in the woodstove changeout arm in comparison with the other study arms. Mean (SD) dPFV during the preintervention winter was 16.3% (11.1) across treatment arms.

Within each arm, preintervention to postintervention changes for PAQLQ and spirometry measures are presented in Table 2. For all treatment arms, the overall PAQLQ composite score, as well as domain scores for symptoms, activity limitation, and emotion, increased in magnitude by 0.2–0.7 on the seven-point scale from one winter to the next. For the spirometry measures, increases in FEV1 and PEF percent predicted measures would indicate improvement in lung function, whereas decreases in dPFV would also indicate improvement in lung function. In general, FEV1 and PEF percent predicted measures decreased (i.e., worsened) from preintervention to postintervention winters in both the air-filter and placebo arms. In contrast, dPFV change in the air-filter arm showed decreased variation, suggesting improved function with a percentage point change of −2.0 (95% (CI)=−4.7 to 0.7), or a 11.8% (95% CI: 27% to −4%) improvement in comparison with baseline in this arm.

Table 2. Preintervention to postintervention mean changes [95% confidence interval (CI)]a within arm and within intervention, relative to placebo (n=114 participants).
Outcome Observations Within group change Placebo Change relative to placebo
Woodstove changeout Air filter Woodstove changeout Air Filter
PAQLQ
 Overall 422 0.47 (0.04, 0.90) 0.22 (−0.07, 0.52) 0.29 (0.01, 0.58) 0.18 (−0.33, 0.69) −0.07 (−0.47, 0.34)
 Symptoms 422 0.61 (0.13, 1.1) 0.19 (−0.14, 0.52) 0.23 (−0.09, 0.55) 0.38 (−0.19, 0.95) −0.04 (−0.49, 0.41)
 Limitation of Activity 422 0.61 (0.09, 1.1) 0.23 (−0.12, 0.59) 0.48 (0.13, 0.82) 0.13 (−0.48, 0.74) −0.24 (−0.73, 0.24)
 Emotion 422 0.18 (−0.25, 0.61) 0.24 (−0.06, 0.53) 0.28 (−0.01, 0.56) −0.10 (−0.61, 0.41) −0.04 (−0.44, 0.36)
Two-week spirometry monitoring
 Evening FEV1 % predicted 406 −0.09 (−8.7, 8.5) −2.7 (−8.7, 3.1) −3.0 (−8.7, 2.6) 2.9 (−7.3, 13) 0.24 (−7.8, 8.3)
 Morning FEV1 % predicted 408 0.96 (−7.8, 9.8) −3.4 (−9.4, 2.7) −2.6 (−8.4, 3.1) 3.6 (−6.8, 14) −0.71 (−8.9, 7.5)
 Evening PEF % predicted 407 0.07 (−7.9, 8.0) −4.6 (−10, 0.98) −7.0 (−12, −1.7) 7.1 (−2.3, 16) 2.4 (−5.0, 9.9)
 Morning PEF % predicted 409 1.1 (−6.9, 9.2) −3.3 (−8.9, 2.3) −6.7 (−12, −1.4) 7.8 (−1.6, 17) 3.4 (−4.1, 11)
 Diurnal PEF variability, % 404 −0.8 (−4.7, 3.0) −2.0 (−4.7, 0.7) 2.2 (−0.4, 4.7) −3.0 (−7.6, 1.6) −4.1 (−7.8, −0.4)

Note: FEV1, forced expiratory volume in the first second; PAQLQ, pediatric asthma quality of life; PEF, peak expiratory flow.

aAdjusted for age and gender.

Efficacy analyses of preintervention to postintervention changes in PAQLQ and spirometry measures for treatment arms relative to placebo are presented in Table 2. Relative to the placebo arm, the air-filter intervention and the woodstove changeout showed no improvement in PAQLQ composite or domain scores. Mean change in PEF and FEV1 were generally better for the air-filter and woodstove changeout arms, in comparison with the placebo arm. For example, change (and 95% CI), relative to placebo, in morning percent predicted PEF were 3.4 (−4.1 to 11) and 7.8 (−1.6 to 17) for air-filter and woodstove changeout arms, respectively. Observations for dPFV showed lesser variability for the air-filter arm, with a 4.1 percentage point greater reduction in dPFV (95% CI=−7.8 to −0.4) from the preintervention to postintervention winters for the air-filter arm, relative to the placebo arm, suggesting improved airway function (Table 2). The woodstove changeout arm showed a 3.0 percentage point reduction for dPFV relative to placebo (95% CI=−7.6 to 1.6).

To further evaluate the influence of asthma severity status on intervention efficacy, we conducted stratified analysis for intervention changes to dPFV according to baseline asthma severity (Table 3). The highest point estimate for reduction in dPFV among the air-filter group was observed among the children with indications of severe asthma at baseline, but no differential effect in dPFV was evident when stratifying by severe or moderate asthma groups. When combining those children with severe and moderate asthma at baseline, the change in dPFV among the air-filter group relative to placebo was similar to the primary analysis (−4.6; 95% CI: −8.6 to −0.5). When including only primary participants, or the child in a given home with the more severe asthma symptoms, the change in dPFV among the air-filter group relative to placebo was lower (−3.2; 95% CI: −7.2 to 0.80; see Table S3). Additional sensitivity analyses, adjusting for baseline health measure and/or ambient temperature, did not change the effect estimates appreciably (see Table S2).

Table 3. Intervention effect on diurnal peak flow variability according to baseline asthma severity, relative to placebo, adjusted for age and gender.
Participant number Observations Woodstove changeout Air filter
Mean change (95% CI) Mean change (95% CI)
All participants 114 404 −3.0 (−7.6, 1.6) −4.1 (−7.8, −0.4)
 Severe 50 170 −1.4 (−9.3, 6.4) −4.7 (−12, 2.6)
 Moderate 51 190 −5.3 (−14, 3.3) −3.8 (−9.2, 1.6)
 Intermittent/mild 13 44 0.07 (−10, 10) −1.4 (−9.4, 10)

As shown previously, we observed a strong overall reduction in indoor PM2.5 concentrations (−67%; 95% CI: −77% to −50%) for the air-filter group relative to the placebo group, but we observed no overall change to indoor PM2.5 for the woodstove exchange (0.0%; 95% CI: −40%, 65%), relative to the placebo group (Ward et al. 2017). Figure 4 shows a smoothed distribution of log-transformed PM2.5 measures, indicating strong overlap in preintervention and postintervention observations. Distributions of preintervention and postintervention PM2.5 measures for the woodstove exchange and the placebo arms were similar, consistent with the previously reported findings of no change for each of these arms. We observed strong reductions in PMc in both the filter (72%) and placebo (57%) intervention arms (Ward et al. 2017).

Kernal density plot showing density of PM sub 2.5 (y-axis) for air filter, improved stove, and placebo across log PM sub 2.5 (x-axis) values observed preintervention and postintervention.
Figure 4. Kernal density plot of pre- (solid line) and postintervention (dashed line) 48-h indoor fine particulate matter (PM2.5) by treatment arm.

To further explore these changes to indoor air quality and impact on health outcomes, we conducted post hoc analyses. Inclusion of PM2.5 and PMc in our primary models attenuated the estimate of the efficacy of the air filter in improving dPFV, relative to placebo (Table 4). Intervention assignment continued to be unassociated with overall PAQLQ or the PAQLQ domain scores when PM measures were included in analyses. However, despite having no indication of efficacy for air-filter interventions on PAQLQ, our exposure response analysis did indicate that PAQLQ was associated with PM2.5 (Table 5). Because PM measures were natural log-transformed, we report changes in each response variable associated with a doubling of PM exposure. A doubling of PM2.5 concentration was associated with small declines in overall PAQLQ [−0.09 (95% CI: −0.16, −0.02)] and emotion domain [−0.10 (95% CI: −0.18, −0.03)] scores. PM2.5 was not associated with changes in most of our assessed spirometry measures. However, consistent with our intent-to-treat analysis, a doubling PM2.5 was associated with increased (i.e., worsening) dPFV, 0.64 (95% CI: −0.05, 1.34).

Table 4. Preintervention to postintervention mean changes (95% CI) for treatments relative to placebo, adjusted for age, gender and indoor particulate matter (n=114 participants).
Outcome Adjusted for age, gender, PM2.5a Adjusted for age, gender, PMcb Adjusted for age, gender, PM2.5, PMcb
Woodstove changeout Air filter Woodstove changeout Air filter Woodstove changeout Air filter
PAQLQ
 Overall 0.18 (−0.34, 0.71) −0.15 (−0.58, 0.27) 0.20 (−0.36, 0.77) −0.13 (−0.57, 0.32) 0.20 (−0.36, 0.77) −0.21 (−0.67, 0.24)
 Symptoms 0.38 (−0.20, 0.96) −0.13 (−0.60, 0.34) 0.39 (−0.23, 1.0) −0.11 (−0.60, 0.37) 0.38 (−0.23, 1.0) −0.20 (−0.70, 0.29)
 Limitation of Activity 0.13 (−0.50, 0.75) −0.27 (−0.78, 0.24) 0.16 (−0.52, 0.83) −0.29 (−0.82, 0.24) 0.16 (−0.52, 0.83) −0.31 (−0.85, 0.24)
 Emotion −0.10 (−0.63, 0.44) −0.14 (−0.57, 0.30) −0.06 (−0.64, 0.51) −0.08 (−0.53, 0.38) −0.06 (−0.64, 0.51) −0.19 (−0.66, 0.27)
Two-week spirometry monitoring
 Evening FEV1 % predicted 2.6 (−7.8, 13) −0.73 (−9.1, 7.7) 1.3 (−10, 13) −0.51 (−9.3, 8.3) 1.2 (−10, 13) −1.3 (−10, 7.8)
 Morning FEV1 % predicted 3.2 (−7.4, 14) −1.7 (−10, 6.9) 2.2 (−9.2, 14) −2.1 (−8.4, 4.2) 2.2 (−9.3, 14) −2.4 (−11, 6.7)
 Evening PEF % predicted 6.9 (−2.7, 17) 0.62 (−7.4, 8.6) 6.3 (−4.0, 16) 2.2 (−5.9, 10) 6.2 (−4.1, 17) 1.2 (−7.2, 9.7)
 Morning PEF % predicted 7.7 (−2.0, 17) 1.7 (−6.3, 9.7) 7.1 (−3.1, 17) 2.5 (−5.6, 11) 7.0 (−3.2, 17) 1.8 (−6.6, 10)
 Diurnal PEF variability, % −2.7 (−7.4, 2.0) −3.6 (−7.6, 0.4) −3.0 (−8.2, 2.3) −3.2 (−7.4, 1.0) −3.0 (−8.3, 2.3) −2.7 (−7.0, 1.7)

Note: FEV1 forces expiratory volume in first second; PAQLQ, Pediatric Asthma Quality of Life Questionnaire; PEF, peak expiratory flow.

aMissing observations for models with PM2.5: n=5 for PAQLQ measures; n=41 for spirometry measures.

bMissing observations for models with PMc: n=6 for PAQLQ measures; n=38 for spirometry measures.

Table 5. Effects of doubling fine fraction particulate matter (PM2.5) concentrations on quality of life measures and two-week spirometry monitoring, adjusting for age, gender, and intervention assignment.
Outcome Obs Estimate 95% CI
PAQLQ
 Overall 415 −0.09 (−0.16, −0.02)
 Symptoms 415 −0.08 (−0.16, 0.001)
 Limitation of activity activity 415 −0.07 (−0.16, 0.03)
 Emotion 415 −0.10 (−0.18, −0.03)
Spirometry
 Evening FEV1 398 0.55 (−0.79, 1.88)
 Morning FEV1 400 0.43 (−0.90, 1.77)
 Evening PEF 399 0.02 (−1.51, 1.55)
 Morning PEF 401 −0.20 (−1.68, 1.28)
 dPFV 396 0.64 (−0.05, 1.34)

Note: dPFV, diurnal peak flow variability; FEV1, forced expiratory volume in first second; Obs, observations; PAQLQ, pediatric asthma quality of life Questionnaire; PEF, peak expiratory flow.

Discussion

To our knowledge, this investigation is the first randomized controlled trial aimed at improving asthma symptoms in children by reducing indoor PM2.5 concentrations generated from woodstoves used for heating. Although the filter intervention resulted in the anticipated reductions in indoor PM2.5 concentrations, we did not observe treatment efficacy according to PAQLQ, our primary health outcome measure (Table 2). Given that this was a subjective metric measured over a two-winter period, it is possible that there was a learning effect due to repeated assessment. Indeed, we observed an improvement in the magnitude of overall PAQLQ score from the preintervention to postintervention winters in all treatment arms. Improvements also were observed in the symptom and activity limitation PAQLQ domains for the woodstove changeout arm and in the activity limitation PAQLQ domains for the placebo arm (Table 2). As in analyses of the overall PAQLQ score, no treatment effect with respect to placebo was observed for these domains. The preintervention to postintervention changes in PAQLQ were modest across all treatment arms with most point estimate changes less than one-half unit, whereas changes in this scale of 0.5 or more points have been translated as a clinically relevant outcome (Juniper et al. 1996).

Assignment to the filter arm was associated with strong reductions in indoor PM2.5 concentrations in this randomized trial (Ward et al. 2017). Reductions in coarse particles also were observed in the filter arm and, unexpectedly, in the placebo arm. This latter observation may explain, in part, why we did not demonstrate efficacy for PAQLQ in the air-filter arm, relative to the placebo arm. The mechanism for the observed improvement in overall PAQLQ score, as well as in scores in the PAQLQ symptom and activity limitation domains, in the woodstove arm is unclear. As demonstrated previously, homes assigned to this intervention experienced no improvement in PM2.5 concentrations or PNCs of any size; they did experience reductions in carbon monoxide concentrations, however (Ward et al. 2017).

Peak flow measures were the secondary health measures for this trial. We observed an improvement in dPFV, specifically a 4.1-percentage point reduction for children in the filter arm, relative to placebo. Of note, the filter arm had higher preintervention dPFV, relative to the placebo, indicating the randomization was not entirely successful in balancing asthma severity between groups. Thus, we cannot rule out completely the possibility that regression to the mean may explain, at least in part, the observed efficacy of the filter in improving dPFV. When analyses were limited to only those with moderate or severe persistent asthma, the improvement in dPFV variability was similar, perhaps even stronger for those with severe asthma. Home-based peak flow monitoring is not well supported as a tool for clinical decision making of asthma therapy (Kamps et al. 2001; Self et al. 2014; Yoos et al. 2002). There also are questions about the validity of dPFV as a clinical tool for asthmatic children, and correlation between measures of peak expiratory flow variation and asthma symptoms may wane over longer periods of observation (i.e., >1 year), suggesting that this measure may be less desirable as a tool for management of asthma in children (Brand et al. 1999). Nonetheless, dPFV does track with hyper-responsiveness, severity of symptoms, unscheduled medical visits, and response to bronchodilators (Brand et al. 1999; Brouwer et al. 2006; Greenberg et al. 2012; Mortimer et al. 2001), and change in dPFV has been used as a measure that reflects the dynamic nature of asthma (Lebowitz et al. 1997). There is also support for using dPFV and similar measures that reflect the natural diurnal variation of airway function and response to exposures and treatments in randomized controlled clinical trials (Frey and Bielicki 2017; Kaminsky et al. 2017). Here we use dPFV in repeated two-week snapshots to indicate short-term change and make no inferences regarding its utility as a self-management tool. Increased dPFV has been shown to correspond to bronchial responsiveness to nonspecific challenges (Cockcroft and Hargreave 1990; Hetzel and Clark 1980). More recently, dPFV among children with asthma was positively associated with indoor measures of specific fungi species (Bundy et al. 2009; Douwes et al. 2000).

Comparisons of PM and other air pollutant exposures with dPFV among children with asthma are limited and have indicated that pollutant effects on dPFV may be stronger among children with mild asthma. A study of children with mild asthma in Korea showed that dPFV increased during Asian dust events in comparison with control days, with corresponding strong correlations observed between PM10 concentrations and dPFV (Yoo et al. 2008). A panel study of children in France showed an association between ambient SO2 and dPFV, but only among those children with mild asthma; no associations were observed for ambient PM13 and dPFV (Segala et al. 1998). In contrast, the present study supports an association between smaller size fraction PM, specifically PM2.5, and our findings of efficacy with respect to dPFV remained robust, perhaps stronger, when limiting the analysis to children with moderate to severe asthma. Our observed response in dPFV among children with moderate to severe asthma was consistent with studies of medical intervention. Measures of PFV were found to change favorably in response to treatment with inhaled corticosteroids among children with moderate to severe asthma (Simons 1997; van Essen-Zandvliet et al. 1992), but not necessarily among children with mild asthma (Waalkens et al. 1991).

The beneficial impact of the air-filter intervention on dPFV was not corroborated with parallel findings in other pulmonary function measures, i.e., morning and evening percent predicted FEV1 and PEF. The within-group improvement in dPFV following the air-filter intervention also was fairly small (i.e., 2 percentage point decline). Nevertheless, this change did indicate a greater than 11% improvement in dPFV, relative to the baseline winter mean in this group. Our exposure response analysis also showed an association between elevated PM2.5 concentrations and higher dPFV, consistent with our overall hypothesis that the air-filter intervention improves health by improving air quality.

This study was relatively large, including more than 400 multiday exposure and health sampling visits among 114 children with asthma from various rural regions in western Montana, Idaho, and Alaska. Extensive information on air pollutant exposures, asthma measures, and relevant covariates was ascertained with the randomized, placebo-controlled design providing protection against confounding. However, we note several limitations. First, to the extent possible, participating filter and placebo homes were blinded to their treatment assignment. This blinding was not possible for the homes receiving the woodstove intervention. Moreover, field staff responsible for collecting exposure and health data were not blinded, as study protocol and efficient use of project resources required them to install intervention filter units, replace filters at prespecified intervals, and collect home-based health and exposure data. Quality-control procedures were employed to protect against any potential influence resulting from failures in blinding at the participant and investigator levels, but this circumstance remains a limitation when considered in the context of rigorous randomized controlled trial designs. Second, performance and recording of self-monitored PEF and FEV1 measures is highly effort dependent. Our procedures included training participants and parents on best-effort procedures. Although within-individual variation in effort is likely, we expect that this error would be random rather than systematic over the two-winter observation period, particularly when considering the most robust finding, change in dPFV, where the smallest unit of measure is percent difference in successive PEF measures in less than 24 h (i.e., evening to morning). Third, despite the randomized design, we observed higher preintervention PM2.5 concentrations and poorer baseline health in participants residing in households assigned to the woodstove changeout arm. This arm of the study was eliminated early due to evidence of failed efficacy with respect to reduction of indoor PM2.5 concentrations. Had this arm continued to full recruitment, it is possible that the baseline exposure and health measures would have balanced with the other arms. This change did result in larger numbers of participants allocated to the air-filter and placebo arms, making comparison between those arms more robust. Nevertheless, observations for the woodstove changeout arm relative to the placebo arm should be interpreted with caution. Fourth, the nature of the intervention strategy being tested in this study called for between-winter, rather than within-winter, evaluations of preintervention-to-postintervention health changes. Asthma health and response to environmental stimuli can change during a one-year period, particularly among adolescents who typically have worsening symptoms during these ages. Given the randomized design, we expect that such intra-individual changes would not be differential with respect to treatment arm. A crossover design was not possible in this study due to the woodstove changeout arm, but this approach may have been a more robust approach for evaluating within-winter filter versus placebo effects for children with asthma. Fifth, the efficacy of the air-filter intervention depended on the home residents operating the unit at the recommended setting. To assess compliance, we measured energy usage with Kilowatt meters attached at the filter units. As demonstrated previously, mean percent compliance (i.e., actual Kilowatt usage as a proportion of predicated Kilowatt usage) was 78%, and PM2.5 exposure reduction was robust to lower levels of compliance (Ward et al. 2017). Sixth, although we conducted a limited number of primary analyses, we performed nine primary statistical tests and might observe an association due to chance alone. As a result, we cannot rule out the possibility that the efficacy of the air filter in improving dPFV is simply due to chance. Finally, unlike patient-level clinical trials where compliant delivery of a drug or procedure accounts for 100% of the intended dose, we must recognize that environmental intervention studies account for some component of dose. Thus, despite substantial reductions in indoor PM2.5 concentration following the introduction of the filter, participating children were still exposed to measurable PM2.5.

Conclusions

Biomass combustion during residential heating is an important source of household air pollution, particularly in rural communities, and evidence-based strategies to improve indoor air quality and asthma outcomes in such settings are needed. This randomized trial targeted a vulnerable population, children with asthma who were chronically exposed to woodsmoke. The interventions tested did not result in improvements to quality-of-life measures, but the air-filtration device, a simple in-home intervention, was efficacious for improving the secondary measure dPFV, an indirect measure of airway hyper-responsiveness. The attenuation of the effect when including PM in the models as well as the exposure–response analysis provide evidence that filter-based PM reductions contribute, at least partially, to the observed improvements in the dPFV. This trial was conducted across several rural communities in three states, but translation of these findings to other settings with similarly exposed child asthma populations would require further study and inquiry into the challenges associated with dissemination of in-home PM reduction strategies.

Acknowledgments

The authors thank J. Balmes and K. Smith for their advice on study design. The authors also thank 3M Company for advice and materials in the design of the placebo filter.

This study was funded by the National Institutes of Health/National Institute of Environmental Health Sciences (NIH/NIEHS) 1R01ES016336-01 and 3R01ES016336-02S1. Additional support was provided by NIGMS (1U54GM104944 and P30GM103338) and NICHD (1UG1HD090902).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

References

Allen R, Larson T, Sheppard L, Wallace L, Liu LJ. 2003. Use of real-time light scattering data to estimate the contribution of infiltrated and indoor-generated particles to indoor air. Environ Sci Technol 37(16):3484–3492, PMID: 12953856, 10.1021/es021007e.

Allen RW, Carlsten C, Karlen B, Leckie S, van Eeden S, Vedal S, et al. 2011. An air filter intervention study of endothelial function among healthy adults in a woodsmoke-impacted community. Am J Respir Crit Care Med 183(9):1222–1230, PMID: 21257787, 10.1164/rccm.201010-1572OC.

Barn P, Larson T, Noullett M, Kennedy S, Copes R, Brauer M. 2008. Infiltration of forest fire and residential wood smoke: an evaluation of air cleaner effectiveness. J Expo Sci Environ Epidemiol 18(5):503–511, PMID: 18059421, 10.1038/sj.jes.7500640.

Brand PL, Duiverman EJ, Waalkens HJ, van Essen-Zandvliet EE, Kerrebijn KF. 1999. Peak flow variation in childhood asthma: correlation with symptoms, airways obstruction, and hyperresponsiveness during long-term treatment with inhaled corticosteroids. Dutch CNSLD Study Group. Thorax 54(2):103–107, PMID: 10325912, 10.1136/thx.54.2.103.

Brouwer AF, Roorda RJ, Brand PL. 2006. Home spirometry and asthma severity in children. Eur Respir J 28(6):1131–1137, PMID: 16870659, 10.1183/09031936.06.00118205.

Bryant-Stephens T. 2009. Asthma disparities in urban environments. J Allergy Clin Immunol 123(6):1199–1206, PMID: 19501229, 10.1016/j.jaci.2009.04.030.

Bundy KW, Gent JF, Beckett W, Bracken MB, Belanger K, Triche E, et al. 2009. Household airborne penicillium associated with peak expiratory flow variability in asthmatic children. Ann Allergy Asthma Immunol 103(1):26–30, PMID: 19663123, 10.1016/S1081-1206(10)60139-1.

Butz AM, Matsui EC, Breysse P, Curtin-Brosnan J, Eggleston P, Diette G, et al. 2011. A randomized trial of air cleaners and a health coach to improve indoor air quality for inner-city children with asthma and secondhand smoke exposure. Arch Pediatr Adolesc Med 165(8):741–748, PMID: 21810636, 10.1001/archpediatrics.2011.111.

Cockcroft DW, Hargreave FE. 1990. Airway hyperresponsiveness. Relevance of random population data to clinical usefulness. Am Rev Respir Dis 142(3):497–500, PMID: 2202243, 10.1164/ajrccm/142.3.497.

Douwes J, Zuidhof A, Doekes G, van der Zee SC, Wouters I, Boezen MH, et al. 2000. (1–>3)-beta-D-glucan and endotoxin in house dust and peak flow variability in children. Am J Respir Crit Care Med 162(4 Pt 1):1348–1354, PMID: 11029343, 10.1164/ajrccm.162.4.9909118.

Eggleston PA, Butz A, Rand C, Curtin-Brosnan J, Kanchanaraksa S, Swartz L, et al. 2005. Home environmental intervention in inner-city asthma: a randomized controlled clinical trial. Ann Allergy Asthma Immunol 95(6):518–524, PMID: 16400889, 10.1016/S1081-1206(10)61012-5.

Frey U, Bielicki JA. 2017. Fluctuation metrics as novel endpoints for clinical trials in asthma. Am J Respir Crit Care Med 195(8):967–968, PMID: 28409680, 10.1164/rccm.201611-2286ED.

Greenberg S, Liu N, Kaur A, Lakshminarayanan M, Zhou Y, Nelsen LM, et al. 2012. Airway obstruction lability helps distinguish levels of disease activity in asthma. Respir Med 106(4):500–507, PMID: 22301379, 10.1016/j.rmed.2011.12.013.

Hetzel MR, Clark TJ. 1980. Comparison of normal and asthmatic circadian rhythms in peak expiratory flow rate. Thorax 35(10):732–738, PMID: 7466721, 10.1136/thx.35.10.732.

Howden-Chapman P, Pierse N, Nicholls S, Gillespie-Bennett J, Viggers H, Cunningham M, et al. 2008. Effects of improved home heating on asthma in community dwelling children: randomised controlled trial. BMJ 337:a1411, PMID: 18812366, 10.1136/bmj.a1411.

Hystad PU, Setton EM, Allen RW, Keller PC, Brauer M. 2009. Modeling residential fine particulate matter infiltration for exposure assessment. J Expos Sci Environ Epidemiol 19(6):570–579, PMID: 18716606, 10.1038/jes.2008.45.

Johnston FH, Hanigan IC, Henderson SB, Morgan GG. 2013. Evaluation of interventions to reduce air pollution from biomass smoke on mortality in Launceston, Australia: retrospective analysis of daily mortality, 1994-2007. BMJ 346:e8446, PMID: 23299843, 10.1136/bmj.e8446.

Juniper EF, Guyatt GH, Feeny DH, Ferrie PJ, Griffith LE, Townsend M. 1996. Measuring quality of life in children with asthma. Qual Life Res 5(1):35–46, PMID: 8901365, 10.1007/BF00435967.

Kaminsky DA, Wang LL, Bates JH, Thamrin C, Shade DM, Dixon AE, et al. 2017. Fluctuation analysis of peak expiratory flow and its association with treatment failure in asthma. Am J Respir Crit Care Med 195(8):993–999, PMID: 27814453, 10.1164/rccm.201601-0076OC.

Kamps AW, Roorda RJ, Brand PL. 2001. Peak flow diaries in childhood asthma are unreliable. Thorax 56(3):180–182, PMID: 11182008, 10.1136/thorax.56.3.180.

Krieger JW, Takaro TK, Song L, Weaver M. 2005. The Seattle-King County Healthy Homes Project: a randomized, controlled trial of a community health worker intervention to decrease exposure to indoor asthma triggers. Am J Public Health 95(4):652–659, PMID: 15798126, 10.2105/AJPH.2004.042994.

Lebowitz MD, Krzyzanowski M, Quackenboss JJ, O’Rourke MK. 1997. Diurnal variation of PEF and its use in epidemiological studies. Eur Respir J Suppl 24:49S–56S, PMID: 9098711.

Levy JI, Brugge D, Peters JL, Clougherty JE, Saddler SS. 2006. A community-based participatory research study of multifaceted in-home environmental interventions for pediatric asthmatics in public housing. Soc Sci Med 63(8):2191–2203, PMID: 16781807, 10.1016/j.socscimed.2006.05.006.

Magzamen S, Mortimer KM, Davis A, Tager IB. 2005. School-based asthma surveillance: a comparison of student and parental report. Pediatr Allergy Immunol 16(8):669–678, PMID: 16343089, 10.1111/j.1399-3038.2005.00304.x.

McGowan JA, Hider RN, Chacko E, Town GI. 2002. Particulate air pollution and hospital admissions in Christchurch, New Zealand. Aust N Z J Public Health 26(1):23–29, PMID: 11895020, 10.1111/j.1467-842X.2002.tb00266.x.

McNamara M, Noonan C, Ward T. 2011. Correction factor for continuous monitoring of wood smoke fine particulate matter. Aerosol Air Qual Res 11(3):315–322, PMID: 25364330, 10.4209/aaqr.2010.08.0072.

Morgan WJ, Crain EF, Gruchalla RS, O’Connor GT, Kattan M, Evans R 3rd
, et al. 2004. Results of a home-based environmental intervention among urban children with asthma. N Engl J Med 351(11):1068–1080, PMID: 15356304, 10.1056/NEJMoa032097.

Mortimer KM, Redline S, Kattan M, Wright EC, Kercsmar CM. 2001. Are peak flow and symptom measures good predictors of asthma hospitalizations and unscheduled visits? Pediatr Pulmonol 31(3):190–197, PMID: 11276131, 10.1002/ppul.1028.

Naeher LP, Brauer M, Lipsett M, Zelikoff JT, Simpson CD, et al. 2007. Woodsmoke health effects: a review. Inhal Toxicol 19(1):67–106, PMID: 17127644, 10.1080/08958370600985875.

National Heart Lung and Blood Institute. 2007. Expert Panel Report 3: Guidelines for the diagnosis and management of asthma. NIH publication 07-4051. Bethesda, MD:National Heart, Lung, and Blood Institute.

Noonan CW, Navidi W, Sheppard L, Palmer CP, Bergauff M, Hooper K, et al. 2012. Residential indoor PM2.5 in wood stove homes: follow-up of the Libby changeout program. Indoor Air 22(6):492–500, PMID: 22607315, 10.1111/j.1600-0668.2012.00789.x.

Noonan CW, Ward TJ. 2012. Asthma randomized trial of indoor wood smoke (ARTIS): rationale and methods. Contemp Clin Trials 33(5):1080–1087, PMID: 22735495, 10.1016/j.cct.2012.06.006.

Noonan CW, Ward TJ, Semmens EO. 2015. Estimating the number of vulnerable people in the United States exposed to residential wood smoke. Environ Health Perspect 123(2):A30, PMID: 25642637, 10.1289/ehp.1409136.

Segala C, Fauroux B, Just J, Pascual L, Grimfeld A, Neukirch F. 1998. Short-term effect of winter air pollution on respiratory health of asthmatic children in Paris. Eur Respir J 11(3):677–685, PMID: 9596121.

Self TH, George CM, Wallace JL, Patterson SJ, Finch CK. 2014. Incorrect use of peak flow meters: are you observing your patients? J Asthma 51(6):566–572, PMID: 24720711, 10.3109/02770903.2014.914218.

Semmens EO, Noonan CW, Allen RW, Weiler EC, Ward TJ. 2015. Indoor particulate matter in rural, wood stove heated homes. Environ Res 138:93–100, PMID: 25701812, 10.1016/j.envres.2015.02.005.

Sigsgaard T, Forsberg B, Annesi-Maesano I, Blomberg A, Bølling A, Boman C, et al. 2015. Health impacts of anthropogenic biomass burning in the developed world. Eur Respir J 46(6):1577–1588, PMID: 26405285, 10.1183/13993003.01865-2014.

Simons FE. 1997. A comparison of beclomethasone, salmeterol, and placebo in children with asthma. Canadian Beclomethasone Dipropionate-Salmeterol Xinafoate Study Group. N Engl J Med 337(23):1659–1665, PMID: 9385125, 10.1056/NEJM199712043372304.

Smith KR, Bruce N, Balakrishnan K, Adair-Rohani H, Balmes J, Chafe Z, et al. 2014. Millions dead: how do we know and what does it mean? Methods used in the comparative risk assessment of household air pollution. Annu Rev Public Health 35:185–206, PMID: 24641558, 10.1146/annurev-publhealth-032013-182356.

Sweet LL, Polivka BJ, Chaudry RV, Bouton P. 2014. The impact of an urban home-based intervention program on asthma outcomes in children. Public Health Nurs 31(3):243–252, PMID: 24720657, 10.1111/phn.12071.

U.S. Department of Energy. 2009. Residential Energy Consumption Survey, Table HC6.7 Space Heating in U.S. Homes, by Census Region. Washington, DC.

van Essen-Zandvliet EE, Hughes MD, Waalkens HJ, Duiverman EJ, Pocock SJ, Kerrebijn KF. 1992. Effects of 22 months of treatment with inhaled corticosteroids and/or beta-2-agonists on lung function, airway responsiveness, and symptoms in children with asthma. The Dutch Chronic Non-specific Lung Disease Study Group. Am Rev Respir Dis 146(3):547–554, PMID: 1355640, 10.1164/ajrccm/146.3.547.

Waalkens HJ, Gerritsen J, Koëter GH, Krouwels FH, van Aalderen WM, Knol K. 1991. Budesonide and terbutaline or terbutaline alone in children with mild asthma: effects on bronchial hyperresponsiveness and diurnal variation in peak flow. Thorax 46(7):499–503, PMID: 1877037, 10.1136/thx.46.7.499.

Ward T, Lange T. 2010. The impact of wood smoke on ambient PM2.5 in northern Rocky Mountain valley communities. Environ Pollut 158(3):723–729, PMID: 19897293, 10.1016/j.envpol.2009.10.016.

Ward T, Boulafentis J, Simpson J, Hester C, Moliga T, Warden K, et al. 2011. Lessons learned from a woodstove changeout on the Nez Perce Reservation. Sci Total Environ 409(4):664–670, PMID: 21144555, 10.1016/j.scitotenv.2010.11.006.

Ward TJ, Rinehart L, Lange T. 2006. The 2003/2004 Libby, Montana PM2.5 Source Apportionment Research Study. Aerosol Sci Tech 40(3):166–177, 10.1080/02786820500494536.

Ward TJ, Semmens EO, Weiler E, Harrar S, Noonan CW. 2017. Efficacy of interventions targeting household air pollution from residential wood stoves. J Expos Sci Environ Epidemiol 27(1):64–71, PMID: 26555475, 10.1038/jes.2015.73.

Yoo Y, Choung JT, Yu J, Kim d. K, Koh YY. 2008. Acute effects of Asian dust events on respiratory symptoms and peak expiratory flow in children with mild asthma. J Korean Med Sci 23(1):66–71, PMID: 18303201, 10.3346/jkms.2008.23.1.66.

Yoos HL, Kitzman H, McMullen A, Henderson C, Sidora K. 2002. Symptom monitoring in childhood asthma: a randomized clinical trial comparing peak expiratory flow rate with symptom monitoring. Ann Allergy Asthma Immunol 88(3):283–291, PMID: 11926622, 10.1016/S1081-1206(10)62010-8.

Children’s Lead Exposure: A Multimedia Modeling Analysis to Guide Public Health Decision-Making

Author Affiliations open

1U.S. Environmental Protection Agency, Office of Research and Development, Research Triangle Park, North Carolina, USA

PDF icon PDF Version (2.4 MB)

  • Background:
    Drinking water and other sources for lead are the subject of public health concerns around the Flint, Michigan, drinking water and East Chicago, Indiana, lead in soil crises. In 2015, the U.S. Environmental Protection Agency (EPA)’s National Drinking Water Advisory Council (NDWAC) recommended establishment of a “health-based, household action level” for lead in drinking water based on children’s exposure.
    Objectives:
    The primary objective was to develop a coupled exposure–dose modeling approach that can be used to determine what drinking water lead concentrations keep children’s blood lead levels (BLLs) below specified values, considering exposures from water, soil, dust, food, and air. Related objectives were to evaluate the coupled model estimates using real-world blood lead data, to quantify relative contributions by the various media, and to identify key model inputs.
    Methods:
    A modeling approach using the EPA’s Stochastic Human Exposure and Dose Simulation (SHEDS)-Multimedia and Integrated Exposure Uptake and Biokinetic (IEUBK) models was developed using available data. This analysis for the U.S. population of young children probabilistically simulated multimedia exposures and estimated relative contributions of media to BLLs across all population percentiles for several age groups.
    Results:
    Modeled BLLs compared well with nationally representative BLLs (0–23% relative error). Analyses revealed relative importance of soil and dust ingestion exposure pathways and associated Pb intake rates; water ingestion was also a main pathway, especially for infants.
    Conclusions:
    This methodology advances scientific understanding of the relationship between lead concentrations in drinking water and BLLs in children. It can guide national health-based benchmarks for lead and related community public health decisions. https://doi.org/10.1289/EHP1605
  • Received: 12 January 2017
    Revised: 09 May 2017
    Accepted: 18 May 2017
    Published: 12 September 2017

    Address correspondence to V. Zartarian, 5 Post Office Square, Suite 100, MC ORA 01-3, Boston, MA 02109-3912 USA. Telephone: (617) 918-1541. Email: zartarian.valerie@epa.gov

    Supplemental Material is available online (https://doi.org/10.1289/EHP1605).

    The views expressed in this article are those of the authors and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (813 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (192 KB)

Introduction

Background

The U.S. Environmental Protection Agency (EPA), Centers for Disease Control and Prevention (CDC), and American Academy of Pediatrics agree that there is no known safe level of lead (Pb) in a child’s blood; even low levels of Pb in the blood can result in behavior and learning problems, lower IQ and hyperactivity, slowed growth, hearing problems, and anemia (www.epa.gov/lead; http://www.cdc.gov/nceh/lead/; Council on Environmental Health 2016). Triantafyllidou et al. (2014) concluded that low levels of Pb in drinking water could pose a human health concern in sensitive population groups (e.g., young children and particularly formula-fed infants). Drinking water and other exposure sources for Pb have recently been the subject of public health concerns around the Flint, Michigan, drinking water (Hanna-Attisha et al. 2016; Laidlaw et al. 2016) and East Chicago, Indiana, Pb in soil (Goodnough 2016) crises. As part of the EPA’s Safe Drinking Water Act assessment of lead in drinking water, the National Drinking Water Advisory Council (NDWAC)’s Lead and Copper Rule (LCR) Working Group was established to provide advice to EPA in considering potential revisions to the LCR. In December 2015, NDWAC recommended establishment of a “health-based, household action level” for Pb in drinking water based on children’s exposure (NDWAC 2015). The NDWAC working group recommended that “water systems would be required to notify the consumer, state drinking water program, and the local public health agency if this level were exceeded. The expectation is that individuals and local officials would use this information to take prompt actions at the household level to mitigate lead risks. …” While the EPA has not yet determined the specific role of a health-based benchmark for Pb in drinking water in the new rule, the agency sees value in providing states with drinking water systems and the public with a greater understanding of the potential health implications for vulnerable populations of specific levels of Pb in drinking water. The EPA anticipates that a health-based benchmark could also help inform other potential elements of a revised LCR, including public education requirements, prioritization of households for lead service line replacement programs or other risk mitigation actions at the household level, and potential requirements related to schools or other priority locations (U.S. EPA 2016a). To guide a potential health-based benchmark for Pb in drinking water, an approach is needed to advance scientific understanding of the relationship between Pb concentrations in drinking water and blood lead levels (BLLs) in infants and young children.

Objectives

The primary objective was to develop a coupled exposure–dose modeling approach that can be used to determine what drinking water Pb concentrations keep exposed children’s BLL below specified target values, considering exposures from multiple media (water, soil, dust, food, air). There is no acceptable level of Pb in children; selected target values here relate to the CDC blood Pb reference value, currently 5 μg/dL at the 97.5th percentile of BLLs in U.S. children (cdc.gov/nceh/lead/acclpp/blood_lead_levels.htm). The CDC is considering changing the reference value to 3.5 μg/dL (ATSDR 2016). Related objectives of this analysis were to evaluate the coupled model estimates using EPA NHEXAS [National Human Exposure Assessment Survey (Clayton et al. 1999)] and CDC National Health and Nutrition Examination Survey [NHANES (CDC 2013a, 2013b, 2016)] BLL data, to quantify relative contributions by the various media, and to identify key model inputs. Our main hypothesis was that the Stochastic Human Exposure and Dose Simulation (SHEDS)-Multimedia Model (https://www.epa.gov/chemical-research/stochastic-human-exposure-and-dose-simulation-sheds-estimate-human-exposure), the probabilistic exposure model that was previously evaluated and applied for other chemicals, coupled with the Integrated Exposure Uptake and Biokinetic (IEUBK) Model (https://www.epa.gov/superfund/lead-superfund-sites-software-and-users-manuals), can estimate BLLs comparable to observed BLL data, i.e., with a relative error <50%. A second hypothesis was that results from this coupled modeling approach can inform a health-based benchmark for Pb in drinking water considering a multimedia risk cup approach (a conceptual approach for estimating total Pb exposures and risks, aggregated from different sources), and provide a better understanding of the relative importance of exposure pathways and data needs to guide public health decisions for reducing childhood Pb risks.

While this work pertains to the U.S. residential (civilian) population, the same approach could be applied to other populations and countries, depending on available data. This analysis was not designed for specific at-risk populations or households, but some evaluation and contribution analysis results are provided with regional scale (NHEXAS Region 5) data in addition to national scale. The focus of this paper is the modeling and multimedia exposure analysis methodology; results are provided for several selected BLLs and percentiles of the population (based in part on the CDC blood Pb reference value mentioned above).

Methods

Models Used

A probabilistic modeling approach was developed and applied to quantify and analyze children’s Pb exposures and BLLs from drinking water and other environmental media (soil, dust, food, air). The analysis used the EPA’s SHEDS-Multimedia (version 4.1; U.S. EPA) coupled with the IEUBK (version 1.1, build 11; U.S. EPA). The SHEDS-Multimedia model is a physically based probabilistic Monte Carlo exposure model that can simulate aggregate or cumulative exposures over time via dietary and residential routes for a variety of multimedia environmental chemicals using real-world data (i.e., human activity diaries, measured concentration data, exposure factors) for model inputs. SHEDS-Multimedia has been applied for various pesticides, metals, and polychlorinated biphenyls in research applications and to inform EPA regulatory decisions (Xue et al. 2010, 2012, 2014a, 2014b; Zartarian et al. 2006, 2012; Glen et al. 2012). It has been well evaluated against real-world data (e.g., blood biomarker measurements), peer reviewed by multiple EPA external scientific advisory panels (www.epa.gov/sap), and published in over 30 journal articles. These published SHEDS-Multimedia sensitivity analysis and model evaluation analysis methods were used in this Pb application. The IEUBK model for estimation of childhood BLLs has also been externally peer reviewed and used for agency regulatory purposes (U.S. EPA 1994a, 1994b; Hogan et al. 1998; White et al. 1998; NRC 2005). It predicts childhood BLLs resulting from multiple pathways of exposure and supports soil clean-up levels at Superfund sites.

The general consensus of a 1999 workshop was that a fully probabilistic version of the IEUBK model would aid in understanding how exposure variability affects the distribution of BLL (NRC 2005, p. 239). SHEDS-Multimedia complements IEUBK by considering human exposures probabilistically. Coupling these models allows for simulating variability in Pb exposures and doses for different pathways, allowing identification of key model input variables and analysis of relative contribution by media and exposure pathways to BLL for different age groups and population percentiles. While SHEDS-Multimedia is a two-stage Monte Carlo model, we chose to not conduct a quantitative uncertainty analysis for the multimedia Pb analysis at this time, given the major effort involved to characterize level of confidence in each key input and conduct uncertainty simulations and analyses (Xue et al. 2006); thus, we present uncertainties and limitations qualitatively in the “Discussion” section of this paper.

Approach Overview

Figure 1 illustrates the general approach for this coupled model analysis. The top three panels illustrate the SHEDS-Multimedia exposure modeling methodology as described in Zartarian et al. (2012). Monte Carlo sampling was applied to obtain population variability distributions of exposures by pathway, from which available intake was determined and summed across pathways to compute uptake. Regression equations derived from IEUBK were applied to convert absorbed dose (uptake) to BLL. See Supplemental Materials Section S1, Table S1, and Figure S1 for details on the SHEDS–IEUBK coupling methodology and model inputs used in the analysis. Modeled BLL was plotted against water Pb concentration for a specified percentile of the population to determine the water Pb concentration that keep BLL below specified values. This process was repeated with different assumed water concentrations to identify the relationship between concentration and resulting BLL at the specified population percentile (Figure 1, bottom left panel). The red horizontal lines in this panel illustrate several target BLL values; the values on the curves intersecting the target BLL values represent the tipping point water Pb concentrations that keep BLL below specified levels.

Flow diagram.
Figure 1. Illustration of Stochastic Human Exposure and Dose Simulation (SHEDS)–Integrated Exposure Uptake and Biokinetic (IEUBK) modeling to inform a health-based benchmark for Pb.

Data Used in the Modeling Analysis

Available data from various sources were used for children’s activity patterns, Pb concentrations in different media, exposure factors, and biokinetic dose factors; distributional inputs were based on measurements collected in EPA and other federal agency field studies, or reported in published literature (see Supplemental Materials S2 and Tables S3–S5). Age-specific model inputs were used where available. We simulated infants (0 to 6 mo of age) per NDWAC recommendation, but there is more uncertainty for this age group. We used activity diaries from Consolidated Human Activity Database and NHANES for this age group, but due to lack of exposure factor data, we assumed the same soil/dust ingestion rate as for 1-y-olds. Separate model analysis results were generated for different scenarios. The age groups considered were 0- to 6-mo-olds, 1- to <2-y, 2- to <6-y, and 0- to 7-y-olds (lifetime average, 0–84 mo). Exposure scenarios considered included Pb in drinking water only and aggregate exposures from Pb from water, soil, food, dust, and air. Several types of SHEDS–IEUBK runs were conducted: a) for model evaluation, a national-scale analysis using NHANES data, and a regional-scale analysis using NHEXAS Region 5 data; b) for analyzing relative contributions by exposure pathway in the United States and NHEXAS Region 5; c) for sensitivity analyses to identify key factors; and d) for national-scale runs with a set of alternative drinking water Pb concentration scenarios to develop the linear relationships between concentration and BLL percentiles, shown in the bottom left of Figure 1.

Model Averaging Time and Addressing Biological Variability in the Coupled Models

Initial analyses were conducted with 2-d model averaging times, given available activity diaries used in SHEDS-Multimedia; we subsequently focused on 30-d averaging time simulations consistent with the IEUBK period, per recommendations of a work-in-progress peer consultation panel (Versar, Inc. 2016). The 30-d analysis results are shown below, and 2-d analysis results are provided in Supplemental Materials for comparison; pros and cons of both are in the Discussion. IEUBK blood Pb estimates do not reflect interindividual behavioral and pharmacokinetic differences; a geometric standard deviation (GSD) of 1.6 is applied to outputs to account for biological variability and measurement error, but does not account for exposure variability (Hogan et al. 1998; White et al. 1998). The SHEDS–IEUBK modeling only represents exposure variability; thus, a variability factor is needed to reflect real-world BLLs that also account for biological variability (this term may also account for other sources of variability, such as measurement and/or model error), and this factor is affected by the model averaging time period.

From the model evaluation results comparing SHEDS–IEUBK BLL estimates vs. NHANES-measured BLLs, the GSDs are 1.64 and 1.62 for 1- to <2-y-old and 2- to <6-y-old groups, respectively, while GSDs for NHANES BLLs are 1.92 and 1.89 for those two age groups, respectively (presented in the “Results” section below). These results indicate that GSDs of the real-world BLL measurements are consistently higher than those of predicted BLLs for both age groups, i.e., GSDs of 1 to <2 y and 2 to <6 y are almost the same for the two age groups, and the difference between NHANES and SHEDS–IEUBK BLL GSDs is ∼0.3 for both age groups. This implies biological variability was missing in our original 30-d averaging time BLL predictions, since only exposure variability is accounted for in coupling SHEDS-Multimedia and IEUBK; the missing variability will affect the distribution of the BLLs and high percentiles. Thus, we used the GSDs of NHANES BLL data as the standard to calculate the missing biological variability as shown in the equations below. We assumed exposure and biological variances are independent, and the distribution is lognormal. In the log–transformed space:

Qtotal2: total variance

Qe2: exposure variance

Qb2: biological variance

This is the formula that was used to calculate the biological variance by age group:


Qb2 for 1- to <2-y-olds and 2- to <6-y-olds are 0.185 and 0.176, respectively, which is generally consistent with biological variance, 0.22, specified by IEUBK {GSD=1.6 and [ln(1.6)]2=0.22}. We redid our original 30-d model assessment with the above calculated biological variances for model evaluation and prediction of BLLs with daily averaged household tap water Pb concentration.

Results

Results below and in Supplemental Materials demonstrate the SHEDS–IEUBK modeling approach.

Model Evaluation

For evaluating or ground-truthing national scale estimates of BLLs using the SHEDS–IEUBK coupled model approach, we used NHANES 2009–2014 BLL data. Performance of the coupled models at the national scale was evaluated by the relative error between estimated BLLs and observed BLLs (i.e., the difference between the estimated and observed BLL divided by the observed BLL value). Available representative Pb concentrations for food (U.S. FDA 2014), soil and dust (HUD 2011), and water (U.S. EPA 2010) were used for model inputs. For 30-d exposure time frame analyses, we used correlated inputs for soil, dust, and water Pb concentrations (see Table S5). The relative error in BLL was 0–23%, depending on age and percentile (see Figure 2 and Table 1; note that for 2-d analyses not considering biological variability and possibly overestimating exposure variability, relative error was <10% for all percentiles and age groups as shown in Table S6, Figure S2). SHEDS–IEUBK modeling underestimated BLL for NHANES 2009–2010 sampling period data, overpredicted for 2013–2014, and better predicted for 2011–2012 (Figure S3 and Figure S4). Additional model evaluation results (e.g., with NHEXAS data) are presented in Table S7; for the regional-scale analysis, relative error was 35–42%, depending on age and percentile.

Three graphs plotting observed and predicted concentration of blood lead in micrograms per deciliter (y-axis) across percentile (x-axis) for age groups 1 to less than 2 years, 2 to less than 6 years, and 1 to less than 6 years.
Figure 2. Evaluation of Stochastic Human Exposure and Dose Simulation (SHEDS)–Integrated Exposure Uptake and Biokinetic (IEUBK) modeled blood lead levels (BLL) vs. National Health and Nutrition Examination Survey (NHANES) 2009–2014 BLL for different age groups. conc., concentration.
Table 1. Stochastic Human Exposure and Dose Simulation (SHEDS)–Integrated Exposure Uptake and Biokinetic (IEUBK) modeling blood lead level (BLL) evaluation with 2009–2014 National Health and Nutrition Examination Survey (NHANES) blood data, longitudinal (30 d) with correlated key inputs.
Age group Source n Mean SD 50th GM GSD 95th 97.5th 99th %>3 μg/dL
1 to <2 y old Observed 475 1.47 1.30 1.12 1.16 1.92 3.60 5.54 7.90 6.95
Predicted 3,000 1.46 1.27 1.13 1.16 1.92a 3.58 4.60 6.41 7.70
Relative error 0% 1% 0% 1% 17% 19%
2 to <6 y old Observed 1,892 1.33 1.60 0.98 1.03 1.89 3.13 4.39 7.15 5.44
Predicted 3,000 1.55 1.28 1.20 1.25 1.88a 3.84 4.94 6.67 8.60
Relative error 17% 23% 21% 23% 12% 7%
Note: Relative error here is absolute value of predicted minus observed, divided by observed, multiplied by 100. GM, geometric mean; GSD, geometric standard deviation; n=sample size; SD, standard deviation.

aThis GSD reflects the effect of exposure and biological variability on BLL.

Relative Exposure Pathway Contributions

The modeled exposure pathway contribution analyses revealed that for children older than 1 y, the soil/dust ingestion and food ingestion pathways contributed more to BLL than drinking water, and the air pathway contributed the least with a small amount. For higher percentiles of the BLL distribution, soil/dust ingestion is the major pathway. Water ingestion is also an important contributor, especially for infants. For context, the CDC currently has a reference BLL for 1- to 5-y-olds of 5 μg/dL based on the distribution of BLLs in the United States; this is the reference level at which CDC recommends public health actions be initiated. As shown in Figure 3, for the national analyses:

  • For 0- to 6-mo-olds, soil/dust and water ingestion pathways predominate at the highest BLL percentiles. At the 90th to 100th percentiles with 2.66 μg/dL median predicted BLL (range: 2.15 to 8.50 μg/dL), soil/dust and water account for ∼52% and ∼39%, respectively. At the 70th–80th percentile or 1.49 μg/dL median predicted BLL (1.35 to 1.66 μg/dL), soil/dust and water ingestion together account for ∼80% of Pb exposure. Soil/dust, food, and water ingestion have similar contributions up to the ∼50th percentile of the population at ∼0.90 μg/dL predicted median BLL (0.83 to 0.97 μg/dL). Food intake is a background exposure accounting for ∼10–25%, depending on the BLL percentile, and food intake accounts for ∼0.1–0.3 μg/dL of BLL.
  • For 1- to <2-y-olds, soil/dust ingestion was the dominant pathway above the ∼80th BLL percentile. Above the 90th BLL percentile or 3.26 μg/dL predicted median BLL (2.39 to 16.7 ug/dL), soil/dust, food intake, and water account for 77%, 16%, and 7%, respectively. Food intake was a major contributor below the ∼70th percentile BLL, and contributed ∼0.6 μg/dL, on average, across all percentiles. Water accounted for ∼10–15% of the BLL, depending on the percentile, and contributed ∼0.2 μg/dL on average.
  • Not illustrated in Figure 3, the pathway contributions for 2- to <6-y-olds were essentially the same as for 1- to <2-y-olds (see Figure S5).
Figures 3A and 3B are stacked bar charts plotting Pb exposure from diet, soil and dust ingestion, water, and inhalation from air in micrograms per day (y-axis) across percentile range of predicted BLL and midpoint of predicted BLL (x-axis) for 0 to 6 month olds and 1 to less than 2 year olds, respectively.
Figure 3. Estimated contribution of exposure pathways to BLL, for national scale. Bar charts provide Pb daily exposure contributions from diet, soil and dust ingestion, water, and inhalation from air for percentiles of the BLL distribution. The bars are 10% increments in the BLL distribution. The median BLL for each increment is indicated under each bar. Exposure in the figure is adjusted for bioavailability of Pb in each exposure pathway. Panel (A), national scale for 0- to 6-mo-olds; Panel (B), national scale for 1- to <2-y-olds.

Additional contribution analysis results with NHEXAS Region 5 data are presented in Figure S6.

Key Model Inputs Identified by Sensitivity Analyses

Model results were most sensitive to dietary inputs for lower percentiles and soil/dust ingestion inputs for higher percentiles of BLL distributions, as illustrated in Figure 3. Sensitivity analyses showed soil/dust ingestion rate, soil Pb concentration, food Pb intake, and bioavailability are key inputs. Food Pb intake was highly sensitive to methods for handling nondetects (see Table S8). For soil/dust ingestion rates, the most influential input considered for the coupled model outputs, we did an additional sensitivity analysis using the central tendency value of 100 mg/d suggested by U.S. EPA (2011) and also 80 mg/d, and found the 3.5 μg/dL BLL targets at the 97.5th percentile were exceeded without drinking water Pb. Similarly, targets were exceeded with a sensitivity analysis using von Lindern et al. (2016) soil/dust ingestion rates for 1 to <2 y (see Table S9 and Figure S7); this analysis also found SHEDS–IEUBK overestimated NHANES BLLs. The sensitivity analyses show that the blood Pb prediction for 0- to 7-y-olds is very sensitive to soil/dust ingestion rate when it was scaled from the input based on Özkaynak et al. (2011) to 80 mg/d; for example, at the 97.5th percentile, the daily averaged tap water Pb concentration that could keep BLL below 5 μg/dL was reduced from 5 ppb to 1 ppb (see Table S10). Details on these inputs and sensitivity analyses are provided in Supplemental Materials. There are three current approaches for estimating soil/dust ingestion rate as described in U.S. EPA (2011); this variable is highly uncertain for children under age 2 y.

Drinking Water Lead Concentrations at Example Target Blood Lead Levels

Figure 4 and Table 2 show SHEDS–IEUBK results for estimated maximum daily average household tap water Pb concentrations that could keep BLL below specified targets (30-d averaging time); these were derived as described with Figure 1. Figure 4 illustrates the predicted BLL at the 97.5th percentile of the U.S. population as a function of daily average household tap Pb water concentration for the different age and exposure scenarios. These plots allowed us to extract the daily average water Pb concentration that could keep BLLs below the specified targets of 3.5 and 5 μg/dL at the 97.5th percentile BLL of the U.S. population of each age range, as shown in Table 2. The dashes in Table 2 for three aggregate scenarios indicate that even with no Pb in water, this target would be exceeded. The robustness of this modeling approach allows consideration of household tap water Pb concentrations for other percentiles. However, the numbers in Table 2 could be conservatively low because key model input values based on the available older data (e.g., soil and dust concentrations, and soil/dust ingestion rate) may be higher than they are currently.

Figures 4A and 4B are graphs plotting modeled BLL versus water Pb concentration for water-only exposure scenario and aggregate exposure scenario, respectively.
Figure 4. Illustrative graphs for determining household tap water Pb concentrations were calculated for different scenarios. y-Axis is modeled blood Pb level at 97.5th percentile of simulated population; x-axis is daily average water Pb concentration. The different colored lines represent different ages: orange is infants age 0–6 mo, dark blue is 1- to <2-y-olds, and light blue is 2- to <6-y-olds.

Table 2. Stochastic Human Exposure and Dose Simulation (SHEDS)–Integrated Exposure Uptake and Biokinetic (IEUBK) results for maximum daily average household tap water Pb concentrations that could keep BLL below specified values (30-d averaging time; accounting for correlations and other external peer consult input described in Supplemental Material).
Age group Exposure scenario BLL: 3.5 μg/dL 97.5th percentile BLL: 5 μg/dL 97.5th percentile BLL: 3.5 μg/dL 95th percentile BLL: 5 μg/dL 95th percentile
0 to 6 mo old Water only 13 ppb 19 ppb 14 ppb 21 ppb
Aggregate 4 ppb 16 ppb 7 ppb 17 ppb
1 to <2 y-old Water only 25 ppb 38 ppb 31 ppb 46 ppb
Aggregate 5 ppb 3 ppb 14 ppb
2 to <6 y-old Water only 24 ppb 35 ppb 29 ppb 44 ppb
Aggregate 3 ppb 1 ppb 12 ppb
0 to 7 y old Water only 20 ppb 30 ppb 27 ppb 41 ppb
Aggregate 5 ppb 2 ppb 13 ppb
Note: Daily average of a distribution reflecting real-world monitoring scheme to be determined. —, BLL will not be below targets even with 0 ppb Pb in water.

Discussion

This paper presents a state-of-the-science methodology that can guide a health-based benchmark for Pb in drinking water and can also be applied to other media. The well-reviewed, published, evaluated models allowed for contribution and sensitivity analyses, and identification of key factors, media, and exposure pathways. The coupled SHEDS–IEUBK estimates compared well against BLL data from NHANES and NHEXAS (0–23% and ∼36–42% relative error, respectively), despite compiling different input data sets not originally intended for this purpose. The ability to probabilistically simulate multimedia exposures for the U.S. population and provide blood lead predictions consistent with NHANES BLLs represents an advance in science and a potential to guide public health decisions. Human exposures and public health outcomes are considered in the EPA’s Pb policies, such as the LCR. For example, revisions underway to strengthen the LCR include a potential health-based benchmark for Pb in drinking water and assessment of the benefits of lead service line replacement programs. Recent surveys [conducted in 2011 and 2013 and discussed in Cornwell et al. (2016)] conducted by the American Water Works Association indicate that between 15 to 22 million people of the 293 million served by U.S. community water systems have either full or partial Pb-containing lines servicing their home (7%) (Cornwell et al. 2016). The SHEDS–IEUBK multimedia exposure modeling analysis approach presented in this paper could inform national rulemaking efforts that translate to the local scale through state and local drinking water programs. If communities with water Pb issues are aware that soil and dust Pb can also be important contributors to children’s BLLs and understand the limits of drinking water program efforts, community-level education and outreach efforts can be targeted to maximize multimedia exposure reduction efforts for minimizing children’s Pb risks.

Another strength of this SHEDS–IEUBK analysis is that it uniquely reports percent contribution to children’s BLL by pathway, population percentile, and age group. The EPA’s 2007 Risk and Exposure Assessment for Lead did provide an urban case study of Pb pathway contributions with estimates of 20.5% of Pb from diet, 11.9% from drinking water, 43.7% from outdoor soil/dust, 23.7% from indoor dust, and 0.1% from air by inhalation (based on average annual uptake from each media until a child is 7-y-old and assuming a 0.05 μg/m3 maximum monthly average airborne Pb) (U.S. EPA 2007). There are a number of papers reporting on the importance of the soil/dust pathway to BLL of children as described in U.S. EPA (2013) and references therein (e.g., Mielke et al. 2011). The relative media contributions at the upper percentiles of SHEDS–IEUBK estimates for >1-y-olds are consistent with the U.S. EPA (2007) results; we estimate dietary contribution greater in lower percentiles, and water contribution higher for infants 0–6 mo of age. However, contributions from pathways are highly dependent on scenarios being considered (e.g., Elwood et al. 1984; Mielke et al. 2011; Zahran et al. 2013). Isolated events and widespread occurrences of drinking water contaminated with Pb have been associated with and thought to be the dominant contributor to elevated BLLs in North Carolina, Maine, Michigan, and Washington, DC (Edwards et al. 2009; Hanna-Attisha et al. 2016). Additionally, underestimates of the contribution to BLL from Pb-contaminated water may occur due to potential indirect exposure from food preparation (Triantafyllidou and Edwards 2012). Other studies have also shown indoor dust sources from both Pb-based paint (Blette 2008) and legacy soil Pb concentrations (Mielke and Reagan 1998) to be major contributors to elevated BLL, and in some cases to be a dominant source of exposure (Gasana et al. 2006).

The SHEDS–IEUBK model evaluation was stronger (lower relative error between observed and modeled values) for earlier NHANES time frames. BLLs have been decreasing over decades and have continued to decrease since 2010 (U.S. EPA 2016b, 2016c; Laidlaw et al. 2016). Whether this recent change can be explained by changing media concentrations, human activity patterns (two main components of human exposure), or both, remains unclear. Certainly, due to federal regulations, the removal or reduction of Pb in gasoline, paint, and plumbing has contributed (Council on Environmental Health 2016). The apparent decline in time spent outdoors by children in the United States (Roberts and Foehr 2008) may also have contributed by reducing Pb soil ingestion. There is also seasonal variation in BLL, with BLLs tending to be increased in the fall (e.g., see Laidlaw et al. 2016), which we could not model using IEUBK.

There are some other limitations and uncertainties of this analysis. Daily model average results for Pb in drinking water related to the CDC reference value may be impacted by temporal changes in NHANES in addition to model inputs changing over time. Our approach involves selecting a BLL benchmark (e.g., CDC reference level that may change). The multimedia Pb modeling analysis results are based on inputs for which available data may not reflect recent exposures [e.g., U.S. Department of Housing and Urban Development (HUD) soil Pb data is 2005–2006]. With additional information from future field studies on temporal changes in model inputs in recent years, further evaluation of model predictions against temporal changes in recent NHANES samplings would be possible. Because NHANES sample size is limited to represent the national population, collecting and analyzing states’ blood Pb data may be useful for further model evaluation.

Our modeling indicates that soil and dust ingestion is a dominant exposure pathway. The soil/dust ingestion rate for children is a key input to which model results are highly sensitive, and for which data are limited and uncertain, especially for children <2 y; for older ages, values are similar between Özkaynak et al. (2011) used in this analysis and von Lindern et al. (2016), developed using different methodologies. If higher soil/dust ingestion rate values were used with this analysis, modeled water Pb concentrations would be lower. Although we applied soil and dust ingestion rates, which are among the lowest in the reported literature, our simulations overpredicted blood Pb for the most recent 2013–2014 NHANES cycle. An analysis of U.S. soil Pb studies from 1970 to 2012 reported no association between year and median soil Pb concentration at a national scale, although within single cities, soil Pb generally declined over time (Datko-Williams et al. 2014); thus, we posit that changing human activity patterns, such as soil/dust ingestion rates, may in part explain the BLL declines.

In addition to soil/dust ingestion rate, other uncertainties in this analysis are not accounting for seasonal variations (due to lack of available data and use of IEUBK), model averaging time, and how the coupled models capture biological and other sources of variability in the GSD of BLLs. Because the 30-d exposure period GSD reflects the effect of exposure variability, but not biological variability on BLL, our original results underpredicted the GSD and upper percentiles of BLLs in NHANES, and accordingly, overestimated Pb in water concentrations. Using a 2-d model averaging time does not align with IEUBK, but shows closer comparison to NHANES BLL data and GSD, as shown in Supplemental Materials (Figures S2–S3, and Table S6). The 2-d results may approximate BLL accounting for biologic variability by overestimating exposure variability. With the approach for addressing the biological variability issue described in the “Methods” section above, GSDs between SHEDS–IEUBK estimated and NHANES measured BLLs are very close (1.92 vs. 1.92 for 1- to <2-y-olds and 1.89 vs. 1.88 for 2- to <6-y-olds; see Table 1), and evaluation with NHANES BLLs has been improved with adding biological variance, especially for higher percentiles (see Table 1 and Figure 2). More BLL data being collected from states (McClure et al. 2016) could help evaluate the biological variance correction factor and which averaging time is more appropriate to guide a health-based benchmark for Pb. State-collected BLL data will also supplement NHANES BLL data, which may not be fully representative of the true distribution of the U.S. population BLLs, particularly at the tails.

While this work pertains to the U.S. population, the same approach could be applied to other populations or countries, but the results might be different. Although we simulated correlations in Pb exposure among dust, soil, and water (using NHEXAS and HUD data), stratified data by housing age, and assessed BLL at upper percentiles of the BLL distribution, our current analyses are not focused on specific at-risk populations, such as Flint, Michigan, and East Chicago, Indiana, or other environmental justice communities or homes with high Pb in soil, dust, or water. The household tap water monitoring scheme is a factor that influences estimated drinking water Pb concentrations and related exposures. Given the spatial and temporal variability of household Pb water concentrations, there are uncertainties in water Pb concentration data collected under the current LCR regulatory sampling that limits the ability to predict Pb exposures from drinking water. Local-scale data for multimedia model inputs and BLLs, preferably collected simultaneously and with geospatial and temporal resolution, would be beneficial for extending the coupled model approach for other applications and specific communities.

Conclusions

This Pb modeling methodology and multimedia analysis advances scientific understanding of the relationship between Pb levels in drinking water and BLLs in infants and young children, and can inform a health-based benchmark for lead in drinking water. The approach can also be applied to soil, dust, food, or other environmental media to guide decision-making, considering exposures aggregated from multiple media. While the focus of this analysis is the national scale, to help inform national rulemaking for Pb policies addressing multimedia exposures and public health outcomes, decisions such as setting a health-based benchmark for Pb in drinking water under the revised LCR would guide local-scale monitoring programs and Pb risk prevention education efforts in communities, and help systematically identify vulnerable communities such as Flint, Michigan, and East Chicago, Indiana, in the United States.

In addition, this modeling approach developed for Pb could apply to other multimedia contaminants for cumulative impact analyses. While model evaluation provides confidence in the results, more up-to-date data and information on key model inputs (e.g., children’s soil/dust ingestion rate and bioavailability) and BLLs would be helpful to refine model estimates for quantifying and reducing uncertainties, and to focus on specific at-risk populations and communities. Modeled estimates of BLL using the SHEDS–IEUBK approach can be extended to quantify health endpoints (e.g., IQ decrements) and to inform benefits analyses for strengthening public health protection (e.g., considering benefits of Pb service line replacement programs under the revised LCR). This modeling approach, together with state-collected BLL data and other data sets, e.g., for environmental justice variables and social determinants of health, could also be applied to help identify the most at-risk communities for Pb exposures and understand key factors for disparities; such analyses could inform decisions for minimizing public health risks from national to local scales.

Acknowledgments

We gratefully acknowledge the following individuals: U.S. EPA Office of Research and Development managers and scientists for providing guidance and review, including J. Garland, J. Orme-Zavaleta, P. Price, K. Isaacs, K. Alapaty, A. Gillespie, M. Slimak, A. Geller, R. Kavlock, T. Burke; H. Huang [via Postdoctoral Program administered by the Oak Ridge Institute for Science and Education (ORISE) through Interagency Agreement between DOE and EPA, IA number DW-89-92431601], J. Frank (ORISE), and A. Poulakos (ASRC Federal Vistronix, contract EP-G131-00143) for assisting with inputs and literature review; U.S. EPA Program Office staff for technical input, including E. Helm, A. Hafez, L. Christ, E. Burneson, S. Foster, K. Raffaele, M. Burgess, D. Murphy, and Z. Pekar; Office of Research and Development staff for technical input and assistance with the peer consult, and Quality Assurance Project Plan, including N. Shao, D. Lytle, M. Schock, T. Speth, R. Daniels, B. Stuart, C. Alvarez; P. Ashley of HUD and J. Spungen from FDA for providing data; Versar, Inc. for managing the peer consult and work-in-progress reviewers K. Bogen, D. Hattis, K. Vork, and E. DesHommes; the peer consult occurred with funding under contract EP-C-12-045-91; CSRA for graphics support. The data reported in this paper are presented or available at EPA’s ScienceHub (https://edg.epa.gov/metadata/catalog/main/home.page).

References

ATSDR (Agency for Toxics Substances and Disease Registry). 2016. Meeting of the Lead Poisoning Prevention Subcommittee of the NCEH/ATSDR Board of Scientific Counselors. 19 September 2016, Atlanta, GA, Record of the Proceedings. https://www.atsdr.cdc.gov/science/lpp/docs/lead_subcommittee_minutes_9_19_2016_508.pdf [accessed 13 July 2017].

Blette V. 2008. Drinking water public right-to-know requirements in the United States. J Water Health 6(suppl 1):43–51, 10.2166/wh.2008.031.

CDC. 2013a. (Centers for Disease Control and Prevention). National Health and Nutrition Examination Survey (NHANES 2009–2010). https://wwwn.cdc.gov/nchs/nhanes/search/datapage.aspx?Component=Laboratory&CycleBeginYear=2009 [accessed 13 July 2017].

CDC. 2013b. National Health and Nutrition Examination Survey (NHANES 2011–2012). https://wwwn.cdc.gov/nchs/nhanes/search/datapage.aspx?Component=Laboratory&CycleBeginYear=2011 [accessed 13 July 2017].

CDC. 2016. National Health and Nutrition Examination Survey (NHANES 2013–2014). https://wwwn.cdc.gov/nchs/nhanes/search/datapage.aspx?Component=Laboratory&CycleBeginYear=2013 [accessed 13 July 2017].

Clayton CA, Pellizzari ED, Whitmore RW, Perritt RL, Quackenboss JJ. 1999. National Human Exposure Assessment Survey (NHEXAS): distributions and associations of lead, arsenic and volatile organic compounds in EPA region 5. J Expo Anal Environ Epidemiol 9(5):381–392, PMID: 10554141, 10.1038/sj.jea.7500055.

Cornwell DA, Brown RA, Via SH. 2016. National survey of lead service line occurrence. J Am Water Works Assoc 108(4):E182–E191, 10.5942/jawwa.2016.108.0086.

Council on Environmental Health. 2016. Prevention of Childhood Lead Toxicity. Pediatrics 138(1):e20161493, PMID: 27325637, 10.1542/peds.2016-1493.

Datko-Williams L, Wilkie A, Richmond-Bryant J. 2014. Analysis of U.S. soil lead (Pb) studies from 1970 to 2012. Sci Tot Env 468–469:854–863, 10.1016/j.scitotenv.2013.08.089.

Edwards M, Triantafyllidou S, Best D. 2009. D. Elevated blood lead in young children due to lead-contaminated drinking water: Washington, DC, 2001–2004. Environ Sci Technol 43(5):1618–1623, 10.1021/es802789w.

Elwood PC, Gallacher JEJ, Phillips KM, Davies BE, Toothill C. 1984. Greater contribution to blood lead from water than from air. Nature 310:138–140, 10.1038/310138a0.

Gasana J, Hlaing WM, Siegel KA, Chamorro A, Niyonsenga T. 2006. Blood lead levels in children and environmental lead contamination in Miami inner city, Florida. Int J Environ Res Public Health 3(3):228–234, PMID: 16968968, 10.3390/ijerph2006030027.

Glen G, Zartarian V, Smith L, Xue J. 2012. The Stochastic Human Exposure and Dose Simulation Model for Multimedia, Multipathway Chemical: Residential Module (SHEDS-Residential version 4, Technical Manual). Washington, DC:U.S. Environmental Protection Agency, Office of Research and Development.

Goodnough A. 2016. Their soil toxic, 1,100 Indiana residents scramble to find new homes. The New York Times (news story). U.S. section, online edition. 31 August 2016. https://www.nytimes.com/2016/08/31/us/lead-contamination-public-housing-east-chicago-indiana.html [accessed 6 July 2017].

Hanna-Attisha M, LaChance J, Sadler RC, Champney Schnepp A. 2016. Elevated blood lead levels in children associated with the Flint drinking water crisis: a spatial analysis of risk and public health response. Am J Public Health 106(2):283–290, PMID: 26691115, 10.2105/AJPH.2015.303003.

Hogan K, Marcus A, Smith R, White P. 1998. Integrated exposure uptake biokinetic model for lead in children: empirical comparisons with epidemiologic data. Environ Health Perspect 106(suppl 6):1557–1567, PMID: 9860915, 10.1289/ehp.98106s61557.

HUD (U.S. Department of Housing and Urban Development). 2011. American Healthy Homes Survey, American Healthy Homes Survey Lead and Arsenic Findings (Lead concentration data provided in 2016 from Policy and Standards Division, Office of Lead Hazard Control and Healthy Homes, U.S. Department of Housing and Urban Development). http://portal.hud.gov/hudportal/documents/huddoc?id=AHHS_Report.pdf [accessed 6 July 2017].

Laidlaw MAS, Filippelli GM, Sadler RC, Gonzales CR, Ball AS, Mielke HW. 2016. Children’s Blood Lead Seasonality in Flint, Michigan (USA), and Soil-Sourced Lead Hazard Risks. Int J Environ Res Public Health 13(4):358, PMID: 27023578, 10.3390/ijerph13040358.

McClure LF, Niles JK, Kaufman HW. 2016. Blood lead levels in young children: US, 2009–2015. J Pediatr 75:173–81, PMID: 27297207, 10.1016/j.jpeds.2016.05.005.

Mielke HW, Gonzales CR, Mielke PW. 2011. The continuing impact of lead dust on children’s blood lead: comparison of public and private properties in New Orleans. Environ Res 111(8):1164–1172, PMID: 21764050, 10.1016/j.envres.2011.06.010.

Mielke HW, Reagan PL. 1998. Soil is an important pathway of human lead exposure. Environ Health Perspect 106(suppl 1):217–229, PMID: 9539015, 10.1289/ehp.98106s1217.

NDWAC (National Drinking Water Advisory). 2015. “Report of the Lead and Copper Rule Working Group to the National Drinking Water Advisory Council. Final. August 24, 2015.” 36–37. https://www.epa.gov/sites/production/files/2016-01/documents/ndwaclcrwgfinalreportaug2015.pdf [accessed 6 July 2017].

NRC (National Research Council). 2005. Superfund and Mining Mega