We used data collected from mothers and newborns enrolled in the Health Outcomes and Measures of the Environment (HOME) Study, an ongoing prospective birth cohort. Subject recruitment and interviews for the HOME Study have been described elsewhere (Braun et al. 2010
). Briefly, we identified and contacted women attending seven prenatal clinics in the Cincinnati metropolitan area between March 2003 and January 2006. Eligibility criteria included being at least 18 years of age; living in a home built before 1978; ≤ 19 weeks of gestation; being HIV negative; living within five surrounding counties; and not receiving thyroid or seizure medications, or chemotherapy or radiation treatments. Institutional review boards of all involved research institutions, hospitals, and laboratories approved the study protocol. All mothers gave written informed consent for themselves and their children before enrolling. Of 1,263 eligible women, 468 enrolled in the study. Sixty-seven women dropped out of the study before delivery. The remaining 401 women gave birth to 389 live singleton babies, as well as 9 sets of twins and 3 still-born infants. Of those delivering singletons, 344 women provided urine at both 16 and 26 weeks gestation; they make up our primary study sample.
Measurement of pesticide exposure.
Participants provided blood and spot urine samples at 16 and 26 weeks gestation, and within 24 hr of delivery. Urine was collected in polypropylene containers and stored at –20°C until shipment to the Centers for Disease Control and Prevention (CDC) for analysis. We measured six urinary DAPs—dimethylphosphate (DMP), dimethylthiophosphate (DMTP), dimethyldithiophosphate (DMDTP), diethylphosphate (DEP), diethylthiophosphate (DETP), and diethyldithiophosphate (DEDTP), which are common metabolites of about 75% of the OP insecticides used in the United States—using a modification of the analytical method of Bravo et al. (2004)
that employs gas chromatography-tandem mass spectrometry with isotope dilution calibration.
Quality control (QC) materials, prepared at CDC from spiked pooled urine, were analyzed with standards, blanks, and study samples. The QC concentrations were evaluated using standard statistical probability rules (Caudill et al. 2008
). The limits of detection (LODs) varied depending on the metabolite and ranged from 0.2 μg/L for DMTP to 0.6 μg/L for DMP. For samples reporting nonzero concentrations below the LOD, we used the reported value. We imputed DAP concentrations that were reported as zero by choosing a random number between zero and the lowest nonzero value for that metabolite.
We converted the metabolite concentrations from micrograms per liter to their molar concentrations (nanomoles per liter) and summed them to obtain overall concentrations of diethyl alkyl phosphates (ΣDEAPs: DEP, DETP, and DEDTP), dimethyl alkyl phosphates (ΣDMAPs: DMP, DMTP, DMDTP), and of all six metabolites (ΣDAPs). We used creatinine as a marker of urine dilution and obtained creatinine-standardized DAP concentrations by dividing metabolite concentrations by creatinine concentration (Larsen 1972
). To avoid confounding due to metabolic changes that occur around the time of delivery, we confined our analysis to the 16- and 26-week results, and restricted the analysis to mothers with data from both time points. Because metabolite concentrations varied widely between time points, we averaged the concentrations of each metabolite across the two samples. Because the distribution of the averaged DAP concentrations was right-skewed, we log10
-transformed the averages to achieve a more normal distribution.
Outcome and covariate measures.
We abstracted birth weight from medical records, and calculated gestational age from the mother’s self-reported date of last menstrual period. When this information was not available, we used the results of an ultrasound (n
= 7) or a Ballard examination performed just after delivery (n
= 3) to determine gestational age. Demographic factors (i.e., maternal age, education, race/ethnicity, parity, income, marital status) were obtained from subject interviews. We also characterized infant size using race-specific birth weight for gestational age z
-scores (e.g., birth weight standardized for gestational age) to determine whether infants’ growth was restricted relative to their length of gestation (Oken et al. 2003
We measured serum cotinine, a metabolite of nicotine that has been validated as a biomarker of secondhand and active tobacco smoke exposure (Benowitz et al. 2009
; DeLorenze et al. 2002
), and whole blood lead concentrations from blood samples taken at the same time as the spot urine samples. As with the urinary DAP concentrations, we averaged serum cotinine and blood lead concentrations across the 16- and 26-week samples and log10
-transformed the results. Several women were missing lead or cotinine measurements for one or both time points; if a lead or cotinine measurement was missing for one time point, we used the value for the other time point instead of the average. Participants who were missing blood lead or serum cotinine concentrations from both time points (n
= 18) were excluded from the analysis. Blood lead and serum cotinine were quantified using previously described methods at the CDC laboratories (Bernert et al. 2000
; CDC 2003
PON genotyping data. DNA was extracted from frozen archived cord blood using the 5PRIME PerfectPure DNA blood kit (5PRIME, Gaithersburg, MD, USA). Genotyping was conducted at the Cincinnati Children’s Hospital Medical Center DNA sequencing and genotyping core facility. We used the Applied Biosystems predesigned TaqMan assays (Applied Biosystems, Carlsbad, CA, USA) for rs662 (192Q/R) and rs705379 (–108C/T) with 15 ng of genomic DNA. The protocol was as per the manufacturer’s recommendations in 384 well format, including 16 blanks and 4 sets of 2 controls for QC purposes. Assay plates mixing reagents and DNA were prepared using a TECAN Evo200 robotic laboratory workstation (TECAN, Durham, NC, USA) and amplified in an ABI 9700 thermocycler (Applied Biosystems). TaqMan results were read on an ABI 7900HT real time PCR system and exported in Excel format using the ABI SDS 2.4 software (Applied Biosystems). Genotype data was available for 312 of 344 (90.7%) of infants, with 300 (87.2%) having identifiable SNPs at both PON1 sites.
For multivariable analysis, we used PROC GLM in SAS version 9.1 (SAS Institute Inc., Cary, NC, USA). We began with an a priori
set of covariates that have been consistently associated with gestational age and birth weight: maternal age, education, race/ethnicity, and marital status; household income; gestational age at initiation of prenatal care; and parity. We also considered the following covariates: depressive symptoms (measured by the Beck Depression Inventory-II) (Beck et al. 1996
), IQ, insurance status, area of residence, maternal body mass index, alcohol use during pregnancy, child sex, tobacco exposure (assessed using serum cotinine), and blood lead concentrations. Any of these covariates showing univariate associations (p
< 0.2) with both DAP concentrations and birth outcomes were included in our multivariable analyses. Only blood lead and serum cotinine concentrations fit these criteria. Our final model included maternal age (continuous), race/ethnicity (black or white), and marital status (married or not married), household income (contiuous), parity (0, 1, or ≥ 2), and blood lead and serum cotinine concentrations (log10
-transformed). We did not include gestational age in the primary birth weight analysis because of its potentially being on the causal pathway. However, for comparison with prior results in other cohorts, we constructed gestational age-adjusted models in a sensitivity analysis.
We assessed the form of the exposure–response relationship using locally weighted regression (LOESS) plots of log-transformed concentrations and outcomes. Because a straight line fit inside the 95% confidence bands, we concluded that it was appropriate to use a linear term to describe the relationship.
Based on our preliminary analyses, we observed that the pattern for associations of DAP concentrations and birth outcomes varied by race/ethnicity. Because almost all mothers identified themselves as either black or white (the “other” group contained only 15 mothers), we restricted our analyses to black and white mothers. We then ran race-stratified models to obtain effect estimates, and formally assessed interactions in separate models using cross-product terms between race and urinary OP pesticide metabolite concentrations, using p < 0.1 as the criteria for a statistically significant interaction.
To examine gene–environment interactions between PON1 genotype and OP pesticides, we ran several PON-stratified analyses, stratifying by PON1192 or PON1–108. We also formally assessed interactions using dummy variables and interaction terms. Because race appeared to be an effect modifier and PON1 genotype can vary across racial groups, we also ran the PON1-stratified analyses separately for black and white mothers whenever possible given sample sizes.
We also performed several sensitivity analyses. First, we examined whether mothers’ pregnancy medical conditions might confound the relationship between urinary DAP concentrations and birth outcomes. These included abruptio placenta, placenta previa, chorioamnionitis, preeclampsia, and pregnancy-induced hypertension. Next, we examined whether relationships between DAP concentrations and birth weight were evident for only term newborns (≥ 37 weeks) (Wilcox 2006
). We considered child sex as an effect modifier and assessed interactions with cross-product terms. Next, we assessed whether different methods of creatinine adjustment altered our results. We used DAP concentrations as our exposure variable without creatinine standardization or adjustment, and also ran a model with log10
-transformed mean urinary creatinine concentrations as a covariate, as suggested by Barr et al. (2005)
. We also examined whether dilute urine samples changed the pattern of our results by excluding women who had at least one sample with creatinine < 20 mg/dL. Finally, to assess the effects of different time periods of sample collection, we ran separate models using the DAP concentrations in samples taken at the 16-week and 26-week visits.