Urinary polycyclic aromatic hydrocarbon (PAH) metabolite concentrations in three pregnancy cohorts from 7 U.S. study sites

Objective Prenatal exposure to polycyclic aromatic hydrocarbons (PAHs) is associated with adverse birth and developmental outcomes in children. We aimed to describe prenatal PAH exposures in a large, multisite U.S. consortium. Methods We measured 12 mono-hydroxylated metabolites (OH-PAHs) of 7 PAHs (naphthalene, fluorene, phenanthrene, pyrene, benzo(c)phenanthrene, chrysene, benz(a)anthracene) in mid-pregnancy urine of 1,892 pregnant individuals from the ECHO PATHWAYS consortium cohorts: CANDLE (n = 988; Memphis), TIDES (n = 664; Minneapolis, Rochester, San Francisco, Seattle) and GAPPS (n = 240; Seattle and Yakima, WA). We described concentrations of 8 OH-PAHs of non-smoking participants (n = 1,695) by site, socioeconomic characteristics, and pregnancy stage (we report intraclass correlation coefficients (ICC) for n = 677 TIDES participants). Results Exposure to the selected PAHs was ubiquitous at all sites. 2-hydroxynaphthalene had the highest average concentrations at all sites. CANDLE had the highest average concentrations of most metabolites. Among non-smoking participants, we observed some patterns by income, education, and race but these were not consistent and varied by site and metabolite. ICCs of repeated OH-PAH measures from TIDES participants were ≤ 0.51. Conclusion In this geographically-diverse descriptive analysis of U.S. pregnancies, we observed ubiquitous exposure to low molecular weight PAHs, highlighting the importance of better understanding PAH sources and their pediatric health outcomes attributed to early life PAH exposure.


Introduction
Polycyclic aromatic hydrocarbons (PAHs) are common toxicants that result from incomplete combustion of organic materials, including coal, crude oil, garbage, wood and gasoline.They occur in mixtures with various other toxic products of combustion.PAHs are found in ambient and indoor air, food, soil, and water, making human exposure possible via inhalation, ingestion, and skin contact [1,2].Exposure occurs primarily via inhalation and diet [1] with key anthropogenic sources including tobacco smoke, vehicular emissions, high-temperature cooking and grilled/roasted/smoked foods, industrial food processing, coal power plants, asphalt-based products, and residential heating, including wood burning [1].Natural sources include forest fires, volcanic activity, petroleum seeps, and coal deposits [1].
PAH exposure during pregnancy has been linked to a number of adverse birth and pediatric outcomes [3][4][5].Because PAHs are generally metabolized within a few hours of exposure and mono-hydroxylated metabolites (OH-PAHs) are detectable in urine, urinary OH-PAH concentrations are frequently considered biomarkers of recent PAH exposure [6][7][8].However, evidence shows that PAHs can accumulate in fatty tissue compartments such as fat and brain [9], thus urinary OH-PAH levels may, to a certain extent, reflect PAHs released from longerterm storage compartments in addition to recent exposures.In general, urinary PAH biomarkers are more convenient and cost-effective to collect and analyze compared to blood, thus they are increasingly being used in epidemiological studies of prenatal exposures and child health outcomes.To date, however, existing data on urinary OH-PAH levels in pregnancy are mostly derived from small studies representing a single geographic region, and/or databases that impute non-detectable observations using a single value, e.g., limit of detection (LOD)/square root of two, which can bias results where detection frequencies are low [10][11][12][13].
We built on these studies and described urinary OH-PAH levels among pregnant individuals from 7 sites in a large U.S. consortium, ECHO PATHWAYS, part of the Environmental Influences on Child Health Outcomes (ECHO) initiative of the U.S. National Institutes of Health (NIH) [14].The ECHO PATHWAYS consortium provided harmonized prenatal exposure data, including urinary OH-PAHs.Maternal urinary concentrations of 8 OH-PAHs were measured in a sample of nearly 2,000 participants from 7 study sites residing in 6 different U.S. metropolitan areas.We compared our results to those of non-smoking females of reproductive age in the U.S. National Health and Nutrition Examination Survey (NHANES) 2011-2012.

OH-PAH measurements
Samples were stored at -80˚C in each cohort's respective biorepository until analysis for OH-PAHs at the New York State Department of Health, Wadsworth Laboratory.We used liquid-liquid extraction and LC-MS/MS to measure 12 mono-hydroxylated metabolites of seven PAH parent compounds (naphthalene, fluorene, phenanthrene, pyrene, benzo(c)phenanthrene, chrysene, benz[a]anthracene) in urine [19].Briefly, samples (500 μL) were transferred into 15-mL glass tubes, spiked with 10 ng each of an isotopically labeled internal standard mixture, and mixed with 1 mL of 0.5 M ammonium acetate buffer containing 200 units/mL of βglucuronidase/sulfatase enzyme (MP Biomedicals, LLC, Solon, OH, USA).Samples were gently mixed and incubated overnight at 37˚C, then diluted (2 mL HPLC water, followed by 7 mL of 80% pentane: 20% toluene, v:v) and shaken on a reciprocating shaker for one hour.After centrifugation, the supernatant was transferred to a new tube, concentrated under a gentle stream of nitrogen at 30˚C, and the final extract reconstituted with 250 μL of methanol.
Chromatographic separation was accomplished using a Waters Acquity I-Class UPLC system (Waters Corporation; Milford, MA, USA) connected with an Acquity UPLC BEH C18 column (50 × 2.1 mm, 1.7 μm, Waters; Milford, MA, USA).Identification and quantification of OH-PAHs was performed on an ABSCIEX 5500 Triple Quadrupole mass spectrometer (Applied Biosystems; Foster City, CA, USA) operated in electrospray ionization negative mode with multiple reaction monitoring.The mobile phases were methanol and water.The peaks of 2-, 3-, 9-hydroxyfluorene and 1-, 9-hydroxyphenanthrene could not be chromatographically separated, so they are quantified as the sum of the individual metabolite concentrations.
Quality assurance/quality control (QA/QC) procedures included processing, for each batch of 100 samples, five duplicates of the following: method blank, matrix (urine) blank, and matrix (urine) spiked samples.Two duplicates of NIST SRM1 (Standard Reference Material) 3672 (Organic Contaminants in Smokers' Urine) and SRM1 3673 (Organic Contaminants in Non-Smokers' Urine), containing certified values of several target OH-PAHs, were also processed.Synthetic urine (Cerilliant, Round Rock, Texas, USA) was used for the matrix blank and matrix spiked samples.HPLC water was used for sample/procedural blanks.Thirteen calibration standards were analyzed with each batch of samples, from which a quadratic calibration curve (0.02 ng/mL-200 ng/mL) with a 1/x weighting was developed.Periodic injections of OH-PAH calibration standards were included throughout the sample run to ensure instrument stability in responses.Periodic instrumental blanks were also injected throughout the sample run to ensure no carryover or contamination.
For the TIDES cohort, s.g. of urine was determined using a handheld refractometer.For the CANDLE and GAPPS cohort data, s.g. of urine was determined using an Atago PAL-3 refractometer (Atago, Tokyo, Japan), which measures BRIX values.These values were further converted to specific gravity, according to the following equation: specific gravity = 1 + 0.004 X (BRIX%) [20,21].
QC results.Analyte-specific LODs were calculated as the lower value of three times the signal-to-noise ratio or the lowest calibration standard concentration.Samples with concentrations above the highest calibration standard were diluted and re-analyzed to fit the calibration range.Some target analytes were detected in the CANDLE and GAPPS method and matrix blanks; samples and matrix spike samples from these cohorts were blank corrected using the mean of the method blanks and matrix blanks, respectively.No analytes were detected in TIDES blanks, so those results were not blank corrected.

Cohort characteristics
Maternal smoking during pregnancy and socioeconomic data (including annual household income at enrollment, education, maternal age at birth, race and ethnicity) were collected through questionnaires administered during pregnancy.We used the names of U.S. metropolitan areas to define the study sites in our description of the data although some participants resided outside of these areas.We relied on gestational age and date of urine sample collection to determine the pregnancy visit timing (e.g., early, mid, late).We classified participants as smokers during pregnancy if they self-reported any tobacco use during pregnancy or if their urinary cotinine level was � 200 ng/mL [22].Cotinine analysis was completed using a method established for measuring perfluorochemicals and has been adapted for cotinine analysis [23].

NHANES comparison
For comparison, we examined OH-PAH data from the U.S. National Health and Nutrition Examination Survey (NHANES) 2011-2012 [24].Urinary OH-PAHs were measured in a subsample of NHANES participants aged 6 years and older using isotope dilution gas chromatography with tandem high-resolution mass spectrometry [24].We restricted the NHANES data to non-smoking females aged 16-46 years (n = 535) to better reflect our PATHWAYS sample.We determined smoking status by self-report and urinary cotinine levels (� 200 ng/mL indicative of active smoking) [22].Nineteen of these NHANES participants had a positive pregnancy test but we did not consider them separately because of the small sample size.We used the unweighted, uncorrected NHANES OH-PAH concentrations (ng/mL) for comparisons with PATHWAYS uncorrected OH-PAH concentrations (ng/mL).

Statistical analyses
We described the analytic sample (N = 1,892) stratified by maternal smoking status during pregnancy given the clearly established association between tobacco smoke exposure and urinary OH-PAH concentrations and by study site [1].We examined descriptive statistics by socioeconomic characteristics using arithmetic means and standard deviations for continuous measures and counts and proportions for categorical measures.We excluded 9 participants who were missing maternal smoking status during pregnancy data (cotinine or self-report) and another 9 participants who were missing specific gravity data.In the GAPPS cohort, cotinine was not measured for 62% of participants.Since only one GAPPS participant (among those with cotinine data) had cotinine � 200 ng/mL, we assumed cotinine < 200 ng/mL for GAPPS participants with unmeasured cotinine data and relied on their self-report to classify smokers.We reported gestational age (in weeks) at urine sample collection for all visits by study site.
We evaluated the proportion of each site-specific sample with OH-PAH results > LOD.We reported uncorrected urine concentrations and those corrected for s.g.: (OH-PAH conc.*[(median s.g.)-1/(s.g.-1)], where s.g.represents the s.g. of the urine sample) to account for variation in dilution of the urinary samples.All concentrations are reported in ng/mL.We calculated descriptive statistics for both uncorrected and s.g.-corrected OH-PAH concentrations by study site and maternal smoking status during pregnancy.Analyses were conducted based on a complete case analysis approach; in the case of missing data, sub-sample sizes were noted in the Results tables and figures.
We characterized OH-PAH concentrations by study site and socioeconomic characteristics for non-smoking participants only (n = 1,695).For TIDES participants with multiple pregnancy urine samples, we also characterized OH-PAHs by stage of pregnancy (n = 677 TIDES participants) and estimated intraclass correlation coefficients (ICC) by study site and in the pooled TIDES sample for metabolites with >95% detection frequencies at all four TIDES sites (2-hydroxynaphthalene, 2-and 3-hydroxyphenanthrene) using the PROC MIXED command in SAS.To reduce bias and account for left-censoring in the OH-PAH data, we used a maximum likelihood estimation (MLE) method and the PROC LIFEREG command in SAS or the R 'fitdistrplus' package to fit lognormal distributions and estimate geometric means (GM) and geometric standard deviations (GSD) of s.g.-corrected OH-PAH values [10][11][12]15,[25][26][27][28].We visually compared each MLE-fitted distribution with the corresponding histogram of the logtransformed OH-PAH values as well as the mode, which we estimated using the parzen function in the R 'modeest' package [29].S2 and S3 Figs show examples of MLE fitted distributions, histograms and modes for analytes with relatively low (i.e., ~10%) and high (i.e., ~35%), respectively, degrees of left-censoring.In our Results figures, we indicated our confidence in the MLE fit and central tendency statistics using regular font (interpret with caution: detection frequency between 60 and 85%) and bold font (very confident: detection frequency > 85%).We did not report central tendency statistics when site-specific detection frequencies were < 60% [30].We used the R clikcorr package, which also uses MLE to account for left-censored observations, to estimate correlation coefficients between log-transformed individual OH-PAH metabolites within sites [31].
Since income and education were previously associated with urinary OH-PAH concentrations in pregnant [32] and non-pregnant people [33][34], we compared GM/GSD OH-PAH concentrations across income and education categories.We also used Tobit regression to formally evaluate associations between income and education and log-transformed OH-PAH concentrations, accounting for left censoring in the data and adjusting for urinary specific gravity.In these regressions, income was modeled following the recommended approach for multi-site PATHWAYS analyses-harmonized household size (2-3, 4, 5, � 6 people) and income (adjusted for region and inflation to 2012 $USD) from the 4-6 year old visit was used instead of the prenatal visit because of the proportion of right-censored prenatal TIDES and GAPPS observations, which was addressed at the 4-6 year old visit with additional upper income categories.Natural log household income interacted with household size ('household size adjusted household income') and the corresponding main effects were included in the models.For the education regressions, the "< High school" and "High school completion" categories were combined to avoid low cell counts in the "< High school" category.Complete case analysis was used for the Tobit regressions.We also explored associations with maternal age (years), race (White, Black/African American, and an "Other" category comprised of Asian, Native Hawaiian/Pacific Islander, American Indian/Alaska Native, Other, and Multiple to avoid low cell counts), and ethnicity (Not Hispanic/Latino, Hispanic/Latino) using the same approach.
Analyses were conducted in SAS version 9.4 (SAS Institute, Cary, NC, USA) and R version 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria).

Study sample description
Our final analytic sample included 1,892 pregnant participants, aged 16-49 years, from seven study sites across six U.S. metropolitan regions: Memphis, TN; San Francisco, CA; Minneapolis, MN; Rochester, NY; Seattle, WA (2 distinct samples: TIDES and GAPPS); and Yakima, WA (Table 1).The mean age was 29 (SD: 6) years, with variation across sites; in general, CAN-DLE, TIDES Rochester, and GAPPS Yakima participants were younger than participants at other sites.Household income also differed across sites.Annual household incomes within the Memphis and Rochester sites tended to be lower than at the other five sites.Overall, most of the study sample completed at least high school (91%) and over half (57%) also completed college or technical school, but this too varied by site.The study sample was predominantly White (51%) and Black/African American (38%), however race was not evenly distributed across the sites.Most of the Black/African American participants resided at the Memphis and Rochester study sites.

OH-PAH results
OH-PAH concentrations.Table 2 shows OH-PAH results by study site and participant smoking status during pregnancy, and uncorrected NHANES 2011-2012 OH-PAH concentrations for females of reproductive age for comparison.We did not report data for smokers at the following TIDES sites due to small sub-sample sizes: San Francisco (n = 2), Minneapolis (n = 4), Seattle (n = 3).Among the four study sites for which we reported OH-PAH concentrations stratified by smoking status, we observed higher median concentrations among smokers compared to non-smokers for 1-and 2-hydroxynaphthalene, 2/3/9-hydroxyfluorene, and 1-hydroxypyrene in the Memphis and Rochester study samples.These differences were less notable and inconsistent for the two smaller GAPPS samples.Differences in median concentrations between smokers and non-smokers were less notable for the phenanthrene metabolites at all study sites.
We detected 2-hydroxynaphthalene in nearly all samples, regardless of site or smoking status, and 2-hydroxynaphthalene detection frequencies were high in all sites.Detection frequencies were also high (> 70%) for 1-hydroxynaphthalene, 1/9-hydroxyphenanthrene, and 2-and 3-hydroxyphenanthrene, with some site differences possibly explained by differences in LODs and/or differences in exposures.The higher LOD for 2/3/9-hydroxyfluorene in the TIDES cohort sites compared to other cohorts likely explains the lower detection frequencies for this analyte, while detection frequencies for 4-hydroxyphenanthrene were < 70% at most sites despite the similar LODs.Despite sensitive LODs, 1-hydroxypyrene detection frequencies       were low at all sites except CANDLE, where it was found in 88.8% and 97.1% of non-smoking and smoking during pregnancy samples, respectively.3-hydroxybenzo(c)phenanthrene, 1and 6-hydroxychrysene, and 1-hydroxybenz(a)anthracene were either not detected or detected in only a few samples and are not discussed further.3).Among the 8 metabolites we analyzed, 2-hydroxynaphthalene was detected in the highest concentrations at all sites, with GMs ranging from 1.73-4.95ng/mL across sites.At      most sites (with the exception of TIDES Minneapolis and TIDES Rochester), 1-hydroxynaphthalene was the next highest concentration detected.The CANDLE (Memphis) GMs for these metabolites (1-and 2-hydroxynaphthalene), as well as 2/3/9-hydroxyfluorene, and 1/ 9-hydroxyphenanthrene, were greater than observed in other study sites.

Repeated OH-PAH measures across pregnancy in the TIDES Cohort
Participants (n = 677) from the four TIDES study sites provided at least two prenatal urine samples.Among them, the median gestational age was 11 weeks at the early pregnancy visit (range: 5-23 weeks), 21 weeks at the mid pregnancy visit (range: 13-35 weeks), and 32 weeks at the late pregnancy visit (range: 26-41 weeks) (Table 1).We did not observe any obvious or consistent trends by pregnancy timing for any OH-PAH at any of the four TIDES sites.For OH-PAHs with high detection frequencies for all study sites, we estimated intra-site  diet, which may further be patterned by demographic characteristics that differ between the cohorts, may help to explain differences in observed concentrations between the other sites.We expected 2-hydroxynaphthalene to be detected in the highest concentrations given naphthalene is the most volatile of the four parent PAHs we studied and accounted for ~87% of the total concentration of 16 PAHs measured in the U.S. Environmental Protection Agency's (EPA) Urban Air Toxic Monitoring program from 1990 to 2014 [40].In addition to its many combustion-related outdoor sources, naphthalene can also be emitted indoors from mothballs and older moth repellant formulations, as well as from building materials such as caulking, carpet pads, and flooring [37].For the LMW PAHs (i.e., 2-3 rings) we studied-naphthalene, fluorene, and phenanthrene-indoor air may dominate outdoor air and diet as exposure sources for the average American [37,41].[37].In this study, the distributions of biomarkerbased intakes most closely resembled the distributions of modeled indoor inhalation intakes for the LMW PAHs (naphthalene, fluorene, phenanthrene) and for pyrene, which with 4 rings is considered a high molecular weight (HMW) PAH but with a molecular weight just above 200 g/mol, may behave similarly to the LMW PAHs [37].[41,44].Several of our observations are consistent with the smaller studies of pregnant individuals described above.These studies also detected OH-PAHs in a high proportion of urine samples [31,41,42].The two that analyzed a wider range of metabolites also observed 1-and 2-hydroxynaphthalene in the highest concentrations [31,42].Cathey, et al. observed significant differences in the concentrations of certain metabolites between their two study sites: 1-hydroxynaphthalene, 2-hydroxyfluorene, 1-hydroxyphenanthrene (higher in participants from Boston) and 2-hydroxynaphthalene (higher in participants from Puerto Rico) [31].Both our study and Cathey, et al. analyzed 1-and 2-hydroxynapthalene, 4-hydroxyphenanthrene, and 1-hydroxyprene.We observed a range of uncorrected GM concentrations in the PATH-WAYS data (among sites with � 65% detection frequency) that included those observed in the Boston and Puerto Rico study samples with two exceptions: 2-hydroxynapthalene and 1-hydroxyprene GMs were higher in Puerto Rico than Boston and all PATHWAYS samples [31].

Comparison with females of reproductive age in NHANES 2011-2012 data
The NHANES 2011-2012 data showed uncorrected median OH-PAH concentrations within the range of median values observed across all PATHWAYS study sites for all metabolites except 1-and 2-hydroxynapthalene where NHANES medians were slightly above the range of median values observed across PATHWAYS sites.Woodruff, et al. described urinary OH-PAH levels among pregnant participants (n = 84-91) in NHANES 2003-2004, 9% of whom were smokers [45].Their median uncorrected OH-PAH concentrations were generally like those in the NHANES 2011-2012 data we examined (i.e., n = 535 non-smoking females of reproductive age) except that the 2-hydroxynaphthalene median was notably higher in the 2011-2012 subsample (4.9 ng/mL) compared to the 2003-2004 pregnant subsample (2.4 ng/mL).

Repeated measures
We did not observe consistent trends based on timing of gestation in the repeated measures of OH-PAHs across pregnancy in the TIDES cohort.The low ICCs we observed across the repeated measures within each TIDES site may illustrate the limitations of using a single spot urine measurement to represent typical PAH exposures but may also be related to the potential influence of season relative to pregnancy stage, to additional variability introduced by changes in glomerular filtration rate and/or other physiological processes across pregnancy [46], or to some combination of factors.Other studies that have also looked at repeated measures of OH-PAHs in pregnancy have not observed significant differences or strong trends across pregnancy [31,41,42].A study of non-pregnant females of reproductive age similarly observed low to moderate ICCs for OH-PAHs (0.23-0.67), suggesting the potential importance of analyzing multiple samples within participants for association studies [47].

Socioeconomic characteristics
Among the PATHWAYS participants who did not smoke during pregnancy, we observed inconsistent patterns in income, education, and race related to OH-PAH concentrations.
Other studies have suggested level of education may be inversely correlated with OH-PAH concentrations in pregnant people [31].It is important that subsequent analyses investigating potential environmental sources of exposure rely on multivariable models and include these important factors as populations may be differentially exposed to PAHs based on socioeconomic characteristics.

Strengths and limitations
Strengths of this study were the large sample size, geographic and socioeconomic diversity, and participation of pregnant individuals from seven U.S. study sites.Additionally, we analyzed OH-PAHs in urine which captures aggregate PAH exposures, instead of focusing on a single medium such as air or diet.For four of the PATHWAYS study sites, we evaluate repeated measures of OH-PAHs during pregnancy.By using an MLE approach, validated with visual diagnostics, to impute values below the detection limit, we generated less biased geometric mean estimates compared to comparable studies using single value (i.e., LOD, LOD/sqrt 2, ½ LOD) imputation [10][11][12]15,[25][26][27].This in turn allowed us to make valid comparisons of average concentrations across sites despite site-specific differences in analytical detection limits.Appropriate handling of left-censored observations is especially important for large multisite studies like ECHO PATHWAYS where detection limits can vary by a factor of 10 or more depending on site.Although we presented data from different study sites, it is important to consider in the interpretation of the data that, with the exception of the Memphis site, which represents Shelby County, Tennessee, the study samples are not statistically representative of the regional population in the area where they were recruited.It is also important to note that the results we presented are descriptive in nature, based on one urine sample.For compounds metabolized in hours vs. days or months, a single spot urine sample does not capture exposures across all of pregnancy.Additionally, our GM estimates did not account for seasonal variation, indoor or outdoor sources, diet, or other factors related to PAH exposure.These will be the focus of future investigations planned to examine associations in subsequent multivariable analyses.

Conclusion
This descriptive analysis built on an existing limited literature on PAH exposure during pregnancy using sensitive and accurate measurements of urinary metabolite biomarkers and utilized a less-biased approach for handling observations below the detection limit compared to single-value imputation.Our overall sample size was comprised of nearly 2,000 pregnant individuals who were non-smokers during pregnancy.We had repeated urinary PAH measurements during pregnancy for nearly 700 participants from four different study sites.We characterized PAH exposure based on 8 metabolites by socioeconomic characteristics in 7 study samples.
In this 7-site analysis, although PAH exposure was ubiquitous at all sites, exposure patterns varied by study site and PAH metabolite.These findings highlight the importance of better understanding PAH sources and their pediatric health outcomes attributed to early life PAH exposure.Future studies will examine the associations between OH-PAHs and potential environmental sources of exposure using multivariable models, controlling for the influence of study sample characteristics such as region, season and demographic factors.

Fig 1
Fig 1 shows the s.g.-correctedGM OH-PAH concentrations by site (exact values are reported in Table3).Among the 8 metabolites we analyzed, 2-hydroxynaphthalene was detected in the highest concentrations at all sites, with GMs ranging from 1.73-4.95ng/mL across sites.At

Table 3 .
(Continued) Shin et al. (2013) compared daily PAH intake estimated from urinary OH-PAH concentrations in NHANES 2001-2002 with daily intakes modeled using county-level emissions data from EPA's National Air Toxics Assessment (NATA), indoor air concentrations from small panel studies, and the CalTOX multimedia exposure model to estimate PAH intakes from food