Article pubs.acs.org/est
Improving Infant Exposure and Health Risk Estimates: Using Serum Data to Predict Polybrominated Diphenyl Ether Concentrations in Breast Milk Satori A. Marchitti,† Judy S. LaKind,‡,§,∥ Daniel Q. Naiman,⊥ Cheston M. Berlin,∥ and John F. Kenneke†,* †
National Exposure Research Laboratory, U.S. Environmental Protection Agency, Athens, Georgia 30605, United States LaKind Associates, LLC, Catonsville, Maryland 21228, United States § Department of Epidemiology and Public Health, University of Maryland School of Medicine, Baltimore, Maryland 21201, United States ∥ Department of Pediatrics, Milton S. Hershey Medical Center, Penn State College of Medicine, Hershey, Pennsylvania, 17033, United States ⊥ Department of Applied Mathematics and Statistics, Whiting School of Engineering, Johns Hopkins University, Baltimore, Maryland 21218, United States ‡
S Supporting Information *
ABSTRACT: Women in the United States have breast milk concentrations of polybrominated diphenyl ethers (PBDEs) that are among the highest in the world, leading to concerns over the potential health implications to breastfeeding infants during critical stages of growth and development. Developing cost-effective and sustainable methods for assessing chemical exposures in infants is a high priority to federal agencies and local communities. PBDE data are available in nationally representative serum samples but not in breast milk. As a method to predict PBDE concentrations in U.S. breast milk, we present the development of congener-specific linear regression partitioning models and their application to U.S. serum data. Models were developed from existing paired milk and serum data and applied to 2003−2004 NHANES serum data for U.S. women. Highest estimated median U.S. breast milk concentrations were for BDE-47 (30.6 ng/g lipid) and BDE-99 (6.1 ng/g lipid) with the median concentration of Σ7PBDEs estimated at 54.2 ng/g lipid. Predictions of breast milk PBDE concentration were consistent with reported concentrations from 11 similarly timed U.S. studies. When applied to NHANES data, these models provide a sustainable method for estimating population-level concentrations of PBDEs in U.S. breast milk and should improve exposure estimates in breastfeeding infants.
■
INTRODUCTION
Animal and in vitro studies suggest that PBDEs are endocrine-disrupting compounds that can cause reproductive, neurodevelopmental, and thyroid toxicities.8,9 In humans, studies have associated PBDE exposure in adults with altered endocrine and reproductive functions.10,11 Associations between prenatal and/or postnatal exposure to PBDEs and neurodevelopmental outcomes are not fully understood; however, there is significant concern over the potential health effects of these compounds in infants and children.12−14 While concentrations of PBDEs may begin to plateau in response to decreased production and use, concentrations of PBDEs in humans over the last several decades dramatically increased. In the U.S. population, PBDE concentrations are
Polybrominated diphenyl ethers (PBDEs) are persistent organic pollutants1 that have been used worldwide for decades as flame retardants in numerous consumer products. Three major PBDE technical products were produced and classified according to average bromine content; that is, pentaBDE, octaBDE, and decaBDE. Due to their longer half-lives and greater potential to bioaccumulate,2,3 the lower brominated penta and octa BDE formulations have been banned in Europe and a voluntary manufacturing phase-out was initiated in the United States (U.S.) in 2004. However, these formulations continue to be produced in developing countries and exported to the U.S. as additives in consumer products. Due to their propensity to leach from consumer products, PBDEs have become widely distributed in the environment,4 food sources,5 household dust,6 and indoor air,7 and thus, human exposures are expected to continue for many years. © 2013 American Chemical Society
Received: Revised: Accepted: Published: 4787
December 20, 2012 April 8, 2013 April 12, 2013 April 12, 2013 dx.doi.org/10.1021/es305229d | Environ. Sci. Technol. 2013, 47, 4787−4795
Environmental Science & Technology
Article
to 2003 no reliable partitioning information existed for PBDEs primarily due to a lack of studies where milk and serum samples were taken close in time.29 Three studies were found from 2003-present that met our study selection criteria.20,30,31 These studies were similarly timed with sample collection occurring in 2003 (Schecter et al. 2006; n = 11), 2004−2005 (LaKind et al. 2009; n = 10), and 2007 (Schecter et al. 2010; n = 29), designated hereafter as U.S. Study 1, 2, and 3, respectively. Breast milk and serum PBDE data were available or provided by the authors. For model development, we only included participants for whom lipid-adjusted (ng/g lipid) milk and serum PBDE concentrations were both above the limit of detection (LOD).29 Data for seven PBDE congeners were available for model development: 2,4,4′-tribromodiphenyl ether (BDE-28), 2,2′,4,4′-tetrabromodiphenyl ether (BDE-47), 2,2′,3,4,4′-pentabromodiphenyl ether (BDE-85), 2,2′,4,4′,5-pentabromodiphenyl ether (BDE-99), 2,2′,4,4′,6-pentabromodiphenyl ether (BDE-100), 2,2′,4,4′,5,5′-hexabromodiphenyl ether (BDE153), and 2,2′,4,4′,5,6′-hexabromodiphenyl ether (BDE-154). These lower brominated congeners are the primary components of the pentaBDE technical mixture.32 Individual PBDE milk:serum partitioning ratios were calculated for each congener by dividing the milk concentration (ng/g lipid) by the serum concentration (ng/g lipid). We found minimal interindividual variability in partitioning ratios among the participants of each study; milk:serum relationships for each congener were consistent across the three studies. Thus, we combined the data from the three studies into one data set we refer to as the “model building population”. Pearson’s r correlation coefficients and least-squares linear regression were used to determine if serum and milk PBDE concentrations for each congener were significantly correlated (i.e., if serum PBDE concentrations could be used to predict breast milk PBDE concentrations). Linear regression of each of the seven PBDE congeners was performed using SigmaPlot Systat Software, version 12.3 (2011):
among the highest in the world with levels approximately 1− 100 fold higher than those reported in Asia and Europe.15,16 This is likely due to the fact that the U.S. was the primary consumer of the BDE technical products. The lower brominated BDE congeners, such as those found in the pentaBDE mixture, are highly lipophilic and can bioaccumulate in adipose tissue. Of particular concern is that PBDEs stored in a mother’s body can be released into breast milk during lactation. Thus, breast milk is believed to be the greatest source of PBDE exposure to breastfeeding infants17,18 and it has been reported that levels of PBDEs in breast milk may result in some infant exposures exceeding the U.S. EPA Reference Doses (RfDs).19 Despite the presence of environmental chemicals in breast milk, the preponderance of evidence indicates that breastfeeding not only confers numerous health benefits on the infant,24 but may also counter effects associated with prenatal chemical exposures.25−27 Therefore, it is critical to accurately evaluate this risk-benefit paradigm with respect to infant health. For a complete assessment of PBDE exposures in U.S. infants, population-level information on PBDE concentrations in U.S. breast milk is needed.20 As one option for obtaining these data, some researchers have advocated for a U.S. breast milk monitoring program.21,22 Other countries have utilized these types of programs to obtain valuable information on early life exposures and temporal trends;23 however, such a program is not likely to be implemented in the U.S. While milk collection is noninvasive and does not require a medical professional, collecting and analyzing breast milk samples for chemicals is resource-intensive and can heighten concerns that breastfeeding is unsafe. The Centers for Disease Control and Prevention’s (CDC) National Center for Environmental Health recently released nationally representative biomonitoring data on serum levels of PBDEs in the U.S. as part of the National Health and Nutrition Examination Survey (NHANES) conducted during 2003− 2004.28 Having the ability to utilize NHANES serum PBDE levels as a surrogate for predicting PBDE concentrations in breast milk would provide an alternative to breast milk monitoring. To develop such predictions, lipid-based, congener-specific, PBDE milk:serum partitioning ratios are required.20,22,29,30 In order to determine accurate partitioning ratios, it is critical for both blood and breast milk samples to be collected postpartum, and as close in time as possible, due to differences in day-to-day exposure patterns and physiological changes during pregnancy and lactation that may affect chemical concentrations.29 Prior to 2006, limited data existed on the partitioning of PBDEs into breast milk and none were available that met these criteria; since then, acceptable data from three U.S. studies have become available.20,30,31 The major objectives of the present study are 2-fold: to develop robust, predictive PBDE partitioning models based on breast milk and serum data and to apply these models to NHANES serum data to estimate the distribution of PBDE concentrations in the breast milk of U.S. women of childbearing age.
yi = β0 + β1x i
(1)
where yi is the breast milk PBDE concentration of the participant (i), β0 is the y-intercept, β1 is the slope of the line, and xi is the serum PBDE concentration of the participant. The predictive power of each regression model was determined by k-fold cross-validation.33 Briefly, milk and serum data pairs for each congener were randomly assigned into equal- (or near equal-) sized segments or nonoverlapping groups of subjects. The number of segments (k) for each congener varied between four and eight, with three to seven data pairs in each segment. For each set of congener segments, a revised model was developed by excluding one segment at a time and generating a model with the remaining segments. The revised model was then used to predict breast milk PBDE concentrations (yi′) using the serum data (xi) from the excluded segment; these steps were repeated k times equal to the number of subsets for each congener. The predictive ability (Q2) of each congener model was quantified by determining the sum of squares for the difference between the measured (yi) and predicted (yi′) milk PBDE concentration values [PRESS (predictive residual sum of squares)], and the sum of squares for the difference between measured (yi) and mean (yi̅ ) breast milk PBDE concentrations [SST (total sum of squares)]:
■
MATERIALS AND METHODS PBDE Milk:Serum Partitioning Model Development. We conducted literature searches using PubMed and Web of Science for studies that measured PBDE concentrations in both breast milk and serum samples taken as close in time as possible from the same women. Preliminary findings indicated that prior 4788
dx.doi.org/10.1021/es305229d | Environ. Sci. Technol. 2013, 47, 4787−4795
Environmental Science & Technology
Article
Figure 1. BDE-47 (A; r2 = 0.98) and BDE-153 (B; r2 = 0.99) linear regression models. Solid lines are the least-squares fit of individual serum and milk concentrations from U.S. studies 1, 2, and 3. Dashed lines indicate unity. Insets show lower-range PBDE concentrations.
Q2 = 1 −
PRESS SST
with similar sampling years (2003−2007) as the NHANES serum sampling years (2003−2004).
■
(2)
RESULTS PBDE Milk:Serum Partitioning Model Development. Serum and milk concentrations for the seven PBDE congeners we examined were significantly, positively correlated (r2 ≥ 0.90; Pearson’s r ≥ 0.95, p values 0.9 were considered to have excellent predictability.33 A subset of data from the LaKind et al. (2009) study was analyzed to evaluate whether PBDE milk:serum partitioning ratios changed with breastfeeding duration. Milk:serum partitioning ratios calculated from additional sets of breast milk and serum samples taken from seven women at the terminus of breastfeeding were compared to those calculated from samples taken at approximately one month postpartum using Student’s t test (p < 0.05). Model Application to NHANES Serum Data. The CDC’s National Center for Health Statistics data files for the NHANES 2003−2004 survey are available at http://www.cdc. gov/nchs/nhanes/nhanes2003-2004/lab03_04.htm. Lipid-adjusted PBDE concentration (ng/g lipid) data are from a subsample (Laboratory 28; 2337 persons) of the total population sampled (10 122 persons). Demographic and weighting data for each NHANES 2003−2004 participant are available at http://www.cdc.gov/nchs/nhanes/nhanes20032004/demo03_04.htm PBDE serum data from women of childbearing age (18−45 yr) were compiled from the NHANES data files (n = 424−428, depending on the congener). This age range was comparable to the age range from the model building population (22−44 yr). When serum PBDE measurements < LOD, a value of LOD divided by the square root of two was assigned.16 NHANES sample weights were applied to the data, resulting in estimates of percentile concentrations meant to be representative of the U.S. population. To determine if PBDE concentrations for the model building population were comparable to the range of concentrations measured in the NHANES population of women, serum PBDE concentrations (ng/g lipid) for both populations were plotted as log-normal box and whisker distributions and overlaid. PBDE milk:serum partitioning models were then applied to the NHANES serum data to predict the distribution of individual and sum PBDE (∑7PBDEs) congeners in U.S. human milk. These predicted concentrations were then compared to those reported for 11 U.S. breast milk studies
Table 1. PBDE Milk:Serum Linear Regression (yi = β1xi) Models and Pearson Correlations.a β1b
PBDE
n
BDE-28 BDE-47 BDE-85 BDE-99 BDE-100 BDE-153 BDE-154
20 49 14 39 48 42 17
1.41 1.46 1.32 1.40 1.26 0.93 0.84
± ± ± ± ± ± ±
0.17 0.07 0.18 0.16 0.02 0.03 0.11
r2
Q2
Pearson’s r
p value
0.95 0.98 0.95 0.90 1.00 0.99 0.95
0.90 0.97 0.93 0.88 1.00 0.99 0.93
0.97 0.99 0.98 0.95 1.00 0.99 0.97
