Environmental Chemicals in Breast Milk

Environmental Chemicals in Breast Milk: Predicting Breast Milk: Serum Partitioning using QSAR Models


Erin Pias Hines


Satori A. Marchitti, Caleb Dillingham, Michael-Rock Goldsmith, Daniel Chang, Marc-Andre Verner, Judy S. LaKind, Erin Hines, Suzanne Fenton, and John F. Kenneke


ORISE Fellow, Office of Research and Development, U.S. Environmental Protection Agency, Athens, GA

Student Services Authority, U.S. Environmental Protection Agency, Athens, GA

Chemical Computing Group Inc., Montreal, QC, Canada

Office of Chemical Safety and Pollution Prevention, U.S. Environmental Protection Agency, Washington D.C.

Department of Occupational and Environmental Health, University of Montreal Public Health Research Institute, Montreal, QC, Canada

LaKind Associates, LLC, Catonsville, MD

Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, NC

National Institute of Environmental Health Sciences, National Institutes of Health, Research Triangle Park, NC

Office of Research and Development, U.S. Environmental Protection Agency, Athens, GA


Increasing concern exists over the presence of environmental chemicals in breast milk. While the benefits of breastfeeding are well known, models for predicting the transfer of environmental chemicals into breast milk would greatly improve the accuracy of infant risk and exposure assessments. However, few current models are applicable to environmental chemicals, which have different physicochemical properties than pharmaceuticals. In addition, for the highest quality milk:serum partitioning data, milk and serum samples should be taken from the same woman as close in time as possible, and only concentrations > LOD should be accepted. To meet these criteria, we curated a dataset composed of available and unpublished data for the development of Quantitative Structure Activity Relationship (QSAR) models. The dataset consisted of 110 pharmaceutical drugs and 72 environmental chemicals, including polychlorinated biphenyls (PCBs), dioxins/furans, organochlorine pesticides, bromobiphenyls, polybrominated diphenyl ethers (PBDEs), phenols, parabens, perfluorinated compounds (PFCs), and several relevant metabolites. For pharmaceutical drugs, mean wet weight milk:serum ratios ranged from 0.0055 to 6.21 with a median value of 0.76 (mean = 1.18), while those of environmental chemicals ranged from 0.37 to 20.47 with a median value of 3.91 (mean = 4.46). QSAR models developed using the entire dataset (n = 182) demonstrated acceptable accuracy and predictability (R2 and Q2 values of 0.56-0.60 and 0.50-0.56, respectively) when created using pH 7.4-adjusted chemical species and 2D and 3D descriptors. Models developed using uncharged compounds only (n = 101; 70 environmental chemicals, 31 drugs) demonstrated the highest accuracy (R2, 0.75-0.80) and predictability (Q2, 0.53-0.59). Chemical descriptors important for predicting milk:serum partitioning included those related to molecular surface area, partial charge, solubility, weight, density, flexibility, and hydrophobicity. Collectively, these results highlight the importance of data quality for QSAR modeling, and using the most abundant chemical structure in the correct charge state at physiological pH. The QSAR models developed in this study should be effective tools for predicting the milk:serum partitioning of both drugs and environmental chemicals.  

Disclaimer: The opinions presented here are those of the authors and do not represent official policy of the US EPA.