UC Berkeley

Endocrine-disrupting chemicals (EDCs) are a class of chemicals that interact with components of the endogenous endocrine system and have been associated with a wide range of outcomes including reproductive anomalies, certain cancers, obesity, and birth weight. Phthalates and environmental phenols are a broad category of EDCs found in a wide array of plastics and personal care products, linked through their estrogen/androgen-related mechanisms and their ubiquity in the environment. The Developmental Origins of Health and Disease (DOHaD) hypothesis emphasizes the role of the early life environment in establishing adult trajectories of health and disease, with epigenetic modifications implicated as a key mediator between early life exposures and postnatal health. By studying epigenetic changes in response to EDC exposure, we may shed light on potential biological mechanisms in human populations and identify early markers indicative of biological influence. However, the epigenetic effects of prenatal phthalate and phenol exposure in humans remains poorly characterized, especially when considering the effects of phthalate and phenol mixtures. In addition to imparting information directly related to gene expression, epigenetic data leveraged from across the genome can provide an invaluable tool for studying complex phenotypic traits including biological aging, burden of environmental exposure, and identification of disease states. Existing epigenetic predictors of biological aging, referred to as epigenetic clocks throughout, have been found to be associated with a myriad of environmental and social exposures in adult populations. However, the environmental determinants of biological aging in children remains relatively poorly understood. Additionally, although existing epigenetic predictors have proven to be invaluable biomarkers for studying a wide array of exposures, limitations in the development process of existing epigenetic biomarkers may limit their application in complex longitudinal cohorts and in the prediction of complex traits. This dissertation addresses key knowledge gaps in the epigenetic mechanisms of prenatal EDC exposure, utilizing both genome-wide DNA methylation measures and epigenetic clocks, while also examining the potential of using ensemble machine learning to improve the development of epigenetic biomarkers. Chapter 1 provides an introduction to endocrine-disrupting chemicals, epigenetics, the CHAMACOS cohort, and summarizes the specific aims for each subsequent chapter. Chapter 2 focuses on identifying associations between prenatal exposure to common phthalates and epigenetic aging from birth through childhood. Phthalates, a group of pervasive endocrine-disrupting chemicals found in plastics and personal care products, have been associated with a wide range of developmental and health outcomes. However, their impact on biomarkers of aging has not been characterized. We tested associations between prenatal exposure to 11 phthalate metabolites on epigenetic aging in children at birth, 7, 9, and 14 years of age. We hypothesized that prenatal phthalate exposure will be associated with epigenetic age acceleration measures at birth and in early childhood, with patterns dependent on sex and timing of DNAm measurement. Among 385 mother-child pairs from the CHAMACOS cohort, we measured DNAm at birth, 7, 9, and 14 years of age, and utilized general linear regression to assess the association between prenatal phthalate exposure and Bohlin’s Gestational Age Acceleration (GAA) at birth and Intrinsic Epigenetic Age Acceleration (IEAA) throughout childhood. Additionally, quantile g-computation was utilized to assess the effect of the phthalate mixture on GAA at birth and IEAA throughout childhood. We found a negative association between prenatal di (2-ethylhexyl) phthalate (DEHP) exposure and IEAA among males at age 7 (-0.62 years; 95% CI:-1.06 to -0.18), and a marginal negative association between the whole phthalate mixture and GAA among males at birth (-1.54 days, 95% CI: -2.79 to -0.28), while most other associations were nonsignificant. Our results regarding DEHP and the whole phthalate mixture suggest that prenatal exposure to certain phthalates is associated with epigenetic aging in children. Additionally, our findings suggest that the influence of prenatal exposures on epigenetic age may only manifest during specific periods of child development, and studies relying on DNAm measurements solely from cord blood or single time points may overlook potential relationships. Chapter 3 examines associations between prenatal exposure to environmental phenols and DNA methylation in cord blood, investigating the impact of both individual pollutant and complex mixtures on individual CpG sites and regions within the epigenome. Epigenetic marks are key biomarkers linking the prenatal environment to health and development. However, DNA methylation associations and persistence of marks for prenatal exposure to multiple Endocrine Disrupting Chemicals (EDCs) in human populations have not been examined in great detail. Bisphenol-A (BPA), triclosan, benzophenone-3 (BP3), methyl-paraben, propyl-paraben, and butyl-paraben, as well as 11 phthalate metabolites, were measured in two pregnancy urine samples, at approximately 13 and 26 weeks of gestation. DNA methylation of cord blood at birth and child peripheral blood at ages 9 and 14 years was measured with 450K and EPIC arrays. Robust linear regression was used to identify differentially methylated probes (DMPs), and comb-p was used to identify differentially methylated regions (DMRs). Quantile g-computation was used to assess associations of the whole phenol/phthalate mixture with both DMPs and DMRs. Prenatal BPA exposure was associated with 1 CpG site among males and Parabens were associated with 10 CpG sites among females at Bonferroni-level significance. Other suggestive DMPs (p < 1x10-6) and several DMRs associated with the individual phenols and whole mixture were also identified. A total of 10 CpG sites associated with BPA, Triclosan, BP3, Parabens, and the whole mixture in cord blood were found to persist into adolescence. Several identified differentially methylated CpG sites and DMRs were located in genes relevant to obesity, neurodevelopment, and cancer. Chapter 4 explores the application of SuperLearner, an ensemble machine learning methodology, towards the development of epigenetic predictors of phenotypic traits. Epigenetic clocks, a broad nomenclature for epigenetic predictors of biological and chronological age, have proven to be invaluable biomarkers for environmental epidemiology research by enabling the study of the determinants and consequences of biological aging in human populations. Epigenetic predictors are traditionally trained using elastic net regression of the CpG matrix, however the reliability of individual CpGs can be influenced by several factors. Recent evidence has suggested noise reduction benefits of training epigenetic predictors using elastic net regression of the principal components generated from CpG-level data. The substantial data reduction conferred by PCA may open the door for the application of more advanced machine learning to the epigenetic predictor process. We developed a pipeline to simultaneously train three epigenetic predictors: a traditional CpG Clock, a PCA Clock, and a SuperLearner PCA Clock (SL PCA). We gathered publicly available DNAm datasets to generate a i) childhood-specific epigenetic clock, ii) reconstructed the Hannum adult blood clock, and iii) as a proof of concept, a predictor of polybrominated biphenyl exposure using the three developmental methodologies. We used correlation coefficients and median absolute error to assess fit between predicted and observed measures, as well as agreement between duplicates. The SL PCA clocks improved fit with observed phenotypes relative to the PCA clocks or CpG clocks across several datasets. We found evidence for higher agreement between duplicate samples run on alternate DNAm arrays when using SL PCA clocks relative to traditional methods. Analyses examining associations between relevant exposures and epigenetic age acceleration (EAA) produced more precise effect estimates when using predictions derived from SL PCA clocks. The improved performance conferred by the SL PCA methodology may be especially relevant for studies with longitudinal designs utilizing multiple array types, as well as for the development of predictors of more complex phenotypic traits. Chapter 5 then summarizes the conclusions derived from each chapter and provides possible directions for future research.