Prenatal exposures to endocrine disrupting chemicals: The role of multi-omics in understanding toxicity

Abstract

Endocrine disrupting chemicals (EDCs) are a diverse group of toxicants detected in populations globally. Prenatal EDC exposures impact birth and childhood outcomes. EDCs work through persistent changes at the molecular, cellular, and organ level. Molecular and biochemical signals or ‘omics’ can be measured at various functional levels – including the epigenome, transcriptome, proteome, metabolome, and the microbiome. In this narrative review, we introduce each omics and give examples of associations with prenatal EDC exposures. There is substantial research on epigenomic modifications in offspring exposed to EDCs during gestation, and a growing number of studies evaluating the transcriptome, proteome, metabolome, or microbiome in response to these exposures. Multi-omics, integrating data across omics layers, may improve understanding of disrupted function pathways related to early life exposures. We highlight several data integration methods to consider in multi-omics studies. Information from multi-omics can improve understanding of the biological processes and mechanisms underlying prenatal EDC toxicity.

1.0. INTRODUCTION

1.1. Background on EDCs

Gestational exposure to toxicants can negatively impact birth outcomes and have lasting effects on child or adult health including adverse effects on neurodevelopment, growth, adiposity, and metabolism [1, 2]. Environmental exposures, especially during early gestation, may perturb epigenetic programming and metabolic homeostasis thereby setting the stage for altered fetal growth and ultimately disease development later in life [3–12]. Endocrine disrupting chemicals (EDCs) are of particular concern given the widespread use of many of these chemicals throughout the world and their known toxicity. Here we broadly define EDCs by the Endocrine Society’s definition, which states that “EDCs are substances in the environment (air, soil, or water supply), food sources, personal care products, and manufactured products that interfere with the normal function of your body’s endocrine system” [13]. EDCs include chemicals found contaminating the environment such as pesticides (i.e. chlorpyrifos, glyphosate), polychlorinated biphenyls and dioxins, and metals with endocrine disrupting properties (i.e. lead, cadmium). EDCs such as phthalates, phenols, brominated flame retardants, and per- and polyfluoroalkyl substances (PFAS) are found in a range of consumer products from personal care products to food packaging to furniture.

EDCs are widely detected in populations around the world, including in pregnant people [14–19]. Advances in targeted and untargeted approaches for exposure assessment enable dozens to hundreds of chemicals to be analyzed in human samples. In studies utilizing this approach, chemicals including many different EDCs are often detected in the majority of samples analyzed. For example, in a Danish pre-pregnancy study that measured 135 chemicals in each participant, on average 92 chemicals were detected including phthalates, metals, and phytoestrogens [15]. This is problematic since both individual EDCs and mixtures of EDCs are linked to adverse pregnancy and birth outcomes. Gestational PFAS exposures are associated with lower birth weight and hypertensive disorders of pregnancy [20–22]. Phthalate exposures are linked to preterm birth [23]. When modeling the influence of a mixture of EDCs, organochlorines and lead were associated with reduced birth weight in a Canadian cohort [17]. These are just several examples of the published links between EDCs and adverse offspring outcomes from epidemiological and rodent studies, topics that have been reviewed previously [24–27].

In line with the developmental origins of health and disease (DOHaD), the impact of prenatal EDC exposures on offspring health extends beyond birth into childhood and even adulthood [28]. Rodent models of gestational exposure including to bisphenol a, di (2-ethylhexyl) phthalate (DEHP), and lead, report health impacts in adult offspring [29–33]. Adverse health outcomes include increased liver tumors and weight gain for BPA, reproductive toxicity and altered liver metabolism for DEHP, and cognitive defects for lead. While it can be difficult to tease apart the influence of prenatal versus postnatal exposures in human studies, longitudinal epidemiological studies have reported associations between prenatal exposures and adverse health outcomes later in life. For example, prenatal phthalate exposures were associated with disrupted timing and tempo of sexual maturation and reproductive hormone levels in boys and girls [34, 35]. Exposures to various PFAS in utero were associated with increased adiposity among girls in childhood [36], and increased body mass index among 20-year old women [37]. Given the latent effects prenatal EDC exposures can have on offspring health, it is imperative that we understand how these exposures leave a lasting impact on the body that eventually result in disease. Such knowledge will enable better risk assessment and mitigation of hazards along with insight into biological pathways that can be targeted to promote health and reverse damage.

Exposures’ impacts are embodied through a series of persistent changes at the molecular, cellular, and organ level. Molecular signals have the potential to serve as predictors of disease and/or mechanistic links to disease [38–40]. Molecular and biochemical signals or ‘omics’ can be measured at various functional levels – including the epigenome, transcriptome, proteome, and metabolome/lipidome. These omics signatures provide a snapshot of the interaction between the environment and genome at a given point in time. The microbiome –including its composition and metabolic output – is also of interest since this community of microorganisms living within us influences disease risk and how the body handles toxic exposures [41–43]. While all of the molecular and biochemical omics signals are important, the epigenome is the most commonly evaluated in human studies. Animal studies focusing on EDCs have likewise generated primarily epigenomics and transcriptomics data. For example, gestational exposures to EDCs including bisphenols, phthalates, and PFAS have been shown to alter the offspring epigenome, and DNA methylation in particular [31, 44–46]. More recently, studies focused on elucidating EDCs’ effects have started to evaluate the proteome, metabolome, and microbiome. Integrating data via multi-omics approaches will move the field towards a systems toxicology approach and vastly improved understanding of exposures’ impacts (Figure 1).

Figure 1.

The impacts of gestational exposures to endocrine disrupting chemicals can be detected across multiple ‘omics’ which may impact health outcomes and development later in life. (A) Exposure to endocrine disrupting chemicals (EDCs) during pregnancy is associated with (B) alterations in multi-omics processes in both the pregnant individual and the fetus that can be detected in samples collected during pregnancy (i.e. blood) or at birth (placenta, cord blood) and (C) adverse health outcomes in the offspring which may manifest at any developmental time point from infancy to elderly years. Knowledge gained from omics analyses can be used to predict, prevent, or mitigate toxicity of in utero exposure to EDCs.

1.2. Purpose of this review

A substantial body of literature in human and animal studies links prenatal EDC exposures to adverse birth outcomes and to long lasting impacts on offspring development and health. Growing evidence suggests that epigenetic programming is one of the mechanisms underlying these persistent effects. However, other functional molecular and biochemical markers such as the proteome and metabolome are not as well studied [12]. Evaluating multi-omics and integrating data across layers may vastly improve understanding of the functional biological pathways that are disrupted by these exposures early in life. In this narrative review, we first broadly highlight what is known about prenatal EDC exposures and their influence on the epigenome, transcriptome, proteome, metabolome, and microbiome. We then describe promising strategies for integrating data across omics to understand toxicity and disease. Finally, we provide recommendations for incorporating multi-omics into prenatal exposure studies in the future. Throughout, we focus on examples from human and experimental mammal studies of prenatal exposures to EDCs, acknowledging that there is growing research in other toxicological models such as zebrafish and cell culture that is not included here.

2.0. EDCs AND INDIVIDUAL OMICS

2.1. EDCs and Epigenomics

The epigenome encompasses modifications on top of the genome that are heritable, at least mitotically, and regulate gene expression without compromising the integrity of the genetic code itself. While the genomic sequence of an individual remains static throughout their life course and across cell types, the epigenome is dynamic, varying from cell to cell and changing in response to their environment and developmental needs [47–50]. Relevant to this review, the epigenome is known to be altered in response to environmental exposures, including prenatal exposures to EDCs [51, 52]. DNA methylation, chromatin architecture, histone modifications, and noncoding RNAs (ncRNAs) are all examples of epigenetic mechanisms that help regulate gene expression [50, 53]. Techniques used to profile the epigenome may focus on any one of these processes, though DNA methylation is the most commonly studied in humans. DNA methylation refers to the covalent addition of a methyl group to cytosines, often upstream of guanines (called CpG sites). DNA methylation is typically associated with gene suppression, though this varies by genomic context [50, 54]. Bisulfite sequencing (e.g. Whole-Genome Bisulfite Sequencing) and array-based methodologies (e.g. Illumina Epic BeadChip) are epigenomic techniques that are frequently utilized by researchers to quantify DNA methylation across the epigenome [55–57]. NcRNA molecules like microRNAs or long ncRNAs also regulate gene expression through post-transcriptional targeting of mRNAs and silencing of transposons by modulating heterochromatin formation [53, 58]. Another common epigenomic technology is Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq), which identifies regions of open chromatin that are associated with active regulatory regions and gene expression [59]. DNA methylation, ncRNA, and chromatin accessibility, among other epigenetic mechanisms, are important modifications that provide unique layers to the same epigenomic story. Each process affects gene expression and associated cellular functioning in different ways. Although examination of the epigenome using one of these techniques alone can help us gain substantial insight regarding the dynamic epigenome and how it interacts with specific exposures, using these techniques in concert with one another can only strengthen our understanding of this highly complex topic.

Epigenetics is not only vital in modulating developmental processes but can also be altered in response to EDC exposure, meaning this class of chemicals has the potential to induce subsequent changes to normal development through epigenetic means. Epigenetic alterations associated with toxic insult may contribute to toxicity (a mechanism), be an adaptive response against an exposure, or have no biological impact [60]. In the case of non-functional changes, these alterations could still serve as biomarkers of past exposures or predictors of future effects from a given exposure [61]. In either case, biomarkers may herald the onset of potential disease or adverse health effects that are more likely to occur later in life. Understanding epigenetic mechanisms, how they relate to prenatal EDC exposures and their associated diseases may provide insight to health care professionals regarding their recommendations for early intervention and/or disease prevention.

Several studies have described epigenetic modifications as mechanisms by which EDCs exert their toxic effects. Due to reprogramming events occurring in the developing germline, in utero EDC exposures can be particularly detrimental to health outcomes later in life [51, 53, 59, 62]. To begin to unravel the complex ways in which prenatal EDC exposure affects various health outcomes, we must first understand the concept of epigenetic reprogramming. This process refers to the erasure and then re-writing of epigenetic marks, such as DNA methylation and histone modification patterns, across the epigenome during critical periods of embryonic development [63]. Epigenomic reprogramming events are essential for normal development and cellular differentiation processes but can be disrupted by prenatal EDC exposures [47, 64]. Dysregulation of epigenetic reprogramming at early stages propagates across tissues, potentially leading to aberrant gene expression patterns and changes to normal cellular functioning that persist throughout one’s life [50, 63, 65]. Prenatal exposure to EDCs has been shown to induce changes in DNA methylation at genes involved in hormone signaling and immune function in the developing fetus. For example, in a sample of 142 mother-infant pairs from an urban-industrial population (Duisburg Birth Cohort Study), researchers quantified DNA methylation levels in cord blood using the Illumina EPIC BeadChip array, as well as levels of the persistent EDCs - PCBs and PCDD/Fs - in blood collected during late pregnancy. An epigenome-wide association study was conducted and identified associations between the exposures and DNA methylation at a variety of genes. Exposure to congener PCDD66, for instance, was associated with differential methylation of PAQR4 in males; this gene encodes a protein that is related to Isolated Growth Hormone Deficiency (Type Ib). In the same study, researchers identified significant associations between exposures and methylation at genes involved in gene expression regulation, including associations between PCB126, PCDD66 or PCDD70 with DNA methylation for H2BC5 in males. This gene encodes one of four core histone proteins responsible for the nucleosome structure, and plays a major role in modulating DNA accessibility. The gene was part of a differentially methylated region also identified by this study which includes transcription factor-binding site motifs, and positions overlapping or directly neighboring YY1 and TBP [51]. In our study, using the Michigan Mother-Infant Pairs (MMIP) cohort, first-trimester maternal exposure to BPA and phthalates were found to be inversely associated with DNA methylation in cord blood at loci important for growth, metabolism, and development—LINE-1 repetitive elements, imprinted genes (H19, IGF2) and non-imprinted (PPARA, ESR1) genes [62]. These and other studies have continued to provide evidence that prenatal EDCs exposures alter the offspring epigenome (see reviews [46, 66–68]).

Epigenomics is a fascinating, highly relevant field of study that has proven to be profoundly useful in exploring questions related to toxicity of prenatal EDC exposures, but as with any scientific research it is met with many challenges and knowledge gaps. One of such challenges is the fact that DNA methylation does not always translate to transcriptional alterations. While much can be inferred about gene expression in epigenomic studies, especially when changes in DNA methylation are observed at regulatory regions of genes such as promoters, researchers are limited in their ability to interpret functional outcomes related to epigenetic changes in the absence of transcriptomic data. The establishment of expression quantitative trait methylation (eQTM) catalogues for tissues common in prenatal studies (i.e. cord blood, placenta) would be useful for inference. eQTMs are specific loci where DNA methylation status is known to regulate expression of a gene. Publicly-available catalogs built using data from cohort studies are available for children’s blood [69], but to our knowledge these have not yet been generated for cord blood. Catalogues of eQTMs are useful for interpretation of biological relevance of EWAS studies, even when transcriptomic data is not available [70].

While DNA methylation is the most researched, it is not the only epigenetic mechanism. Hydroxymethylation of DNA, for example, is a frequently overlooked epigenetic modification, which has its own role in regulating gene expression [71]. Common methods used to assess DNA methylation including the Infinium arrays require bisulfite conversion of DNA first. Methods based on bisulfite conversion cannot distinguish between 5-methylcytosine (5-mC) and 5-hydroxymethylcytosine (5-hmC) [72]. While the structures of 5-mCs and 5-hmCs are comparable to one another, the two marks have distinct and often opposing effects on gene expression. If both marks are present at a given site, this creates the potential for them to mask the effects of one another if jointly quantified, thus reducing the number of significantly differentially methylated sites identified in a total methylation EWAS [72–74]. In our recent study in the MMIP cohort, for example, PFAS concentrations were measured in maternal plasma collected during the first trimester and DNA methylation was quantified in cord blood at birth using a method that distinguishes between 5-mC and 5-hmC. To accomplish this, samples were treated with sodium bisulfite and oxidative bisulfite treatment in parallel prior to hybridization to the Infinium EPIC array. While PFAS exposures were associated with total methylation at only a few loci, when the two marks were made distinguishable from one another the exposures were associated with 5-hmC or 5-mC at hundreds of loci, showcasing the importance of evaluating them as specific marks [72]. For the most part, PFAS were associated with higher levels of 5-hmC and lower levels of 5-mC. These associations essentially cancel each other out when the outcome is total methylation, resulting in false negatives. In addition to challenges related to results interpretation and the different types of DNA modifications, the literature is currently lacking in explorations of other epigenetic processes, such as those involving histones and small ncRNA molecules.

The complex, dynamic nature of the epigenetic regulation of gene expression is another major challenge to the field. In addition to there being multiple epigenetic processes influencing one another, they can also be influenced by various other factors, including genetic background, developmental stage, environmental factors, and exposure to chemical mixtures [60]. The limited availability and access to human samples from relevant tissues is a challenge. Even in instances where human samples are available, researchers are often limited to tissues such as cord blood, blood leukocytes, saliva, or placenta, which are more accessible and used as proxies for the tissue of interest [52, 62, 75–80]. For instance, researchers interested in the effects of EDCs on neurodevelopment may be interested in the epigenetic alterations occurring in the brain, which is not feasible or ethical to collect in population studies. Due to these ethical considerations, most research projects utilize proxy tissues like cord blood or placenta. One common solution to this quandary is to pair epidemiological studies with in vivo or in vitro experiments that can focus on the exact tissue of interest in a model organism or cell line. These techniques are burdened with their own set of assumptions regarding topics such as inter-species variability and complexities of a true physiological environment. Despite these limitations, these types of experiments partnered with human data can help to build a stronger understanding of EDC toxicity through epigenomic alteration. While basic research on the long-term effects of prenatal EDC exposure on human health outcomes has been conducted in animal and cell models through consortia such as TaRGET II (Toxicant Exposures and Responses by Genomic and Epigenomic Regulators of Transcription Program), longitudinal follow-up of exposed individuals in human populations for persistence of effects are less common [81]. Such longitudinal analyses are essential and would be possible in both future and pre-existing cohorts, such as those comprising the US-based NIH ECHO (National Institute of Health’s Environmental influences on Child Health Outcomes) consortium or the international Pregnancy and Childhood Epigenetics (PACE) consortium [82–84]. Extant data and biospecimens can be used to corroborate findings from in vitro and in vivo analyses by assessing the relationships between prenatal EDCs exposures, and adverse health outcomes persisting through childhood, adolescence and eventually adulthood in these cohorts.

Epigenomic techniques are also associated with some technical challenges, like sample preparation, sequencing depth, and data analysis, which can influence the quality of the data obtained by epigenomic techniques. Obtaining and preparing samples for ATAC-seq, for example, can be particularly challenging, as samples need to be processed in a time-sensitive manner to preserve nuclei prior to freezing [85]. The field of epigenomics can make efforts to overcome these challenges by developing better sampling techniques and pairing them with pre-existing technologies to build on the strengths and weaknesses of different approaches. Furthermore, protocols for collection of biological materials, sample processing, and data analysis should be standardized to ensure the reproducibility of results across studies.

2.2. EDCs and transcriptomics or proteomics

Transcriptomics involves the study of all the ribonucleic acid (RNA) molecules present in a given cell, tissue, or whole organism. RNAs, or “transcripts” of the DNA, are evaluated using transcriptomics techniques to better understand their location, expression, function, and degradation patterns under varying conditions [86]. Transcriptomic techniques may quantify or detect all types of transcripts including protein-coding messenger RNAs (mRNAs) and non-coding RNAs (transfer RNAs (tRNA), ribosomal RNA (rRNA), small nuclear RNA (snRNA), small nucleolar RNA (snoRNA), short interfering RNA (siRNA), microRNA (miRNA), piwi-interaction RNA (piRNA), long non-coding RNA and pseudogenes) that perform a wide range of cellular activities [87]. RNA and protein expression are not always reliably related depending on differing mechanisms of post-translational regulation and degradation. Thus, proteomics, or the study of sets of proteins (also referred to as the “proteome”) provides an additional layer of context in the evaluation within a biological system [88]. The proteome describes the post-translational impact of exposure and can serve as a surrogate composite measure of the biological systems’ response to stressors, the functional output of the epigenome and transcriptome, and may provide mechanisms linking exposure to disease. Both transcriptomics and proteomics have vast potential for implementation in clinical diagnostics [89], therapeutic target identification, and improvement of risk assessment including for occupational and environmental exposures [90–92]. Similar to other omics data, transcriptomic and proteomic analyses help to infer the biological processes being impacted by EDC exposure in utero and how they might relate to adverse health outcomes in the child, both in the immediate (birth outcomes) and long-term (metabolic disease or cancer). Unlike epigenomic analyses, the epidemiological literature reports limited examples of associations between prenatal exposure to EDCs and transcriptomics or proteomics.

Remy and colleagues evaluated cord blood transcriptomics in relation to prenatal exposures to dichlorodiphenyldichloroethylene (p,p′-DDE), polychlorinated biphenyl-153 (PCB-153), perfluorooctanoic acid (PFOA), and perfluorooctane sulfonate (PFOS) from a birth cohort (n=195) initiated as part of the 2nd Flemish Environment and Health Survey (FLEHSII, 2007–2011). A total of 39 metabolically relevant transcription factors were identified as significantly enriched in samples with elevated EDC exposures in a whole genome oligo microarray. Their results included enrichment of the glucocorticoid receptor (with higher exposures to p,p′-DDE and PCB-153) and the progesterone receptor (with PFOA and PFOS). Pathways associated with metabolic disorders were noted as significantly enriched in relation to p,p’-DDE and the authors suggest a potential link between specific metabolic targets of prenatal EDC exposure and adverse health outcomes later in life [93]. As p,p’-DDE is known to accumulate in fat tissue, the interpretation of transcriptomics findings merits careful consideration of a given EDC’s toxicokinetic and toxicodynamic properties. In a recent follow-up analysis of the FLEHSII birth cohort (n=192), expression of five metabolism-related genes (BCAT2, IVD, SLC25a16, HAS3, and MBOAT2) were identified as associated with an exposure (either p,p-DDE, PCB-153, PFOA, or PFOS) and birth weight [94].

PFAS were evaluated for associations between maternal exposure and transcriptomic profiles in neonatal cord blood from the Norwegian BraMat cohort (n=66), a subset within the Norwegian Mother and Child Cohort Study (MoBa) (2007–2008). Study authors identified 52 PFAS exposure-associated gene transcripts including some that are commonly associated with rubella titers and/or common cold episodes following targeted microarray gene expression profiling. Several immunomodulatory gene transcripts and other immune-associated genes were suggestive of a connection between prenatal PFAS exposure and impaired immune function in early childhood with potential PPAR or NF-κβ-mediated mechanisms of action [95]. Chemicals impacting the success of response to vaccination and the body’s ability to amount a sufficient immune response contribute to concerns regarding the prevalence of, potentially, vaccine-resistant infectious diseases. Transcriptomics analyses will aid in improving our understanding of this novel EDC-associated adverse health effect.

In a study of pregnant women in Guiyu (e-waste exposed) and Haojiang (non-exposed group), China, placental concentrations of lead and cadmium were evaluated for proteomic associations and various birth outcomes including gestational age, neonatal weight, and neonatal body length. Elevated levels of cadmium in the exposed group were associated with a significant decrease in neonatal weight and length and an increase in gestational age. A total of 54 proteins were differentially expressed in the exposed placental tissue compared to the reference group including 23 up-regulated and 31 down-regulated proteins. These proteins were predominantly associated with metabolic processes (43.75%) that relate to fetal growth restriction. The study authors describe how exposure in utero shows metal-induced oxidative stress that alters amino acid transfer, enzyme activity, and hormonal secretion that may ultimately lead to intrauterine growth and developmental restraint. Proteomic analysis identified biomarkers essential to better understanding the full picture of how prenatal exposure impact health [96].

Overall, epidemiological transcriptomic and proteomic analyses conducted in studies focused on prenatal exposures to EDCs are scarce and much of the available data utilizes controlled animal models. For example, neurotoxic effects were observed in the brains of offspring from pregnant rats exposed to the organophosphate flame retardant triphenyl phosphate (TPHP) and a mixture of organophosphate flame retardants. Transcript levels of genes responsible for the transport and synthesis of serotonin and metabolic processes related to neuroactive metabolites in the serotonin and kynurenine pathways were upregulated [97]. Proteomic analysis of hippocampal tissue from mice prenatally exposed to TPHP, revealed 531 differentially expressed proteins involved in axon guidance, synaptic function, neurotransmitter transport, exocytosis, and energy metabolism [98]. Female mice exposed to BPA in utero had dose-response dependent differential expression in 17 proteins within the thyroid tissue. Two identified proteins (Anxa6 and Vcp) were identified as potential thyroid cancer proteomic biomarkers for BPA-early life exposure [99]. Female offspring of rats exposed to BPA in utero had 21 proteins that were differentially expressed in a proteomic analysis of rat mammary glands at 21 and 50 days old. Several of the proteins identified have been previously associated with an increased susceptibility to cancer development and were indicated as possible biomarkers for breast cancer [100, 101]. Male and female rats exposed to BPA prenatally showed disrupted cell cycle, lipid homeostasis, and hormone balance in liver proteomic and transcriptomic profile analyses. These alterations in transcript and protein levels were concordant with the observed increases in body weight and anogenital distance in offspring [102]. Thirty-five differentially expressed genes were identified in the transcriptomic profile of male offspring exposed to di-2-ethylhexyl phthalate (DEHP) in utero. Among these transcriptomic changes were genes associated with insulin resistance, oxidative stress, and hypothyroidism [103]. These and other animal-based transcriptomic and proteomic analyses provide an important basis for further investigation among existing and future human cohorts.

Proteomic techniques have also been applied in non-prenatal evaluations of EDCs. For example, data from the Child Panel Study (n=156 school-aged children), a subpopulation of the Human Early-Life Exposome (HELIX) project including participants from five of the European population-based birth cohorts that are part of HELIX, was evaluated for urinary concentrations of non-persistent EDCs (as a pooled sample of 15 urines) and associated changes in 36 plasma proteins. Reproducible, statistically significant associations between urinary organophosphate pesticides and granulocyte-colony stimulating factor, triclosan and leptin, and phthalate metabolite methylbenzylpiperazine (MBzP) and epidermal growth factor were described in the study findings. Biomarkers identified during proteomic analysis of plasma samples from children may have some involvement in explaining the relationship between EDCs and obesity or insulin resistance [104]. These results give us insight into how the application of proteomic techniques in existing human cohorts warrant further use in prenatal EDC exposure studies.

Compared to genomic analysis, evaluation of the transcriptome and proteome is inherently more complex. Much like the epigenome, the transcriptome and proteome differ from cell to cell and between timepoints. Specific genes are expressed in distinct cell types, and this introduces additional variation in interpreting the results of transcriptomic and proteomic analyses. As discussed in the epigenomics section, tissue types evaluated in human studies are often limited to blood and implications for health must be interpreted carefully. Serum is often used for proteomics; analytes can represent proteins secreted or shed from cells and tissue types throughout the body, and results interpretation should take that into account [105]. RNA expression levels can be impacted by common, frequent experiences like eating, physical activity, or diurnal cycles. In addition, various forms of techniques exist for analyzing the transcriptome and proteome. These methods vary depending on the research question and design. Consideration of differing platforms and technologies (e.g., microarrays, RNA-seq, single-cell RNA-seq, 2-D gel electrophoresis, high performance liquid chromatography [HPLC]) is needed with consideration of advantages and limitations and how these impact the ability to address a given research question [106, 107]. In epidemiological studies, samples collected in the clinic or field need to be properly preserved for downstream RNA or protein prior to freezing which is one reason many longstanding cohorts cannot add these analyses using archived samples.

Further, many proteins form complexes with other proteins or RNA molecules and may only function in the presence of these molecules; thus, their independent detection does not necessarily indicate a biologically relevant change. Another major concern in proteomics identification is that the protein sequence libraries often leave out alternative splice isoforms or include them with limited annotation – limiting peptide and protein identification due to the presence of peptide fragments that may correspond to more than one protein in a given database [108, 109]. Many proteins and protein isoforms identified during sequencing lack any obvious function as our understanding of the cellular and biochemical role of the majority of proteins remains quite limited. Techniques that determine the subcellular location of each transcript or protein (e.g., spatial transcriptomics) will continue to aid in the identification of protein complexation and potential biological function. Artificial intelligence and machine learning techniques are also being employed to improve biocuration efforts for extracting and organizing biological information into a structured form to interpret the relationships between transcripts, proteins, drugs, physiochemical properties and functions, and any disease interactions [110].

2.3. EDCs, metabolomics, and lipidomics

Metabolomic analyses are one of the most common omic techniques employed in the evaluation of EDCs as metabolic changes are intrinsic to the expected adverse health effects associated with endocrine disruption. Defined as the measurement of all metabolites and low-molecular-weight molecules within a biological system (e.g. cell, tissue, or biofluids), the metabolome encompasses a wide array of compounds that serve as substrates, intermediates, or end products of enzyme reactions [111–113]. Due to their variety of physiochemical properties, metabolites are often subcategorized during analysis by their polarity, functional groups, or structural similarities [111, 113]. Lipidomics, an emerging subcategory of metabolomics, studies cellular lipids and lipoproteins and provides a more specific window into essential biological functions including cellular membrane and barrier dynamics, signaling, and energy storage/utilization [114–116]. As measurable changes in metabolites often coincide with pathological profiles of disease and some monogenic disorders, metabolomic and lipidomic techniques provide a substantial opportunity for an improved understanding of disease origins, presentation, and potential treatment modalities.

Metabolomics approaches have the ability to describe the molecular profile of a physiological activity that may be related to fetal programming and can also characterize internal exposures or health states during pregnancy [117]. As such, the evaluation of metabolite and lipid changes has been used as a screening method for specific prenatal/pregnancy disorders including congenital anatomic defects, aneuploidy, single gene disorders, preterm labor, fetal growth restriction, preeclampsia, and normal pregnancy physiology [118]. EDCs are known to disrupt the regulation of natural hormones and metabolite profiles provide insight into any potentially associated adverse metabolic phenotypes following exposure [119, 120].

Examples of metabolomic applications in prenatal exposure studies have become increasingly popular over the past five years. After identifying a strong positive association between PFAS exposure and offspring breast cancer risk, Hu and colleagues evaluated the association between prenatal exposure to two PFAS compounds, 2-(N-Ethyl-perfluorooctane sulfonamido) acetic acid (EtFOSAA) and PFOS, and changes in the maternal metabolome [121]. EtFOSAA and PFOS were associated with impacts to critical metabolites of the urea cycle and some non-essential amino acids. PFOS exposure was associated with alterations in carnitine metabolism, supporting evidence of impacts to fatty acid metabolism [122]. The study authors describe this finding as a potential link between prenatal PFAS exposure and breast cancer risk in offspring due to the essential role of urea cycle intermediates and related amino acids in cancer cells to reprogram metabolism [123].

Another study highlighting the impactful application of metabolomics techniques evaluated the association between early pregnancy phthalate exposure and sex steroid hormones in relation to newborn genital outcomes in The Infant Development and the Environment Study (TIDES) cohort (n=591 mother-infant dyads). Sathyanarayana and colleagues collected first trimester urine samples and measured specific-gravity adjusted concentrations of the DEHP metabolites, mono-2-ethylhexyl phthalate (MEHP), mono-2-ethyl-5-hydroxy-hexyl phthalate (MEHHP), mono-2-ethyl-5-oxy-hexyl phthalate (MEOHP), and mono-2-ethyl-5-carboxypentyl phthalate (MECPP). Other phthalate metabolites measured included monoethyl phthalate (MEP), monoisobutyl phthalate (MiBP), monobenzyl phthalate (MBzP), monobutyl phthalate (MBP), mono(carboxynonyl) phthalate (MCNP), and mono(carboxy-isooctyl) phthalate (MCOP). Associations between first trimester MiBP, MBzP, MEOHP, and MEHHP metabolite concentrations and elevated levels of estrone and estradiol in the maternal blood were identified. Circulating free testosterone was decreased with increasing concentrations of MCNP and MECPP. In addition, the study authors found an inverse relationship between maternal free testosterone and the prevalence of male genital abnormalities at birth [124]. These results continue to implicate the sensitivity of the maternal metabolome to EDC exposures and its subsequent impact on fetal development.

Existing evidence of an association between in utero phthalate exposure and cognitive ability in children prompted further evaluation by Jones and colleagues using an untargeted metabolomics approach. In a subset of the Growing Up in Singapore Towards Healthy Outcomes (GUSTO) cohort (n=373), the metabolomic profile of maternal hair samples collected at 26–28 weeks gestation were analyzed using gas chromatography-mass spectrometry. The study authors identified 27 maternal metabolites that were statistically significantly associated with lower language learning ability in children [125]. This study highlights the utility of metabolomic approaches that incorporate non-invasive sampling methods for identifying significant associations between the maternal metabolome and offspring outcomes.

Despite its extensive power as a high-throughput and precision health analytic technique, metabolomics comes with its share of challenges. Similar to the epigenome, transcriptome, and proteome, the metabolome exhibits tissue specificity and temporal sensitivity. Large human interindividual variabilities exist among measurements as the metabolome is highly differential depending on age, diet, exercise, genetic variation, and other exposures. In addition, metabolome profiling is conducted using an array of nuclear magnetic resonance and/or mass spectroscopy techniques; but, due to the complexity of the metabolome and dynamic concentrations, no single analytical platform exists that can be employed to detect all metabolites in a biological system [126]. This lack of congruency across analysis platforms has limited the development of a robust reference library for metabolite identification and corresponding biomolecular pathways. Similarly, the heterogeneity of sampling methods reduces the validity of data pooling techniques across cohorts. Incomplete understanding of a given metabolite’s biological function in regulating pregnancy or fetal development often requires addition experimentation and limits the holistic interpretation of a metabolomic analysis in EDC exposure studies. However, as future studies continue to implement and improve metabolomic and lipidomic techniques, the benefits of including a multifaceted, high-throughput approach will inevitably improve our understanding of how ongoing prenatal exposures to EDCs impact the health of future generations.

2.4. EDCs and microbiome and metagenomics

The community of Bacteria, Archaea and other microorganisms found in the gastrointestinal tract is collectively referred to as the gut microbiome [42]. This complex ecological system has been shown to affect health and disease risk, nutrient, drug, and toxicant metabolism, and mental health outcomes; and has thus been referred to as an additional organ of the body [127]. Specifically, differential abundances of key taxa and changes in taxa over the lifespan within an individual have been shown to impact human health [128]. Taxonomic classification is most frequently completed through 16S RNA gene sequencing, which provides a species-level resolution of the Bacteria and Archaea present in a sample. Increasingly, metagenomic sequencing is implemented to obtain a species-level resolution of Bacteria, Fungi, Viruses, and other microorganisms. Shotgun metagenomic sequencing therefore provides a more complete classification of the gut microbiome, including rare taxa. This is relevant because rare taxa have been identified as drivers of microbial ecology and their abundance has been linked to human health outcomes [128–130]. When considering changes to the gut microbiome, metrics of the ecological structure are commonly used. Alpha diversity refers to the species richness, or the number of different taxa present in a sample. Beta diversity refers to how different each sample is from each other. These metrics help to compare differences in the entire community of microbes present in a given sample and how they interact with one another. The gut microbiome is a crucial component of nutrient metabolism, for example, Vitamin K is synthesized by gut bacteria [131]. It also plays a role in prescription drug metabolism and impacts host response to these interventions [132, 133]. Unsurprisingly, toxicant exposure has been shown to alter the microbiome in non-pregnant adult populations. Exposure to EDCs such as PFAS, pesticides, DEHP, triclosan, and triclocarban is associated with perturbations in microbial diversity, community composition, and microbial metabolites which are metrics with known relevance to human health outcomes [43, 134–136].

Given the intricate relationship between commensal gut microbes and human health, the gut microbiome must be considered when examining the effect of prenatal exposure to EDCs. Of particular relevance is the impact of the maternal gut microbiome on infant health. The maternal microbiome has been shown to play a major role in the initial seeding of the infant gut microbiome through contact with the vagina, mouth, skin, and through breastmilk [137]. Hormonal changes during pregnancy can also impact the relative abundances of certain bacterial genera, further highlighting the need for comprehensive investigation of the effect of EDCs on the gut microbiome of the pregnant person and the offspring [138].

There are currently only a few studies that have assessed relationships between prenatal EDCs and the microbiome of the pregnant person or offspring. Laue et al. measured polybrominated diphenyl ethers (PBDEs) and polychlorinated biphenols (PCBs) in maternal plasma in early pregnancy (n = 18) and at delivery (n = 25) in women who participated in the Gestation and Environment Cohort in Quebec, Canada [139]. PBDEs measured included PBDE-47, PBDE-99, PBDE-100, PBDE-153, and PCBs included PCB-153, PCB-180, and PCB-138. PBDEs are used as flame retardants and had historically been applied to many consumer products, leading to nearly universal exposure in human populations. They are associated with hormonal, thyroid, and neurodevelopmental conditions, but the mechanism for these outcomes is largely unknown [140]. PCBs, which were banned under the Toxic Substances Control Act in 1979, have also been associated with decreased growth in utero and immunological and thyroid deficits [141–144]. Both of these groups of chemicals have been associated with changes in the microbiome in in vitro and animal models, but their effects on the maternal microbiome and child health have not been examined [145, 146]. Children were followed up at 6–8 years of age and provided a fecal sample for 16S RNA sequencing of the microbiome. While neither group of chemicals was associated with microbiome alpha (inter-sample) or beta (intra-sample) diversity at either time point, exposure to PBDEs and PCBs in early pregnancy was significantly associated with differences in bacterial structure. This means that the relative abundances of certain bacterial families were different in more highly exposed children. Specifically, higher exposure to PCB-153, PCB-180, and the sum of PCBs was associated with higher relative abundance of Propionibacteriales and the family Propionibacteriaceae [139]. This study did not evaluate clinical outcomes related to these differences in bacterial structure, but the results do draw attention to the potential importance of early pregnancy PBDE and PCB exposure on the child microbiome. Detection of these changes in such a small sample as much as 8 years after the exposure assessment also demonstrates the potential for these chemical to impact the long-term ecological structure of the gut microbiome, which tends to be relatively stable after the third year of life [147]. Whether this is related to modifications of the maternal gut microbiome which are then passed onto the infant shortly after birth, effects of PCDEs and PCBs crossing the placenta, or continued exposures to these toxicants during childhood is unknown. This pilot study suggests that prenatal exposures play a role in shaping the child’s gut microbiome, which could impact long term health outcomes. However, given the small sample size of this study, caution should be exercised in generalizing the results.

3.0. INTEGRATING MULTIPLE LAYERS OF OMICS

EDCs can impact the epigenome, transcriptome, proteome, metabolome, and lipidome. Evidence is emerging that EDCs also affect the microbiome. However, making sense of EDC-omics associations and applying them to understand toxicity and disease risk is difficult, especially when most studies only examine one or two omics at a time. Incorporating a systems toxicology approach by evaluating multiple levels of omics biomarkers can improve decision-making about safe exposure levels, aid discovery of biomarkers of disease, and help to identify susceptible individuals or therapeutic targets [148, 149]. This approach can provide a holistic picture of biological mechanisms underlying health risks from prenatal exposures to EDCs. Multi-omics can reveal common disrupted biological pathways among EDCs that impact children’s health and could inform adverse outcome pathway development. Moreover, these methods may reveal common signals among certain endpoints (i.e., child growth) or diseases that could contribute to risk assessment.

But how do multi-omics analyses really work? Hopeful toxicologists may envision multi-omics as a magic box that reveals one or two clear biological pathways impacted by exposure. In this scenario, scientists read in omics data across multiple levels (for example, blood DNA methylome, transcriptome, proteome, and metabolome) from newborns with low and high exposure to an EDC of interest. A machine learning program then relates the data together and identifies biological pathways that are changing across omics layers. The identified pathway(s) (for example, PPARα signaling) would enhance understanding of toxicity and become the target for future intervention development. Other EDCs with similar chemical properties could be screened against this target for efficient evaluation of understudied toxicants. If studies used omics data from a diverse population, multi-omics analyses could identify several biological pathways that are altered in subgroups of people with varying susceptibility to the exposure.

The application of multi-omics to toxicology and perinatal epidemiology is in its infancy but can take advantage of methods used throughout the biomedical sciences. At the time of writing this review, the number of bioinformatic and statistical methods to analyze multi-omics data are increasingly rapidly and improving over time. These methods grapple with inherent challenges including high dimensionality, missing data, and heterogeneity across omics datasets in terms of number of variables, data type, and data distribution. There are a variety of data integration strategies which differ by underlying statistical methods and the stage of the analysis during which omics layers are combined. A simplified categorization of multi-omics methods is listed in Table 1. We will introduce each approach along with some general considerations when conducting a multi-omics analysis. Describing all the methods developed for these approaches and their mathematical and statistical details are beyond the scope of this review; see other reviews for in depth information on a variety of methods and approaches [150–153].

Table 1.

Categories of Multi-Omics Data Integration^*

Type	Advantages	Limitations	Example method
Early Integration	Can identify a great number of independent signals	Imbalanced when numbers of variables or data distributions are dissimilar across type of omics data	LUCID [154]
Mixed	Removes heterogeneity between datasets to facilitate analysis with a variety of methods	Most methods assume equal importance of each type of omics data	OmicsNet [155, 156]
Intermediate	Produces omics-specific and multi-omics outputs	Pre-processing and feature selection necessary first to prevent issues from data heterogeneity	JIVE [157]
Late Integration	Picks up related signals from each omics layer	Does not consider interactions between omics	MOGONET [158]
Hierarchical	Incorporates knowledge on regulatory relationships between omics into the integration	Methods designed to fit omics with specific relationships (i.e. epigenetics with gene expression)	iBAG [159]

^*Categorizations as first introduced by Picard et al. 2021 [150].

Early integration approaches involve concatenating all omics datasets from the same samples into one large data matrix. Analyses are conducted on the concatenated dataset. Advantages of this approach include relative simplicity and the application of machine learning methods to make connections across omics layers. Disadvantages of this approach include imbalances when one omics dataset contains many more data points then another and when features have different types of data distributions. For example, data from the DNA methylome (>800,000 data points via the EPIC array), transcriptome (up to 20,000 genes via RNA-seq), and proteome (1,000 proteins via SomaScan) functionally complement each other, but commonly used laboratory methods return datasets of vastly different dimensions. Feature selection can be used on the concatenated dataset to reduce the number of features. Various methods have been developed for early integration. One that has been used in environmental research is Latent Unknown Clustering Integrating Multi-Omics Data (LUCID) [154]. This method estimates latent unknown clusters that serve two purposes; these clusters distinguish subgroups of participants with varying risk for the outcome and identify sets of exposures and omics biomarkers that are together related to the outcome. LUCID works with three types of data: 1) expected causal factors (genomics and exposure data), 2) expected mediators or biomarkers (omics data beyond genomics), and 3) outcomes. For each type of data, input can be low or high dimensional. LUCID can be used for outcome prediction, but it also provides estimates for the association between exposures and subgroups determined by clustering the omics data.

Mixed integration approaches transform each omics dataset first before combining and then analyzing. The transformation process reduces dimensionality and noise in each dataset while also ensuring that all data types are similar before combining [150]. Many published methods exist to handle transformation, and these methods are often built around kernel-based, graph-based, or deep learning based methods [150]. By reducing heterogeneity between omics datasets, the transformed outputs can be combined and perform well in machine-learning models. However, the interpretation of the transformed data is not straightforward. Intermediate integration approaches jointly integrate multiple omics datasets without simply concatenating them (early integration) or first transforming all layers in a similar manner (mixed integration). Outputs from these methods include omics-specific and across-omics variables, enabling retention of some omics-specific information while also identifying interactions across layers [150]. One such method is Joint and Individual Variation Explained (JIVE) [157]. This method can read in multiple omics datasets and break them down into a joint structure shared across omics, unique structures for each omic dataset, and noise. To test associations between exposures and multi-omics, the low-dimensional joint and individual structures are used.

Late integration involves modeling the relationship of interest separately on each omics dataset and then combining the results [150]. This approach takes advantage of well-developed statistical methods for each omics dataset. However, this approach cannot consider interactions between omics layers. Much like the early integration approaches, late integration is user friendly and results interpretation is intuitive. However, late integration lags behind other approaches when it comes to harnessing complex relationships between omics layers. Hierarchical approaches bring in prior knowledge about the regulatory relationship between omics layers. For example, DNA methylation regulates gene expression and translation (proteomics) follows gene expression. Hierarchical approaches incorporate directionality and known information about gene networks into the integration. While these methods are relatively more complicated and limited to omics with well-characterized inter-relationships, results point to clear genes or biological pathways that are affected [159].

Cross-cutting themes across approaches include the need to reduce noisy or redundant variables within the omics datasets. Feature selection or feature extraction methods can be used to achieve this. Feature selection can improve computing power, model efficiency, reduce model overfitting, and make the number of features within each omics dataset equal. Feature extraction reduces noise and redundancy in the data, but results can be harder to interpret since new variables are created from the original omics features. In addition to considering dimension reduction approaches, available multi-omics methods include both unsupervised and supervised methods. Many of the multi-omics methods were designed for downstream prediction (i.e. identifying a set of omics biomarkers that predicts disease). Some but not all methods are able to estimate associations between exposures with omics and disease, and this is often the question of interest for environmental studies.

4.0. SUMMARY AND FUTURE DIRECTIONS

Prenatal exposures to EDCs are of global concern as associations between these chemicals and adverse infant and child outcomes are increasingly reported in the scientific literature. EDCs such as BPA, DEHP, and PFOA have been restricted or banned in some nations, but there remain a large and growing number of EDCs with limited toxicity information and no restrictions. Due to the demonstrated potential for gestational exposures to program long-term health impacts on offspring, there is an urgent need to understand how and why these chemicals lead to toxicity. Early life exposures cause subtle and persistent changes at the molecular and cellular levels. In this review, we provided examples of EDCs’ effects on the offspring epigenome, transcriptome, proteome, and metabolome. We also discussed emerging research on the environmental impact on the microbiome. It is clear there are multiple avenues through which EDCs impact child health. However, epigenomics is the most widely studied. By drawing information from other omics layers, such as transcriptomics, metabolomics, and proteomics, researchers can create a more comprehensive understanding of the biological processes and mechanisms underlying prenatal EDC toxicity.

The analysis of multiple layers of omics signatures in individuals exposed to EDCs has the potential to vastly improve our understanding of toxicity. Integration across these omics layers will accelerate advancement. The promise for systems toxicology and precision environmental health is ultimately to inform risk assessment, aid prediction of long-term adverse outcomes, and inform protective interventions, if necessary, for children previously exposed [160–162]. While multi-omics studies in environmental health are still rare, the field is ready to launch into this space. Consortia focused on early life exposures and child health in both animal models and humans are starting to amass omics data. For example, the National Institute of Environmental Health Sciences’ TaRGET II Consortium, uses human relevant mouse models to explore whether epigenetic signatures induced by developmental exposures to EDCs including lead and DEHP occur in both surrogate (e.g., blood) and target tissues (e.g., brain or liver) [81]. Importantly, TaRGET II captures data at multiple time points in life, from multiple tissues, and across omics layers (transcriptomics, chromatin accessibility via ATAC-seq, DNA methylation via WGBS, and histone modifications via ChIP-Seq). Consortia of birth cohorts and children’s studies, such as the PACE Consortium [84] and the National Institutes of Health ECHO Program [163], are starting to combine data across cohorts to identify epigenetic associations with early life exposures. Some of the cohorts included in these consortia have at least two omics datasets (DNA methylome, metabolome, transcriptome, microbiome), with additional cohorts planning to generate these data on archived samples.

Existing consortia and other ongoing or new studies focused on the impact of prenatal EDC exposures on child health can take advantage of rapidly advancing methods and strategies in multi-omics. When designing the study, selection of omics to measure, the timing and type of sample omics will be analyzed in, and the data analysis strategy should be carefully planned to fit the study question. Sample collection protocols for new studies should capture samples that are suitable for exposure assessment and all omics endpoints that would benefit the study in the future. While it is relatively easy to store and collect adequate samples for DNA analyses (i.e. frozen cord blood preserved in EDTA for genomics or DNA methylation analysis) or metabolomics (frozen plasma), analysis of proteins, chromatin structure, and RNA require additional steps prior to freezing to preserve samples for successful analysis. The research question will dictate the ideal timing for sample collection and from who (parent or child). For example, a study focused on understanding the molecular mechanisms linking an EDC exposure to preterm birth could assess multi-omics of the placenta – an organ essential to fetal development. Alternatively, omics of the pregnant participant could be assessed instead at multiple time points throughout gestation to determine whether changes to the metabolome, microbiome, etc. lead to preterm birth and its subsequent comorbidities. If the research question is focused on how a gestational exposure leads to developmental delays observed in childhood, omics at birth and in childhood could be screened to determine whether the initial exposure led to molecular alterations that then persisted.

While few studies focused on the impact of environmental exposures have published multi-omics analyses to date [154, 162, 164], data analysis approaches for handling these data are already quite advanced and growing [150]. Toxicologists, statisticians, and bioinformaticists can work together to apply or adapt existing methods to address research gaps. Many multi-omics analytical methods are developed with disease prediction as the ultimate goal. In studies focused on prenatal EDCs, the primary goal is often different. Methods that identify molecular signatures or common biological pathways across omics layers that are altered by exposures will be important. Studies of EDCs may also make use of methods that test mediation of the relationship between exposure and a child health outcome by clusters of omics features.

In addition to selecting appropriate methods for multi-omics analysis based on the goals of the study, human studies on this topic must acknowledge limitations and challenges of epidemiology and mitigate these whenever possible. Studies focused on the impact of gestational exposures are often observational, and as such are subject to various biases and confounding factors that influence effect estimates and impede causal inference. Careful study design can reduce some types of bias (i.e. sampling or selection bias) and appropriate statistical methods can account for others (i.e. confounding bias) [165–167]. Challenges of exposure assessment include obtaining the appropriate sample type and time point(s) to estimate either acute or chronic exposure, short half-lives for some chemicals (i.e. phthalates) leading to variable levels from day to day, and the impact of organ function on elimination of chemicals from the body. Statistical or study design solutions to reducing exposure misclassification should be considered [168, 169]. Challenges inherent to environmental epidemiology may be augmented when extending traditional analyses to high dimensional omics outcomes. The field could benefit from adaptation of statistical methods that reduce major biases to multi-omics approaches.

Overall, application of multi-omics to toxicology and epidemiology studies focused on early life EDC exposures will accelerate our understanding of toxicity and improve risk assessment, prevention, and intervention strategies to protect health. Advancements in the omics field at large – including standardization of laboratory and data processing protocols and development of less computationally intensive data integration approaches - will also benefit our field and increase feasibility of incorporating these measures into environmental studies.