Multiple receptors shape the estrogen response pathway and are critical considerations for the future of in vitro-based risk assessment efforts

Abstract

Current in life toxicity testing paradigms are being challenged as the future of risk assessment moves towards more comprehensive mode of action/adverse outcome pathway based approaches. In particular, endocrine disruption screening is now a global activity and key initiatives in the United States focus on the use of high throughput in vitro assays to prioritize compounds for further testing of estrogen, androgen or thyroid disruption. Of these pathways, much of the emphasis to date has been on high-throughput methods for estrogenic activity primarily using ligand binding and trans-activation assays. However, as the knowledge regarding estrogen receptor signaling pathways continues to evolve, it is clear that the assumption of a simple one-receptor pathway underlying current in vitro screening assays is out of date. To develop more accurate models for estrogen-initiated pathways useful for quantitative safety assessments, we must design assays that account for the key signaling processes driving cellular dose response based on up-to-date understanding of the biological network. In this review, we summarize the state of the science for the estrogen receptor signaling network, particularly with regard to proliferative effects, and highlight gaps in current high throughput approaches. From the sum of this literature, we propose a model for the estrogen-signaling pathway that should serve as a starting point for development of new in vitro methods fit for the purpose of predicting dose response for estrogenic chemicals in the human.

Keywords:

• TT21C
• endocrine disruption
• human health safety assessment
• alternative models
• in vitro models
• estrogen receptor

View correction statement:

Erratum

Introduction

Classical paradigms for risk assessment using in life toxicity testing are undergoing changes as more organizations worldwide push for a move from in vivo animal testing to a greater reliance on in vitro methods. In 2007, the National Research Council published a report, “Toxicity Testing in the 21st Century (TT21C): A Vision and a Strategy” (NRC 2007). This report identified the need for advances in toxicity testing to incorporate modern methods in biology and integrate advanced computational approaches. These changes promise to reduce or even eliminate animal based in vivo assays, replacing them with human cell-based in vitro methods, and thereby improving the relevance of testing for predicting human responses and streamlining toxicity testing. Initiatives to move to in vitro only methods will require a systematic approach to categorizing chemicals for more rapid testing but the manner in which these new directions will incorporate exposure estimates, use categories, or identification of specific adverse event pathways is still in discussion. It is clear, however, that assessing human health risks associated with chemical exposures using mode of action (MOA) or adverse outcome pathway (AOP) approaches will require broad suites of pathway-specific in vitro assays. Of the known chemical toxicity pathways, disruption of the endocrine pathway represents a major area of concern for human safety assessment. The estrogen, androgen, and thyroid systems are among the most studied hormone pathways in toxicity testing. Of these, the most comprehensive testing strategies currently in use relate to the estrogen signaling pathways.

Estrogen signaling represents a key component of regulation in mammalian endocrine systems, with important physiologic roles in both sexes. The estrogen signaling pathway is initiated by various endogenous estrogens including estriol, estrone, and the primary physiologic estrogen, 17β-estradiol (E2). In addition to the endogenous estrogens, naturally occurring phytoestrogens from vegetables or berries, and synthetic compounds with estrogen activity found in manufactured materials or the environment can initiate signaling (Tapiero et al. 2002; Dixon 2004). These environmental estrogens have been linked to adverse health effects including breast cancer, uterine cancer, and alteration in sexual development (Safe et al. 2001). In the past few decades, heightened public awareness of the potential dangers of synthetic estrogen compounds, such as Bisphenol A (BPA) and octylphenol, has prompted scientific research into the toxicologic mechanisms of environmental estrogens and reevaluation of the methods of regulation for potential estrogen active compounds (Soto et al. 1991; Krishnan et al. 1993). Traditional testing strategies are slow, have significant costs, and require the use of large numbers of animals. Many of these study designs span multiple life stages and evaluate effects across more than one generation. These approaches do not provide a feasible path forward for testing the large number of environmentally relevant chemicals for endocrine disruption.

In an effort to increase the throughput for estrogen screening, the USEPA recently recommended a suite of sixteen in vitro assays to prioritize compounds for more rigorous evaluation of estrogenic activity (Table 1; Browne et al. 2015). These assays combine in vitro tests from the endocrine disruptor screening program (EDSP), and the Toxicity Forcaster (ToxCast) and Tox21 programs. The EDSP program was established to screen environmental chemicals as part of a tiered system of tests. The first tier originally consisted of a combination of in vitro and short-term in vivo tests to prioritize chemicals for further, more involved in vivo testing under Tier 2. After the first battery of chemicals was tested, it was clear that even these short-term animal tests were too labor, time, and animal use intensive to address the large numbers of environmental compounds needing testing (Judson et al. 2009). As an alternative, the USEPA and NICEATM (NIH; NTP Interagency Center for the Evaluation of Alternative Toxicological Methods) evaluated the feasibility of using a suite of assays from the ToxCast and Tox21 efforts (Table 1) as a possible replacement for the EDSP Tier 1 screening (Browne et al. 2015).

Table 1. Current in vitro tests used by the EPA (EDSP/ToxCast) and NIEHS (Tox21) for estrogen screening.

Assay description	Cell type	Tissue	Measurement	EDSP	ToxCast	Tox21
Steroidogenesis	H295R	Adrenal	Steroidogenesis	X	–	–
Aromatase inhibition, viability	H295R	Adrenal	Steroidogenesis	–	X	X
Aromatase inhibition, recombinant	H295R	Adrenal	Steroidogenesis	X	X	X
Rat uterine cell lysate, E2 displacement	n/a	n/a	Ligand binding	X	–	–
Human uterine cell lysate, E2 displacement	n/a	n/a	Ligand binding	–	X	–
Murine uterine cell lysate, E2 displacement	n/a	n/a	Ligand binding	–	X	–
Bovine uterine cell lysate, E2 displacement	n/a	n/a	Ligand binding	–	X	–
ERα:α dimerization (Odyssey Thera)	HEK293T	Kidney	Dimerization	–	X	–
ERα:β dimerization (Odyssey Thera)	HEK293T	Kidney	Dimerization	–	X	–
ERβ:β dimerization (Odyssey Thera)	HEK293T	Kidney	Dimerization	–	X	–
ERα, reporter gene luciferase, agonist	MCF7	Ovarian	ERE activity	–	X	X
ERα reporter gene luciferase, antagonist	MCF7	Ovarian	ERE activity	–	X	X
ERα reporter gene luciferase, viability	MCF7	Ovarian	n/a	–	X	X
ERα LBD with ERE β-lactamase reporter, agonist	HEK293T	Kidney	ERE binding	–	X	X
ERα LBD with ERE β-lactamase reporter, antagonist	HEK293T	Kidney	ERE binding	–	X	X
ERα LBD with ERE β-lactamase reporter, viability	HEK293T	Kidney	n/a	–	X	X
ERα transfected, ERE-GFP reporter (Odyssey Thera)	HeLa	Cervix	ERE activity	–	X	–
ERα transfected, ERE-luciferase reporter (Odyssey Thera)	HeLa	Cervix	ERE activity	–	X	–
ERα transfected, ERE-luciferase reporter	HeLa	Cervix	ERE activity	X		–
ERE cis (Attagene)	HepG2	Liver	ERE activity	–	X	–
ERα trans (Attagene)	HepG2	Liver	ERE binding	–	X	–
Cell growth using real-time measurements (ACEA)	T47D	Breast	proliferation	–	X	–

These in vitro assays were originally sourced from estrogen tests already in use primarily in the pharmaceutical community and, as such, contain broadly overlapping endpoints. They fall into five categories that focus on one of the following endpoints: hormone production in adrenal tissues (steroidogenesis), chemical/ligand association with estrogen receptors (ligand binding), formation of Estrogen Receptor (ER) dimers (dimerization), receptor association with estrogen response elements (ERE binding or ERE activity), and proliferation in a breast cancer cell line (Table 1). These assays have a high sensitivity and selectivity for classifying compounds as estrogenic or not estrogenic when compared with rodent uterotrophic assays, thus supporting their use for prioritization of compounds (Browne et al. 2015). There are notable differences between rodent and human systems with regard to chemical safety such as the classic example of PPAR-mediated carcinogenicity (Mukherjee et al. 1994). As such, the ability of these tests to predict animal responses is an area of uncertain relevance to human safety decisions. As we move toward animal-free safety assessments, this issue becomes particularly important and is further complicated by the question of whether these assays are useful for quantitative risk assessment, e.g. setting a point of departure (PoD) for endocrine active compounds in humans. Successfully predicting concentrations at which compounds could be expected to induce estrogenic activity in the intact human requires an assay strategy that incorporates enough of the biology of the estrogen pathway to recapitulate chemical dose–response behaviors.

The ToxCast/Tox21 assays were originally designed for high throughput and ease of use, with the goal of obtaining qualitative (hit/no hit) information. As such, the majority of the assays were developed in cell-free systems or systems based in cell types with questionable relevance for assessing in-life effects. The only assay developed directly in an estrogen responsive cell, with no manipulation of the expression of receptors within that cell line, is the ACEA proliferation assay in the breast cancer cell line (T47D) – a cell that cannot be expected to predict responses in other tissues sensitive to estrogens, such as the uterus. Therefore, while these assays have been highly successful as an in vitro alternative for identifying perturbations in estrogen signaling, there is a still a need for assays which generate quantitative information about biologic dose responses. In a tiered testing strategy, these ToxCast/Tox21 assays could be used as an early suite to identify potential hits and prioritize chemical testing in more, biologically relevant in vitro systems designed and validated for estimating chemical PoDs in support of quantitative safety assessments. Incorporating such experimental systems designed to provide robust, in vivo relevant dose–response information is critical as we move forward toward an in vitro-only risk assessment paradigm.

Cell-free systems do not account for the interplay between the multiple estrogen receptors and kinases acting as feedback and feed forward control components known to drive behavior in other cell signaling systems (Zhang et al. 2014). These components are critical for modulating cellular response to transient changes in hormone levels. One such cellular response, proliferation, is considered a key adverse outcome of estrogen signaling dysregulation, with potential implications for uterine and breast cancer. The dose–response for estrogen is notably tissue specific, as demonstrated in the clinical use of tamoxifen, likely due to differences in the expression of estrogen receptor isoforms and cofactors (Zhang et al. 2005). Thus, while substantial information has been generated from high-throughput screening efforts and there is clear utility in using such screens for chemical prioritization, these tests oversimplify the estrogen signaling process and are likely to provide an incomplete assessment of adverse effect levels in humans, a key requirement for replacing in vivo methods intended to identify chemical PoDs.

To move towards use of in vitro assay results in more quantitative safety assessments, we need to characterize the key processes in estrogen receptor signaling that lead to downstream adverse effects. Understanding of these key events will then enable development of fit-for-purpose assays to assist predictions of the dose response for estrogen dysregulation. This paper reviews the current knowledge of cellular estrogen signaling processes and incorporates this information into a data-driven framework for the estrogen signaling network, with a focus on tissue-specific proliferative responses as in vitro markers of adversity. We first define the current understanding of the estrogen receptors that modulate estrogen-mediated proliferative signaling in human cells through an extensive literature review. We then bring together data and define the interactions between these unique receptors and their putative functions, before putting all of this within the context of cell or tissue type. Finally, we summarize the findings as a simplified schematic, providing suggestions for incorporating this information into an in vitro safety assessment strategy. Based on this review, we identify a need for tissue-specific models of estrogen signaling. In particular, we suggest a model for uterine signaling that should be distinct from existing breast tissue systems. The development of fit-for-purpose assays that incorporate tissue specificity to address specific toxicity pathway responses has advantages beyond shoring up confidence in screening applications. By ensuring that in vitro platforms recapitulate key biology and cellular dose–response characteristics, we can move forward with more confidence in developing quantitative in vitro-based safety assessments.

Materials and methods

An in-depth search and review of articles related to estrogen receptors and estrogen receptor signaling was performed using PubMed, Scopus, JSTOR, Academic Search Complete (EBSCO), and Google Scholar databases. The last search was performed 15 February 2016. A preliminary search using the terms “estrogen receptor”, “estrogen signaling”, or “in vitro estrogen” was used to identify broadly applicable papers. For estrogen signaling, search terms “ERE”, “estrogen response element”, “non-genomic”, and “genomic” were used in conjunction with “ERα”, “ERβ”, and “GPER OR GPR30”. For specific sections of the review covering the signaling associated with each receptor the following search terms were applied: “ER isoforms”, “ERα isoforms”, “ERα66 OR ER alpha 66”, “ERα46 OR ER alpha 46”, “ERα36 OR ER alpha 36”, “ERβ OR ER beta”, “GPER OR GPR30 OR G protein-coupled estrogen receptor”, and “EGFR AND estrogen”. Two people performed independent examinations of the titles and abstracts in order to select articles that were appropriate for inclusion in this review. Only those articles published in English in peer reviewed publications were considered. The articles selected were then critically reviewed and any additional papers cited within these publications were evaluated for inclusion as well. Redundant articles were omitted in preference for the earliest publication with these results. Where applicable, studies were categorized by the tissue type of the experimental system. For sections of the paper reporting activity specific to uterine tissue, non-uterine systems were not included. After curating the publication list in this manner, 118 articles were identified for review. The in vitro assays reviewed for ToxCast/Tox21 and EDSP were identified directly from the program descriptions on the EPA or NIEHS websites respectively, and supporting publications were identified from those listed on these agency sites as well.

Estrogen receptors

Uterine and breast tissues express several receptors that bind estrogens. The first receptor binding estrogen in rat uterine cells was characterized in detail by Toft and Gorski in the 1960s, building on the contemporary nuclear hormone work of Jensen et al. (Toft & Gorski 1966). Early studies described the receptor as a 66 kDa protein that localized to the nucleus and bound 17β-estradiol (E2) with a high degree of specificity (Noteboom & Gorski 1965; Toft et al. 1967). A link between estrogen receptor expression and breast cancer was quickly identified and certain tumor tissues were found to be E2-responsive. However, this relationship remained largely enigmatic (Jensen et al. 1971; Sluyser & Van Nie 1974; Block et al. 1975). It was not until 1986 that two research groups published the first data identifying the estrogen receptor (ER) coding sequence and provided analysis of the estrogen receptor (ER) structure and potential ligand binding sites (Green et al. 1986a, 1986b; MacGregor & Jordan 1998). At the time, it was believed that only one type of receptor was responsible for estrogen signaling.

Ten years later a homologous, yet novel, ER was identified that also bound E2 with high affinity (Kuiper et al. 1996). This new receptor was designated ERβ (ESR2) and the first receptor identified by Toft and Gorski became known as ERα (ESR1). Additional signaling receptors in the estrogen response pathway were observed with the discovery of a G protein-coupled receptor (G-protein estrogen receptor; GPER) that demonstrated a functional response to E2 (Filardo et al. 2000). More recent research has shown that ERα exists as one of the three isoforms distinguished as ERα66, ERα46, and ERα36 based on their respective molecular weights, adding to a total of five distinct receptors that are activated by estrogen and appear to regulate E2-mediated proliferation (Flouriot et al. 2000; Wang et al. 2005). However, the specific signaling events associated with each of these receptors, the interactions between them, and the role each plays in cellular proliferation remain unclear. This uncertainty about the pathways in estrogen signaling represents a significant knowledge gap that has consequences for using upstream signaling events, such as ligand binding and trans-activation, for more quantitative safety assessments based on proliferation. Here, we outline the current literature for each of these receptors, highlighting areas needing further evaluation and examining signaling interactions between the multiple receptor pathways that are important for controlling cellular responses to estrogenic compounds.

ERα characterization and signaling overview

While all five receptors drive cellular changes in response to estrogen, the traditional focus of estrogen signaling research has been the activation of ERα66. In the “genomic” description of ERα signaling, the binding of ligand causes a conformational change in cytosolic ERs, allowing the formation of dimers and exposure of the nuclear localization sequence (NLS). Dimer formation by two ERs following ligand binding is important for full transcriptional activation (Tamrazi et al. 2002). Activated dimers bound to ligand traffic from the cytoplasm into the nucleus, bind to estrogen response elements (ERE) at specific gene promoters, and regulate transcriptional activation through recruitment of preinitiation complexes and additional transcription factors. The ERE consensus for ER recognition has been identified as GGTCAnnnTGACC, but many EREs deviate from this sequence and approximately one-third of known ER target genes associate with ERs indirectly through adaptor or scaffold proteins (O'Lone et al. 2004). ERα interacts with coactivators such as AP1 (cFos/cJun), Sp1, and SNCG or, conversely, with co-repressors such as Sin3A to modify transcriptional responses to estrogen (Kushner et al. 2000; Jiang et al. 2003; Schultz et al. 2003; Ellison-Zelski et al. 2009; Zhao et al. 2010). Global gene expression profiling of breast tumor cell lines following E2 stimulation has identified as many as 1500 genes that are differentially regulated in response to estrogen (Katzenellenbogen et al. 2000; Klinge 2000; Harrington et al. 2003; Kininis & Kraus 2008; Pendse et al. 2017). Only a small fraction of the differentially expressed genes are directly modulated by the ERα and ERβ transcription factors (Madak-Erdogan et al. 2008). In addition to the ERs, other transcription factors including Foxm1, E2F1, E2F7, and ZNF217 are responsible for initiating expression of genes associated with cell cycle, cell growth, and differentiation (Shen et al. 2011; Pendse et al. 2017). This low degree of overlap between estrogen receptor binding and gene expression regulation is in sharp contrast to other nuclear receptors that have been shown to account for approximately half of the affected genes’ expression (van der Meer et al. 2010; McMullen et al. 2014) and highlights the contribution of the so-called “non-genomic” signaling (i.e. transcriptional activation not mediated through the transcription factor activity of the ER itself).

Mechanistic studies have confirmed the importance of these non-genomic signaling pathways. Estrogen–dendrimer conjugates that are incapable of entering the nucleus can still initiate transcription of a large number of estrogen-responsive genes, indicating that recruitment of ERα or ERβ to the ERE is not an essential step for estrogen-mediated signaling activation of transcription (Madak-Erdogan et al. 2008). Non-genomic effects of estrogen signaling occur rapidly following stimulation and are thought to involve distinct membrane-associated mechanisms (Revelli et al. 1998). Downstream activation of protein kinase A (PKA) or phospholipase C (PLC) and calcium release also occur in cells expressing ERs following estrogen stimulation (Revelli et al. 1998). There is also associated growth factor signaling such as activation of ERK1/2 and the MAPK pathway by ligand-activated ERα (Zhang et al. 2002; Martin et al. 2005). This non-genomic signaling encompasses modes of activation that have ER activity taking place outside of the nucleus and transcriptional regulation mediated by different transcription factors, such as those described above. Thus, while the common nomenclature for this signaling is “non-genomic”, there is still a significant transcriptional component to the pathway.

Because the identification of GPER, ERα46 and ERα36 is fairly recent, the majority of research on the non-genomic effects of ER activation does not distinguish between responses associated with specific receptor types or isoforms. The localization, relative expression and unique signal transduction of each estrogen receptor isoform in response to estrogens are still under investigation. Regardless, it is becoming clear that these earlier studies provide an important framework for evaluating ERα functions. Later in this review, we evaluate these observations and mechanisms of ERα activity to highlight contributions from each of the isoforms and to provide a much more detailed mechanistic view of this cellular pathway controlling proliferation.

ERα isoform gene and protein structures

The ERα gene contains multiple promoters and splice sites that produce the alternate isoforms ERα46 and ERα36. Compared with ERα66, each isoform transcript has deletions in functional domain coding regions that confer receptor activity. The specific role of these regions in ERα function has been extensively investigated using mouse models and changes in coding regions at the RNA level result in dramatic changes in downstream ER-mediated responses (Arao et al. 2011). Alternative splicing of the ERα gene appears to be responsible for formation of ERα46 transcripts that are missing the first exon of the full length ERα66 in their coding sequence as shown in Figure 1(A,B) (Flouriot et al. 2000). A favorable Kozak sequence at +752 in exon 2 serves as a translational start for ERα46, creating a 46 kDa protein that lacks the AF1 domain encoded by exon 1, but is otherwise identical to ERα66 (Figure 1(C); Flouriot et al. 2000). ERα36 originates from a novel intronic promoter within ESR1, producing an mRNA transcript that lacks exon 1 of the ERα66, skips exons 7 and 8, but acquires an additional exon at the 3′ end on the coding sequence through a unique alternative splicing event (Figure 1(B); Wang et al. 2005; Zou et al. 2009; Gu et al. 2014). ERα36 mRNA contains the same +752 translation initiation site in exon 2 as ERα46 and thus encodes a 36 kDa protein identical to ERα46 at the N terminus, but which lacks the AF1 domain. The deletion of exons 7 and 8 also removes the AF2 domain and the C terminal end of ERα36 is capped by 27 unique amino acids translated from exon 9 (Figure 1(C); Gu et al. 2014).

Figure 1. ERα gene, mRNA, and protein isoform structures. (A) The ERα gene has two potential transcriptional start sites. The ERα66 and ERα46 transcript variants arise from the promoter identified by a 5-point star. The ERα36 variant arises from a unique intronic promoter denoted by a 4-point star. The ERα36 transcript also uses an alternative splice site denoted by a 4-point star. (B) ERα66, ERα46, and ERα36 mRNA maps are aligned. (C) Corresponding protein maps for ERα66, ERα46, and ERα36 are aligned and both structural domains (A–F) and functional domains (AF-1, DBD: DNA binding domain, Hinge, LBD: ligand binding domain, AF-2) are identified.

ERα transcription involves a degree of autoregulation in vitro; ERα directly binds to its own promoters and alters transcriptional activity (deGraffenried et al. 2004). The specific promoter – and, therefore, isoform – affected and whether this interaction drives or inhibits transcription occurs in a cell-type specific manner as osteoclasts, MCF7 breast cancer cells and cervical HeLa cells all exhibit different patterns of autoregulation at the ERα66 promoters (Castles et al. 1997; Denger et al. 2001; Lambertini et al. 2003). Regulation of the shorter ERα isoforms has also been demonstrated. ERα46 transcription is stimulated in the human macrophage cell line RAW264.7 following E2 treatment, and this regulation is determined by a switch in specific promoter usage (Murphy et al. 2009; Figure 1(A)). In contrast, the human endothelial cell line EA.hy926 expresses ERα46 with serum starvation, and only begins expressing ERα66 following E2 stimulation, suggesting that signaling through the estrogen pathway positively regulates full length ERα transcription while ERα46 is transcribed at a steady state level (Li et al. 2003). The ERα36 promoter sequence in intron 1 contains binding sites consistent with other ERα promoters including AP1, NFκB, and a partial ERE site that may play a role in autoregulation (Zou et al. 2009). A reporter construct with the ERα36 promoter driving luciferase expression in HEK293 cells, which lack endogenous ERs, has minimal transcriptional response to estrogen treatment. However, the expression of full length ERα66 is sufficient to drive transcription from the ERα36 promoter in response to treatment. Interestingly, co-expression of ERα36 or ERα46 with ERα66 interfered with transcriptional activity from the luciferase reporter construct, suggesting a negative feedback or antagonistic role for these truncated forms of ERα. Together, these data form an incomplete and conflicting picture of the regulatory feedback loops in regards to regulation of transcription by ERα isoforms. Although it is clear that there are differences in the levels of transcription for each isoform according to cell type, a robust understanding of these auto regulatory mechanisms is still missing (Zhao et al. 2009; Zheng et al. 2010; Wallacides et al. 2012).

ERα isoform localization

Full length ERα protein primarily resides in the cytoplasm and upon activation, localizes to the nucleus to initiate genomic estrogen signaling. However, for non-genomic signaling, plasma membrane associated ERs play a critical role. Removal of the nuclear localization sequence in ERα or treatment with a membrane insoluble E2BSA conjugate does not abolish estrogen-mediated responses, indicating that direct ERE-targeted genomic regulation cannot account for all of the estrogen-mediated signaling events (Zhang et al. 2002; Chen et al. 2004). Membrane-localized ERs may initiate rapid non-genomic signals, while traditional genomic signaling by cytoplasmic and nuclear compartmentalized ERs occurs by a distinct, slower pathway (Kelly & Levin 2001; Levin 2005). Structural differences between ER isoforms likely cause functional differences in the ability of each isoform to localize to the plasma membrane or to engage in specific interactions at the cell surface that are necessary for induction of non-genomic signaling. Tissue-specific expression or preferential induction of isoforms in response to exposures may, therefore, skew the response towards either genomic or non-genomic associated cellular events. For this reason, both expression levels and localization of ERα isoforms are of great importance in characterizing the responses of cells to estrogen. ERα interacts with caveolin and can localize to the cytoplasmic side of the plasma membrane in these intracellular membrane pockets (isoform not specified; Kim et al. 1999; Chambliss et al. 2000; Razandi et al. 2002). However, the ERα localized within caveolae represents a small fraction of total membrane associated ERα (Li et al. 2003).

All three ERα isoforms (ERα66, ERα46, and ERα36) associate with the cell surface (Razandi et al. 1999; Acconcia et al. 2004). Specific targeting of myc-tagged ERα46 to the plasma membrane appears to require cysteine palmitoylation (Li et al. 2003; Acconcia et al. 2004; Kim et al. 2011). A subset of ERα66 is palmitoylated at this same site (Cys447), which results in targeting of the protein to the plasma membrane, producing a small population of membrane associated ERα66 (Acconcia et al. 2004; Levin 2005). Nonetheless, the majority of ERα66 and ERα46 are still found in the nuclear and cytoplasmic fraction, indicating that surface localized ERα66 represents a minor population of the protein (Chambliss et al. 2000; Kelly & Levin 2001; Levin 2005). Early studies of ERα46 consistently detected an “unknown” band at 36 kDa in the membrane-associated protein fraction from an endothelial cell line (Li et al. 2003). The 36 kDa isoform of ERα had not been reported at the time. Based on the current knowledge, it is likely that this membrane associated protein was ERα36. Since then, antibodies directed at the unique 27 amino acid tail of ERα36 in cells expressing only the ERα36 isoform have shown that the majority of ERα36 localizes to the plasma membrane (Wang et al. 2006). Endogenous ERα36 has also been identified at the plasma membrane of both breast cancer and endometrial cancer cells (Lee et al. 2008; Lin et al. 2009). Across cell types, ERα36 is found to varying degrees in each cellular compartment but with a preferential localization to the plasma membrane. From these data, a picture of cellular trafficking begins to emerge for each of the three ERα isoforms, with ERα36 mainly residing in the plasma membrane but incidentally found in the cytoplasm and nucleus; ERα66 mainly in the cytoplasm and nucleus but with some surface localization; and ERα46 distributed in a similar manner to ERα66. The evidence for surface localization of the three isoforms is listed in Table 2.

Table 2. ERα isoform localization literature summary.

Isoform	Cell type	Method	Finding	Ref.
ERα66	MCF-7	Mass Spec	Nuclear, some membrane associated	Pedram et al. (2006)
ERα66	ERα66 transfected HeLa	–	Nuclear, some membrane associated	Acconcia et al. (2004)
ERα66/46	Transfected MCF-7	Immunofluorescence	Both are predominantly nuclear	Metivier et al. (2004)
ERα46	Endothelial	Immunofluorescence	Nuclear, some membrane associated	Li et al. (2003)
ERα46	Endothelial	Fluorescence Microscopy	Nuclear, some membrane associated	Kim et al. (2011)
ERα36	ERα36 transfected HEK293	Cell fractionation, Western	Cell membrane	Wang et al. (2006)
ERα36	Ishikawa	Immunofluorescence	Cell membrane	Tong et al. (2010)
ERα36	TCam-2	Immunofluorescence	Nuclear and membrane associated	Wallacides et al. (2012)

ERα isoform signaling

The classical model for estrogen signaling entails estrogen binding to cytoplasmic ERα, receptor dimer formation, nuclear translocation, and binding to the ERE to promote transcription of estrogen responsive genes. This first-order description of estrogen signaling may be useful in outlining the basics of the estrogen response, but the reality of estrogen receptor action involves additional pathways and receptor activities, as evidenced by the existence of several isoforms with different cellular locations and transcriptional activities. Location of the receptors within the cell has practical consequences for the cellular signaling network. On one hand, genomic signaling is associated with classical understanding of estrogen signaling, wherein the cytoplasmic ER translocates to the nucleus and initiates transcription. Non-genomic signaling, on the other hand, is associated with membrane-bound ERs and initiation of these pathways is mediated through proteins such as Src or Ras that also reside at the plasma membrane (Edwards & Boonyaratanakornkit 2003). Although activated ER dimers in the cytoplasm have similar structural properties to those monomers that are anchored to the membrane, the initiation factors for non-genomic signaling are not accessible to dimers in the nucleus. Therefore, at any one point in time, activated dimers are either at the surface engaged in non-genomic signaling or in the nucleus engaged in genomic signaling. For this reason, we have separated genomic and non-genomic signaling actions into two separate sections for the following overview of ERα signaling mechanisms.

Genomic signaling of ERα

The structural differences among ERα66, ERα46, and ERα36 become important considerations in relation to ER signaling interactions at estrogen responsive gene promoters. The three isoforms all contain identical DNA binding domains for ERE recognition, nuclear localization signals, and dimer interface regions (Figure 1). However, differences in transcriptional activity between each isoform at ERE-driven promoters have been extensively documented (Flouriot et al. 2000; Li et al. 2003; Metivier et al. 2004; Penot et al. 2005). ERα46 is much less efficient at inducing genomic estrogen responses, conferring significantly lower values for ERE activity than ERα66, as noted with E2 treatment when these receptors are overexpressed in ER negative cells with an ERE-luciferase reporter (Flouriot et al. 2000; Li et al. 2003; Penot et al. 2005). Importantly, since ERα46 does not have a functional AF1 domain, transcriptional events that are initiated through the AF1 domain of full length ERα are not mediated by ERα46 signaling (Penot et al. 2005). With the missing AF1 domain, ERα46 interferes with ERα66 activity by binding to, and potentially displacing ERα66 from EREs leading to decreased genomic signaling. The AF1 domain deletion also recruits different cofactors to ERα46 complexes compared to ERα66. ERα46 preferentially recruits factors such as Sin3 resulting in complexes that further dampen genomic responses (Metivier et al. 2004). Thus, it appears that ERα46 has a repressive role in ERE-mediated transcription, either through interference with ERα66 binding or through recruitment of transcriptional repressors. ERα36 does not induce any observable transcription from an ERE-luciferase reporter construct following E2 treatment (10 nM) and co-expression of ERα36 with ERα66 prevents the transcriptional activity of ERα66 from the ERE reporter (Wang et al. 2006). Given that ERα36 lacks AF1 and AF2 domains, which are important for cofactor recruitment to the DNA, the activated ERα36 homodimer may recognize EREs in the nucleus, but be unable to initiate signaling once bound, thereby acting as an inhibitor of genomic estrogen signaling. Taken together, these experiments indicate that estrogen-activated ERα66 dimers result in maximal ERE-driven transcriptional responses, ERα46 dimers result in minimal to moderate responses, which are largely restricted to AF2 permissive genes, and ERα36 dimers do not directly drive any transcriptional activity. These results are summarized in Table 3.

Table 3. ERα isoforms genomic signaling literature summary.

Isoform	Cell type	Method	Finding	Ref.
ERα46	Transfected Cos-7	ERE-luciferase reporter	Has half the activity of ERα66	Li et al. (2003)
ERα46	Transfected MCF-7	ERE-luciferase reporter	Has less activity than ERα66	Flouriot et al. (2000)
ERα46	Transfected HeLa	ERE-luciferase reporter	Has less activity than ERα66	Penot et al. (2005)
ERα46	Transfected MB231	Co-IP, ChIP	Repressive cofactors recruited to ERE	Metivier et al. (2004)
ERα36	Transfected HEK293T	ERE-luciferase reporter	Interferes with ERα66 transcriptional activity	Wang et al. (2006)

Non-genomic signaling of ERα

Non-genomic estrogen signaling in cells occurs very rapidly following estrogen exposure and can be observed in vitro following treatment with cell impermeable compounds such as BSA-conjugated E2. These rapid responses include calcium mobilization, kinase activation, and pro-survival signaling (Revelli et al. 1998). Non-genomic signaling is also associated with slower downstream cellular responses, including alterations in cell cycle and proliferation (Kushner et al. 2000). Activated membrane-bound ERα receptors engage specific factors such as Src, Ras, and Raf, which are also found at the surface of the cell (Edwards & Boonyaratanakornkit 2003). Thus, cellular location of the ERα isoforms becomes a critical aspect in initiation of non-genomic events. All three ERαs are found at the surface of the plasma membrane, but only ERα36 preferentially localizes to this area. The contribution of endogenous receptor isoforms to non-genomic signaling is difficult to define as the relative amount of each isoform varies depending on the cell type. However, overexpression or knockdown systems have provided a basis for understanding distinct roles of the isoforms. The following discussion summarizes the work to date on the role of the individual isoforms in non-genomic induction of proliferation.

Estrogen induction of non-genomic responses performed in intact cells does not distinguish between ERα isoforms. In the experiments summarized above, researchers focused on the presence of ERα66 and did not provide sufficient information to determine the presence of ERα46 or ERα36 in these systems. However, using ERα-deficient cells in combination with specific isoform overexpression provides a method for separating out the unique contributions of the isoforms to these pathways. E2BSA treatment in Ishikawa endometrial cells expressing ERα66 resulted in activation of ERK phosphorylation pathways and cellular proliferation (Wang et al. 2006). Treatment with siRNA directed to ERα66 did not block these responses, which points to a role for non-genomic signaling in proliferation (Wang et al. 2006).

The majority of work on proliferative responses associated with ERα46driven signaling has been performed in breast cancer cells. Induction of ERα46 from an inducible promoter construct leads to a rapid decrease in the fraction of cells entering S phase and an overall decrease in proliferation after E2 treatment (10 nM) when compared to cells expressing ERα66 only (Penot et al. 2005). Similarly, overexpression of ERα46 in breast cancer cells inhibited ligand-activated proliferative responses (Klinge et al. 2010). The consistency of these results showing reduced transcriptional activity argue for a greater role of ERα46 in genomic signaling than non-genomic signaling, but data limitations impair our ability to make a definitive distinction. Inclusion of BSA-conjugated ligands and measures of kinase pathway activation in future proliferation assays could help differentiate the non-genomic and genomic aspects of ERα46 signaling.

ERα36 overexpressed in ER-negative HEK293T cells drives a proliferative response following E2 (10 nM) stimulation through a MAPK/ERK-mediated mechanism (Wang et al. 2006). ERα36 shRNA knockdown studies show that ERα36 association with Src/Shc leads to phosphorylation of ERK, increased c-myc and cyclin D1 expression, and cellular proliferation (Wang et al. 2005, 2006). Further, E2BSA treatment in cells expressing both ERα36 and ERα66 results in phosphorylation of PKC delta and ERK1/2, upregulation of cyclin D1 (CCND1) and CDK4, and cellular proliferation that is blocked by ERα36 knockdown, but not ERα66 knockdown (Tong et al. 2010). Based on these results, ERα36 appears to play a significant role in non-genomic responses to estrogen, while ERα46 and ERα66 are more likely involved with regulation of genomic signaling events. These findings are summarized in Table 4.

Table 4. ERα isoforms nongenomic signaling literature summary.

Isoform	Cell Type	Method	Finding	Ref.
ERα46	Transfected MCF-7	Flow cytometry	Prevents estrogen-mediated cell-cycle progression	Penot et al. (2005)
ERα46	Transfected MCF-7	BrdU ELISA	Prevents estrogen-mediated cell-cycle progression	Klinge et al. (2010)
ERα36	Transfected HEK293T	Western, MTT	Drives estrogen-mediated proliferation through MAPK/ERK	Wang et al. (2006)
ERα36	TCam-2	Viability assay	Knockdown prevents estrogen-mediated proliferation	Wallacides et al. (2012)
ERα36	Ishikawa	Western, MTT	Mediates proliferation through PKCδ and ERK1/2 signaling	Tong et al. (2010)

ERβ, GPER, and EGFR in estrogen signaling

Estrogen signaling has been shown to involve activation of many receptors that can directly feed into the ER-mediated signaling cascade, cross-react with intermediate steps in the pathway, or may become activated due to transcriptional cross-talk downstream of estrogen signaling. Among these are estrogen receptor β (ER β), epidermal growth factor receptor (EGFR), G-protein estrogen receptor (GPER), aryl hydrocarbon receptor (AhR), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and insulin-like growth factor receptor (IGF). Each of these receptors could play a role in adverse responses to estrogen within the context of the biological system. ERβ and GPER have been shown to directly bind E2 and initiate signaling events. Additionally, ligand-independent signaling of EGFR has been documented following estrogen stimulation and is attributed to interactions with one or more of the ERs. The importance of these three receptors in the direct signaling pathway initiated by the ERs warrants a more detailed review. Here, we will briefly outline the overlap in signaling between each of these receptors and how this relates to induction of cellular proliferation following estrogen exposure.

ERß

The human ERβ gene is located on chromosome 14, consists of eight exons, and is translated into a protein that is only 47% homologous to ERα (Enmark et al. 1997; Zhao et al. 2010). ERβ mRNA has been isolated from multiple cell types including uterine, mammary gland, and breast tumor cells and is often co-expressed with ERα at varying ratios (Enmark et al. 1997). ERβ overexpressed in ER negative cells partially localizes to the cell membrane (Pedram et al. 2006). There are also multiple splice variants of ERβ with various N-terminal extensions or in-frame insertions, but in contrast to the ERα variants, the major domains are conserved in all of the ERβ isoforms suggesting that the activities of the isoforms are likely similar (Leygue et al. 1998; Petersen et al. 1998). Additional ERβ isoforms with unique C termini have been isolated including ERβcx, a form with an altered ligand binding pocket that does not appear to bind to either E2 or to the ERE (Moore et al. 1998; Ogawa et al. 1998; Leung et al. 2006). Gene expression analyses have identified similar ERα and ERβ targets in the genome as both receptors recognize EREs, but co-expression studies have provided evidence that ERα demonstrates preferential binding that can displace ERβ at these locations (Chang et al. 2006; Lin et al. 2007; Charn et al. 2010). ERα and ERβ also have unique promoter binding sites as shown by chromatin immunoprecipitation (ChIP) analysis, where a subset of genes was actively bound only by ERβ (Grober et al. 2011). Like ERα, ERβ can indirectly bind AP1, Sp1, and CRE sites, but it inhibits rather than promotes transcription of AP1 targets (Paech et al. 1997; Sabbah et al. 1999; Kushner et al. 2000; Saville et al. 2000; Liu et al. 2002). ERβ acts primarily through the AF2 domain and has negligible AF1 activity, which may result in alternative cofactor recruitment and differential regulation of the same promoters (Cowley & Parker 1999). The cell type-specific expression of cofactors and physical differences in the AF1 domains of ERα and ERβ themselves likely confound these results. Nonetheless, ERα and ERβ clearly produce different biologic responses to estrogenic ligands (Paech et al. 1997; Barkhem et al. 1998; Pike et al. 1999).

Furthermore, ERβ interferes with ERα-mediated cellular proliferation and ERβ may have a negative regulatory effect when forming heterodimers with ERα (Omoto et al. 2003; Chang et al. 2006; Lin et al. 2007; Zhao et al. 2009). All ERα and ERβ isotypes investigated participate in homo-or heterodimer formation, but the functional activity of each dimer differs and there appears to be a strong preference for homodimer formation. ERβ appears to have a negative regulatory role in non-genomic estrogen signaling, either by sequestering estrogen from ERα or through a more active role in heterodimer formation. There is also a distinct role for ERβ in genomic signaling as compared with ERα, with ERβ repressing transcription of genes related to proliferation, supporting a role for ERβ in antagonizing ERα-mediated signals (Zhao et al. 2009).

GPER and EGFR

GPER is an orphan receptor in the G protein-coupled receptor (GPCR) superfamily with the characteristic 7 membrane-spanning loop structure. Early GPER studies demonstrated a positive correlation between GPER and ER expression in breast cancer tissues, indicating that this protein has a role in hormone-responsive cells (Carmeci et al. 1997). The SKBr3 and MDAMB231 breast cancer cell lines express GPER and respond to estrogen by activation of ERK1/2 – a response that is lost following treatment with GPER siRNA (Filardo et al. 2000; Thomas et al. 2005; Ford et al. 2011). Although estrogen directly binds GPER, GPER does not initiate ERK/PI3K/adenylyl cyclase activity or changes in DNA synthesis in response to E2 in the absence of ERα (Filardo et al. 2002; Kanda & Watanabe 2003; Revankar et al. 2005; Thomas et al. 2005; Pedram et al. 2006). Competitive binding studies have shown specific binding of estrogenic compounds to GPER that activates adenylyl cyclase signaling pathways and increases cAMP in the presence of ERα (Thomas et al. 2005). These data demonstrate that while GPER plays a role in estrogen-initiated non-genomic pathways, it does not function independently of ERα.

The development of a GPER-specific agonist, G1, enabled a fresh look at GPER signaling in cells also expressing ERα (Bologa et al. 2006). Specific activation of GPER alone in ovarian tumor cells upregulated some expected estrogen responsive genes (including c fos, cyclin A, cyclin D, and cyclin E), but not others (most notably, progesterone receptor) (Albanito et al. 2007). Further, inhibition of GPER signaling abolishes E2 induced ERK signaling, highlighting the crosstalk between other ERs and GPER. Expression of GPER is sufficient for ERα36 induction in a human breast carcinoma derived cell line. The ability of MEK/Src inhibitors to block this GPER activity indicates that the increased ERα36 expression is a downstream effect of GPER activation (Kang et al. 2003, 2010).

GPER also stimulates EGFR in the presence of estrogen, by releasing HBEGF and MMP from the plasma membrane and inducing ERK1/2 signaling and subsequent genomic responses. This interaction between GPER and EGFR takes place in the absence of classical ERs (Maggiolini et al. 2004). EGFR is also activated by membrane-associated ERα36. There is evidence in breast cancer cells that ERα36 may engage in a positive feedback loop with EGFR signaling, resulting in increased transcription of ERα36 (Zhang et al. 2011). However, it is unclear whether this autoregulation is specifically activated during oncogenic signaling or if it is present in non-cancerous cells as well. EGFR activation results in activation of NFkB through phospholipase D, calcium mobilization from the endoplasmic reticulum via phospholipase C activity, and induction of the MAP kinase cascade through Ras/Raf. As all of these pathways are part of critical non-genomic effects of estrogen, it is apparent that EGFR is a consistent component of estrogen-initiated signaling. The full mechanism of the GPER, EGFR, and ERα36 signaling network has yet to be unraveled, but they appear to work together to promote non-genomic proliferative responses to estrogen.

Developing a network structure for ER-mediated proliferative response

While estrogen signaling has a role in many physiological systems, including cardiac, bone, vasculature, immune, endocrine, and neurological systems, the focus of this review was the role of estrogen receptors in proliferative response, particularly with respect to the uterus and mammary tissue (Vaananen & Harkonen 1996; Babiker et al. 2002; Nalbandian & Kovats 2005; Mermelstein & Micevych 2008; Miller & Duckles 2008). We have, therefore, compiled the current body of information from overexpression, gene silencing, breast cancer models, and other in vitro systems into one comprehensive schematic of signaling interactions for the receptors discussed in this paper (Figure 2). This schematic is a general representation of estrogen signaling in estrogen responsive human tissues. It does not differentiate between mammary and uterine tissue, and includes only information specific to humans. The overwhelming majority of information included in this figure is derived from cells of breast tissue origin which highlights the need for a deeper understanding of the signaling mechanisms within the context of the uterus.

Figure 2. Schematic representation of estrogen signaling literature review. Pathways described in the published body of information on estrogen initiated signaling have been summarized here with arrows to denote signaling interactions leading to cellular responses. Each ER contributes to specific aspects of the overall signaling network in response to estrogen. Those interactions at the cell surface lead to activation of pathways associated with survival, metabolism, protein synthesis, proliferation, and changes in cell morphology or migration. These signaling cascades overlap with those of EGFR as shown. Genomic signaling in the nucleus consists of direct association of activated ER dimers with specific gene promoters through recognition of EREs or recruitment of additional transcription factors as scaffolding. Genomic homo- and hetero-dimers can for from each receptor combination. A gray box (?) is used to denote any of the three isoforms. [References: ¹Pedram et al. (2006); ²Acconcia et al. (2004); ²Metivier et al. (2004); ³Wang et al. (2006); ⁴Zhang et al. (2011); ⁵Edwards & Boonyaratanakornkit (2003); ⁶Wang et al. (2005); ⁷Sanchez et al. (2010); ⁷Yun et al. (2012); ⁸Thomas et al. (2005); ⁸Revankar et al. (2005); ⁹Wang et al. (2006); ¹⁰Hernandez-Sotomayor & Carpenter (1992); ¹¹Campbell et al. (2001); ¹²Wang et al. (2006); ¹³Tong et al. (2010); ¹⁴Hernandez-Sotomayor & Carpenter (1992); ¹⁴Villalobo et al. (2000); ¹⁵Habib et al. (2001); ¹⁶Wang et al. (2006); ¹⁷Maggiolini et al. (2004); ¹⁸Filardo et al. (2000); ¹⁹Li et al. (2003); ¹⁹Penot et al. (2005); ²⁰Wang et al. (2006); ²¹Cowley & Parker (1999)].

In using this information to draw conclusions about the regulatory structure of ER signaling, we have to recognize that many of the systems described above are artificially manipulated; ER isoforms were overexpressed or repressed and estrogen concentrations were typically far beyond expected physiologic levels. Physiologic exposure levels of uterine cells to endogenous estrogens approaches a maximum at approximately 1 nM range, according to the Mayo Clinic’s established reference levels for circulating estradiol: 551200 pM in premenopausal women and less than 36 pM in postmenopausal women. The in vitro studies described in this review typically used treatments in the 1100 nM range and the use of excessively high concentrations of chemicals may alter physiologic response. The various ERs will have different binding affinities for ligands and saturating the system with such high doses could obscure contributions of the individual receptors to overall cellular response. In addition, comparisons across the published studies can be difficult. The studies use different cell types that have very different estrogen signaling profiles in vitro and in vivo (Diel 2002). Nonetheless, the large body of literature investigating the signaling network for estrogen signaling provides important insights. We leveraged the knowledge gained through decades of study by using a weight-of-evidence approach to identify key processes in a generic estrogen-mediated proliferative signaling pathway appropriate for representing estrogen-mediated effects in uterine tissues (Figure 3(A)). These signaling processes are supported by data derived from human uterine cell studies, but further research is still necessary to confirm the accuracy of these interactions.

Figure 3. Coordinated receptor model for estrogen-initiated signaling in the human endometrium. (A) The interactions of the five ERs and the EGF receptor in the uterus (Figure 2) are displayed as a signaling map of estrogen-mediated proliferation. Each estrogen receptor or receptor isoform provides a unique contribution to the proposed network and either initiates or inhibits specific steps in receptor-mediated signaling cascades, ultimately resulting in either activation or suppression of proliferation. The arrow from ERα66 to GPER denotes a hypothetical interaction that has been suggested by multiple publications but yet to be definitively demonstrated. (B) Implementation of a mathematical description of the model structure depicted in A. To demonstrate the potential impact of the interactions between receptors, we implemented the directional signaling into a mathematical model (see Supplemental materials). Model parameters representing activation of individual receptors were varied to simulate selective modulation (affinity) by ligands. Depending on the relative affinity of chemicals for the various receptors, the shape of the dose–response curve may vary significantly resulting in proliferation, inhibition, or non-monotonic dose–response.

While the traditionally recognized ERα (ERα66) clearly has an important role in estrogen-mediated response, it is accompanied by a suite of estrogen receptors that are also involved in regulating cellular response to estrogen. ERβ, ERα66, and ERα46 act as nuclear receptors that modulate estrogenic activities. Further, non-genomic signaling through GPER, ERα36, and EGFR is critical to mediating the proliferative response. This model also incorporates the directionality of the interactions between receptors (Figure 3(A)). It is important to acknowledge that additional signaling processes such as those initiated through progesterone receptor or the aryl hydrocarbon receptor can affect estrogen signaling (Migliaccio et al. 1998; Swedenborg & Pongratz 2010). This cross-talk could be an important factor to consider during model development but is not included in the scope of this review.

The overlapping signaling processes inherent in the system point to a complex system of activation and inhibition that provides multiple levels of control of the dose-response for estrogen-mediated cellular responses. Binding of the estrogen receptors can either initiate or inhibit a specific signaling event and it is the combination of these events that ultimately shapes proliferative response curve. The ligand specificity of each ER isoform is, therefore, of great importance for determining the unique activation profile for individual compounds. Some work has been performed to this effect, but a more comprehensive exploration of binding activity for each of the receptors including GPER should be performed as part of a developmental strategy for a uterine specific assay (Lin et al. 2013).

To explore the expected behavior of this signaling system, we created a mathematical representation of this model based on mass-action kinetics with Hill functions to describe receptor–ligand interactions (described in Supplemental material). We then used the mathematical model to explore how proliferation varies for different substrates with selectivity for ERα66 (blue), ERα46 (purple), ERβ (yellow), and GPER (red) (Figure 3(B)). To demonstrate the complexities that arise from activation of multiple ERs in this model, we also simulated an estrogenic compound with affinity for both ERα66 and ERβ (green). In this simulation, there is a high affinity, low capacity for ERβ (the K_m of the ligand for ERβ is lower than for ERα66, but the V^max of ERα66 activation is higher than that of ERβ). At low concentrations, the negative influence of ERβ on proliferation dominates. At higher concentrations, however, the higher capacity of ERα66 compensates for its lower affinity, resulting in a non-monotonic dose–response curve. In a similar vein, differences in expression of the ER isoforms or the activating cofactors would also lead to different responses to ligand activation. Thus, this relatively simple biologically based model demonstrates how the complex estrogen signaling networks present in the cell could lead to widely different dose–response curves depending on the activating ligand. Viewing the process of receptor activation of proliferation through this simple two-component system shows the difficulties in interpreting dose response through the lens of a single ERα66-mediated process.

Implications for current estrogen disruptor screening efforts

The toxicology and risk assessment fields are moving quickly towards the use of in vitro biology coupled with computational modeling to replace animal-based toxicity testing. In the ToxCast program at the US EPA, large-scale, high-throughput screening (HTS) tests thousands of chemicals and prioritize them for further animal testing based on relative risk (i.e. margin of exposure determined from in vitro activity). A more long-term goal is to move towards replacement of animal tests (Krewski et al. 2010; Reif et al. 2010; Rotroff et al. 2013; Rotroff et al. 2014; Browne et al. 2015) where the results from in vitro test platforms inform the process to set safe exposure levels.

Unfortunately, the majority of current HTS assays were not designed to probe known toxicity pathways but rather to quantify one specific event outside the context of an intact signaling network. Little consideration was given to incorporating assays to cover each key event associated with the toxicity or adverse outcome pathways and instead, the assay suites were expected to throw a broad net across potential chemical activities. In the case of estrogen, the ToxCast and Tox21 assays include a series of 16 assays purported to test key events along the entire estrogen AOP. However, when considered in light of the current knowledge of the estrogen pathway (Table 1), these estrogen-related assays are built upon a very simple picture of classical estrogen receptor-mediated signaling, in which estrogen binds to ERα (i.e. ERα66) and/or ERβ, leading to dimer formation and direct transcriptional regulation of proliferation. When considered in terms of “hit” or “no-hit” screening, these assays have a high success rate for predicting in vivo rodent tests (Browne et al. 2015). However, the feedback mechanisms, noted throughout this review, clearly show that the dose–-response profile for proliferation is dependent on a more nuanced signaling network (Figure 3) that involves many more governing factors than the classical paradigm (ligand binding-dimerization-transcription-proliferation) would require. The negative feedback provided by ERβ and ERα46 are likely present to maintain homeostasis in the face of transient hormone fluctuations. Likewise, the existence of genomic and non-genomic pathways would amplify proliferation in the presence of sustained estrogen signaling. As demonstrated by the computational model, these overlapping signaling events have implications for chemical dose–response, and they should be incorporated into any in vitro model whose goal is to evaluate safety of chemicals. The following discussion addresses how the current knowledge of estrogen signaling and larger set of biological data could be leveraged to design a fit-for-purpose assay for estrogen-mediated proliferation. We also discuss key data gaps that should be addressed in order to ensure utility of such assays for human safety decisions.

A data driven approach to in vitro assay design for assessing human risk

Designing in vitro assays to replace in vivo methods requires selection of cell models that incorporate key signaling components that lead to a specific phenotypic cellular outcome and accurately recapitulate the dose–response relationship expected in the intact system. Thus, the context of the in vitro system is important. Does the system use an intact cell expressing the key proteins in the signaling pathway? Does the media formulation provide needed cofactors (EGF, Ca++, etc.)? Does the cell model recapitulate known responses in human tissues? And finally, is the model capable of predicting dose response behavior consistent with that seen for the intact human?

Figure 3(A) provides an overview of key receptors that have been shown to be important in cellular response to estrogen and should be present in cellular models whose purpose is to predict dose response for chemical perturbation of the estrogen pathway. To incorporate this mechanistic outline into a useful in vitro model for safety assessment, several additional issues need consideration. Of particular importance is the question of tissue type. Tissue-specific differences in the response to exogenous estrogenic compounds do occur in the human. Tamoxifen, a compound used to treat breast cancer, is a clear example. Tamoxifen is highly effective at inhibiting growth of estrogen responsive breast cancers, but causes endometrial cancer in some patients (Barakat 1995). Tissue-specific dose response likely results from tissue-dependent expression of the receptors, their cofactors, and differential binding affinity of estrogenic ligands to the various receptors in the signaling pathway. Additionally, the context of the cell within the organ system will need to be further evaluated. In particular, uterine responses are a coordinated network of tissue types engaged in both auto- and paracrine signaling to achieve a unified activity such as thickening of the uterine wall. These external factors should be evaluated for significance to the endpoint of concern and incorporated into the experimental system if necessary.

Differential binding affinity has been demonstrated for a number of compounds including endogenous estrogen. ERα36 demonstrates no specific binding affinity for E2 within physiologic ranges, while ERα46 and ERα66 bind E2 with Kd values in the 6070 picomolar range (Lin et al. 2013). Several exogenous ligands bind ERα46 and ERα66 with very different affinities, however, indicating that chemicals could preferentially activate individual ERα isoforms (Lin et al. 2013). These differences are likely a result of structural changes within the ligand binding pocket as ERα36 is missing a section of the binding domain and ERα46 may have altered tertiary interactions which will result in a conformational change effecting binding. However, a study directly evaluating the structure of the binding pockets in each of these isoforms has yet to be performed. The potential for each isoform to have unique activation profiles to ligands is a critical piece of the signaling process and must be included in future screening efforts.

When combined with tissue-specific expression of the receptor isoforms, this preferential activation of specific receptors could have a substantial impact on the shape of the dose–response curves for environmental compounds (as demonstrated by the mathematical model in Figure 3(B)). The observed disparity in tissue response indicates that it is likely necessary to include breast and uterine-specific cellular assays in both screening and dose–response evaluations. Further, these breast and uterine cell models should insure the presence of the key signaling components that drive the tissue-specific dose response.

Key data gaps

Incorporating tissue-specific expression of the ER isoforms into cell proliferation models presents a challenge, as studies evaluating isoform concentrations in the human breast and endometrium are limited. While many studies have identified ERα and ERβ proteins in normal and malignant human endometrial tissue, the majority of these have not distinguished between ERα isoforms (Fujimoto et al. 1999; Mylonas et al. 2004; Mylonas et al. 2005; Knapp et al. 2013). Irsik et al. (2013) measured ERα66, ERα46, ERα36, and ERβ in various murine tissues. The four receptors were found to varying degrees in mammary, ovarian and uterine tissue samples, with ERα66 highly expressed in the ovaries and uterus, ERα46 expressed at low levels in the uterus with minimal ovarian and mammary expression, and ERα36 highly expressed in all three tissues. Expression of ERα36 has also been identified in the human endometrial cancer cell line (Ishikawa) and was found to be essential for E2induced cellular proliferation through activation of PKCδ/ERK pathways with ERα66 expression alone being insufficient for proliferative responses (Tong et al. 2010). ERα36 is expressed in human uterine tissue and there is a positive correlation between endometrial cancer and uterine ERα36 expression (Tu et al. 2011). Direct quantification of ERα46 expression in primary human uterine cells has not been published but if the mouse data holds true for human cells, some level of ERα46 would be expected in the endometrium (Irsik et al. 2013). Full length ERα66, by contrast, is ubiquitously expressed in the human uterus as identified by western blot (Hiroi et al. 1999; Shao et al. 2007; Knapp et al. 2013). Little is known on cell-type (epithelial, stromal, etc.) or menstrual cycle phase dependence of the expression of these receptors. What little work has been published on the subject used immunohistochemistry, a method that is less reliable for distinguishing ERα isoforms due to a lack of specificity of available ERα antibodies (Supplemental Figure 1). The regulation of expression may also be controlled by epigenetic patterns in the cells themselves (Hervouet et al. 2013). This will be a particular concern for reproductive toxicity as generational effects are a concern for estrogen active compounds (National Toxicology 2010). A careful study is needed to determine the expression of the ERα isoforms in human estrogen responsive cells if in vitro assays are to be predictive of responses in intact human cells/tissues.

In addition to incorporating data about ERα expression into the design, the differential expression of receptors other than ERα that have the potential for crosstalk with the ERα signaling pathways is important in developing cellular assays for more formal safety assessment. ERβ is ubiquitously expressed throughout the estrous cycle in various cell types, including the uterus (Critchley et al. 2002). GPER is also in the endometrium of healthy human subjects, with increased mRNA and protein levels during peak proliferative phases of the menstrual cycle (Kolkova et al. 2010; Plante et al. 2012). EGFR, like ERβ, is expressed at all points of the estrous cycle in the uterus (Smith et al. 1991). The ubiquitous and tissue specific expression of these receptors in women, together with the evidence described above for their roles in estrogen-mediated signaling, argues for a better consideration of the biology in in vitro systems that are meant to predict human safety.

Summary and perspectives

Inappropriate estrogenic signaling is associated with cancer and alteration in sexual development and is a major focus of toxicological assessment for chemicals and compounds in consumer products. The push for an in vitro model of estrogen signaling to replace the current in vivo methods, such as the rat uterotrophic assay, has gained momentum in recent years. However, current HTS in vitro assays are designed for screening and prioritization, not evaluation of dose response and prediction of human safety (Krewski et al. 2010). The aim of this paper was to review the current literature related to estrogen-mediated proliferative signaling and provide a framework with which to move forward with development of an in vitro strategy to quantitatively assess human risk. Here, we have described the current literature related to the three major isoforms of ERα, with detailed analysis of tissue specific signaling mechanisms and interactions of additional types of estrogen receptors including GPER and ERβ. We have identified key data gaps, including research specific to receptor expression in normal human tissues, evaluation of tissue specific differences in signaling interactions, and studies with realistic estrogen exposure levels. We organized the existing data into a signaling network that accounts for the contributions of ERα66, ERα46, ERα36, ERβ, GPER, and EGFR to estrogen-mediated proliferation and used this knowledge of the ER signaling pathway to recommend an in vitro assay design fit for the purpose of predicting estrogenic chemical dose–response in the human uterus. The future of in vitro safety assessments will benefit from the use of systems biology approaches, such as those outlined here, to inform in vitro assay design. Careful consideration of the key biological mediators of cellular outcome (i.e. key events in adverse outcome pathways) will be important to support a move to in vitro-based safety testing while ensuring predictivity and accuracy in risk assessment.

The purpose of this review was primarily to provide an example of how the current biological knowledge could be brought together to design an in vitro assay fit for the purpose of defining points of departure using estrogenic compounds as an example. The broader discussion of how these assays will be used in the risk assessment process was outside our scope. In particular, with endocrine active chemicals, there is much debate on how such chemicals should be regulated, and even whether classic dose-paradigms apply to endocrine response (Vandenberg et al. 2012; Autrup et al. 2015). Here, we show how the complex circuitry can be incorporated into both in vitro and in silico models to allow evaluation of the complex networks controlling cellular responses. As demonstrated by the computational model, the feedforward and feedback processes can lead to a variety of dose-response behaviors, depending on the relative affinity of the chemical(s) for the receptors. Varied affinities for different receptors would essentially lead to a “mixture” effect, where individual dose-response curves lead to composite behavior that could be monotonic or non-monotonic. Importantly, however, this paper also demonstrates how these circuits can be defined and incorporated into a biologically based testing paradigm, where the processes governing chemical dose–response (even complex composite dose–response curves) are tractable and predictable, and the assay readouts can be used with confidence for decisions about chemical safety.

Declaration of interest

The current employment affiliation of the authors is as shown on the cover page. Prior to 1 January 2016, all the authors were employed by the Hamner Institutes for Health Sciences as part of the Institute for Chemical Safety Sciences (ICSS). The Hamner Institutes was a non-profit toxicology research institution that ceased operations on 31 December 2015. SciMetrika, a privately owned population health consulting commercial entity, acquired the assets of the ICSS division, including former members of ICSS leadership and staff, through its wholly owned subsidiary, ScitoVation. ScitoVation is a privately owned research company dedicated to advancing human health through the development of research tools, processes and technologies using human cell-based and computational methods. The authors have sole responsibility for the writing and content of the paper. This work was supported by the Long-Range Research Initiative (LRI) of the American Chemistry Council and was also performed as part of the Hamner Institutes’ “TT21C Consortium” with funding from Agilent Technologies, CropLife America, Dow Chemical Company, Dow Corning, and ExxonMobil Biomedical Sciences Foundation. The funding organizations were able to view early drafts of the paper. The funding organizations did not have direct influence on this review and did not contribute to any part of the writing or editorial process. The final work product, including conclusions drawn and recommendations offered, is the exclusive professional work product of the authors and may not necessarily represent the views of the sponsoring organizations or ScitoVation. None of the authors has appeared during the past five years in any legal or regulatory proceedings related to the material covered in this review.

Supplemental material

Supplemental data for this article can be accessed here.

Acknowledgements

The authors would like to thank Dr. Rebecca Alyea for her assistance in selecting publications with appropriate content for this review. The authors are grateful for the extensive and knowledgeable feedback provided by tehe anonymous peer reviewers selected by they Editor of the journal.