The Need to Decide If All Estrogens Are Intrinsically Similar

We used gene expression profiling to investigate whether the molecular effects induced by estrogens of different provenance are intrinsically similar. In this article we show that the physiologic estrogen 17β-estradiol, the phytoestrogen genistein, and the synthetic estrogen diethylstilbestrol alter the expression of the same 179 genes in the intact immature mouse uterus under conditions where each chemical has produced an equivalent gravimetric and histologic uterotrophic effect, using the standard 3-day assay protocol. Data are also presented indicating the limitations associated with comparison of gene expression profiles for different chemicals at times before the uterotrophic effects are fully realized. We conclude that the case has yet to be made for regarding synthetic estrogens as presenting a unique human hazard compared with phytoestrogens and physiologic estrogens.

The question of whether phytoestrogens and synthetic estrogens are toxicologically similar, or intrinsically different, presents a challenge to all involved in human hazard and risk assessments. Although there is a general concern that exposure to nanogram or microgram amounts of environmental estrogens may be associated with adverse health effects, in the public mind there is a widespread belief that foods and dietary supplements containing milligram quantities of phytoestrogens confer only health benefits. An implicit distinction therefore seems to have been drawn between synthetic and plant-derived estrogens-a belief sustained in the public mind by the assumption that natural is good and synthetic is bad-but an untested and potentially misleading notion for those involved with sciencebased human hazard/risk assessments.
Phytoestrogens and synthetic estrogens are generally considered separately in the literature. For example, Howdeshell et al. (1999) suggested a possible association between the advance in first estrus observed in mice exposed in utero to 2.4 µg/kg of the synthetic environmental estrogen bisphenol A and reports of an increased incidence of hypospadias in boys (Paulozzi et al. 1997) and the earlier sexual maturation of girls (Herman-Giddens et al. 1997)-the implication being that synthetic estrogens present a greater hazard than the much higher levels of phytoestrogens being consumed by those same children. In contrast, there are reports of an increased incidence of hypospadias in boys born to vegetarians (North and Golding 2000), of alterations in the menstrual cycle (Cassidy et al. 1994), and of reduced breast cancer incidences (Messina 1999) among women eating diets rich in phytoestrogens. Support for these epidemiologic observations comes from experimental studies indicating that advances in sexual development in rodents can be induced by their exposure to phytoestrogens (Casanova et al. 1999;Cassidy and Faughnan 2000;Safe et al. 2002). In contrast to these separate lines of inquiry, Newbold and colleagues have evaluated potential similarities between natural and synthetic estrogens. In seminal studies, they demonstrated that neonatal exposure of female mice to equipotent uterotrophic doses of the phytoestrogen genistein (GEN; Figure 1) or the synthetic estrogen diethylstilbestrol (DES) leads to an identical incidence of uterine adenomas at 18 months of age (Newbold et al. 2001). However, in attempting to draw parallels, or distinctions, between phytoestrogens and synthetic estrogens, it is imperative to consider growing awareness of the complexity of estrogen signaling pathway and the pleuripotential biologic activities of most organic chemicalsirrespective of their origin.
Estrogen signaling in mammalian cells is primarily mediated at the molecular level by two members of the nuclear receptor superfamily-estrogen receptors alpha (ER-α) and beta (ER-β). Ligand-activated ER-α and ER-β function as transcription factors, in conjunction with numerous coregulatory proteins, in order to activate or repress the transcription of ER-responsive genes (Hall et al. 2001;Moggs and Orphanides 2001). There is considerable variation in the binding affinity of ER-α and ER-β among different estrogens (Kuiper et al. 1998). In the case of the chemicals studied here, the physiologic estrogen 17β-estradiol (E 2 ) and DES bind with a similar affinity to ER-α and ER-β, whereas GEN binds with approximately 20-fold higher affinity to ER-β than to ER-α (Kuiper et al. 1998). Concerning nonhormonal properties of the test chemicals (most of which have only be defined in vitro), GEN inhibits a range of enzymes, including tyrosine kinases (Akiyama et al. 1987), nitric oxide synthase (Duarte et al. 1997), and topoisomerase II (Okura et al. 1988), and also decreases calcium-channel activity (Potier and Rovira 1999), lipid peroxidation (Arora et al. 1998), and diacylglycerol synthesis (Dean et al. 1989). Likewise, DES is reported to induce aneuploidy in mammalian cells (Aardema et al. 1998) and to bind to rat liver DNA (Williams et al. 1993). More recently, some phytoestrogens were reported to inhibit the aromatase-mediated conversion of testosterone to E 2 in vitro (Almstrup et al. 2002), and equol, the major circulating estrogenic metabolite associated with the dietary ingestion of phytoestrogens, is reported to selectively sequester dihydrotestosterone and thereby to act as a functional antiandrogen in vivo (Lund et al. 2004).
In order to advance understanding in this area, we decided to compare the genes expressed in the immature mouse uterus when it had grown in response to treatment with the estrogens E 2 , DES, and GEN. The immature mouse uterus was selected for our analysis because it is a major estrogen-responsive organ and forms the basis for a reference assay of estrogenic activity (Owens and Ashby 2002), including carcinogenesis (Newbold et al. 2001). Furthermore, it expresses both ER-α and ER-β (Weihua et al. 2000) and the androgen receptor (Frasor et al. 2003). We initially conducted a global analysis of gene expression in the mouse uterus at 1, 2, 4, 8, 24, 48, and 72 hr after exposure to a single high dose of either GEN (250 mg/kg) or E 2 (400 µg/kg). These single high doses yielded a sustained uterotrophic response over 72 hr ( Figure 2A) and were selected to avoid the complex transcriptional program that may result from the standard uterotrophic assay exposure regime in which each test compound is dosed by repeated administration on 3 consecutive days (Odum et al. 1997 subcutaneous injection of each compound or the test vehicle [arachis oil (AO); 5 mL/kg], and uterine RNA was isolated and pooled by group at each of the seven time points to determine gene expression levels among the 12,488 mouse genes represented on the Affymetrix MG-U74Av2 GeneChip (Affymetrix, High Wycombe, UK). Transcript profiling was performed using MG-U74Av2 GeneChip and Microarray Analysis Suite 5.0 (Affymetrix). Normalization and hierarchical clustering were performed with GeneSpring 6.0 (Silicon Genetics, Redwood City, CA, USA). MIAME (Minimum Information About a Microarray Experiment)-compliant microarray data are available as supplementary information and submitted to the Gene Expression Omnibus (GEO) database (GEO 2004). These data were analyzed using unsupervised hierarchical clustering and yielded temporal relationships between the expression profiles of 3,450 genes that were either up-or down-regulated (> 1.5-fold) by E 2 and/or GEN ( Figure 2B). Each chemical induced a similar, multistage transcriptional response ( Figure 2B), although it is noteworthy that we observed variations in the magnitude and timing of both early (e.g., c-fos) and late (e.g., lactotransferrin) ER-responsive genes during the uterotrophic responses induced by E 2 and GEN ( Figure 2C).
A detailed description of the molecular functions of the genes affected, together with their association with physiologic changes during uterine growth, has been reported (Orphanides et al. 2003) and will be described VOLUME 112 | NUMBER 11 | August 2004 • Environmental Health Perspectives (A) Blotted uterine weights (mean ± SD) of sexually immature mice (n = 10/group) at different times after a single subcutaneous dose of E 2 (400 µg/kg), GEN (250 mg/kg), or AO (control; 5 mL/kg). See text for details of experiments. (B) Temporal expression profiles of 3,450 genes up-regulated or repressed (> 1.5-fold) by either E 2 (400 µg/kg) or GEN (250 mg/kg) at one or more of seven different time points. The magnitude of altered gene expression (fold change vs. time-matched vehicle control) is indicated by color; genes are grouped according to similarity of their temporal expression profiles (Pearson correlation-based hierarchical clustering). (C) Northern blot analysis of temporal expression pattern of early (c-fos) and late (lactotransferrin) estrogen-responsive genes; the fold induction of gene expression relative to time-matched vehicle controls was calculated after data were normalized to the expression of the control gene RPB1 (accession number NM_009089). *p < 0.05; **p < 0.01; two-sided Student t-test.  in more detail in a future publication (Moggs et al., unpublished data). These observations suggest that GEN does not induce "off-target" ER-independent transcriptional responses, that is, those associated with the properties of GEN other than estrogenicity. Furthermore, there was no evidence for the topoisomerase II-inhibiting properties of GEN in the bone marrow of the present mice despite demonstration of the sensitivity of that tissue to the potent micronucleus-inducing activity of the topoisomerase II inhibitor etoposide (data not shown). Together, these data led us to question whether a synthetic estrogen such as DES would also induce similar transcriptional responses in the immature mouse uterus.

Time (hr) Time (hr)
In order to avoid temporal vagaries in gene expression (e.g., Figure 2C), we decided to anchor our transcript profiling data to the phenotype of the grown uterus by employing equipotent uterotrophic doses of E 2 , GEN, and DES. We compared the global gene expression profiles in the uteri of intact immature mice stimulated with three daily low doses of either GEN, DES, or E 2 , with an exposure regimen the same as that used in a standard 3-day uterotrophic assay (Odum et al. 1997). The route of administration and the doses of GEN and DES used were as described by Newbold et al. (2001) in their equivalent-outcome carcinogenicity bioassays of these two chemicals. Three independent replicates of four groups of sexually immature mice (Alpk:APfCD-1; 19/20 days of age; maintained on RM1 diet) received three daily subcutaneous injections of GEN (50 mg/kg), E 2 (2.5 µg/kg), or DES (2 µg/kg). Control animals received the vehicle, AO (5 mL/kg). These doses elicited similar uterotrophic responses (72 hr after the initial dose; Figure 3A, Table 1) and identical histologic changes in the uteri of the treated animals (Table 1). Uterine RNA was isolated and pooled for each of the 12 groups and analyzed for changes in gene expression levels using the same Affymetrix microarray of 12,488 mouse genes. The data were analyzed using two independent statistical methods. First, unsupervised hierarchical clustering defined the global relationships (Euclidean distances) between the 12 gene expression profiles ( Figure 3B). The three control groups clustered under one node, whereas the chemical treatment groups formed a separate node of compoundindependent clusters, indicating equal similarity within and between the transcriptional responses induced by the three estrogens ( Figure 3B). One-way analysis of variance (ANOVA), with Bonferroni (Holm 1979) correction (familywise error rate < 0.05) to minimize false positives, identified 179 genes where expression levels were altered by one or more chemical treatments ( Figure 3C). Remarkably, Tukey post hoc testing revealed that all of these genes were affected in all nine compound treatment groups. Table 2 highlights the high degree of similarity between the transcriptional responses to each of the three estrogens. These include established estrogen-responsive genes such as lactotransferrin, complement component 3, c-fos, small proline-rich protein 2A, and keratoepithelin (Hewitt et al. 2003;Naciff et al. 2003), together with many genes that have not  previously been associated with estrogenicity (Table 2).
Although these three estrogens can alter the expression of some genes with different magnitudes [e.g., peptidyl arginine deiminase II is up-regulated to a lesser extent by E 2 (1.86-fold ± 0.27) relative to GEN (9.11-fold ± 0.33) and DES (5.15-fold ± 1.53); Table 2], the present data show that the same genes are affected during equivalent uterotrophic responses. Previous studies have revealed both similarities and differences between transcriptional responses induced at a single time point after exposure to E 2 and DES in the uteri of immature ovariectomized mice (Watanabe et al. 2003) and after exposure to either GEN, bisphenol A, or 17α-ethynyl estradiol in the reproductive tract of intact adult rats (Naciff et al. 2002). We suggest that these reported differences most probably arise from dose-dependent variations in the magnitude and kinetics of gene expression ( Figure 2C), rather than from the operation of distinct mechanisms of estrogenic action.
Our data indicate that estrogens of differing provenance may have in common the potential for both beneficial and adverse health effects. This highlights the need for an holistic approach to hazard assessment wherein preconceptions are replaced by an objective assessment of the likely perturbations of physiologic functions caused by combined exposures to physiologic, synthetic, and plantderived estrogens. This need is reinforced by data showing that plasma concentrations of isoflavones in infants fed soy formula are approximately 200 times higher than for those fed human milk (Setchell et al. 1997 Liu et al. 2003), GenBank accession numbers (GenBank 2004), and mean (± SD) fold induction/repression of gene expression are shown in the same order as the gene cluster in Figure 3C. b Genes mentioned in the text.