The mice were given a single subcutaneous injection of E₂ (400 μg/kg) or arachis oil (AO; vehicle control) using a dosing volume of 5 mL/kg body weight. A single dose of E₂ was used to avoid the complex transcriptional program that may result from the standard uterotrophic assay exposure regime (i.e., repeated administration on 3 consecutive days; Odum et al. 1997). The relatively high dose level of 400 μg/kg was chosen to ensure a sustained and significant increase in blotted uterine weight during the 72-hr sampling period (Supplemental Data, Figure 1). No overt toxicity was observed during the 72-hr exposure to E₂ (400 μg/kg). All animals were terminated at the appropriate time using an overdose of halothane (Concord Pharmaceuticals Ltd., Essex, UK) followed by cervical dislocation. Vaginal opening was recorded, and the uterus was then removed, trimmed free of fat, gently blotted, and weighed. Blotted uterine weights were analyzed by covariance with terminal body weights (SAS Institute Inc. 1999). Half of each left uterine horn was fixed in 10% formol saline and processed to paraffin wax for histologic analysis (Odum et al. 1997). The mean thickness of the endometrial and epithelial cell layers, indicators of cellular hypertrophy, were calculated based on the assessment of 10 locations on hemotoxylin- and eosin-stained longitudinal uterine sections for each animal. All hypertrophy data were assessed for statistical significance by analysis of variance (ANOVA). The remainder of the uterus was snap frozen in liquid nitrogen and stored at −70°C for RNA extraction.

The total number of mitotic figures in each uterus section was counted, noting the tissue location, and the area of the section was measured using a KS400 image analysis system (Imaging Associates, Bicester, UK). The number of mitotic figures per square millimeter was calculated, and the frequency after administration of E₂ was compared with the frequency seen after the administration of AO using an appropriate statistical procedure. The number of mitoses per square millimeter was considered by a fixed-effects ANOVA allowing for treatment, time, and the treatment by time interaction. Analyses were carried out using the MIXED procedure in SAS, version 8.2 (SAS Institute Inc. 1999). Contrasts within the treatment by time interaction provided estimates of differences in E₂ and control response at each time point. These were compared statistically using a two-sided Student t-test based on the error mean square in the ANOVA.

Three independent biologic replicates of the entire time course study for E₂-treated and time-matched AO-treated groups of animals were used to generate transcript profiling data and for subsequent statistical analysis. To minimize the effect of any interanimal variability, total RNA was isolated from the pooled uteri for each treatment group (n = 10 in the first two studies; reduced to n = 5 for the last study because of highly similar transcriptional responses being obtained in replicate studies 1 and 2) using RNeasy Midi kits (Qiagen Ltd., Crawley, West Sussex, UK). Biotin-labeled complementary RNAs were synthesized using the Enzo Bioarray HighYield RNA transcript labeling kit and hybridized to Affymetrix murine U74-Av2 GeneChips as described previously (Zhu et al. 2001) and in the Affymetrix GeneChip expression analysis technical manual (Affymetrix, Inc. 2002). Probe arrays were scanned and the intensities were averaged using Microarray Analysis Suite 5.0 (Affymetrix, High Wycombe, UK). The mean signal intensity of each array was globally scaled to a target signal value of 500. To select E₂-responsive genes, each gene was subjected to a mixed-model ANOVA allowing for treatment, time, and the treatment by time interaction as fixed effects and replicate study as a random effect. The use of mixed ANOVA models for the analysis of differential gene expression in microarray experiments has been previously described (Churchill 2004; Cui and Churchill 2003). Analyses were carried out using the MIXED procedure in SAS, version 8.2 (SAS Institute Inc. 1999). Contrasts within the treatment by time interaction provided estimates of differences in E₂ and control response at each time point. These were compared statistically using a two-sided Student t-test based on the error mean square in the ANOVA [Supplemental Data, Table 1 (http://ehp.niehs.nih.gov/txg/members/2004/7345/supplemental.pdf)]. Data for genes exhibiting significant changes in expression (p < 0.01, two-sided) at one or more time points were then exported into GeneSpring 6.0 (SiliconGenetics, Redwood City, CA, USA), and a data transformation (values < 0.01 set to 0.01) and per-chip normalization (to the 50th percentile) were applied. Genes that did not have a Present detection call (Affymetrix) in any of the 14 treatment groups were removed from further analysis. Ratios of changes in gene expression were then calculated by normalizing each E₂-treated sample to its corresponding time-matched vehicle (AO)-treated control. GeneChip data sets for the three independent biologic replicates were interpreted in log of ratio analysis mode, with biologic replicates being selected as a noncontinuous parameter. A total of 3,538 E₂-responsive genes exhibiting a minimum of 1.5-fold up- or down-regulation in at least one time point were then subjected to gene tree–based hierarchical clustering (Pearson correlation). To identify genes that function in specific biologic pathways, these 3,538 genes were further filtered using functional annotations derived from the NetAffx database‚ Analysis Center (Liu et al. 2003; http://www.affymetrix.com/analysis/index.affx), together with manual annotations from published literature, before hierarchical clustering using GeneSpring. Gene names used in this article (see Appendix) were derived by homology searching of nucleotide sequence databases (BLASTn; http://www.ncbi.nih.gov/BLAST/) using Affymetrix probe target sequences and the interrogation of NetAffx (Liu et al. 2003) database. All genes described in the figures and text showed statistically significant alterations in expression in all three replicate studies. MIAME (Minimum Information About a Microarray Experiment)-compliant microarray data for the three independent replicate studies are available as supplementary information and have been submitted to the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/).

Uterine RNA from the seven time points for each of the E₂-treated and time-matched vehicle control groups was analyzed using Affymetrix MG-U74Av2 GeneChips. A total of 42 microarray data sets were collected for the three replicate experiments. We used a multistep method to analyze the microarray gene expression data (Figure 3A). First, data were filtered and subjected to statistical analyses to identify the 3,538 genes with altered expression in E₂-treated mice (p < 0.01 and > 1.5-fold) during at least one time point (see “Materials and Methods”). Unsupervised hierarchical clustering was then used to group these genes into co-regulated clusters (Quackenbush 2002; Figure 3B), revealing a complex multistage transcriptional response to E₂ in the uterus (gene clusters A–I in Figure 3B). To gain an overview of the predominant molecular functions and biologic pathways that were regulated at the transcriptional level during the uterotrophic response to E₂, we interrogated the 3,538 E₂-responsive genes using the GOStat gene ontology mining tool (http://gostat.wehi.edu.au) (Beissbarth and Speed 2004). This approach revealed that E₂ targets predominantly genes involved in protein metabolism, cell cycle, cell proliferation, DNA replication, RNA metabolism, mRNA transcription, and blood vessel development [Supplemental Data, Table 2 (http://ehp.niehs.nih.gov/txg/members/2004/7345/supplemental.pdf)]. Next, we used a supervised clustering approach using customized gene ontology definitions (see “Materials and Methods”) to identify gene functions that were predominant in each co-regulated cluster in Figure 3B. This revealed that E₂ regulates each class of gene during a narrow window of time and suggests that E₂ induces uterine growth and maturation by regulating successively the activities of different biologic pathways (described below). Finally, we analyzed the temporal associations between the gene expression program and alterations in uterine weight and histology to anchor the gene expression changes to alterations in uterine phenotype. These associations are described below.

The first 4 hr of the uterotrophic response is characterized by the influx into the uterus of fluid that provides the nutrients and ions required for growth (Clark and Mani 1994). This leads to decompaction of stromal cells (Figure 4A) and thickening of the stromal endometrial layer at 4 hr (Figure 2B). This first phase of the uterotrophic response is accompanied by the rapid and transient regulation of genes encoding components of intra- and inter-cellular signaling pathways (Figure 4B) and sequence-specific transcriptional regulators (Figure 4C). Most of these genes show maximal expression between 1 and 4 hr, suggesting that the transcriptional effects of E ₂, mediated via ER- αand ER-β, are amplified rapidly through the induction or modulation of multiple transcriptional and nontranscriptional signaling pathways.

The signaling genes rapidly up-regulated by E₂ function in a broad array of signal transduction pathways (Figure 4B). These genes include protein kinases (AKT, MEK1, PIM3), growth factors (VEGF, PLGF), GTPases (RHOC, RAB11A, DEXRAS1), cytokine signaling proteins (MCP1, SOCS1, SOCS3, WSB1, IL17R), and a Wnt signaling factor (WNT4). Several E ₂ -induced genes may act to attenuate initial signaling events (e.g., the protein phosphatase MKP1 negatively modulates MAP kinase activity). Strikingly, many of the signaling genes induced within 4 hr of E₂ exposure have roles in the regulation of vascular permeability in other tissues, suggesting that they may be involved directly in initiating the influx of fluid into the uterus at this time (Figure 4B). These genes include angiogenic/vascular cell growth factors (VEGF, PLGF, ADM, ANGPT2, TGFB2), vasoactive serine proteases (KLK2, KLK6, KLK9, KLK22), and vascular endothelial receptors (IL17R, BDKRB1, ENG, GNA13). Furthermore, the vascular growth factor receptors TIE1 and TIE2 are rapidly down-regulated in response to E₂ (Figure 4B), which may serve to attenuate the uptake of fluid after 4 hr. Collectively, these genes shed light on the mechanism by which E₂ promotes fluid uptake in the uterus and provide a clear link between gene expression changes and histologic changes occurring at this time.

Our data also provide insight into the mechanisms by which E₂ releases cells of the immature uterus from quiescence and promotes cell division. The E₂-induced expression profile of E2F1, a key transcriptional regulator of DNA replication genes (Ohtani 1999), closely parallels the induction of the chromosome replication genes (Figure 6B), consistent with the proposal that E2F1 regulates the expression of components of the DNA replication fork in human breast cancer cell lines exposed to E₂ (Lobenhofer et al. 2002). Our data indicate that release from quiescence also involves the E₂-induced down-regulation of genes that maintain cells in a growth-arrested state (KIP1, CCNG2, CCNG1). The principle way in which mitogens induce proliferation of quiescent cells involves a reduction in levels of the Kip1 protein, which inhibits the activities of cyclin–cdk complexes and induces cell cycle arrest (Olashaw and Pledger 2002). We found that KIP1 was down-regulated within 1 hr of E₂ exposure and remains repressed over a period of at least 24 hr, only reaching control levels when cell proliferation has ceased (Figure 6C). Furthermore, E₂ may promote degradation of the Kip1 protein via the induction of CDC34 (Figure 6C), a gene that has been implicated in the ubiquitin-mediated degradation of Kip1 (Koepp et al. 1999). These data suggest that E₂ promotes cell proliferation by coordinately reducing Kip1 mRNA and protein levels. It is not clear whether KIP1 is a direct or indirect target of the activated ERs. However, KIP1 gene expression is controlled by ras-mediated PI3K signaling pathways (Olashaw and Pledger 2002), components of which are up-regulated rapidly in response to E₂ (e.g., DEXRAS1, RASSF1; Figure 4B).

E₂ protects against apoptosis in a number of tissues, including brain, testes, and uterus (Thompson 1994). Although the anti-apoptotic activity of estrogen in the uterus is thought to play a crucial role in the maintenance of uterine homeostasis, the mechanistic basis for this action has not been defined. Our data reveal that E₂ induces the expression of anti-apoptotic genes (BAG2, BAG3, DAD1) while simultaneously down-regulating the expression of pro-apoptotic genes (CASP2, NIX; Figure 6D). Thus, apoptosis appears to be suppressed through transcriptional mechanisms during E₂-induced uterine growth. Consistent with these observations, E₂ also induces the apoptotic regulators BCL2 and BAG1 in cultured breast cancer cells (Perillo et al. 2000; Soulez and Parker 2001). It will be important to determine whether estrogens elicit similar changes in the expression of apoptosis-regulating genes in other tissues.

The period from 24 to 72 hr after E ₂ exposure is associated with remodeling of the luminal epithelial cell layer, including the formation of secretory epithelial cells and a glycocalyx layer consisting of glycoproteins (Paria et al. 2003; Weitlauf 1994). These changes result in the formation of a highly differentiated epithelial layer that is primed for cell recognition and adhesion events necessary for embryo attachment and implantation.

The final phase of the uterotrophic response coincides with the induction of a battery of genes involved in the cytoarchitectural remodeling of proliferating uterine cells, thus providing a further link between phenotypic and gene expression changes (Figure 7A). These genes encode components of desmosomes (DSG2), gap junctions (CX26), tight junctions (CLDN4, CLDN7), the cornified envelope (SPRR1A, 2A, 2B, 2E, 2F, 2G, 2I, 2J), intermediate filaments (KRT19), and a variety of cell-surface and extracellular-matrix glycoproteins (SPP1, BGP1, BGP2, MUC1, TROP2, CLU). The latter class of genes is likely to contribute to the formation of the glycocalyx layer present on differentiated uterine epithelium (Weitlauf 1994). The concomitant E₂-dependent induction of a number of enzymes required for carbohydrate metabolism (MAN2B1, GALNT3) may provide the increase in sugar metabolism necessary for the production of these glycoproteins. E₂ also induces genes encoding ion channels that regulate the balance of Na⁺ absorption and Cl⁻ secretion across the endometrial epithelium to maintain a luminal fluid microenvironment suitable for implantation (CFTR, CLCA3, MAT8; Figure 7A).