Drug and Cell Type-Specific Regulation of Genes with Different Classes of Estrogen Receptor β-Selective Agonists

Estrogens produce biological effects by interacting with two estrogen receptors, ERα and ERβ. Drugs that selectively target ERα or ERβ might be safer for conditions that have been traditionally treated with non-selective estrogens. Several synthetic and natural ERβ-selective compounds have been identified. One class of ERβ-selective agonists is represented by ERB-041 (WAY-202041) which binds to ERβ much greater than ERα. A second class of ERβ-selective agonists derived from plants include MF101, nyasol and liquiritigenin that bind similarly to both ERs, but only activate transcription with ERβ. Diarylpropionitrile represents a third class of ERβ-selective compounds because its selectivity is due to a combination of greater binding to ERβ and transcriptional activity. However, it is unclear if these three classes of ERβ-selective compounds produce similar biological activities. The goals of these studies were to determine the relative ERβ selectivity and pattern of gene expression of these three classes of ERβ-selective compounds compared to estradiol (E2), which is a non-selective ER agonist. U2OS cells stably transfected with ERα or ERβ were treated with E2 or the ERβ-selective compounds for 6 h. Microarray data demonstrated that ERB-041, MF101 and liquiritigenin were the most ERβ-selective agonists compared to estradiol, followed by nyasol and then diarylpropionitrile. FRET analysis showed that all compounds induced a similar conformation of ERβ, which is consistent with the finding that most genes regulated by the ERβ-selective compounds were similar to each other and E2. However, there were some classes of genes differentially regulated by the ERβ agonists and E2. Two ERβ-selective compounds, MF101 and liquiritigenin had cell type-specific effects as they regulated different genes in HeLa, Caco-2 and Ishikawa cell lines expressing ERβ. Our gene profiling studies demonstrate that while most of the genes were commonly regulated by ERβ-selective agonists and E2, there were some genes regulated that were distinct from each other and E2, suggesting that different ERβ-selective agonists might produce distinct biological and clinical effects.


Introduction
Estrogens exert their biological effects by interacting with two known ERs, ERa and ERb [1,2,3,4]. ERs are involved in development of the reproductive tract and regulation of reproductive processes [5]. In addition to their role in reproduction, ERs also have important roles in the breast, bone, brain and the cardiovascular system [1,2,3,4]. Studies with ERa and ERb knockout mice demonstrated that ERa is required for the development of certain tissues in the reproductive tract and mammary gland [6]. ERb knockout mice (bERKO) show other defects. There are fewer corpora lutea in the bERKO mice, which likely accounts for the observation that these mice are subfertile [7]. In luminal mammary epithelial cells of bERKO mice there was a widespread increase in the proliferation marker, Ki-67, suggesting that ERb is important for terminal differentiation of mammary epithelial cells [8]. Prostate and myelogenous hyperplasia have been observed in bERKO mice [9,10]. These mice also show a loss of anxiety [11] and spatial learning [12], and developed depression-like behavior [13]. These observations support a role for ERb in behavior, mood and affective disorders.
Estrogens have been used extensively to treat menopausal symptoms and osteoporosis in postmenopausal women. The Women's Health Initiative (WHI) trial found that the risks outweighed the benefits of hormone therapy (HT) [14,15,16,17,18]. It is now thought that some adverse effects of HT observed in the WHI were due to an older subject population in the trial [19,20]. However, there remains an intense effort to discover safer estrogens that selectively regulate ERa or ERb, as alternatives to the estrogens currently used in HT regimens that non-selectively regulate both ERs.
ERb-selective estrogens might be more desirable for HT than ERa-selective estrogens, because studies indicate that ERa mediates cell proliferation that contributes to breast and endometrial cancer whereas ERb generally is thought to counteract ERa-dependent cell proliferation and tumor formation [21,22,23]. The first reported ERb-selective estrogen synthesized and studied was diarylpropionitrile (DPN). DPN has a 70-fold higher in vitro binding affinity and 170-fold higher potency in transcription assays with ERb compared to ERa [24]. Other ERbselective ligands have been synthesized in both academic and industrial settings, of which ERB-041 is among the most studied [7,25]. In addition to synthetic ERb ligands, a plant extract, MF101 [26] and a flavanone derived from a single plant in MF101, liquiritigenin [27] are highly ERb-selective compounds.
Studies with ERb-selective compounds indicate that there are at least three classes of ERb-selective agonists. ERB-041 is the prototype of a ligand that is an ERb-selective binder, because it binds to ERb with a much higher affinity than ERa. In contrast, we showed that MF101 and liquiritigenin bind similarly to ERa and ERb, but do not regulate gene transcription in the presence of ERa or stimulate uterine growth or breast cancer tumor formation in mouse models [27]. These studies established that some ligands can act as highly ERb-selective transcriptional activators, even though they bind nonselectively to both ERa and ERb. A third class of ERb-selective agonists is represented by DPN, which is selective by a combination of preferential binding to ERb and increased transcriptional activity [24]. An unanswered question is whether different ERb-selective agonists produce biological effects that are distinct from each other and non-selective ER agonists used in HT, such as estradiol. To investigate this issue, we determined if these ERb-selective compounds regulate the same or different genes.

Cell lines and culture
Tetracycline-inducible U2OS-ERa and U2OS-ERb cells were characterized and maintained as previously described [30]. U2OS, Caco-2, HeLa, and Ishikawa cells were obtained from the UCSF cell culture facility and maintained as previously described [28,31]. All experiments were done with cells containing 5% charcoalstripped fetal bovine serum.

Fö rster resonance energy transfer (FRET)
U2OS cells (n = 500,000) were plated into six-well dish containing a borosilicate glass coverslip and grown in phenol red-free DMEM/F12 media supplemented with 5% charcoalstripped fetal bovine serum and 2 mM glutamine. The following day the cells were transfected with 500 ng/well of CFP-ERa-YFP [32] or CFP-ERb-YFP [26] using Lipofectamine TM 2000 according to manufacturer's protocol (Invitrogen, Carlsbad, CA). After 6 h the medium was replaced with complete medium containing 10% stripped fetal bovine serum, 2 mM glutamine, 50 U/ml penicillin, 50 mg/ml streptomycin and the cells were incubated overnight. One day after transfection cells were treated with the indicated amounts of ligand for 30 minutes before image collection. Within each independent experiment, an average of 124 cells were collected for each ligand at each concentration and the amount of FRET averaged by comparing the amounts of fluorescence in the acceptor bleedthrough corrected FRET channel to the amount in the Donor channel; the conversion of these values to the percentage of Energy transferred from CFP to YFP was done using the calibration methods we have previously described [33]. For each ligand, the dose response curves were conducted twice on independent days and presented at the mean+/2range (Figure 1, open bars). Measurements at 1 mM of ligand were repeated on four independent days and presented as the mean+/2standard deviation (Figure 1, closed bars). In total, FRET measurements were collected from 35,396 cells expressing CFP-ERa-YFP or CFP-ERb-YFP and from an additional 4,432 control cells expressing ERa or ERb attached to CFP or YFP alone.

Microarrays
U2OS-ERa and U2OS-ERb cells were maintained in 5% charcoal-stripped fetal bovine serum and plated in 6-well plates. When the cells reached 80% confluent, they were treated with 1 mg/ml doxycycline for 12 h to induce ERs. The cells were then treated with 10 nM E 2 , 125 mg/ml MF101, or 1 mM liquiritigenin, nyasol or DPN for 6 h at 37 C. Total cellular RNA was isolated with the Aurum RNA isolation kit (Bio-Rad, Hercules, CA) per the manufacturer's protocol. RNA was first quantified by standard spectrophotometry, and then qualitatively evaluated by capillary electrophoresis employing the Bio-Rad Experion system (Hercules, CA). Biotin-labeled cRNA samples were prepared with 750 ng of total RNA template. Following synthesis and purification, the biotin-labeled samples were evaluated by both 260/280 absorbance spectrophotometry and capillary electrophoresis. The final labeled cRNA samples were hybridized overnight against Human genome HG U133A-2.0 GeneChip arrays containing more than 22,200 probe sets (Affymetrix, Santa Clara, CA) or 48,000 transcripts HumanWG-6 BeadChip (Illumina, San Diego, CA) arrays. For the U133A-2.0 GeneChips the array hybridizations, washing, staining, as well as scanning were performed by the J.D. Gladstone Genomics Core, (San Francisco, CA), whereas the Ilumina microarrays were processed at the UCSF Genomics Core. The drug studies were done with the U133A-2.0 GeneChips and the four cell type study was done with WG-6 BeadChips. Same batch of microarrays were used for all treatments and most treatments were done in triplicate except for NYA treatment in U2OS-ERa samples in 2 replicates, and E2, MF101, and LIQ treatment in U2OS-ERb samples in four replicates.

Microarray data analysis
The Affymetrix expression arrays were pre-processed using a variant of GCRMA [34]. The microarrays were preprocessed with a procedure similar to GCRMA, except that the background adjustment step is modified. Instead of using the probe sequence to predict non-specific binding (as in GCRMA), the non-specific binding for each probe is estimated from a database composed of hybridization data on the same platform of microarrays used in a variety of experiments. The new procedure is therefore dubbed dbRMA. Background parameters were estimated for each probe separately in dbRMA and avoided borrowing information across probes sharing similar but not identical sequences. More specifically, the probe intensity across all the samples in the database was modeled as a mixture distribution with the first component as background and estimated using normal approximation. Assessment on calibration data (Affymetrix Latin Square experiment) showed better accuracy of background parameters compared to those predicted by sequence. The normalization and summarization steps in the preprocessing procedures remain the same as GCRMA. The details of dbRMA procedure will be presented in a separate manuscript.
The Illumina expression arrays were pre-processed using lumi package [35]. The differential expression analysis was performed using limma package [36]. These packages are all available in R/ BioConductor. For drug screen data, probesets were selected for further analysis if the fold change was greater than 2 and multiple testing adjusted p-value using Benjamini and Hochberg procedure (BH-adjusted p-value) was less than 0.05 [37]. For the three cell line data, fold change threshold 1.5 was used. The heatmaps of log intensities of genes across different experiments were produced using Cluster and TreeView software [38]. Cluster software was used to perform the hierarchical clustering based on Pearson correlation coefficients (PCC) to find clusters of genes with similar expression patterns. TreeView was then used to visualize the clusters and produce the figures.

Functional enrichment analysis of target genes
To elucidate the biological processes of target genes, we searched enriched GO annotations using GOstat software [39]. For each annotated GO term, GOstat counted the number of overlapping genes from the input gene list, and compared it with the one expected from a reference list (GO annotation human (http://www.ebi.ac.uk/GOA/human_release.html). Fisher's exact test was performed to compute a p-value for each GO category and BH-adjusted p-values were calculated. Results for significant GO ''biological process'' categories were reported. To compare the enriched GO terms cross different experiments, the scores {log 10 of BH-adjusted p-values for each GO term were combined into one table with GO terms shown in rows and different experiments shown in columns. Cluster and TreeView software [38] were then used to produce the GO charts.

RNA extraction and quantitative real-time PCR
Caco-2, HeLa and Ishikawa cells were infected with an adenovirus (100 MOI) that expresses ERb [21]. After 20 h, the cells were treated for 6 h with MF101 or LIQ. Total RNA was extracted with Aurum total RNA mini kit and cDNA synthesis was performed with the iScript cDNA synthesis kit (Bio-Rad, Hercules, CA). Real-time PCR analysis was performed in duplicates using iQ SYBR Green Mix with an iCycler thermal cycler (Bio-Rad, Hercules, CA). U2OS-ERa and U2OS-ERb were treated with 1 mg/ml doxycycline for 12 h to induce ERs. The cells were then treated for increasing times with the drugs and real-time PCR was done using primers for keratin 19 (K19), A kinase (PRKA) anchor protein 1 (AKAP1), interleukin 17 receptor B (IL17RB). The sequences of primers used are listed in Table S1.

ERb-selective compounds produce conformational changes in both ERa and ERb
One goal of this study was to compare the relative ERb-selectivity of three classes of ERb agonists and to determine if they produce similar effects on gene expression to each other and E 2 . The structures of the compounds are shown in Figure S1. ERB-041 is an ERb-selective binder because it binds 200-fold greater to ERb than ERa [40]. MF101, liquiritigenin and nyasol are ERb-selective activators, because they bind similarly to ERa and ERb, but activate genes only with ERb [26,27]. DPN is a combined ERb-selective binder and activator because of greater binding to ERb and transcriptional activity with ERb [24]. For comparison, we chose to study the effects of these drugs on gene expression at saturating concentrations of the compounds. FRET was used to determine the concentration required for saturation of the ligands to ERa and ERb. The amount of FRET between CFP and YFP attached on opposite termini of each ER was shown to be a measure of ligand binding in intact cells [26,32,41]. U2OS cells were transfected with CFP-ERa-YFP or CFP-ERb-YFP [26,32] and then treated with the compounds. All of the compounds produced a dose-dependent enhancement of FRET with both ERa and ERb when added to the cell culture medium at concentrations ranging from 0.3 nM to 3 mM (data not shown). The maximal amount of energy transfer at saturating amounts of ligand is shown for ERa ( Figure 1A, open bars) or ERb ( Figure 1B, open bars) and is compared to the amounts of energy transfer detected at the 1 mM concentration (closed bars). All compounds produced equivalent amounts of energy transfer, above the no ligand controls, with both ERa and ERb when provided at saturating levels. Note that the large error bars for ERB-041 at ERa ( Figure 1A, open bars) reflects the variations in the extrapolation of the dose-response because maximal energy transfer was not achieved at 3 mM ERB-041 (the highest concentration used). Thus at 1 mM, all compounds except ERB-041 were able to saturate both ERa and ERb. Similarly 125 mg/ml of the crude MF101 extract was sufficient to activate both ERa ( Figure 1C) and ERb ( Figure 1D). We previously showed that 1 mM liquiritigenin (LIQ) and 125 mg/ml MF101 was the concentration that maximally activated reporter genes [26,27]. Furthermore, 1 mM of nyasol (NYA), ERB-041 and DPN produced a maximal activation of ERE-tkLuc with ERb in transfection assays ( Figure S2). Based on the transfection and FRET studies, 1 mM of each compound and 125 mg/ml of MF101 extract was used for the subsequent studies to establish the ER subtypeselectivity of each compound.

MF101, liquiritigenin and ERB-041 are the most ERbselective compounds
To investigate the ERb-selectivity of synthetic and natural compounds, we used the previously characterized human U2OS cells that are stably transfected with a doxycycline-inducible expression vector for ERa or ERb [30]. After the cells were treated with doxycycline to induce ERs, they were treated with E 2 and the plant-derived ERb-agonists, MF101, NYA and LIQ, and the synthetic ERb-agonists, DPN [24] and ERB-041 [40]. We previously showed that MF101 is a selective ERb agonist despite being a complex, crude plant extract [26]. LIQ was isolated from Glycyrrhizae uralensis Fisch and is ERb-selective [27]. NYA is a diphenylpentane norlignan that was purified from the plant Anemarrhena asphodeloides in MF101 and has ERb-selectivity using transfection assays (data not shown). For each compound we defined a regulated gene to be activated by 2.0-fold or greater or repressed by 50% or greater and statistically different from the untreated control cells (BH-adjusted p-value,0.05). The regulated genes and magnitude of regulation in U2OS-ERa and U2OS-ERb cells by each drug are found in Table S2. The heatmaps show the genes that are significantly regulated by the drugs compared to the control cells. The compounds produced a distinct pattern of regulated genes in the U2OS-ERa (Figure 2A) cells compared to U2OS-ERb cells ( Figure 2B). The non-ER selective agonist E 2 , which was used as a positive control, regulated 489 specific genes in the U2OS-ERa cells relative to the control cells (Table 1). In the U2OS-ERa cells, there were a total of 238 genes regulated by DPN and 152 genes regulated by nyasol. The Gene Ontology (GO) analysis showed that the major classes of genes commonly regulated in U2OS-ERa cells by E 2 , nyasol and DPN were involved in anatomical structure development, multicellular organismal process and developmental process ( Figure S3). ERB-041 regulated 2 genes in the ERa cells, whereas LIQ and MF101 weakly regulated (between 2-3 fold) 3 and 16 genes in the ERa cells, respectively. These results demonstrate that relative to E 2 , only DPN and NYA showed ERa activity. In contrast, all the drugs regulated about 400 genes in the U2OS-ERb cells ( Table 1). The heatmap shows that overall the genes regulated by the ERb agonists were similar to each other and to E 2 ( Figure 2B). By comparing the results in the U2OS-ERa and U2OS-ERb cells the most ERb-selective agonists at saturating levels were ERB-041, LIQ and MF101 followed by NYA, and then DPN. To investigate that the possibility that the different genes regulated by ERa and ERb were related to the 6 hour treatment time, we performed time-courses on three regulated genes ( Figure 3). In the U2OS-ERa cells, E 2 and DPN maximally activated AKAP1 ( Figure 3A), IL-17 ( Figure 3C), and K19 ( Figure 3E) at 6 hour. No regulation was observed with other drugs at all time points. In contrast, all the drugs activated AKAP1 ( Figure 3B), IL-17 ( Figure 3D), and K19 ( Figure 3F) in the U2OS-ERb cells. The maximal activation of AKAP1 and IL-17 occurred at 6 hours, whereas K19 was maximally activated by the drugs at 12 h. All of drugs produced the maximal activation of these three genes at the same time-point in both U2OS-ERa and U2OS-ERb cells. These findings indicate that the differences in regulation by drugs in the microarrays were not due to the selection of the 6 hour time-point.

ERb-selective compounds regulate some different genes in U2OS-ERb cells
Further analysis of the microarray data was done to determine if the three classes of ERb-selective agonists regulate different genes in the U2OS-ERb cells. Overall most of genes were commonly regulated with the ERb-selective compounds ( Table 2). The list of the regulated genes by each compound is found in Table S2. However, some genes were uniquely regulated by the ERb-selective  Table 2). The ERb-selective agonists regulated more genes in common with each other compared to E 2 in the U2OS-ERb cells. The greatest difference in commonly regulated genes occurred with MF101 and E 2 , whereas LIQ and DPN showed no difference in the gene expression profiles. Some genes regulated by E 2 in the ERb cells were also different from those regulated by the ERb-selective compounds. We performed GO analysis to determine what classes of genes were regulated similarly and differently by the ERb agonists. Most of the classes of genes were regulated similarly, such as developmental process, multicellular organismal development, system development, organ development, biological regulation, and negative regulation of cellular process ( Figure S4). However, some classes of genes were differentially regulated by the ERb-selective drugs and E 2 (Figure 4). For example, E 2 uniquely regulated RNA metabolic process genes, whereas NYA regulated embryonic development genes, MF101 regulated gland development genes, LIQ regulated extracellular structure organization genes and biogenesis, and DPN regulated the regulation of phosphorylation genes (Figure 4). The magnitude of regulation by the drugs of several differentially regulated genes is shown in Figure 5. For comparison, the COL gene was regulated similarly by all the drugs ( Figure 5A). The highest activation of the GPER gene was observed with MF101 and NYA ( Figure 5B), and E 2 , LIQ, DPN and ERB-041 for the SOX9 gene ( Figure 5C). The ID1 was repressed the most with E 2 , MF101, NYA and ERB-041 ( Figure 5D). These results demonstrate that while most of the genes are commonly regulated there are some differences in class of genes regulated and the magnitude of regulation by the different drugs, which might be important in producing biological effects.

Cell type-specific regulation of genes with ERb-selective ligands
To examine whether the ERb-selective ligands regulate genes in a cell-specific manner, Caco-2, HeLa and Ishikawa cells were infected with an adenovirus that expresses ERb. These three cell lines did not express ERa or ERb ( Figure 6). The expression of ERb after the cells were infected with Ad-ERb was similar in the three cell lines. For microarray analysis, we chose to focus on MF101 and LIQ, because this allowed us to evaluate if the effects of a crude extract were similar to a single active compound. The cells were treated for 6 h with MF101 or LIQ and the gene expression profiles were determined. Surprisingly, there was very little overlap in the regulated genes in the three cell lines (Table 3). Only 3 genes were commonly regulated by MF101 and no genes were commonly regulated by LIQ in the three cell types. Because only a few genes were commonly regulated by MF101 and LIQ in three cell lines, we compared the number of genes commonly regulated by these drugs in two cell lines ( Table 4). The most overlap with MF101 treatment occurred in the Caco-2 and HeLa cells with 17 genes commonly regulated. The list of the regulated genes by MF101 or LIQ in three cell lines is found in Table S3. The GO analysis showed that not only do MF101 (Figure 7) and LIQ ( Figure S5) regulate different genes, but also that the regulated genes are involved with different biological processes. These data demonstrate that there is a remarkable cell-type specificity of genes regulated by two of the ERb-selective agonists. We used real-time PCR to examine the regulation by MF101 or LIQ in the three cell lines infected with Ad-ERb. MF101 or LIQ increased mRNA levels for ADAMTS-like 5 (ADAMTSL5), protein tyrosine phosphatase, receptor type, E (PTPRE), retinoic acid receptor, alpha (RARA), and transglutaminase 2 (TGM2) genes in HeLa cells ( Figure 8A), hydroxysteroid (11-beta) dehydrogenase 2 (HSD), ectodysplasin-A receptor (EDAR), chromosome 3 open reading frame 59 (C3orf59) and OTU domain, ubiquitin aldehyde binding 2 (OTUB2) in Ishikawa cells ( Figure 8B), cytochrome P450, family 1, subfamily A, polypeptide 1, (CYP1A1), cytochrome P450, family 1, subfamily B, polypeptide 1 (CYP1B1), baculoviral IAP repeat-containing 3 (BIRC3) and fibroblast growth factor binding protein 1 (FGFBP1) in Caco-2 cells ( Figure 8C). These results confirm the regulation observed in the microarrays.

Discussion
The biological effects of estrogens are mediated by ERa and ERb. All the current estrogens approved for hormone therapy non-selectively bind to and regulate both ERs. ERa has an important role in preventing osteoporosis, because males with a defective ERa develop severe osteoporosis and the increased bone turnover is not reversed by high-dose estrogen treatment [42]. However, the activation of ERa by estrogens also causes the proliferation of cells, which increases the risk of breast and endometrial cancer [18]. The pro-proliferative properties of non-ER selective estrogens has prevented their use in non-hysterectomized women, and caused an intense effort to discover more selective estrogens. Drugs that selectively activate ERb are a particularly attractive alternative for HT, because ERb acts as a tumor suppressor that inhibits the growth of breast cancer cells [21,22,23]. The lack of proliferative effects of ERb were also demonstrated by the observations ERB-041 did not exhibit any proliferative effects on the mammary glands and uterus of rats [40], and MF101 and LIQ did not stimulate uterine growth or breast cancer tumor formation in a mouse xenograft model [26,27]. Whereas these results indicate that ERb-selective agonists will not elicit the same proliferative effects as the non-selective estrogens, it is unclear if they will be beneficial for treating menopausal symptoms or osteoporosis.
Some ERb-selective compounds did not show any benefits on hot flashes in rat models indicating that ERb-selective agonists might not be effective for this classical indication for HT [43]. In contrast, DPN reduced hot flashes as measured by a reversal of the elevation in of basal tail skin temperature that occurs after ovariectomy [44]. The ERb-selective agonist MF101 showed a statistically significant reduction in hot flashes in a phase 2 randomized placebo controlled study [45]. One possible explanation for these findings is that different classes of ERb-selective agonists might regulate distinct genes and thereby produce different biological effects. To examine this possibility, we compared the ERb-selectivity of synthetic and plant-derived ERb-selective agonists in U20S cells that express ERa or ERb using microarrays to study their selectivity over a broad range of ER target genes. We found that ERB-041, LIQ and MF101 were the most ERb-selective, followed by NYA, and DPN. The precise mechanism for the ERb-selectivity of the compounds is unclear. ERB-041 is considered to be an ERbselective agonist because it binds to ERb with about a 200-fold higher affinity compared to ERa [40]. DPN has a 70-fold higher affinity to ERb, whereas LIQ bound to ERb with a 20-fold higher affinity [27]. MF101 and NYA bound to ERa and ERb with a similar affinity [26]. All of these binding studies used in vitro competition binding assays. To explore the relative binding of the compounds in living cells, we performed FRET studies in U2OS cells. Our FRET studies showing that ERB-041 was the only compound that did not produce any conformational change in ERa at 1 mM demonstrated that ERB-041 is a selective ERb binder. In contrast, conformational changes in ERa and ERb were induced at similar concentrations with MF101, LIQ, NYA and DPN, demonstrating that these compounds can bind to both ERa and ERb. However, the gene expression data showed that even though they bound similar to ERa and ERb at 1 mM, these compounds regulated genes selectively with ERb at this concentration. These results indicate that the conformation of ERa induced by MF101 and LIQ is essentially inactive, whereas the conformation induced by NYA and DPN was weakly active. It is clear that at saturating levels the ERb-selectivity of these compounds is not related to differential binding to ERb, but results from events that occur after ligand binding. We previously showed that MF101 and LIQ did not recruit coactivators to ERa [26,27], suggesting that compound-bound ERa was in a conformation that was incapable of binding coactivators. Our FRET data shows that the conformations produced by all ERbselective agonists were similar despite that they showed different patterns of gene regulation. The FRET study measures the position of YFP relative to CFP, which appeared to be very similar when ERb is bound with the different compounds. It is likely that FRET is not sensitive enough to detect subtle changes in conformation that led to the differences in gene expression profiles with the compounds.
One of the most interesting findings of our study is that some genes regulated by the ERb-selective compounds were not regulated by E 2 in the U2OS-ERb cells. The number of genes differentially regulated by the ERb agonists compared to E 2 , range from 31 with DPN to 168 with MF101. These results demonstrate that the ERb-selective compounds do not entirely mimic the action of E 2 after binding to ERb, suggesting that they might elicit different biological effects than E 2 . While there was no difference in FRET with E 2 and the ERb-selective compounds it is likely that subtle differences in conformation not detectable by FRET might lead to a differential recruitment of coregulatory proteins and ultimately different genes regulated. This issue is difficult to address experimentally because the regulatory elements in the genes that are differentially regulated by E 2 and ERb-selective agonists as observed with the microarrays are not known.
Our study also demonstrated that two ERb-selective compounds regulated different genes in the three cell lines. Although the cells were exposed to the same amount of Ad-ERb, concentration of drugs, and time of drug treatment there was very little overlap in the regulated genes in the these cell lines. Unexpectedly, only 3 genes were commonly regulated in all cell types. The reason for the cell-specific regulation is unknown. It has been proposed that the differential expression of coactivators in different cell types might be responsible for cell-specific regulation [46,47]. Our microarrays showed similar expression of SRC-1, SRC-2 and SRC-3 in the three cell lines (data not shown). These findings indicate that the differential expression of these three classes of coactivators is an unlikely explanation for the different pattern of gene regulation in the cell lines. Genome-wide tiling arrays demonstrate that ER binding sites are associated with different transcription factors that are important for gene activation [48,49,50]. We also showed that the activation of the NKG2E gene requires multiple transcription factors (32). These findings suggest that differential expression of transcription factors in the cells might lead to the differences in gene regulation. Another explanation is that there are different epigenetic changes in the regulated genes in each cell type that allow the recruitment of cell specific transcription factors as shown with FOXA1 [51]. It is also possible that the drugs are differentially metabolized in the three cells. If the metabolites are active this might account for some of the differences in the genes regulated.
Our study shows several important features of ERb-selective agonists that could have important clinical ramifications. First, although most of the genes regulated by the three different classes of ERb-selective agonists were the same, there were some classes of genes that were differentially regulated and the magnitude of regulation of some regulated genes differed. These findings suggest that different ERb-selective drugs might exert distinct clinical effects and that it can not be assumed that if one drug fails or succeeds in clinical trials that other ERb-selective drugs will behave similarly. Second, the ERb-selective agonists regulate different genes than E 2 . These findings suggest that ERb-selective agonists will have a different side-effect profile than currently hormone therapy regimens. Although the effect of the ERbselective compounds on thromboembolic events is unknown, their benign effect on the uterus and mammary gland in preclinical models is a potentially differentiating factor from the non-selective estrogens. Our hypothesis that different classes of ERb-selective agonists will produce distinct biological effects needs to be tested in clinical trials with postmenopausal women.     whereas the lightest gray implies the corresponding GO term is not significantly enriched. GO terms significantly enriched in at least three conditions are shown. Found at: doi:10.1371/journal.pone.0006271.s004 (0.55 MB TIF) Figure S5 GO charts for genes regulated by LIQ in HeLa, Caco-2 or Ishikawa cells. Analysis of biological processes enriched among genes regulated by LIQ in HeLa, Caco-2 or Ishi cells. Gene ontology terms significantly enriched in genes regulated by LIQ in each of the fours cell lines are shown. A threshold 0.001 was used for selecting GO terms using BHadjusted p-values. (p-value) was used as an enrichment score. Darker shading denotes more significantly enriched GO terms, whereas the lightest gray implies the corresponding GO term is not significantly enriched.