|Home | About | Journals | Submit | Contact Us | Français|
Ligand structure can affect the activation of nuclear receptors, such as estrogen receptors (ERs), and their control of signaling pathways for cellular responses including death and differentiation. We hypothesized that distinct biological functions of similar estradiol (E2) analogs could be identified by integrating gene expression patterns obtained from human tumor cell lines with receptor binding and functional data for the purpose of developing compounds for treatment of a variety of diseases. We compared the estrogen receptor subtype selectivity and impact on signaling pathways for three distinct, but structurally similar, analogs of E2. Modifications in the core structure of E2 led to pronounced changes in subtype selectivity for estrogen receptors, ER-α or ER-β, along with varying degrees of ER dimerization and activation. While all three E2 analogs are predominantly ER-β agonists, the cell growth inhibitory activity commonly associated with this class of compounds was detected for only two of the analogs and might be explained by a ligand-specific pattern of gene transcription. Microarray studies using three different human tumor cell lines demonstrated that the analogs distinctly affect the transcription of genes in signaling pathways for chromosome replication, cell death, and oligodendrocyte progenitor cell differentiation. That the E2 analogs could lower tumor cell viability and stimulate neuronal differentiation confirmed that gene expression data could accurately distinguish biological activity of the E2 analogs. The findings reported here confirm that cellular responses can be regulated by making key structural alterations to the core structure of endogenous ER ligands.
Nuclear receptors are transcription factors that play a key role in both embryonic development (Chung & Cooney 2003) and adult homeostasis (Strauss et al. 2009, Colasanti et al. 2011, Ribas et al. 2011), and are a common target for drug developers treating human diseases such as cardiovascular, cancer, and autoimmune disorders (Deroo & Korach 2006). Drugs in the market that modulate the activity of nuclear receptors represent over $30 billion in pharmaceutical sales (Moore et al. 2006, Via 2010). Such drugs include prednisolone and dexamethasone for modulating glucocorticoid receptors (GRs), diethylstilbestrol and 4-hydroxytamoxifen for modulating estrogen receptors (ERs), and spironolactone (Aldactone) and eplerenone (Inspra) for modulating mineralocorticoid receptors (Moore et al. 2006). Interestingly, drugs that modulate the control of the same nuclear receptors are used to treat a diverse set of diseases, an effect that is attributed to the role of nuclear receptors in regulating many distinct intracellular pathways at the gene transcription level (Moore et al. 2006).
ERs are members of the nuclear receptor steroid family and have long been targets for therapeutic development, including the selective ER modulators (SERMs). The most prominent and best understood forms of the estrogen receptor are ER-α and ER-β, which share significant sequence homology (Barkhem et al. 1998), but display tissue-specific gene expression patterns (Couse et al. 1997), which can determine their use as effective therapeutic targets. In the case where ER-α and ER-β are coexpressed in the same tissues, they can form active homo- or heterodimers upon binding ER ligands (Deroo & Korach 2006). While several ER isoforms, such as ER-β2, ER-β3, ER-β4, and ER-β5 are known to exist, the biological significance of isoforms remains to be determined, although they may dimerize with ERs and enhance ER transactivation in a ligand-dependent manner (Paulmurugan et al. 2011). Estrogens and SERMs are known to bind and activate both ER-α and ER-β, making them useful for treating diseases, such as breast cancer, osteoporosis, and menopausal symptoms (Moore et al. 2006).
Binding of structurally dissimilar ER ligands can affect ER conformation along with the composition of the ER transcription complexes (Brzozowski et al. 1997). These phenomena may depend on the size and shape of the ligand-binding domains (LBDs), which vary between subtypes. The LBD for ER-α is substantially larger than for ER-β, 450 vs 390Å3 respectively (Ruff et al. 2000). This difference likely impacts the ER ligand-binding affinity, orientation of the ER ligand in the ligand-binding pocket, conformation of the ligand-bound ER, ER dimerization, coactivator and corepressor binding, and function of the receptors in their transcription complexes (Ruff et al. 2000). Helix 12 of both ER-α and ER-β has been shown to be essential for transactivation of gene expression, as loss of function mutations within this region results in a nonfunctional receptor even in the presence of ligand (Danielian et al. 1992). Notably, the position of helix 12 in the ERs is thought to be dependent on the particular bound ER ligand, and the helix 12 position affects the receptor function. For example, when 17β-estradiol (E2) is bound within the ligand-binding pocket of ER-α, helix 12 is oriented over the ligand-binding cavity and is packed against helixes 3, 5/6, and 11 in an agonist conformation (Brzozowski et al. 1997). By contrast, the binding of raloxifene within the ligand-binding pocket orients helix 12 to lie in a groove formed by helix 5 and the carboxy-terminal end of helix 3 in an antagonist conformation (Brzozowski et al. 1997).
Binding of different ligands to nuclear receptors, such as the ERs, can alter their conformation (Brzozowski et al. 1997, Shiau et al. 1998, Connor et al. 2001, Moore et al. 2006) resulting in ligand-dependent gene transcription profiles (Frasor et al. 2004, Paruthiyil et al. 2009). Studies with ER ligands have confirmed that slight changes in stereochemistry or modifications to small substituents can alter subtype selectivity and shift receptor-binding potency by orders of magnitude (Katzenellenbogen 2011). Accordingly, a plethora of ER ligands have been synthesized that are built around a diversity of core structures including polycyclic cores, acyclic cores, macrocyclic cores, and heterocyclic cores (Katzenellenbogen 2011). However, analogs designed for specificity to a particular ER subtype have not always aligned with the desired biological activity.
In this study, we used gene expression data obtained from three different human tumor cell lines as a predictor of biological function for closely related E2 analogs that were designed in-house. We employed ER-binding and functional assays to further examine the attributes of each ligand, along with cell-based assays to confirm the effect that specific ligand structures impart on the fate of cells. Beginning with the core E2 structure, analogs were devised with new substituents at the C-6 of ring B or the C-18 of ring D. For the three E2 analogs studied here, we show that subtle structural changes in the core E2 can affect both ER selectivity and gene expression pathways responsible for cell growth and differentiation. The potential of the three E2 analogs studied here for treating human diseases, such as multiple sclerosis and cancer, supports the viability of this approach for drug discovery.
Details for synthesis of the E2 analogs (Fig. 1) are found in Supplementary Figure 1A, B and C, see section on supplementary data given at the end of this article. Briefly, NDC-1022 was prepared starting from E2 (Supplementary Figure 1A) and was obtained as a white solid with a purity of 98%. As the starting material was E2, care was taken in column chromatography to ensure that <10ppm E2 was present in the final material used in the assays for these studies. The chiral purity of the 17β-isomer was at least 95%. The compound was kept stored at under −18°C. NDC-1308 was prepared according to the scheme in Supplementary Figure 1B. Starting from E2, NDC-1308 was obtained as a solid, white foam with a purity of 95% and a chiral purity of 96%. Since the starting material was E2, care was taken in column chromatography to ensure that <10ppm E2 was present in the final material. The compound was kept stored at under −18°C. NDC-1352 was prepared starting from NDC-1022 according to the scheme in Supplementary Figure 1C. NDC-1352 was obtained as a white solid with a purity of 95% starting with NDC-1022. The compound was kept stored at under −18°C.
For the biological assays, E2 and the E2 analogs were first dissolved in DMSO to a stock concentration of 10–100mM and then serially diluted to the appropriate concentrations.
Cell lines used for the cytotoxicity assay include NIH-H23, CAPAN-1, CAPAN-2, MDA-MB-231, MCF-7, and OVCAR-3 obtained from ATCC (Manassas, VA, USA) or SK-OV-3, A-549 obtained from the NCBI (Bethesda, MD, USA). Cell lines were grown in RPMI (SK-OV-3, NCI-H23, MDA-MB231, A-549, MCF-7), RPMI with 10mM HEPES/20%FBS (OVCAR-3), Iscove's Modified Dulbecco's Media (CAPAN-2), or McCoy's 5a Modified Media (CAPAN-1).
Modeling studies were performed with ICM-Pro v.3.6-1e licensed from Molsoft, LLC (San Diego, CA, USA). The coordinates for ER-α and ER-β in various conformations were obtained from the Protein Data Bank (PDB, www.rcsb.org). The evaluation of several different coordinate files allowed us to explore protein flexibility. Generation of protein–ligand complexes, using different receptor conformations, was achieved using the Docking module in ICM-Pro. Evaluation of fitness of docks, and effects of the C-18 methyl group, was done by first generating docked complexes between the ligand of interest and ER-α (derived from PDB entry 1KQT) and ER-β (derived from PDB entry 2J7X). These docks were then energy minimized to convergence using the Molecular Mechanics function using the ‘boundary element’ with a dielectric constant=8 for the electrostatic term. In this minimization, the ligand and protein residues within 5Å of the ligand were free to move while the rest of the protein was fixed. The energetics between the protein and the ligand of the minimized complex was then evaluated using the ‘show energy’ function using the ‘vw, sf, el and en’ terms.
Recombinant ER-α and ER-β were supplied by PanVera (Madison, WI, USA). The 3H-E2 was obtained from New England Nuclear (Boston, MA, USA). The binding assay was carried out by incubating the recombinant ER-α or ER-β with appropriate amount of 3H-E2 in the presence or absence of various concentrations of test compounds at 4°C overnight in 200μl binding buffer with pH 7.4 (10mM Tris–HCl, 10% glycerol, 1% albumin, 1mM phenylmethylsulfonyl fluoride, 25nM leupeptin). Various concentrations of nonlabeled E2 were tested in the same assay. Nonspecific binding was determined using a concentration of the nonradiolabeled E2 at 100 times that of the 3H-E2 in the same assay tube. At the end of incubation, the bound 3H-E2 was separated from the free unbounded 3H-E2 using 60% hydroxyapatite (HAP) in binding buffer. An aliquot of the supernatant was mixed with 4ml ScintiSafe (Fisher Scientific, Hampton, NH, USA) and the radioactivity was measured on a scintillation counter. The relative binding affinity of the test article was the amount of nonradioactive E2 required to displace 50% of radioactive E2 bound to ER relative to the amount of nonradioactive test compound required to displace 50% of radioactive E2 bound to ER×100.
Bioluminescent resonance energy transfer (BRET) assays were performed as previously described (Powell & Xu 2008). Briefly, HEK293 cells were transfected with pCMX-ERα-RLuc or pCMX-RLuc-ERβ BRET fusion constructs in combination with empty vector, pCMX-ERα-YFP and pCMX-YFP-ERβ. The following cotransfections were used to assess dimer formation: pCMX-ERα-RLuc/pCMX-ERα-YFP for ER-α homodimers, pCMX-ERα-RLuc/pCMX-YFP-ERβ for heterodimers, and pCMX-RLuc-ERβ/pCMX-YFP-ERβ for ER-β homodimers. Twenty-four hours after transfection, ~50000 cells/well suspended in PBS were seeded in triplicate onto a white 96-well plate and treated with the corresponding compounds or vehicle (0.6% DMSO). After an 1h incubation with ligand, coelenterazine h (Promega) was added to a final concentration of 5μM, and emission at 460 and 530nm was detected using a Victor X5 microplate reader. BRET ratios were calculated as described previously (Powell & Xu 2008), and the ratios represent the energy transfer between the donor and acceptor molecules. Higher BRET ratios are indicative of dimerization.
The plasmids for ER-binding element luciferase reporter (pERE) and for ER-β cDNA expression vector (pER-β) were supplied by Addgene (Cambridge, MA, USA). The plasmid for the ER-α cDNA expression vector (pER-α) was obtained from OriGene Technologies, Inc. (Rockville, MD, USA). The Dual-Luciferase kit for measuring luminescence was obtained from Promega. MDA-MB-231 cells were seeded in 12-well plates at a density of 1.5×105 cells/well in medium without antibiotics. Cells were allowed to grow overnight to 50–70% confluence before transfection. For transfection, an appropriate amount of pERE, pER-α or pER-β reporter plasmid and internal control plasmid pRL-TK was cotransfected transiently into cells. Transfection was performed using TurboFectin8 (OriGene) according to the protocol provided by the supplier. Cells were then treated with either E2 or the E2 analogs. After incubating for 24h at 37°C, cells were washed once with PBS, and then lysed with buffer provided by the supplier. The ratio of firefly compared with renilla luminescence was calculated for each treatment group. The functional expression of each compound was expressed as a percentage of the E2 ratio (100%) measured at 10nM.
Three human tumor cell lines, A-549, Panc-1, and SK-OV-3, were selected based on their endogenous levels of ER-α and ER-β and each grown in two flasks cultured to roughly 50% confluence. Cells were treated for 24h with either the E2 analogs or 10% DMSO at concentrations of compound specified in Table 1. At termination, the cells were scraped free and washed in ice-cold PBS, collected by centrifugation, and immediately frozen at −80°C. Total RNA was prepared from the frozen tissue samples using Trizol-based cell lysis followed by 65°C hot phenol extraction and RNeasy chromatography purification (Qiagen). The concentration of RNA was determined by measuring the absorbance at 260nm (A260). RNA purity was assessed by confirming that extracted RNAs had an A260/A280 ratio of 1.8 or greater. The RNA was then tested for relative integrity by determining the ratio of intact 28S and 18S rRNAs by capillary electrophoresis with the Agilent 2100 BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA). All RNAs accepted for array analysis had ratios exceeding 1.0.
RNAs were labeled using 1μg RNA as input to Low Input labeling reaction (Agilent Technologies) with Cy5 (650nm emitter) and reference RNA was labeled with Cy3 (550nm emitter) nucleotides. Labeling, hybridizations, and subsequent washings were carried out on with the Human GE 4x44K v2 G4110B (NDC-1308, NDC-1022) or the Human GE 4x44K v2 G4845A (NDC-1352) microarray kits from Agilent Technologies, according to the manufacturer's instructions. The resulting hybridized chips were scanned (G2505 Scanner, Agilent Technologies), and the intensity information for each detector spot was extracted from the scanned image using Agilent feature extraction software version 10.5.1.1. The filtered raw data for each microarray spot can be accessed in Supplementary Table 1, see section on supplementary data given at the end of this article. For analysis, a threshold of P≤0.001 was used as the cutoff point for significant change in mRNA abundance between the compound-treated and vehicle-treated samples.
NDC-1022, NDC-1308, NDC-1352, E2 (Sigma), and tamoxifen (Sigma) were each dissolved in DMSO at 100mM. For dilutions, the 100mM DMSO stock was added to the appropriate media and serially diluted in the same media such that the final DMSO contribution was ≤1% (v/v).
Each cell line was maintained at 5% CO2, 37°C, and 95% relative humidity in the appropriate media including all necessary supplements. The cells were subcultured every 2–3 days and plated in tissue culture-treated, white/clear-bottom 96-well plates and incubated at 5% CO2 and 37°C overnight before initiation of the assay. The media in the cell plate (100μl) was replaced with fresh media (100μl) immediately before adding the 100μl test articles; the media were not estrogen-depleted. Wells containing no cells were used for background controls. Tamoxifen was used as a known positive control for each assay and DMSO was included as the vehicle control. The cells were incubated at ~37°C in humidified 5% CO2 atmosphere for 72h.
Cell number was determined by measuring the ATP levels using the CellTiter-Glo Kit (Promega). Luminescence was measured with the Infinite M200 plate reader (Tecan, Durham, NC, USA). The mean baseline controls (wells with no cells) were subtracted from the total luminescence to determine the net luminescence for that well. This total was compared with the vehicle control of DMSO only. Both EC50 and EC90 values were calculated using Sigma Plot 11.0 (Systat Software, Inc., San Jose, CA, USA). The lower EC90/50 values help to identify cell lines that are more susceptible to compounds.
Oligodendrocyte precursor cell (OPC) cultures were prepared as previously described (Pedraza et al. 2008). Briefly, brains were removed from E14.5 C57Bl6/J (expressing PLP-EGFP) mice (Mallon et al. 2002), cleaned, and cortical hemispheres were isolated. Tissue was then triturated and seeded into T-25cm2 flasks at a density of one brain (two cortical hemispheres) per flask. Neurospheres were passed once after 3 days. Cells from passage two were used to prepare two 96-well plates for the primary screen. Following the initial 48h incubation in OPC media, the OPCs were treated with 10μM of E2, NDC-1022, NDC-1308, and NDC-1352 along with vehicle control DMSO and positive control compounds including 20ng/ml ciliary neurotrophic factor (CNTF) and 1μM ERK kinase inhibitor (MEKi). Cells were treated for a total of 4 days and the media was replaced once at the 48h mark with fresh compounds. After the 4 day treatment, the cells were fixed with 4% paraformaldehyde and stained with Hoechst 33342 to visualize nuclei. Cells were also stained with anti-GFAP antibodies to identify astrocytes. Briefly, cells were blocked with 3% normal goat serum followed by an overnight incubation of anti-rabbit GFAP antibody (1:500). Cells were labeled with a 1:1000 concentration of secondary goat anti-rabbit Alexa 647 fluorochrome antibody. Images were acquired using the Cellomics Arrayscan VTI. Twenty fields at 10× magnification were acquired per well and the cells expressing EGFP (mature oligodendrocytes) and GFAP (astrocytes) were evaluated by neuronal profiling algorithm.
For the BRET assays, the values for each compound were compared with DMSO control for each dimer condition using a one-tailed Student's t-test with significant difference P<0.05. For the DNA microarray studies, significant (P<0.001) up- or downregulated genes were identified by comparing the E2 analog-treated and vehicle-treated samples from each tumor cell line. Genes were included in the relevant signaling pathways (Table 1) only if they were consistently regulated in the same manner for all three tumor cell lines. For the oligodendrocyte differentiation studies, ANOVA showed statistically significant differences (P<0.001) between the different treatment groups. Comparisons of E2 analogs, MEKi, and CNTF were then made to control DMSO using a Student's t-test with significant difference P<0.001.
The structures of E2 and the synthesized E2 analogs are shown in Fig. 1. Alkoxyalkyl substituents were added to the C-6 position of the E2 B-ring for all three analogs while the C-18 methyl was removed to create NDC-1352 (Supplementary Figure 1A, B and C).
Docking studies of these compounds to various receptor conformations, allowing for the exploration of protein flexibility of both ER-α and ER-β, were carried out using the docking algorithm implemented in ICMpro and generated two possible docking poses. Using NDC-1022 as an example, in one pose (Fig. 2A and B) the steroid core would bind in the same manner that E2 does (E2-like). In the second pose (Fig. 2C) the E2 analogs bind in an ‘upside down’ manner similar to the way both 16-α, E2 (DOI: 10.2210/pdb2j7y/pdb) (Pike et al. 2001) and ICI 164384 (DOI: 10.2210/pdb1hj1/pdb) bind. These two different binding modes represent isomeric complexes of a single compound, differing in both the relative position of the ligand in the binding site and the conformation of the protein. It should be noted that for ER-β, the binding affinity decreases with ligand volume, whereas for ER-α, binding affinity increases with ligand volume, which is consistent with this model.
We propose the E2-like docking pose as the most likely pose for the E2 analogs studied here. In the E2-like orientation for NDC-1308 (Fig. 2D), the C-6 alkoxyalkyl groups of the E2 analogs could fold on top of the steroid core of the ligand, binding to a pocket of the ERs that E2 would not be able to interact. The C-6 groups in this orientation could fit into a small hydrophobic pocket formed by the side chain of Met-291, Leu-294, and Met-295 of ER-β (corresponding to Leu-384, Leu-387, and Met-388 of ER-α) although this appears to be a rather crowded fit. We predict that this crowded fit results in subtle conformational changes in the ER leading to altered transcriptional activities.
In the E2-like bound orientation of NDC-1022, the C-6 side chain could make van der Waals interactions with the C-18 methyl. In NDC-1352, the C-18 methyl is absent. To simplify the modeling analysis, the effect of the C-18 methyl was examined by comparing the binding of E2 with nor-C-18-methyl-E2 (norMe_E2). For ER-β, norMe_E2 had slightly more favorable van der Waals interactions than did E2. For ER-α, the situation was reversed in that norMe_E2 had slightly less favorable van der Waals interactions than did E2. It should be noted that for ER-α, the C-18 methyl group of E2 makes van der Waals interactions with the side chain of Leu-384, whereas with norMe_E2 with a hydrogen at this position, a small gap between ligand and protein is created. For ER-β, the C-18 methyl group of E2 interacts with the side chain Met-291. As the methionine side chain is somewhat flexible, it is better able to fill in the gap between the C-18 hydrogen in norMe_E2 and the ER-β protein. Similarly, the somewhat flexible nature of Met-291 in ER-β might allow for a better induced fit of the C-6 side chain of NDC-1352 within ER-β compared with the less flexible Leu-384 of ER-α. Thus, the substitution of the C-18 methyl group of NDC-1022 for a hydrogen in NDC-1352 is expected to result in an increased β-selectivity of NDC-1352. The E2-like binding pose represents the agonist conformation of the ER and provides a better proposed fit for the E2 analogs compared with the ‘upside down’ binding pose (Fig. 2C).
Each E2 analog along with reference E2 was assessed for binding affinity to ER-α and ER-β using in vitro recombinant receptor binding assays. The analogs were found to possess distinctly different binding affinities for ER-α and ER-β (Table 2). Although it has a large alkoxyalkyl moiety at C-6, the 1/β:α ratio for NDC-1308 shows that there is no significant preference for binding either ER-α (IC50 70nM) or ER-β (IC50 100nM), similar to E2. On the other hand, NDC-1022 and NDC-1352 each have a smaller alkoxyalkyl moiety at C-6 and are structurally more similar to E2 than NDC-1308, yet both preferentially bind ER-β over ER-α. NDC-1022 had a greater than sixfold higher binding affinity to ER-β than ER-α, while NDC-1352 had a >100-fold higher affinity to ER-β. The different ER-β/ER-α binding ratios for NDC-1022 compared with NDC-1352 demonstrate the impact that a seemingly innocuous hydrogen substitution (NDC-1352) for a methyl group (NDC-1022) at C-18 can make on receptor binding in combination with an alkoxyalkyl moiety at C-6.
While some ligands have been found to cross-react with different classes of hormone receptors (Kimbrel & McDonnell 2003), there was no significant binding of either NDC-1308 or NDC-1022 to the androgen (AR), progesterone (PR), or GR receptors. The IC50s for DHT (AR control), progesterone (PR control), and dexamethasone (GR control) were 8.5, 1.3, and 6.6nM respectively. By contrast, the IC50 of NDC-1022 was >3000nM for all three receptors, while the IC50 of NDC-1308 was >3000nM for AR and GR and 1200nM for PR. NDC-1352 was not evaluated in this assay.
The binding of exogenous ligands to ER-α and ER-β and subsequent dimer formation is an initial step required for transcriptional activation. In cells expressing both ER-α and ER-β, homo- and heterodimers can readily form following the binding of estrogens (Deroo & Korach 2006). NDC-1022 and NDC-1308 were tested for their ability to cause homo- and heterodimer formation in vitro using a BRET assay that measures the ERα/β heterodimerization in a cell-based physiological environment in real time (Pfleger & Eidne 2006, Powell & Xu 2008). Although NDC-1022 caused all combinations of homo- and heterodimers to form at concentrations of 1μM or greater, α/β heterodimers and β/β homodimers were preferred at 100nM (Fig. 3A). By contrast, NDC-1308 induces all dimer formations starting at 1μM but with no preference for either form (Fig. 3B). BRET assays for ER-β homodimerization typically show 1.5- to 2-fold inductions with E2 treatment because of high ligand-independent dimerization, which has been previously described by Powell & Xu (2008). In addition, conformational changes within the receptor fusion proteins allow for efficient energy transfer, and the BRET ratios are dependent on the conformational changes induced by different ligands, thereby affecting the BRET ratio output. Although NDC-1022 and NDC-1308 do not induce BRET ratios to the same extent as E2, there is significant ER dimerization induced by both compounds. Since ERs are transcription factors, the ability of these E2 analogs to form ER dimers in vitro suggested that they would drive gene expression in cells.
Compounds NDC-1022, NDC-1308, and NDC-1352 were assayed for transcriptional activity from both ER subtypes, and compared to E2 for reference. Reporter gene transfection assays were conducted in MDA-MB-231 cells using expression plasmids for either full-length ER-α or ER-β and an estrogen-responsive luciferase reporter gene system.
Overall, the E2 analogs were more potent agonists of ER-β than ER-α (refer to β:α functional selectivity in Table 2), but less potent ER activators compared with E2. Functional selectivity was dose-dependent for NDC-1022 and NDC-1308 as both favored ER-β over ER-α at 10 and 100nM. The selectivity was not observed at, or above, 1μM of ligand (Table 2). Treatment of cells with NDC-1352 caused 15- to 20-fold more transcription from the reporter gene for ER-β than ER-α, even up to the highest concentration of ligand tested (10μM, data not shown), which is consistent with the results from the binding affinity assay.
In a separate follow-up study, the addition of an ER antagonist, ICI 182780 (100nM), to the culture media of the cells for the luciferase assay effectively competed away the transcriptional activity of the reporter gene by NDC-1022 and NDC-1308 at all concentrations, for both ER-α and ER-β (Supplementary Figure 2, see section on supplementary data given at the end of this article). This demonstrates that the E2 analogs are indeed functioning through the ERs.
As the gene activity measured by the luciferase functional assays are associated with only one estrogen response element (ERE) and nuclear receptors influence many genes and pathways from various EREs, the global transcriptional effect of the E2 analogs was assessed by DNA microarray. Gene expression profiles were determined by microarray in three different human tumor cell lines, which vary in their endogenous amounts of ER-α and ER-β. A-549 is an ER-α negative, ER-β positive human lung adenocarcinoma cell line (Dougherty et al. 2006). Panc-1 is an ER-α positive, ER-β positive pancreatic tumor cell line (Mariani 2005). SK-OV-3 is an ER-α positive, ER-β positive ovarian tumor cell line; however, there is a 32-bp deletion in ER-α rendering it nonfunctional (Lau et al. 1999). Cells were treated with varying amounts of each E2 analog at concentrations ranging from 10 to 100μM in order to ensure complete dimer formation (Fig. 3) and full activation of the ERs (Table 2).
Table 1 summarizes the microarray dataset for differentially expressed genes found in key signaling pathways. Each gene listed in Table 1 was selected due to its common change in expression across all three tumor cell lines following treatment with each compound. The values in the table represent the mean expression (log2) across all three cell lines. For NDC-1022 and NDC-1308, all three cell lines showed on average a i) 2- to 75-fold upregulation of genes in signaling pathways related to OPC differentiation, ii) one- to fivefold downregulation of genes for signaling pathways related to chromosome replication, and iii) 1- to 25-fold upregulation of genes in signaling pathways controlling cell death. For both analogs, the response of the cells at the transcriptional level was dose-dependent, yet the potency of NDC-1308 was significantly greater than NDC-1022. This is best demonstrated by comparing the magnitude of gene expression changes measured at 10μM for NDC-1308 to 100μM for NDC-1022 (Table 1). By contrast, genes within these signaling pathways were relatively unaffected by even the highest concentration of NDC-1352 (100μM). Instead, with NDC-1352, chemokine genes possibly related to the immune system pathways or the pro-inflammatory pathways were upregulated (3- to 200-fold). As expected, housekeeping genes were not significantly up-regulated or uniformly downregulated by the E2 analogs.
As the changes in gene expression lead to distinct signaling pathways for each E2 analog, we reasoned that the pathway information derived from the tumor cell lines could be used to further distinguish the similar E2 analogs from each other in cell-based assays for viability (tumor cell death) and differentiation (neuronal precursor assays).
The gene expression profiles (Table 1) predict that NDC-1308 would be the most potent of the E2 analogs for inducing cell death. To test this, ten different human tumor cell lines obtained from six different tissue types were treated separately, with each of the three E2 analogs compared with tamoxifen and E2 as a reference (Table 3). On average, NDC-1308 (EC50 18.0±2.2) was the most potent of the E2 analogs followed by NDC-1022 (EC50 57.1±5.5) and NDC-1352 (EC50 117.1±11.5). Treating cancer cells with NDC-1308 showed a dose-dependent ability to kill tumor cell lines with an EC50 ranging from 7 to 31μM, which is similar to the potency of tamoxifen and about twofold more potent than E2 (18–52μM). NDC-1022 also possessed a dose-dependent cytotoxic activity with an EC50 ranging from 33 to 85μM, but as predicted from the microarray data (Table 3), was about threefold less potent than NDC-1308. Consistent with the microarray data, the potency of NDC-1352 (EC50 50–150μM) was about twofold less than NDC-1022, although the chemical structures are very similar (Fig. 1). The EC90 value was on average two- to tenfold lower for NDC-1308 compared with either E2 or the other E2 analogs, confirming that the structural changes made in the E2 core affect cell viability.
To test whether the E2 analogs directly induce cell death via the ERs, we used NDC-1352 as a competitive inhibitor for blocking the tumor cell death mediated by NDC-1308. We chose NDC-1352 because it binds almost exclusively to ER-β (Table 2), but is five- to tenfold less potent than NDC-1308 in the tumor cell viability assays (Table 3). To simplify the potential intracellular interactions of having both ERs present, we used the A549 tumor cell line as it endogenously expresses only a functional ER-β. We found that addition of 5μM NDC-1352 significantly raised the IC50 (1.4-fold) and IC90 (1.9-fold) of NDC-1308. This means that tumor cell death induced by NDC-1308 is mediated through ER-β.
The gene expression data (Table 1) show that NDC-1022 and NDC-1308 upregulate genes in signaling pathways involved in myelin production and the differentiation of OPCs. To directly test for this activity, mouse OPCs were treated for 96h with 10μM of either NDC-1022, NDC-1308, or NDC-1352 and the extent of oligodendrocyte maturation measured by the level of the PLP-EGFP reporter signal (Fig. 4). Proteolipid protein (PLP) was used as a biomarker, because it is known to be expressed in mature oligodendrocytes and is a component of the myelin sheath. As shown in Fig. 4A, NDC-1308 was the most potent E2 analog having threefold more activity to differentiate OPCs compared with DMSO and 1.5- to 2.0-fold more activity than the positive controls, CNTF (Stankoff et al. 2002) and MEKi (Younes-Rapozo et al. 2009) respectively. The number of pyknotic nuclei for the E2 analog-treated cells (~20% of total) was the same as the control DMSO- and CNTF-treated cells, indicating that the OPCs were not adversely affected by the drug treatment. As predicted from the microarray data, NDC-1352 did not cause significant OPC differentiation at 10μM.
Previous reports have described a number of diverse ER ligands without adequately assigning a biological function to different substituents present in the chemical structures (Frasor et al. 2004, Paulmurugan et al. 2011). Towards this end, we integrated gene expression data derived from human tumor cell lines along with receptor binding and cell-based functional assays to ascertain the effect that a change in ER ligand structure imparts on ER function. The three E2 analogs studied here were designed with different substituents added or removed from the E2 core. While all three analogs are ER-β agonists, the structural modifications present in each compound led to unique ER dimerization, different subtype selectivity, and receptor-binding affinities, along with differential effects on gene expression, cell growth, and cell differentiation.
Structure–function relationships with modifications at the C-6 position (B-ring) of the E2 core have not previously been reported. However, our ER modeling data (Fig. 2) suggested that an alkoxyalkyl group addition containing at least four carbons at the C-6 position would fold over the B-ring of the E2 core. This could still allow the analogs to bind in the E2 orientation, but perhaps bind more compactly in the LBD of the receptor compared with E2. This provided the rationale to synthesize NDC-1308 with a six carbon alkoxyalkyl group at the C-6 position of E2 (Fig. 1, B-ring). As a comparator, NDC-1022 was constructed with a single carbon alkoxyalkyl group added to the E2 core. An additional modification was made at the C-18 (Fig. 1, D-ring) of NDC-1022 by changing the methyl group to a hydrogen atom. The resulting ligand, NDC-1352, is predicted to be primarily oriented in the LBD by virtue of the C-3 and C-17 alcohol groups in the E2 core.
While we originally viewed some of the structural differences in the three E2 analogs to be relatively innocuous compared with E2 and also with each other, it was surprising that each compound had vastly different binding affinity and ER subtype selectivity. NDC-1308 lacked a preferential affinity for either ER-α or ER-β, while NDC-1022 preferentially bound ER-β (greater than fivefold). Remarkably, NDC-1352 bound almost exclusively to ER-β (>100-fold). For NDC-1308, the impact of the longer alkoxyalkyl side chain (six carbons) fitting more compactly in the LBD may explain the loss of ER-β selectivity for this E2 ligand compared with the other two E2 analogs. The striking gain in ER-β selectivity for NDC-1352 and the different gene expression patterns compared with NDC-1022 are solely attributed to the hydrogen substitution for the methyl at C-18 of the D-ring. The competitive binding assays showed that E2 had at least tenfold better binding to ER-β and 35-fold better binding to ER-α than the best E2 analog, probably due to the overall bulkiness of the alkoxyalkyl side chains added to the C-6 position of each E2 analog.
Moreover, a plausible explanation for why NDC-1308 is more potent and less selective than NDC-1022 is because the longer side chain of NDC-1308 makes additional van der Waals interactions to stabilize it within the ligand-binding cleft of both receptors. The enhanced ER-β selectivity of NDC-1022 may be due to the C-6 side chain making van der Waals contacts with residues that are different between ER-α (Leu-384) and ER-β (Met-291). Likewise, the remarkable ER-β selectivity of NDC-1352 may be explained by the loss of the C-18 methyl from the core structure and the interaction it normally makes with these same residues.
ERs bind ligand and dimerize to become part of an active transcription complex. We evaluated the potential of NDC-1022 and NDC-1308 to form ER-α and ER-β homo- and heterodimers in vitro as a predictor of their ability to activate and regulate transcription in cell lines. Compounds NDC-1022 and NDC-1308 act similar to other estrogens by causing the formation of ER homo- and heterodimers (Deroo & Korach 2006). The results of the dimer formation assays showed NDC-1308 to be more promiscuous than NDC-1022 for ERs, consistent with the receptor binding data. That the E2 analogs could specifically activate transcription via ERs was confirmed by functional assays showing that each functions as ER agonists, with no detectable cross-reactivity for the ARs, PRs, and GRs.
Based on their different ER subtype selectivity and knowing that ER activation is not an ‘on-off’ switch, we anticipated that the E2 analogs would elicit different cellular responses. We used microarray data from human tumor cell lines to distinguish the structure of each ligand for key intracellular signaling pathways and ultimately to predict biological responses. Choosing tumor cell lines derived from diverse tissues (lung, pancreas, and ovary) ensured that common signaling pathways would be detected independent of genetic background or levels of ER expression. This strategy also filters away significant gene expression changes that are due to cell line-specific coactivators and corepressors. In addition, the human tumor cell lines were treated with micromolar concentrations of each E2 analog to ensure full dimerization (Fig. 3) and activation (Table 2) of the ERs within the 24h incubation period. We were able to identify common signaling pathways in all three human tumor cell lines for each E2 analog.
Identifying the dramatic regulation of key genes within several specific signaling pathways helped us predict the biological response for each E2 analogs. For example, the dramatic upregulation of MBP, OLIG2, and MOG at the lowest dose of NDC-1308 (10μM) predicted an activity for differentiation of OPCs. This was confirmed in vitro using a mouse OPC differentiation assay (Fig. 4) where NDC-1308 had the most potent activity (EC50 1.0μM, data not shown). Conversely, at higher concentrations both NDC-1308 (50μM) and NDC-1022 (100μM) caused dramatic downregulation of MCM4, MCM10, GTSE1, SPC25, and other genes essential for chromosome replication. Moreover, NDC-1308 and NDC-1022 cause a substantial upregulation of genes for cell death including GDF15, GZMK, and GADD45A. This predicted cytotoxic activity at higher concentrations of the E2 analogs, which was confirmed in the cell-based viability assay for both NDC-1308 and NDC-1022 (Table 3). That NDC-1352 is less cytotoxic and does not cause OPC differentiation is consistent with the gene expression data, and points out the dramatic effect that a hydrogen substitution for a methyl group can make at C-18 of the core E2. NDC-1352 does elicit a dramatic upregulation of chemokine, metallopeptidases, and a complement component suggesting a potential role in stimulating inflammation. Notably, this specific transcriptional upregulation is absent in NDC-1308 and NDC-1022.
As described above, the altered transcriptional activities of these E2 analogs could be attributed to conformational changes in the ER resulting from adding the different C-6 alkoxyalkyl side chains. Likewise, removing the C-18 methyl group of the E2 core affected ER activation. Except for a hydrogen atom substituted in place of a methyl at C-18, NDC-1352 is structurally identical to NDC-1022. The modeling data suggest that the absence of the C-18 methyl group of NDC-1352 allows for a better fit of the C-6 side chain in ER-β than in ER-α. This could explain why remarkably different gene expression differences were obtained in key signaling pathways for NDC-1352 compared with NDC-1022. Taken together, the comparison of gene expression and functional data for NDC-1352 and NDC-1022 demonstrates how a critical alteration in the core E2 structure can dramatically alter the cellular response.
Our data confirm and extend the notion that the inherent structure of specific ER ligands has a major impact on gene expression events that control cell fate. Structural modifications of E2 analogs at the C-6 and C-18 positions led to dramatic differences in the activation of the ERs that ultimately led to tumor cell death or neuronal cell differentiation. The use of ER competitors in transcriptional assays and the cell viability assay is consistent with the activity of the E2 analogs being directed through the ERs. Furthermore, the activities reported here only require the presence of functional ER-β as two (A-549 and SK-OV-3) of the three tumor cell lines used for the microarray study do not make a functional ER-α protein. These studies support the notion that by using this approach for drug discovery, ligands for nuclear receptors may be designed for treating a variety of diseases, such as multiple sclerosis and cancer.
This is linked to the online version of the paper at http://dx.doi.org/10.1530/JME-12-0083.
The authors thank Brian Bai and Caroline Lego of Renovo Neural for assistance with the OPC cultures and oligodendrocyte maturation assays, Andrea Hubbel of Ricerca Biosciences for assistance with the cytotoxicity assays, and Wan-Ru Chao of SRI International for assistance with the receptor-binding and functional assays. They also thank Mark Verhaar, Erik Keller, and Martin Poelert of Syncom for the synthesis of the compounds used in this study. Finally, they thank Bruce Trapp for critically reviewing the manuscript.
The authors declare that R E B, M B, E S, W X, and P H have nothing to disclose that could be perceived as prejudicing the impartiality of the research reported. J G Y and S H N are shareholders of ENDECE, LLC.
This research did not receive any specific grant from any funding agency in the public, commercial, or not-for-profit sector.