A Structure-based Approach to Retinoid X Receptor-α Inhibition*

In this paper we describe a structure-based approach designed to identify novel ligands for retinoid X receptor-α (RXRα). By using a virtual approach based on a modified scoring function, we have selected 200 potential candidates on the basis of their predicted ability of docking into the ligand-binding site of the target. Subsequent experimental verification of the compounds in in vitro and cell-based assays led to the identification of a number of novel high affinity ligands for RXRα. The compounds are capable of displacing 9-cis-retinoic acid with IC50 values in the 10 nm and 5 μm range and exhibit marked antagonistic activity in cellular assays. The inhibitory scaffolds discovered with this method form the basis for the development of novel RXRα ligands with potential therapeutic properties.

In this paper we describe a structure-based approach designed to identify novel ligands for retinoid X receptor-␣ (RXR␣). By using a virtual approach based on a modified scoring function, we have selected 200 potential candidates on the basis of their predicted ability of docking into the ligand-binding site of the target. Subsequent experimental verification of the compounds in in vitro and cell-based assays led to the identification of a number of novel high affinity ligands for RXR␣. The compounds are capable of displacing 9-cis-retinoic acid with IC 50 values in the 10 nM and 5 M range and exhibit marked antagonistic activity in cellular assays. The inhibitory scaffolds discovered with this method form the basis for the development of novel RXR␣ ligands with potential therapeutic properties.
Retinoid X receptor-␣ (RXR␣) 3 is a member of the nuclear receptor superfamily, which mediates the biological effects of many hormones, vitamins, and drugs (1)(2)(3)(4)(5). A unique property of RXR␣ is its exceptional capacity to form heterodimers with other nuclear receptors, including retinoic acid receptors (RARs), thyroid hormone receptor, vitamin D receptor, and peroxisome proliferator-activated receptor (1)(2)(3)(4)(5). In addition, RXR␣ can homodimerize in response to its ligands (1)(2)(3)(4)(5). Heterodimerization of RXR␣ with its partners can mediate diverse endocrine signaling pathways that when altered can lead to the development of cancer (6). Genetic disruption of RXR␣ targeted to the prostatic epithelium results in intraepithelial neoplasia in mice (7), whereas diminished RXR␣ protein expression may represent an early event in the development of human cancer (8). Like other nuclear receptors, RXR␣ acts as a transcriptional factor to positively or negatively regulate expression of target genes (1)(2)(3)(4)(5). On binding ligands, RXR␣ undergoes conformational changes to recruit transcriptional corepressors or coactivators, leading to suppression or activation of transcriptional program on target gene promoters (3,4). Recent progress indicates that RXR␣ also exerts certain nongenotropic actions, including its mitochondrial targeting to induce apoptosis (9 -11) and its interaction with ␤-catenin to inhibit the Wnt/␤-catenin signaling (12). RXR␣ was found to cotranslocate with NGFI-B (also known as Nur77 or TR3) from the nucleus to the cytoplasm in response to nerve growth factor treatment (13), a process implicated in the differentiation of PC12 pheochromocytoma cells. In response to apoptotic stimuli, RXR␣ and Nur77 associate with mitochondria as a Nur77/RXR␣ heterodimer in LNCaP prostate cancer and H460 lung cancer cells (10). RXR␣ also cotranslocates with Nur77 from the nucleus to mitochondria in response to IGFBP-3 (insulin-like growth factor-binding protein-3) (9), presumably its interaction with IGFBP-3.
A number of natural and synthetic molecules with different structural features and diverse biological effects have been identified as ligands for RXR␣. 9-cis-Retinoic acid (9-cis-RA) was the first compound known to bind RXR␣. Recently, several dietary fatty acids were found to bind RXR␣ and appear to act as natural RXR␣ ligands (14 -16). A nonsteroidal anti-inflammatory drug, (R)-etodolac, also binds RXR␣ and acts as an RXR␣ antagonist to inhibit its transactivation, an event which is associated with its tumor-selective induction of apoptosis in animal (11). The ability of RXR␣ to bind various ligands that have diverse biological effects suggests that RXR␣ acts as an important intracellular mediator regulating multiple signal transduction pathways. Such a complexity of RXR␣ signalings calls for the need to identify additional compounds that selectively regulate a specific RXR␣ signal transduction pathway. Such molecules could be useful for their therapeutic properties but also will facilitate mechanistic studies on the role of RXR␣ in signal transduction. In this study, we undertook a structurebased approach for the identification of new RXR␣ ligands. By using a virtual docking approach based on the three-dimensional structure of RXR␣, we identified a number of potential RXR␣ ligands. Following experimental in vitro and in cell-based assays with top scoring compounds, we have finally identified ligands with nanomicromolar binding affinity to RXR␣ ligand binding domain (LBD) and selectivity when tested against RAR␣.

EXPERIMENTAL PROCEDURES
Virtual Docking-A target binding site was derived from the crystal structure of the ternary complex involving BMS-649 compound in complex with (h)RXR␣-LBD (Protein Data Bank code 1MVC) (17). The protein active site was defined, including those residues within 6.5 Å from the benzoic acid substructure of BMS-649 found in the complex. Hydrogen atoms were calculated using FlexX (BioSolveIT, Sankt Augustin, Germany) (18,19), and water molecules were eliminated. Docking geometries were obtained by using FlexX (five solutions per molecule) implemented on a 20 ϫ 3.2-GHz CPUs Linux cluster. Different parameters of the scoring functions were varied (Table 1), and a collection of random 1026 compounds that included the BMS-649 compound was used to estimate the ranking ability of each function. 50,000 compounds (ChemBridge, San Diego, CA) were subsequently docked and ranked according to the software FlexX using DJ score. Top 200 ranking compounds were selected (Chembridge, San Diego) and experimentally tested in the in vitro displacement assay. Hit compounds * This work was supported in part by National Institutes of Health Grants CA113318 and CA109345 and DoD Grant W81 XWH-04-1-0161. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 These authors contributed equally to this work. 2  were repurchased and tested in additional NMR binding and cell-based assays. When the six validated hits were included in the collection of 1025 compounds, however, the average ranking of the compounds was only slightly better when using DJ score compared with FlexX, although both functions performed much better than Screenscore and Chemscore, with this particular set.
Ligand Binding Assays-His-tagged (h)RXR␣-LBD (amino acids 223-462) was expressed in Escherichia coli and purified as a polyhistidine-tagged fusion protein for use in competition binding assays (20,21). Briefly, (h)RXR␣-LBD (100 ng) was incubated in binding buffer (0.15 M KCl in 10 mM Tris-HCl, pH 7.4, containing 0.5% CHAPS detergent (Roche Diagnostics) and 8% glycerol; 300 liters) with 7.5 nM 9-cis-[ 3 H]RA (44 Ci/mmol; PerkinElmer Life Sciences) in the absence or presence of increasing concentrations of nonlabeled compound for 16 -18 h at 4°C. Next, His-Bind resin (500 g; Novagen) was added, and incubation with shaking was continued for 1 h at room temperature. The His-Bind resin was washed (three times with 1 ml of binding buffer) to separate receptor-bound from nonbound label, suspended (500 ml of buffer), and then transferred for scintillation counting (3.5 ml of EcoLume liquid scintillation fluid, ICN; Beckman Coulter LS 3801 counter). Experiments were performed in duplicate, and specific binding was calculated as the average of the percentage of the total bound  indicates the root mean square deviation between all heavy atoms of the crystal structure and docked structures generated by the different scoring functions. j Ranking of the crystal structure of the BMS-649 compound when included in a data base of randomly selected 1025 compounds. The ranking was obtained by using the following free energy of binding:

Structures of hit compounds and their in vitro characterization
counts/min remaining defined as counts/min/(total bound counts/ min) ϫ 100. NMR Experiments-All experiments were carried out at 298 K by using a 600-or a 500-MHz Bruker Avance spectrometer, equipped with four and three rf channels and pulse-field gradients along the z axis, respectively. NMR samples were consisted of 100 M of compound in presence or absence of 10 M of protein, in 40 mM D 2 O phosphate buffer, pH 7.5.
A spin-lock pulse of variable length (1, 10, 100, and 200 ms) was used for the acquisition of one-dimensional 1 H T 1 series (22,23). Each spec-trum was acquired with 256 scans. Water suppression was achieved by means of a WATERGATE pulse scheme. Spectra were processed and analyzed with the software Mestrec-C.
RXRa Transactivation Assays-CV-1 cells were maintained in Dulbecco's minimal essential medium supplemented with 10% fetal calf serum, 100 units/ml penicillin, and 100 g/ml streptomycin. For transfection assays, cells were seeded at 1.0 ϫ 10 5 cells/ml in 24-well plates for 16 -24 h before transfection. Cells were then transfected using FuGENE 6 Transfection Reagent (Roche Diagnostics) with (TREpal) 2tk-CAT with either RXRa or RAR␣ (25 ng). In addition, cells were also transfected with ␤-galactosidase expression vector (pCH 110, Amersham Biosciences) and carrier DNA (pBluescript, Stratagene) to a final concentration of 1000 mg/well. At 20 h after transfection, the medium was changed to Dulbecco's minimal essential medium containing charcoal-stripped fetal calf serum, and cells were treated for 24 h with either 10 nM 9-cis-RA for RXRa or 10 nM all-trans-RA for RARa or with varying concentrations of a compound of interest. CAT activity was expressed relative to ␤-galactosidase activity to normalize for transfection efficiency (1,2). Reactions were performed in triplicate and the results averaged.
Western Blotting-Cells were harvested and lysed in lysis buffer (50 mM Tris-HCl, pH 8.0, and 150 mM NaCl, with 0.1% Triton X-100, 0.25% sodium deoxycholate, 1 mM EDTA, 1 mM phenylmethanesulfonyl fluoride, 1 g/ml aprotinin, 1 g/ml leupeptin, and 1 mM sodium orthovanadate; Sigma). Equivalent protein extracts from each sample were separated on 8% SDS-polyacrylamide gels. Protein was quantitated by a total protein assay (Bio-Rad). Proteins were transferred onto nitrocellulose membranes (Trans-Blot, Bio-Rad). Nitrocellulose membranes were preblocked with 5% nonfat milk powder in phosphate-buffered saline containing 0.05% Tween 20 detergent for 1 h at room temperature. Following phosphate-buffered saline/Tween washes, preblocked membranes were incubated with 1 g/ml equivalent of anti-RAR␤ polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA). RAR␤ protein was detected by horseradish peroxidase-conjugated secondary antibody against immunoglobulins (Amersham Biosciences) after 1 h of incubation at room temperature, and specific bands were visualized by ECL (Amersham Biosciences). Equivalent loading of samples was deter-    Table 2. A-F, report ligands 1-6. In each panel, intermolecular hydrogen bondings between a given docked inhibitor and the ligand binding domain of RXR␣ are highlighted.

RESULTS AND DISCUSSION
In order to identify novel RXR␣ ligands, we adopted a structurebased approach based on virtual docking of a large library of drug-like small organic molecules followed by experimental verification of top scoring compounds. This approach is based on first obtaining reliable docking geometries for each compound and subsequently being able to rank order the compounds for their predicted binding affinity according to a scoring function.
Although most of the available virtual docking software packages have demonstrated their utility in several different cases (25)(26)(27), in practical terms, it has been shown that their success and failures are target-dependent (25). In addition, our recent experience with ranking compounds according to a given scoring function is not too encouraging (28). Although a consensus score between several functions is a possible solution, we found that intermolecular hydrogen bonding and visual inspection of the docked geometries significantly increase the success rate of the ranked compounds (28). The formation of hydrogen bonds with Arg-316 is a very important feature that is common to virtually all the reported RXR␣ ligands as well as its natural substrate, 9-cis-RA (29). For these reasons, we tried to optimize the parameters for the scoring functions in a way that would emphasize the importance of hydrogen bonding interactions. By using in-house data from high throughput screening (30) and smaller scale screening projects (28,31,32), we scaled the parameters of the empirical terms of Chemscore (33) that produced the most reliable ranking ( Table 1). As internal control we used the x-ray-derived docked structure of the compound BMS-649 in complex with RXR␣ (Protein Data Bank code 1MVC) (17). In the Screening score (34) and Chemscore (33), the authors decreased the factors of lipophilic and "ambiguous interactions" to increase the contribution of "matching interactions." In addition, Eldridge et al. (33) reported on a modified FlexX score (similar to Chemscore) that was obtained by changing the parameterizations of the entropy term and the definition of hydrophobic groups (35). Here we afford another small improvement on this work (Table 1) in order to produce docked geometries and ranking that better reflect the importance of hydrogen bond formation in RXR␣. By using the BMS-649 as test ligand, we could verify that all the different scoring functions produced a docked conformation that is remarkably close to that in the crystallographic coordinates (Table 1 and Fig. 1). However, when the compound was inserted in a data base of random 1025 compounds, BMS-649 ranked first only when using our modified scoring function, which we named DJ score (Table  1). These data gave us the confidence that this approach would lead to reliable docked geometries and that high affinity ligands could be found among the top ranking compounds. Therefore, by using this strategy we first obtained the docked geometries of each of the 50,000 compounds from the entire Chembridge DiverseSet collection (Chembridge, San Diego) and subsequently examined experimentally a subset of compounds constituted by the top ranking 200 molecules.
To experimentally verify which of the selected top scoring compounds effectively binds RXR␣, competitive ligand binding studies using recombinant histidine-tagged human (h)RXR␣-LBD were conducted in duplicate, with compounds initially tested at 10 M concentration. Dose-response curves were subsequently obtained for those compounds that reproducibly displaced more than 50% of 9-cis-[ 3 H]RA at 10 M concentration. Our results show that 6 of 200 top scoring compounds were able to compete the binding of 9-cis-[ 3 H]RA with relative binding affinities, as measured by IC 50 values, ranging from 10 nM to 5 M (Table 2). A representative binding curve along with the data for the 9-cis-RA control is shown in Fig. 2. In order to exclude possible nonspecific binding of the compounds to the reference ligand, additional binding assays of the compounds to (h)RXR␣-LBD were performed by acquiring one-dimensional 1 H NMR T 1 series for each compound (at 100 M concentration) (22,23) in the absence and presence of a substoichiometric amount of protein (10 M). However, because 100 M concentration is a lower detection limit for these experiments with our current instrumentation, we could perform this NMR binding assay only on compounds that are soluble at this concentration (Table 2). Representative T 1 experiments (spin lock duration equal to 200 ms) are reported in Fig. 3. As can be seen, the T 1 spectrum of compound 6 exhibits an obvious decrease in signal intensity when compared with the same spectrum measured in the presence of (h)RXR␣-LBD (Fig. 3A), thus indicating that the compound interacts with the target. Compounds 4 and 5 proved to be stronger RXR␣ ligands according to such an assay as indicated by the significant reduction of the signal intensity that could be observed in the T 1 spectra (Fig. 3, B and C). Thus, for the compounds that could be tested by NMR, there is a close parallel between the in vitro displacement assay and the direct interaction of the ligands with (h)RXR␣-LBD ( Table 2).
The docked structures of these compounds are reported in Fig. 4, in which the structure of each is shown and superimposed to the x-ray coordinates of BMS-649 as reference. Hydrogen bonding between each compound and Arg-316 and the backbone amide of Ala-327 are also highlighted. Interestingly, these compounds lack the typical carboxylate group involved in hydrogen bonding and salt-bridge interactions with the residues mentioned above, which is characteristic of most RXR␣ ligands reported thus far (29). Instead, these interactions seem to involve the barbituric acid ring in compounds 1 and 3-5 and a nitrogroup in compounds 2 and 6, making these structures novel RXR␣binding motifs.
To further evaluate the effect of these compounds in cells, an RXR␣ transcriptional activation activity was assessed in classical cotransfection assays. Transcriptional activation in CV-1 cells was determined using cotransfected RXR␣ expression vector and RXR␣-responsive CAT reporter construct ((TREpal) 2 -tk-CAT) (1,37). The TREpal is a palindromic response element that is activated by RXR-agonist complexes (1, 2). Dose-dependent antagonism of the RXR␣ homodimer activated by 9-cis-RA was observed with compounds 1-5 (Fig. 5A). The specificity of these compounds for RXR␣ could be inferred by the fact that they had no effect on the ability of RAR␣ to stimulate expression of the same reporter in response to all-trans-RA (Fig. 5B). These data should further validate the discovered compounds as bona fide RXR␣ ligands.
To further elucidate the antagonist effect of these compounds, we studied whether they could inhibit expression of RAR␤, which is regulated by various RXR␣-containing complexes (36,(37)(38)(39) in ZR-75-1 breast cancer cells. Treatment of ZR-75-1 cells with all-trans-RA strongly induced the expression of RAR␤, which was completely inhibited by each compound with the exception of compound 6, consistent with the inhibitory effects witnessed on TREpal reporter gene activity (Fig. 5C). Taken together, these data demonstrate that compounds 1-5 are effective RXR␣ antagonists.
In conclusion, we report here a successful virtual docking approach in which 5 of 200 selected compounds showed RXR␣ transcriptional antagonistic activity in vitro and in cells with binding affinities in the nanomolar range in vitro and low micromolar cellular activity. These compounds with their structural features differing from the existing RXR ligands may represent valuable tools to assist with the dissection of the complex RXR signal transduction pathways. The novel structural scaffolds reported here could be useful for the development of potential lead compounds targeting members of the nuclear receptor superfamily in the development of novel cancer therapeutics.