Minireview: Not Picking Pockets: Nuclear Receptor Alternate-Site Modulators (NRAMs)

Abstract

Because of their central importance in gene regulation and mediating the actions of many hormones, the nuclear receptors (NRs) have long been recognized as very important biological and pharmaceutical targets. Of all the surfaces available on a given NR, the singular site for regulation of receptor activity has almost invariably been the ligand-binding pocket of the receptor, the site where agonists, antagonists, and selective NR modulators interact. With our increasing understanding of the multiple molecular components involved in NR action, researchers have recently begun to look to additional interaction sites on NRs for regulating their activities by novel mechanisms. The alternate NR-associated interaction sites that have been targeted include the coactivator-binding groove and allosteric sites in the ligand-binding domain, the zinc fingers of the DNA-binding domain, and the NR response element in DNA. The studies thus far have been performed with the estrogen receptors, the androgen receptor (AR), the thyroid hormone receptors, and the pregnane X receptor. Phenotypic and conformation-based screens have also identified small molecule modulators that are believed to function through the NRs but have, as yet, unknown sites and mechanisms of action. The rewards from investigation of these NR alternate-site modulators should be the discovery of new therapeutic approaches and novel agents for regulating the activities of these important NR proteins.

Nuclear Receptor Alternate-site Modulators (NRAMs) are molecules that affect nuclear receptor activity by binding at sites distinct from the ligand-binding pocket.

Nuclear Receptor (NR) Activities, Interaction Partners, and Small Molecule Modulators

Our understanding of NR biology has changed greatly since the first reports in the 1960s of a protein in estrogen target tissues that bound the endogenous hormone 17β-estradiol and appeared responsible for mediating its biological effects (1,2,3). For instance, it is now known that NRs comprise a superfamily of approximately 50 proteins that mediate critical actions in reproduction, metabolic regulation, xenobiotic metabolism, etc (4). Additionally, we know that there are a myriad of other factors involved in NR-associated signal transduction, including dimerization partners, coactivators, corepressors, promoter regions (response elements), chaperones, ubiquitin ligases, kinases, phosphatases, and others (5). Within the same time frame, however, the means by which molecular therapeutics target the NRs has progressed comparably less: What is largely available are competitors for natural ligands in the form of NR synthetic agonists, selective NR modulators and NR antagonists. All of these molecules interact with the ligand-binding pocket of the NR, prevent binding of the endogenous ligand, and induce conformations of the NR that either favor or, to varying degrees, disfavor interaction with associated coregulators. More recently developed have been biosynthesis inhibitors that deny the NR its endogenous ligand (e.g. aromatase inhibitors), as well as ligands that bind in the ligand-binding pocket but accelerate receptor degradation [e.g. selective estrogen receptor (ER) down-regulators (6,7)].

There are good reasons for the disparity between our knowledge of NR interacting and modifying partners and the availability of small molecule inhibitors that target them: Although the NRs have been known for many years, the identification and characterization of the other interacting partners have occurred only more recently. Also, from a medicinal chemist’s perspective, NR modulators that act by displacing hormone from the ligand-binding pocket have tractable properties. It is relatively easy to synthesize organic compounds that are hydrophobic and have low water solubility; the ability of a receptor to bind a number of different structural classes, often with low to subnanomolar affinity, makes identifying positive hits, as well as patenting them, simpler; and finally, the ability of a ligand to bind at a site with high affinity translates to fewer off-target effects.

Because of the growing wealth of knowledge of the molecular elements associated with NR function, researchers have begun to consider whether interaction sites other than the ligand-binding pocket could be targeted with new probes, with the ultimate hope that such probes might address some of the limitations of current therapeutics [e.g. inadequate selectivity or resistance to endocrine blockage of NR signaling (8,9,10)]. Some aspects of these NR alternate-site modulators (NRAMs) have been reviewed elsewhere (11,12,13,14), but our goal here is not to concentrate on structural studies. Rather, it is to focus on function, giving as much biological information about these ligands and their targets as is known, so that the molecular endocrinology community might become more aware of their existence and be able to see opportunities for new and more sophisticated experiments. We hope this review will be especially useful to those involved in high-content screening campaigns, because, as whole-cell, and even whole-organism phenotypic screening become more common and increasingly more high throughput (15,16,17,18,19,20,21,22), and as these technologies replace the traditional ligand-binding or coactivator recruitment-screening assays, it becomes likely that molecules targeting these secondary sites will be discovered. Hopefully, these alternate-site modulators will be seen as agents with which one can study new biology and develop new therapeutics.

The sites we will discuss are exemplified in Fig. 1, which shows them displayed on a recent crystal structure of a NR dimer [peroxisome proliferator-activated receptor γ (PPARγ)/retinoid X receptor α] in interaction with a DNA response element (green) and with coactivators (violet) (23). One of the NRs (PPARγ) is represented as black skeletal sticks, to allow buried elements to be visible, whereas the other (retinoid X receptor α) is depicted as a yellow surface. Both NRs are liganded, although only one ligand (rosiglitazone, blue) is visible. Additionally, both receptors bind to the PPAR response element through the zinc fingers in each of their DNA-binding domains; two of the four zinc ions (red) are visible. Each of these sites has been targeted in certain NRs. (Another site termed “BF-3,” not illustrated in Fig. 1, has been described for the AR; see below.)

Figure 1.

Illustration of the different interaction sites for the regulation of NR activity. Multicolor and multimode rendered diagram of a representative NR heterodimer [retinoid X receptor α (yellow)/peroxisome proliferator-activated receptor γ (black)], showing one of two ligands [rosiglitazone (blue)], two of four zinc ions (red) in the zinc fingers, a DNA response element (green), and sequences from a NR interaction box [NR-box (violet)] of SRC 2. (Figure prepared from PDB accession code 3DZY).

In this minireview, we have listed the known NRAMs first by site, and then by NR. Additionally, we have focused on molecules that are likely to be active in cell culture and in vivo experiments; thus, we have not included natural and nonnatural peptides that have been reported (24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43), unless specific actions were taken to make the peptides cell permeable. In many cases, it is also important to rule out the conventional mechanism of NR antagonism with each of these modulators; fortunately, these competition-based experiments are easy to perform and are often among the first done to test for a novel mechanism.

Coregulator Binding Groove

The NRs are currently known to interact with nearly 300 coregulators that have both activating (coactivators) and repressive (corepressors) activities. When the first coregulators were discovered in the 1990s (44,45,46,47,48), they were originally thought to function as mere transcriptional adaptors that linked the NR transcription factor to the RNA polymerase II holocomplex, but it has since become clear that the coregulators have important enzymatic functions as well, including those related to chromatin remodeling (49,50).

The most common corepressors, NR corepressors, are recruited to type II receptors (e.g. TRs, PPARs) when unliganded, via corepressor NR interaction boxes. Because many type II receptors are bound to chromatin and transcriptionally active even in the absence of ligands, corepressor recruitment results in silencing (i.e. repression of basal gene transcription). The interaction of corepressors with NRs has thus far not been targeted, but one would expect that blocking this interaction would result in gene derepression. It is of note that unlike agonists for the majority of type II receptors, which bind within the ligand-binding pocket, dismiss corepressors, and recruit coactivators (5,51), certain TR antagonists, upon binding, dismiss corepressors, but fail to recruit coactivators (52,53,54,55).

The members of the steroid receptor coactivator (SRC) family are among the most important of the NR coactivators and contain similar functional domains: an N-terminal basic helix-loop-helix-Per/ARNT/Sim domain; a flanking serine-/threonine-rich region; an NR interaction domain containing multiple LXXLL motifs; transcriptional activation domain (AD)1, the interaction site for CREB-binding protein , and p300; and transcriptional AD2, the interaction site for protein arginine methyltransferases (e.g. coactivator-associated arginine methyltransferase 1 and Protein Arginine N-Methyl Transferase). Additionally, AD1 and AD2 possess histone acetyltransferase activity, although, because the histone acetyltransferase activity of SRCs is weaker than that of CREB-binding protein and p300, the importance of such activity is unclear (50,56).

In addition to interacting with the NRs, the SRCs also interact with other transcription factors, including activator protein-1 (AP-1) (57), nuclear factor-κB (58), signal transducers and activators of transcription (59), and E2F1 (60). The SRCs are not the only coactivators to interact with NRs; in fact, there are many more. One recent review of the literature placed the number of NR coregulators at nearly 300 (49).

The SRCs, and SRC-3 in particular, are overexpressed in a number of cancers, including ovarian, prostate, and breast cancers (49). This overexpression may contribute to carcinogenesis, enhance tumor progression, and even alter the effects of therapeutic NR ligands. Because attenuating NR activity in cancer cells is desirable, the NR-coactivator interface has been posited as a potential target for pharmaceuticals (41,61). In structures of agonists cocrystallized with NRs and fragments of these coactivators (NR boxes), the receptor envelops the ligand in a newly formed hydrophobic binding pocket, whereas the NR Box peptide reinforces this conformation by binding to the NR in a hydrophobic groove formed by helices 3, 4, 5, and 12, directly above the ligand-binding pocket on the surface of the receptor (62,63). In binding to this groove, coactivators use a conserved LXXLL motif, or NR interaction (NR) box, where L represents leucine and X represents any amino acid (64,65). (An important exception is the ability of the AR alone to also bind an F/WXXLF motif; see below.) Comprising two turns of an α-helix, the LXXLL motif has been shown to be both necessary and sufficient for binding of the coactivator to the NR. Furthermore, the first and last leucines of the motif appear to make deeper hydrophobic interactions in the groove; therefore, it is believed these residues are most important for coactivator binding. In addition, most structures show that the intrinsic dipole moment of the coactivator α-helix is matched on the NR by a negatively charged residue (glutamate) at the N terminus (dipole positive end) and a positively charged residue (lysine) at its C terminus (dipole negative end), which forms a charge clamp (63). Mutational studies have verified the importance of this charge clamp in the functional interaction between NRs and SRC NR boxes (63,66); however, given that the charge clamp glutamate and lysine are on the protein surface, it is not clear that they would need to be engaged in the binding of nonpeptidic molecules designed to fit this site but lacking an intrinsic dipole (see Fig. 2A).

Figure 2.

Ribbon diagrams of three NR ligand-binding domains illustrating different interaction sites. A, ERα (blue) liganded with diethylstilbestrol (red) in the ligand-binding pocket showing interaction of an SRC coactivator peptide (magenta) in the coactivator-binding groove. B, TRβ (green) liganded with T₃ (red) and Michael acceptor HPPE (gray/cyan). The sulfhydryl group of cysteine residue 298 (yellow) later forms a covalent bond with the α,β-unsaturated ketone group (shown in cyan) of HPPE. C, AR (lavender) liganded with DHT (red), with flufenamic acid (orange) located at binding function 3. (Panels A, B, and C prepared from PDB accession codes 3ERD, 2PIN, and 2PIX, respectively).

ERs

The earliest examples of small molecules known to alter the effects of any NR outside of the ligand-biding pocket were those described for ERα in 2004. Rodriguez et al. (67) described pyrimidines (see Fig. 3) that block the interaction of purified E₂-activated ERα with a labeled SRC1 Box II peptide in a fluorescence polarization (FP) assay. To improve the potency of these coactivator binding inhibitors (CBIs), Parent et al. (68) synthesized a larger library of these pyrimidines, with the best exhibiting K_i values of 2–3 μm in a time-resolved fluorescence resonance energy transfer assay (FRET) (69). Additionally, the compounds were shown to inhibit ERα-mediated transcription in HEC-1 cells that had been transiently cotransfected with an ERα plasmid and a luciferase reporter gene, with IC₅₀ values comparable to those from the time-resolved FRET assay. These methods have also been employed by Gunther et al. (70) in describing amphipathic benzene CBIs exhibiting median inhibitory concentrations of 1.7 μm.

Figure 3.

ER coactivator-binding inhibitors.

In 2004, researchers from Wyeth Pharmaceuticals (71) described CBIs discovered through both a high-throughput screen and a virtual screen. Although both series of compounds disclosed were found to inhibit the interaction of ERα with SRC-1, -2, and -3 in an ELISA, only the guanylhydrazone compound ERI-05, discovered through high-throughput screening, was capable of inhibiting the interactions of Gal4 DNA-binding domain/hERα ligand binding domain fusion and SRC-1, SRC-3 or SRC-3/VP16 fusion in a mammalian two-hybrid assay performed in COS7 cells (IC₅₀ = 5.5 μm). Additionally, in an ELISA, ERI-05 was found capable of inhibiting the interaction of SRCs with ERβ, but not with progesterone receptor. ERI-05 also inhibited endogenous expression of the ERα-regulated gene pS2 in MCF-7 cell line at 20 μm, but concentrations greater than this were toxic to cells. LaFrate et al. (72) made a library of these compounds, with some showing improved potency over ERI-05.

In 2006, Wang et al. (73) published a provocative crystal structure of ERβ bound to two molecules of the selective ER modulator (SERM) hydroxytamoxifen (4-OH-TAM). One of these molecules was bound, as expected, in the ligand-binding pocket of the receptor, but the other was bound in the coactivator-binding groove. This finding seems to substantiate older biochemical data that showed that the total binding capacity of 4-OH-TAM for ERα was 2 times that of estradiol (12,74,75), and it further illustrates at a structural level that this coactivator-binding groove is capable of accommodating other small molecules. It is curious, however, to find a second molecule of 4-OH-TAM bound in a groove that, at least in ERα, is normally occluded by helix-12 in ER complexes with SERMs (63,76). It is possible that at high concentrations, this second molecule is able to enforce a secondary inhibition of coactivator binding; however, because SERM-ER complexes themselves are unable to bind coactivators in vitro, it would be difficult to ascertain experimentally whether inhibition of coactivator binding is the result of occupancy of the coactivator groove by the second molecule of ligand in this 4-OH-TAM-ERβ complex or a consequence of the conventional SERM mechanism.

Carraz et al. (77) have recently described CBI peptides that have been tagged with biotin (for imaging by recruitment of fluorescently labeled antistreptavidin) and a nona-Arg TAT peptide (for cell permeability). These molecules bound ERα with low micromolar IC₅₀ values; the sequence of the most efficacious of the peptides in a reporter gene assay was taken from Tif2 (SRC-2) Box 2. Additionally, administration of the peptide (70 μm) to MCF-7 cells down-regulated expression of the estrogen-regulated pS2 gene to non-E₂-treated levels. Oligo-Arg sequences are known to facilitate sequestration of peptides labeled with the sequence to the nucleolus (78), and confocal laser scanning microscopy studies in U2OS cells (transiently transfected with full-length ERα-cyan fluorescent protein) revealed that the compounds bound ERα and sequestered it in the nucleolus. Thus, these molecules function by a two-pronged mechanism: direct inhibition of the ER-coactivator interaction and sequestration of the ER from its ERE.

Other compounds have been shown to directly inhibit the interaction of ER with coactivators, although the biological analysis of these small molecules is less complete (79,80). Additionally, some CBI peptides have been designed to be cell permeable (27,29), but because of the untimely death of the leader of this project, Arno Spatola, the ability of the peptides to traverse the cell membrane has not been examined.

Thyroid hormone receptor (TR)

Using a FP-based high-throughput screen, Arnold et al. (81) discovered a chemically unstable compound (a β-amino ketone or Mannich base) that eliminates an amine to give an α,β-unsaturated ketone, more commonly known as a Michael acceptor [e.g. 1-(4-hexylphenyl)-prop-2-en-1-one (HPPE); see Fig. 4]. Soft nucleophiles, such as the sulfhydryl group of cysteine, can react with Michael acceptors to form a new covalent bond. Mutation, pull-down, fragmentation, and structural studies (82) suggested that the target of the in situ-formed Michael acceptor is Cys298, a residue on the surface of the receptor near the coactivator-binding groove (see Fig. 2B). The reactive Michael acceptor appears to be generated preferentially whereas the precursor Mannich base is bound, initially reversibly, at its site of eventual covalent attachment, because its modification of Cys298 is much more specific than that of general thiol modifying reagents. This selective alkylation is reminiscent of the preferential target site reaction of tamoxifen (TAM) aziridine with cysteine residues in the ligand-binding pocket of ER (83).

Figure 4.

TR coactivator-binding inhibitors.

Synthesis of analogs of these compounds showed no clear structure-activity relationship, but the best of the compounds had IC₅₀ values as low as 1.5 μm in the FP assay, and as low as 2.5 μm in cytotoxicity assays in both TR-positive (ARO) and TR-negative (U2OS) cancer cell lines (84). A second-generation library of these β-amino ketones, aimed at improving the pharmacological profile of the original, was recently reported by Hwang et al. (85). Several of these compounds showed improved potency in an FP assay (IC₅₀ = 0.6 μm), while having low toxicity in HepG2 cells (EC₅₀ = 101 μm). Additionally, these TR CBIs were largely inactive on the hERG channel, a marker for a cardiac ion channel activity often associated with drug toxicity.

AR

As mentioned above and shown in Fig. 5, the AR is unique among NRs in its ability to bind larger motifs (e.g. FXXLF and WXXVW) (25,87). It has been proposed that this size-exclusion principle could be used to selectively target the AR over other NRs (61).

Figure 5.

Representation of coactivator-binding grooves of ERα ligand-binding domain (A) bound to an LXXLL motif (yellow) and AR ligand-binding domain (B) bound to an FXXLF motif (green). A distance map (red = 0.5 Å; dark gray > 7 Å) from the backbone of the coactivator helix has been applied to the surface of each receptor, demonstrating that the interior of the ER groove is shallower (pink), on average, than that of the AR (gray). The charge clamp residues (K362/E542 for ER and K720/E897 for androgen receptor) are denoted. (Figure prepared using PDB accession codes 3ERD and 1XOW.)

Leveraging this ability of AR to selectively bind sterically demanding motifs, Gunther et al. (88) reassayed their previously published pyrimidine ER CBIs (68) to search for AR-selective CBIs. Because the display of substituents on the FXXLF motif is equivalent to that of the LXXLL motif, their hypothesis was that those pyrimidines that displayed larger, aromatic substituents would be accommodated by AR, but not ER. Using the previously described luciferase assay, Gunther et al. discovered that some of the larger CBIs were, indeed, selective for AR (IC₅₀ values as low as 1.9 μm), to the complete exclusion of ER. Additionally, most of the compounds were active against the T877A LNCaP mutant that is present in approximately 30% of patients with metastatic prostate cancer, a significant finding when one considers that this mutation renders the widely used prostate cancer drug hydroxyflutamide a weak agonist, rather than an antagonist (89,90). This size-expansion approach could be used in guiding substituent hopping from known ER CBIs to potential AR CBIs.

Pregnane X receptor (PXR)

The PXR is one of the NRs known to mediate the metabolism of xenobiotics (constitutive androstane receptor and farnesol X receptor are others). Antagonists of PXR could be useful when coadministered with a metabolically unstable drug as a means to extend the half-life of the drug (11). Typical antagonists of PXR that displace the agonist rifampicin from the ligand-binding pocket are not commonly known, but in 2002 Takeshita et al. (91) reported that the antifungal ketoconazole (see Fig. 6) acted as an antagonist of PXR. Later, Huang et al. (92) reported that ketoconazole does not affect ligand binding, receptor dimerization, or DNA binding. In part, because its IC₅₀ of 74 μm in a radiometric ligand-binding assay is higher than the concentrations (6–25 μm) needed to induce biological effects, the authors have asserted that ketoconazole directly or allosterically blocks the interaction of PXR with SRCs. Indeed, loss of righting reflex (LORR) studies induced by 2,2,2-tribromoethanol in C57BL/6 mice demonstrated a LORR of short duration when pregnenolone carbonitrile or paclitaxel (known PXR activators) alone were given, and reversal of these effects when these activators were given with ketoconazole. Ketoconazole inhibited the induction of cyp3a11 and mdr-1, two genes regulated by PXR, in the livers of PXR^+/+ mice. These effects were not seen in PXR^−/− mice, suggesting that the effect was attributable to PXR. Das et al. (93) have prepared analogs of ketoconazole, some showing affinities comparable to ketoconazole itself. Ekins et al. (94,95) have found other compounds, including the antirheumatic compound leflunomide (IC₅₀ = 6.8 μm) and some electrophilic thioester compounds, that appear to be PXR CBIs.

Figure 6.

PXR coactivator-binding inhibitors.

The phytoestrogen coumestrol has also been shown to be a PXR CBI (96). Coumestrol binds to the ligand-binding pocket of PXR with an IC₅₀ of 13 μm, but it is believed to be a CBI because it antagonizes PXRs that have been mutated to have ligand pocket-filling residues, and it does so regardless of whether an agonist is present. A direct FP assay to determine whether coumestrol directly blocks the PXR/SRC-1 interaction was inconclusive because of the low solubility of coumestrol in the assay buffer. Coumestrol (25 μm) was able to block the induction of cyp3a4 in primary human hepatocytes. LORR experiments, carried out as above, demonstrated that, in mice carrying the human PXR allele, coumestrol was able to significantly increase the duration of LORR. Because coumestrol is a known phytoestrogen, at the high concentrations required to disrupt PXR activity as a CBI, it would be expected to have considerable estrogenic effects (97).

Binding Function 3 (BF-3)

AR

Estébanez-Perpiñá et al. (98) have reported a novel crystallographic screening approach that has identified compounds that allosterically, rather than directly, inhibit the interaction of AR with coactivators, by binding at a previously unknown site. This hydrophobic binding site, which is almost as large as AF-2 (but is distinct from it), is near the junction of H1, the H3-H5 loop, and H9; it is termed “binding function 3” (BF-3; see Fig. 2C). Seven compounds, including T₃, triiodothyroacetic acid, and flufenamic acid (99) (see Fig. 7), were cocrystallized with the AR at the BF-3 site. The molecules seem to reorganize residues in both BF-3 and AF-2 and induce a large-scale repositioning of Arg726 at the AF-2 boundary. Although the median inhibitory concentrations of these compounds were rather high in an FP assay (i.e. ≥ 50 μm), they were active at 10–30 μm in reporter gene assays. Many of these compounds would be active at their original targets (e.g. triiodothyroacetic acid and T₃ on TR) and thus have off-target effects, but they are among the only known compounds of their type for which structural data are available, and, as such, provide a powerful starting point for drug discovery.

Figure 7.

AR-binding function 3 inhibitors.

Nuclear Response Elements

NRs mediate the expression of genes, in part, by binding to their response elements, which can occur in promoter sequences proximal to the gene or at more distant enhancer elements that can be very far from the transcription start site. Although there is a good deal of promiscuity in recognized sequences, most NRs have a consensus response element composed of two half-sites separated by a variable number of nucleotides; the consensus androgen response element (ARE) is 5′-GGTACAnnnTGTTCT-3′ (100), and the consensus estrogen response element (ERE) is 5′-AGGTCAnnnTGACCT-3′ (101). The DNA-binding domain of the NR interacts with the response element through the DNA-binding domain constructed from two Cys₄ zinc fingers (102) (see below).

ER

Gearhart et al. (103) have reported pyrrole-imidazole polyamides, based on the generalized approach developed by Dervan and co-workers, that bind to the minor groove of an ERE. (The sequence of the pyrrole-imidazole polyamide can be controlled in such a way that specific DNA sequences can be targeted with high specificity and affinity; for a review see Ref. 104.) The ERE is capable of binding either the ERs or the estrogen-related receptors (ERRs), although their binding modes are different: ERRs bind as monomers to ERE half-sites, making contacts in both the major and minor grooves, whereas the ERs bind as dimers to the ERE, making contacts in only the major groove. Gearhart et al. examined whether these high-affinity polyamides [e.g. polyamide 4 binds to the ERα consensus 5′-AGGTCA-3′ half-site with a dissociation constant (K_d) of 1.0 nm], which bind the minor groove, would be selective in inhibiting the ERR-ERE interaction (minor groove) or would be capable of also allosterically inhibiting the ER/ERE interaction (major groove). The authors found that both hERR2 and hERα were blocked from interacting with an ERE, suggesting either a direct or an allosteric mechanism is operable; however, they also found that the ERR/ERE interaction could be selectively inhibited not by targeting the consensus 5′-AGGTCA-3′ half-site, but by targeting the C-terminal extension recognized by the DNA-binding domains of ERR but not ERα.

AR

Nickols and Dervan (105) have described hairpin polyamides, similar to those mentioned above, that target the minor groove of the androgen response element (ARE) consensus 5′-AGAACAGCAAGTGCT-3′ from the prostate-specific antigen (PSA) promoter (K_a = 8.3 ± 1.7 × 10⁹ m⁻¹). In an EMSA, polyamide 5 inhibits AR binding to the 5′-³²P-labeled duplex corresponding to the ARE site in the PSA promoter at concentrations as low as 10 nm, as assayed using a nuclear extract isolated from DHT-stimulated LNCaP cells. In LNCaP cells, polyamide 5 inhibits the DHT-induced expression of PSA and FKBP5, two strongly androgen-induced genes in a dose-dependent manner (up to 60–70% at 10 μm), and chromatin immunoprecipitation (ChIP) experiments indicate decreased AR occupancy at the PSA promoter and enhancer in the presence of 10 μm polyamide 5. Of more than 50,000 transcripts assayed in LNCaP cells, polyamide 5 (10 μm) affected the expression of 1053 transcripts by at least 2-fold. Of the 122 transcripts down-regulated by both bicalutamide and polyamide 5, 117 are also observed to be induced by DHT at the same thresholds. Of the 90 transcripts up-regulated by both bicalutamide and polyamide 5, 59 are observed to be repressed by DHT. These data suggest that the ARE is varied to an extent that a particular polyamide would be unlikely to affect all AR-regulated genes equally, but they also suggest that it might be possible to target selectively a particular cadre of genes, based on some unifying characteristic of their AREs.

Reasoning that a more constrained, cyclic, rather than hairpin, polyamide would have higher DNA affinity, Chenoweth et al. (106) prepared a cyclized version (not shown) of polyamide 5 that demonstrated higher thermal stabilization (ΔT_m of 23.6 and 18.4 C for cyclic and hairpin congeners, respectively), and had a greater effect on decreasing PSA mRNA levels in LNCaP cells. The cyclic polyamide exhibited low CaCo-2 cell permeability (suggesting low oral availability) and was protein bound in plasma, but it was nontoxic to HepG2 cells at 100 μm. It also had a half-life longer than 3 h in human and rat liver microsomes and did not any inhibit any of the P450s examined.

Zinc Fingers of the DNA-Binding Domain

ER

A DNA-binding inhibitor that targets one of the zinc fingers of the ER has been described by Wang et al. (107). The DNA-binding domain of the ER contains two nonequivalent Cys₄ zinc fingers; of these two, the C-terminal zinc finger is structurally disordered but is stabilized upon dimerization. A 2,2′-dithiobisbenzamide (DIBA; see Fig. 9), caused release of zinc with purified full-length ERα, but not with PR B, and kinetic studies suggested that the C-terminal zinc finger underwent ejection at a rate faster than the N-terminal zinc finger. This is consistent with modeling studies that suggested that the C-terminal zinc finger is less sterically and electrostatically protected than the N-terminal zinc finger, and, moreover, less protected than C-terminal zinc fingers in other NRs [i.e. retinoid X receptor and glucocorticoid receptor (GR)]. DIBA inhibited proliferation of ER-positive breast cancer cell lines (i.e. MCF-7, T-47D, and ZR-75 cells), with IC₅₀ values of 10–25 μm. MCF-7 xenografts in nude mice showed a dose-dependent decrease in growth; a high dose (30 mg/kg) of DIBA reduced tumor mass to undetectable levels, without any resulting kidney or liver toxicity. Significantly, MDA-MB-231 (ER-negative breast cancer cell line) xenografts showed no difference in tumor growth. In MCF-7 cells, DIBA inhibited the expression of mRNA of the ER-regulated gene e2f-1 and decreased binding of purified ERα to an ERE, the latter being reversed by addition of zinc. DIBA appeared to be selective for ER, and it did not affect the expression of zinc-dependent HDAC, nor did it affect nuclear factor-κB, a nonzinc finger transcription factor. It did, however, inhibit retroviral Cys₃His zinc fingers (108).

Figure 9.

ER zinc finger-binding inhibitors.

Notably, the addition of DIBA to TAM-resistant cell lines (i.e. MCF-7/LCC2, MCF-7/HER2-18, BT474, and ZR-75) restored the antagonist activity of TAM (107), and DIBA decreased tumor volume in MCF-7 xenografts when given with 4-OH-TAM, whereas 4-OH-TAM alone had no effect on tumor growth. Small interfering RNA knockdown showed that ER was necessary for the cellular effects of DIBA and that DIBA functioned by inhibiting ER from binding to its ERE, but did not affect AR/ARE binding activity. DIBA also did not block estradiol- or TAM-stimulated activator protein (AP)-1 binding activity, which relies on ER binding to DNA through AP-1 at an AP-1 chromatin binding site, rather than an ERE, and thus does not require a functional DNA-binding domain. DIBA blocked occupancy of ERα to promoter regions of the ER-regulated genes pS2, c-myc, and cathepsin D, and also decreased the recruitment of the coactivator AIB1 (SRC-3) to ERα mediated by SERM 4-OH-TAM and increased the recruitment of NR corepressor.

Kim et al. (109), have found that a ligand for PPARγ (15d-PGJ₂) inhibits proliferation of breast cancer cells, but that it does so without working through PPARγ. Based on transactivation assays, the compound seems to inhibit ERα, but not ERβ, at low micromolar concentrations. On the basis of proteolytic digestion, pull-down, and gel mobility shift assays, the cyclopentenone ring system of 15d-PGJ₂ is believed to undergo Michael addition at Cys277 and Cys240 in the C-terminal zinc finger of ERα. A control compound lacking the key double bond (a cyclopentanone) was inactive.

Mao et al. (110) have described a small molecule [theophylline, 8-[(benzylthio)methyl](TPBM)], discovered through a FP-based high-throughput screen, which interrupts the E₂-activated interaction of ERα with its ERE with an IC₅₀ value of 3 μm and is somewhat specific for ERα (i.e. IC₅₀ values for AR and progesterone receptor (PR) are 7.6 and 9.5 μm, respectively). TPBM was also active in a luciferase reporter gene assay (IC₅₀ of 11.5 μm), and it did not inhibit PR or GR transactivation to a significant extent in similar assays. It exhibited dose-dependent inhibition of proliferation of BG-1 cells (IC₅₀ = 5 μm), but was not toxic to MDA-MB-231 cells (ER negative) nor to a panel of 60 cell lines tested through the NIH Developmental Therapeutics Program. TPBM is effective in inhibiting the E₂- and 4-OH-TAM-induced expression of PI-9 in a TAM-resistant model (MCF7/ERαHA). ChIP experiments demonstrated that TPBM inhibited the binding of E₂-ERα to the PI-9 estrogen-responsive unit, but does not act as a zinc chelator.

Unknown Binding Sites/Targets

AR

Using a conformation-based high-throughput screen, Jones et al. (111) and Jones and Diamond (112) discovered two high-affinity compounds (pyrvinium pamoate and harmol hydrochloride) (see Fig. 10). that inhibit the action of AR in in vitro and mouse models of AR activity, although it is unclear what the molecular targets of these compounds are. Both compounds acted synergistically with hydroxyflutamide in reporter gene assays in LNCaP and LAPC4 cell lines, implying that they did not bind at the same site as hydroxyflutamide, a known selective androgen receptor modulator. Also, they did not compete with DHT to inhibit AR-driven reporter gene activity, and pyrvinium pamoate did not compete at all in a radiometric ligand-binding assay.

Figure 10.

AR antagonists/inverse agonists that function at unknown binding sites.

Pyrvinium pamoate and harmol hydrochloride were synergistic with each other, suggesting that they bind at distinct sites. ChIP assays showed that pyrvinium pamoate did not affect DHT-induced occupancy of AR at its DNA-binding sites, but that harmol hydrochloride did reduce occupancy of the receptor at six sites examined in LNCaP cells. Both compounds, however, reduced RNA polymerase II occupancy at locations along the PSA promoter. Thus, harmol hydrochloride is believed to block AR promoter occupancy, whereas pyrvinium pamoate seems to block recruitment of RNA polymerase II.

Pyrvinium pamoate was more effective than either harmol hydrochloride or bicalutamide at inhibiting the proliferation of LNCaP cells. Human embryonic kidney 293 (control) cells were not affected by any of the compounds, implying that the effect was AR dependent. Although harmol hydrochloride was metabolized quickly, the plasma concentration of pyrvinium pamoate in rodents remained at 20–150 nm, within the efficacious range. Pyrvinium pamoate alone caused a 9% reduction in prostate weight (not statistically significant), and bicalutamide alone caused a 35% reduction; the combination of pyrvinium pamoate and bicalutamide caused a 63% reduction in prostate weight, almost as effective as castration.

Jones et al. (113) have also used a multimodal approach that combined the FRET screen, mentioned above, to examine AR conformation and high-throughput microscopy to examine AR nuclear accumulation. Most of the compounds that were found to have activity are known or suspected to target AR-accessory or regulatory factors [e.g., HSP90 (radicicol) and JAK/signal transducer and activator of transcription 3 (cucurbitacin)], but this assay could be used to find alternate-site modulators as well.

Gelsolin is an AR-binding protein that interacts with AR in the presence of both agonists and antagonists, and its overexpression seems to increase the partial agonist activities of casodex and hydroxyflutamide (114). Joseph et al. (115) and Norris et al. (116) have developed a mammalian two-hybrid assay to search for AR inverse agonists/antagonists that disrupt the AR/gelsolin interaction in the presence of R1881. Two compounds, D36 and D80, not only inhibited this interaction at low micromolar concentrations but were also able to inhibit the transcription of a luciferase reporter gene in LNCaP cells. D36 and D80 also inhibited the expression of endogenous genes regulated by AR (e.g. PSA) and inhibited the basal and androgen-stimulated proliferation of LAPC4 prostate cancer cells; additionally, D36 and D80 inhibited proliferation of two different cellular models of hormone-refractory prostate cancer (VCaP and SRαAR LNCaP). Mechanistically, D36 and D80 appeared to act at the DNA-binding or cofactor-recruitment level, because the compounds did not recruit AR to the PSA enhancer, nor did they affect the subcellular localization of AR.

Conclusions

It is clear from the aforementioned examples that there is an expanding array of inhibitors of NR signaling that operate through binding sites that are distinct from the ligand-binding pocket itself. In some cases, these novel inhibitors interrupt other key interactions required for the gene-regulatory activities of the NRs, such as those that block NR-coactivator binding or interfere with NR-DNA binding, either by binding to the DNA-binding domain of the NR or to the response element itself. In other cases, the sites of interaction are as yet unknown. Additionally, there are other binding sites that have been posited and examined as targets [e.g. the NR dimer interface (117)] and still others where binding seems to have no immediate biological consequence (37,41). Figure 1 illustrated the locations of some of these new sites of interaction, mapping these on the most complete x-ray structure of a NR dimer-DNA complex known (23).

Nevertheless, despite considerable progress that has been made in the field of NRAMs, it is still early days and much is yet to be accomplished. What is most seriously lacking are NRAMs with low nanomolar potencies. Most conventional ligands (i.e. agonists, antagonists, and selective nuclear receptor modulators) have potencies in this range, which undoubtedly contribute to their in vivo selectivity and even clinical utility. Few NRAMs, however, have comparable potency; most have IC₅₀ values that are, at best, micromolar. Screening larger compound libraries, especially those enriched in compounds of higher molecular weight, more appropriate for disrupting protein-protein and protein-DNA interactions (86), might be a fruitful approach to discovering NRAMs of higher potency, particularly CBIs. Broad screens that were conducted to identify conventional, competitive ligand antagonists by their ability to block coactivator recruitment could be reexamined, because CBIs will also be active in such assays although by a different mechanism; CBIs and conventional antagonists can be readily distinguished by appropriate competition assays because inhibition by CBIs is not reversed by increasing agonist concentrations whereas inhibition by competitive antagonists is (68,69,88). In general, little is also known about the NR selectivity of NRAMs; typically, they have been developed for one alternate site on a specific NR, and investigation of their off-target activities has been limited. Their activities against greater panels of NRs and other transcription factors could be investigated. Also very limited at this point have been structural biology studies of NRAMs as well as studies of their activities in vivo in appropriate experimental animal model systems.

High-throughput screening assays based on specific ligand-protein or protein-protein interactions have been the mainstay of discovery of small molecule NR inhibitors in the past two decades. Although these methods have been very successful, they are specific or focused and might even be considered somewhat myopic in that the molecular components from which the assay is constituted largely define the mechanism by which active hits will operate; in essence, the pockets are picked during assay design. What is newer is the advent of automated ways to perform high-content screens, often based on monitoring the morphology or phenotype of cells, and even whole organisms, when exposed to compound libraries (15,16,17). Other microscopy-based screening approaches follow the conformation, movement, or interactions of genetically tagged proteins in cells by microspectrometric analysis. These assays are not limited to a predetermined target; therefore, when a hit is found it is usually not known where and how it acts. Although establishing mechanisms and sites of action of hits found through such high-content phenotypic screens can be challenging, the ultimate rewards from not picking pockets should be the discovery of compounds with novel activities that may uncover new mechanisms and lead to new therapeutic agents for the treatment of the many disease states in which NRs play a role. Clinical utility would be the ultimate goal for NRAMs, but some of them could be used now as mechanistic probes to disentangle the complex pathways associated with NR biology.

Figure 8.

ERE- and ARE-binding inhibitors.