Abstract
While estrogen-related receptors (ERRα, ERRβ, and ERRγ) share a high amino acid sequence homology with estrogen receptors (ERs), estrogens are not ligands of ERRs. Structure-function studies from this and other laboratories have revealed that ERRs have small ligand-binding pockets and have provided evidence to show that these receptors can activate gene transcription in a constitutive manner. To address the question as to whether there is any agonist for ERRs, our laboratory recently performed virtual ligand screening on ERRα that predicted flavone and isoflavone phytoestrogens to be ligands of this receptor. Our mammalian cell transfection and mammalian two-hybrid experiments revealed that three isoflavones (genistein, daidzein, and biochanin A) and one flavone (6,3′,4′-trihydroxyflavone) behaved as agonists of ERRs. These phytoestrogens induced the activity of ERRα at concentrations that are comparable to those for the activation of ERα and ERβ. In this study, we also used the results of ERRα ligand-binding site mutant, F232A, to verify our ERRα hypothetical computer model. Our recent ERR research has determined for the first time that flavone and isoflavone phytoestrogens are agonists of ERRs. In addition, our studies have demonstrated that an approach that combines structure-based virtual screening and receptor functional assays can identify novel ligands of orphan nuclear receptors.
Introduction
There are three members in the estrogen-related receptor (ERR) family, that is, ERRα, ERRβ, and ERRγ. The cDNAs for ERRα and ERRβ were first isolated by screening cDNA libraries using probes corresponding to the DNA-binding domain of the human estrogen receptor α (ERα; 1). ERRγ was identified during an analysis of the critical region of type IIa Usher syndrome (2), and was also identified by yeast two-hybrid screening, using the transcriptional coactivator glucocorticoid receptor interacting protein 1 (GRIP1) as bait (3). While ERRs share a high amino acid sequence homology with ERs, estrogens are not ligands of ERRs. In fact, ERRs are transcriptionally active in the absence of exogenous ligand. In a recent study, we generated results that lead us to propose that Phe-2322
Please note that ERRα was numbered according to the protein sequence recently updated in GenBank (NM_004451; 4). The ERRα numbering will be used throughout the text, if not otherwise mentioned.
During the last several years, four compounds have been shown to act as antagonists of ERRs. Using yeast-based assays and mammalian transient transfection assays, we have previously found that two organochlorine pesticides, toxaphene and chlordane, can act as antagonists of ERRα, suppressing its constitutive activity (7). Diethylstilbestrol (DES) and 4-hydroxytamoxifen (4-OHT) have also been found to be antagonists of ERRs by two laboratories (8, 9). Because ERRs are constitutively active, it is easier to search for antagonists that suppress the basal activity than to identify agonists, which have to be able to augment the basal activity. To address the important question as to whether there are any agonists of ERRs, we used an approach that combines structure-based virtual screening (SVS) and receptor functional assays to search for agonists. Our studies have revealed that flavone and isoflavone phytoestrogens can act as agonists of ERRs (Fig. 1). Results from these studies will be presented and discussed.
Results
Computer Modeling of the Human ERRα Ligand-Binding Domain
Virtual ligand screening by docking requires a good representative three-dimensional structure of the considered target. To screen for agonists of ERRα, it would be advantageous to use an agonist-bound ERRα crystal structure. While an ERRα ligand-bound structure has not yet been reported, a ligand-free human ERRγ structure (PDB code 1KV6) was reported by Greschik et al. (6) that shares 57% amino acid sequence identity with the human ERRα ligand-binding domain (LBD). We prepared a homology model of ERRα based on the ERRγ free protein X-ray structure and found that it possessed a very small “ligand”-binding pocket. More specifically, the side chain of Phe-399 on helix 11 protruded into the ligand-binding pocket, partially filled the cavity, and interfered with the formation of hydrogen bonds between His-398 and the ligand. The position of this Phe-399, together with the pull-in of Leu-305, Phe-286, as well as helices 3, 5, 7, 11, and 12, led to a tightly packed pocket that is only about half of the size of ERα as indicated by Greschik et al. (6). Due to the small size of the pocket, we decided not to use this model for ligand screening.
A homology model of ERRα using the DES-bound agonist form of human ERα LBD as the template was previously generated from our laboratory and was used to explain the constitutive activity of ERRα (5). Since then, several high-resolution X-ray crystal structures of the agonist-bound human ERα have been published. Sequence homology between the human ERRα LBD (235 residues) and the human ERα LBD is 34% amino acid identity and 55% similarity. The homology is better conserved in the ligand-binding pocket formed by 22 residues (10), where the two receptors share amino acid sequence identity of 45% and similarity of 74%. For the purpose of generating an accurate, diversified, and unbiased ligand-binding pocket for the screening of agonists of ERRα, three human ERα crystal structures with different agonist complexes [i.e., DES, (R,R)-5,11-cis-diethyl-5,6,11,12-tetrahydrochrysene-2,8-diol, and 17β-estradiol] were used as templates (PDB codes 3ERD, 1L2I, and 1GWR, respectively). This second-generation structure model has the common classical three-dimensional features of the nuclear receptor family. The 22 residues forming the ligand-binding cavity are mostly hydrophobic (Fig. 2). The model was minimized as described in “Materials and Methods” and was used to screen for ERRα agonists. Our previous study suggested that Phe-232 on helix 3, one of the amino acids in the ligand-binding pocket of ERRα, is responsible for the constitutive activity of ERRα (5). The mutant F232A, noted as ERRαM, has a significantly lower activity than the wild-type receptor. A homology model of ERRαM was built by the same procedure as described for ERRα and used to investigate the role of Phe-232 in the protein activity and ligand binding. It is interesting to note that, like ERα and ERβ, ERRβ and ERRγ also have an alanine residue in this site. Phe-232 is a unique structural feature in the ligand-binding site of ERRα.
The quality of our ERRα and ERRαM models was evaluated by checking the stereochemistry, local geometry, solvent accessible surface areas, and side chain conformational probabilities with the ProTable module from SYBYL (11). Analysis of the Ramachandran plot showed that 97.5% of the residues in the two models have the Φ/ψ dihedral angles in the most favored or allowed regions. Only Thr-191 at the NH2 termini and Glu-343 in the loop region connecting H8 and H9, which are far from the ligand-binding cavity, sit in disallowed regions of the Ramachandran plot, and were corrected later by energy minimization. The overall average energies of the models calculated by MatchMaker energy plot (11) were below zero, indicating that there is no major problem with the structures and the models could be used for further analysis.
Structural Features of the Ligand-Binding Pocket of ERRα
Compared with the ER family, members of the ERR family have ligand-binding pockets that contain bulkier residues (Fig. 2). As a result, human ERRγ (ligand-binding pocket volume of 220 Å3, 6) and human ERRα (295 Å3 from our modeling study, Fig. 3B) have smaller ligand-binding pockets than human ERα (450 Å3) and human ERβ (390 Å; 11). Our new ERRα model based on ERα ligand-bound structures indicates that three phenylalanine residues, Phe-232 from H3, Phe-399 from H11, and Phe-414 from H12, line up on one side of the ERRα ligand-binding cavity and stack with each other (Fig. 3, B and C). These stacked aromatic side chains fill the upper portion of the β-face of the cavity and change the pocket from the “wedge” shape in ER (Fig. 3A) to a more flattened shape (Fig. 3B).
A conserved water molecule has been indicated to be important for ligand binding to human ERα and ERβ (12–14), and is also structurally conserved among other steroid receptors (15). The water molecule was manually added to our ERRα model by superimposing the modeled structure with the ERα template, followed by unrestrained energy minimization. This water resides at the narrow end of the β-face of the ligand-binding cavity of ERRα receptors and could stabilize the ligand-receptor complex by forming hydrogen bonds with hydroxyl groups of the ligand and Glu-235 and Arg-276 of the receptor. The hydrogen bonding between the water and the ligand could help the ligand dock into the pocket (Fig. 3, B, D, E, and F).
Screening of a Virtual Database for Agonists of ERRα
While our previous computer model of ERRα had revealed that the side chain of Phe-232 might be able to mimic bound ligand in the free protein, our new model suggests that in the presence of ligand, Phe-232 took a different rotamer to stack with Phe-414 and opened up the pocket for ligand binding. This new model was used in the search for ERRα agonists. A virtual screening of 603 compounds in the Indofine catalog was performed (see “Materials and Methods”). Each compound in the Indofine database was flexibly docked into the modeled ERRα agonist-binding pocket with the conserved water present. Those ligands that failed to dock into the pocket were filtered out using a TRIPOS SPL script. To verify our findings from the virtual screening, 50 compounds in the hit list were purchased and tested. Mammalian cell transfection and mammalian two-hybrid functional assays were used to test these compounds (discussed in the next section). Four ligands, ranked within the top 30 in the hit list, including three isoflavones (genistein, daidzein, and biochanin A) and a flavone (6,3′,4′-trihydroxyflavone), demonstrated the ability to enhance ERRα activity and were thus identified as ERRα agonists (Fig. 1). Two known antagonists of ERRs, 4-OHT and DES, and the ERα agonist, E2, failed to dock in our virtual screening. E2 is known not to be a ligand of ERRs. 4-OHT and DES have been shown to be antagonists of ERRβ and ERRγ, thus they should not bind to the agonist-bound form of ERRα (16).
Confirmation of the Interaction of Phytoestrogens With ERRα by Mammalian Transfection and Mammalian Two-Hybrid Experiments
Our computer docking analyses have predicted that flavone and isoflavone phytoestrogens are ligands of ERRα. Using mammalian transfection and mammalian two-hybrid experiments, we have found that three isoflavones (genistein, daidzein, and biochanin A) and one flavone (6,3′,4′-trihydroxyflavone; Fig. 1) can act as agonists of ERRα. The mammalian transfection experiments demonstrated that these four compounds also act as agonists of ERRβ (Fig. 4). However, the induction of the activation of ERRγ by genistein and daidzein were not statistically significant, suggesting that these two isoflavones are relatively poor ligands of ERRγ. Dose-response studies to evaluate the binding of these compounds to ERRα were performed, and the results were compared to those generated with ERα and ERβ (Fig. 5). Although ERRα has a high constitutive activity, a dose-dependent increase of the reporter activity in the presence of phytoestrogens was observed. In addition, the maximal activity after phytoestrogen treatment was similar among ERRα, ERα, and ERβ (Fig. 5).
The interaction of phytoestrogens with ERRs was further examined using mammalian two-hybrid analysis. The three isoflavones were shown to enhance the interaction between ERRs and the coactivator PNRC (Fig. 6). These results confirmed that the compounds tested here are indeed agonists of ERRs. It is recognized that the increase in the interaction between ERRα and PNRC by genistein is not statistically significant. Furthermore, while 6,3′,4′-trihydroxyflavone was shown to be an agonist of ERRs using the mammalian transfection assays, this compound was not able to enhance the interaction between ERRs and PNRC (results not shown). Several studies (17–19) have found that different ligands of a nuclear receptor can induce different conformational changes of the receptor that lead to differential coactivator recruitment capacities. It is possible that genistein and 6,3′,4′-trihydroxyflavone induce conformational changes that reduce ERR interaction with PNRC or enhance ERR interaction with endogenous coactivators in HeLa cells.
As discussed above, our laboratory has previously reported that Phe-232 (analogous to Ala-350 in ERα) is responsible for the constitutive activity of ERRα. In this study, we predicted that Phe-232 is also playing a critical role in defining the ligand-binding pocket of ERRα. Therefore, we examined the interaction of the four phytoestrogens with the ERRα mutant F232A using mammalian transfection experiments. As reported by Chen et al. (5), the mutant F232A is a dominant negative mutant. The LUC activity of the mutant F232A (without ligand) was determined to be approximately 25% of that of the wild-type ERRα. Our experiments revealed that genistein and daidzein were more effective inducing agents for F232A than for the wild-type ERRα (Fig. 7). However, the mutation reduced the binding of biochanin A and 6,3′,4′-trihydroxyflavone. These results further confirmed that these phytoestrogens indeed bind to the ligand-binding pocket of ERRα and their interactions are modified by the mutation F232A.
Interpretation of the Assay Results by Computer Modeling
The identified agonists, genistein, biochanin A, daidzein, and 6,3′,4′-trihydroxyflavone, were carefully checked by visual inspection in the ligand-binding pocket for their shape complementarity, hydrogen bonding network, and van der Waals clashes with the receptor. The ERRα agonists are isoflavone and flavone analogues with extended aromatic rings (Fig. 1). These four compounds are different from most of the other flavones and isoflavones because they have two hydroxyl or methoxyl groups lined up at the C7 and C4′ end positions with C2 symmetry. Our initial computer docking analysis with ERRα also predicted several other phytochemicals as the ligands of this receptor, such as 4′-hydroxy-α-naphthoflavone, 5,7,3′,4′-tetrahydroxyflavone, phloretin, and 7-hydroxy-3(4′-methoxyphenyl)-4-methylcoumarin. But these compounds failed to activate the receptor as indicated by our bioassay. All of the failed compounds either do not have hydroxyl groups at the two ends of the molecule or the hydroxyl groups do not line up symmetrically. Those two hydroxyl groups are predicted to provide key hydrogen bonding interactions with the protein to stabilize the complex formation. Ligands with bulky substituents, such as methyl or ethyl, on the flavone rings, which likely cause steric hindrance with the protein's bulky phenylalanine residues, were also found not to be ligands from our bioassay. As indicated above, for the purpose of generating an unbiased ligand-binding pocket for the screening of agonists of ERRα, three human ERα crystal structures with different agonist complexes were used as templates. Because ERα is known to have a larger binding pocket than ERRα, that model could accept ligands which were larger than the true ligands of this receptor. The templates that were used to generate the model had a direct impact on the results of virtual screening by docking. The inclusion of a water molecule in the ligand-binding site (see discussion below) helped us greatly by placing the ligands in a more definitive orientation. We believe that our ERRα model displays a significant degree of accuracy because we have been able to identify the true agonists (by bioassays) from the top 5% of our predicted list.
Our docking results showed that all of the four agonists were tightly packed due to the small size of the cavity, and deeply docked into the ligand-binding site, partially due to the termini water-bridged hydrogen bonds between hydroxyl group(s) of the ligands and the conserved water molecule. The hydrogen bonding groups on the two ends of the agonists form hydrogen bonds with residues Glu-235, Arg-276, and His-398 at the corresponding ends of the ligand-binding cavity (Fig. 3, B, D, E, and F). Phe-286 on sheet 1 was fixed in an orientation by its aromatic ring current interaction with the flat aromatic rings on the agonists. The three phenylalanines, Phe-232, Phe-399, and Phe-414, sit at the β face, while Phe-286 sits at the opposite α face of the cavity to sandwich the agonists (Fig. 3, B, D, E, and F). The hydrogen bonding, aromatic, and hydrophobic interactions between the ligands and helices 3, 5, 7, 11, and 12 of the receptor stabilized the complexes of ERRα with the phytoestrogens.
In summary, our modeling study indicated that the ERRα agonist-binding cavity is rather flat due to the phenylalanines in the ligand-binding site. This suggests that ligands with extended aromatic structures make more effective agonists. Our receptor functional analysis further indicates that an effective agonist should be a molecule without bulky substituents in the middle and with a pair of lined-up hydrogen bond forming groups at the ends. The derived pharmacophore information will enable us to identify additional agonists of ERRα by screening larger commercial three-dimensional chemical databases and verifying by mutagenesis and functional assays.
Discussion
Virtual Screening of the ERRα Agonists
SVS using a homology model has been proven to be a valuable technique for nuclear hormone receptors, an important therapeutic target family (20). Here, we identified for the first time that flavone and isoflavone phytoestrogens are agonists of all three isoforms of ERR through SVS and verified by mammalian cell transfection and two-hybrid functional assays. An important element of determining the success of the SVS is the choice of the specific agonist conformation of nuclear receptors used for screening. Our functional assay results validate the effectiveness of using the homology model generated from agonist-bound ERα X-ray complex structures to automatically screen more than 600 structurally diverse phytoestrogens and score using an effective consensus scoring approach. Our studies have provided key structural information to identify additional agonists of ERRα. We believe that our study represents an excellent example of an approach that combines SVS and receptor functional assays to identify novel ligands of orphan nuclear receptors.
This study was aimed at the identification of ERRα novel agonists and we presented a successful strategy to meet that goal. An in-depth evaluation of the agonist binding mode and structure-activity relationship requires extended work, such as molecular dynamics and mutagenesis experiments, and that is beyond the scope of this discussion. Enormous advances in genomics have identified a significant number of orphan nuclear receptors as potential therapeutic targets. We would like to place our emphasis on aspects of combining the reliable and inexpensive SVS technique with nuclear receptor functional assays for lead discovery.
Functional Significance of Phe-232 of ERRα
Our previous modeling study indicated that Phe-232 is responsible for the constitutive activity of ERRα (5). The current modeling study suggested that the Phe-232 could contribute to the protein activity by stacking with Phe-414 of H12 to help position and stabilize H12 in the agonist position and uphold the constitutive activity of ERRα. It was shown in the ERRα model that Phe-414 was sandwiched by stacking to Phe-232 of H3 and Phe-399 of H11 on each side (Fig. 3, B and C). A close inspection of the free ERRγ structure (1KV6, 6) indicates that the phenylalanine on H12 (Phe-450 in ERRγ) and the phenylalanine on H11 (Phe-435 in ERRγ), the two conserved phenylalanines in all three ERR isoforms, sit very close to each other and fall in the range of van der Waals interactions (about 2.5 Å proton-proton distance). A similar type of aromatic interaction has been observed recently from the structural analysis of the constitutively active orphan nuclear receptor Nurr1 (21). Together with results from previous studies (5), we hypothesize that the constitutive activity of ERRα results from (a) the side chain of Phe-232 mimicking bound ligand, and (b) helix 12 being kept at the agonist-bound conformation through an aromatic interaction among Phe-232 (H3), Phe-399 (H11), and Phe-414, which is in helix 12.
The receptor functional assays carried out in our laboratory clearly indicated that the mutation of one of the three stacked phenylalanines of ERRα, F232A, could significantly reduce the constitutive activity of the receptor. Furthermore, genistein and daidzein were found to be better agonists for the mutant F232A, while biochanin A and 6,3′,4′-trihydroxyflavone were more effective to activate wild-type ERRα (Fig. 7). This raised the question of how the mutation, F232A, affected the binding of each agonist and the protein activity in the presence of the agonist. Unlike the ER-E2 complex, in the ERRα-agonist system, ERRα has more bulky aromatic residues at its ligand-binding site (Fig. 2), and the agonists are molecules with extended aromatic ring structures (Fig. 1). These new structure features imply that aromatic interactions between the receptor and its agonists could play a very important role in ligand binding and receptor activity. Modeling of the mutant F232A (i.e., ERRαM) indicated that the side chain of Phe-414 took a different rotamer from the stacking position in ERRα to fill part of the space left out by Phe-232 in the mutant (Fig. 3, D and F). Phe-414 was then stabilized by aromatic interactions with Phe-399 of H11 (Fig. 3, D and F). H12 could be further stabilized by interacting with the agonists. The protein activity changes induced by the mutation are the consequences of the adjusted protein and ligand conformations. One of the many factors accountable for the changes could be the orientation change of the agonists in the ligand-binding site.
A detailed structure and activity analysis requires further work considering the very similar structural features of the four agonists. However, there are a few points that can be addressed from this study. It was observed that the agonists of ERRα were able to take two different docking orientations in the binding pocket. One is the classical orientation of the nuclear receptor ligand, where the phenolic ring sits at the A-ring end (Fig. 3, E and F) as in the E2-ERα complex (22). The alternate orientation is the flavone ring, instead, residing at this end (Fig. 3, B and D). It was shown that each agonist had its favored orientation in the binding site, depending on the specific structural characteristics of the agonist. Genistein and daidzein, which have only one functional group, the 4′-OH, on the phenolic ring, favor the flavone ring at the A-ring end in both the wild type and the mutant (Fig. 3, B and D) and are better agonists for the mutant. On the other hand, 6,3′,4′-trihydroxyflavone, which has an extra hydroxyl group on the phenolic ring and is a better agonist for the wild-type protein, tended to favor the phenolic ring at the A-ring end in both the wild-type and the mutant receptors (Fig. 3, E and F). Unlike 6,3′,4′-trihydroxyflavone, biochanin A, which acts as a better agonist for the wild-type protein and has a bulkier methoxyl group lacking the favored hydrogen donor for the A-ring end (23), kept its orientation with the flavone ring at the A-ring end in both the mutated and wild-type proteins. The preference of different orientations under specific circumstances was not well understood.
Implication of Physiological Significance of Our Findings of Phytoestrogens as Ligands of ERRα
As indicated in the introduction, it is not easy to identify agonists for a receptor that is constitutively active, because they have to be able to augment the basal activity. This work suggested that ERRα stays at a stable agonist conformation through the unique Phe-232 that mimics ligand and contributes to aromatic interactions with Phe-399 in helix 11 and Phe-414 in helix 12. It is important to point out that the binding of flavone and isoflavone ligands to ERRα increases its activity to the levels of ERα and ERβ in the presence of the same compounds at similar concentrations. These results suggest that ERRα has been stimulated to its maximal level in the presence of the ligands.
ERRα is expressed in breast tissue and its expression has recently been shown to associate with unfavorable biomarkers in breast cancer (24). The action of ERRα and enhancement of its activity by phytoestrogens in breast cancer should not be overlooked. In addition, ERRα mRNA has been found to be more highly expressed in rat calvaria (RC) cell cultures than either ERα and ERβ (25), and a decrease in the synthesis of ERRα (through the use of antisense oligonucleotides) led to an inhibition of RC proliferation and bone nodule formation in vitro (26). The interaction of phytoestrogens with ERRα can play important roles in breast and bone, a newly identified action of phytoestrogens. The effect of phytoestrogens on ERRα may not be easily observed when functionally active ERα or ERβ is present because these phytochemicals are also ligands of ERs. The effect of phytoestrogens (through the interaction with ERRα) in breast and bone in women could be detected when the function of ERs is suppressed, such as through the use of antiestrogens.
Our study has provided a novel approach to identifying environmental chemicals that interact with ERRα as agonists through the use of SVS and cell culture-based functional analyses. We will further screen larger commercial chemical databases for both agonists and antagonists of ERRα and verify these results by mutagenesis and functional assays. This research has a translational impact because the results generated can be used for designing prevention strategies against breast cancer. One should avoid exposure to those chemicals shown to act as agonists of ERRs that may induce breast cancer development. On the other hand, chemicals that act as antagonists of these receptors may have value as preventative agents against breast cancer. However, through the interaction with ERRα in bone, agonists of ERRα may prevent bone loss in women who use antiestrogens.
Materials and Methods
Materials
5,7-Dihydroxy-4′-methoxyisoflavone (biochanin A), 7,4′-dihydroxyisoflavone (daidzein), 5,7,4′-trihydroxyisoflavone (genistein), and 6,3′,4′-trihydroxyflavone were purchased from Indofine Chemical Co., Inc. (Somerville, NJ). The structures of these compounds are shown in Fig. 1. hERRβ cDNA and HeLa cervix adenocarcinoma cells were purchased from American Type Culture Collection (ATCC, Manassas, VA). HeLa cells were maintained in Eagle's MEM with nonessential amino acids, sodium pyruvate, and 10% fetal bovine serum at 37°C and 5% CO2. Charcoal/dextran-treated serum was obtained from Gemini Bio-Products (Woodland, CA). Lipofectin was purchased from Invitrogen Life Technologies (Palo Alto, CA).
Plasmid
All recombinant DNA and plasmid construction experiments were performed according to standard procedures. The sequence and orientation of inserted DNA fragments in plasmid constructs were verified by standard DNA sequencing. The expression plasmids, pSG5-ERRα and pSG5-ERRα mutant F232A, were constructed by Chen et al. (5). The full-length coding region of human ERRβ was generated by PCR using forward primer 5′-GCTGGAATTCATGTCGTCCGAAGACAGGCA-3′ and reverse primer 5′-TGCGGAATTCTCACA CCTTGGCCTCCAGCA-3′, with the template cDNA, hERR2 (ATCC). The PCR product was digested with EcoRI and subcloned into vector pSG5 through the EcoRI sites. The full-length coding region of human ERRγ was generated by PCR using forward primer 5′-CCGGGAATTCATGGATTCGGTAGAACTTTG-3′ and reverse primer 5′-GAGCGAATTCTCAGACCTTGGCCTCCAACA-3′, with the template cDNA prepared from H295 cells (ATCC). The PCR product was digested with EcoRI and subcloned into vector pSG5 through the EcoRI sites. The luciferase reporter plasmid, pGL3 (ERE)3-Luciferase, which contains three copies of the ERE sequence, was constructed by Chen et al. (5). The pSG5-GRIP1 plasmid was kindly provided by Dr. Michael R. Stallcup (University of Southern California, Los Angeles, CA), and pCI-ERα and pCI-ERβ were kindly provided by Dr. Y. Kinoshita (Beckman Research Institute of the City of Hope, Duarte, CA).
Generation of the expression plasmids for mammalian two-hybrid analysis, that is, PM-ERRα LBD, PM-ERRβ LBD, PM-ERRγ LBD, and pVP-PNRC coactivator fragment (aa 270–327), is briefly described below: The cDNA fragments of human ERRα LBD were generated by PCR using forward primer 5′-CCCCGAATTCACAGCAGCCCCCAGTGAATGC-3′ and reverse primer 5′-ACCCGGATCCTCAGTCCATCATGGCCTCGA-3′, with the template DNA, pSG5 ERRα. The cDNA fragment of human ERRγ LBD was generated by forward primer 5′-TCCCGACGCTAAAAAGCCATTGACTAA-3′ and reverse primer 5′-TGCGGGATCCTCACACCTTGGCCTCCAGCA-3′ with the template DNA, pSG5 ERRβ. Finally, the cDNA fragment of human ERRγ LBD was generated by forward primer 5′-GGTTGAATTCGCCAAAAAGCCATATAACAA-3′ and reverse primer 5′-GAGCGGATCCTCAGACCTTGGCCTCCAACA-3′, with the template DNA, pSG5 ERRγ. The PCR products were subcloned into vector pM through the EcoRI and BamHI sites. For pVP-PNRC270–327, the cDNA fragment of PNRC270–327 (27) was generated by PCR using forward primer 5′-GCCGGATCCTAATGACTGAAGTGAGCCAAAAGGAA-3′ and reverse primer 5′-CGCTCGGATCCCTAAGTTTGAACTTTTGAGGAG-3′, and the PCR product was inserted in proper reading frame into pVP16 activation domain vector (Clontech Laboratories, Inc., Palo Alto, CA) at the BamHI site.
Mammalian Cell Transfection and Luciferase Assays
HeLa cells were cultured in MEM Earle's salts medium supplemented with 5% charcoal/dextran-treated fetal bovine serum. Cells were divided and cultured in six-well plates until 80% confluent. The cells were transfected with 4 μg Lipofectin, and an equal amount of total DNA was used in all transfections by including appropriate amounts of the empty vector, pSG5, in addition to specific amounts of the test plasmids indicated in each experiment. After overnight incubation, medium containing Lipofectin and DNA was removed, and the cells were cultured in growth medium containing 5% charcoal/dextran-treated fetal bovine serum with or without ligands. After a 24-h incubation, the cells were harvested from the plates by scraping, and the luciferase activities in the cell lysates (with the same amounts of protein) were measured according to the manufacturer's instructions (Promega, Madison, WI). All experiments were performed in triplicate.
Mammalian Two-Hybrid Assays
HeLa cells were transiently transfected with 0.5 μg reporter plasmid, Gal4-luciferase (Clontech), and 0.5 μg PM-ERRα LBD, PM-ERRβ LBD, or PM-ERRγ LBD along with 0.5 μg pVP-PNRC270–327.
Building the Models of ERRα LBD
Three-dimensional homology modeling and SVS were performed using the SYBYL program package (11), version 6.9 (Tripos, Inc., St. Louis, MO) on a Silicon Graphics O2+ workstation with the IRIX 6.5 operating system. The sequence of hERRα was obtained from GenBank (NM_004451). The FASTA search option available in the Protein Data Bank (28) helped to identify several structural templates on which to base the homology model. The homology models of hERRα LBD and its mutant F232A were generated using the SYBYL COMPOSER module (29). Three different agonist-bound X-ray complex structures of the human ERα LBD templates (PDB codes 3ERD, 1L2I, 1GWR) were chosen because of their high resolution (2.03, 1.95, and 2.40 Å). The structures were then refined by torsional minimization and a series of energy minimization steps first with ligand-binding pocket side chains, followed by all protein side chains and finally the entire protein. All the energy minimizations were performed using the Tripos force field with the cutoff of nonbonded (NB) interactions at 8.0 Å and the distance dielectric constant set at 4.0 following the gradient termination of the Powell method with RMS of 0.005 kcal/mol Å or the maximum 1000 iterations. The volume of the ERRα ligand-binding pocket was estimated using SYBYL MOLCAD separated surfaces with a grid width of 1.0 and a probe radius of 1.4 Å.
Database Preparation
The Flavonoid and Coumarin Catalog was obtained from Indofine Chemical Company. Three-dimensional structures of the Indofine compounds were imported from the Available Chemicals Directory (ACD; Molecular Design Limited, San Leandro, CA) using ISIS/Base (MDL). For the known ligands of ERs and ERRs not already included in the Indofine Catalog, such as E2, DES, and 4-OHT, two-dimensional structures were prepared by ISIS/Draw (MDL). Three-dimensional structures of those molecules were generated using the Concord (Tripos) conversion program. The new structures were then added to the Indofine Catalog to create a final Indofine database of 603 molecules. Final coordinates were stored in a SYBYL database. A subset of this database, containing 37 molecules (known ERα agonists together with randomly chosen molecules), was created as a test set to fine tune parameters for docking the Indofine database and scoring the ligands for ERα (data not shown).
Receptor-Ligand Docking
The SYBYL FlexX program version 1.10 interfaced within TRIPOS SYBYL 6.9 (11) was used to dock compounds to the ligand-binding sites of ERRα and ERRαM. FlexX is a fast-automated docking program that considers ligand conformational flexibility into a rigid protein structure by an incremental fragment placing technique (30, 31). A structurally conserved water molecule has been included in the binding pocket for the docking. Standard parameters and FlexX scores implemented in the program were used for docking and scoring of FlexX poses. The ligand-protein complex was relaxed by torsional minimization and a series of constrained energy minimization steps. The ligand was first minimized within the complex to an RMS of 0.001 kcal/mol Å to remove bad contacts. The side chains of amino acids in the ligand-binding site and the entire complex were then minimized respectively. During the minimization processes, hydrogen bond constraints with 50 kcal/(mol Å)2 force constant were applied. Energy minimizations were carried out using the Tripos force field with an NB cutoff of 8.0 Å and the distance dielectric constant set at 4.0 following gradient termination using the Powell method to an RMS of 0.005 kcal/mol Å or the maximum 1000 iterations.
Virtual Ligand Screening of the Indofine Database
Each flexible ligand of the Indofine databases composed of flavonoids and coumarins was docked automatically into the receptor. The FlexX score was used to guide the growing of the ligand and was assigned to each successfully docked compound to measure the goodness of its fit with the receptor. The FlexX scoring function includes both polar (hydrogen bond and charge-charge) and non-polar (hydrophobic) interactions that are used to dock the ligand into the active site. The screening of the Indofine database of 603 compounds took less than 5 h on a Silicon Graphics O2+ workstation. Out of 603 screened compounds, 426 were claimed to be “successfully docked” and up to 30 top scored poses of each docked compound were saved. These 426 compounds were checked by an SPL (Sybyl Programming Language) script to find out whether these compounds were actually docked into the ligand-binding pocket (flexx_pocket_mss.spl). Two hundred twenty-seven out of these 426 “successful” compounds were found docked into the pocket. ChemScore, Dock, FlexX, Gold, and PMF scores were recalculated for the docked ligands using the CScore module of SYBYL 6.9 for consensus scoring. Consensus scoring has found to outperform any of the single functions comparing the results from the test set. A hit list containing the top scores from the three single scoring functions (PMF, FlexX, and Gold) which were able to successfully predict the non-binder such as E2 in the test set was generated. The top 100 ligands in the hit list were visually inspected and 50 of them were selected, purchased, and experimentally tested.
Acknowledgments
We thank Dr. Dujin Zhou and TRIPOS Application Scientists for their comments and suggestions to the research described here.
References
National Institutes of Health Grants ES08258 and CA44735.
Note: Masatomo Suetsugi and Leila Su contributed equally to this work.