Structural Insights into Corepressor Recognition by Antagonist-bound Estrogen Receptors*

Direct recruitment of transcriptional corepressors to estrogen receptors (ER) is thought to contribute to the tissue-specific effects of clinically important ER antagonists. Here, we present the crystal structures of two affinity-selected peptides in complex with antagonist-bound ERα ligand-binding domain. Both peptides adopt helical conformations, bind along the activation function 2 coregulator interaction surface, and mimic corepressor (CoRNR) sequence motif binding. Peptide binding is weak in a wild-type context but significantly enhanced by removal of ER helix 12. This region contains a previously unrecognized CoRNR motif that is able to compete with corepressors for binding to activation function 2, thereby providing a structural explanation for the poor ability of ER to directly interact with classical corepressors. Furthermore, the ability of other sequence motifs to mimic corepressor binding raises the possibility that coregulators do not necessarily require CoRNR motifs for direct recruitment to antagonist-bound ER.

The biological significance of corepressor recruitment by nuclear receptors (NRs) 5 in gene repression is well documented (reviewed in Ref. 1). Known NR corepressors have been isolated in complexes together with histone deacetylases, which facili-tate gene repression through deacetylation of histone tails (2). Even though the components of several complexes have been identified, less is known about the processes controlling their recruitment by NRs. Unlike most NRs, the estrogen receptor (ER) is unusual in that it does not appear to be repressed in the absence of hormone; consequently, the importance of NR corepressors in ER-mediated transcriptional signaling remains controversial. However, a number of recent studies have demonstrated that both agonist-and antagonist-bound ERs are able to recruit a variety of proteins that can repress its activity (3). Differential coregulator recruitment is also known to contribute to the tissue-specific effects of selective ER modulators (SERMs), a therapeutically important class of ER ligands that exhibit characteristics of both estrogens and anti-estrogens depending on the tissue (4,5).
Direct coactivator/corepressor recruitment by NRs is primarily mediated by the receptor's activation function 2 (AF2) located in the C-terminal ligand-binding domain (LBD) between helices H3, H5, and H12. Structural analyses have demonstrated that AF2 activity is dictated by the orientation of the mobile C-terminal AF2 activation helix (H12) (6). Agonist binding stabilizes an "active" orientation of H12 resulting in the formation of a specific binding site for the LXXLL interaction motifs of NR coactivators (7). NR antagonists prevent proper alignment of H12 and induce receptor conformations in which H12 is repositioned so that it occludes the AF2 binding site (7,8), is completely dissociated from the body of the LBD (9), or is bound to another site outside the AF2 region (10). In the latter cases, the AF2 binding site is accessible and can interact with the extended LXXXIXXXL CoRNR consensus motifs of NR corepressors.
Affinity-selected peptides that recognize SERM-bound ERs have been isolated from both random (11,12) and focused (13) peptide libraries. These studies have revealed a variety of hydrophobic sequence motifs that act as highly specific conformational probes and are good predictors of the biological effects of a particular ligand (14). Such motifs also provide information regarding potential ligand-specific ER-coregulator interaction sites (11,15,16). McDonnell and co-workers (12,13) have reported the isolation and identification of a number of short peptides using phage display that specifically recognize 4-hydroxytamoxifen (OHT)-bound ER. This study focuses on two such peptides. The OHT-specific ␣␤V peptide (SPG-SREWFKDMLS) was isolated from a random peptide library * 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. and contains a novel interaction motif (12). A second peptide, bT1 (hereafter referred to as CoRNR ER box; sequence DAFQL-RQLILRGLQDD) was isolated from a focused library based on the corepressor consensus motif (13).
To understand the structural basis for the interaction between ER-and SERM-specific motifs, we have crystallized and solved the structure of SERM-bound ER␣LBD in complex with both an affinity-selected CoRNR box peptide and a tamoxifen (OHT)-specific peptide. This study extends previous structural information on corepressor binding to NRs and provides novel insights into the specific recognition of the antagonistbound state of ER.
Cell Culture and Transient Transfections-Mammalian twohybrid experiments were performed as described in Ref. 15. HuH7 (human liver) cells were maintained in Dulbecco's modified Eagle's medium high glucose (Invitrogen) supplemented with 10% fetal bovine serum and 2 mM L-glutamine. For transient transfection, cells were seeded into 24-well plates 24 h before transfection in phenol red-free medium supplemented with 10% Dextran Charcoal-stripped fetal bovine serum and 2 mM L-glutamine. Cells were transfected using Lipofectamine 2000 according to the manufacturer's instructions (Invitrogen Corp.). After transfection, cells were treated with ligands for 16 h before assaying luciferase and ␤-galactosidase activity.
Surface Plasmon Resonance (SPR)-Measurements were performed using a Biacore X instrument and streptavidin-coated sensor chips. All experiments were carried out at 25°C in 50 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.05% Tween 20. 50 -120 response units of biotinylated ␣␤V peptide (biotin-SGSGPGSREWFKDML) was immobilized onto the chip surface. Qualitative binding experiments were performed by flowing liganded ER␣LBDs (1 M dimeric concentration) over the sensor chip for 2 min at 5 l min Ϫ1 . For the competition experiments, OHT-liganded ER␣⌬⌯12 was preincubated with CoRNR ER box peptide (DAFQLRQLILRGLQDD) at the desired molar ratio for 30 min prior to injection of the proteinpeptide mixture over the ␣␤V sensor chip. Preincubation with an LXXLL-containing peptide (EKHKILHRLLQDS) was used as a control.
Crystallography-A truncated ER␣LBD (ER␣⌬H12; residues 305-533) mutant was used to facilitate crystallization. OHTand RAL-liganded ER␣⌬H12 LBD was prepared as previously described (15). Peptide complexes were assembled by incubating protein with a 1.5-fold molar excess of peptide followed by concentration using ultrafiltration. Additional peptide was added to obtain a final peptide:LBD molar ratio of 3:1. Initial screening of crystallization conditions was performed at 19°C in a 300-nl, 96-well sitting drop format using a Mosquito liquid-handling robot (TTP Labtech). A single crystal of the ER␣⌬H12⅐RAL⅐CoRNR box complex was grown from a 300-nl drop containing an equal mixture of protein (7 mg/ml) and reservoir solution of 0.35 M (NH 4 ) 2 SO 4 , 0.7 M Li 2 SO 4 , 0.07 M tri-sodium citrate, pH 5.6. Crystals of the ER␣⌬H12⅐OHT⅐␣␤V complex were grown in hanging drops comprising equal volumes of protein (10 mg/ml) and reservoir solution of 2.5% (v/v) polyethylene glycol 550 monomethylether, 2.5% (w/v) polyethylene glycol 20000, 0.06 M calcium acetate, 0.1 M Tris, pH 8.5.
Crystals were cryoprotected by passing through a mother liquor solution supplemented with 25-30% (v/v) ethylene glycol prior to vitrification in liquid nitrogen. X-ray diffraction data were recorded on a Quantum-4 CCD detector at 100 K at the European Synchrotron Radiation Facility (Grenoble, France) and were processed using the HKL suite of programs (17). The structures were solved by molecular replacement with AMoRe (18) using the coordinates of ER␣⌬H12⅐OHT (15); Protein Data Bank code 2BJ4) as a search model. The peptides were clearly visible in the initial sigmaA-weighted electron density maps. Model building was performed with QUANTA (Accelrys, San Diego, CA), and the complexes were refined with REF-MAC (19) using appropriate non-crystallographic symmetry and translation/libration/screw parameter restraints. Data collection and refinement statistics are given in Table 1.
Analysis-Coordinates were superposed in QUANTA, and secondary structure assignment was carried out using DSSP. The peptide binding interfaces were analyzed with the help of the protein interfaces PISA server at the European Bioinformatics Institute (www.ebi.ac.uk/msd-srv/prot_int/pistart. html). Figs. 2 and 3, A-C, were generated using the program PyMol (Delano Scientific; pymol.sourceforge.net/).

RESULTS AND DISCUSSION
Corepressor Motifs Bind to a Common Site in ER-Previous studies have reported the isolation and identification of a number of short peptides using phage display that specifically rec-ognize OHT-bound ER (11)(12)(13). The OHT-specific ␣␤V peptide was isolated from a random phage display peptide library (12), whereas the CoRNR ER box peptide was isolated from a focused library based on the corepressor consensus motif (13). Although both peptides were affinity-selected using OHT-bound ERs, they exhibit differing abilities to interact with full-length ERs in a mammalian two-hybrid assay (Fig.  1B). The ␣␤V motif displayed a robust interaction with full-length ER in the presence of OHT comparable with 17␤-estradiol (E2)dependent LXXLL motif binding (Fig. 1B). In comparison, the CoRN-R ER box peptide provoked a much weaker reporter response. Removal of helix H12 from full-length ER (ER␣⌬H12) significantly enhanced binding of both peptides and had a dramatic effect on the CoRNR ER motif interaction (Fig. 1C). Furthermore, real-time binding analysis between ER-LBD and an immobilized ␣␤V peptide using SPR demonstrated that, in the context of the LBD alone, no interaction occurs unless H12 is removed (Fig. 1D). Competition studies using SPR indicated that both peptides target similar binding sites on the LBD surface, as preincubation of the LBD with one peptide reduced the binding to a peptide sensor chip (Fig. 1E). Similar binding behavior was observed with ER␤ (data not shown).
Previous studies by a number of groups have concluded that regions outside the LBD contribute to corepressor binding by NRs (10,(21)(22)(23). We also performed extensive domain deletion experiments as well as single site mutations to further investigate the architecture of the corepressor binding site (Fig. 5C and supplemental material). Domain mapping experiments suggest that the presence of ER's N-terminal AB domain is also required for interaction as constructs lacking this region exhibit no apparent binding to either CoRNR ER or ␣␤V peptide in our M2H assay. As removal of H12 potentiates corepressor binding (Fig. 1, C-E), the role of the AB domain appears to be indirect, presumably through an effect on the positioning of H12 in the intact receptor that alters the accessibility of the AF2 binding site in the SERM-bound state. Interestingly, even though structural evidence does not exist, there are data indicating a functional communication between distant receptor N-terminal domains and the LBD (24).
Structure Determination-Initial co-crystallization trials with peptide and either OHT-liganded ER␣or ER␤LBD did not yield any crystals suitable for structural studies. Based on the SPR observations that minimal peptide binding to the isolated LBD occurs in the presence of H12, an H12-truncated ER␣LBD (ER␣⌬H12) mutant, in which 21 amino acids at the C terminus of the LBD encompassing the H11-12 loop and H12 were removed, was used to facilitate crystallization of the Data presented as % activity with the peptide interaction with wild-type ER in the presence of OHT set to 100%. D, interaction of ␣␤V peptide with ER LBD using SPR. Sensorgrams shown were obtained from injection of 1 M wild-type or H12-truncated (⌬H12) ER␣LBD, liganded to either 17␤-estradiol or OHT, over a surface immobilized with ␣␤V peptide. E, competitive binding studies using SPR. Sensorgrams obtained from injection of 1 M ER␣⌬H12⅐OHT over an ␣␤V surface. ER␣⌬H12 was incubated with increasing molar ratios of CoRNR peptide prior to injection.  Table 1). Structure of ER␣⌬H12⅐RAL⅐ CoRNR ER Box Complex-The overall structure of the ⌬H12 LBD and the conformation of raloxifene (Fig. 1A, RAL) is identical to that observed in the full-length ER␣LBD⅐RAL complex (8). The CoRNR ER box peptide is bound along the AF2 groove between H3 and H5 as an extended, 3.5-turn amphipathic ␣-helix (Fig.  2). The peptide, which is 20 Å in length, occupies the entire AF2 cleft with its N terminus extending into the position normally occupied by H12 in its agonist orientation. The canonical Leu/Ile residues of the CoRNR ER motif (Leu5 (Lϩ1), Ile9 (Iϩ5), Leu13 (Lϩ9) lie along one face of the helix and are buried at the interface between the LBD and peptide (Fig. 2B). Contacts with the LBD are almost exclusively non-polar in character. The N terminus of the peptide is packed against the antagonist side chain of RAL. Lϩ1, Iϩ5, and Lϩ9 form a hydrophobic keel that anchors the peptide in the AF2 groove. Lϩ1 contacts Leu-354, Trp-383, and the terminal piperidine ring of the side chain of RAL. The central Iϩ5 makes contact with Ile-358, Val-376, and Leu-379. Lϩ9 is bound in a shallow depression formed by the side chains of Ile-358, Phe-367, Leu-372, and Val-376. These van der Waals interactions are complemented by two polar interactions. Lys-362 is hydrogenbonded to the main chain carbonyl group of Gly-12 and provides electrostatic stabilization to the C-terminal end of the CoRNR ER box. In addition, Glu-380 participates in a salt bridge with the side chain of Arg-6 (Fig. 2B).
Comparison of the ER CoRNR ER complex with the structure of PPAR␣LBD bound to antagonist and the ID2 CoRNR box motif of   (10)). B, ER␣ H12 (green) in its antagonist conformation (Protein Data Bank code 3ERT (7)). C, ␣␤V peptide (green). D, structure-based sequence alignment of corepressor motifs and ER␣ H12. Boxed regions indicate key contact points with the AF2 binding groove. The lengths of the various helical elements are depicted above the alignment. SMRT (10) reveals that the general principles of corepressor motif binding to the AF2 region of NRs are conserved. Nonetheless, several notable differences are apparent because of the differing surface topology of the AF2 regions of these two NR LBDs (Fig. 3A). The SMRT ID2 helix is shorter than the ER motif and is severely distorted at its N terminus so as to maintain favorable packing contacts with the AF2 cleft of peroxisome proliferators activated receptor (PPAR). In particular, differences in the amino acid composition of the H5/6 junction dictate that the Leuϩ1 (Leu-685) residue of the SMRT motif binds closer to the LBD surface. In addition, whereas the bulky side chain of RAL protrudes from the ligand binding cavity and interacts with Lϩ1 of the CoRNR ER motif, the bound antagonist in the PPAR-SMRT structure contributes very little to the immediate CoRNR binding surface.
Role of H12 in CoRNR Box Binding-The inhibitory properties of H12 on CoRNR box binding to ER (Fig. 1C) are readily apparent when one compares the interaction modes of these two elements. Both peptides interact with the LBD in a fashion similar to that observed for H12 in complexes of the intact ER␣LBD bound to SERM AF2 antagonists such as RAL and OHT (Fig. 3B). A structure-based sequence alignment clearly highlights the similarity between the affinity-selected CoRNR box motif and the ER␣ sequence in the vicinity of H12 (Fig. 3D). The CoRNR ER box motif's Leuϩ1, Ileϩ5, and Leuϩ9 perfectly mimic the equivalent interactions made by the Leu-536, Leu-540, and Leu-544 of H12. This observation may also explain the apparent inability of ER to bind CoRNR box sequences found in bona fide NR corepressors such as N-CoR and SMRT (13,25). In effect, ER possesses its own, highly effective CoRNR box surrogate within H12 that preferentially occupies the AF2 site in the presence of SERMs and passive antagonists. Consequently, H12 would need to be displaced from AF2 before any CoRNR box-mediated, corepressor binding could occur. Nonetheless, corepressors are understood to play a significant role in the biological effects of ER antagonists. SERM-bound ER␣ has been shown to be associated with N-CoR/SMRT in vivo (5,26,27); however, based on our study it seems highly unlikely that such associations are directly mediated through the AF2 region. Importantly, both N-CoR and SMRT have been isolated as part of multiprotein complexes (28,29) and are more likely to be recruited to ER␣ via indirect mechanisms that require additional factors.
H12 Length as a Predictor of Corepressor Binding-Examination of the sequences in the vicinity of H12 suggests that this mechanism to resist corepressor binding may be quite common within the NR superfamily (Fig. 4). NRs that exhibit poor corepressor binding, such as RXR, have a H12 sequence that resembles the CoRNR consensus motif and are able to adopt a relatively long amphipathic helix that, like ER, would be able to occlude the entire AF2 binding site (CoRNRϪ class in Fig. 4). In the case of DAX-1 and SHP there is strong support for a CoRNR box corepressor-independent repression (reviewed in Refs. 30,31). In contrast, NRs that exhibit good corepressor binding, such as thyroid receptor, peroxisome proliferators receptor, liver ϫ receptor, retinoid acid receptor (10,22,(32)(33)(34) exhibit much less similarity to the CoRNR box sequence. Critically, these NRs all possess a H12 sequence that is incompatible with the formation of a long AF2-blocking helix due to the presence of a proline residue that restricts the length of H12 (CoRNRϩ class). The resultant shorter H12 would constitute no barrier to corepressor binding as CoRNR box-containing corepressors are more likely to be able to displace H12 and bind along the AF2 groove of these NRs. Structural data to support such a mechanism to resist corepressor binding are limited as there are relatively few examples of crystal structures of antagonistbound NRs, other than with ER, in which H12 is observed to occupy the AF2 cleft. Nonetheless, in the structure of the retinoid ϫ receptor/retinoid acid receptor heterodimer (35), H12 of both partners lies in the antagonist position but the helix of retinoid ϫ receptor is considerably longer and buries 20% more accessible surface area than that of retinoid acid receptor. Consequently, corepressors are more likely to be able to displace retinoid acid receptor H12 and preferentially bind along its AF2 groove within this heterodimer. Notably, androgen receptor (AR), glucocorticoid receptor (GR), mineral-corticoid receptor (MR), and progesterone receptor (PR) contain an intermediate H12 length that does not harbor a complete CoRNR box-like sequence and would allow corepressor binding. In light of recent reports describing interactions ("direct" or "indirect") between NCoR/SMRT and these steroid receptors, it would be interesting to reinvestigate the possibility of direct recruitment to LBD (23, 36 -39). The role of sequences C-terminal to H12 is unknown, but long extensions may influence both H12 stability and coregulator access.
Structure of ER␣⌬H12⅐OHT⅐␣␤V Peptide Complex-Despite bearing little sequence homology to the CoRNR box consensus sequence, the OHT-specific ␣␤V motif acts as an effective Sequences are shown beginning at the proline residue (red) that typically defines the N-terminal extent of H12. The core region of H12 is highlighted. The corepressor CoRNR consensus motif is boxed. NRs that are known to be poor binders of corepressors show a strong similarity between the sequence of H12 and the CoRNR motif. Sequences are subdivided into two classes based on H12 length (CoRNRϪ and CoRNRϩ). Sequences indicated by asterisks have a C-terminal extension of either 20 amino acids (*) or more than 30 amino acids (**). structural mimic of a corepressor motif and binds along ER's AF2 groove in a similar fashion ( Fig. 2A). The ␣␤V peptide adopts a compact two-turn helix with a non-helical, N-terminal extension that sits atop the dimethylamino headgroup of the basic side chain of OHT (Figs. 3C and 5A). Whereas the CoRNR ER box peptide uses three hydrophobic Leu/Ile residues to bind to AF2, the ␣␤V motif utilizes only two. Phe-8 and Leu-12, which are characteristic of the ␣␤V class of motif (16), occupy analogous spatial positions to CoRNR ER box Iϩ5 and Lϩ9 residues. Trp-7 packs against the face of Phe-8, clamping the phenylalanine into a deep pocket in the AF2 groove (Fig.  5A). This structural feature appears to be important for the binding of the motif and explains why this position is always occupied by an aromatic residue (16). The helical nature of the ␣␤V motif is further stabilized by the side chains of Pro-2, Trp-7, and Met-11, which interlock and form a hydrophobic stack that interacts with the H3 edge of the binding groove (Fig.  5A). Analysis of other SERM-specific peptides reveals that the majority contain a CoRNR-like consensus sequence enriched with bulky hydrophobic residues replacing the canonical Leu/ Ile residues (14).
The ability of ␣␤V to adopt a shorter, compact ␣-helical structure compared with the CoRNR ER motif may account for its ability to efficiently compete with H12 for the AF2 site in full-length ER (Fig. 1B). Furthermore, the structure of the ␣␤V complex suggests that direct contact with the protruding side chain of OHT contributes to the observed ligand specificity of the peptide and provides a molecular basis for SERM-specific coregulator interactions. This ligand-dependent interaction is supported by mutagenesis data that show that either replacement of residues at the N terminus of ␣␤V that are in direct contact with OHT's terminal dimethylamino moiety, or mutation of the ER side chain involved in the positioning of the OHT side chain (Asp-351), abolish peptide interaction (Fig. 5B). The ER␣ mutation D351Y is the only naturally occurring receptor mutation found in a tumor cell line responsive to OHT-stimulated growth (40). This residue is suggested to be able to regulate both estrogenic and antiestrogenic properties of OHTcomplexed ER (20,41). Although the precise contribution of this residue to ER ligand pharmacology is not clear, it has been suggested that, by affecting the positioning of the basic side chain of OHT, this may provide an extra coactivator interaction site (41). An alternative explanation of the regulatory properties attributed to this residue is that its mutation leads to the disruption of a potential ␣␤V-like corepressor interaction surface.
To further investigate the specificity determinants of the novel ␣␤V motif, we performed alanine scanning mutagenesis and evaluated binding to full-length ER using a M2H interaction assay (Fig.  5B). Replacements of residues that lie on the hydrophobic face of the ␣␤V helix (Trp-7, Phe-8, Met-11, Leu-12) abolish binding. Similarly, mutations that disrupt the AF2 surface (L358R, L379R, L372R) or remove the "charge clamp" lysine (K362A) also abolished binding of both the CoRNR ER and the ␣␤V peptides (Fig. 5C). Intriguingly, mutation of Asp-351 has a differential effect on peptide binding and significantly enhances the interaction of the CoRNR ER box-containing motif while abolishing ␣/␤V binding (Fig. 5C).
Concluding Remarks-The two structures presented here demonstrate that the AF2 region of ER is, in principle, capable of interacting with coregulator proteins that recognize the SERM-bound conformational state of the receptor. However, the internal CoRNR box motif within ER␣'s H12 serves as an effective "corepressor surrogate" and provides a considerable barrier to binding. In addition, this study shows that amino acid sequences, other than the classical CoRNR box, can bind to the corepressor site of ER. These observations raise the possibility that cofactors with the binding characteristics of ␣␤V may exist in vivo and contribute to the effects of SERMs as this interaction motif, in contrast to CoRNR box-containing factors, is readily recruited to full-length antagonist-bound ER.