Characterization of the Interaction between Retinoic Acid Receptor/Retinoid X Receptor (RAR/RXR) Heterodimers and Transcriptional Coactivators through Structural and Fluorescence Anisotropy Studies*

Retinoid receptors (RARs and RXRs) are ligand-activated transcription factors that regulate the transcription of target genes by recruiting coregulator complexes at cognate promoters. To understand the effects of heterodimerization and ligand binding on coactivator recruitment, we solved the crystal structure of the complex between the RARβ/RXRα ligand-binding domain heterodimer, its 9-cis retinoic acid ligand, and an LXXLL-containing peptide (termed NR box 2) derived from the nuclear receptor interaction domain (NID) of the TRAP220 coactivator. In parallel, we measured the binding affinities of the isolated NR box 2 peptide or the full-length NID of the coactivator SRC-1 for retinoid receptors in the presence of various types of ligands. Our correlative analysis of three-dimensional structures and fluorescence data reveals that heterodimerization does not significantly alter the structure of individual subunits or their intrinsic capacity to interact with NR box 2. Similarly, we show that the ability of a protomer to recruit NR box 2 does not vary as a function of the ligand binding status of the partner receptor. In contrast, the strength of the overall association between the heterodimer and the full-length SRC-1 NID is dictated by the combinatorial action of RAR and RXR ligands, the simultaneous presence of the two receptor agonists being required for highest binding affinity. We identified an LXXLL peptide-driven mechanism by which the concerted reorientation of three phenylalanine side chains generates an “aromatic clamp” that locks the RXR activation helix H12 in the transcriptionally active conformation. Finally, we show how variations of helix H11-ligand interactions can alter the communication pathway linking helices H11, H12, and the connecting loop L11-12 to the coactivator-binding site. Together, our results reveal molecular and structural features that impact on the ligand-dependent interaction of the RAR/RXR heterodimer with nuclear receptor coactivators.


Retinoid receptors (RARs and RXRs) are ligand-activated transcription factors that regulate the transcription of target genes by recruiting coregulator complexes at cognate promoters.
To understand the effects of heterodimerization and ligand binding on coactivator recruitment, we solved the crystal structure of the complex between the RAR␤/RXR␣ ligand-binding domain heterodimer, its 9-cis retinoic acid ligand, and an LXXLLcontaining peptide (termed NR box 2) derived from the nuclear receptor interaction domain (NID) of the TRAP220 coactivator. In parallel, we measured the binding affinities of the isolated NR box 2 peptide or the fulllength NID of the coactivator SRC-1 for retinoid receptors in the presence of various types of ligands. Our correlative analysis of three-dimensional structures and fluorescence data reveals that heterodimerization does not significantly alter the structure of individual subunits or their intrinsic capacity to interact with NR box 2. Similarly, we show that the ability of a protomer to recruit NR box 2 does not vary as a function of the ligand binding status of the partner receptor. In contrast, the strength of the overall association between the heterodimer and the full-length SRC-1 NID is dictated by the combinatorial action of RAR and RXR ligands, the simultaneous presence of the two receptor agonists being required for highest binding affinity. We identified an LXXLL peptidedriven mechanism by which the concerted reorientation of three phenylalanine side chains generates an "aromatic clamp" that locks the RXR activation helix H12 in the transcriptionally active conformation. Finally, we show how variations of helix H11-ligand interactions can alter the communication pathway linking helices H11, H12, and the connecting loop L11-12 to the coactivatorbinding site. Together, our results reveal molecular and structural features that impact on the ligand-dependent interaction of the RAR/RXR heterodimer with nuclear receptor coactivators.
Retinoids are the active metabolites of vitamin A that regulate complex gene networks involved in cell differentiation, proliferation, and apoptosis (1)(2)(3)(4). Notably, they are effective inhibitors of tumor cell growth, thus supporting their actual or potential use in cancer therapy and prevention (5)(6)(7). The biological activity of all-trans and 9-cis retinoic acid (9C-RA) 1 signals are mediated by two families of nuclear receptors that act as ligand-dependent transcriptional regulators: the retinoic acid receptors (RAR␣, -␤, and -␥) and the retinoid X receptors (RXR␣, -␤, and -␥). Although RARs bind and are activated by both the 9-cis and the all-trans isomers of the retinoic acid, RXRs are exclusively activated by the 9-cis isomer (8 -10). Like other members of the nuclear receptor superfamily, retinoid receptors are modular and include an N-terminal A/B activation domain, a central DNA-binding domain (region C), a linker region D, a ligand-binding domain (region E), and a C-terminal region F of unknown structure and function. The ligand-binding domain (LBD) is a multifunctional domain, capable of ligand binding, dimerization, and interaction with transcriptional coregulators.
Previous studies have shown that retinoid actions are predominantly mediated through the formation of heterodimers between the various RAR and RXR isotypes. The RAR/RXR heterodimers can operate as either transcriptional repressors or activators. The silencing activity involves the establishment of transcription repressing complexes in which corepressors are recruited to receptor heterodimers in the absence of ligand (apo-receptors) (11)(12)(13)(14)(15). Binding of agonists (holo-receptors) triggers a mechanism by which the most C-terminal LBD helix H12, encompassing the core of the functionally conserved activation function 2 (AF-2), is repositioned in a so-called active-or holo-conformation, thereby creating a binding surface that facilitates interaction with coactivator proteins and concomitantly decreases the affinity for corepressors (16 -18). Conversely, binding of antagonists prevents the formation of this specific surface by interfering directly or indirectly with the holo-conformation of helix H12 (16,19). Transcriptional coactivators including those of the SRC-1/p160 family and the TRAP complex are involved in chromatin remodeling or recruitment of the basal transcription machinery and interact with nuclear receptors via short LXXLL helical motifs, socalled NR boxes, present as multiple copies in their nuclear receptor interaction domain (NID) (20,21). The ligand-induced exchange of corepressor and coactivator complexes is believed to underlie the molecular mechanism controlling target gene repression and activation (22,23).
In an effort to gain new insights into the molecular mechanism of the RAR/RXR-coactivator interaction, we solved the crystal structure of the RAR␤/RXR␣ LBD heterodimer in complex with 9C-RA and a 13-residue fragment containing the NR box 2 of the coactivator TRAP220. Furthermore, we used fluorescence anisotropy to characterize in a quantitative manner the interaction between the transcriptional coactivator SRC-1 and monomeric or heterodimeric retinoid receptors bound to a variety of ligands. Based on these structural and biophysical studies, we propose a model of the RAR/RXR-SRC-1 interaction to understand how dimerization and ligand binding influence coactivator recruitment.

EXPERIMENTAL PROCEDURES
Peptides and Ligands-The human SRC-1 570 -780 coactivator fragment was purified and labeled with Alexa Fluor 488 as described (24). Fluorescent peptide (fluorescein-␤A-686 RHKILHRLLQEGS 698 ) corresponding to the NR box 2-binding motif of SRC-1 was purchased from Neosystem (Strasbourg, France). Unlabeled peptides corresponding to NR box 1 ( 629 TSHKLVQLLTTTA 641 ), NR box 2 ( 686 RHKILHR-LLQEGS 698 ), NR box 3 ( 745 DHQLLRYLLDKDE 757 ) of SRC-1, and NR box 2 of human TRAP220 coactivator ( 641 NHPMLMNLLKDNPA 654 ) were prepared on a Pioneer peptide synthesis system from Applied Biosystems (Solid Phase Peptide Synthesis) using 2-(1H-benzotriazole-1-yl)-1,1,3,3-tetramethyluronium tetrafluoroborate (TBTU) activation for Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry; coupling was carried out with a 4-fold excess of activated amino acid for a minimum of 30 min. The peptides were synthesized on Pal linker polyethyleneglycol (Pal-Peg-PS) resin (Applied Biosystems). After synthesis, peptides were treated with a mixture of 88% trifluoroacetic acid, 5% water, 5% phenol, and 2% triisopropylsilane for cleavage from the resin and side-chain deprotection and purified on analytical high pressure liquid chromatography using Waters C-18 columns (Waters). Finally, the mass and purity were checked by electrospray ionization mass spectrometry. 9C-RA was purchased from ICN Biomedicals. AM80 and TTNPB were kindly provided by Patrick Balaguer (INSERM U540, Montpellier, France). BMS614 and CD3254 were kindly provided by Chris Zusi (Bristol-Myers Squibb, Princeton, NJ) and Galderma (Sophia-Antipolis, France), respectively. UVI3003 was synthesized and characterized. 2 Protein Expression, Purification, and Crystallization-The histidinetagged LBD of mouse RAR␤2 (residues 146 -448 in pET15b vector) and the LBD of mouse RXR␣1 (residues 227-467 in pET3a vector) were expressed in Escherichia coli BL21(DE3), copurified, and crystallized as an RAR/RXR heterodimer as described previously (25). Briefly, after a first step of a nickel affinity column (HiTrap chelating column, Amersham Biosciences), the fractions containing the eluted heterodimer were pooled and incubated overnight at 4°C with 5-fold molar excess of 9C-RA and thrombin to remove the His tag. The protein was further purified using a Superdex 75 26/60 gel filtration column (Amersham Biosciences) and a buffer containing 10 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 5 mM dithiothreitol. The fractions were analyzed by SDS-PAGE, pooled, and mixed with 2-fold molar excess of 9C-RA and 3-fold molar excess of TRAP220 NR box 2 peptide ( 641 NHPMLMNLLKD-NPA 654 ). The purified complex was concentrated to 15 mg⅐ml Ϫ1 and centrifuged for 30 min at 13,000 rpm prior to crystallization assays. Crystals with unit cell parameters a ϭ b ϭ 115.68 Å and c ϭ 247.19 Å contain two heterodimers per asymmetric unit and belong to the P3 1 21 space group. They were grown from 100 mM sodium formate and 20% polyethylene glycol 3350. The crystals grew in 4 days to a maximum final size of 0.45 ϫ 0.25 ϫ 0.25 mm and appear bipyramidal with a hexagonal base.
Data Collection, Structure Solution, and Refinement-The protein crystal was mounted from the mother liquor onto a cryoloop (Hampton Research) and then sequentially soaked in the reservoir solution containing an additional 5-25% glycerol in five steps and finally quickly frozen in liquid nitrogen. Diffraction data ( ϭ 0.9797 Å) were collected at 100 K using a MAR CCD detector at the French beamline for Investigation of Proteins (BM30A) at European Synchrotron Radiation Facility (ESRF) (Grenoble, France). The crystal diffracted beyond 2.6 Å with useful data to 2.9 Å. Diffraction data were processed using MOS-FLM (26) and scaled with SCALA from the CCP4 suite of programs (27). Data collection statistics are given in Table I. The structure was solved by molecular replacement, using the MOLREP software from the CCP4 package, with a truncated version of the RAR␣/RXR␣ LBD heterodimer (28) as the search model. In this model, all water and ligand molecules were removed as well as the RAR␣ and RXR␣ C-terminal helices H12 that were expected to adopt different conformations in the two dimers. The phases from the molecular replacement solution were refined with solvent flattening and 2-fold non-crystallographic symmetry averaging as implemented in the CCP4 dm program. The produced map was very clear for the RAR␤ and the RXR␣ LBDs, the TRAP220 peptides, and the 9C-RA ligands. Multiple cycles of manual model building, including conversion of side chains from RAR␣ to RAR␤ sequence, were carried out with the program O (29). Structure refinements were processed with CNS (30), using the maximum likelihood target and non-crystallographic summary constraints. The statistics of the structure refinement are summarized in Table I. The stereochemical quality of the model was assessed by use of the program PROCHEK (31). Fig. 1B was generated using SETOR (32). Figs. 1A, 2A, 4, and 5A were generated using PyMOL (33).
Fluorescence Assays-Fluorescence anisotropy assays were performed using a BEACON 2000 polarization instrument (Panvera Corp.) regulated at 4°C, using filters for fluorescein at a peptide concentration of 2 nM. The buffer solution for assays was 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM EDTA, 5 mM dithiothreitol, and 10% (v/v) glycerol. All ligands (Table II) were present at concentrations above their respective K d values for binding to RAR and RXR and also at sufficient concentration to saturate both receptors. For the fluorescent peptide, due to the rapid establishment of the equilibrium, the measurements were initiated at the highest concentration of protein, and then the sample 2 A. de Lera and H. Gronemeyer, manuscript in preparation.
d R free is as R cryst but calculated over 3% of data that where excluded from the refinement process. e r.m.s.d. is the root-mean-square deviation from ideal geometry.
was diluted successively by a factor of 0.75 with the buffer containing 2 nM of fluorescent peptide, allowing us to establish the titration curve. In contrast, for titration of the labeled SRC-1 570 -780 fragment, samples corresponding to each point of the titration curve were prepared in advance and incubated for 1 h. In all cases, for each titration point, anisotropy was measured successively until stabilized, and the reported values are the average of four measurements after stabilization. Binding data were analyzed using the package BIOEQS (34). Binding profiles were fit using a simple model assuming the stoichiometry of one SRC-1 by species (RAR/RXR heterodimer or RXR and RAR monomers).
To determine affinities of other coactivator peptides for RAR and RXR, we used a strategy of competition. Prior to performing the assays, we estimated the ideal working concentrations of both protein partner and competitor peptide by simulating curves with BIOEQS. Finally, the fluorescent peptide was competed with increasing concentrations of unlabeled competitor peptide in the presence of a constant concentration of partner receptor. AM80 and CD3254 were used in the titrations as reference agonist ligands for RAR␣ and RXR␣, respectively. All K d values are reported in Tables III and IV. Transactivation Assay-Cos1 cells plated in 24-well plates were transfected as described previously (35) using phosphate calcium precipitates. Ten nanograms of expression vectors pSG5-mRXR␣ wild type, F282A, F442A, and F443A were transfected together with 100 ng of DR1-tk-CAT reporter construct and 25 ng of CMV-␤-galactosidase used as internal control to normalize for transfection efficiency. The total quantity of DNA was adjusted to 1 g. 15 h after transfection, cells were washed with phosphate-buffered saline and supplemented with fresh medium with or without 9C-RA for an additional 24 h. Intracellular levels of CAT were measured using a CAT enzymelinked immunosorbent assay kit (Roche Applied Science) according to the manufacturer's recommendations.

RESULTS AND DISCUSSION
Structure Determination-The purified heterodimer of E. coli-expressed LBDs of mouse RAR␤2 (fragment 146 -448, containing the entire D, E, and F regions) and mouse RXR␣1 (fragment 227-467, containing the entire E region) was crystallized in the presence of 9C-RA and a synthetic peptide containing the second LXXLL motif (NR box 2) of the NID of the human TRAP220 coactivator (fragment 641-654). The structure was solved by molecular replacement using a truncated version of the previously determined RAR␣/RXR␣ heterodimer structure (see "Experimental Procedures") as the search probe. Two solutions were obtained from the molecular replacement search with a correlation coefficient of 0.269 (next highest solution 0.189) and an R-factor of 0.575 consistent with two complexes in the asymmetric unit. The final solution compris-ing two heterodimers had a correlation coefficient of 0.472 (next highest solution 0.240) and an R-factor of 0.492. As the unliganded and H12-deleted heterodimer had been used for phasing, the very clear densities for the ligand in RAR␤ and RXR␣ LBDs and the continuous electron density observed for helix H12 of the two protomers confirmed the accuracy of the solution. The asymmetric unit contains two heterodimers, with each subunit binding one TRAP220 NR box 2 peptide and one ligand. Helix H2, the connection between H2 and H3 in RXR␣, the D-region and the last 30 residues of the F-region in RAR␤, could not be modeled due to poor electron density in these regions. The final model, refined to 2.9-Å resolution, comprises 462 protein and 21 peptide residues, 2 ligands, and 5 water molecules. The collection data and refinement statistics are summarized in Table I. The final R cryst and R free were 0.253 and 0.296, respectively, using all the data in the resolution range of 30 -2.9 Å. PRO-CHECK analysis (31) showed that there were no residues in the disallowed regions of the Ramachandran plot. The geometry was very good, and the overall G factor was 0.17.
Overall Structure of the RAR␤/RXR␣ Heterodimer-The RAR␤ and RXR␣ LBDs adopt the canonical three-layered ␣-helical sandwich structure seen in other nuclear receptor LBD structures with the ligands buried within the core of the proteins (17,36,37). A ribbon representation of the complex is shown in Fig. 1A. In contrast to the RAR␣/RXR␣ heterodimer (28), both protomers of the 9C-RA-bound RAR␤/RXR␣ heterodimer adopt the active conformation with the "activation helix" H12 positioned against the protein, thereby essentially sealing the ligand-binding pockets (LBPs) of each protomer. This conformation generates a recognition surface constituted by mostly hydrophobic residues from helices H3, H4, and H12 of RAR␤ and RXR␣, which allow one TRAP220 coactivator peptide to bind per protomer. The TRAP220 peptide folds as a two-turn amphipathic ␣-helix with the hydrophobic side chains packed against the agonist-induced RAR␤ and RXR␣ surfaces. In principle, the overall RAR␤/RXR␣ heterodimeric arrangement closely resembles that of RAR␣/RXR␣ (28), peroxisome proliferator-activated receptor ␥ (PPAR␥)/RXR␣ (38), and LXR␣/RXR␤ (39) heterodimers with residues from helices H7, H9, H10, as well as loops L8 -9 and L9 -10 of each protomer forming the interface and a nearly two-fold axis relating the two subunits. The binding mode of 9C-RA to RXR␣ is identical to that observed previously (38,40). Briefly, the ligand is completely buried within the ligand-binding pocket formed by residues located on helices H3, H5, H7, H11, and the ␤-turn (Fig.  1B, left panel). The majority of the contacts are van der Waals interactions except for the carboxylate group of the ligand that forms a salt bridge with the conserved Arg 321 in helix H5 and a hydrogen bond with the backbone amide of Ala 332 of the ␤-turn. Similarly, the binding mode of 9C-RA to RAR␤ (Fig. 1B,  right panel) closely resembles that already observed in RAR␥ (41) and involves the formation of a salt bridge between Arg 269 of H5 and the carboxylate group, which in addition participates in a network of hydrogen bonds with Ser 280 (␤-turn) and a water molecule. The remaining residues of the LBP are hydrophobic and belong to helices H1, H3, H5, H11, H12, the connecting loop L6 -7, and the ␤-turn. Electron density was observed for only 10 residues (out of 40) of the RAR␤ C-terminal F region. These residues adopt an extended loop conformation projecting toward RXR␣, establishing some interactions with helices H7 and H11 as well as loop L6 -7 (Fig. 1A). In particular, Glu 414 is engaged in two hydrogen bonds with the backbone amides of RXR␣ Ala 349 and Ile 350 , whereas Pro 415 makes van der Waals contacts with His 440 , Phe 443 , and Phe 444 from H11 and Gly 346 of L6 -7 (not shown). However, the proximity of crystallographic or non-crystallographic symmetry-related molecules stabilizing the C-terminal extension in the observed conformation strongly suggests that this RAR-RXR interaction is crystal packing-dependent. If this interaction would indeed occur in solution, it should stabilize the active conformation of RAR␤ helix H12 (in the absence or in the presence of agonist) and therefore modify coregulator recruitment by RAR␤. However, no significant differences were observed for the binding affinities of fluorescent SRC-1 NR box 2 peptide to RAR␤ with or without the F-region (Table IV).
The LXXLL-binding Site on RAR␤ and RXR␣-The binding mode of the LXXLL motif to RAR␤ and RXR␣ LBDs is similar to that described in other LBD-coactivator peptide complex structures (38,42,43). Briefly, TRAP220 peptides are held in place through interactions of their leucine residues (Leu 645 , Leu 648 , and Leu 649 ) with the hydrophobic groove generated by the C-terminal part of H3, the loop L3-4, H4, and H12 of RAR␤ and RXR␣ (Figs. 1A and 2A). In addition, each receptor shows two conserved residues that are hydrogen-bonded to a mainchain peptide bond of the LXXLL motif and generate a charge clamp that defines the precise length of the helical motif that can be docked to the cleft (not shown). These amino acids are a lysine at the C terminus of H3 (Lys 237 and Lys 284 in RAR␤ and RXR␣, respectively) and a glutamate in H12 (Glu 405 and Glu 453 in RAR␤ and RXR␣, respectively). Binding affinities of the interactions between the TRAP220 NR box 2 peptide and agonist-bound RAR␣ and RXR␣ are reported in Table III.
Interestingly, comparison of RXR holo-structures obtained in the absence or in the presence of coactivator NR box highlighted the important role played by three RXR phenylalanine residues during the peptide binding process. In the unbound receptor, Phe 282 (H3), Phe 442 (H11), and Phe 455 (H12) are almost completely solvent exposed and make very little or no contact at all with other LBD residues ( Fig. 2A). Upon NR box binding, we observed a concerted reorientation of the three  1. Overall structure of the heterodimer and the ligand-binding pocket interactions. A, two orthogonal views of the heterodimer. On the left panel, the complex is viewed perpendicular to the dimer axis, whereas on the right panel, it is viewed along the dimer axis. The RXR␣ and the RAR␤ subunits are shown in green and blue, respectively, with the activation helices H12 depicted in red. The two TRAP220 NR box 2 peptides are colored in orange. The two ligands are in space-filling representation, with carbons and oxygens colored in yellow and red, respectively. Selected secondary structural elements are labeled according to the nomenclature of other nuclear receptors. "F " denotes the 10 residues of the RAR␤ C-terminal F region that could be modeled. B, 9C-RA interactions with the RXR␣ (left panel) and RAR␤ (right panel) LBDs. The 2F o Ϫ F c annealed "omit" map for the 9C-RA molecules was calculated in the absence of the ligand and contoured at 1.0 . The protein C␣ traces are represented as gray thick lines. The ligand-contacting side chains (carbon, yellow; oxygen, red; nitrogen, blue; and sulfur, green) and the ligands (green in RXR␣ and blue in RAR␤) are represented by thick lines. The secondary structural elements as well as the residues delineating the ligand-binding pockets are labeled. phenylalanine side chains that brings them in close contact to H12 and the peptide. Due to steric restrictions applied by Leu 648 of the peptide, Phe 282 rotates around the C␣-C␤ bond by 110°and forces Phe 455 from H12 to pivot by 104°. A rigid-body translation of helix H12 toward H3 and the coactivator peptide of almost 2.0 Å allows Phe 455 to be accommodated in a small, well suited hydrophobic pocket composed of residues from helices H3 (Phe 282 , Leu 281 , Leu 285 ), H12 (Met 459 ), and the peptide (Leu 645 , Met 644 , Leu 648 ). Finally, Phe 442 (H11) flips by 112°to contact Leu 460 (H12) and forms together with Phe 282 of H3 an "aromatic clamp" that locks H12 in the optimal conformation. Phe 442 and Phe 443 (H11) were previously found as being important for the stabilization of the unliganded form of RXR by partially filling the ligand-binding pocket of the apo-structure (44,45). Single point mutations show that substitution of Phe 443 with alanine has no effect on transcriptional activity of RXR homodimers in response to increasing concentrations of 9C-RA (Fig. 2B). In vivid contrast, disrupting the aromatic clamp by mutating Phe 442 or Phe 282 (H3) severely impairs maximal transactivation efficacy of 9C-RA-bound RXR homodimers, without modifying 9C-RA EC 50 . These structural and functional data demonstrate the importance of the aromatic clamp for formation and stabilization of the transcriptionally active RXR conformation and highlight the pivotal role of Phe 442 in maintaining the integrity of both the apo-and the holoRXR structures.
Ligand Binding and Coactivator Recruitment-We used fluorescence anisotropy to quantify the ligand-dependent interaction between RXR␣, RAR␣, and SRC-1. We first determined the affinities of all three NR boxes of SRC-1 for RAR␣ and RXR␣ bound to their respective synthetic agonists AM80 and CD3254 (Table II). Fluorescein-labeled NR box 2 peptide was first titrated with either RAR␣ or RXR␣, and its affinity was determined to be 0.47 (K d ) and 1.87 M (K d ), respectively (Table IV). These complexes were then competed with unlabeled NR box 1, 2, and 3 peptides. As shown in Fig. 3A, the NR box 2 motif competes with itself and exhibits the highest affinity for RAR␣ and RXR␣ followed by NR box 3 and finally NR box 1, thereby confirming previous studies demonstrating that the presence of an intact NR box 2 in the NID of TIF-2 is sufficient for both efficient interaction with holoLBDs and stimulation of AF-2 activity, whereas NR boxes 1 and 3 are poorly efficient on their own (46).
To analyze the effect of ligands on the recruitment of coactivators by RAR␣ and RXR␣, we measured the affinity of the interaction between monomeric receptors and either fluorescein-labeled NR box 2 peptide or the Alexa Fluor 488-labeled NID of the coactivator SRC-1 containing all three NR boxes (amino acid residues 570 -780; SRC-1 570 -780 ). Titrations were carried out in the absence of ligand and in the presence of agonists (AM80 and CD3254 for RAR␣ and RXR␣, respectively) or antagonists (BMS614 and UVI3003 for RAR␣ and RXR␣, respectively) (Table  II). Fig. 3B shows that apoRAR␣ binds SRC-1 fragments with higher affinity than does apoRXR␣. In keeping with their agonistic properties, the binding of AM80 or CD3254 to their respective receptors strongly enhances the binding affinity of both the SRC-1 NR box 2 peptide and the SRC-1 570 -780 to RAR␣ and RXR␣; note that RAR␣-AM80 is slightly more efficient than RXR␣-CD3254 in recruiting the coactivator peptides. Interestingly, regardless of the receptor type (RAR␣ or RXR␣), no significant difference in affinity is observed between the NR box 2 peptide and the full NID fragment, suggesting very similar binding modes. As expected, the binding of the antagonists BMS614 or UVI3003 to their specific receptors further reduces the binding affinity of SRC-1 NR box 2 to RAR␣ and RXR␣ when compared with the apo-receptors. Overall, it appears that the short 13residue-peptide SRC-1 NR box 2 is sufficient to reproduce the essential aspects of the behavior of the entire SRC-1 NID fragment for monomeric receptor binding.
Heterodimerization and Coactivator Recruitment-Given the importance of heterodimerization in the retinoid signaling pathway, we addressed the question whether association of RAR and RXR LBDs modifies the intrinsic ability of the receptors to interact with coregulators. To this end, we examined coactivator binding by RAR␣ and RXR␣ in the context of the heterodimer and in response to a variety of ligands IV Dissociation constants of SRC-1 fragments complexed with monomeric or heterodimeric retinoid receptor LBDs Experiments were carried out using the fluorescein-labeled NR box 2 peptide or the Alexa Fluor 488-labeled NID of the coactivator SRC-1 ( fluo SRC-1 NR2 and A488 SRC-1 570 -780 , respectively). Uncertainties on the recovered dissociation constants were obtained by rigorous confidence limit testing. This approach takes into consideration all of the possible correlations between parameters in the fit. ( Table II). RAR␣/RXR␣ heterodimers liganded with the RXRselective antagonist UVI3003 and the RAR-selective agonist AM80 ( Fig. 3C; AM80/UVI3003) recruit SRC-1 NR box 2 and SRC-1 570 -780 with binding affinities almost identical to those observed for the monomeric RAR␣-AM80 complex (Fig. 3B), indicating that RXR␣ antagonist conformation does not affect the capacity of holoRAR␣ to recruit coactivators. Likewise, SRC-1 570 -780 or its isolated NR box 2 displayed similar affinities for CD3254-liganded RXR␣ as monomer (Fig. 3B) or as heterodimer with antagonist-bound RAR␣ ( Fig. 3C; BMS614/ CD3254). Together, these results demonstrate that, as far a coactivator recruitment is concerned, RAR␣ and RXR␣ act, within the heterodimer, independently from each other, like their monomers, which is in full agreement with the observation of only minimal structural alteration of the receptors upon dimerization. Indeed, except for some necessary sidechain reorientations at the dimerization interface, the structural superposition of the monomeric RAR␤ (47)  It has been previously observed that RAR and RXR ligands act synergistically (48) and that RAR/RXR heterodimers can cooperatively interact with the NR boxes of TIF2 (49). Indeed, heterodimers in which both RAR␣ and RXR␣ are bound to selective agonists show enhanced binding affinity of SRC-1 570 -780 (Fig. 3C; compare AM80/CD3254 with the other ligand combinations), but the affinity of the isolated SRC-1 NR box 2 remains unchanged in the presence of these various ligands. This result underscores the above observation that heterodimerization does not modify the intrinsic capacity of individual receptors to interact with single LXXLL motifs and supports earlier observations suggesting that the synergy between receptor agonists could result from the formation of two binding sites on the heterodimer surface, which would interact with two LXXLL motifs of the same coactivator molecule (24,43,49). Furthermore, it appears that within the heterodimer, unliganded RAR and RXR contribute differently to the interaction with the coactivator. Although apoRAR is able to synergize with CD3254 ( Fig. 3C; apo/CD3254) to strongly enhance the overall binding of SRC-1 570 -780 as compared with SRC-1 NR box 2 or to the heterodimer bound to the RAR antagonist BMS614 (BMS614/ CD3254), the similar recruitment of SRC-1 NR box 2 and SRC-1 570 -780 by the AM80/apo or AM80/UVI3003 complexes clearly suggests that the unliganded RXR only marginally interacts with LXXLL motifs. Note in this respect that in contrast to cell-based assays in which serum-borne retinoic acid may complicate data analysis, we can exclude such interference in the present in vitro assays.
Having shown that dimerization does not significantly affect the structure of RAR and RXR LBDs or their binding affinity for isolated LXXLL motifs, we then addressed the question whether the dimerization interface is modified depending on the type of bound ligand, thereby providing a structural basis for the proposed allosteric communication between some heterodimeric receptors (50 -52). The resolution of the structure of the agonistbound RAR␤/RXR␣ heterodimer allowed us to compare it with the previously determined crystal structure of the heterodimeric complex of RAR␣ and RXR␣ LBDs in which the two subunits exhibit the antagonist conformation. The dimerization interface of the RAR␣-BMS614/RXR␣-oleat complex (28) was superposed onto the 9C-RA-bound RAR␤/RXR␣ heterodimer (Fig. 4). The C␣ atoms of the dimerization surface (H7-H10) of the RXR␣ protomers were used to fix the superposition. This allowed us to compare the relative positions of the dimerization surfaces of the associated RAR protomers in each heterodimer. The r.m.s.d. for the 90 matched RXR C␣ atoms within the H7-H10 region is 0.41 Å, indicating that the two RXR dimerization surfaces exhibit almost identical structures. The resulting r.m.s.d. between the dimerization surfaces of the RAR␣ and RAR␤ protomers also exhibits a low value (0.59 Å, over 86 C␣), indicating that overall, the two RAR/RXR heterodimer interfaces are very similar. A similar approach using the unliganded RXR␣ homodimer structure (44) revealed an r.m.s.d. of 0.48 Å between the dimerization surfaces of apoRXR in the homodimer and holoRXR in the heterodimer (not shown). Thus, no major conformational differences could be observed for the dimerization surfaces of RAR and RXR in the absence or the presence of agonist or antagonist ligands. In keeping with the above structural analysis, the binding affinity of the RAR␣/AM80 subunit to SRC-1 NR box 2 remains unchanged irrespective of the presence or absence of RXR agonist or antagonist ( Fig. 3C; compare AM80/CD3254 with AM80/apo and AM80/UVI3003).
Helix H11 Is a Sensor of Ligand Activity-We compared the 9C-RA-bound RAR␤ holo-structure with the recently reported structure of RAR␤ bound to the synthetic agonist TTNPB (Table II) (47). Despite the overall similarity of the receptor conformations and the similar positioning of the ligands that bind with similar affinities to RAR␤, some differences with important functional implications can be observed. When compared with 9C-RA, TTNPB is much less flexible and binds to RAR␤ in a more elongated conformation (Fig. 5A). Moreover, TTNPB contains two additional methyl groups attached to the tetrahydronaphthalene ring that interact with Val 388 and push helix H11 away from the pocket. In contrast, 9C-RA adopts a more bent conformation, allowing helix H11 to shift toward H12, therefore bringing Ala 385 and Val 388 too close to Leu 407 from H12 (1.92 and 2.57 Å, respectively). As a consequence, helix H12 slightly translates along its helical axis toward helix H3,  Table III. B, dissociation constants of complexes between SRC-1 NR box 2 (white bars) or the SRC-1 570 -780 fragment (gray bars) and monomeric RAR␣ or RXR␣ LBDs obtained from fluorescence anisotropy-based titrations. Experiments were carried out in the absence (apo) or in the presence of agonists (AM80 and CD3254) or antagonists (BMS614 and UVI3003). ND, not determined. C, the same experiment as in B but carried out with the RAR␣/RXR␣ heterodimer. and the side chains of Met 406 and Leu 407 undergo large conformational changes. Thus, the interaction surface with coactivators, of which helix H12 is an integral part, is slightly different in the two complexes. Equilibrium binding of the fluoresceinlabeled SRC-1 NR box 2 peptide to the RAR␤ LBD alone or bound to 9C-RA or TTNPB was analyzed to evaluate the impact of the structural differences described above on the recruitment of coactivators. As shown in Fig. 5B, although both TTNPB and 9C-RA can be considered as agonists because they facilitate interaction with the coactivator fragment, they show different efficiencies. Indeed, the SRC-1 NR box 2 fluorescent peptide is recruited to the RAR␤/TTNPB complex with a K d of 0.5 M, whereas in the presence of 9C-RA, the affinity of the interaction is only 1.6 M. Similarly, experiments conducted with RAR␣ also reveal 3-fold lower efficiency of 9C-RA-bound RAR␣ for recruiting SRC-1 NR box 2 when compared with RAR␣ bound to TTNPB (Fig. 5C). In contrast, the RAR␣-specific agonist AM80 induces similar high affinity recruitment of SRC-1 NR box 2 when compared with TTNPB. Accordingly, both li-gands share the important structural and chemical features discussed above. Therefore, it appears that because of suboptimal H11-ligand interactions, 9C-RA is incapable of stabilizing helix H12 in the conformation observed in the RAR␤/TTNPB complex structure, resulting in a less efficient recruitment of LXXLL motifs. Together, these data demonstrate how subtle chemical differences between various ligands can be sensed by helix H11 and transferred to the activation helix through intimate packing contacts between helix H11, loop L11-12, and H12.
Concluding Remarks-In this study, we used x-ray crystallography and fluorescence anisotropy to get new insights into the ligand-dependent regulation of the interaction between RAR/ RXR heterodimers and coactivators. Our data suggest a model of RAR/RXR heterodimer action in which the intrinsic capacity of each subunit to interact with LXXLL motifs of coactivators can be independently regulated by ligands without affecting the activity of the partner receptor. However, in full agreement with previous reports (49), we observed that the strength of the overall associ- FIG. 4. Structural comparison of fully agonistic and antagonistic conformations of RAR/RXR LBD heterodimers. The RAR␣-BMS614/RXR␣oleat heterodimer (Protein Data Bank ID code: 1DKF) displaying a fully antagonistic conformation was superposed onto the RAR␤-9C-RA/RXR␣-9C-RA LBD complex. The agonist-and the antagonist-bound heterodimers are shown in blue and green, respectively. In both complexes, helices H12 are colored in red. The TRAP220 peptide is depicted in yellow. Selected secondary structural elements are labeled. This figure shows that the structural differences between the two dimers are primarily located in the extreme C-terminal portion of the LBD comprising helices H11 and H12 as well as loop L11-12. The remaining parts of the dimers, including the dimerization interfaces, perfectly align.
FIG. 5. Helix H11 mediates ligand activity. A, overlay of RAR␤ LBPs bound to 9C-RA and TTNPB, colored in blue and magenta, respectively. Note that whereas helices H5 and H10 superpose perfectly, ligand-induced conformational changes can be seen for helices H11 and H12. Titrations of fluorescein-labeled SRC-1 NR box 2 by RAR␤ (B) and RAR␣ LBDs (C), in the absence (apo) or in the presence of various agonists. In keeping with the structural differences observed between the RAR␤/9C-RA and RAR␤/TTNPB complexes, synthetic retinoids (TTNPB or AM80) activate RAR␣ and RAR␤ more efficiently than does 9C-RA. Dissociation constants derived from the curves are reported in Table IV. ation with the coactivator is dictated by the combinatorial action of RAR and RXR ligands, the simultaneous presence of the two receptor agonists being required for highest binding affinity of coactivators. Our structural and energetic data argue against an allosteric mechanism in which communication across the dimer interface would enhance the affinity of RAR or RXR for individual LXXLL motifs. Rather, the synergy between receptor agonists would result from the enlargement of the contact area between the heterodimer and the coactivator through formation of one interacting surface on each heterodimer subunit. In this context, two NR boxes would mediate the optimal assembly of one coactivator molecule onto heterodimeric receptors. Of note, different receptor-coactivator pairs have been shown to use different combinations of NR boxes, providing both receptor-and ligand-specific assembly of coactivator complexes (46,(53)(54)(55), thus allowing for the design of selective competing peptide (or peptidomimetic) antagonists (56,57). Interestingly, the structure reveals that when bound to NR box 2, RXR helix H12 in the holo-conformation is capped by an aromatic clamp on the surface of the receptor formed by two phenylalanine residues on H3 and H11. Site-directed mutagenesis confirmed the functional importance of these residues for RXR transcriptional activity.
Comparison of the structures of RAR␤ LBD in complex with 9C-RA or TTNPB reveals that the position of helix H11 is dictated by the type of bound ligand and impacts on the conformation of helix H12 and in turn coactivator recruitment. This structural knowledge should be considered in the rational drug design of new "mixed" agonists/antagonists (i.e. weaker agonists than TTNPB or AM80). A particular feature of such ligands is that, due to their inability to efficiently stabilize holoLBD conformations, their activity on nuclear receptor activation depends on the cellular context, thereby giving rise to some degree of tissue specificity. Indeed, evidence exists that specific coactivator and corepressor levels in cells contribute to the relative agonist versus antagonist activity of such ligands (49,58). In cell lines characterized by high concentrations of coactivators, the ligand acts as an agonist because the active conformation of the receptor is further stabilized through interaction with the coregulator proteins, whereas in cells containing low levels of coactivators, the ligand is neutral or acts as an antagonist.