Control of Retinoic Acid Receptor Heterodimerization by Ligand-induced Structural Transitions

Heterodimerization of retinoic acid receptors (RARs) with 9-cis-retinoic receptors (RXRs) is a prerequisite for binding of RXR·RAR dimers to DNA and for retinoic acid-induced gene regulation. Whether retinoids control RXR/RAR solution interaction remains a debated question, and we have usedin vitro and in vivo protein interaction assays to investigate the role of ligand in modulating RXR/RAR interaction in the absence of DNA. Two-hybrid assay in mammalian cells demonstrated that only RAR agonists were able to increase significantly RAR interaction with RXR, whereas RAR antagonists inhibited RXR binding to RAR. Quantitative glutathione S-transferase pull-down assays established that there was a strict correlation between agonist binding affinity for the RAR monomer and the affinity of RXR for liganded RAR, but RAR antagonists were inactive in inducing RXR recruitment to RAR in vitro. Alteration of coactivator- or corepressor-binding interfaces of RXR or RAR did not alter ligand-enhanced dimerization. In contrast, preventing the formation of a stable holoreceptor structure upon agonist binding strongly altered RXR·RAR dimerization. Finally, we observed that RAR interaction with RXR silenced RXR ligand-dependent activation function. We propose that ligand-controlled dimerization of RAR with RXR is an important step in the RXR·RAR activation process. This interaction is dependent upon adequate remodeling of the AF-2 structure and amenable to pharmacological inhibition by structurally modified retinoids.

The ligand-dependent transcriptional activation of nuclear receptors is a DNA-dependent multistep process in which the ligand/receptor interaction is the key initiating event. Ligand binding to nuclear receptors leads to dramatic structural transitions of the polypeptidic chain that alter several functions of the ligand-bound receptor. These functions are specified by structural domains, the organization of which is shared in this gene superfamily. Another general property of nuclear receptors is their capacity to engage protein/protein interactions, which serves essentially two purposes: (i) increasing the spec-ificity and affinity for hormone response elements through homodimerization or heterodimerization with other nuclear receptors and (ii) modulating nuclear receptor transcriptional activity through ligand-modulated recruitment of nuclear coactivators and corepressors.
Although the role of ligand in inducing conformational alterations of the receptor that either disrupt (nuclear corepressor) or stabilize (nuclear coactivator) protein/protein binding interfaces is abundantly documented (reviewed in Refs. 1 and 2), the biological relevance of ligand-induced nuclear receptor homodimerization and heterodimerization in the absence of DNA is more controversial. Retinoic acid receptors (RARs) 1 and 9-cis-retinoic acid receptors (RXRs) have two dimerization interfaces, which are located in the DNA-binding domain and the ligand-binding domain (LBD) (3). These two domains are functionally autonomous, and isolated RAR and RXR DNA-binding domains bind as dimers specifically and cooperatively to retinoic acid response elements (RAREs) similarly to wild-type (WT) receptors (4). We have shown that the RAR LBD contains a strong dimerization interface in helix 9 enabling RXR/RAR interaction in the absence of DNA (3). These observations and others also demonstrate that receptor dimerization and DNA binding are, at least in vitro, ligand-independent events. This consequently questions the role of ligand as a regulator of receptor dimerization, despite the presence of a strong dimerization interface in the LBD, which is likely to be altered, directly or indirectly, upon ligand binding.
Ligand-dependent dimerization of some nuclear receptors has, however, been reported for various systems. 1,25-Dihydroxyvitamin D 3 favors the dissociation of vitamin D 3 receptor homodimers into monomers, rendering vitamin D 3 receptor molecules available for dimerization with RXR (5). Thyroid hormone has been shown to promote thyroid hormone receptor dimerization with RXR prior to DNA binding both in vivo and in vitro (6), and 9-cis-retinoic acid was shown to induce RXR homodimerization, providing a molecular basis for the observed inhibition of vitamin D 3 receptor-and thyroid hormone receptor-enhanced transcription by 9-cis-retinoic acid (5)(6)(7). We have also shown that RAR agonists promote the formation of hRAR␣ LBD homodimers (8). Moreover, the importance of ligand structure in regulating receptor dimerization is poorly understood, and conclusions drawn from the study of synthetic ligands for sex steroid receptors such as antiestrogens (9) and antipro-gestins (10), which are clinically important molecules, suggested that transcriptional inhibition stems either from a repressing activity of antagonist-bound monomer or from impaired dimerization (11,12).
Since the role of ligand in receptor dimerization has not yet been studied for the RXR/RAR system, we decided to investigate the role of retinoids in the regulation of RXR⅐RAR dimerization in vivo and in vitro using a panel of natural and synthetic retinoids with distinct biological properties. Results obtained using quantitative GST pull-down experiments and a mammalian two-hybrid system demonstrated that RAR agonists stabilized hRXR␣⅐hRAR␣ heterodimers as a function of their affinity for monomeric hRAR␣, whereas antagonists were unable to promote heterodimer formation. This lack of activity may result, at least in part, from an impaired AF-2 (activation function-2) folding since a mutation designed to prevent the stabilization of the holo-hRAR␣ structure similarly decreased the affinity of hRAR␣ for hRXR␣. hRXR␣ AF-2 did not display any significant contribution to dimerization, suggesting that the hRAR␣ monomer has a predominant role in regulating the dimerization rate with RXR.

EXPERIMENTAL PROCEDURES
Materials-atRA was obtained from Sigma (Saint-Quentin Fallavier, France). Synthetic retinoids were a kind gift from Dr. U. Reichert. Ro 41-5253 was a gift from Hoffman-La Roche. Ligand structures and properties have been described in detail (13). DNA restriction and modification enzymes were from Promega (Charbonnières, France). Polyethyleneimine (ExGen 500) was from EuroMedex (Souffelweyersheim, France).
Plasmids-pGEX2TKhRXR␣ and pGEX2ThRAR␣ vectors were obtained by cloning the RXR␣ and RAR␣ cDNAs into pGEX2TK and pGEX2T, respectively. hRXR␣ was truncated from the first 44 N-terminal amino acids. Constructs containing either the wild-type hRAR␣ or hRXR␣ cDNA subcloned into pSG5 (Stratagene) have been described (8,14). The UAS-tk-Luc reporter gene was a gift from V. K. Chatterjee and contains two 17-mer UAS Gal4 response elements (6). pCMV-Gal4-hRXR␣ LBD and pCMV-VP16-hRAR␣ were obtained from T. Perlmann (15) (see Fig. 1). The (GRARE) 3 -tk-Luc reporter gene is a pGL3-based vector (Promega) containing three repeats of a composite response element made up of a glucocorticoid response element half-site separated from a RARE half-site by a 5-bp spacer (AGAACAccgaaAGGTCA). This response element was cloned upstream of the thymidine kinase promoter (Ϫ105 to ϩ50). This reporter gene was activable only in the presence of overexpressed RARs and of a mutant hRXR␣, RXGR: mutations introduced into hRXR␣ converted the P box sequence scEGckGff to ccGSckVff, hence conferring to hRXR␣ the ability to bind specifically glucocorticoid response elements. 2 The VP16-hRAR␣ AHT mutant was constructed by site-directed mutagenesis (QuickChange, Stratagene) using the following oligonucleotides: 5Ј-ggtgcgcaaagggggccaggaaacctttcctgccctctgcc-3Ј and 5Ј-ggcagagggcaggaaaggtttcctggccccctttgcgcacc-3Ј. This mutagenesis procedure led to the introduction of two mutations, converting Ala 194 into Gly and His 195 into Gly. The hRAR␣ L409A mutant was constructed according to the same protocol using the following mutagenic primers: 5Ј-catgccgcctgccatccaggaaatgctcgagaactcagagg-3Ј and 5Ј-cctctgagttctcgagcatttcctggatggcaggcggcatg-3Ј. Other plasmids were constructed using standard subcloning procedures (16). All constructs were verified by automatic sequencing.
Cell Culture and Transfections-HeLa Tet-On cells were cultured and transfected as previously described (17).
Protein/Protein Interaction Assays-DNA-independent protein/protein interactions were performed as described (18).
Restriction Site (SmaI) Accessibility Assay-Cells were grown and treated as indicated and collected in ice-cold 1ϫ phosphate-buffered saline. The cell pellet was resuspended in 5 volumes of buffer A (20 mM Tris-HCl (pH 7.5), 1 mM EDTA, 50 mM KCl, 1% Triton X-100, 0.15 mM spermine, 0.5 mM spermidine, 5 mM sodium butyrate, and protease mixture inhibitor (1:100 dilution; Sigma)). Cells were lysed by several strokes in a Dounce with an A-type pestle, and the homogenate was layered onto a 0.8 M sucrose cushion in buffer A. Nuclei (ϳ3-5 A 260 ) were digested with 80 -100 units of SmaI (Promega) in a final volume of 200 l in SmaI digestion buffer. The reaction was stopped by adding an equal volume of stop buffer (20 mM Tris-HCl (pH 7.5), 1% SDS, 25 mM EDTA, and 5 mg/ml proteinase K (Promega)) and incubated for 8 h at 37°C. DNA was purified, and 50 -80 g was digested with 25 units of PstI. Cleaved DNAs were analyzed on native 1% agarose gels and stained with ethidium bromide. Gels were transferred to nylon membranes (Hybond N ϩ , Amersham Pharmacia Biotech), and DNA was UV-cross-linked to the membrane. Hybridization was carried out with a 600-bp RAR␤2 32 P-labeled probe (Rediprime II, Amersham Pharmacia Biotech), which was amplified from mouse genomic DNA using the following set of primers: 5Ј-ttggccaggaacaggagagc-3Ј and 5Ј-tttaccattttccaggctcg-3Ј (nucleotides 418 -437 and 1031 to 1012, respectively, in the mouse retinoic acid receptor-␤ gene promoter region; GenBank TM / EBI accession number X56850.1). After hybridization, blots were washed, and radioactive material was detected using a Storm 860 Phosphoimager.
Reverse Transcription-PCR Analysis of mRAR␤2 and Actin Transcripts in P19 Cells-Total RNA was prepared using RNAble reagent (Eurobio, Les Ulis, France) according to the manufacturer's protocol. Total RNA (50 g) was then treated with 10 units of RNase-free DNase I (Genhunter, Nashville, TN) for 1 h at 37°C to digest genomic DNA. The purified RNA was adjusted to 1 g/l and checked for integrity by agarose gel electrophoresis. Reverse transcription was performed using random primers (Promega) as recommended by the manufacturer. PCR primers were designed as follows: RAR␤, 5Ј-aagtggtaggaagtgagctg-3Ј and 5Ј-ctacattgagcagtatgccg-3; and actin, 5Ј-atcatgtttgagaccttcaa-3Ј and 5Ј-catctcttgctcgaagtcca-3Ј. PCR conditions were 40 cycles of 30 s at 94°C, 1 min at 58°C, and 1.5 min at 72°C followed by an elongation step at 72°C for 7 min (RAR␤) and 30 cycles of 30 s at 94°C, 1 min at 56°C, and 1.5 min at 72°C followed by an elongation step at 72°C for 7 min (actin).
Statistical Analysis-All incubations or assays were performed at least in triplicate. Measured values were used to calculate mean Ϯ S.D. Calculations were carried out using Prism software (GraphPAD Inc., San Diego, CA)

RESULTS
hRAR␣ Interacts with hRXR␣ in Vivo in a Ligand-dependent Manner-Interaction of hRAR␣ with hRXR␣ was first characterized using a mammalian two-hybrid interaction assay. In this system, the acidic VP16 activation domain (AD) is fused to the N terminus of full-length hRAR␣, whereas the Gal4 DNAbinding domain is linked to the N-terminal end of the hRXR␣ LBD (Fig. 1A). The activity of a reporter gene containing two Gal4 consensus DNA-binding sites (UAS-tk-Luc) was the biological readout used to monitor the interaction between these two receptors in HeLa cells. Transfection of VP16-hRAR␣ did not promote a significant increase in reporter gene activity, 2 but Gal4-hRXR␣ LBD alone turned out to be responsive to the RXR-specific ligand CD2425, which induced a 4 -5-fold increased activity (Fig. 1B). When WT hRAR␣ was overexpressed together with Gal4-hRXR␣ LBD, a very modest increase (1.5fold) in the level of luciferase activity was detected in cell extracts in response to retinoid treatment, suggesting that the hRAR␣ AF-2 region is unable to activate this system. Since Gal4-hRXR␣ LBD, WT hRAR␣, and VP16-hRAR␣ were expressed at similar levels in HeLa cells as judged by whole cell ligand binding assays (from 30,000 to 100,000 sites/cell), 2 this clearly suggests that WT hRAR␣ is unable per se to mediate a significant ligand-dependent activation of the system. Coexpression of Gal4-hRXR␣ LBD and VP16-hRAR␣ led to a detectable ligand-independent activation of the reporter gene (3-5fold), and addition of RAR-specific ligands strongly increased reporter gene activity (10 -15-fold). However, the ligand-mediated activation of the two-hybrid system appeared to be restricted to RAR␣-binding agonists such as atRA and Am580. As expected, RAR␤-selective (CD417) and RAR␥-selective ligands ( Fig. 2) 2 had no effect in our system, in agreement with their very poor affinity for hRAR␣. The RAR antagonist Ro 41-5253 did not activate this system, and Ro 41-5253 competitively inhibited atRA-induced activity. 2 More surprisingly, the RXRspecific ligand, which was able to induce a 4 -5-fold increase in reporter gene activity when RXR was expressed alone, was inactive when WT hRAR␣ or VP16-hRAR␣ was coexpressed with Gal4-hRXR␣.
The ligand-dependent activation paralleled the pattern of induction observed when using full-length hRAR␣, hRXR␣, and a DR5 RARE-driven reporter gene ((GRARE) 3 -tk-Luc) (Fig.  1B). In the same assay, VP16-hRAR␣ responded to ligands its wild-type counterpart, but promoted a much higher transcriptional activity of the (GRARE) 3 -tk-Luc reporter, as expected from the presence of the strong VP16 AD acidic activator.
These observations thus revealed several features of the two-hybrid system. First, they suggest that RXR/RAR interaction occurs in the absence of ligand. In line with this observation, coexpression of RAR and RXR inactivated RXR responsiveness to ligand, suggesting that RAR interaction inhibits RXR function. Second, only RAR␣-specific agonists are able to increase significantly the activity of this system. Third, RARdependent activation is enhanced by the VP16 AD, consistent with the hypothesis of ligand-dependent dimerization. However, our data do not rule out at this stage that RAR agonist requirement merely reflects a synergism of RAR AF-2 with the VP16 AD and/or a repression exerted by unliganded hRAR␣ on VP16 activation function.

Tethering of Coactivators and Corepressors to RAR and RXR Does Not Alter the Ligand-induced Response of the Two-hybrid
System-Unliganded, heterodimeric hRAR␣ interacts with the corepressor SMRT, which dissociates upon agonist binding, and this interaction is stabilized in the presence of antagonists (19). To test whether the agonist-induced activity of the twohybrid system is due to corepressor release, we assayed the activity of a hRAR␣ mutant containing a double mutation (A194G/H195G, hRAR␣ AHT) in the nuclear corepressor box (20), which considerably lessened hRAR␣ interaction with SMRT (10 -12-fold) ( Fig. 2A) without altering the ligand binding activity of the receptor. 3 Interaction with SRC-1 was also diminished, yet clearly detectable. Such a mutation would thus predictably lead to an enhanced basal level of reporter gene activity and/or to a decreased ligand-induced activity. While showing a low constitutive activity, as already reported for an analogous mutation in the thyroid hormone receptor (6), hRAR␣ AHT displayed a ligand-induced activity similar to that of WT hRAR␣, arguing against a possible role for corepressors and coactivators in VP16 activity control. In keeping with this, we observed that truncation of the RXR AF-2 AD region dominant negative (RXR), which strongly increases corepressor binding to RXR (21), 2 did not affect the response of the system to atRA treatment (see Fig. 4).
Another possibility is that the observed luciferase activity results from the synergistic contribution of the hRAR␣ AF-2 AD and VP16 AD to the transcriptional activation of the reporter gene. Leucine 409 of hRAR␣ is located in helix 12, and its mutation leads to transcriptional inactivation of hRAR␣ while maintaining WT ligand binding activity (Fig. 3C) (22). 2 This amino acid becomes part of the hydrophobic groove formed upon agonist-induced AF-2 AD folding to which coactivators bind through their LXXLL motif. The L409A substitution abolished ligand-induced SRC-1 recruitment to hRAR␣, whereas receptor binding to SMRT was left unchanged ( Fig. 2A). RAR␣ L409A induced both basal and ligand-induced activities comparable to WT hRAR␣ in the two-hybrid system ( Fig. 2A), suggesting again that atRA-induced activity is indeed due to receptor heterodimerization and not to the concerted activation of the hRAR␣ AF-2 AD and VP16 AD.
Antagonists Do Not Inactivate the VP16 Activation Domain-The observed inactivity of the two-hybrid system in the presence of antagonists may stem from the inhibition of the VP16 moiety, as suggested for progesterone antagonists (10). To test this hypothesis, we monitored the activity of WT hRAR␣ and VP16-hRAR␣ in a standard transactivation assay using a DR5driven reporter gene (Fig. 2B). As expected, atRA promoted a very strong increase (8 -15-fold induction) in reporter gene activity in the presence of overexpressed hRAR␣ or VP16-hRAR␣, whereas a very modest increase (2-fold) was observed when cells were challenged by two antagonists, CD3105 and CD3106 (23). Competitive inhibition of atRA-induced transcription was observed when each antagonist was used together with atRA. The basal level of transcription was higher (8 -10-fold increase) when VP16-hRAR␣ was expressed in HeLa cells together with hRXR␣, as expected from the strong transactivating potential of the VP16 AD. The basal level of transcription was again slightly increased (2-fold) in the presence of both antagonists and was not decreased, as it could be 3 A. Mouchon and P. Lefebvre, unpublished observations. predicted if VP16 function was conditioned by RAR activity, demonstrating that VP16 AD transactivating potential is not altered upon antagonist binding to hRAR␣.
A RAR Antagonist Restores the RXR Response to a RXRspecific Ligand-The inhibition of RXR activity upon expression of WT hRAR␣ or VP16-hRAR␣ suggested an allosteric control of RXR function by RAR and implied that a fraction of RAR and RXR populations is involved in ligand-independent dimerization. We thus tested whether antagonist binding to hRAR␣ affected RXR response to a specific ligand (Fig. 2C). When Gal4-hRXR␣ was expressed alone, the system was responsive to RXR ligand, and this responsiveness was not affected by RAR ligands. Overexpression of hRAR␣ conferred responsiveness to a RAR agonist (Am580), and the RXR-induced activity was repressed, as shown above. Cotreatment of cells with the RAR-specific agonist Am580 and the RXR-specific agonist CD2425 did not reveal any synergy between both receptors. Very interestingly, addition of the RAR antagonist CD3106 together with the RXR-specific agonist restored a RXR-induced response, similar to that observed in the presence of Gal4-hRXR␣ alone, suggesting that RXR is able, under these conditions, to function autonomously. These results thus support our view that RAR antagonists prevent RAR interaction with RXR in vivo.
Agonist-mediated Stabilization of hRXR␣⅐hRAR␣ Dimers in Vivo-The results shown in Fig. 1 suggested several important conclusions. (i) Ligand binding is required to stabilize RXR⅐RAR dimers in vivo; (ii) antagonists, despite having a high affinity for RAR, do not induce detectable interactions between these two receptors; and (iii) dimerization is dependent solely on ligand occupancy of the hRAR␣ ligand-binding pocket. We extended our observation to a panel of synthetic retinoids with distinct biological properties: isotype-selective RAR agonists, a partial agonist, antagonists, and dissociated (anti-AP-1) retinoids (Fig. 3A). Comparing the maximal induction of the GAL4 reporter gene in the presence of these retinoids allowed us to distinguish two classes of molecules: active and inactive compounds. A closer examination showed that active compounds were referenced, based on transactivation assays, as RAR␣ agonists (atRA, CD367, Am580, and Am80). Inactive compounds were either RAR␤-selective (CD417) or RAR␥-selective (CD666) retinoids or RAR antagonists (CD3106, CD2331, CD2905, and CD2856). The recently described anti-AP-1 retinoid CD2409 (24) was unable to promote RXR⅐RAR dimerization, and the partial agonist CD2665 had an intermediate ability to induce RXR⅐RAR heterodimerization in vivo. We then monitored the activity of the VP16-RAR␣ L409A mutant in response to several retinoids in the same system (Fig. 3B). This transcriptionally defective RAR␣ mutant displayed a similar response profile to agonists and antagonists compared with WT RAR␣, establishing that RAR␣ AF-2 does not contribute to the activity of the system. Agonist-induced luciferase activity was fully inhibited, in a dose-dependent manner, by CD3106, suggesting that the observed inhibition results from direct competition for the ligand-binding pocket of RAR␣. Moreover, limited proteolytic assays carried out in parallel using WT RAR␣ and RAR␣ L409A (Fig. 3C) showed that both receptors underwent similar structural transitions upon ligand binding, establishing that the L409A mutation did not significantly alter RAR␣ three-dimensional structure. Thus, our data and others show that the L409A mutation is compatible with antagonist binding and ligand-induced dimerization.
Retinoids able to bind hRAR␣ with high affinity can be classified according to their ability to induce hRXR␣⅐hRAR␣ dimerization: agonists and others, this latter category including, most notably, anti-retinoids. The two-hybrid system thus discriminates between agonists and potential antagonists on the basis of ligand ability to induce dimerization in vivo.
Ligand-induced AF-2 AD Folding of hRAR␣ Is Necessary for RXR⅐RAR Dimerization-Major structural transitions occur in the LBD of nuclear receptors upon ligand binding and lead to the repositioning of several ␣-helices. The AF-2 AD region (helix 12), which is fully exposed to solvent in the unliganded state (aporeceptor), establishes salt bridges with basic amino acids located in helices 3 and 4. This repositioning along the LBD creates a hydrophobic groove to which nuclear coactivators bind via LXXLL motifs (holoreceptor). RAR AF-2 AD repositioning may thus have a role in remodeling dimerization interfaces of RAR, and we tested this hypothesis using both in vitro and in vivo protein/protein interaction assays (Fig. 4).
The RAR␣ holo-LBD three-dimensional structure designates Lys 244 and Lys 262 as establishing salt bridges with acidic residues located in the AF-2 AD region (Glu 412 and Glu 415 ), and mutation of each lysine yields receptor mutants with impaired transactivating properties (17). Mutating both of these lysines yielded an inactive receptor (hRAR␣ K244A/K262A) in the transactivation assay (Fig. 4A). This RAR mutant was unable to interact with nuclear coactivators and corepressors in vitro. 2 GST pull-down assays revealed that hRAR␣ K244A/K262A was still able to interact with hRXR␣ in a ligand-dependent manner, with an estimated K d of 40 -60 nM when complexed to atRA. This double mutation induced a 4 -5-fold decrease in the affinity of hRAR␣ for RXR, which was also reflected in the mammalian two-hybrid system by only a 2-fold increase in the activity of the Gal4/VP16 system, in sharp contrast to WT hRAR␣, which promoted a 5-7-fold increase (Fig. 4A). Thus, a 4-fold decreased activity in the two-hybrid system is correlated with a 4-fold decrease in the hRAR␣ affinity for hRXR␣ in vitro.
The RXR AF-2 AD motif has been proposed to regulate nega-tively nuclear coactivator access to RAR by direct interaction with the LBD hydrophobic groove (25), suggesting that this domain may be involved in the stabilization of the RXR⅐RAR complex. We thus tested whether truncation of the hRXR␣ AF-2 region perturbed RXR/RAR interaction. Although dnRXR was a strong inhibitor of RAR-mediated transactivation (Fig. 4B), it displayed normal affinity for WT hRAR␣ in the in vitro interaction assay (K d Ϸ 20 nM). The two-hybrid assay showed that hRAR␣ interacted with a similar efficiency with WT hRXR␣ and dnRXR (Fig. 4A), leading to the conclusion that the RXR AF-2 AD has no major influence on RXR/ RAR interaction.

Affinity of Liganded hRAR␣ for hRXR␣ Correlates with Agonist Binding Affinity for Monomeric hRAR␣ in Vitro-
To determine which parameter(s) are critical for triggering hRXR␣/hRAR␣ interaction, we carried out an in vitro interaction assay in which GSTtagged full-length hRXR␣ was used as a bait to trap in vitro translated, 35 S-labeled hRAR␣ (Fig. 5). Labeled hRAR␣ was incubated for 1 h at 4°C with increasing concentrations of the indicated ligand, and then the GST-hRXR␣ affinity matrix was added to the binding mixture. In the absence of ligand, a weak constitutive interaction of RAR with RXR was observed (Fig. 5A). Increasing the atRA concentration from 10 Ϫ10 to 10 Ϫ6 M caused a dose-dependent increase in the RXR/RAR interaction (up to 5-7-fold). A similar pattern was observed with CD367, a synthetic RAR agonist. In contrast, the antagonist CD3106, which was inactive in the twohybrid assay, was unable to stabilize RXR/RAR interaction. We noted that at concentrations exceeding 50 nM, this ligand even promoted a clear decrease in the constitutive association of the heterodimer. Finally, a RAR␤-selective ligand 6-(5,6,7,8-tetrahydro-5,5,8,8-tetramethyl-2-naphtalenyl)-2-naphtale n-carboxylic acid, did not exhibit a significant effect on RXR/RAR interaction.
The results shown above (Fig. 3B) documented the ability of a retinoid antagonist to inhibit, in a dose-dependent manner,

FIG. 3. Ligand biological properties correlate with ability to induce RXR⅐RAR dimerization in vivo.
A, ligand-induced activation of the two-hybrid system in the presence of WT RAR␣. HeLa cells were transiently transfected with the UAS-tk-Luc reporter gene and the CMV-Gal4-hRXR␣ LBD and the CMV-VP16-hRAR␣ expression vectors. Cells were treated with retinoids, and luciferase activity was assayed as described in the legend to Fig. 1. Results are expressed as the means Ϯ S.D. of at least three individual experiments, with the basal level of luciferase activity (usually between 3000 and 10,000 RLU) scaled to 1. Biological activities of retinoids are indicated at the top of the bar graph. B, ligand-induced activation of the two-hybrid system in the presence of RAR␣ L409A. Experiments and analysis results are presented exactly as described for A using VP19-RAR␣ L409A in place of VP16-RAR␣. C, limited proteolytic clipping assays. Assays were carried out exactly as described (44). 35  the interaction of RAR with RXR in intact cells. We therefore carried out an analogous competition experiment using the in vitro GST pull-down assay (Fig. 5B). The RAR␣-selective ligand Am580 was, as expected from its biological classification, able to promote RXR⅐RAR dimerization, whereas Ro 41-5253, a RAR antagonist, was inactive in this assay. When RAR␣ was incubated simultaneously with 25 nM Am580 and increasing concentrations of Ro 41-5253 (from 0.1 nM to 0.1 M), a dose-dependent decrease in the rate of heterodimer formation was observed, establishing a correlation between this in vitro assay and the two-hybrid assay.
Since we were able to observe a dose-dependent increase in the RXR/RAR interaction in this system, we assayed the affinity of hRAR␣ for hRXR␣ in the presence of the retinoids used previously. Values are tabulated in Fig. 5C along with relative affinities of hRAR␣ for each ligand as determined by in vitro ligand binding assay using hRAR␣ overexpressed in Escherichia coli (13). Quite strikingly, a classification similar to that established on the basis of the mammalian two-hybrid system could be proposed: active and inactive compounds. The first category comprised only high affinity ligands classified as agonists. Within this category, a clear correlation between ligand binding affinities and the affinity of liganded hRAR␣ for hRXR␣ was observed. Inactive compounds could be subdivided into two groups: high and low affinity ligands. The first subgroup included retinoids classified as antagonists, whereas the second includes ␤ and ␥ isotype-selective retinoids. Thus, RAR antagonists can be identified on the basis of two criteria: high affinity for monomeric hRAR␣ and failure to promote RXR/ RAR interaction in vitro and in vivo.
atRA, but Not a Retinoid Antagonist, Induces Chromatin Structure Alterations in the Endogenous RAR␤2 Promoter-P19 embryonic carcinoma cells are a well documented cellular model to study molecular responses to retinoids (reviewed in Refs. 26 and 27). The mRAR␤2 gene is, in this cell line, highly inducible in response to atRA treatment (28), and the chroma-tin organization of this promoter undergoes structural alterations within minutes after retinoid treatment of P19 cells (29,30). These alterations are correlated with receptor binding to the DR5 RARE present in this promoter, which is organized around a nucleosome core both in vitro (31) and in vivo. 4 Such alterations can be monitored by variation of restriction site accessibility, and this assay has been used to document glucocorticoid receptor-mediated activation of the murine mammary tumor virus promoter (32) and RXR⅐RAR-mediated activation of the mRAR␤2 promoter (29). Since this chromatin remodeling activity is correlated with RARE occupancy as determined by in vivo footprinting (30), this assay may be considered as an indirect assessment of receptor binding to DNA in vivo.
P19 cells were thus treated with atRA or CD3106 (a retinoid antagonist), and SmaI accessibility to the endogenous mRAR␤2 promoter was assayed in each condition (Fig. 6B). In parallel, the rate of expression of the mRAR␤2 gene was assayed by reverse transcription-PCR (Fig. 6A), which was, as expected, strongly induced after a 2-h induction with 1 M atRA, but not with 1 M CD3106. P19 cell DNA was thus extracted and cleaved by PstI, generating a 1185-bp fragment. atRA treatment strongly increased SmaI access to its cutting sites located in the vicinity of the DR5 RARE (608-bp fragment) and in the 5Ј-untranslated region (424-bp fragment), whereas the antagonist CD3106 did not induce major structural transitions in the mRAR␤2 promoter. We note that the accessibility of both SmaI sites is increased in this assay, suggesting that retinoid receptor binding triggers a general opening of the chromatin in the vicinity of the DR5 RARE, which may be related to the ordered occupancy of the various cis-acting elements located along this promoter (30). The receptor was incubated with a GSH-GST-hRXR␣-Sepharose matrix, and the amount of RAR bound to the RXR affinity matrix was quantified by 8% SDS-polyacrylamide gel electrophoresis followed by autoradiography. Right panel, in vivo protein interaction assay. HeLa cells were transfected with the UAS-tk-Luc reporter gene and the CMV-Gal4-hRXR␣ LBD and CMV-VP16-hRAR␣ or hRAR␣ K244A/K262A expression vectors. 24 h after transfection, cells were treated with 1 M for 16 h, and luciferase activity was assayed. Results are expressed as the means Ϯ S.D. of at least three independent experiments. B, a C-terminally truncated hRXR␣ is transcriptionally inactive, but heterodimerizes with hRAR␣ as does WT hRXR␣. The transcriptional activity and the ability of the truncated RXR (dnRXR) to interact both in vitro and in vivo with hRAR␣ were assayed as described for A.

DISCUSSION
In this study, we show that RAR agonists promote RXR recruitment to liganded RAR in solution. A comparative study of amino acid substitutions in hRAR␣ helices 3, 4, and 12 revealed that a mutation in helix 12 (L409A) that was designed to inactivate the RAR␣ coactivator-binding interface did not affect agonist-induced RAR␣ dimerization with RXR␣ in vivo. In contrast, mutations in helices 3 and 4 (K244A/K262A) that prevent AF-2 AD positioning along the holo-LBD structure strongly inhibited RXR␣ recruitment to RAR␣. From this we conclude that RXR⅐RAR heterodimerization is not dependent on the integrity of hRXR␣ and hRAR␣ AF-2 ADs, but that proper repositioning of the hRAR␣ AF-2 AD region is important for RXR⅐RAR dimerization.
The agonist-enhanced activity of the two-hybrid system could potentially be interpreted as a contribution of the AF-2 inherent in either RXR or RAR, which may synergize with the VP16 AD. Manipulating the level of interaction of either hRAR␣ or hRXR␣ with corepressors or coactivators by amino acid changes or deletion in LBDs did not modify the responsiveness of this system to retinoids, arguing against such a hypothesis. Moreover, the perfect correlation between in vivo and in vitro data, for which a contribution of additional modulators cannot be invoked, strongly argues for an agonist-dependent dimerization in vivo, as previously demonstrated for the thyroid hormone receptor (6).
Very intriguingly, the RXR-specific ligand CD2425 induced a clear RXR response when this receptor was expressed alone. This ligand-induced activity was inhibited in the presence of hRAR␣, suggesting that RAR binds to RXR and represses its autonomous activation function. However, RAR antagonists restored RXR function within this experimental setting, strongly arguing for a blockade of retinoid receptor heterodimerization in the presence of a RAR antagonist. RXR has been reported to occur as a stable tetrameric complex that is destabilized into RXR dimers and monomers upon agonist binding. However, RXR tetramers did not dissociate in the presence of unliganded RAR (33). The active fraction of RXR molecules is thus likely to occur in our system as a monomeric subunit, available for dimerization. RAR-mediated repression of RXR has been widely documented for RXR⅐RAR het- FIG. 6. Impeded dimerization is correlated with transcriptional inactivity and chromatin structure stability of the mRAR␤2 promoter. A, reverse transcription-PCR analysis of mRAR␤2 and actin transcripts in P19 cells. P19 cells were treated for 2 h with vehicle (dimethyl sulfoxide (DMSO)), 1 M atRA, or 1 M CD3106, and total RNA was extracted. mRNA expression levels of the mRAR␤2 gene and ␤-actin were assayed by semiquantitative reverse transcription-PCR. Amplification products were resolved on 1% agarose gel and stained with ethidium bromide. Molecular size markers are also shown (lane M). B, effect of retinoid treatment on SmaI accessibility to the mRAR␤2 promoter. P19 cells were treated as described for A, and nuclei were purified. DNA was digested with 80 -100 units of SmaI for 10 min at 25°C and purified. After a secondary digestion with PstI, DNAs were analyzed by Southern blotting and detected using a mRAR␤2 probe abutting the 3Ј-PstI site. UTR, untranslated region. erodimers bound to DNA, and our data suggest that DNA binding is not necessary to observe allosteric regulation of RXR function by RAR.
The effect of ligand on RXR/RAR interaction was similar in both in vitro and in vivo assays, in which a full-length hRXR␣ or an isolated hRXR␣ LBD was used, respectively, suggesting that RXR and RAR LBDs are the main contributors to the measured dimerization activity. We note also that similar results were obtained when using full-length proteins with RAR used as a bait in the in vitro assay. 2 This is in keeping with our previous observations showing that amino acids located in helix 9 of hRAR␣ control both homodimerization and heterodimerization of RAR (3). Since dimerization through LBDs is important for heterodimer stabilization, we conclude that this activity would be of the utmost importance at physiological levels of expression of both receptors, mostly by modulating the DNA binding activity of receptors. Heterodimerization has been shown to increase strongly the DNA binding affinity of several nuclear receptors (5, 34 -37). As a consequence, one would predict that a decreased affinity for DNA may be caused by impaired dimerization. The retinoid antagonist CD3106 did not trigger structural alteration of the chromatin structure of the mRAR␤2 promoter in intact cells, in opposition to atRA. This may be interpreted as an impaired interaction with nucleosomal DNA. This observation therefore strongly supports our hypothesis and is in agreement with previous reports showing that the retinoid antagonists Ro 41-5253, BMS453, and BMS411 did not induce a dimethyl sulfate genomic footprint over the DR5 RARE region in P19 and NB4 cells (29,30,38).
The three-dimensional structure of the heterodimer RXR LBD⅐RAR LBD has been published recently (39). As for other nuclear receptors (40), ligand binding does not promote detectable variation of the structure of the dimerization interface, which is essentially composed of helices 7-9. However, biochemical studies showed that ligand is a strong promoter of estrogen (41), progesterone (42), RAR (8), and RXR (7) homodimerization, and our data, as well as others (5,6,43), support the concept that ligand induces receptor heterodimerization with RXR. Thus, ligand exerts structural constraint(s) on region(s) of hRAR␣ lying outside of the LBD, and further investigations will be aimed at identifying receptor subdomains involved in the ligand-dependent enhancement of RXR⅐RAR heterodimerization.
As the most decisive message, we would like to propose that ligand plays a critical role in an early step of receptor activation, i.e. receptor dimerization, and that the ability to interfere with this process is a key feature of retinoid antagonists. Finally, since several studies documented similar properties for estrogen antagonists (9,11) and since all retinoid antagonists tested so far blocked atRA-induced dimerization, 2 we would like to suggest that inhibition of dimerization may be a important feature of pure nuclear receptor antagonists.