Dimerization with Retinoid X Receptors and Phosphorylation Modulate the Retinoic Acid-induced Degradation of Retinoic Acid Receptors α and γ through the Ubiquitin-Proteasome Pathway*

In eukaryotic cells, the ubiquitin-proteasome pathway is the major mechanism for targeted degradation of proteins. We show that, in F9 cells and in transfected COS-1 cells, the nuclear retinoid receptors, retinoic acid receptor γ2 (RARγ2), RARα1, and retinoid X receptor α1 (RXRα1) are degraded in a retinoic acid-dependent manner through the ubiquitin-proteasome pathway. The degradation of RARγ2 is entirely dependent on its phosphorylation and on its heterodimerization with liganded RXRα1. In contrast, RARα1 degradation can occur in the absence of heterodimerization, whereas it is inhibited by phosphorylation, and heterodimerization reverses that inhibition. RXRα1 degradation is also modulated by heterodimerization. Thus, each partner of RARγ/RXRα and RARα/RXRα heterodimers modulates the degradation of the other. We conclude that the ligand-dependent degradation of RARs and RXRs by the ubiquitin-proteasome pathway, which is regulated by heterodimerization and by phosphorylation, could be important for the regulation of the magnitude and duration of the effects of retinoid signals.

␤, and ␥) and three RXR isotypes (␣, ␤, and ␥) encoded by distinct genes, and for each isotype there are at least two main isoforms, which differ in their amino-terminal region (1)(2)(3)(4). Each RAR and RXR isoform possesses two transcription activation functions: AF-1, which is located in the N-terminal A/B region, and the ligand-dependent AF-2, which is associated with the ligand binding domain (LBD) and requires the integrity of a highly conserved amphipatic ␣-helix, the AF-2 AD core that corresponds to helix 12 of the LBD (1). RAR␣ and RAR␥ are constitutively phosphorylated at conserved serines located in their N-terminal region by Cdk7 within the general transcription factor TFIIH, also involved in DNA repair (5,6). RAR␣ and RAR␥ are also phosphorylated by protein kinase A at the C-terminal end of their LBD (7). For both RAR␣ and RAR␥, phosphorylation of these residues located either in the N-terminal region or in the LBD modulates the activity of AF-1 and AF-2, respectively (5)(6)(7)(8). Finally, RAR/RXR heterodimers are the functional units that preferentially transduce the retinoid signal in vitro and in vivo (Refs. 1, 4, and 9 -13 and references therein), and the ligand-dependent activity of RXR is subordinated to that of RAR (Refs. 14 -16 and references therein).
The F9 embryonal carcinoma (EC) cells provide an interesting cell-autonomous model system for the analysis of retinoid signaling in vivo, since upon RA treatment and depending on culture conditions, they differentiate into three distinct cell types resembling the primitive, parietal, and visceral extraembryonic cells (17)(18)(19). F9 cells express all RAR and RXR isotypes, with RAR␣1 and RAR␥2 being the main RAR isoforms (20 -22), and their respective roles in the response of F9 EC cells to RA treatment is now well established. RAR␥ is indispensable for RA-induced differentiation of F9 EC cells into primitive endoderm (8,10,23), while RAR␣ is additionally required for parietal endodermal differentiation that occurs in the presence of RA and cAMP, with primitive endodermal differentiation being a prerequisite for parietal endodermal differentiation (8,24). RAR␥ phosphorylation at its N-terminal region is indispensable for differentiation of F9 cells into primitive endoderm, while RAR␣ phosphorylation at its protein kinase A site is required for parietal endodermal differentiation (8). This F9 EC cell RA-induced differentiation is also accompanied by the induction of expression of a number of genes (8). Thus, one of the key issues for understanding the regulation of cell differentiation and gene expression by retinoid receptors is how the duration and magnitude of their activity in response to RA can be controlled.
The ubiquitin-proteasome pathway is the major system in eukaryotic cells for selective degradation of short lived regula-* This work was supported by funds from CNRS, INSERM, the Collège de France, the Hôpital Universitaire de Strasbourg, the Association pour la Recherche sur le Cancer, and Bristol-Myers Squibb. 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. ʈ To whom correspondence should be addressed: Tel.: 33 3 88 65 34 59; Fax: 33 3 88 65 32 01; E-mail: cegly@igbmc.u-strasbg.fr. 1 The abbreviations used are: RA, retinoic acid; RAR, RA receptor; tRA, all-trans-retinoic acid; 9cRA, 9-cis-retinoic acid; RXR, retinoid X receptor; LBD, ligand binding domain; AD, activation domain; EC, embryonal carcinoma; mRXR and mRAR, mouse RXR and RAR, respectively; PCR, polymerase chain reaction; RT-PCR, reverse transcriptase-PCR; EMSA, electrophoretic mobility shift assay; WT, wild type; ALLN, N-acetyl-L-leucyl-L-leucinyl-norleucinal. tory proteins and transcription factors (25,26). A common feature of proteasome-mediated protein degradation is the covalent attachment of ubiquitin, a highly conserved 8.6-kDa protein, to lysine residues of proteins targeted for degradation, followed by the formation of polyubiquitin chains attached covalently to the targeted protein. Ubiquitinated proteins are recognized and degraded by the multisubunit protease complex, the 26 S proteasome (Refs. 27 and 28 and references therein).
Several members of the nuclear hormone receptor superfamily have been reported to interact with some of the enzymes involved in ubiquitination (29 -31). In addition, liganded nuclear receptors recruit SUG1 (32), which is not only a DNA helicase (33) but also a component of the 26 S proteasome (34). Since certain nuclear hormone receptors have been shown recently to be degraded in a ligand-dependent manner through the ubiquitin-proteasome pathway (30,35,36), we addressed the question of whether liganded RARs are degraded through this same pathway. We report here that, in F9 cells and in transfected COS-1 cells, RAR␥2 is degraded in a RA-dependent manner and that this degradation involves the ubiquitin-proteasome pathway. This degradation requires phosphorylation of the receptor in its N-terminal region and is completely dependent on its heterodimerization with liganded RXR. RAR␣1 and RXR␣1 are also degraded by the ubiquitin-proteasome pathway upon ligand binding, but the requirements for their degradation are different from those for RAR␥2 degradation. Finally, each partner of RXR/RAR heterodimers was shown to modulate the degradation of the other. These results suggest that the retinoid-induced degradation of RARs and RXRs could play an important role in the control of the activity of RAR/RXR heterodimers.

EXPERIMENTAL PROCEDURES
Chemicals-Calpain inhibitor II (ALLN) was from Roche Molecular Biochemicals, and MG-132 was from France Biochem.
Electrophoretic Mobility Shift Assays (EMSA)-EMSAs were done as described by Zechel et al. (45,46) with 32 P-labeled DR5G oligonucleotides. Recombinant proteins used were Escherichia coli-expressed histidine-tagged hRAR␣1 and mRXR␣1 obtained by in vitro transcription and translation using the TNT-coupled reticulocyte lysate system (Promega). DNA-protein complexes were separated on a 5% nondenaturing polyacrylamide gel and detected by autoradiography.
Cells, Transfections, Immunoprecipitations, and Western Blotting-F9 cells were cultured as monolayer on gelatinized surfaces (23). RAR␥ Ϫ/Ϫ F9 cells reexpressing RAR␥2WT (clone 51) and RAR␥2S66A/ S68A were as described (8). COS-1 cells were grown and transiently transfected by using the calcium phosphate technique (47). In addition to the expression vectors (5 g), all transfections contained Bluescript as a carrier and 1 g of the ␤-galactosidase expression vector pCH110 to correct for variations in transfection efficiency (37). After a 16-h incubation with calcium phosphate-precipitated DNA, the cells were washed, maintained for 8 h in the appropriate medium, and incubated for an additional 15 h with ligand. Cells were harvested 40 h after transfection.
Whole cell extracts were prepared from F9 cells or transfected COS-1 cells by four cycles of freeze-thaw in 50-200 l of 10 mM Tris-HCl (pH 8) containing 0.6 M KCl, 1.5 mM EDTA, and a protease inhibitor mixture and resolved by SDS-10% polyacrylamide gel electrophoresis. After electrotransfer onto nitrocellulose membranes, RARs and RXRs were detected by immunoblotting and chemiluminescence according to the manufacturer's protocol (Amersham Pharmacia Biotech). For immunoprecipitations, whole cell extracts were incubated with Protein A-Sepharose beads cross-linked with the appropriate antibodies (5), and the immunoprecipitated proteins were detected by immunoblotting and chemiluminescence.

RA-induced Degradation of RAR␥2 in Transfected COS-1
Cells Depends on Its Heterodimerization with RXR␣-To investigate whether retinoids decrease mRAR␥2 protein levels, transiently transfected COS-1 cells were treated for 15 h with tRA (10 Ϫ7 M) or vehicle. Under these conditions, mRAR␥2 is transcriptionally active upon tRA treatment (6,37). Aliquots of whole cell extracts normalized for ␤-galactosidase were analyzed by immunoblotting with a RAR␥ antibody and revealed substantial mRAR␥2 protein, while this receptor was undetectable in control untransfected cells. However, RAR␥2 levels were not affected up to 24 h of tRA treatment (Fig. 1A, lanes 1 and 2). The same observation was made with a RAR␥-specific agonist (BMS961), added either alone or in association with a pan-RXR agonist (BMS649) (Fig. 1A, lanes 4 -6) (see Ref. 50 for BMS retinoids).
RAR␥2 Degradation Involves the Ubiquitin Proteasome Pathway and Is Modulated by Phosphorylation-To determine whether the decrease in RAR␥2 involves the ubiquitin-proteasome pathway, COS cells overexpressing both RAR␥2 and RXR␣ were incubated with either vehicle or inhibitors of the 26 S proteasome activity, in the absence and presence of tRA. Either ALLN, lactacystin (data not shown), or MG132 (Fig. 1B, lanes 7-9) blocked the decrease in RAR␥2 protein induced by the BMS961/BMS649 combination, indicating that the disappearance of RAR␥2 involves the proteasome pathway.
As most protein substrates for the 26 S proteasome are ubiquitinated, we then investigated whether RAR␥2 was polyubiquitinated in transfected COS cells. RAR␥2 was immunoprecipitated from untreated cells and immunoblotted with ubiquitin antibodies. A ladder of polyubiquitinated forms of RAR␥2 could be detected in control cells (Fig. 1C, lanes 1-3). Ubiquitination of RAR␥2 was increased in intensity upon tRA, 9cRA, and pan-RXR agonist (BMS649) treatment only when RXR␣ was coexpressed (Fig. 1C, compare lanes 1-4, and data not shown). Thus, the retinoid-induced degradation of RAR␥2 can be correlated to an increase in the ubiquitination of the receptor that is dependent on heterodimerization with RXR␣.
In most cases, phosphorylation is a positive signal for ubiquitination and targeting of several known substrates to the 26 S proteasome (28). RAR␥2 is constitutively phosphorylated by TFIIH at two serines that are located in the B region (serines 66 and 68) (6) and belong to a PEST sequence (see Table I).
Since phosphorylation of PEST sequences has been involved in protein instability (53), we studied whether this phosphorylation affects RAR␥2 degradation. Thus, the degradation of RAR␥2 mutated at these phosphorylation sites (RAR␥2S66A/ S68A) was studied in transfected COS cells upon ligand treatment. In the absence of cotransfected RXR␣, RAR␥2S66A/S68A was as stable as RAR␥2WT upon treatment with either tRA or BMS961 (data not shown). However, when heterodimerized with RXR␣, it was not degraded upon treatment with tRA or the synthetic retinoids BMS649 and BMS961 either individu- either alone or in association, as indicated. Whole cell extracts were prepared, and aliquots normalized for ␤-galactosidase were resolved by SDS-polyacrylamide gel electrophoresis and immunoblotted with RP␥(F). B, COS-1 cells were cotransfected with mRAR␥2 and mRXR␣1 expression vectors and treated with the various retinoids, at the same concentrations as in A, either alone or in association as indicated. In lanes 8 and 9, MG132 at 0.1 and 1 M was added. Whole cell extracts were analyzed for RAR␥2 protein levels as in A. C, COS-1 cells were cotransfected with the mRAR␥2WT, mRAR␥2S66A/S68A, or mRAR␥2⌬AF-2 expression vectors in association or not with the mRXR␣1 expression vector and treated with BMS961 (10 Ϫ7 M) and BMS649 (10 Ϫ6 M). Whole cell extracts containing equal amounts of mRAR␥2 were immunoprecipitated with monoclonal antibodies against RAR␥2, followed by immunoblotting with anti-ubiquitin antibodies (upper panel) or RP␥(F) (lower panel). D, COS-1 cells were transfected with mRAR␥2WT, mRAR␥2S66A/S68A, mRAR␥⌬AB, or mRAR␥2⌬AF-2 expression vectors, in association with mRXR␣1WT (lanes 1-8) or mRXR␣⌬AF-2 expression vectors (lanes 9 and 10) and treated for 15 h with the BMS961/649 combination as in C. In lanes 11 and 12, RAR␥2 was transfected in the absence of RXR␣1. RAR␥2 levels were analyzed by immunoblotting as in A.
Note that RAR␥2 deleted for the A/B region (RAR␥⌬AB) was also resistant to ligand-induced degradation and was more stable in the absence of ligand (Fig. 1D, lanes 7 and 8, and Fig.  3A, lanes 7 and 8). Altogether, these results indicate that the proteasome-dependent degradation of RAR␥2 induced by RA requires phosphorylation of the receptor in its B region.
Upon ligand binding to RARs and RXRs, there is a major transconformation change of helix 12 (referred as the AF-2 AD core), located in the C-terminal end of the LBD, which creates a new surface allowing the binding of cofactors such as SUG1, which is a component of the 26 S proteasome (32,34). Thus, the consequences of deleting the AF-2 AD core of RXR␣ (RXR␣⌬AF-2) on RAR␥2 degradation were studied. In the presence of cotransfected RXR␣⌬AF-2, RAR␥2 was resistant to the degradation induced by tRA (data not shown) or by BMS649 either alone or in combination with BMS961 (Fig. 1D, lanes 9 and 10, and Fig. 3A, lanes 9 and 10). In contrast, RAR␥2⌬AF-2 (cotransfected with RXR␣WT) was still degraded (Fig. 1D, lanes 3 and 4, and Fig. 3A, lanes 3 and 4) and polyubiquitinated (Fig. 1C, lanes 5 and 6) as efficiently as RAR␥2WT. Altogether, these results indicate that the transconformational change of RXR␣ helix 12 that occurs upon ligand binding is required for RAR␥2 degradation.

RAR␣1 Degradation in Transfected COS-1 Cells Is Inhibited by Phosphorylation and Increased by Heterodimerization-The
effects of retinoids were also tested on mRAR␣1 protein levels. In transfected COS-1 cells, RAR␣1 was only mildly decreased (by 30 -40%) upon tRA, 9cRA, or RAR␣ agonist (BMS753) treatment, at either 10 Ϫ7 or 10 Ϫ6 M (Fig. 4A, lanes 1-4, and data not shown), while the pan-RXR agonist (BMS649) and the RAR␣ antagonist (BMS614) had no effect (Fig. 4A, lanes 5 and  6). Note, however, that BMS614 reversed the BMS753-induced degradation of RAR␣ (Fig. 4A, lane 8), indicating that the transconformational change of RAR␣ that occurs upon BMS753 binding is required for RAR␣ degradation. In a parallel study (54), this degradation was found to involve ubiquitination and to be blocked by proteasome inhibitors such as ALLN, MG132, and lactacystin, confirming that it occurs through the proteasome pathway.
In order to determine the relative contribution of the AF-2 AD cores from each partner, the same experiments were performed with either RAR␣⌬AF-2 or RXR␣⌬AF-2. Fig. 4C demonstrates that RAR␣⌬AF-2 (in the presence of RXR␣WT) was not degraded in the presence of tRA, 9cRA, or the BMS753/649 combination (see also Fig. 3B, lanes 9 and 10). The same results were obtained in the absence of coexpressed RXR␣ (data not shown). In contrast, RAR␣1 was still degraded in the presence of RXR␣⌬AF-2 ( Fig. 4C and Fig. 3B, lanes 9 and 10). Altogether, these results indicate that the retinoid-induced degradation of RAR␣1 depends on the integrity of its own helix 12 and does not require the AF-2 AD core of its partner, RXR␣.
As in the case of RAR␥2, RAR␣1 is constitutively phosphorylated in its B region at serine 77 (5), which also belongs to a PEST sequence (see Table I). In the absence of cotransfected RXR␣1, RAR␣ deleted for the A/B region (RAR␣⌬AB) or mutated at the phosphorylation site located in this region (RAR␣1S77A) was drastically decreased in the presence of either tRA (data not shown) or the BMS753/649 combination (Fig. 4D, lanes 1 and 2), suggesting that phosphorylation of the A/B region may be inhibitory for RAR␣1 degradation. On the other hand, when heterodimerized with RXR␣, RAR␣1WT, RAR␣⌬AB, and RAR␣1S77A were similarly degraded (Fig. 4D,  lanes 3 and 4, and Fig. 3B, lanes 4 -8). Altogether, these results indicate that, in the absence of heterodimerization, RAR␣1 degradation is inhibited by phosphorylation, while upon dimerization, RAR␣1 degradation occurs independently of its phosphorylation. Interestingly, these results are opposite to those described above for RAR␥2.
The Degradation of RXR␣ in Transfected COS Cells Is Also Dependent on Heterodimerization-Since RXR␣ contributes to the tRA-induced degradation of RAR␥2 and RAR␣1, we also examined the effects of retinoids and of RAR␥2 and RAR␣1 on mRXR␣ levels. In the absence of a heterodimeric partner, RXR␣ (the RXR␣1 isoform) was slightly degraded in the presence of tRA, whereas it was reduced by 50 and 70% by 9cRA and BMS649, respectively (Fig. 5A, lanes 1-4).
When heterodimerized with RAR␥2, RXR␣ became efficiently degraded (by 90 -95%) in the presence of any of the three ligands (Fig. 5B, lanes 1-4). The RAR␥-selective agonist (BMS961) on its own was unable to promote the degradation of heterodimerized RXR␣ (Fig. 5B, lane 5) and did not enhance the effect of the BMS649 pan-RXR agonist (Fig. 5B, lane 7). RXR␣ degradation involved the ubiquitin proteasome pathway, since it was reversed by the proteasome inhibitor MG132 (Fig.  5B, lanes 7-9) and was preceded by an increase in ubiquitination (data not shown). Similar observations were made when a The sequences of the mouse, human, or chicken (represented by m, h, or c) were analyzed using the computer program "WWW PESTfind Analysis" (EMBnet AUSTRIA). Previous phosphorylation sites described (RAR␥ and RAR␣) or potential phosphorylation sites (RAR␤) are underlined. The phosphorylated serines are indicated by boldface type. RXR␣1 was heterodimerized with RAR␣1 (Fig. 5C). In both cases, neither the RAR␣ antagonist BMS614 (Fig. 5C, lane 8) or the pan-RAR antagonist BMS493 (data not shown) reversed the effects of BMS649. Altogether, these results indicate that the degradation of RXR␣ is triggered by its own ligand and is modulated by the presence of the nonliganded heterodimerization partner, RAR␥2 or RAR␣1. Note that this ligand-induced degradation of RXR␣ did not occur when the receptor was mutated in its heterodimerization interface (Fig. 2, A and B), confirming that heterodimerization is also important for RXR␣ degradation.
Deletion of the AF-2 AD core in the RAR␥2 or RAR␣1 partner did not affect the retinoid-induced degradation of RXR␣ (Fig. 5,  D and E, lanes 3 and 4). In contrast, RXR␣ deleted for its AF-2 AD core (RXR␣⌬AF-2) was not degraded (Fig. 5, D and E, lanes 9 and 10). Thus, our results indicate that the retinoid-induced degradation of RXR␣ requires the integrity of its helix 12 but not that of its heterodimeric partner. Note that RXR␣ was less efficiently degraded when heterodimerized with RAR␥2S66A/ S68A or with RAR␥⌬AB (Fig. 5D, lanes 5-8), indicating that phosphorylation of RAR␥2 in its A/B region modulates not only its own degradation but also that of RXR␣. In contrast, the analogous mutations in RAR␣1 did not affect the degradation of RXR␣ (Fig. 5E, lanes 5-8).
In F9 Cells, RA Differentially Decreases the Levels of RAR␥2 and RAR␣1-Wild type (WT) F9 EC cells that endogenously express the mouse RAR␣1, RAR␥2, and RXR␣ proteins (see Introduction) provide an interesting model system for the analysis of their retinoid-induced degradation. Thus, F9 cells were cultured without or with tRA (10 Ϫ7 M) for 8, 24, 48, and 96 h, and mRAR␥2 levels were analyzed by immunoblotting after immunoprecipitation of equal amounts of whole cell extracts with a RAR␥2-specific monoclonal antibody. RAR␥2 was strongly reduced within 48 h and completely disappeared at 96 h (Fig. 6A, compare lanes 6 and 8 with lanes 5 and 7), while it was unaffected in control cells. Similar results were obtained with RAR␥ Ϫ/Ϫ F9 cells reexpressing RAR␥2WT from an heterologous promoter (Fig. 6A; this RAR␥2WT rescue cell line was described by Taneja et al. (8)). Under the same experimental conditions, the level of RAR␥2 RNA transcripts was not af-fected by tRA treatment in the RAR␥2WT rescue line (Fig. 6B,  compare lanes 7-14) and was only reduced by 60% at 96 h in WT F9 cells (Fig. 6B, lane 6), indicating that in F9 cells, the tRA-induced decrease of RAR␥2 results from increased degradation of the protein.
This decrease in RAR␥2 involves the ubiquitin-proteasome pathway, since it was blocked by either ALLN (Fig. 7A, compare lanes 1-4), MG132, or lactacystin (data not shown). Moreover, when exposing the gels for a longer time than in Fig. 7A, a ladder of higher molecular weight species of immunoprecipitated RAR␥2 could be seen in control untreated F9 cells (Fig.  7B, lane 1). This ladder increased in intensity in cells treated with the MG132 inhibitor (Fig. 7B, lanes 1-5). When immunoprecipitated RAR␥2 was immunoblotted with ubiquitin antibodies, a ladder of polyubiquitinated forms of RAR␥2 could be detected even in control cells (Fig. 7, B, lane 6, and C, lane 3) and was increased in intensity upon treatment with MG132 (Fig. 7B, lanes 6 -10). Polyubiquitination of RAR␥2 was also increased after 24 h of tRA treatment before a decrease in the receptor could be noticed (Fig. 7C, lanes 1-4). Thus, in F9 cells, RAR␥2 is polyubiquitinated, and an increase in ubiquitination precedes its tRA-induced degradation by the 26 S proteasome complex.
In F9 cells, as in transfected COS cells, RAR␥2 is constitutively phosphorylated at serines 66 and 68 (6,8). Whether, in these cells, the tRA-induced decrease in RAR␥2 was affected by phosphorylation was investigated by using the RAR␥ Ϫ/Ϫ rescue line reexpressing RAR␥2 mutated at these phosphorylation sites (the RAR␥2S66A/S68A rescue cell line described in Taneja et al. (8)). While RAR␥2WT migrated as a doublet, the "phosphorylation mutant" RAR␥2S66A/S68A migrated as a single species corresponding to the faster migrating, nonphosphorylated form of RAR␥2 (Fig. 6A, compare RAR␥2WT and RAR␥2S66A/S68A) that was completely resistant to tRA-induced degradation (Fig. 6A and Fig. 7A, lanes 5 and 6) and was not affected by the addition of ALLN (Fig. 7A, lanes 5-8). The observation that the upper migrating species of RAR␥2, which correspond to the phosphoreceptor, are preferentially degraded (Fig.6A,comparelanes7and8)corroboratesthattheproteasomedependent degradation of RAR␥2 induced by tRA requires phosphorylation of the receptor. Moreover, treatment of the cells with okadaic acid, which increases the overall phosphorylation of RAR␥2, accelerated its degradation. 2 It must be stressed that phosphorylated RAR␥2 is indispensable for RA-induced differentiation of F9 cells into primitive endoderm, since the RAR␥2S66A/S68A rescue cell line does not differentiate upon tRA treatment (8). Thus, we could not exclude the possibility that the tRA-induced degradation of RAR␥2 in F9WT cells that differentiate upon tRA addition may reflect events related to the differentiation status of the cells rather than related to the liganded status of RAR␥2. In that respect, note that in F9 cells, RAR␥2 degradation occurred at 48 h, when cells start to differentiate into primitive endoderm (8).
RAR␣1 protein levels were also examined after immunoprecipitation with a RAR␣1-specific monoclonal antibody. In cells treated with tRA, RAR␣1 was decreased by only 40% at 48 h (Fig. 6C, compare lanes 4 and 5) and by 60% at 96 h (Fig. 6C,  compare lanes 7 and 8). When combined with cAMP, tRA was more efficient in decreasing RAR␣1 protein levels, which were reduced by 80% at 96 h (Fig. 6C, compare lanes 7 and 9). However, in contrast to RAR␥2, RAR␣1 did not disappear.  1 and 2) or in association with mRXR␣1 expression vector (lanes 3 and 4) and treated with the BMS753/649 combination for 15 h. RAR␣1 levels were analyzed by immunoblotting with RP␣(F) as in A.
Under the same experimental conditions, the level of RAR␣1 RNA transcript was not affected (Fig. 6D), indicating that the decrease in RAR␣1 may also reflect increased degradation of the protein in the presence of tRA. Note that in F9 cells, RAR␣ is dispensable for primitive endodermal differentiation but is required in addition to RAR␥2 and cAMP for the subsequent differentiation into parietal endoderm (8). Thus, as mentioned above, the possibility cannot be excluded that in F9 cells the loss of the receptor (either RAR␥2 or RAR␣1) generally correlates with RA-induced differentiation.

DISCUSSION
It is now well established that RXR/RAR heterodimers are the functional units involved in the transduction of the retinoid signal. Indeed, RAR␥2/RXR␣ heterodimers play a key role in the RA-induced differentiation of the F9 murine embryonal carcinoma cell line into primitive endoderm-like cells and in the induction of many endogenous RA-responsive genes (9 -12). However, one of the unresolved issues for understanding the regulation of cell differentiation and gene expression by retinoid receptors has to do with how cells restrict the duration and magnitude of their activity in response to RA.
In the present study, we have shown that, in F9 cells, RAR␥2 is degraded subsequently to RA treatment. This effect involves the ubiquitin-proteasome pathway, since it is reversed by proteasome inhibitors and is preceded by an increase in the ubiquitination of the receptor. We also found that RAR␥2 degradation requires phosphorylation of the receptor as previously reported for other substrates such as the large subunit of POL II, IB, and ␤-catenin (28, 55,56). Such a role for phosphorylation was essentially supported by the fact that the RAR␥2 phosphorylated species was preferentially degraded compared with the nonphosphorylated one. However, since the cell line FIG. 6. In F9 cells, RAR␥2 protein levels are decreased upon tRA treatment. A, WT F9 cells and RAR␥ Ϫ/Ϫ cells reexpressing either RAR␥2WT or RAR␥2S66A/S68A were treated with either vehicle or tRA (10 Ϫ7 M) for 8, 24, 48, or 96 h. Whole cell extracts were prepared, and equal amounts (as checked by immunoblotting of an unrelated protein, SUG1) were immunoprecipitated with a specific RAR␥2 monoclonal antibody (Ab10␥(A2)) followed by immunoblotting with RP␥(F). B, total RNA from WT F9 cells (lanes 1-6) and from the RAR␥2WT rescue line (lanes 7-14), treated as in A, was subjected to RT-PCR analysis for RAR␥2 using transcripts of the 36B4 gene as an internal control to normalize the amounts of RNA. C, WT F9 cells were treated with vehicle or with tRA (10 Ϫ7 M) combined or not with cAMP (250 M) for 24, 48, and 96 h. Whole cell extracts were prepared and immunoprecipitated with a specific RAR␣1 monoclonal antibody (Ab10␣(A)) followed by immunoblotting with RP␣(F). D, total RNA from WT F9 cells treated as in C was subjected to RT-PCR analysis for RAR␣1 as described in B. All presented results correspond to a representative experiment among two. expressing RAR␥2 mutated at its phosphorylation sites does not differentiate upon RA treatment (8), it is difficult to conclude whether the RA-induced degradation of RAR␥2 in F9WT cells is related to F9 cell differentiation or reflects a distinct property of liganded RAR␥2 per se. Accordingly, RAR␣, which is not involved in RA-induced primitive endodermal differentiation of F9 cells, was not efficiently degraded. Thus, more studies will be required to discriminate between RA-induced differentiation and receptor degradation in these cells.
The mechanism underlying the RA-induced degradation of RARs was therefore further dissected in transfected COS cells, which are "insensitive" to RA. Interestingly, we found that the ligand-dependent RAR␥2 degradation depends strictly on its heterodimerization with RXR␣ and on phosphorylation of its B region. RAR␣1 and RXR␣1 were also found to be degraded by the ubiquitin-proteasome pathway upon ligand binding, but the role played by heterodimerization and phosphorylation appears to be different.
The Ligand-dependent Degradation of RAR␥2 Depends on both Its Phosphorylation and Dimerization with RXR␣-In transfected COS cells, RAR␥2 is degraded upon RA treatment only when coexpressed with RXR␣. In order to dissect the role of each partner in this process, their degradation was studied following treatment with specific agonists and/or antagonists of each receptor. By using this strategy, we found that a RXRspecific, but not a RAR␥-specific ligand, induced the degradation of the two receptors and that there was no synergism between the two ligands. These results are in contrast to those previously reported concerning transactivation by RAR/RXR heterodimers (14 -16), and thus there is low probability for the activation of a cascade of RA-inducible genes leading to the synthesis of components of the proteasome. However, they raised the question of how liganded RXR can induce the degradation of its dimerization partner, RAR␥2.
One possibility would be that liganded RXR␣ induces an allosteric transconformation change in its partner RAR␥2 (57,58), allowing the generation of an interaction surface and the subsequent recruitment of the proteasome through some of its components, e.g. SUG1 (32). However, our results are not consistent with this idea, since RAR␥2 deleted for its helix 12 (AF-2 AD core) was efficiently degraded. In fact, the degradation of RAR␥2 (as that of RXR␣) appears to require the integrity of the AF-2 AD core of RXR␣. Determination of whether the proteasome is in fact recruited by RXR␣ through SUG1, as recently suggested in a report describing the degradation of the vitamin D receptor (35), would require further investigations. In an alternative model, the degradation apparatus might recognize a complex of coactivators bound to ligand-activated RXR. Note that similar possibilities have been recently proposed concerning the ligand-induced degradation of another nuclear receptor, Peroxisome Proliferator-activated Receptor ␥ (59).
Interestingly, our results highlight the role of RAR␥2 phosphorylation and ubiquitination in the degradation of the two partners of RAR␥2/RXR␣ heterodimers. Indeed, we found that RAR␥2 responds to liganded RXR␣ through an increase in its ubiquitination that is completely dependent on the phosphorylation of RAR␥2 at serines 66 and 68 located in its B region (6). Liganded RXR␣ may cause an allosteric transconformation in its partner, which would result in an increase in RAR␥2 ubiquitination that is dependent on the recognition of phosphorylated serines 66 and 68 by the ubiquitin machinery, thus falicitating its degradation by the proteasome. Note in this respect that dimerization has also been shown to be required for ubiquitination and degradation of the ATF-2 transcription factor (60) and that phosphorylation has been reported to be a signal for ubiquitination and subsequent degradation of various substrates (28, 55,56). Since phosphorylation of the AF-1 containing A/B region of other nuclear receptors, such as ER␤ and SF-1 has been shown to lead to the ligand-independent recruitment of coactivators by this domain (61,62), phosphorylation of the A/B region of RAR␥2 may similarly participate to the recruitment of components of the ubiquitination machinery as in the case of ␤-catenin, IB, and E2F-1 (28, 63) or of the proteasome. Interestingly, as already reported for RAR␣ (5), RAR␥2 residues 66 and 68 are constitutively phosphorylated by the Cdk7 component of the general transcription/DNA repair factor TFIIH (6), and TFIIH is known to interact with SUG1 and to recruit the proteasome complex (64). This raises the interesting possibility that the degradation of RAR␥2 could also be related to its interaction/phosphorylation by the TFIIH general transcription factor.
In conclusion, our results indicate that in RAR␥2/RXR␣ heterodimers, the degradation of RAR␥2 is entirely dependent on its heterodimerization with RXR␣. In addition, they highlight the crucial role played by RAR␥2 phosphorylation and ubiquitination on its degradation.
Phosphorylation Modulates the Heterodimerization Requirement for Retinoid-induced RAR␣1 Degradation-We also addressed the question whether the requirements for RAR␥2 degradation could be generalized to other RARs. In contrast to what we observed for RAR␥2, RAR␣1 degradation could be triggered by its own ligand in the absence of heterodimerization, according to previous reports (54,65), but was dependent on the integrity of its AF-2 AD core (helix 12). Moreover, we found that, in the absence of RXR, RAR␣1 was more efficiently degraded when either deleted for its A/B region or nonphosphorylated in this region, thus suggesting that phosphorylation of the A/B region prevents RAR␣1 degradation, as previously described for c-Jun (66). Interestingly, heterodimerization with RXR␣ increased the degradation of phosphorylated RAR␣1 but had no further effect on that of unphosphorylated RAR␣1.
Thus, the requirements for retinoid-induced degradation of RAR␣ appear to be different from those of RAR␥, since heterodimerization was required only when RAR␣ was phosphorylated. Further studies are necessary to establish whether and how these differences may reflect specific functions of RAR␣ and RAR␥ in retinoid signaling under physiological conditions. RXR␣ Degradation Is Dependent on Heterodimerization and Differentially Modulated by Phosphorylation of Its RAR Partners-Our present data demonstrated that RXR␣ degradation is also dependent on its heterodimerization with RAR in both RAR␣1/RXR␣ and RAR␥2/RXR␣ heterodimers. Indeed, RXR␣ degradation, although dependent on its own ligand, was potentiated by the presence of the unliganded RAR. However, RXR␣ degradation was dependent on the integrity of its AF-2 AD core. In addition, phosphorylation of the RAR partner appears to modulate RXR␣ degradation differently, since phosphorylation of RAR␥, but not that of RAR␣, is critical for RXR␣ degradation.
Relationship between RAR/RXR Degradation and Their Functional Activity-The ubiquitin-proteasome pathway plays a crucial role in the regulation of the intracellular levels of a wide range of regulatory proteins, including transcriptional activators, implicated in the control of key cellular functions such as cell cycle progression, signal transduction, cell differentiation, and cell death (Ref. 67 and references therein). The present study shows that, in F9 cells, RAR␥/RXR␣ heterodimers that play a key role in RA-induced primitive differentiation and in the induction of expression of many endogenous RA-responsive genes, are preferentially degraded, compared with RAR␣/RXR␣ heterodimers that have more restricted effects on these events (8). These results support the suggestion that, in vivo, highly potent inducers of transcription may be favorite targets for degradation by the proteasome (67).
We therefore propose a model in which the ubiquitin-proteasome pathway may contribute to the regulation of duration and magnitude of retinoid action. Unliganded RAR␥ is phosphorylated (6,8) and weakly ubiquitinated (this study). Upon ligand binding, the RAR␥/RXR heterodimers bound to cognate response elements recruit coactivators (68,69), and transcription is activated. The present study shows that there is a subsequent increase in ubiquitination of RAR␥, which is dependent on its phosphorylation in the N-terminal B region and involves the heterodimeric RXR partner. This ubiquitination paves the way to the degradation of both partners of the heterodimer by the proteasome, possibly through interaction with its SUG1 component. Taken together, our results support a model in which several events, phosphorylation, heterodimerization, and ligand-dependent ubiquitination, act in concert to trigger proteasome-mediated receptor degradation, thus controlling the magnitude of the retinoid effect by modulating the intracellular levels of RAR/RXR heterodimers.