Negative Modulation of RXRα Transcriptional Activity by Small Ubiquitin-related Modifier (SUMO) Modification and Its Reversal by SUMO-specific Protease SUSP1*

Retinoid X receptor α (RXRα) belongs to a family of ligand-activated transcription factors that regulate many aspects of metazoan life. Here we demonstrate that RXRα is a target substrate of a small ubiquitin-related modifier (SUMO)-specific protease, SUSP1, which is capable of controlling the transcriptional activity of RXRα. RXRα was modified by SUMO-1 in vivo as well as in vitro, and the Lys-108 residue within the IKPP sequence of RXRα AF-1 domain was identified as the major SUMO-1 acceptor site. Prevention of SUMO modification by Lys-to-Arg mutation led to an increase not only in the transcriptional activity of RXRα but also in the activity of its heterodimeric complex with retinoic acid receptor-α or peroxisome proliferator-activated receptor-γ (PPARγ). SUSP1 co-localized with RXRα in the nucleus and removed SUMO-1 from RXRα but not from androgen receptor or PPARγ. Moreover, overexpression of SUSP1 caused an increase in the transcriptional activity of RXRα, whereas small hairpin RNA-mediated knockdown of endogenous SUSP1 led to a decrease in RXRα activity. These results suggest that SUSP1 plays an important role in the control of the transcriptional activity of RXRα and thus in the RXRα-mediated cellular processes.

many cell types (1,2). RA exerts its effects through two classes of nuclear receptors acting as ligand-dependent transcriptional regulators: the retinoic acid receptors (RARs), which bind either all-trans-retinoic acid or 9-cis-RA (9cRA); and the retinoid X receptors (RXRs), which bind 9cRA only. There are three RAR isotypes (␣, ␤, 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 N-terminal region. Each receptor has an N-terminal A/B region that harbors the ligand-independent activation function-1 (AF-1), a central C region containing a DNA-binding domain, and a C-terminal E region containing a ligand-binding domain and a ligand-dependent AF-2.
RXRs play important roles in numerous nuclear receptor-dependent signaling pathways (3). Not only can an RXR function as a homodimer but also as an obligate heterodimeric partner for many other receptors, including those for retinoic acid, thyroid hormone, vitamin D, prostanoids, oxysterols, bile acids, and xenobiotics, as well as several orphan receptors. The homoand heterodimeric complexes of RXR target specific DNA sequences known as hormone response elements. Correct receptor dimerization on target DNA is required for the recruitment of necessary co-activator or co-repressor proteins to the transcription complex (2,4).
Sumoylation is a reversible process, and several SUMO-specific proteases have been described (6 -13). Sequence analyses have suggested the presence of seven SUMO protease genes in mammals, which encode proteins with diverse N-terminal domain and conserved catalytic C-terminal domain. These enzymes include SENP3 (SMT3IP1), which localizes to the nucleolus (33); SUSP1 (SENP6), found primarily in the cytoplasm when GFP is fused to its N terminus (34); SENP1, which localizes to foci in the nucleus and the nuclear rim (35); and SENP2, found in at least in three different isoforms derived from alternatively spliced mRNAs (Axam, SMT3IP2/Axam2, SuPr-1). Axam localizes to the nucleoplasmic face of the nuclear pore complex (36,37), and Axam2 and SuPr-1 have been detected in the cytosol and PML bodies, respectively (38,39). Substrate specificity between different SUMO-specific proteases has been suggested based on the observed differences in subcellular localization (12).
In the present studies we show that RXR␣ is covalently modified by SUMO-1 in vivo as well as in vitro and that this SUMO-1 modification negatively regulates the transcriptional activity of the receptor. We further show that SUSP1 specifically removes SUMO-1 from RXR␣ and thereby enhances the transcriptional activity of the nuclear receptor. Thus, it appears that reversible SUMO modification serves as an important mechanism for regulation of RXR␣-mediated transcriptional activation.

EXPERIMENTAL PROCEDURES
Plasmid Constructions-pcDNA3-FLAG-RXR␣ was constructed by amplification of the RXR␣ coding sequence by PCR using a 5Ј primer containing BglII site and a 3Ј primer containing NotI site. The PCR fragment was ligated into BamHI and NotI sites of pcDNA3-FLAG. To eliminate the SUMO-1 acceptor site(s) in RXR␣, Lys residues were replaced by Arg upon site-directed mutagenesis using pcDNA3-FLAG-RXR␣ as the template. To generate RXR␣ deletion mutants (amino acid sequences 1-100, 1-200, 1-300, 1-400, 101-462, 201-462, 301-462 and 401-462), the PCR fragments corresponding to the sequences were ligated into BamHI and NotI sites in pcDNA3-FLAG. To obtain vectors expressing SUMO-1, the coding region for the mature form of SUMO-1 was amplified by PCR using primers containing BamHI and XhoI sites. The PCR fragments were ligated into pGEX4T-2 (Amersham Biosciences) and pcDNA4-HisMax/C (Invitrogen), and the resulting vectors were referred to as pGEX-SUMO-1 and pcDNA4-HisMax-SUMO-1, respectively. Expression vectors for GAL4-fused RXR␣ deletion mutants were constructed by ligating the PCR fragments for the correspond-ing regions into pM vector (Clontech). pcDNA3.1-SUSP1-V5 expressing SUSP1 fused to the N terminus of V5 tag was prepared by TA cloning of its coding sequence into pcDNA3.1/V5/His-TOPO (Invitrogen) as specified by the manufacturer. pcDNA3.1-SUSP1/C1030S-V5 was generated by replacement of active site Cys-1030 by Ser upon sitedirected mutagenesis. pEGFP-C1-SUSP1 was prepared as described previously (34).
Protein Purification-GST-SUMO-1, GST-SAE2/SAE1, and GST-RXR␣ were overproduced in BL21-CodonPlus bacteria (Stratagene). Each protein was then purified by using a glutathione-Sepharose 4B column. Purified GST-SUMO-1 was treated with thrombin and applied again onto the GST-affinity column. SUMO-1 was then recovered as the unbound protein.
GST Pulldown Assay-GST (10 g) or GST-RXR␣ (10 g) was incubated with 10 l of glutathione-Sepharose 4B beads for 1 h at 4°C. The samples were further incubated with 1 g of His-UBC9 in the presence or absence of 1 M 9cRA for 2 h at 4°C. Beads were washed five times with 1 ml of the pulldown buffer consisting of 50 mM Tris-HCl (pH 7.8), 150 mM NaCl, 0.1% Nonidet P-40, 0.1% Triton X-100, 1 mM EDTA, 10% glycerol, 5 mM MgCl 2 , and 1ϫ protease inhibitor mixture (Roche Applied Science). Proteins bound to the beads were released by boiling in 2ϫ SDS-PAGE sampling buffer and analyzed by immunoblot with anti-His monoclonal antibody (Qiagen).
Cell Culture and Transfections-All cells were cultured in DMEM supplemented with 10% fetal bovine serum (FBS), 25 units/ml penicillin, and 25 units/ml streptomycin. Cells were transfected with the appropriate vectors using Lipofectamine Plus reagent (Invitrogen). The total amounts of transfected vectors in each experiment were equalized by supplementing empty vector DNA.
In Vitro Sumoylation and Desumoylation Assays-For the in vitro sumoylation assay, RXR␣ or its mutant forms were radiolabeled using an in vitro transcription and translation system (Promega) in the presence of [ 35 S]Met. The labeled proteins were then subjected to SUMO modification by incubation at 37°C for 2 h with 10 g of SUMO-1, 10 g of His-UBC9, 3 g of GST-SAE2/SAE1, and an ATP-regenerating system (50 mM Tris-HCl, pH 7.6, 5 mM MgCl 2 , 2 mM ATP, 10 mM creatine phosphate, 3.5 units/ml creatine kinase, and 0.6 unit/ml inorganic pyrophosphatase). The reaction was terminated by the addition of 2 ϫ SDS-PAGE sampling buffer followed by boiling. The samples were resolved by SDS-PAGE and visualized by autoradiography.
For in vitro desumoylation assay, purified GST-RXR␣ was sumoylated as described above followed by incubation at 37°C with partially purified SUSP1 or SUSP1/C1030S in 10 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 1 mM dithiothreitol. The reaction was terminated by addition of 2ϫ SDS-PAGE sampling buffer and boiling. The samples were resolved by SDS-PAGE followed by immunoblot with anti-RXR␣ antibody (Santa Cruz Biotechnology).
Immunocytochemistry-HeLa cells were grown on coverslips and transfected with the appropriate vectors. After incubation for 36 h, cells were fixed by incubation with 2% formaldehyde in PBS for 30 min. They were washed three times with PBS containing 0.1% Triton X-100, permeabilized with PBS containing 0.5% Triton X-100 for 5 min, and incubated with PBS containing 0.1% Triton X-100, 10% normal goat serum, 1% bovine serum albumin, and 1% gelatin. Cells were incubated for 1 h with rabbit anti-FLAG antibody (Sigma), mouse anti-V5 antibody (Invitrogen), or anti-SUSP1 antibody (Abgent) in PBS containing 3% bovine serum albumin and 0.1% Triton X-100. After washing three times with PBS containing 0.1% Triton X-100, cells were incubated for 1 h with FITC-labeled goat antimouse IgG and TRITC-labeled goat anti-rabbit IgG or FITClabeled goat anti-rabbit IgG in PBS containing 3% bovine serum albumin and 0.1% Triton X-100. After washing, cells were observed under a Carl Zeiss LSM510 confocal microscope or a Zeiss Axioplan II microscope.
Knockdown of SUSP1 mRNA-pSM2c-SUSP1 expressing SUSP1-specific short hairpin RNAs (nucleotides 2679 -2698 from NM_015571), referred to as shRNA, was purchased from Open Biosystems. To knock down SUSP1 mRNA, HEK293T cells were transfected with shRNA or a negative control vector (shControl). Total RNAs were prepared from cells by extracting them with Trizol (Invitrogen) and were subjected to reverse transcription-PCR. Reverse transcription reactions were performed using Superscript III reverse transcriptase (Invitrogen) and oligo(dT) primer by following the manufacturer's instruction. SUSP1 mRNA was amplified with primers (5Ј-GCT GTA ATT GAT TCC AAT CC-3Ј and 5Ј-AGT CAA TCT GAG ATA CTA TTG ACA C-3Ј), and ␤-actin mRNA was amplified as an expression control.

RESULTS
SUMO Modification of RXR␣-To identify novel proteins that interact with RXR␣, we performed yeast two-hybrid screening by using RXR␣ as bait from a Xenopus embryo cDNA library. Among 5 independently identified clones, 2 clones encoded sequences identical to a SUMO-specific protease, SUSP1. Both clones contained the N-terminal regions of Xenopus SUSP1 spanning the amino acid sequence 5-826, which corresponds to that of the human SUSP1 sequence 5-813 containing a part of catalytic domain called the His box (34). Because SUSP1 interacted with RXR␣ (see below), we hypothesized that RXR␣ could be a target for SUMO modification. To test this hypothesis, we first examined whether RXR␣ could interact with UBC9, a SUMO-conjugating E2 enzyme. Purified GST-RXR␣ was incubated with His-UBC9 in the absence or presence of 9cRA. A pulldown assay using glutathione-conjugated Sepharose resin revealed that GST-RXR␣ could co-precipitate with UBC9 whether or not 9cRA was present (Fig. 1A), indicating that RXR␣ directly interacts with UBC9 and 9cRA does not influence their interaction. These results also suggest that RXR␣ is a target of SUMO modification.
To determine whether RXR␣ can be sumoylated, 35 S-labeled RXR␣ was prepared and incubated with purified SUMO-1, E1 (SAE1/SAE2), and E2 (UBC9). Fig. 1B shows that RXR␣ can be modified by SUMO-1 in vitro. We also examined whether RXR␣ could be sumoylated in vivo. FLAG-RXR␣ was expressed in HEK293T cells with HisMax-SUMO-1, FLAG-UBC9, or both. Cell lysates were prepared and subjected to immunoblot with anti-RXR␣ antibody. A slow migrating band could be seen in the lysates from the cells expressing SUMO-1 (Fig. 1C). Moreover, co-expression of UBC9 led to an increase in the intensity of the slow migrating band as well as of SUMOconjugated cellular proteins, suggesting that RXR␣ can be modified by SUMO-1. We then examined whether endogenous RXR␣ could be modified by SUMO-1. Lysates obtained from HEK293T cells were subjected to immunoprecipitation with anti-RXR␣ antibody followed by immunoblot with anti-FIGURE 1. Interaction of RXR␣ with UBC9 and sumoylation of RXR␣. A, purified His-UBC9 was incubated with GST or GST-RXR␣ at 4°C for 2 h and then with glutathione-conjugated Sepharose 4B for the next hour. Proteins bound to the resin were pulled down (PD), and subjected to SDS-PAGE followed by immunoblot with anti-His antibody. B, in vitro translated 35 S-labeled RXR␣ was incubated with purified SUMO-1, His-UBC9, and GST-SAE2/SAE1 at 4°C for 2 h. The samples were then subjected to SDS-PAGE followed by autoradiography. C, pcDNA3-FLAG-RXR␣ was transfected to HEK293T cells with pcDNA4-HisMax-SUMO-1, pcDNA3-FLAG-UBC9, or both. After incubation for 36 h, cells were collected and their lysates subjected to immunoblot with anti-FLAG antibody. The lysates were also probed with anti-Xpress antibody. D, lysates were obtained from HEK293T cells and subjected to immunoprecipitation (IP) with anti-RXR␣ antibody followed by immunoblot with anti-SUMO-1 antibody (left panel ). The same cell lysates were subjected to immunoprecipitation with anti-SUMO-1 antibody followed by immunoblot with anti-RXR␣ antibody (right panel ).
SUMO-1 antibody. A SUMO-conjugated band could be seen in the lysates precipitated by anti-RXR␣ antibody but not in those precipitated by control IgG (Fig. 1D, left panel), suggesting that endogenous RXR␣ can be sumoylated. As a confirmation, the same cell lysates were subjected to immunoprecipitation with anti-SUMO-1 antibody followed by immunoblot with anti-RXR␣ antibody. A SUMO-conjugated band again appeared in the lysates that had been precipitated by anti-SUMO-1 antibody but not in those precipitated by control IgG (Fig. 1D, right  panel). These results indicate that RXR␣ is a natural substrate for SUMO modification.
Determination of SUMO Acceptor Site in RXR␣-SUMO modification to target proteins occurs at specific Lys residues, which are commonly embedded in a consensus sequence, -Lys-X-Glu (20,40). RXR␣ contains two similar sequences, MK 201 RE and PK 245 TE in the hinge region and ligand-binding domain, respectively. Therefore, we first examined whether the Lys residues might serve as the SUMO acceptor sites upon substitution of each residue with Arg by site-directed mutagenesis.
In vitro sumoylation assay revealed that neither the K201R nor the K245R mutation affected the SUMO modification of RXR␣ (data not shown). Neither did the double mutation (K210R/ K245R) show any effect on RXR␣ sumoylation, indicating that Lys-210 and Lys-245 in the putative consensus sequences are not the SUMO acceptor sites in RXR␣.
The Lys residues in non-consensus sequences have also been shown to serve as the SUMO acceptor sites (41,42). To identify the SUMO acceptor site(s) in RXR␣, we constructed serial deletion mutants ( Fig. 2A) and subjected them to an in vitro sumoylation assay. As shown in Fig. 2B, slow migrating, sumoylated bands were not detected with the RXR␣ mutant proteins lacking the amino acid sequence from 101 to 200. These results suggest that the SUMO acceptor site is located in the RXR␣ sequence 101-200, which contains 10 Lys residues. To identify the SUMO acceptor Lys residue(s) in the 101-200 sequence, each of 10 Lys residues was replaced by Arg, and the resulting mutant proteins were subjected to in vitro sumoylation assay. Among the Lys-to-Arg mutations, only the K108R mutation blocked the sumoylation of RXR␣ (Fig. 2C), indicating that Lys-108 is the major acceptor site for SUMO under in vitro conditions. To determine whether the K108R mutation could prevent in vivo sumoylation of RXR␣, HisMax-SUMO-1 was expressed in HEK293T cells with FLAG-RXR␣ or FLAG-RXR␣/K108R. Immunoprecipitation analysis using anti-FLAG antibody demonstrated that wild-type RXR␣ (wt), but not RXR␣/K108R (mt), can be sumoylated (Fig. 2D). These results indicate that Lys-108 serves as the major SUMO acceptor site in RXR␣ under both in vivo and in vitro conditions.
Effect of K108R Mutation on Transcriptional Activity of RXR␣-SUMO modification has been shown to influence the activity of many transcription factors. To determine whether the K108R mutation might affect the transcriptional activity of RXR␣, the DNA-binding domain of GAL4 was fused to various deletions of RXR␣ or RXR␣/K108R (Fig. 3A). HEK293T cells were transfected with vectors expressing the GAL4-fused RXR␣ mutants and a luciferase reporter vector. They were then cultured in the absence or presence of 9cRA. The transcriptional activity of GAL4-AF1/K108R was more than 1.5-fold higher than that of GAL4-AF1; this effect of the K108R mutation was independent of 9cRA, as the constructs do not contain the ligand-binding domain (Fig. 3B, lower panel). The 9cRAstimulated activities of GAL4-RXR␣/K108R and GAL4-AF1/ K108R-LBD lacking the DNA-binding domain were ϳ2-fold higher than those of their parental forms, respectively. The upper panel of Fig. 3B shows the expression levels of RXR␣ deletions as a control for transfection efficiencies. These results suggest that the transcriptional activity of RXR␣ is negatively regulated by SUMO modification.
RXR␣ is a combinational partner in the nuclear receptor family and can form a homodimer by itself and heterodimers with a variety of hormone and orphan receptors. The dimerization partners of RXR␣ include RAR, PPAR, TR (thyroid hormone receptor), VDR (vitamin D receptor), and so on (4). To determine whether the K108R mutation might also influence the transcriptional activity of heterodimeric RXR␣, FIGURE 2. Determination of SUMO-1 conjugation site in RXR␣. A, various deletion constructs of RXR␣ were generated as shown. A summary of the results from B is shown on the right: ϩ, constructs that were SUMO-conjugated; Ϫ, those that were not SUMO-conjugated. B, 35 S-labeled RXR␣ and its deletion mutants were subjected to a sumoylation assay in the absence or presence of SUMO-1 as described under "Experimental Procedures." Proteins were resolved by SDS-PAGE and visualized by autoradiography. The arrowheads indicate the sumoylated proteins. C, each of 10 Lys residues in the RXR␣ sequence 101-200 was replaced by Arg, and the resulting mutant proteins were subjected to a sumoylation assay as described in B. The numerals shown above the gels indicate the position of each Lys residue in the sequence. D, pcDNA3-FLAG-RXR␣ (wt) or pcDNA3-FLAG-RXR␣/K108R (mt) was transfected to HEK293T cells with or without pc-DNA4-HisMax-SUMO-1. After incubation for 36 h, cells were subjected to immunoprecipitation (IP) with anti-FLAG antibody followed by immunoblot with anti-Xpress antibody. Cell lysates were also directly probed with anti-FLAG antibody.
HEK293T cells were transfected with luciferase reporter vectors containing various response elements and vectors expressing the RXR␣ binding partners. They were then cultured in the absence or presence of corresponding ligands (i.e. 9cRA for RXR␣ homodimer; all-trans-RA for RAR␣/ RXR␣ heterodimer, and rosiglitazone for PPAR␥/RXR␣ heterodimer). As a control, we first examined the effect of K108R mutation on the transcriptional activity of RXR␣ homodimer using the cells transfected with p3xPPRE-Luc and a RXR␣ expression vector. RXR␣/K108R had about 3-fold higher activity than wild-type RXR␣ (Fig. 4A), again indicating that SUMO modification negatively regulates the transcriptional activity of RXR␣ homodimer. In the cells transfected with pRARE-tk-Luc and expression vectors for both RXR␣ and RAR␣, the K108R mutation caused about a 1.5-fold increase in the transcriptional activity (Fig. 4B).
Similarly, the same mutation caused about a 2-fold increase in the transcriptional activity in cells transfected with p3xPPRE-Luc and expression vectors for both RXR␣ and PPAR␥ (Fig. 4C). The insets in Fig. 4 show the expression levels of RXR␣ and RXR␣/K108R as a control for transfection efficiencies. Taken together, these results suggest that SUMO modification negatively regulates the transcriptional activity of RXR␣ heterodimers as well as of RXR␣ homodimer.
Interaction of RXR␣ with SUSP1 and Their Subcellular Localization-Interaction of RXR␣ with SUSP1 was initially identified by yeast two-hybrid screening. To confirm the interaction of RXR␣ with SUSP1, FLAG-RXR␣, SUSP1-V5, or both were expressed in HEK293T cells. Immunoprecipitation of cell lysates with anti-FLAG antibody shows that SUSP1 can be coprecipitated with RXR␣ and RXR␣/K108R (Fig. 5A), indicating that these proteins interact with each other in vivo and that  SUMO modification of RXR␣ is not required for its binding to SUSP1. We then examined whether endogenous RXR␣ and SUSP1 could interact with each other. Lysates were prepared from HEK293T cells and subjected to immunoprecipitation with anti-RXR␣ antibody followed by immunoblot with anti-SUSP1-antibody. Fig. 5B shows that RXR␣ co-precipitates with SUSP1, indicating that the endogenous proteins interact with each other in cells.
RXR␣ localizes in the nucleus. However, we have previously reported that SUSP1 fused to the C terminus of GFP (i.e. GFP-SUSP1) resides in the cytoplasm (34). These findings provoke a discrepancy in the interaction of two proteins that are localized in different cellular compartments. To clarify this discrepancy, we first expressed GFP-SUSP1 in HeLa cells with FLAG-RXR␣. Cells were then fixed and stained with anti-FLAG antibody. FLAG-RXR␣ appeared exclusively in the nucleus as expected, whereas GFP-SUSP1 localized predominantly in the cytoplasm (Fig. 5C, upper panels) in accord with our previous report (34). Therefore, we suspected that the presence of GFP at the N terminus of SUSP1 might interfere with the translocation of SUSP1 to the nucleus. To test this possibility, FLAG-RXR␣ was expressed in cells with SUSP1 fused to the N terminus of V5 (i.e. SUSP1-V5). Immunostaining of the cells shows that FLAG-RXR␣ and SUSP1-V5 co-localize in the nucleus (Fig. 5C, lower  panels). SUSP1-GFP, unlike GFP-SUSP1, also co-localized with FLAG-RXR␣ in the nucleus (data not shown), indicating that the position of tagged proteins alters the subcellular localization of SUSP1. We then examined the subcellular localization of endogenous SUSP1 by staining cells with anti-SUSP1 antibody. SUSP1 was stained almost exclusively in the nucleus (Fig.  5D). Collectively, these results indicate that SUSP1 is a nuclear protein.
Desumoylation of RXR␣ by SUSP1-Because SUSP1 interacts with RXR␣, we examined whether SUSP1 shows desumoylating activity toward RXR␣ in vitro. GST-RXR␣ that had been modified by SUMO-1 was incubated with partially purified SUSP1 or SUSP1/C1030S, of which the active site Cys-1030 was replaced by Ser. Fig. 6A shows that wild-type SUSP1 (Wt), but not SUSP1/C1030S (C/S), is capable of cleaving SUMO-1-conjugated RXR␣ in a time-dependent fashion, suggesting that RXR␣ is a target substrate of SUSP1. To determine whether FIGURE 5. Interaction of RXR␣ with SUSP1 and their subcellular localization. A, pcDNA3-FLAG-RXR␣ or pcDNA3-FLAG-RXR␣/K108R was transfected to HEK293T cells with or without pcDNA3.1-SUSP1-V5. Cell lysates were prepared and subjected to immunoprecipitation (IP) by anti-FLAG antibody followed by immunoblot with anti-V5 antibody. The lysates were also directly probed with anti-FLAG or anti-V5 antibody. Wt, wild type. B, lysates were prepared from HEK293T cells and subjected to immunoprecipitation with anti-RXR␣ antibody followed by immunoblot with anti-SUSP1-antibody. C, pcDNA3-FLAG-RXR␣ was transfected to HeLa cells with pEGFP-C1-SUSP1 (upper panels) or pcDNA3.1-SUSP1-V5 (lower panels). Cells were then fixed and stained with anti-FLAG or anti-V5 antibody. D, HeLa cells were stained withanti-SUSP1antibody.Thenucleiwerestainedwith4,6-diamidino-2-phenylindole (DAPI). The bars indicate 10 m. SUSP1 could also cleave off SUMO-1 from RXR␣ under in vivo conditions, FLAG-RXR␣, HisMax-SUMO-1, and FLAG-UBC9 were expressed in HEK293T cells with SUSP1 or SUSP1/ C1030S. Immunoblot analysis of cell lysates with anti-FLAG antibody shows that SUSP1, but not SUSP1/C1030S, could cleave SUMO-1-conjugated RXR␣ (Fig. 6B). On the other hand, SENP1 showed relatively little activity toward SUMO-1conjugated RXR␣, suggesting that RXR␣ is a specific target of SUSP1. We also examined whether SUSP1 could remove SUMO from other nuclear receptors. However, neither AR nor PPAR␥ was desumoylated by SUSP1 (Fig. 6C), again suggesting that SUSP1 acts specifically on RXR␣.
Effect of SUSP1 on Transcriptional Activity of RXR␣-To determine whether SUSP1 is involved in the control of RXR␣ transcriptional activity, SUSP1 was expressed in HEK293T cells with GAL4-RXR␣ or GAL4-RXR␣/K108R. Increased expression of SUSP1 led to a gradual increase in the transcriptional activity of GAL4-RXR␣ (Fig. 7A). On the other hand, SUSP1 at all of the concentrations tested showed little or no effect on the transcriptional activity of GAL4-RXR␣/K108R. These results again show that the transcriptional activity of RXR␣ is negatively regulated by SUMO modification. These results also suggest that SUSP1 specifically acts on RXR␣ but not on other target proteins, in which SUMO modification might influence the transcriptional activity of RXR␣.
We then examined whether the desumoylating activity of SUSP1 is required for the increase in transcriptional activity of RXR␣. Expression of catalytically inactive SUSP1/C1030, unlike that of wild-type enzyme, showed little or no effect on the transcriptional activity of RXR␣ (Fig. 7B). On the other hand, neither the wild-type nor mutant form of SUSP1 showed any effect on the activity of RXR␣/K108R, which lacks the SUMO acceptor site. In addition, SENP1 did not affect the transcriptional activity of either RXR␣ or RXR␣/K108R, again demonstrating that RXR␣ is a specific target substrate of SUSP1. These results indicate that the desumoylating activity of SUSP1 is required for the positive control of RXR␣ activity.
To determine whether endogenous SUSP1 is indeed involved in the control of transcriptional activity of RXR␣, HEK293T cells were transfected with a SUSP1-specific small hairpin RNA. Transfection of shRNA, but not a control RNA vector (shControl), led to a marked reduction in the level of SUSP1 protein as well as in its transcript level (Fig. 7C, left panel). Consistently, the level of sumoylated RXR␣ in shRNA-transfected cells was significantly higher than that in cells transfected with shControl. Furthermore, SUSP1 knockdown caused a significant decrease in the transcriptional activity of RXR␣ (Fig. 7C, right panel). These results demonstrate that reversible SUMO modification could be a mechanism that regulates the transcriptional activity of RXR␣.

DISCUSSION
In the present studies we have demonstrated that RXR␣ is covalently modified by SUMO-1 and that this SUMO modification negatively regulates the transcriptional activity of RXR␣. We further demonstrated that a SUMO-specific protease, SUSP1, removed SUMO-1 from RXR␣, thereby reversing the sumoylation-mediated repression of RXR␣ activity. Therefore, we suggest that reversible SUMO modification could serve as an important mechanism for the control of RXR␣-dependent transcription. Of note was the finding that the SUMO acceptor site of RXR␣ resides within the IKPP sequence of the AF-1 domain, which appears distinct from the common ⌿KXE sequence. Interestingly, RXR␤ and RXR␥ contain VKPP and IKPL, respectively, in their AF-1 domains, suggesting that the atypical sequences may also serve as the SUMO acceptor sites of the RXR␣ isoforms. Unlike the retinoid receptors, other were transfected to HEK293T cells with pGAL4-RXR␣ or pGAL4-RXR␣/K108R. Cells were also transfected with pGAL4-UAS-Luc and pCMV-␤-galactosidase. After incubation for 24 h, cells were further cultured in DMEM containing 0.2% charcoal-stripped FBS and 1 M 9cRA. Cell lysates were then subjected to a luciferase assay (upper panel ). To confirm the expression of SUSP1, the same lysates were subjected to immunoblot with anti-V5 antibody (lower panel ). The numerals on the top of the gel indicate the amounts of pDNA3.1-SUSP1-V5 DNA transfected to cells. B, pGAL4-RXR␣ or pGAL4-RXR␣/K108R was transfected to cells with pcDNA3.1-SUSP1-V5 (wt), pcDNA3.1-SUSP1/C1030S-V5 (C/S), or pcDNA3.1-SENP1-V5. Cells were then cultured and subjected to luciferase assay as described in A. C, pGAL4-RXR␣ was transfected to cells with pSM2c-SUSP1 (shRNA) or pSM2c (shControl ). Total RNAs levels were obtained and subjected to reverse transcription-PCR to determine the mRNA levels for SUSP1 and ␤-actin. Lysates were prepared from cells and subjected to immunoblot with anti-SUSP1 or anti-␤-actin antibody. Lysates were also subjected to immunoprecipitation (IP) with anti-RXR␣ antibody followed by immunoblot with anti-SUMO-1 antibody (left panels). After transfection, duplicated cells were further cultured in the absence or presence of 1 M 9cRA and subjected to a transcription assay as in A (right panel ). Data represent mean Ϯ S.D. of triplicates.
nuclear receptors, including PPAR␥, AR, and the glucocorticoid and progesterone receptors, have the consensus ⌿KXE sequence as their SUMO acceptor sites (29 -32, 43). However, the sumoylation sites of the latter receptors also are located in their AF-1 domain. Therefore, it appears that sumoylation-mediated transcriptional repression occurs mainly (if not exclusively) through the modification of Lys residue(s) in the AF-1 domains of the nuclear receptors.
We have previously reported that GFP-SUSP1 localizes predominantly in the cytoplasm (34). However, lines of evidence provided in this study indicate that SUSP1 is a nuclear protein. First, upon yeast two-hybrid screening, SUSP1 was found to interact with RXR␣, which is known to reside exclusively in the nucleus. Second, endogenous SUSP1 could be co-immunoprecipitated with RXR␣. Third, SUSP-V5 could also be co-immunoprecipitated with FLAG-RXR␣. Fourth, ectopically expressed SUSP-V5 or SUSP1-GFP, but not GFP-SUSP1, co-localized with FLAG-RXR␣ in the nucleus. Finally, endogenous SUSP1 was found to locate exclusively in the nucleus upon staining with anti-SUSP1 antibody. These results indicate that GFP tagged to the N terminus of SUSP1, but not to its C terminus, somehow interferes with the translocation of the protease into the nucleus. Therefore, we wondered whether any nuclear localization signal sequence might be located immediately adjacent to the N terminus of SUSP1. However, three putative nuclear localization signal sequences were found in the sequences 188 -205 (KKTEESESQVEPEIKRK), 419 -422 (KRRK), and 1086 -1090 (KRKHK), all of which are quite remote from the N terminus of SUSP1. Although it remains unclear why GFP-SUSP1 cannot be translocated into the nucleus, unlike SUSP1-V5 or SUSP1-GFP, we should have more carefully analyzed the localization of SUSP1 in our previous report by alternating the position of the tags.
An increasing number of sumoylated transcription factors and co-regulators have been identified. In most cases, SUMO modification appears to repress the activity of targeted transcription factors through altering their subcellular localization and/or their interaction with co-repressors. For examples, sumoylation reduces the transcriptional activity of Sp3 by translocating it to PML nuclear bodies (23). Sumoylation of Elk-1 not only regulates the nucleocytoplasmic shuttling of the transcription factor but also recruits histone deacetylase-2 (HDAC-2) to Elk-1-regulated promoters, thereby repressing their transcription (44,45). Sumoylation of transcriptional coactivator also recruits HADC-6, leading to repression of p300mediated transcriptional activity (24). Replacement of the Smad4 sumoylation site Lys-159 by Arg blocks Smad4-Daxx interaction and relieves Daxx-mediated repression of Smad4 transcriptional activity (46). However, SUMO modification of RXR␣ showed little or no effect on its subcellular localization or interaction with co-regulators tested thus far, such as HDACs, Daxx, and SRC-1 (data not shown). Thus, the mechanism for sumoylation-mediated control of RXR␣ activity remains to be investigated.
In addition to SUSP1, several SUMO-specific proteases have been shown to reverse the sumoylation-mediated transcriptional repression of nuclear factors (5). SuPr-1, a spliced form of SENP2, was shown to induce c-Jun-dependent transcription independently of c-Jun phosphorylation (39). The mechanism underlying SuPr-1 action on c-Jun activity is through SuPr-1 binding of SUMO-modified PML, thereby altering the distribution of PML in nuclear bodies and nuclear body-associated proteins. SENP1 was also shown to enhance c-Jun-dependent transcription, independently, however, of the sumoylation and phosphorylation status of c-Jun (47). SENP1 action on c-Jun activity is through desumoylation of the CRD1 domain of p300, thereby releasing the cis-repression of CRD1 of p300. It has been demonstrated that SENP1 also enhances AR-dependent transcription (48). This stimulatory effect of SENP1 is not mediated by desumoylation of AR but rather through its ability to remove SUMO from HDAC-1, thereby reversing the repressive function of the deacetylase on AR-dependent transcription. Thus, it appears that SuPr-1 and SENP1 act on co-regulators, but not directly on nuclear factors, for the control of transcriptional activity.
Unlike these SUMO-specific proteases, SUSP1 appears to directly act on RXR␣ for the control of transcriptional activity. Increased expression of SUSP1 resulted in the enhancement of RXR␣ activity up to the extent seen with the mutant protease (i.e. SUSP1/C1030S), which lacks desumoylating activity (Fig.  7A). Thus, it is likely that SUSP1-dependent control of RXR␣ transcriptional activity is mediated by direct desumoylation of the nuclear receptor other than its co-regulators, although the possibility that SUSP1 might also desumoylate the transcriptional co-regulators of RXR␣ cannot be excluded. Taken together, we suggest that reversible SUMO modification RXR␣ is a potential mechanism for the control of RXR␣ transcriptional activity and that SUSP1 is involved in the positive control of the nuclear receptor function.