Downstream Codons in the Retinoic Acid Receptor β-2 and β-4 mRNAs Initiate Translation of a Protein Isoform That Disrupts Retinoid-activated Transcription*

Retinoic acid receptors (RARs) are essential for the differentiation and maintenance of normal epithelium. In studies of RARs in breast cancer, there are striking differences in the expression of certain protein isoforms of the RARβ gene between cells derived from normal human mammary glands and those derived from breast tumors. While the protein isoforms RARβ2 and RARβ4 consist of the longest open reading frames of the RARβ2 and RARβ4 mRNAs, respectively, we find that a fraction of scanning ribosomes bypass these upstream RARβ2 and RARβ4 protein start codons and initiate translation downstream. This downstream translation initiation site is identical in the RARβ2 and RARβ4 transcripts and generates a third RARβ protein isoform, here termed RARβ′ (formerly human RARβ4). RARβ′ lacks protein domains found in the N terminus of RARβ2 and RARβ4, including one of two zinc fingers required for DNA binding. However, RARβ′ retains the ability to heterodimerize with RXRα and interact with transcription cofactors. In reporter gene assays, RARβ′ repressed retinoic acid-activated transcription of co-transfected RARβ2, RARβ4, and RARα. This repression required the presence of acidic amino acids within the AF2 domain. These findings demonstrate an antagonistic role for RARβ′ in signaling by retinoic acid.

Vitamin A derivatives, called retinoids, are required for epithelial cell development and differentiation. Retinoid signals are transduced largely through ligand activation of two families of retinoid-activated transcription factors belonging to the nuclear receptor superfamily, retinoic acid receptors (RARs) 1 and retinoid X receptors (RXRs). Receptors of both the RAR and RXR subfamilies share conserved protein regions designated A through F. The A and B regions include a ligand-independent transactivation domain (AF-1) whereas region C encompasses the DNA-binding domain. The D region acts as a conformational hinge between the DNA-binding domain and the remain-der of the protein. The ligand binding, dimerization, and liganddependent transactivation (AF-2) domains are all found within regions E and F. RAR-RXR heterodimers bind to cis-acting DNA elements in gene promoter regions called retinoic acid response elements (RAREs) and transactivate target genes in the presence of a retinoid ligand. Both the RAR and RXR families are comprised of three genes (␣, ␤, and ␥) that generate multiple isoforms via the usage of two promoters, P1 and P2, and alternative splicing (1).
As the primary effectors of retinoid signaling, the RARs and RXRs themselves appear to be targets for disruption in tumorigenesis (2), including loss of heterozygosity (3), gene rearrangements (4,5), mutations (6), and aberrant mRNA expression (7,8). Retinoic acid receptor ␤ (RAR␤) in particular has been extensively studied in human carcinomas, and a body of evidence indicates that it may play a role in tumor suppression. Such findings include the loss of heterozygosity of its genetic locus (3p24) in primary tumors of the breast (3) and the loss of RAR␤ mRNA expression in primary breast (9,10), lung (8,11), and esophageal carcinomas (12). Loss of RAR␤ transcript expression has been linked to epigenetic silencing by methylation of the RAR␤ P2 promoter in primary breast (13)(14)(15)(16)(17) and lung tumors (18). Further evidence of RAR␤ suppression of tumor cell proliferation comes from cell culture studies, where ectopic expression of the RAR␤2 isoform resulted in decreased proliferation of tumor cells lines derived from the breast (19,20), lung (21,22), cervix (23), pancreas (24), and squamous cell carcinomas (25).
In the mouse, the RAR␤ gene generates four distinct transcripts: splice variants RAR␤1 and RAR␤3 from transcription at promoter P1, and RAR␤2 and RAR␤4 from the RARE-containing P2 promoter (26,27). In the human, only RAR␤2 and RAR␤4 transcripts have been identified in normal adult cells (28). Human RAR␤1 is expressed in fetal tissues and some small cell lung carcinoma cell lines (29), whereas a human homologue of the RAR␤3 isoform has not been detected. The RAR␤2 and RAR␤4 transcripts differ only in the content of their 5Ј-most exon (exon 5 relative to the entire RAR␤ gene), a result of alternative splicing (26). RAR␤4 lacks 357 nucleosides encoded by this exon that are present in the RAR␤2 mRNA, including the translation initiation site for the RAR␤2 protein.
We have previously reported several unique findings regarding RAR␤ expression in cultured cells derived from normal and neoplastic mammary epithelium. Only two mRNA products of the RAR␤ gene (RAR␤2 and RAR␤4) had been detected in these cells by Northern blot analyses (30), and these transcripts were expressed at low levels in most breast tumor cell lines compared with human mammary epithelial cells (HMECs) isolated from normal breast tissue. However, the levels of RAR␤ protein in these cells could not be predicted on the basis of their transcript levels. Some cell lines with abundant RAR␤2 and RAR␤4 mRNAs expressed low levels of RAR␤ protein. Conversely, other cells, such as the breast carcinoma cell line MCF-7, expressed high levels of RAR␤ protein with very low transcript levels by Northern blot and RT-PCR analyses. We also noted the presence of three distinct RAR␤ protein species in these cell lines, although only two RAR␤ transcripts were present. Finally, we observed differential expression of these RAR␤ protein isoforms in cells derived from normal breast tissue compared with those from neoplastic mammary tissue (31).
In the present study, we introduced mutations into RAR␤2 and RAR␤4 5Ј transcript leader sequences to determine which candidate codons operate to initiate protein synthesis within the cell. We performed a functional characterization of the protein generated from a downstream initiation codon of both the RAR␤2 and RAR␤4 transcripts, RAR␤Ј, to determine its ability to bind a RARE, heterodimerize, and interact with transcription cofactors. We then tested the effect of this protein on retinoid-activated transcription in transient transfection assays. Our studies show distinct and opposing functions for RAR␤2 and RAR␤4 compared with RAR␤Ј, and we propose that RAR␤Ј acts as an inhibitor of transcriptional activation mediated by RARs.

EXPERIMENTAL PROCEDURES
Plasmid Construction-To synthesize expression constructs of RAR␤2 or RAR␤4 transcripts including all nucleosides from ϩ1 throughout the coding region, the promoter construct p␤2-747luc (32) was PCR-amplified between ϩ1 and ϩ155 (HindIII/RAR␤ forward primer, 5Ј-GAT AAG CTT GTG ACA GAA GTA GTA GGA AGT-3Ј, HindIII site in bold; reverse primer, 5Ј-AGT GGA TCC TAC CCC GAC GGT GCC CAG AC-3Ј). Following digestion with HindIII and BamHI, the amplicon was ligated into the HindIII and BamHI sites of the pBluescript II vector (Stratagene, La Jolla, CA), and the resulting plasmid was linearized with BamHI digestion. RAR␤2 and RAR␤4 cDNA fragments from ϩ155 through ϩ2107 or ϩ155 through ϩ1750, respectively, were obtained by the digestion of pCR(RAR␤2) and pCR(RAR␤4) (31) with BamHI. These RAR␤ fragments were ligated into the linearized plasmid described above to obtain plasmids pB-S(RAR␤2) and pBS(RAR␤4).
For translation studies, RAR␤ cDNAs were cloned into the first polylinker of vector pCS2ϩMT (33) to achieve read-through translation of 6 Myc epitopes at the protein C terminus. pCS2ϩMT was first modified to contain a SmaI site by the insertion of a double-stranded oligonucleotide (top, 5Ј-GAT CCA GAT CTA CTG CAG ATC CCG GGT CAT-3Ј; bottom, 5Ј-CGA TGA CCC GGG ATC TGC AGT AGA TCT G-3Ј; SmaI sites in bold) between the vector BamHI and ClaI sites. This modified vector was then digested with HindIII and SmaI for the following cloning steps. Plasmids pCS(RAR␤2) and pCS(RAR␤4), which contain nucleotides corresponding to ϩ1 of the transcript through the entire RAR␤2 or RAR␤4 coding sequence (excluding the stop codon) were created by PCR amplification of pBS(RAR␤2) or pBS(RAR␤4) with primers containing either a HindIII site (HindIII/RAR␤ forward primer) or an XbaI site (reverse primer, 5Ј-CCCTCTAGATTGCAC-GAGTGGTGACTG-3Ј; XbaI site in bold). All PCR amplifications were performed using the Expand High Fidelity PCR System (Roche Molecular Biochemicals) and cycled using parameters detailed previously (31). Amplicons were digested with XbaI followed by a Klenow fill-in reaction and then subsequently digested with HindIII. The resulting fragments were ligated into the digested pCS2ϩMT vector.
A series of constructs derived from plasmids pCS(RAR␤2) and pCS-(RAR␤4) were created with point mutations at putative translation initiation sites to either eliminate or to enhance ribosomal recognition as translation initiation sites. Site-directed mutagenesis was performed using the QuikChange kit (Stratagene, La Jolla, CA) according to directions. Site-directed mutagenesis plasmids and their targeted mutations are detailed in Table I. For mammalian expression, protein-coding sequences for the RAR␤ isoforms were cloned into the pTag vector (31) and included stop codons following the coding sequence so that the C-terminal (His) 6 and Myc tags were not translated. A fragment containing the RAR␤2 coding sequence was amplified from plasmid pCS(RAR␤2 T2Ϫ) (see Table I) using forward primer BamHI-RAR␤2 (31) and reverse primer 3ЈBamHI RP (5Ј-TTGGGGATCCTTATTGCACGAGTGGTGAC-3Ј; BamHI site in bold). Following BamHI digestion, the insert was ligated into the BamHI site of pTag to create pTag(RAR␤2/stop). Forward primer BamHI-mut-RAR␤4 259 (31) and reverse primer 3ЈNotI RP (5Ј-TT-GGGCGGCCGCTTATTGCACGAGTGGTGAC-3Ј, NotI site in bold) were used to amplify the RAR␤4 coding sequence from pCS(RAR␤4 T2Ϫ). RAR␤Ј protein coding sequence was amplified from pCS(RAR␤2) with forward primer (5Ј-CCTGGATCCCAGAAGAATATGGTTTA-CACTTGTC-3Ј, BamHI site in bold) and reverse primer 3ЈNotI RP. Following digestion with BamHI and NotI, agarose gel-purified RAR␤4 and RAR␤Ј coding sequences were ligated into the BamHI and NotI sites of pTag, generating plasmids pTag(RAR␤4/stop) and pTag(RAR␤Ј/ stop), respectively. pTag(RAR␤2⌬AF2) was constructed by PCR amplification of pSP(RAR␤2⌬AF2) (see below) with primers BamHI-RAR␤2 and 3ЈBamHI RP, followed by BamHI digestion and ligation into the BamHI site of pTag. pTag(RAR␤Ј⌬AF2) was created by PCR amplification of pSP(RAR␤Ј⌬AF2) (below) using primers PstI-RAR␤Ј FP (5Ј-CCACTGCAGAAGAATATGATTTACACTTGTCACC-3Ј, PstI site in bold)and 3ЈNotI RP, followed by PstI and NotI digestion and ligation into the PstI and NotI sites of pTag.
For translation in vitro of a C-terminal-truncated RAR␤2 isoform, plasmid pSP(RAR␤2) was PCR-amplified with forward primer BamHI-RAR␤2 (31) and the following reverse primer with an added SacI site (in bold): 5Ј-GAGCTCTTAATTCTTCTGAATACTTCTGCGG-3Ј. The amplicon was gel purified and ligated into the BamHI and SacI sites of pSP64 (Promega, Madison, WI) to create plasmid pSP(RAR␤2N). Three mutations were introduced into the coding sequence of both RAR␤2 and RAR␤Ј by site-directed mutagenesis of plasmids pSP(RAR␤2) and pS-P(RAR␤4 448) (31) to replace codons for three glutamic acid residues in the AF2 domain with those encoding neutral amino acids alanine and valine (Fig. 8A). Site-directed mutagenesis was performed using the QuikChange kit as described by the manufacturer and resulted in plasmids pSP(RAR␤2⌬AF2) and pSP(RAR␤Ј⌬AF2). For pSP(RXR␣), pL(RXR␣SN) (34), a gift from Steven Collins, was amplified (forward primer, 5Ј-GAATTCGTCGCAGACATGGACACCAAACATTTCC-3Ј and reverse primer, 5Ј-TCTAGACCCGCAGGCCTAAGTCATTTGGTG-CGG-3Ј; EcoRI and XbaI digestion sites in bold, respectively), followed by EcoRI digestion, Klenow fill-in, XbaI digestion, and ligation into the HindIII (blunted by Klenow) and XbaI sites of pSP64.
Double-stranded, synthetic oligonucleotides corresponding to nucleotides Ϫ83 to Ϫ37 of the RAR␤2 promoter (5Ј-ggg tca TTT GAa ggt taG CAG CCC GGG TAG GGT TCA CCG AAA GTT CA-3Ј; RARE in bold, non-consensus RARE in lowercase) were ligated into a BglII-and NheIdigested pGL-Promoter (Promega) to give pGL(␤RARE), a retinoic acidresponsive reporter construct for use in the luciferase assays. All plasmid constructs described above were verified by automated sequence analysis.
Cell Culture and Transfection-The AG11132 human mammary epithelial cell (HMEC) strain was obtained from the Coriell Institute (National Institutes of Aging Repository, Camden, NJ). MCF-7 human breast cancer cells were obtained from the American Type Culture Collection (ATCC, Manassas, VA). AG11132 and MCF-7 cells were cultured as previously described (37).
For expression of Myc-tagged proteins, 1.0 ϫ 10 6 AG11132 or MCF-7 cells were plated in 60-mm 2 dishes. After 24 h, cells were transfected with 10 g of one of the above pCS plasmids using 20 l of SuperFect transfection reagent (Qiagen, Valencia, CA) following the manufacturer's protocol. Forty-eight hours post-transfection, cells were scraped in lysis buffer (150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 10 mM Tris-HCl, pH 7.4, 5 mM dithiothreitol, 0.1 mM phenylmethylsulfonyl fluoride). Following incubation on ice for 10 min, the cells were centrifuged at 13,000 ϫ g for 10 min and supernatants were collected and stored at Ϫ80°C for Western blot analysis. For luciferase assays, MCF-7 cells were plated in 24-well plates at 70,000 or 100,000 cells/ well. The cells were transfected with a total of 0.24 -0.27 g DNA per well using Effectene transfection reagent (Qiagen) in medium containing 5% charcoal-dextran-treated fetal calf serum (Hyclone, Logan, UT). The amount of DNA per transfection was equalized using the pTag vector. The transfection control plasmid pRL (Renilla luciferase) was co-transfected at one-tenth the amount of pGL(␤RARE). Twenty-four hours following transfection, 0.1% ethanol or all-trans retinoic acid (atRA) (to final concentrations of 0.01 or 1 M) was added. After another 24 h, the cells were lysed in 1ϫ Passive Lysis Buffer (Promega) and immediately assayed for luciferase activity or stored at Ϫ20°C.
Western Blot Analysis-12 l of each cell lysate from MCF-7 and AG11132 cells transfected with pCS-derived plasmids (above) were separated on 12% SDS-PAGE gels and transferred to polyvinylidene difluoride membrane as described previously (31). Membranes were incubated in 1% milk in phosphate-buffered saline with 0.1% Tween-20 (PBS-T) for 1 h at room temperature and then incubated with a monoclonal anti-Myc antibody (1:200 in PBS-T) produced as a medium supernatant from the Myc1-9E10.2 hybridoma line (38) for 1 h at room temperature. After extensive washing in PBS-T, membranes were incubated 1 h at room temperature with anti-mouse IgG-horseradish peroxidase secondary antibody (1:20,000 in PBS-T; Pierce, Rockford, IL), re-washed in PBS-T, then incubated 5 min at room temperature in SuperSignal West Pico chemiluminescent substrate (Pierce), followed by autoradiography.
Luciferase Assay-Firefly and Renilla luciferase activities were measured on a Zylux FB15 luminometer (Zylux Corp., Maryville, TN) using the Dual-Luciferase Reporter Assay System (Promega). For each experiment, duplicate transfections were assayed. Firefly relative light units were divided by Renilla relative light units and normalized to vector control (without addition of atRA). The average of the 2-3 experiments was calculated, and statistical significance was determined using the two-tailed Student's t test for samples with equal variance. Statistical analyses compared the relative luciferase activity obtained for samples with an added component, such as a cofactor, with those obtained for identical samples without the added component.

The RAR␤2 and RAR␤4 Transcripts Each Generate Two
Protein Isoforms-We have previously reported the expression of three discrete RAR␤ protein isoforms in HMECs and breast tumor cells that express only two RAR␤ mRNAs (RAR␤2 and RAR␤4) ( Fig. 1) (31). We hypothesized that at least one of the RAR␤ transcripts directs the synthesis of multiple proteins by the mechanism of leaky scanning (39,40). Expression plasmids pCS(RAR␤2) and pCS(RAR␤4) were created containing the entire transcript leader region and coding sequence of each transcript cloned in-frame to six C-terminal Myc epitopes ( Fig.  2A). These plasmids were transiently transfected into both HMECs (AG11132) and MCF-7 cells, and soluble protein extracts were analyzed by Western blots using an antibody raised against the Myc epitope. We found that the pCS(RAR␤2) and pCS(RAR␤4) transcripts each generated two proteins in both cell lines assayed (Fig. 2B). The apparent masses of these proteins were comparable to the endogenous RAR␤ proteins when the ϳ14-kDa tag at the C terminus of the ectopically expressed proteins is accounted for. Note that in vitro translated proteins generated from each of these putative translation start sites migrated in SDS-PAGE with one of the three endogenous RAR␤ proteins found in HMECs or MCF-7 cells (31). The lowest molecular weight protein appeared to be common to both transcripts, whereas the higher molecular weight protein made by pCS(RAR␤4) and pCS(RAR␤2) transcripts were dissimilar in size (Fig. 2B). For the RAR␤4-derived transcript, the lower molecular weight protein was expressed at higher levels than that of the larger protein in MCF-7 cells, and at approximately equivalent amounts in the HMEC lysates ( Fig. 2, B and C). The higher molecular weight protein was the predominant protein generated from the RAR␤2-derived transcript in both MCF-7 cells and HMECs (Fig. 2, B and C). In identical experiments, these patterns of protein expression directed by transiently transfected pCS(RAR␤2) and pCS-(RAR␤4) plasmids into MCF-7 cells were replicated in two other breast tumor cell lines, MDA-MB-231 and ZR-75-1 (data not shown).
To identify the translation start sites utilized by each transcript, we introduced single mutations into pCS(RAR␤2) and pCS(RAR␤4) to abolish ribosome recognition of putative translation start sites (Table I and Fig. 2A). Mutations were targeted to eradicate translation of either the putative 5Ј-most translation initiation codon (termed T1 for the first translation initiation site) or the putative downstream initiation codon (T2) in both the RAR␤2-and RAR␤4-derived transcripts. In both HMECs and MCF-7 cells, mutation of T1, a previously identified translation initiation codon at nucleoside ϩ469 of the RAR␤2 transcript, from AUG to AUA altered the ratio of expression of the two proteins. The proportion of the larger protein decreased, whereas that of the smaller protein increased (Fig. 2C). Conversely, mutation of an AUG at ϩ805 of the RAR␤2 transcript (T2) to AUU resulted in a loss of expression of the lower molecular weight pCS(RAR␤2) protein, whereas expression of the higher molecular weight protein was maintained (Fig. 2C). T2 of the RAR␤2 mRNA is found within sequence that is identical to a translation start site previously identified in the human RAR␤4 transcript (31).
In a similar fashion, mutation of a CUG to a CUU at ϩ259 (T1) of the RAR␤4 transcript, a CUG analogous to one that initiates translation of the mouse RAR␤4 (26), eradicated the expression of the larger pCS(RAR␤4) protein in both HMECs and MCF-7 cells (Fig. 2C). It is interesting to note that mutation of the AUG translation start site to a codon (AUA) that should not initiate translation decreased, but did not completely abolish, the presence of the larger protein. Mutation of the previously identified translation start site at ϩ448 relative to the RAR␤4 transcript (T2; equivalent to the AUG located at ϩ805 in the RAR␤2 transcript) to AUU abolished the expression of the lower molecular weight protein in both cell lines tested. These results clearly indicated that the RAR␤2 and RAR␤4 transcripts each produce two RAR␤ proteins. RAR␤2 initiated translation at AUGs located at ϩ469 (T1) and ϩ805 (T2) of the RAR␤2 transcript, whereas RAR␤4 used a CUG at ϩ259 (T1) and an AUG at ϩ448 (T2) of the RAR␤4 transcript for translation initiation (Fig. 2C). The lower molecular weight protein generated at T2 in both RAR␤2-and RAR␤4-derived transcripts, previously called human RAR␤4, is identical on the basis of the coding sequence downstream of T2 in both transcripts; we therefore re-designated this protein as RAR␤Ј. We refer to the protein product of translation initiation at the CUG codon at T1 of the RAR␤4 transcript as RAR␤4, the analogous isoform to the previously reported mouse RAR␤4 protein (26). RAR␤Ј was the primary protein product of the RAR␤4-derived mRNA in the MCF-7 breast cancer cell line, whereas RAR␤4 and RAR␤Ј were generated at approximately equivalent amounts in HMECs. In contrast, RAR␤Ј was produced at significantly lower levels relative to the RAR␤2 protein in both cell types when the RAR␤2-derived transcript was expressed (Fig. 2C).
The CUG T1 translation initiation site of RAR␤4 does not conform to the Kozak consensus sequence predicting robust recognition by the scanning ribosome (GCC(A/G)CCAUGG, translation initiation codon in bold; Ref. 41). The most important bases for ribosome recognition in this sequence are the initiation codon and the nucleosides in the Ϫ3 and ϩ4 positions relative to the start of translation. Since the RAR␤4 start site is not an ideal initiation codon, we hypothesized that two proteins were generated from the human RAR␤4 transcript by a mechanism known as leaky scanning. To determine whether translation of RAR␤Ј at T2 in the RAR␤4 mRNA was a result of leaky scanning past the non-canonical CUG initiation codon at T1 (42), we performed site-directed mutagenesis to create RAR␤4 translation initiation sites with a greater identity to the Kozak consensus sequence (Table I; Fig. 3A). As shown in Fig.  3B, mutation of T1 of the RAR␤4-derived transcript from a CUG to an AUG resulted in an increased usage of the 5Ј-most translation initiation site along with a decrease in RAR␤Ј translation, consistent with a leaky scanning model. This effect was observed identically in MCF-7 cells and HMECs.
Although the Kozak sequence of the T1 translation initiation site of RAR␤2 should be adequate for translation initiation (Table I), we created two mutants of the pCS(RAR␤2) construct to determine if improving the Kozak sequence at T1 would abolish translation at T2. The first mutation (T1ϩ) exchanged a guanosine for the uracil at ϩ4 relative to the start of translation, a key nucleoside for translation initiation. The second construct (T1ϩϩ) contained mutations altering the entire context of the T1 site to conform exactly to the Kozak sequence (Table I). As shown in Fig. 3B, mutation of the ϩ4 nucleoside from uracil to guanosine (T1ϩ) did not alter the expression of the RAR␤Ј isoform at T2 in either HMECs or MCF-7 cells. However, the RAR␤2-derived transcript having the T1 initiation site in a perfect Kozak context (T1ϩϩ) displayed no trans-FIG. 2. The RAR␤2 and RAR␤4 transcripts can each generate two proteins by the utilization of discrete translation initiation sites. A, diagram of RAR␤2 and RAR␤4 transcripts produced from the pCS vectors used in panels B and C. Each transcript includes coding sequence for six Myc tags that are in-frame with the RAR␤ coding sequence at the C terminus, represented by stippled boxes. In some constructs, single point mutations were introduced into putative translation initiation sites in RAR␤2 or RAR␤4 (X over T1 or T2). Minus signs designate mutations that were intended to eliminate ribosome recognition at either T1 or T2. B, anti-Myc immunoblots of HMEC or MCF-7 whole cell lysates transfected with pCS vectors expressing wild type RAR␤2 or RAR␤4 transcripts, or C, RAR␤2 and RAR␤4 transcripts containing point mutations to abolish ribosomal recognition of either T1 or T2. The plasmids transfected are identified above each sample. Migration of molecular weight markers is indicated to the left of each gel. lation product of the downstream T2 site in both cell lines utilized (Fig. 3C). These results provide evidence that, like the RAR␤4 transcript, some ribosomes scan past the T1 of RAR␤2 and initiate translation downstream at T2.
RAR␤Ј Does Not Bind to a RARE-Because RAR␤Ј lacks one of the two zinc-finger motifs characteristic of full-length RAR DNA-binding domains, we hypothesized that RAR␤Ј would be unable to bind a RARE. We performed electrophoretic mobility shift assays (EMSAs) with non-radioactive, in vitro translated RAR␤2, RAR␤4, RAR␤Ј, and RXR␣ proteins. Since the presence of retinoic acid in nuclear extracts from HMECs and breast cancer cell lines had no effect on the shifting pattern or intensity of shifted bands (data not shown), the following EM-SAs were performed in the absence of retinoic acid. In combination with RXR␣, both RAR␤2 and RAR␤4, which have fulllength DNA-binding domains, bound a DR-5 RARE (␤RARE) found within the P2 promoter of the RAR␤ gene (Fig. 4A, lanes  5 and 8, respectively). Visible supershifts in the presence of RAR␤ and RXR␣ antibodies confirmed the binding of RAR␤2: RXR␣ and RAR␤4:RXR␣ heterodimers to the ␤RARE (Fig. 4A,  lanes 6 and 7 and lanes 9 and 10, respectively). Unlike RAR␤2 and RAR␤4, the RAR␤Ј protein did not result in any shifted bands when incubated with RXR␣ (Fig. 4A, lane 11). We therefore concluded that RAR␤Ј does not bind the ␤RARE element either as a homodimer or heterodimer with RXR␣.
RAR␤2, RAR␤4, and RAR␤Ј Interact with Nuclear Receptor Cofactors-Although RAR␤Ј lacks a full-length, functional DNA-binding domain, it retains domains known to be involved in heterodimer and transcription cofactor binding. We verified the ability of this shortened isoform to interact with cofactors by GST pull-down assays. We incubated [ 35 S]methionine-labeled in vitro translated RAR␤2, RAR␤4, and RAR␤Ј proteins with GST fusion proteins of full-length RXR␣ and C-terminal regions of AIB1 co-activator and SMRT co-repressor. These regions of the cofactors have been previously characterized to contain domains required for interaction with nuclear receptors (35,36). As shown in Fig. 5, GST alone did not pull down any of the RAR␤ isoforms tested; conversely, none of the GST fusion proteins were able to pull down RAR␤2N, a negative control containing only the N-terminal 112 amino acids of RAR␤2 (includes domains A, B, and the N-terminal half of domain C). GST-RXR␣ showed an equal ability to pull down RAR␤2, RAR␤4, and RAR␤Ј in the absence or presence of atRA (Fig. 5, A-C). The amount of RAR␤2, RAR␤4, and RAR␤Ј pulled down by the C terminus of AIB1 was enhanced with the addition of atRA. All three isoforms were eluted at equivalent levels from glutathione beads conjugated with the C terminus of the SMRT in the absence but not in the presence of atRA (Fig. 5, A-C). Since RAR␤Ј interacts with RXR␣, AIB1, and SMRT, it could potentially compete with RAR␤2 for limiting amounts of their heterodimeric partners and/or transcriptional cofactors. RAR␤Ј Antagonizes Transcriptional Activation by RAR␤2, RAR␤4, and RAR␣-To test the ability of the various RAR␤ isoforms to activate transcription, a fragment of the RAR␤ P2 promoter from position Ϫ83 to Ϫ37 (relative to the start of transcription) was linked upstream of a SV40 promoter and a firefly luciferase reporter gene. The portion of the RAR␤ promoter included contains recognition elements of a non-consensus RARE and the DR-5 ␤RARE, previously shown to bind RAR-RXR heterodimers in vitro (43,44) and mediate RA-activated transcription in vivo (45,46). Normal HMECs and ZR-75-1 and MCF-7 breast cancer cells were transiently transfected with the reporter construct along with varying amounts of RAR␤2, RAR␤4, and RAR␤Ј expression plasmids. The T2 initiation codons of the RAR␤2 and RAR␤4 expression constructs were mutated (AUG to AUU) to prevent translation of FIG. 3. Improving the Kozak context of the RAR␤2 and RAR␤4 T1 abolishes translation initiation downstream at T2. A, diagram of RAR␤2and RAR␤4-derived transcripts generated from pCS vectors harboring point mutations to enhance ribosome recognition of the putative 5Ј-most (T1) translation initiation sites. The 5Ј transcript leader sequence is represented by a line and protein coding sequence by a box. The positions of the T1 and T2 sites found on each transcript are indicated by arrows. The six Myc tags in-frame with RAR␤ coding sequence at the C terminus is represented by stippled boxes. Plus signs connote constructs having mutations at the T1 site to increase the identity of the surrounding nucleosides to that of the optimal Kozak sequence. Two plus signs designate an RAR␤2 T1 mutation that is exactly homologous to the optimal Kozak sequence. The size of the arrowhead, T1 or T2, indicates relative optimal Kozak consensus sequence. B and C, anti-Myc immunoblots derived from whole cell lysates of AG11132 HMEC and MCF-7 breast carcinoma cells transfected with the above pCS plasmids. The plasmid transfected is identified above each sample. The migration of molecular weight markers is indicated to the left of each gel.
RAR␤Ј from these plasmids. Additionally, the CUG translation initiation codon of RAR␤4 T1 was altered to an AUG to ensure high levels of translation. A Renilla luciferase reporter construct was co-transfected as a measure of transfection efficiency, and total amounts of DNA transfected per sample were equalized using the empty expression vector. The following experiments were performed using two different concentrations of atRA for induction: a pharmacological dosage of 1 M and one approximating physiological levels at 0.01 M. Interestingly, both dosages of atRA produced equivalent results, including degree of atRA induction (2-3-fold), for all samples transfected with RAR␤2 or RAR␤4 (Fig. 6). In the following experiments, we report results of transfections in the MCF-7 cell line, although equivalent results were obtained in the other two cell lines tested. We also induced reporter gene expression using 1 M of atRA.
When RAR␤2, RAR␤4, and RAR␤Ј expression vectors were transfected individually, cells with the highest amounts (10ϫ) of RAR␤2 and RAR␤4 did not display significantly greater levels of reporter gene activity than those with the lowest amount, either in the absence or presence of atRA (Fig. 6). However, the reporter gene activity of the cells transfected with the highest amount of RAR␤Ј was slightly lower than that with the lowest amount of RAR␤Ј with atRA (Fig. 6, p Ͻ 0.05). RAR␤2-and RAR␤4-transfected cells showed a 1.2-1.5-fold elevated level of reporter gene activity compared with in the absence of atRA ( Fig. 6; p Ͻ 0.05). With the addition of 1 M atRA, RAR␤2-and RAR␤4-transfected cells displayed a 2-fold greater reporter gene activity than vector-transfected cells (p Ͻ 0.01). There was no significant difference between RAR␤2-and RAR␤4-transfected cells in the amount of reporter gene activity, with or without atRA addition. AtRA-treated cells transfected with RAR␤Ј, however, showed similar levels of reporter gene activity as those transfected with the empty expression vector (Fig. 6).
Co-transfection of increasing amounts of RAR␤Ј with either RAR␤2 or RAR␤4 resulted in a decrease of reporter gene activity, with the highest amounts of RAR␤Ј (10ϫ) diminishing the expression of the reporter to the level of the empty vector (Fig. 7, A and B; p Ͻ 0.05 and p Ͻ 0.01, respectively, in the presence of atRA). The inhibitory effect of RAR␤Ј on reporter gene expression was opposite to what was observed with cotransfection of RAR␤4. Increasing the amount of RAR␤4 cotransfected with constant amounts of RAR␤2 resulted in an increase in reporter gene activity (Fig. 7A, p Ͻ 0.01 with the highest amounts of RAR␤4). Because the AUG to AUA at T2 created a conserved methionine to isoleucine substitution in the C region of the full-length RAR␤2 and RAR␤4 proteins, these experimental results were replicated in assays using expression plasmids without the T2-null mutations, and the same effect was seen (data not shown). We tested the effect of RAR␤Ј overexpression on atRA activation mediated by a second RAR subfamily, RAR␣. Similar to its effect on RAR␤2-activated transcription, increasing the amount of co-transfected RAR␤Ј decreased reporter gene expression mediated by RAR␣ (Fig.  7C, p Ͻ 0.01). Conversely, co-transfection of increasing amounts of RAR␤2 or RAR␤4 plasmids with a constant amount of RAR␣ increased activity of the luciferase reporter (data not shown). These findings demonstrated that ectopic expression of the RAR␤Ј isoform in MCF-7 cells inhibited atRA-activated transcription mediated by RAR␤2, RAR␤4, and RAR␣.
The levels of reporter gene expression in the absence of exogenous retinoids were higher than expected and resulted in a low apparent induction of expression with atRA treatment (ϳ2-fold). To determine whether reporter gene expression in the absence of added atRA is due to the presence of endogenous retinoic acids and retinoid precursors in the medium or an artifact reflecting either poor transfection or improperly packaged transiently transfected promoter elements with histone proteins, we performed stable transfection of reporter genes into MCF-7 cells. Vectors stably transfected included either a promoter-less luciferase gene, or one of two vectors with reti-  5-13, panel A; lanes 5-12, panel B). Duplicate samples containing RAR-RXR mixtures were incubated with either a RAR␤ or RXR␣ antibody. The antibodies used are indicated above each sample; the RAR␤ antibody is abbreviated as ␤, and RXR␣ as X. P designates samples incubated in the absence of added proteins and antibodies. The arrowhead indicates supershifted bands. The shifted band corresponding to the RAR-RXR heterodimer is indicated by an arrow. A, wild type RAR␤ isoforms. B, comparison of ␤RARE binding between wild type and mutant RAR␤2 and RAR␤Ј proteins. noid responsive promoters: pGL[␤RARE] Luc or p␤2-747luc (consisting of the RAR␤ gene P2 promoter, from Ϫ747 to ϩ155 bp, upstream of a luciferase gene, Ref. 32). Mass cultures of cells stably transfected with either of the two vectors with luciferase gene expression driven by an atRA-inducible promoter displayed significant reporter gene expression without atRA addition and increased reporter gene expression between 2.6-and 6.6-fold after exposure to 1 M atRA for 24 h, similar to what is observed in transient transfection assays. Cells transfected with the control vector exhibited a slight decline in luciferase expression with atRA induction (data not shown).
Negatively Charged Amino Acids within the AF2 Domain Are Required for RAR␤Ј Inhibition of atRA-activated Transcription-Evidence that some transcription co-activators are present at limiting concentrations within the cell (47)(48)(49) led us to hypothesize that RAR␤Ј represses transcription by competing with full-length RARs for essential cofactors for transcriptional activation. For RARs, protein interactions with transcription co-activators are dependent upon negatively charged amino acids within the AF2 domain (50). We performed site-directed mutagenesis of the expression plasmids pTag(RAR␤2) and pTag(RAR␤Ј) to alter codons for the three glutamic acid residues within the AF2 domain to encode the neutral amino acids valine and alanine (see Fig. 8A). These three amino acid substitutions had no effect on RAR␤2 binding as a heterodimer with RXR␣ to a RARE in vitro (Fig. 4B, lanes 6 -8) when compared with EMSAs with the wild type protein (Fig. 4B, lane

5)
. RAR␤Ј proteins with AF2 domain mutations (RAR␤Ј⌬AF2) also behaved similarly to the wild type RAR␤Ј in EMSAs, remaining unable to bind the RARE in vitro, with or without RXR␣ (Fig. 4B, lanes 4 and 10). GST pull-down assays were performed to examine the effect of the AF2 mutations on heterodimer, co-repressor, and co-activator interactions. As expected, binding of RAR␤2⌬AF2 and RAR␤Ј⌬AF2 proteins to GST-RXR␣ and GST-SMRT was equivalent between the wild type proteins in the presence and absence of atRA (compare 10-fold excess of the RAR␤Ј plasmid. C, RAR␣ (a) expression construct was co-transfected with a 1-, 5-, or 10-fold excess of RAR␤Ј plasmid. In all panels, statistics were performed comparing samples co-transfected with RAR␤Ј to equivalent samples without RAR␤Ј addition. *, p Ͻ 0.05 and **, p Ͻ 0.01.

. Transcriptional inhibition by RAR␤ is relieved by mutation of negatively charged amino acids located within the AF2 domain.
A, schematic diagram of structural domains present in the RAR␤2, RAR␤4, and RAR␤Ј isoforms. The amino acid sequence of domains represented by the same markings have identical amino acid sequences between the three isoforms; however, the N-terminal portion of the AF1 and DNA-binding domains are absent from RAR␤4 and RAR␤Ј, respectively, and are represented here as shorter boxes in the diagram. Two unique amino acids found in RAR␤4 are indicated by dark shading at the N terminus of the protein. Included is the amino acid sequence of the AF2 domain (common to all three isoforms) as well as the amino acid changes generated by site-directed mutagenesis to create RAR␤2⌬AF2 and RAR␤Ј⌬AF2. B, MCF-7 cells were transiently transfected with a ␤RARE-firefly luciferase reporter plasmid, Renilla luciferase reporter, and expression plasmids of RAR␤2 (b2), RAR␤Ј (bЈ), RAR␤2⌬AF2 (b2M), RAR␤Ј⌬AF2 (bЈM), or empty vector (v) as a control. In samples 6 -8, 1-, 5-, or 10-fold excess RAR␤Ј⌬AF2 plasmid over RAR␤2 plasmid were co-transfected. Transfected cells were treated with ethanol vehicle (white bars) or 1 M atRA (shaded bars). For each sample, the value of the firefly luciferase activity was normalized to that of the Renilla luciferase transfection control. Each bar represents the average of three experiments performed in duplicate, normalized to values obtained by samples transfected with vector alone (without atRA). Error bars represent S.D. *, p Ͻ 0.05 versus RAR␤2 without RAR␤Ј⌬AF2 co-transfection and induced with 1 M atRA (sample 2). Fig. 5, panels A and E, and C and F). Unlike their wild type counterparts, RAR␤2⌬AF2 and RAR␤Ј⌬AF2 did not bind GST-AIB1 either before or after atRA addition (Fig. 5, panels E and F) confirming that these mutations indeed disrupt co-activator interactions. This conclusion was also suggested by the results of transient transfection assays. An equivalent amount of atRA-induced (␤RARE)Luc reporter gene expression was obtained following the introduction of RAR␤2⌬AF2 into MCF-7 cells as a vector only control (Fig. 8B).
To determine whether the ability to bind co-activators is a requirement for the repression of retinoic acid-induced transcription, we compared the effect of RAR␤Ј and RAR␤Ј⌬AF2 expression in transient transfection assays. As seen with RAR␤2⌬AF2, transfection of MCF-7 cells with a RAR␤Ј⌬AF2 expression vector did not affect atRA-induced reporter gene expression beyond that observed with the empty expression vector (Fig. 8B). We then co-transfected vectors expressing RAR␤2 and 1-, 5-, and 10-fold excess RAR␤Ј⌬AF2 into MCF-7 cells. Whereas increasing RAR␤Ј represses the retinoic acid induction of the (␤RARE)Luc reporter gene, increasing the amount of RAR␤Ј⌬AF2 transfected with RAR␤2 had a positive effect on the atRA-induction of luciferase activity (Fig. 8B, p Ͻ 0.05 with the highest amounts of RAR␤Ј⌬AF2 in the presence of atRA). This result demonstrated that negatively charged residues within the AF2 domain, which are involved in coactivator binding, are essential for RAR␤Ј inhibition of RAR␤2activated transcription. DISCUSSION We previously noted the presence of three RAR␤ proteins in cell lines that express only the RAR␤2 and RAR␤4 transcripts of the RAR␤ gene (31). We had hypothesized that a posttranslational modification might account for the presence of three discrete RAR␤ proteins where only two RAR␤ mRNA isoforms were expressed. However, we were unable to detect any differential phosphorylation or glycosylation of the RAR␤ proteins found in HMECs and MCF-7 cells (data not shown).
Here we show that three RAR␤ proteins are created from the two individual RAR␤ transcripts found in human breast epithelial cells. Two of these proteins, RAR␤2 and RAR␤4, are unique to their respective transcripts and represent translation initiation products from their 5Ј-most translation initiation sites (T1). We also show that the third RAR␤ protein, previously termed RAR␤4 but which we now call RAR␤Ј, is generated from both transcripts by a downstream AUG (T2). As this AUG is located 3Ј of exon 5, the only variably spliced exon in RAR␤2 and RAR␤4, this AUG is present identically in both mRNAs. It is unlikely that the RAR␤Ј isoform present in our transient transfection assays (Figs. 2 and 3) is a protease cleavage artifact, as the amount of RAR␤Ј dramatically increases in RAR␤2-and RAR␤4-derived transcripts having mutations that abrogate translation initiation upstream (Fig. 2C). Additionally, strengthening the Kozak consensus sequence of the RAR␤2 and RAR␤4 initiation codons (T1) abolishes expression of RAR␤Ј (T2) from the RAR␤2-and RAR␤4-derived transcripts (Fig. 3, B and C). These results are contradictory to what one would expect if RAR␤Ј were a cleavage product of the full-length RAR␤2 or RAR␤4.
RAR␤Ј was the primary protein product of the RAR␤4-derived transcript found in the MCF-7 breast cancer cell line in our transient transfection assays (Fig. 2, B and C), a result observed in MDA-MB-231 and ZR-75-1 breast cancer cell lines (data not shown). We observed equivalent translation initiation at the RAR␤4 mRNA T1 and T2 in HMECs. While RAR␤Ј was also generated from the RAR␤2-derived transcript, it was produced at much lower levels than that of the RAR␤2 protein (Fig. 2B). We tested whether leaky scanning played a role in the expression of the downstream translation initiation site. According to the scanning model of translation, after recognition and binding to an RNA cap, the 40 S ribosomal subunit will scan in the 3Ј direction along the transcript until it reaches an AUG that is in an appropriate sequence context to act as a signal for translation initiation. Deviations from the Kozak sequence at key positions (the initiating AUG or Ϫ3 and ϩ4 relative to the start of translation) may decrease or abolish ribosome recognition of the translation start site. In this event, a proportion of scanning ribosomes will bypass such a translation start site and continue to scan down the mRNA until reaching an AUG within an appropriate Kozak context, a phenomenon termed leaky scanning (reviewed in Ref. 39). In the case of the RAR␤4 mRNA, we found that leaky scanning is likely a consequence of the weak recognition by ribosomes of the CUG utilized at T1 as a translation start site. There is a markedly increased usage of this T1 upon mutation to AUG, and a resulting loss of translation downstream at T2 (Fig. 3B). Non-AUG codons only rarely initiate translation, and are inefficiently recognized by the scanning ribosome. When non-AUG codons are utilized to initiate translation, translation at the first AUG codon is usually also observed as well. It has been proposed that translation initiation at non-AUG codons may be a mechanism of creating an N-terminally extended version of the encoded protein (42).
Leaky scanning may also have a role in the production of RAR␤Ј from the RAR␤2 mRNA, as mutation of the nucleosides surrounding the RAR␤2 translation start site to exactly conform to the Kozak consensus sequence abolished translation initiation downstream. It is interesting to note that the RAR␤2derived transcript having an AUG to AUA mutation at the T1 initiation codon did not abolish translation at T1 as expected, although it did result in greater usage of the downstream T2 as expected by the leaky scanning model (Fig. 2C). We are currently investigating the possibility that structural or sequence elements in the transcript leader sequence are also involved in the RAR␤2 mRNA T1 site recognition. We do not yet understand the mechanism by which the endogenous RAR␤Ј isoform is expressed in breast tumor cells at higher levels than in normal HMECs. It has been demonstrated previously that the ratio of C/EBP␣ and C/EBP␤ protein isoforms that are generated by leaky scanning can be dramatically shifted during mammary gland lactation (51) and breast epithelial cell tumorigenesis (52). This shift in translation start site usage is regulated in adipocytes by the eukaryotic initiation factors (eIF) eIF2␣ and eIF4E (53). eIF4E is up-regulated in a high percentage of primary human breast tumors (54), and the levels of eIF4E expression in breast tumors are thought to be of prognostic significance in determining the risk of tumor recurrence (55). It is interesting to note that the presence of evolutionarily conserved uORFs in the C/EBP␣ and C/EBP␤ transcripts are required for proper regulation of isoform expression by leaky scanning (53). As upstream open reading frames of the RAR␤2 transcript leader region have been shown to regulate tissuespecific expression of the mouse RAR␤2 protein at the level of translation initiation (56,57), future experiments will also be performed to identify whether the levels of RAR␤ isoforms are regulated by the cell at the level of translation initiation.
Structurally, the RAR␤4 protein is identical to RAR␤2 except that it lacks the N-terminal 49 amino acids that make up the AF-1 domain of RAR␤2. Also identical to RAR␤2 at the C terminus, RAR␤Ј lacks the first 113 amino acids of RAR␤2. This 113-amino acid region contains the AF-1 domain and one of two zinc finger motifs of the DNA-binding site. Other naturally occurring transcription factor isoforms have been described that also lack a DNA-binding domain. Most analogous to RAR␤Ј is the 11.4 kb progesterone receptor C mRNA that encodes an N-terminally truncated progesterone receptor. Like RAR␤Ј, progesterone receptor C is made up of amino acids that comprise the dimerization, ligand-binding, and ligand-activated transactivation domains of the full-length progesterone receptor but lacks all peptides N-terminal to the second zinc finger of the DNA-binding domain (58). Another example is the Id family of proteins. Id (or "inhibitor of DNA binding") proteins are closely related to the basic helix-loop-helix family of transcription factors that play key roles in cell type determination and differentiation. Id proteins contain a conserved helix-loophelix dimerization motif but lack an upstream basic domain that mediates DNA binding (59). Id proteins inhibit DNA binding of other basic helix-loop-helix transcription factors to their DNA elements and inhibit transactivation mediated by basic helix-loop-helix transcription factors in reporter gene assays (59).
In transient transfection assays, atRA induced expression directed by the retinoid-inducible promoter ϳ2-fold in cells having only endogenous RARs and ϳ3.4-fold in cells that were transfected with full-length RARs (e.g. Fig. 6). In comparison, and representative of the literature, Durand et al. (50) report an increase in atRA activation of a DR-5 RARE-activated promoter from 2-fold to 15-30-fold after ectopic expression of RAR␣ into COS-1 cells. There are several lines of evidence to suggest that the amplitude of atRA-induced gene expression that we find in our transient transfection assays is likely an accurate reflection of the degree of atRA-induction of RAR␤ in the MCF-7 cells and is not a result of transfection or chromosome packaging artifacts. Quantification of endogenous RAR␤ expression in MCF-7 cells by RNase protection assays revealed a 6-fold increase in the transcript after 3 days of exposure to 1 M atRA (60). The level of atRA-induced activation that we observe in our cells is consistent between reporter genes that are stably integrated into chromosomes and reporter genes that are present as transiently transfected plasmids. Additionally, MCF-7 cells express significant quantities of RAR␣, RAR␤, RAR␥, and RXR␣ proteins by Western blot analyses (data not shown), and would be expected to activate reporter gene expression without further addition of receptors upon retinoid activation. Reporter gene expression without atRA addition to the cell culture media likely results from activation effected by the presence of endogenous retinoic acids in serum, which are ϳ10 nM for atRA alone in fasting human and rat serum (61). Although treatment of serum with charcoal-dextran reduces the steroid hormones in serum, it only poorly removes the hydrophilic retinoic acids. Note that the ED 50 of atRA activation of RAR␣-mediated transcription is 0.6 nM (62) and that addition of atRA to a concentration of only 10 nM induced similar amounts of reporter gene expression as that of the pharmacological dose of 1 M routinely used in transient transfection assays (Fig. 6).
In functional characterizations of the three RAR␤ isoforms that are expressed in normal and breast tumor-derived mammary cell lines, we find that the RAR␤4 protein behaves similarly to RAR␤2 in binding the ␤RARE as a heterodimer with RXR␣ (Fig. 4A). Additionally, both RAR␤2 and RAR␤4 show equivalent transactivation of a reporter gene containing a DR-5 RARE in the presence of physiological (0.01 M) and pharmacological (1 M) levels of atRA (Fig. 6). In contrast, RAR␤Ј does not bind the DR-5 RARE in vitro (Fig. 4). Importantly, RAR␤Ј expression antagonizes retinoic acid-activated transcription mediated by RAR␤2, RAR␤4, and another RAR family member, RAR␣ (Fig. 7). The inhibition by RAR␤Ј of transcription mediated by other RAR subtypes (Fig. 7) demonstrates that RAR␤Ј may exert a general repression of transcription by other nu-clear receptors. Because RAR␤Ј with mutations in the AF2 cofactor interacting domain failed to repress atRA-induced transcription of RAR␤2 (Fig. 8B), it is likely that the ability to compete for transcription cofactors is a requirement for RAR␤Ј inhibition of retinoid-activated transcription.
We hypothesize that the RAR␤2 and RAR␤4 transcripts generate both positive-acting (RAR␤2 and RAR␤4 proteins) and negative-acting (RAR␤Ј protein) factors for transactivation at RAREs as a mechanism of fine-tuning atRA-induced transcription. Unlike some other reported dominant negative nuclear receptors (6,50,63), RAR␤Ј does not bind cis-acting DNA elements and therefore cannot directly inactivate gene transcription. RAR␤Ј likely represses by stoichiometric competition, away from the RARE, against other transcription factors within the cell (e.g. RAR␣, RAR␤, and RAR␥) for transcription cofactors. Additionally, although the nuclear localization signal is retained in the RAR␤Ј protein, a significant fraction of RAR␤Ј is located in the cytoplasm (31), decreasing the amount of RAR␤Ј available to compete for transcription cofactors.
Although previous reports suggest that RAR␤4 mRNA expression may be tumorigenic (64), we find that the RAR␤4 protein activated transcription from a RARE at similar levels as RAR␤2 in our reporter assays (Figs. 6 and 7). We also find that the RAR␤4 transcript is present at higher levels in normal mammary epithelial cells than most breast tumor cell lines (31). It is possible that tumorigenic effects reported for RAR␤4 expression may in fact be due to overexpression of the RAR␤Ј isoform at T2 in the RAR␤4 transcript. Whereas the mechanism regulating expression of these proteins in HMECs and breast cancer cells awaits further investigation, it is clear that future expression studies of the RAR␤ gene products must include an analysis of the RAR␤ isoforms at the protein level.
Our finding of RAR␤Ј as an inhibitor of retinoic acid-mediated transcriptional activation reveals an additional layer of complexity to retinoid signaling. As the evidence presented here suggests that RAR␤Ј competes with other RARs for transcription co-activators, it would likely inhibit other nuclear receptors that partner with these co-activators, implicating RAR␤Ј as a potential regulator of signaling by other pathways in addition to the RAR family.