Specific Activation of Retinoic Acid Receptors (RARs) and Retinoid X Receptors Reveals a Unique Role for RARγ in Induction of Differentiation and Apoptosis of S91 Melanoma Cells*

Retinoic acid (RA) and 9-cis-RA induce growth arrest and differentiation of S91 melanoma cells. RA activates retinoic acid receptors (RARs), whereas 9-cis-RA activates both RARs and retinoid X receptors (RXRs). Both classes of receptors function as ligand-dependent transcription factors. S91 melanoma cells contain mRNA for RXRα, RXRβ, RARα, RARγ, and RARβ in low levels. Among these, only RARβ gene transcription is induced by retinoids. However, at present the individual role(s) for each RXR and RAR isoform in these processes is unclear. We assessed the function of all isoforms in the S91 melanoma model by using RXR and RAR isoform-specific retinoids to study their effects on cell growth, RARβ expression, and differentiation. Activation of each of the endogenous RXR or RAR isoforms induces RARβ gene expression, and blocks cellular proliferation. However, only the RARγ-ligands cause additional differentiation toward a melanocytic phenotype, which coincides with substantial apoptosis well before morphological changes are apparent. Apoptosis is completely dependent on de novo protein synthesis but cannot be induced by changes in activities of AP-1, protein kinase C, and protein kinase A, nor can it be blocked by the presence of the antioxidant glutathione. These results argue against a specific role for RARβ, but suggest that RARγ has a critical role in a genetic switch between melanocytes and melanoma, and induction of ligand-dependent apoptosis.

stereoisomer, 9-cis-RA (5,6), whereas RXRs only bind the latter. Ligand-bound receptors can modulate the expression of genes containing appropriate response elements (RA response element, RARE). A productive RARE generally consists of the direct repeat AGGTCA spaced by two or five nucleotides (DR2 and DR5, respectively), although other functional configurations exist (Ref. 1, and references therein). RXRs activate through retinoid X response elements (RXREs), which resemble RAREs except that the spacing for an optimal response is one nucleotide (DR1) (7)(8)(9). It is generally assumed that RARs are heterodimerized with RXRs inside the cell (7, 8, 10 -16), but RXRs can also form homodimers in the presence of its own ligand, 9-cis-RA (8,9). RA especially is known to have profound effects on vertebrate development and differentiation in vivo (17,18), and it can also induce differentiation of a large number of tumor cell lines in vitro (19). Thus, RARs and RXRs are at the peak of a pyramid of an extremely complex genetic network to control cell growth and differentiation. Conceivably, mutations or aberrant expression levels of these receptors might lead to neoplasia (20 -22). S91 murine melanoma cells constitutively express RAR␣, RAR␥, RXR␣, RXR␤, and very low levels of RAR␤ (23)(24)(25). Upon treatment with RA or 9-cis-RA, a reversible conversion of malignant melanoma into a benign, melanocytic phenotype takes place suggesting that a specific genetic program is induced and maintained by RARs and/or RXRs (23,(25)(26)(27). Administration of either of these two retinoids results in rapid up-regulation of RAR␤ expression, followed by cessation of cell division, and morphological differentiation (23)(24)(25). A correlation was observed between retinoid-dependent induction of the RAR␤ gene and growth arrest, indicating that RAR␤ might be involved (23). The RAR␤ promoter contains a typical RARE (␤RARE) of the DR5 type (28 -30), but at least in S91 cells, also appears to be regulated by RXR (25). Unfortunately, it is difficult to assess the importance of individual RAR isoforms and RXRs in the above described processes using pan-RAR and pan-RAR-RXR agonists like RA and 9-cis-RA, respectively. A more detailed comprehension of the underlying molecular events leading to neoplastic change in this model would be advantageous to our basic understanding of cancer, and it might benefit the development of improved therapies (31).
We set out to answer these questions by using a number of RXR and RAR isoform-specific agonists to evaluate their effects on cell growth, RAR␤ expression, and morphological differentiation. Our results suggest that all receptors can induce growth arrest and transcriptional activation of the RAR␤ gene. In contrast, only the RAR␥ ligands cause morphological differentiation which, interestingly, occurs concomitantly with sub-stantial apoptosis well before phenotypic changes become apparent. The mechanism of apoptosis is dependent on newly synthesized proteins, suggesting that new gene transcription is a necessary prerequisite. Changes in activities of AP-1, PKC, or PKA do not lead to apoptosis, whereas high concentrations of the free oxygen radical scavenger glutathione (GSH) do not block RAR␥ agonist-induced cell death. These results suggests that redundant mechanism(s) inhibit cellular proliferation and RAR␤ expression. Activation of the RAR␤ promoter and/or receptor is not a necessary condition for growth arrest in these cells. It also shows that there is no default mechanism leading to phenotypic differentiation after growth arrest. Instead, only RAR␥ appears to be able to activate the differentiation program, and induce ligand-dependent apoptosis.

Cell Culture and Cell Proliferation Assay
All tissue culture plates and flasks were from Costar (Cambridge, MA). S91 cells (ATTC CCL 53.1) were grown in Dulbecco's modified Eagle's medium with 10% (v/v) fetal calf serum at 37°C in 5% CO 2 in humidified air. Retinoid stock solutions (10 or 1 mM in Me 2 SO) were added to media to achieve the desired concentration. Final Me 2 SO concentrations in cell culture medium never exceeded 0.1% (v/v). For the cell proliferation assay, cells were seeded at a density of about 7,500 cells/well in triplicate in 96-well plates, and treated for 5 days in total with the indicated retinoids. Media with/without retinoids was changed every 24 h. At the fifth day, relative viable cell numbers were obtained using the CellTiter 96 TM AQ ueous Non-Radioactive Cell Proliferation Assay kit from Promega (Madison, WI). Absorbance at 490 nm was determined in a UV max kinetic microplate reader (Molecular Devices (Sunnyvale, CA). Morphology of cells was monitored on an Olympus IMT-2 phase-contrast microscope, and photographed through an attached Olympus OM-2s program camera (Tokyo, Japan) on Kodak T-Max 400 film (Rochester, NY).

Plasmids, Transfection, and Luciferase/␤-Galactosidase Assay
A plasmid with a 4.4-kilobase chromosomal fragment containing the RAR␤2 promoter linked to the luciferase (Luc) gene, pW1RAR␤2pr-lucif (32), was kindly provided by Dr. E. Linney with permission of Dr. P. Chambon. Dr. Chambon also provided us with pSGRAR␤2 (33). Plasmids were grown in Escherichia coli strain DH5␣ in LB media with ampicillin, and purified using a Qiagen plasmid maxi kit (Chatsworth, CA). For the transfection with pW1RAR␤2pr-lucif, cells were grown in large flasks (165 cm 2 ) until 75% confluent. Per flask, cells were harvested by trypsinization, washed in ice-cold Dulbecco's modified Eagle's medium, resuspended in 0.5 ml of ice-cold Dulbecco's modified Eagle's medium in a 0.4-cm electroporation cuvette (Invitrogen, San Diego, CA), and incubated for 5 min on ice with 40 g of pW1RAR␤2pr-lucif and 12 g of CH110 (Pharmacia Biotech Inc., Piscataway, NJ). CH110 constitutively expresses ␤-galactosidase activity which serves as an internal control for the transfection efficiency. Cells were then electroporated using a Electroporator II (Invitrogen, San Diego, CA) set at 200 V, 1000 microfarads, incubated on ice for 10 min, and resuspended in Dulbecco's modified Eagle's medium, 10% fetal calf serum at 37°C, and plated on a 6-well plate. After 36 h, cells were treated with retinoids and harvested 12 h later. Luciferase assays and ␤-galactosidase assays were performed as described (34). Values from the luciferase assay were normalized for ␤-galactosidase activity. Fold induction means normalized luciferase activity in the presence of retinoid/normalized luciferase activity in the absence of retinoid (control, Me 2 SO).

Northern Blot Analysis
Cells were grown in 150-cm 2 plates until 75% confluent. Medium was changed containing the appropriate amounts of retinoid or control (Me 2 SO), and/or 10 g/ml cycloheximide (Sigma). Cells were harvested after 8 h, washed twice with phosphate-buffered saline, and poly(A) ϩ mRNA extracted using a Qiagen Direct mRNA Midi Kit (Chatsworth, CA). Two g of mRNA was subjected to electrophoresis through a denaturing formaldehyde-agarose gel (1%). Blotting procedures, hybridization/washing conditions, preparation of radioactively labeled probe, and quantitation were all performed as described (25). Cyclophilin probe serves as internal control for loading and blotting, and used to normalize values obtained from the RAR␤ probe. Fold-induction means normalized value of RAR␤ probe (in the presence of retinoid and pres-ence or absence of cycloheximide)/value of RAR␤ probe (in the absence of retinoid and absence of cycloheximide).

Melanin Assay
Cells were grown in 10-cm plates, and treated for 5 days with the required amount of indicated retinoids or control (Me 2 SO). Medium with/without retinoids was changed every 24 h. Adherent cells were harvested by trypsinization and washed twice with phosphate-buffered saline. 2 ϫ 10 4 cells were pelleted and lysed overnight in 100 l, 1 N KOH at 80°C, and absorbance at 490 nM was determined.

Apoptosis Assays
DNA "Laddering"-Cells were grown until 50% confluent in 10-cm plates and treated for 24 h with the indicated retinoids. 4 ϫ 10 6 cells (adherent and floating) were harvested by trypsinization, washed twice in phosphate-buffered saline, and pellets were lysed in 400 l of lysis buffer (10 mM EDTA, 50 mM Tris, pH 8.0, 0.5% (w/v) Sarcosyl, 0.5 mg/ml proteinase K) for 3 h at 50°C. Next, protein was removed by treatment with an equal volume of phenol/chloroform/isoamyl alcohol (50:48:2), and once with chloroform/isoamyl alcohol (24:1). DNA was precipitated, dissolved in 100 l of TE buffer with 0.25 mg/ml RNase A, and incubated at 50°C for 1 h followed by 1 h at 37°C. 25 l were then loaded onto a 1.5% agarose gel in TAE buffer with 0.5 g/ml ethidium bromide and separated by electrophoresis. DNA was visualized by UV light at 305 nM and photographed.
Cell Death Detection by Enzyme-linked Immunosorbent Assay-Cells were grown in 6-well plates, and adherent and floating cells were harvested by trypsinization, washed twice in phosphate-buffered saline, and counted. Pelleted cells were then treated following the manufacturer's protocol with the Cell Death Detection ELISA (Boehringer Mannheim, Indianapolis, IN). 1,000 cells were used in the enzyme-linked immunosorbent assay, and absorbance was determined at 405 nM in a UV max kinetic microplate reader (Molecular Devices, Sunnyvale, CA).

Binding Assay
Equilibrium dissociation constants (K d values) for the interaction of the different retinoids with the three RAR isoforms were determined in vitro at 4°C by measuring displacement of radiolabeled reference retinoid CD367 by increasing dose of non-radioactive ligands. The assays were performed as described before, using RARs that were obtained by overexpression in COS-7 cells (35). At least three independent determinations for binding were performed. The average standard error of the mean was within 25%. K d values and RXR-specific transactivation properties of the pan-RXR ligand CD2624 were described previously (36). Except for CD2624, none of the compounds listed in Table I have significant binding affinity for RXR, nor do they transcriptionally activate this receptor (not shown).

Retinoids
The chemical nomenclature for the synthetic ligands is: Am80  (91). RA was obtained from Sigma. All other retinoids were synthesized at CIRD Galderma (Valbonne, France). All retinoids were kept as 30 mM stock solutions in Me 2 SO, and kept at Ϫ20°C to Ϫ80°C.

Ligands Which Specifically Bind and Activate RAR␣, RAR␤, RAR␥, or RXR Cause Growth
Arrest-For the studies described here, we selected the following synthetic retinoids (listed in Table I), and determined their K d values for each RAR isoform (35): Am80 and Am580, which have a strong preference for RAR␣; CD417 and CD2314, which are selective for RAR␤; CD437 (also named SR11248 or AHPN) and CD2325, which bind to RAR␥. These results are generally in good agreement with their transactivational properties as reported previously (39,40). Finally, we used CD2624, which binds and activates all RXR isoforms, but not RARs (36).
First, the efficacy of these retinoids in inducing growth arrest was examined. Cells were treated for 5 days with different concentrations of retinoids with or without CD2624 as indicated ( Fig. 1), or CD2624 alone, to study the effects of single activated RAR isoforms in the presence or absence of activated RXR, respectively. After the last day, relative cell numbers were obtained by means of a colorimetric assay, in which the absorbance at 490 nM directly correlates with the number of viable cells. The absorbance values of control-treated cells was arbitrarily set at 100%, and all other values were normalized accordingly. Media and retinoids were changed every 24 h to minimize loss of concentration of intact ligands due to chemical decomposition, as well as to minimize potential aberrant specificities by those byproducts (41).
As shown in Fig  these compounds (42). Simultaneous addition of RAR and RXR ligands renders the cells about 10-fold more sensitive, indicative of a synergistic rather than an additive effect. These results suggests that all RARs and RXR␣ and/or RXR␤ can, directly or indirectly, inhibit cellular proliferation. Induction of RAR␤ Gene Expression by Ligands Specific for RAR␣, RAR␤, RAR␥, and RXR-Because RAR␤ has been implicated in the above described processes, we next studied the induction of the RAR␤ gene by two different methods. Transcription of this gene is rapidly induced by RA (within 2 h), even in the absence of protein synthesis (23,24). Cells were transiently transfected with a plasmid, pW1RAR␤2pr-lucif, which contains the chromosomal fragment harboring the RAR␤2 promoter linked to the luciferase (Luc) gene (32). Transcriptional activation of this promoter is then detected as enzymatic luciferase activity in the cell extract ( Fig. 2A). Three concentrations of each ligand were chosen, varying between 10 Ϫ8 and 10 Ϫ5 M, based on their displayed efficacy in the cell proliferation assay. Thirty-six hours after transfection, cells were exposed to all ligands for an additional 12 h because this is the earliest time point at which significant luciferase activity could be detected (data not shown). Fig. 2A shows that the promoter is activated by all ligands, including CD2624, in a dose-dependent manner. Maximum induction levels are generally reached between 6-and 15-fold at 10 Ϫ7 and 10 Ϫ6 M. These data correlate well with their activity in the cell proliferation assay, and suggest that all expressed receptors can activate this promoter.
The second method measures induction of the endogenous promoter. Cells were treated for 8 h with a representative group of retinoids at a concentration of 10 Ϫ6 M for maximum induction. In addition, cells were incubated at the same time in the absence or presence of the protein synthesis inhibitor cycloheximide to test whether induction is dependent on on-going protein synthesis. After treatment, poly(A) ϩ RNA was isolated and separated by agarose gel electrophoresis, and RAR␤ mRNA was detected by Northern blot analysis (Fig. 2B). We again see that the RAR␤ gene is rapidly induced by all tested agonists. Also, the level of induction (10 -15-fold) corresponds well with the values found in the transfection assay. Thus, regulation of expression of this gene can be mediated by ligands specific for all receptors, including RXR␣ and/or RXR␤. The latter result confirms our previous report that RXR can transcriptionally activate this promoter despite its unfavorable (to RXRs) ␤RARE (28 -30). Fig. 2B also shows that cycloheximide does not abrogate transcription of this gene. This suggests that all receptors, including RAR␤ whose mRNA levels are very low (but still detectable) in the absence of added retinoids ( Fig. 2B and Refs. [23][24][25], are likely already present in sufficient numbers to initiate a functional response. Interestingly, cycloheximide does not abrogate, but rather enhances induction by all tested agonists in a strictly retinoid-dependent fashion; on average, RNA levels increase to between 28-and 61-fold higher levels as compared with control treated cells. Superinduction of the RAR␤ promoter by cycloheximide was previously reported for treatment with RA (23, 24) but our results show that isoform-specific ligands also synergize with cycloheximide.
Thus, all ligands and, therefore, most likely all RARs as well as RXR␣ and/or RXR␤ can activate RAR␤ expression. Even though we also observe a correlation between growth arrest and RAR␤ expression, our results show that the accumulation of RAR␤ occurs in the absence of RAR␤-activating ligands. These data argue against a critical role for ligand-bound RAR␤ in S91 cells.
RAR␥ Agonists Specifically Induce Morphological Differentiation and Melanin Synthesis-After 5 days of treatment with the RAR␥-specific ligands CD437 and CD2325 at 10 Ϫ6 M, we observed a substantial loss in the number of adherent cells (see below). In addition, compared with all other treated cells, the morphology of the remaining adherent cells is that of a differentiated, melanocytic phenotype including dramatic cell flattening and growth of dendritic extensions (Fig. 3A), which is independent of initial cell density, although very sparsely seeded cells are more spindle-shaped than more densely growing cells (compare Fig. 3, B with A, respectively). The RAR␥ligand-induced morphology resembles that which is obtained after treatment with RA (Ref. 23, and data not shown). These attached cells also display another marker for melanocytic differentiation, increased intracellular melanin (23,27) (Fig.  3C). In contrast, the other ligands induce no or much less melanin. This suggests that RAR␥ uniquely among the expressed retinoid receptors mediates the differentiation of these melanoma cells. Our results also imply that growth arrest may be a necessary but not sufficient condition for morphological differentiation, indicative of two distinct and separate pathways. A similar principle was postulated for RA-dependent differentiation of neuroblastoma (NB) cells (43).

Concomitant with Differentiation, RAR␥-specific Ligands
Cause Rapid Apoptosis-Finally, we examined the mechanism(s) leading to the loss of adherent cells upon addition of the RAR-specific ligands, CD437 and CD2325, at 10 Ϫ6 M. At this concentration, substantial numbers of floating cells can be observed after about 24 h treatment. These cells largely retain dye after staining with trypan blue, suggesting that they are dead or dying (data not shown). We reasoned that the two ligands could either be toxic to the cells, which could lead to necrosis, or might induce genetically programmed cell death (apoptosis). Retinoid-induced apoptosis has been observed in a number of other retinoid-treated tumor cell lines (44 -57). To examine whether the latter mechanism would also apply here, we tested for two different markers of apoptosis.
First, genomic DNA was isolated from all cells (adherent and floating) that were treated for 24 h with our agonists, followed by analysis by agarose gel electrophoresis. DNA laddering due to internucleosomal fragmentation of DNA in multiples of 180 nucleotides, typical of apoptotic cells, can be observed in the lanes with RAR␥ agonist-treated cells, but not in any of the other lanes (Fig. 4A).
We also applied a different method based on the sandwich enzyme immunoassay principle which detects and quantifies accumulating mono-and oligonucleosomes in the cytoplasmic fraction of apoptotic cells. Our results show that apoptosis occurs between 8 and 16 h of treatment with CD2325, after which it no longer increases (Fig. 4B). Consistent with the morphological observations and the amount of DNA laddering in Fig. 4A, similar final absorbance values in this assay are measured after treatment with CD437. In contrast, Am580, CD2314, and CD2624 show little or no increase in absorbance after 24 h (Fig. 4B), and RA does not cause significant apoptosis either.
Evidence That the Mechanism of RAR␥ Agonist-induced Apoptosis Is Directly Mediated by RAR␥-To initially investigate whether the mechanism of apoptosis requires newly synthesized proteins, we simultaneously incubated cells for 16 h in the presence or absence of CD437 and CD2325, and with or without cycloheximide. As shown in Fig. 5A, these agonists again induce extensive apoptosis as detected by our sandwich enzyme immunoassay. However, apoptosis can be completely blocked by the addition of cycloheximide, whereas cycloheximide alone has no significant effect on cell viability. This experiment shows that the protein(s) which mediate apoptosis are not yet present in the untreated melanoma cells, but are induced by the RAR␥ agonists and suggest that new gene transcription is required. Since functional RAR␥ is already present in the melanoma cells, and CD437 and CD2325 are strong activators of this receptor, it is quite conceivable that RAR␥ is responsible for induction of these putative genes.
However, this could still be an indirect, rather than a direct transcriptional effect. It is known for some time that liganded RARs can interfere with AP-1 (Jun)-mediated gene transcription, possibly by directly interfering with DNA binding by AP-1 through protein-protein interactions. Conversely, AP-1 can also block DNA binding of RAR⅐RXR heterodimers (58). It was recently reported that the anti-AP-1 activity of liganded RARs, and not their transactivational properties, is responsible for the growth-inhibitory effects of retinoids on certain tumor cell lines (59 -61). In our model, RA-liganded RARs could block AP-1-mediated transcription, which could prevent induction of apoptosis by AP-1. Possibly, CD437 (and CD2325) could bind to RAR␥ in a different fashion and alter its conformation such that it would no longer interact with AP-1, and this release may then lead to cycloheximide-sensitive apoptosis. To investigate this possibility, we performed an EMSA, shown in Fig. 5B. As a source of proteins, RAR␥ and RXR␣ were obtained by in vitro translation in the rabbit reticulocyte system, whereas recombinant AP-1 (Jun) was commercially obtained. As probe, we used an oligonucleotide with the well established ␤RARE (25). As expected, unprogrammed and RAR␥or RXR␣-programmed lysate displayed little DNA binding activity (lanes 1, 9, and 10 5 and 7, respectively), which indicates that there are no major differences in receptor conformation induced by these two ligands. Next, to imitate the situation in our cells after addition of ligands, we added pure Jun protein to the liganded heterodimer complex. As seen in lane 6, we too observe that Jun completely inhibits DNA-binding of RA-liganded RXR␣⅐RAR␥ heterodimers as their is no shifted complex visible anymore. The same observation is made when Jun is added to CD437-bound RXR␣⅐RAR␥ heterodimers. Thus, in vitro there is no discernable differential ligand-specific effect in the behavior of liganded-RXR␣⅐RAR␥ heterodimers with respect to their interaction with AP-1 (Jun). The same appears to be true in vivo as well. If cells undergo apoptosis because of unmasking of AP-1-mediated transcriptional activity, it can be expected that direct activation of AP-1 activity might lead to apoptosis. For this purpose, cells were treated for 16 h with increasing doses of phorbol 12-myristate 13-acetate (also called 12-O-tetradecanoylphorbol-13-acetate). Phorbol 12-myristate 13-acetate is a well known activator of PKC and AP-1 activity. However, no apoptosis can be detected over this time frame in the presence of up to 100 ng/ml (Fig. 5C). In the same experiment, apoptosis starts to become measurable at 10 Ϫ7 M CD437, and is fully established at our working concentration of 10 Ϫ6 M (Fig. 5C). Similarly, there is also no effect in the presence of the PKA activator 8-Br-cAMP (Fig. 5C). These results all suggest that AP-1, PKC, and PKA activities are not of critical importance for the mechanism of CD437 and CD2325-induced apoptosis.
Finally, another mechanism could be at work here which would be independent of RARs. It is possible that these retinoids induce oxidative stress due to the formation of reactive oxygen radicals, and this may induce gene transcription. Transforming growth factor-␤ has been shown to induce apoptosis in certain cells through this mechanism (Ref. 62, and references therein). To investigate this possibility, we incubated the cells for 16 h in the presence of CD437 and increasing doses of up to 1 mM GSH. GSH is a free oxygen radical scavenger, and if this mechanism is applicable here it might be expected that GSH could inhibit CD437-induced apoptosis. However, as shown in Fig. 5C, addition of GSH has no measurable effect on CD437-induced apoptosis. This makes this mechanism less probable here. DISCUSSION Retinoids are known to have profound effects on growth and differentiation of cells in vivo and in vitro (17)(18)(19)(20). S91 melanoma cells provide a model system that allows the study of neoplastic change of melanocytic cells into melanoma in a controlled experimental setting. The entire process is strictly retinoid-dependent, and, with the data presented here as well as in other reports (23)(24)(25), all active and expressed RARs and FIG. 5. The mechanism of RAR␥ agonist-induced apoptosis requires protein synthesis, and appears independent of changes in AP-1, PKC, and PKA activities or induction of oxidative stress. A, similar immunosandwich assay as in Fig. 4B. Apoptosis by CD437 and CD2325 is dependent on de novo protein synthesis because it can be blocked by cycloheximide (CHX). Cells were treated for 16 h with 1 M CD437, CD2325, or Me 2 SO (control) in the presence or absence of cycloheximide (2.5 g/ml). Absorption is given for 10 3 cells in each case. Shown is mean of representative experiment done in triplicate which was repeated at least three times with similar results. RXRs in S91 cells have now been determined. However, a list of all potential players involved in the first step in the genetic cascade which leads to growth arrest and differentiation does not reveal the importance of each RXR and RAR isoform. This study attempts to dissect the individual roles of all receptors in these processes by using ligands that have high specificity for certain isoforms only. In accordance with these data, our results show that these agonists indeed have very specific effects on cellular proliferation, viability, morphological differentiation, induction of RAR␤ gene expression, and apoptosis, as discussed below.
First, we show that inhibition of cell division and transcriptional induction of the RAR␤ gene can be mediated by all ligands, and thus, most likely all RAR isoforms and RXR␣ and/or RXR␤. Functional redundancy in RAR function has been observed before, both in vivo and in vitro. For instance, RAR␣ and RAR␥ gene knock-outs in transgenic mice do not cause the embryos to die in utero, but they can develop relatively normally until birth (63)(64)(65). RAR␤ knock-out mice have no discernable phenotype at all (66). RAR␣ and RAR␥ knock-outs in F9 embryo carcinoma (EC) cells do not abrogate RA-dependent induction of RAR␤ expression (41,67). Presumably, in all those cases as well as in others (68 -71), alternative receptors can substitute for functions performed by the deleted RAR genes.
It is conceivable that RAR␤ could play a major role in the RA-induced differentiation of S91 cells (23). Indeed, exogenously introduced RAR␤ can halt the growth of lung cancer cells (71), and RAR␤ gene expression and/or RAR␤ mediated activity appears to be altered in lung tumor cells (72)(73)(74) and other (pre)cancerous epithelial cells (22,75,76). However, our results show that, in S91 cells, ligand-bound RAR␤ is not required for growth arrest or differentiation, because ligands that do not activate RAR␤ also cause growth arrest. It is interesting that the RAR␤ ligands are about equipotent in inducing RAR␤ gene transcription as the RAR␣, RAR␥, and RXR ligands. In the absence of exogenously added retinoids, expression of RAR␤ is extremely low. Thus, it appears that the base-line level of RAR␤ in untreated cells is high enough to induce transcription, which would also explain why inhibition of protein synthesis by cycloheximide does not abrogate RAR␤mediated transcription. It is unknown why cycloheximide enhances retinoid-dependent transcription, but our results indicate that all RAR isoforms and RXR might interact with the same putative labile transcriptional repressor protein(s) (23).
Second, our results show that only the RAR␥-ligands induce morphological differentiation toward the melanocytic phenotype which is normally only seen after treatment with RA and 9-cis-RA (Ref. 23, and data not shown). This identifies RAR␥ as most likely to be the critical RAR isoform which maintains these melanocytes in their differentiated state, and prevents them from converting to melanoma. We postulate that RAR␥ can uniquely modulate the expression of a set of "differentiation" genes, which cannot be regulated by RAR␣, RAR␤, or RXR. This conclusion is based on the assumption that the selectivity of the ligands, which is determined in vitro at 4°C, is preserved in vivo at 37°C. This is difficult to determine directly due to the lack of mammalian cell lines which do not express RARs and/or RXRs. However, the available evidence suggest that ligand-binding selectivity is indeed maintained over this temperature range (38,77).
RAR␥-induced differentiation may be mediated by a novel type of RAR␥-responsive RARE, present in genes which are activated or repressed by only this particular receptor, or could involve new isoform-specific cofactors similar in function to those in the AP-1 family (78). Differential activation of certain RAREs by different receptor isoforms has been shown before (Ref. 33, and see below).
Previously, a unique role for RAR␥ among RARs in the differentiation of tumor cells was established in the more "primitive" embryo carcinoma (EC) cell lines, NTera2 and F9, which differentiate into a neuronal-looking, phenotype-type and parietal endoderm after administration of RA, respectively (41,79). Separate transfection of RAR␣, RAR␤, or RAR␥ expression vectors into NTera2 cells showed that only the latter transfectant gives rise to a cell with a differentiated phenotype. Somewhat puzzling, however, is that this RAR␥-expressing transfectant will now differentiate in the absence of exogenously added RA (80). Gene ablation experiments in F9 EC cells showed that RAR␥ is mostly responsible for differentiation. This laboratory also showed that different RAR isoform knock-outs differentially affect the expression of RA-regulated genes in F9 cells (41,67). However, reintroduction of plasmids overexpressing either RAR␣ or RAR␥ (but not RAR␤) in RAR␥ Ϫ/Ϫ cells showed that both RAR isoforms could restore differentiation (25). Consistent with our data, as well as that of others, they suggested that certain genetic functions are redundant (63-65, 68, 69, 81) but others may be regulated by specific RAR isoforms (33,67,(82)(83)(84), perhaps through the aforementioned isoform-responsive RAREs. A unique role for RAR␥ in neuroblastoma cells is less well established than that in EC cells. Marshall et al. (43) showed that RAR␥ expression is significantly higher in primary neuroblastoma tumor tissue compared with advanced or disseminated neuroblastoma tumors. Also, transfection of an RAR␥ expression vector into the neuroblastoma cell line BE(2)-C altered its neuritic differentiation potential. However, other reports suggest that RAR␣ and/or RAR␤ is involved in differentiation of neuroblastoma cells (85)(86)(87).
Finally, we observed that the RAR␥ ligands, CD437 and CD2325, not only induced differentiation, but also apoptosis. Retinoids have been shown to induce apoptosis in several other different tumor cell lines, and can occur during or after differentiation, or occur in the absence of any differentiation (44 -57). Recently, induction of growth arrest and apoptosis by CD437 (or AHPN) was reported in the breast cancer lines MCF-7 and MDA-MB-231, which appears to take place with similar kinetics as in our cells (56). This mechanism(s) is independent of p53, bax, and bcl-2, and may be dependent on regulation of WAF/CIP1. In addition, the authors suggest that it is at least partly independent of RARs because promyelocytic HL-60R cells, a RA-resistant subclone of HL-60 cells, are also growth arrested by CD437 (56), although the authors do not report whether HL-60R cells will also undergo apoptosis after treatment with CD437. On the other hand, a report by Liu et al. (50) showed the involvement of RAR␤ in retinoid-mediated apoptosis in breast cancer cells. It is clear that one should be cautious in extrapolating results from one cell line to another. For instance, with regard to HL-60 cells, apoptosis is mediated by RXRs (53) whereas RAR␣ mediates their differentiation (68). According to our data, these receptors do not appear to be very important in the S91 melanoma model. Another example is RA-dependent apoptosis in the tracheobronchial cell line SPOC-1. Here, RAR␣ agonists cause apoptosis but CD437 has no effect (57). It appears that the cell type strongly influences which pathway leading to differentiation and/or apoptosis is activated.
Our evidence suggests that in S91 cells RAR␥ is activated by CD437 and CD2325, and our experiments with cycloheximide suggest that transcription and synthesis of new protein(s) which mediate apoptosis needs to take place. This would require some time to develop, and our observation that apoptosis becomes apparent about 8 h after administration of agonist would certainly not be inconsistent with this hypothesis. Moreover, we found little or no evidence which would support alternative, unexplained receptor-independent mechanisms, although we cannot completely rule these out.
Activated RARs can interfere with AP-1-mediated transcription, and we investigated the possibility whether the RAR␥ agonists would alter the conformation of the receptor in a different way than RA. We speculated that RA-bound receptors might block AP-1 activity and subsequent apoptosis, whereas the RAR␥ agonist-bound receptors may not. This could explain why RA does not cause apoptosis in contrast to the RAR␥ agonists. However, as judged by EMSA, both types of ligands allow the activated receptor to interact with AP-1. We were also unable to directly induce apoptosis by activation of AP-1 and PKC by phorbol 12-myristate 13-acetate (and PKA by 8-Br-cAMP). In addition, it was reported that increased activity of AP-1 takes place during cAMP-induced melanogenesis in the related B16 cells (88). These results all argue against involvement of AP-1 during apoptosis. Another possibility could be that CD437 and CD2325, and none of the other retinoids tested here, could lead to oxidative stress due to the formation of reactive oxygen intermediates. This could also lead to induction of gene transcription, and transforming growth factor-␤ is known to cause apoptosis through this mechanism (62). However, we found that GSH, a well established oxygen radical scavenger, does not block CD437-induced apoptosis. Again, these results certainly do not favor this mechanism, and make it less likely than a receptor-mediated mechanism. In addition, the observed requirement for new protein synthesis argues against an interleukin 1␤-converting enzyme-type mediated mechanism, as these proteins are ubiquitously present in the cytoplasm.
One important question which remains to be answered is why RA induces differentiation, but not apoptosis, in contrast to the synthetic RAR␥ agonists, even though they both must activate RAR␥. It is not due to an intrinsic inability of RA to cause cell death: in two other reports of concomitant apoptosis and differentiation by retinoids, one in a particular subclone of F9 EC cells (45), and the other in P39 myelomonocytic leukemia cells (44), the effects are induced by RA, albeit with slower kinetics and in a relatively small number of cells. We hypothesize that, under certain conditions, the RAR␥ agonists can be stronger activators of RAR␥ than RA, perhaps mediated by new cofactors (78). We speculate that there may be genes which have an RARE whose degree of transcriptional activation is dependent on the type of agonist bound to the receptor. In case of CD437 and CD2325 this may then lead to apoptosis. It has been shown before that conformationally restricted agonists can exhibit gene selectivity in the context of a given RARE, which could have important physiological consequences (89). For instance, CD437 and CD2325 are more powerful inducers of F9 EC cell differentiation than RA (38).
The unraveling of the complex retinoid-mediated pathways in the S91 melanoma model will be important for our understanding of neoplastic transformation, and this work is currently in progress in our laboratory. In this report, we identified RAR␥ as the responsible RAR isoform in S91 cells which operates a genetic switch of the molecular processes causing the malignant transformation of melanocytes into melanoma cells, and probably apoptosis. Hopefully, this may eventually lead to the design of more effective treatments and/or diagnostics in vivo.