Interaction between retinoic acid and vitamin D signaling pathways.

The nuclear signaling pathways for retinoids and vitamin D differ in the specificity of the respective receptors for response elements. Two pathways for the action of both retinoic acid receptors (RARs) and vitamin D receptors (VDRs) have been identified, one being retinoid X receptor (RXR)-dependent and the other being RXR-independent. Moreover, RXRs were found to function as homodimers. In several steps we converted the retinoid specific response element of the human retinoic acid receptor beta promoter into the vitamin D/retinoic acid response element of the human osteocalcin promoter. We found that VDR homodimers only bind to the motif RGGTGA. The extended osteocalcin element also contains an imperfect direct repeat based on the motif RGGTGA spaced by three nucleotides, which is bound by RXR homodimers and activated by 9-cis-retinoic acid. The responsiveness of the osteocalcin element to all-trans-retinoic acid is mediated neither by RAR homodimers nor by RAR-RXR heterodimers. However, a VDR-RAR heterodimer binds to the osteocalcin response element and mediates activation by all-trans-retinoic acid. This heterodimer also binds to pure retinoid response elements, but it does not mediate activation by vitamin D alone. In combination with all-trans-retinoic acid, however, vitamin D enhances VDR-RAR heterodimer-mediated gene expression. This finding suggests a direct interaction between nuclear signaling by retinoic acid and vitamin D increasing the combinatorial possibilities for gene regulation by the nuclear receptors involved.

Interaction between Retinoic Acid and Vitamin D Signaling Pathways* (Received for publication, March 16, 1993)

I~
The nuclear signaling pathways for retinoids and vitamin D differ in the specificity of the respective receptors for response elements. Two pathways for the action of both retinoic acid receptors (RARs) and vitamin D receptors (VDRs) have been identified, one being retinoid X receptor (RXR)-dependent and the other being RXR-independent. Moreover, RXRs were found to function as homodimers. In several steps we converted the retinoid specific response element of the human retinoic acid receptor 6 promoter into the vitamin D/retinoic acid response element of the human osteocalcin promoter. We found that VDR homodimers only bind to the motif RGGTGA. The extended osteocalcin element also contains an imperfect direct repeat based on the motif RGGTGA spaced by three nucleotides, which is bound by RXR homodimers and activated by 9-cis-retinoic acid. The responsiveness of the osteocalcin element to all-trans-retinoic acid is mediated neither by RAR homodimers nor by RAR-RXR heterodimers. However, a VDR-RAR heterodimer binds to the osteocalcin response element and mediates activation by all-trans-retinoic acid. This heterodimer also binds to pure retinoid response elements, but it does not mediate activation by vitamin D alone. In combination with all-trans-retinoic acid, however, vitamin D enhances VDR-RAR heterodimer-mediated gene expression. This finding suggests a direct interaction between nuclear signaling by retinoic acid and vitamin D increasing the combinatorial possibilities for gene regulation by the nuclear receptors involved.
The biologically active metabolites of vitamin A (retinol) and vitamin D, all-trans-retinoic acid (RA)' and 1,25-dihydroxyvitamin DS (VD), have profound effects on cellular differentiation, embryonic development and calcium homeostasis (1)(2)(3)(4). They exert their effects by binding as ligands to their respective intracellular receptors, RARs and VDRs, which are members of the large family of ligand-activated transcription factors termed the nuclear receptor family (5,6). Studies of estrogen and glucocorticoid receptors indicated that the classical steroid hormone receptors form homodimers and modulate transcription of their target genes via cis-acting palindromic response elements (7,8). The concept that each * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. hormone binds to a specific receptor homodimer was found to be oversimplified for RARs and VDRs. Both receptors form heterodimeric complexes with RXRs to bind with high affinity to their response elements (9)(10)(11)(12)(13)(14). Thus, for transcriptional activation by all-trans-RA and -VD, RXR-dependent pathways were postulated. However, recent evidence suggested that nuclear signaling of all-trans-RA and -VD is also transmitted through RXR-independent pathways (15,17)' presumably mediated by RAR and VDR homodimers. The specific ligand for RXRs was found to be a stereoisomer of all-truns-RA, 19). The notion that homodimers of RXRs also bind to DNA (20) constituted a third retinoid signaling pathway. All three RXR-dependent pathways are transcriptionally activated by 9-cis-RA (15,17,20).' Analogously, RAR-dependent pathways are induced by all-trans-RA (17); and VDR-dependent pathways are induced by VD (15). Interestingly, 9-cis-RA was also found to bind RARs (18, 19,21) but seems to act as a specific antagonist of their RXRindependent action.' In this way 9-cis-RA would not only positively coordinate the action of nuclear signaling of VD and all-trans-RA, it would also negatively regulate the activity of RAR homodimers.
Unlike for steroid hormone receptors, the natural response elements for RARs, RXRs, and VDRs include directly repeated rather than palindromic half-sites based on the hexamer RGGTCA. It was suggested that the spacing between the half-sites directs the binding specificity of RXR, VDR, and RAR. This rule (22,23) predicts that directly repeated motifs with spacings of one, three, and five base pairs are RXR, VDR, and RAR binding sites, respectively. Accordingly, the RA response element of the P2 promoter of the human RARB gene is a direct repeat of two RGTTCA motifs separated by five nucleotides (24). This very strong response element is bound by RAR homodimers, RXR homodimers, and RAR-RXR heterodimers (17)' and, therefore, responsive to all three retinoid signaling pathways.
The response element in the promoter of the human bone osteoblast-specific protein osteocalcin has been reported to be located at positions -510 to -488 (25) and -513 to -493 (26) mediating VD and all-trans-RA action (27). This response element contains two directly repeated copies of the half-site motif GGGTGA separated by six spacing nucleotides. Apparently, the spacing number is not in accordance with the spacing rule. Additional sequences downstream, however, with a third imperfect repeat separated by three nucleotides were also associated with VD responsiveness (28). Recently, it was reported that this extended osteocalcin response element (-510 to -485) is bound and activated by VDR homodimers but not by VDR-RXR heterodimers (15).
Here, we show the critical sequence parameters for the switch in specificity from retinoid to VD signaling pathways. We converted the RARB RA response element in several steps into the osteocalcin VD/RA response element. We found that RGGTGA motifs with five or six intervening nucleotides were bound by VDR homodimers. Neither RAR-RXR heterodimers nor RAR homodimers bound to such response elements. Moreover, only heterodimers of VDRs and RARs conferred responsiveness to all-trans-RA of the osteocalcin element. The interaction between VDRs and RARs suggests, in addition to the RXR-dependent action of VDRs, an alternative cross-talk between the nuclear signaling of VD and retinoids. Therefore, our data provide further combinatorial possibilities for the regulation of VD-and retinoid-responsive genes.

MATERIALS AND METHODS
Cell Culture-The human breast cancer cell line, MCF-7, was grown in RPMI 1640 medium supplemented with 10% fetal calf serum (FCS). The Drosophila cell line, SL-3 (29), was grown at room temperature in Schneider's medium (Life Technologies, Inc.) supplemented with 15% FCS.
DNA Constructs-The two 42-mer oligonucleotides containing the RARB gene promoter RA response element (24)

OLIGONUCLEOTIDES 1 AND 2
were synthesized on a Pharmacia Gene Assembler Plus apparatus, purified, phosphorylated, and annealed to yield the indicated doublestranded DNA fragment. This fragment was subcloned into the XbaI site of pBLCAT2 (30) upstream of the tk promoter driving the expression of the chloramphenicol acetyltransferase (CAT) gene. All other response elements (see Fig. 1 for their core sequences) had the same flanking sequences and were synthesized and cloned in the same way. RARa and RXRa cDNAs subcloned into pSG5 (Stratagene) were kindly provided by J. F. Grippo (Roche Nutley).
Transfection and CATAssays-For MCF-7 cells, 2 X lo5 cells were seeded into 6-well plates and grown overnight in phenol red-free RPMI supplemented with 10% charcoal-treated FCS. For SL-3 cells, 5 X 10' cells/6-well plate were grown overnight in Schneider's medium supplemented with 15% charcoal-treated FCS.
Liposomes were formed by incubating 1 pg of the reporter plasmid and 1 pg of the reference plasmid pCHllO (Pharmacia LKB Biotechnology Inc.) with 30 pg of N-[l-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium methylsulfate (DOTAP, Boehringer Mannheim) for 15 min at room temperature in a total volume of 100 pl. For the transfection of SL-3 cells, 0.25 pg of a pSG5-based (Stratagene) VDR-, RARa-, and RXRa-expression vector were also included. After dilution with 0.9 ml of phenol red-free RPMI (or Schneider's medium for SL-3 cells), the liposomes were added to the cells. One ml of phenol red-free RPMI supplemented with 20% charcoal-treated FCS (or 1 ml of Schneider's medium supplemented with 30% charcoaltreated FCS for SL-3 cells) was added 8 h after transfection. At this time VD, all-trans-RA, g-cis-RA, combinations of them (final concentrations, 1 p M each), or solvent was also added. The cells were harvested 16 h after initiation of the treatment and lysed in 100 p1 of 250 mM Tris-HC1, pH 7.8, by three cycles of freezing and thawing. For the CAT assay, 75 p1 of the supernatant were incubated with acetyl-coA and [14C]chloramphenicol for 1-5 h. The acetylation products were separated on TLC plates and quantitated by liquid scintillation counting. Protein concentrations were determined by the Bradford method (Bio-Rad), and @-galactosidase activity was measured at 420 nm after incubation of 20 pl of supernatant with onitrophenyl 8-D-galactopyranoside. Both the protein concentration and the 8-galactosidase activity were used for normalization of the CAT activity and yielded similar results. Stimulation factors were calculated as the ratio of CAT activity of ligand-stimulated cells to that of mock-induced control. Each condition was analyzed in triplicate, and data are shown as mean with standard deviation.
In Vitro Translation of Nuclear Receptors and DNA Binding Assays-Five pg of linearized cDNAs for VDR, RARa, and RXRm were used for in vitro transcription in a total volume of 50 pl containing 0.5 mM dNTPs, 40 units of RNasin, 40 mM Tris-HC1, pH 7.5, 6 m M MgCl,, 2 mM spermidine, 10 mM NaCl, 10 mM dithiothreitol, and 20 units of T7 RNA polymerase (all from Promega Corp.). The tran-scription reactions were incubated for 60 min at 37 "C, then 10 units of RQ1 RNase-free DNase I (Promega Corp.) were added, and the mixture was further incubated for 5 min at 37 "C. Five pg of each RNA were mixed with 175 pl of rabbit reticulocyte lysate, 100 units of RNasin and 20 p~ complete amino acid mixture (all from Promega Corp.) in a total volume of 250 p1 and incubated at 30 "C for 90 min. The response element probes were prepared by double digestion of the respective plasmid DNA with Hind111 and BamHI, purified by gel electrophoresis, and labeled by a fill-in reaction using [cx-~'P] dCTP and T7 DNA polymerase (Pharmacia). Five pl of in vitro translated receptors were preincubated for 10 min at room temperature in a total volume of 20 pl of binding buffer (10 mM Hepes pH 7.9, 80 mM KCl, 1 mM dithiothreitol, 0.2 pg/pl poly(dI/dC), and 5% glycerol). About 1 ng of labeled probe (25,000 cpm) was added, and the incubation was continued for 30 min. The protein-DNA complexes were resolved on a 5% non-denaturing polyacrylamide gel (at room temperature) in 0.5 X TBE (45 pM Tris borate, 1 pM EDTA, pH 8.3) (prerun for 2 h). The dried gel was exposed to a Kodak XAR-5 film.

RESULTS
Conversion of the RARB RA Response Element into the Osteocalcin V D Response Element-To define the sequences that are necessary and sufficient to provide a VD response, we mutagenized the RA response element within the promoter of the human RARp gene promoter (24) to convert it in several steps into the VD response element of the osteocalcin gene promoter (25,26). The RARB RA response element is a direct repeat of two RGTTCA motifs with five intervening nucleotides (Fig. hi). The osteocalcin response element, contains two directly repeated GGGTGA motifs with six spacing nucleotides (Fig. 1H). Therefore, we initially mutagenized the binding motifs and subsequently increased the spacing distance. The T residue in the third position was converted in both motifs into a G residue (B: DR5 reporter gene. Their transactivation potential was analyzed in the human breast cancer cell line MCF-7. This cell line expresses VDRs, RARs, and RXRs endogenously (15,17) and, therefore, can be induced by the specific ligands of these receptors, VD, all-trans-RA and 9-cis-RA. Transfected cells were stimulated with the three ligands, each alone and in the three possible combinations (Fig. 2). Short-term incubations (16 h) were performed with the ligands to diminish isomerization of the retinoids.' The maximal induction was 4.9-fold for VD, 10.1-fold for all-trans-RA, 9.8-fold for g-cis-RA, 10.3fold for VD and g-cis-RA, 19.8-fold for all-trans-and 9-cis-RA, and 10.2-fold for VD and all-trans-RA.
A Direct Repeat of the Motif RGGTGA Mediates Inducibility by VD- Fig. 2a shows that responsiveness to VD is linked to the presence of RGGTGA motifs within the response element (D-H). Spacing of five and six nucleotides mediates VD response with similar efficiency. However, the induction is increased by an additional 25% when the spacer contains an inverted copy of the RGGTGA motif, as in the two natural sequences, G and H. This augmentation could be related to the AP-1 site formed by these sequences (27). The two segments of the osteocalcin promoter, G and H, showed nearly identical responses to VD suggesting that, in contrast to former results (28), the third motif (GGGGCA) might not be involved in activation by VDRs.
The Osteocalcin Element Is Also Responsive to All-trans-RA-As shown previously (17) the first two mutations, B and C, of the RARp RA response element decreased its responsiveness to all-trans-RA (Fig. 26). The five VD response elements based on RGGTGA motifs (D-H) conferred induction by all-trans-RA. The response to 9-cis-RA of elements B a b and C was similar to that to all-trans-RA (Fig. 2c). As previously reported (15), the extended osteocalcin element (H) was also weakly induced by 9-cis-RA; however, the shorter osteocalcin element (G) and the elements D, E, and F did not respond to this ligand. The response of the eight elements to combined stimulation by VD and 9-cis-RA (Fig. ad) appears to represent a superposition of their responsiveness to the individual ligands (Figs. 1, a and c). The relative induction of the set of response elements by a combination of all-transand 9-cis-RA (Fig. 2e) was very similar to that obtained with 9-cis-RA alone (Fig. 2c); only the RARP element (A) and mutant B responded strongly. The stimulation pattern with VD and all-trans-RA (Fig. 2 f ) was nearly identical to that obtained with all-trans-RA alone (Fig. 2b).
Homodimer Activity of VDRs, RARs, and RXRs-The presence or absence of VDRs, RARs, and/or RXRs and their specific ligands seems to determine the signaling pathway through which VD-and retinoid-regulated genes are modulated. However, detailed studies in MCF-7 cells are hampered by the fact that these, as almost all mammalian cells, express the two retinoid receptor subfamilies. We have shown previously (15,17)'that in Drosophila SL-3 cells, which are devoid of endogenous VDRs, RARs, and RXRs, the VD and retinoid pathways can be reconstituted separately by transfection with appropriate receptor expression plasmids.
For all response elements tested, transient expression of VDR in SL-3 cells (Fig. 3a) gave a response pattern very similar to that of MCF-7 cells stimulated with VD (Fig. 2a). This not only substantiated the finding that the osteocalcin response element (H) is activated by VDRs in a RXR-independent way (15) but also established this for the other   elements with RGGTGA motifs (D-G). Gel shift experiments with in vitro translated VDRs suggested that these types of VD response elements were bound by VDR homodimers. When RARa was expressed, induction with all-trans-RA (Fig.  3b) yielded significant CAT activity only with the RARB RA response element (A) and mutant B, in accordance to the gel shift experiments. As reported recently (17), RAR homodimers bind to RA response elements, similar to that of the RARB gene promoter, but apparently they do not mediate the responsiveness to all-trans-RA of the osteocalcin element. The response pattern to 9-cis-RA of SL-3 cells expressing RXRa (Fig. 3c) resembles that established for 9-cis-RA-induced MCF-7 cells (Fig. 2c). RXR homodimers transactivated the RARB RA response element (A) and mutant B and, although less efficiently, the extended osteocalcin element (H) but not its shorter version (G). This suggested that the imperfect direct repeat, but not the RGGTGA motifs, promotes the binding of RXR homodimers. Gel shifts support this suggestion. As observed previously (15, 17); the in vitro binding of VDRs and RARs to their response elements is for some unknown reason much weaker than that of RXR homodimers or heterodimers.

FIG. 2. Ligand-dependent transactivation by endogenous receptors in MCF
VDR-RXR and RAR-RXR Heterodimer Activity--In SL-3 cells we expressed RAR, RXR, and VDR in the three combinations of two, thus creating by free association variations of homodimeric and heterodimeric receptor species. Subsequently, we treated these cells with the responsive ligands either individually or in appropriate combinations of two. In the presence of VDR and RXRa the response pattern after stimulation with VD alone (Fig. 4a, black columns) was nearly identical to that established in the presence of VDR alone (Fig. 3a). The response pattern to 9-cis-RA alone (Fig. 4a, light gray columns) was also very similar to that established after expression of RXRa alone (Fig. 8c). The response pattern after treatment with both ligands (Fig. 4a, white columns) was essentially identical to that observed for the response elements A-C after treatment with 9-cis-RA alone and to that observed for the response elements D-G after treatment with VD alone. Only the extended osteocalcin element (H), which is responsive to each of the two ligands, showed an increased induction by combined stimulation with VD and 9-cis-RA. This induction was only additive and not synergistic, as expected for the action of VDR-RXR heterodimers (15). Since homodimers of in vitro translated RXRs migrate in gel shift experiments slightly faster than homo-and heterodimers of VDRs and RARs, RXR homodimers could be discriminated in gel shift experiments from other protein complexes (Fig.  4a). A plausible interpretation is that the response elements A-C were bound by RXR homodimers, the elements D-G by VDR homodimers, and the extended osteocalcin element by both VDR and RXR homodimers but not by VDR-RXR heterodimers. The combined expression of RARa and RXRa followed by stimulation with all-trans-RA (Fig. 4b, dark gray columns) resulted in a response pattern similar to that observed in the presence of RARa alone (Fig. 3b). Likewise, the 9-cis-RA response pattern (light gray columns) is comparable with that observed in the presence of RXRa alone (Fig. 3c). For a detailed comparison the higher absolute stimulation rate for RAR-RXR heterodimers as compared with RAR and RXR homodimers has to be taken into account. The stimulation with both ligands (white columns) resulted in a synergistic response for element B, whereas for elements A, C, D, and E, only an additive effect was observed. As already shown in the previously mentioned experiments, the extended osteocalcin element (H) was not activated by all-trans-RA, and its weak activation by 9-cis-RA was mediated by RXR homodi- VDR-RAR Mediate the All-trans-RA Responsiueness of the Osteocalcin VD Response Element-Since potential isomerization of retinoids in our experimental system was kept at a minimum, responsiveness to all-trans-RA should be considered specifically mediated by the action of RARs. The experiments in Figs. 3b and 46 clearly indicated that RARs do not bind to the osteocalcin response element, neither as homodimers nor as RAR-RXR heterodimers. Therefore, we tested the third combinatorial possibility for heterodimer formation between VDRs, RARs, and RXRs, i.e. VDR-RAR heterodimers. When SL-3 cells were transfected with expression plasmids encoding VDR and RARcu, we observed for the elements D-G a response to VD (Fig. 4c compared with Figs. 3a and  4a; all black columns) probably mediated by VDR homodimers. In addition, these elements showed a response to all-trans-RA (Fig. 4c, darkgray column), which cannot be explained by the action of RAR homodimers (Fig. 3b). The three retinoid response elements A-C mediated the expected response by RAR homodimers. In the presence of both ligands (Fig. 4c, white columns) the response was clearly augmented, with the exception of element C. In comparison with the rather weak protein-DNA complexes formed by VDR and RAR homodimers (Figs. 3, a and b), a mixture of VDR and RAR provided strong binding complexes with the various response elements, indicative of their heterodimeric character. Surprisingly, these VDR-RAR heterodimers bound even better to the two retinoid response elements A and B than to the five VD response elements D-H. Since the retinoid response elements (A-C) had been shown not to be responsive to VD (Fig. 3a) even in the presence of endogenous receptors in MCF-7 cells (Fig.  Za), VDR-RAR heterodimers can be considered not to be inducible by VD (Fig. 4c, black columns for response elements A-C). However, the action of all-trans-RA-stimulated VDR-RAR heterodimers appears to be enhanced by VD (Fig. 4c, white columns).

DISCUSSION
We report the conversion of a response element specific for the three retinoid pathways into a response element that is activated through the RXR-independent pathway of VD action as well as through a pathway of coupled action of VD and all-trans-RA. Since in mammalian cell systems, like MCF-7 cells (Fig. 2), one cannot discriminate between the transcriptional effects of homo-and heterodimers of VDRs, RARs, and RXRs, we have used for our studies Drosophila SL-3 cells. We have previously shown (15,17) that SL-3 cells do not contain nuclear receptors that can functionally substitute for the mammalian RXR even though they express the ultraspiracle gene product, which is closely related to RXRs (31). Also putative "accessory" factors (32) do not bind 9-cis-RA or VD and therefore cannot be responsible for the observed synergistic effectg related to transcriptionally active heterodimers.
We found that activation by VDR homodimers is mediated only by response elements that are based on direct repeats of two RGGTGA motifs separated by five or six nucleotides (Fig.  3a). From the few known VD response elements only the human osteocalcin gene promoter contains this kind of direct repeat. Other VD response elements, as those in the promoters of the mouse osteopontin gene (33) or of the rat osteocalcin gene (34), consist of direct repeats with only three spacing nucleotides and are activated by VDR-RXR heterodimers (15). The extended human osteocalcin response element additionally contains an imperfect direct repeat with three intervening nucleotides and was, therefore, believed to bind VDR-RXR heterodimers responsible for VD action on the gene (28). However, we found this third motif not contributing to any enhanced response to VD but enabling the binding of RXR homodimers, thus evoking weak inducibility by 9-cis-RA to the response element.
Neither VDR-RXR and RAR-RXR heterodimers nor RAR homodimers bound the osteocalcin response element (Figs. 3b  and 4, a and b). The inducibility by all-trans-RA (Fig. 2b) of this response element can be explained, however, by the binding of VDR-RAR heterodimers (Fig. 4c). Before the discovery of RXRs as co-factors for efficient DNA binding of VDR (9, 11) there were several indications that VDR needed an accessory factor for its function (35-37). This co-factor had an estimated molecular mass of 55 kDa (36) being close to that of RXR and RAR. Furthermore, the accessory factor was found to enhance the binding of VDRs to the mouse RARP RA response element (36), as we observed the binding of VDR-RAR heterodimers to the human homologue of this response element (Fig. 4c). We believe therefore that, depending on the response element, VDR may select one of the two retinoid receptors as a co-factor. This idea is supported by the finding that, like RARs and RXRs, the unknown nuclear accessory factor was present in almost all tissues (36). Beside its ability to form heterodimers with RXRs, RARa was found to interact with several cell-type-specific proteins (38). Since the short as well as the extended human osteocalcin elements showed no difference in the binding of and transactivation by VDR-RAR heterodimers, we conclude that the heterodimeric complex binds preferentially to the direct repeat with a sixnucleotide spacer rather than to the imperfect direct repeat with a three nucleotide spacer similar to the VDR homodimer.
The notion that several members of the nuclear hormone receptor superfamily form heterodimers with RXR increased the number of transcription factors with potentially unique specificities. In this respect our observation of the biological function of VDR-RAR heterodimers on at least one natural response element extends the list of combinatorially active gene regulators (16). Therefore, one can think of up to six separate hormone response pathways that are governed by VD, all-trans-RA, and 9-cis-RA acting individually or by the three binary combinations. In cells in which only one of these ligands is present, the respective target genes will be activated only moderately. In the presence of a second ligand, however, some of these genes, whose promoters bind the respective heterodimers, will be more sensitive and achieve a greater maximal response. A possible function of such a mechanism could be controlling the switch from proliferation to differentiation, though the reality is certainly more complicated. There might be undiscovered VD-or retinoid-sensitive receptors or orphan receptors that interfere with the nuclear signaling by these ligands.