Cyclin D3 is a cofactor of retinoic acid receptors, modulating their activity in the presence of cellular retinoic acid-binding protein II.

Ligand-induced transcription activation of retinoic acid (RA) target genes by nuclear receptors (retinoic acid (RAR) and retinoid X (RXR) receptors) depends on the recruitment of coactivators. We have previously demonstrated that the small 15-kDa cellular RA-binding protein II (CRABPII) is a coactivator present in the RA-dependent nuclear complex. As identifying cell-specific partners of CRABPII might help to understand the novel control of RA signaling, we performed a yeast two-hybrid screen of a hematopoietic HL-60 cDNA library using human CRABPII as bait and have subsequently identified human cyclin D3 as a partner of CRABPII. Cyclin D3 interacted with CRABPII in a ligand-independent manner and equally bound RAR alpha, but not RXR alpha, and only in the presence of RA. We further show that cyclin D3 positively modulated RA-mediated transcription through CRABPII. Therefore, cyclin D3 may be part of a ternary complex with CRABPII and RAR. Finally, we show that cyclin D3 expression paralleled HL-60 differentiation and arrest of cell growth. These findings led us to speculate that control of cell proliferation during induction of differentiation may directly involve, at the transcriptional level, nuclear receptors, coactivators, and proteins of the cell cycle in a cell- and nuclear receptor-specific manner.

with ligand-binding (LBD) and DNA-binding (DBD) domains. In addition, they possess two autonomous transcription activation functions, AF-1 and AF-2. AF-1, located at the N-terminal end (A/B region), is ligand-independent; in contrast, AF-2, located in the C-terminal E region, is ligand-dependent (1).
Ligand binding induces conformational changes in the LBD involving helix H12, which encompasses the core of AF-2 of the nuclear receptor, resulting in the creation of a new surface for the recruitment of coactivators, such as proteins of the p160 family (SRC-1 (steroid receptor coactivator-1)/NCoA1, TIF2/ GRIP-1/SRC-2, and p/CIP/RAC3/ACTR/AIB-1/TRAM-1) and CBP (cAMP-response element-binding protein-binding protein)/p300 (2)(3)(4). These coactivators are also associated with other large histone acetyltransferase complexes (such as the p/CAF complex) that lead to chromatin decondensation (5,6). Nuclear receptors further recruit an additional complex variously termed TRAP, DRIP, SMCC, or Mediator, which establishes contacts with RNA polymerase II and the general transcription factors (7,8).
We previously showed that a small protein (15 kDa) belonging to the family of intracellular lipid-binding proteins that bind small hydrophobic molecules such as retinoids and fatty acids (9), cellular retinoic acid-binding protein II (CRABPII), also acts as a coactivator of nuclear retinoid receptors (10). Indeed, CRABPII can be found in the nucleus (10,11) and more specifically in the RA-dependent nuclear complex (10). Moreover, we (10) and others (12)(13)(14) have shown that overexpression of CRABPII enhances transactivation of RA target genes by RXR⅐RAR heterodimers either in transfected cells or in human hematopoietic cells. This coactivator effect results from a physical interaction of CRABPII with the LBD of RAR␣ or RXR␣ (15), which releases all-trans-RA from CRABPII to RAR␣ (13). To explain the function of this novel ligand-binding coactivator, we proposed the following scenario (15). First, holo-CRABPII docks to the apo-receptors bound to their promoters in the nucleus; the docking occurs around key structures of the ligand entrance pockets of CRABPII and the nuclear receptor. This establishes a channel that allows the release of RA from holo-CRABPII to apo-RAR␣ or apo-RXR␣. Because the CRAB-PII-nuclear RAR interaction does not require the presence of RA (10), apo-CRABPII could remain bound to the holo-receptor, preventing dissociation of RA from the nuclear receptor. Thus, CRABPII would increase the stability of the DNA-bound RXR⅐RAR complex (10), further contributing to the enhancement of RA-mediated transcription.
As identifying cell-specific partners of CRABPII in the RAdependent nuclear complex may help to understand the novel control of RA signaling, we performed a yeast two-hybrid screen of a hematopoietic HL-60 cDNA library using human CRABPII as bait and have subsequently identified human cyclin D3 as a novel partner of CRABPII. We found that cyclin D3 interacted with CRABPII in a ligand-independent manner. No interaction was noted with other D-type cyclins (D1 or D2). Interestingly, cyclin D3 was also found to interact with RAR␣, but not with RXR␣, and only in a RA-dependent manner. Cyclin D3 enhanced RA-mediated transactivation of RA target genes by increasing the interaction of CRABPII with RAR␣. Furthermore, we show that cyclin D3 expression paralleled the induction of differentiation in HL-60 cells. Our results identify cyclin D3 as a partner of a ternary complex with the coactivator CRABPII and the nuclear receptor RAR␣ and demonstrate another level of transcriptional control during RA-induced differentiation and arrest of cell growth.
Yeast Strains and Transformation-The yeast reporter strain Y190 (HIS3, lacZ, trp1, leu2, cyhr2) was previously described by Harper et al. (20). Transformation was carried out using the lithium acetate procedure (21), and expression of the fusion proteins was checked by immunoblotting. Standard media (Clontech) were used for yeast growth according to the manufacturer's protocols.
Two-hybrid Screening-An HL-60 Gal4AAD fusion library (a gift from J.-M. Garnier and P. Chambon) was inserted into the pACT2 vector, which contains the LEU2 marker and a cassette expressing nuclear localized Gal4AAD preceding a polylinker with cloning sites, and introduced into the Y190 reporter strain expressing hCRABPII from the pAS2-1 vector. The cells were then spread on Trp Ϫ /Leu Ϫ /His Ϫ plates containing 25 mM 3-amino-1,2,4-triazole (3-AT) and 1 M alltrans-RA. Plasmids from the isolated positive clones were recovered into Escherichia coli strain HB101 according to the protocol provided by Clontech and subjected to restriction analysis. pACT2 plasmids containing cDNA were identified by colony PCR using primers specific for the LEU2 gene contained in pACT2. The inserts were then sequenced.
Protein-Protein Interaction Using the Yeast Two-hybrid System-The yeast strain Y190 was co-transformed with pACT2 expressing cyclin D3 and with either hCRABPII or hRAR␣DEF in pAS2-1. The cells were plated in selective medium lacking tryptophan, leucine, and histidine and supplemented with 25 mM 3-AT and 1 M all-trans-RA. After 3 days, a qualitative ␤-galactosidase assay was performed.
Immunoblot Analysis-Protein extracts were quantified by the BCA protein assay (Pierce) and by Coomassie staining after separation by SDS-PAGE. The proteins were transferred onto a nitrocellulose membrane and incubated with a 1:200 dilution of anti-cyclin D1, D2, or D3 antibody; a 1:1000 dilution of anti-RAR antibody; a 1:500 dilution of anti-CRABPII antibody; or a 1:4000 dilution of anti-actin antibody. The proteins were identified by chemiluminescence.
In Vitro GST Pull-down Assays-GST and GST fusion proteins (GST-hRAR␣, GST-mRAR␣AB, GST-hRAR␣DEF, GST-hRAR␣DEF⌬(408 -416), GST-mRXR␣, and GST-hCRABPII) were expressed in E. coli and bound to glutathione-Sepharose beads (Amersham Biosciences). The GST proteins bound to the beads were then incubated with rabbit reticulocyte lysates containing translated 35 S-labeled cyclin D3 (T7 Quick coupled transcription/translation system, Promega, Madison, WI) or with extracts from COS-6 cells overexpressing cyclin D3 as described (10). Reactions were performed in the presence or absence of 1 M all-trans-RA. Bound proteins were recovered in SDS loading buffer, resolved by 12% SDS-PAGE, and analyzed by immunoblotting or by autoradiography of dried gels.
Immunoprecipitation-COS-6 cells were transiently transfected in the presence or absence of 1 M of all-trans-RA using the calcium phosphate precipitation technique as described (22). Whole cell extracts were prepared (10) and immunoprecipitated as follows. The extracts were first incubated with goat anti-cyclin D3 or D2 antibody for 1 h and then with protein G-Sepharose (Amersham Biosciences) for an additional hour. The immunocomplexes were recovered by centrifugation, washed, and resolved by SDS-PAGE. The immunoprecipitated proteins were revealed by immunoblotting and chemiluminescence.
HL-60 Cell Transfection-HL-60 cells were routinely maintained in RPMI 1640 medium (Invitrogen) supplemented with 15% fetal bovine serum, 2 mM glutamine, and 100 g/ml penicillin/streptomycin (all from BioWhittaker Europe, Verviers, Belgium). HL-60 cells were electroporated, as previously described (10), with the luciferase reporter gene (hRAR␤2-luciferase) in the presence or absence of the expression vectors for cyclin D or CRABPII. All transfections also contained the ␤-galactosidase expression vector (pCH110) as an internal standard and pRC/CMV as a carrier. Cells were treated or not with 1 M alltrans-RA for 24 h, and luciferase assays were performed according to a standard procedure (Promega). All results are expressed as -fold induction based on the basal activity of the reporter gene in the absence of RA and of any expression vector.
Differentiation of HL-60 Cells-Differentiation of HL-60 cells into granulocytes was induced by treatment with 1 M all-trans-RA or 9-cis-RA or 1.3% Me 2 SO. Nuclear extracts of control or treated cells were prepared after 1, 3, or 6 days; resolved by 12% SDS-PAGE; and tested for D-type cyclin and actin protein expression by immunoblotting. Cell differentiation was assayed by scoring the percentage of nitro blue tetrazolium-positive cells in treated cells versus untreated controls at 6 days after treatment. Cell growth was determined by counting the number of viable cells at the different time intervals with the trypan blue exclusion dye test.

Identification of Cyclin D3 as a CRABPII-binding Protein in
Yeast Two-hybrid Screening-The yeast two-hybrid system was used to identify proteins that interact with hCRABPII. Gal4DBD fused amino-terminally to hCRABPII in the pAS2-1 vector was expressed in yeast strain Y190, which contains two reporter genes: a HIS3 reporter gene and a lacZ reporter gene, both under the control of Gal4-binding sites. No growth was observed on plates lacking histidine (either in the absence or presence of 3-AT, a competitive inhibitor of the HIS3 gene product), indicating that hCRABPII on its own does not transactivate the HIS3 gene (data not shown). An HL-60 cDNA library was constructed in the pACT2 vector that directs synthesis of polypeptides fused to Gal4AAD. Y190 yeast cells expressing hCRABPII were transformed with this library. Approximately 1.0 ϫ 10 6 Y190 yeast transformants were spread on His Ϫ plates containing 25 mM 3-AT in the absence or presence of 1 M all-trans-RA. We analyzed colonies that grew on selective medium in the presence of 1 M all-trans-RA and that turned blue when tested in a ␤-galactosidase assay. Plasmids were recovered from the few positive clones, amplified, subjected to restriction analysis, and sequenced. Sequence comparison with the GenBank TM /EBI Data Bank and the Swiss Protein Database identified one 0.651-kb cDNA insert as the sequence encoding human cyclin D3 (accession number NM_001760). This cDNA fragment depicted the known open reading frame for cyclin D3, but lacked the nucleotides encoding the first 26 N-terminal amino acids and the last 33 Cterminal amino acids (Fig. 1A).
Cyclin D3 Interacts Specifically with CRABPII and the DEF Regions of RAR␣ in Yeast-To confirm the interaction of cyclin D3 with CRABPII, the Gal4AAD-cyclin D3 hybrid protein was expressed in the Y190 yeast strain in combination with pAS2-1CRABPII or the empty vector. The transformants were spread on His Ϫ plates containing 25 mM 3-AT and 1 M all-trans-RA. Both proteins interacted as evidenced by growth of colonies on histidine-deficient medium and activation of ␤-galactosidase activity (Fig. 1B, area 3). No clones were obtained with the unrelated bait vector pAS2-1 (Fig. 1B, area 2). These findings demonstrate that cyclin D3 interacts with CRABPII.
We also tested whether cyclin D3 is able to interact with the DEF regions of RAR␣. The Gal4DBD-RAR␣DEF hybrid protein was expressed in the Y190 yeast strain either alone or in combination with the Gal4AAD-cyclin D3 hybrid protein. Both proteins interacted as evidenced by growth of colonies on histidine-deficient medium and activation of ␤-galactosidase activity (Fig. 1B). No clones were obtained when Y190 yeast cells were transformed with the empty vector (pACT2) and either pAS2-1CRABPII or pAS2-1RAR␣DEF (data not shown). Together, these results indicate that cyclin D3 can interact with both CRABPII and the DEF regions of RAR␣ in yeast.
Cyclin D3 Interacts with CRABPII in the Presence or Absence of RA in Vitro-To study further the data obtained with the yeast two-hybrid system, binding assays with human cyclin D3 and hCRABPII were performed in vitro with recombinant proteins. 35 S-Labeled cyclin D3 produced by in vitro translation in rabbit reticulocytes was incubated with the GST-hCRABPII fusion protein attached to glutathione-Sepharose beads. Bound cyclin D3 was revealed by autoradiography after SDS-PAGE.
In agreement with the two-hybrid data, human cyclin D3 interacted with hCRABPII (Fig. 2). Beads loaded with the control GST protein did not retain cyclin D3 (Fig. 2, lane 2). This interaction occurred either in the presence (data not shown) or absence (Fig. 2, lane 3) of RA. No interactions were identified with various hCRABPII deletion mutants fused to GST corresponding to the nuclear receptor-interacting domains and the LBD of hCRABPII (15) (data not shown), suggesting that the complete structure of hCRABPII is essential for its interaction with cyclin D3. Interestingly, no interactions were observed under the same conditions with cyclins D1 and D2 (Fig. 2), suggesting that CRABPII interacts with only one type of D-type cyclin, cyclin D3. Cyclin D3 Interacts with the DEF Regions of RAR␣ in the Presence of RA in Vitro-Binding assays with cyclin D3 and the DEF regions of RAR␣ were also performed in vitro. Extracts from COS-6 cells overexpressing the different cyclins (D1-D3) were incubated with the GST-RAR␣ fusion protein attached to glutathione-Sepharose beads in the presence or absence of all-FIG. 1. Cyclin D3 interacts with CRABPII and RAR␣ in the yeast twohybrid system. A, the yeast two-hybrid screen using CRABPII as bait allowed us to isolate a clone encoded by a 0.651-kb cDNA fragment (yellow) with 100% identity to human cyclin D3. This cDNA fragment depicts the known open reading frame for cyclin D3. Protein sequence alignment revealed that the identified clone lacks the first 26 N-terminal residues as well as the last 33 C-terminal residues (yellow). B, cyclin D3 interacts with CRABPII and the DEF regions of RAR␣ in yeast. Individual colonies were streaked onto medium containing histidine (panel a) or lacking histidine (panels b and c) to characterize the plasmid combinations resulting in activation of the HIS3 gene. In panel c, ␤-galactosidase activity was tested to measure the activation of the lacZ reporter gene. Area 1, pAS2-1RAR␣DEF plus pACT2-cyclinD3; area 2, pAS2-1 plus pACT2-cyclinD3; area 3, pAS2-1CRABPII plus pACT2-cyclinD3. Specific interactions were observed only with the following hybrid pairs: RAR␣DEF plus cyclin D3 (area 2) and CRABPII plus cyclin D3 (area 3).
trans-RA (Fig. 3). Bound cyclins were revealed by immunoblotting. No interaction was detected in the absence of RA (Fig. 3A,  lane 2). However, a RA-dependent interaction was observed between RAR␣ and cyclin D3 (Fig. 3A, lane 3). No interaction was detected with cyclins D1 and D2 (data not shown).
To identify which RAR␣ domain is involved in the interaction with cyclin D3, several GST fusion proteins carrying deletions of RAR␣ were tested in the presence or absence of RA. Compared with full-length RAR␣, RAR␣ with its DBD and LBD deleted (RAR␣AB) (Fig. 3B) was impaired in its ability to interact with cyclin D3 (Fig. 3C, lanes 5 and 6). In contrast, the LBD of RAR␣ (RAR␣DEF) retained its capacity to bind cyclin D3 in the presence of RA (Fig. 3C, lanes 7 and 8). Similarly, a RAR␣DEF mutant bearing an internal deletion of the core motif (RAR␣DEF⌬(408 -416)) ( Fig. 3B) involved in the recruitment of coactivators retained the RA-dependent interaction with cyclin D3 (Fig. 3C, lanes 9 and 10). Altogether, these results suggest that the interaction with cyclin D3 involves a motif located in the LBD, distinct from the AF-2 domain core, but whose accessibility depends on RA binding. It must be noted that, under the same conditions, cyclin D3 did not interact with GST-RXR␣ even in the presence of 9-cis-RA (Fig. 3A,  lanes 4 and 5), indicating that cyclin D3 interacts in vitro specifically with RAR␣.
Human Cyclin D3 Co-immunoprecipitates with Both CRAB-PII and RAR␣ in Vivo-The interaction of cyclin D3 with CRABPII and RAR␣ was further investigated in co-immunoprecipitation experiments using COS-6 cells overexpressing cyclin D3 either alone or in combination with RAR␣ or CRAB-PII and treated or not with RA. Extracts were immunoprecipitated with anti-cyclin D3 antibody and analyzed by SDS-PAGE/immunoblotting (Fig. 4).
Anti-cyclin D3 antibody immunoprecipitated both cyclin D3 and RAR␣ from extracts of COS-6 cells cotransfected with the corresponding expression vectors and treated with RA (Fig.  4A). These proteins were not revealed in control immunoprecipitations. Similarly, anti-cyclin D3 antibody co-immunoprecipitated CRABPII when both proteins were coexpressed in COS-6 cells (data not shown). Such a co-immunoprecipitation occurred whether the cells were treated or not with RA, corroborating the results obtained in two-hybrid and GST pulldown assays.
When both RAR␣ and CRABPII expression vectors were cotransfected with cyclin D3 in COS-6 cells, anti-cyclin D3 antibody co-immunoprecipitated the three proteins (Fig. 4B). It must be noted that this co-immunoprecipitation was observed whether or not the cells were treated with RA. Such a ligandindependent co-immunoprecipitation of RAR␣ with cyclin D3 and CRABPII may reflect the RA-independent interaction of the receptor with CRABPII (10, 15). Thus, one can hypothesize that, in vivo in the absence of RA, CRABPII forms a bridge between RAR␣ and cyclin D3.
Cyclin D3 Enhances the Transactivation of RA Target Genes through CRABPII-We previously demonstrated that CRAB-PII acts as a coactivator for RA-mediated transactivation of target genes in HL-60 cells through its binding to RARs (10). Having shown herein that RAR␣ forms a complex not only with CRABPII, but also with cyclin D3, we studied the effects of cyclin D3 on RAR␣ transactivation in either the absence or  5) immobilized on glutathione-Sepharose beads in either the absence (lanes 1, 2, and 4) or presence of all-trans-RA (lane 3) or 9-cis-RA (lane 5). Bound human cyclin D3 was detected by immunoblotting. The cyclin D3 input corresponds to a doublet because anti-cyclin D3 antibody recognizes both exogenous human cyclin D3 (upper band) and endogenous simian cyclin D3 (lower band). B, schematic representation of the different RAR␣ deletion mutants fused to the GST protein (not drawn to scale). The gray boxes correspond to the AF-2 domain core (amino acids 408 -416) of RAR␣. C, the cyclin D3-RAR␣ interaction involves the DEF regions of RAR␣, but not the AF-2 activation domain core. Extracts from COS-6 cells transfected with the cyclin D3 expression vector were incubated with GST (lane 2) or the various GST-RAR␣ fusion proteins immobilized on glutathione-Sepharose beads: GST-RAR␣ (lanes 3 and 4), GST-RAR␣AB (lanes 5 and 6), GST-RAR␣DEF (lanes 7 and 8), or GST-RAR␣DEF(⌬408 -416) (lanes 9 and 10) in the absence (lanes 3, 5, 7, and 9) or presence (lanes 4, 6, 8, and 10) of all-trans-RA. Bound cyclin D3 was detected by immunoblotting.
presence of CRABPII. HL-60 cells were transiently transfected with the hRAR␤2-luciferase reporter construct in the absence or presence of the CRABPII and/or cyclin D3 expression vectors and treated with RA (Fig. 5). Surprisingly, the RA-induced luciferase activity was not affected upon overexpression of cyclin D3 (Fig. 5A). Similar results were obtained with different amounts of cyclin D3 expression vector (Fig. 5B). However, cyclin D3 significantly enhanced the increase in luciferase activity induced by CRABPII (Fig. 5). This effect was proportional to the amount of cyclin D3 expression vector (Fig. 5B). Cyclin D2 had no such effect (Fig. 5A). Altogether, these data indicate that cyclin D3 modulates the coactivator effect of CRABPII on the transactivation of RA target genes.
Cyclin D3 Stabilizes the CRABPII-RAR␣ Interaction-To investigate further how cyclin D3 enhances the transactivation of RA target genes through CRABPII, in vitro GST pull-down assays were performed. 35 S-Labeled RAR␣ produced by in vitro translation in rabbit reticulocytes was incubated with the GST-hCRABPII fusion protein immobilized on glutathione-Sepharose beads in the presence or absence of COS-6 extracts overexpressing cyclin D3 (Fig. 6). As expected, in the absence of cyclin D3, RAR␣ interacted with CRABPII, independently of RA (Fig. 6, lanes 2 and 3) (10). However, higher amounts of RAR␣ were retained in the presence of cyclin D3 in a RAindependent manner (Fig. 6, lanes 4 and 5). Therefore, cyclin D3 appears to increase the efficiency of the formation of the CRABPII⅐RAR complexes in vitro. These data confirm the results from the in vivo assay showing the formation of the ternary complex involving RAR, CRABPII, and cyclin D3 (Fig.  4B). Such a process might be required for the increase in the transactivation of RA target genes observed with cyclin D3 in the presence of CRABPII.  4). The immunocomplexes were resolved by SDS-12% PAGE, transferred to nitrocellulose membranes, and immunoblotted with anti-cyclin D3 or anti-RAR␣ antibody. B, COS-6 cells were cotransfected with the CRABPII, RAR␣, and cyclin D3 expression vectors and treated or not with all-trans-RA as indicated. Whole cell extracts were immunoprecipitated as described for A and immunoblotted with anti-cyclin D3, anti-RAR␣, or anti-CRABPII antibody.
FIG. 5. Cyclin D3 enhances RA-mediated transactivation through CRABPII. A, HL-60 cells were electroporated with the RAR␤2-luciferase reporter gene in association or not with the cyclin D3, cyclin D2, or CRABPII expression vector as indicated and treated or not with all-trans-RA (0.1 or 1 M) for 24 h. All transfections also contained the ␤-galactosidase expression vector (pCH110) as an internal standard and pRC/CMV as a carrier. Extracts were tested for luciferase activity. Results are expressed as -fold induction relative to the luciferase activity displayed in the absence of RA and in the absence of cyclins and CRABPII. All experiments were normalized to ␤-galactosidase. The data are representative of the average of three similar experiments done in triplicates. Means Ϯ S.D. are shown. B, HL-60 cells were transfected with the luciferase reporter gene in association or not with the expression vectors for CRABPII (5 g) and cyclin D3 (5 or 10 ng). Cells were treated with 1 M all-trans-RA, and extracts were analyzed for luciferase activity. Results are expressed as described for A.
Expression of Cyclin D3 Is Up-regulated by Retinoids in HL-60 Cells-To unravel the physiological relevance of the interactions between RAR␣, CRABPII, and cyclin D3, the nu-clear expression of cyclin D3 was examined in several RAsensitive hematopoietic cell lines: myeloblastic HL-60, promyelocytic NB4, and monoblastic U-937 (Fig. 7A). As shown by Western blot analysis, cyclin D3 was expressed in the nuclear compartment of all three cell lines, with the highest amounts being detected in U-937 cells (Fig. 7A). Cyclin D2 and, to a lesser extent, cyclin D1 were also expressed in the three cell lines (Fig. 7A). Because the induction of expression of cyclin D3 had previously been observed during differentiation of HL-60 cells by 12-O-tetradecanoylphorbol-13-acetate or Me 2 SO (23) and in F9 teratocarcinoma cells treated with different retinoids (24), we studied its expression in HL-60 cells after treatment with all-trans-RA (1 M), 9-cis-RA (1 M), or Me 2 SO (1.3%). Differentiation and cell growth were analyzed after 6 days of culture. Nuclear extracts of control and treated cells were prepared after 1, 3, or 6 days and immunoblotted with anti-cyclin D3 antibody (Fig. 7B). Both retinoids and Me 2 SO differentiated HL-60 cells as evidenced by the acquisition of redox function (nitro blue tetrazolium test). A weaker efficacy for all-trans-RA compared with 9-cis-RA was noted, as previously reported (25). Arrest of proliferation was achieved after 6 days of treatment (data not shown). Me 2 SO increased the expression of cyclin D3 from days 1 to 6 ( Fig. 7B, lanes 4, 8, and 12). Cyclin D3 expression was modulated by both retinoids, with higher levels achieved with 9-cis-RA compared with all-trans-RA from day 3 (Fig. 7B, lanes 7 and 11 and lanes 6 and 10, respectively). Likewise, the cyclin D3 up-regulation correlated with alltrans-RA or 9-cis-RA differentiation efficiency (Fig. 7B, lanes  10 and 11, respectively) and cell proliferation inhibition (data not shown).

DISCUSSION
In this study, we isolated human cyclin D3 as a CRABPIIinteracting protein. A physical RA-independent interaction between cyclin D3 and CRABPII was evidenced both in vitro and in vivo. No interactions were observed with various CRABPII deletion mutants, suggesting that the complete structure of CRABPII is essential for its interaction with cyclin D3. Furthermore, no interactions were identified between CRABPII and cyclin D1 or D2 in vitro. These data suggest that the retinoic acid-independent CRABPII-cyclin D3 interaction is specific for only one member of D-type cyclins.
Interestingly, we also demonstrated that cyclin D3 interacted with RAR␣, but not with RXR␣. Once again, cyclin D3 was the only D-type cyclin that could interact with RAR␣. This interaction involved the DEF region of RAR␣ and was observed in vitro as well as in yeast cells. Although the CRABPII-cyclin  1-4), day 3 (lanes 5-8), and day 6 (lanes 9 -12), nuclear extracts were immunoblotted with anti-cyclin D3, anti-D1, or anti-D2 antibody. Lanes 1, 5, and 9 correspond to control mock-treated cells. HL-60 differentiation was assayed by scoring the percentage of nitro blue tetrazolium (NBT)-positive cells in treated cells versus untreated controls at 6 days after treatment. D3 interaction was RA-independent, the cyclin D3-RAR␣ interaction required the presence of RA both in vitro and in vivo. Although cyclin D3 interacted physically with RAR␣ in a RAdependent manner, this interaction did not require the AF-2 "core" harboring the LLXXXL motif of RAR␣ generally implicated in the ligand-dependent interaction between the nuclear receptors and the LXXLL motifs of their coactivators. RA may, however, be required to induce conformational changes around the AF-2 domain to facilitate the cyclin D3-RAR␣ interaction. Surprisingly, the presence of CRABPII allowed cyclin D3 and RAR␣ to interact in the absence of RA. Because CRABPII interacted with both RAR␣ and cyclin D3 in a RA-independent manner, we hypothesize that CRABPII facilitates the recruitment of cyclin D3 on RAR␣ both in the absence and presence of RA.
As we have previously identified a coactivator function of CRABPII for the nuclear RARs (10), the interaction between cyclin D3, CRABPII, and RAR␣ led us to investigate the role of cyclin D3 in RA-mediated transactivation. On its own, cyclin D3 had no effect on RA-mediated transactivation. By contrast, in the presence of CRABPII, cyclin D3 positively modulated transcription activation. This may be due to an increase in the recruitment of CRABPII on RAR␣ or an increase in the stability of the CRABPII-RAR␣ interaction. Indeed, cyclin D3 appeared to increase the efficiency of the formation of the CRABPII⅐RAR complexes in vitro. Thus, cyclin D3 belongs to a ternary complex with RAR␣ and CRABPII and enhances the CRABPII coactivator transactivation activity. This is the first study demonstrating that cyclin D3 is involved in nuclear receptor transcriptional regulation. Another D-type cyclin (D1) has been described as a bridging factor between the estrogen receptor and its coactivators to positively regulate nuclear receptor-mediated transactivation. Similar functions can be found between cyclins D1 and D3, as cyclin D1 binds estrogen receptor-␣ in the absence of ligand and enhances estrogen-dependent transactivation (16,26,27) and binds the estrogen receptor and one of its coactivators, SRC-1 (28). Cyclin D1 may also promote or stabilize the association between the estrogen receptor and p/CAF, thereby increasing the histone acetyltransferase activity of the transcriptional machinery (29). Conversely, cyclin D1 has also been reported to bind the androgen receptor through the N-terminal domain and to inhibit transactivation by directly competing for p/CAF binding (30,31). Apart from being specific for RAR␣, cyclin D3 has other distinct features. In contrast to the interaction between cyclin D1 and SRC-1, which involves the LLXXXL motif of cyclin D1 and the LXXLL motif of SRC-1 (32), the interaction between cyclin D3 and CRABPII is different, as cyclin D3 does not harbor any LLXXXL motif and thus will not bind potential or candidate coactivators through their LXXLL motifs. Interestingly, CRABPII does not have a LXXLL motif, strongly suggesting that, for its interaction with cyclin D3, other recognition motifs are involved.
Hence, the involvement of a D-type cyclin in the transcriptional control of ligands directly implicated in cell growth and differentiation provides a new concept in which cell cycle regulatory proteins could play a dual function. Indeed, although D-type cyclins are essential regulatory subunits of the cyclindependent kinases (33)(34)(35) and operate in mid-to-late G 1 to allow cell progression in S phase (34, 36 -38), the transactivation function of cyclin D1 was shown to be independent of CDK4 and Rb phosphorylation (16,26,27). Although cyclins D1 and D2 have been extensively studied, few reports are available for cyclin D3. Structurally related to cyclins D1 and D2, cyclin D3 appears nevertheless to be more ubiquitously expressed. Furthermore, although its role in cell proliferation is clearly evidenced in lymphocytes, it is mainly shown to accumulate in terminally differentiated tissues (23). This characteristic was corroborated in vitro during differentiation of rodent myoblasts in which cyclin D3 expression was induced and reported to play an important role in irreversible cell cycle arrest of differentiated myocytes (39 -41). In this study, we confirm and extend previous data on HL-60 myeloid or F9 teratocarcinoma cells. Although all three cyclins could be detected in the nuclear compartment of different hematopoietic cells (HL-60, NB4, and U-937), only cyclin D3 expression paralleled the acquisition of differentiation features and the arrest of cell growth, as previously noted (23,24). Retinoids act as regulators of cell growth, differentiation, and apoptosis and have been shown to specifically arrest myeloid cells in the G 1 phase of the cell cycle (42). In the ML-1 myeloid cell line, the cell cycle arrest associated with all-trans-RA-induced differentiation involves regulation of the expression of cyclin D3 and cyclin kinase inhibitors (p18 and p21), affects the phosphorylation status of cyclin-dependent kinases, and ultimately triggers dephosphorylation of the Rb protein (43). Thus, in RAmediated myeloid differentiation, the increased levels of cyclin D3 contrast with the cells' progressive arrest in G 1 and point to a distinct effect on the establishment and/or maintenance of the differentiated status of the cell. The former may be explained by titration of cyclin D3 by either the cyclin D3-RAR or CRABPII-cyclin D3 interaction, whereas the latter may be related to cyclin D3 transcription activity on RA target genes involved in either differentiation (RAR, CRABPII, CD11b, Hox, and granulocyte colony-stimulating factor genes (G-CSF)) and/or cell cycle inhibition (p21 and p27 genes) (44).
In summary, we have identified in this study a novel level of molecular control associating three distinct partners of RAmediated transcription (cyclin D3, CRABPII, and RAR␣) that connects three key cellular processes: the cell cycle, ligand bioavailability, and gene expression. From our results and other published data (10,15,45), we may establish a physiological model (Fig. 8). After binding to RA, CRABPII translocates from the cytoplasm to the nucleus (45). Holo-CRABPII then docks to the apo-nuclear receptors bound to their promoters. Cyclin D3 forms a ternary complex and allows the stability of the CRABPII-RAR-RA interaction to increase and thus to enhance the transactivation activity of the complex.